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 *Magic = "!<arch>\n";
26 static bool isInternalMember(const ArchiveMemberHeader &amh) {
27 static const char *const internals[] = {
32 StringRef name = amh.getName();
33 for (std::size_t i = 0; i < sizeof(internals) / sizeof(*internals); ++i) {
34 if (name == internals[i])
40 void Archive::anchor() { }
42 StringRef ArchiveMemberHeader::getName() const {
44 if (Name[0] == '/' || Name[0] == '#')
48 llvm::StringRef::size_type end =
49 llvm::StringRef(Name, sizeof(Name)).find(EndCond);
50 if (end == llvm::StringRef::npos)
52 assert(end <= sizeof(Name) && end > 0);
53 // Don't include the EndCond if there is one.
54 return llvm::StringRef(Name, end);
57 uint32_t ArchiveMemberHeader::getSize() const {
59 if (llvm::StringRef(Size, sizeof(Size)).rtrim(" ").getAsInteger(10, Ret))
60 llvm_unreachable("Size is not a decimal number.");
64 sys::fs::perms ArchiveMemberHeader::getAccessMode() const {
66 if (StringRef(AccessMode, sizeof(AccessMode)).rtrim(" ").getAsInteger(8, Ret))
67 llvm_unreachable("Access mode is not an octal number.");
68 return static_cast<sys::fs::perms>(Ret);
71 sys::TimeValue ArchiveMemberHeader::getLastModified() const {
73 if (StringRef(LastModified, sizeof(LastModified)).rtrim(" ")
74 .getAsInteger(10, Seconds))
75 llvm_unreachable("Last modified time not a decimal number.");
78 Ret.fromEpochTime(Seconds);
82 unsigned ArchiveMemberHeader::getUID() const {
84 if (StringRef(UID, sizeof(UID)).rtrim(" ").getAsInteger(10, Ret))
85 llvm_unreachable("UID time not a decimal number.");
89 unsigned ArchiveMemberHeader::getGID() const {
91 if (StringRef(GID, sizeof(GID)).rtrim(" ").getAsInteger(10, Ret))
92 llvm_unreachable("GID time not a decimal number.");
96 static const ArchiveMemberHeader *toHeader(const char *base) {
97 return reinterpret_cast<const ArchiveMemberHeader *>(base);
100 Archive::Child::Child(const Archive *Parent, const char *Start)
105 const ArchiveMemberHeader *Header = toHeader(Start);
106 Data = StringRef(Start, sizeof(ArchiveMemberHeader) + Header->getSize());
108 // Setup StartOfFile and PaddingBytes.
109 StartOfFile = sizeof(ArchiveMemberHeader);
110 // Don't include attached name.
111 StringRef Name = Header->getName();
112 if (Name.startswith("#1/")) {
114 if (Name.substr(3).rtrim(" ").getAsInteger(10, NameSize))
115 llvm_unreachable("Long name length is not an integer");
116 StartOfFile += NameSize;
120 Archive::Child Archive::Child::getNext() const {
121 size_t SpaceToSkip = Data.size();
122 // If it's odd, add 1 to make it even.
126 const char *NextLoc = Data.data() + SpaceToSkip;
128 // Check to see if this is past the end of the archive.
129 if (NextLoc >= Parent->Data->getBufferEnd())
130 return Child(Parent, NULL);
132 return Child(Parent, NextLoc);
135 error_code Archive::Child::getName(StringRef &Result) const {
136 StringRef name = getRawName();
137 // Check if it's a special name.
138 if (name[0] == '/') {
139 if (name.size() == 1) { // Linker member.
141 return object_error::success;
143 if (name.size() == 2 && name[1] == '/') { // String table.
145 return object_error::success;
150 if (name.substr(1).rtrim(" ").getAsInteger(10, offset))
151 llvm_unreachable("Long name offset is not an integer");
152 const char *addr = Parent->StringTable->Data.begin()
153 + sizeof(ArchiveMemberHeader)
156 if (Parent->StringTable == Parent->end_children()
157 || addr < (Parent->StringTable->Data.begin()
158 + sizeof(ArchiveMemberHeader))
159 || addr > (Parent->StringTable->Data.begin()
160 + sizeof(ArchiveMemberHeader)
161 + Parent->StringTable->getSize()))
162 return object_error::parse_failed;
164 // GNU long file names end with a /.
165 if (Parent->kind() == K_GNU) {
166 StringRef::size_type End = StringRef(addr).find('/');
167 Result = StringRef(addr, End);
171 return object_error::success;
172 } else if (name.startswith("#1/")) {
174 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size))
175 llvm_unreachable("Long name length is not an ingeter");
176 Result = Data.substr(sizeof(ArchiveMemberHeader), name_size);
177 return object_error::success;
179 // It's a simple name.
180 if (name[name.size() - 1] == '/')
181 Result = name.substr(0, name.size() - 1);
184 return object_error::success;
187 error_code Archive::Child::getMemoryBuffer(OwningPtr<MemoryBuffer> &Result,
188 bool FullPath) const {
190 if (error_code ec = getName(Name))
192 SmallString<128> Path;
193 Result.reset(MemoryBuffer::getMemBuffer(
194 getBuffer(), FullPath ? (Twine(Parent->getFileName()) + "(" + Name + ")")
198 return error_code::success();
201 error_code Archive::Child::getAsBinary(OwningPtr<Binary> &Result) const {
202 OwningPtr<Binary> ret;
203 OwningPtr<MemoryBuffer> Buff;
204 if (error_code ec = getMemoryBuffer(Buff))
206 if (error_code ec = createBinary(Buff.take(), ret))
209 return object_error::success;
212 Archive::Archive(MemoryBuffer *source, error_code &ec)
213 : Binary(Binary::ID_Archive, source) {
214 // Check for sufficient magic.
