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/COFF.h"
21#include "llvm/Object/COFFImportFile.h"
22#include "llvm/Object/Error.h"
23#include "llvm/Object/IRObjectFile.h"
24#include "llvm/Object/MachO.h"
25#include "llvm/Object/ObjectFile.h"
26#include "llvm/Object/SymbolicFile.h"
27#include "llvm/Object/XCOFFObjectFile.h"
28#include "llvm/Support/Alignment.h"
29#include "llvm/Support/EndianStream.h"
30#include "llvm/Support/Errc.h"
31#include "llvm/Support/ErrorHandling.h"
32#include "llvm/Support/Format.h"
33#include "llvm/Support/MathExtras.h"
34#include "llvm/Support/Path.h"
35#include "llvm/Support/SmallVectorMemoryBuffer.h"
36#include "llvm/Support/raw_ostream.h"
37
38#include <cerrno>
39#include <map>
40
41#if !defined(_MSC_VER) && !defined(__MINGW32__)
42#include <unistd.h>
43#else
44#include <io.h>
45#endif
46
47using namespace llvm;
48using namespace llvm::object;
49
50struct SymMap {
51 bool UseECMap = false;
52 std::map<std::string, uint16_t> Map;
53 std::map<std::string, uint16_t> ECMap;
54};
55
56NewArchiveMember::NewArchiveMember(MemoryBufferRef BufRef)
57 : Buf(MemoryBuffer::getMemBuffer(Ref: BufRef, RequiresNullTerminator: false)),
58 MemberName(BufRef.getBufferIdentifier()) {}
59
60object::Archive::Kind NewArchiveMember::detectKindFromObject() const {
61 auto MemBufferRef = this->Buf->getMemBufferRef();
62 Expected<std::unique_ptr<object::ObjectFile>> OptionalObject =
63 object::ObjectFile::createObjectFile(Object: MemBufferRef);
64
65 if (OptionalObject) {
66 if (isa<object::MachOObjectFile>(Val: **OptionalObject))
67 return object::Archive::K_DARWIN;
68 if (isa<object::XCOFFObjectFile>(Val: **OptionalObject))
69 return object::Archive::K_AIXBIG;
70 if (isa<object::COFFObjectFile>(Val: **OptionalObject) ||
71 isa<object::COFFImportFile>(Val: **OptionalObject))
72 return object::Archive::K_COFF;
73 return object::Archive::K_GNU;
74 }
75
76 // Squelch the error in case we had a non-object file.
77 consumeError(Err: OptionalObject.takeError());
78
79 // If we're adding a bitcode file to the archive, detect the Archive kind
80 // based on the target triple.
81 LLVMContext Context;
82 if (identify_magic(magic: MemBufferRef.getBuffer()) == file_magic::bitcode) {
83 if (auto ObjOrErr = object::SymbolicFile::createSymbolicFile(
84 Object: MemBufferRef, Type: file_magic::bitcode, Context: &Context)) {
85 auto &IRObject = cast<object::IRObjectFile>(Val&: **ObjOrErr);
86 auto TargetTriple = Triple(IRObject.getTargetTriple());
87 return object::Archive::getDefaultKindForTriple(T: TargetTriple);
88 } else {
89 // Squelch the error in case this was not a SymbolicFile.
90 consumeError(Err: ObjOrErr.takeError());
91 }
92 }
93
94 return object::Archive::getDefaultKind();
95}
96
97Expected<NewArchiveMember>
98NewArchiveMember::getOldMember(const object::Archive::Child &OldMember,
99 bool Deterministic) {
100 Expected<llvm::MemoryBufferRef> BufOrErr = OldMember.getMemoryBufferRef();
101 if (!BufOrErr)
102 return BufOrErr.takeError();
103
104 NewArchiveMember M;
105 M.Buf = MemoryBuffer::getMemBuffer(Ref: *BufOrErr, RequiresNullTerminator: false);
106 M.MemberName = M.Buf->getBufferIdentifier();
107 if (!Deterministic) {
108 auto ModTimeOrErr = OldMember.getLastModified();
109 if (!ModTimeOrErr)
110 return ModTimeOrErr.takeError();
111 M.ModTime = ModTimeOrErr.get();
112 Expected<unsigned> UIDOrErr = OldMember.getUID();
113 if (!UIDOrErr)
114 return UIDOrErr.takeError();
115 M.UID = UIDOrErr.get();
116 Expected<unsigned> GIDOrErr = OldMember.getGID();
117 if (!GIDOrErr)
118 return GIDOrErr.takeError();
119 M.GID = GIDOrErr.get();
120 Expected<sys::fs::perms> AccessModeOrErr = OldMember.getAccessMode();
121 if (!AccessModeOrErr)
122 return AccessModeOrErr.takeError();
123 M.Perms = AccessModeOrErr.get();
124 }
125 return std::move(M);
126}
127
128Expected<NewArchiveMember> NewArchiveMember::getFile(StringRef FileName,
129 bool Deterministic) {
130 sys::fs::file_status Status;
131 auto FDOrErr = sys::fs::openNativeFileForRead(Name: FileName);
132 if (!FDOrErr)
133 return FDOrErr.takeError();
134 sys::fs::file_t FD = *FDOrErr;
135 assert(FD != sys::fs::kInvalidFile);
136
137 if (auto EC = sys::fs::status(FD, Result&: Status))
138 return errorCodeToError(EC);
139
140 // Opening a directory doesn't make sense. Let it fail.
141 // Linux cannot open directories with open(2), although
142 // cygwin and *bsd can.
143 if (Status.type() == sys::fs::file_type::directory_file)
144 return errorCodeToError(EC: make_error_code(E: errc::is_a_directory));
145
146 ErrorOr<std::unique_ptr<MemoryBuffer>> MemberBufferOrErr =
147 MemoryBuffer::getOpenFile(FD, Filename: FileName, FileSize: Status.getSize(), RequiresNullTerminator: false);
148 if (!MemberBufferOrErr)
149 return errorCodeToError(EC: MemberBufferOrErr.getError());
150
151 if (auto EC = sys::fs::closeFile(F&: FD))
152 return errorCodeToError(EC);
153
154 NewArchiveMember M;
155 M.Buf = std::move(*MemberBufferOrErr);
156 M.MemberName = M.Buf->getBufferIdentifier();
157 if (!Deterministic) {
158 M.ModTime = std::chrono::time_point_cast<std::chrono::seconds>(
159 t: Status.getLastModificationTime());
160 M.UID = Status.getUser();
161 M.GID = Status.getGroup();
162 M.Perms = Status.permissions();
163 }
164 return std::move(M);
165}
166
167template <typename T>
168static void printWithSpacePadding(raw_ostream &OS, T Data, unsigned Size) {
169 uint64_t OldPos = OS.tell();
170 OS << Data;
171 unsigned SizeSoFar = OS.tell() - OldPos;
172 assert(SizeSoFar <= Size && "Data doesn't fit in Size");
173 OS.indent(NumSpaces: Size - SizeSoFar);
174}
175
176static bool isDarwin(object::Archive::Kind Kind) {
177 return Kind == object::Archive::K_DARWIN ||
178 Kind == object::Archive::K_DARWIN64;
179}
180
181static bool isAIXBigArchive(object::Archive::Kind Kind) {
182 return Kind == object::Archive::K_AIXBIG;
183}
184
185static bool isCOFFArchive(object::Archive::Kind Kind) {
186 return Kind == object::Archive::K_COFF;
187}
188
189static bool isBSDLike(object::Archive::Kind Kind) {
190 switch (Kind) {
191 case object::Archive::K_GNU:
192 case object::Archive::K_GNU64:
193 case object::Archive::K_AIXBIG:
194 case object::Archive::K_COFF:
195 case object::Archive::K_ZOS:
196 return false;
197 case object::Archive::K_BSD:
198 case object::Archive::K_DARWIN:
199 case object::Archive::K_DARWIN64:
200 return true;
201 }
202 llvm_unreachable("not supported for writting");
203}
204
205template <class T>
206static void print(raw_ostream &Out, object::Archive::Kind Kind, T Val) {
207 support::endian::write(Out, Val,
208 isBSDLike(Kind) ? llvm::endianness::little
209 : llvm::endianness::big);
210}
211
212template <class T> static void printLE(raw_ostream &Out, T Val) {
213 support::endian::write(Out, Val, llvm::endianness::little);
214}
215
216static void printRestOfMemberHeader(
217 raw_ostream &Out, const sys::TimePoint<std::chrono::seconds> &ModTime,
218 unsigned UID, unsigned GID, unsigned Perms, uint64_t Size) {
219 printWithSpacePadding(OS&: Out, Data: sys::toTimeT(TP: ModTime), Size: 12);
220
221 // The format has only 6 chars for uid and gid. Truncate if the provided
222 // values don't fit.
