1 //===- Archive.cpp - ar File Format implementation ------------------------===// 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/Optional.h" 16 #include "llvm/ADT/SmallString.h" 17 #include "llvm/ADT/StringRef.h" 18 #include "llvm/ADT/Twine.h" 19 #include "llvm/Object/Binary.h" 20 #include "llvm/Object/Error.h" 21 #include "llvm/Support/Chrono.h" 22 #include "llvm/Support/Endian.h" 23 #include "llvm/Support/Error.h" 24 #include "llvm/Support/ErrorOr.h" 25 #include "llvm/Support/FileSystem.h" 26 #include "llvm/Support/MemoryBuffer.h" 27 #include "llvm/Support/Path.h" 28 #include "llvm/Support/raw_ostream.h" 29 #include <algorithm> 30 #include <cassert> 31 #include <cstddef> 32 #include <cstdint> 33 #include <cstring> 34 #include <memory> 35 #include <string> 36 #include <system_error> 37 38 using namespace llvm; 39 using namespace object; 40 using namespace llvm::support::endian; 41 42 static const char *const Magic = "!<arch>\n"; 43 static const char *const ThinMagic = "!<thin>\n"; 44 45 void Archive::anchor() {} 46 47 static Error 48 malformedError(Twine Msg) { 49 std::string StringMsg = "truncated or malformed archive (" + Msg.str() + ")"; 50 return make_error<GenericBinaryError>(std::move(StringMsg), 51 object_error::parse_failed); 52 } 53 54 ArchiveMemberHeader::ArchiveMemberHeader(const Archive *Parent, 55 const char *RawHeaderPtr, 56 uint64_t Size, Error *Err) 57 : Parent(Parent), 58 ArMemHdr(reinterpret_cast<const ArMemHdrType *>(RawHeaderPtr)) { 59 if (RawHeaderPtr == nullptr) 60 return; 61 ErrorAsOutParameter ErrAsOutParam(Err); 62 63 if (Size < sizeof(ArMemHdrType)) { 64 if (Err) { 65 std::string Msg("remaining size of archive too small for next archive " 66 "member header "); 67 Expected<StringRef> NameOrErr = getName(Size); 68 if (!NameOrErr) { 69 consumeError(NameOrErr.takeError()); 70 uint64_t Offset = RawHeaderPtr - Parent->getData().data(); 71 *Err = malformedError(Msg + "at offset " + Twine(Offset)); 72 } else 73 *Err = malformedError(Msg + "for " + NameOrErr.get()); 74 } 75 return; 76 } 77 if (ArMemHdr->Terminator[0] != '`' || ArMemHdr->Terminator[1] != '\n') { 78 if (Err) { 79 std::string Buf; 80 raw_string_ostream OS(Buf); 81 OS.write_escaped(StringRef(ArMemHdr->Terminator, 82 sizeof(ArMemHdr->Terminator))); 83 OS.flush(); 84 std::string Msg("terminator characters in archive member \"" + Buf + 85 "\" not the correct \"`\\n\" values for the archive " 86 "member header "); 87 Expected<StringRef> NameOrErr = getName(Size); 88 if (!NameOrErr) { 89 consumeError(NameOrErr.takeError()); 90 uint64_t Offset = RawHeaderPtr - Parent->getData().data(); 91 *Err = malformedError(Msg + "at offset " + Twine(Offset)); 92 } else 93 *Err = malformedError(Msg + "for " + NameOrErr.get()); 94 } 95 return; 96 } 97 } 98 99 // This gets the raw name from the ArMemHdr->Name field and checks that it is 100 // valid for the kind of archive. If it is not valid it returns an Error. 101 Expected<StringRef> ArchiveMemberHeader::getRawName() const { 102 char EndCond; 103 auto Kind = Parent->kind(); 104 if (Kind == Archive::K_BSD || Kind == Archive::K_DARWIN64) { 105 if (ArMemHdr->Name[0] == ' ') { 106 uint64_t Offset = reinterpret_cast<const char *>(ArMemHdr) - 107 Parent->getData().data(); 108 return malformedError("name contains a leading space for archive member " 109 "header at offset " + Twine(Offset)); 110 } 111 EndCond = ' '; 112 } 113 else if (ArMemHdr->Name[0] == '/' || ArMemHdr->Name[0] == '#') 114 EndCond = ' '; 115 else 116 EndCond = '/'; 117 StringRef::size_type end = 118 StringRef(ArMemHdr->Name, sizeof(ArMemHdr->Name)).find(EndCond); 119 if (end == StringRef::npos) 120 end = sizeof(ArMemHdr->Name); 121 assert(end <= sizeof(ArMemHdr->Name) && end > 0); 122 // Don't include the EndCond if there is one. 123 return StringRef(ArMemHdr->Name, end); 124 } 125 126 // This gets the name looking up long names. Size is the size of the archive 127 // member including the header, so the size of any name following the header 128 // is checked to make sure it does not overflow. 129 Expected<StringRef> ArchiveMemberHeader::getName(uint64_t Size) const { 130 131 // This can be called from the ArchiveMemberHeader constructor when the 132 // archive header is truncated to produce an error message with the name. 133 // Make sure the name field is not truncated. 