SemaChecking.cpp revision 2def77399ab3088106a2d61372344f5b7104e1a2
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/Initialization.h" 19#include "clang/Sema/ScopeInfo.h" 20#include "clang/Analysis/Analyses/FormatString.h" 21#include "clang/AST/ASTContext.h" 22#include "clang/AST/CharUnits.h" 23#include "clang/AST/DeclCXX.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/ExprCXX.h" 26#include "clang/AST/ExprObjC.h" 27#include "clang/AST/EvaluatedExprVisitor.h" 28#include "clang/AST/DeclObjC.h" 29#include "clang/AST/StmtCXX.h" 30#include "clang/AST/StmtObjC.h" 31#include "clang/Lex/Preprocessor.h" 32#include "llvm/ADT/BitVector.h" 33#include "llvm/ADT/STLExtras.h" 34#include "llvm/Support/raw_ostream.h" 35#include "clang/Basic/TargetBuiltins.h" 36#include "clang/Basic/TargetInfo.h" 37#include "clang/Basic/ConvertUTF.h" 38#include <limits> 39using namespace clang; 40using namespace sema; 41 42SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 43 unsigned ByteNo) const { 44 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 45 PP.getLangOptions(), PP.getTargetInfo()); 46} 47 48 49/// CheckablePrintfAttr - does a function call have a "printf" attribute 50/// and arguments that merit checking? 51bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 52 if (Format->getType() == "printf") return true; 53 if (Format->getType() == "printf0") { 54 // printf0 allows null "format" string; if so don't check format/args 55 unsigned format_idx = Format->getFormatIdx() - 1; 56 // Does the index refer to the implicit object argument? 57 if (isa<CXXMemberCallExpr>(TheCall)) { 58 if (format_idx == 0) 59 return false; 60 --format_idx; 61 } 62 if (format_idx < TheCall->getNumArgs()) { 63 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 64 if (!Format->isNullPointerConstant(Context, 65 Expr::NPC_ValueDependentIsNull)) 66 return true; 67 } 68 } 69 return false; 70} 71 72/// Checks that a call expression's argument count is the desired number. 73/// This is useful when doing custom type-checking. Returns true on error. 74static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 75 unsigned argCount = call->getNumArgs(); 76 if (argCount == desiredArgCount) return false; 77 78 if (argCount < desiredArgCount) 79 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 80 << 0 /*function call*/ << desiredArgCount << argCount 81 << call->getSourceRange(); 82 83 // Highlight all the excess arguments. 84 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 85 call->getArg(argCount - 1)->getLocEnd()); 86 87 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 88 << 0 /*function call*/ << desiredArgCount << argCount 89 << call->getArg(1)->getSourceRange(); 90} 91 92/// CheckBuiltinAnnotationString - Checks that string argument to the builtin 93/// annotation is a non wide string literal. 94static bool CheckBuiltinAnnotationString(Sema &S, Expr *Arg) { 95 Arg = Arg->IgnoreParenCasts(); 96 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 97 if (!Literal || !Literal->isAscii()) { 98 S.Diag(Arg->getLocStart(), diag::err_builtin_annotation_not_string_constant) 99 << Arg->getSourceRange(); 100 return true; 101 } 102 return false; 103} 104 105ExprResult 106Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 107 ExprResult TheCallResult(Owned(TheCall)); 108 109 // Find out if any arguments are required to be integer constant expressions. 110 unsigned ICEArguments = 0; 111 ASTContext::GetBuiltinTypeError Error; 112 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 113 if (Error != ASTContext::GE_None) 114 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 115 116 // If any arguments are required to be ICE's, check and diagnose. 117 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 118 // Skip arguments not required to be ICE's. 119 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 120 121 llvm::APSInt Result; 122 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 123 return true; 124 ICEArguments &= ~(1 << ArgNo); 125 } 126 127 switch (BuiltinID) { 128 case Builtin::BI__builtin___CFStringMakeConstantString: 129 assert(TheCall->getNumArgs() == 1 && 130 "Wrong # arguments to builtin CFStringMakeConstantString"); 131 if (CheckObjCString(TheCall->getArg(0))) 132 return ExprError(); 133 break; 134 case Builtin::BI__builtin_stdarg_start: 135 case Builtin::BI__builtin_va_start: 136 if (SemaBuiltinVAStart(TheCall)) 137 return ExprError(); 138 break; 139 case Builtin::BI__builtin_isgreater: 140 case Builtin::BI__builtin_isgreaterequal: 141 case Builtin::BI__builtin_isless: 142 case Builtin::BI__builtin_islessequal: 143 case Builtin::BI__builtin_islessgreater: 144 case Builtin::BI__builtin_isunordered: 145 if (SemaBuiltinUnorderedCompare(TheCall)) 146 return ExprError(); 147 break; 148 case Builtin::BI__builtin_fpclassify: 149 if (SemaBuiltinFPClassification(TheCall, 6)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_isfinite: 153 case Builtin::BI__builtin_isinf: 154 case Builtin::BI__builtin_isinf_sign: 155 case Builtin::BI__builtin_isnan: 156 case Builtin::BI__builtin_isnormal: 157 if (SemaBuiltinFPClassification(TheCall, 1)) 158 return ExprError(); 159 break; 160 case Builtin::BI__builtin_shufflevector: 161 return SemaBuiltinShuffleVector(TheCall); 162 // TheCall will be freed by the smart pointer here, but that's fine, since 163 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 164 case Builtin::BI__builtin_prefetch: 165 if (SemaBuiltinPrefetch(TheCall)) 166 return ExprError(); 167 break; 168 case Builtin::BI__builtin_object_size: 169 if (SemaBuiltinObjectSize(TheCall)) 170 return ExprError(); 171 break; 172 case Builtin::BI__builtin_longjmp: 173 if (SemaBuiltinLongjmp(TheCall)) 174 return ExprError(); 175 break; 176 177 case Builtin::BI__builtin_classify_type: 178 if (checkArgCount(*this, TheCall, 1)) return true; 179 TheCall->setType(Context.IntTy); 180 break; 181 case Builtin::BI__builtin_constant_p: 182 if (checkArgCount(*this, TheCall, 1)) return true; 183 TheCall->setType(Context.IntTy); 184 break; 185 case Builtin::BI__sync_fetch_and_add: 186 case Builtin::BI__sync_fetch_and_add_1: 187 case Builtin::BI__sync_fetch_and_add_2: 188 case Builtin::BI__sync_fetch_and_add_4: 189 case Builtin::BI__sync_fetch_and_add_8: 190 case Builtin::BI__sync_fetch_and_add_16: 191 case Builtin::BI__sync_fetch_and_sub: 192 case Builtin::BI__sync_fetch_and_sub_1: 193 case Builtin::BI__sync_fetch_and_sub_2: 194 case Builtin::BI__sync_fetch_and_sub_4: 195 case Builtin::BI__sync_fetch_and_sub_8: 196 case Builtin::BI__sync_fetch_and_sub_16: 197 case Builtin::BI__sync_fetch_and_or: 198 case Builtin::BI__sync_fetch_and_or_1: 199 case Builtin::BI__sync_fetch_and_or_2: 200 case Builtin::BI__sync_fetch_and_or_4: 201 case Builtin::BI__sync_fetch_and_or_8: 202 case Builtin::BI__sync_fetch_and_or_16: 203 case Builtin::BI__sync_fetch_and_and: 204 case Builtin::BI__sync_fetch_and_and_1: 205 case Builtin::BI__sync_fetch_and_and_2: 206 case Builtin::BI__sync_fetch_and_and_4: 207 case Builtin::BI__sync_fetch_and_and_8: 208 case Builtin::BI__sync_fetch_and_and_16: 209 case Builtin::BI__sync_fetch_and_xor: 210 case Builtin::BI__sync_fetch_and_xor_1: 211 case Builtin::BI__sync_fetch_and_xor_2: 212 case Builtin::BI__sync_fetch_and_xor_4: 213 case Builtin::BI__sync_fetch_and_xor_8: 214 case Builtin::BI__sync_fetch_and_xor_16: 215 case Builtin::BI__sync_add_and_fetch: 216 case Builtin::BI__sync_add_and_fetch_1: 217 case Builtin::BI__sync_add_and_fetch_2: 218 case Builtin::BI__sync_add_and_fetch_4: 219 case Builtin::BI__sync_add_and_fetch_8: 220 case Builtin::BI__sync_add_and_fetch_16: 221 case Builtin::BI__sync_sub_and_fetch: 222 case Builtin::BI__sync_sub_and_fetch_1: 223 case Builtin::BI__sync_sub_and_fetch_2: 224 case Builtin::BI__sync_sub_and_fetch_4: 225 case Builtin::BI__sync_sub_and_fetch_8: 226 case Builtin::BI__sync_sub_and_fetch_16: 227 case Builtin::BI__sync_and_and_fetch: 228 case Builtin::BI__sync_and_and_fetch_1: 229 case Builtin::BI__sync_and_and_fetch_2: 230 case Builtin::BI__sync_and_and_fetch_4: 231 case Builtin::BI__sync_and_and_fetch_8: 232 case Builtin::BI__sync_and_and_fetch_16: 233 case Builtin::BI__sync_or_and_fetch: 234 case Builtin::BI__sync_or_and_fetch_1: 235 case Builtin::BI__sync_or_and_fetch_2: 236 case Builtin::BI__sync_or_and_fetch_4: 237 case Builtin::BI__sync_or_and_fetch_8: 238 case Builtin::BI__sync_or_and_fetch_16: 239 case Builtin::BI__sync_xor_and_fetch: 240 case Builtin::BI__sync_xor_and_fetch_1: 241 case Builtin::BI__sync_xor_and_fetch_2: 242 case Builtin::BI__sync_xor_and_fetch_4: 243 case Builtin::BI__sync_xor_and_fetch_8: 244 case Builtin::BI__sync_xor_and_fetch_16: 245 case Builtin::BI__sync_val_compare_and_swap: 246 case Builtin::BI__sync_val_compare_and_swap_1: 247 case Builtin::BI__sync_val_compare_and_swap_2: 248 case Builtin::BI__sync_val_compare_and_swap_4: 249 case Builtin::BI__sync_val_compare_and_swap_8: 250 case Builtin::BI__sync_val_compare_and_swap_16: 251 case Builtin::BI__sync_bool_compare_and_swap: 252 case Builtin::BI__sync_bool_compare_and_swap_1: 253 case Builtin::BI__sync_bool_compare_and_swap_2: 254 case Builtin::BI__sync_bool_compare_and_swap_4: 255 case Builtin::BI__sync_bool_compare_and_swap_8: 256 case Builtin::BI__sync_bool_compare_and_swap_16: 257 case Builtin::BI__sync_lock_test_and_set: 258 case Builtin::BI__sync_lock_test_and_set_1: 259 case Builtin::BI__sync_lock_test_and_set_2: 260 case Builtin::BI__sync_lock_test_and_set_4: 261 case Builtin::BI__sync_lock_test_and_set_8: 262 case Builtin::BI__sync_lock_test_and_set_16: 263 case Builtin::BI__sync_lock_release: 264 case Builtin::BI__sync_lock_release_1: 265 case Builtin::BI__sync_lock_release_2: 266 case Builtin::BI__sync_lock_release_4: 267 case Builtin::BI__sync_lock_release_8: 268 case Builtin::BI__sync_lock_release_16: 269 case Builtin::BI__sync_swap: 270 case Builtin::BI__sync_swap_1: 271 case Builtin::BI__sync_swap_2: 272 case Builtin::BI__sync_swap_4: 273 case Builtin::BI__sync_swap_8: 274 case Builtin::BI__sync_swap_16: 275 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 276 case Builtin::BI__atomic_load: 277 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Load); 278 case Builtin::BI__atomic_store: 279 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Store); 280 case Builtin::BI__atomic_exchange: 281 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xchg); 282 case Builtin::BI__atomic_compare_exchange_strong: 283 return SemaAtomicOpsOverloaded(move(TheCallResult), 284 AtomicExpr::CmpXchgStrong); 285 case Builtin::BI__atomic_compare_exchange_weak: 286 return SemaAtomicOpsOverloaded(move(TheCallResult), 287 AtomicExpr::CmpXchgWeak); 288 case Builtin::BI__atomic_fetch_add: 289 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Add); 290 case Builtin::BI__atomic_fetch_sub: 291 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Sub); 292 case Builtin::BI__atomic_fetch_and: 293 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::And); 294 case Builtin::BI__atomic_fetch_or: 295 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Or); 296 case Builtin::BI__atomic_fetch_xor: 297 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::Xor); 298 case Builtin::BI__builtin_annotation: 299 if (CheckBuiltinAnnotationString(*this, TheCall->getArg(1))) 300 return ExprError(); 301 break; 302 } 303 304 // Since the target specific builtins for each arch overlap, only check those 305 // of the arch we are compiling for. 306 if (BuiltinID >= Builtin::FirstTSBuiltin) { 307 switch (Context.getTargetInfo().getTriple().getArch()) { 308 case llvm::Triple::arm: 309 case llvm::Triple::thumb: 310 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 311 return ExprError(); 312 break; 313 default: 314 break; 315 } 316 } 317 318 return move(TheCallResult); 319} 320 321// Get the valid immediate range for the specified NEON type code. 322static unsigned RFT(unsigned t, bool shift = false) { 323 NeonTypeFlags Type(t); 324 int IsQuad = Type.isQuad(); 325 switch (Type.getEltType()) { 326 case NeonTypeFlags::Int8: 327 case NeonTypeFlags::Poly8: 328 return shift ? 7 : (8 << IsQuad) - 1; 329 case NeonTypeFlags::Int16: 330 case NeonTypeFlags::Poly16: 331 return shift ? 15 : (4 << IsQuad) - 1; 332 case NeonTypeFlags::Int32: 333 return shift ? 31 : (2 << IsQuad) - 1; 334 case NeonTypeFlags::Int64: 335 return shift ? 63 : (1 << IsQuad) - 1; 336 case NeonTypeFlags::Float16: 337 assert(!shift && "cannot shift float types!"); 338 return (4 << IsQuad) - 1; 339 case NeonTypeFlags::Float32: 340 assert(!shift && "cannot shift float types!"); 341 return (2 << IsQuad) - 1; 342 } 343 return 0; 344} 345 346/// getNeonEltType - Return the QualType corresponding to the elements of 347/// the vector type specified by the NeonTypeFlags. This is used to check 348/// the pointer arguments for Neon load/store intrinsics. 349static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 350 switch (Flags.getEltType()) { 351 case NeonTypeFlags::Int8: 352 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 353 case NeonTypeFlags::Int16: 354 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 355 case NeonTypeFlags::Int32: 356 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 357 case NeonTypeFlags::Int64: 358 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 359 case NeonTypeFlags::Poly8: 360 return Context.SignedCharTy; 361 case NeonTypeFlags::Poly16: 362 return Context.ShortTy; 363 case NeonTypeFlags::Float16: 364 return Context.UnsignedShortTy; 365 case NeonTypeFlags::Float32: 366 return Context.FloatTy; 367 } 368 return QualType(); 369} 370 371bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 372 llvm::APSInt Result; 373 374 unsigned mask = 0; 375 unsigned TV = 0; 376 int PtrArgNum = -1; 377 bool HasConstPtr = false; 378 switch (BuiltinID) { 379#define GET_NEON_OVERLOAD_CHECK 380#include "clang/Basic/arm_neon.inc" 381#undef GET_NEON_OVERLOAD_CHECK 382 } 383 384 // For NEON intrinsics which are overloaded on vector element type, validate 385 // the immediate which specifies which variant to emit. 386 unsigned ImmArg = TheCall->getNumArgs()-1; 387 if (mask) { 388 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 389 return true; 390 391 TV = Result.getLimitedValue(64); 392 if ((TV > 63) || (mask & (1 << TV)) == 0) 393 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 394 << TheCall->getArg(ImmArg)->getSourceRange(); 395 } 396 397 if (PtrArgNum >= 0) { 398 // Check that pointer arguments have the specified type. 399 Expr *Arg = TheCall->getArg(PtrArgNum); 400 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 401 Arg = ICE->getSubExpr(); 402 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 403 QualType RHSTy = RHS.get()->getType(); 404 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 405 if (HasConstPtr) 406 EltTy = EltTy.withConst(); 407 QualType LHSTy = Context.getPointerType(EltTy); 408 AssignConvertType ConvTy; 409 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 410 if (RHS.isInvalid()) 411 return true; 412 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 413 RHS.get(), AA_Assigning)) 414 return true; 415 } 416 417 // For NEON intrinsics which take an immediate value as part of the 418 // instruction, range check them here. 419 unsigned i = 0, l = 0, u = 0; 420 switch (BuiltinID) { 421 default: return false; 422 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 423 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 424 case ARM::BI__builtin_arm_vcvtr_f: 425 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 426#define GET_NEON_IMMEDIATE_CHECK 427#include "clang/Basic/arm_neon.inc" 428#undef GET_NEON_IMMEDIATE_CHECK 429 }; 430 431 // Check that the immediate argument is actually a constant. 432 if (SemaBuiltinConstantArg(TheCall, i, Result)) 433 return true; 434 435 // Range check against the upper/lower values for this isntruction. 436 unsigned Val = Result.getZExtValue(); 437 if (Val < l || Val > (u + l)) 438 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 439 << l << u+l << TheCall->getArg(i)->getSourceRange(); 440 441 // FIXME: VFP Intrinsics should error if VFP not present. 442 return false; 443} 444 445/// CheckFunctionCall - Check a direct function call for various correctness 446/// and safety properties not strictly enforced by the C type system. 447bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 448 // Get the IdentifierInfo* for the called function. 449 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 450 451 // None of the checks below are needed for functions that don't have 452 // simple names (e.g., C++ conversion functions). 453 if (!FnInfo) 454 return false; 455 456 // FIXME: This mechanism should be abstracted to be less fragile and 457 // more efficient. For example, just map function ids to custom 458 // handlers. 459 460 // Printf and scanf checking. 461 for (specific_attr_iterator<FormatAttr> 462 i = FDecl->specific_attr_begin<FormatAttr>(), 463 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 464 465 const FormatAttr *Format = *i; 466 const bool b = Format->getType() == "scanf"; 467 if (b || CheckablePrintfAttr(Format, TheCall)) { 468 bool HasVAListArg = Format->getFirstArg() == 0; 469 CheckPrintfScanfArguments(TheCall, HasVAListArg, 470 Format->getFormatIdx() - 1, 471 HasVAListArg ? 0 : Format->getFirstArg() - 1, 472 !b); 473 } 474 } 475 476 for (specific_attr_iterator<NonNullAttr> 477 i = FDecl->specific_attr_begin<NonNullAttr>(), 478 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 479 CheckNonNullArguments(*i, TheCall->getArgs(), 480 TheCall->getCallee()->getLocStart()); 481 } 482 483 // Builtin handling 484 int CMF = -1; 485 switch (FDecl->getBuiltinID()) { 486 case Builtin::BI__builtin_memset: 487 case Builtin::BI__builtin___memset_chk: 488 case Builtin::BImemset: 489 CMF = CMF_Memset; 490 break; 491 492 case Builtin::BI__builtin_memcpy: 493 case Builtin::BI__builtin___memcpy_chk: 494 case Builtin::BImemcpy: 495 CMF = CMF_Memcpy; 496 break; 497 498 case Builtin::BI__builtin_memmove: 499 case Builtin::BI__builtin___memmove_chk: 500 case Builtin::BImemmove: 501 CMF = CMF_Memmove; 502 break; 503 504 case Builtin::BIstrlcpy: 505 case Builtin::BIstrlcat: 506 CheckStrlcpycatArguments(TheCall, FnInfo); 507 break; 508 509 case Builtin::BI__builtin_memcmp: 510 CMF = CMF_Memcmp; 511 break; 512 513 case Builtin::BI__builtin_strncpy: 514 case Builtin::BI__builtin___strncpy_chk: 515 case Builtin::BIstrncpy: 516 CMF = CMF_Strncpy; 517 break; 518 519 case Builtin::BI__builtin_strncmp: 520 CMF = CMF_Strncmp; 521 break; 522 523 case Builtin::BI__builtin_strncasecmp: 524 CMF = CMF_Strncasecmp; 525 break; 526 527 case Builtin::BI__builtin_strncat: 528 case Builtin::BIstrncat: 529 CMF = CMF_Strncat; 530 break; 531 532 case Builtin::BI__builtin_strndup: 533 case Builtin::BIstrndup: 534 CMF = CMF_Strndup; 535 break; 536 537 default: 538 if (FDecl->getLinkage() == ExternalLinkage && 539 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 540 if (FnInfo->isStr("memset")) 541 CMF = CMF_Memset; 542 else if (FnInfo->isStr("memcpy")) 543 CMF = CMF_Memcpy; 544 else if (FnInfo->isStr("memmove")) 545 CMF = CMF_Memmove; 546 else if (FnInfo->isStr("memcmp")) 547 CMF = CMF_Memcmp; 548 else if (FnInfo->isStr("strncpy")) 549 CMF = CMF_Strncpy; 550 else if (FnInfo->isStr("strncmp")) 551 CMF = CMF_Strncmp; 552 else if (FnInfo->isStr("strncasecmp")) 553 CMF = CMF_Strncasecmp; 554 else if (FnInfo->isStr("strncat")) 555 CMF = CMF_Strncat; 556 else if (FnInfo->isStr("strndup")) 557 CMF = CMF_Strndup; 558 } 559 break; 560 } 561 562 // Memset/memcpy/memmove handling 563 if (CMF != -1) 564 CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo); 565 566 return false; 567} 568 569bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 570 // Printf checking. 571 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 572 if (!Format) 573 return false; 574 575 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 576 if (!V) 577 return false; 578 579 QualType Ty = V->getType(); 580 if (!Ty->isBlockPointerType()) 581 return false; 582 583 const bool b = Format->getType() == "scanf"; 584 if (!b && !CheckablePrintfAttr(Format, TheCall)) 585 return false; 586 587 bool HasVAListArg = Format->getFirstArg() == 0; 588 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 589 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 590 591 return false; 592} 593 594ExprResult 595Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, AtomicExpr::AtomicOp Op) { 596 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 597 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 598 599 // All these operations take one of the following four forms: 600 // T __atomic_load(_Atomic(T)*, int) (loads) 601 // T* __atomic_add(_Atomic(T*)*, ptrdiff_t, int) (pointer add/sub) 602 // int __atomic_compare_exchange_strong(_Atomic(T)*, T*, T, int, int) 603 // (cmpxchg) 604 // T __atomic_exchange(_Atomic(T)*, T, int) (everything else) 605 // where T is an appropriate type, and the int paremeterss are for orderings. 