1 //===- ArchiveWriter.cpp - ar File Format implementation --------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file defines the writeArchive function. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "llvm/Object/ArchiveWriter.h" 14 #include "llvm/ADT/ArrayRef.h" 15 #include "llvm/ADT/StringMap.h" 16 #include "llvm/ADT/StringRef.h" 17 #include "llvm/BinaryFormat/Magic.h" 18 #include "llvm/IR/LLVMContext.h" 19 #include "llvm/Object/Archive.h" 20 #include "llvm/Object/Error.h" 21 #include "llvm/Object/ObjectFile.h" 22 #include "llvm/Object/SymbolicFile.h" 23 #include "llvm/Support/Alignment.h" 24 #include "llvm/Support/EndianStream.h" 25 #include "llvm/Support/Errc.h" 26 #include "llvm/Support/ErrorHandling.h" 27 #include "llvm/Support/Format.h" 28 #include "llvm/Support/Path.h" 29 #include "llvm/Support/SmallVectorMemoryBuffer.h" 30 #include "llvm/Support/raw_ostream.h" 31 32 #include <map> 33 34 #if !defined(_MSC_VER) && !defined(__MINGW32__) 35 #include <unistd.h> 36 #else 37 #include <io.h> 38 #endif 39 40 using namespace llvm; 41 42 NewArchiveMember::NewArchiveMember(MemoryBufferRef BufRef) 43 : Buf(MemoryBuffer::getMemBuffer(BufRef, false)), 44 MemberName(BufRef.getBufferIdentifier()) {} 45 46 Expected<NewArchiveMember> 47 NewArchiveMember::getOldMember(const object::Archive::Child &OldMember, 48 bool Deterministic) { 49 Expected<llvm::MemoryBufferRef> BufOrErr = OldMember.getMemoryBufferRef(); 50 if (!BufOrErr) 51 return BufOrErr.takeError(); 52 53 NewArchiveMember M; 54 M.Buf = MemoryBuffer::getMemBuffer(*BufOrErr, false); 55 M.MemberName = M.Buf->getBufferIdentifier(); 56 if (!Deterministic) { 57 auto ModTimeOrErr = OldMember.getLastModified(); 58 if (!ModTimeOrErr) 59 return ModTimeOrErr.takeError(); 60 M.ModTime = ModTimeOrErr.get(); 61 Expected<unsigned> UIDOrErr = OldMember.getUID(); 62 if (!UIDOrErr) 63 return UIDOrErr.takeError(); 64 M.UID = UIDOrErr.get(); 65 Expected<unsigned> GIDOrErr = OldMember.getGID(); 66 if (!GIDOrErr) 67 return GIDOrErr.takeError(); 68 M.GID = GIDOrErr.get(); 69 Expected<sys::fs::perms> AccessModeOrErr = OldMember.getAccessMode(); 70 if (!AccessModeOrErr) 71 return AccessModeOrErr.takeError(); 72 M.Perms = AccessModeOrErr.get(); 73 } 74 return std::move(M); 75 } 76 77 Expected<NewArchiveMember> NewArchiveMember::getFile(StringRef FileName, 78 bool Deterministic) { 79 sys::fs::file_status Status; 80 auto FDOrErr = sys::fs::openNativeFileForRead(FileName); 81 if (!FDOrErr) 82 return FDOrErr.takeError(); 83 sys::fs::file_t FD = *FDOrErr; 84 assert(FD != sys::fs::kInvalidFile); 85 86 if (auto EC = sys::fs::status(FD, Status)) 87 return errorCodeToError(EC); 88 89 // Opening a directory doesn't make sense. Let it fail. 90 // Linux cannot open directories with open(2), although 91 // cygwin and *bsd can. 92 if (Status.type() == sys::fs::file_type::directory_file) 93 return errorCodeToError(make_error_code(errc::is_a_directory)); 94 95 ErrorOr<std::unique_ptr<MemoryBuffer>> MemberBufferOrErr = 96 MemoryBuffer::getOpenFile(FD, FileName, Status.getSize(), false); 97 if (!MemberBufferOrErr) 98 return errorCodeToError(MemberBufferOrErr.getError()); 99 100 if (auto EC = sys::fs::closeFile(FD)) 101 return errorCodeToError(EC); 102 103 NewArchiveMember M; 104 M.Buf = std::move(*MemberBufferOrErr); 105 M.MemberName = M.Buf->getBufferIdentifier(); 106 if (!Deterministic) { 107 M.ModTime = std::chrono::time_point_cast<std::chrono::seconds>( 108 Status.getLastModificationTime()); 109 M.UID = Status.getUser(); 110 M.GID = Status.getGroup(); 111 M.Perms = Status.permissions(); 112 } 113 return std::move(M); 114 } 115 116 template <typename T> 117 static void printWithSpacePadding(raw_ostream &OS, T Data, unsigned Size) { 118 uint64_t OldPos = OS.tell(); 119 OS << Data; 120 unsigned SizeSoFar = OS.tell() - OldPos; 121 assert(SizeSoFar <= Size && "Data doesn't fit in Size"); 122 OS.indent(Size - SizeSoFar); 123 } 124 125 static bool isDarwin(object::Archive::Kind Kind) { 126 return Kind == object::Archive::K_DARWIN || 127 Kind == object::Archive::K_DARWIN64; 128 } 129 130 static bool isBSDLike(object::Archive::Kind Kind) { 131 switch (Kind) { 132 case object::Archive::K_GNU: 133 case object::Archive::K_GNU64: 134 return false; 135 case object::Archive::K_BSD: 136 case object::Archive::K_DARWIN: 137 case object::Archive::K_DARWIN64: 138 return true; 139 case object::Archive::K_AIXBIG: 140 case object::Archive::K_COFF: 141 break; 142 } 143 llvm_unreachable("not supported for writting"); 144 } 145 146 template <class T> 147 static void print(raw_ostream &Out, object::Archive::Kind Kind, T Val) { 148 support::endian::write(Out, Val, 149 isBSDLike(Kind) ? support::little : support::big); 150 } 151 152 static void printRestOfMemberHeader( 153 raw_ostream &Out, const sys::TimePoint<std::chrono::seconds> &ModTime, 154 unsigned UID, unsigned GID, unsigned Perms, uint64_t Size) { 155 printWithSpacePadding(Out, sys::toTimeT(ModTime), 12); 156 157 // The format has only 6 chars for uid and gid. Truncate if the provided 158 // values don't fit. 159 printWithSpacePadding(Out, UID % 1000000, 6); 160 printWithSpacePadding(Out, GID % 1000000, 6); 161 162 printWithSpacePadding(Out, format("%o", Perms), 8); 163 printWithSpacePadding(Out, Size, 10); 164 Out << "`\n"; 165 } 166 167 static void 168 printGNUSmallMemberHeader(raw_ostream &Out, StringRef Name, 169 const sys::TimePoint<std::chrono::seconds> &ModTime, 170 unsigned UID, unsigned GID, unsigned Perms, 171 uint64_t Size) { 172 printWithSpacePadding(Out, Twine(Name) + "/", 16); 173 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms, Size); 174 } 175 176 static void 177 printBSDMemberHeader(raw_ostream &Out, uint64_t Pos, StringRef Name, 178 const sys::TimePoint<std::chrono::seconds> &ModTime, 179 unsigned UID, unsigned GID, unsigned Perms, uint64_t Size) { 180 uint64_t PosAfterHeader = Pos + 60 + Name.size(); 181 // Pad so that even 64 bit object files are aligned. 182 unsigned Pad = offsetToAlignment(PosAfterHeader, Align(8)); 183 unsigned NameWithPadding = Name.size() + Pad; 184 printWithSpacePadding(Out, Twine("#1/") + Twine(NameWithPadding), 16); 185 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms, 186 NameWithPadding + Size); 187 Out << Name; 188 while (Pad--) 189 Out.write(uint8_t(0)); 190 } 191 192 static bool useStringTable(bool Thin, StringRef Name) { 193 return Thin || Name.size() >= 16 || Name.contains('/'); 194 } 195 196 static bool is64BitKind(object::Archive::Kind Kind) { 197 switch (Kind) { 198 case object::Archive::K_GNU: 199 case object::Archive::K_BSD: 200 case object::Archive::K_DARWIN: 201 case object::Archive::K_COFF: 202 case object::Archive::K_AIXBIG: 203 return false; 204 case object::Archive::K_DARWIN64: 205 case object::Archive::K_GNU64: 206 return true; 207 } 208 llvm_unreachable("not supported for writting"); 209 } 210 211 static void 212 printMemberHeader(raw_ostream &Out, uint64_t Pos, raw_ostream &StringTable, 213 StringMap<uint64_t> &MemberNames, object::Archive::Kind Kind, 214 bool Thin, const NewArchiveMember &M, 215 sys::TimePoint<std::chrono::seconds> ModTime, uint64_t Size) { 216 if (isBSDLike(Kind)) 217 return printBSDMemberHeader(Out, Pos, M.MemberName, ModTime, M.UID, M.GID, 218 M.Perms, Size); 219 if (!useStringTable(Thin, M.MemberName)) 220 return printGNUSmallMemberHeader(Out, M.MemberName, ModTime, M.UID, M.GID, 