1 //===- Archive.cpp - ar File Format implementation ------------------------===//
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 ArchiveObjectFile class.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/Object/Archive.h"
14 #include "llvm/ADT/SmallString.h"
15 #include "llvm/ADT/StringRef.h"
16 #include "llvm/ADT/Twine.h"
17 #include "llvm/Object/Binary.h"
18 #include "llvm/Object/Error.h"
19 #include "llvm/Support/Chrono.h"
20 #include "llvm/Support/Endian.h"
21 #include "llvm/Support/EndianStream.h"
22 #include "llvm/Support/Error.h"
23 #include "llvm/Support/ErrorOr.h"
24 #include "llvm/Support/FileSystem.h"
25 #include "llvm/Support/MathExtras.h"
26 #include "llvm/Support/MemoryBuffer.h"
27 #include "llvm/Support/Path.h"
28 #include "llvm/Support/raw_ostream.h"
29 #include "llvm/TargetParser/Host.h"
30 #include <algorithm>
31 #include <cassert>
32 #include <cstddef>
33 #include <cstdint>
34 #include <memory>
35 #include <string>
36 #include <system_error>
37 
38 using namespace llvm;
39 using namespace object;
40 using namespace llvm::support::endian;
41 
anchor()42 void Archive::anchor() {}
43 
malformedError(Twine Msg)44 static Error malformedError(Twine Msg) {
45   std::string StringMsg = "truncated or malformed archive (" + Msg.str() + ")";
46   return make_error<GenericBinaryError>(std::move(StringMsg),
47                                         object_error::parse_failed);
48 }
49 
50 static Error
createMemberHeaderParseError(const AbstractArchiveMemberHeader * ArMemHeader,const char * RawHeaderPtr,uint64_t Size)51 createMemberHeaderParseError(const AbstractArchiveMemberHeader *ArMemHeader,
52                              const char *RawHeaderPtr, uint64_t Size) {
53   StringRef Msg("remaining size of archive too small for next archive "
54                 "member header ");
55 
56   Expected<StringRef> NameOrErr = ArMemHeader->getName(Size);
57   if (NameOrErr)
58     return malformedError(Msg + "for " + *NameOrErr);
59 
60   consumeError(NameOrErr.takeError());
61   uint64_t Offset = RawHeaderPtr - ArMemHeader->Parent->getData().data();
62   return malformedError(Msg + "at offset " + Twine(Offset));
63 }
64 
65 template <class T, std::size_t N>
getFieldRawString(const T (& Field)[N])66 StringRef getFieldRawString(const T (&Field)[N]) {
67   return StringRef(Field, N).rtrim(" ");
68 }
69 
70 template <class T>
getRawAccessMode() const71 StringRef CommonArchiveMemberHeader<T>::getRawAccessMode() const {
72   return getFieldRawString(ArMemHdr->AccessMode);
73 }
74 
75 template <class T>
getRawLastModified() const76 StringRef CommonArchiveMemberHeader<T>::getRawLastModified() const {
77   return getFieldRawString(ArMemHdr->LastModified);
78 }
79 
getRawUID() const80 template <class T> StringRef CommonArchiveMemberHeader<T>::getRawUID() const {
81   return getFieldRawString(ArMemHdr->UID);
82 }
83 
getRawGID() const84 template <class T> StringRef CommonArchiveMemberHeader<T>::getRawGID() const {
85   return getFieldRawString(ArMemHdr->GID);
86 }
87 
getOffset() const88 template <class T> uint64_t CommonArchiveMemberHeader<T>::getOffset() const {
89   return reinterpret_cast<const char *>(ArMemHdr) - Parent->getData().data();
90 }
91 
92 template class object::CommonArchiveMemberHeader<UnixArMemHdrType>;
93 template class object::CommonArchiveMemberHeader<BigArMemHdrType>;
94 
ArchiveMemberHeader(const Archive * Parent,const char * RawHeaderPtr,uint64_t Size,Error * Err)95 ArchiveMemberHeader::ArchiveMemberHeader(const Archive *Parent,
96                                          const char *RawHeaderPtr,
97                                          uint64_t Size, Error *Err)
98     : CommonArchiveMemberHeader<UnixArMemHdrType>(
99           Parent, reinterpret_cast<const UnixArMemHdrType *>(RawHeaderPtr)) {
100   if (RawHeaderPtr == nullptr)
101     return;
102   ErrorAsOutParameter ErrAsOutParam(Err);
103 
104   if (Size < getSizeOf()) {
105     *Err = createMemberHeaderParseError(this, RawHeaderPtr, Size);
106     return;
107   }
108   if (ArMemHdr->Terminator[0] != '`' || ArMemHdr->Terminator[1] != '\n') {
109     if (Err) {
110       std::string Buf;
111       raw_string_ostream OS(Buf);
112       OS.write_escaped(
113           StringRef(ArMemHdr->Terminator, sizeof(ArMemHdr->Terminator)));
114       OS.flush();
115       std::string Msg("terminator characters in archive member \"" + Buf +
116                       "\" not the correct \"`\\n\" values for the archive "
117                       "member header ");
118       Expected<StringRef> NameOrErr = getName(Size);
119       if (!NameOrErr) {
120         consumeError(NameOrErr.takeError());
121         uint64_t Offset = RawHeaderPtr - Parent->getData().data();
122         *Err = malformedError(Msg + "at offset " + Twine(Offset));
123       } else
124         *Err = malformedError(Msg + "for " + NameOrErr.get());
125     }
126     return;
127   }
128 }
129 
BigArchiveMemberHeader(const Archive * Parent,const char * RawHeaderPtr,uint64_t Size,Error * Err)130 BigArchiveMemberHeader::BigArchiveMemberHeader(const Archive *Parent,
131                                                const char *RawHeaderPtr,
132                                                uint64_t Size, Error *Err)
133     : CommonArchiveMemberHeader<BigArMemHdrType>(
134           Parent, reinterpret_cast<const BigArMemHdrType *>(RawHeaderPtr)) {
135   if (RawHeaderPtr == nullptr)
136     return;
137   ErrorAsOutParameter ErrAsOutParam(Err);
138 
139   if (RawHeaderPtr + getSizeOf() >= Parent->getData().end()) {
140     if (Err)
141       *Err = malformedError("malformed AIX big archive: remaining buffer is "
142                             "unable to contain next archive member");
143     return;
144   }
145 
146   if (Size < getSizeOf()) {
147     Error SubErr = createMemberHeaderParseError(this, RawHeaderPtr, Size);
148     if (Err)
149       *Err = std::move(SubErr);
150   }
151 }
152 
153 // This gets the raw name from the ArMemHdr->Name field and checks that it is
154 // valid for the kind of archive.  If it is not valid it returns an Error.
getRawName() const155 Expected<StringRef> ArchiveMemberHeader::getRawName() const {
156   char EndCond;
157   auto Kind = Parent->kind();
158   if (Kind == Archive::K_BSD || Kind == Archive::K_DARWIN64) {
159     if (ArMemHdr->Name[0] == ' ') {
160       uint64_t Offset =
161           reinterpret_cast<const char *>(ArMemHdr) - Parent->getData().data();
162       return malformedError("name contains a leading space for archive member "
163                             "header at offset " +
164                             Twine(Offset));
165     }
166     EndCond = ' ';
167   } else if (ArMemHdr->Name[0] == '/' || ArMemHdr->Name[0] == '#')
168     EndCond = ' ';
169   else
170     EndCond = '/';
171   StringRef::size_type end =
172       StringRef(ArMemHdr->Name, sizeof(ArMemHdr->Name)).find(EndCond);
173   if (end == StringRef::npos)
174     end = sizeof(ArMemHdr->Name);
175   assert(end <= sizeof(ArMemHdr->Name) && end > 0);
176   // Don't include the EndCond if there is one.
