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"
20 #include "llvm/Support/Path.h"
23 using namespace object;
24 using namespace llvm::support::endian;
26 static const char *const Magic = "!<arch>\n";
27 static const char *const ThinMagic = "!<thin>\n";
29 void Archive::anchor() { }
31 StringRef ArchiveMemberHeader::getName() const {
33 if (Name[0] == '/' || Name[0] == '#')
37 llvm::StringRef::size_type end =
38 llvm::StringRef(Name, sizeof(Name)).find(EndCond);
39 if (end == llvm::StringRef::npos)
41 assert(end <= sizeof(Name) && end > 0);
42 // Don't include the EndCond if there is one.
43 return llvm::StringRef(Name, end);
46 uint32_t ArchiveMemberHeader::getSize() const {
48 if (llvm::StringRef(Size, sizeof(Size)).rtrim(" ").getAsInteger(10, Ret))
49 llvm_unreachable("Size is not a decimal number.");
53 sys::fs::perms ArchiveMemberHeader::getAccessMode() const {
55 if (StringRef(AccessMode, sizeof(AccessMode)).rtrim(" ").getAsInteger(8, Ret))
56 llvm_unreachable("Access mode is not an octal number.");
57 return static_cast<sys::fs::perms>(Ret);
60 sys::TimeValue ArchiveMemberHeader::getLastModified() const {
62 if (StringRef(LastModified, sizeof(LastModified)).rtrim(" ")
63 .getAsInteger(10, Seconds))
64 llvm_unreachable("Last modified time not a decimal number.");
67 Ret.fromEpochTime(Seconds);
71 unsigned ArchiveMemberHeader::getUID() const {
73 if (StringRef(UID, sizeof(UID)).rtrim(" ").getAsInteger(10, Ret))
74 llvm_unreachable("UID time not a decimal number.");
78 unsigned ArchiveMemberHeader::getGID() const {
80 if (StringRef(GID, sizeof(GID)).rtrim(" ").getAsInteger(10, Ret))
81 llvm_unreachable("GID time not a decimal number.");
85 Archive::Child::Child(const Archive *Parent, const char *Start)
90 const ArchiveMemberHeader *Header =
91 reinterpret_cast<const ArchiveMemberHeader *>(Start);
92 uint64_t Size = sizeof(ArchiveMemberHeader);
93 if (!Parent->IsThin || Header->getName() == "/" || Header->getName() == "//")
94 Size += Header->getSize();
95 Data = StringRef(Start, Size);
97 // Setup StartOfFile and PaddingBytes.
98 StartOfFile = sizeof(ArchiveMemberHeader);
99 // Don't include attached name.
100 StringRef Name = Header->getName();
101 if (Name.startswith("#1/")) {
103 if (Name.substr(3).rtrim(" ").getAsInteger(10, NameSize))
104 llvm_unreachable("Long name length is not an integer");
105 StartOfFile += NameSize;
109 uint64_t Archive::Child::getSize() const {
111 return getHeader()->getSize();
112 return Data.size() - StartOfFile;
115 uint64_t Archive::Child::getRawSize() const {
116 return getHeader()->getSize();
119 ErrorOr<StringRef> Archive::Child::getBuffer() const {
121 return StringRef(Data.data() + StartOfFile, getSize());
122 ErrorOr<StringRef> Name = getName();
123 if (std::error_code EC = Name.getError())
125 SmallString<128> FullName = sys::path::parent_path(
126 Parent->getMemoryBufferRef().getBufferIdentifier());
127 sys::path::append(FullName, *Name);
128 ErrorOr<std::unique_ptr<MemoryBuffer>> Buf = MemoryBuffer::getFile(FullName);
129 if (std::error_code EC = Buf.getError())
131 Parent->ThinBuffers.push_back(std::move(*Buf));
132 return Parent->ThinBuffers.back()->getBuffer();
135 Archive::Child Archive::Child::getNext() const {
136 size_t SpaceToSkip = Data.size();
137 // If it's odd, add 1 to make it even.
141 const char *NextLoc = Data.data() + SpaceToSkip;
143 // Check to see if this is past the end of the archive.
