SemaChecking.cpp revision 7864435ef2bce200224120bd1df3aed98ea5b99a
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/STLExtras.h" 34#include "llvm/Support/raw_ostream.h" 35#include "clang/Basic/TargetBuiltins.h" 36#include "clang/Basic/TargetInfo.h" 37#include "clang/Basic/ConvertUTF.h" 38#include <limits> 39using namespace clang; 40using namespace sema; 41 42SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 43 unsigned ByteNo) const { 44 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 45 PP.getLangOptions(), PP.getTargetInfo()); 46} 47 48 49/// CheckablePrintfAttr - does a function call have a "printf" attribute 50/// and arguments that merit checking? 51bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 52 if (Format->getType() == "printf") return true; 53 if (Format->getType() == "printf0") { 54 // printf0 allows null "format" string; if so don't check format/args 55 unsigned format_idx = Format->getFormatIdx() - 1; 56 // Does the index refer to the implicit object argument? 57 if (isa<CXXMemberCallExpr>(TheCall)) { 58 if (format_idx == 0) 59 return false; 60 --format_idx; 61 } 62 if (format_idx < TheCall->getNumArgs()) { 63 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 64 if (!Format->isNullPointerConstant(Context, 65 Expr::NPC_ValueDependentIsNull)) 66 return true; 67 } 68 } 69 return false; 70} 71 72/// Checks that a call expression's argument count is the desired number. 73/// This is useful when doing custom type-checking. Returns true on error. 74static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 75 unsigned argCount = call->getNumArgs(); 76 if (argCount == desiredArgCount) return false; 77 78 if (argCount < desiredArgCount) 79 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 80 << 0 /*function call*/ << desiredArgCount << argCount 81 << call->getSourceRange(); 82 83 // Highlight all the excess arguments. 84 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 85 call->getArg(argCount - 1)->getLocEnd()); 86 87 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 88 << 0 /*function call*/ << desiredArgCount << argCount 89 << call->getArg(1)->getSourceRange(); 90} 91 92/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 93/// annotation is a non wide string literal. 94static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 95 Arg = Arg->IgnoreParenCasts(); 96 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 97 if (!Literal || !Literal->isAscii()) { 98 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 99 << Arg->getSourceRange(); 100 return true; 101 } 102 return false; 103} 104 105ExprResult 106Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 107 ExprResult TheCallResult(Owned(TheCall)); 108 109 // Find out if any arguments are required to be integer constant expressions. 110 unsigned ICEArguments = 0; 111 ASTContext::GetBuiltinTypeError Error; 112 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 113 if (Error != ASTContext::GE_None) 114 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 115 116 // If any arguments are required to be ICE's, check and diagnose. 117 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 118 // Skip arguments not required to be ICE's. 119 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 120 121 llvm::APSInt Result; 122 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 123 return true; 124 ICEArguments &= ~(1 << ArgNo); 125 } 126 127 switch (BuiltinID) { 128 case Builtin::BI__builtin___CFStringMakeConstantString: 129 assert(TheCall->getNumArgs() == 1 && 130 "Wrong # arguments to builtin CFStringMakeConstantString"); 131 if (CheckObjCString(TheCall->getArg(0))) 132 return ExprError(); 133 break; 134 case Builtin::BI__builtin_stdarg_start: 135 case Builtin::BI__builtin_va_start: 136 if (SemaBuiltinVAStart(TheCall)) 137 return ExprError(); 138 break; 139 case Builtin::BI__builtin_isgreater: 140 case Builtin::BI__builtin_isgreaterequal: 141 case Builtin::BI__builtin_isless: 142 case Builtin::BI__builtin_islessequal: 143 case Builtin::BI__builtin_islessgreater: 144 case Builtin::BI__builtin_isunordered: 145 if (SemaBuiltinUnorderedCompare(TheCall)) 146 return ExprError(); 147 break; 148 case Builtin::BI__builtin_fpclassify: 149 if (SemaBuiltinFPClassification(TheCall, 6)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_isfinite: 153 case Builtin::BI__builtin_isinf: 154 case Builtin::BI__builtin_isinf_sign: 155 case Builtin::BI__builtin_isnan: 156 case Builtin::BI__builtin_isnormal: 157 if (SemaBuiltinFPClassification(TheCall, 1)) 158 return ExprError(); 159 break; 160 case Builtin::BI__builtin_shufflevector: 161 return SemaBuiltinShuffleVector(TheCall); 162 // TheCall will be freed by the smart pointer here, but that's fine, since 163 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 164 case Builtin::BI__builtin_prefetch: 165 if (SemaBuiltinPrefetch(TheCall)) 166 return ExprError(); 167 break; 168 case Builtin::BI__builtin_object_size: 169 if (SemaBuiltinObjectSize(TheCall)) 170 return ExprError(); 171 break; 172 case Builtin::BI__builtin_longjmp: 173 if (SemaBuiltinLongjmp(TheCall)) 174 return ExprError(); 175 break; 176 177 case Builtin::BI__builtin_classify_type: 178 if (checkArgCount(*this, TheCall, 1)) return true; 179 TheCall->setType(Context.IntTy); 180 break; 181 case Builtin::BI__builtin_constant_p: 182 if (checkArgCount(*this, TheCall, 1)) return true; 183 TheCall->setType(Context.IntTy); 184 break; 185 case Builtin::BI__sync_fetch_and_add: 186 case Builtin::BI__sync_fetch_and_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. 247static 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 273bool 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. 327bool 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 449bool 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 474ExprResult 475Sema::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. 621static 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, 646ExprResult 647Sema::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). 862bool 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. 892bool 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. 957bool 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. 1001bool 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. 1040ExprResult 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. 1125bool 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. 1162bool 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. 1181bool 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. 1199bool 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. 1215bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1216 bool HasVAListArg, 1217 unsigned format_idx, unsigned firstDataArg, 1218 bool isPrintf, bool inFunctionCall) { 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 inFunctionCall) 1232 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 1233 format_idx, firstDataArg, isPrintf, 1234 inFunctionCall); 1235 } 1236 1237 case Stmt::IntegerLiteralClass: 1238 // Technically -Wformat-nonliteral does not warn about this case. 1239 // The behavior of printf and friends in this case is implementation 1240 // dependent. Ideally if the format string cannot be null then 1241 // it should have a 'nonnull' attribute in the function prototype. 1242 return true; 1243 1244 case Stmt::ImplicitCastExprClass: { 1245 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1246 goto tryAgain; 1247 } 1248 1249 case Stmt::OpaqueValueExprClass: 1250 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1251 E = src; 1252 goto tryAgain; 1253 } 1254 return false; 1255 1256 case Stmt::PredefinedExprClass: 1257 // While __func__, etc., are technically not string literals, they 1258 // cannot contain format specifiers and thus are not a security 1259 // liability. 1260 return true; 1261 1262 case Stmt::DeclRefExprClass: { 1263 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1264 1265 // As an exception, do not flag errors for variables binding to 1266 // const string literals. 1267 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1268 bool isConstant = false; 1269 QualType T = DR->getType(); 1270 1271 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1272 isConstant = AT->getElementType().isConstant(Context); 1273 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1274 isConstant = T.isConstant(Context) && 1275 PT->getPointeeType().isConstant(Context); 1276 } 1277 1278 if (isConstant) { 1279 if (const Expr *Init = VD->getAnyInitializer()) 1280 return SemaCheckStringLiteral(Init, TheCall, 1281 HasVAListArg, format_idx, firstDataArg, 1282 isPrintf, /*inFunctionCall*/false); 1283 } 1284 1285 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1286 // special check to see if the format string is a function parameter 1287 // of the function calling the printf function. If the function 1288 // has an attribute indicating it is a printf-like function, then we 1289 // should suppress warnings concerning non-literals being used in a call 1290 // to a vprintf function. For example: 1291 // 1292 // void 1293 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1294 // va_list ap; 1295 // va_start(ap, fmt); 1296 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1297 // ... 1298 // 1299 // 1300 // FIXME: We don't have full attribute support yet, so just check to see 1301 // if the argument is a DeclRefExpr that references a parameter. We'll 1302 // add proper support for checking the attribute later. 1303 if (HasVAListArg) 1304 if (isa<ParmVarDecl>(VD)) 1305 return true; 1306 } 1307 1308 return false; 1309 } 1310 1311 case Stmt::CallExprClass: { 1312 const CallExpr *CE = cast<CallExpr>(E); 1313 if (const ImplicitCastExpr *ICE 1314 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1315 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1316 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1317 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1318 unsigned ArgIndex = FA->getFormatIdx(); 1319 const Expr *Arg = CE->getArg(ArgIndex - 1); 1320 1321 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1322 format_idx, firstDataArg, isPrintf, 1323 inFunctionCall); 1324 } 1325 } 1326 } 1327 } 1328 1329 return false; 1330 } 1331 case Stmt::ObjCStringLiteralClass: 1332 case Stmt::StringLiteralClass: { 1333 const StringLiteral *StrE = NULL; 1334 1335 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1336 StrE = ObjCFExpr->getString(); 1337 else 1338 StrE = cast<StringLiteral>(E); 1339 1340 if (StrE) { 1341 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1342 firstDataArg, isPrintf, inFunctionCall); 1343 return true; 1344 } 1345 1346 return false; 1347 } 1348 1349 default: 1350 return false; 1351 } 1352} 1353 1354void 1355Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1356 const Expr * const *ExprArgs, 1357 SourceLocation CallSiteLoc) { 1358 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1359 e = NonNull->args_end(); 1360 i != e; ++i) { 1361 const Expr *ArgExpr = ExprArgs[*i]; 1362 if (ArgExpr->isNullPointerConstant(Context, 1363 Expr::NPC_ValueDependentIsNotNull)) 1364 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1365 } 1366} 1367 1368/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1369/// functions) for correct use of format strings. 1370void 1371Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1372 unsigned format_idx, unsigned firstDataArg, 1373 bool isPrintf) { 1374 1375 const Expr *Fn = TheCall->getCallee(); 1376 1377 // The way the format attribute works in GCC, the implicit this argument 1378 // of member functions is counted. However, it doesn't appear in our own 1379 // lists, so decrement format_idx in that case. 1380 if (isa<CXXMemberCallExpr>(TheCall)) { 1381 const CXXMethodDecl *method_decl = 1382 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1383 if (method_decl && method_decl->isInstance()) { 1384 // Catch a format attribute mistakenly referring to the object argument. 1385 if (format_idx == 0) 1386 return; 1387 --format_idx; 1388 if(firstDataArg != 0) 1389 --firstDataArg; 1390 } 1391 } 1392 1393 // CHECK: printf/scanf-like function is called with no format string. 1394 if (format_idx >= TheCall->getNumArgs()) { 1395 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1396 << Fn->getSourceRange(); 1397 return; 1398 } 1399 1400 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1401 1402 // CHECK: format string is not a string literal. 1403 // 1404 // Dynamically generated format strings are difficult to 1405 // automatically vet at compile time. Requiring that format strings 1406 // are string literals: (1) permits the checking of format strings by 1407 // the compiler and thereby (2) can practically remove the source of 1408 // many format string exploits. 1409 1410 // Format string can be either ObjC string (e.g. @"%d") or 1411 // C string (e.g. "%d") 1412 // ObjC string uses the same format specifiers as C string, so we can use 1413 // the same format string checking logic for both ObjC and C strings. 1414 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1415 firstDataArg, isPrintf)) 1416 return; // Literal format string found, check done! 1417 1418 // If there are no arguments specified, warn with -Wformat-security, otherwise 1419 // warn only with -Wformat-nonliteral. 1420 if (TheCall->getNumArgs() == format_idx+1) 1421 Diag(TheCall->getArg(format_idx)->getLocStart(), 1422 diag::warn_format_nonliteral_noargs) 1423 << OrigFormatExpr->getSourceRange(); 1424 else 1425 Diag(TheCall->getArg(format_idx)->getLocStart(), 1426 diag::warn_format_nonliteral) 1427 << OrigFormatExpr->getSourceRange(); 1428} 1429 1430namespace { 1431class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1432protected: 1433 Sema &S; 1434 const StringLiteral *FExpr; 1435 const Expr *OrigFormatExpr; 1436 const unsigned FirstDataArg; 1437 const unsigned NumDataArgs; 1438 const bool IsObjCLiteral; 1439 const char *Beg; // Start of format string. 