1 //===- ArchiveWriter.cpp - ar File Format implementation --------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file defines the writeArchive function.
10 //
11 //===----------------------------------------------------------------------===//
12
13 #include "llvm/Object/ArchiveWriter.h"
14 #include "llvm/ADT/ArrayRef.h"
15 #include "llvm/ADT/StringMap.h"
16 #include "llvm/ADT/StringRef.h"
17 #include "llvm/BinaryFormat/Magic.h"
18 #include "llvm/IR/LLVMContext.h"
19 #include "llvm/Object/Archive.h"
20 #include "llvm/Object/Error.h"
21 #include "llvm/Object/IRObjectFile.h"
22 #include "llvm/Object/MachO.h"
23 #include "llvm/Object/ObjectFile.h"
24 #include "llvm/Object/SymbolicFile.h"
25 #include "llvm/Object/XCOFFObjectFile.h"
26 #include "llvm/Support/Alignment.h"
27 #include "llvm/Support/EndianStream.h"
28 #include "llvm/Support/Errc.h"
29 #include "llvm/Support/ErrorHandling.h"
30 #include "llvm/Support/Format.h"
31 #include "llvm/Support/MathExtras.h"
32 #include "llvm/Support/Path.h"
33 #include "llvm/Support/SmallVectorMemoryBuffer.h"
34 #include "llvm/Support/raw_ostream.h"
35
36 #include <map>
37
38 #if !defined(_MSC_VER) && !defined(__MINGW32__)
39 #include <unistd.h>
40 #else
41 #include <io.h>
42 #endif
43
44 using namespace llvm;
45
NewArchiveMember(MemoryBufferRef BufRef)46 NewArchiveMember::NewArchiveMember(MemoryBufferRef BufRef)
47 : Buf(MemoryBuffer::getMemBuffer(BufRef, false)),
48 MemberName(BufRef.getBufferIdentifier()) {}
49
detectKindFromObject() const50 object::Archive::Kind NewArchiveMember::detectKindFromObject() const {
51 auto MemBufferRef = this->Buf->getMemBufferRef();
52 Expected<std::unique_ptr<object::ObjectFile>> OptionalObject =
53 object::ObjectFile::createObjectFile(MemBufferRef);
54
55 if (OptionalObject)
56 return isa<object::MachOObjectFile>(**OptionalObject)
57 ? object::Archive::K_DARWIN
58 : (isa<object::XCOFFObjectFile>(**OptionalObject)
59 ? object::Archive::K_AIXBIG
60 : object::Archive::K_GNU);
61
62 // Squelch the error in case we had a non-object file.
63 consumeError(OptionalObject.takeError());
64
65 // If we're adding a bitcode file to the archive, detect the Archive kind
66 // based on the target triple.
67 LLVMContext Context;
68 if (identify_magic(MemBufferRef.getBuffer()) == file_magic::bitcode) {
69 if (auto ObjOrErr = object::SymbolicFile::createSymbolicFile(
70 MemBufferRef, file_magic::bitcode, &Context)) {
71 auto &IRObject = cast<object::IRObjectFile>(**ObjOrErr);
72 return Triple(IRObject.getTargetTriple()).isOSDarwin()
73 ? object::Archive::K_DARWIN
74 : object::Archive::K_GNU;
75 } else {
76 // Squelch the error in case this was not a SymbolicFile.
77 consumeError(ObjOrErr.takeError());
78 }
79 }
80
81 return object::Archive::getDefaultKindForHost();
82 }
83
84 Expected<NewArchiveMember>
getOldMember(const object::Archive::Child & OldMember,bool Deterministic)85 NewArchiveMember::getOldMember(const object::Archive::Child &OldMember,
86 bool Deterministic) {
87 Expected<llvm::MemoryBufferRef> BufOrErr = OldMember.getMemoryBufferRef();
88 if (!BufOrErr)
89 return BufOrErr.takeError();
90
91 NewArchiveMember M;
92 M.Buf = MemoryBuffer::getMemBuffer(*BufOrErr, false);
93 M.MemberName = M.Buf->getBufferIdentifier();
94 if (!Deterministic) {
95 auto ModTimeOrErr = OldMember.getLastModified();
96 if (!ModTimeOrErr)
97 return ModTimeOrErr.takeError();
98 M.ModTime = ModTimeOrErr.get();
99 Expected<unsigned> UIDOrErr = OldMember.getUID();
100 if (!UIDOrErr)
101 return UIDOrErr.takeError();
102 M.UID = UIDOrErr.get();
103 Expected<unsigned> GIDOrErr = OldMember.getGID();
104 if (!GIDOrErr)
105 return GIDOrErr.takeError();
106 M.GID = GIDOrErr.get();
107 Expected<sys::fs::perms> AccessModeOrErr = OldMember.getAccessMode();
108 if (!AccessModeOrErr)
109 return AccessModeOrErr.takeError();
110 M.Perms = AccessModeOrErr.get();
111 }
112 return std::move(M);
113 }
114
getFile(StringRef FileName,bool Deterministic)115 Expected<NewArchiveMember> NewArchiveMember::getFile(StringRef FileName,
116 bool Deterministic) {
117 sys::fs::file_status Status;
118 auto FDOrErr = sys::fs::openNativeFileForRead(FileName);
119 if (!FDOrErr)
120 return FDOrErr.takeError();
121 sys::fs::file_t FD = *FDOrErr;
122 assert(FD != sys::fs::kInvalidFile);
123
124 if (auto EC = sys::fs::status(FD, Status))
125 return errorCodeToError(EC);
126
127 // Opening a directory doesn't make sense. Let it fail.