177   return StringRef(ArMemHdr->Name, end);
178 }
179 
180 Expected<uint64_t>
getArchiveMemberDecField(Twine FieldName,const StringRef RawField,const Archive * Parent,const AbstractArchiveMemberHeader * MemHeader)181 getArchiveMemberDecField(Twine FieldName, const StringRef RawField,
182                          const Archive *Parent,
183                          const AbstractArchiveMemberHeader *MemHeader) {
184   uint64_t Value;
185   if (RawField.getAsInteger(10, Value)) {
186     uint64_t Offset = MemHeader->getOffset();
187     return malformedError("characters in " + FieldName +
188                           " field in archive member header are not "
189                           "all decimal numbers: '" +
190                           RawField +
191                           "' for the archive "
192                           "member header at offset " +
193                           Twine(Offset));
194   }
195   return Value;
196 }
197 
198 Expected<uint64_t>
getArchiveMemberOctField(Twine FieldName,const StringRef RawField,const Archive * Parent,const AbstractArchiveMemberHeader * MemHeader)199 getArchiveMemberOctField(Twine FieldName, const StringRef RawField,
200                          const Archive *Parent,
201                          const AbstractArchiveMemberHeader *MemHeader) {
202   uint64_t Value;
203   if (RawField.getAsInteger(8, Value)) {
204     uint64_t Offset = MemHeader->getOffset();
205     return malformedError("characters in " + FieldName +
206                           " field in archive member header are not "
207                           "all octal numbers: '" +
208                           RawField +
209                           "' for the archive "
210                           "member header at offset " +
211                           Twine(Offset));
212   }
213   return Value;
214 }
215 
getRawName() const216 Expected<StringRef> BigArchiveMemberHeader::getRawName() const {
217   Expected<uint64_t> NameLenOrErr = getArchiveMemberDecField(
218       "NameLen", getFieldRawString(ArMemHdr->NameLen), Parent, this);
219   if (!NameLenOrErr)
220     // TODO: Out-of-line.
221     return NameLenOrErr.takeError();
222   uint64_t NameLen = NameLenOrErr.get();
223 
224   // If the name length is odd, pad with '\0' to get an even length. After
225   // padding, there is the name terminator "`\n".
226   uint64_t NameLenWithPadding = alignTo(NameLen, 2);
227   StringRef NameTerminator = "`\n";
228   StringRef NameStringWithNameTerminator =
229       StringRef(ArMemHdr->Name, NameLenWithPadding + NameTerminator.size());
230   if (!NameStringWithNameTerminator.ends_with(NameTerminator)) {
231     uint64_t Offset =
232         reinterpret_cast<const char *>(ArMemHdr->Name + NameLenWithPadding) -
233         Parent->getData().data();
234     // TODO: Out-of-line.
235     return malformedError(
236         "name does not have name terminator \"`\\n\" for archive member"
237         "header at offset " +
238         Twine(Offset));
239   }
240   return StringRef(ArMemHdr->Name, NameLen);
241 }
242 
243 // member including the header, so the size of any name following the header
244 // is checked to make sure it does not overflow.
getName(uint64_t Size) const245 Expected<StringRef> ArchiveMemberHeader::getName(uint64_t Size) const {
246 
247   // This can be called from the ArchiveMemberHeader constructor when the
248   // archive header is truncated to produce an error message with the name.
249   // Make sure the name field is not truncated.
250   if (Size < offsetof(UnixArMemHdrType, Name) + sizeof(ArMemHdr->Name)) {
251     uint64_t ArchiveOffset =
252         reinterpret_cast<const char *>(ArMemHdr) - Parent->getData().data();
253     return malformedError("archive header truncated before the name field "
254                           "for archive member header at offset " +
255                           Twine(ArchiveOffset));
256   }
257 
258   // The raw name itself can be invalid.
259   Expected<StringRef> NameOrErr = getRawName();
260   if (!NameOrErr)
261     return NameOrErr.takeError();
262   StringRef Name = NameOrErr.get();
263 
264   // Check if it's a special name.
265   if (Name[0] == '/') {
266     if (Name.size() == 1) // Linker member.
267       return Name;
268     if (Name.size() == 2 && Name[1] == '/') // String table.
269       return Name;
270     // System libraries from the Windows SDK for Windows 11 contain this symbol.
271     // It looks like a CFG guard: we just skip it for now.
272     if (Name.equals("/<XFGHASHMAP>/"))
273       return Name;
274     // Some libraries (e.g., arm64rt.lib) from the Windows WDK
275     // (version 10.0.22000.0) contain this undocumented special member.
276     if (Name.equals("/<ECSYMBOLS>/"))
277       return Name;
278     // It's a long name.
279     // Get the string table offset.
280     std::size_t StringOffset;
281     if (Name.substr(1).rtrim(' ').getAsInteger(10, StringOffset)) {
282       std::string Buf;
283       raw_string_ostream OS(Buf);
284       OS.write_escaped(Name.substr(1).rtrim(' '));
285       OS.flush();
286       uint64_t ArchiveOffset =
287           reinterpret_cast<const char *>(ArMemHdr) - Parent->getData().data();
288       return malformedError("long name offset characters after the '/' are "
289                             "not all decimal numbers: '" +
290                             Buf + "' for archive member header at offset " +
291                             Twine(ArchiveOffset));
292     }
293 
294     // Verify it.
295     if (StringOffset >= Parent->getStringTable().size()) {
296       uint64_t ArchiveOffset =
297           reinterpret_cast<const char *>(ArMemHdr) - Parent->getData().data();
298       return malformedError("long name offset " + Twine(StringOffset) +
299                             " past the end of the string table for archive "
300                             "member header at offset " +
301                             Twine(ArchiveOffset));
302     }
303 
304     // GNU long file names end with a "/\n".
305     if (Parent->kind() == Archive::K_GNU ||
306         Parent->kind() == Archive::K_GNU64) {
307       size_t End = Parent->getStringTable().find('\n', /*From=*/StringOffset);
308       if (End == StringRef::npos || End < 1 ||
309           Parent->getStringTable()[End - 1] != '/') {
310         return malformedError("string table at long name offset " +
311                               Twine(StringOffset) + "not terminated");
312       }
313       return Parent->getStringTable().slice(StringOffset, End - 1);
314     }
315     return Parent->getStringTable().begin() + StringOffset;
316   }
317 
318   if (Name.starts_with("#1/")) {
319     uint64_t NameLength;
320     if (Name.substr(3).rtrim(' ').getAsInteger(10, NameLength)) {
321       std::string Buf;
322       raw_string_ostream OS(Buf);
323       OS.write_escaped(Name.substr(3).rtrim(' '));
324       OS.flush();
325       uint64_t ArchiveOffset =
326           reinterpret_cast<const char *>(ArMemHdr) - Parent->getData().data();
327       return malformedError("long name length characters after the #1/ are "
328                             "not all decimal numbers: '" +
329                             Buf + "' for archive member header at offset " +
330                             Twine(ArchiveOffset));
331     }
332     if (getSizeOf() + NameLength > Size) {
333       uint64_t ArchiveOffset =
334           reinterpret_cast<const char *>(ArMemHdr) - Parent->getData().data();
335       return malformedError("long name length: " + Twine(NameLength) +
336                             " extends past the end of the member or archive "
337                             "for archive member header at offset " +
338                             Twine(ArchiveOffset));
339     }
340     return StringRef(reinterpret_cast<const char *>(ArMemHdr) + getSizeOf(),
341                      NameLength)
342         .rtrim('\0');
343   }
344 
345   // It is not a long name so trim the blanks at the end of the name.
346   if (Name[Name.size() - 1] != '/')
347     return Name.rtrim(' ');
348 
349   // It's a simple name.