134 if (Size < offsetof(ArMemHdrType, Name) + sizeof(ArMemHdr->Name)) { 135 uint64_t ArchiveOffset = reinterpret_cast<const char *>(ArMemHdr) - 136 Parent->getData().data(); 137 return malformedError("archive header truncated before the name field " 138 "for archive member header at offset " + 139 Twine(ArchiveOffset)); 140 } 141 142 // The raw name itself can be invalid. 143 Expected<StringRef> NameOrErr = getRawName(); 144 if (!NameOrErr) 145 return NameOrErr.takeError(); 146 StringRef Name = NameOrErr.get(); 147 148 // Check if it's a special name. 149 if (Name[0] == '/') { 150 if (Name.size() == 1) // Linker member. 151 return Name; 152 if (Name.size() == 2 && Name[1] == '/') // String table. 153 return Name; 154 // It's a long name. 155 // Get the string table offset. 156 std::size_t StringOffset; 157 if (Name.substr(1).rtrim(' ').getAsInteger(10, StringOffset)) { 158 std::string Buf; 159 raw_string_ostream OS(Buf); 160 OS.write_escaped(Name.substr(1).rtrim(' ')); 161 OS.flush(); 162 uint64_t ArchiveOffset = reinterpret_cast<const char *>(ArMemHdr) - 163 Parent->getData().data(); 164 return malformedError("long name offset characters after the '/' are " 165 "not all decimal numbers: '" + Buf + "' for " 166 "archive member header at offset " + 167 Twine(ArchiveOffset)); 168 } 169 170 // Verify it. 171 if (StringOffset >= Parent->getStringTable().size()) { 172 uint64_t ArchiveOffset = reinterpret_cast<const char *>(ArMemHdr) - 173 Parent->getData().data(); 174 return malformedError("long name offset " + Twine(StringOffset) + " past " 175 "the end of the string table for archive member " 176 "header at offset " + Twine(ArchiveOffset)); 177 } 178 179 // GNU long file names end with a "/\n". 180 if (Parent->kind() == Archive::K_GNU || 181 Parent->kind() == Archive::K_GNU64) { 182 size_t End = Parent->getStringTable().find('\n', /*From=*/StringOffset); 183 if (End == StringRef::npos || End < 1 || 184 Parent->getStringTable()[End - 1] != '/') { 185 return malformedError("string table at long name offset " + 186 Twine(StringOffset) + "not terminated"); 187 } 188 return Parent->getStringTable().slice(StringOffset, End - 1); 189 } 190 return Parent->getStringTable().begin() + StringOffset; 191 } 192 193 if (Name.startswith("#1/")) { 194 uint64_t NameLength; 195 if (Name.substr(3).rtrim(' ').getAsInteger(10, NameLength)) { 196 std::string Buf; 197 raw_string_ostream OS(Buf); 198 OS.write_escaped(Name.substr(3).rtrim(' ')); 199 OS.flush(); 200 uint64_t ArchiveOffset = reinterpret_cast<const char *>(ArMemHdr) - 201 Parent->getData().data(); 202 return malformedError("long name length characters after the #1/ are " 203 "not all decimal numbers: '" + Buf + "' for " 204 "archive member header at offset " + 205 Twine(ArchiveOffset)); 206 } 207 if (getSizeOf() + NameLength > Size) { 208 uint64_t ArchiveOffset = reinterpret_cast<const char *>(ArMemHdr) - 209 Parent->getData().data(); 210 return malformedError("long name length: " + Twine(NameLength) + 211 " extends past the end of the member or archive " 212 "for archive member header at offset " + 213 Twine(ArchiveOffset)); 214 } 215 return StringRef(reinterpret_cast<const char *>(ArMemHdr) + getSizeOf(), 216 NameLength).rtrim('\0'); 217 } 218 219 // It is not a long name so trim the blanks at the end of the name. 220 if (Name[Name.size() - 1] != '/') 221 return Name.rtrim(' '); 222 223 // It's a simple name. 224 return Name.drop_back(1); 225 } 226 227 Expected<uint32_t> ArchiveMemberHeader::getSize() const { 228 uint32_t Ret; 229 if (StringRef(ArMemHdr->Size, 230 sizeof(ArMemHdr->Size)).rtrim(" ").getAsInteger(10, Ret)) { 231 std::string Buf; 232 raw_string_ostream OS(Buf); 233 OS.write_escaped(StringRef(ArMemHdr->Size, 234 sizeof(ArMemHdr->Size)).rtrim(" ")); 235 OS.flush(); 236 uint64_t Offset = reinterpret_cast<const char *>(ArMemHdr) - 237 Parent->getData().data(); 238 return malformedError("characters in size field in archive header are not " 239 "all decimal numbers: '" + Buf + "' for archive " 240 "member header at offset " + Twine(Offset)); 241 } 242 return Ret; 243 } 244 245 