128 // Linux cannot open directories with open(2), although
129 // cygwin and *bsd can.
130 if (Status.type() == sys::fs::file_type::directory_file)
131 return errorCodeToError(make_error_code(errc::is_a_directory));
132
133 ErrorOr<std::unique_ptr<MemoryBuffer>> MemberBufferOrErr =
134 MemoryBuffer::getOpenFile(FD, FileName, Status.getSize(), false);
135 if (!MemberBufferOrErr)
136 return errorCodeToError(MemberBufferOrErr.getError());
137
138 if (auto EC = sys::fs::closeFile(FD))
139 return errorCodeToError(EC);
140
141 NewArchiveMember M;
142 M.Buf = std::move(*MemberBufferOrErr);
143 M.MemberName = M.Buf->getBufferIdentifier();
144 if (!Deterministic) {
145 M.ModTime = std::chrono::time_point_cast<std::chrono::seconds>(
146 Status.getLastModificationTime());
147 M.UID = Status.getUser();
148 M.GID = Status.getGroup();
149 M.Perms = Status.permissions();
150 }
151 return std::move(M);
152 }
153
154 template <typename T>
printWithSpacePadding(raw_ostream & OS,T Data,unsigned Size)155 static void printWithSpacePadding(raw_ostream &OS, T Data, unsigned Size) {
156 uint64_t OldPos = OS.tell();
157 OS << Data;
158 unsigned SizeSoFar = OS.tell() - OldPos;
159 assert(SizeSoFar <= Size && "Data doesn't fit in Size");
160 OS.indent(Size - SizeSoFar);
161 }
162
isDarwin(object::Archive::Kind Kind)163 static bool isDarwin(object::Archive::Kind Kind) {
164 return Kind == object::Archive::K_DARWIN ||
165 Kind == object::Archive::K_DARWIN64;
166 }
167
isAIXBigArchive(object::Archive::Kind Kind)168 static bool isAIXBigArchive(object::Archive::Kind Kind) {
169 return Kind == object::Archive::K_AIXBIG;
170 }
171
isBSDLike(object::Archive::Kind Kind)172 static bool isBSDLike(object::Archive::Kind Kind) {
173 switch (Kind) {
174 case object::Archive::K_GNU:
175 case object::Archive::K_GNU64:
176 case object::Archive::K_AIXBIG:
177 return false;
178 case object::Archive::K_BSD:
179 case object::Archive::K_DARWIN:
180 case object::Archive::K_DARWIN64:
181 return true;
182 case object::Archive::K_COFF:
183 break;
184 }
185 llvm_unreachable("not supported for writting");
186 }
187
188 template <class T>
print(raw_ostream & Out,object::Archive::Kind Kind,T Val)189 static void print(raw_ostream &Out, object::Archive::Kind Kind, T Val) {
190 support::endian::write(Out, Val,
191 isBSDLike(Kind) ? support::little : support::big);
192 }
193
printRestOfMemberHeader(raw_ostream & Out,const sys::TimePoint<std::chrono::seconds> & ModTime,unsigned UID,unsigned GID,unsigned Perms,uint64_t Size)194 static void printRestOfMemberHeader(
195 raw_ostream &Out, const sys::TimePoint<std::chrono::seconds> &ModTime,
196 unsigned UID, unsigned GID, unsigned Perms, uint64_t Size) {
197 printWithSpacePadding(Out, sys::toTimeT(ModTime), 12);
198
199 // The format has only 6 chars for uid and gid. Truncate if the provided
200 // values don't fit.
201 printWithSpacePadding(Out, UID % 1000000, 6);
202 printWithSpacePadding(Out, GID % 1000000, 6);
203
204 printWithSpacePadding(Out, format("%o", Perms), 8);
205 printWithSpacePadding(Out, Size, 10);
206 Out << "`\n";
207 }
208
209 static void
printGNUSmallMemberHeader(raw_ostream & Out,StringRef Name,const sys::TimePoint<std::chrono::seconds> & ModTime,unsigned UID,unsigned GID,unsigned Perms,uint64_t Size)210 printGNUSmallMemberHeader(raw_ostream &Out, StringRef Name,
211 const sys::TimePoint<std::chrono::seconds> &ModTime,
212 unsigned UID, unsigned GID, unsigned Perms,
213 uint64_t Size) {
214 printWithSpacePadding(Out, Twine(Name) + "/", 16);
215 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms, Size);
216 }
217
218 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)219 printBSDMemberHeader(raw_ostream &Out, uint64_t Pos, StringRef Name,
220 const sys::TimePoint<std::chrono::seconds> &ModTime,
221 unsigned UID, unsigned GID, unsigned Perms, uint64_t Size) {
222 uint64_t PosAfterHeader = Pos + 60 + Name.size();
223 // Pad so that even 64 bit object files are aligned.