350   return Name.drop_back(1);
351 }
352 
getName(uint64_t Size) const353 Expected<StringRef> BigArchiveMemberHeader::getName(uint64_t Size) const {
354   return getRawName();
355 }
356 
getSize() const357 Expected<uint64_t> ArchiveMemberHeader::getSize() const {
358   return getArchiveMemberDecField("size", getFieldRawString(ArMemHdr->Size),
359                                   Parent, this);
360 }
361 
getSize() const362 Expected<uint64_t> BigArchiveMemberHeader::getSize() const {
363   Expected<uint64_t> SizeOrErr = getArchiveMemberDecField(
364       "size", getFieldRawString(ArMemHdr->Size), Parent, this);
365   if (!SizeOrErr)
366     return SizeOrErr.takeError();
367 
368   Expected<uint64_t> NameLenOrErr = getRawNameSize();
369   if (!NameLenOrErr)
370     return NameLenOrErr.takeError();
371 
372   return *SizeOrErr + alignTo(*NameLenOrErr, 2);
373 }
374 
getRawNameSize() const375 Expected<uint64_t> BigArchiveMemberHeader::getRawNameSize() const {
376   return getArchiveMemberDecField(
377       "NameLen", getFieldRawString(ArMemHdr->NameLen), Parent, this);
378 }
379 
getNextOffset() const380 Expected<uint64_t> BigArchiveMemberHeader::getNextOffset() const {
381   return getArchiveMemberDecField(
382       "NextOffset", getFieldRawString(ArMemHdr->NextOffset), Parent, this);
383 }
384 
getAccessMode() const385 Expected<sys::fs::perms> AbstractArchiveMemberHeader::getAccessMode() const {
386   Expected<uint64_t> AccessModeOrErr =
387       getArchiveMemberOctField("AccessMode", getRawAccessMode(), Parent, this);
388   if (!AccessModeOrErr)
389     return AccessModeOrErr.takeError();
390   return static_cast<sys::fs::perms>(*AccessModeOrErr);
391 }
392 
393 Expected<sys::TimePoint<std::chrono::seconds>>
getLastModified() const394 AbstractArchiveMemberHeader::getLastModified() const {
395   Expected<uint64_t> SecondsOrErr = getArchiveMemberDecField(
396       "LastModified", getRawLastModified(), Parent, this);
397 
398   if (!SecondsOrErr)
399     return SecondsOrErr.takeError();
400 
401   return sys::toTimePoint(*SecondsOrErr);
402 }
403 
getUID() const404 Expected<unsigned> AbstractArchiveMemberHeader::getUID() const {
405   StringRef User = getRawUID();
406   if (User.empty())
407     return 0;
408   return getArchiveMemberDecField("UID", User, Parent, this);
409 }
410 
getGID() const411 Expected<unsigned> AbstractArchiveMemberHeader::getGID() const {
412   StringRef Group = getRawGID();
413   if (Group.empty())
414     return 0;
415   return getArchiveMemberDecField("GID", Group, Parent, this);
416 }
417 
isThin() const418 Expected<bool> ArchiveMemberHeader::isThin() const {
419   Expected<StringRef> NameOrErr = getRawName();
420   if (!NameOrErr)
421     return NameOrErr.takeError();
422   StringRef Name = NameOrErr.get();
423   return Parent->isThin() && Name != "/" && Name != "//" && Name != "/SYM64/";
424 }
425 
getNextChildLoc() const426 Expected<const char *> ArchiveMemberHeader::getNextChildLoc() const {
427   uint64_t Size = getSizeOf();
428   Expected<bool> isThinOrErr = isThin();
429   if (!isThinOrErr)
430     return isThinOrErr.takeError();
431 
432   bool isThin = isThinOrErr.get();
433   if (!isThin) {
434     Expected<uint64_t> MemberSize = getSize();
435     if (!MemberSize)
436       return MemberSize.takeError();
437 
438     Size += MemberSize.get();
439   }
440 
441   // If Size is odd, add 1 to make it even.
442   const char *NextLoc =
443       reinterpret_cast<const char *>(ArMemHdr) + alignTo(Size, 2);
444 
445   if (NextLoc == Parent->getMemoryBufferRef().getBufferEnd())
446     return nullptr;
447 
448   return NextLoc;
449 }
450 
getNextChildLoc() const451 Expected<const char *> BigArchiveMemberHeader::getNextChildLoc() const {
452   if (getOffset() ==
453       static_cast<const BigArchive *>(Parent)->getLastChildOffset())
454     return nullptr;
455 
456   Expected<uint64_t> NextOffsetOrErr = getNextOffset();
457   if (!NextOffsetOrErr)
458     return NextOffsetOrErr.takeError();
459   return Parent->getData().data() + NextOffsetOrErr.get();
460 }
461 
Child(const Archive * Parent,StringRef Data,uint16_t StartOfFile)462 Archive::Child::Child(const Archive *Parent, StringRef Data,
463                       uint16_t StartOfFile)
464     : Parent(Parent), Data(Data), StartOfFile(StartOfFile) {
465   Header = Parent->createArchiveMemberHeader(Data.data(), Data.size(), nullptr);
466 }
467 
Child(const Archive * Parent,const char * Start,Error * Err)468 Archive::Child::Child(const Archive *Parent, const char *Start, Error *Err)
469     : Parent(Parent) {
470   if (!Start) {
471     Header = nullptr;
472     StartOfFile = -1;
473     return;
474   }
475 
476   Header = Parent->createArchiveMemberHeader(
477       Start,
478       Parent ? Parent->getData().size() - (Start - Parent->getData().data())
479              : 0,
480       Err);
481 
482   // If we are pointed to real data, Start is not a nullptr, then there must be
483   // a non-null Err pointer available to report malformed data on.  Only in
484   // the case sentinel value is being constructed is Err is permitted to be a
485   // nullptr.
486   assert(Err && "Err can't be nullptr if Start is not a nullptr");
487 
488   ErrorAsOutParameter ErrAsOutParam(Err);
489 
490   // If there was an error in the construction of the Header
491   // then just return with the error now set.
492   if (*Err)
493     return;
494 
495   uint64_t Size = Header->getSizeOf();
496   Data = StringRef(Start, Size);
497   Expected<bool> isThinOrErr = isThinMember();
498   if (!isThinOrErr) {
499     *Err = isThinOrErr.takeError();
500     return;
501   }
502   bool isThin = isThinOrErr.get();
503   if (!isThin) {
504     Expected<uint64_t> MemberSize = getRawSize();
505     if (!MemberSize) {
506       *Err = MemberSize.takeError();
507       return;
508     }
509     Size += MemberSize.get();
510     Data = StringRef(Start, Size);
511   }
512 
513   // Setup StartOfFile and PaddingBytes.
514   StartOfFile = Header->getSizeOf();
515   // Don't include attached name.
516   Expected<StringRef> NameOrErr = getRawName();
517   if (!NameOrErr) {
518     *Err = NameOrErr.takeError();
519     return;
520   }
521   StringRef Name = NameOrErr.get();
522 
523   if (Parent->kind() == Archive::K_AIXBIG) {
524     // The actual start of the file is after the name and any necessary
525     // even-alignment padding.