144 if (NextLoc >= Parent->Data.getBufferEnd())
145 return Child(Parent, nullptr);
147 return Child(Parent, NextLoc);
150 uint64_t Archive::Child::getChildOffset() const {
151 const char *a = Parent->Data.getBuffer().data();
152 const char *c = Data.data();
153 uint64_t offset = c - a;
157 ErrorOr<StringRef> Archive::Child::getName() const {
158 StringRef name = getRawName();
159 // Check if it's a special name.
160 if (name[0] == '/') {
161 if (name.size() == 1) // Linker member.
163 if (name.size() == 2 && name[1] == '/') // String table.
168 if (name.substr(1).rtrim(" ").getAsInteger(10, offset))
169 llvm_unreachable("Long name offset is not an integer");
170 const char *addr = Parent->StringTable->Data.begin()
171 + sizeof(ArchiveMemberHeader)
174 if (Parent->StringTable == Parent->child_end()
175 || addr < (Parent->StringTable->Data.begin()
176 + sizeof(ArchiveMemberHeader))
177 || addr > (Parent->StringTable->Data.begin()
178 + sizeof(ArchiveMemberHeader)
179 + Parent->StringTable->getSize()))
180 return object_error::parse_failed;
182 // GNU long file names end with a "/\n".
183 if (Parent->kind() == K_GNU || Parent->kind() == K_MIPS64) {
184 StringRef::size_type End = StringRef(addr).find('\n');
185 return StringRef(addr, End - 1);
187 return StringRef(addr);
188 } else if (name.startswith("#1/")) {
190 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size))
191 llvm_unreachable("Long name length is not an ingeter");
192 return Data.substr(sizeof(ArchiveMemberHeader), name_size)
193 .rtrim(StringRef("\0", 1));
195 // It's a simple name.
196 if (name[name.size() - 1] == '/')
197 return name.substr(0, name.size() - 1);
201 ErrorOr<MemoryBufferRef> Archive::Child::getMemoryBufferRef() const {
202 ErrorOr<StringRef> NameOrErr = getName();
203 if (std::error_code EC = NameOrErr.getError())
205 StringRef Name = NameOrErr.get();
206 ErrorOr<StringRef> Buf = getBuffer();
207 if (std::error_code EC = Buf.getError())
209 return MemoryBufferRef(*Buf, Name);
212 ErrorOr<std::unique_ptr<Binary>>
213 Archive::Child::getAsBinary(LLVMContext *Context) const {
214 ErrorOr<MemoryBufferRef> BuffOrErr = getMemoryBufferRef();
215 if (std::error_code EC = BuffOrErr.getError())
218 return createBinary(BuffOrErr.get(), Context);
221 ErrorOr<std::unique_ptr<Archive>> Archive::create(MemoryBufferRef Source) {
223 std::unique_ptr<Archive> Ret(new Archive(Source, EC));
226 return std::move(Ret);
229 Archive::Archive(MemoryBufferRef Source, std::error_code &ec)
230 : Binary(Binary::ID_Archive, Source), SymbolTable(child_end()),
231 StringTable(child_end()), FirstRegular(child_end()) {
232 StringRef Buffer = Data.getBuffer();
233 // Check for sufficient magic.
234 if (Buffer.startswith(ThinMagic)) {
236 } else if (Buffer.startswith(Magic)) {
239 ec = object_error::invalid_file_type;
243 // Get the special members.
244 child_iterator i = child_begin(false);
245 child_iterator e = child_end();
248 ec = std::error_code();
252 StringRef Name = i->getRawName();
254 // Below is the pattern that is used to figure out the archive format
255 // GNU archive format
256 // First member : / (may exist, if it exists, points to the symbol table )
257 // Second member : // (may exist, if it exists, points to the string table)
258 // Note : The string table is used if the filename exceeds 15 characters
259 // BSD archive format
260 // First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table)
261 // There is no string table, if the filename exceeds 15 characters or has a
262 // embedded space, the filename has #1/<size>, The size represents the size
263 // of the filename that needs to be read after the archive header
264 // COFF archive format
266 // Second member : / (provides a directory of symbols)
267 // Third member : // (may exist, if it exists, contains the string table)
268 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
269 // even if the string table is empty. However, lib.exe does not in fact
270 // seem to create the third member if there's no member whose filename
271 // exceeds 15 characters. So the third member is optional.