1440 const bool HasVAListArg; 1441 const CallExpr *TheCall; 1442 unsigned FormatIdx; 1443 llvm::BitVector CoveredArgs; 1444 bool usesPositionalArgs; 1445 bool atFirstArg; 1446 bool inFunctionCall; 1447public: 1448 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1449 const Expr *origFormatExpr, unsigned firstDataArg, 1450 unsigned numDataArgs, bool isObjCLiteral, 1451 const char *beg, bool hasVAListArg, 1452 const CallExpr *theCall, unsigned formatIdx, 1453 bool inFunctionCall) 1454 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1455 FirstDataArg(firstDataArg), 1456 NumDataArgs(numDataArgs), 1457 IsObjCLiteral(isObjCLiteral), Beg(beg), 1458 HasVAListArg(hasVAListArg), 1459 TheCall(theCall), FormatIdx(formatIdx), 1460 usesPositionalArgs(false), atFirstArg(true), 1461 inFunctionCall(inFunctionCall) { 1462 CoveredArgs.resize(numDataArgs); 1463 CoveredArgs.reset(); 1464 } 1465 1466 void DoneProcessing(); 1467 1468 void HandleIncompleteSpecifier(const char *startSpecifier, 1469 unsigned specifierLen); 1470 1471 virtual void HandleInvalidPosition(const char *startSpecifier, 1472 unsigned specifierLen, 1473 analyze_format_string::PositionContext p); 1474 1475 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1476 1477 void HandleNullChar(const char *nullCharacter); 1478 1479 template <typename Range> 1480 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1481 const Expr *ArgumentExpr, 1482 PartialDiagnostic PDiag, 1483 SourceLocation StringLoc, 1484 bool IsStringLocation, Range StringRange, 1485 FixItHint Fixit = FixItHint()); 1486 1487protected: 1488 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1489 const char *startSpec, 1490 unsigned specifierLen, 1491 const char *csStart, unsigned csLen); 1492 1493 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1494 const char *startSpec, 1495 unsigned specifierLen); 1496 1497 SourceRange getFormatStringRange(); 1498 CharSourceRange getSpecifierRange(const char *startSpecifier, 1499 unsigned specifierLen); 1500 SourceLocation getLocationOfByte(const char *x); 1501 1502 const Expr *getDataArg(unsigned i) const; 1503 1504 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1505 const analyze_format_string::ConversionSpecifier &CS, 1506 const char *startSpecifier, unsigned specifierLen, 1507 unsigned argIndex); 1508 1509 template <typename Range> 1510 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1511 bool IsStringLocation, Range StringRange, 1512 FixItHint Fixit = FixItHint()); 1513 1514 void CheckPositionalAndNonpositionalArgs( 1515 const analyze_format_string::FormatSpecifier *FS); 1516}; 1517} 1518 1519SourceRange CheckFormatHandler::getFormatStringRange() { 1520 return OrigFormatExpr->getSourceRange(); 1521} 1522 1523CharSourceRange CheckFormatHandler:: 1524getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1525 SourceLocation Start = getLocationOfByte(startSpecifier); 1526 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1527 1528 // Advance the end SourceLocation by one due to half-open ranges. 1529 End = End.getLocWithOffset(1); 1530 1531 return CharSourceRange::getCharRange(Start, End); 1532} 1533 1534SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1535 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1536} 1537 1538void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1539 unsigned specifierLen){ 1540 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1541 getLocationOfByte(startSpecifier), 1542 /*IsStringLocation*/true, 1543 getSpecifierRange(startSpecifier, specifierLen)); 1544} 1545 1546void 1547CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1548 analyze_format_string::PositionContext p) { 1549 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1550 << (unsigned) p, 1551 getLocationOfByte(startPos), /*IsStringLocation*/true, 1552 getSpecifierRange(startPos, posLen)); 1553} 1554 1555void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1556 unsigned posLen) { 1557 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1558 getLocationOfByte(startPos), 1559 /*IsStringLocation*/true, 1560 getSpecifierRange(startPos, posLen)); 1561} 1562 1563void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1564 if (!IsObjCLiteral) { 1565 // The presence of a null character is likely an error. 1566 EmitFormatDiagnostic( 1567 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1568 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1569 getFormatStringRange()); 1570 } 1571} 1572 1573const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1574 return TheCall->getArg(FirstDataArg + i); 1575} 1576 1577void CheckFormatHandler::DoneProcessing() { 1578 // Does the number of data arguments exceed the number of 1579 // format conversions in the format string? 1580 if (!HasVAListArg) { 1581 // Find any arguments that weren't covered. 1582 CoveredArgs.flip(); 1583 signed notCoveredArg = CoveredArgs.find_first(); 1584 if (notCoveredArg >= 0) { 1585 assert((unsigned)notCoveredArg < NumDataArgs); 1586 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1587 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1588 /*IsStringLocation*/false, getFormatStringRange()); 1589 } 1590 } 1591} 1592 1593bool 1594CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1595 SourceLocation Loc, 1596 const char *startSpec, 1597 unsigned specifierLen, 1598 const char *csStart, 1599 unsigned csLen) { 1600 1601 bool keepGoing = true; 1602 if (argIndex < NumDataArgs) { 1603 // Consider the argument coverered, even though the specifier doesn't 1604 // make sense. 1605 CoveredArgs.set(argIndex); 1606 } 1607 else { 1608 // If argIndex exceeds the number of data arguments we 1609 // don't issue a warning because that is just a cascade of warnings (and 1610 // they may have intended '%%' anyway). We don't want to continue processing 1611 // the format string after this point, however, as we will like just get 1612 // gibberish when trying to match arguments. 1613 keepGoing = false; 1614 } 1615 1616 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1617 << StringRef(csStart, csLen), 1618 Loc, /*IsStringLocation*/true, 1619 getSpecifierRange(startSpec, specifierLen)); 1620 1621 return keepGoing; 1622} 1623 1624void 1625CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1626 const char *startSpec, 1627 unsigned specifierLen) { 1628 EmitFormatDiagnostic( 1629 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1630 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1631} 1632 1633bool 1634CheckFormatHandler::CheckNumArgs( 1635 const analyze_format_string::FormatSpecifier &FS, 1636 const analyze_format_string::ConversionSpecifier &CS, 1637 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1638 1639 if (argIndex >= NumDataArgs) { 1640 PartialDiagnostic PDiag = FS.usesPositionalArg() 1641 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1642 << (argIndex+1) << NumDataArgs) 1643 : S.PDiag(diag::warn_printf_insufficient_data_args); 1644 EmitFormatDiagnostic( 1645 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1646 getSpecifierRange(startSpecifier, specifierLen)); 1647 return false; 1648 } 1649 return true; 1650} 1651 1652template<typename Range> 1653void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1654 SourceLocation Loc, 1655 bool IsStringLocation, 1656 Range StringRange, 1657 FixItHint FixIt) { 1658 EmitFormatDiagnostic(S, inFunctionCall, TheCall->getArg(FormatIdx), PDiag, 1659 Loc, IsStringLocation, StringRange, FixIt); 1660} 1661 1662/// \brief If the format string is not within the funcion call, emit a note 1663/// so that the function call and string are in diagnostic messages. 1664/// 1665/// \param inFunctionCall if true, the format string is within the function 1666/// call and only one diagnostic message will be produced. Otherwise, an 1667/// extra note will be emitted pointing to location of the format string. 1668/// 1669/// \param ArgumentExpr the expression that is passed as the format string 1670/// argument in the function call. Used for getting locations when two 1671/// diagnostics are emitted. 1672/// 1673/// \param PDiag the callee should already have provided any strings for the 1674/// diagnostic message. This function only adds locations and fixits 1675/// to diagnostics. 1676/// 1677/// \param Loc primary location for diagnostic. If two diagnostics are 1678/// required, one will be at Loc and a new SourceLocation will be created for 1679/// the other one. 1680/// 1681/// \param IsStringLocation if true, Loc points to the format string should be 1682/// used for the note. Otherwise, Loc points to the argument list and will 1683/// be used with PDiag. 1684/// 1685/// \param StringRange some or all of the string to highlight. This is 1686/// templated so it can accept either a CharSourceRange or a SourceRange. 1687/// 1688/// \param Fixit optional fix it hint for the format string. 1689template<typename Range> 1690void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1691 const Expr *ArgumentExpr, 1692 PartialDiagnostic PDiag, 1693 SourceLocation Loc, 1694 bool IsStringLocation, 1695 Range StringRange, 1696 FixItHint FixIt) { 1697 if (InFunctionCall) 1698 S.Diag(Loc, PDiag) << StringRange << FixIt; 1699 else { 1700 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1701 << ArgumentExpr->getSourceRange(); 1702 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1703 diag::note_format_string_defined) 1704 << StringRange << FixIt; 1705 } 1706} 1707 1708//===--- CHECK: Printf format string checking ------------------------------===// 1709 1710namespace { 1711class CheckPrintfHandler : public CheckFormatHandler { 1712public: 1713 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1714 const Expr *origFormatExpr, unsigned firstDataArg, 1715 unsigned numDataArgs, bool isObjCLiteral, 1716 const char *beg, bool hasVAListArg, 1717 const CallExpr *theCall, unsigned formatIdx, 1718 bool inFunctionCall) 1719 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1720 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1721 theCall, formatIdx, inFunctionCall) {} 1722 1723 1724 bool HandleInvalidPrintfConversionSpecifier( 1725 const analyze_printf::PrintfSpecifier &FS, 1726 const char *startSpecifier, 1727 unsigned specifierLen); 1728 1729 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1730 const char *startSpecifier, 1731 unsigned specifierLen); 1732 1733 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1734 const char *startSpecifier, unsigned specifierLen); 1735 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1736 const analyze_printf::OptionalAmount &Amt, 1737 unsigned type, 1738 const char *startSpecifier, unsigned specifierLen); 1739 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1740 const analyze_printf::OptionalFlag &flag, 1741 const char *startSpecifier, unsigned specifierLen); 1742 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1743 const analyze_printf::OptionalFlag &ignoredFlag, 1744 const analyze_printf::OptionalFlag &flag, 1745 const char *startSpecifier, unsigned specifierLen); 1746}; 1747} 1748 1749bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1750 const analyze_printf::PrintfSpecifier &FS, 1751 const char *startSpecifier, 1752 unsigned specifierLen) { 1753 const analyze_printf::PrintfConversionSpecifier &CS = 1754 FS.getConversionSpecifier(); 1755 1756 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1757 getLocationOfByte(CS.getStart()), 1758 startSpecifier, specifierLen, 1759 CS.getStart(), CS.getLength()); 1760} 1761 1762bool CheckPrintfHandler::HandleAmount( 1763 const analyze_format_string::OptionalAmount &Amt, 1764 unsigned k, const char *startSpecifier, 1765 unsigned specifierLen) { 1766 1767 if (Amt.hasDataArgument()) { 1768 if (!HasVAListArg) { 1769 unsigned argIndex = Amt.getArgIndex(); 1770 if (argIndex >= NumDataArgs) { 1771 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 1772 << k, 1773 getLocationOfByte(Amt.getStart()), 1774 /*IsStringLocation*/true, 1775 getSpecifierRange(startSpecifier, specifierLen)); 1776 // Don't do any more checking. We will just emit 1777 // spurious errors. 1778 return false; 1779 } 1780 1781 // Type check the data argument. It should be an 'int'. 1782 // Although not in conformance with C99, we also allow the argument to be 1783 // an 'unsigned int' as that is a reasonably safe case. GCC also 1784 // doesn't emit a warning for that case. 1785 CoveredArgs.set(argIndex); 1786 const Expr *Arg = getDataArg(argIndex); 1787 QualType T = Arg->getType(); 1788 1789 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1790 assert(ATR.isValid()); 1791 1792 if (!ATR.matchesType(S.Context, T)) { 1793 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 1794 << k << ATR.getRepresentativeType(S.Context) 1795 << T << Arg->getSourceRange(), 1796 getLocationOfByte(Amt.getStart()), 1797 /*IsStringLocation*/true, 1798 getSpecifierRange(startSpecifier, specifierLen)); 1799 // Don't do any more checking. We will just emit 1800 // spurious errors. 1801 return false; 1802 } 1803 } 1804 } 1805 return true; 1806} 1807 1808void CheckPrintfHandler::HandleInvalidAmount( 1809 const analyze_printf::PrintfSpecifier &FS, 1810 const analyze_printf::OptionalAmount &Amt, 1811 unsigned type, 1812 const char *startSpecifier, 1813 unsigned specifierLen) { 1814 const analyze_printf::PrintfConversionSpecifier &CS = 1815 FS.getConversionSpecifier(); 1816 1817 FixItHint fixit = 1818 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 1819 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1820 Amt.getConstantLength())) 1821 : FixItHint(); 1822 1823 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 1824 << type << CS.toString(), 1825 getLocationOfByte(Amt.getStart()), 1826 /*IsStringLocation*/true, 1827 getSpecifierRange(startSpecifier, specifierLen), 1828 fixit); 1829} 1830 1831void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1832 const analyze_printf::OptionalFlag &flag, 1833 const char *startSpecifier, 1834 unsigned specifierLen) { 1835 // Warn about pointless flag with a fixit removal. 1836 const analyze_printf::PrintfConversionSpecifier &CS = 1837 FS.getConversionSpecifier(); 1838 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 1839 << flag.toString() << CS.toString(), 1840 getLocationOfByte(flag.getPosition()), 1841 /*IsStringLocation*/true, 1842 getSpecifierRange(startSpecifier, specifierLen), 1843 FixItHint::CreateRemoval( 1844 getSpecifierRange(flag.getPosition(), 1))); 1845} 1846 1847void CheckPrintfHandler::HandleIgnoredFlag( 1848 const analyze_printf::PrintfSpecifier &FS, 1849 const analyze_printf::OptionalFlag &ignoredFlag, 1850 const analyze_printf::OptionalFlag &flag, 1851 const char *startSpecifier, 1852 unsigned specifierLen) { 1853 // Warn about ignored flag with a fixit removal. 1854 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 1855 << ignoredFlag.toString() << flag.toString(), 1856 getLocationOfByte(ignoredFlag.getPosition()), 1857 /*IsStringLocation*/true, 1858 getSpecifierRange(startSpecifier, specifierLen), 1859 FixItHint::CreateRemoval( 1860 getSpecifierRange(ignoredFlag.getPosition(), 1))); 1861} 1862 1863bool 1864CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1865 &FS, 1866 const char *startSpecifier, 1867 unsigned specifierLen) { 1868 1869 using namespace analyze_format_string; 1870 using namespace analyze_printf; 1871 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1872 1873 if (FS.consumesDataArgument()) { 1874 if (atFirstArg) { 1875 atFirstArg = false; 1876 usesPositionalArgs = FS.usesPositionalArg(); 1877 } 1878 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1879 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 1880 startSpecifier, specifierLen); 1881 return false; 1882 } 1883 } 1884 1885 // First check if the field width, precision, and conversion specifier 1886 // have matching data arguments. 