223 printWithSpacePadding(OS&: Out, Data: UID % 1000000, Size: 6);
224 printWithSpacePadding(OS&: Out, Data: GID % 1000000, Size: 6);
225
226 printWithSpacePadding(OS&: Out, Data: format(Fmt: "%o", Vals: Perms), Size: 8);
227 printWithSpacePadding(OS&: Out, Data: Size, Size: 10);
228 Out << "`\n";
229}
230
231static void
232printGNUSmallMemberHeader(raw_ostream &Out, StringRef Name,
233 const sys::TimePoint<std::chrono::seconds> &ModTime,
234 unsigned UID, unsigned GID, unsigned Perms,
235 uint64_t Size) {
236 printWithSpacePadding(OS&: Out, Data: Twine(Name) + "/", Size: 16);
237 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms, Size);
238}
239
240static void
241printBSDMemberHeader(raw_ostream &Out, uint64_t Pos, StringRef Name,
242 const sys::TimePoint<std::chrono::seconds> &ModTime,
243 unsigned UID, unsigned GID, unsigned Perms, uint64_t Size) {
244 uint64_t PosAfterHeader = Pos + 60 + Name.size();
245 // Pad so that even 64 bit object files are aligned.
246 unsigned Pad = offsetToAlignment(Value: PosAfterHeader, Alignment: Align(8));
247 unsigned NameWithPadding = Name.size() + Pad;
248 printWithSpacePadding(OS&: Out, Data: Twine("#1/") + Twine(NameWithPadding), Size: 16);
249 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms,
250 Size: NameWithPadding + Size);
251 Out << Name;
252 while (Pad--)
253 Out.write(C: uint8_t(0));
254}
255
256static void
257printBigArchiveMemberHeader(raw_ostream &Out, StringRef Name,
258 const sys::TimePoint<std::chrono::seconds> &ModTime,
259 unsigned UID, unsigned GID, unsigned Perms,
260 uint64_t Size, uint64_t PrevOffset,
261 uint64_t NextOffset) {
262 unsigned NameLen = Name.size();
263
264 printWithSpacePadding(OS&: Out, Data: Size, Size: 20); // File member size
265 printWithSpacePadding(OS&: Out, Data: NextOffset, Size: 20); // Next member header offset
266 printWithSpacePadding(OS&: Out, Data: PrevOffset, Size: 20); // Previous member header offset
267 printWithSpacePadding(OS&: Out, Data: sys::toTimeT(TP: ModTime), Size: 12); // File member date
268 // The big archive format has 12 chars for uid and gid.
269 printWithSpacePadding(OS&: Out, Data: UID % 1000000000000, Size: 12); // UID
270 printWithSpacePadding(OS&: Out, Data: GID % 1000000000000, Size: 12); // GID
271 printWithSpacePadding(OS&: Out, Data: format(Fmt: "%o", Vals: Perms), Size: 12); // Permission
272 printWithSpacePadding(OS&: Out, Data: NameLen, Size: 4); // Name length
273 if (NameLen) {
274 printWithSpacePadding(OS&: Out, Data: Name, Size: NameLen); // Name
275 if (NameLen % 2)
276 Out.write(C: uint8_t(0)); // Null byte padding
277 }
278 Out << "`\n"; // Terminator
279}
280
281static bool useStringTable(bool Thin, StringRef Name) {
282 return Thin || Name.size() >= 16 || Name.contains(C: '/');
283}
284
285static bool is64BitKind(object::Archive::Kind Kind) {
286 switch (Kind) {
287 case object::Archive::K_GNU:
288 case object::Archive::K_BSD:
289 case object::Archive::K_DARWIN:
290 case object::Archive::K_COFF:
291 case object::Archive::K_ZOS:
292 return false;
293 case object::Archive::K_AIXBIG:
294 case object::Archive::K_DARWIN64:
295 case object::Archive::K_GNU64:
296 return true;
297 }
298 llvm_unreachable("not supported for writting");
299}
300
301static void
302printMemberHeader(raw_ostream &Out, uint64_t Pos, raw_ostream &StringTable,
303 StringMap<uint64_t> &MemberNames, object::Archive::Kind Kind,
304 bool Thin, const NewArchiveMember &M,
305 sys::TimePoint<std::chrono::seconds> ModTime, uint64_t Size) {
306 if (isBSDLike(Kind))
307 return printBSDMemberHeader(Out, Pos, Name: M.MemberName, ModTime, UID: M.UID, GID: M.GID,
308 Perms: M.Perms, Size);
309 if (!useStringTable(Thin, Name: M.MemberName))
310 return printGNUSmallMemberHeader(Out, Name: M.MemberName, ModTime, UID: M.UID, GID: M.GID,
311 Perms: M.Perms, Size);
312 Out << '/';
313 uint64_t NamePos;
314 if (Thin) {
315 NamePos = StringTable.tell();
316 StringTable << M.MemberName << "/\n";
317 } else {
318 auto Insertion = MemberNames.insert(KV: {M.MemberName, uint64_t(0)});
319 if (Insertion.second) {
320 Insertion.first->second = StringTable.tell();
321 StringTable << M.MemberName;
322 if (isCOFFArchive(Kind))
323 StringTable << '\0';
324 else
325 StringTable << "/\n";
326 }
327 NamePos = Insertion.first->second;
328 }
329 printWithSpacePadding(OS&: Out, Data: NamePos, Size: 15);
330 printRestOfMemberHeader(Out, ModTime, UID: M.UID, GID: M.GID, Perms: M.Perms, Size);
331}
332
333namespace {
334struct MemberData {
335 std::vector<unsigned> Symbols;
336 std::string Header;
337 StringRef Data;
338 StringRef Padding;
339 uint64_t PreHeadPadSize = 0;
340 std::unique_ptr<SymbolicFile> SymFile = nullptr;
341};
342} // namespace
343
344static MemberData computeStringTable(StringRef Names) {
345 unsigned Size = Names.size();
346 unsigned Pad = offsetToAlignment(Value: Size, Alignment: Align(2));
347 std::string Header;
348 raw_string_ostream Out(Header);
349 printWithSpacePadding(OS&: Out, Data: "//", Size: 48);
350 printWithSpacePadding(OS&: Out, Data: Size + Pad, Size: 10);
351 Out << "`\n";
352 return {.Symbols: {}, .Header: std::move(Header), .Data: Names, .Padding: Pad ? "\n" : ""};
353}
354
355static sys::TimePoint<std::chrono::seconds> now(bool Deterministic) {
356 using namespace std::chrono;
357
358 if (!Deterministic)
359 return time_point_cast<seconds>(t: system_clock::now());
360 return sys::TimePoint<seconds>();
361}
362
363static bool isArchiveSymbol(const object::BasicSymbolRef &S) {
364 Expected<uint32_t> SymFlagsOrErr = S.getFlags();
365 if (!SymFlagsOrErr)
366 // TODO: Actually report errors helpfully.
