1 //===- Archive.cpp - ar File Format implementation --------------*- C++ -*-===// 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 defines the ArchiveObjectFile class. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/Object/Archive.h" 15 #include "llvm/ADT/APInt.h" 16 #include "llvm/ADT/SmallString.h" 17 #include "llvm/ADT/Twine.h" 18 #include "llvm/Support/Endian.h" 19 #include "llvm/Support/MemoryBuffer.h" 20 #include "llvm/Support/Path.h" 21 22 using namespace llvm; 23 using namespace object; 24 using namespace llvm::support::endian; 25 26 static const char *const Magic = "!<arch>\n"; 27 static const char *const ThinMagic = "!<thin>\n"; 28 29 void Archive::anchor() { } 30 31 StringRef ArchiveMemberHeader::getName() const { 32 char EndCond; 33 if (Name[0] == '/' || Name[0] == '#') 34 EndCond = ' '; 35 else 36 EndCond = '/'; 37 llvm::StringRef::size_type end = 38 llvm::StringRef(Name, sizeof(Name)).find(EndCond); 39 if (end == llvm::StringRef::npos) 40 end = sizeof(Name); 41 assert(end <= sizeof(Name) && end > 0); 42 // Don't include the EndCond if there is one. 43 return llvm::StringRef(Name, end); 44 } 45 46 ErrorOr<uint32_t> ArchiveMemberHeader::getSize() const { 47 uint32_t Ret; 48 if (llvm::StringRef(Size, sizeof(Size)).rtrim(" ").getAsInteger(10, Ret)) 49 return object_error::parse_failed; // Size is not a decimal number. 50 return Ret; 51 } 52 53 sys::fs::perms ArchiveMemberHeader::getAccessMode() const { 54 unsigned Ret; 55 if (StringRef(AccessMode, sizeof(AccessMode)).rtrim(" ").getAsInteger(8, Ret)) 56 llvm_unreachable("Access mode is not an octal number."); 57 return static_cast<sys::fs::perms>(Ret); 58 } 59 60 sys::TimeValue ArchiveMemberHeader::getLastModified() const { 61 unsigned Seconds; 62 if (StringRef(LastModified, sizeof(LastModified)).rtrim(" ") 63 .getAsInteger(10, Seconds)) 64 llvm_unreachable("Last modified time not a decimal number."); 65 66 sys::TimeValue Ret; 67 Ret.fromEpochTime(Seconds); 68 return Ret; 69 } 70 71 unsigned ArchiveMemberHeader::getUID() const { 72 unsigned Ret; 73 if (StringRef(UID, sizeof(UID)).rtrim(" ").getAsInteger(10, Ret)) 74 llvm_unreachable("UID time not a decimal number."); 75 return Ret; 76 } 77 78 unsigned ArchiveMemberHeader::getGID() const { 79 unsigned Ret; 80 if (StringRef(GID, sizeof(GID)).rtrim(" ").getAsInteger(10, Ret)) 81 llvm_unreachable("GID time not a decimal number."); 82 return Ret; 83 } 84 85 Archive::Child::Child(const Archive *Parent, StringRef Data, 86 uint16_t StartOfFile) 87 : Parent(Parent), Data(Data), StartOfFile(StartOfFile) {} 88 89 Archive::Child::Child(const Archive *Parent, const char *Start, 90 std::error_code *EC) 91 : Parent(Parent) { 92 if (!Start) 93 return; 94 95 uint64_t Size = sizeof(ArchiveMemberHeader); 96 Data = StringRef(Start, Size); 97 if (!isThinMember()) { 98 ErrorOr<uint64_t> MemberSize = getRawSize(); 99 if ((*EC = MemberSize.getError())) 100 return; 101 Size += MemberSize.get(); 102 Data = StringRef(Start, Size); 103 } 104 105 // Setup StartOfFile and PaddingBytes. 106 StartOfFile = sizeof(ArchiveMemberHeader); 107 // Don't include attached name. 108 StringRef Name = getRawName(); 109 if (Name.startswith("#1/")) { 110 uint64_t NameSize; 111 if (Name.substr(3).rtrim(" ").getAsInteger(10, NameSize)) 112 llvm_unreachable("Long name length is not an integer"); 113 StartOfFile += NameSize; 114 } 115 } 116 117 ErrorOr<uint64_t> Archive::Child::getSize() const { 118 if (Parent->IsThin) { 119 ErrorOr<uint32_t> Size = getHeader()->getSize(); 120 if (std::error_code EC = Size.getError()) 121 return EC; 122 return Size.get(); 123 } 124 return Data.size() - StartOfFile; 125 } 126 127 ErrorOr<uint64_t> Archive::Child::getRawSize() const { 128 ErrorOr<uint32_t> Size = getHeader()->getSize(); 129 if (std::error_code EC = Size.getError()) 130 return EC; 131 return Size.get(); 132 } 133 134 bool Archive::Child::isThinMember() const { 135 StringRef Name = getHeader()->getName(); 