1887 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1888 startSpecifier, specifierLen)) { 1889 return false; 1890 } 1891 1892 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1893 startSpecifier, specifierLen)) { 1894 return false; 1895 } 1896 1897 if (!CS.consumesDataArgument()) { 1898 // FIXME: Technically specifying a precision or field width here 1899 // makes no sense. Worth issuing a warning at some point. 1900 return true; 1901 } 1902 1903 // Consume the argument. 1904 unsigned argIndex = FS.getArgIndex(); 1905 if (argIndex < NumDataArgs) { 1906 // The check to see if the argIndex is valid will come later. 1907 // We set the bit here because we may exit early from this 1908 // function if we encounter some other error. 1909 CoveredArgs.set(argIndex); 1910 } 1911 1912 // Check for using an Objective-C specific conversion specifier 1913 // in a non-ObjC literal. 1914 if (!IsObjCLiteral && CS.isObjCArg()) { 1915 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1916 specifierLen); 1917 } 1918 1919 // Check for invalid use of field width 1920 if (!FS.hasValidFieldWidth()) { 1921 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1922 startSpecifier, specifierLen); 1923 } 1924 1925 // Check for invalid use of precision 1926 if (!FS.hasValidPrecision()) { 1927 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1928 startSpecifier, specifierLen); 1929 } 1930 1931 // Check each flag does not conflict with any other component. 1932 if (!FS.hasValidThousandsGroupingPrefix()) 1933 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1934 if (!FS.hasValidLeadingZeros()) 1935 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1936 if (!FS.hasValidPlusPrefix()) 1937 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1938 if (!FS.hasValidSpacePrefix()) 1939 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1940 if (!FS.hasValidAlternativeForm()) 1941 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1942 if (!FS.hasValidLeftJustified()) 1943 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1944 1945 // Check that flags are not ignored by another flag 1946 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1947 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1948 startSpecifier, specifierLen); 1949 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1950 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1951 startSpecifier, specifierLen); 1952 1953 // Check the length modifier is valid with the given conversion specifier. 1954 const LengthModifier &LM = FS.getLengthModifier(); 1955 if (!FS.hasValidLengthModifier()) 1956 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 1957 << LM.toString() << CS.toString(), 1958 getLocationOfByte(LM.getStart()), 1959 /*IsStringLocation*/true, 1960 getSpecifierRange(startSpecifier, specifierLen), 1961 FixItHint::CreateRemoval( 1962 getSpecifierRange(LM.getStart(), 1963 LM.getLength()))); 1964 1965 // Are we using '%n'? 1966 if (CS.getKind() == ConversionSpecifier::nArg) { 1967 // Issue a warning about this being a possible security issue. 1968 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 1969 getLocationOfByte(CS.getStart()), 1970 /*IsStringLocation*/true, 1971 getSpecifierRange(startSpecifier, specifierLen)); 1972 // Continue checking the other format specifiers. 1973 return true; 1974 } 1975 1976 // The remaining checks depend on the data arguments. 1977 if (HasVAListArg) 1978 return true; 1979 1980 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1981 return false; 1982 1983 // Now type check the data expression that matches the 1984 // format specifier. 1985 const Expr *Ex = getDataArg(argIndex); 1986 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1987 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1988 // Check if we didn't match because of an implicit cast from a 'char' 1989 // or 'short' to an 'int'. This is done because printf is a varargs 1990 // function. 1991 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1992 if (ICE->getType() == S.Context.IntTy) { 1993 // All further checking is done on the subexpression. 1994 Ex = ICE->getSubExpr(); 1995 if (ATR.matchesType(S.Context, Ex->getType())) 1996 return true; 1997 } 1998 1999 // We may be able to offer a FixItHint if it is a supported type. 2000 PrintfSpecifier fixedFS = FS; 2001 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2002 2003 if (success) { 2004 // Get the fix string from the fixed format specifier 2005 llvm::SmallString<128> buf; 2006 llvm::raw_svector_ostream os(buf); 2007 fixedFS.toString(os); 2008 2009 // FIXME: getRepresentativeType() perhaps should return a string 2010 // instead of a QualType to better handle when the representative 2011 // type is 'wint_t' (which is defined in the system headers). 2012 EmitFormatDiagnostic( 2013 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2014 << ATR.getRepresentativeType(S.Context) << Ex->getType() 2015 << Ex->getSourceRange(), 2016 getLocationOfByte(CS.getStart()), 2017 /*IsStringLocation*/true, 2018 getSpecifierRange(startSpecifier, specifierLen), 2019 FixItHint::CreateReplacement( 2020 getSpecifierRange(startSpecifier, specifierLen), 2021 os.str())); 2022 } 2023 else { 2024 S.Diag(getLocationOfByte(CS.getStart()), 2025 diag::warn_printf_conversion_argument_type_mismatch) 2026 << ATR.getRepresentativeType(S.Context) << Ex->getType() 2027 << getSpecifierRange(startSpecifier, specifierLen) 2028 << Ex->getSourceRange(); 2029 } 2030 } 2031 2032 return true; 2033} 2034 2035//===--- CHECK: Scanf format string checking ------------------------------===// 2036 2037namespace { 2038class CheckScanfHandler : public CheckFormatHandler { 2039public: 2040 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2041 const Expr *origFormatExpr, unsigned firstDataArg, 2042 unsigned numDataArgs, bool isObjCLiteral, 2043 const char *beg, bool hasVAListArg, 2044 const CallExpr *theCall, unsigned formatIdx, 2045 bool inFunctionCall) 2046 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2047 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2048 theCall, formatIdx, inFunctionCall) {} 2049 2050 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2051 const char *startSpecifier, 2052 unsigned specifierLen); 2053 2054 bool HandleInvalidScanfConversionSpecifier( 2055 const analyze_scanf::ScanfSpecifier &FS, 2056 const char *startSpecifier, 2057 unsigned specifierLen); 2058 2059 void HandleIncompleteScanList(const char *start, const char *end); 2060}; 2061} 2062 2063void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2064 const char *end) { 2065 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2066 getLocationOfByte(end), /*IsStringLocation*/true, 2067 getSpecifierRange(start, end - start)); 2068} 2069 2070bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2071 const analyze_scanf::ScanfSpecifier &FS, 2072 const char *startSpecifier, 2073 unsigned specifierLen) { 2074 2075 const analyze_scanf::ScanfConversionSpecifier &CS = 2076 FS.getConversionSpecifier(); 2077 2078 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2079 getLocationOfByte(CS.getStart()), 2080 startSpecifier, specifierLen, 2081 CS.getStart(), CS.getLength()); 2082} 2083 2084bool CheckScanfHandler::HandleScanfSpecifier( 2085 const analyze_scanf::ScanfSpecifier &FS, 2086 const char *startSpecifier, 2087 unsigned specifierLen) { 2088 2089 using namespace analyze_scanf; 2090 using namespace analyze_format_string; 2091 2092 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2093 2094 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2095 // be used to decide if we are using positional arguments consistently. 2096 if (FS.consumesDataArgument()) { 2097 if (atFirstArg) { 2098 atFirstArg = false; 2099 usesPositionalArgs = FS.usesPositionalArg(); 2100 } 2101 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2102 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2103 startSpecifier, specifierLen); 2104 return false; 2105 } 2106 } 2107 2108 // Check if the field with is non-zero. 2109 const OptionalAmount &Amt = FS.getFieldWidth(); 2110 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2111 if (Amt.getConstantAmount() == 0) { 2112 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2113 Amt.getConstantLength()); 2114 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2115 getLocationOfByte(Amt.getStart()), 2116 /*IsStringLocation*/true, R, 2117 FixItHint::CreateRemoval(R)); 2118 } 2119 } 2120 2121 if (!FS.consumesDataArgument()) { 2122 // FIXME: Technically specifying a precision or field width here 2123 // makes no sense. Worth issuing a warning at some point. 2124 return true; 2125 } 2126 2127 // Consume the argument. 2128 unsigned argIndex = FS.getArgIndex(); 2129 if (argIndex < NumDataArgs) { 2130 // The check to see if the argIndex is valid will come later. 2131 // We set the bit here because we may exit early from this 2132 // function if we encounter some other error. 2133 CoveredArgs.set(argIndex); 2134 } 2135 2136 // Check the length modifier is valid with the given conversion specifier. 2137 const LengthModifier &LM = FS.getLengthModifier(); 2138 if (!FS.hasValidLengthModifier()) { 2139 S.Diag(getLocationOfByte(LM.getStart()), 2140 diag::warn_format_nonsensical_length) 2141 << LM.toString() << CS.toString() 2142 << getSpecifierRange(startSpecifier, specifierLen) 2143 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 2144 LM.getLength())); 2145 } 2146 2147 // The remaining checks depend on the data arguments. 2148 if (HasVAListArg) 2149 return true; 2150 2151 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2152 return false; 2153 2154 // FIXME: Check that the argument type matches the format specifier. 2155 2156 return true; 2157} 2158 2159void Sema::CheckFormatString(const StringLiteral *FExpr, 2160 const Expr *OrigFormatExpr, 2161 const CallExpr *TheCall, bool HasVAListArg, 2162 unsigned format_idx, unsigned firstDataArg, 2163 bool isPrintf, bool inFunctionCall) { 2164 2165 // CHECK: is the format string a wide literal? 2166 if (!FExpr->isAscii()) { 2167 CheckFormatHandler::EmitFormatDiagnostic( 2168 *this, inFunctionCall, TheCall->getArg(format_idx), 2169 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2170 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2171 return; 2172 } 2173 2174 // Str - The format string. NOTE: this is NOT null-terminated! 2175 StringRef StrRef = FExpr->getString(); 2176 const char *Str = StrRef.data(); 2177 unsigned StrLen = StrRef.size(); 2178 const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg; 2179 2180 // CHECK: empty format string? 2181 if (StrLen == 0 && numDataArgs > 0) { 2182 CheckFormatHandler::EmitFormatDiagnostic( 2183 *this, inFunctionCall, TheCall->getArg(format_idx), 2184 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2185 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2186 return; 2187 } 2188 2189 if (isPrintf) { 2190 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2191 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2192 Str, HasVAListArg, TheCall, format_idx, 2193 inFunctionCall); 2194 2195 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 2196 H.DoneProcessing(); 2197 } 2198 else { 2199 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2200 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2201 Str, HasVAListArg, TheCall, format_idx, 2202 inFunctionCall); 2203 2204 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 2205 H.DoneProcessing(); 2206 } 2207} 2208 2209//===--- CHECK: Standard memory functions ---------------------------------===// 2210 2211/// \brief Determine whether the given type is a dynamic class type (e.g., 2212/// whether it has a vtable). 2213static bool isDynamicClassType(QualType T) { 2214 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2215 if (CXXRecordDecl *Definition = Record->getDefinition()) 2216 if (Definition->isDynamicClass()) 2217 return true; 2218 2219 return false; 2220} 2221 2222/// \brief If E is a sizeof expression, returns its argument expression, 2223/// otherwise returns NULL. 2224static const Expr *getSizeOfExprArg(const Expr* E) { 2225 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2226 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2227 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2228 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2229 2230 return 0; 2231} 2232 2233/// \brief If E is a sizeof expression, returns its argument type. 2234static QualType getSizeOfArgType(const Expr* E) { 2235 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2236 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2237 if (SizeOf->getKind() == clang::UETT_SizeOf) 2238 return SizeOf->getTypeOfArgument(); 2239 2240 return QualType(); 2241} 2242 2243/// \brief Check for dangerous or invalid arguments to memset(). 2244/// 2245/// This issues warnings on known problematic, dangerous or unspecified 2246/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2247/// function calls. 2248/// 2249/// \param Call The call expression to diagnose. 2250void Sema::CheckMemaccessArguments(const CallExpr *Call, 2251 CheckedMemoryFunction CMF, 2252 IdentifierInfo *FnName) { 2253 // It is possible to have a non-standard definition of memset. Validate 2254 // we have enough arguments, and if not, abort further checking. 2255 unsigned ExpectedNumArgs = (CMF == CMF_Strndup ? 2 : 3); 2256 if (Call->getNumArgs() < ExpectedNumArgs) 2257 return; 2258 2259 unsigned LastArg = (CMF == CMF_Memset || CMF == CMF_Strndup ? 1 : 2); 2260 unsigned LenArg = (CMF == CMF_Strndup ? 1 : 2); 2261 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2262 2263 // We have special checking when the length is a sizeof expression. 2264 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2265 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2266 llvm::FoldingSetNodeID SizeOfArgID; 2267 2268 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2269 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2270 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2271 2272 QualType DestTy = Dest->getType(); 2273 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2274 QualType PointeeTy = DestPtrTy->getPointeeType(); 2275 2276 // Never warn about void type pointers. This can be used to suppress 2277 // false positives. 2278 if (PointeeTy->isVoidType()) 2279 continue; 2280 2281 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2282 // actually comparing the expressions for equality. Because computing the 2283 // expression IDs can be expensive, we only do this if the diagnostic is 2284 // enabled. 