Expected<sys::fs::perms> ArchiveMemberHeader::getAccessMode() const { 246 unsigned Ret; 247 if (StringRef(ArMemHdr->AccessMode, 248 sizeof(ArMemHdr->AccessMode)).rtrim(' ').getAsInteger(8, Ret)) { 249 std::string Buf; 250 raw_string_ostream OS(Buf); 251 OS.write_escaped(StringRef(ArMemHdr->AccessMode, 252 sizeof(ArMemHdr->AccessMode)).rtrim(" ")); 253 OS.flush(); 254 uint64_t Offset = reinterpret_cast<const char *>(ArMemHdr) - 255 Parent->getData().data(); 256 return malformedError("characters in AccessMode field in archive header " 257 "are not all decimal numbers: '" + Buf + "' for the " 258 "archive member header at offset " + Twine(Offset)); 259 } 260 return static_cast<sys::fs::perms>(Ret); 261 } 262 263 Expected<sys::TimePoint<std::chrono::seconds>> 264 ArchiveMemberHeader::getLastModified() const { 265 unsigned Seconds; 266 if (StringRef(ArMemHdr->LastModified, 267 sizeof(ArMemHdr->LastModified)).rtrim(' ') 268 .getAsInteger(10, Seconds)) { 269 std::string Buf; 270 raw_string_ostream OS(Buf); 271 OS.write_escaped(StringRef(ArMemHdr->LastModified, 272 sizeof(ArMemHdr->LastModified)).rtrim(" ")); 273 OS.flush(); 274 uint64_t Offset = reinterpret_cast<const char *>(ArMemHdr) - 275 Parent->getData().data(); 276 return malformedError("characters in LastModified field in archive header " 277 "are not all decimal numbers: '" + Buf + "' for the " 278 "archive member header at offset " + Twine(Offset)); 279 } 280 281 return sys::toTimePoint(Seconds); 282 } 283 284 Expected<unsigned> ArchiveMemberHeader::getUID() const { 285 unsigned Ret; 286 StringRef User = StringRef(ArMemHdr->UID, sizeof(ArMemHdr->UID)).rtrim(' '); 287 if (User.empty()) 288 return 0; 289 if (User.getAsInteger(10, Ret)) { 290 std::string Buf; 291 raw_string_ostream OS(Buf); 292 OS.write_escaped(User); 293 OS.flush(); 294 uint64_t Offset = reinterpret_cast<const char *>(ArMemHdr) - 295 Parent->getData().data(); 296 return malformedError("characters in UID field in archive header " 297 "are not all decimal numbers: '" + Buf + "' for the " 298 "archive member header at offset " + Twine(Offset)); 299 } 300 return Ret; 301 } 302 303 Expected<unsigned> ArchiveMemberHeader::getGID() const { 304 unsigned Ret; 305 StringRef Group = StringRef(ArMemHdr->GID, sizeof(ArMemHdr->GID)).rtrim(' '); 306 if (Group.empty()) 307 return 0; 308 if (Group.getAsInteger(10, Ret)) { 309 std::string Buf; 310 raw_string_ostream OS(Buf); 311 OS.write_escaped(Group); 312 OS.flush(); 313 uint64_t Offset = reinterpret_cast<const char *>(ArMemHdr) - 314 Parent->getData().data(); 315 return malformedError("characters in GID field in archive header " 316 "are not all decimal numbers: '" + Buf + "' for the " 317 "archive member header at offset " + Twine(Offset)); 318 } 319 return Ret; 320 } 321 322 Archive::Child::Child(const Archive *Parent, StringRef Data, 323 uint16_t StartOfFile) 324 : Parent(Parent), Header(Parent, Data.data(), Data.size(), nullptr), 325 Data(Data), StartOfFile(StartOfFile) { 326 } 327 328 Archive::Child::Child(const Archive *Parent, const char *Start, Error *Err) 329 : Parent(Parent), 330 Header(Parent, Start, 331 Parent 332 ? Parent->getData().size() - (Start - Parent->getData().data()) 333 : 0, Err) { 334 if (!Start) 335 return; 336 337 // If we are pointed to real data, Start is not a nullptr, then there must be 338 // a non-null Err pointer available to report malformed data on. Only in 339 // the case sentinel value is being constructed is Err is permitted to be a 340 // nullptr. 341 assert(Err && "Err can't be nullptr if Start is not a nullptr"); 342 343 ErrorAsOutParameter ErrAsOutParam(Err); 344 345 // If there was an error in the construction of the Header 346 // then just return with the error now set. 347 if (*Err) 348 return; 349 350 uint64_t Size = Header.getSizeOf(); 351 Data = StringRef(Start, Size); 352 Expected<bool> isThinOrErr = isThinMember(); 353 if (!isThinOrErr) { 354 *Err = isThinOrErr.takeError(); 355 return; 356 } 357 bool isThin = isThinOrErr.get(); 358 if (!isThin) { 359 Expected<uint64_t> MemberSize = getRawSize(); 360 if (!MemberSize) { 361 *Err = MemberSize.takeError(); 362 return; 363 } 364 Size += MemberSize.get(); 365 Data = StringRef(Start, Size); 366 } 367 368 // Setup StartOfFile and PaddingBytes. 