136 return Parent->IsThin && Name != "/" && Name != "//"; 137 } 138 139 ErrorOr<StringRef> Archive::Child::getBuffer() const { 140 if (!isThinMember()) { 141 ErrorOr<uint32_t> Size = getSize(); 142 if (std::error_code EC = Size.getError()) 143 return EC; 144 return StringRef(Data.data() + StartOfFile, Size.get()); 145 } 146 ErrorOr<StringRef> Name = getName(); 147 if (std::error_code EC = Name.getError()) 148 return EC; 149 SmallString<128> FullName = sys::path::parent_path( 150 Parent->getMemoryBufferRef().getBufferIdentifier()); 151 sys::path::append(FullName, *Name); 152 ErrorOr<std::unique_ptr<MemoryBuffer>> Buf = MemoryBuffer::getFile(FullName); 153 if (std::error_code EC = Buf.getError()) 154 return EC; 155 Parent->ThinBuffers.push_back(std::move(*Buf)); 156 return Parent->ThinBuffers.back()->getBuffer(); 157 } 158 159 ErrorOr<Archive::Child> Archive::Child::getNext() const { 160 size_t SpaceToSkip = Data.size(); 161 // If it's odd, add 1 to make it even. 162 if (SpaceToSkip & 1) 163 ++SpaceToSkip; 164 165 const char *NextLoc = Data.data() + SpaceToSkip; 166 167 // Check to see if this is at the end of the archive. 168 if (NextLoc == Parent->Data.getBufferEnd()) 169 return Child(Parent, nullptr, nullptr); 170 171 // Check to see if this is past the end of the archive. 172 if (NextLoc > Parent->Data.getBufferEnd()) 173 return object_error::parse_failed; 174 175 std::error_code EC; 176 Child Ret(Parent, NextLoc, &EC); 177 if (EC) 178 return EC; 179 return Ret; 180 } 181 182 uint64_t Archive::Child::getChildOffset() const { 183 const char *a = Parent->Data.getBuffer().data(); 184 const char *c = Data.data(); 185 uint64_t offset = c - a; 186 return offset; 187 } 188 189 ErrorOr<StringRef> Archive::Child::getName() const { 190 StringRef name = getRawName(); 191 // Check if it's a special name. 192 if (name[0] == '/') { 193 if (name.size() == 1) // Linker member. 194 return name; 195 if (name.size() == 2 && name[1] == '/') // String table. 196 return name; 197 // It's a long name. 198 // Get the offset. 199 std::size_t offset; 200 if (name.substr(1).rtrim(" ").getAsInteger(10, offset)) 201 llvm_unreachable("Long name offset is not an integer"); 202 203 // Verify it. 204 if (offset >= Parent->StringTable.size()) 205 return object_error::parse_failed; 206 const char *addr = Parent->StringTable.begin() + offset; 207 208 // GNU long file names end with a "/\n". 209 if (Parent->kind() == K_GNU || Parent->kind() == K_MIPS64) { 210 StringRef::size_type End = StringRef(addr).find('\n'); 211 return StringRef(addr, End - 1); 212 } 213 return StringRef(addr); 214 } else if (name.startswith("#1/")) { 215 uint64_t name_size; 216 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size)) 217 llvm_unreachable("Long name length is not an ingeter"); 218 return Data.substr(sizeof(ArchiveMemberHeader), name_size) 219 .rtrim(StringRef("\0", 1)); 220 } 221 // It's a simple name. 222 if (name[name.size() - 1] == '/') 223 return name.substr(0, name.size() - 1); 224 return name; 225 } 226 227 ErrorOr<MemoryBufferRef> Archive::Child::getMemoryBufferRef() const { 228 ErrorOr<StringRef> NameOrErr = getName(); 229 if (std::error_code EC = NameOrErr.getError()) 230 return EC; 231 StringRef Name = NameOrErr.get(); 232 ErrorOr<StringRef> Buf = getBuffer(); 233 if (std::error_code EC = Buf.getError()) 234 return EC; 235 return MemoryBufferRef(*Buf, Name); 236 } 237 238 ErrorOr<std::unique_ptr<Binary>> 239 Archive::Child::getAsBinary(LLVMContext *Context) const { 240 ErrorOr<MemoryBufferRef> BuffOrErr = getMemoryBufferRef(); 241 if (std::error_code EC = BuffOrErr.getError()) 242 return EC; 243 244 return createBinary(BuffOrErr.get(), Context); 245 } 246 247 ErrorOr<std::unique_ptr<Archive>> Archive::create(MemoryBufferRef Source) { 248 std::error_code EC; 249 std::unique_ptr<Archive> Ret(new Archive(Source, EC)); 250 if (EC) 251 return EC; 252 return std::move(Ret); 253 } 254 255 void Archive::setFirstRegular(const Child &C) { 256 FirstRegularData = C.Data; 257 FirstRegularStartOfFile = C.StartOfFile; 258 } 259 260 Archive::Archive(MemoryBufferRef Source, std::error_code &ec) 261 : Binary(Binary::ID_Archive, Source) { 262 StringRef Buffer = Data.getBuffer(); 263 // Check for sufficient magic. 