221 M.Perms, Size); 222 Out << '/'; 223 uint64_t NamePos; 224 if (Thin) { 225 NamePos = StringTable.tell(); 226 StringTable << M.MemberName << "/\n"; 227 } else { 228 auto Insertion = MemberNames.insert({M.MemberName, uint64_t(0)}); 229 if (Insertion.second) { 230 Insertion.first->second = StringTable.tell(); 231 StringTable << M.MemberName << "/\n"; 232 } 233 NamePos = Insertion.first->second; 234 } 235 printWithSpacePadding(Out, NamePos, 15); 236 printRestOfMemberHeader(Out, ModTime, M.UID, M.GID, M.Perms, Size); 237 } 238 239 namespace { 240 struct MemberData { 241 std::vector<unsigned> Symbols; 242 std::string Header; 243 StringRef Data; 244 StringRef Padding; 245 }; 246 } // namespace 247 248 static MemberData computeStringTable(StringRef Names) { 249 unsigned Size = Names.size(); 250 unsigned Pad = offsetToAlignment(Size, Align(2)); 251 std::string Header; 252 raw_string_ostream Out(Header); 253 printWithSpacePadding(Out, "//", 48); 254 printWithSpacePadding(Out, Size + Pad, 10); 255 Out << "`\n"; 256 Out.flush(); 257 return {{}, std::move(Header), Names, Pad ? "\n" : ""}; 258 } 259 260 static sys::TimePoint<std::chrono::seconds> now(bool Deterministic) { 261 using namespace std::chrono; 262 263 if (!Deterministic) 264 return time_point_cast<seconds>(system_clock::now()); 265 return sys::TimePoint<seconds>(); 266 } 267 268 static bool isArchiveSymbol(const object::BasicSymbolRef &S) { 269 Expected<uint32_t> SymFlagsOrErr = S.getFlags(); 270 if (!SymFlagsOrErr) 271 // TODO: Actually report errors helpfully. 272 report_fatal_error(SymFlagsOrErr.takeError()); 273 if (*SymFlagsOrErr & object::SymbolRef::SF_FormatSpecific) 274 return false; 275 if (!(*SymFlagsOrErr & object::SymbolRef::SF_Global)) 276 return false; 277 if (*SymFlagsOrErr & object::SymbolRef::SF_Undefined) 278 return false; 279 return true; 280 } 281 282 static void printNBits(raw_ostream &Out, object::Archive::Kind Kind, 283 uint64_t Val) { 284 if (is64BitKind(Kind)) 285 print<uint64_t>(Out, Kind, Val); 286 else 287 print<uint32_t>(Out, Kind, Val); 288 } 289 290 static uint64_t computeSymbolTableSize(object::Archive::Kind Kind, 291 uint64_t NumSyms, uint64_t OffsetSize, 292 StringRef StringTable, 293 uint32_t *Padding = nullptr) { 294 assert((OffsetSize == 4 || OffsetSize == 8) && "Unsupported OffsetSize"); 295 uint64_t Size = OffsetSize; // Number of entries 296 if (isBSDLike(Kind)) 297 Size += NumSyms * OffsetSize * 2; // Table 298 else 299 Size += NumSyms * OffsetSize; // Table 300 if (isBSDLike(Kind)) 301 Size += OffsetSize; // byte count 302 Size += StringTable.size(); 303 // ld64 expects the members to be 8-byte aligned for 64-bit content and at 304 // least 4-byte aligned for 32-bit content. Opt for the larger encoding 305 // uniformly. 306 // We do this for all bsd formats because it simplifies aligning members. 307 uint32_t Pad = offsetToAlignment(Size, Align(isBSDLike(Kind) ? 8 : 2)); 308 Size += Pad; 309 if (Padding) 310 *Padding = Pad; 311 return Size; 312 } 313 314 static void writeSymbolTableHeader(raw_ostream &Out, object::Archive::Kind Kind, 315 bool Deterministic, uint64_t Size) { 316 if (isBSDLike(Kind)) { 317 const char *Name = is64BitKind(Kind) ? "__.SYMDEF_64" : "__.SYMDEF"; 318 printBSDMemberHeader(Out, Out.tell(), Name, now(Deterministic), 0, 0, 0, 319 Size); 320 } else { 321 const char *Name = is64BitKind(Kind) ? "/SYM64" : ""; 322 printGNUSmallMemberHeader(Out, Name, now(Deterministic), 0, 0, 0, Size); 323 } 324 } 325 326 static void writeSymbolTable(raw_ostream &Out, object::Archive::Kind Kind, 327 bool Deterministic, ArrayRef<MemberData> Members, 328 StringRef StringTable) { 329 // We don't write a symbol table on an archive with no members -- except on 330 // Darwin, where the linker will abort unless the archive has a symbol table. 