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
47 using namespace llvm;
48 using namespace llvm::object;
49
50 struct SymMap {
51 bool UseECMap = false;
52 std::map<std::string, uint16_t> Map;
53 std::map<std::string, uint16_t> ECMap;
54 };
55
NewArchiveMember(MemoryBufferRef BufRef)56 NewArchiveMember::NewArchiveMember(MemoryBufferRef BufRef)
57 : Buf(MemoryBuffer::getMemBuffer(BufRef, false)),
58 MemberName(BufRef.getBufferIdentifier()) {}
59
detectKindFromObject() const60 object::Archive::Kind NewArchiveMember::detectKindFromObject() const {
61 auto MemBufferRef = this->Buf->getMemBufferRef();
62 Expected<std::unique_ptr<object::ObjectFile>> OptionalObject =
63 object::ObjectFile::createObjectFile(MemBufferRef);
64
65 if (OptionalObject) {
66 if (isa<object::MachOObjectFile>(**OptionalObject))
67 return object::Archive::K_DARWIN;
68 if (isa<object::XCOFFObjectFile>(**OptionalObject))
69 return object::Archive::K_AIXBIG;
70 if (isa<object::COFFObjectFile>(**OptionalObject) ||
71 isa<object::COFFImportFile>(**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(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(MemBufferRef.getBuffer()) == file_magic::bitcode) {
83 if (auto ObjOrErr = object::SymbolicFile::createSymbolicFile(
84 MemBufferRef, file_magic::bitcode, &Context)) {
85 auto &IRObject = cast<object::IRObjectFile>(**ObjOrErr);
86 auto TargetTriple = Triple(IRObject.getTargetTriple());
87 return object::Archive::getDefaultKindForTriple(TargetTriple);
88 } else {
89 // Squelch the error in case this was not a SymbolicFile.
90 consumeError(ObjOrErr.takeError());
91 }
92 }
93
94 return object::Archive::getDefaultKind();
95 }
96
97 Expected<NewArchiveMember>
getOldMember(const object::Archive::Child & OldMember,bool Deterministic)98 NewArchiveMember::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(*BufOrErr, 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
getFile(StringRef FileName,bool Deterministic)128 Expected<NewArchiveMember> NewArchiveMember::getFile(StringRef FileName,
129 bool Deterministic) {
130 sys::fs::file_status Status;
131 auto FDOrErr = sys::fs::openNativeFileForRead(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, 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(make_error_code(errc::is_a_directory));
145
146 ErrorOr<std::unique_ptr<MemoryBuffer>> MemberBufferOrErr =
147 MemoryBuffer::getOpenFile(FD, FileName, Status.getSize(), false);
148 if (!MemberBufferOrErr)
149 return errorCodeToError(MemberBufferOrErr.getError());
150
151 if (auto EC = sys::fs::closeFile(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 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
167 template <typename T>
printWithSpacePadding(raw_ostream & OS,T Data,unsigned Size)168 static 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(Size - SizeSoFar);
174 }
175
isDarwin(object::Archive::Kind Kind)176 static bool isDarwin(object::Archive::Kind Kind) {
177 return Kind == object::Archive::K_DARWIN ||
178 Kind == object::Archive::K_DARWIN64;
179 }
180
isAIXBigArchive(object::Archive::Kind Kind)181 static bool isAIXBigArchive(object::Archive::Kind Kind) {
182 return Kind == object::Archive::K_AIXBIG;
183 }
184
isCOFFArchive(object::Archive::Kind Kind)185 static bool isCOFFArchive(object::Archive::Kind Kind) {
186 return Kind == object::Archive::K_COFF;
187 }
188
isBSDLike(object::Archive::Kind Kind)189 static 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
204 template <class T>
print(raw_ostream & Out,object::Archive::Kind Kind,T Val)205 static 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
printLE(raw_ostream & Out,T Val)211 template <class T> static void printLE(raw_ostream &Out, T Val) {
212 support::endian::write(Out, Val, llvm::endianness::little);
213 }
214
printRestOfMemberHeader(raw_ostream & Out,const sys::TimePoint<std::chrono::seconds> & ModTime,unsigned UID,unsigned GID,unsigned Perms,uint64_t Size)215 static 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(Out, sys::toTimeT(ModTime), 12);
219
220 // The format has only 6 chars for uid and gid. Truncate if the provided
221 // values don't fit.
222 printWithSpacePadding(Out, UID % 1000000, 6);
223 printWithSpacePadding(Out, GID % 1000000, 6);
224
225 printWithSpacePadding(Out, format("%o", Perms), 8);
226 printWithSpacePadding(Out, Size, 10);
227 Out << "`\n";
228 }
229
230 static void
printGNUSmallMemberHeader(raw_ostream & Out,StringRef Name,const sys::TimePoint<std::chrono::seconds> & ModTime,unsigned UID,unsigned GID,unsigned Perms,uint64_t Size)231 printGNUSmallMemberHeader(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(Out, Twine(Name) + "/", 16);
236 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms, Size);
237 }
238
239 static void
printBSDMemberHeader(raw_ostream & Out,uint64_t Pos,StringRef Name,const sys::TimePoint<std::chrono::seconds> & ModTime,unsigned UID,unsigned GID,unsigned Perms,uint64_t Size)240 printBSDMemberHeader(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(PosAfterHeader, Align(8));
246 unsigned NameWithPadding = Name.size() + Pad;
247 printWithSpacePadding(Out, Twine("#1/") + Twine(NameWithPadding), 16);
248 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms,
249 NameWithPadding + Size);
250 Out << Name;
251 while (Pad--)
252 Out.write(uint8_t(0));
253 }
254
255 static void
printBigArchiveMemberHeader(raw_ostream & Out,StringRef Name,const sys::TimePoint<std::chrono::seconds> & ModTime,unsigned UID,unsigned GID,unsigned Perms,uint64_t Size,uint64_t PrevOffset,uint64_t NextOffset)256 printBigArchiveMemberHeader(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(Out, Size, 20); // File member size
264 printWithSpacePadding(Out, NextOffset, 20); // Next member header offset
265 printWithSpacePadding(Out, PrevOffset, 20); // Previous member header offset
266 printWithSpacePadding(Out, sys::toTimeT(ModTime), 12); // File member date
267 // The big archive format has 12 chars for uid and gid.