606 unsigned NumVals = 1; 607 unsigned NumOrders = 1; 608 if (Op == AtomicExpr::Load) { 609 NumVals = 0; 610 } else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) { 611 NumVals = 2; 612 NumOrders = 2; 613 } 614 615 if (TheCall->getNumArgs() < NumVals+NumOrders+1) { 616 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 617 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 618 << TheCall->getCallee()->getSourceRange(); 619 return ExprError(); 620 } else if (TheCall->getNumArgs() > NumVals+NumOrders+1) { 621 Diag(TheCall->getArg(NumVals+NumOrders+1)->getLocStart(), 622 diag::err_typecheck_call_too_many_args) 623 << 0 << NumVals+NumOrders+1 << TheCall->getNumArgs() 624 << TheCall->getCallee()->getSourceRange(); 625 return ExprError(); 626 } 627 628 // Inspect the first argument of the atomic operation. This should always be 629 // a pointer to an _Atomic type. 630 Expr *Ptr = TheCall->getArg(0); 631 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 632 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 633 if (!pointerType) { 634 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 635 << Ptr->getType() << Ptr->getSourceRange(); 636 return ExprError(); 637 } 638 639 QualType AtomTy = pointerType->getPointeeType(); 640 if (!AtomTy->isAtomicType()) { 641 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 642 << Ptr->getType() << Ptr->getSourceRange(); 643 return ExprError(); 644 } 645 QualType ValType = AtomTy->getAs<AtomicType>()->getValueType(); 646 647 if ((Op == AtomicExpr::Add || Op == AtomicExpr::Sub) && 648 !ValType->isIntegerType() && !ValType->isPointerType()) { 649 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 650 << Ptr->getType() << Ptr->getSourceRange(); 651 return ExprError(); 652 } 653 654 if (!ValType->isIntegerType() && 655 (Op == AtomicExpr::And || Op == AtomicExpr::Or || Op == AtomicExpr::Xor)){ 656 Diag(DRE->getLocStart(), diag::err_atomic_op_logical_needs_atomic_int) 657 << Ptr->getType() << Ptr->getSourceRange(); 658 return ExprError(); 659 } 660 661 switch (ValType.getObjCLifetime()) { 662 case Qualifiers::OCL_None: 663 case Qualifiers::OCL_ExplicitNone: 664 // okay 665 break; 666 667 case Qualifiers::OCL_Weak: 668 case Qualifiers::OCL_Strong: 669 case Qualifiers::OCL_Autoreleasing: 670 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 671 << ValType << Ptr->getSourceRange(); 672 return ExprError(); 673 } 674 675 QualType ResultType = ValType; 676 if (Op == AtomicExpr::Store) 677 ResultType = Context.VoidTy; 678 else if (Op == AtomicExpr::CmpXchgWeak || Op == AtomicExpr::CmpXchgStrong) 679 ResultType = Context.BoolTy; 680 681 // The first argument --- the pointer --- has a fixed type; we 682 // deduce the types of the rest of the arguments accordingly. Walk 683 // the remaining arguments, converting them to the deduced value type. 684 for (unsigned i = 1; i != NumVals+NumOrders+1; ++i) { 685 ExprResult Arg = TheCall->getArg(i); 686 QualType Ty; 687 if (i < NumVals+1) { 688 // The second argument to a cmpxchg is a pointer to the data which will 689 // be exchanged. The second argument to a pointer add/subtract is the 690 // amount to add/subtract, which must be a ptrdiff_t. The third 691 // argument to a cmpxchg and the second argument in all other cases 692 // is the type of the value. 693 if (i == 1 && (Op == AtomicExpr::CmpXchgWeak || 694 Op == AtomicExpr::CmpXchgStrong)) 695 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 696 else if (!ValType->isIntegerType() && 697 (Op == AtomicExpr::Add || Op == AtomicExpr::Sub)) 698 Ty = Context.getPointerDiffType(); 699 else 700 Ty = ValType; 701 } else { 702 // The order(s) are always converted to int. 703 Ty = Context.IntTy; 704 } 705 InitializedEntity Entity = 706 InitializedEntity::InitializeParameter(Context, Ty, false); 707 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 708 if (Arg.isInvalid()) 709 return true; 710 TheCall->setArg(i, Arg.get()); 711 } 712 713 SmallVector<Expr*, 5> SubExprs; 714 SubExprs.push_back(Ptr); 715 if (Op == AtomicExpr::Load) { 716 SubExprs.push_back(TheCall->getArg(1)); // Order 717 } else if (Op != AtomicExpr::CmpXchgWeak && Op != AtomicExpr::CmpXchgStrong) { 718 SubExprs.push_back(TheCall->getArg(2)); // Order 719 SubExprs.push_back(TheCall->getArg(1)); // Val1 720 } else { 721 SubExprs.push_back(TheCall->getArg(3)); // Order 722 SubExprs.push_back(TheCall->getArg(1)); // Val1 723 SubExprs.push_back(TheCall->getArg(2)); // Val2 724 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 725 } 726 727 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 728 SubExprs.data(), SubExprs.size(), 729 ResultType, Op, 730 TheCall->getRParenLoc())); 731} 732 733 734/// checkBuiltinArgument - Given a call to a builtin function, perform 735/// normal type-checking on the given argument, updating the call in 736/// place. This is useful when a builtin function requires custom 737/// type-checking for some of its arguments but not necessarily all of 738/// them. 739/// 740/// Returns true on error. 741static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 742 FunctionDecl *Fn = E->getDirectCallee(); 743 assert(Fn && "builtin call without direct callee!"); 744 745 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 746 InitializedEntity Entity = 747 InitializedEntity::InitializeParameter(S.Context, Param); 748 749 ExprResult Arg = E->getArg(0); 750 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 751 if (Arg.isInvalid()) 752 return true; 753 754 E->setArg(ArgIndex, Arg.take()); 755 return false; 756} 757 758/// SemaBuiltinAtomicOverloaded - We have a call to a function like 759/// __sync_fetch_and_add, which is an overloaded function based on the pointer 760/// type of its first argument. The main ActOnCallExpr routines have already 761/// promoted the types of arguments because all of these calls are prototyped as 762/// void(...). 763/// 764/// This function goes through and does final semantic checking for these 765/// builtins, 766ExprResult 767Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 768 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 769 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 770 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 771 772 // Ensure that we have at least one argument to do type inference from. 773 if (TheCall->getNumArgs() < 1) { 774 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 775 << 0 << 1 << TheCall->getNumArgs() 776 << TheCall->getCallee()->getSourceRange(); 777 return ExprError(); 778 } 779 780 // Inspect the first argument of the atomic builtin. This should always be 781 // a pointer type, whose element is an integral scalar or pointer type. 782 // Because it is a pointer type, we don't have to worry about any implicit 783 // casts here. 784 // FIXME: We don't allow floating point scalars as input. 785 Expr *FirstArg = TheCall->getArg(0); 786 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 787 if (!pointerType) { 788 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 789 << FirstArg->getType() << FirstArg->getSourceRange(); 790 return ExprError(); 791 } 792 793 QualType ValType = pointerType->getPointeeType(); 794 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 795 !ValType->isBlockPointerType()) { 796 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 797 << FirstArg->getType() << FirstArg->getSourceRange(); 798 return ExprError(); 799 } 800 801 switch (ValType.getObjCLifetime()) { 802 case Qualifiers::OCL_None: 803 case Qualifiers::OCL_ExplicitNone: 804 // okay 805 break; 806 807 case Qualifiers::OCL_Weak: 808 case Qualifiers::OCL_Strong: 809 case Qualifiers::OCL_Autoreleasing: 810 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 811 << ValType << FirstArg->getSourceRange(); 812 return ExprError(); 813 } 814 815 // Strip any qualifiers off ValType. 816 ValType = ValType.getUnqualifiedType(); 817 818 // The majority of builtins return a value, but a few have special return 819 // types, so allow them to override appropriately below. 820 QualType ResultType = ValType; 821 822 // We need to figure out which concrete builtin this maps onto. For example, 823 // __sync_fetch_and_add with a 2 byte object turns into 824 // __sync_fetch_and_add_2. 825#define BUILTIN_ROW(x) \ 826 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 827 Builtin::BI##x##_8, Builtin::BI##x##_16 } 828 829 static const unsigned BuiltinIndices[][5] = { 830 BUILTIN_ROW(__sync_fetch_and_add), 831 BUILTIN_ROW(__sync_fetch_and_sub), 832 BUILTIN_ROW(__sync_fetch_and_or), 833 BUILTIN_ROW(__sync_fetch_and_and), 834 BUILTIN_ROW(__sync_fetch_and_xor), 835 836 BUILTIN_ROW(__sync_add_and_fetch), 837 BUILTIN_ROW(__sync_sub_and_fetch), 838 BUILTIN_ROW(__sync_and_and_fetch), 839 BUILTIN_ROW(__sync_or_and_fetch), 840 BUILTIN_ROW(__sync_xor_and_fetch), 841 842 BUILTIN_ROW(__sync_val_compare_and_swap), 843 BUILTIN_ROW(__sync_bool_compare_and_swap), 844 BUILTIN_ROW(__sync_lock_test_and_set), 845 BUILTIN_ROW(__sync_lock_release), 846 BUILTIN_ROW(__sync_swap) 847 }; 848#undef BUILTIN_ROW 849 850 // Determine the index of the size. 851 unsigned SizeIndex; 852 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 853 case 1: SizeIndex = 0; break; 854 case 2: SizeIndex = 1; break; 855 case 4: SizeIndex = 2; break; 856 case 8: SizeIndex = 3; break; 857 case 16: SizeIndex = 4; break; 858 default: 859 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 860 << FirstArg->getType() << FirstArg->getSourceRange(); 861 return ExprError(); 862 } 863 864 // Each of these builtins has one pointer argument, followed by some number of 865 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 866 // that we ignore. Find out which row of BuiltinIndices to read from as well 867 // as the number of fixed args. 868 unsigned BuiltinID = FDecl->getBuiltinID(); 869 unsigned BuiltinIndex, NumFixed = 1; 870 switch (BuiltinID) { 871 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 872 case Builtin::BI__sync_fetch_and_add: 873 case Builtin::BI__sync_fetch_and_add_1: 874 case Builtin::BI__sync_fetch_and_add_2: 875 case Builtin::BI__sync_fetch_and_add_4: 876 case Builtin::BI__sync_fetch_and_add_8: 877 case Builtin::BI__sync_fetch_and_add_16: 878 BuiltinIndex = 0; 879 break; 880 881 case Builtin::BI__sync_fetch_and_sub: 882 case Builtin::BI__sync_fetch_and_sub_1: 883 case Builtin::BI__sync_fetch_and_sub_2: 884 case Builtin::BI__sync_fetch_and_sub_4: 885 case Builtin::BI__sync_fetch_and_sub_8: 886 case Builtin::BI__sync_fetch_and_sub_16: 887 BuiltinIndex = 1; 888 break; 889 890 case Builtin::BI__sync_fetch_and_or: 891 case Builtin::BI__sync_fetch_and_or_1: 892 case Builtin::BI__sync_fetch_and_or_2: 893 case Builtin::BI__sync_fetch_and_or_4: 894 case Builtin::BI__sync_fetch_and_or_8: 895 case Builtin::BI__sync_fetch_and_or_16: 896 BuiltinIndex = 2; 897 break; 898 899 case Builtin::BI__sync_fetch_and_and: 900 case Builtin::BI__sync_fetch_and_and_1: 901 case Builtin::BI__sync_fetch_and_and_2: 902 case Builtin::BI__sync_fetch_and_and_4: 903 case Builtin::BI__sync_fetch_and_and_8: 904 case Builtin::BI__sync_fetch_and_and_16: 905 BuiltinIndex = 3; 906 break; 907 908 case Builtin::BI__sync_fetch_and_xor: 909 case Builtin::BI__sync_fetch_and_xor_1: 910 case Builtin::BI__sync_fetch_and_xor_2: 911 case Builtin::BI__sync_fetch_and_xor_4: 912 case Builtin::BI__sync_fetch_and_xor_8: 913 case Builtin::BI__sync_fetch_and_xor_16: 914 BuiltinIndex = 4; 915 break; 916 917 case Builtin::BI__sync_add_and_fetch: 918 case Builtin::BI__sync_add_and_fetch_1: 919 case Builtin::BI__sync_add_and_fetch_2: 920 case Builtin::BI__sync_add_and_fetch_4: 921 case Builtin::BI__sync_add_and_fetch_8: 922 case Builtin::BI__sync_add_and_fetch_16: 923 BuiltinIndex = 5; 924 break; 925 926 case Builtin::BI__sync_sub_and_fetch: 927 case Builtin::BI__sync_sub_and_fetch_1: 928 case Builtin::BI__sync_sub_and_fetch_2: 929 case Builtin::BI__sync_sub_and_fetch_4: 930 case Builtin::BI__sync_sub_and_fetch_8: 931 case Builtin::BI__sync_sub_and_fetch_16: 932 BuiltinIndex = 6; 933 break; 934 935 case Builtin::BI__sync_and_and_fetch: 936 case Builtin::BI__sync_and_and_fetch_1: 937 case Builtin::BI__sync_and_and_fetch_2: 938 case Builtin::BI__sync_and_and_fetch_4: 939 case Builtin::BI__sync_and_and_fetch_8: 940 case Builtin::BI__sync_and_and_fetch_16: 941 BuiltinIndex = 7; 942 break; 943 944 case Builtin::BI__sync_or_and_fetch: 945 case Builtin::BI__sync_or_and_fetch_1: 946 case Builtin::BI__sync_or_and_fetch_2: 947 case Builtin::BI__sync_or_and_fetch_4: 948 case Builtin::BI__sync_or_and_fetch_8: 949 case Builtin::BI__sync_or_and_fetch_16: 950 BuiltinIndex = 8; 951 break; 952 953 case Builtin::BI__sync_xor_and_fetch: 954 case Builtin::BI__sync_xor_and_fetch_1: 955 case Builtin::BI__sync_xor_and_fetch_2: 956 case Builtin::BI__sync_xor_and_fetch_4: 957 case Builtin::BI__sync_xor_and_fetch_8: 958 case Builtin::BI__sync_xor_and_fetch_16: 959 BuiltinIndex = 9; 960 break; 961 962 case Builtin::BI__sync_val_compare_and_swap: 963 case Builtin::BI__sync_val_compare_and_swap_1: 964 case Builtin::BI__sync_val_compare_and_swap_2: 965 case Builtin::BI__sync_val_compare_and_swap_4: 966 case Builtin::BI__sync_val_compare_and_swap_8: 967 case Builtin::BI__sync_val_compare_and_swap_16: 968 BuiltinIndex = 10; 969 NumFixed = 2; 970 break; 971 972 case Builtin::BI__sync_bool_compare_and_swap: 973 case Builtin::BI__sync_bool_compare_and_swap_1: 974 case Builtin::BI__sync_bool_compare_and_swap_2: 975 case Builtin::BI__sync_bool_compare_and_swap_4: 976 case Builtin::BI__sync_bool_compare_and_swap_8: 977 case Builtin::BI__sync_bool_compare_and_swap_16: 978 BuiltinIndex = 11; 979 NumFixed = 2; 980 ResultType = Context.BoolTy; 981 break; 982 983 case Builtin::BI__sync_lock_test_and_set: 984 case Builtin::BI__sync_lock_test_and_set_1: 985 case Builtin::BI__sync_lock_test_and_set_2: 986 case Builtin::BI__sync_lock_test_and_set_4: 987 case Builtin::BI__sync_lock_test_and_set_8: 988 case Builtin::BI__sync_lock_test_and_set_16: 989 BuiltinIndex = 12; 990 break; 991 992 case Builtin::BI__sync_lock_release: 993 case Builtin::BI__sync_lock_release_1: 994 case Builtin::BI__sync_lock_release_2: 995 case Builtin::BI__sync_lock_release_4: 996 case Builtin::BI__sync_lock_release_8: 997 case Builtin::BI__sync_lock_release_16: 998 BuiltinIndex = 13; 999 NumFixed = 0; 1000 ResultType = Context.VoidTy; 1001 break; 1002 1003 case Builtin::BI__sync_swap: 1004 case Builtin::BI__sync_swap_1: 1005 case Builtin::BI__sync_swap_2: 1006 case Builtin::BI__sync_swap_4: 1007 case Builtin::BI__sync_swap_8: 1008 case Builtin::BI__sync_swap_16: 1009 BuiltinIndex = 14; 1010 break; 1011 } 1012 1013 // Now that we know how many fixed arguments we expect, first check that we 1014 // have at least that many. 1015 if (TheCall->getNumArgs() < 1+NumFixed) { 1016 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1017 << 0 << 1+NumFixed << TheCall->getNumArgs() 1018 << TheCall->getCallee()->getSourceRange(); 1019 return ExprError(); 1020 } 1021 1022 // Get the decl for the concrete builtin from this, we can tell what the 1023 // concrete integer type we should convert to is. 1024 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1025 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1026 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 1027 FunctionDecl *NewBuiltinDecl = 1028 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 1029 TUScope, false, DRE->getLocStart())); 1030 1031 // The first argument --- the pointer --- has a fixed type; we 1032 // deduce the types of the rest of the arguments accordingly. Walk 1033 // the remaining arguments, converting them to the deduced value type. 1034 for (unsigned i = 0; i != NumFixed; ++i) { 1035 ExprResult Arg = TheCall->getArg(i+1); 1036 1037 // GCC does an implicit conversion to the pointer or integer ValType. This 1038 // can fail in some cases (1i -> int**), check for this error case now. 1039 // Initialize the argument. 1040 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1041 ValType, /*consume*/ false); 1042 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1043 if (Arg.isInvalid()) 1044 return ExprError(); 1045 1046 // Okay, we have something that *can* be converted to the right type. Check 1047 // to see if there is a potentially weird extension going on here. This can 1048 // happen when you do an atomic operation on something like an char* and 1049 // pass in 42. The 42 gets converted to char. This is even more strange 1050 // for things like 45.123 -> char, etc. 1051 // FIXME: Do this check. 1052 TheCall->setArg(i+1, Arg.take()); 1053 } 1054 1055 ASTContext& Context = this->getASTContext(); 1056 1057 // Create a new DeclRefExpr to refer to the new decl. 1058 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1059 Context, 1060 DRE->getQualifierLoc(), 1061 NewBuiltinDecl, 1062 DRE->getLocation(), 1063 NewBuiltinDecl->getType(), 1064 DRE->getValueKind()); 1065 1066 // Set the callee in the CallExpr. 1067 // FIXME: This leaks the original parens and implicit casts. 1068 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 1069 if (PromotedCall.isInvalid()) 1070 return ExprError(); 1071 TheCall->setCallee(PromotedCall.take()); 1072 1073 // Change the result type of the call to match the original value type. This 1074 // is arbitrary, but the codegen for these builtins ins design to handle it 1075 // gracefully. 1076 TheCall->setType(ResultType); 1077 1078 return move(TheCallResult); 1079} 1080 1081/// CheckObjCString - Checks that the argument to the builtin 1082/// CFString constructor is correct 1083/// Note: It might also make sense to do the UTF-16 conversion here (would 1084/// simplify the backend). 1085bool Sema::CheckObjCString(Expr *Arg) { 1086 Arg = Arg->IgnoreParenCasts(); 1087 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1088 1089 if (!Literal || !Literal->isAscii()) { 1090 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1091 << Arg->getSourceRange(); 1092 return true; 1093 } 1094 1095 if (Literal->containsNonAsciiOrNull()) { 1096 StringRef String = Literal->getString(); 1097 unsigned NumBytes = String.size(); 1098 SmallVector<UTF16, 128> ToBuf(NumBytes); 1099 const UTF8 *FromPtr = (UTF8 *)String.data(); 1100 UTF16 *ToPtr = &ToBuf[0]; 1101 1102 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1103 &ToPtr, ToPtr + NumBytes, 1104 strictConversion); 1105 // Check for conversion failure. 1106 if (Result != conversionOK) 1107 Diag(Arg->getLocStart(), 1108 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1109 } 1110 return false; 1111} 1112 1113/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1114/// Emit an error and return true on failure, return false on success. 1115bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1116 Expr *Fn = TheCall->getCallee(); 1117 if (TheCall->getNumArgs() > 2) { 1118 Diag(TheCall->getArg(2)->getLocStart(), 1119 diag::err_typecheck_call_too_many_args) 1120 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1121 << Fn->getSourceRange() 1122 << SourceRange(TheCall->getArg(2)->getLocStart(), 1123 (*(TheCall->arg_end()-1))->getLocEnd()); 1124 return true; 1125 } 1126 1127 if (TheCall->getNumArgs() < 2) { 1128 return Diag(TheCall->getLocEnd(), 1129 diag::err_typecheck_call_too_few_args_at_least) 1130 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1131 } 1132 1133 // Type-check the first argument normally. 1134 if (checkBuiltinArgument(*this, TheCall, 0)) 1135 return true; 1136 1137 // Determine whether the current function is variadic or not. 1138 BlockScopeInfo *CurBlock = getCurBlock(); 1139 bool isVariadic; 1140 if (CurBlock) 1141 isVariadic = CurBlock->TheDecl->isVariadic(); 1142 else if (FunctionDecl *FD = getCurFunctionDecl()) 1143 isVariadic = FD->isVariadic(); 1144 else 1145 isVariadic = getCurMethodDecl()->isVariadic(); 1146 1147 if (!isVariadic) { 1148 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1149 return true; 1150 } 1151 1152 // Verify that the second argument to the builtin is the last argument of the 1153 // current function or method. 1154 bool SecondArgIsLastNamedArgument = false; 1155 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1156 1157 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1158 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1159 // FIXME: This isn't correct for methods (results in bogus warning). 1160 // Get the last formal in the current function. 