331 if (StringTable.empty() && !isDarwin(Kind)) 332 return; 333 334 unsigned NumSyms = 0; 335 for (const MemberData &M : Members) 336 NumSyms += M.Symbols.size(); 337 338 uint64_t OffsetSize = is64BitKind(Kind) ? 8 : 4; 339 uint32_t Pad; 340 uint64_t Size = computeSymbolTableSize(Kind, NumSyms, OffsetSize, StringTable, &Pad); 341 writeSymbolTableHeader(Out, Kind, Deterministic, Size); 342 343 uint64_t Pos = Out.tell() + Size; 344 345 if (isBSDLike(Kind)) 346 printNBits(Out, Kind, NumSyms * 2 * OffsetSize); 347 else 348 printNBits(Out, Kind, NumSyms); 349 350 for (const MemberData &M : Members) { 351 for (unsigned StringOffset : M.Symbols) { 352 if (isBSDLike(Kind)) 353 printNBits(Out, Kind, StringOffset); 354 printNBits(Out, Kind, Pos); // member offset 355 } 356 Pos += M.Header.size() + M.Data.size() + M.Padding.size(); 357 } 358 359 if (isBSDLike(Kind)) 360 // byte count of the string table 361 printNBits(Out, Kind, StringTable.size()); 362 Out << StringTable; 363 364 while (Pad--) 365 Out.write(uint8_t(0)); 366 } 367 368 static Expected<std::vector<unsigned>> 369 getSymbols(MemoryBufferRef Buf, raw_ostream &SymNames, bool &HasObject) { 370 std::vector<unsigned> Ret; 371 372 // In the scenario when LLVMContext is populated SymbolicFile will contain a 373 // reference to it, thus SymbolicFile should be destroyed first. 374 LLVMContext Context; 375 std::unique_ptr<object::SymbolicFile> Obj; 376 377 const file_magic Type = identify_magic(Buf.getBuffer()); 378 // Treat unsupported file types as having no symbols. 379 if (!object::SymbolicFile::isSymbolicFile(Type, &Context)) 380 return Ret; 381 if (Type == file_magic::bitcode) { 382 auto ObjOrErr = object::SymbolicFile::createSymbolicFile( 383 Buf, file_magic::bitcode, &Context); 384 if (!ObjOrErr) 385 return ObjOrErr.takeError(); 386 Obj = std::move(*ObjOrErr); 387 } else { 388 auto ObjOrErr = object::SymbolicFile::createSymbolicFile(Buf); 389 if (!ObjOrErr) 390 return ObjOrErr.takeError(); 391 Obj = std::move(*ObjOrErr); 392 } 393 394 HasObject = true; 395 for (const object::BasicSymbolRef &S : Obj->symbols()) { 396 if (!isArchiveSymbol(S)) 397 continue; 398 Ret.push_back(SymNames.tell()); 399 if (Error E = S.printName(SymNames)) 400 return std::move(E); 401 SymNames << '\0'; 402 } 403 return Ret; 404 } 405 406 static Expected<std::vector<MemberData>> 407 computeMemberData(raw_ostream &StringTable, raw_ostream &SymNames, 408 object::Archive::Kind Kind, bool Thin, bool Deterministic, 409 bool NeedSymbols, ArrayRef<NewArchiveMember> NewMembers) { 410 static char PaddingData[8] = {'\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n'}; 411 412 // This ignores the symbol table, but we only need the value mod 8 and the 413 // symbol table is aligned to be a multiple of 8 bytes 414 uint64_t Pos = 0; 415 416 std::vector<MemberData> Ret; 417 bool HasObject = false; 418 419 // Deduplicate long member names in the string table and reuse earlier name 420 // offsets. This especially saves space for COFF Import libraries where all 421 // members have the same name. 422 StringMap<uint64_t> MemberNames; 423 424 // UniqueTimestamps is a special case to improve debugging on Darwin: 425 // 426 // The Darwin linker does not link debug info into the final 427 // binary. Instead, it emits entries of type N_OSO in in the output 428 // binary's symbol table, containing references to the linked-in 429 // object files. Using that reference, the debugger can read the 430 // debug data directly from the object files. Alternatively, an 431 // invocation of 'dsymutil' will link the debug data from the object 432 // files into a dSYM bundle, which can be loaded by the debugger, 433 // instead of the object files. 