367 report_fatal_error(Err: SymFlagsOrErr.takeError());
368 if (*SymFlagsOrErr & object::SymbolRef::SF_FormatSpecific)
369 return false;
370 if (!(*SymFlagsOrErr & object::SymbolRef::SF_Global))
371 return false;
372 if (*SymFlagsOrErr & object::SymbolRef::SF_Undefined)
373 return false;
374 return true;
375}
376
377static void printNBits(raw_ostream &Out, object::Archive::Kind Kind,
378 uint64_t Val) {
379 if (is64BitKind(Kind))
380 print<uint64_t>(Out, Kind, Val);
381 else
382 print<uint32_t>(Out, Kind, Val);
383}
384
385static uint64_t computeSymbolTableSize(object::Archive::Kind Kind,
386 uint64_t NumSyms, uint64_t OffsetSize,
387 uint64_t StringTableSize,
388 uint32_t *Padding = nullptr) {
389 assert((OffsetSize == 4 || OffsetSize == 8) && "Unsupported OffsetSize");
390 uint64_t Size = OffsetSize; // Number of entries
391 if (isBSDLike(Kind))
392 Size += NumSyms * OffsetSize * 2; // Table
393 else
394 Size += NumSyms * OffsetSize; // Table
395 if (isBSDLike(Kind))
396 Size += OffsetSize; // byte count
397 Size += StringTableSize;
398 // ld64 expects the members to be 8-byte aligned for 64-bit content and at
399 // least 4-byte aligned for 32-bit content. Opt for the larger encoding
400 // uniformly.
401 // We do this for all bsd formats because it simplifies aligning members.
402 // For the big archive format, the symbol table is the last member, so there
403 // is no need to align.
404 uint32_t Pad = isAIXBigArchive(Kind)
405 ? 0
406 : offsetToAlignment(Value: Size, Alignment: Align(isBSDLike(Kind) ? 8 : 2));
407
408 Size += Pad;
409 if (Padding)
410 *Padding = Pad;
411 return Size;
412}
413
414static uint64_t computeSymbolMapSize(uint64_t NumObj, SymMap &SymMap,
415 uint32_t *Padding = nullptr) {
416 uint64_t Size = sizeof(uint32_t) * 2; // Number of symbols and objects entries
417 Size += NumObj * sizeof(uint32_t); // Offset table
418
419 for (auto S : SymMap.Map)
420 Size += sizeof(uint16_t) + S.first.length() + 1;
421
422 uint32_t Pad = offsetToAlignment(Value: Size, Alignment: Align(2));
423 Size += Pad;
424 if (Padding)
425 *Padding = Pad;
426 return Size;
427}
428
429static uint64_t computeECSymbolsSize(SymMap &SymMap,
430 uint32_t *Padding = nullptr) {
431 uint64_t Size = sizeof(uint32_t); // Number of symbols
432
433 for (auto S : SymMap.ECMap)
434 Size += sizeof(uint16_t) + S.first.length() + 1;
435
436 uint32_t Pad = offsetToAlignment(Value: Size, Alignment: Align(2));
437 Size += Pad;
438 if (Padding)
439 *Padding = Pad;
440 return Size;
441}
442
443static void writeSymbolTableHeader(raw_ostream &Out, object::Archive::Kind Kind,
444 bool Deterministic, uint64_t Size,
445 uint64_t PrevMemberOffset = 0,
446 uint64_t NextMemberOffset = 0) {
447 if (isBSDLike(Kind)) {
448 const char *Name = is64BitKind(Kind) ? "__.SYMDEF_64" : "__.SYMDEF";
449 printBSDMemberHeader(Out, Pos: Out.tell(), Name, ModTime: now(Deterministic), UID: 0, GID: 0, Perms: 0,
450 Size);
451 } else if (isAIXBigArchive(Kind)) {
452 printBigArchiveMemberHeader(Out, Name: "", ModTime: now(Deterministic), UID: 0, GID: 0, Perms: 0, Size,
453 PrevOffset: PrevMemberOffset, NextOffset: NextMemberOffset);
454 } else {
455 const char *Name = is64BitKind(Kind) ? "/SYM64" : "";
456 printGNUSmallMemberHeader(Out, Name, ModTime: now(Deterministic), UID: 0, GID: 0, Perms: 0, Size);
457 }
458}
459
460static uint64_t computeHeadersSize(object::Archive::Kind Kind,
461 uint64_t NumMembers,
462 uint64_t StringMemberSize, uint64_t NumSyms,
463 uint64_t SymNamesSize, SymMap *SymMap) {
464 uint32_t OffsetSize = is64BitKind(Kind) ? 8 : 4;
465 uint64_t SymtabSize =
466 computeSymbolTableSize(Kind, NumSyms, OffsetSize, StringTableSize: SymNamesSize);
467 auto computeSymbolTableHeaderSize = [=] {
468 SmallString<0> TmpBuf;
469 raw_svector_ostream Tmp(TmpBuf);
470 writeSymbolTableHeader(Out&: Tmp, Kind, Deterministic: true, Size: SymtabSize);
471 return TmpBuf.size();
472 };
473 uint32_t HeaderSize = computeSymbolTableHeaderSize();
474 uint64_t Size = strlen(s: "!<arch>\n") + HeaderSize + SymtabSize;
475
476 if (SymMap) {
477 Size += HeaderSize + computeSymbolMapSize(NumObj: NumMembers, SymMap&: *SymMap);
478 if (SymMap->ECMap.size())
479 Size += HeaderSize + computeECSymbolsSize(SymMap&: *SymMap);
480 }
481
482 return Size + StringMemberSize;
483}
484
485static Expected<std::unique_ptr<SymbolicFile>>
486getSymbolicFile(MemoryBufferRef Buf, LLVMContext &Context,
487 object::Archive::Kind Kind, function_ref<void(Error)> Warn) {
488 const file_magic Type = identify_magic(magic: Buf.getBuffer());
489 // Don't attempt to read non-symbolic file types.
490 if (!object::SymbolicFile::isSymbolicFile(Type, Context: &Context))
491 return nullptr;
492 if (Type == file_magic::bitcode) {
493 auto ObjOrErr = object::SymbolicFile::createSymbolicFile(
494 Object: Buf, Type: file_magic::bitcode, Context: &Context);
495 // An error reading a bitcode file most likely indicates that the file
496 // was created by a compiler from the future. Normally we don't try to
497 // implement forwards compatibility for bitcode files, but when creating an
498 // archive we can implement best-effort forwards compatibility by treating
499 // the file as a blob and not creating symbol index entries for it. lld and
500 // mold ignore the archive symbol index, so provided that you use one of
501 // these linkers, LTO will work as long as lld or the gold plugin is newer
502 // than the compiler. We only ignore errors if the archive format is one
503 // that is supported by a linker that is known to ignore the index,
504 // otherwise there's no chance of this working so we may as well error out.
505 // We print a warning on read failure so that users of linkers that rely on
506 // the symbol index can diagnose the issue.
507 //
508 // This is the same behavior as GNU ar when the linker plugin returns an
509 // error when reading the input file. If the bitcode file is actually
510 // malformed, it will be diagnosed at link time.
511 if (!ObjOrErr) {
512 switch (Kind) {
513 case object::Archive::K_BSD:
514 case object::Archive::K_GNU:
515 case object::Archive::K_GNU64:
516 Warn(ObjOrErr.takeError());
517 return nullptr;
518 case object::Archive::K_AIXBIG:
519 case object::Archive::K_COFF:
520 case object::Archive::K_DARWIN:
521 case object::Archive::K_DARWIN64:
522 case object::Archive::K_ZOS:
523 return ObjOrErr.takeError();
524 }
525 }
526 return std::move(*ObjOrErr);
527 } else {
528 auto ObjOrErr = object::SymbolicFile::createSymbolicFile(Object: Buf);
529 if (!ObjOrErr)
530 return ObjOrErr.takeError();
531 return std::move(*ObjOrErr);
532 }
533}
534
535static bool is64BitSymbolicFile(const SymbolicFile *SymObj) {
536 return SymObj != nullptr ? SymObj->is64Bit() : false;
537}
538
539// Log2 of PAGESIZE(4096) on an AIX system.
540static const uint32_t Log2OfAIXPageSize = 12;
541
542// In the AIX big archive format, since the data content follows the member file
543// name, if the name ends on an odd byte, an extra byte will be added for
544// padding. This ensures that the data within the member file starts at an even
545// byte.
546static const uint32_t MinBigArchiveMemDataAlign = 2;
547
548template <typename AuxiliaryHeader>
549uint16_t getAuxMaxAlignment(uint16_t AuxHeaderSize, AuxiliaryHeader *AuxHeader,
550 uint16_t Log2OfMaxAlign) {
551 // If the member doesn't have an auxiliary header, it isn't a loadable object
552 // and so it just needs aligning at the minimum value.