1161 const ParmVarDecl *LastArg; 1162 if (CurBlock) 1163 LastArg = *(CurBlock->TheDecl->param_end()-1); 1164 else if (FunctionDecl *FD = getCurFunctionDecl()) 1165 LastArg = *(FD->param_end()-1); 1166 else 1167 LastArg = *(getCurMethodDecl()->param_end()-1); 1168 SecondArgIsLastNamedArgument = PV == LastArg; 1169 } 1170 } 1171 1172 if (!SecondArgIsLastNamedArgument) 1173 Diag(TheCall->getArg(1)->getLocStart(), 1174 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1175 return false; 1176} 1177 1178/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1179/// friends. This is declared to take (...), so we have to check everything. 1180bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1181 if (TheCall->getNumArgs() < 2) 1182 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1183 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1184 if (TheCall->getNumArgs() > 2) 1185 return Diag(TheCall->getArg(2)->getLocStart(), 1186 diag::err_typecheck_call_too_many_args) 1187 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1188 << SourceRange(TheCall->getArg(2)->getLocStart(), 1189 (*(TheCall->arg_end()-1))->getLocEnd()); 1190 1191 ExprResult OrigArg0 = TheCall->getArg(0); 1192 ExprResult OrigArg1 = TheCall->getArg(1); 1193 1194 // Do standard promotions between the two arguments, returning their common 1195 // type. 1196 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1197 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1198 return true; 1199 1200 // Make sure any conversions are pushed back into the call; this is 1201 // type safe since unordered compare builtins are declared as "_Bool 1202 // foo(...)". 1203 TheCall->setArg(0, OrigArg0.get()); 1204 TheCall->setArg(1, OrigArg1.get()); 1205 1206 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1207 return false; 1208 1209 // If the common type isn't a real floating type, then the arguments were 1210 // invalid for this operation. 1211 if (!Res->isRealFloatingType()) 1212 return Diag(OrigArg0.get()->getLocStart(), 1213 diag::err_typecheck_call_invalid_ordered_compare) 1214 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1215 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1216 1217 return false; 1218} 1219 1220/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1221/// __builtin_isnan and friends. This is declared to take (...), so we have 1222/// to check everything. We expect the last argument to be a floating point 1223/// value. 1224bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1225 if (TheCall->getNumArgs() < NumArgs) 1226 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1227 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1228 if (TheCall->getNumArgs() > NumArgs) 1229 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1230 diag::err_typecheck_call_too_many_args) 1231 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1232 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1233 (*(TheCall->arg_end()-1))->getLocEnd()); 1234 1235 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1236 1237 if (OrigArg->isTypeDependent()) 1238 return false; 1239 1240 // This operation requires a non-_Complex floating-point number. 1241 if (!OrigArg->getType()->isRealFloatingType()) 1242 return Diag(OrigArg->getLocStart(), 1243 diag::err_typecheck_call_invalid_unary_fp) 1244 << OrigArg->getType() << OrigArg->getSourceRange(); 1245 1246 // If this is an implicit conversion from float -> double, remove it. 1247 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1248 Expr *CastArg = Cast->getSubExpr(); 1249 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1250 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1251 "promotion from float to double is the only expected cast here"); 1252 Cast->setSubExpr(0); 1253 TheCall->setArg(NumArgs-1, CastArg); 1254 OrigArg = CastArg; 1255 } 1256 } 1257 1258 return false; 1259} 1260 1261/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1262// This is declared to take (...), so we have to check everything. 1263ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1264 if (TheCall->getNumArgs() < 2) 1265 return ExprError(Diag(TheCall->getLocEnd(), 1266 diag::err_typecheck_call_too_few_args_at_least) 1267 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1268 << TheCall->getSourceRange()); 1269 1270 // Determine which of the following types of shufflevector we're checking: 1271 // 1) unary, vector mask: (lhs, mask) 1272 // 2) binary, vector mask: (lhs, rhs, mask) 1273 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1274 QualType resType = TheCall->getArg(0)->getType(); 1275 unsigned numElements = 0; 1276 1277 if (!TheCall->getArg(0)->isTypeDependent() && 1278 !TheCall->getArg(1)->isTypeDependent()) { 1279 QualType LHSType = TheCall->getArg(0)->getType(); 1280 QualType RHSType = TheCall->getArg(1)->getType(); 1281 1282 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1283 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1284 << SourceRange(TheCall->getArg(0)->getLocStart(), 1285 TheCall->getArg(1)->getLocEnd()); 1286 return ExprError(); 1287 } 1288 1289 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1290 unsigned numResElements = TheCall->getNumArgs() - 2; 1291 1292 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1293 // with mask. If so, verify that RHS is an integer vector type with the 1294 // same number of elts as lhs. 1295 if (TheCall->getNumArgs() == 2) { 1296 if (!RHSType->hasIntegerRepresentation() || 1297 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1298 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1299 << SourceRange(TheCall->getArg(1)->getLocStart(), 1300 TheCall->getArg(1)->getLocEnd()); 1301 numResElements = numElements; 1302 } 1303 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1304 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1305 << SourceRange(TheCall->getArg(0)->getLocStart(), 1306 TheCall->getArg(1)->getLocEnd()); 1307 return ExprError(); 1308 } else if (numElements != numResElements) { 1309 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1310 resType = Context.getVectorType(eltType, numResElements, 1311 VectorType::GenericVector); 1312 } 1313 } 1314 1315 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1316 if (TheCall->getArg(i)->isTypeDependent() || 1317 TheCall->getArg(i)->isValueDependent()) 1318 continue; 1319 1320 llvm::APSInt Result(32); 1321 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1322 return ExprError(Diag(TheCall->getLocStart(), 1323 diag::err_shufflevector_nonconstant_argument) 1324 << TheCall->getArg(i)->getSourceRange()); 1325 1326 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1327 return ExprError(Diag(TheCall->getLocStart(), 1328 diag::err_shufflevector_argument_too_large) 1329 << TheCall->getArg(i)->getSourceRange()); 1330 } 1331 1332 SmallVector<Expr*, 32> exprs; 1333 1334 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1335 exprs.push_back(TheCall->getArg(i)); 1336 TheCall->setArg(i, 0); 1337 } 1338 1339 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1340 exprs.size(), resType, 1341 TheCall->getCallee()->getLocStart(), 1342 TheCall->getRParenLoc())); 1343} 1344 1345/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1346// This is declared to take (const void*, ...) and can take two 1347// optional constant int args. 1348bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1349 unsigned NumArgs = TheCall->getNumArgs(); 1350 1351 if (NumArgs > 3) 1352 return Diag(TheCall->getLocEnd(), 1353 diag::err_typecheck_call_too_many_args_at_most) 1354 << 0 /*function call*/ << 3 << NumArgs 1355 << TheCall->getSourceRange(); 1356 1357 // Argument 0 is checked for us and the remaining arguments must be 1358 // constant integers. 1359 for (unsigned i = 1; i != NumArgs; ++i) { 1360 Expr *Arg = TheCall->getArg(i); 1361 1362 llvm::APSInt Result; 1363 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1364 return true; 1365 1366 // FIXME: gcc issues a warning and rewrites these to 0. These 1367 // seems especially odd for the third argument since the default 1368 // is 3. 1369 if (i == 1) { 1370 if (Result.getLimitedValue() > 1) 1371 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1372 << "0" << "1" << Arg->getSourceRange(); 1373 } else { 1374 if (Result.getLimitedValue() > 3) 1375 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1376 << "0" << "3" << Arg->getSourceRange(); 1377 } 1378 } 1379 1380 return false; 1381} 1382 1383/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1384/// TheCall is a constant expression. 1385bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1386 llvm::APSInt &Result) { 1387 Expr *Arg = TheCall->getArg(ArgNum); 1388 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1389 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1390 1391 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1392 1393 if (!Arg->isIntegerConstantExpr(Result, Context)) 1394 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1395 << FDecl->getDeclName() << Arg->getSourceRange(); 1396 1397 return false; 1398} 1399 1400/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1401/// int type). This simply type checks that type is one of the defined 1402/// constants (0-3). 1403// For compatibility check 0-3, llvm only handles 0 and 2. 1404bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1405 llvm::APSInt Result; 1406 1407 // Check constant-ness first. 1408 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1409 return true; 1410 1411 Expr *Arg = TheCall->getArg(1); 1412 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1413 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1414 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1415 } 1416 1417 return false; 1418} 1419 1420/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1421/// This checks that val is a constant 1. 1422bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1423 Expr *Arg = TheCall->getArg(1); 1424 llvm::APSInt Result; 1425 1426 // TODO: This is less than ideal. Overload this to take a value. 1427 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1428 return true; 1429 1430 if (Result != 1) 1431 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1432 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1433 1434 return false; 1435} 1436 1437// Handle i > 1 ? "x" : "y", recursively. 1438bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1439 bool HasVAListArg, 1440 unsigned format_idx, unsigned firstDataArg, 1441 bool isPrintf, bool inFunctionCall) { 1442 tryAgain: 1443 if (E->isTypeDependent() || E->isValueDependent()) 1444 return false; 1445 1446 E = E->IgnoreParens(); 1447 1448 switch (E->getStmtClass()) { 1449 case Stmt::BinaryConditionalOperatorClass: 1450 case Stmt::ConditionalOperatorClass: { 1451 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1452 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 1453 format_idx, firstDataArg, isPrintf, 1454 inFunctionCall) 1455 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 1456 format_idx, firstDataArg, isPrintf, 1457 inFunctionCall); 1458 } 1459 1460 case Stmt::IntegerLiteralClass: 1461 // Technically -Wformat-nonliteral does not warn about this case. 1462 // The behavior of printf and friends in this case is implementation 1463 // dependent. Ideally if the format string cannot be null then 1464 // it should have a 'nonnull' attribute in the function prototype. 1465 return true; 1466 1467 case Stmt::ImplicitCastExprClass: { 1468 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1469 goto tryAgain; 1470 } 1471 1472 case Stmt::OpaqueValueExprClass: 1473 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1474 E = src; 1475 goto tryAgain; 1476 } 1477 return false; 1478 1479 case Stmt::PredefinedExprClass: 1480 // While __func__, etc., are technically not string literals, they 1481 // cannot contain format specifiers and thus are not a security 1482 // liability. 1483 return true; 1484 1485 case Stmt::DeclRefExprClass: { 1486 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1487 1488 // As an exception, do not flag errors for variables binding to 1489 // const string literals. 1490 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1491 bool isConstant = false; 1492 QualType T = DR->getType(); 1493 1494 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1495 isConstant = AT->getElementType().isConstant(Context); 1496 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1497 isConstant = T.isConstant(Context) && 1498 PT->getPointeeType().isConstant(Context); 1499 } 1500 1501 if (isConstant) { 1502 if (const Expr *Init = VD->getAnyInitializer()) 1503 return SemaCheckStringLiteral(Init, TheCall, 1504 HasVAListArg, format_idx, firstDataArg, 1505 isPrintf, /*inFunctionCall*/false); 1506 } 1507 1508 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1509 // special check to see if the format string is a function parameter 1510 // of the function calling the printf function. If the function 1511 // has an attribute indicating it is a printf-like function, then we 1512 // should suppress warnings concerning non-literals being used in a call 1513 // to a vprintf function. For example: 1514 // 1515 // void 1516 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1517 // va_list ap; 1518 // va_start(ap, fmt); 1519 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1520 // ... 1521 // 1522 // 1523 // FIXME: We don't have full attribute support yet, so just check to see 1524 // if the argument is a DeclRefExpr that references a parameter. We'll 1525 // add proper support for checking the attribute later. 1526 if (HasVAListArg) 1527 if (isa<ParmVarDecl>(VD)) 1528 return true; 1529 } 1530 1531 return false; 1532 } 1533 1534 case Stmt::CallExprClass: { 1535 const CallExpr *CE = cast<CallExpr>(E); 1536 if (const ImplicitCastExpr *ICE 1537 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1538 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1539 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1540 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1541 unsigned ArgIndex = FA->getFormatIdx(); 1542 const Expr *Arg = CE->getArg(ArgIndex - 1); 1543 1544 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1545 format_idx, firstDataArg, isPrintf, 1546 inFunctionCall); 1547 } 1548 } 1549 } 1550 } 1551 1552 return false; 1553 } 1554 case Stmt::ObjCStringLiteralClass: 1555 case Stmt::StringLiteralClass: { 1556 const StringLiteral *StrE = NULL; 1557 1558 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1559 StrE = ObjCFExpr->getString(); 1560 else 1561 StrE = cast<StringLiteral>(E); 1562 1563 if (StrE) { 1564 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1565 firstDataArg, isPrintf, inFunctionCall); 1566 return true; 1567 } 1568 1569 return false; 1570 } 1571 1572 default: 1573 return false; 1574 } 1575} 1576 1577void 1578Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1579 const Expr * const *ExprArgs, 1580 SourceLocation CallSiteLoc) { 1581 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1582 e = NonNull->args_end(); 1583 i != e; ++i) { 1584 const Expr *ArgExpr = ExprArgs[*i]; 1585 if (ArgExpr->isNullPointerConstant(Context, 1586 Expr::NPC_ValueDependentIsNotNull)) 1587 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1588 } 1589} 1590 1591/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1592/// functions) for correct use of format strings. 1593void 1594Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1595 unsigned format_idx, unsigned firstDataArg, 1596 bool isPrintf) { 1597 1598 const Expr *Fn = TheCall->getCallee(); 1599 1600 // The way the format attribute works in GCC, the implicit this argument 1601 // of member functions is counted. However, it doesn't appear in our own 1602 // lists, so decrement format_idx in that case. 1603 if (isa<CXXMemberCallExpr>(TheCall)) { 1604 const CXXMethodDecl *method_decl = 1605 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1606 if (method_decl && method_decl->isInstance()) { 1607 // Catch a format attribute mistakenly referring to the object argument. 1608 if (format_idx == 0) 1609 return; 1610 --format_idx; 1611 if(firstDataArg != 0) 1612 --firstDataArg; 1613 } 1614 } 1615 1616 // CHECK: printf/scanf-like function is called with no format string. 1617 if (format_idx >= TheCall->getNumArgs()) { 1618 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1619 << Fn->getSourceRange(); 1620 return; 1621 } 1622 1623 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1624 1625 // CHECK: format string is not a string literal. 1626 // 1627 // Dynamically generated format strings are difficult to 1628 // automatically vet at compile time. Requiring that format strings 1629 // are string literals: (1) permits the checking of format strings by 1630 // the compiler and thereby (2) can practically remove the source of 1631 // many format string exploits. 1632 1633 // Format string can be either ObjC string (e.g. @"%d") or 1634 // C string (e.g. "%d") 1635 // ObjC string uses the same format specifiers as C string, so we can use 1636 // the same format string checking logic for both ObjC and C strings. 1637 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1638 firstDataArg, isPrintf)) 1639 return; // Literal format string found, check done! 1640 1641 // If there are no arguments specified, warn with -Wformat-security, otherwise 1642 // warn only with -Wformat-nonliteral. 1643 if (TheCall->getNumArgs() == format_idx+1) 1644 Diag(TheCall->getArg(format_idx)->getLocStart(), 1645 diag::warn_format_nonliteral_noargs) 1646 << OrigFormatExpr->getSourceRange(); 1647 else 1648 Diag(TheCall->getArg(format_idx)->getLocStart(), 1649 diag::warn_format_nonliteral) 1650 << OrigFormatExpr->getSourceRange(); 1651} 1652 1653namespace { 1654class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1655protected: 1656 Sema &S; 1657 const StringLiteral *FExpr; 1658 const Expr *OrigFormatExpr; 1659 const unsigned FirstDataArg; 1660 const unsigned NumDataArgs; 1661 const bool IsObjCLiteral; 1662 const char *Beg; // Start of format string. 1663 const bool HasVAListArg; 1664 const CallExpr *TheCall; 1665 unsigned FormatIdx; 1666 llvm::BitVector CoveredArgs; 1667 bool usesPositionalArgs; 1668 bool atFirstArg; 1669 bool inFunctionCall; 1670public: 1671 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1672 const Expr *origFormatExpr, unsigned firstDataArg, 1673 unsigned numDataArgs, bool isObjCLiteral, 1674 const char *beg, bool hasVAListArg, 1675 const CallExpr *theCall, unsigned formatIdx, 1676 bool inFunctionCall) 1677 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1678 FirstDataArg(firstDataArg), 1679 NumDataArgs(numDataArgs), 1680 IsObjCLiteral(isObjCLiteral), Beg(beg), 1681 HasVAListArg(hasVAListArg), 1682 TheCall(theCall), FormatIdx(formatIdx), 1683 usesPositionalArgs(false), atFirstArg(true), 1684 inFunctionCall(inFunctionCall) { 1685 CoveredArgs.resize(numDataArgs); 1686 CoveredArgs.reset(); 1687 } 1688 1689 void DoneProcessing(); 1690 1691 void HandleIncompleteSpecifier(const char *startSpecifier, 1692 unsigned specifierLen); 1693 1694 virtual void HandleInvalidPosition(const char *startSpecifier, 1695 unsigned specifierLen, 1696 analyze_format_string::PositionContext p); 1697 1698 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1699 1700 void HandleNullChar(const char *nullCharacter); 1701 1702 template <typename Range> 1703 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1704 const Expr *ArgumentExpr, 1705 PartialDiagnostic PDiag, 1706 SourceLocation StringLoc, 1707 bool IsStringLocation, Range StringRange, 1708 FixItHint Fixit = FixItHint()); 1709 1710protected: 1711 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1712 const char *startSpec, 1713 unsigned specifierLen, 1714 const char *csStart, unsigned csLen); 1715 1716 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1717 const char *startSpec, 1718 unsigned specifierLen); 1719 1720 SourceRange getFormatStringRange(); 1721 CharSourceRange getSpecifierRange(const char *startSpecifier, 1722 unsigned specifierLen); 1723 SourceLocation getLocationOfByte(const char *x); 1724 1725 const Expr *getDataArg(unsigned i) const; 1726 1727 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1728 const analyze_format_string::ConversionSpecifier &CS, 1729 const char *startSpecifier, unsigned specifierLen, 1730 unsigned argIndex); 1731 1732 template <typename Range> 1733 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1734 bool IsStringLocation, Range StringRange, 1735 FixItHint Fixit = FixItHint()); 1736 1737 void CheckPositionalAndNonpositionalArgs( 1738 const analyze_format_string::FormatSpecifier *FS); 1739}; 1740} 1741 1742SourceRange CheckFormatHandler::getFormatStringRange() { 1743 return OrigFormatExpr->getSourceRange(); 1744} 1745 1746CharSourceRange CheckFormatHandler:: 1747getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1748 SourceLocation Start = getLocationOfByte(startSpecifier); 1749 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1750 1751 // Advance the end SourceLocation by one due to half-open ranges. 