369 StartOfFile = Header.getSizeOf(); 370 // Don't include attached name. 371 Expected<StringRef> NameOrErr = getRawName(); 372 if (!NameOrErr){ 373 *Err = NameOrErr.takeError(); 374 return; 375 } 376 StringRef Name = NameOrErr.get(); 377 if (Name.startswith("#1/")) { 378 uint64_t NameSize; 379 if (Name.substr(3).rtrim(' ').getAsInteger(10, NameSize)) { 380 std::string Buf; 381 raw_string_ostream OS(Buf); 382 OS.write_escaped(Name.substr(3).rtrim(' ')); 383 OS.flush(); 384 uint64_t Offset = Start - Parent->getData().data(); 385 *Err = malformedError("long name length characters after the #1/ are " 386 "not all decimal numbers: '" + Buf + "' for " 387 "archive member header at offset " + 388 Twine(Offset)); 389 return; 390 } 391 StartOfFile += NameSize; 392 } 393 } 394 395 Expected<uint64_t> Archive::Child::getSize() const { 396 if (Parent->IsThin) { 397 Expected<uint32_t> Size = Header.getSize(); 398 if (!Size) 399 return Size.takeError(); 400 return Size.get(); 401 } 402 return Data.size() - StartOfFile; 403 } 404 405 Expected<uint64_t> Archive::Child::getRawSize() const { 406 return Header.getSize(); 407 } 408 409 Expected<bool> Archive::Child::isThinMember() const { 410 Expected<StringRef> NameOrErr = Header.getRawName(); 411 if (!NameOrErr) 412 return NameOrErr.takeError(); 413 StringRef Name = NameOrErr.get(); 414 return Parent->IsThin && Name != "/" && Name != "//"; 415 } 416 417 Expected<std::string> Archive::Child::getFullName() const { 418 Expected<bool> isThin = isThinMember(); 419 if (!isThin) 420 return isThin.takeError(); 421 assert(isThin.get()); 422 Expected<StringRef> NameOrErr = getName(); 423 if (!NameOrErr) 424 return NameOrErr.takeError(); 425 StringRef Name = *NameOrErr; 426 if (sys::path::is_absolute(Name)) 427 return Name; 428 429 SmallString<128> FullName = sys::path::parent_path( 430 Parent->getMemoryBufferRef().getBufferIdentifier()); 431 sys::path::append(FullName, Name); 432 return StringRef(FullName); 433 } 434 435 Expected<StringRef> Archive::Child::getBuffer() const { 436 Expected<bool> isThinOrErr = isThinMember(); 437 if (!isThinOrErr) 438 return isThinOrErr.takeError(); 439 bool isThin = isThinOrErr.get(); 440 if (!isThin) { 441 Expected<uint32_t> Size = getSize(); 442 if (!Size) 443 return Size.takeError(); 444 return StringRef(Data.data() + StartOfFile, Size.get()); 445 } 446 Expected<std::string> FullNameOrErr = getFullName(); 447 if (!FullNameOrErr) 448 return FullNameOrErr.takeError(); 449 const std::string &FullName = *FullNameOrErr; 450 ErrorOr<std::unique_ptr<MemoryBuffer>> Buf = MemoryBuffer::getFile(FullName); 451 if (std::error_code EC = Buf.getError()) 452 return errorCodeToError(EC); 453 Parent->ThinBuffers.push_back(std::move(*Buf)); 454 return Parent->ThinBuffers.back()->getBuffer(); 455 } 456 457 Expected<Archive::Child> Archive::Child::getNext() const { 458 size_t SpaceToSkip = Data.size(); 459 // If it's odd, add 1 to make it even. 460 if (SpaceToSkip & 1) 461 ++SpaceToSkip; 462 463 const char *NextLoc = Data.data() + SpaceToSkip; 464 465 // Check to see if this is at the end of the archive. 466 if (NextLoc == Parent->Data.getBufferEnd()) 467 return Child(nullptr, nullptr, nullptr); 468 469 // Check to see if this is past the end of the archive. 470 if (NextLoc > Parent->Data.getBufferEnd()) { 471 std::string Msg("offset to next archive member past the end of the archive " 472 "after member "); 473 Expected<StringRef> NameOrErr = getName(); 474 if (!NameOrErr) { 475 consumeError(NameOrErr.takeError()); 476 uint64_t Offset = Data.data() - Parent->getData().data(); 477 return malformedError(Msg + "at offset " + Twine(Offset)); 478 } else 479 return malformedError(Msg + NameOrErr.get()); 480 } 481 482 Error Err = Error::success(); 483 Child Ret(Parent, NextLoc, &Err); 484 if (Err) 485 return std::move(Err); 486 return Ret; 487 } 488 489 uint64_t Archive::Child::getChildOffset() const { 490 const char *a = Parent->Data.getBuffer().data(); 491 const char *c = Data.data(); 492 uint64_t offset = c - a; 493 return offset; 494 } 495 496 Expected<StringRef> Archive::Child::getName() const { 497 Expected<uint64_t> RawSizeOrErr = getRawSize(); 498 if (!RawSizeOrErr) 