268 printWithSpacePadding(Out, UID % 1000000000000, 12); // UID
269 printWithSpacePadding(Out, GID % 1000000000000, 12); // GID
270 printWithSpacePadding(Out, format("%o", Perms), 12); // Permission
271 printWithSpacePadding(Out, NameLen, 4); // Name length
272 if (NameLen) {
273 printWithSpacePadding(Out, Name, NameLen); // Name
274 if (NameLen % 2)
275 Out.write(uint8_t(0)); // Null byte padding
276 }
277 Out << "`\n"; // Terminator
278 }
279
useStringTable(bool Thin,StringRef Name)280 static bool useStringTable(bool Thin, StringRef Name) {
281 return Thin || Name.size() >= 16 || Name.contains('/');
282 }
283
is64BitKind(object::Archive::Kind Kind)284 static 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
299 static void
printMemberHeader(raw_ostream & Out,uint64_t Pos,raw_ostream & StringTable,StringMap<uint64_t> & MemberNames,object::Archive::Kind Kind,bool Thin,const NewArchiveMember & M,sys::TimePoint<std::chrono::seconds> ModTime,uint64_t Size)300 printMemberHeader(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, M.MemberName, ModTime, M.UID, M.GID,
306 M.Perms, Size);
307 if (!useStringTable(Thin, M.MemberName))
308 return printGNUSmallMemberHeader(Out, M.MemberName, ModTime, M.UID, M.GID,
309 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({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(Out, NamePos, 15);
328 printRestOfMemberHeader(Out, ModTime, M.UID, M.GID, M.Perms, Size);
329 }
330
331 namespace {
332 struct 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
computeStringTable(StringRef Names)342 static MemberData computeStringTable(StringRef Names) {
343 unsigned Size = Names.size();
344 unsigned Pad = offsetToAlignment(Size, Align(2));
345 std::string Header;
346 raw_string_ostream Out(Header);
347 printWithSpacePadding(Out, "//", 48);
348 printWithSpacePadding(Out, Size + Pad, 10);
349 Out << "`\n";
350 Out.flush();
351 return {{}, std::move(Header), Names, Pad ? "\n" : ""};
352 }
353
now(bool Deterministic)354 static sys::TimePoint<std::chrono::seconds> now(bool Deterministic) {
355 using namespace std::chrono;
356
357 if (!Deterministic)
358 return time_point_cast<seconds>(system_clock::now());
359 return sys::TimePoint<seconds>();
360 }
361
isArchiveSymbol(const object::BasicSymbolRef & S)362 static 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(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
printNBits(raw_ostream & Out,object::Archive::Kind Kind,uint64_t Val)376 static 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
computeSymbolTableSize(object::Archive::Kind Kind,uint64_t NumSyms,uint64_t OffsetSize,uint64_t StringTableSize,uint32_t * Padding=nullptr)384 static 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(Size, Align(isBSDLike(Kind) ? 8 : 2));
406
407 Size += Pad;
408 if (Padding)
409 *Padding = Pad;
410 return Size;
411 }
412
computeSymbolMapSize(uint64_t NumObj,SymMap & SymMap,uint32_t * Padding=nullptr)413 static 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(Size, Align(2));
422 Size += Pad;
423 if (Padding)
424 *Padding = Pad;
425 return Size;
426 }
427
computeECSymbolsSize(SymMap & SymMap,uint32_t * Padding=nullptr)428 static 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(Size, Align(2));
436 Size += Pad;
437 if (Padding)
438 *Padding = Pad;
439 return Size;
440 }
441
writeSymbolTableHeader(raw_ostream & Out,object::Archive::Kind Kind,bool Deterministic,uint64_t Size,uint64_t PrevMemberOffset=0,uint64_t NextMemberOffset=0)442 static 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, Out.tell(), Name, now(Deterministic), 0, 0, 0,
449 Size);
450 } else if (isAIXBigArchive(Kind)) {
451 printBigArchiveMemberHeader(Out, "", now(Deterministic), 0, 0, 0, Size,
452 PrevMemberOffset, NextMemberOffset);
453 } else {
454 const char *Name = is64BitKind(Kind) ? "/SYM64" : "";
455 printGNUSmallMemberHeader(Out, Name, now(Deterministic), 0, 0, 0, Size);
456 }
457 }
458
computeHeadersSize(object::Archive::Kind Kind,uint64_t NumMembers,uint64_t StringMemberSize,uint64_t NumSyms,uint64_t SymNamesSize,SymMap * SymMap)459 static 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, SymNamesSize);
466 auto computeSymbolTableHeaderSize = [=] {
467 SmallString<0> TmpBuf;
468 raw_svector_ostream Tmp(TmpBuf);
469 writeSymbolTableHeader(Tmp, Kind, true, SymtabSize);
470 return TmpBuf.size();
471 };
472 uint32_t HeaderSize = computeSymbolTableHeaderSize();
473 uint64_t Size = strlen("!<arch>\n") + HeaderSize + SymtabSize;
474
475 if (SymMap) {
476 Size += HeaderSize + computeSymbolMapSize(NumMembers, *SymMap);
477 if (SymMap->ECMap.size())
478 Size += HeaderSize + computeECSymbolsSize(*SymMap);
479 }
480
481 return Size + StringMemberSize;
482 }
483
484 static Expected<std::unique_ptr<SymbolicFile>>
getSymbolicFile(MemoryBufferRef Buf,LLVMContext & Context,object::Archive::Kind Kind,function_ref<void (Error)> Warn)485 getSymbolicFile(MemoryBufferRef Buf, LLVMContext &Context,
486 object::Archive::Kind Kind, function_ref<void(Error)> Warn) {
487 const file_magic Type = identify_magic(Buf.getBuffer());
488 // Don't attempt to read non-symbolic file types.
489 if (!object::SymbolicFile::isSymbolicFile(Type, &Context))
490 return nullptr;
491 if (Type == file_magic::bitcode) {
492 auto ObjOrErr = object::SymbolicFile::createSymbolicFile(
493 Buf, file_magic::bitcode, &Context);
494 // An error reading a bitcode file most likely indicates that the file
495 // was created by a compiler from the future. Normally we don't try to
496 // implement forwards compatibility for bitcode files, but when creating an
497 // archive we can implement best-effort forwards compatibility by treating
498 // the file as a blob and not creating symbol index entries for it. lld and
499 // mold ignore the archive symbol index, so provided that you use one of
500 // these linkers, LTO will work as long as lld or the gold plugin is newer
501 // than the compiler. We only ignore errors if the archive format is one
502 // that is supported by a linker that is known to ignore the index,
503 // otherwise there's no chance of this working so we may as well error out.
504 // We print a warning on read failure so that users of linkers that rely on
505 // the symbol index can diagnose the issue.
506 //
507 // This is the same behavior as GNU ar when the linker plugin returns an
508 // error when reading the input file. If the bitcode file is actually
509 // malformed, it will be diagnosed at link time.
510 if (!ObjOrErr) {
511 switch (Kind) {
512 case object::Archive::K_BSD:
513 case object::Archive::K_GNU:
514 case object::Archive::K_GNU64:
515 Warn(ObjOrErr.takeError());
516 return nullptr;
517 case object::Archive::K_AIXBIG:
518 case object::Archive::K_COFF:
519 case object::Archive::K_DARWIN:
520 case object::Archive::K_DARWIN64:
521 return ObjOrErr.takeError();
522 }
523 }
524 return std::move(*ObjOrErr);
525 } else {
526 auto ObjOrErr = object::SymbolicFile::createSymbolicFile(Buf);
527 if (!ObjOrErr)
528 return ObjOrErr.takeError();
529 return std::move(*ObjOrErr);
530 }
531 }
532
is64BitSymbolicFile(const SymbolicFile * SymObj)533 static bool is64BitSymbolicFile(const SymbolicFile *SymObj) {
534 return SymObj != nullptr ? SymObj->is64Bit() : false;
535 }
536
537 // Log2 of PAGESIZE(4096) on an AIX system.