1752 End = End.getLocWithOffset(1); 1753 1754 return CharSourceRange::getCharRange(Start, End); 1755} 1756 1757SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1758 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1759} 1760 1761void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1762 unsigned specifierLen){ 1763 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1764 getLocationOfByte(startSpecifier), 1765 /*IsStringLocation*/true, 1766 getSpecifierRange(startSpecifier, specifierLen)); 1767} 1768 1769void 1770CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1771 analyze_format_string::PositionContext p) { 1772 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1773 << (unsigned) p, 1774 getLocationOfByte(startPos), /*IsStringLocation*/true, 1775 getSpecifierRange(startPos, posLen)); 1776} 1777 1778void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1779 unsigned posLen) { 1780 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1781 getLocationOfByte(startPos), 1782 /*IsStringLocation*/true, 1783 getSpecifierRange(startPos, posLen)); 1784} 1785 1786void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1787 if (!IsObjCLiteral) { 1788 // The presence of a null character is likely an error. 1789 EmitFormatDiagnostic( 1790 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1791 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1792 getFormatStringRange()); 1793 } 1794} 1795 1796const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1797 return TheCall->getArg(FirstDataArg + i); 1798} 1799 1800void CheckFormatHandler::DoneProcessing() { 1801 // Does the number of data arguments exceed the number of 1802 // format conversions in the format string? 1803 if (!HasVAListArg) { 1804 // Find any arguments that weren't covered. 1805 CoveredArgs.flip(); 1806 signed notCoveredArg = CoveredArgs.find_first(); 1807 if (notCoveredArg >= 0) { 1808 assert((unsigned)notCoveredArg < NumDataArgs); 1809 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1810 getDataArg((unsigned) notCoveredArg)->getLocStart(), 1811 /*IsStringLocation*/false, getFormatStringRange()); 1812 } 1813 } 1814} 1815 1816bool 1817CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1818 SourceLocation Loc, 1819 const char *startSpec, 1820 unsigned specifierLen, 1821 const char *csStart, 1822 unsigned csLen) { 1823 1824 bool keepGoing = true; 1825 if (argIndex < NumDataArgs) { 1826 // Consider the argument coverered, even though the specifier doesn't 1827 // make sense. 1828 CoveredArgs.set(argIndex); 1829 } 1830 else { 1831 // If argIndex exceeds the number of data arguments we 1832 // don't issue a warning because that is just a cascade of warnings (and 1833 // they may have intended '%%' anyway). We don't want to continue processing 1834 // the format string after this point, however, as we will like just get 1835 // gibberish when trying to match arguments. 1836 keepGoing = false; 1837 } 1838 1839 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 1840 << StringRef(csStart, csLen), 1841 Loc, /*IsStringLocation*/true, 1842 getSpecifierRange(startSpec, specifierLen)); 1843 1844 return keepGoing; 1845} 1846 1847void 1848CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 1849 const char *startSpec, 1850 unsigned specifierLen) { 1851 EmitFormatDiagnostic( 1852 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 1853 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 1854} 1855 1856bool 1857CheckFormatHandler::CheckNumArgs( 1858 const analyze_format_string::FormatSpecifier &FS, 1859 const analyze_format_string::ConversionSpecifier &CS, 1860 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1861 1862 if (argIndex >= NumDataArgs) { 1863 PartialDiagnostic PDiag = FS.usesPositionalArg() 1864 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 1865 << (argIndex+1) << NumDataArgs) 1866 : S.PDiag(diag::warn_printf_insufficient_data_args); 1867 EmitFormatDiagnostic( 1868 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 1869 getSpecifierRange(startSpecifier, specifierLen)); 1870 return false; 1871 } 1872 return true; 1873} 1874 1875template<typename Range> 1876void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 1877 SourceLocation Loc, 1878 bool IsStringLocation, 1879 Range StringRange, 1880 FixItHint FixIt) { 1881 EmitFormatDiagnostic(S, inFunctionCall, TheCall->getArg(FormatIdx), PDiag, 1882 Loc, IsStringLocation, StringRange, FixIt); 1883} 1884 1885/// \brief If the format string is not within the funcion call, emit a note 1886/// so that the function call and string are in diagnostic messages. 1887/// 1888/// \param inFunctionCall if true, the format string is within the function 1889/// call and only one diagnostic message will be produced. Otherwise, an 1890/// extra note will be emitted pointing to location of the format string. 1891/// 1892/// \param ArgumentExpr the expression that is passed as the format string 1893/// argument in the function call. Used for getting locations when two 1894/// diagnostics are emitted. 1895/// 1896/// \param PDiag the callee should already have provided any strings for the 1897/// diagnostic message. This function only adds locations and fixits 1898/// to diagnostics. 1899/// 1900/// \param Loc primary location for diagnostic. If two diagnostics are 1901/// required, one will be at Loc and a new SourceLocation will be created for 1902/// the other one. 1903/// 1904/// \param IsStringLocation if true, Loc points to the format string should be 1905/// used for the note. Otherwise, Loc points to the argument list and will 1906/// be used with PDiag. 1907/// 1908/// \param StringRange some or all of the string to highlight. This is 1909/// templated so it can accept either a CharSourceRange or a SourceRange. 1910/// 1911/// \param Fixit optional fix it hint for the format string. 1912template<typename Range> 1913void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 1914 const Expr *ArgumentExpr, 1915 PartialDiagnostic PDiag, 1916 SourceLocation Loc, 1917 bool IsStringLocation, 1918 Range StringRange, 1919 FixItHint FixIt) { 1920 if (InFunctionCall) 1921 S.Diag(Loc, PDiag) << StringRange << FixIt; 1922 else { 1923 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 1924 << ArgumentExpr->getSourceRange(); 1925 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 1926 diag::note_format_string_defined) 1927 << StringRange << FixIt; 1928 } 1929} 1930 1931//===--- CHECK: Printf format string checking ------------------------------===// 1932 1933namespace { 1934class CheckPrintfHandler : public CheckFormatHandler { 1935public: 1936 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1937 const Expr *origFormatExpr, unsigned firstDataArg, 1938 unsigned numDataArgs, bool isObjCLiteral, 1939 const char *beg, bool hasVAListArg, 1940 const CallExpr *theCall, unsigned formatIdx, 1941 bool inFunctionCall) 1942 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1943 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1944 theCall, formatIdx, inFunctionCall) {} 1945 1946 1947 bool HandleInvalidPrintfConversionSpecifier( 1948 const analyze_printf::PrintfSpecifier &FS, 1949 const char *startSpecifier, 1950 unsigned specifierLen); 1951 1952 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1953 const char *startSpecifier, 1954 unsigned specifierLen); 1955 1956 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1957 const char *startSpecifier, unsigned specifierLen); 1958 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1959 const analyze_printf::OptionalAmount &Amt, 1960 unsigned type, 1961 const char *startSpecifier, unsigned specifierLen); 1962 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1963 const analyze_printf::OptionalFlag &flag, 1964 const char *startSpecifier, unsigned specifierLen); 1965 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1966 const analyze_printf::OptionalFlag &ignoredFlag, 1967 const analyze_printf::OptionalFlag &flag, 1968 const char *startSpecifier, unsigned specifierLen); 1969}; 1970} 1971 1972bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1973 const analyze_printf::PrintfSpecifier &FS, 1974 const char *startSpecifier, 1975 unsigned specifierLen) { 1976 const analyze_printf::PrintfConversionSpecifier &CS = 1977 FS.getConversionSpecifier(); 1978 1979 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1980 getLocationOfByte(CS.getStart()), 1981 startSpecifier, specifierLen, 1982 CS.getStart(), CS.getLength()); 1983} 1984 1985bool CheckPrintfHandler::HandleAmount( 1986 const analyze_format_string::OptionalAmount &Amt, 1987 unsigned k, const char *startSpecifier, 1988 unsigned specifierLen) { 1989 1990 if (Amt.hasDataArgument()) { 1991 if (!HasVAListArg) { 1992 unsigned argIndex = Amt.getArgIndex(); 1993 if (argIndex >= NumDataArgs) { 1994 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 1995 << k, 1996 getLocationOfByte(Amt.getStart()), 1997 /*IsStringLocation*/true, 1998 getSpecifierRange(startSpecifier, specifierLen)); 1999 // Don't do any more checking. We will just emit 2000 // spurious errors. 2001 return false; 2002 } 2003 2004 // Type check the data argument. It should be an 'int'. 2005 // Although not in conformance with C99, we also allow the argument to be 2006 // an 'unsigned int' as that is a reasonably safe case. GCC also 2007 // doesn't emit a warning for that case. 2008 CoveredArgs.set(argIndex); 2009 const Expr *Arg = getDataArg(argIndex); 2010 QualType T = Arg->getType(); 2011 2012 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 2013 assert(ATR.isValid()); 2014 2015 if (!ATR.matchesType(S.Context, T)) { 2016 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2017 << k << ATR.getRepresentativeTypeName(S.Context) 2018 << T << Arg->getSourceRange(), 2019 getLocationOfByte(Amt.getStart()), 2020 /*IsStringLocation*/true, 2021 getSpecifierRange(startSpecifier, specifierLen)); 2022 // Don't do any more checking. We will just emit 2023 // spurious errors. 2024 return false; 2025 } 2026 } 2027 } 2028 return true; 2029} 2030 2031void CheckPrintfHandler::HandleInvalidAmount( 2032 const analyze_printf::PrintfSpecifier &FS, 2033 const analyze_printf::OptionalAmount &Amt, 2034 unsigned type, 2035 const char *startSpecifier, 2036 unsigned specifierLen) { 2037 const analyze_printf::PrintfConversionSpecifier &CS = 2038 FS.getConversionSpecifier(); 2039 2040 FixItHint fixit = 2041 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2042 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2043 Amt.getConstantLength())) 2044 : FixItHint(); 2045 2046 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2047 << type << CS.toString(), 2048 getLocationOfByte(Amt.getStart()), 2049 /*IsStringLocation*/true, 2050 getSpecifierRange(startSpecifier, specifierLen), 2051 fixit); 2052} 2053 2054void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2055 const analyze_printf::OptionalFlag &flag, 2056 const char *startSpecifier, 2057 unsigned specifierLen) { 2058 // Warn about pointless flag with a fixit removal. 2059 const analyze_printf::PrintfConversionSpecifier &CS = 2060 FS.getConversionSpecifier(); 2061 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2062 << flag.toString() << CS.toString(), 2063 getLocationOfByte(flag.getPosition()), 2064 /*IsStringLocation*/true, 2065 getSpecifierRange(startSpecifier, specifierLen), 2066 FixItHint::CreateRemoval( 2067 getSpecifierRange(flag.getPosition(), 1))); 2068} 2069 2070void CheckPrintfHandler::HandleIgnoredFlag( 2071 const analyze_printf::PrintfSpecifier &FS, 2072 const analyze_printf::OptionalFlag &ignoredFlag, 2073 const analyze_printf::OptionalFlag &flag, 2074 const char *startSpecifier, 2075 unsigned specifierLen) { 2076 // Warn about ignored flag with a fixit removal. 2077 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2078 << ignoredFlag.toString() << flag.toString(), 2079 getLocationOfByte(ignoredFlag.getPosition()), 2080 /*IsStringLocation*/true, 2081 getSpecifierRange(startSpecifier, specifierLen), 2082 FixItHint::CreateRemoval( 2083 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2084} 2085 2086bool 2087CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2088 &FS, 2089 const char *startSpecifier, 2090 unsigned specifierLen) { 2091 2092 using namespace analyze_format_string; 2093 using namespace analyze_printf; 2094 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2095 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 // First check if the field width, precision, and conversion specifier 2109 // have matching data arguments. 2110 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2111 startSpecifier, specifierLen)) { 2112 return false; 2113 } 2114 2115 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2116 startSpecifier, specifierLen)) { 2117 return false; 2118 } 2119 2120 if (!CS.consumesDataArgument()) { 2121 // FIXME: Technically specifying a precision or field width here 2122 // makes no sense. Worth issuing a warning at some point. 2123 return true; 2124 } 2125 2126 // Consume the argument. 2127 unsigned argIndex = FS.getArgIndex(); 2128 if (argIndex < NumDataArgs) { 2129 // The check to see if the argIndex is valid will come later. 2130 // We set the bit here because we may exit early from this 2131 // function if we encounter some other error. 2132 CoveredArgs.set(argIndex); 2133 } 2134 2135 // Check for using an Objective-C specific conversion specifier 2136 // in a non-ObjC literal. 2137 if (!IsObjCLiteral && CS.isObjCArg()) { 2138 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2139 specifierLen); 2140 } 2141 2142 // Check for invalid use of field width 2143 if (!FS.hasValidFieldWidth()) { 2144 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2145 startSpecifier, specifierLen); 2146 } 2147 2148 // Check for invalid use of precision 2149 if (!FS.hasValidPrecision()) { 2150 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2151 startSpecifier, specifierLen); 2152 } 2153 2154 // Check each flag does not conflict with any other component. 2155 if (!FS.hasValidThousandsGroupingPrefix()) 2156 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2157 if (!FS.hasValidLeadingZeros()) 2158 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2159 if (!FS.hasValidPlusPrefix()) 2160 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2161 if (!FS.hasValidSpacePrefix()) 2162 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2163 if (!FS.hasValidAlternativeForm()) 2164 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2165 if (!FS.hasValidLeftJustified()) 2166 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2167 2168 // Check that flags are not ignored by another flag 2169 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2170 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2171 startSpecifier, specifierLen); 2172 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2173 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2174 startSpecifier, specifierLen); 2175 2176 // Check the length modifier is valid with the given conversion specifier. 2177 const LengthModifier &LM = FS.getLengthModifier(); 2178 if (!FS.hasValidLengthModifier()) 2179 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2180 << LM.toString() << CS.toString(), 2181 getLocationOfByte(LM.getStart()), 2182 /*IsStringLocation*/true, 2183 getSpecifierRange(startSpecifier, specifierLen), 2184 FixItHint::CreateRemoval( 2185 getSpecifierRange(LM.getStart(), 2186 LM.getLength()))); 2187 2188 // Are we using '%n'? 2189 if (CS.getKind() == ConversionSpecifier::nArg) { 2190 // Issue a warning about this being a possible security issue. 2191 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2192 getLocationOfByte(CS.getStart()), 2193 /*IsStringLocation*/true, 2194 getSpecifierRange(startSpecifier, specifierLen)); 2195 // Continue checking the other format specifiers. 2196 return true; 2197 } 2198 2199 // The remaining checks depend on the data arguments. 2200 if (HasVAListArg) 2201 return true; 2202 2203 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2204 return false; 2205 2206 // Now type check the data expression that matches the 2207 // format specifier. 2208 const Expr *Ex = getDataArg(argIndex); 2209 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 2210 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2211 // Check if we didn't match because of an implicit cast from a 'char' 2212 // or 'short' to an 'int'. This is done because printf is a varargs 2213 // function. 2214 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2215 if (ICE->getType() == S.Context.IntTy) { 2216 // All further checking is done on the subexpression. 2217 Ex = ICE->getSubExpr(); 2218 if (ATR.matchesType(S.Context, Ex->getType())) 2219 return true; 2220 } 2221 2222 // We may be able to offer a FixItHint if it is a supported type. 2223 PrintfSpecifier fixedFS = FS; 2224 bool success = fixedFS.fixType(Ex->getType(), S.getLangOptions()); 2225 2226 if (success) { 2227 // Get the fix string from the fixed format specifier 2228 llvm::SmallString<128> buf; 2229 llvm::raw_svector_ostream os(buf); 2230 fixedFS.toString(os); 2231 2232 EmitFormatDiagnostic( 2233 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2234 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2235 << Ex->getSourceRange(), 2236 getLocationOfByte(CS.getStart()), 2237 /*IsStringLocation*/true, 2238 getSpecifierRange(startSpecifier, specifierLen), 2239 FixItHint::CreateReplacement( 2240 getSpecifierRange(startSpecifier, specifierLen), 2241 os.str())); 2242 } 2243 else { 2244 S.Diag(getLocationOfByte(CS.getStart()), 2245 diag::warn_printf_conversion_argument_type_mismatch) 2246 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2247 << getSpecifierRange(startSpecifier, specifierLen) 2248 << Ex->getSourceRange(); 2249 } 2250 } 2251 2252 return true; 2253} 2254 2255//===--- CHECK: Scanf format string checking ------------------------------===// 2256 2257namespace { 2258class CheckScanfHandler : public CheckFormatHandler { 2259public: 2260 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2261 const Expr *origFormatExpr, unsigned firstDataArg, 2262 unsigned numDataArgs, bool isObjCLiteral, 2263 const char *beg, bool hasVAListArg, 2264 const CallExpr *theCall, unsigned formatIdx, 2265 bool inFunctionCall) 2266 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2267 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2268 theCall, formatIdx, inFunctionCall) {} 2269 2270 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2271 const char *startSpecifier, 2272 unsigned specifierLen); 2273 2274 bool HandleInvalidScanfConversionSpecifier( 2275 const analyze_scanf::ScanfSpecifier &FS, 2276 const char *startSpecifier, 2277 unsigned specifierLen); 2278 2279 void HandleIncompleteScanList(const char *start, const char *end); 2280}; 2281} 2282 2283void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2284 const char *end) { 2285 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2286 getLocationOfByte(end), /*IsStringLocation*/true, 2287 getSpecifierRange(start, end - start)); 2288} 2289 2290bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2291 const analyze_scanf::ScanfSpecifier &FS, 2292 const char *startSpecifier, 2293 unsigned specifierLen) { 2294 2295 const analyze_scanf::ScanfConversionSpecifier &CS = 2296 FS.getConversionSpecifier(); 2297 2298 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2299 getLocationOfByte(CS.getStart()), 2300 startSpecifier, specifierLen, 2301 CS.getStart(), CS.getLength()); 2302} 2303 2304bool CheckScanfHandler::HandleScanfSpecifier( 2305 const analyze_scanf::ScanfSpecifier &FS, 2306 const char *startSpecifier, 2307 unsigned specifierLen) { 2308 2309 using namespace analyze_scanf; 2310 using namespace analyze_format_string; 2311 2312 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2313 2314 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2315 // be used to decide if we are using positional arguments consistently. 2316 if (FS.consumesDataArgument()) { 2317 if (atFirstArg) { 2318 atFirstArg = false; 2319 usesPositionalArgs = FS.usesPositionalArg(); 2320 } 2321 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2322 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2323 startSpecifier, specifierLen); 2324 return false; 2325 } 2326 } 2327 2328 // Check if the field with is non-zero. 2329 const OptionalAmount &Amt = FS.getFieldWidth(); 2330 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2331 if (Amt.getConstantAmount() == 0) { 2332 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2333 Amt.getConstantLength()); 2334 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2335 getLocationOfByte(Amt.getStart()), 2336 /*IsStringLocation*/true, R, 2337 FixItHint::CreateRemoval(R)); 2338 } 2339 } 2340 2341 if (!FS.consumesDataArgument()) { 2342 // FIXME: Technically specifying a precision or field width here 2343 // makes no sense. Worth issuing a warning at some point. 