499 return RawSizeOrErr.takeError(); 500 uint64_t RawSize = RawSizeOrErr.get(); 501 Expected<StringRef> NameOrErr = Header.getName(Header.getSizeOf() + RawSize); 502 if (!NameOrErr) 503 return NameOrErr.takeError(); 504 StringRef Name = NameOrErr.get(); 505 return Name; 506 } 507 508 Expected<MemoryBufferRef> Archive::Child::getMemoryBufferRef() const { 509 Expected<StringRef> NameOrErr = getName(); 510 if (!NameOrErr) 511 return NameOrErr.takeError(); 512 StringRef Name = NameOrErr.get(); 513 Expected<StringRef> Buf = getBuffer(); 514 if (!Buf) 515 return Buf.takeError(); 516 return MemoryBufferRef(*Buf, Name); 517 } 518 519 Expected<std::unique_ptr<Binary>> 520 Archive::Child::getAsBinary(LLVMContext *Context) const { 521 Expected<MemoryBufferRef> BuffOrErr = getMemoryBufferRef(); 522 if (!BuffOrErr) 523 return BuffOrErr.takeError(); 524 525 auto BinaryOrErr = createBinary(BuffOrErr.get(), Context); 526 if (BinaryOrErr) 527 return std::move(*BinaryOrErr); 528 return BinaryOrErr.takeError(); 529 } 530 531 Expected<std::unique_ptr<Archive>> Archive::create(MemoryBufferRef Source) { 532 Error Err = Error::success(); 533 std::unique_ptr<Archive> Ret(new Archive(Source, Err)); 534 if (Err) 535 return std::move(Err); 536 return std::move(Ret); 537 } 538 539 void Archive::setFirstRegular(const Child &C) { 540 FirstRegularData = C.Data; 541 FirstRegularStartOfFile = C.StartOfFile; 542 } 543 544 Archive::Archive(MemoryBufferRef Source, Error &Err) 545 : Binary(Binary::ID_Archive, Source) { 546 ErrorAsOutParameter ErrAsOutParam(&Err); 547 StringRef Buffer = Data.getBuffer(); 548 // Check for sufficient magic. 549 if (Buffer.startswith(ThinMagic)) { 550 IsThin = true; 551 } else if (Buffer.startswith(Magic)) { 552 IsThin = false; 553 } else { 554 Err = make_error<GenericBinaryError>("File too small to be an archive", 555 object_error::invalid_file_type); 556 return; 557 } 558 559 // Make sure Format is initialized before any call to 560 // ArchiveMemberHeader::getName() is made. This could be a valid empty 561 // archive which is the same in all formats. So claiming it to be gnu to is 562 // fine if not totally correct before we look for a string table or table of 563 // contents. 564 Format = K_GNU; 565 566 // Get the special members. 567 child_iterator I = child_begin(Err, false); 568 if (Err) 569 return; 570 child_iterator E = child_end(); 571 572 // See if this is a valid empty archive and if so return. 573 if (I == E) { 574 Err = Error::success(); 575 return; 576 } 577 const Child *C = &*I; 578 579 auto Increment = [&]() { 580 ++I; 581 if (Err) 582 return true; 583 C = &*I; 584 return false; 585 }; 586 587 Expected<StringRef> NameOrErr = C->getRawName(); 588 if (!NameOrErr) { 589 Err = NameOrErr.takeError(); 590 return; 591 } 592 StringRef Name = NameOrErr.get(); 593 594 // Below is the pattern that is used to figure out the archive format 595 // GNU archive format 596 // First member : / (may exist, if it exists, points to the symbol table ) 597 // Second member : // (may exist, if it exists, points to the string table) 598 // Note : The string table is used if the filename exceeds 15 characters 599 // BSD archive format 600 // First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table) 601 // There is no string table, if the filename exceeds 15 characters or has a 602 // embedded space, the filename has #1/<size>, The size represents the size 603 // of the filename that needs to be read after the archive header 604 // COFF archive format 605 // First member : / 606 // Second member : / (provides a directory of symbols) 607 // Third member : // (may exist, if it exists, contains the string table) 608 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present 609 // even if the string table is empty. However, lib.exe does not in fact 610 // seem to create the third member if there's no member whose filename 611 // exceeds 15 characters. So the third member is optional. 612 613 if (Name == "__.SYMDEF" || Name == "__.SYMDEF_64") { 614 if (Name == "__.SYMDEF") 615 Format = K_BSD; 616 else // Name == "__.SYMDEF_64" 617 Format = K_DARWIN64; 618 // We know that the symbol table is not an external file, but we still must 619 // check any Expected<> return value. 