434 // 435 // For an object file, the N_OSO entries contain the absolute path 436 // path to the file, and the file's timestamp. For an object 437 // included in an archive, the path is formatted like 438 // "/absolute/path/to/archive.a(member.o)", and the timestamp is the 439 // archive member's timestamp, rather than the archive's timestamp. 440 // 441 // However, this doesn't always uniquely identify an object within 442 // an archive -- an archive file can have multiple entries with the 443 // same filename. (This will happen commonly if the original object 444 // files started in different directories.) The only way they get 445 // distinguished, then, is via the timestamp. But this process is 446 // unable to find the correct object file in the archive when there 447 // are two files of the same name and timestamp. 448 // 449 // Additionally, timestamp==0 is treated specially, and causes the 450 // timestamp to be ignored as a match criteria. 451 // 452 // That will "usually" work out okay when creating an archive not in 453 // deterministic timestamp mode, because the objects will probably 454 // have been created at different timestamps. 455 // 456 // To ameliorate this problem, in deterministic archive mode (which 457 // is the default), on Darwin we will emit a unique non-zero 458 // timestamp for each entry with a duplicated name. This is still 459 // deterministic: the only thing affecting that timestamp is the 460 // order of the files in the resultant archive. 461 // 462 // See also the functions that handle the lookup: 463 // in lldb: ObjectContainerBSDArchive::Archive::FindObject() 464 // in llvm/tools/dsymutil: BinaryHolder::GetArchiveMemberBuffers(). 465 bool UniqueTimestamps = Deterministic && isDarwin(Kind); 466 std::map<StringRef, unsigned> FilenameCount; 467 if (UniqueTimestamps) { 468 for (const NewArchiveMember &M : NewMembers) 469 FilenameCount[M.MemberName]++; 470 for (auto &Entry : FilenameCount) 471 Entry.second = Entry.second > 1 ? 1 : 0; 472 } 473 474 for (const NewArchiveMember &M : NewMembers) { 475 std::string Header; 476 raw_string_ostream Out(Header); 477 478 MemoryBufferRef Buf = M.Buf->getMemBufferRef(); 479 StringRef Data = Thin ? "" : Buf.getBuffer(); 480 481 // ld64 expects the members to be 8-byte aligned for 64-bit content and at 482 // least 4-byte aligned for 32-bit content. Opt for the larger encoding 483 // uniformly. This matches the behaviour with cctools and ensures that ld64 484 // is happy with archives that we generate. 485 unsigned MemberPadding = 486 isDarwin(Kind) ? offsetToAlignment(Data.size(), Align(8)) : 0; 487 unsigned TailPadding = 488 offsetToAlignment(Data.size() + MemberPadding, Align(2)); 489 StringRef Padding = StringRef(PaddingData, MemberPadding + TailPadding); 490 491 sys::TimePoint<std::chrono::seconds> ModTime; 492 if (UniqueTimestamps) 493 // Increment timestamp for each file of a given name. 494 ModTime = sys::toTimePoint(FilenameCount[M.MemberName]++); 495 else 496 ModTime = M.ModTime; 497 498 uint64_t Size = Buf.getBufferSize() + MemberPadding; 499 if (Size > object::Archive::MaxMemberSize) { 500 std::string StringMsg = 501 "File " + M.MemberName.str() + " exceeds size limit"; 502 return make_error<object::GenericBinaryError>( 503 std::move(StringMsg), object::object_error::parse_failed); 504 } 505 506 printMemberHeader(Out, Pos, StringTable, MemberNames, Kind, Thin, M, 507 ModTime, Size); 508 Out.flush(); 509 510 std::vector<unsigned> Symbols; 511 if (NeedSymbols) { 