553 if (AuxHeader == nullptr)
554 return MinBigArchiveMemDataAlign;
555
556 // If the auxiliary header does not have both MaxAlignOfData and
557 // MaxAlignOfText field, it is not a loadable shared object file, so align at
558 // the minimum value. The 'ModuleType' member is located right after
559 // 'MaxAlignOfData' in the AuxiliaryHeader.
560 if (AuxHeaderSize < offsetof(AuxiliaryHeader, ModuleType))
561 return MinBigArchiveMemDataAlign;
562
563 // If the XCOFF object file does not have a loader section, it is not
564 // loadable, so align at the minimum value.
565 if (AuxHeader->SecNumOfLoader == 0)
566 return MinBigArchiveMemDataAlign;
567
568 // The content of the loadable member file needs to be aligned at MAX(maximum
569 // alignment of .text, maximum alignment of .data) if there are both fields.
570 // If the desired alignment is > PAGESIZE, 32-bit members are aligned on a
571 // word boundary, while 64-bit members are aligned on a PAGESIZE(2^12=4096)
572 // boundary.
573 uint16_t Log2OfAlign =
574 std::max(AuxHeader->MaxAlignOfText, AuxHeader->MaxAlignOfData);
575 return 1 << (Log2OfAlign > Log2OfAIXPageSize ? Log2OfMaxAlign : Log2OfAlign);
576}
577
578// AIX big archives may contain shared object members. The AIX OS requires these
579// members to be aligned if they are 64-bit and recommends it for 32-bit
580// members. This ensures that when these members are loaded they are aligned in
581// memory.
582static uint32_t getMemberAlignment(SymbolicFile *SymObj) {
583 XCOFFObjectFile *XCOFFObj = dyn_cast_or_null<XCOFFObjectFile>(Val: SymObj);
584 if (!XCOFFObj)
585 return MinBigArchiveMemDataAlign;
586
587 // If the desired alignment is > PAGESIZE, 32-bit members are aligned on a
588 // word boundary, while 64-bit members are aligned on a PAGESIZE boundary.
589 return XCOFFObj->is64Bit()
590 ? getAuxMaxAlignment(AuxHeaderSize: XCOFFObj->fileHeader64()->AuxHeaderSize,
591 AuxHeader: XCOFFObj->auxiliaryHeader64(),
592 Log2OfMaxAlign: Log2OfAIXPageSize)
593 : getAuxMaxAlignment(AuxHeaderSize: XCOFFObj->fileHeader32()->AuxHeaderSize,
594 AuxHeader: XCOFFObj->auxiliaryHeader32(), Log2OfMaxAlign: 2);
595}
596
597static void writeSymbolTable(raw_ostream &Out, object::Archive::Kind Kind,
598 bool Deterministic, ArrayRef<MemberData> Members,
599 StringRef StringTable, uint64_t MembersOffset,
600 unsigned NumSyms, uint64_t PrevMemberOffset = 0,
601 uint64_t NextMemberOffset = 0,
602 bool Is64Bit = false) {
603 // We don't write a symbol table on an archive with no members -- except on
604 // Darwin, where the linker will abort unless the archive has a symbol table.
605 if (StringTable.empty() && !isDarwin(Kind) && !isCOFFArchive(Kind))
606 return;
607
608 uint64_t OffsetSize = is64BitKind(Kind) ? 8 : 4;
609 uint32_t Pad;
610 uint64_t Size = computeSymbolTableSize(Kind, NumSyms, OffsetSize,
611 StringTableSize: StringTable.size(), Padding: &Pad);
612 writeSymbolTableHeader(Out, Kind, Deterministic, Size, PrevMemberOffset,
613 NextMemberOffset);
614
615 if (isBSDLike(Kind))
616 printNBits(Out, Kind, Val: NumSyms * 2 * OffsetSize);
617 else
618 printNBits(Out, Kind, Val: NumSyms);
619
620 uint64_t Pos = MembersOffset;
621 for (const MemberData &M : Members) {
622 if (isAIXBigArchive(Kind)) {
623 Pos += M.PreHeadPadSize;
624 if (is64BitSymbolicFile(SymObj: M.SymFile.get()) != Is64Bit) {
625 Pos += M.Header.size() + M.Data.size() + M.Padding.size();
626 continue;
627 }
628 }
629
630 for (unsigned StringOffset : M.Symbols) {
631 if (isBSDLike(Kind))
632 printNBits(Out, Kind, Val: StringOffset);
633 printNBits(Out, Kind, Val: Pos); // member offset
634 }
635 Pos += M.Header.size() + M.Data.size() + M.Padding.size();
636 }
637
638 if (isBSDLike(Kind))
639 // byte count of the string table
640 printNBits(Out, Kind, Val: StringTable.size());
641 Out << StringTable;
642
643 while (Pad--)
644 Out.write(C: uint8_t(0));
645}
646
647static void writeSymbolMap(raw_ostream &Out, object::Archive::Kind Kind,
648 bool Deterministic, ArrayRef<MemberData> Members,
649 SymMap &SymMap, uint64_t MembersOffset) {
650 uint32_t Pad;
651 uint64_t Size = computeSymbolMapSize(NumObj: Members.size(), SymMap, Padding: &Pad);
652 writeSymbolTableHeader(Out, Kind, Deterministic, Size, PrevMemberOffset: 0);
653
654 uint32_t Pos = MembersOffset;
655
656 printLE<uint32_t>(Out, Val: Members.size());
657 for (const MemberData &M : Members) {
658 printLE(Out, Val: Pos); // member offset
659 Pos += M.Header.size() + M.Data.size() + M.Padding.size();
660 }
661
662 printLE<uint32_t>(Out, Val: SymMap.Map.size());
663
664 for (auto S : SymMap.Map)
665 printLE(Out, Val: S.second);
666 for (auto S : SymMap.Map)
667 Out << S.first << '\0';
668
669 while (Pad--)
670 Out.write(C: uint8_t(0));
671}
672
673static void writeECSymbols(raw_ostream &Out, object::Archive::Kind Kind,
674 bool Deterministic, ArrayRef<MemberData> Members,
675 SymMap &SymMap) {
676 uint32_t Pad;
677 uint64_t Size = computeECSymbolsSize(SymMap, Padding: &Pad);
678 printGNUSmallMemberHeader(Out, Name: "/<ECSYMBOLS>", ModTime: now(Deterministic), UID: 0, GID: 0, Perms: 0,
679 Size);
680
681 printLE<uint32_t>(Out, Val: SymMap.ECMap.size());
682
683 for (auto S : SymMap.ECMap)
684 printLE(Out, Val: S.second);
685 for (auto S : SymMap.ECMap)
686 Out << S.first << '\0';
687 while (Pad--)
688 Out.write(C: uint8_t(0));
689}
690
691static bool isECObject(object::SymbolicFile &Obj) {
692 if (Obj.isCOFF())
693 return cast<llvm::object::COFFObjectFile>(Val: &Obj)->getMachine() !=
694 COFF::IMAGE_FILE_MACHINE_ARM64;
695
696 if (Obj.isCOFFImportFile())
697 return cast<llvm::object::COFFImportFile>(Val: &Obj)->getMachine() !=
698 COFF::IMAGE_FILE_MACHINE_ARM64;
699
700 if (Obj.isIR()) {
701 Expected<std::string> TripleStr =
702 getBitcodeTargetTriple(Buffer: Obj.getMemoryBufferRef());
703 if (!TripleStr)
704 return false;
705 Triple T(std::move(*TripleStr));
706 return T.isWindowsArm64EC() || T.getArch() == Triple::x86_64;
707 }
708
709 return false;
710}
711
712static bool isAnyArm64COFF(object::SymbolicFile &Obj) {
713 if (Obj.isCOFF())
714 return COFF::isAnyArm64(Machine: cast<COFFObjectFile>(Val: &Obj)->getMachine());
715
716 if (Obj.isCOFFImportFile())
717 return COFF::isAnyArm64(Machine: cast<COFFImportFile>(Val: &Obj)->getMachine());
718
719 if (Obj.isIR()) {
720 Expected<std::string> TripleStr =
721 getBitcodeTargetTriple(Buffer: Obj.getMemoryBufferRef());
722 if (!TripleStr)
723 return false;
724 Triple T(std::move(*TripleStr));
725 return T.isOSWindows() && T.getArch() == Triple::aarch64;
726 }
727
728 return false;
729}
730
731bool isImportDescriptor(StringRef Name) {
732 return Name.starts_with(Prefix: ImportDescriptorPrefix) ||
733 Name == StringRef{NullImportDescriptorSymbolName} ||
734 (Name.starts_with(Prefix: NullThunkDataPrefix) &&
735 Name.ends_with(Suffix: NullThunkDataSuffix));
736}
737
738static Expected<std::vector<unsigned>> getSymbols(SymbolicFile *Obj,
739 uint16_t Index,
740 raw_ostream &SymNames,
741 SymMap *SymMap) {
742 std::vector<unsigned> Ret;
743
744 if (Obj == nullptr)
745 return Ret;
746
747 std::map<std::string, uint16_t> *Map = nullptr;
748 if (SymMap)
749 Map = SymMap->UseECMap && isECObject(Obj&: *Obj) ? &SymMap->ECMap : &SymMap->Map;
750
751 for (const object::BasicSymbolRef &S : Obj->symbols()) {
752 if (!isArchiveSymbol(S))
753 continue;
754 if (Map) {
755 std::string Name;
756 raw_string_ostream NameStream(Name);
757 if (Error E = S.printName(OS&: NameStream))
758 return std::move(E);
759 if (!Map->try_emplace(k: Name, args&: Index).second)
760 continue; // ignore duplicated symbol
761 if (Map == &SymMap->Map) {
762 Ret.push_back(x: SymNames.tell());
763 SymNames << Name << '\0';
764 // If EC is enabled, then the import descriptors are NOT put into EC
765 // objects so we need to copy them to the EC map manually.