2285 if (SizeOfArg && 2286 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2287 SizeOfArg->getExprLoc())) { 2288 // We only compute IDs for expressions if the warning is enabled, and 2289 // cache the sizeof arg's ID. 2290 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2291 SizeOfArg->Profile(SizeOfArgID, Context, true); 2292 llvm::FoldingSetNodeID DestID; 2293 Dest->Profile(DestID, Context, true); 2294 if (DestID == SizeOfArgID) { 2295 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2296 // over sizeof(src) as well. 2297 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2298 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2299 if (UnaryOp->getOpcode() == UO_AddrOf) 2300 ActionIdx = 1; // If its an address-of operator, just remove it. 2301 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2302 ActionIdx = 2; // If the pointee's size is sizeof(char), 2303 // suggest an explicit length. 2304 unsigned DestSrcSelect = (CMF == CMF_Strndup ? 1 : ArgIdx); 2305 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2306 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2307 << FnName << DestSrcSelect << ActionIdx 2308 << Dest->getSourceRange() 2309 << SizeOfArg->getSourceRange()); 2310 break; 2311 } 2312 } 2313 2314 // Also check for cases where the sizeof argument is the exact same 2315 // type as the memory argument, and where it points to a user-defined 2316 // record type. 2317 if (SizeOfArgTy != QualType()) { 2318 if (PointeeTy->isRecordType() && 2319 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2320 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2321 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2322 << FnName << SizeOfArgTy << ArgIdx 2323 << PointeeTy << Dest->getSourceRange() 2324 << LenExpr->getSourceRange()); 2325 break; 2326 } 2327 } 2328 2329 // Always complain about dynamic classes. 2330 if (isDynamicClassType(PointeeTy)) 2331 DiagRuntimeBehavior( 2332 Dest->getExprLoc(), Dest, 2333 PDiag(diag::warn_dyn_class_memaccess) 2334 << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy 2335 // "overwritten" if we're warning about the destination for any call 2336 // but memcmp; otherwise a verb appropriate to the call. 2337 << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF) 2338 << Call->getCallee()->getSourceRange()); 2339 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 2340 DiagRuntimeBehavior( 2341 Dest->getExprLoc(), Dest, 2342 PDiag(diag::warn_arc_object_memaccess) 2343 << ArgIdx << FnName << PointeeTy 2344 << Call->getCallee()->getSourceRange()); 2345 else 2346 continue; 2347 2348 DiagRuntimeBehavior( 2349 Dest->getExprLoc(), Dest, 2350 PDiag(diag::note_bad_memaccess_silence) 2351 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2352 break; 2353 } 2354 } 2355} 2356 2357// A little helper routine: ignore addition and subtraction of integer literals. 2358// This intentionally does not ignore all integer constant expressions because 2359// we don't want to remove sizeof(). 2360static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2361 Ex = Ex->IgnoreParenCasts(); 2362 2363 for (;;) { 2364 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2365 if (!BO || !BO->isAdditiveOp()) 2366 break; 2367 2368 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2369 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2370 2371 if (isa<IntegerLiteral>(RHS)) 2372 Ex = LHS; 2373 else if (isa<IntegerLiteral>(LHS)) 2374 Ex = RHS; 2375 else 2376 break; 2377 } 2378 2379 return Ex; 2380} 2381 2382// Warn if the user has made the 'size' argument to strlcpy or strlcat 2383// be the size of the source, instead of the destination. 2384void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2385 IdentifierInfo *FnName) { 2386 2387 // Don't crash if the user has the wrong number of arguments 2388 if (Call->getNumArgs() != 3) 2389 return; 2390 2391 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2392 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2393 const Expr *CompareWithSrc = NULL; 2394 2395 // Look for 'strlcpy(dst, x, sizeof(x))' 2396 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2397 CompareWithSrc = Ex; 2398 else { 2399 // Look for 'strlcpy(dst, x, strlen(x))' 2400 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2401 if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen 2402 && SizeCall->getNumArgs() == 1) 2403 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2404 } 2405 } 2406 2407 if (!CompareWithSrc) 2408 return; 2409 2410 // Determine if the argument to sizeof/strlen is equal to the source 2411 // argument. In principle there's all kinds of things you could do 2412 // here, for instance creating an == expression and evaluating it with 2413 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2414 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2415 if (!SrcArgDRE) 2416 return; 2417 2418 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2419 if (!CompareWithSrcDRE || 2420 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2421 return; 2422 2423 const Expr *OriginalSizeArg = Call->getArg(2); 2424 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2425 << OriginalSizeArg->getSourceRange() << FnName; 2426 2427 // Output a FIXIT hint if the destination is an array (rather than a 2428 // pointer to an array). This could be enhanced to handle some 2429 // pointers if we know the actual size, like if DstArg is 'array+2' 2430 // we could say 'sizeof(array)-2'. 2431 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2432 QualType DstArgTy = DstArg->getType(); 2433 2434 // Only handle constant-sized or VLAs, but not flexible members. 2435 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2436 // Only issue the FIXIT for arrays of size > 1. 2437 if (CAT->getSize().getSExtValue() <= 1) 2438 return; 2439 } else if (!DstArgTy->isVariableArrayType()) { 2440 return; 2441 } 2442 2443 llvm::SmallString<128> sizeString; 2444 llvm::raw_svector_ostream OS(sizeString); 2445 OS << "sizeof("; 2446 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2447 OS << ")"; 2448 2449 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2450 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2451 OS.str()); 2452} 2453 2454//===--- CHECK: Return Address of Stack Variable --------------------------===// 2455 2456static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2457static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2458 2459/// CheckReturnStackAddr - Check if a return statement returns the address 2460/// of a stack variable. 2461void 2462Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2463 SourceLocation ReturnLoc) { 2464 2465 Expr *stackE = 0; 2466 SmallVector<DeclRefExpr *, 8> refVars; 2467 2468 // Perform checking for returned stack addresses, local blocks, 2469 // label addresses or references to temporaries. 2470 if (lhsType->isPointerType() || 2471 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2472 stackE = EvalAddr(RetValExp, refVars); 2473 } else if (lhsType->isReferenceType()) { 2474 stackE = EvalVal(RetValExp, refVars); 2475 } 2476 2477 if (stackE == 0) 2478 return; // Nothing suspicious was found. 2479 2480 SourceLocation diagLoc; 2481 SourceRange diagRange; 2482 if (refVars.empty()) { 2483 diagLoc = stackE->getLocStart(); 2484 diagRange = stackE->getSourceRange(); 2485 } else { 2486 // We followed through a reference variable. 'stackE' contains the 2487 // problematic expression but we will warn at the return statement pointing 2488 // at the reference variable. We will later display the "trail" of 2489 // reference variables using notes. 2490 diagLoc = refVars[0]->getLocStart(); 2491 diagRange = refVars[0]->getSourceRange(); 2492 } 2493 2494 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2495 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2496 : diag::warn_ret_stack_addr) 2497 << DR->getDecl()->getDeclName() << diagRange; 2498 } else if (isa<BlockExpr>(stackE)) { // local block. 2499 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2500 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2501 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2502 } else { // local temporary. 2503 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2504 : diag::warn_ret_local_temp_addr) 2505 << diagRange; 2506 } 2507 2508 // Display the "trail" of reference variables that we followed until we 2509 // found the problematic expression using notes. 2510 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2511 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2512 // If this var binds to another reference var, show the range of the next 2513 // var, otherwise the var binds to the problematic expression, in which case 2514 // show the range of the expression. 2515 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2516 : stackE->getSourceRange(); 2517 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2518 << VD->getDeclName() << range; 2519 } 2520} 2521 2522/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2523/// check if the expression in a return statement evaluates to an address 2524/// to a location on the stack, a local block, an address of a label, or a 2525/// reference to local temporary. The recursion is used to traverse the 2526/// AST of the return expression, with recursion backtracking when we 2527/// encounter a subexpression that (1) clearly does not lead to one of the 2528/// above problematic expressions (2) is something we cannot determine leads to 2529/// a problematic expression based on such local checking. 2530/// 2531/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2532/// the expression that they point to. Such variables are added to the 2533/// 'refVars' vector so that we know what the reference variable "trail" was. 2534/// 2535/// EvalAddr processes expressions that are pointers that are used as 2536/// references (and not L-values). EvalVal handles all other values. 2537/// At the base case of the recursion is a check for the above problematic 2538/// expressions. 2539/// 2540/// This implementation handles: 2541/// 2542/// * pointer-to-pointer casts 2543/// * implicit conversions from array references to pointers 2544/// * taking the address of fields 2545/// * arbitrary interplay between "&" and "*" operators 2546/// * pointer arithmetic from an address of a stack variable 2547/// * taking the address of an array element where the array is on the stack 2548static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2549 if (E->isTypeDependent()) 2550 return NULL; 2551 2552 // We should only be called for evaluating pointer expressions. 2553 assert((E->getType()->isAnyPointerType() || 2554 E->getType()->isBlockPointerType() || 2555 E->getType()->isObjCQualifiedIdType()) && 2556 "EvalAddr only works on pointers"); 2557 2558 E = E->IgnoreParens(); 2559 2560 // Our "symbolic interpreter" is just a dispatch off the currently 2561 // viewed AST node. We then recursively traverse the AST by calling 2562 // EvalAddr and EvalVal appropriately. 2563 switch (E->getStmtClass()) { 2564 case Stmt::DeclRefExprClass: { 2565 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2566 2567 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2568 // If this is a reference variable, follow through to the expression that 2569 // it points to. 2570 if (V->hasLocalStorage() && 2571 V->getType()->isReferenceType() && V->hasInit()) { 2572 // Add the reference variable to the "trail". 2573 refVars.push_back(DR); 2574 return EvalAddr(V->getInit(), refVars); 2575 } 2576 2577 return NULL; 2578 } 2579 2580 case Stmt::UnaryOperatorClass: { 2581 // The only unary operator that make sense to handle here 2582 // is AddrOf. All others don't make sense as pointers. 2583 UnaryOperator *U = cast<UnaryOperator>(E); 2584 2585 if (U->getOpcode() == UO_AddrOf) 2586 return EvalVal(U->getSubExpr(), refVars); 2587 else 2588 return NULL; 2589 } 2590 2591 case Stmt::BinaryOperatorClass: { 2592 // Handle pointer arithmetic. All other binary operators are not valid 2593 // in this context. 2594 BinaryOperator *B = cast<BinaryOperator>(E); 2595 BinaryOperatorKind op = B->getOpcode(); 2596 2597 if (op != BO_Add && op != BO_Sub) 2598 return NULL; 2599 2600 Expr *Base = B->getLHS(); 2601 2602 // Determine which argument is the real pointer base. It could be 2603 // the RHS argument instead of the LHS. 2604 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2605 2606 assert (Base->getType()->isPointerType()); 2607 return EvalAddr(Base, refVars); 2608 } 2609 2610 // For conditional operators we need to see if either the LHS or RHS are 2611 // valid DeclRefExpr*s. If one of them is valid, we return it. 2612 case Stmt::ConditionalOperatorClass: { 2613 ConditionalOperator *C = cast<ConditionalOperator>(E); 2614 2615 // Handle the GNU extension for missing LHS. 2616 if (Expr *lhsExpr = C->getLHS()) { 2617 // In C++, we can have a throw-expression, which has 'void' type. 2618 if (!lhsExpr->getType()->isVoidType()) 2619 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2620 return LHS; 2621 } 2622 2623 // In C++, we can have a throw-expression, which has 'void' type. 2624 if (C->getRHS()->getType()->isVoidType()) 2625 return NULL; 2626 2627 return EvalAddr(C->getRHS(), refVars); 2628 } 2629 2630 case Stmt::BlockExprClass: 2631 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2632 return E; // local block. 2633 return NULL; 2634 2635 case Stmt::AddrLabelExprClass: 2636 return E; // address of label. 2637 2638 // For casts, we need to handle conversions from arrays to 2639 // pointer values, and pointer-to-pointer conversions. 2640 case Stmt::ImplicitCastExprClass: 2641 case Stmt::CStyleCastExprClass: 2642 case Stmt::CXXFunctionalCastExprClass: 2643 case Stmt::ObjCBridgedCastExprClass: { 2644 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2645 QualType T = SubExpr->getType(); 2646 2647 if (SubExpr->getType()->isPointerType() || 2648 SubExpr->getType()->isBlockPointerType() || 2649 SubExpr->getType()->isObjCQualifiedIdType()) 2650 return EvalAddr(SubExpr, refVars); 2651 else if (T->isArrayType()) 2652 return EvalVal(SubExpr, refVars); 2653 else 2654 return 0; 2655 } 2656 2657 // C++ casts. For dynamic casts, static casts, and const casts, we 2658 // are always converting from a pointer-to-pointer, so we just blow 2659 // through the cast. In the case the dynamic cast doesn't fail (and 2660 // return NULL), we take the conservative route and report cases 2661 // where we return the address of a stack variable. For Reinterpre 2662 // FIXME: The comment about is wrong; we're not always converting 2663 // from pointer to pointer. I'm guessing that this code should also 2664 // handle references to objects. 2665 case Stmt::CXXStaticCastExprClass: 2666 case Stmt::CXXDynamicCastExprClass: 2667 case Stmt::CXXConstCastExprClass: 2668 case Stmt::CXXReinterpretCastExprClass: { 2669 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2670 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2671 return EvalAddr(S, refVars); 2672 else 2673 return NULL; 2674 } 2675 2676 case Stmt::MaterializeTemporaryExprClass: 2677 if (Expr *Result = EvalAddr( 2678 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2679 refVars)) 2680 return Result; 2681 2682 return E; 2683 2684 // Everything else: we simply don't reason about them. 2685 default: 2686 return NULL; 2687 } 2688} 2689 2690 2691/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2692/// See the comments for EvalAddr for more details. 