512 Expected<std::vector<unsigned>> SymbolsOrErr = 513 getSymbols(Buf, SymNames, HasObject); 514 if (auto E = SymbolsOrErr.takeError()) 515 return std::move(E); 516 Symbols = std::move(*SymbolsOrErr); 517 } 518 519 Pos += Header.size() + Data.size() + Padding.size(); 520 Ret.push_back({std::move(Symbols), std::move(Header), Data, Padding}); 521 } 522 // If there are no symbols, emit an empty symbol table, to satisfy Solaris 523 // tools, older versions of which expect a symbol table in a non-empty 524 // archive, regardless of whether there are any symbols in it. 525 if (HasObject && SymNames.tell() == 0) 526 SymNames << '\0' << '\0' << '\0'; 527 return Ret; 528 } 529 530 namespace llvm { 531 532 static ErrorOr<SmallString<128>> canonicalizePath(StringRef P) { 533 SmallString<128> Ret = P; 534 std::error_code Err = sys::fs::make_absolute(Ret); 535 if (Err) 536 return Err; 537 sys::path::remove_dots(Ret, /*removedotdot*/ true); 538 return Ret; 539 } 540 541 // Compute the relative path from From to To. 542 Expected<std::string> computeArchiveRelativePath(StringRef From, StringRef To) { 543 ErrorOr<SmallString<128>> PathToOrErr = canonicalizePath(To); 544 ErrorOr<SmallString<128>> DirFromOrErr = canonicalizePath(From); 545 if (!PathToOrErr || !DirFromOrErr) 546 return errorCodeToError(std::error_code(errno, std::generic_category())); 547 548 const SmallString<128> &PathTo = *PathToOrErr; 549 const SmallString<128> &DirFrom = sys::path::parent_path(*DirFromOrErr); 550 551 // Can't construct a relative path between different roots 552 if (sys::path::root_name(PathTo) != sys::path::root_name(DirFrom)) 553 return sys::path::convert_to_slash(PathTo); 554 555 // Skip common prefixes 556 auto FromTo = 557 std::mismatch(sys::path::begin(DirFrom), sys::path::end(DirFrom), 558 sys::path::begin(PathTo)); 559 auto FromI = FromTo.first; 560 auto ToI = FromTo.second; 561 562 // Construct relative path 563 SmallString<128> Relative; 564 for (auto FromE = sys::path::end(DirFrom); FromI != FromE; ++FromI) 565 sys::path::append(Relative, sys::path::Style::posix, ".."); 566 567 for (auto ToE = sys::path::end(PathTo); ToI != ToE; ++ToI) 568 sys::path::append(Relative, sys::path::Style::posix, *ToI); 569 570 return std::string(Relative.str()); 571 } 572 573 static Error writeArchiveToStream(raw_ostream &Out, 574 ArrayRef<NewArchiveMember> NewMembers, 575 bool WriteSymtab, object::Archive::Kind Kind, 576 bool Deterministic, bool Thin) { 577 assert((!Thin || !isBSDLike(Kind)) && "Only the gnu format has a thin mode"); 578 579 SmallString<0> SymNamesBuf; 580 raw_svector_ostream SymNames(SymNamesBuf); 581 SmallString<0> StringTableBuf; 582 raw_svector_ostream StringTable(StringTableBuf); 583 584 Expected<std::vector<MemberData>> DataOrErr = 585 computeMemberData(StringTable, SymNames, Kind, Thin, Deterministic, 586 WriteSymtab, NewMembers); 587 if (Error E = DataOrErr.takeError()) 588 return E; 589 std::vector<MemberData> &Data = *DataOrErr; 590 591 if (!StringTableBuf.empty()) 592 Data.insert(Data.begin(), computeStringTable(StringTableBuf)); 593 594 // We would like to detect if we need to switch to a 64-bit symbol table. 595 if (WriteSymtab) { 596 uint64_t MaxOffset = 8; // For the file signature. 597 uint64_t LastOffset = MaxOffset; 598 uint64_t NumSyms = 0; 599 for (const auto &M : Data) { 600 // Record the start of the member's offset 601 LastOffset = MaxOffset; 602 // Account for the size of each part associated with the member. 