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 uint64_t Size = sizeof(ArchiveMemberHeader);
91 Data = StringRef(Start, Size);
92 if (!isThinMember()) {
94 Data = StringRef(Start, Size);
97 // Setup StartOfFile and PaddingBytes.
98 StartOfFile = sizeof(ArchiveMemberHeader);
99 // Don't include attached name.
100 StringRef Name = getRawName();
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 bool Archive::Child::isThinMember() const {
120 StringRef Name = getHeader()->getName();
121 return Parent->IsThin && Name != "/" && Name != "//";
124 ErrorOr<StringRef> Archive::Child::getBuffer() const {
126 return StringRef(Data.data() + StartOfFile, getSize());
127 ErrorOr<StringRef> Name = getName();
128 if (std::error_code EC = Name.getError())
130 SmallString<128> FullName = sys::path::parent_path(
131 Parent->getMemoryBufferRef().getBufferIdentifier());
132 sys::path::append(FullName, *Name);
133 ErrorOr<std::unique_ptr<MemoryBuffer>> Buf = MemoryBuffer::getFile(FullName);
134 if (std::error_code EC = Buf.getError())
136 Parent->ThinBuffers.push_back(std::move(*Buf));
137 return Parent->ThinBuffers.back()->getBuffer();
140 Archive::Child Archive::Child::getNext() const {
141 size_t SpaceToSkip = Data.size();
142 // If it's odd, add 1 to make it even.
146 const char *NextLoc = Data.data() + SpaceToSkip;
148 // Check to see if this is past the end of the archive.
149 if (NextLoc >= Parent->Data.getBufferEnd())
150 return Child(Parent, nullptr);
152 return Child(Parent, NextLoc);
155 uint64_t Archive::Child::getChildOffset() const {
156 const char *a = Parent->Data.getBuffer().data();
157 const char *c = Data.data();
158 uint64_t offset = c - a;
162 ErrorOr<StringRef> Archive::Child::getName() const {
163 StringRef name = getRawName();
164 // Check if it's a special name.
165 if (name[0] == '/') {
166 if (name.size() == 1) // Linker member.
168 if (name.size() == 2 && name[1] == '/') // String table.
173 if (name.substr(1).rtrim(" ").getAsInteger(10, offset))
174 llvm_unreachable("Long name offset is not an integer");
175 const char *addr = Parent->StringTable->Data.begin()
176 + sizeof(ArchiveMemberHeader)
179 if (Parent->StringTable == Parent->child_end()
180 || addr < (Parent->StringTable->Data.begin()
181 + sizeof(ArchiveMemberHeader))
182 || addr > (Parent->StringTable->Data.begin()
183 + sizeof(ArchiveMemberHeader)
184 + Parent->StringTable->getSize()))
185 return object_error::parse_failed;
187 // GNU long file names end with a "/\n".
188 if (Parent->kind() == K_GNU || Parent->kind() == K_MIPS64) {
189 StringRef::size_type End = StringRef(addr).find('\n');
190 return StringRef(addr, End - 1);
192 return StringRef(addr);
193 } else if (name.startswith("#1/")) {
195 if (name.substr(3).rtrim(" ").getAsInteger(10, name_size))
196 llvm_unreachable("Long name length is not an ingeter");
197 return Data.substr(sizeof(ArchiveMemberHeader), name_size)
198 .rtrim(StringRef("\0", 1));
200 // It's a simple name.
201 if (name[name.size() - 1] == '/')
202 return name.substr(0, name.size() - 1);
206 ErrorOr<MemoryBufferRef> Archive::Child::getMemoryBufferRef() const {
207 ErrorOr<StringRef> NameOrErr = getName();
208 if (std::error_code EC = NameOrErr.getError())
210 StringRef Name = NameOrErr.get();
211 ErrorOr<StringRef> Buf = getBuffer();
212 if (std::error_code EC = Buf.getError())
214 return MemoryBufferRef(*Buf, Name);
217 ErrorOr<std::unique_ptr<Binary>>
218 Archive::Child::getAsBinary(LLVMContext *Context) const {
219 ErrorOr<MemoryBufferRef> BuffOrErr = getMemoryBufferRef();
220 if (std::error_code EC = BuffOrErr.getError())
223 return createBinary(BuffOrErr.get(), Context);
226 ErrorOr<std::unique_ptr<Archive>> Archive::create(MemoryBufferRef Source) {
228 std::unique_ptr<Archive> Ret(new Archive(Source, EC));
231 return std::move(Ret);
234 Archive::Archive(MemoryBufferRef Source, std::error_code &ec)
235 : Binary(Binary::ID_Archive, Source), SymbolTable(child_end()),
236 StringTable(child_end()), FirstRegular(child_end()) {
237 StringRef Buffer = Data.getBuffer();
238 // Check for sufficient magic.
239 if (Buffer.startswith(ThinMagic)) {
241 } else if (Buffer.startswith(Magic)) {
244 ec = object_error::invalid_file_type;
248 // Get the special members.