538 static const uint32_t Log2OfAIXPageSize = 12;
539
540 // In the AIX big archive format, since the data content follows the member file
541 // name, if the name ends on an odd byte, an extra byte will be added for
542 // padding. This ensures that the data within the member file starts at an even
543 // byte.
544 static const uint32_t MinBigArchiveMemDataAlign = 2;
545
546 template <typename AuxiliaryHeader>
getAuxMaxAlignment(uint16_t AuxHeaderSize,AuxiliaryHeader * AuxHeader,uint16_t Log2OfMaxAlign)547 uint16_t getAuxMaxAlignment(uint16_t AuxHeaderSize, AuxiliaryHeader *AuxHeader,
548 uint16_t Log2OfMaxAlign) {
549 // If the member doesn't have an auxiliary header, it isn't a loadable object
550 // and so it just needs aligning at the minimum value.
551 if (AuxHeader == nullptr)
552 return MinBigArchiveMemDataAlign;
553
554 // If the auxiliary header does not have both MaxAlignOfData and
555 // MaxAlignOfText field, it is not a loadable shared object file, so align at
556 // the minimum value. The 'ModuleType' member is located right after
557 // 'MaxAlignOfData' in the AuxiliaryHeader.
558 if (AuxHeaderSize < offsetof(AuxiliaryHeader, ModuleType))
559 return MinBigArchiveMemDataAlign;
560
561 // If the XCOFF object file does not have a loader section, it is not
562 // loadable, so align at the minimum value.
563 if (AuxHeader->SecNumOfLoader == 0)
564 return MinBigArchiveMemDataAlign;
565
566 // The content of the loadable member file needs to be aligned at MAX(maximum
567 // alignment of .text, maximum alignment of .data) if there are both fields.
568 // If the desired alignment is > PAGESIZE, 32-bit members are aligned on a
569 // word boundary, while 64-bit members are aligned on a PAGESIZE(2^12=4096)
570 // boundary.
571 uint16_t Log2OfAlign =
572 std::max(AuxHeader->MaxAlignOfText, AuxHeader->MaxAlignOfData);
573 return 1 << (Log2OfAlign > Log2OfAIXPageSize ? Log2OfMaxAlign : Log2OfAlign);
574 }
575
576 // AIX big archives may contain shared object members. The AIX OS requires these
577 // members to be aligned if they are 64-bit and recommends it for 32-bit
578 // members. This ensures that when these members are loaded they are aligned in
579 // memory.
getMemberAlignment(SymbolicFile * SymObj)580 static uint32_t getMemberAlignment(SymbolicFile *SymObj) {
581 XCOFFObjectFile *XCOFFObj = dyn_cast_or_null<XCOFFObjectFile>(SymObj);
582 if (!XCOFFObj)
583 return MinBigArchiveMemDataAlign;
584
585 // If the desired alignment is > PAGESIZE, 32-bit members are aligned on a
586 // word boundary, while 64-bit members are aligned on a PAGESIZE boundary.
587 return XCOFFObj->is64Bit()
588 ? getAuxMaxAlignment(XCOFFObj->fileHeader64()->AuxHeaderSize,
589 XCOFFObj->auxiliaryHeader64(),
590 Log2OfAIXPageSize)
591 : getAuxMaxAlignment(XCOFFObj->fileHeader32()->AuxHeaderSize,
592 XCOFFObj->auxiliaryHeader32(), 2);
593 }
594
writeSymbolTable(raw_ostream & Out,object::Archive::Kind Kind,bool Deterministic,ArrayRef<MemberData> Members,StringRef StringTable,uint64_t MembersOffset,unsigned NumSyms,uint64_t PrevMemberOffset=0,uint64_t NextMemberOffset=0,bool Is64Bit=false)595 static void writeSymbolTable(raw_ostream &Out, object::Archive::Kind Kind,
596 bool Deterministic, ArrayRef<MemberData> Members,
597 StringRef StringTable, uint64_t MembersOffset,
598 unsigned NumSyms, uint64_t PrevMemberOffset = 0,
599 uint64_t NextMemberOffset = 0,
600 bool Is64Bit = false) {
601 // We don't write a symbol table on an archive with no members -- except on
602 // Darwin, where the linker will abort unless the archive has a symbol table.
603 if (StringTable.empty() && !isDarwin(Kind) && !isCOFFArchive(Kind))
604 return;
605
606 uint64_t OffsetSize = is64BitKind(Kind) ? 8 : 4;
607 uint32_t Pad;
608 uint64_t Size = computeSymbolTableSize(Kind, NumSyms, OffsetSize,
609 StringTable.size(), &Pad);
610 writeSymbolTableHeader(Out, Kind, Deterministic, Size, PrevMemberOffset,
611 NextMemberOffset);
612
613 if (isBSDLike(Kind))
614 printNBits(Out, Kind, NumSyms * 2 * OffsetSize);
615 else
616 printNBits(Out, Kind, NumSyms);
617
618 uint64_t Pos = MembersOffset;
619 for (const MemberData &M : Members) {
620 if (isAIXBigArchive(Kind)) {
621 Pos += M.PreHeadPadSize;
622 if (is64BitSymbolicFile(M.SymFile.get()) != Is64Bit) {
623 Pos += M.Header.size() + M.Data.size() + M.Padding.size();
624 continue;
625 }
626 }
627
628 for (unsigned StringOffset : M.Symbols) {
629 if (isBSDLike(Kind))
630 printNBits(Out, Kind, StringOffset);
631 printNBits(Out, Kind, Pos); // member offset
632 }
633 Pos += M.Header.size() + M.Data.size() + M.Padding.size();
634 }
635
636 if (isBSDLike(Kind))
637 // byte count of the string table
638 printNBits(Out, Kind, StringTable.size());
639 Out << StringTable;
640
641 while (Pad--)
642 Out.write(uint8_t(0));
643 }
644
writeSymbolMap(raw_ostream & Out,object::Archive::Kind Kind,bool Deterministic,ArrayRef<MemberData> Members,SymMap & SymMap,uint64_t MembersOffset)645 static void writeSymbolMap(raw_ostream &Out, object::Archive::Kind Kind,
646 bool Deterministic, ArrayRef<MemberData> Members,
647 SymMap &SymMap, uint64_t MembersOffset) {
648 uint32_t Pad;
649 uint64_t Size = computeSymbolMapSize(Members.size(), SymMap, &Pad);
650 writeSymbolTableHeader(Out, Kind, Deterministic, Size, 0);
651
652 uint32_t Pos = MembersOffset;
653
654 printLE<uint32_t>(Out, Members.size());
655 for (const MemberData &M : Members) {
656 printLE(Out, Pos); // member offset
657 Pos += M.Header.size() + M.Data.size() + M.Padding.size();
658 }