264 if (Buffer.startswith(ThinMagic)) { 265 IsThin = true; 266 } else if (Buffer.startswith(Magic)) { 267 IsThin = false; 268 } else { 269 ec = object_error::invalid_file_type; 270 return; 271 } 272 273 // Get the special members. 274 child_iterator I = child_begin(false); 275 if ((ec = I->getError())) 276 return; 277 child_iterator E = child_end(); 278 279 if (I == E) { 280 ec = std::error_code(); 281 return; 282 } 283 const Child *C = &**I; 284 285 auto Increment = [&]() { 286 ++I; 287 if ((ec = I->getError())) 288 return true; 289 C = &**I; 290 return false; 291 }; 292 293 StringRef Name = C->getRawName(); 294 295 // Below is the pattern that is used to figure out the archive format 296 // GNU archive format 297 // First member : / (may exist, if it exists, points to the symbol table ) 298 // Second member : // (may exist, if it exists, points to the string table) 299 // Note : The string table is used if the filename exceeds 15 characters 300 // BSD archive format 301 // First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table) 302 // There is no string table, if the filename exceeds 15 characters or has a 303 // embedded space, the filename has #1/<size>, The size represents the size 304 // of the filename that needs to be read after the archive header 305 // COFF archive format 306 // First member : / 307 // Second member : / (provides a directory of symbols) 308 // Third member : // (may exist, if it exists, contains the string table) 309 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present 310 // even if the string table is empty. However, lib.exe does not in fact 311 // seem to create the third member if there's no member whose filename 312 // exceeds 15 characters. So the third member is optional. 313 314 if (Name == "__.SYMDEF") { 315 Format = K_BSD; 316 // We know that the symbol table is not an external file, so we just assert 317 // there is no error. 318 SymbolTable = *C->getBuffer(); 319 if (Increment()) 320 return; 321 setFirstRegular(*C); 322 323 ec = std::error_code(); 324 return; 325 } 326 327 if (Name.startswith("#1/")) { 328 Format = K_BSD; 329 // We know this is BSD, so getName will work since there is no string table. 330 ErrorOr<StringRef> NameOrErr = C->getName(); 331 ec = NameOrErr.getError(); 332 if (ec) 333 return; 334 Name = NameOrErr.get(); 335 if (Name == "__.SYMDEF SORTED" || Name == "__.SYMDEF") { 336 // We know that the symbol table is not an external file, so we just 337 // assert there is no error. 338 SymbolTable = *C->getBuffer(); 339 if (Increment()) 340 return; 341 } 342 setFirstRegular(*C); 343 return; 344 } 345 346 // MIPS 64-bit ELF archives use a special format of a symbol table. 347 // This format is marked by `ar_name` field equals to "/SYM64/". 348 // For detailed description see page 96 in the following document: 349 // http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf 350 351 bool has64SymTable = false; 352 if (Name == "/" || Name == "/SYM64/") { 353 // We know that the symbol table is not an external file, so we just assert 354 // there is no error. 355 SymbolTable = *C->getBuffer(); 356 if (Name == "/SYM64/") 357 has64SymTable = true; 358 359 if (Increment()) 360 return; 361 if (I == E) { 362 ec = std::error_code(); 363 return; 364 } 365 Name = C->getRawName(); 366 } 367 368 if (Name == "//") { 369 Format = has64SymTable ? K_MIPS64 : K_GNU; 370 // The string table is never an external member, so we just assert on the 371 // ErrorOr. 