2344 return true; 2345 } 2346 2347 // Consume the argument. 2348 unsigned argIndex = FS.getArgIndex(); 2349 if (argIndex < NumDataArgs) { 2350 // The check to see if the argIndex is valid will come later. 2351 // We set the bit here because we may exit early from this 2352 // function if we encounter some other error. 2353 CoveredArgs.set(argIndex); 2354 } 2355 2356 // Check the length modifier is valid with the given conversion specifier. 2357 const LengthModifier &LM = FS.getLengthModifier(); 2358 if (!FS.hasValidLengthModifier()) { 2359 S.Diag(getLocationOfByte(LM.getStart()), 2360 diag::warn_format_nonsensical_length) 2361 << LM.toString() << CS.toString() 2362 << getSpecifierRange(startSpecifier, specifierLen) 2363 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 2364 LM.getLength())); 2365 } 2366 2367 // The remaining checks depend on the data arguments. 2368 if (HasVAListArg) 2369 return true; 2370 2371 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2372 return false; 2373 2374 // FIXME: Check that the argument type matches the format specifier. 2375 2376 return true; 2377} 2378 2379void Sema::CheckFormatString(const StringLiteral *FExpr, 2380 const Expr *OrigFormatExpr, 2381 const CallExpr *TheCall, bool HasVAListArg, 2382 unsigned format_idx, unsigned firstDataArg, 2383 bool isPrintf, bool inFunctionCall) { 2384 2385 // CHECK: is the format string a wide literal? 2386 if (!FExpr->isAscii()) { 2387 CheckFormatHandler::EmitFormatDiagnostic( 2388 *this, inFunctionCall, TheCall->getArg(format_idx), 2389 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2390 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2391 return; 2392 } 2393 2394 // Str - The format string. NOTE: this is NOT null-terminated! 2395 StringRef StrRef = FExpr->getString(); 2396 const char *Str = StrRef.data(); 2397 unsigned StrLen = StrRef.size(); 2398 const unsigned numDataArgs = TheCall->getNumArgs() - firstDataArg; 2399 2400 // CHECK: empty format string? 2401 if (StrLen == 0 && numDataArgs > 0) { 2402 CheckFormatHandler::EmitFormatDiagnostic( 2403 *this, inFunctionCall, TheCall->getArg(format_idx), 2404 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2405 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2406 return; 2407 } 2408 2409 if (isPrintf) { 2410 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2411 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2412 Str, HasVAListArg, TheCall, format_idx, 2413 inFunctionCall); 2414 2415 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 2416 H.DoneProcessing(); 2417 } 2418 else { 2419 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2420 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2421 Str, HasVAListArg, TheCall, format_idx, 2422 inFunctionCall); 2423 2424 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 2425 H.DoneProcessing(); 2426 } 2427} 2428 2429//===--- CHECK: Standard memory functions ---------------------------------===// 2430 2431/// \brief Determine whether the given type is a dynamic class type (e.g., 2432/// whether it has a vtable). 2433static bool isDynamicClassType(QualType T) { 2434 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2435 if (CXXRecordDecl *Definition = Record->getDefinition()) 2436 if (Definition->isDynamicClass()) 2437 return true; 2438 2439 return false; 2440} 2441 2442/// \brief If E is a sizeof expression, returns its argument expression, 2443/// otherwise returns NULL. 2444static const Expr *getSizeOfExprArg(const Expr* E) { 2445 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2446 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2447 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2448 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2449 2450 return 0; 2451} 2452 2453/// \brief If E is a sizeof expression, returns its argument type. 2454static QualType getSizeOfArgType(const Expr* E) { 2455 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2456 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2457 if (SizeOf->getKind() == clang::UETT_SizeOf) 2458 return SizeOf->getTypeOfArgument(); 2459 2460 return QualType(); 2461} 2462 2463/// \brief Check for dangerous or invalid arguments to memset(). 2464/// 2465/// This issues warnings on known problematic, dangerous or unspecified 2466/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2467/// function calls. 2468/// 2469/// \param Call The call expression to diagnose. 2470void Sema::CheckMemaccessArguments(const CallExpr *Call, 2471 CheckedMemoryFunction CMF, 2472 IdentifierInfo *FnName) { 2473 // It is possible to have a non-standard definition of memset. Validate 2474 // we have enough arguments, and if not, abort further checking. 2475 unsigned ExpectedNumArgs = (CMF == CMF_Strndup ? 2 : 3); 2476 if (Call->getNumArgs() < ExpectedNumArgs) 2477 return; 2478 2479 unsigned LastArg = (CMF == CMF_Memset || CMF == CMF_Strndup ? 1 : 2); 2480 unsigned LenArg = (CMF == CMF_Strndup ? 1 : 2); 2481 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2482 2483 // We have special checking when the length is a sizeof expression. 2484 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2485 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2486 llvm::FoldingSetNodeID SizeOfArgID; 2487 2488 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2489 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2490 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2491 2492 QualType DestTy = Dest->getType(); 2493 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2494 QualType PointeeTy = DestPtrTy->getPointeeType(); 2495 2496 // Never warn about void type pointers. This can be used to suppress 2497 // false positives. 2498 if (PointeeTy->isVoidType()) 2499 continue; 2500 2501 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2502 // actually comparing the expressions for equality. Because computing the 2503 // expression IDs can be expensive, we only do this if the diagnostic is 2504 // enabled. 2505 if (SizeOfArg && 2506 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2507 SizeOfArg->getExprLoc())) { 2508 // We only compute IDs for expressions if the warning is enabled, and 2509 // cache the sizeof arg's ID. 2510 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2511 SizeOfArg->Profile(SizeOfArgID, Context, true); 2512 llvm::FoldingSetNodeID DestID; 2513 Dest->Profile(DestID, Context, true); 2514 if (DestID == SizeOfArgID) { 2515 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2516 // over sizeof(src) as well. 2517 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2518 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2519 if (UnaryOp->getOpcode() == UO_AddrOf) 2520 ActionIdx = 1; // If its an address-of operator, just remove it. 2521 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2522 ActionIdx = 2; // If the pointee's size is sizeof(char), 2523 // suggest an explicit length. 2524 unsigned DestSrcSelect = (CMF == CMF_Strndup ? 1 : ArgIdx); 2525 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2526 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2527 << FnName << DestSrcSelect << ActionIdx 2528 << Dest->getSourceRange() 2529 << SizeOfArg->getSourceRange()); 2530 break; 2531 } 2532 } 2533 2534 // Also check for cases where the sizeof argument is the exact same 2535 // type as the memory argument, and where it points to a user-defined 2536 // record type. 2537 if (SizeOfArgTy != QualType()) { 2538 if (PointeeTy->isRecordType() && 2539 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2540 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2541 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2542 << FnName << SizeOfArgTy << ArgIdx 2543 << PointeeTy << Dest->getSourceRange() 2544 << LenExpr->getSourceRange()); 2545 break; 2546 } 2547 } 2548 2549 // Always complain about dynamic classes. 2550 if (isDynamicClassType(PointeeTy)) 2551 DiagRuntimeBehavior( 2552 Dest->getExprLoc(), Dest, 2553 PDiag(diag::warn_dyn_class_memaccess) 2554 << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy 2555 // "overwritten" if we're warning about the destination for any call 2556 // but memcmp; otherwise a verb appropriate to the call. 2557 << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF) 2558 << Call->getCallee()->getSourceRange()); 2559 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 2560 DiagRuntimeBehavior( 2561 Dest->getExprLoc(), Dest, 2562 PDiag(diag::warn_arc_object_memaccess) 2563 << ArgIdx << FnName << PointeeTy 2564 << Call->getCallee()->getSourceRange()); 2565 else 2566 continue; 2567 2568 DiagRuntimeBehavior( 2569 Dest->getExprLoc(), Dest, 2570 PDiag(diag::note_bad_memaccess_silence) 2571 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2572 break; 2573 } 2574 } 2575} 2576 2577// A little helper routine: ignore addition and subtraction of integer literals. 2578// This intentionally does not ignore all integer constant expressions because 2579// we don't want to remove sizeof(). 2580static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2581 Ex = Ex->IgnoreParenCasts(); 2582 2583 for (;;) { 2584 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2585 if (!BO || !BO->isAdditiveOp()) 2586 break; 2587 2588 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2589 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2590 2591 if (isa<IntegerLiteral>(RHS)) 2592 Ex = LHS; 2593 else if (isa<IntegerLiteral>(LHS)) 2594 Ex = RHS; 2595 else 2596 break; 2597 } 2598 2599 return Ex; 2600} 2601 2602// Warn if the user has made the 'size' argument to strlcpy or strlcat 2603// be the size of the source, instead of the destination. 2604void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2605 IdentifierInfo *FnName) { 2606 2607 // Don't crash if the user has the wrong number of arguments 2608 if (Call->getNumArgs() != 3) 2609 return; 2610 2611 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2612 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2613 const Expr *CompareWithSrc = NULL; 2614 2615 // Look for 'strlcpy(dst, x, sizeof(x))' 2616 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2617 CompareWithSrc = Ex; 2618 else { 2619 // Look for 'strlcpy(dst, x, strlen(x))' 2620 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2621 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2622 && SizeCall->getNumArgs() == 1) 2623 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2624 } 2625 } 2626 2627 if (!CompareWithSrc) 2628 return; 2629 2630 // Determine if the argument to sizeof/strlen is equal to the source 2631 // argument. In principle there's all kinds of things you could do 2632 // here, for instance creating an == expression and evaluating it with 2633 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2634 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2635 if (!SrcArgDRE) 2636 return; 2637 2638 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2639 if (!CompareWithSrcDRE || 2640 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2641 return; 2642 2643 const Expr *OriginalSizeArg = Call->getArg(2); 2644 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2645 << OriginalSizeArg->getSourceRange() << FnName; 2646 2647 // Output a FIXIT hint if the destination is an array (rather than a 2648 // pointer to an array). This could be enhanced to handle some 2649 // pointers if we know the actual size, like if DstArg is 'array+2' 2650 // we could say 'sizeof(array)-2'. 2651 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2652 QualType DstArgTy = DstArg->getType(); 2653 2654 // Only handle constant-sized or VLAs, but not flexible members. 2655 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2656 // Only issue the FIXIT for arrays of size > 1. 2657 if (CAT->getSize().getSExtValue() <= 1) 2658 return; 2659 } else if (!DstArgTy->isVariableArrayType()) { 2660 return; 2661 } 2662 2663 llvm::SmallString<128> sizeString; 2664 llvm::raw_svector_ostream OS(sizeString); 2665 OS << "sizeof("; 2666 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2667 OS << ")"; 2668 2669 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2670 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2671 OS.str()); 2672} 2673 2674//===--- CHECK: Return Address of Stack Variable --------------------------===// 2675 2676static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2677static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2678 2679/// CheckReturnStackAddr - Check if a return statement returns the address 2680/// of a stack variable. 2681void 2682Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2683 SourceLocation ReturnLoc) { 2684 2685 Expr *stackE = 0; 2686 SmallVector<DeclRefExpr *, 8> refVars; 2687 2688 // Perform checking for returned stack addresses, local blocks, 2689 // label addresses or references to temporaries. 2690 if (lhsType->isPointerType() || 2691 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2692 stackE = EvalAddr(RetValExp, refVars); 2693 } else if (lhsType->isReferenceType()) { 2694 stackE = EvalVal(RetValExp, refVars); 2695 } 2696 2697 if (stackE == 0) 2698 return; // Nothing suspicious was found. 2699 2700 SourceLocation diagLoc; 2701 SourceRange diagRange; 2702 if (refVars.empty()) { 2703 diagLoc = stackE->getLocStart(); 2704 diagRange = stackE->getSourceRange(); 2705 } else { 2706 // We followed through a reference variable. 'stackE' contains the 2707 // problematic expression but we will warn at the return statement pointing 2708 // at the reference variable. We will later display the "trail" of 2709 // reference variables using notes. 2710 diagLoc = refVars[0]->getLocStart(); 2711 diagRange = refVars[0]->getSourceRange(); 2712 } 2713 2714 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2715 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2716 : diag::warn_ret_stack_addr) 2717 << DR->getDecl()->getDeclName() << diagRange; 2718 } else if (isa<BlockExpr>(stackE)) { // local block. 2719 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2720 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2721 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2722 } else { // local temporary. 2723 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2724 : diag::warn_ret_local_temp_addr) 2725 << diagRange; 2726 } 2727 2728 // Display the "trail" of reference variables that we followed until we 2729 // found the problematic expression using notes. 2730 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2731 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2732 // If this var binds to another reference var, show the range of the next 2733 // var, otherwise the var binds to the problematic expression, in which case 2734 // show the range of the expression. 2735 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2736 : stackE->getSourceRange(); 2737 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2738 << VD->getDeclName() << range; 2739 } 2740} 2741 2742/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2743/// check if the expression in a return statement evaluates to an address 2744/// to a location on the stack, a local block, an address of a label, or a 2745/// reference to local temporary. The recursion is used to traverse the 2746/// AST of the return expression, with recursion backtracking when we 2747/// encounter a subexpression that (1) clearly does not lead to one of the 2748/// above problematic expressions (2) is something we cannot determine leads to 2749/// a problematic expression based on such local checking. 2750/// 2751/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2752/// the expression that they point to. Such variables are added to the 2753/// 'refVars' vector so that we know what the reference variable "trail" was. 2754/// 2755/// EvalAddr processes expressions that are pointers that are used as 2756/// references (and not L-values). EvalVal handles all other values. 2757/// At the base case of the recursion is a check for the above problematic 2758/// expressions. 2759/// 2760/// This implementation handles: 2761/// 2762/// * pointer-to-pointer casts 2763/// * implicit conversions from array references to pointers 2764/// * taking the address of fields 2765/// * arbitrary interplay between "&" and "*" operators 2766/// * pointer arithmetic from an address of a stack variable 2767/// * taking the address of an array element where the array is on the stack 2768static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2769 if (E->isTypeDependent()) 2770 return NULL; 2771 2772 // We should only be called for evaluating pointer expressions. 2773 assert((E->getType()->isAnyPointerType() || 2774 E->getType()->isBlockPointerType() || 2775 E->getType()->isObjCQualifiedIdType()) && 2776 "EvalAddr only works on pointers"); 2777 2778 E = E->IgnoreParens(); 2779 2780 // Our "symbolic interpreter" is just a dispatch off the currently 2781 // viewed AST node. We then recursively traverse the AST by calling 2782 // EvalAddr and EvalVal appropriately. 2783 switch (E->getStmtClass()) { 2784 case Stmt::DeclRefExprClass: { 2785 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2786 2787 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2788 // If this is a reference variable, follow through to the expression that 2789 // it points to. 2790 if (V->hasLocalStorage() && 2791 V->getType()->isReferenceType() && V->hasInit()) { 2792 // Add the reference variable to the "trail". 2793 refVars.push_back(DR); 2794 return EvalAddr(V->getInit(), refVars); 2795 } 2796 2797 return NULL; 2798 } 2799 2800 case Stmt::UnaryOperatorClass: { 2801 // The only unary operator that make sense to handle here 2802 // is AddrOf. All others don't make sense as pointers. 2803 UnaryOperator *U = cast<UnaryOperator>(E); 2804 2805 if (U->getOpcode() == UO_AddrOf) 2806 return EvalVal(U->getSubExpr(), refVars); 2807 else 2808 return NULL; 2809 } 2810 2811 case Stmt::BinaryOperatorClass: { 2812 // Handle pointer arithmetic. All other binary operators are not valid 2813 // in this context. 2814 BinaryOperator *B = cast<BinaryOperator>(E); 2815 BinaryOperatorKind op = B->getOpcode(); 2816 2817 if (op != BO_Add && op != BO_Sub) 2818 return NULL; 2819 2820 Expr *Base = B->getLHS(); 2821 2822 // Determine which argument is the real pointer base. It could be 2823 // the RHS argument instead of the LHS. 2824 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2825 2826 assert (Base->getType()->isPointerType()); 2827 return EvalAddr(Base, refVars); 2828 } 2829 2830 // For conditional operators we need to see if either the LHS or RHS are 2831 // valid DeclRefExpr*s. If one of them is valid, we return it. 2832 case Stmt::ConditionalOperatorClass: { 2833 ConditionalOperator *C = cast<ConditionalOperator>(E); 2834 2835 // Handle the GNU extension for missing LHS. 2836 if (Expr *lhsExpr = C->getLHS()) { 2837 // In C++, we can have a throw-expression, which has 'void' type. 2838 if (!lhsExpr->getType()->isVoidType()) 2839 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2840 return LHS; 2841 } 2842 2843 // In C++, we can have a throw-expression, which has 'void' type. 2844 if (C->getRHS()->getType()->isVoidType()) 2845 return NULL; 2846 2847 return EvalAddr(C->getRHS(), refVars); 2848 } 2849 2850 case Stmt::BlockExprClass: 2851 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2852 return E; // local block. 2853 return NULL; 2854 2855 case Stmt::AddrLabelExprClass: 2856 return E; // address of label. 2857 2858 case Stmt::ExprWithCleanupsClass: 2859 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2860 2861 // For casts, we need to handle conversions from arrays to 2862 // pointer values, and pointer-to-pointer conversions. 2863 case Stmt::ImplicitCastExprClass: 2864 case Stmt::CStyleCastExprClass: 2865 case Stmt::CXXFunctionalCastExprClass: 2866 case Stmt::ObjCBridgedCastExprClass: { 2867 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2868 QualType T = SubExpr->getType(); 2869 2870 if (SubExpr->getType()->isPointerType() || 2871 SubExpr->getType()->isBlockPointerType() || 2872 SubExpr->getType()->isObjCQualifiedIdType()) 2873 return EvalAddr(SubExpr, refVars); 2874 else if (T->isArrayType()) 2875 return EvalVal(SubExpr, refVars); 2876 else 2877 return 0; 2878 } 2879 2880 // C++ casts. For dynamic casts, static casts, and const casts, we 2881 // are always converting from a pointer-to-pointer, so we just blow 2882 // through the cast. In the case the dynamic cast doesn't fail (and 2883 // return NULL), we take the conservative route and report cases 2884 // where we return the address of a stack variable. For Reinterpre 2885 // FIXME: The comment about is wrong; we're not always converting 2886 // from pointer to pointer. I'm guessing that this code should also 2887 // handle references to objects. 