215 if (!source || source->getBufferSize()
216 < (8 + sizeof(ArchiveMemberHeader)) // Smallest archive.
217 || StringRef(source->getBufferStart(), 8) != Magic) {
218 ec = object_error::invalid_file_type;
222 // Get the special members.
223 child_iterator i = begin_children(false);
224 child_iterator e = end_children();
227 ec = object_error::parse_failed;
231 StringRef Name = i->getRawName();
233 // Below is the pattern that is used to figure out the archive format
234 // GNU archive format
235 // First member : / (may exist, if it exists, points to the symbol table )
236 // Second member : // (may exist, if it exists, points to the string table)
237 // Note : The string table is used if the filename exceeds 15 characters
238 // BSD archive format
239 // First member : __.SYMDEF (points to the symbol table)
240 // There is no string table, if the filename exceeds 15 characters or has a
241 // embedded space, the filename has #1/<size>, The size represents the size
242 // of the filename that needs to be read after the archive header
243 // COFF archive format
245 // Second member : / (provides a directory of symbols)
246 // Third member : // (may exist, if it exists, contains the string table)
247 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
248 // even if the string table is empty. However, lib.exe does not in fact
249 // seem to create the third member if there's no member whose filename
250 // exceeds 15 characters. So the third member is optional.
252 if (Name == "__.SYMDEF") {
255 ec = object_error::success;
264 ec = object_error::parse_failed;
267 Name = i->getRawName();
273 ec = object_error::success;
277 if (Name[0] != '/') {
279 ec = object_error::success;
284 ec = object_error::parse_failed;
293 ec = object_error::success;
297 Name = i->getRawName();
302 ec = object_error::success;
305 Archive::child_iterator Archive::begin_children(bool skip_internal) const {
306 const char *Loc = Data->getBufferStart() + strlen(Magic);
308 // Skip internals at the beginning of an archive.
309 if (skip_internal && isInternalMember(*toHeader(Loc)))
314 Archive::child_iterator Archive::end_children() const {
315 return Child(this, NULL);
318 error_code Archive::Symbol::getName(StringRef &Result) const {
319 Result = StringRef(Parent->SymbolTable->getBuffer().begin() + StringIndex);
320 return object_error::success;
323 error_code Archive::Symbol::getMember(child_iterator &Result) const {
324 const char *Buf = Parent->SymbolTable->getBuffer().begin();
325 const char *Offsets = Buf + 4;
327 if (Parent->kind() == K_GNU) {
328 Offset = *(reinterpret_cast<const support::ubig32_t*>(Offsets)
330 } else if (Parent->kind() == K_BSD) {
331 llvm_unreachable("BSD format is not supported");
333 uint32_t MemberCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
336 Buf += sizeof(support::ulittle32_t)
337 + (MemberCount * sizeof(support::ulittle32_t));
339 uint32_t SymbolCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
341 if (SymbolIndex >= SymbolCount)
342 return object_error::parse_failed;
344 // Skip SymbolCount to get to the indices table.
345 const char *Indices = Buf + sizeof(support::ulittle32_t);
347 // Get the index of the offset in the file member offset table for this
349 uint16_t OffsetIndex =
350 *(reinterpret_cast<const support::ulittle16_t*>(Indices)
352 // Subtract 1 since OffsetIndex is 1 based.
355 if (OffsetIndex >= MemberCount)
356 return object_error::parse_failed;
358 Offset = *(reinterpret_cast<const support::ulittle32_t*>(Offsets)
362 const char *Loc = Parent->getData().begin() + Offset;
363 Result = Child(Parent, Loc);
365 return object_error::success;
368 Archive::Symbol Archive::Symbol::getNext() const {
370 // Go to one past next null.
372 Parent->SymbolTable->getBuffer().find('\0', t.StringIndex) + 1;
377 Archive::symbol_iterator Archive::begin_symbols() const {
378 const char *buf = SymbolTable->getBuffer().begin();
379 if (kind() == K_GNU) {
380 uint32_t symbol_count = 0;
381 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
382 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
383 } else if (kind() == K_BSD) {
384 llvm_unreachable("BSD archive format is not supported");
386 uint32_t member_count = 0;
387 uint32_t symbol_count = 0;
388 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
389 buf += 4 + (member_count * 4); // Skip offsets.
390 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
391 buf += 4 + (symbol_count * 2); // Skip indices.
393 uint32_t string_start_offset = buf - SymbolTable->getBuffer().begin();
394 return symbol_iterator(Symbol(this, 0, string_start_offset));
397 Archive::symbol_iterator Archive::end_symbols() const {
398 const char *buf = SymbolTable->getBuffer().begin();
399 uint32_t symbol_count = 0;
400 if (kind() == K_GNU) {
401 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
402 } else if (kind() == K_BSD) {
403 llvm_unreachable("BSD archive format is not supported");
405 uint32_t member_count = 0;
406 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
407 buf += 4 + (member_count * 4); // Skip offsets.
408 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
410 return symbol_iterator(
411 Symbol(this, symbol_count, 0));
414 Archive::child_iterator Archive::findSym(StringRef name) const {
415 Archive::symbol_iterator bs = begin_symbols();
416 Archive::symbol_iterator es = end_symbols();
417 Archive::child_iterator result;
420 for (; bs != es; ++bs) {
421 if (bs->getName(symname))
422 return end_children();
423 if (symname == name) {
424 if (bs->getMember(result))
425 return end_children();
429 return end_children();