603 MaxOffset += M.Header.size() + M.Data.size() + M.Padding.size(); 604 NumSyms += M.Symbols.size(); 605 } 606 607 // We assume 32-bit offsets to see if 32-bit symbols are possible or not. 608 uint64_t SymtabSize = computeSymbolTableSize(Kind, NumSyms, 4, SymNamesBuf); 609 auto computeSymbolTableHeaderSize = 610 [=] { 611 SmallString<0> TmpBuf; 612 raw_svector_ostream Tmp(TmpBuf); 613 writeSymbolTableHeader(Tmp, Kind, Deterministic, SymtabSize); 614 return TmpBuf.size(); 615 }; 616 LastOffset += computeSymbolTableHeaderSize() + SymtabSize; 617 618 // The SYM64 format is used when an archive's member offsets are larger than 619 // 32-bits can hold. The need for this shift in format is detected by 620 // writeArchive. To test this we need to generate a file with a member that 621 // has an offset larger than 32-bits but this demands a very slow test. To 622 // speed the test up we use this environment variable to pretend like the 623 // cutoff happens before 32-bits and instead happens at some much smaller 624 // value. 625 uint64_t Sym64Threshold = 1ULL << 32; 626 const char *Sym64Env = std::getenv("SYM64_THRESHOLD"); 627 if (Sym64Env) 628 StringRef(Sym64Env).getAsInteger(10, Sym64Threshold); 629 630 // If LastOffset isn't going to fit in a 32-bit varible we need to switch 631 // to 64-bit. Note that the file can be larger than 4GB as long as the last 632 // member starts before the 4GB offset. 633 if (LastOffset >= Sym64Threshold) { 634 if (Kind == object::Archive::K_DARWIN) 635 Kind = object::Archive::K_DARWIN64; 636 else 637 Kind = object::Archive::K_GNU64; 638 } 639 } 640 641 if (Thin) 642 Out << "!<thin>\n"; 643 else 644 Out << "!<arch>\n"; 645 646 if (WriteSymtab) 647 writeSymbolTable(Out, Kind, Deterministic, Data, SymNamesBuf); 648 649 for (const MemberData &M : Data) 650 Out << M.Header << M.Data << M.Padding; 651 652 Out.flush(); 653 return Error::success(); 654 } 655 656 Error writeArchive(StringRef ArcName, ArrayRef<NewArchiveMember> NewMembers, 657 bool WriteSymtab, object::Archive::Kind Kind, 658 bool Deterministic, bool Thin, 659 std::unique_ptr<MemoryBuffer> OldArchiveBuf) { 660 Expected<sys::fs::TempFile> Temp = 661 sys::fs::TempFile::create(ArcName + ".temp-archive-%%%%%%%.a"); 662 if (!Temp) 663 return Temp.takeError(); 664 raw_fd_ostream Out(Temp->FD, false); 665 666 if (Error E = writeArchiveToStream(Out, NewMembers, WriteSymtab, Kind, 667 Deterministic, Thin)) { 668 if (Error DiscardError = Temp->discard()) 669 return joinErrors(std::move(E), std::move(DiscardError)); 670 return E; 671 } 672 673 // At this point, we no longer need whatever backing memory 674 // was used to generate the NewMembers. On Windows, this buffer 675 // could be a mapped view of the file we want to replace (if 676 // we're updating an existing archive, say). In that case, the 677 // rename would still succeed, but it would leave behind a 678 // temporary file (actually the original file renamed) because 679 // a file cannot be deleted while there's a handle open on it, 680 // only renamed. So by freeing this buffer, this ensures that 681 // the last open handle on the destination file, if any, is 682 // closed before we attempt to rename. 683 OldArchiveBuf.reset(); 684 685 return Temp->keep(ArcName); 686 } 687 688 Expected<std::unique_ptr<MemoryBuffer>> 689 writeArchiveToBuffer(ArrayRef<NewArchiveMember> NewMembers, bool WriteSymtab, 690 object::Archive::Kind Kind, bool Deterministic, 691 bool Thin) { 692 SmallVector<char, 0> ArchiveBufferVector; 693 raw_svector_ostream ArchiveStream(ArchiveBufferVector); 694 695 if (Error E = writeArchiveToStream(ArchiveStream, NewMembers, WriteSymtab, 696 Kind, Deterministic, Thin)) 697 return std::move(E); 698 699 return std::make_unique<SmallVectorMemoryBuffer>( 700 std::move(ArchiveBufferVector), /*RequiresNullTerminator=*/false); 701 } 702 703 } // namespace llvm 704