659
660 printLE<uint32_t>(Out, SymMap.Map.size());
661
662 for (auto S : SymMap.Map)
663 printLE(Out, S.second);
664 for (auto S : SymMap.Map)
665 Out << S.first << '\0';
666
667 while (Pad--)
668 Out.write(uint8_t(0));
669 }
670
writeECSymbols(raw_ostream & Out,object::Archive::Kind Kind,bool Deterministic,ArrayRef<MemberData> Members,SymMap & SymMap)671 static void writeECSymbols(raw_ostream &Out, object::Archive::Kind Kind,
672 bool Deterministic, ArrayRef<MemberData> Members,
673 SymMap &SymMap) {
674 uint32_t Pad;
675 uint64_t Size = computeECSymbolsSize(SymMap, &Pad);
676 printGNUSmallMemberHeader(Out, "/<ECSYMBOLS>", now(Deterministic), 0, 0, 0,
677 Size);
678
679 printLE<uint32_t>(Out, SymMap.ECMap.size());
680
681 for (auto S : SymMap.ECMap)
682 printLE(Out, S.second);
683 for (auto S : SymMap.ECMap)
684 Out << S.first << '\0';
685 while (Pad--)
686 Out.write(uint8_t(0));
687 }
688
isECObject(object::SymbolicFile & Obj)689 static bool isECObject(object::SymbolicFile &Obj) {
690 if (Obj.isCOFF())
691 return cast<llvm::object::COFFObjectFile>(&Obj)->getMachine() !=
692 COFF::IMAGE_FILE_MACHINE_ARM64;
693
694 if (Obj.isCOFFImportFile())
695 return cast<llvm::object::COFFImportFile>(&Obj)->getMachine() !=
696 COFF::IMAGE_FILE_MACHINE_ARM64;
697
698 if (Obj.isIR()) {
699 Expected<std::string> TripleStr =
700 getBitcodeTargetTriple(Obj.getMemoryBufferRef());
701 if (!TripleStr)
702 return false;
703 Triple T(*TripleStr);
704 return T.isWindowsArm64EC() || T.getArch() == Triple::x86_64;
705 }
706
707 return false;
708 }
709
isAnyArm64COFF(object::SymbolicFile & Obj)710 static bool isAnyArm64COFF(object::SymbolicFile &Obj) {
711 if (Obj.isCOFF())
712 return COFF::isAnyArm64(cast<COFFObjectFile>(&Obj)->getMachine());
713
714 if (Obj.isCOFFImportFile())
715 return COFF::isAnyArm64(cast<COFFImportFile>(&Obj)->getMachine());
716
717 if (Obj.isIR()) {
718 Expected<std::string> TripleStr =
719 getBitcodeTargetTriple(Obj.getMemoryBufferRef());
720 if (!TripleStr)
721 return false;
722 Triple T(*TripleStr);
723 return T.isOSWindows() && T.getArch() == Triple::aarch64;
724 }
725
726 return false;
727 }
728
isImportDescriptor(StringRef Name)729 bool isImportDescriptor(StringRef Name) {
730 return Name.starts_with(ImportDescriptorPrefix) ||
731 Name == StringRef{NullImportDescriptorSymbolName} ||
732 (Name.starts_with(NullThunkDataPrefix) &&
733 Name.ends_with(NullThunkDataSuffix));
734 }
735
getSymbols(SymbolicFile * Obj,uint16_t Index,raw_ostream & SymNames,SymMap * SymMap)736 static Expected<std::vector<unsigned>> getSymbols(SymbolicFile *Obj,
737 uint16_t Index,
738 raw_ostream &SymNames,
739 SymMap *SymMap) {
740 std::vector<unsigned> Ret;
741
742 if (Obj == nullptr)
743 return Ret;
744
745 std::map<std::string, uint16_t> *Map = nullptr;
746 if (SymMap)
747 Map = SymMap->UseECMap && isECObject(*Obj) ? &SymMap->ECMap : &SymMap->Map;
748
749 for (const object::BasicSymbolRef &S : Obj->symbols()) {
750 if (!isArchiveSymbol(S))
751 continue;
752 if (Map) {
753 std::string Name;
754 raw_string_ostream NameStream(Name);
755 if (Error E = S.printName(NameStream))
756 return std::move(E);
757 if (Map->find(Name) != Map->end())
758 continue; // ignore duplicated symbol
759 (*Map)[Name] = Index;
760 if (Map == &SymMap->Map) {
761 Ret.push_back(SymNames.tell());
762 SymNames << Name << '\0';
763 // If EC is enabled, then the import descriptors are NOT put into EC
764 // objects so we need to copy them to the EC map manually.
765 if (SymMap->UseECMap && isImportDescriptor(Name))
766 SymMap->ECMap[Name] = Index;
767 }
768 } else {
769 Ret.push_back(SymNames.tell());
770 if (Error E = S.printName(SymNames))
771 return std::move(E);
772 SymNames << '\0';
773 }
774 }
775 return Ret;
776 }
777
778 static Expected<std::vector<MemberData>>
computeMemberData(raw_ostream & StringTable,raw_ostream & SymNames,object::Archive::Kind Kind,bool Thin,bool Deterministic,SymtabWritingMode NeedSymbols,SymMap * SymMap,LLVMContext & Context,ArrayRef<NewArchiveMember> NewMembers,std::optional<bool> IsEC,function_ref<void (Error)> Warn)779 computeMemberData(raw_ostream &StringTable, raw_ostream &SymNames,
780 object::Archive::Kind Kind, bool Thin, bool Deterministic,
781 SymtabWritingMode NeedSymbols, SymMap *SymMap,
782 LLVMContext &Context, ArrayRef<NewArchiveMember> NewMembers,
783 std::optional<bool> IsEC, function_ref<void(Error)> Warn) {
784 static char PaddingData[8] = {'\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n'};
785 uint64_t MemHeadPadSize = 0;
786 uint64_t Pos =
787 isAIXBigArchive(Kind) ? sizeof(object::BigArchive::FixLenHdr) : 0;
788
789 std::vector<MemberData> Ret;
790 bool HasObject = false;
791
792 // Deduplicate long member names in the string table and reuse earlier name
793 // offsets. This especially saves space for COFF Import libraries where all
794 // members have the same name.
795 StringMap<uint64_t> MemberNames;
796
797 // UniqueTimestamps is a special case to improve debugging on Darwin:
798 //
799 // The Darwin linker does not link debug info into the final
800 // binary. Instead, it emits entries of type N_OSO in the output
801 // binary's symbol table, containing references to the linked-in
802 // object files. Using that reference, the debugger can read the
803 // debug data directly from the object files. Alternatively, an
804 // invocation of 'dsymutil' will link the debug data from the object
805 // files into a dSYM bundle, which can be loaded by the debugger,
806 // instead of the object files.