526     StartOfFile += ((Name.size() + 1) >> 1) << 1;
527   } else if (Name.starts_with("#1/")) {
528     uint64_t NameSize;
529     StringRef RawNameSize = Name.substr(3).rtrim(' ');
530     if (RawNameSize.getAsInteger(10, NameSize)) {
531       uint64_t Offset = Start - Parent->getData().data();
532       *Err = malformedError("long name length characters after the #1/ are "
533                             "not all decimal numbers: '" +
534                             RawNameSize +
535                             "' for archive member header at offset " +
536                             Twine(Offset));
537       return;
538     }
539     StartOfFile += NameSize;
540   }
541 }
542 
getSize() const543 Expected<uint64_t> Archive::Child::getSize() const {
544   if (Parent->IsThin)
545     return Header->getSize();
546   return Data.size() - StartOfFile;
547 }
548 
getRawSize() const549 Expected<uint64_t> Archive::Child::getRawSize() const {
550   return Header->getSize();
551 }
552 
isThinMember() const553 Expected<bool> Archive::Child::isThinMember() const { return Header->isThin(); }
554 
getFullName() const555 Expected<std::string> Archive::Child::getFullName() const {
556   Expected<bool> isThin = isThinMember();
557   if (!isThin)
558     return isThin.takeError();
559   assert(isThin.get());
560   Expected<StringRef> NameOrErr = getName();
561   if (!NameOrErr)
562     return NameOrErr.takeError();
563   StringRef Name = *NameOrErr;
564   if (sys::path::is_absolute(Name))
565     return std::string(Name);
566 
567   SmallString<128> FullName = sys::path::parent_path(
568       Parent->getMemoryBufferRef().getBufferIdentifier());
569   sys::path::append(FullName, Name);
570   return std::string(FullName);
571 }
572 
getBuffer() const573 Expected<StringRef> Archive::Child::getBuffer() const {
574   Expected<bool> isThinOrErr = isThinMember();
575   if (!isThinOrErr)
576     return isThinOrErr.takeError();
577   bool isThin = isThinOrErr.get();
578   if (!isThin) {
579     Expected<uint64_t> Size = getSize();
580     if (!Size)
581       return Size.takeError();
582     return StringRef(Data.data() + StartOfFile, Size.get());
583   }
584   Expected<std::string> FullNameOrErr = getFullName();
585   if (!FullNameOrErr)
586     return FullNameOrErr.takeError();
587   const std::string &FullName = *FullNameOrErr;
588   ErrorOr<std::unique_ptr<MemoryBuffer>> Buf = MemoryBuffer::getFile(FullName);
589   if (std::error_code EC = Buf.getError())
590     return errorCodeToError(EC);
591   Parent->ThinBuffers.push_back(std::move(*Buf));
592   return Parent->ThinBuffers.back()->getBuffer();
593 }
594 
getNext() const595 Expected<Archive::Child> Archive::Child::getNext() const {
596   Expected<const char *> NextLocOrErr = Header->getNextChildLoc();
597   if (!NextLocOrErr)
598     return NextLocOrErr.takeError();
599 
600   const char *NextLoc = *NextLocOrErr;
601 
602   // Check to see if this is at the end of the archive.
603   if (NextLoc == nullptr)
604     return Child(nullptr, nullptr, nullptr);
605 
606   // Check to see if this is past the end of the archive.
607   if (NextLoc > Parent->Data.getBufferEnd()) {
608     std::string Msg("offset to next archive member past the end of the archive "
609                     "after member ");
610     Expected<StringRef> NameOrErr = getName();
611     if (!NameOrErr) {
612       consumeError(NameOrErr.takeError());
613       uint64_t Offset = Data.data() - Parent->getData().data();
614       return malformedError(Msg + "at offset " + Twine(Offset));
615     } else
616       return malformedError(Msg + NameOrErr.get());
617   }
618 
619   Error Err = Error::success();
620   Child Ret(Parent, NextLoc, &Err);
621   if (Err)
622     return std::move(Err);
623   return Ret;
624 }
625 
getChildOffset() const626 uint64_t Archive::Child::getChildOffset() const {
627   const char *a = Parent->Data.getBuffer().data();
628   const char *c = Data.data();
629   uint64_t offset = c - a;
630   return offset;
631 }
632 
getName() const633 Expected<StringRef> Archive::Child::getName() const {
634   Expected<uint64_t> RawSizeOrErr = getRawSize();
635   if (!RawSizeOrErr)
636     return RawSizeOrErr.takeError();
637   uint64_t RawSize = RawSizeOrErr.get();
638   Expected<StringRef> NameOrErr =
639       Header->getName(Header->getSizeOf() + RawSize);
640   if (!NameOrErr)
641     return NameOrErr.takeError();
642   StringRef Name = NameOrErr.get();
643   return Name;
644 }
645 
getMemoryBufferRef() const646 Expected<MemoryBufferRef> Archive::Child::getMemoryBufferRef() const {
647   Expected<StringRef> NameOrErr = getName();
648   if (!NameOrErr)
649     return NameOrErr.takeError();
650   StringRef Name = NameOrErr.get();
651   Expected<StringRef> Buf = getBuffer();
652   if (!Buf)
653     return createFileError(Name, Buf.takeError());
654   return MemoryBufferRef(*Buf, Name);
655 }
656 
657 Expected<std::unique_ptr<Binary>>
getAsBinary(LLVMContext * Context) const658 Archive::Child::getAsBinary(LLVMContext *Context) const {
659   Expected<MemoryBufferRef> BuffOrErr = getMemoryBufferRef();
660   if (!BuffOrErr)
661     return BuffOrErr.takeError();
662 
663   auto BinaryOrErr = createBinary(BuffOrErr.get(), Context);
664   if (BinaryOrErr)
665     return std::move(*BinaryOrErr);
666   return BinaryOrErr.takeError();
667 }
668 
create(MemoryBufferRef Source)669 Expected<std::unique_ptr<Archive>> Archive::create(MemoryBufferRef Source) {
670   Error Err = Error::success();
671   std::unique_ptr<Archive> Ret;
672   StringRef Buffer = Source.getBuffer();
673 
674   if (Buffer.starts_with(BigArchiveMagic))
675     Ret = std::make_unique<BigArchive>(Source, Err);
676   else
677     Ret = std::make_unique<Archive>(Source, Err);
678 
679   if (Err)
680     return std::move(Err);
681   return std::move(Ret);
682 }
683 
684 std::unique_ptr<AbstractArchiveMemberHeader>
createArchiveMemberHeader(const char * RawHeaderPtr,uint64_t Size,Error * Err) const685 Archive::createArchiveMemberHeader(const char *RawHeaderPtr, uint64_t Size,
686                                    Error *Err) const {
687   ErrorAsOutParameter ErrAsOutParam(Err);
688   if (kind() != K_AIXBIG)
689     return std::make_unique<ArchiveMemberHeader>(this, RawHeaderPtr, Size, Err);
690   return std::make_unique<BigArchiveMemberHeader>(this, RawHeaderPtr, Size,
691                                                   Err);
692 }
693 
getArchiveMagicLen() const694 uint64_t Archive::getArchiveMagicLen() const {
695   if (isThin())
696     return sizeof(ThinArchiveMagic) - 1;
697 
698   if (Kind() == K_AIXBIG)
699     return sizeof(BigArchiveMagic) - 1;
700 
701   return sizeof(ArchiveMagic) - 1;
702 }
703 
setFirstRegular(const Child & C)704 void Archive::setFirstRegular(const Child &C) {
705   FirstRegularData = C.Data;
706   FirstRegularStartOfFile = C.StartOfFile;
707 }
708 
Archive(MemoryBufferRef Source,Error & Err)709 Archive::Archive(MemoryBufferRef Source, Error &Err)
710     : Binary(Binary::ID_Archive, Source) {
711   ErrorAsOutParameter ErrAsOutParam(&Err);
712   StringRef Buffer = Data.getBuffer();
713   // Check for sufficient magic.
714   if (Buffer.starts_with(ThinArchiveMagic)) {
715     IsThin = true;
716   } else if (Buffer.starts_with(ArchiveMagic)) {
717     IsThin = false;
718   } else if (Buffer.starts_with(BigArchiveMagic)) {
719     Format = K_AIXBIG;
720     IsThin = false;
721     return;
722   } else {
723     Err = make_error<GenericBinaryError>("file too small to be an archive",
724                                          object_error::invalid_file_type);
725     return;
726   }
727 
728   // Make sure Format is initialized before any call to
729   // ArchiveMemberHeader::getName() is made.  This could be a valid empty
730   // archive which is the same in all formats.  So claiming it to be gnu to is
731   // fine if not totally correct before we look for a string table or table of
732   // contents.