2693static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2694do { 2695 // We should only be called for evaluating non-pointer expressions, or 2696 // expressions with a pointer type that are not used as references but instead 2697 // are l-values (e.g., DeclRefExpr with a pointer type). 2698 2699 // Our "symbolic interpreter" is just a dispatch off the currently 2700 // viewed AST node. We then recursively traverse the AST by calling 2701 // EvalAddr and EvalVal appropriately. 2702 2703 E = E->IgnoreParens(); 2704 switch (E->getStmtClass()) { 2705 case Stmt::ImplicitCastExprClass: { 2706 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2707 if (IE->getValueKind() == VK_LValue) { 2708 E = IE->getSubExpr(); 2709 continue; 2710 } 2711 return NULL; 2712 } 2713 2714 case Stmt::DeclRefExprClass: { 2715 // When we hit a DeclRefExpr we are looking at code that refers to a 2716 // variable's name. If it's not a reference variable we check if it has 2717 // local storage within the function, and if so, return the expression. 2718 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2719 2720 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2721 if (V->hasLocalStorage()) { 2722 if (!V->getType()->isReferenceType()) 2723 return DR; 2724 2725 // Reference variable, follow through to the expression that 2726 // it points to. 2727 if (V->hasInit()) { 2728 // Add the reference variable to the "trail". 2729 refVars.push_back(DR); 2730 return EvalVal(V->getInit(), refVars); 2731 } 2732 } 2733 2734 return NULL; 2735 } 2736 2737 case Stmt::UnaryOperatorClass: { 2738 // The only unary operator that make sense to handle here 2739 // is Deref. All others don't resolve to a "name." This includes 2740 // handling all sorts of rvalues passed to a unary operator. 2741 UnaryOperator *U = cast<UnaryOperator>(E); 2742 2743 if (U->getOpcode() == UO_Deref) 2744 return EvalAddr(U->getSubExpr(), refVars); 2745 2746 return NULL; 2747 } 2748 2749 case Stmt::ArraySubscriptExprClass: { 2750 // Array subscripts are potential references to data on the stack. We 2751 // retrieve the DeclRefExpr* for the array variable if it indeed 2752 // has local storage. 2753 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2754 } 2755 2756 case Stmt::ConditionalOperatorClass: { 2757 // For conditional operators we need to see if either the LHS or RHS are 2758 // non-NULL Expr's. If one is non-NULL, we return it. 2759 ConditionalOperator *C = cast<ConditionalOperator>(E); 2760 2761 // Handle the GNU extension for missing LHS. 2762 if (Expr *lhsExpr = C->getLHS()) 2763 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2764 return LHS; 2765 2766 return EvalVal(C->getRHS(), refVars); 2767 } 2768 2769 // Accesses to members are potential references to data on the stack. 2770 case Stmt::MemberExprClass: { 2771 MemberExpr *M = cast<MemberExpr>(E); 2772 2773 // Check for indirect access. We only want direct field accesses. 2774 if (M->isArrow()) 2775 return NULL; 2776 2777 // Check whether the member type is itself a reference, in which case 2778 // we're not going to refer to the member, but to what the member refers to. 2779 if (M->getMemberDecl()->getType()->isReferenceType()) 2780 return NULL; 2781 2782 return EvalVal(M->getBase(), refVars); 2783 } 2784 2785 case Stmt::MaterializeTemporaryExprClass: 2786 if (Expr *Result = EvalVal( 2787 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2788 refVars)) 2789 return Result; 2790 2791 return E; 2792 2793 default: 2794 // Check that we don't return or take the address of a reference to a 2795 // temporary. This is only useful in C++. 2796 if (!E->isTypeDependent() && E->isRValue()) 2797 return E; 2798 2799 // Everything else: we simply don't reason about them. 2800 return NULL; 2801 } 2802} while (true); 2803} 2804 2805//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2806 2807/// Check for comparisons of floating point operands using != and ==. 2808/// Issue a warning if these are no self-comparisons, as they are not likely 2809/// to do what the programmer intended. 2810void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 2811 bool EmitWarning = true; 2812 2813 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 2814 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 2815 2816 // Special case: check for x == x (which is OK). 2817 // Do not emit warnings for such cases. 2818 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2819 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2820 if (DRL->getDecl() == DRR->getDecl()) 2821 EmitWarning = false; 2822 2823 2824 // Special case: check for comparisons against literals that can be exactly 2825 // represented by APFloat. In such cases, do not emit a warning. This 2826 // is a heuristic: often comparison against such literals are used to 2827 // detect if a value in a variable has not changed. This clearly can 2828 // lead to false negatives. 2829 if (EmitWarning) { 2830 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2831 if (FLL->isExact()) 2832 EmitWarning = false; 2833 } else 2834 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2835 if (FLR->isExact()) 2836 EmitWarning = false; 2837 } 2838 } 2839 2840 // Check for comparisons with builtin types. 2841 if (EmitWarning) 2842 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2843 if (CL->isBuiltinCall(Context)) 2844 EmitWarning = false; 2845 2846 if (EmitWarning) 2847 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2848 if (CR->isBuiltinCall(Context)) 2849 EmitWarning = false; 2850 2851 // Emit the diagnostic. 2852 if (EmitWarning) 2853 Diag(Loc, diag::warn_floatingpoint_eq) 2854 << LHS->getSourceRange() << RHS->getSourceRange(); 2855} 2856 2857//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2858//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2859 2860namespace { 2861 2862/// Structure recording the 'active' range of an integer-valued 2863/// expression. 2864struct IntRange { 2865 /// The number of bits active in the int. 2866 unsigned Width; 2867 2868 /// True if the int is known not to have negative values. 2869 bool NonNegative; 2870 2871 IntRange(unsigned Width, bool NonNegative) 2872 : Width(Width), NonNegative(NonNegative) 2873 {} 2874 2875 /// Returns the range of the bool type. 2876 static IntRange forBoolType() { 2877 return IntRange(1, true); 2878 } 2879 2880 /// Returns the range of an opaque value of the given integral type. 2881 static IntRange forValueOfType(ASTContext &C, QualType T) { 2882 return forValueOfCanonicalType(C, 2883 T->getCanonicalTypeInternal().getTypePtr()); 2884 } 2885 2886 /// Returns the range of an opaque value of a canonical integral type. 2887 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2888 assert(T->isCanonicalUnqualified()); 2889 2890 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2891 T = VT->getElementType().getTypePtr(); 2892 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2893 T = CT->getElementType().getTypePtr(); 2894 2895 // For enum types, use the known bit width of the enumerators. 2896 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2897 EnumDecl *Enum = ET->getDecl(); 2898 if (!Enum->isCompleteDefinition()) 2899 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2900 2901 unsigned NumPositive = Enum->getNumPositiveBits(); 2902 unsigned NumNegative = Enum->getNumNegativeBits(); 2903 2904 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2905 } 2906 2907 const BuiltinType *BT = cast<BuiltinType>(T); 2908 assert(BT->isInteger()); 2909 2910 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2911 } 2912 2913 /// Returns the "target" range of a canonical integral type, i.e. 2914 /// the range of values expressible in the type. 2915 /// 2916 /// This matches forValueOfCanonicalType except that enums have the 2917 /// full range of their type, not the range of their enumerators. 2918 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2919 assert(T->isCanonicalUnqualified()); 2920 2921 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2922 T = VT->getElementType().getTypePtr(); 2923 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2924 T = CT->getElementType().getTypePtr(); 2925 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2926 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 2927 2928 const BuiltinType *BT = cast<BuiltinType>(T); 2929 assert(BT->isInteger()); 2930 2931 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2932 } 2933 2934 /// Returns the supremum of two ranges: i.e. their conservative merge. 2935 static IntRange join(IntRange L, IntRange R) { 2936 return IntRange(std::max(L.Width, R.Width), 2937 L.NonNegative && R.NonNegative); 2938 } 2939 2940 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2941 static IntRange meet(IntRange L, IntRange R) { 2942 return IntRange(std::min(L.Width, R.Width), 2943 L.NonNegative || R.NonNegative); 2944 } 2945}; 2946 2947IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2948 if (value.isSigned() && value.isNegative()) 2949 return IntRange(value.getMinSignedBits(), false); 2950 2951 if (value.getBitWidth() > MaxWidth) 2952 value = value.trunc(MaxWidth); 2953 2954 // isNonNegative() just checks the sign bit without considering 2955 // signedness. 2956 return IntRange(value.getActiveBits(), true); 2957} 2958 2959IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2960 unsigned MaxWidth) { 2961 if (result.isInt()) 2962 return GetValueRange(C, result.getInt(), MaxWidth); 2963 2964 if (result.isVector()) { 2965 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2966 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2967 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2968 R = IntRange::join(R, El); 2969 } 2970 return R; 2971 } 2972 2973 if (result.isComplexInt()) { 2974 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2975 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2976 return IntRange::join(R, I); 2977 } 2978 2979 // This can happen with lossless casts to intptr_t of "based" lvalues. 2980 // Assume it might use arbitrary bits. 2981 // FIXME: The only reason we need to pass the type in here is to get 2982 // the sign right on this one case. It would be nice if APValue 2983 // preserved this. 2984 assert(result.isLValue()); 2985 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2986} 2987 2988/// Pseudo-evaluate the given integer expression, estimating the 2989/// range of values it might take. 2990/// 2991/// \param MaxWidth - the width to which the value will be truncated 2992IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2993 E = E->IgnoreParens(); 2994 2995 // Try a full evaluation first. 2996 Expr::EvalResult result; 2997 if (E->EvaluateAsRValue(result, C)) 2998 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2999 3000 // I think we only want to look through implicit casts here; if the 3001 // user has an explicit widening cast, we should treat the value as 3002 // being of the new, wider type. 3003 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3004 if (CE->getCastKind() == CK_NoOp) 3005 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3006 3007 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3008 3009 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3010 3011 // Assume that non-integer casts can span the full range of the type. 3012 if (!isIntegerCast) 3013 return OutputTypeRange; 3014 3015 IntRange SubRange 3016 = GetExprRange(C, CE->getSubExpr(), 3017 std::min(MaxWidth, OutputTypeRange.Width)); 3018 3019 // Bail out if the subexpr's range is as wide as the cast type. 3020 if (SubRange.Width >= OutputTypeRange.Width) 3021 return OutputTypeRange; 3022 3023 // Otherwise, we take the smaller width, and we're non-negative if 3024 // either the output type or the subexpr is. 3025 return IntRange(SubRange.Width, 3026 SubRange.NonNegative || OutputTypeRange.NonNegative); 3027 } 3028 3029 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3030 // If we can fold the condition, just take that operand. 3031 bool CondResult; 3032 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3033 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3034 : CO->getFalseExpr(), 3035 MaxWidth); 3036 3037 // Otherwise, conservatively merge. 3038 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3039 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3040 return IntRange::join(L, R); 3041 } 3042 3043 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3044 switch (BO->getOpcode()) { 3045 3046 // Boolean-valued operations are single-bit and positive. 3047 case BO_LAnd: 3048 case BO_LOr: 3049 case BO_LT: 3050 case BO_GT: 3051 case BO_LE: 3052 case BO_GE: 3053 case BO_EQ: 3054 case BO_NE: 3055 return IntRange::forBoolType(); 3056 3057 // The type of the assignments is the type of the LHS, so the RHS 3058 // is not necessarily the same type. 3059 case BO_MulAssign: 3060 case BO_DivAssign: 3061 case BO_RemAssign: 3062 case BO_AddAssign: 3063 case BO_SubAssign: 3064 case BO_XorAssign: 3065 case BO_OrAssign: 3066 // TODO: bitfields? 3067 return IntRange::forValueOfType(C, E->getType()); 3068 3069 // Simple assignments just pass through the RHS, which will have 3070 // been coerced to the LHS type. 3071 case BO_Assign: 3072 // TODO: bitfields? 3073 return GetExprRange(C, BO->getRHS(), MaxWidth); 3074 3075 // Operations with opaque sources are black-listed. 3076 case BO_PtrMemD: 3077 case BO_PtrMemI: 3078 return IntRange::forValueOfType(C, E->getType()); 3079 3080 // Bitwise-and uses the *infinum* of the two source ranges. 3081 case BO_And: 3082 case BO_AndAssign: 3083 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3084 GetExprRange(C, BO->getRHS(), MaxWidth)); 3085 3086 // Left shift gets black-listed based on a judgement call. 3087 case BO_Shl: 3088 // ...except that we want to treat '1 << (blah)' as logically 3089 // positive. It's an important idiom. 3090 if (IntegerLiteral *I 3091 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3092 if (I->getValue() == 1) { 3093 IntRange R = IntRange::forValueOfType(C, E->getType()); 3094 return IntRange(R.Width, /*NonNegative*/ true); 3095 } 3096 } 3097 // fallthrough 3098 3099 case BO_ShlAssign: 3100 return IntRange::forValueOfType(C, E->getType()); 3101 3102 // Right shift by a constant can narrow its left argument. 3103 case BO_Shr: 3104 case BO_ShrAssign: { 3105 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3106 3107 // If the shift amount is a positive constant, drop the width by 3108 // that much. 3109 llvm::APSInt shift; 3110 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3111 shift.isNonNegative()) { 3112 unsigned zext = shift.getZExtValue(); 3113 if (zext >= L.Width) 3114 L.Width = (L.NonNegative ? 