807 //
808 // For an object file, the N_OSO entries contain the absolute path
809 // path to the file, and the file's timestamp. For an object
810 // included in an archive, the path is formatted like
811 // "/absolute/path/to/archive.a(member.o)", and the timestamp is the
812 // archive member's timestamp, rather than the archive's timestamp.
813 //
814 // However, this doesn't always uniquely identify an object within
815 // an archive -- an archive file can have multiple entries with the
816 // same filename. (This will happen commonly if the original object
817 // files started in different directories.) The only way they get
818 // distinguished, then, is via the timestamp. But this process is
819 // unable to find the correct object file in the archive when there
820 // are two files of the same name and timestamp.
821 //
822 // Additionally, timestamp==0 is treated specially, and causes the
823 // timestamp to be ignored as a match criteria.
824 //
825 // That will "usually" work out okay when creating an archive not in
826 // deterministic timestamp mode, because the objects will probably
827 // have been created at different timestamps.
828 //
829 // To ameliorate this problem, in deterministic archive mode (which
830 // is the default), on Darwin we will emit a unique non-zero
831 // timestamp for each entry with a duplicated name. This is still
832 // deterministic: the only thing affecting that timestamp is the
833 // order of the files in the resultant archive.
834 //
835 // See also the functions that handle the lookup:
836 // in lldb: ObjectContainerBSDArchive::Archive::FindObject()
837 // in llvm/tools/dsymutil: BinaryHolder::GetArchiveMemberBuffers().
838 bool UniqueTimestamps = Deterministic && isDarwin(Kind);
839 std::map<StringRef, unsigned> FilenameCount;
840 if (UniqueTimestamps) {
841 for (const NewArchiveMember &M : NewMembers)
842 FilenameCount[M.MemberName]++;
843 for (auto &Entry : FilenameCount)
844 Entry.second = Entry.second > 1 ? 1 : 0;
845 }
846
847 std::vector<std::unique_ptr<SymbolicFile>> SymFiles;
848
849 if (NeedSymbols != SymtabWritingMode::NoSymtab || isAIXBigArchive(Kind)) {
850 for (const NewArchiveMember &M : NewMembers) {
851 Expected<std::unique_ptr<SymbolicFile>> SymFileOrErr = getSymbolicFile(
852 M.Buf->getMemBufferRef(), Context, Kind, [&](Error Err) {
853 Warn(createFileError(M.MemberName, std::move(Err)));
854 });
855 if (!SymFileOrErr)
856 return createFileError(M.MemberName, SymFileOrErr.takeError());
857 SymFiles.push_back(std::move(*SymFileOrErr));
858 }
859 }
860
861 if (SymMap) {
862 if (IsEC) {
863 SymMap->UseECMap = *IsEC;
864 } else {
865 // When IsEC is not specified by the caller, use it when we have both
866 // any ARM64 object (ARM64 or ARM64EC) and any EC object (ARM64EC or
867 // AMD64). This may be a single ARM64EC object, but may also be separate
868 // ARM64 and AMD64 objects.
869 bool HaveArm64 = false, HaveEC = false;
870 for (std::unique_ptr<SymbolicFile> &SymFile : SymFiles) {
871 if (!SymFile)
872 continue;
873 if (!HaveArm64)
874 HaveArm64 = isAnyArm64COFF(*SymFile);
875 if (!HaveEC)
876 HaveEC = isECObject(*SymFile);
877 if (HaveArm64 && HaveEC) {
878 SymMap->UseECMap = true;
879 break;
880 }
881 }
882 }
883 }
884
885 // The big archive format needs to know the offset of the previous member
886 // header.
887 uint64_t PrevOffset = 0;
888 uint64_t NextMemHeadPadSize = 0;
889
890 for (uint32_t Index = 0; Index < NewMembers.size(); ++Index) {
891 const NewArchiveMember *M = &NewMembers[Index];
892 std::string Header;
893 raw_string_ostream Out(Header);
894
895 MemoryBufferRef Buf = M->Buf->getMemBufferRef();
896 StringRef Data = Thin ? "" : Buf.getBuffer();
897
898 // ld64 expects the members to be 8-byte aligned for 64-bit content and at
899 // least 4-byte aligned for 32-bit content. Opt for the larger encoding
900 // uniformly. This matches the behaviour with cctools and ensures that ld64
901 // is happy with archives that we generate.
902 unsigned MemberPadding =
903 isDarwin(Kind) ? offsetToAlignment(Data.size(), Align(8)) : 0;
904 unsigned TailPadding =
905 offsetToAlignment(Data.size() + MemberPadding, Align(2));
906 StringRef Padding = StringRef(PaddingData, MemberPadding + TailPadding);
907
908 sys::TimePoint<std::chrono::seconds> ModTime;
909 if (UniqueTimestamps)
910 // Increment timestamp for each file of a given name.
911 ModTime = sys::toTimePoint(FilenameCount[M->MemberName]++);
912 else
913 ModTime = M->ModTime;
914
915 uint64_t Size = Buf.getBufferSize() + MemberPadding;
916 if (Size > object::Archive::MaxMemberSize) {
917 std::string StringMsg =
918 "File " + M->MemberName.str() + " exceeds size limit";
919 return make_error<object::GenericBinaryError>(
920 std::move(StringMsg), object::object_error::parse_failed);
921 }
922
923 std::unique_ptr<SymbolicFile> CurSymFile;
924 if (!SymFiles.empty())
925 CurSymFile = std::move(SymFiles[Index]);
926
927 // In the big archive file format, we need to calculate and include the next
928 // member offset and previous member offset in the file member header.
929 if (isAIXBigArchive(Kind)) {
930 uint64_t OffsetToMemData = Pos + sizeof(object::BigArMemHdrType) +
931 alignTo(M->MemberName.size(), 2);
932
933 if (M == NewMembers.begin())
934 NextMemHeadPadSize =
935 alignToPowerOf2(OffsetToMemData,
936 getMemberAlignment(CurSymFile.get())) -
937 OffsetToMemData;
938
939 MemHeadPadSize = NextMemHeadPadSize;
940 Pos += MemHeadPadSize;
941 uint64_t NextOffset = Pos + sizeof(object::BigArMemHdrType) +
942 alignTo(M->MemberName.size(), 2) + alignTo(Size, 2);
943
944 // If there is another member file after this, we need to calculate the
945 // padding before the header.
