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 std::error_code Archive::Child::getName(StringRef &Result) 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 return object_error::success;
126 if (name.size() == 2 && name[1] == '/') { // String table.
128 return object_error::success;
133 if (name.substr(1).rtrim(" ").getAsInteger(10, offset))
134 llvm_unreachable("Long name offset is not an integer");
135 const char *addr = Parent->StringTable->Data.begin()
136 + sizeof(ArchiveMemberHeader)
139 if (Parent->StringTable == Parent->child_end()
140 || addr < (Parent->StringTable->Data.begin()
141 + sizeof(ArchiveMemberHeader))
142 || addr > (Parent->StringTable->Data.begin()
143 + sizeof(ArchiveMemberHeader)
144 + Parent->StringTable->getSize()))
145 return object_error::parse_failed;
147 // GNU long file names end with a /.
148 if (Parent->kind() == K_GNU) {
149 StringRef::size_type End = StringRef(addr).find('/');
150 Result = StringRef(addr, End);
154 return object_error::success;
155 } else if (name.startswith("#1/")) {
157 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size))
158 llvm_unreachable("Long name length is not an ingeter");
159 Result = Data.substr(sizeof(ArchiveMemberHeader), name_size)
160 .rtrim(StringRef("\0", 1));
161 return object_error::success;
163 // It's a simple name.
164 if (name[name.size() - 1] == '/')
165 Result = name.substr(0, name.size() - 1);
168 return object_error::success;
172 Archive::Child::getMemoryBuffer(std::unique_ptr<MemoryBuffer> &Result,
173 bool FullPath) const {
175 if (std::error_code ec = getName(Name))
177 SmallString<128> Path;
178 Result.reset(MemoryBuffer::getMemBuffer(
179 getBuffer(), FullPath ? (Twine(Parent->getFileName()) + "(" + Name + ")")
183 return std::error_code();
186 std::error_code Archive::Child::getAsBinary(std::unique_ptr<Binary> &Result,
187 LLVMContext *Context) const {
188 std::unique_ptr<Binary> ret;
189 std::unique_ptr<MemoryBuffer> Buff;
190 if (std::error_code ec = getMemoryBuffer(Buff))
192 ErrorOr<Binary *> BinaryOrErr = createBinary(Buff.release(), Context);
193 if (std::error_code EC = BinaryOrErr.getError())
195 Result.reset(BinaryOrErr.get());
196 return object_error::success;
199 ErrorOr<Archive*> Archive::create(MemoryBuffer *Source) {
201 std::unique_ptr<Archive> Ret(new Archive(Source, EC));
204 return Ret.release();
207 Archive::Archive(MemoryBuffer *source, std::error_code &ec)
208 : Binary(Binary::ID_Archive, source), SymbolTable(child_end()) {
209 // Check for sufficient magic.
211 if (source->getBufferSize() < 8 ||
212 StringRef(source->getBufferStart(), 8) != Magic) {
213 ec = object_error::invalid_file_type;
217 // Get the special members.
218 child_iterator i = child_begin(false);
219 child_iterator e = child_end();
222 ec = object_error::success;
226 StringRef Name = i->getRawName();
228 // Below is the pattern that is used to figure out the archive format
229 // GNU archive format
230 // First member : / (may exist, if it exists, points to the symbol table )
231 // Second member : // (may exist, if it exists, points to the string table)
232 // Note : The string table is used if the filename exceeds 15 characters
233 // BSD archive format
234 // First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table)
235 // There is no string table, if the filename exceeds 15 characters or has a
236 // embedded space, the filename has #1/<size>, The size represents the size
237 // of the filename that needs to be read after the archive header
238 // COFF archive format
240 // Second member : / (provides a directory of symbols)
241 // Third member : // (may exist, if it exists, contains the string table)
242 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
243 // even if the string table is empty. However, lib.exe does not in fact
244 // seem to create the third member if there's no member whose filename
245 // exceeds 15 characters. So the third member is optional.
247 if (Name == "__.SYMDEF") {
252 ec = object_error::success;
256 if (Name.startswith("#1/")) {
258 // We know this is BSD, so getName will work since there is no string table.