2888 case Stmt::CXXStaticCastExprClass: 2889 case Stmt::CXXDynamicCastExprClass: 2890 case Stmt::CXXConstCastExprClass: 2891 case Stmt::CXXReinterpretCastExprClass: { 2892 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2893 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2894 return EvalAddr(S, refVars); 2895 else 2896 return NULL; 2897 } 2898 2899 case Stmt::MaterializeTemporaryExprClass: 2900 if (Expr *Result = EvalAddr( 2901 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2902 refVars)) 2903 return Result; 2904 2905 return E; 2906 2907 // Everything else: we simply don't reason about them. 2908 default: 2909 return NULL; 2910 } 2911} 2912 2913 2914/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2915/// See the comments for EvalAddr for more details. 2916static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2917do { 2918 // We should only be called for evaluating non-pointer expressions, or 2919 // expressions with a pointer type that are not used as references but instead 2920 // are l-values (e.g., DeclRefExpr with a pointer type). 2921 2922 // Our "symbolic interpreter" is just a dispatch off the currently 2923 // viewed AST node. We then recursively traverse the AST by calling 2924 // EvalAddr and EvalVal appropriately. 2925 2926 E = E->IgnoreParens(); 2927 switch (E->getStmtClass()) { 2928 case Stmt::ImplicitCastExprClass: { 2929 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2930 if (IE->getValueKind() == VK_LValue) { 2931 E = IE->getSubExpr(); 2932 continue; 2933 } 2934 return NULL; 2935 } 2936 2937 case Stmt::ExprWithCleanupsClass: 2938 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars); 2939 2940 case Stmt::DeclRefExprClass: { 2941 // When we hit a DeclRefExpr we are looking at code that refers to a 2942 // variable's name. If it's not a reference variable we check if it has 2943 // local storage within the function, and if so, return the expression. 2944 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2945 2946 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2947 if (V->hasLocalStorage()) { 2948 if (!V->getType()->isReferenceType()) 2949 return DR; 2950 2951 // Reference variable, follow through to the expression that 2952 // it points to. 2953 if (V->hasInit()) { 2954 // Add the reference variable to the "trail". 2955 refVars.push_back(DR); 2956 return EvalVal(V->getInit(), refVars); 2957 } 2958 } 2959 2960 return NULL; 2961 } 2962 2963 case Stmt::UnaryOperatorClass: { 2964 // The only unary operator that make sense to handle here 2965 // is Deref. All others don't resolve to a "name." This includes 2966 // handling all sorts of rvalues passed to a unary operator. 2967 UnaryOperator *U = cast<UnaryOperator>(E); 2968 2969 if (U->getOpcode() == UO_Deref) 2970 return EvalAddr(U->getSubExpr(), refVars); 2971 2972 return NULL; 2973 } 2974 2975 case Stmt::ArraySubscriptExprClass: { 2976 // Array subscripts are potential references to data on the stack. We 2977 // retrieve the DeclRefExpr* for the array variable if it indeed 2978 // has local storage. 2979 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2980 } 2981 2982 case Stmt::ConditionalOperatorClass: { 2983 // For conditional operators we need to see if either the LHS or RHS are 2984 // non-NULL Expr's. If one is non-NULL, we return it. 2985 ConditionalOperator *C = cast<ConditionalOperator>(E); 2986 2987 // Handle the GNU extension for missing LHS. 2988 if (Expr *lhsExpr = C->getLHS()) 2989 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2990 return LHS; 2991 2992 return EvalVal(C->getRHS(), refVars); 2993 } 2994 2995 // Accesses to members are potential references to data on the stack. 2996 case Stmt::MemberExprClass: { 2997 MemberExpr *M = cast<MemberExpr>(E); 2998 2999 // Check for indirect access. We only want direct field accesses. 3000 if (M->isArrow()) 3001 return NULL; 3002 3003 // Check whether the member type is itself a reference, in which case 3004 // we're not going to refer to the member, but to what the member refers to. 3005 if (M->getMemberDecl()->getType()->isReferenceType()) 3006 return NULL; 3007 3008 return EvalVal(M->getBase(), refVars); 3009 } 3010 3011 case Stmt::MaterializeTemporaryExprClass: 3012 if (Expr *Result = EvalVal( 3013 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3014 refVars)) 3015 return Result; 3016 3017 return E; 3018 3019 default: 3020 // Check that we don't return or take the address of a reference to a 3021 // temporary. This is only useful in C++. 3022 if (!E->isTypeDependent() && E->isRValue()) 3023 return E; 3024 3025 // Everything else: we simply don't reason about them. 3026 return NULL; 3027 } 3028} while (true); 3029} 3030 3031//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3032 3033/// Check for comparisons of floating point operands using != and ==. 3034/// Issue a warning if these are no self-comparisons, as they are not likely 3035/// to do what the programmer intended. 3036void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3037 bool EmitWarning = true; 3038 3039 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3040 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3041 3042 // Special case: check for x == x (which is OK). 3043 // Do not emit warnings for such cases. 3044 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3045 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3046 if (DRL->getDecl() == DRR->getDecl()) 3047 EmitWarning = false; 3048 3049 3050 // Special case: check for comparisons against literals that can be exactly 3051 // represented by APFloat. In such cases, do not emit a warning. This 3052 // is a heuristic: often comparison against such literals are used to 3053 // detect if a value in a variable has not changed. This clearly can 3054 // lead to false negatives. 3055 if (EmitWarning) { 3056 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3057 if (FLL->isExact()) 3058 EmitWarning = false; 3059 } else 3060 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3061 if (FLR->isExact()) 3062 EmitWarning = false; 3063 } 3064 } 3065 3066 // Check for comparisons with builtin types. 3067 if (EmitWarning) 3068 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3069 if (CL->isBuiltinCall()) 3070 EmitWarning = false; 3071 3072 if (EmitWarning) 3073 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3074 if (CR->isBuiltinCall()) 3075 EmitWarning = false; 3076 3077 // Emit the diagnostic. 3078 if (EmitWarning) 3079 Diag(Loc, diag::warn_floatingpoint_eq) 3080 << LHS->getSourceRange() << RHS->getSourceRange(); 3081} 3082 3083//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3084//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3085 3086namespace { 3087 3088/// Structure recording the 'active' range of an integer-valued 3089/// expression. 3090struct IntRange { 3091 /// The number of bits active in the int. 3092 unsigned Width; 3093 3094 /// True if the int is known not to have negative values. 3095 bool NonNegative; 3096 3097 IntRange(unsigned Width, bool NonNegative) 3098 : Width(Width), NonNegative(NonNegative) 3099 {} 3100 3101 /// Returns the range of the bool type. 3102 static IntRange forBoolType() { 3103 return IntRange(1, true); 3104 } 3105 3106 /// Returns the range of an opaque value of the given integral type. 3107 static IntRange forValueOfType(ASTContext &C, QualType T) { 3108 return forValueOfCanonicalType(C, 3109 T->getCanonicalTypeInternal().getTypePtr()); 3110 } 3111 3112 /// Returns the range of an opaque value of a canonical integral type. 3113 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3114 assert(T->isCanonicalUnqualified()); 3115 3116 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3117 T = VT->getElementType().getTypePtr(); 3118 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3119 T = CT->getElementType().getTypePtr(); 3120 3121 // For enum types, use the known bit width of the enumerators. 3122 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3123 EnumDecl *Enum = ET->getDecl(); 3124 if (!Enum->isCompleteDefinition()) 3125 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3126 3127 unsigned NumPositive = Enum->getNumPositiveBits(); 3128 unsigned NumNegative = Enum->getNumNegativeBits(); 3129 3130 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3131 } 3132 3133 const BuiltinType *BT = cast<BuiltinType>(T); 3134 assert(BT->isInteger()); 3135 3136 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3137 } 3138 3139 /// Returns the "target" range of a canonical integral type, i.e. 3140 /// the range of values expressible in the type. 3141 /// 3142 /// This matches forValueOfCanonicalType except that enums have the 3143 /// full range of their type, not the range of their enumerators. 3144 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3145 assert(T->isCanonicalUnqualified()); 3146 3147 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3148 T = VT->getElementType().getTypePtr(); 3149 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3150 T = CT->getElementType().getTypePtr(); 3151 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3152 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3153 3154 const BuiltinType *BT = cast<BuiltinType>(T); 3155 assert(BT->isInteger()); 3156 3157 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3158 } 3159 3160 /// Returns the supremum of two ranges: i.e. their conservative merge. 3161 static IntRange join(IntRange L, IntRange R) { 3162 return IntRange(std::max(L.Width, R.Width), 3163 L.NonNegative && R.NonNegative); 3164 } 3165 3166 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3167 static IntRange meet(IntRange L, IntRange R) { 3168 return IntRange(std::min(L.Width, R.Width), 3169 L.NonNegative || R.NonNegative); 3170 } 3171}; 3172 3173IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 3174 if (value.isSigned() && value.isNegative()) 3175 return IntRange(value.getMinSignedBits(), false); 3176 3177 if (value.getBitWidth() > MaxWidth) 3178 value = value.trunc(MaxWidth); 3179 3180 // isNonNegative() just checks the sign bit without considering 3181 // signedness. 3182 return IntRange(value.getActiveBits(), true); 3183} 3184 3185IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3186 unsigned MaxWidth) { 3187 if (result.isInt()) 3188 return GetValueRange(C, result.getInt(), MaxWidth); 3189 3190 if (result.isVector()) { 3191 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3192 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3193 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3194 R = IntRange::join(R, El); 3195 } 3196 return R; 3197 } 3198 3199 if (result.isComplexInt()) { 3200 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3201 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3202 return IntRange::join(R, I); 3203 } 3204 3205 // This can happen with lossless casts to intptr_t of "based" lvalues. 3206 // Assume it might use arbitrary bits. 3207 // FIXME: The only reason we need to pass the type in here is to get 3208 // the sign right on this one case. It would be nice if APValue 3209 // preserved this. 3210 assert(result.isLValue()); 3211 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3212} 3213 3214/// Pseudo-evaluate the given integer expression, estimating the 3215/// range of values it might take. 3216/// 3217/// \param MaxWidth - the width to which the value will be truncated 3218IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3219 E = E->IgnoreParens(); 3220 3221 // Try a full evaluation first. 3222 Expr::EvalResult result; 3223 if (E->EvaluateAsRValue(result, C)) 3224 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3225 3226 // I think we only want to look through implicit casts here; if the 3227 // user has an explicit widening cast, we should treat the value as 3228 // being of the new, wider type. 3229 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3230 if (CE->getCastKind() == CK_NoOp) 3231 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3232 3233 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3234 3235 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3236 3237 // Assume that non-integer casts can span the full range of the type. 3238 if (!isIntegerCast) 3239 return OutputTypeRange; 3240 3241 IntRange SubRange 3242 = GetExprRange(C, CE->getSubExpr(), 3243 std::min(MaxWidth, OutputTypeRange.Width)); 3244 3245 // Bail out if the subexpr's range is as wide as the cast type. 3246 if (SubRange.Width >= OutputTypeRange.Width) 3247 return OutputTypeRange; 3248 3249 // Otherwise, we take the smaller width, and we're non-negative if 3250 // either the output type or the subexpr is. 3251 return IntRange(SubRange.Width, 3252 SubRange.NonNegative || OutputTypeRange.NonNegative); 3253 } 3254 3255 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3256 // If we can fold the condition, just take that operand. 3257 bool CondResult; 3258 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3259 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3260 : CO->getFalseExpr(), 3261 MaxWidth); 3262 3263 // Otherwise, conservatively merge. 3264 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3265 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3266 return IntRange::join(L, R); 3267 } 3268 3269 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3270 switch (BO->getOpcode()) { 3271 3272 // Boolean-valued operations are single-bit and positive. 3273 case BO_LAnd: 3274 case BO_LOr: 3275 case BO_LT: 3276 case BO_GT: 3277 case BO_LE: 3278 case BO_GE: 3279 case BO_EQ: 3280 case BO_NE: 3281 return IntRange::forBoolType(); 3282 3283 // The type of the assignments is the type of the LHS, so the RHS 3284 // is not necessarily the same type. 3285 case BO_MulAssign: 3286 case BO_DivAssign: 3287 case BO_RemAssign: 3288 case BO_AddAssign: 3289 case BO_SubAssign: 3290 case BO_XorAssign: 3291 case BO_OrAssign: 3292 // TODO: bitfields? 3293 return IntRange::forValueOfType(C, E->getType()); 3294 3295 // Simple assignments just pass through the RHS, which will have 3296 // been coerced to the LHS type. 3297 case BO_Assign: 3298 // TODO: bitfields? 3299 return GetExprRange(C, BO->getRHS(), MaxWidth); 3300 3301 // Operations with opaque sources are black-listed. 3302 case BO_PtrMemD: 3303 case BO_PtrMemI: 3304 return IntRange::forValueOfType(C, E->getType()); 3305 3306 // Bitwise-and uses the *infinum* of the two source ranges. 3307 case BO_And: 3308 case BO_AndAssign: 3309 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3310 GetExprRange(C, BO->getRHS(), MaxWidth)); 3311 3312 // Left shift gets black-listed based on a judgement call. 3313 case BO_Shl: 3314 // ...except that we want to treat '1 << (blah)' as logically 3315 // positive. It's an important idiom. 3316 if (IntegerLiteral *I 3317 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3318 if (I->getValue() == 1) { 3319 IntRange R = IntRange::forValueOfType(C, E->getType()); 3320 return IntRange(R.Width, /*NonNegative*/ true); 3321 } 3322 } 3323 // fallthrough 3324 3325 case BO_ShlAssign: 3326 return IntRange::forValueOfType(C, E->getType()); 3327 3328 // Right shift by a constant can narrow its left argument. 3329 case BO_Shr: 3330 case BO_ShrAssign: { 3331 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3332 3333 // If the shift amount is a positive constant, drop the width by 3334 // that much. 3335 llvm::APSInt shift; 3336 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3337 shift.isNonNegative()) { 3338 unsigned zext = shift.getZExtValue(); 3339 if (zext >= L.Width) 3340 L.Width = (L.NonNegative ? 0 : 1); 3341 else 3342 L.Width -= zext; 3343 } 3344 3345 return L; 3346 } 3347 3348 // Comma acts as its right operand. 3349 case BO_Comma: 3350 return GetExprRange(C, BO->getRHS(), MaxWidth); 3351 3352 // Black-list pointer subtractions. 3353 case BO_Sub: 3354 if (BO->getLHS()->getType()->isPointerType()) 3355 return IntRange::forValueOfType(C, E->getType()); 3356 break; 3357 3358 // The width of a division result is mostly determined by the size 3359 // of the LHS. 3360 case BO_Div: { 3361 // Don't 'pre-truncate' the operands. 3362 unsigned opWidth = C.getIntWidth(E->getType()); 3363 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3364 3365 // If the divisor is constant, use that. 3366 llvm::APSInt divisor; 3367 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3368 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3369 if (log2 >= L.Width) 3370 L.Width = (L.NonNegative ? 0 : 1); 3371 else 3372 L.Width = std::min(L.Width - log2, MaxWidth); 3373 return L; 3374 } 3375 3376 // Otherwise, just use the LHS's width. 3377 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3378 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3379 } 3380 3381 // The result of a remainder can't be larger than the result of 3382 // either side. 3383 case BO_Rem: { 3384 // Don't 'pre-truncate' the operands. 3385 unsigned opWidth = C.getIntWidth(E->getType()); 3386 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3387 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3388 3389 IntRange meet = IntRange::meet(L, R); 3390 meet.Width = std::min(meet.Width, MaxWidth); 3391 return meet; 3392 } 3393 3394 // The default behavior is okay for these. 3395 case BO_Mul: 3396 case BO_Add: 3397 case BO_Xor: 3398 case BO_Or: 3399 break; 3400 } 3401 3402 // The default case is to treat the operation as if it were closed 3403 // on the narrowest type that encompasses both operands. 3404 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3405 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3406 return IntRange::join(L, R); 3407 } 3408 3409 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3410 switch (UO->getOpcode()) { 3411 // Boolean-valued operations are white-listed. 3412 case UO_LNot: 3413 return IntRange::forBoolType(); 3414 3415 // Operations with opaque sources are black-listed. 3416 case UO_Deref: 3417 case UO_AddrOf: // should be impossible 3418 return IntRange::forValueOfType(C, E->getType()); 3419 3420 default: 3421 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3422 } 3423 } 3424 3425 if (dyn_cast<OffsetOfExpr>(E)) { 3426 IntRange::forValueOfType(C, E->getType()); 3427 } 3428 3429 if (FieldDecl *BitField = E->getBitField()) 3430 return IntRange(BitField->getBitWidthValue(C), 3431 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3432 3433 return IntRange::forValueOfType(C, E->getType()); 3434} 3435 3436IntRange GetExprRange(ASTContext &C, Expr *E) { 3437 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3438} 3439 3440/// Checks whether the given value, which currently has the given 3441/// source semantics, has the same value when coerced through the 3442/// target semantics. 3443bool IsSameFloatAfterCast(const llvm::APFloat &value, 3444 const llvm::fltSemantics &Src, 3445 const llvm::fltSemantics &Tgt) { 3446 llvm::APFloat truncated = value; 3447 3448 bool ignored; 3449 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3450 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3451 3452 return truncated.bitwiseIsEqual(value); 3453} 3454 3455/// Checks whether the given value, which currently has the given 3456/// source semantics, has the same value when coerced through the 3457/// target semantics. 3458/// 3459/// The value might be a vector of floats (or a complex number). 3460bool IsSameFloatAfterCast(const APValue &value, 3461 const llvm::fltSemantics &Src, 3462 const llvm::fltSemantics &Tgt) { 3463 if (value.isFloat()) 3464 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3465 3466 if (value.isVector()) { 3467 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3468 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3469 return false; 3470 return true; 3471 } 3472 3473 assert(value.isComplexFloat()); 3474 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3475 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3476} 3477 3478void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3479 3480static bool IsZero(Sema &S, Expr *E) { 3481 // Suppress cases where we are comparing against an enum constant. 3482 if (const DeclRefExpr *DR = 3483 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3484 if (isa<EnumConstantDecl>(DR->getDecl())) 3485 return false; 3486 3487 // Suppress cases where the '0' value is expanded from a macro. 3488 if (E->getLocStart().isMacroID()) 3489 return false; 3490 3491 llvm::APSInt Value; 3492 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3493} 3494 3495static bool HasEnumType(Expr *E) { 3496 // Strip off implicit integral promotions. 