1 //===- Archive.cpp - ar File Format implementation --------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the ArchiveObjectFile class.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/Object/Archive.h"
15 #include "llvm/ADT/APInt.h"
16 #include "llvm/ADT/SmallString.h"
17 #include "llvm/ADT/Twine.h"
18 #include "llvm/Support/Endian.h"
19 #include "llvm/Support/MemoryBuffer.h"
22 using namespace object;
24 static const char *const Magic = "!<arch>\n";
26 void Archive::anchor() { }
28 StringRef ArchiveMemberHeader::getName() const {
30 if (Name[0] == '/' || Name[0] == '#')
34 llvm::StringRef::size_type end =
35 llvm::StringRef(Name, sizeof(Name)).find(EndCond);
36 if (end == llvm::StringRef::npos)
38 assert(end <= sizeof(Name) && end > 0);
39 // Don't include the EndCond if there is one.
40 return llvm::StringRef(Name, end);
43 uint32_t ArchiveMemberHeader::getSize() const {
45 if (llvm::StringRef(Size, sizeof(Size)).rtrim(" ").getAsInteger(10, Ret))
46 llvm_unreachable("Size is not a decimal number.");
50 sys::fs::perms ArchiveMemberHeader::getAccessMode() const {
52 if (StringRef(AccessMode, sizeof(AccessMode)).rtrim(" ").getAsInteger(8, Ret))
53 llvm_unreachable("Access mode is not an octal number.");
54 return static_cast<sys::fs::perms>(Ret);
57 sys::TimeValue ArchiveMemberHeader::getLastModified() const {
59 if (StringRef(LastModified, sizeof(LastModified)).rtrim(" ")
60 .getAsInteger(10, Seconds))
61 llvm_unreachable("Last modified time not a decimal number.");
64 Ret.fromEpochTime(Seconds);
68 unsigned ArchiveMemberHeader::getUID() const {
70 if (StringRef(UID, sizeof(UID)).rtrim(" ").getAsInteger(10, Ret))
71 llvm_unreachable("UID time not a decimal number.");
75 unsigned ArchiveMemberHeader::getGID() const {
77 if (StringRef(GID, sizeof(GID)).rtrim(" ").getAsInteger(10, Ret))
78 llvm_unreachable("GID time not a decimal number.");
82 Archive::Child::Child(const Archive *Parent, const char *Start)
87 const ArchiveMemberHeader *Header =
88 reinterpret_cast<const ArchiveMemberHeader *>(Start);
89 Data = StringRef(Start, sizeof(ArchiveMemberHeader) + Header->getSize());
91 // Setup StartOfFile and PaddingBytes.
92 StartOfFile = sizeof(ArchiveMemberHeader);
93 // Don't include attached name.
94 StringRef Name = Header->getName();
95 if (Name.startswith("#1/")) {
97 if (Name.substr(3).rtrim(" ").getAsInteger(10, NameSize))
98 llvm_unreachable("Long name length is not an integer");
99 StartOfFile += NameSize;
103 Archive::Child Archive::Child::getNext() const {
104 size_t SpaceToSkip = Data.size();
105 // If it's odd, add 1 to make it even.
109 const char *NextLoc = Data.data() + SpaceToSkip;
111 // Check to see if this is past the end of the archive.
112 if (NextLoc >= Parent->Data.getBufferEnd())
113 return Child(Parent, nullptr);
115 return Child(Parent, NextLoc);
118 ErrorOr<StringRef> Archive::Child::getName() const {
119 StringRef name = getRawName();
120 // Check if it's a special name.
121 if (name[0] == '/') {
122 if (name.size() == 1) // Linker member.
124 if (name.size() == 2 && name[1] == '/') // String table.
129 if (name.substr(1).rtrim(" ").getAsInteger(10, offset))
130 llvm_unreachable("Long name offset is not an integer");
131 const char *addr = Parent->StringTable->Data.begin()
132 + sizeof(ArchiveMemberHeader)
135 if (Parent->StringTable == Parent->child_end()
136 || addr < (Parent->StringTable->Data.begin()
137 + sizeof(ArchiveMemberHeader))
138 || addr > (Parent->StringTable->Data.begin()
139 + sizeof(ArchiveMemberHeader)
140 + Parent->StringTable->getSize()))
141 return object_error::parse_failed;
143 // GNU long file names end with a /.
144 if (Parent->kind() == K_GNU) {
145 StringRef::size_type End = StringRef(addr).find('/');
146 return StringRef(addr, End);
148 return StringRef(addr);
149 } else if (name.startswith("#1/")) {
151 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size))
152 llvm_unreachable("Long name length is not an ingeter");
153 return Data.substr(sizeof(ArchiveMemberHeader), name_size)
154 .rtrim(StringRef("\0", 1));
156 // It's a simple name.