224 unsigned Pad = offsetToAlignment(PosAfterHeader, Align(8));
225 unsigned NameWithPadding = Name.size() + Pad;
226 printWithSpacePadding(Out, Twine("#1/") + Twine(NameWithPadding), 16);
227 printRestOfMemberHeader(Out, ModTime, UID, GID, Perms,
228 NameWithPadding + Size);
229 Out << Name;
230 while (Pad--)
231 Out.write(uint8_t(0));
232 }
233
234 static void
printBigArchiveMemberHeader(raw_ostream & Out,StringRef Name,const sys::TimePoint<std::chrono::seconds> & ModTime,unsigned UID,unsigned GID,unsigned Perms,uint64_t Size,unsigned PrevOffset,unsigned NextOffset)235 printBigArchiveMemberHeader(raw_ostream &Out, StringRef Name,
236 const sys::TimePoint<std::chrono::seconds> &ModTime,
237 unsigned UID, unsigned GID, unsigned Perms,
238 uint64_t Size, unsigned PrevOffset,
239 unsigned NextOffset) {
240 unsigned NameLen = Name.size();
241
242 printWithSpacePadding(Out, Size, 20); // File member size
243 printWithSpacePadding(Out, NextOffset, 20); // Next member header offset
244 printWithSpacePadding(Out, PrevOffset, 20); // Previous member header offset
245 printWithSpacePadding(Out, sys::toTimeT(ModTime), 12); // File member date
246 // The big archive format has 12 chars for uid and gid.
247 printWithSpacePadding(Out, UID % 1000000000000, 12); // UID
248 printWithSpacePadding(Out, GID % 1000000000000, 12); // GID
249 printWithSpacePadding(Out, format("%o", Perms), 12); // Permission
250 printWithSpacePadding(Out, NameLen, 4); // Name length
251 if (NameLen) {
252 printWithSpacePadding(Out, Name, NameLen); // Name
253 if (NameLen % 2)
254 Out.write(uint8_t(0)); // Null byte padding
255 }
256 Out << "`\n"; // Terminator
257 }
258
useStringTable(bool Thin,StringRef Name)259 static bool useStringTable(bool Thin, StringRef Name) {
260 return Thin || Name.size() >= 16 || Name.contains('/');
261 }
262
is64BitKind(object::Archive::Kind Kind)263 static bool is64BitKind(object::Archive::Kind Kind) {
264 switch (Kind) {
265 case object::Archive::K_GNU:
266 case object::Archive::K_BSD:
267 case object::Archive::K_DARWIN:
268 case object::Archive::K_COFF:
269 return false;
270 case object::Archive::K_AIXBIG:
271 case object::Archive::K_DARWIN64:
272 case object::Archive::K_GNU64:
273 return true;
274 }
275 llvm_unreachable("not supported for writting");
276 }
277
278 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)279 printMemberHeader(raw_ostream &Out, uint64_t Pos, raw_ostream &StringTable,
280 StringMap<uint64_t> &MemberNames, object::Archive::Kind Kind,
281 bool Thin, const NewArchiveMember &M,
282 sys::TimePoint<std::chrono::seconds> ModTime, uint64_t Size) {
283 if (isBSDLike(Kind))
284 return printBSDMemberHeader(Out, Pos, M.MemberName, ModTime, M.UID, M.GID,
285 M.Perms, Size);
286 if (!useStringTable(Thin, M.MemberName))
287 return printGNUSmallMemberHeader(Out, M.MemberName, ModTime, M.UID, M.GID,
288 M.Perms, Size);
289 Out << '/';
290 uint64_t NamePos;
291 if (Thin) {
292 NamePos = StringTable.tell();
293 StringTable << M.MemberName << "/\n";
294 } else {
295 auto Insertion = MemberNames.insert({M.MemberName, uint64_t(0)});
296 if (Insertion.second) {
297 Insertion.first->second = StringTable.tell();
298 StringTable << M.MemberName << "/\n";
299 }
300 NamePos = Insertion.first->second;
301 }
302 printWithSpacePadding(Out, NamePos, 15);
303 printRestOfMemberHeader(Out, ModTime, M.UID, M.GID, M.Perms, Size);
304 }
305
306 namespace {
307 struct MemberData {
308 std::vector<unsigned> Symbols;
309 std::string Header;
310 StringRef Data;
311 StringRef Padding;
312 };
313 } // namespace
314
computeStringTable(StringRef Names)315 static MemberData computeStringTable(StringRef Names) {
316 unsigned Size = Names.size();
317 unsigned Pad = offsetToAlignment(Size, Align(2));
318 std::string Header;
319 raw_string_ostream Out(Header);
320 printWithSpacePadding(Out, "//", 48);
321 printWithSpacePadding(Out, Size + Pad, 10);
322 Out << "`\n";
323 Out.flush();
324 return {{}, std::move(Header), Names, Pad ? "\n" : ""};
325 }
326
now(bool Deterministic)327 static sys::TimePoint<std::chrono::seconds> now(bool Deterministic) {
328 using namespace std::chrono;
329
330 if (!Deterministic)
331 return time_point_cast<seconds>(system_clock::now());
332 return sys::TimePoint<seconds>();
333 }
334
isArchiveSymbol(const object::BasicSymbolRef & S)335 static bool isArchiveSymbol(const object::BasicSymbolRef &S) {
336 Expected<uint32_t> SymFlagsOrErr = S.getFlags();
337 if (!SymFlagsOrErr)
338 // TODO: Actually report errors helpfully.