0 : 1); 3115 else 3116 L.Width -= zext; 3117 } 3118 3119 return L; 3120 } 3121 3122 // Comma acts as its right operand. 3123 case BO_Comma: 3124 return GetExprRange(C, BO->getRHS(), MaxWidth); 3125 3126 // Black-list pointer subtractions. 3127 case BO_Sub: 3128 if (BO->getLHS()->getType()->isPointerType()) 3129 return IntRange::forValueOfType(C, E->getType()); 3130 break; 3131 3132 // The width of a division result is mostly determined by the size 3133 // of the LHS. 3134 case BO_Div: { 3135 // Don't 'pre-truncate' the operands. 3136 unsigned opWidth = C.getIntWidth(E->getType()); 3137 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3138 3139 // If the divisor is constant, use that. 3140 llvm::APSInt divisor; 3141 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3142 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3143 if (log2 >= L.Width) 3144 L.Width = (L.NonNegative ? 0 : 1); 3145 else 3146 L.Width = std::min(L.Width - log2, MaxWidth); 3147 return L; 3148 } 3149 3150 // Otherwise, just use the LHS's width. 3151 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3152 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3153 } 3154 3155 // The result of a remainder can't be larger than the result of 3156 // either side. 3157 case BO_Rem: { 3158 // Don't 'pre-truncate' the operands. 3159 unsigned opWidth = C.getIntWidth(E->getType()); 3160 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3161 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3162 3163 IntRange meet = IntRange::meet(L, R); 3164 meet.Width = std::min(meet.Width, MaxWidth); 3165 return meet; 3166 } 3167 3168 // The default behavior is okay for these. 3169 case BO_Mul: 3170 case BO_Add: 3171 case BO_Xor: 3172 case BO_Or: 3173 break; 3174 } 3175 3176 // The default case is to treat the operation as if it were closed 3177 // on the narrowest type that encompasses both operands. 3178 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3179 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3180 return IntRange::join(L, R); 3181 } 3182 3183 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3184 switch (UO->getOpcode()) { 3185 // Boolean-valued operations are white-listed. 3186 case UO_LNot: 3187 return IntRange::forBoolType(); 3188 3189 // Operations with opaque sources are black-listed. 3190 case UO_Deref: 3191 case UO_AddrOf: // should be impossible 3192 return IntRange::forValueOfType(C, E->getType()); 3193 3194 default: 3195 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3196 } 3197 } 3198 3199 if (dyn_cast<OffsetOfExpr>(E)) { 3200 IntRange::forValueOfType(C, E->getType()); 3201 } 3202 3203 if (FieldDecl *BitField = E->getBitField()) 3204 return IntRange(BitField->getBitWidthValue(C), 3205 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3206 3207 return IntRange::forValueOfType(C, E->getType()); 3208} 3209 3210IntRange GetExprRange(ASTContext &C, Expr *E) { 3211 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3212} 3213 3214/// Checks whether the given value, which currently has the given 3215/// source semantics, has the same value when coerced through the 3216/// target semantics. 3217bool IsSameFloatAfterCast(const llvm::APFloat &value, 3218 const llvm::fltSemantics &Src, 3219 const llvm::fltSemantics &Tgt) { 3220 llvm::APFloat truncated = value; 3221 3222 bool ignored; 3223 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3224 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3225 3226 return truncated.bitwiseIsEqual(value); 3227} 3228 3229/// Checks whether the given value, which currently has the given 3230/// source semantics, has the same value when coerced through the 3231/// target semantics. 3232/// 3233/// The value might be a vector of floats (or a complex number). 3234bool IsSameFloatAfterCast(const APValue &value, 3235 const llvm::fltSemantics &Src, 3236 const llvm::fltSemantics &Tgt) { 3237 if (value.isFloat()) 3238 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3239 3240 if (value.isVector()) { 3241 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3242 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3243 return false; 3244 return true; 3245 } 3246 3247 assert(value.isComplexFloat()); 3248 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3249 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3250} 3251 3252void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3253 3254static bool IsZero(Sema &S, Expr *E) { 3255 // Suppress cases where we are comparing against an enum constant. 3256 if (const DeclRefExpr *DR = 3257 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3258 if (isa<EnumConstantDecl>(DR->getDecl())) 3259 return false; 3260 3261 // Suppress cases where the '0' value is expanded from a macro. 3262 if (E->getLocStart().isMacroID()) 3263 return false; 3264 3265 llvm::APSInt Value; 3266 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3267} 3268 3269static bool HasEnumType(Expr *E) { 3270 // Strip off implicit integral promotions. 3271 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3272 if (ICE->getCastKind() != CK_IntegralCast && 3273 ICE->getCastKind() != CK_NoOp) 3274 break; 3275 E = ICE->getSubExpr(); 3276 } 3277 3278 return E->getType()->isEnumeralType(); 3279} 3280 3281void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3282 BinaryOperatorKind op = E->getOpcode(); 3283 if (E->isValueDependent()) 3284 return; 3285 3286 if (op == BO_LT && IsZero(S, E->getRHS())) { 3287 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3288 << "< 0" << "false" << HasEnumType(E->getLHS()) 3289 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3290 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3291 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3292 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3293 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3294 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3295 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3296 << "0 >" << "false" << HasEnumType(E->getRHS()) 3297 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3298 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3299 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3300 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3301 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3302 } 3303} 3304 3305/// Analyze the operands of the given comparison. Implements the 3306/// fallback case from AnalyzeComparison. 3307void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3308 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3309 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3310} 3311 3312/// \brief Implements -Wsign-compare. 3313/// 3314/// \param E the binary operator to check for warnings 3315void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3316 // The type the comparison is being performed in. 3317 QualType T = E->getLHS()->getType(); 3318 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3319 && "comparison with mismatched types"); 3320 3321 // We don't do anything special if this isn't an unsigned integral 3322 // comparison: we're only interested in integral comparisons, and 3323 // signed comparisons only happen in cases we don't care to warn about. 3324 // 3325 // We also don't care about value-dependent expressions or expressions 3326 // whose result is a constant. 3327 if (!T->hasUnsignedIntegerRepresentation() 3328 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3329 return AnalyzeImpConvsInComparison(S, E); 3330 3331 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3332 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3333 3334 // Check to see if one of the (unmodified) operands is of different 3335 // signedness. 3336 Expr *signedOperand, *unsignedOperand; 3337 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3338 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3339 "unsigned comparison between two signed integer expressions?"); 3340 signedOperand = LHS; 3341 unsignedOperand = RHS; 3342 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3343 signedOperand = RHS; 3344 unsignedOperand = LHS; 3345 } else { 3346 CheckTrivialUnsignedComparison(S, E); 3347 return AnalyzeImpConvsInComparison(S, E); 3348 } 3349 3350 // Otherwise, calculate the effective range of the signed operand. 3351 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3352 3353 // Go ahead and analyze implicit conversions in the operands. Note 3354 // that we skip the implicit conversions on both sides. 3355 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3356 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3357 3358 // If the signed range is non-negative, -Wsign-compare won't fire, 3359 // but we should still check for comparisons which are always true 3360 // or false. 3361 if (signedRange.NonNegative) 3362 return CheckTrivialUnsignedComparison(S, E); 3363 3364 // For (in)equality comparisons, if the unsigned operand is a 3365 // constant which cannot collide with a overflowed signed operand, 3366 // then reinterpreting the signed operand as unsigned will not 3367 // change the result of the comparison. 3368 if (E->isEqualityOp()) { 3369 unsigned comparisonWidth = S.Context.getIntWidth(T); 3370 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3371 3372 // We should never be unable to prove that the unsigned operand is 3373 // non-negative. 3374 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3375 3376 if (unsignedRange.Width < comparisonWidth) 3377 return; 3378 } 3379 3380 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3381 << LHS->getType() << RHS->getType() 3382 << LHS->getSourceRange() << RHS->getSourceRange(); 3383} 3384 3385/// Analyzes an attempt to assign the given value to a bitfield. 3386/// 3387/// Returns true if there was something fishy about the attempt. 3388bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3389 SourceLocation InitLoc) { 3390 assert(Bitfield->isBitField()); 3391 if (Bitfield->isInvalidDecl()) 3392 return false; 3393 3394 // White-list bool bitfields. 3395 if (Bitfield->getType()->isBooleanType()) 3396 return false; 3397 3398 // Ignore value- or type-dependent expressions. 3399 if (Bitfield->getBitWidth()->isValueDependent() || 3400 Bitfield->getBitWidth()->isTypeDependent() || 3401 Init->isValueDependent() || 3402 Init->isTypeDependent()) 3403 return false; 3404 3405 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3406 3407 Expr::EvalResult InitValue; 3408 if (!OriginalInit->EvaluateAsRValue(InitValue, S.Context) || 3409 !InitValue.Val.isInt()) 3410 return false; 3411 3412 const llvm::APSInt &Value = InitValue.Val.getInt(); 3413 unsigned OriginalWidth = Value.getBitWidth(); 3414 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3415 3416 if (OriginalWidth <= FieldWidth) 3417 return false; 3418 3419 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3420 3421 // It's fairly common to write values into signed bitfields 3422 // that, if sign-extended, would end up becoming a different 3423 // value. We don't want to warn about that. 3424 if (Value.isSigned() && Value.isNegative()) 3425 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3426 else 3427 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3428 3429 if (Value == TruncatedValue) 3430 return false; 3431 3432 std::string PrettyValue = Value.toString(10); 3433 std::string PrettyTrunc = TruncatedValue.toString(10); 3434 3435 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3436 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3437 << Init->getSourceRange(); 3438 3439 return true; 3440} 3441 3442/// Analyze the given simple or compound assignment for warning-worthy 3443/// operations. 3444void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3445 // Just recurse on the LHS. 3446 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3447 3448 // We want to recurse on the RHS as normal unless we're assigning to 3449 // a bitfield. 3450 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3451 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3452 E->getOperatorLoc())) { 3453 // Recurse, ignoring any implicit conversions on the RHS. 3454 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3455 E->getOperatorLoc()); 3456 } 3457 } 3458 3459 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3460} 3461 3462/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3463void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3464 SourceLocation CContext, unsigned diag) { 3465 S.Diag(E->getExprLoc(), diag) 3466 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3467} 3468 3469/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3470void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3471 unsigned diag) { 3472 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3473} 3474 3475/// Diagnose an implicit cast from a literal expression. Does not warn when the 3476/// cast wouldn't lose information. 3477void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3478 SourceLocation CContext) { 3479 // Try to convert the literal exactly to an integer. If we can, don't warn. 3480 bool isExact = false; 3481 const llvm::APFloat &Value = FL->getValue(); 3482 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3483 T->hasUnsignedIntegerRepresentation()); 3484 if (Value.convertToInteger(IntegerValue, 3485 llvm::APFloat::rmTowardZero, &isExact) 3486 == llvm::APFloat::opOK && isExact) 3487 return; 3488 3489 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3490 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3491} 3492 3493std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3494 if (!Range.Width) return "0"; 3495 3496 llvm::APSInt ValueInRange = Value; 3497 ValueInRange.setIsSigned(!Range.NonNegative); 3498 ValueInRange = ValueInRange.trunc(Range.Width); 3499 return ValueInRange.toString(10); 3500} 3501 3502static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3503 SourceManager &smgr = S.Context.getSourceManager(); 3504 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3505} 3506 3507void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3508 SourceLocation CC, bool *ICContext = 0) { 3509 if (E->isTypeDependent() || E->isValueDependent()) return; 3510 3511 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3512 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3513 if (Source == Target) return; 3514 if (Target->isDependentType()) return; 3515 3516 // If the conversion context location is invalid don't complain. We also 3517 // don't want to emit a warning if the issue occurs from the expansion of 3518 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3519 // delay this check as long as possible. Once we detect we are in that 3520 // scenario, we just return. 3521 if (CC.isInvalid()) 3522 return; 3523 3524 // Diagnose implicit casts to bool. 3525 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3526 if (isa<StringLiteral>(E)) 3527 // Warn on string literal to bool. Checks for string literals in logical 3528 // expressions, for instances, assert(0 && "error here"), is prevented 3529 // by a check in AnalyzeImplicitConversions(). 