3497 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3498 if (ICE->getCastKind() != CK_IntegralCast && 3499 ICE->getCastKind() != CK_NoOp) 3500 break; 3501 E = ICE->getSubExpr(); 3502 } 3503 3504 return E->getType()->isEnumeralType(); 3505} 3506 3507void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3508 BinaryOperatorKind op = E->getOpcode(); 3509 if (E->isValueDependent()) 3510 return; 3511 3512 if (op == BO_LT && IsZero(S, E->getRHS())) { 3513 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3514 << "< 0" << "false" << HasEnumType(E->getLHS()) 3515 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3516 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3517 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3518 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3519 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3520 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3521 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3522 << "0 >" << "false" << HasEnumType(E->getRHS()) 3523 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3524 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3525 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3526 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3527 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3528 } 3529} 3530 3531/// Analyze the operands of the given comparison. Implements the 3532/// fallback case from AnalyzeComparison. 3533void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3534 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3535 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3536} 3537 3538/// \brief Implements -Wsign-compare. 3539/// 3540/// \param E the binary operator to check for warnings 3541void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3542 // The type the comparison is being performed in. 3543 QualType T = E->getLHS()->getType(); 3544 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3545 && "comparison with mismatched types"); 3546 3547 // We don't do anything special if this isn't an unsigned integral 3548 // comparison: we're only interested in integral comparisons, and 3549 // signed comparisons only happen in cases we don't care to warn about. 3550 // 3551 // We also don't care about value-dependent expressions or expressions 3552 // whose result is a constant. 3553 if (!T->hasUnsignedIntegerRepresentation() 3554 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3555 return AnalyzeImpConvsInComparison(S, E); 3556 3557 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3558 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3559 3560 // Check to see if one of the (unmodified) operands is of different 3561 // signedness. 3562 Expr *signedOperand, *unsignedOperand; 3563 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3564 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3565 "unsigned comparison between two signed integer expressions?"); 3566 signedOperand = LHS; 3567 unsignedOperand = RHS; 3568 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3569 signedOperand = RHS; 3570 unsignedOperand = LHS; 3571 } else { 3572 CheckTrivialUnsignedComparison(S, E); 3573 return AnalyzeImpConvsInComparison(S, E); 3574 } 3575 3576 // Otherwise, calculate the effective range of the signed operand. 3577 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3578 3579 // Go ahead and analyze implicit conversions in the operands. Note 3580 // that we skip the implicit conversions on both sides. 3581 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3582 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3583 3584 // If the signed range is non-negative, -Wsign-compare won't fire, 3585 // but we should still check for comparisons which are always true 3586 // or false. 3587 if (signedRange.NonNegative) 3588 return CheckTrivialUnsignedComparison(S, E); 3589 3590 // For (in)equality comparisons, if the unsigned operand is a 3591 // constant which cannot collide with a overflowed signed operand, 3592 // then reinterpreting the signed operand as unsigned will not 3593 // change the result of the comparison. 3594 if (E->isEqualityOp()) { 3595 unsigned comparisonWidth = S.Context.getIntWidth(T); 3596 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3597 3598 // We should never be unable to prove that the unsigned operand is 3599 // non-negative. 3600 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3601 3602 if (unsignedRange.Width < comparisonWidth) 3603 return; 3604 } 3605 3606 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3607 << LHS->getType() << RHS->getType() 3608 << LHS->getSourceRange() << RHS->getSourceRange(); 3609} 3610 3611/// Analyzes an attempt to assign the given value to a bitfield. 3612/// 3613/// Returns true if there was something fishy about the attempt. 3614bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3615 SourceLocation InitLoc) { 3616 assert(Bitfield->isBitField()); 3617 if (Bitfield->isInvalidDecl()) 3618 return false; 3619 3620 // White-list bool bitfields. 3621 if (Bitfield->getType()->isBooleanType()) 3622 return false; 3623 3624 // Ignore value- or type-dependent expressions. 3625 if (Bitfield->getBitWidth()->isValueDependent() || 3626 Bitfield->getBitWidth()->isTypeDependent() || 3627 Init->isValueDependent() || 3628 Init->isTypeDependent()) 3629 return false; 3630 3631 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3632 3633 Expr::EvalResult InitValue; 3634 if (!OriginalInit->EvaluateAsRValue(InitValue, S.Context) || 3635 !InitValue.Val.isInt()) 3636 return false; 3637 3638 const llvm::APSInt &Value = InitValue.Val.getInt(); 3639 unsigned OriginalWidth = Value.getBitWidth(); 3640 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3641 3642 if (OriginalWidth <= FieldWidth) 3643 return false; 3644 3645 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3646 3647 // It's fairly common to write values into signed bitfields 3648 // that, if sign-extended, would end up becoming a different 3649 // value. We don't want to warn about that. 3650 if (Value.isSigned() && Value.isNegative()) 3651 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3652 else 3653 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3654 3655 if (Value == TruncatedValue) 3656 return false; 3657 3658 std::string PrettyValue = Value.toString(10); 3659 std::string PrettyTrunc = TruncatedValue.toString(10); 3660 3661 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3662 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3663 << Init->getSourceRange(); 3664 3665 return true; 3666} 3667 3668/// Analyze the given simple or compound assignment for warning-worthy 3669/// operations. 3670void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3671 // Just recurse on the LHS. 3672 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3673 3674 // We want to recurse on the RHS as normal unless we're assigning to 3675 // a bitfield. 3676 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3677 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3678 E->getOperatorLoc())) { 3679 // Recurse, ignoring any implicit conversions on the RHS. 3680 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3681 E->getOperatorLoc()); 3682 } 3683 } 3684 3685 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3686} 3687 3688/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3689void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3690 SourceLocation CContext, unsigned diag) { 3691 S.Diag(E->getExprLoc(), diag) 3692 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3693} 3694 3695/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3696void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3697 unsigned diag) { 3698 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3699} 3700 3701/// Diagnose an implicit cast from a literal expression. Does not warn when the 3702/// cast wouldn't lose information. 3703void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3704 SourceLocation CContext) { 3705 // Try to convert the literal exactly to an integer. If we can, don't warn. 3706 bool isExact = false; 3707 const llvm::APFloat &Value = FL->getValue(); 3708 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3709 T->hasUnsignedIntegerRepresentation()); 3710 if (Value.convertToInteger(IntegerValue, 3711 llvm::APFloat::rmTowardZero, &isExact) 3712 == llvm::APFloat::opOK && isExact) 3713 return; 3714 3715 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3716 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3717} 3718 3719std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3720 if (!Range.Width) return "0"; 3721 3722 llvm::APSInt ValueInRange = Value; 3723 ValueInRange.setIsSigned(!Range.NonNegative); 3724 ValueInRange = ValueInRange.trunc(Range.Width); 3725 return ValueInRange.toString(10); 3726} 3727 3728static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3729 SourceManager &smgr = S.Context.getSourceManager(); 3730 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3731} 3732 3733void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3734 SourceLocation CC, bool *ICContext = 0) { 3735 if (E->isTypeDependent() || E->isValueDependent()) return; 3736 3737 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3738 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3739 if (Source == Target) return; 3740 if (Target->isDependentType()) return; 3741 3742 // If the conversion context location is invalid don't complain. We also 3743 // don't want to emit a warning if the issue occurs from the expansion of 3744 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3745 // delay this check as long as possible. Once we detect we are in that 3746 // scenario, we just return. 3747 if (CC.isInvalid()) 3748 return; 3749 3750 // Diagnose implicit casts to bool. 3751 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 3752 if (isa<StringLiteral>(E)) 3753 // Warn on string literal to bool. Checks for string literals in logical 3754 // expressions, for instances, assert(0 && "error here"), is prevented 3755 // by a check in AnalyzeImplicitConversions(). 3756 return DiagnoseImpCast(S, E, T, CC, 3757 diag::warn_impcast_string_literal_to_bool); 3758 if (Source->isFunctionType()) { 3759 // Warn on function to bool. Checks free functions and static member 3760 // functions. Weakly imported functions are excluded from the check, 3761 // since it's common to test their value to check whether the linker 3762 // found a definition for them. 3763 ValueDecl *D = 0; 3764 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 3765 D = R->getDecl(); 3766 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 3767 D = M->getMemberDecl(); 3768 } 3769 3770 if (D && !D->isWeak()) { 3771 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 3772 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 3773 << F << E->getSourceRange() << SourceRange(CC); 3774 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 3775 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 3776 QualType ReturnType; 3777 UnresolvedSet<4> NonTemplateOverloads; 3778 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 3779 if (!ReturnType.isNull() 3780 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 3781 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 3782 << FixItHint::CreateInsertion( 3783 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 3784 return; 3785 } 3786 } 3787 } 3788 return; // Other casts to bool are not checked. 3789 } 3790 3791 // Strip vector types. 3792 if (isa<VectorType>(Source)) { 3793 if (!isa<VectorType>(Target)) { 3794 if (isFromSystemMacro(S, CC)) 3795 return; 3796 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3797 } 3798 3799 // If the vector cast is cast between two vectors of the same size, it is 3800 // a bitcast, not a conversion. 3801 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3802 return; 3803 3804 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3805 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3806 } 3807 3808 // Strip complex types. 3809 if (isa<ComplexType>(Source)) { 3810 if (!isa<ComplexType>(Target)) { 3811 if (isFromSystemMacro(S, CC)) 3812 return; 3813 3814 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3815 } 3816 3817 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3818 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3819 } 3820 3821 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3822 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3823 3824 // If the source is floating point... 3825 if (SourceBT && SourceBT->isFloatingPoint()) { 3826 // ...and the target is floating point... 3827 if (TargetBT && TargetBT->isFloatingPoint()) { 3828 // ...then warn if we're dropping FP rank. 3829 3830 // Builtin FP kinds are ordered by increasing FP rank. 3831 if (SourceBT->getKind() > TargetBT->getKind()) { 3832 // Don't warn about float constants that are precisely 3833 // representable in the target type. 3834 Expr::EvalResult result; 3835 if (E->EvaluateAsRValue(result, S.Context)) { 3836 // Value might be a float, a float vector, or a float complex. 3837 if (IsSameFloatAfterCast(result.Val, 3838 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3839 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3840 return; 3841 } 3842 3843 if (isFromSystemMacro(S, CC)) 3844 return; 3845 3846 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3847 } 3848 return; 3849 } 3850 3851 // If the target is integral, always warn. 3852 if ((TargetBT && TargetBT->isInteger())) { 3853 if (isFromSystemMacro(S, CC)) 3854 return; 3855 3856 Expr *InnerE = E->IgnoreParenImpCasts(); 3857 // We also want to warn on, e.g., "int i = -1.234" 3858 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3859 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3860 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3861 3862 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3863 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3864 } else { 3865 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3866 } 3867 } 3868 3869 return; 3870 } 3871 3872 if (!Source->isIntegerType() || !Target->isIntegerType()) 3873 return; 3874 3875 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3876 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3877 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3878 << E->getSourceRange() << clang::SourceRange(CC); 3879 return; 3880 } 3881 3882 IntRange SourceRange = GetExprRange(S.Context, E); 3883 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3884 3885 if (SourceRange.Width > TargetRange.Width) { 3886 // If the source is a constant, use a default-on diagnostic. 3887 // TODO: this should happen for bitfield stores, too. 3888 llvm::APSInt Value(32); 3889 if (E->isIntegerConstantExpr(Value, S.Context)) { 3890 if (isFromSystemMacro(S, CC)) 3891 return; 3892 3893 std::string PrettySourceValue = Value.toString(10); 3894 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3895 3896 S.DiagRuntimeBehavior(E->getExprLoc(), E, 3897 S.PDiag(diag::warn_impcast_integer_precision_constant) 3898 << PrettySourceValue << PrettyTargetValue 3899 << E->getType() << T << E->getSourceRange() 3900 << clang::SourceRange(CC)); 3901 return; 3902 } 3903 3904 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3905 if (isFromSystemMacro(S, CC)) 3906 return; 3907 3908 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3909 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3910 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3911 } 3912 3913 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3914 (!TargetRange.NonNegative && SourceRange.NonNegative && 3915 SourceRange.Width == TargetRange.Width)) { 3916 3917 if (isFromSystemMacro(S, CC)) 3918 return; 3919 3920 unsigned DiagID = diag::warn_impcast_integer_sign; 3921 3922 // Traditionally, gcc has warned about this under -Wsign-compare. 3923 // We also want to warn about it in -Wconversion. 3924 // So if -Wconversion is off, use a completely identical diagnostic 3925 // in the sign-compare group. 3926 // The conditional-checking code will 3927 if (ICContext) { 3928 DiagID = diag::warn_impcast_integer_sign_conditional; 3929 *ICContext = true; 3930 } 3931 3932 return DiagnoseImpCast(S, E, T, CC, DiagID); 3933 } 3934 3935 // Diagnose conversions between different enumeration types. 3936 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3937 // type, to give us better diagnostics. 3938 QualType SourceType = E->getType(); 3939 if (!S.getLangOptions().CPlusPlus) { 3940 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3941 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3942 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3943 SourceType = S.Context.getTypeDeclType(Enum); 3944 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3945 } 3946 } 3947 3948 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3949 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3950 if ((SourceEnum->getDecl()->getIdentifier() || 3951 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3952 (TargetEnum->getDecl()->getIdentifier() || 3953 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3954 SourceEnum != TargetEnum) { 3955 if (isFromSystemMacro(S, CC)) 3956 return; 3957 3958 return DiagnoseImpCast(S, E, SourceType, T, CC, 3959 diag::warn_impcast_different_enum_types); 3960 } 3961 3962 return; 3963} 3964 3965void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3966 3967void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3968 SourceLocation CC, bool &ICContext) { 3969 E = E->IgnoreParenImpCasts(); 3970 3971 if (isa<ConditionalOperator>(E)) 3972 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3973 3974 AnalyzeImplicitConversions(S, E, CC); 3975 if (E->getType() != T) 3976 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3977 return; 3978} 3979 3980void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3981 SourceLocation CC = E->getQuestionLoc(); 3982 3983 AnalyzeImplicitConversions(S, E->getCond(), CC); 3984 3985 bool Suspicious = false; 3986 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3987 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3988 3989 // If -Wconversion would have warned about either of the candidates 3990 // for a signedness conversion to the context type... 3991 if (!Suspicious) return; 3992 3993 // ...but it's currently ignored... 3994 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3995 CC)) 3996 return; 3997 3998 // ...then check whether it would have warned about either of the 3999 // candidates for a signedness conversion to the condition type. 4000 if (E->getType() == T) return; 4001 4002 Suspicious = false; 4003 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4004 E->getType(), CC, &Suspicious); 4005 if (!Suspicious) 4006 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4007 E->getType(), CC, &Suspicious); 4008} 4009 4010/// AnalyzeImplicitConversions - Find and report any interesting 4011/// implicit conversions in the given expression. There are a couple 4012/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4013void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4014 QualType T = OrigE->getType(); 4015 Expr *E = OrigE->IgnoreParenImpCasts(); 4016 4017 if (E->isTypeDependent() || E->isValueDependent()) 4018 return; 4019 4020 // For conditional operators, we analyze the arguments as if they 4021 // were being fed directly into the output. 4022 if (isa<ConditionalOperator>(E)) { 4023 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4024 CheckConditionalOperator(S, CO, T); 4025 return; 4026 } 4027 4028 // Go ahead and check any implicit conversions we might have skipped. 4029 // The non-canonical typecheck is just an optimization; 4030 // CheckImplicitConversion will filter out dead implicit conversions. 4031 if (E->getType() != T) 4032 CheckImplicitConversion(S, E, T, CC); 4033 4034 // Now continue drilling into this expression. 4035 4036 // Skip past explicit casts. 4037 if (isa<ExplicitCastExpr>(E)) { 4038 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4039 return AnalyzeImplicitConversions(S, E, CC); 4040 } 4041 4042 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4043 // Do a somewhat different check with comparison operators. 4044 if (BO->isComparisonOp()) 4045 return AnalyzeComparison(S, BO); 4046 4047 // And with assignments and compound assignments. 4048 if (BO->isAssignmentOp()) 4049 return AnalyzeAssignment(S, BO); 4050 } 4051 4052 // These break the otherwise-useful invariant below. Fortunately, 4053 // we don't really need to recurse into them, because any internal 4054 // expressions should have been analyzed already when they were 4055 // built into statements. 4056 if (isa<StmtExpr>(E)) return; 4057 4058 // Don't descend into unevaluated contexts. 4059 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4060 4061 // Now just recurse over the expression's children. 