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