946 if (Index + 1 != SymFiles.size()) {
947 uint64_t OffsetToNextMemData =
948 NextOffset + sizeof(object::BigArMemHdrType) +
949 alignTo(NewMembers[Index + 1].MemberName.size(), 2);
950 NextMemHeadPadSize =
951 alignToPowerOf2(OffsetToNextMemData,
952 getMemberAlignment(SymFiles[Index + 1].get())) -
953 OffsetToNextMemData;
954 NextOffset += NextMemHeadPadSize;
955 }
956 printBigArchiveMemberHeader(Out, M->MemberName, ModTime, M->UID, M->GID,
957 M->Perms, Size, PrevOffset, NextOffset);
958 PrevOffset = Pos;
959 } else {
960 printMemberHeader(Out, Pos, StringTable, MemberNames, Kind, Thin, *M,
961 ModTime, Size);
962 }
963 Out.flush();
964
965 std::vector<unsigned> Symbols;
966 if (NeedSymbols != SymtabWritingMode::NoSymtab) {
967 Expected<std::vector<unsigned>> SymbolsOrErr =
968 getSymbols(CurSymFile.get(), Index + 1, SymNames, SymMap);
969 if (!SymbolsOrErr)
970 return createFileError(M->MemberName, SymbolsOrErr.takeError());
971 Symbols = std::move(*SymbolsOrErr);
972 if (CurSymFile)
973 HasObject = true;
974 }
975
976 Pos += Header.size() + Data.size() + Padding.size();
977 Ret.push_back({std::move(Symbols), std::move(Header), Data, Padding,
978 MemHeadPadSize, std::move(CurSymFile)});
979 }
980 // If there are no symbols, emit an empty symbol table, to satisfy Solaris
981 // tools, older versions of which expect a symbol table in a non-empty
982 // archive, regardless of whether there are any symbols in it.
983 if (HasObject && SymNames.tell() == 0 && !isCOFFArchive(Kind))
984 SymNames << '\0' << '\0' << '\0';
985 return std::move(Ret);
986 }
987
988 namespace llvm {
989
canonicalizePath(StringRef P)990 static ErrorOr<SmallString<128>> canonicalizePath(StringRef P) {
991 SmallString<128> Ret = P;
992 std::error_code Err = sys::fs::make_absolute(Ret);
993 if (Err)
994 return Err;
995 sys::path::remove_dots(Ret, /*removedotdot*/ true);
996 return Ret;
997 }
998
999 // Compute the relative path from From to To.
computeArchiveRelativePath(StringRef From,StringRef To)1000 Expected<std::string> computeArchiveRelativePath(StringRef From, StringRef To) {
1001 ErrorOr<SmallString<128>> PathToOrErr = canonicalizePath(To);
1002 ErrorOr<SmallString<128>> DirFromOrErr = canonicalizePath(From);
1003 if (!PathToOrErr || !DirFromOrErr)
1004 return errorCodeToError(errnoAsErrorCode());
1005
1006 const SmallString<128> &PathTo = *PathToOrErr;
1007 const SmallString<128> &DirFrom = sys::path::parent_path(*DirFromOrErr);
1008
1009 // Can't construct a relative path between different roots
1010 if (sys::path::root_name(PathTo) != sys::path::root_name(DirFrom))
1011 return sys::path::convert_to_slash(PathTo);
1012
1013 // Skip common prefixes
1014 auto FromTo =
1015 std::mismatch(sys::path::begin(DirFrom), sys::path::end(DirFrom),
1016 sys::path::begin(PathTo));
1017 auto FromI = FromTo.first;
1018 auto ToI = FromTo.second;
1019
1020 // Construct relative path
1021 SmallString<128> Relative;
1022 for (auto FromE = sys::path::end(DirFrom); FromI != FromE; ++FromI)
1023 sys::path::append(Relative, sys::path::Style::posix, "..");
1024
1025 for (auto ToE = sys::path::end(PathTo); ToI != ToE; ++ToI)
1026 sys::path::append(Relative, sys::path::Style::posix, *ToI);
1027
1028 return std::string(Relative);
1029 }
1030
writeArchiveToStream(raw_ostream & Out,ArrayRef<NewArchiveMember> NewMembers,SymtabWritingMode WriteSymtab,object::Archive::Kind Kind,bool Deterministic,bool Thin,std::optional<bool> IsEC,function_ref<void (Error)> Warn)1031 Error writeArchiveToStream(raw_ostream &Out,
1032 ArrayRef<NewArchiveMember> NewMembers,
1033 SymtabWritingMode WriteSymtab,
1034 object::Archive::Kind Kind, bool Deterministic,
1035 bool Thin, std::optional<bool> IsEC,
1036 function_ref<void(Error)> Warn) {
1037 assert((!Thin || !isBSDLike(Kind)) && "Only the gnu format has a thin mode");
1038
1039 SmallString<0> SymNamesBuf;
1040 raw_svector_ostream SymNames(SymNamesBuf);
1041 SmallString<0> StringTableBuf;
1042 raw_svector_ostream StringTable(StringTableBuf);
1043 SymMap SymMap;
1044 bool ShouldWriteSymtab = WriteSymtab != SymtabWritingMode::NoSymtab;
1045
1046 // COFF symbol map uses 16-bit indexes, so we can't use it if there are too
1047 // many members. COFF format also requires symbol table presence, so use
1048 // GNU format when NoSymtab is requested.
1049 if (isCOFFArchive(Kind) && (NewMembers.size() > 0xfffe || !ShouldWriteSymtab))
1050 Kind = object::Archive::K_GNU;
1051
1052 // In the scenario when LLVMContext is populated SymbolicFile will contain a
1053 // reference to it, thus SymbolicFile should be destroyed first.
1054 LLVMContext Context;
1055
1056 Expected<std::vector<MemberData>> DataOrErr = computeMemberData(
1057 StringTable, SymNames, Kind, Thin, Deterministic, WriteSymtab,
1058 isCOFFArchive(Kind) ? &SymMap : nullptr, Context, NewMembers, IsEC, Warn);
1059 if (Error E = DataOrErr.takeError())
1060 return E;
1061 std::vector<MemberData> &Data = *DataOrErr;
1062
1063 uint64_t StringTableSize = 0;
1064 MemberData StringTableMember;
1065 if (!StringTableBuf.empty() && !isAIXBigArchive(Kind)) {
1066 StringTableMember = computeStringTable(StringTableBuf);
1067 StringTableSize = StringTableMember.Header.size() +
1068 StringTableMember.Data.size() +
1069 StringTableMember.Padding.size();
1070 }
1071
1072 // We would like to detect if we need to switch to a 64-bit symbol table.
1073 uint64_t LastMemberEndOffset = 0;
1074 uint64_t LastMemberHeaderOffset = 0;
1075 uint64_t NumSyms = 0;
1076 uint64_t NumSyms32 = 0; // Store symbol number of 32-bit member files.
1077
1078 for (const auto &M : Data) {
1079 // Record the start of the member's offset
1080 LastMemberEndOffset += M.PreHeadPadSize;
1081 LastMemberHeaderOffset = LastMemberEndOffset;
1082 // Account for the size of each part associated with the member.