766 if (SymMap->UseECMap && isImportDescriptor(Name))
767 SymMap->ECMap[Name] = Index;
768 }
769 } else {
770 Ret.push_back(x: SymNames.tell());
771 if (Error E = S.printName(OS&: SymNames))
772 return std::move(E);
773 SymNames << '\0';
774 }
775 }
776 return Ret;
777}
778
779static Expected<std::vector<MemberData>>
780computeMemberData(raw_ostream &StringTable, raw_ostream &SymNames,
781 object::Archive::Kind Kind, bool Thin, bool Deterministic,
782 SymtabWritingMode NeedSymbols, SymMap *SymMap,
783 LLVMContext &Context, ArrayRef<NewArchiveMember> NewMembers,
784 std::optional<bool> IsEC, function_ref<void(Error)> Warn) {
785 static char PaddingData[8] = {'\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n'};
786 uint64_t MemHeadPadSize = 0;
787 uint64_t Pos =
788 isAIXBigArchive(Kind) ? sizeof(object::BigArchive::FixLenHdr) : 0;
789
790 std::vector<MemberData> Ret;
791 bool HasObject = false;
792
793 // Deduplicate long member names in the string table and reuse earlier name
794 // offsets. This especially saves space for COFF Import libraries where all
795 // members have the same name.
796 StringMap<uint64_t> MemberNames;
797
798 // UniqueTimestamps is a special case to improve debugging on Darwin:
799 //
800 // The Darwin linker does not link debug info into the final
801 // binary. Instead, it emits entries of type N_OSO in the output
802 // binary's symbol table, containing references to the linked-in
803 // object files. Using that reference, the debugger can read the
804 // debug data directly from the object files. Alternatively, an
805 // invocation of 'dsymutil' will link the debug data from the object
806 // files into a dSYM bundle, which can be loaded by the debugger,
807 // instead of the object files.
808 //
809 // For an object file, the N_OSO entries contain the absolute path
810 // path to the file, and the file's timestamp. For an object
811 // included in an archive, the path is formatted like
812 // "/absolute/path/to/archive.a(member.o)", and the timestamp is the
813 // archive member's timestamp, rather than the archive's timestamp.
814 //
815 // However, this doesn't always uniquely identify an object within
816 // an archive -- an archive file can have multiple entries with the
817 // same filename. (This will happen commonly if the original object
818 // files started in different directories.) The only way they get
819 // distinguished, then, is via the timestamp. But this process is
820 // unable to find the correct object file in the archive when there
821 // are two files of the same name and timestamp.
822 //
823 // Additionally, timestamp==0 is treated specially, and causes the
824 // timestamp to be ignored as a match criteria.
825 //
826 // That will "usually" work out okay when creating an archive not in
827 // deterministic timestamp mode, because the objects will probably
828 // have been created at different timestamps.
829 //
830 // To ameliorate this problem, in deterministic archive mode (which
831 // is the default), on Darwin we will emit a unique non-zero
832 // timestamp for each entry with a duplicated name. This is still
833 // deterministic: the only thing affecting that timestamp is the
834 // order of the files in the resultant archive.
835 //
836 // See also the functions that handle the lookup:
837 // in lldb: ObjectContainerBSDArchive::Archive::FindObject()
838 // in llvm/tools/dsymutil: BinaryHolder::GetArchiveMemberBuffers().
839 bool UniqueTimestamps = Deterministic && isDarwin(Kind);
840 std::map<StringRef, unsigned> FilenameCount;
841 if (UniqueTimestamps) {
842 for (const NewArchiveMember &M : NewMembers)
843 FilenameCount[M.MemberName]++;
844 for (auto &Entry : FilenameCount)
845 Entry.second = Entry.second > 1 ? 1 : 0;
846 }
847
848 std::vector<std::unique_ptr<SymbolicFile>> SymFiles;
849
850 if (NeedSymbols != SymtabWritingMode::NoSymtab || isAIXBigArchive(Kind)) {
851 for (const NewArchiveMember &M : NewMembers) {
852 Expected<std::unique_ptr<SymbolicFile>> SymFileOrErr = getSymbolicFile(
853 Buf: M.Buf->getMemBufferRef(), Context, Kind, Warn: [&](Error Err) {
854 Warn(createFileError(F: M.MemberName, E: std::move(Err)));
855 });
856 if (!SymFileOrErr)
857 return createFileError(F: M.MemberName, E: SymFileOrErr.takeError());
858 SymFiles.push_back(x: std::move(*SymFileOrErr));
859 }
860 }
861
862 if (SymMap) {
863 if (IsEC) {
864 SymMap->UseECMap = *IsEC;
865 } else {
866 // When IsEC is not specified by the caller, use it when we have both
867 // any ARM64 object (ARM64 or ARM64EC) and any EC object (ARM64EC or
868 // AMD64). This may be a single ARM64EC object, but may also be separate
869 // ARM64 and AMD64 objects.
870 bool HaveArm64 = false, HaveEC = false;
871 for (std::unique_ptr<SymbolicFile> &SymFile : SymFiles) {
872 if (!SymFile)
873 continue;
874 if (!HaveArm64)
875 HaveArm64 = isAnyArm64COFF(Obj&: *SymFile);
876 if (!HaveEC)
877 HaveEC = isECObject(Obj&: *SymFile);
878 if (HaveArm64 && HaveEC) {
879 SymMap->UseECMap = true;
880 break;
881 }
882 }
883 }
884 }
885
886 // The big archive format needs to know the offset of the previous member
887 // header.
888 uint64_t PrevOffset = 0;
889 uint64_t NextMemHeadPadSize = 0;
890
891 for (uint32_t Index = 0; Index < NewMembers.size(); ++Index) {
892 const NewArchiveMember *M = &NewMembers[Index];
893 std::string Header;
894 raw_string_ostream Out(Header);
895
896 MemoryBufferRef Buf = M->Buf->getMemBufferRef();
897 StringRef Data = Thin ? "" : Buf.getBuffer();
898
899 // ld64 expects the members to be 8-byte aligned for 64-bit content and at
900 // least 4-byte aligned for 32-bit content. Opt for the larger encoding
901 // uniformly. This matches the behaviour with cctools and ensures that ld64
902 // is happy with archives that we generate.