733   Format = K_GNU;
734 
735   // Get the special members.
736   child_iterator I = child_begin(Err, false);
737   if (Err)
738     return;
739   child_iterator E = child_end();
740 
741   // See if this is a valid empty archive and if so return.
742   if (I == E) {
743     Err = Error::success();
744     return;
745   }
746   const Child *C = &*I;
747 
748   auto Increment = [&]() {
749     ++I;
750     if (Err)
751       return true;
752     C = &*I;
753     return false;
754   };
755 
756   Expected<StringRef> NameOrErr = C->getRawName();
757   if (!NameOrErr) {
758     Err = NameOrErr.takeError();
759     return;
760   }
761   StringRef Name = NameOrErr.get();
762 
763   // Below is the pattern that is used to figure out the archive format
764   // GNU archive format
765   //  First member : / (may exist, if it exists, points to the symbol table )
766   //  Second member : // (may exist, if it exists, points to the string table)
767   //  Note : The string table is used if the filename exceeds 15 characters
768   // BSD archive format
769   //  First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table)
770   //  There is no string table, if the filename exceeds 15 characters or has a
771   //  embedded space, the filename has #1/<size>, The size represents the size
772   //  of the filename that needs to be read after the archive header
773   // COFF archive format
774   //  First member : /
775   //  Second member : / (provides a directory of symbols)
776   //  Third member : // (may exist, if it exists, contains the string table)
777   //  Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
778   //  even if the string table is empty. However, lib.exe does not in fact
779   //  seem to create the third member if there's no member whose filename
780   //  exceeds 15 characters. So the third member is optional.
781 
782   if (Name == "__.SYMDEF" || Name == "__.SYMDEF_64") {
783     if (Name == "__.SYMDEF")
784       Format = K_BSD;
785     else // Name == "__.SYMDEF_64"
786       Format = K_DARWIN64;
787     // We know that the symbol table is not an external file, but we still must
788     // check any Expected<> return value.
789     Expected<StringRef> BufOrErr = C->getBuffer();
790     if (!BufOrErr) {
791       Err = BufOrErr.takeError();
792       return;
793     }
794     SymbolTable = BufOrErr.get();
795     if (Increment())
796       return;
797     setFirstRegular(*C);
798 
799     Err = Error::success();
800     return;
801   }
802 
803   if (Name.starts_with("#1/")) {
804     Format = K_BSD;
805     // We know this is BSD, so getName will work since there is no string table.
806     Expected<StringRef> NameOrErr = C->getName();
807     if (!NameOrErr) {
808       Err = NameOrErr.takeError();
809       return;
810     }
811     Name = NameOrErr.get();
812     if (Name == "__.SYMDEF SORTED" || Name == "__.SYMDEF") {
813       // We know that the symbol table is not an external file, but we still
814       // must check any Expected<> return value.
815       Expected<StringRef> BufOrErr = C->getBuffer();
816       if (!BufOrErr) {
817         Err = BufOrErr.takeError();
818         return;
819       }
820       SymbolTable = BufOrErr.get();
821       if (Increment())
822         return;
823     } else if (Name == "__.SYMDEF_64 SORTED" || Name == "__.SYMDEF_64") {
824       Format = K_DARWIN64;
825       // We know that the symbol table is not an external file, but we still
826       // must check any Expected<> return value.
827       Expected<StringRef> BufOrErr = C->getBuffer();
828       if (!BufOrErr) {
829         Err = BufOrErr.takeError();
830         return;
831       }
832       SymbolTable = BufOrErr.get();
833       if (Increment())
834         return;
835     }
836     setFirstRegular(*C);
837     return;
838   }
839 
840   // MIPS 64-bit ELF archives use a special format of a symbol table.
841   // This format is marked by `ar_name` field equals to "/SYM64/".
842   // For detailed description see page 96 in the following document:
843   // http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
844 
845   bool has64SymTable = false;
846   if (Name == "/" || Name == "/SYM64/") {
847     // We know that the symbol table is not an external file, but we still
848     // must check any Expected<> return value.
849     Expected<StringRef> BufOrErr = C->getBuffer();
850     if (!BufOrErr) {
851       Err = BufOrErr.takeError();
852       return;
853     }
854     SymbolTable = BufOrErr.get();
855     if (Name == "/SYM64/")
856       has64SymTable = true;
857 
858     if (Increment())
859       return;
860     if (I == E) {
861       Err = Error::success();
862       return;
863     }
864     Expected<StringRef> NameOrErr = C->getRawName();
865     if (!NameOrErr) {
866       Err = NameOrErr.takeError();
867       return;
868     }
869     Name = NameOrErr.get();
870   }
871 
872   if (Name == "//") {
873     Format = has64SymTable ? K_GNU64 : K_GNU;
874     // The string table is never an external member, but we still
875     // must check any Expected<> return value.
876     Expected<StringRef> BufOrErr = C->getBuffer();
877     if (!BufOrErr) {
878       Err = BufOrErr.takeError();
879       return;
880     }
881     StringTable = BufOrErr.get();
882     if (Increment())
883       return;
884     setFirstRegular(*C);
885     Err = Error::success();
886     return;
887   }
888 
889   if (Name[0] != '/') {
890     Format = has64SymTable ? K_GNU64 : K_GNU;
891     setFirstRegular(*C);
892     Err = Error::success();
893     return;
894   }
895 
896   if (Name != "/") {
897     Err = errorCodeToError(object_error::parse_failed);
898     return;
899   }
900 
901   Format = K_COFF;
902   // We know that the symbol table is not an external file, but we still
903   // must check any Expected<> return value.
904   Expected<StringRef> BufOrErr = C->getBuffer();
905   if (!BufOrErr) {
906     Err = BufOrErr.takeError();
907     return;
908   }
909   SymbolTable = BufOrErr.get();
910 
911   if (Increment())
912     return;
913 
914   if (I == E) {
915     setFirstRegular(*C);
916     Err = Error::success();
917     return;
918   }
919 
920   NameOrErr = C->getRawName();
921   if (!NameOrErr) {
922     Err = NameOrErr.takeError();
923     return;
924   }
925   Name = NameOrErr.get();
926 
927   if (Name == "//") {
928     // The string table is never an external member, but we still
929     // must check any Expected<> return value.
930     Expected<StringRef> BufOrErr = C->getBuffer();
931     if (!BufOrErr) {
932       Err = BufOrErr.takeError();
933       return;
934     }
935     StringTable = BufOrErr.get();
936     if (Increment())
937       return;
938 
939     if (I == E) {
940       setFirstRegular(*C);
941       Err = Error::success();
942       return;
943     }
944 
945     NameOrErr = C->getRawName();
946     if (!NameOrErr) {
947       Err = NameOrErr.takeError();
948       return;
949     }
950     Name = NameOrErr.get();
951   }
952 
953   if (Name == "/<ECSYMBOLS>/") {
954     // ARM64EC-aware libraries contain an additional special member with
955     // an EC symbol map after the string table. Its format is similar to a
956     // regular symbol map, except it doesn't contain member offsets. Its indexes
957     // refer to member offsets from the regular symbol table instead.