3530 return DiagnoseImpCast(S, E, T, CC, 3531 diag::warn_impcast_string_literal_to_bool); 3532 return; // Other casts to bool are not checked. 3533 } 3534 3535 // Strip vector types. 3536 if (isa<VectorType>(Source)) { 3537 if (!isa<VectorType>(Target)) { 3538 if (isFromSystemMacro(S, CC)) 3539 return; 3540 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3541 } 3542 3543 // If the vector cast is cast between two vectors of the same size, it is 3544 // a bitcast, not a conversion. 3545 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3546 return; 3547 3548 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3549 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3550 } 3551 3552 // Strip complex types. 3553 if (isa<ComplexType>(Source)) { 3554 if (!isa<ComplexType>(Target)) { 3555 if (isFromSystemMacro(S, CC)) 3556 return; 3557 3558 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3559 } 3560 3561 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3562 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3563 } 3564 3565 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3566 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3567 3568 // If the source is floating point... 3569 if (SourceBT && SourceBT->isFloatingPoint()) { 3570 // ...and the target is floating point... 3571 if (TargetBT && TargetBT->isFloatingPoint()) { 3572 // ...then warn if we're dropping FP rank. 3573 3574 // Builtin FP kinds are ordered by increasing FP rank. 3575 if (SourceBT->getKind() > TargetBT->getKind()) { 3576 // Don't warn about float constants that are precisely 3577 // representable in the target type. 3578 Expr::EvalResult result; 3579 if (E->EvaluateAsRValue(result, S.Context)) { 3580 // Value might be a float, a float vector, or a float complex. 3581 if (IsSameFloatAfterCast(result.Val, 3582 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3583 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3584 return; 3585 } 3586 3587 if (isFromSystemMacro(S, CC)) 3588 return; 3589 3590 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3591 } 3592 return; 3593 } 3594 3595 // If the target is integral, always warn. 3596 if ((TargetBT && TargetBT->isInteger())) { 3597 if (isFromSystemMacro(S, CC)) 3598 return; 3599 3600 Expr *InnerE = E->IgnoreParenImpCasts(); 3601 // We also want to warn on, e.g., "int i = -1.234" 3602 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3603 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3604 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3605 3606 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3607 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3608 } else { 3609 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3610 } 3611 } 3612 3613 return; 3614 } 3615 3616 if (!Source->isIntegerType() || !Target->isIntegerType()) 3617 return; 3618 3619 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3620 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3621 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3622 << E->getSourceRange() << clang::SourceRange(CC); 3623 return; 3624 } 3625 3626 IntRange SourceRange = GetExprRange(S.Context, E); 3627 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3628 3629 if (SourceRange.Width > TargetRange.Width) { 3630 // If the source is a constant, use a default-on diagnostic. 3631 // TODO: this should happen for bitfield stores, too. 3632 llvm::APSInt Value(32); 3633 if (E->isIntegerConstantExpr(Value, S.Context)) { 3634 if (isFromSystemMacro(S, CC)) 3635 return; 3636 3637 std::string PrettySourceValue = Value.toString(10); 3638 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3639 3640 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3641 S.PDiag(diag::warn_impcast_integer_precision_constant) 3642 << PrettySourceValue << PrettyTargetValue 3643 << E->getType() << T << E->getSourceRange() 3644 << clang::SourceRange(CC)); 3645 return; 3646 } 3647 3648 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3649 if (isFromSystemMacro(S, CC)) 3650 return; 3651 3652 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3653 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3654 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3655 } 3656 3657 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3658 (!TargetRange.NonNegative && SourceRange.NonNegative && 3659 SourceRange.Width == TargetRange.Width)) { 3660 3661 if (isFromSystemMacro(S, CC)) 3662 return; 3663 3664 unsigned DiagID = diag::warn_impcast_integer_sign; 3665 3666 // Traditionally, gcc has warned about this under -Wsign-compare. 3667 // We also want to warn about it in -Wconversion. 3668 // So if -Wconversion is off, use a completely identical diagnostic 3669 // in the sign-compare group. 3670 // The conditional-checking code will 3671 if (ICContext) { 3672 DiagID = diag::warn_impcast_integer_sign_conditional; 3673 *ICContext = true; 3674 } 3675 3676 return DiagnoseImpCast(S, E, T, CC, DiagID); 3677 } 3678 3679 // Diagnose conversions between different enumeration types. 3680 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3681 // type, to give us better diagnostics. 3682 QualType SourceType = E->getType(); 3683 if (!S.getLangOptions().CPlusPlus) { 3684 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3685 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3686 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3687 SourceType = S.Context.getTypeDeclType(Enum); 3688 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3689 } 3690 } 3691 3692 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3693 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3694 if ((SourceEnum->getDecl()->getIdentifier() || 3695 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3696 (TargetEnum->getDecl()->getIdentifier() || 3697 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3698 SourceEnum != TargetEnum) { 3699 if (isFromSystemMacro(S, CC)) 3700 return; 3701 3702 return DiagnoseImpCast(S, E, SourceType, T, CC, 3703 diag::warn_impcast_different_enum_types); 3704 } 3705 3706 return; 3707} 3708 3709void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3710 3711void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3712 SourceLocation CC, bool &ICContext) { 3713 E = E->IgnoreParenImpCasts(); 3714 3715 if (isa<ConditionalOperator>(E)) 3716 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3717 3718 AnalyzeImplicitConversions(S, E, CC); 3719 if (E->getType() != T) 3720 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3721 return; 3722} 3723 3724void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3725 SourceLocation CC = E->getQuestionLoc(); 3726 3727 AnalyzeImplicitConversions(S, E->getCond(), CC); 3728 3729 bool Suspicious = false; 3730 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3731 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3732 3733 // If -Wconversion would have warned about either of the candidates 3734 // for a signedness conversion to the context type... 3735 if (!Suspicious) return; 3736 3737 // ...but it's currently ignored... 3738 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3739 CC)) 3740 return; 3741 3742 // ...then check whether it would have warned about either of the 3743 // candidates for a signedness conversion to the condition type. 3744 if (E->getType() == T) return; 3745 3746 Suspicious = false; 3747 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3748 E->getType(), CC, &Suspicious); 3749 if (!Suspicious) 3750 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3751 E->getType(), CC, &Suspicious); 3752} 3753 3754/// AnalyzeImplicitConversions - Find and report any interesting 3755/// implicit conversions in the given expression. There are a couple 3756/// of competing diagnostics here, -Wconversion and -Wsign-compare. 3757void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3758 QualType T = OrigE->getType(); 3759 Expr *E = OrigE->IgnoreParenImpCasts(); 3760 3761 if (E->isTypeDependent() || E->isValueDependent()) 3762 return; 3763 3764 // For conditional operators, we analyze the arguments as if they 3765 // were being fed directly into the output. 3766 if (isa<ConditionalOperator>(E)) { 3767 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3768 CheckConditionalOperator(S, CO, T); 3769 return; 3770 } 3771 3772 // Go ahead and check any implicit conversions we might have skipped. 3773 // The non-canonical typecheck is just an optimization; 3774 // CheckImplicitConversion will filter out dead implicit conversions. 3775 if (E->getType() != T) 3776 CheckImplicitConversion(S, E, T, CC); 3777 3778 // Now continue drilling into this expression. 3779 3780 // Skip past explicit casts. 3781 if (isa<ExplicitCastExpr>(E)) { 3782 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3783 return AnalyzeImplicitConversions(S, E, CC); 3784 } 3785 3786 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3787 // Do a somewhat different check with comparison operators. 3788 if (BO->isComparisonOp()) 3789 return AnalyzeComparison(S, BO); 3790 3791 // And with assignments and compound assignments. 3792 if (BO->isAssignmentOp()) 3793 return AnalyzeAssignment(S, BO); 3794 } 3795 3796 // These break the otherwise-useful invariant below. Fortunately, 3797 // we don't really need to recurse into them, because any internal 3798 // expressions should have been analyzed already when they were 3799 // built into statements. 3800 if (isa<StmtExpr>(E)) return; 3801 3802 // Don't descend into unevaluated contexts. 3803 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3804 3805 // Now just recurse over the expression's children. 3806 CC = E->getExprLoc(); 3807 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 3808 bool IsLogicalOperator = BO && BO->isLogicalOp(); 3809 for (Stmt::child_range I = E->children(); I; ++I) { 3810 Expr *ChildExpr = cast<Expr>(*I); 3811 if (IsLogicalOperator && 3812 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 3813 // Ignore checking string literals that are in logical operators. 3814 continue; 3815 AnalyzeImplicitConversions(S, ChildExpr, CC); 3816 } 3817} 3818 3819} // end anonymous namespace 3820 3821/// Diagnoses "dangerous" implicit conversions within the given 3822/// expression (which is a full expression). Implements -Wconversion 3823/// and -Wsign-compare. 3824/// 3825/// \param CC the "context" location of the implicit conversion, i.e. 3826/// the most location of the syntactic entity requiring the implicit 3827/// conversion 3828void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3829 // Don't diagnose in unevaluated contexts. 3830 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3831 return; 3832 3833 // Don't diagnose for value- or type-dependent expressions. 3834 if (E->isTypeDependent() || E->isValueDependent()) 3835 return; 3836 3837 // Check for array bounds violations in cases where the check isn't triggered 3838 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 3839 // ArraySubscriptExpr is on the RHS of a variable initialization. 3840 CheckArrayAccess(E); 3841 3842 // This is not the right CC for (e.g.) a variable initialization. 3843 AnalyzeImplicitConversions(*this, E, CC); 3844} 3845 3846void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3847 FieldDecl *BitField, 3848 Expr *Init) { 3849 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3850} 3851 3852/// CheckParmsForFunctionDef - Check that the parameters of the given 3853/// function are appropriate for the definition of a function. This 3854/// takes care of any checks that cannot be performed on the 3855/// declaration itself, e.g., that the types of each of the function 3856/// parameters are complete. 3857bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3858 bool CheckParameterNames) { 3859 bool HasInvalidParm = false; 3860 for (; P != PEnd; ++P) { 3861 ParmVarDecl *Param = *P; 3862 3863 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3864 // function declarator that is part of a function definition of 3865 // that function shall not have incomplete type. 3866 // 3867 // This is also C++ [dcl.fct]p6. 3868 if (!Param->isInvalidDecl() && 3869 RequireCompleteType(Param->getLocation(), Param->getType(), 3870 diag::err_typecheck_decl_incomplete_type)) { 3871 Param->setInvalidDecl(); 3872 HasInvalidParm = true; 3873 } 3874 3875 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3876 // declaration of each parameter shall include an identifier. 3877 if (CheckParameterNames && 3878 Param->getIdentifier() == 0 && 3879 !Param->isImplicit() && 3880 !getLangOptions().CPlusPlus) 3881 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3882 3883 // C99 6.7.5.3p12: 3884 // If the function declarator is not part of a definition of that 3885 // function, parameters may have incomplete type and may use the [*] 3886 // notation in their sequences of declarator specifiers to specify 3887 // variable length array types. 3888 QualType PType = Param->getOriginalType(); 3889 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3890 if (AT->getSizeModifier() == ArrayType::Star) { 3891 // FIXME: This diagnosic should point the the '[*]' if source-location 3892 // information is added for it. 3893 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3894 } 3895 } 3896 } 3897 3898 return HasInvalidParm; 3899} 3900 3901/// CheckCastAlign - Implements -Wcast-align, which warns when a 3902/// pointer cast increases the alignment requirements. 3903void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3904 // This is actually a lot of work to potentially be doing on every 3905 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3906 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3907 TRange.getBegin()) 3908 == DiagnosticsEngine::Ignored) 3909 return; 3910 3911 // Ignore dependent types. 3912 if (T->isDependentType() || Op->getType()->isDependentType()) 3913 return; 3914 3915 // Require that the destination be a pointer type. 3916 const PointerType *DestPtr = T->getAs<PointerType>(); 3917 if (!DestPtr) return; 3918 3919 // If the destination has alignment 1, we're done. 3920 QualType DestPointee = DestPtr->getPointeeType(); 3921 if (DestPointee->isIncompleteType()) return; 3922 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3923 if (DestAlign.isOne()) return; 3924 3925 // Require that the source be a pointer type. 3926 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3927 if (!SrcPtr) return; 3928 QualType SrcPointee = SrcPtr->getPointeeType(); 3929 3930 // Whitelist casts from cv void*. We already implicitly 3931 // whitelisted casts to cv void*, since they have alignment 1. 3932 // Also whitelist casts involving incomplete types, which implicitly 3933 // includes 'void'. 