903 unsigned MemberPadding =
904 isDarwin(Kind) ? offsetToAlignment(Value: Data.size(), Alignment: Align(8)) : 0;
905 unsigned TailPadding =
906 offsetToAlignment(Value: Data.size() + MemberPadding, Alignment: Align(2));
907 StringRef Padding = StringRef(PaddingData, MemberPadding + TailPadding);
908
909 sys::TimePoint<std::chrono::seconds> ModTime;
910 if (UniqueTimestamps)
911 // Increment timestamp for each file of a given name.
912 ModTime = sys::toTimePoint(T: FilenameCount[M->MemberName]++);
913 else
914 ModTime = M->ModTime;
915
916 uint64_t Size = Buf.getBufferSize() + MemberPadding;
917 if (Size > object::Archive::MaxMemberSize) {
918 std::string StringMsg =
919 "File " + M->MemberName.str() + " exceeds size limit";
920 return make_error<object::GenericBinaryError>(
921 Args: std::move(StringMsg), Args: object::object_error::parse_failed);
922 }
923
924 std::unique_ptr<SymbolicFile> CurSymFile;
925 if (!SymFiles.empty())
926 CurSymFile = std::move(SymFiles[Index]);
927
928 // In the big archive file format, we need to calculate and include the next
929 // member offset and previous member offset in the file member header.
930 if (isAIXBigArchive(Kind)) {
931 uint64_t OffsetToMemData = Pos + sizeof(object::BigArMemHdrType) +
932 alignTo(Value: M->MemberName.size(), Align: 2);
933
934 if (M == NewMembers.begin())
935 NextMemHeadPadSize =
936 alignToPowerOf2(Value: OffsetToMemData,
937 Align: getMemberAlignment(SymObj: CurSymFile.get())) -
938 OffsetToMemData;
939
940 MemHeadPadSize = NextMemHeadPadSize;
941 Pos += MemHeadPadSize;
942 uint64_t NextOffset = Pos + sizeof(object::BigArMemHdrType) +
943 alignTo(Value: M->MemberName.size(), Align: 2) + alignTo(Value: Size, Align: 2);
944
945 // If there is another member file after this, we need to calculate the
946 // padding before the header.
947 if (Index + 1 != SymFiles.size()) {
948 uint64_t OffsetToNextMemData =
949 NextOffset + sizeof(object::BigArMemHdrType) +
950 alignTo(Value: NewMembers[Index + 1].MemberName.size(), Align: 2);
951 NextMemHeadPadSize =
952 alignToPowerOf2(Value: OffsetToNextMemData,
953 Align: getMemberAlignment(SymObj: SymFiles[Index + 1].get())) -
954 OffsetToNextMemData;
955 NextOffset += NextMemHeadPadSize;
956 }
957 printBigArchiveMemberHeader(Out, Name: M->MemberName, ModTime, UID: M->UID, GID: M->GID,
958 Perms: M->Perms, Size, PrevOffset, NextOffset);
959 PrevOffset = Pos;
960 } else {
961 printMemberHeader(Out, Pos, StringTable, MemberNames, Kind, Thin, M: *M,
962 ModTime, Size);
963 }
964
965 std::vector<unsigned> Symbols;
966 if (NeedSymbols != SymtabWritingMode::NoSymtab) {
967 Expected<std::vector<unsigned>> SymbolsOrErr =
968 getSymbols(Obj: CurSymFile.get(), Index: Index + 1, SymNames, SymMap);
969 if (!SymbolsOrErr)
970 return createFileError(F: M->MemberName, E: SymbolsOrErr.takeError());
971 Symbols = std::move(*SymbolsOrErr);
972 if (CurSymFile)
973 HasObject = true;
974 }
975
976 Pos += Header.size() + Data.size() + Padding.size();
977 Ret.push_back(x: {.Symbols: std::move(Symbols), .Header: std::move(Header), .Data: Data, .Padding: Padding,
978 .PreHeadPadSize: MemHeadPadSize, .SymFile: std::move(CurSymFile)});
979 }
980 // If there are no symbols, emit an empty symbol table, to satisfy Solaris
981 // tools, older versions of which expect a symbol table in a non-empty
982 // archive, regardless of whether there are any symbols in it.
983 if (HasObject && SymNames.tell() == 0 && !isCOFFArchive(Kind))
984 SymNames << '\0' << '\0' << '\0';
985 return std::move(Ret);
986}
987
988namespace llvm {
989
990static ErrorOr<SmallString<128>> canonicalizePath(StringRef P) {
991 SmallString<128> Ret = P;
992 std::error_code Err = sys::fs::make_absolute(path&: Ret);
993 if (Err)
994 return Err;
995 sys::path::remove_dots(path&: Ret, /*removedotdot*/ remove_dot_dot: true);
996 return Ret;
997}
998
999// Compute the relative path from From to To.
1000Expected<std::string> computeArchiveRelativePath(StringRef From, StringRef To) {
1001 ErrorOr<SmallString<128>> PathToOrErr = canonicalizePath(P: To);
1002 ErrorOr<SmallString<128>> DirFromOrErr = canonicalizePath(P: From);
1003 if (!PathToOrErr || !DirFromOrErr)
1004 return errorCodeToError(EC: errnoAsErrorCode());
1005
1006 const SmallString<128> &PathTo = *PathToOrErr;
1007 const SmallString<128> &DirFrom = sys::path::parent_path(path: *DirFromOrErr);
1008
1009 // Can't construct a relative path between different roots
1010 if (sys::path::root_name(path: PathTo) != sys::path::root_name(path: DirFrom))
1011 return sys::path::convert_to_slash(path: PathTo);
1012
1013 // Skip common prefixes
1014 auto FromTo =
1015 std::mismatch(first1: sys::path::begin(path: DirFrom), last1: sys::path::end(path: DirFrom),
1016 first2: sys::path::begin(path: PathTo));
1017 auto FromI = FromTo.first;
1018 auto ToI = FromTo.second;
1019
1020 // Construct relative path
1021 SmallString<128> Relative;
1022 for (auto FromE = sys::path::end(path: DirFrom); FromI != FromE; ++FromI)
1023 sys::path::append(path&: Relative, style: sys::path::Style::posix, a: "..");
1024
1025 for (auto ToE = sys::path::end(path: PathTo); ToI != ToE; ++ToI)
1026 sys::path::append(path&: Relative, style: sys::path::Style::posix, a: *ToI);
1027
1028 return std::string(Relative);
1029}
1030
1031Error writeArchiveToStream(raw_ostream &Out,
1032 ArrayRef<NewArchiveMember> NewMembers,
1033 SymtabWritingMode WriteSymtab,
1034 object::Archive::Kind Kind, bool Deterministic,
1035 bool Thin, std::optional<bool> IsEC,
1036 function_ref<void(Error)> Warn) {
1037 assert((!Thin || !isBSDLike(Kind)) && "Only the gnu format has a thin mode");
1038
1039 SmallString<0> SymNamesBuf;
1040 raw_svector_ostream SymNames(SymNamesBuf);
1041 SmallString<0> StringTableBuf;
1042 raw_svector_ostream StringTable(StringTableBuf);
1043 SymMap SymMap;
1044 bool ShouldWriteSymtab = WriteSymtab != SymtabWritingMode::NoSymtab;
1045
1046 // COFF symbol map uses 16-bit indexes, so we can't use it if there are too
1047 // many members. COFF format also requires symbol table presence, so use
1048 // GNU format when NoSymtab is requested.
1049 if (isCOFFArchive(Kind) && (NewMembers.size() > 0xfffe || !ShouldWriteSymtab))
1050 Kind = object::Archive::K_GNU;
1051
1052 // In the scenario when LLVMContext is populated SymbolicFile will contain a
1053 // reference to it, thus SymbolicFile should be destroyed first.