958     Expected<StringRef> BufOrErr = C->getBuffer();
959     if (!BufOrErr) {
960       Err = BufOrErr.takeError();
961       return;
962     }
963     ECSymbolTable = BufOrErr.get();
964     if (Increment())
965       return;
966   }
967 
968   setFirstRegular(*C);
969   Err = Error::success();
970 }
971 
getDefaultKindForHost()972 object::Archive::Kind Archive::getDefaultKindForHost() {
973   Triple HostTriple(sys::getProcessTriple());
974   return HostTriple.isOSDarwin()
975              ? object::Archive::K_DARWIN
976              : (HostTriple.isOSAIX() ? object::Archive::K_AIXBIG
977                                      : object::Archive::K_GNU);
978 }
979 
child_begin(Error & Err,bool SkipInternal) const980 Archive::child_iterator Archive::child_begin(Error &Err,
981                                              bool SkipInternal) const {
982   if (isEmpty())
983     return child_end();
984 
985   if (SkipInternal)
986     return child_iterator::itr(
987         Child(this, FirstRegularData, FirstRegularStartOfFile), Err);
988 
989   const char *Loc = Data.getBufferStart() + getFirstChildOffset();
990   Child C(this, Loc, &Err);
991   if (Err)
992     return child_end();
993   return child_iterator::itr(C, Err);
994 }
995 
child_end() const996 Archive::child_iterator Archive::child_end() const {
997   return child_iterator::end(Child(nullptr, nullptr, nullptr));
998 }
999 
isECSymbol() const1000 bool Archive::Symbol::isECSymbol() const {
1001   // Symbols use SymbolCount..SymbolCount+getNumberOfECSymbols() for EC symbol
1002   // indexes.
1003   uint32_t SymbolCount = Parent->getNumberOfSymbols();
1004   return SymbolCount <= SymbolIndex &&
1005          SymbolIndex < SymbolCount + Parent->getNumberOfECSymbols();
1006 }
1007 
getName() const1008 StringRef Archive::Symbol::getName() const {
1009   if (isECSymbol())
1010     return Parent->ECSymbolTable.begin() + StringIndex;
1011   return Parent->getSymbolTable().begin() + StringIndex;
1012 }
1013 
getMember() const1014 Expected<Archive::Child> Archive::Symbol::getMember() const {
1015   const char *Buf = Parent->getSymbolTable().begin();
1016   const char *Offsets = Buf;
1017   if (Parent->kind() == K_GNU64 || Parent->kind() == K_DARWIN64 ||
1018       Parent->kind() == K_AIXBIG)
1019     Offsets += sizeof(uint64_t);
1020   else
1021     Offsets += sizeof(uint32_t);
1022   uint64_t Offset = 0;
1023   if (Parent->kind() == K_GNU) {
1024     Offset = read32be(Offsets + SymbolIndex * 4);
1025   } else if (Parent->kind() == K_GNU64 || Parent->kind() == K_AIXBIG) {
1026     Offset = read64be(Offsets + SymbolIndex * 8);
1027   } else if (Parent->kind() == K_BSD) {
1028     // The SymbolIndex is an index into the ranlib structs that start at
1029     // Offsets (the first uint32_t is the number of bytes of the ranlib
1030     // structs).  The ranlib structs are a pair of uint32_t's the first
1031     // being a string table offset and the second being the offset into
1032     // the archive of the member that defines the symbol.  Which is what
1033     // is needed here.
1034     Offset = read32le(Offsets + SymbolIndex * 8 + 4);
1035   } else if (Parent->kind() == K_DARWIN64) {
1036     // The SymbolIndex is an index into the ranlib_64 structs that start at
1037     // Offsets (the first uint64_t is the number of bytes of the ranlib_64
1038     // structs).  The ranlib_64 structs are a pair of uint64_t's the first
1039     // being a string table offset and the second being the offset into
1040     // the archive of the member that defines the symbol.  Which is what
1041     // is needed here.
1042     Offset = read64le(Offsets + SymbolIndex * 16 + 8);
1043   } else {
1044     // Skip offsets.
1045     uint32_t MemberCount = read32le(Buf);
1046     Buf += MemberCount * 4 + 4;
1047 
1048     uint32_t SymbolCount = read32le(Buf);
1049     uint16_t OffsetIndex;
1050     if (SymbolIndex < SymbolCount) {
1051       // Skip SymbolCount to get to the indices table.
1052       const char *Indices = Buf + 4;
1053 
1054       // Get the index of the offset in the file member offset table for this
1055       // symbol.
1056       OffsetIndex = read16le(Indices + SymbolIndex * 2);
1057     } else if (isECSymbol()) {
1058       // Skip SymbolCount to get to the indices table.
1059       const char *Indices = Parent->ECSymbolTable.begin() + 4;
1060 
1061       // Get the index of the offset in the file member offset table for this
1062       // symbol.
1063       OffsetIndex = read16le(Indices + (SymbolIndex - SymbolCount) * 2);
1064     } else {
1065       return errorCodeToError(object_error::parse_failed);
1066     }
1067     // Subtract 1 since OffsetIndex is 1 based.
1068     --OffsetIndex;
1069 
1070     if (OffsetIndex >= MemberCount)
1071       return errorCodeToError(object_error::parse_failed);
1072 
1073     Offset = read32le(Offsets + OffsetIndex * 4);
1074   }
1075 
1076   const char *Loc = Parent->getData().begin() + Offset;
1077   Error Err = Error::success();
1078   Child C(Parent, Loc, &Err);
1079   if (Err)
1080     return std::move(Err);
1081   return C;
1082 }
1083 
getNext() const1084 Archive::Symbol Archive::Symbol::getNext() const {
1085   Symbol t(*this);
1086   if (Parent->kind() == K_BSD) {
1087     // t.StringIndex is an offset from the start of the __.SYMDEF or
1088     // "__.SYMDEF SORTED" member into the string table for the ranlib
1089     // struct indexed by t.SymbolIndex .  To change t.StringIndex to the
1090     // offset in the string table for t.SymbolIndex+1 we subtract the
1091     // its offset from the start of the string table for t.SymbolIndex
1092     // and add the offset of the string table for t.SymbolIndex+1.
1093 
1094     // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
1095     // which is the number of bytes of ranlib structs that follow.  The ranlib
1096     // structs are a pair of uint32_t's the first being a string table offset
1097     // and the second being the offset into the archive of the member that
1098     // define the symbol. After that the next uint32_t is the byte count of
1099     // the string table followed by the string table.
1100     const char *Buf = Parent->getSymbolTable().begin();
1101     uint32_t RanlibCount = 0;
1102     RanlibCount = read32le(Buf) / 8;
1103     // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount)
1104     // don't change the t.StringIndex as we don't want to reference a ranlib
1105     // past RanlibCount.
1106     if (t.SymbolIndex + 1 < RanlibCount) {
1107       const char *Ranlibs = Buf + 4;
1108       uint32_t CurRanStrx = 0;
1109       uint32_t NextRanStrx = 0;
1110       CurRanStrx = read32le(Ranlibs + t.SymbolIndex * 8);
1111       NextRanStrx = read32le(Ranlibs + (t.SymbolIndex + 1) * 8);
1112       t.StringIndex -= CurRanStrx;
1113       t.StringIndex += NextRanStrx;
1114     }
1115   } else if (t.isECSymbol()) {
1116     // Go to one past next null.
1117     t.StringIndex = Parent->ECSymbolTable.find('\0', t.StringIndex) + 1;
1118   } else {
1119     // Go to one past next null.
1120     t.StringIndex = Parent->getSymbolTable().find('\0', t.StringIndex) + 1;
1121   }
1122   ++t.SymbolIndex;
1123   return t;
1124 }
1125 
symbol_begin() const1126 Archive::symbol_iterator Archive::symbol_begin() const {
1127   if (!hasSymbolTable())
1128     return symbol_iterator(Symbol(this, 0, 0));
1129 
1130   const char *buf = getSymbolTable().begin();
1131   if (kind() == K_GNU) {
1132     uint32_t symbol_count = 0;
1133     symbol_count = read32be(buf);
1134     buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
1135   } else if (kind() == K_GNU64) {
1136     uint64_t symbol_count = read64be(buf);
1137     buf += sizeof(uint64_t) + (symbol_count * (sizeof(uint64_t)));
1138   } else if (kind() == K_BSD) {
1139     // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
1140     // which is the number of bytes of ranlib structs that follow.  The ranlib
1141     // structs are a pair of uint32_t's the first being a string table offset
1142     // and the second being the offset into the archive of the member that
1143     // define the symbol. After that the next uint32_t is the byte count of
1144     // the string table followed by the string table.