339 report_fatal_error(SymFlagsOrErr.takeError());
340 if (*SymFlagsOrErr & object::SymbolRef::SF_FormatSpecific)
341 return false;
342 if (!(*SymFlagsOrErr & object::SymbolRef::SF_Global))
343 return false;
344 if (*SymFlagsOrErr & object::SymbolRef::SF_Undefined)
345 return false;
346 return true;
347 }
348
printNBits(raw_ostream & Out,object::Archive::Kind Kind,uint64_t Val)349 static void printNBits(raw_ostream &Out, object::Archive::Kind Kind,
350 uint64_t Val) {
351 if (is64BitKind(Kind))
352 print<uint64_t>(Out, Kind, Val);
353 else
354 print<uint32_t>(Out, Kind, Val);
355 }
356
computeSymbolTableSize(object::Archive::Kind Kind,uint64_t NumSyms,uint64_t OffsetSize,StringRef StringTable,uint32_t * Padding=nullptr)357 static uint64_t computeSymbolTableSize(object::Archive::Kind Kind,
358 uint64_t NumSyms, uint64_t OffsetSize,
359 StringRef StringTable,
360 uint32_t *Padding = nullptr) {
361 assert((OffsetSize == 4 || OffsetSize == 8) && "Unsupported OffsetSize");
362 uint64_t Size = OffsetSize; // Number of entries
363 if (isBSDLike(Kind))
364 Size += NumSyms * OffsetSize * 2; // Table
365 else
366 Size += NumSyms * OffsetSize; // Table
367 if (isBSDLike(Kind))
368 Size += OffsetSize; // byte count
369 Size += StringTable.size();
370 // ld64 expects the members to be 8-byte aligned for 64-bit content and at
371 // least 4-byte aligned for 32-bit content. Opt for the larger encoding
372 // uniformly.
373 // We do this for all bsd formats because it simplifies aligning members.
374 // For the big archive format, the symbol table is the last member, so there
375 // is no need to align.
376 uint32_t Pad = isAIXBigArchive(Kind)
377 ? 0
378 : offsetToAlignment(Size, Align(isBSDLike(Kind) ? 8 : 2));
379 Size += Pad;
380 if (Padding)
381 *Padding = Pad;
382 return Size;
383 }
384
writeSymbolTableHeader(raw_ostream & Out,object::Archive::Kind Kind,bool Deterministic,uint64_t Size,uint64_t PrevMemberOffset=0)385 static void writeSymbolTableHeader(raw_ostream &Out, object::Archive::Kind Kind,
386 bool Deterministic, uint64_t Size,
387 uint64_t PrevMemberOffset = 0) {
388 if (isBSDLike(Kind)) {
389 const char *Name = is64BitKind(Kind) ? "__.SYMDEF_64" : "__.SYMDEF";
390 printBSDMemberHeader(Out, Out.tell(), Name, now(Deterministic), 0, 0, 0,
391 Size);
392 } else if (isAIXBigArchive(Kind)) {
393 printBigArchiveMemberHeader(Out, "", now(Deterministic), 0, 0,
394 0, Size, PrevMemberOffset, 0);
395 } else {
396 const char *Name = is64BitKind(Kind) ? "/SYM64" : "";
397 printGNUSmallMemberHeader(Out, Name, now(Deterministic), 0, 0, 0, Size);
398 }
399 }
400
writeSymbolTable(raw_ostream & Out,object::Archive::Kind Kind,bool Deterministic,ArrayRef<MemberData> Members,StringRef StringTable,uint64_t PrevMemberOffset=0)401 static void writeSymbolTable(raw_ostream &Out, object::Archive::Kind Kind,
402 bool Deterministic, ArrayRef<MemberData> Members,
403 StringRef StringTable,
404 uint64_t PrevMemberOffset = 0) {
405 // We don't write a symbol table on an archive with no members -- except on
406 // Darwin, where the linker will abort unless the archive has a symbol table.
407 if (StringTable.empty() && !isDarwin(Kind))
408 return;
409
410 unsigned NumSyms = 0;
411 for (const MemberData &M : Members)
412 NumSyms += M.Symbols.size();
413
414 uint64_t OffsetSize = is64BitKind(Kind) ? 8 : 4;
415 uint32_t Pad;
416 uint64_t Size = computeSymbolTableSize(Kind, NumSyms, OffsetSize, StringTable, &Pad);
417 writeSymbolTableHeader(Out, Kind, Deterministic, Size, PrevMemberOffset);
418
419 uint64_t Pos = isAIXBigArchive(Kind) ? sizeof(object::BigArchive::FixLenHdr)
420 : Out.tell() + Size;
421
422 if (isBSDLike(Kind))
423 printNBits(Out, Kind, NumSyms * 2 * OffsetSize);
424 else
425 printNBits(Out, Kind, NumSyms);
426
427 for (const MemberData &M : Members) {
428 for (unsigned StringOffset : M.Symbols) {
429 if (isBSDLike(Kind))
430 printNBits(Out, Kind, StringOffset);
431 printNBits(Out, Kind, Pos); // member offset
432 }
433 Pos += M.Header.size() + M.Data.size() + M.Padding.size();
434 }
435
436 if (isBSDLike(Kind))
437 // byte count of the string table
438 printNBits(Out, Kind, StringTable.size());
439 Out << StringTable;
440
441 while (Pad--)
442 Out.write(uint8_t(0));
443 }
444
445 static Expected<std::vector<unsigned>>
getSymbols(MemoryBufferRef Buf,raw_ostream & SymNames,bool & HasObject)446 getSymbols(MemoryBufferRef Buf, raw_ostream &SymNames, bool &HasObject) {
447 std::vector<unsigned> Ret;
448
449 // In the scenario when LLVMContext is populated SymbolicFile will contain a
450 // reference to it, thus SymbolicFile should be destroyed first.