620 Expected<StringRef> BufOrErr = C->getBuffer(); 621 if (!BufOrErr) { 622 Err = BufOrErr.takeError(); 623 return; 624 } 625 SymbolTable = BufOrErr.get(); 626 if (Increment()) 627 return; 628 setFirstRegular(*C); 629 630 Err = Error::success(); 631 return; 632 } 633 634 if (Name.startswith("#1/")) { 635 Format = K_BSD; 636 // We know this is BSD, so getName will work since there is no string table. 637 Expected<StringRef> NameOrErr = C->getName(); 638 if (!NameOrErr) { 639 Err = NameOrErr.takeError(); 640 return; 641 } 642 Name = NameOrErr.get(); 643 if (Name == "__.SYMDEF SORTED" || Name == "__.SYMDEF") { 644 // We know that the symbol table is not an external file, but we still 645 // must check any Expected<> return value. 646 Expected<StringRef> BufOrErr = C->getBuffer(); 647 if (!BufOrErr) { 648 Err = BufOrErr.takeError(); 649 return; 650 } 651 SymbolTable = BufOrErr.get(); 652 if (Increment()) 653 return; 654 } 655 else if (Name == "__.SYMDEF_64 SORTED" || Name == "__.SYMDEF_64") { 656 Format = K_DARWIN64; 657 // We know that the symbol table is not an external file, but we still 658 // must check any Expected<> return value. 659 Expected<StringRef> BufOrErr = C->getBuffer(); 660 if (!BufOrErr) { 661 Err = BufOrErr.takeError(); 662 return; 663 } 664 SymbolTable = BufOrErr.get(); 665 if (Increment()) 666 return; 667 } 668 setFirstRegular(*C); 669 return; 670 } 671 672 // MIPS 64-bit ELF archives use a special format of a symbol table. 673 // This format is marked by `ar_name` field equals to "/SYM64/". 674 // For detailed description see page 96 in the following document: 675 // http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf 676 677 bool has64SymTable = false; 678 if (Name == "/" || Name == "/SYM64/") { 679 // We know that the symbol table is not an external file, but we still 680 // must check any Expected<> return value. 681 Expected<StringRef> BufOrErr = C->getBuffer(); 682 if (!BufOrErr) { 683 Err = BufOrErr.takeError(); 684 return; 685 } 686 SymbolTable = BufOrErr.get(); 687 if (Name == "/SYM64/") 688 has64SymTable = true; 689 690 if (Increment()) 691 return; 692 if (I == E) { 693 Err = Error::success(); 694 return; 695 } 696 Expected<StringRef> NameOrErr = C->getRawName(); 697 if (!NameOrErr) { 698 Err = NameOrErr.takeError(); 699 return; 700 } 701 Name = NameOrErr.get(); 702 } 703 704 if (Name == "//") { 705 Format = has64SymTable ? K_GNU64 : K_GNU; 706 // The string table is never an external member, but we still 707 // must check any Expected<> return value. 708 Expected<StringRef> BufOrErr = C->getBuffer(); 709 if (!BufOrErr) { 710 Err = BufOrErr.takeError(); 711 return; 712 } 713 StringTable = BufOrErr.get(); 714 if (Increment()) 715 return; 716 setFirstRegular(*C); 717 Err = Error::success(); 718 return; 719 } 720 721 if (Name[0] != '/') { 722 Format = has64SymTable ? K_GNU64 : K_GNU; 723 setFirstRegular(*C); 724 Err = Error::success(); 725 return; 726 } 727 728 if (Name != "/") { 729 Err = errorCodeToError(object_error::parse_failed); 730 return; 731 } 732 733 Format = K_COFF; 734 // We know that the symbol table is not an external file, but we still 735 // must check any Expected<> return value. 736 Expected<StringRef> BufOrErr = C->getBuffer(); 737 if (!BufOrErr) { 738 Err = BufOrErr.takeError(); 739 return; 740 } 741 SymbolTable = BufOrErr.get(); 742 743 if (Increment()) 744 return; 745 746 if (I == E) { 747 setFirstRegular(*C); 748 Err = Error::success(); 749 return; 750 } 751 752 NameOrErr = C->getRawName(); 753 if (!NameOrErr) { 754 Err = NameOrErr.takeError(); 755 return; 756 } 757 Name = NameOrErr.get(); 758 759 if (Name == "//") { 760 // The string table is never an external member, but we still 761 // must check any Expected<> return value. 