1145     uint32_t ranlib_count = 0;
1146     ranlib_count = read32le(buf) / 8;
1147     const char *ranlibs = buf + 4;
1148     uint32_t ran_strx = 0;
1149     ran_strx = read32le(ranlibs);
1150     buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t))));
1151     // Skip the byte count of the string table.
1152     buf += sizeof(uint32_t);
1153     buf += ran_strx;
1154   } else if (kind() == K_DARWIN64) {
1155     // The __.SYMDEF_64 or "__.SYMDEF_64 SORTED" member starts with a uint64_t
1156     // which is the number of bytes of ranlib_64 structs that follow.  The
1157     // ranlib_64 structs are a pair of uint64_t's the first being a string
1158     // table offset and the second being the offset into the archive of the
1159     // member that define the symbol. After that the next uint64_t is the byte
1160     // count of the string table followed by the string table.
1161     uint64_t ranlib_count = 0;
1162     ranlib_count = read64le(buf) / 16;
1163     const char *ranlibs = buf + 8;
1164     uint64_t ran_strx = 0;
1165     ran_strx = read64le(ranlibs);
1166     buf += sizeof(uint64_t) + (ranlib_count * (2 * (sizeof(uint64_t))));
1167     // Skip the byte count of the string table.
1168     buf += sizeof(uint64_t);
1169     buf += ran_strx;
1170   } else if (kind() == K_AIXBIG) {
1171     buf = getStringTable().begin();
1172   } else {
1173     uint32_t member_count = 0;
1174     uint32_t symbol_count = 0;
1175     member_count = read32le(buf);
1176     buf += 4 + (member_count * 4); // Skip offsets.
1177     symbol_count = read32le(buf);
1178     buf += 4 + (symbol_count * 2); // Skip indices.
1179   }
1180   uint32_t string_start_offset = buf - getSymbolTable().begin();
1181   return symbol_iterator(Symbol(this, 0, string_start_offset));
1182 }
1183 
symbol_end() const1184 Archive::symbol_iterator Archive::symbol_end() const {
1185   return symbol_iterator(Symbol(this, getNumberOfSymbols(), 0));
1186 }
1187 
ec_symbols() const1188 Expected<iterator_range<Archive::symbol_iterator>> Archive::ec_symbols() const {
1189   uint32_t Count = 0;
1190 
1191   // Validate EC symbol table.
1192   if (!ECSymbolTable.empty()) {
1193     if (ECSymbolTable.size() < sizeof(uint32_t))
1194       return malformedError("invalid EC symbols size (" +
1195                             Twine(ECSymbolTable.size()) + ")");
1196     if (SymbolTable.size() < sizeof(uint32_t))
1197       return malformedError("invalid symbols size (" +
1198                             Twine(ECSymbolTable.size()) + ")");
1199 
1200     Count = read32le(ECSymbolTable.begin());
1201     size_t StringIndex = sizeof(uint32_t) + Count * sizeof(uint16_t);
1202     if (ECSymbolTable.size() < StringIndex)
1203       return malformedError("invalid EC symbols size. Size was " +
1204                             Twine(ECSymbolTable.size()) + ", but expected " +
1205                             Twine(StringIndex));
1206 
1207     uint32_t MemberCount = read32le(SymbolTable.begin());
1208     const char *Indexes = ECSymbolTable.begin() + sizeof(uint32_t);
1209 
1210     for (uint32_t i = 0; i < Count; ++i) {
1211       uint16_t Index = read16le(Indexes + i * sizeof(uint16_t));
1212       if (!Index)
1213         return malformedError("invalid EC symbol index 0");
1214       if (Index > MemberCount)
1215         return malformedError("invalid EC symbol index " + Twine(Index) +
1216                               " is larger than member count " +
1217                               Twine(MemberCount));
1218 
1219       StringIndex = ECSymbolTable.find('\0', StringIndex);
1220       if (StringIndex == StringRef::npos)
1221         return malformedError("malformed EC symbol names: not null-terminated");
1222       ++StringIndex;
1223     }
1224   }
1225 
1226   uint32_t SymbolCount = getNumberOfSymbols();
1227   return make_range(
1228       symbol_iterator(Symbol(this, SymbolCount,
1229                              sizeof(uint32_t) + Count * sizeof(uint16_t))),
1230       symbol_iterator(Symbol(this, SymbolCount + Count, 0)));
1231 }
1232 
getNumberOfSymbols() const1233 uint32_t Archive::getNumberOfSymbols() const {
1234   if (!hasSymbolTable())
1235     return 0;
1236   const char *buf = getSymbolTable().begin();
1237   if (kind() == K_GNU)
1238     return read32be(buf);
1239   if (kind() == K_GNU64 || kind() == K_AIXBIG)
1240     return read64be(buf);
1241   if (kind() == K_BSD)
1242     return read32le(buf) / 8;
1243   if (kind() == K_DARWIN64)
1244     return read64le(buf) / 16;
1245   uint32_t member_count = 0;
1246   member_count = read32le(buf);
1247   buf += 4 + (member_count * 4); // Skip offsets.
1248   return read32le(buf);
1249 }
1250 
getNumberOfECSymbols() const1251 uint32_t Archive::getNumberOfECSymbols() const {
1252   if (ECSymbolTable.size() < sizeof(uint32_t))
1253     return 0;
1254   return read32le(ECSymbolTable.begin());
1255 }
1256 
findSym(StringRef name) const1257 Expected<std::optional<Archive::Child>> Archive::findSym(StringRef name) const {
1258   Archive::symbol_iterator bs = symbol_begin();
1259   Archive::symbol_iterator es = symbol_end();
1260 
1261   for (; bs != es; ++bs) {
1262     StringRef SymName = bs->getName();
1263     if (SymName == name) {
1264       if (auto MemberOrErr = bs->getMember())
1265         return Child(*MemberOrErr);
1266       else
1267         return MemberOrErr.takeError();
1268     }
1269   }
1270   return std::nullopt;
1271 }
1272 
1273 // Returns true if archive file contains no member file.
isEmpty() const1274 bool Archive::isEmpty() const {
1275   return Data.getBufferSize() == getArchiveMagicLen();
1276 }
1277 
hasSymbolTable() const1278 bool Archive::hasSymbolTable() const { return !SymbolTable.empty(); }
1279 
getGlobalSymtabLocAndSize(const MemoryBufferRef & Data,uint64_t GlobalSymtabOffset,const char * & GlobalSymtabLoc,uint64_t & Size,const char * BitMessage)1280 static Error getGlobalSymtabLocAndSize(const MemoryBufferRef &Data,
1281                                        uint64_t GlobalSymtabOffset,
1282                                        const char *&GlobalSymtabLoc,
1283                                        uint64_t &Size, const char *BitMessage) {
1284   uint64_t BufferSize = Data.getBufferSize();
1285   uint64_t GlobalSymtabContentOffset =
1286       GlobalSymtabOffset + sizeof(BigArMemHdrType);
1287   if (GlobalSymtabContentOffset > BufferSize)
1288     return malformedError(
1289         Twine(BitMessage) + " global symbol table header at offset 0x" +
1290         Twine::utohexstr(GlobalSymtabOffset) + " and size 0x" +
1291         Twine::utohexstr(sizeof(BigArMemHdrType)) +
1292         " goes past the end of file");
1293 
1294   GlobalSymtabLoc = Data.getBufferStart() + GlobalSymtabOffset;
1295   const BigArMemHdrType *GlobalSymHdr =
1296       reinterpret_cast<const BigArMemHdrType *>(GlobalSymtabLoc);
1297   StringRef RawOffset = getFieldRawString(GlobalSymHdr->Size);
1298   if (RawOffset.getAsInteger(10, Size))
1299     return malformedError(Twine(BitMessage) + " global symbol table size \"" +
1300                           RawOffset + "\" is not a number");
1301 
1302   if (GlobalSymtabContentOffset + Size > BufferSize)
1303     return malformedError(
1304         Twine(BitMessage) + " global symbol table content at offset 0x" +
1305         Twine::utohexstr(GlobalSymtabContentOffset) + " and size 0x" +
1306         Twine::utohexstr(Size) + " goes past the end of file");
1307 
1308   return Error::success();
1309 }
1310 
1311 struct GlobalSymtabInfo {
1312   uint64_t SymNum;
1313   StringRef SymbolTable;
1314   StringRef SymbolOffsetTable;
1315   StringRef StringTable;
1316 };
1317 
1318 static void
appendGlobalSymbolTableInfo(SmallVector<GlobalSymtabInfo> & SymtabInfos,const char * GlobalSymtabLoc,uint64_t Size)1319 appendGlobalSymbolTableInfo(SmallVector<GlobalSymtabInfo> &SymtabInfos,
1320                             const char *GlobalSymtabLoc, uint64_t Size) {
1321   // In a big archive, a global symbol table contains the following information:
1322   // - The number of symbols.