273 if (Name == "__.SYMDEF") {
278 ec = std::error_code();
282 if (Name.startswith("#1/")) {
284 // We know this is BSD, so getName will work since there is no string table.
285 ErrorOr<StringRef> NameOrErr = i->getName();
286 ec = NameOrErr.getError();
289 Name = NameOrErr.get();
290 if (Name == "__.SYMDEF SORTED" || Name == "__.SYMDEF") {
298 // MIPS 64-bit ELF archives use a special format of a symbol table.
299 // This format is marked by `ar_name` field equals to "/SYM64/".
300 // For detailed description see page 96 in the following document:
301 // http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
303 bool has64SymTable = false;
304 if (Name == "/" || Name == "/SYM64/") {
306 if (Name == "/SYM64/")
307 has64SymTable = true;
311 ec = std::error_code();
314 Name = i->getRawName();
318 Format = has64SymTable ? K_MIPS64 : K_GNU;
322 ec = std::error_code();
326 if (Name[0] != '/') {
327 Format = has64SymTable ? K_MIPS64 : K_GNU;
329 ec = std::error_code();
334 ec = object_error::parse_failed;
344 ec = std::error_code();
348 Name = i->getRawName();
356 ec = std::error_code();
359 Archive::child_iterator Archive::child_begin(bool SkipInternal) const {
360 if (Data.getBufferSize() == 8) // empty archive.
366 const char *Loc = Data.getBufferStart() + strlen(Magic);
371 Archive::child_iterator Archive::child_end() const {
372 return Child(this, nullptr);
375 StringRef Archive::Symbol::getName() const {
376 return Parent->getSymbolTable().begin() + StringIndex;
379 ErrorOr<Archive::child_iterator> Archive::Symbol::getMember() const {
380 const char *Buf = Parent->getSymbolTable().begin();
381 const char *Offsets = Buf;
382 if (Parent->kind() == K_MIPS64)
383 Offsets += sizeof(uint64_t);
385 Offsets += sizeof(uint32_t);
387 if (Parent->kind() == K_GNU) {
388 Offset = read32be(Offsets + SymbolIndex * 4);
389 } else if (Parent->kind() == K_MIPS64) {
390 Offset = read64be(Offsets + SymbolIndex * 8);
391 } else if (Parent->kind() == K_BSD) {
392 // The SymbolIndex is an index into the ranlib structs that start at
393 // Offsets (the first uint32_t is the number of bytes of the ranlib
394 // structs). The ranlib structs are a pair of uint32_t's the first
395 // being a string table offset and the second being the offset into
396 // the archive of the member that defines the symbol. Which is what
398 Offset = read32le(Offsets + SymbolIndex * 8 + 4);
401 uint32_t MemberCount = read32le(Buf);
402 Buf += MemberCount * 4 + 4;
404 uint32_t SymbolCount = read32le(Buf);
405 if (SymbolIndex >= SymbolCount)
406 return object_error::parse_failed;
408 // Skip SymbolCount to get to the indices table.
409 const char *Indices = Buf + 4;
411 // Get the index of the offset in the file member offset table for this
413 uint16_t OffsetIndex = read16le(Indices + SymbolIndex * 2);
414 // Subtract 1 since OffsetIndex is 1 based.
417 if (OffsetIndex >= MemberCount)
418 return object_error::parse_failed;
420 Offset = read32le(Offsets + OffsetIndex * 4);
423 const char *Loc = Parent->getData().begin() + Offset;
424 child_iterator Iter(Child(Parent, Loc));
428 Archive::Symbol Archive::Symbol::getNext() const {
430 if (Parent->kind() == K_BSD) {
431 // t.StringIndex is an offset from the start of the __.SYMDEF or
432 // "__.SYMDEF SORTED" member into the string table for the ranlib
433 // struct indexed by t.SymbolIndex . To change t.StringIndex to the
434 // offset in the string table for t.SymbolIndex+1 we subtract the
435 // its offset from the start of the string table for t.SymbolIndex
436 // and add the offset of the string table for t.SymbolIndex+1.