372 StringTable = *C->getBuffer(); 373 if (Increment()) 374 return; 375 setFirstRegular(*C); 376 ec = std::error_code(); 377 return; 378 } 379 380 if (Name[0] != '/') { 381 Format = has64SymTable ? K_MIPS64 : K_GNU; 382 setFirstRegular(*C); 383 ec = std::error_code(); 384 return; 385 } 386 387 if (Name != "/") { 388 ec = object_error::parse_failed; 389 return; 390 } 391 392 Format = K_COFF; 393 // We know that the symbol table is not an external file, so we just assert 394 // there is no error. 395 SymbolTable = *C->getBuffer(); 396 397 if (Increment()) 398 return; 399 400 if (I == E) { 401 setFirstRegular(*C); 402 ec = std::error_code(); 403 return; 404 } 405 406 Name = C->getRawName(); 407 408 if (Name == "//") { 409 // The string table is never an external member, so we just assert on the 410 // ErrorOr. 411 StringTable = *C->getBuffer(); 412 if (Increment()) 413 return; 414 } 415 416 setFirstRegular(*C); 417 ec = std::error_code(); 418 } 419 420 Archive::child_iterator Archive::child_begin(bool SkipInternal) const { 421 if (Data.getBufferSize() == 8) // empty archive. 422 return child_end(); 423 424 if (SkipInternal) 425 return Child(this, FirstRegularData, FirstRegularStartOfFile); 426 427 const char *Loc = Data.getBufferStart() + strlen(Magic); 428 std::error_code EC; 429 Child c(this, Loc, &EC); 430 if (EC) 431 return child_iterator(EC); 432 return child_iterator(c); 433 } 434 435 Archive::child_iterator Archive::child_end() const { 436 return Child(this, nullptr, nullptr); 437 } 438 439 StringRef Archive::Symbol::getName() const { 440 return Parent->getSymbolTable().begin() + StringIndex; 441 } 442 443 ErrorOr<Archive::Child> Archive::Symbol::getMember() const { 444 const char *Buf = Parent->getSymbolTable().begin(); 445 const char *Offsets = Buf; 446 if (Parent->kind() == K_MIPS64) 447 Offsets += sizeof(uint64_t); 448 else 449 Offsets += sizeof(uint32_t); 450 uint32_t Offset = 0; 451 if (Parent->kind() == K_GNU) { 452 Offset = read32be(Offsets + SymbolIndex * 4); 453 } else if (Parent->kind() == K_MIPS64) { 454 Offset = read64be(Offsets + SymbolIndex * 8); 455 } else if (Parent->kind() == K_BSD) { 456 // The SymbolIndex is an index into the ranlib structs that start at 457 // Offsets (the first uint32_t is the number of bytes of the ranlib 458 // structs). The ranlib structs are a pair of uint32_t's the first 459 // being a string table offset and the second being the offset into 460 // the archive of the member that defines the symbol. Which is what 461 // is needed here. 462 Offset = read32le(Offsets + SymbolIndex * 8 + 4); 463 } else { 464 // Skip offsets. 465 uint32_t MemberCount = read32le(Buf); 466 Buf += MemberCount * 4 + 4; 467 468 uint32_t SymbolCount = read32le(Buf); 469 if (SymbolIndex >= SymbolCount) 470 return object_error::parse_failed; 471 472 // Skip SymbolCount to get to the indices table. 473 const char *Indices = Buf + 4; 474 475 // Get the index of the offset in the file member offset table for this 476 // symbol. 477 uint16_t OffsetIndex = read16le(Indices + SymbolIndex * 2); 478 // Subtract 1 since OffsetIndex is 1 based. 479 --OffsetIndex; 480 481 if (OffsetIndex >= MemberCount) 482 return object_error::parse_failed; 483 484 Offset = read32le(Offsets + OffsetIndex * 4); 485 } 486 487 const char *Loc = Parent->getData().begin() + Offset; 488 std::error_code EC; 489 Child C(Parent, Loc, &EC); 490 if (EC) 491 return EC; 492 return C; 493 } 494 495 Archive::Symbol Archive::Symbol::getNext() const { 496 Symbol t(*this); 497 if (Parent->kind() == K_BSD) { 498 // t.StringIndex is an offset from the start of the __.SYMDEF or 499 // "__.SYMDEF SORTED" member into the string table for the ranlib 500 // struct indexed by t.SymbolIndex . To change t.StringIndex to the 501 // offset in the string table for t.SymbolIndex+1 we subtract the 502 // its offset from the start of the string table for t.SymbolIndex 503 // and add the offset of the string table for t.SymbolIndex+1. 