1054 LLVMContext Context;
1055
1056 Expected<std::vector<MemberData>> DataOrErr = computeMemberData(
1057 StringTable, SymNames, Kind, Thin, Deterministic, NeedSymbols: WriteSymtab,
1058 SymMap: isCOFFArchive(Kind) ? &SymMap : nullptr, Context, NewMembers, IsEC, Warn);
1059 if (Error E = DataOrErr.takeError())
1060 return E;
1061 std::vector<MemberData> &Data = *DataOrErr;
1062
1063 uint64_t StringTableSize = 0;
1064 MemberData StringTableMember;
1065 if (!StringTableBuf.empty() && !isAIXBigArchive(Kind)) {
1066 StringTableMember = computeStringTable(Names: StringTableBuf);
1067 StringTableSize = StringTableMember.Header.size() +
1068 StringTableMember.Data.size() +
1069 StringTableMember.Padding.size();
1070 }
1071
1072 // We would like to detect if we need to switch to a 64-bit symbol table.
1073 uint64_t LastMemberEndOffset = 0;
1074 uint64_t LastMemberHeaderOffset = 0;
1075 uint64_t NumSyms = 0;
1076 uint64_t NumSyms32 = 0; // Store symbol number of 32-bit member files.
1077
1078 for (const auto &M : Data) {
1079 // Record the start of the member's offset
1080 LastMemberEndOffset += M.PreHeadPadSize;
1081 LastMemberHeaderOffset = LastMemberEndOffset;
1082 // Account for the size of each part associated with the member.
1083 LastMemberEndOffset += M.Header.size() + M.Data.size() + M.Padding.size();
1084 NumSyms += M.Symbols.size();
1085
1086 // AIX big archive files may contain two global symbol tables. The
1087 // first global symbol table locates 32-bit file members that define global
1088 // symbols; the second global symbol table does the same for 64-bit file
1089 // members. As a big archive can have both 32-bit and 64-bit file members,
1090 // we need to know the number of symbols in each symbol table individually.
1091 if (isAIXBigArchive(Kind) && ShouldWriteSymtab) {
1092 if (!is64BitSymbolicFile(SymObj: M.SymFile.get()))
1093 NumSyms32 += M.Symbols.size();
1094 }
1095 }
1096
1097 std::optional<uint64_t> HeadersSize;
1098
1099 // The symbol table is put at the end of the big archive file. The symbol
1100 // table is at the start of the archive file for other archive formats.
1101 if (ShouldWriteSymtab && !is64BitKind(Kind)) {
1102 // We assume 32-bit offsets to see if 32-bit symbols are possible or not.
1103 HeadersSize = computeHeadersSize(Kind, NumMembers: Data.size(), StringMemberSize: StringTableSize,
1104 NumSyms, SymNamesSize: SymNamesBuf.size(),
1105 SymMap: isCOFFArchive(Kind) ? &SymMap : nullptr);
1106
1107 // The SYM64 format is used when an archive's member offsets are larger than
1108 // 32-bits can hold. The need for this shift in format is detected by
1109 // writeArchive. To test this we need to generate a file with a member that
1110 // has an offset larger than 32-bits but this demands a very slow test. To
1111 // speed the test up we use this environment variable to pretend like the
1112 // cutoff happens before 32-bits and instead happens at some much smaller
1113 // value.
1114 uint64_t Sym64Threshold = 1ULL << 32;
1115 const char *Sym64Env = std::getenv(name: "SYM64_THRESHOLD");
1116 if (Sym64Env)
1117 StringRef(Sym64Env).getAsInteger(Radix: 10, Result&: Sym64Threshold);
1118
1119 // If LastMemberHeaderOffset isn't going to fit in a 32-bit varible we need
1120 // to switch to 64-bit. Note that the file can be larger than 4GB as long as
1121 // the last member starts before the 4GB offset.
1122 if (*HeadersSize + LastMemberHeaderOffset >= Sym64Threshold) {
1123 switch (Kind) {
1124 case object::Archive::K_COFF:
1125 // COFF format has no 64-bit version, so we use GNU64 instead.
1126 if (!SymMap.Map.empty() && !SymMap.ECMap.empty())
1127 // Only the COFF format supports the ECSYMBOLS section, so don’t use
1128 // GNU64 when two symbol maps are required.
1129 return make_error<object::GenericBinaryError>(
1130 Args: "Archive is too large: ARM64X does not support archives larger "
1131 "than 4GB");
1132 // Since this changes the headers, we need to recalculate everything.
1133 return writeArchiveToStream(Out, NewMembers, WriteSymtab,
1134 Kind: object::Archive::K_GNU64, Deterministic,
1135 Thin, IsEC, Warn);
1136 case object::Archive::K_DARWIN:
1137 Kind = object::Archive::K_DARWIN64;
1138 break;
1139 default:
1140 Kind = object::Archive::K_GNU64;
1141 break;
1142 }
1143 HeadersSize.reset();
1144 }
1145 }
1146
1147 if (Thin)
1148 Out << "!<thin>\n";
1149 else if (isAIXBigArchive(Kind))
1150 Out << "<bigaf>\n";
1151 else
1152 Out << "!<arch>\n";
1153
1154 if (!isAIXBigArchive(Kind)) {
1155 if (ShouldWriteSymtab) {
1156 if (!HeadersSize)
1157 HeadersSize = computeHeadersSize(
1158 Kind, NumMembers: Data.size(), StringMemberSize: StringTableSize, NumSyms, SymNamesSize: SymNamesBuf.size(),
1159 SymMap: isCOFFArchive(Kind) ? &SymMap : nullptr);
1160 writeSymbolTable(Out, Kind, Deterministic, Members: Data, StringTable: SymNamesBuf,
1161 MembersOffset: *HeadersSize, NumSyms);
1162
1163 if (isCOFFArchive(Kind))
1164 writeSymbolMap(Out, Kind, Deterministic, Members: Data, SymMap, MembersOffset: *HeadersSize);
1165 }
1166
1167 if (StringTableSize)
1168 Out << StringTableMember.Header << StringTableMember.Data
1169 << StringTableMember.Padding;
1170
1171 if (ShouldWriteSymtab && SymMap.ECMap.size())
1172 writeECSymbols(Out, Kind, Deterministic, Members: Data, SymMap);
1173
1174 for (const MemberData &M : Data)
1175 Out << M.Header << M.Data << M.Padding;
1176 } else {
1177 HeadersSize = sizeof(object::BigArchive::FixLenHdr);
1178 LastMemberEndOffset += *HeadersSize;
1179 LastMemberHeaderOffset += *HeadersSize;
1180
1181 // For the big archive (AIX) format, compute a table of member names and
1182 // offsets, used in the member table.
1183 uint64_t MemberTableNameStrTblSize = 0;
1184 std::vector<size_t> MemberOffsets;
1185 std::vector<StringRef> MemberNames;
1186 // Loop across object to find offset and names.
1187 uint64_t MemberEndOffset = sizeof(object::BigArchive::FixLenHdr);
1188 for (size_t I = 0, Size = NewMembers.size(); I != Size; ++I) {
1189 const NewArchiveMember &Member = NewMembers[I];
1190 MemberTableNameStrTblSize += Member.MemberName.size() + 1;
1191 MemberEndOffset += Data[I].PreHeadPadSize;
1192 MemberOffsets.push_back(x: MemberEndOffset);
1193 MemberNames.push_back(x: Member.MemberName);
1194 // File member name ended with "`\n". The length is included in
1195 // BigArMemHdrType.
1196 MemberEndOffset += sizeof(object::BigArMemHdrType) +
1197 alignTo(Value: Data[I].Data.size(), Align: 2) +
1198 alignTo(Value: Member.MemberName.size(), Align: 2);
1199 }
1200
1201 // AIX member table size.
1202 uint64_t MemberTableSize = 20 + // Number of members field
1203 20 * MemberOffsets.size() +
1204 MemberTableNameStrTblSize;
1205
1206 SmallString<0> SymNamesBuf32;
1207 SmallString<0> SymNamesBuf64;
1208 raw_svector_ostream SymNames32(SymNamesBuf32);
1209 raw_svector_ostream SymNames64(SymNamesBuf64);
1210
1211 if (ShouldWriteSymtab && NumSyms)
1212 // Generate the symbol names for the members.