438 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
439 // which is the number of bytes of ranlib structs that follow. The ranlib
440 // structs are a pair of uint32_t's the first being a string table offset
441 // and the second being the offset into the archive of the member that
442 // define the symbol. After that the next uint32_t is the byte count of
443 // the string table followed by the string table.
444 const char *Buf = Parent->getSymbolTable().begin();
445 uint32_t RanlibCount = 0;
446 RanlibCount = read32le(Buf) / 8;
447 // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount)
448 // don't change the t.StringIndex as we don't want to reference a ranlib
450 if (t.SymbolIndex + 1 < RanlibCount) {
451 const char *Ranlibs = Buf + 4;
452 uint32_t CurRanStrx = 0;
453 uint32_t NextRanStrx = 0;
454 CurRanStrx = read32le(Ranlibs + t.SymbolIndex * 8);
455 NextRanStrx = read32le(Ranlibs + (t.SymbolIndex + 1) * 8);
456 t.StringIndex -= CurRanStrx;
457 t.StringIndex += NextRanStrx;
460 // Go to one past next null.
461 t.StringIndex = Parent->getSymbolTable().find('\0', t.StringIndex) + 1;
467 Archive::symbol_iterator Archive::symbol_begin() const {
468 if (!hasSymbolTable())
469 return symbol_iterator(Symbol(this, 0, 0));
471 const char *buf = getSymbolTable().begin();
472 if (kind() == K_GNU) {
473 uint32_t symbol_count = 0;
474 symbol_count = read32be(buf);
475 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
476 } else if (kind() == K_MIPS64) {
477 uint64_t symbol_count = read64be(buf);
478 buf += sizeof(uint64_t) + (symbol_count * (sizeof(uint64_t)));
479 } else if (kind() == K_BSD) {
480 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
481 // which is the number of bytes of ranlib structs that follow. The ranlib
482 // structs are a pair of uint32_t's the first being a string table offset
483 // and the second being the offset into the archive of the member that
484 // define the symbol. After that the next uint32_t is the byte count of
485 // the string table followed by the string table.
486 uint32_t ranlib_count = 0;
487 ranlib_count = read32le(buf) / 8;
488 const char *ranlibs = buf + 4;
489 uint32_t ran_strx = 0;
490 ran_strx = read32le(ranlibs);
491 buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t))));
492 // Skip the byte count of the string table.
493 buf += sizeof(uint32_t);
496 uint32_t member_count = 0;
497 uint32_t symbol_count = 0;
498 member_count = read32le(buf);
499 buf += 4 + (member_count * 4); // Skip offsets.
500 symbol_count = read32le(buf);
501 buf += 4 + (symbol_count * 2); // Skip indices.
503 uint32_t string_start_offset = buf - getSymbolTable().begin();
504 return symbol_iterator(Symbol(this, 0, string_start_offset));
507 Archive::symbol_iterator Archive::symbol_end() const {
508 if (!hasSymbolTable())
509 return symbol_iterator(Symbol(this, 0, 0));
510 return symbol_iterator(Symbol(this, getNumberOfSymbols(), 0));
513 uint32_t Archive::getNumberOfSymbols() const {
514 const char *buf = getSymbolTable().begin();
516 return read32be(buf);
517 if (kind() == K_MIPS64)
518 return read64be(buf);
520 return read32le(buf) / 8;
521 uint32_t member_count = 0;
522 member_count = read32le(buf);
523 buf += 4 + (member_count * 4); // Skip offsets.
524 return read32le(buf);
527 Archive::child_iterator Archive::findSym(StringRef name) const {
528 Archive::symbol_iterator bs = symbol_begin();
529 Archive::symbol_iterator es = symbol_end();
531 for (; bs != es; ++bs) {
532 StringRef SymName = bs->getName();
533 if (SymName == name) {
534 ErrorOr<Archive::child_iterator> ResultOrErr = bs->getMember();
535 // FIXME: Should we really eat the error?
536 if (ResultOrErr.getError())
538 return ResultOrErr.get();
544 bool Archive::hasSymbolTable() const {
545 return SymbolTable != child_end();