157 if (name[name.size() - 1] == '/')
158 return name.substr(0, name.size() - 1);
162 ErrorOr<MemoryBufferRef> Archive::Child::getMemoryBufferRef() const {
163 ErrorOr<StringRef> NameOrErr = getName();
164 if (std::error_code EC = NameOrErr.getError())
166 StringRef Name = NameOrErr.get();
167 return MemoryBufferRef(getBuffer(), Name);
170 ErrorOr<std::unique_ptr<Binary>>
171 Archive::Child::getAsBinary(LLVMContext *Context) const {
172 ErrorOr<MemoryBufferRef> BuffOrErr = getMemoryBufferRef();
173 if (std::error_code EC = BuffOrErr.getError())
176 return createBinary(BuffOrErr.get(), Context);
179 ErrorOr<std::unique_ptr<Archive>> Archive::create(MemoryBufferRef Source) {
181 std::unique_ptr<Archive> Ret(new Archive(Source, EC));
184 return std::move(Ret);
187 Archive::Archive(MemoryBufferRef Source, std::error_code &ec)
188 : Binary(Binary::ID_Archive, Source), SymbolTable(child_end()) {
189 // Check for sufficient magic.
190 if (Data.getBufferSize() < 8 ||
191 StringRef(Data.getBufferStart(), 8) != Magic) {
192 ec = object_error::invalid_file_type;
196 // Get the special members.
197 child_iterator i = child_begin(false);
198 child_iterator e = child_end();
201 ec = object_error::success;
205 StringRef Name = i->getRawName();
207 // Below is the pattern that is used to figure out the archive format
208 // GNU archive format
209 // First member : / (may exist, if it exists, points to the symbol table )
210 // Second member : // (may exist, if it exists, points to the string table)
211 // Note : The string table is used if the filename exceeds 15 characters
212 // BSD archive format
213 // First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table)
214 // There is no string table, if the filename exceeds 15 characters or has a
215 // embedded space, the filename has #1/<size>, The size represents the size
216 // of the filename that needs to be read after the archive header
217 // COFF archive format
219 // Second member : / (provides a directory of symbols)
220 // Third member : // (may exist, if it exists, contains the string table)
221 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
222 // even if the string table is empty. However, lib.exe does not in fact
223 // seem to create the third member if there's no member whose filename
224 // exceeds 15 characters. So the third member is optional.
226 if (Name == "__.SYMDEF") {
231 ec = object_error::success;
235 if (Name.startswith("#1/")) {
237 // We know this is BSD, so getName will work since there is no string table.
238 ErrorOr<StringRef> NameOrErr = i->getName();
239 ec = NameOrErr.getError();
242 Name = NameOrErr.get();
243 if (Name == "__.SYMDEF SORTED") {
256 ec = object_error::parse_failed;
259 Name = i->getRawName();
267 ec = object_error::success;
271 if (Name[0] != '/') {
274 ec = object_error::success;
279 ec = object_error::parse_failed;
289 ec = object_error::success;
293 Name = i->getRawName();
301 ec = object_error::success;
304 Archive::child_iterator Archive::child_begin(bool SkipInternal) const {
305 if (Data.getBufferSize() == 8) // empty archive.
311 const char *Loc = Data.getBufferStart() + strlen(Magic);
316 Archive::child_iterator Archive::child_end() const {
317 return Child(this, nullptr);
320 StringRef Archive::Symbol::getName() const {
321 return Parent->SymbolTable->getBuffer().begin() + StringIndex;
324 ErrorOr<Archive::child_iterator> Archive::Symbol::getMember() const {
325 const char *Buf = Parent->SymbolTable->getBuffer().begin();
326 const char *Offsets = Buf + 4;
328 if (Parent->kind() == K_GNU) {
329 Offset = *(reinterpret_cast<const support::ubig32_t*>(Offsets)
331 } else if (Parent->kind() == K_BSD) {
332 // The SymbolIndex is an index into the ranlib structs that start at
333 // Offsets (the first uint32_t is the number of bytes of the ranlib
334 // structs). The ranlib structs are a pair of uint32_t's the first
335 // being a string table offset and the second being the offset into
336 // the archive of the member that defines the symbol. Which is what
338 Offset = *(reinterpret_cast<const support::ulittle32_t *>(Offsets) +
339 (SymbolIndex * 2) + 1);
341 uint32_t MemberCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
344 Buf += sizeof(support::ulittle32_t)
345 + (MemberCount * sizeof(support::ulittle32_t));
347 uint32_t SymbolCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
349 if (SymbolIndex >= SymbolCount)
350 return object_error::parse_failed;
352 // Skip SymbolCount to get to the indices table.
353 const char *Indices = Buf + sizeof(support::ulittle32_t);
355 // Get the index of the offset in the file member offset table for this
357 uint16_t OffsetIndex =
358 *(reinterpret_cast<const support::ulittle16_t*>(Indices)
360 // Subtract 1 since OffsetIndex is 1 based.