504 505 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t 506 // which is the number of bytes of ranlib structs that follow. The ranlib 507 // structs are a pair of uint32_t's the first being a string table offset 508 // and the second being the offset into the archive of the member that 509 // define the symbol. After that the next uint32_t is the byte count of 510 // the string table followed by the string table. 511 const char *Buf = Parent->getSymbolTable().begin(); 512 uint32_t RanlibCount = 0; 513 RanlibCount = read32le(Buf) / 8; 514 // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount) 515 // don't change the t.StringIndex as we don't want to reference a ranlib 516 // past RanlibCount. 517 if (t.SymbolIndex + 1 < RanlibCount) { 518 const char *Ranlibs = Buf + 4; 519 uint32_t CurRanStrx = 0; 520 uint32_t NextRanStrx = 0; 521 CurRanStrx = read32le(Ranlibs + t.SymbolIndex * 8); 522 NextRanStrx = read32le(Ranlibs + (t.SymbolIndex + 1) * 8); 523 t.StringIndex -= CurRanStrx; 524 t.StringIndex += NextRanStrx; 525 } 526 } else { 527 // Go to one past next null. 528 t.StringIndex = Parent->getSymbolTable().find('\0', t.StringIndex) + 1; 529 } 530 ++t.SymbolIndex; 531 return t; 532 } 533 534 Archive::symbol_iterator Archive::symbol_begin() const { 535 if (!hasSymbolTable()) 536 return symbol_iterator(Symbol(this, 0, 0)); 537 538 const char *buf = getSymbolTable().begin(); 539 if (kind() == K_GNU) { 540 uint32_t symbol_count = 0; 541 symbol_count = read32be(buf); 542 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t))); 543 } else if (kind() == K_MIPS64) { 544 uint64_t symbol_count = read64be(buf); 545 buf += sizeof(uint64_t) + (symbol_count * (sizeof(uint64_t))); 546 } else if (kind() == K_BSD) { 547 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t 548 // which is the number of bytes of ranlib structs that follow. The ranlib 549 // structs are a pair of uint32_t's the first being a string table offset 550 // and the second being the offset into the archive of the member that 551 // define the symbol. After that the next uint32_t is the byte count of 552 // the string table followed by the string table. 553 uint32_t ranlib_count = 0; 554 ranlib_count = read32le(buf) / 8; 555 const char *ranlibs = buf + 4; 556 uint32_t ran_strx = 0; 557 ran_strx = read32le(ranlibs); 558 buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t)))); 559 // Skip the byte count of the string table. 560 buf += sizeof(uint32_t); 561 buf += ran_strx; 562 } else { 563 uint32_t member_count = 0; 564 uint32_t symbol_count = 0; 565 member_count = read32le(buf); 566 buf += 4 + (member_count * 4); // Skip offsets. 567 symbol_count = read32le(buf); 568 buf += 4 + (symbol_count * 2); // Skip indices. 569 } 570 uint32_t string_start_offset = buf - getSymbolTable().begin(); 571 return symbol_iterator(Symbol(this, 0, string_start_offset)); 572 } 573 574 Archive::symbol_iterator Archive::symbol_end() const { 575 return symbol_iterator(Symbol(this, getNumberOfSymbols(), 0)); 576 } 577 578 uint32_t Archive::getNumberOfSymbols() const { 579 if (!hasSymbolTable()) 580 return 0; 581 const char *buf = getSymbolTable().begin(); 582 if (kind() == K_GNU) 583 return read32be(buf); 584 if (kind() == K_MIPS64) 585 return read64be(buf); 586 if (kind() == K_BSD) 587 return read32le(buf) / 8; 588 uint32_t member_count = 0; 589 member_count = read32le(buf); 590 buf += 4 + (member_count * 4); // Skip offsets. 591 return read32le(buf); 592 } 593 594 Archive::child_iterator Archive::findSym(StringRef name) const { 595 Archive::symbol_iterator bs = symbol_begin(); 596 Archive::symbol_iterator es = symbol_end(); 597 598 for (; bs != es; ++bs) { 599 StringRef SymName = bs->getName(); 600 if (SymName == name) { 601 ErrorOr<Archive::child_iterator> ResultOrErr = bs->getMember(); 602 // FIXME: Should we really eat the error? 603 if (ResultOrErr.getError()) 604 return child_end(); 605 return ResultOrErr.get(); 606 } 607 } 608 return child_end(); 609 } 610 611 bool Archive::hasSymbolTable() const { return !SymbolTable.empty(); } 612