259 ec = i->getName(Name);
262 if (Name == "__.SYMDEF SORTED") {
275 ec = object_error::parse_failed;
278 Name = i->getRawName();
286 ec = object_error::success;
290 if (Name[0] != '/') {
293 ec = object_error::success;
298 ec = object_error::parse_failed;
308 ec = object_error::success;
312 Name = i->getRawName();
320 ec = object_error::success;
323 Archive::child_iterator Archive::child_begin(bool SkipInternal) const {
324 if (Data->getBufferSize() == 8) // empty archive.
330 const char *Loc = Data->getBufferStart() + strlen(Magic);
335 Archive::child_iterator Archive::child_end() const {
336 return Child(this, nullptr);
339 std::error_code Archive::Symbol::getName(StringRef &Result) const {
340 Result = StringRef(Parent->SymbolTable->getBuffer().begin() + StringIndex);
341 return object_error::success;
344 std::error_code Archive::Symbol::getMember(child_iterator &Result) const {
345 const char *Buf = Parent->SymbolTable->getBuffer().begin();
346 const char *Offsets = Buf + 4;
348 if (Parent->kind() == K_GNU) {
349 Offset = *(reinterpret_cast<const support::ubig32_t*>(Offsets)
351 } else if (Parent->kind() == K_BSD) {
352 llvm_unreachable("BSD format is not supported");
354 uint32_t MemberCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
357 Buf += sizeof(support::ulittle32_t)
358 + (MemberCount * sizeof(support::ulittle32_t));
360 uint32_t SymbolCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
362 if (SymbolIndex >= SymbolCount)
363 return object_error::parse_failed;
365 // Skip SymbolCount to get to the indices table.
366 const char *Indices = Buf + sizeof(support::ulittle32_t);
368 // Get the index of the offset in the file member offset table for this
370 uint16_t OffsetIndex =
371 *(reinterpret_cast<const support::ulittle16_t*>(Indices)
373 // Subtract 1 since OffsetIndex is 1 based.
376 if (OffsetIndex >= MemberCount)
377 return object_error::parse_failed;
379 Offset = *(reinterpret_cast<const support::ulittle32_t*>(Offsets)
383 const char *Loc = Parent->getData().begin() + Offset;
384 Result = Child(Parent, Loc);
386 return object_error::success;
389 Archive::Symbol Archive::Symbol::getNext() const {
391 // Go to one past next null.
393 Parent->SymbolTable->getBuffer().find('\0', t.StringIndex) + 1;
398 Archive::symbol_iterator Archive::symbol_begin() const {
399 if (!hasSymbolTable())
400 return symbol_iterator(Symbol(this, 0, 0));
402 const char *buf = SymbolTable->getBuffer().begin();
403 if (kind() == K_GNU) {
404 uint32_t symbol_count = 0;
405 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
406 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
407 } else if (kind() == K_BSD) {
408 llvm_unreachable("BSD archive format is not supported");
410 uint32_t member_count = 0;
411 uint32_t symbol_count = 0;
412 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
413 buf += 4 + (member_count * 4); // Skip offsets.
414 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
415 buf += 4 + (symbol_count * 2); // Skip indices.
417 uint32_t string_start_offset = buf - SymbolTable->getBuffer().begin();
418 return symbol_iterator(Symbol(this, 0, string_start_offset));
421 Archive::symbol_iterator Archive::symbol_end() const {
422 if (!hasSymbolTable())
423 return symbol_iterator(Symbol(this, 0, 0));
425 const char *buf = SymbolTable->getBuffer().begin();
426 uint32_t symbol_count = 0;
427 if (kind() == K_GNU) {
428 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
429 } else if (kind() == K_BSD) {
430 llvm_unreachable("BSD archive format is not supported");
432 uint32_t member_count = 0;
433 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
434 buf += 4 + (member_count * 4); // Skip offsets.
435 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
437 return symbol_iterator(
438 Symbol(this, symbol_count, 0));
441 Archive::child_iterator Archive::findSym(StringRef name) const {
442 Archive::symbol_iterator bs = symbol_begin();
443 Archive::symbol_iterator es = symbol_end();
444 Archive::child_iterator result;
447 for (; bs != es; ++bs) {
448 if (bs->getName(symname))
450 if (symname == name) {
451 if (bs->getMember(result))
459 bool Archive::hasSymbolTable() const {
460 return SymbolTable != child_end();