3934 if (SrcPointee->isIncompleteType()) return; 3935 3936 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3937 if (SrcAlign >= DestAlign) return; 3938 3939 Diag(TRange.getBegin(), diag::warn_cast_align) 3940 << Op->getType() << T 3941 << static_cast<unsigned>(SrcAlign.getQuantity()) 3942 << static_cast<unsigned>(DestAlign.getQuantity()) 3943 << TRange << Op->getSourceRange(); 3944} 3945 3946static const Type* getElementType(const Expr *BaseExpr) { 3947 const Type* EltType = BaseExpr->getType().getTypePtr(); 3948 if (EltType->isAnyPointerType()) 3949 return EltType->getPointeeType().getTypePtr(); 3950 else if (EltType->isArrayType()) 3951 return EltType->getBaseElementTypeUnsafe(); 3952 return EltType; 3953} 3954 3955/// \brief Check whether this array fits the idiom of a size-one tail padded 3956/// array member of a struct. 3957/// 3958/// We avoid emitting out-of-bounds access warnings for such arrays as they are 3959/// commonly used to emulate flexible arrays in C89 code. 3960static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 3961 const NamedDecl *ND) { 3962 if (Size != 1 || !ND) return false; 3963 3964 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 3965 if (!FD) return false; 3966 3967 // Don't consider sizes resulting from macro expansions or template argument 3968 // substitution to form C89 tail-padded arrays. 3969 ConstantArrayTypeLoc TL = 3970 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 3971 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 3972 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 3973 return false; 3974 3975 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 3976 if (!RD || !RD->isStruct()) 3977 return false; 3978 3979 // See if this is the last field decl in the record. 3980 const Decl *D = FD; 3981 while ((D = D->getNextDeclInContext())) 3982 if (isa<FieldDecl>(D)) 3983 return false; 3984 return true; 3985} 3986 3987void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 3988 bool isSubscript, bool AllowOnePastEnd) { 3989 const Type* EffectiveType = getElementType(BaseExpr); 3990 BaseExpr = BaseExpr->IgnoreParenCasts(); 3991 IndexExpr = IndexExpr->IgnoreParenCasts(); 3992 3993 const ConstantArrayType *ArrayTy = 3994 Context.getAsConstantArrayType(BaseExpr->getType()); 3995 if (!ArrayTy) 3996 return; 3997 3998 if (IndexExpr->isValueDependent()) 3999 return; 4000 llvm::APSInt index; 4001 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 4002 return; 4003 4004 const NamedDecl *ND = NULL; 4005 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4006 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4007 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4008 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4009 4010 if (index.isUnsigned() || !index.isNegative()) { 4011 llvm::APInt size = ArrayTy->getSize(); 4012 if (!size.isStrictlyPositive()) 4013 return; 4014 4015 const Type* BaseType = getElementType(BaseExpr); 4016 if (BaseType != EffectiveType) { 4017 // Make sure we're comparing apples to apples when comparing index to size 4018 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4019 uint64_t array_typesize = Context.getTypeSize(BaseType); 4020 // Handle ptrarith_typesize being zero, such as when casting to void* 4021 if (!ptrarith_typesize) ptrarith_typesize = 1; 4022 if (ptrarith_typesize != array_typesize) { 4023 // There's a cast to a different size type involved 4024 uint64_t ratio = array_typesize / ptrarith_typesize; 4025 // TODO: Be smarter about handling cases where array_typesize is not a 4026 // multiple of ptrarith_typesize 4027 if (ptrarith_typesize * ratio == array_typesize) 4028 size *= llvm::APInt(size.getBitWidth(), ratio); 4029 } 4030 } 4031 4032 if (size.getBitWidth() > index.getBitWidth()) 4033 index = index.sext(size.getBitWidth()); 4034 else if (size.getBitWidth() < index.getBitWidth()) 4035 size = size.sext(index.getBitWidth()); 4036 4037 // For array subscripting the index must be less than size, but for pointer 4038 // arithmetic also allow the index (offset) to be equal to size since 4039 // computing the next address after the end of the array is legal and 4040 // commonly done e.g. in C++ iterators and range-based for loops. 4041 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4042 return; 4043 4044 // Also don't warn for arrays of size 1 which are members of some 4045 // structure. These are often used to approximate flexible arrays in C89 4046 // code. 4047 if (IsTailPaddedMemberArray(*this, size, ND)) 4048 return; 4049 4050 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4051 if (isSubscript) 4052 DiagID = diag::warn_array_index_exceeds_bounds; 4053 4054 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4055 PDiag(DiagID) << index.toString(10, true) 4056 << size.toString(10, true) 4057 << (unsigned)size.getLimitedValue(~0U) 4058 << IndexExpr->getSourceRange()); 4059 } else { 4060 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4061 if (!isSubscript) { 4062 DiagID = diag::warn_ptr_arith_precedes_bounds; 4063 if (index.isNegative()) index = -index; 4064 } 4065 4066 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4067 PDiag(DiagID) << index.toString(10, true) 4068 << IndexExpr->getSourceRange()); 4069 } 4070 4071 if (ND) 4072 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4073 PDiag(diag::note_array_index_out_of_bounds) 4074 << ND->getDeclName()); 4075} 4076 4077void Sema::CheckArrayAccess(const Expr *expr) { 4078 int AllowOnePastEnd = 0; 4079 while (expr) { 4080 expr = expr->IgnoreParenImpCasts(); 4081 switch (expr->getStmtClass()) { 4082 case Stmt::ArraySubscriptExprClass: { 4083 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4084 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true, 4085 AllowOnePastEnd > 0); 4086 return; 4087 } 4088 case Stmt::UnaryOperatorClass: { 4089 // Only unwrap the * and & unary operators 4090 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4091 expr = UO->getSubExpr(); 4092 switch (UO->getOpcode()) { 4093 case UO_AddrOf: 4094 AllowOnePastEnd++; 4095 break; 4096 case UO_Deref: 4097 AllowOnePastEnd--; 4098 break; 4099 default: 4100 return; 4101 } 4102 break; 4103 } 4104 case Stmt::ConditionalOperatorClass: { 4105 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4106 if (const Expr *lhs = cond->getLHS()) 4107 CheckArrayAccess(lhs); 4108 if (const Expr *rhs = cond->getRHS()) 4109 CheckArrayAccess(rhs); 4110 return; 4111 } 4112 default: 4113 return; 4114 } 4115 } 4116} 4117 4118//===--- CHECK: Objective-C retain cycles ----------------------------------// 4119 4120namespace { 4121 struct RetainCycleOwner { 4122 RetainCycleOwner() : Variable(0), Indirect(false) {} 4123 VarDecl *Variable; 4124 SourceRange Range; 4125 SourceLocation Loc; 4126 bool Indirect; 4127 4128 void setLocsFrom(Expr *e) { 4129 Loc = e->getExprLoc(); 4130 Range = e->getSourceRange(); 4131 } 4132 }; 4133} 4134 4135/// Consider whether capturing the given variable can possibly lead to 4136/// a retain cycle. 4137static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4138 // In ARC, it's captured strongly iff the variable has __strong 4139 // lifetime. In MRR, it's captured strongly if the variable is 4140 // __block and has an appropriate type. 4141 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4142 return false; 4143 4144 owner.Variable = var; 4145 owner.setLocsFrom(ref); 4146 return true; 4147} 4148 4149static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 4150 while (true) { 4151 e = e->IgnoreParens(); 4152 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4153 switch (cast->getCastKind()) { 4154 case CK_BitCast: 4155 case CK_LValueBitCast: 4156 case CK_LValueToRValue: 4157 case CK_ARCReclaimReturnedObject: 4158 e = cast->getSubExpr(); 4159 continue; 4160 4161 default: 4162 return false; 4163 } 4164 } 4165 4166 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4167 ObjCIvarDecl *ivar = ref->getDecl(); 4168 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4169 return false; 4170 4171 // Try to find a retain cycle in the base. 4172 if (!findRetainCycleOwner(ref->getBase(), owner)) 4173 return false; 4174 4175 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4176 owner.Indirect = true; 4177 return true; 4178 } 4179 4180 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4181 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4182 if (!var) return false; 4183 return considerVariable(var, ref, owner); 4184 } 4185 4186 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4187 owner.Variable = ref->getDecl(); 4188 owner.setLocsFrom(ref); 4189 return true; 4190 } 4191 4192 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4193 if (member->isArrow()) return false; 4194 4195 // Don't count this as an indirect ownership. 4196 e = member->getBase(); 4197 continue; 4198 } 4199 4200 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4201 // Only pay attention to pseudo-objects on property references. 4202 ObjCPropertyRefExpr *pre 4203 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4204 ->IgnoreParens()); 4205 if (!pre) return false; 4206 if (pre->isImplicitProperty()) return false; 4207 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4208 if (!property->isRetaining() && 4209 !(property->getPropertyIvarDecl() && 4210 property->getPropertyIvarDecl()->getType() 4211 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4212 return false; 4213 4214 owner.Indirect = true; 4215 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4216 ->getSourceExpr()); 4217 continue; 4218 } 4219 4220 // Array ivars? 4221 4222 return false; 4223 } 4224} 4225 4226namespace { 4227 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4228 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4229 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4230 Variable(variable), Capturer(0) {} 4231 4232 VarDecl *Variable; 4233 Expr *Capturer; 4234 4235 void VisitDeclRefExpr(DeclRefExpr *ref) { 4236 if (ref->getDecl() == Variable && !Capturer) 4237 Capturer = ref; 4238 } 4239 4240 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4241 if (ref->getDecl() == Variable && !Capturer) 4242 Capturer = ref; 4243 } 4244 4245 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4246 if (Capturer) return; 4247 Visit(ref->getBase()); 4248 if (Capturer && ref->isFreeIvar()) 4249 Capturer = ref; 4250 } 4251 4252 void VisitBlockExpr(BlockExpr *block) { 4253 // Look inside nested blocks 4254 if (block->getBlockDecl()->capturesVariable(Variable)) 4255 Visit(block->getBlockDecl()->getBody()); 4256 } 4257 }; 4258} 4259 4260/// Check whether the given argument is a block which captures a 4261/// variable. 4262static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4263 assert(owner.Variable && owner.Loc.isValid()); 4264 4265 e = e->IgnoreParenCasts(); 4266 BlockExpr *block = dyn_cast<BlockExpr>(e); 4267 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4268 return 0; 4269 4270 FindCaptureVisitor visitor(S.Context, owner.Variable); 4271 visitor.Visit(block->getBlockDecl()->getBody()); 4272 return visitor.Capturer; 4273} 4274 4275static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4276 RetainCycleOwner &owner) { 4277 assert(capturer); 4278 assert(owner.Variable && owner.Loc.isValid()); 4279 4280 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4281 << owner.Variable << capturer->getSourceRange(); 4282 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4283 << owner.Indirect << owner.Range; 4284} 4285 4286/// Check for a keyword selector that starts with the word 'add' or 4287/// 'set'. 4288static bool isSetterLikeSelector(Selector sel) { 4289 if (sel.isUnarySelector()) return false; 4290 4291 StringRef str = sel.getNameForSlot(0); 4292 while (!str.empty() && str.front() == '_') str = str.substr(1); 4293 if (str.startswith("set") || str.startswith("add")) 4294 str = str.substr(3); 4295 else 4296 return false; 4297 4298 if (str.empty()) return true; 4299 return !islower(str.front()); 4300} 4301 4302/// Check a message send to see if it's likely to cause a retain cycle. 4303void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4304 // Only check instance methods whose selector looks like a setter. 4305 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4306 return; 4307 4308 // Try to find a variable that the receiver is strongly owned by. 4309 RetainCycleOwner owner; 4310 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4311 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 4312 return; 4313 } else { 4314 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4315 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4316 owner.Loc = msg->getSuperLoc(); 4317 owner.Range = msg->getSuperLoc(); 4318 } 4319 4320 // Check whether the receiver is captured by any of the arguments. 4321 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4322 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4323 return diagnoseRetainCycle(*this, capturer, owner); 4324} 4325 4326/// Check a property assign to see if it's likely to cause a retain cycle. 4327void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4328 RetainCycleOwner owner; 4329 if (!findRetainCycleOwner(receiver, owner)) 4330 return; 4331 4332 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4333 diagnoseRetainCycle(*this, capturer, owner); 4334} 4335 4336bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4337 QualType LHS, Expr *RHS) { 4338 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4339 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4340 return false; 4341 // strip off any implicit cast added to get to the one arc-specific 4342 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4343 if (cast->getCastKind() == CK_ARCConsumeObject) { 4344 Diag(Loc, diag::warn_arc_retained_assign) 4345 << (LT == Qualifiers::OCL_ExplicitNone) 4346 << RHS->getSourceRange(); 4347 return true; 4348 } 4349 RHS = cast->getSubExpr(); 4350 } 4351 return false; 4352} 4353 4354void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4355 Expr *LHS, Expr *RHS) { 4356 QualType LHSType = LHS->getType(); 4357 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4358 return; 4359 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4360 // FIXME. Check for other life times. 4361 if (LT != Qualifiers::OCL_None) 4362 return; 4363 4364 if (ObjCPropertyRefExpr *PRE 4365 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens())) { 4366 if (PRE->isImplicitProperty()) 4367 return; 4368 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4369 if (!PD) 4370 return; 4371 4372 unsigned Attributes = PD->getPropertyAttributes(); 4373 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4374 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4375 if (cast->getCastKind() == CK_ARCConsumeObject) { 4376 Diag(Loc, diag::warn_arc_retained_property_assign) 4377 << RHS->getSourceRange(); 4378 return; 4379 } 4380 RHS = cast->getSubExpr(); 4381 } 4382 } 4383} 4384