4062 CC = E->getExprLoc(); 4063 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4064 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4065 for (Stmt::child_range I = E->children(); I; ++I) { 4066 Expr *ChildExpr = cast<Expr>(*I); 4067 if (IsLogicalOperator && 4068 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4069 // Ignore checking string literals that are in logical operators. 4070 continue; 4071 AnalyzeImplicitConversions(S, ChildExpr, CC); 4072 } 4073} 4074 4075} // end anonymous namespace 4076 4077/// Diagnoses "dangerous" implicit conversions within the given 4078/// expression (which is a full expression). Implements -Wconversion 4079/// and -Wsign-compare. 4080/// 4081/// \param CC the "context" location of the implicit conversion, i.e. 4082/// the most location of the syntactic entity requiring the implicit 4083/// conversion 4084void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4085 // Don't diagnose in unevaluated contexts. 4086 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4087 return; 4088 4089 // Don't diagnose for value- or type-dependent expressions. 4090 if (E->isTypeDependent() || E->isValueDependent()) 4091 return; 4092 4093 // Check for array bounds violations in cases where the check isn't triggered 4094 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4095 // ArraySubscriptExpr is on the RHS of a variable initialization. 4096 CheckArrayAccess(E); 4097 4098 // This is not the right CC for (e.g.) a variable initialization. 4099 AnalyzeImplicitConversions(*this, E, CC); 4100} 4101 4102void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4103 FieldDecl *BitField, 4104 Expr *Init) { 4105 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4106} 4107 4108/// CheckParmsForFunctionDef - Check that the parameters of the given 4109/// function are appropriate for the definition of a function. This 4110/// takes care of any checks that cannot be performed on the 4111/// declaration itself, e.g., that the types of each of the function 4112/// parameters are complete. 4113bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4114 bool CheckParameterNames) { 4115 bool HasInvalidParm = false; 4116 for (; P != PEnd; ++P) { 4117 ParmVarDecl *Param = *P; 4118 4119 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4120 // function declarator that is part of a function definition of 4121 // that function shall not have incomplete type. 4122 // 4123 // This is also C++ [dcl.fct]p6. 4124 if (!Param->isInvalidDecl() && 4125 RequireCompleteType(Param->getLocation(), Param->getType(), 4126 diag::err_typecheck_decl_incomplete_type)) { 4127 Param->setInvalidDecl(); 4128 HasInvalidParm = true; 4129 } 4130 4131 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4132 // declaration of each parameter shall include an identifier. 4133 if (CheckParameterNames && 4134 Param->getIdentifier() == 0 && 4135 !Param->isImplicit() && 4136 !getLangOptions().CPlusPlus) 4137 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4138 4139 // C99 6.7.5.3p12: 4140 // If the function declarator is not part of a definition of that 4141 // function, parameters may have incomplete type and may use the [*] 4142 // notation in their sequences of declarator specifiers to specify 4143 // variable length array types. 4144 QualType PType = Param->getOriginalType(); 4145 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4146 if (AT->getSizeModifier() == ArrayType::Star) { 4147 // FIXME: This diagnosic should point the the '[*]' if source-location 4148 // information is added for it. 4149 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4150 } 4151 } 4152 } 4153 4154 return HasInvalidParm; 4155} 4156 4157/// CheckCastAlign - Implements -Wcast-align, which warns when a 4158/// pointer cast increases the alignment requirements. 4159void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4160 // This is actually a lot of work to potentially be doing on every 4161 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4162 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4163 TRange.getBegin()) 4164 == DiagnosticsEngine::Ignored) 4165 return; 4166 4167 // Ignore dependent types. 4168 if (T->isDependentType() || Op->getType()->isDependentType()) 4169 return; 4170 4171 // Require that the destination be a pointer type. 4172 const PointerType *DestPtr = T->getAs<PointerType>(); 4173 if (!DestPtr) return; 4174 4175 // If the destination has alignment 1, we're done. 4176 QualType DestPointee = DestPtr->getPointeeType(); 4177 if (DestPointee->isIncompleteType()) return; 4178 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4179 if (DestAlign.isOne()) return; 4180 4181 // Require that the source be a pointer type. 4182 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4183 if (!SrcPtr) return; 4184 QualType SrcPointee = SrcPtr->getPointeeType(); 4185 4186 // Whitelist casts from cv void*. We already implicitly 4187 // whitelisted casts to cv void*, since they have alignment 1. 4188 // Also whitelist casts involving incomplete types, which implicitly 4189 // includes 'void'. 4190 if (SrcPointee->isIncompleteType()) return; 4191 4192 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4193 if (SrcAlign >= DestAlign) return; 4194 4195 Diag(TRange.getBegin(), diag::warn_cast_align) 4196 << Op->getType() << T 4197 << static_cast<unsigned>(SrcAlign.getQuantity()) 4198 << static_cast<unsigned>(DestAlign.getQuantity()) 4199 << TRange << Op->getSourceRange(); 4200} 4201 4202static const Type* getElementType(const Expr *BaseExpr) { 4203 const Type* EltType = BaseExpr->getType().getTypePtr(); 4204 if (EltType->isAnyPointerType()) 4205 return EltType->getPointeeType().getTypePtr(); 4206 else if (EltType->isArrayType()) 4207 return EltType->getBaseElementTypeUnsafe(); 4208 return EltType; 4209} 4210 4211/// \brief Check whether this array fits the idiom of a size-one tail padded 4212/// array member of a struct. 4213/// 4214/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4215/// commonly used to emulate flexible arrays in C89 code. 4216static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4217 const NamedDecl *ND) { 4218 if (Size != 1 || !ND) return false; 4219 4220 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4221 if (!FD) return false; 4222 4223 // Don't consider sizes resulting from macro expansions or template argument 4224 // substitution to form C89 tail-padded arrays. 4225 ConstantArrayTypeLoc TL = 4226 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 4227 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 4228 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4229 return false; 4230 4231 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4232 if (!RD) return false; 4233 if (RD->isUnion()) return false; 4234 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4235 if (!CRD->isStandardLayout()) return false; 4236 } 4237 4238 // See if this is the last field decl in the record. 4239 const Decl *D = FD; 4240 while ((D = D->getNextDeclInContext())) 4241 if (isa<FieldDecl>(D)) 4242 return false; 4243 return true; 4244} 4245 4246void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4247 bool isSubscript, bool AllowOnePastEnd) { 4248 const Type* EffectiveType = getElementType(BaseExpr); 4249 BaseExpr = BaseExpr->IgnoreParenCasts(); 4250 IndexExpr = IndexExpr->IgnoreParenCasts(); 4251 4252 const ConstantArrayType *ArrayTy = 4253 Context.getAsConstantArrayType(BaseExpr->getType()); 4254 if (!ArrayTy) 4255 return; 4256 4257 if (IndexExpr->isValueDependent()) 4258 return; 4259 llvm::APSInt index; 4260 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 4261 return; 4262 4263 const NamedDecl *ND = NULL; 4264 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4265 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4266 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4267 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4268 4269 if (index.isUnsigned() || !index.isNegative()) { 4270 llvm::APInt size = ArrayTy->getSize(); 4271 if (!size.isStrictlyPositive()) 4272 return; 4273 4274 const Type* BaseType = getElementType(BaseExpr); 4275 if (BaseType != EffectiveType) { 4276 // Make sure we're comparing apples to apples when comparing index to size 4277 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4278 uint64_t array_typesize = Context.getTypeSize(BaseType); 4279 // Handle ptrarith_typesize being zero, such as when casting to void* 4280 if (!ptrarith_typesize) ptrarith_typesize = 1; 4281 if (ptrarith_typesize != array_typesize) { 4282 // There's a cast to a different size type involved 4283 uint64_t ratio = array_typesize / ptrarith_typesize; 4284 // TODO: Be smarter about handling cases where array_typesize is not a 4285 // multiple of ptrarith_typesize 4286 if (ptrarith_typesize * ratio == array_typesize) 4287 size *= llvm::APInt(size.getBitWidth(), ratio); 4288 } 4289 } 4290 4291 if (size.getBitWidth() > index.getBitWidth()) 4292 index = index.sext(size.getBitWidth()); 4293 else if (size.getBitWidth() < index.getBitWidth()) 4294 size = size.sext(index.getBitWidth()); 4295 4296 // For array subscripting the index must be less than size, but for pointer 4297 // arithmetic also allow the index (offset) to be equal to size since 4298 // computing the next address after the end of the array is legal and 4299 // commonly done e.g. in C++ iterators and range-based for loops. 4300 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 4301 return; 4302 4303 // Also don't warn for arrays of size 1 which are members of some 4304 // structure. These are often used to approximate flexible arrays in C89 4305 // code. 4306 if (IsTailPaddedMemberArray(*this, size, ND)) 4307 return; 4308 4309 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4310 if (isSubscript) 4311 DiagID = diag::warn_array_index_exceeds_bounds; 4312 4313 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4314 PDiag(DiagID) << index.toString(10, true) 4315 << size.toString(10, true) 4316 << (unsigned)size.getLimitedValue(~0U) 4317 << IndexExpr->getSourceRange()); 4318 } else { 4319 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4320 if (!isSubscript) { 4321 DiagID = diag::warn_ptr_arith_precedes_bounds; 4322 if (index.isNegative()) index = -index; 4323 } 4324 4325 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4326 PDiag(DiagID) << index.toString(10, true) 4327 << IndexExpr->getSourceRange()); 4328 } 4329 4330 if (!ND) { 4331 // Try harder to find a NamedDecl to point at in the note. 4332 while (const ArraySubscriptExpr *ASE = 4333 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4334 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4335 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4336 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4337 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4338 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4339 } 4340 4341 if (ND) 4342 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4343 PDiag(diag::note_array_index_out_of_bounds) 4344 << ND->getDeclName()); 4345} 4346 4347void Sema::CheckArrayAccess(const Expr *expr) { 4348 int AllowOnePastEnd = 0; 4349 while (expr) { 4350 expr = expr->IgnoreParenImpCasts(); 4351 switch (expr->getStmtClass()) { 4352 case Stmt::ArraySubscriptExprClass: { 4353 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4354 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true, 4355 AllowOnePastEnd > 0); 4356 return; 4357 } 4358 case Stmt::UnaryOperatorClass: { 4359 // Only unwrap the * and & unary operators 4360 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4361 expr = UO->getSubExpr(); 4362 switch (UO->getOpcode()) { 4363 case UO_AddrOf: 4364 AllowOnePastEnd++; 4365 break; 4366 case UO_Deref: 4367 AllowOnePastEnd--; 4368 break; 4369 default: 4370 return; 4371 } 4372 break; 4373 } 4374 case Stmt::ConditionalOperatorClass: { 4375 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4376 if (const Expr *lhs = cond->getLHS()) 4377 CheckArrayAccess(lhs); 4378 if (const Expr *rhs = cond->getRHS()) 4379 CheckArrayAccess(rhs); 4380 return; 4381 } 4382 default: 4383 return; 4384 } 4385 } 4386} 4387 4388//===--- CHECK: Objective-C retain cycles ----------------------------------// 4389 4390namespace { 4391 struct RetainCycleOwner { 4392 RetainCycleOwner() : Variable(0), Indirect(false) {} 4393 VarDecl *Variable; 4394 SourceRange Range; 4395 SourceLocation Loc; 4396 bool Indirect; 4397 4398 void setLocsFrom(Expr *e) { 4399 Loc = e->getExprLoc(); 4400 Range = e->getSourceRange(); 4401 } 4402 }; 4403} 4404 4405/// Consider whether capturing the given variable can possibly lead to 4406/// a retain cycle. 4407static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4408 // In ARC, it's captured strongly iff the variable has __strong 4409 // lifetime. In MRR, it's captured strongly if the variable is 4410 // __block and has an appropriate type. 4411 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4412 return false; 4413 4414 owner.Variable = var; 4415 owner.setLocsFrom(ref); 4416 return true; 4417} 4418 4419static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 4420 while (true) { 4421 e = e->IgnoreParens(); 4422 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4423 switch (cast->getCastKind()) { 4424 case CK_BitCast: 4425 case CK_LValueBitCast: 4426 case CK_LValueToRValue: 4427 case CK_ARCReclaimReturnedObject: 4428 e = cast->getSubExpr(); 4429 continue; 4430 4431 default: 4432 return false; 4433 } 4434 } 4435 4436 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4437 ObjCIvarDecl *ivar = ref->getDecl(); 4438 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4439 return false; 4440 4441 // Try to find a retain cycle in the base. 4442 if (!findRetainCycleOwner(ref->getBase(), owner)) 4443 return false; 4444 4445 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4446 owner.Indirect = true; 4447 return true; 4448 } 4449 4450 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4451 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4452 if (!var) return false; 4453 return considerVariable(var, ref, owner); 4454 } 4455 4456 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 4457 owner.Variable = ref->getDecl(); 4458 owner.setLocsFrom(ref); 4459 return true; 4460 } 4461 4462 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4463 if (member->isArrow()) return false; 4464 4465 // Don't count this as an indirect ownership. 4466 e = member->getBase(); 4467 continue; 4468 } 4469 4470 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4471 // Only pay attention to pseudo-objects on property references. 4472 ObjCPropertyRefExpr *pre 4473 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4474 ->IgnoreParens()); 4475 if (!pre) return false; 4476 if (pre->isImplicitProperty()) return false; 4477 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4478 if (!property->isRetaining() && 4479 !(property->getPropertyIvarDecl() && 4480 property->getPropertyIvarDecl()->getType() 4481 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4482 return false; 4483 4484 owner.Indirect = true; 4485 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4486 ->getSourceExpr()); 4487 continue; 4488 } 4489 4490 // Array ivars? 4491 4492 return false; 4493 } 4494} 4495 4496namespace { 4497 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4498 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4499 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4500 Variable(variable), Capturer(0) {} 4501 4502 VarDecl *Variable; 4503 Expr *Capturer; 4504 4505 void VisitDeclRefExpr(DeclRefExpr *ref) { 4506 if (ref->getDecl() == Variable && !Capturer) 4507 Capturer = ref; 4508 } 4509 4510 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 4511 if (ref->getDecl() == Variable && !Capturer) 4512 Capturer = ref; 4513 } 4514 4515 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4516 if (Capturer) return; 4517 Visit(ref->getBase()); 4518 if (Capturer && ref->isFreeIvar()) 4519 Capturer = ref; 4520 } 4521 4522 void VisitBlockExpr(BlockExpr *block) { 4523 // Look inside nested blocks 4524 if (block->getBlockDecl()->capturesVariable(Variable)) 4525 Visit(block->getBlockDecl()->getBody()); 4526 } 4527 }; 4528} 4529 4530/// Check whether the given argument is a block which captures a 4531/// variable. 4532static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4533 assert(owner.Variable && owner.Loc.isValid()); 4534 4535 e = e->IgnoreParenCasts(); 4536 BlockExpr *block = dyn_cast<BlockExpr>(e); 4537 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4538 return 0; 4539 4540 FindCaptureVisitor visitor(S.Context, owner.Variable); 4541 visitor.Visit(block->getBlockDecl()->getBody()); 4542 return visitor.Capturer; 4543} 4544 4545static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4546 RetainCycleOwner &owner) { 4547 assert(capturer); 4548 assert(owner.Variable && owner.Loc.isValid()); 4549 4550 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4551 << owner.Variable << capturer->getSourceRange(); 4552 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4553 << owner.Indirect << owner.Range; 4554} 4555 4556/// Check for a keyword selector that starts with the word 'add' or 4557/// 'set'. 4558static bool isSetterLikeSelector(Selector sel) { 4559 if (sel.isUnarySelector()) return false; 4560 4561 StringRef str = sel.getNameForSlot(0); 4562 while (!str.empty() && str.front() == '_') str = str.substr(1); 4563 if (str.startswith("set")) 4564 str = str.substr(3); 4565 else if (str.startswith("add")) { 4566 // Specially whitelist 'addOperationWithBlock:'. 4567 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4568 return false; 4569 str = str.substr(3); 4570 } 4571 else 4572 return false; 4573 4574 if (str.empty()) return true; 4575 return !islower(str.front()); 4576} 4577 4578/// Check a message send to see if it's likely to cause a retain cycle. 4579void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4580 // Only check instance methods whose selector looks like a setter. 4581 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4582 return; 4583 4584 // Try to find a variable that the receiver is strongly owned by. 4585 RetainCycleOwner owner; 4586 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4587 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 4588 return; 4589 } else { 4590 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4591 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4592 owner.Loc = msg->getSuperLoc(); 4593 owner.Range = msg->getSuperLoc(); 4594 } 4595 4596 // Check whether the receiver is captured by any of the arguments. 4597 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4598 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4599 return diagnoseRetainCycle(*this, capturer, owner); 4600} 4601 4602/// Check a property assign to see if it's likely to cause a retain cycle. 4603void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4604 RetainCycleOwner owner; 4605 if (!findRetainCycleOwner(receiver, owner)) 4606 return; 4607 4608 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4609 diagnoseRetainCycle(*this, capturer, owner); 4610} 4611 4612bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4613 QualType LHS, Expr *RHS) { 4614 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4615 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4616 return false; 4617 // strip off any implicit cast added to get to the one arc-specific 4618 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4619 if (cast->getCastKind() == CK_ARCConsumeObject) { 4620 Diag(Loc, diag::warn_arc_retained_assign) 4621 << (LT == Qualifiers::OCL_ExplicitNone) 4622 << RHS->getSourceRange(); 4623 return true; 4624 } 4625 RHS = cast->getSubExpr(); 4626 } 4627 return false; 4628} 4629 4630void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4631 Expr *LHS, Expr *RHS) { 4632 QualType LHSType = LHS->getType(); 4633 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4634 return; 4635 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4636 // FIXME. Check for other life times. 4637 if (LT != Qualifiers::OCL_None) 4638 return; 4639 4640 if (ObjCPropertyRefExpr *PRE 4641 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens())) { 4642 if (PRE->isImplicitProperty()) 4643 return; 4644 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4645 if (!PD) 4646 return; 4647 4648 unsigned Attributes = PD->getPropertyAttributes(); 4649 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4650 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4651 if (cast->getCastKind() == CK_ARCConsumeObject) { 4652 Diag(Loc, diag::warn_arc_retained_property_assign) 4653 << RHS->getSourceRange(); 4654 return; 4655 } 4656 RHS = cast->getSubExpr(); 4657 } 4658 } 4659} 4660