1213 for (const auto &M : Data) {
1214 Expected<std::vector<unsigned>> SymbolsOrErr = getSymbols(
1215 Obj: M.SymFile.get(), Index: 0,
1216 SymNames&: is64BitSymbolicFile(SymObj: M.SymFile.get()) ? SymNames64 : SymNames32,
1217 SymMap: nullptr);
1218 if (!SymbolsOrErr)
1219 return SymbolsOrErr.takeError();
1220 }
1221
1222 uint64_t MemberTableEndOffset =
1223 LastMemberEndOffset +
1224 alignTo(Value: sizeof(object::BigArMemHdrType) + MemberTableSize, Align: 2);
1225
1226 // In AIX OS, The 'GlobSymOffset' field in the fixed-length header contains
1227 // the offset to the 32-bit global symbol table, and the 'GlobSym64Offset'
1228 // contains the offset to the 64-bit global symbol table.
1229 uint64_t GlobalSymbolOffset =
1230 (ShouldWriteSymtab &&
1231 (WriteSymtab != SymtabWritingMode::BigArchive64) && NumSyms32 > 0)
1232 ? MemberTableEndOffset
1233 : 0;
1234
1235 uint64_t GlobalSymbolOffset64 = 0;
1236 uint64_t NumSyms64 = NumSyms - NumSyms32;
1237 if (ShouldWriteSymtab && (WriteSymtab != SymtabWritingMode::BigArchive32) &&
1238 NumSyms64 > 0) {
1239 if (GlobalSymbolOffset == 0)
1240 GlobalSymbolOffset64 = MemberTableEndOffset;
1241 else
1242 // If there is a global symbol table for 32-bit members,
1243 // the 64-bit global symbol table is after the 32-bit one.
1244 GlobalSymbolOffset64 =
1245 GlobalSymbolOffset + sizeof(object::BigArMemHdrType) +
1246 (NumSyms32 + 1) * 8 + alignTo(Value: SymNamesBuf32.size(), Align: 2);
1247 }
1248
1249 // Fixed Sized Header.
1250 printWithSpacePadding(OS&: Out, Data: NewMembers.size() ? LastMemberEndOffset : 0,
1251 Size: 20); // Offset to member table
1252 // If there are no file members in the archive, there will be no global
1253 // symbol table.
1254 printWithSpacePadding(OS&: Out, Data: GlobalSymbolOffset, Size: 20);
1255 printWithSpacePadding(OS&: Out, Data: GlobalSymbolOffset64, Size: 20);
1256 printWithSpacePadding(OS&: Out,
1257 Data: NewMembers.size()
1258 ? sizeof(object::BigArchive::FixLenHdr) +
1259 Data[0].PreHeadPadSize
1260 : 0,
1261 Size: 20); // Offset to first archive member
1262 printWithSpacePadding(OS&: Out, Data: NewMembers.size() ? LastMemberHeaderOffset : 0,
1263 Size: 20); // Offset to last archive member
1264 printWithSpacePadding(
1265 OS&: Out, Data: 0,
1266 Size: 20); // Offset to first member of free list - Not supported yet
1267
1268 for (const MemberData &M : Data) {
1269 Out << std::string(M.PreHeadPadSize, '\0');
1270 Out << M.Header << M.Data;
1271 if (M.Data.size() % 2)
1272 Out << '\0';
1273 }
1274
1275 if (NewMembers.size()) {
1276 // Member table.
1277 printBigArchiveMemberHeader(Out, Name: "", ModTime: sys::toTimePoint(T: 0), UID: 0, GID: 0, Perms: 0,
1278 Size: MemberTableSize, PrevOffset: LastMemberHeaderOffset,
1279 NextOffset: GlobalSymbolOffset ? GlobalSymbolOffset
1280 : GlobalSymbolOffset64);
1281 printWithSpacePadding(OS&: Out, Data: MemberOffsets.size(), Size: 20); // Number of members
1282 for (uint64_t MemberOffset : MemberOffsets)
1283 printWithSpacePadding(OS&: Out, Data: MemberOffset,
1284 Size: 20); // Offset to member file header.
1285 for (StringRef MemberName : MemberNames)
1286 Out << MemberName << '\0'; // Member file name, null byte padding.
1287
1288 if (MemberTableNameStrTblSize % 2)
1289 Out << '\0'; // Name table must be tail padded to an even number of
1290 // bytes.
1291
1292 if (ShouldWriteSymtab) {
1293 // Write global symbol table for 32-bit file members.
1294 if (GlobalSymbolOffset) {
1295 writeSymbolTable(Out, Kind, Deterministic, Members: Data, StringTable: SymNamesBuf32,
1296 MembersOffset: *HeadersSize, NumSyms: NumSyms32, PrevMemberOffset: LastMemberEndOffset,
1297 NextMemberOffset: GlobalSymbolOffset64);
1298 // Add padding between the symbol tables, if needed.
1299 if (GlobalSymbolOffset64 && (SymNamesBuf32.size() % 2))
1300 Out << '\0';
1301 }
1302
1303 // Write global symbol table for 64-bit file members.
1304 if (GlobalSymbolOffset64)
1305 writeSymbolTable(Out, Kind, Deterministic, Members: Data, StringTable: SymNamesBuf64,
1306 MembersOffset: *HeadersSize, NumSyms: NumSyms64,
1307 PrevMemberOffset: GlobalSymbolOffset ? GlobalSymbolOffset
1308 : LastMemberEndOffset,
1309 NextMemberOffset: 0, Is64Bit: true);
1310 }
1311 }
1312 }
1313 Out.flush();
1314 return Error::success();
1315}
1316
1317void warnToStderr(Error Err) {
1318 llvm::logAllUnhandledErrors(E: std::move(Err), OS&: llvm::errs(), ErrorBanner: "warning: ");
1319}
1320
1321Error writeArchive(StringRef ArcName, ArrayRef<NewArchiveMember> NewMembers,
1322 SymtabWritingMode WriteSymtab, object::Archive::Kind Kind,
1323 bool Deterministic, bool Thin,
1324 std::unique_ptr<MemoryBuffer> OldArchiveBuf,
1325 std::optional<bool> IsEC, function_ref<void(Error)> Warn) {
1326 Expected<sys::fs::TempFile> Temp =
1327 sys::fs::TempFile::create(Model: ArcName + ".temp-archive-%%%%%%%.a");
1328 if (!Temp)
1329 return Temp.takeError();
1330 raw_fd_ostream Out(Temp->FD, false);
1331
1332 if (Error E = writeArchiveToStream(Out, NewMembers, WriteSymtab, Kind,
1333 Deterministic, Thin, IsEC, Warn)) {
1334 if (Error DiscardError = Temp->discard())
1335 return joinErrors(E1: std::move(E), E2: std::move(DiscardError));
1336 return E;
1337 }
1338
1339 // At this point, we no longer need whatever backing memory
1340 // was used to generate the NewMembers. On Windows, this buffer
1341 // could be a mapped view of the file we want to replace (if
1342 // we're updating an existing archive, say). In that case, the
1343 // rename would still succeed, but it would leave behind a
1344 // temporary file (actually the original file renamed) because
1345 // a file cannot be deleted while there's a handle open on it,
1346 // only renamed. So by freeing this buffer, this ensures that
1347 // the last open handle on the destination file, if any, is
1348 // closed before we attempt to rename.
1349 OldArchiveBuf.reset();
1350
1351 return Temp->keep(Name: ArcName);
1352}
1353
1354Expected<std::unique_ptr<MemoryBuffer>>
1355writeArchiveToBuffer(ArrayRef<NewArchiveMember> NewMembers,
1356 SymtabWritingMode WriteSymtab, object::Archive::Kind Kind,
1357 bool Deterministic, bool Thin,
1358 function_ref<void(Error)> Warn) {
1359 SmallVector<char, 0> ArchiveBufferVector;
1360 raw_svector_ostream ArchiveStream(ArchiveBufferVector);
1361
1362 if (Error E =
1363 writeArchiveToStream(Out&: ArchiveStream, NewMembers, WriteSymtab, Kind,
1364 Deterministic, Thin, IsEC: std::nullopt, Warn))
1365 return std::move(E);
1366
1367 return std::make_unique<SmallVectorMemoryBuffer>(
1368 args: std::move(ArchiveBufferVector), /*RequiresNullTerminator=*/args: false);
1369}
1370
1371} // namespace llvm
1372