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

source code of llvm/lib/Object/ArchiveWriter.cpp