1083 LastMemberEndOffset += M.Header.size() + M.Data.size() + M.Padding.size();
1084 NumSyms += M.Symbols.size();
1085
1086 // AIX big archive files may contain two global symbol tables. The
1087 // first global symbol table locates 32-bit file members that define global
1088 // symbols; the second global symbol table does the same for 64-bit file
1089 // members. As a big archive can have both 32-bit and 64-bit file members,
1090 // we need to know the number of symbols in each symbol table individually.
1091 if (isAIXBigArchive(Kind) && ShouldWriteSymtab) {
1092 if (!is64BitSymbolicFile(M.SymFile.get()))
1093 NumSyms32 += M.Symbols.size();
1094 }
1095 }
1096
1097 std::optional<uint64_t> HeadersSize;
1098
1099 // The symbol table is put at the end of the big archive file. The symbol
1100 // table is at the start of the archive file for other archive formats.
1101 if (ShouldWriteSymtab && !is64BitKind(Kind)) {
1102 // We assume 32-bit offsets to see if 32-bit symbols are possible or not.
1103 HeadersSize = computeHeadersSize(Kind, Data.size(), StringTableSize,
1104 NumSyms, SymNamesBuf.size(),
1105 isCOFFArchive(Kind) ? &SymMap : nullptr);
1106
1107 // The SYM64 format is used when an archive's member offsets are larger than
1108 // 32-bits can hold. The need for this shift in format is detected by
1109 // writeArchive. To test this we need to generate a file with a member that
1110 // has an offset larger than 32-bits but this demands a very slow test. To
1111 // speed the test up we use this environment variable to pretend like the
1112 // cutoff happens before 32-bits and instead happens at some much smaller
1113 // value.
1114 uint64_t Sym64Threshold = 1ULL << 32;
1115 const char *Sym64Env = std::getenv("SYM64_THRESHOLD");
1116 if (Sym64Env)
1117 StringRef(Sym64Env).getAsInteger(10, Sym64Threshold);
1118
1119 // If LastMemberHeaderOffset isn't going to fit in a 32-bit varible we need
1120 // to switch to 64-bit. Note that the file can be larger than 4GB as long as
1121 // the last member starts before the 4GB offset.
1122 if (*HeadersSize + LastMemberHeaderOffset >= Sym64Threshold) {
1123 if (Kind == object::Archive::K_DARWIN)
1124 Kind = object::Archive::K_DARWIN64;
1125 else
1126 Kind = object::Archive::K_GNU64;
1127 HeadersSize.reset();
1128 }
1129 }
1130
1131 if (Thin)
1132 Out << "!<thin>\n";
1133 else if (isAIXBigArchive(Kind))
1134 Out << "<bigaf>\n";
1135 else
1136 Out << "!<arch>\n";
1137
1138 if (!isAIXBigArchive(Kind)) {
1139 if (ShouldWriteSymtab) {
1140 if (!HeadersSize)
1141 HeadersSize = computeHeadersSize(
1142 Kind, Data.size(), StringTableSize, NumSyms, SymNamesBuf.size(),
1143 isCOFFArchive(Kind) ? &SymMap : nullptr);
1144 writeSymbolTable(Out, Kind, Deterministic, Data, SymNamesBuf,
1145 *HeadersSize, NumSyms);
1146
1147 if (isCOFFArchive(Kind))
1148 writeSymbolMap(Out, Kind, Deterministic, Data, SymMap, *HeadersSize);
1149 }
1150
1151 if (StringTableSize)
1152 Out << StringTableMember.Header << StringTableMember.Data
1153 << StringTableMember.Padding;
1154
1155 if (ShouldWriteSymtab && SymMap.ECMap.size())
1156 writeECSymbols(Out, Kind, Deterministic, Data, SymMap);
1157
1158 for (const MemberData &M : Data)
1159 Out << M.Header << M.Data << M.Padding;
1160 } else {
1161 HeadersSize = sizeof(object::BigArchive::FixLenHdr);
1162 LastMemberEndOffset += *HeadersSize;
1163 LastMemberHeaderOffset += *HeadersSize;
1164
1165 // For the big archive (AIX) format, compute a table of member names and
1166 // offsets, used in the member table.
1167 uint64_t MemberTableNameStrTblSize = 0;
1168 std::vector<size_t> MemberOffsets;
1169 std::vector<StringRef> MemberNames;
1170 // Loop across object to find offset and names.
1171 uint64_t MemberEndOffset = sizeof(object::BigArchive::FixLenHdr);
1172 for (size_t I = 0, Size = NewMembers.size(); I != Size; ++I) {
1173 const NewArchiveMember &Member = NewMembers[I];
1174 MemberTableNameStrTblSize += Member.MemberName.size() + 1;
1175 MemberEndOffset += Data[I].PreHeadPadSize;
1176 MemberOffsets.push_back(MemberEndOffset);
1177 MemberNames.push_back(Member.MemberName);
1178 // File member name ended with "`\n". The length is included in
1179 // BigArMemHdrType.
1180 MemberEndOffset += sizeof(object::BigArMemHdrType) +
1181 alignTo(Data[I].Data.size(), 2) +
1182 alignTo(Member.MemberName.size(), 2);
1183 }
1184
1185 // AIX member table size.
1186 uint64_t MemberTableSize = 20 + // Number of members field
1187 20 * MemberOffsets.size() +
1188 MemberTableNameStrTblSize;
1189
1190 SmallString<0> SymNamesBuf32;
1191 SmallString<0> SymNamesBuf64;
1192 raw_svector_ostream SymNames32(SymNamesBuf32);
1193 raw_svector_ostream SymNames64(SymNamesBuf64);
1194
1195 if (ShouldWriteSymtab && NumSyms)
1196 // Generate the symbol names for the members.
1197 for (const auto &M : Data) {
1198 Expected<std::vector<unsigned>> SymbolsOrErr = getSymbols(
1199 M.SymFile.get(), 0,
1200 is64BitSymbolicFile(M.SymFile.get()) ? SymNames64 : SymNames32,
1201 nullptr);
1202 if (!SymbolsOrErr)
1203 return SymbolsOrErr.takeError();
1204 }
1205
1206 uint64_t MemberTableEndOffset =
1207 LastMemberEndOffset +
1208 alignTo(sizeof(object::BigArMemHdrType) + MemberTableSize, 2);
1209
1210 // In AIX OS, The 'GlobSymOffset' field in the fixed-length header contains
1211 // the offset to the 32-bit global symbol table, and the 'GlobSym64Offset'
1212 // contains the offset to the 64-bit global symbol table.
1213 uint64_t GlobalSymbolOffset =
1214 (ShouldWriteSymtab &&
1215 (WriteSymtab != SymtabWritingMode::BigArchive64) && NumSyms32 > 0)
1216 ? MemberTableEndOffset
1217 : 0;
1218
1219 uint64_t GlobalSymbolOffset64 = 0;
1220 uint64_t NumSyms64 = NumSyms - NumSyms32;
1221 if (ShouldWriteSymtab && (WriteSymtab != SymtabWritingMode::BigArchive32) &&
1222 NumSyms64 > 0) {
1223 if (GlobalSymbolOffset == 0)
1224 GlobalSymbolOffset64 = MemberTableEndOffset;
1225 else
1226 // If there is a global symbol table for 32-bit members,
1227 // the 64-bit global symbol table is after the 32-bit one.