762 Expected<StringRef> BufOrErr = C->getBuffer(); 763 if (!BufOrErr) { 764 Err = BufOrErr.takeError(); 765 return; 766 } 767 StringTable = BufOrErr.get(); 768 if (Increment()) 769 return; 770 } 771 772 setFirstRegular(*C); 773 Err = Error::success(); 774 } 775 776 Archive::child_iterator Archive::child_begin(Error &Err, 777 bool SkipInternal) const { 778 if (isEmpty()) 779 return child_end(); 780 781 if (SkipInternal) 782 return child_iterator(Child(this, FirstRegularData, 783 FirstRegularStartOfFile), 784 &Err); 785 786 const char *Loc = Data.getBufferStart() + strlen(Magic); 787 Child C(this, Loc, &Err); 788 if (Err) 789 return child_end(); 790 return child_iterator(C, &Err); 791 } 792 793 Archive::child_iterator Archive::child_end() const { 794 return child_iterator(Child(nullptr, nullptr, nullptr), nullptr); 795 } 796 797 StringRef Archive::Symbol::getName() const { 798 return Parent->getSymbolTable().begin() + StringIndex; 799 } 800 801 Expected<Archive::Child> Archive::Symbol::getMember() const { 802 const char *Buf = Parent->getSymbolTable().begin(); 803 const char *Offsets = Buf; 804 if (Parent->kind() == K_GNU64 || Parent->kind() == K_DARWIN64) 805 Offsets += sizeof(uint64_t); 806 else 807 Offsets += sizeof(uint32_t); 808 uint64_t Offset = 0; 809 if (Parent->kind() == K_GNU) { 810 Offset = read32be(Offsets + SymbolIndex * 4); 811 } else if (Parent->kind() == K_GNU64) { 812 Offset = read64be(Offsets + SymbolIndex * 8); 813 } else if (Parent->kind() == K_BSD) { 814 // The SymbolIndex is an index into the ranlib structs that start at 815 // Offsets (the first uint32_t is the number of bytes of the ranlib 816 // structs). The ranlib structs are a pair of uint32_t's the first 817 // being a string table offset and the second being the offset into 818 // the archive of the member that defines the symbol. Which is what 819 // is needed here. 820 Offset = read32le(Offsets + SymbolIndex * 8 + 4); 821 } else if (Parent->kind() == K_DARWIN64) { 822 // The SymbolIndex is an index into the ranlib_64 structs that start at 823 // Offsets (the first uint64_t is the number of bytes of the ranlib_64 824 // structs). The ranlib_64 structs are a pair of uint64_t's the first 825 // being a string table offset and the second being the offset into 826 // the archive of the member that defines the symbol. Which is what 827 // is needed here. 828 Offset = read64le(Offsets + SymbolIndex * 16 + 8); 829 } else { 830 // Skip offsets. 831 uint32_t MemberCount = read32le(Buf); 832 Buf += MemberCount * 4 + 4; 833 834 uint32_t SymbolCount = read32le(Buf); 835 if (SymbolIndex >= SymbolCount) 836 return errorCodeToError(object_error::parse_failed); 837 838 // Skip SymbolCount to get to the indices table. 839 const char *Indices = Buf + 4; 840 841 // Get the index of the offset in the file member offset table for this 842 // symbol. 843 uint16_t OffsetIndex = read16le(Indices + SymbolIndex * 2); 844 // Subtract 1 since OffsetIndex is 1 based. 845 --OffsetIndex; 846 847 if (OffsetIndex >= MemberCount) 848 return errorCodeToError(object_error::parse_failed); 849 850 Offset = read32le(Offsets + OffsetIndex * 4); 851 } 852 853 const char *Loc = Parent->getData().begin() + Offset; 854 Error Err = Error::success(); 855 Child C(Parent, Loc, &Err); 856 if (Err) 857 return std::move(Err); 858 return C; 859 } 860 861 Archive::Symbol Archive::Symbol::getNext() const { 862 Symbol t(*this); 863 if (Parent->kind() == K_BSD) { 864 // t.StringIndex is an offset from the start of the __.SYMDEF or 865 // "__.SYMDEF SORTED" member into the string table for the ranlib 866 // struct indexed by t.SymbolIndex . To change t.StringIndex to the 867 // offset in the string table for t.SymbolIndex+1 we subtract the 868 // its offset from the start of the string table for t.SymbolIndex 869 // and add the offset of the string table for t.SymbolIndex+1. 870 871 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t 872 // which is the number of bytes of ranlib structs that follow. The ranlib 873 // structs are a pair of uint32_t's the first being a string table offset 874 // and the second being the offset into the archive of the member that 875 // define the symbol. After that the next uint32_t is the byte count of 876 // the string table followed by the string table. 877 const char *Buf = Parent->getSymbolTable().begin(); 878 uint32_t RanlibCount = 0; 879 RanlibCount = read32le(Buf) / 8; 880 // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount) 881 // don't change the t.StringIndex as we don't want to reference a ranlib 882 // past RanlibCount. 