451 LLVMContext Context;
452 std::unique_ptr<object::SymbolicFile> Obj;
453
454 const file_magic Type = identify_magic(Buf.getBuffer());
455 // Treat unsupported file types as having no symbols.
456 if (!object::SymbolicFile::isSymbolicFile(Type, &Context))
457 return Ret;
458 if (Type == file_magic::bitcode) {
459 auto ObjOrErr = object::SymbolicFile::createSymbolicFile(
460 Buf, file_magic::bitcode, &Context);
461 if (!ObjOrErr)
462 return ObjOrErr.takeError();
463 Obj = std::move(*ObjOrErr);
464 } else {
465 auto ObjOrErr = object::SymbolicFile::createSymbolicFile(Buf);
466 if (!ObjOrErr)
467 return ObjOrErr.takeError();
468 Obj = std::move(*ObjOrErr);
469 }
470
471 HasObject = true;
472 for (const object::BasicSymbolRef &S : Obj->symbols()) {
473 if (!isArchiveSymbol(S))
474 continue;
475 Ret.push_back(SymNames.tell());
476 if (Error E = S.printName(SymNames))
477 return std::move(E);
478 SymNames << '\0';
479 }
480 return Ret;
481 }
482
483 static Expected<std::vector<MemberData>>
computeMemberData(raw_ostream & StringTable,raw_ostream & SymNames,object::Archive::Kind Kind,bool Thin,bool Deterministic,bool NeedSymbols,ArrayRef<NewArchiveMember> NewMembers)484 computeMemberData(raw_ostream &StringTable, raw_ostream &SymNames,
485 object::Archive::Kind Kind, bool Thin, bool Deterministic,
486 bool NeedSymbols, ArrayRef<NewArchiveMember> NewMembers) {
487 static char PaddingData[8] = {'\n', '\n', '\n', '\n', '\n', '\n', '\n', '\n'};
488
489 uint64_t Pos =
490 isAIXBigArchive(Kind) ? sizeof(object::BigArchive::FixLenHdr) : 0;
491
492 std::vector<MemberData> Ret;
493 bool HasObject = false;
494
495 // Deduplicate long member names in the string table and reuse earlier name
496 // offsets. This especially saves space for COFF Import libraries where all
497 // members have the same name.
498 StringMap<uint64_t> MemberNames;
499
500 // UniqueTimestamps is a special case to improve debugging on Darwin:
501 //
502 // The Darwin linker does not link debug info into the final
503 // binary. Instead, it emits entries of type N_OSO in in the output
504 // binary's symbol table, containing references to the linked-in
505 // object files. Using that reference, the debugger can read the
506 // debug data directly from the object files. Alternatively, an
507 // invocation of 'dsymutil' will link the debug data from the object
508 // files into a dSYM bundle, which can be loaded by the debugger,
509 // instead of the object files.
510 //
511 // For an object file, the N_OSO entries contain the absolute path
512 // path to the file, and the file's timestamp. For an object
513 // included in an archive, the path is formatted like
514 // "/absolute/path/to/archive.a(member.o)", and the timestamp is the
515 // archive member's timestamp, rather than the archive's timestamp.
516 //
517 // However, this doesn't always uniquely identify an object within
518 // an archive -- an archive file can have multiple entries with the
519 // same filename. (This will happen commonly if the original object
520 // files started in different directories.) The only way they get
521 // distinguished, then, is via the timestamp. But this process is
522 // unable to find the correct object file in the archive when there
523 // are two files of the same name and timestamp.
524 //
525 // Additionally, timestamp==0 is treated specially, and causes the
526 // timestamp to be ignored as a match criteria.
527 //
528 // That will "usually" work out okay when creating an archive not in
529 // deterministic timestamp mode, because the objects will probably
530 // have been created at different timestamps.
531 //
532 // To ameliorate this problem, in deterministic archive mode (which
533 // is the default), on Darwin we will emit a unique non-zero
534 // timestamp for each entry with a duplicated name. This is still
535 // deterministic: the only thing affecting that timestamp is the
536 // order of the files in the resultant archive.
537 //
538 // See also the functions that handle the lookup:
539 // in lldb: ObjectContainerBSDArchive::Archive::FindObject()
540 // in llvm/tools/dsymutil: BinaryHolder::GetArchiveMemberBuffers().
541 bool UniqueTimestamps = Deterministic && isDarwin(Kind);
542 std::map<StringRef, unsigned> FilenameCount;
543 if (UniqueTimestamps) {
544 for (const NewArchiveMember &M : NewMembers)
545 FilenameCount[M.MemberName]++;
546 for (auto &Entry : FilenameCount)
547 Entry.second = Entry.second > 1 ? 1 : 0;
548 }
549
550 // The big archive format needs to know the offset of the previous member
551 // header.