1323   // - The array of offsets into the archive file. The length is eight
1324   //   times the number of symbols.
1325   // - The name-string table. The size is:
1326   //   Size-(8*(the number of symbols + 1)).
1327 
1328   StringRef SymbolTable =
1329       StringRef(GlobalSymtabLoc + sizeof(BigArMemHdrType), Size);
1330   uint64_t SymNum = read64be(GlobalSymtabLoc + sizeof(BigArMemHdrType));
1331   StringRef SymbolOffsetTable = StringRef(SymbolTable.data() + 8, 8 * SymNum);
1332   unsigned SymOffsetsSize = 8 * (SymNum + 1);
1333   uint64_t SymbolTableStringSize = Size - SymOffsetsSize;
1334   StringRef StringTable =
1335       StringRef(SymbolTable.data() + SymOffsetsSize, SymbolTableStringSize);
1336   SymtabInfos.push_back({SymNum, SymbolTable, SymbolOffsetTable, StringTable});
1337 }
1338 
BigArchive(MemoryBufferRef Source,Error & Err)1339 BigArchive::BigArchive(MemoryBufferRef Source, Error &Err)
1340     : Archive(Source, Err) {
1341   ErrorAsOutParameter ErrAsOutParam(&Err);
1342   StringRef Buffer = Data.getBuffer();
1343   ArFixLenHdr = reinterpret_cast<const FixLenHdr *>(Buffer.data());
1344   uint64_t BufferSize = Data.getBufferSize();
1345 
1346   if (BufferSize < sizeof(FixLenHdr)) {
1347     Err = malformedError("malformed AIX big archive: incomplete fixed length "
1348                          "header, the archive is only" +
1349                          Twine(BufferSize) + " byte(s)");
1350     return;
1351   }
1352 
1353   StringRef RawOffset = getFieldRawString(ArFixLenHdr->FirstChildOffset);
1354   if (RawOffset.getAsInteger(10, FirstChildOffset))
1355     // TODO: Out-of-line.
1356     Err = malformedError("malformed AIX big archive: first member offset \"" +
1357                          RawOffset + "\" is not a number");
1358 
1359   RawOffset = getFieldRawString(ArFixLenHdr->LastChildOffset);
1360   if (RawOffset.getAsInteger(10, LastChildOffset))
1361     // TODO: Out-of-line.
1362     Err = malformedError("malformed AIX big archive: last member offset \"" +
1363                          RawOffset + "\" is not a number");
1364 
1365   uint64_t GlobSymtab32Offset = 0;
1366   RawOffset = getFieldRawString(ArFixLenHdr->GlobSymOffset);
1367   if (RawOffset.getAsInteger(10, GlobSymtab32Offset)) {
1368     Err = malformedError("global symbol table "
1369                          "offset of 32-bit members \"" +
1370                          RawOffset + "\" is not a number");
1371     return;
1372   }
1373 
1374   uint64_t GlobSymtab64Offset = 0;
1375   RawOffset = getFieldRawString(ArFixLenHdr->GlobSym64Offset);
1376   if (RawOffset.getAsInteger(10, GlobSymtab64Offset)) {
1377     Err = malformedError("global symbol table "
1378                          "offset of 64-bit members\"" +
1379                          RawOffset + "\" is not a number");
1380     return;
1381   }
1382 
1383   const char *GlobSymtab32Loc = nullptr;
1384   const char *GlobSymtab64Loc = nullptr;
1385   uint64_t GlobSymtab32Size = 0;
1386   uint64_t GlobSymtab64Size = 0;
1387   const MemoryBufferRef &MemBuffRef = getMemoryBufferRef();
1388 
1389   if (GlobSymtab32Offset) {
1390     Err =
1391         getGlobalSymtabLocAndSize(MemBuffRef, GlobSymtab32Offset,
1392                                   GlobSymtab32Loc, GlobSymtab32Size, "32-bit");
1393     if (Err)
1394       return;
1395 
1396     Has32BitGlobalSymtab = true;
1397   }
1398 
1399   if (GlobSymtab64Offset) {
1400     Err =
1401         getGlobalSymtabLocAndSize(MemBuffRef, GlobSymtab64Offset,
1402                                   GlobSymtab64Loc, GlobSymtab64Size, "64-bit");
1403     if (Err)
1404       return;
1405 
1406     Has64BitGlobalSymtab = true;
1407   }
1408 
1409   SmallVector<GlobalSymtabInfo> SymtabInfos;
1410 
1411   if (GlobSymtab32Offset)
1412     appendGlobalSymbolTableInfo(SymtabInfos, GlobSymtab32Loc, GlobSymtab32Size);
1413   if (GlobSymtab64Offset)
1414     appendGlobalSymbolTableInfo(SymtabInfos, GlobSymtab64Loc, GlobSymtab64Size);
1415 
1416   if (SymtabInfos.size() == 1) {
1417     SymbolTable = SymtabInfos[0].SymbolTable;
1418     StringTable = SymtabInfos[0].StringTable;
1419   } else if (SymtabInfos.size() == 2) {
1420     // In order to let the Archive::Symbol::getNext() work for both 32-bit and
1421     // 64-bit global symbol tables, we need to merge them into a single table.
1422     raw_string_ostream Out(MergedGlobalSymtabBuf);
1423     uint64_t SymNum = SymtabInfos[0].SymNum + SymtabInfos[1].SymNum;
1424     write(Out, SymNum, llvm::endianness::big);
1425     // Merge symbol offset.
1426     Out << SymtabInfos[0].SymbolOffsetTable;
1427     Out << SymtabInfos[1].SymbolOffsetTable;
1428     // Merge string table.
1429     Out << SymtabInfos[0].StringTable;
1430     Out << SymtabInfos[1].StringTable;
1431     SymbolTable = MergedGlobalSymtabBuf;
1432     // The size of the symbol offset to the member file is 8 bytes.
1433     StringTable = StringRef(SymbolTable.begin() + (SymNum + 1) * 8,
1434                             SymtabInfos[0].StringTable.size() +
1435                                 SymtabInfos[1].StringTable.size());
1436   }
1437 
1438   child_iterator I = child_begin(Err, false);
1439   if (Err)
1440     return;
1441   child_iterator E = child_end();
1442   if (I == E) {
1443     Err = Error::success();
1444     return;
1445   }
1446   setFirstRegular(*I);
1447   Err = Error::success();
1448 }
1449