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 uint64_t ArchiveMemberHeader::getSize() const {
59 if (llvm::StringRef(Size, sizeof(Size)).rtrim(" ").getAsInteger(10, ret))
60 llvm_unreachable("Size is not an integer.");
64 Archive::Child::Child(const Archive *Parent, const char *Start)
69 const ArchiveMemberHeader *Header = ToHeader(Start);
70 Data = StringRef(Start, sizeof(ArchiveMemberHeader) + Header->getSize());
72 // Setup StartOfFile and PaddingBytes.
73 StartOfFile = sizeof(ArchiveMemberHeader);
74 // Don't include attached name.
75 StringRef Name = Header->getName();
76 if (Name.startswith("#1/")) {
78 if (Name.substr(3).rtrim(" ").getAsInteger(10, NameSize))
79 llvm_unreachable("Long name length is not an integer");
80 StartOfFile += NameSize;
84 Archive::Child Archive::Child::getNext() const {
85 size_t SpaceToSkip = Data.size();
86 // If it's odd, add 1 to make it even.
90 const char *NextLoc = Data.data() + SpaceToSkip;
92 // Check to see if this is past the end of the archive.
93 if (NextLoc >= Parent->Data->getBufferEnd())
94 return Child(Parent, NULL);
96 return Child(Parent, NextLoc);
99 error_code Archive::Child::getName(StringRef &Result) const {
100 StringRef name = getRawName();
101 // Check if it's a special name.
102 if (name[0] == '/') {
103 if (name.size() == 1) { // Linker member.
105 return object_error::success;
107 if (name.size() == 2 && name[1] == '/') { // String table.
109 return object_error::success;
114 if (name.substr(1).rtrim(" ").getAsInteger(10, offset))
115 llvm_unreachable("Long name offset is not an integer");
116 const char *addr = Parent->StringTable->Data.begin()
117 + sizeof(ArchiveMemberHeader)
120 if (Parent->StringTable == Parent->end_children()
121 || addr < (Parent->StringTable->Data.begin()
122 + sizeof(ArchiveMemberHeader))
123 || addr > (Parent->StringTable->Data.begin()
124 + sizeof(ArchiveMemberHeader)
125 + Parent->StringTable->getSize()))
126 return object_error::parse_failed;
128 // GNU long file names end with a /.
129 if (Parent->kind() == K_GNU) {
130 StringRef::size_type End = StringRef(addr).find('/');
131 Result = StringRef(addr, End);
135 return object_error::success;
136 } else if (name.startswith("#1/")) {
138 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size))
139 llvm_unreachable("Long name length is not an ingeter");
140 Result = Data.substr(sizeof(ArchiveMemberHeader), name_size);
141 return object_error::success;
143 // It's a simple name.
144 if (name[name.size() - 1] == '/')
145 Result = name.substr(0, name.size() - 1);
148 return object_error::success;
151 error_code Archive::Child::getMemoryBuffer(OwningPtr<MemoryBuffer> &Result,
152 bool FullPath) const {
154 if (error_code ec = getName(Name))
156 SmallString<128> Path;
157 Result.reset(MemoryBuffer::getMemBuffer(
158 getBuffer(), FullPath ? (Twine(Parent->getFileName()) + "(" + Name + ")")
162 return error_code::success();
165 error_code Archive::Child::getAsBinary(OwningPtr<Binary> &Result) const {
166 OwningPtr<Binary> ret;
167 OwningPtr<MemoryBuffer> Buff;
168 if (error_code ec = getMemoryBuffer(Buff))
170 if (error_code ec = createBinary(Buff.take(), ret))
173 return object_error::success;
176 Archive::Archive(MemoryBuffer *source, error_code &ec)
177 : Binary(Binary::ID_Archive, source) {
178 // Check for sufficient magic.
179 if (!source || source->getBufferSize()
180 < (8 + sizeof(ArchiveMemberHeader)) // Smallest archive.
181 || StringRef(source->getBufferStart(), 8) != Magic) {
182 ec = object_error::invalid_file_type;
186 // Get the special members.
187 child_iterator i = begin_children(false);
188 child_iterator e = end_children();
191 ec = object_error::parse_failed;
195 StringRef Name = i->getRawName();
197 // Below is the pattern that is used to figure out the archive format
198 // GNU archive format
199 // First member : / (may exist, if it exists, points to the symbol table )
200 // Second member : // (may exist, if it exists, points to the string table)
201 // Note : The string table is used if the filename exceeds 15 characters
202 // BSD archive format
203 // First member : __.SYMDEF (points to the symbol table)
204 // There is no string table, if the filename exceeds 15 characters or has a
205 // embedded space, the filename has #1/<size>, The size represents the size
206 // of the filename that needs to be read after the archive header
207 // COFF archive format
209 // Second member : / (provides a directory of symbols)
210 // Third member : // (may exist, if it exists, contains the string table)
211 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
212 // even if the string table is empty. However, lib.exe does not in fact
213 // seem to create the third member if there's no member whose filename
214 // exceeds 15 characters. So the third member is optional.