552 unsigned PrevOffset = 0;
553 for (const NewArchiveMember &M : NewMembers) {
554 std::string Header;
555 raw_string_ostream Out(Header);
556
557 MemoryBufferRef Buf = M.Buf->getMemBufferRef();
558 StringRef Data = Thin ? "" : Buf.getBuffer();
559
560 // ld64 expects the members to be 8-byte aligned for 64-bit content and at
561 // least 4-byte aligned for 32-bit content. Opt for the larger encoding
562 // uniformly. This matches the behaviour with cctools and ensures that ld64
563 // is happy with archives that we generate.
564 unsigned MemberPadding =
565 isDarwin(Kind) ? offsetToAlignment(Data.size(), Align(8)) : 0;
566 unsigned TailPadding =
567 offsetToAlignment(Data.size() + MemberPadding, Align(2));
568 StringRef Padding = StringRef(PaddingData, MemberPadding + TailPadding);
569
570 sys::TimePoint<std::chrono::seconds> ModTime;
571 if (UniqueTimestamps)
572 // Increment timestamp for each file of a given name.
573 ModTime = sys::toTimePoint(FilenameCount[M.MemberName]++);
574 else
575 ModTime = M.ModTime;
576
577 uint64_t Size = Buf.getBufferSize() + MemberPadding;
578 if (Size > object::Archive::MaxMemberSize) {
579 std::string StringMsg =
580 "File " + M.MemberName.str() + " exceeds size limit";
581 return make_error<object::GenericBinaryError>(
582 std::move(StringMsg), object::object_error::parse_failed);
583 }
584
585 if (isAIXBigArchive(Kind)) {
586 unsigned NextOffset = Pos + sizeof(object::BigArMemHdrType) +
587 alignTo(M.MemberName.size(), 2) + alignTo(Size, 2);
588 printBigArchiveMemberHeader(Out, M.MemberName, ModTime, M.UID, M.GID,
589 M.Perms, Size, PrevOffset, NextOffset);
590 PrevOffset = Pos;
591 } else {
592 printMemberHeader(Out, Pos, StringTable, MemberNames, Kind, Thin, M,
593 ModTime, Size);
594 }
595 Out.flush();
596
597 std::vector<unsigned> Symbols;
598 if (NeedSymbols) {
599 Expected<std::vector<unsigned>> SymbolsOrErr =
600 getSymbols(Buf, SymNames, HasObject);
601 if (!SymbolsOrErr)
602 return createFileError(M.MemberName, SymbolsOrErr.takeError());
603 Symbols = std::move(*SymbolsOrErr);
604 }
605
606 Pos += Header.size() + Data.size() + Padding.size();
607 Ret.push_back({std::move(Symbols), std::move(Header), Data, Padding});
608 }
609 // If there are no symbols, emit an empty symbol table, to satisfy Solaris
610 // tools, older versions of which expect a symbol table in a non-empty
611 // archive, regardless of whether there are any symbols in it.
612 if (HasObject && SymNames.tell() == 0)
613 SymNames << '\0' << '\0' << '\0';
614 return Ret;
615 }
616
617 namespace llvm {
618
canonicalizePath(StringRef P)619 static ErrorOr<SmallString<128>> canonicalizePath(StringRef P) {
620 SmallString<128> Ret = P;
621 std::error_code Err = sys::fs::make_absolute(Ret);
622 if (Err)
623 return Err;
624 sys::path::remove_dots(Ret, /*removedotdot*/ true);
625 return Ret;
626 }
627
628 // Compute the relative path from From to To.
computeArchiveRelativePath(StringRef From,StringRef To)629 Expected<std::string> computeArchiveRelativePath(StringRef From, StringRef To) {
630 ErrorOr<SmallString<128>> PathToOrErr = canonicalizePath(To);
631 ErrorOr<SmallString<128>> DirFromOrErr = canonicalizePath(From);
632 if (!PathToOrErr || !DirFromOrErr)
633 return errorCodeToError(std::error_code(errno, std::generic_category()));
634
635 const SmallString<128> &PathTo = *PathToOrErr;
636 const SmallString<128> &DirFrom = sys::path::parent_path(*DirFromOrErr);
637
638 // Can't construct a relative path between different roots
639 if (sys::path::root_name(PathTo) != sys::path::root_name(DirFrom))
640 return sys::path::convert_to_slash(PathTo);
641
642 // Skip common prefixes
643 auto FromTo =
644 std::mismatch(sys::path::begin(DirFrom), sys::path::end(DirFrom),
645 sys::path::begin(PathTo));
646 auto FromI = FromTo.first;
647 auto ToI = FromTo.second;
648
649 // Construct relative path
650 SmallString<128> Relative;
651 for (auto FromE = sys::path::end(DirFrom); FromI != FromE; ++FromI)
652 sys::path::append(Relative, sys::path::Style::posix, "..");
653
654 for (auto ToE = sys::path::end(PathTo); ToI != ToE; ++ToI)
655 sys::path::append(Relative, sys::path::Style::posix, *ToI);
656
657 return std::string(Relative.str());
658 }
659
writeArchiveToStream(raw_ostream & Out,ArrayRef<NewArchiveMember> NewMembers,bool WriteSymtab,object::Archive::Kind Kind,bool Deterministic,bool Thin)660 static Error writeArchiveToStream(raw_ostream &Out,
661 ArrayRef<NewArchiveMember> NewMembers,
662 bool WriteSymtab, object::Archive::Kind Kind,
663 bool Deterministic, bool Thin) {
664 assert((!Thin || !isBSDLike(Kind)) && "Only the gnu format has a thin mode");
665
666 SmallString<0> SymNamesBuf;
667 raw_svector_ostream SymNames(SymNamesBuf);
668 SmallString<0> StringTableBuf;
669 raw_svector_ostream StringTable(StringTableBuf);
670
671 Expected<std::vector<MemberData>> DataOrErr =
672 computeMemberData(StringTable, SymNames, Kind, Thin, Deterministic,
673 WriteSymtab, NewMembers);
674 if (Error E = DataOrErr.takeError())
675 return E;
676 std::vector<MemberData> &Data = *DataOrErr;
677
678 if (!StringTableBuf.empty() && !isAIXBigArchive(Kind))
679 Data.insert(Data.begin(), computeStringTable(StringTableBuf));
680
681 // We would like to detect if we need to switch to a 64-bit symbol table.