363 if (OffsetIndex >= MemberCount)
364 return object_error::parse_failed;
366 Offset = *(reinterpret_cast<const support::ulittle32_t*>(Offsets)
370 const char *Loc = Parent->getData().begin() + Offset;
371 child_iterator Iter(Child(Parent, Loc));
375 Archive::Symbol Archive::Symbol::getNext() const {
377 if (Parent->kind() == K_BSD) {
378 // t.StringIndex is an offset from the start of the __.SYMDEF or
379 // "__.SYMDEF SORTED" member into the string table for the ranlib
380 // struct indexed by t.SymbolIndex . To change t.StringIndex to the
381 // offset in the string table for t.SymbolIndex+1 we subtract the
382 // its offset from the start of the string table for t.SymbolIndex
383 // and add the offset of the string table for t.SymbolIndex+1.
385 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
386 // which is the number of bytes of ranlib structs that follow. The ranlib
387 // structs are a pair of uint32_t's the first being a string table offset
388 // and the second being the offset into the archive of the member that
389 // define the symbol. After that the next uint32_t is the byte count of
390 // the string table followed by the string table.
391 const char *Buf = Parent->SymbolTable->getBuffer().begin();
392 uint32_t RanlibCount = 0;
393 RanlibCount = (*reinterpret_cast<const support::ulittle32_t *>(Buf)) /
394 (sizeof(uint32_t) * 2);
395 // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount)
396 // don't change the t.StringIndex as we don't want to reference a ranlib
398 if (t.SymbolIndex + 1 < RanlibCount) {
399 const char *Ranlibs = Buf + 4;
400 uint32_t CurRanStrx = 0;
401 uint32_t NextRanStrx = 0;
402 CurRanStrx = *(reinterpret_cast<const support::ulittle32_t *>(Ranlibs) +
403 (t.SymbolIndex * 2));
404 NextRanStrx = *(reinterpret_cast<const support::ulittle32_t *>(Ranlibs) +
405 ((t.SymbolIndex + 1) * 2));
406 t.StringIndex -= CurRanStrx;
407 t.StringIndex += NextRanStrx;
410 // Go to one past next null.
412 Parent->SymbolTable->getBuffer().find('\0', t.StringIndex) + 1;
418 Archive::symbol_iterator Archive::symbol_begin() const {
419 if (!hasSymbolTable())
420 return symbol_iterator(Symbol(this, 0, 0));
422 const char *buf = SymbolTable->getBuffer().begin();
423 if (kind() == K_GNU) {
424 uint32_t symbol_count = 0;
425 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
426 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
427 } else if (kind() == K_BSD) {
428 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
429 // which is the number of bytes of ranlib structs that follow. The ranlib
430 // structs are a pair of uint32_t's the first being a string table offset
431 // and the second being the offset into the archive of the member that
432 // define the symbol. After that the next uint32_t is the byte count of
433 // the string table followed by the string table.
434 uint32_t ranlib_count = 0;
435 ranlib_count = (*reinterpret_cast<const support::ulittle32_t *>(buf)) /
436 (sizeof(uint32_t) * 2);
437 const char *ranlibs = buf + 4;
438 uint32_t ran_strx = 0;
439 ran_strx = *(reinterpret_cast<const support::ulittle32_t *>(ranlibs));
440 buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t))));
441 // Skip the byte count of the string table.
442 buf += sizeof(uint32_t);
445 uint32_t member_count = 0;
446 uint32_t symbol_count = 0;
447 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
448 buf += 4 + (member_count * 4); // Skip offsets.
449 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
450 buf += 4 + (symbol_count * 2); // Skip indices.
452 uint32_t string_start_offset = buf - SymbolTable->getBuffer().begin();
453 return symbol_iterator(Symbol(this, 0, string_start_offset));
456 Archive::symbol_iterator Archive::symbol_end() const {
457 if (!hasSymbolTable())
458 return symbol_iterator(Symbol(this, 0, 0));
460 const char *buf = SymbolTable->getBuffer().begin();
461 uint32_t symbol_count = 0;
462 if (kind() == K_GNU) {
463 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
464 } else if (kind() == K_BSD) {
465 symbol_count = (*reinterpret_cast<const support::ulittle32_t *>(buf)) /
466 (sizeof(uint32_t) * 2);
468 uint32_t member_count = 0;
469 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
470 buf += 4 + (member_count * 4); // Skip offsets.
471 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
473 return symbol_iterator(
474 Symbol(this, symbol_count, 0));
477 Archive::child_iterator Archive::findSym(StringRef name) const {
478 Archive::symbol_iterator bs = symbol_begin();
479 Archive::symbol_iterator es = symbol_end();
481 for (; bs != es; ++bs) {
482 StringRef SymName = bs->getName();
483 if (SymName == name) {
484 ErrorOr<Archive::child_iterator> ResultOrErr = bs->getMember();
485 // FIXME: Should we really eat the error?
486 if (ResultOrErr.getError())
488 return ResultOrErr.get();
494 bool Archive::hasSymbolTable() const {
495 return SymbolTable != child_end();