216 if (Name == "__.SYMDEF") {
219 ec = object_error::success;
228 ec = object_error::parse_failed;
231 Name = i->getRawName();
237 ec = object_error::success;
241 if (Name[0] != '/') {
243 ec = object_error::success;
248 ec = object_error::parse_failed;
257 ec = object_error::success;
261 Name = i->getRawName();
266 ec = object_error::success;
269 Archive::child_iterator Archive::begin_children(bool skip_internal) const {
270 const char *Loc = Data->getBufferStart() + strlen(Magic);
272 // Skip internals at the beginning of an archive.
273 if (skip_internal && isInternalMember(*ToHeader(Loc)))
278 Archive::child_iterator Archive::end_children() const {
279 return Child(this, NULL);
282 error_code Archive::Symbol::getName(StringRef &Result) const {
283 Result = StringRef(Parent->SymbolTable->getBuffer().begin() + StringIndex);
284 return object_error::success;
287 error_code Archive::Symbol::getMember(child_iterator &Result) const {
288 const char *Buf = Parent->SymbolTable->getBuffer().begin();
289 const char *Offsets = Buf + 4;
291 if (Parent->kind() == K_GNU) {
292 Offset = *(reinterpret_cast<const support::ubig32_t*>(Offsets)
294 } else if (Parent->kind() == K_BSD) {
295 llvm_unreachable("BSD format is not supported");
297 uint32_t MemberCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
300 Buf += sizeof(support::ulittle32_t)
301 + (MemberCount * sizeof(support::ulittle32_t));
303 uint32_t SymbolCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
305 if (SymbolIndex >= SymbolCount)
306 return object_error::parse_failed;
308 // Skip SymbolCount to get to the indices table.
309 const char *Indices = Buf + sizeof(support::ulittle32_t);
311 // Get the index of the offset in the file member offset table for this
313 uint16_t OffsetIndex =
314 *(reinterpret_cast<const support::ulittle16_t*>(Indices)
316 // Subtract 1 since OffsetIndex is 1 based.
319 if (OffsetIndex >= MemberCount)
320 return object_error::parse_failed;
322 Offset = *(reinterpret_cast<const support::ulittle32_t*>(Offsets)
326 const char *Loc = Parent->getData().begin() + Offset;
327 Result = Child(Parent, Loc);
329 return object_error::success;
332 Archive::Symbol Archive::Symbol::getNext() const {
334 // Go to one past next null.
336 Parent->SymbolTable->getBuffer().find('\0', t.StringIndex) + 1;
341 Archive::symbol_iterator Archive::begin_symbols() const {
342 const char *buf = SymbolTable->getBuffer().begin();
343 if (kind() == K_GNU) {
344 uint32_t symbol_count = 0;
345 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
346 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
347 } else if (kind() == K_BSD) {
348 llvm_unreachable("BSD archive format is not supported");
350 uint32_t member_count = 0;
351 uint32_t symbol_count = 0;
352 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
353 buf += 4 + (member_count * 4); // Skip offsets.
354 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
355 buf += 4 + (symbol_count * 2); // Skip indices.
357 uint32_t string_start_offset = buf - SymbolTable->getBuffer().begin();
358 return symbol_iterator(Symbol(this, 0, string_start_offset));
361 Archive::symbol_iterator Archive::end_symbols() const {
362 const char *buf = SymbolTable->getBuffer().begin();
363 uint32_t symbol_count = 0;
364 if (kind() == K_GNU) {
365 symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
366 } else if (kind() == K_BSD) {
367 llvm_unreachable("BSD archive format is not supported");
369 uint32_t member_count = 0;
370 member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
371 buf += 4 + (member_count * 4); // Skip offsets.
372 symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
374 return symbol_iterator(
375 Symbol(this, symbol_count, 0));
378 Archive::child_iterator Archive::findSym(StringRef name) const {
379 Archive::symbol_iterator bs = begin_symbols();
380 Archive::symbol_iterator es = end_symbols();
381 Archive::child_iterator result;
384 for (; bs != es; ++bs) {
385 if (bs->getName(symname))
386 return end_children();
387 if (symname == name) {
388 if (bs->getMember(result))
389 return end_children();
393 return end_children();