682 uint64_t LastMemberEndOffset =
683 isAIXBigArchive(Kind) ? sizeof(object::BigArchive::FixLenHdr) : 8;
684 uint64_t LastMemberHeaderOffset = LastMemberEndOffset;
685 uint64_t NumSyms = 0;
686 for (const auto &M : Data) {
687 // Record the start of the member's offset
688 LastMemberHeaderOffset = LastMemberEndOffset;
689 // Account for the size of each part associated with the member.
690 LastMemberEndOffset += M.Header.size() + M.Data.size() + M.Padding.size();
691 NumSyms += M.Symbols.size();
692 }
693
694 // The symbol table is put at the end of the big archive file. The symbol
695 // table is at the start of the archive file for other archive formats.
696 if (WriteSymtab && !isAIXBigArchive(Kind)) {
697 // We assume 32-bit offsets to see if 32-bit symbols are possible or not.
698 uint64_t SymtabSize = computeSymbolTableSize(Kind, NumSyms, 4, SymNamesBuf);
699 auto computeSymbolTableHeaderSize =
700 [=] {
701 SmallString<0> TmpBuf;
702 raw_svector_ostream Tmp(TmpBuf);
703 writeSymbolTableHeader(Tmp, Kind, Deterministic, SymtabSize);
704 return TmpBuf.size();
705 };
706 LastMemberHeaderOffset += computeSymbolTableHeaderSize() + SymtabSize;
707
708 // The SYM64 format is used when an archive's member offsets are larger than
709 // 32-bits can hold. The need for this shift in format is detected by
710 // writeArchive. To test this we need to generate a file with a member that
711 // has an offset larger than 32-bits but this demands a very slow test. To
712 // speed the test up we use this environment variable to pretend like the
713 // cutoff happens before 32-bits and instead happens at some much smaller
714 // value.
715 uint64_t Sym64Threshold = 1ULL << 32;
716 const char *Sym64Env = std::getenv("SYM64_THRESHOLD");
717 if (Sym64Env)
718 StringRef(Sym64Env).getAsInteger(10, Sym64Threshold);
719
720 // If LastMemberHeaderOffset isn't going to fit in a 32-bit varible we need
721 // to switch to 64-bit. Note that the file can be larger than 4GB as long as
722 // the last member starts before the 4GB offset.
723 if (LastMemberHeaderOffset >= Sym64Threshold) {
724 if (Kind == object::Archive::K_DARWIN)
725 Kind = object::Archive::K_DARWIN64;
726 else
727 Kind = object::Archive::K_GNU64;
728 }
729 }
730
731 if (Thin)
732 Out << "!<thin>\n";
733 else if (isAIXBigArchive(Kind))
734 Out << "<bigaf>\n";
735 else
736 Out << "!<arch>\n";
737
738 if (!isAIXBigArchive(Kind)) {
739 if (WriteSymtab)
740 writeSymbolTable(Out, Kind, Deterministic, Data, SymNamesBuf);
741 for (const MemberData &M : Data)
742 Out << M.Header << M.Data << M.Padding;
743 } else {
744 // For the big archive (AIX) format, compute a table of member names and
745 // offsets, used in the member table.
746 uint64_t MemberTableNameStrTblSize = 0;
747 std::vector<size_t> MemberOffsets;
748 std::vector<StringRef> MemberNames;
749 // Loop across object to find offset and names.
750 uint64_t MemberEndOffset = sizeof(object::BigArchive::FixLenHdr);
751 for (size_t I = 0, Size = NewMembers.size(); I != Size; ++I) {
752 const NewArchiveMember &Member = NewMembers[I];
753 MemberTableNameStrTblSize += Member.MemberName.size() + 1;
754 MemberOffsets.push_back(MemberEndOffset);
755 MemberNames.push_back(Member.MemberName);
756 // File member name ended with "`\n". The length is included in
757 // BigArMemHdrType.
758 MemberEndOffset += sizeof(object::BigArMemHdrType) +
759 alignTo(Data[I].Data.size(), 2) +
760 alignTo(Member.MemberName.size(), 2);
761 }
762
763 // AIX member table size.