1228 GlobalSymbolOffset64 =
1229 GlobalSymbolOffset + sizeof(object::BigArMemHdrType) +
1230 (NumSyms32 + 1) * 8 + alignTo(SymNamesBuf32.size(), 2);
1231 }
1232
1233 // Fixed Sized Header.
1234 printWithSpacePadding(Out, NewMembers.size() ? LastMemberEndOffset : 0,
1235 20); // Offset to member table
1236 // If there are no file members in the archive, there will be no global
1237 // symbol table.
1238 printWithSpacePadding(Out, GlobalSymbolOffset, 20);
1239 printWithSpacePadding(Out, GlobalSymbolOffset64, 20);
1240 printWithSpacePadding(Out,
1241 NewMembers.size()
1242 ? sizeof(object::BigArchive::FixLenHdr) +
1243 Data[0].PreHeadPadSize
1244 : 0,
1245 20); // Offset to first archive member
1246 printWithSpacePadding(Out, NewMembers.size() ? LastMemberHeaderOffset : 0,
1247 20); // Offset to last archive member
1248 printWithSpacePadding(
1249 Out, 0,
1250 20); // Offset to first member of free list - Not supported yet
1251
1252 for (const MemberData &M : Data) {
1253 Out << std::string(M.PreHeadPadSize, '\0');
1254 Out << M.Header << M.Data;
1255 if (M.Data.size() % 2)
1256 Out << '\0';
1257 }
1258
1259 if (NewMembers.size()) {
1260 // Member table.
1261 printBigArchiveMemberHeader(Out, "", sys::toTimePoint(0), 0, 0, 0,
1262 MemberTableSize, LastMemberHeaderOffset,
1263 GlobalSymbolOffset ? GlobalSymbolOffset
1264 : GlobalSymbolOffset64);
1265 printWithSpacePadding(Out, MemberOffsets.size(), 20); // Number of members
1266 for (uint64_t MemberOffset : MemberOffsets)
1267 printWithSpacePadding(Out, MemberOffset,
1268 20); // Offset to member file header.
1269 for (StringRef MemberName : MemberNames)
1270 Out << MemberName << '\0'; // Member file name, null byte padding.
1271
1272 if (MemberTableNameStrTblSize % 2)
1273 Out << '\0'; // Name table must be tail padded to an even number of
1274 // bytes.
1275
1276 if (ShouldWriteSymtab) {
1277 // Write global symbol table for 32-bit file members.
1278 if (GlobalSymbolOffset) {
1279 writeSymbolTable(Out, Kind, Deterministic, Data, SymNamesBuf32,
1280 *HeadersSize, NumSyms32, LastMemberEndOffset,
1281 GlobalSymbolOffset64);
1282 // Add padding between the symbol tables, if needed.
1283 if (GlobalSymbolOffset64 && (SymNamesBuf32.size() % 2))
1284 Out << '\0';
1285 }
1286
1287 // Write global symbol table for 64-bit file members.
1288 if (GlobalSymbolOffset64)
1289 writeSymbolTable(Out, Kind, Deterministic, Data, SymNamesBuf64,
1290 *HeadersSize, NumSyms64,
1291 GlobalSymbolOffset ? GlobalSymbolOffset
1292 : LastMemberEndOffset,
1293 0, true);
1294 }
1295 }
1296 }
1297 Out.flush();
1298 return Error::success();
1299 }
1300
warnToStderr(Error Err)1301 void warnToStderr(Error Err) {
1302 llvm::logAllUnhandledErrors(std::move(Err), llvm::errs(), "warning: ");
1303 }
1304
writeArchive(StringRef ArcName,ArrayRef<NewArchiveMember> NewMembers,SymtabWritingMode WriteSymtab,object::Archive::Kind Kind,bool Deterministic,bool Thin,std::unique_ptr<MemoryBuffer> OldArchiveBuf,std::optional<bool> IsEC,function_ref<void (Error)> Warn)1305 Error writeArchive(StringRef ArcName, ArrayRef<NewArchiveMember> NewMembers,
1306 SymtabWritingMode WriteSymtab, object::Archive::Kind Kind,
1307 bool Deterministic, bool Thin,
1308 std::unique_ptr<MemoryBuffer> OldArchiveBuf,
1309 std::optional<bool> IsEC, function_ref<void(Error)> Warn) {
1310 Expected<sys::fs::TempFile> Temp =
1311 sys::fs::TempFile::create(ArcName + ".temp-archive-%%%%%%%.a");
1312 if (!Temp)
1313 return Temp.takeError();
1314 raw_fd_ostream Out(Temp->FD, false);
1315
1316 if (Error E = writeArchiveToStream(Out, NewMembers, WriteSymtab, Kind,
1317 Deterministic, Thin, IsEC, Warn)) {
1318 if (Error DiscardError = Temp->discard())
1319 return joinErrors(std::move(E), std::move(DiscardError));
1320 return E;
1321 }
1322
1323 // At this point, we no longer need whatever backing memory
1324 // was used to generate the NewMembers. On Windows, this buffer
1325 // could be a mapped view of the file we want to replace (if
1326 // we're updating an existing archive, say). In that case, the
1327 // rename would still succeed, but it would leave behind a
1328 // temporary file (actually the original file renamed) because
1329 // a file cannot be deleted while there's a handle open on it,
1330 // only renamed. So by freeing this buffer, this ensures that
1331 // the last open handle on the destination file, if any, is
1332 // closed before we attempt to rename.
1333 OldArchiveBuf.reset();
1334
1335 return Temp->keep(ArcName);
1336 }
1337
1338 Expected<std::unique_ptr<MemoryBuffer>>
writeArchiveToBuffer(ArrayRef<NewArchiveMember> NewMembers,SymtabWritingMode WriteSymtab,object::Archive::Kind Kind,bool Deterministic,bool Thin,function_ref<void (Error)> Warn)1339 writeArchiveToBuffer(ArrayRef<NewArchiveMember> NewMembers,
1340 SymtabWritingMode WriteSymtab, object::Archive::Kind Kind,
1341 bool Deterministic, bool Thin,
1342 function_ref<void(Error)> Warn) {
1343 SmallVector<char, 0> ArchiveBufferVector;
1344 raw_svector_ostream ArchiveStream(ArchiveBufferVector);
1345
1346 if (Error E =
1347 writeArchiveToStream(ArchiveStream, NewMembers, WriteSymtab, Kind,
1348 Deterministic, Thin, std::nullopt, Warn))
1349 return std::move(E);
1350
1351 return std::make_unique<SmallVectorMemoryBuffer>(
1352 std::move(ArchiveBufferVector), /*RequiresNullTerminator=*/false);
1353 }
1354
1355 } // namespace llvm
1356