883 if (t.SymbolIndex + 1 < RanlibCount) { 884 const char *Ranlibs = Buf + 4; 885 uint32_t CurRanStrx = 0; 886 uint32_t NextRanStrx = 0; 887 CurRanStrx = read32le(Ranlibs + t.SymbolIndex * 8); 888 NextRanStrx = read32le(Ranlibs + (t.SymbolIndex + 1) * 8); 889 t.StringIndex -= CurRanStrx; 890 t.StringIndex += NextRanStrx; 891 } 892 } else { 893 // Go to one past next null. 894 t.StringIndex = Parent->getSymbolTable().find('\0', t.StringIndex) + 1; 895 } 896 ++t.SymbolIndex; 897 return t; 898 } 899 900 Archive::symbol_iterator Archive::symbol_begin() const { 901 if (!hasSymbolTable()) 902 return symbol_iterator(Symbol(this, 0, 0)); 903 904 const char *buf = getSymbolTable().begin(); 905 if (kind() == K_GNU) { 906 uint32_t symbol_count = 0; 907 symbol_count = read32be(buf); 908 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t))); 909 } else if (kind() == K_GNU64) { 910 uint64_t symbol_count = read64be(buf); 911 buf += sizeof(uint64_t) + (symbol_count * (sizeof(uint64_t))); 912 } else if (kind() == K_BSD) { 913 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t 914 // which is the number of bytes of ranlib structs that follow. The ranlib 915 // structs are a pair of uint32_t's the first being a string table offset 916 // and the second being the offset into the archive of the member that 917 // define the symbol. After that the next uint32_t is the byte count of 918 // the string table followed by the string table. 919 uint32_t ranlib_count = 0; 920 ranlib_count = read32le(buf) / 8; 921 const char *ranlibs = buf + 4; 922 uint32_t ran_strx = 0; 923 ran_strx = read32le(ranlibs); 924 buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t)))); 925 // Skip the byte count of the string table. 926 buf += sizeof(uint32_t); 927 buf += ran_strx; 928 } else if (kind() == K_DARWIN64) { 929 // The __.SYMDEF_64 or "__.SYMDEF_64 SORTED" member starts with a uint64_t 930 // which is the number of bytes of ranlib_64 structs that follow. The 931 // ranlib_64 structs are a pair of uint64_t's the first being a string 932 // table offset and the second being the offset into the archive of the 933 // member that define the symbol. After that the next uint64_t is the byte 934 // count of the string table followed by the string table. 935 uint64_t ranlib_count = 0; 936 ranlib_count = read64le(buf) / 16; 937 const char *ranlibs = buf + 8; 938 uint64_t ran_strx = 0; 939 ran_strx = read64le(ranlibs); 940 buf += sizeof(uint64_t) + (ranlib_count * (2 * (sizeof(uint64_t)))); 941 // Skip the byte count of the string table. 942 buf += sizeof(uint64_t); 943 buf += ran_strx; 944 } else { 945 uint32_t member_count = 0; 946 uint32_t symbol_count = 0; 947 member_count = read32le(buf); 948 buf += 4 + (member_count * 4); // Skip offsets. 949 symbol_count = read32le(buf); 950 buf += 4 + (symbol_count * 2); // Skip indices. 951 } 952 uint32_t string_start_offset = buf - getSymbolTable().begin(); 953 return symbol_iterator(Symbol(this, 0, string_start_offset)); 954 } 955 956 Archive::symbol_iterator Archive::symbol_end() const { 957 return symbol_iterator(Symbol(this, getNumberOfSymbols(), 0)); 958 } 959 960 uint32_t Archive::getNumberOfSymbols() const { 961 if (!hasSymbolTable()) 962 return 0; 963 const char *buf = getSymbolTable().begin(); 964 if (kind() == K_GNU) 965 return read32be(buf); 966 if (kind() == K_GNU64) 967 return read64be(buf); 968 if (kind() == K_BSD) 969 return read32le(buf) / 8; 970 if (kind() == K_DARWIN64) 971 return read64le(buf) / 16; 972 uint32_t member_count = 0; 973 member_count = read32le(buf); 974 buf += 4 + (member_count * 4); // Skip offsets. 975 return read32le(buf); 976 } 977 978 Expected<Optional<Archive::Child>> Archive::findSym(StringRef name) const { 979 Archive::symbol_iterator bs = symbol_begin(); 980 Archive::symbol_iterator es = symbol_end(); 981 982 for (; bs != es; ++bs) { 983 StringRef SymName = bs->getName(); 984 if (SymName == name) { 985 if (auto MemberOrErr = bs->getMember()) 986 return Child(*MemberOrErr); 987 else 988 return MemberOrErr.takeError(); 989 } 990 } 991 return Optional<Child>(); 992 } 993 994 // Returns true if archive file contains no member file. 995 bool Archive::isEmpty() const { return Data.getBufferSize() == 8; } 996 997 bool Archive::hasSymbolTable() const { return !SymbolTable.empty(); } 998