764 unsigned MemberTableSize = 20 + // Number of members field
765 20 * MemberOffsets.size() +
766 MemberTableNameStrTblSize;
767
768 unsigned GlobalSymbolOffset =
769 (WriteSymtab && NumSyms > 0)
770 ? LastMemberEndOffset +
771 alignTo(sizeof(object::BigArMemHdrType) + MemberTableSize, 2)
772 : 0;
773
774 // Fixed Sized Header.
775 printWithSpacePadding(Out, NewMembers.size() ? LastMemberEndOffset : 0,
776 20); // Offset to member table
777 // If there are no file members in the archive, there will be no global
778 // symbol table.
779 printWithSpacePadding(Out, NewMembers.size() ? GlobalSymbolOffset : 0, 20);
780 printWithSpacePadding(
781 Out, 0,
782 20); // Offset to 64 bits global symbol table - Not supported yet
783 printWithSpacePadding(
784 Out, NewMembers.size() ? sizeof(object::BigArchive::FixLenHdr) : 0,
785 20); // Offset to first archive member
786 printWithSpacePadding(Out, NewMembers.size() ? LastMemberHeaderOffset : 0,
787 20); // Offset to last archive member
788 printWithSpacePadding(
789 Out, 0,
790 20); // Offset to first member of free list - Not supported yet
791
792 for (const MemberData &M : Data) {
793 Out << M.Header << M.Data;
794 if (M.Data.size() % 2)
795 Out << '\0';
796 }
797
798 if (NewMembers.size()) {
799 // Member table.
800 printBigArchiveMemberHeader(Out, "", sys::toTimePoint(0), 0, 0, 0,
801 MemberTableSize, LastMemberHeaderOffset,
802 GlobalSymbolOffset);
803 printWithSpacePadding(Out, MemberOffsets.size(), 20); // Number of members
804 for (uint64_t MemberOffset : MemberOffsets)
805 printWithSpacePadding(Out, MemberOffset,
806 20); // Offset to member file header.
807 for (StringRef MemberName : MemberNames)
808 Out << MemberName << '\0'; // Member file name, null byte padding.
809
810 if (MemberTableNameStrTblSize % 2)
811 Out << '\0'; // Name table must be tail padded to an even number of
812 // bytes.
813
814 if (WriteSymtab && NumSyms > 0)
815 writeSymbolTable(Out, Kind, Deterministic, Data, SymNamesBuf,
816 LastMemberEndOffset);
817 }
818 }
819 Out.flush();
820 return Error::success();
821 }
822
writeArchive(StringRef ArcName,ArrayRef<NewArchiveMember> NewMembers,bool WriteSymtab,object::Archive::Kind Kind,bool Deterministic,bool Thin,std::unique_ptr<MemoryBuffer> OldArchiveBuf)823 Error writeArchive(StringRef ArcName, ArrayRef<NewArchiveMember> NewMembers,
824 bool WriteSymtab, object::Archive::Kind Kind,
825 bool Deterministic, bool Thin,
826 std::unique_ptr<MemoryBuffer> OldArchiveBuf) {
827 Expected<sys::fs::TempFile> Temp =
828 sys::fs::TempFile::create(ArcName + ".temp-archive-%%%%%%%.a");
829 if (!Temp)
830 return Temp.takeError();
831 raw_fd_ostream Out(Temp->FD, false);
832
833 if (Error E = writeArchiveToStream(Out, NewMembers, WriteSymtab, Kind,
834 Deterministic, Thin)) {
835 if (Error DiscardError = Temp->discard())
836 return joinErrors(std::move(E), std::move(DiscardError));
837 return E;
838 }
839
840 // At this point, we no longer need whatever backing memory
841 // was used to generate the NewMembers. On Windows, this buffer
842 // could be a mapped view of the file we want to replace (if
843 // we're updating an existing archive, say). In that case, the
844 // rename would still succeed, but it would leave behind a
845 // temporary file (actually the original file renamed) because
846 // a file cannot be deleted while there's a handle open on it,
847 // only renamed. So by freeing this buffer, this ensures that
848 // the last open handle on the destination file, if any, is
849 // closed before we attempt to rename.
850 OldArchiveBuf.reset();
851
852 return Temp->keep(ArcName);
853 }
854
855 Expected<std::unique_ptr<MemoryBuffer>>
writeArchiveToBuffer(ArrayRef<NewArchiveMember> NewMembers,bool WriteSymtab,object::Archive::Kind Kind,bool Deterministic,bool Thin)856 writeArchiveToBuffer(ArrayRef<NewArchiveMember> NewMembers, bool WriteSymtab,
857 object::Archive::Kind Kind, bool Deterministic,
858 bool Thin) {
859 SmallVector<char, 0> ArchiveBufferVector;
860 raw_svector_ostream ArchiveStream(ArchiveBufferVector);
861
862 if (Error E = writeArchiveToStream(ArchiveStream, NewMembers, WriteSymtab,
863 Kind, Deterministic, Thin))
864 return std::move(E);
865
866 return std::make_unique<SmallVectorMemoryBuffer>(
867 std::move(ArchiveBufferVector), /*RequiresNullTerminator=*/false);
868 }
869
870 } // namespace llvm
871