249 child_iterator i = child_begin(false);
250 child_iterator e = child_end();
253 ec = std::error_code();
257 StringRef Name = i->getRawName();
259 // Below is the pattern that is used to figure out the archive format
260 // GNU archive format
261 // First member : / (may exist, if it exists, points to the symbol table )
262 // Second member : // (may exist, if it exists, points to the string table)
263 // Note : The string table is used if the filename exceeds 15 characters
264 // BSD archive format
265 // First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table)
266 // There is no string table, if the filename exceeds 15 characters or has a
267 // embedded space, the filename has #1/<size>, The size represents the size
268 // of the filename that needs to be read after the archive header
269 // COFF archive format
271 // Second member : / (provides a directory of symbols)
272 // Third member : // (may exist, if it exists, contains the string table)
273 // Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
274 // even if the string table is empty. However, lib.exe does not in fact
275 // seem to create the third member if there's no member whose filename
276 // exceeds 15 characters. So the third member is optional.
278 if (Name == "__.SYMDEF") {
283 ec = std::error_code();
287 if (Name.startswith("#1/")) {
289 // We know this is BSD, so getName will work since there is no string table.
290 ErrorOr<StringRef> NameOrErr = i->getName();
291 ec = NameOrErr.getError();
294 Name = NameOrErr.get();
295 if (Name == "__.SYMDEF SORTED" || Name == "__.SYMDEF") {
303 // MIPS 64-bit ELF archives use a special format of a symbol table.
304 // This format is marked by `ar_name` field equals to "/SYM64/".
305 // For detailed description see page 96 in the following document:
306 // http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
308 bool has64SymTable = false;
309 if (Name == "/" || Name == "/SYM64/") {
311 if (Name == "/SYM64/")
312 has64SymTable = true;
316 ec = std::error_code();
319 Name = i->getRawName();
323 Format = has64SymTable ? K_MIPS64 : K_GNU;
327 ec = std::error_code();
331 if (Name[0] != '/') {
332 Format = has64SymTable ? K_MIPS64 : K_GNU;
334 ec = std::error_code();
339 ec = object_error::parse_failed;
349 ec = std::error_code();
353 Name = i->getRawName();
361 ec = std::error_code();
364 Archive::child_iterator Archive::child_begin(bool SkipInternal) const {
365 if (Data.getBufferSize() == 8) // empty archive.
371 const char *Loc = Data.getBufferStart() + strlen(Magic);
376 Archive::child_iterator Archive::child_end() const {
377 return Child(this, nullptr);
380 StringRef Archive::Symbol::getName() const {
381 return Parent->getSymbolTable().begin() + StringIndex;
384 ErrorOr<Archive::child_iterator> Archive::Symbol::getMember() const {
385 const char *Buf = Parent->getSymbolTable().begin();
386 const char *Offsets = Buf;
387 if (Parent->kind() == K_MIPS64)
388 Offsets += sizeof(uint64_t);
390 Offsets += sizeof(uint32_t);
392 if (Parent->kind() == K_GNU) {
393 Offset = read32be(Offsets + SymbolIndex * 4);
394 } else if (Parent->kind() == K_MIPS64) {
395 Offset = read64be(Offsets + SymbolIndex * 8);
396 } else if (Parent->kind() == K_BSD) {
397 // The SymbolIndex is an index into the ranlib structs that start at
398 // Offsets (the first uint32_t is the number of bytes of the ranlib
399 // structs). The ranlib structs are a pair of uint32_t's the first
400 // being a string table offset and the second being the offset into
401 // the archive of the member that defines the symbol. Which is what
403 Offset = read32le(Offsets + SymbolIndex * 8 + 4);
406 uint32_t MemberCount = read32le(Buf);
407 Buf += MemberCount * 4 + 4;
409 uint32_t SymbolCount = read32le(Buf);
410 if (SymbolIndex >= SymbolCount)
411 return object_error::parse_failed;
413 // Skip SymbolCount to get to the indices table.
414 const char *Indices = Buf + 4;
416 // Get the index of the offset in the file member offset table for this
418 uint16_t OffsetIndex = read16le(Indices + SymbolIndex * 2);
419 // Subtract 1 since OffsetIndex is 1 based.
422 if (OffsetIndex >= MemberCount)
423 return object_error::parse_failed;
425 Offset = read32le(Offsets + OffsetIndex * 4);
428 const char *Loc = Parent->getData().begin() + Offset;
429 child_iterator Iter(Child(Parent, Loc));
433 Archive::Symbol Archive::Symbol::getNext() const {
435 if (Parent->kind() == K_BSD) {
436 // t.StringIndex is an offset from the start of the __.SYMDEF or
437 // "__.SYMDEF SORTED" member into the string table for the ranlib
438 // struct indexed by t.SymbolIndex . To change t.StringIndex to the
439 // offset in the string table for t.SymbolIndex+1 we subtract the
440 // its offset from the start of the string table for t.SymbolIndex
441 // and add the offset of the string table for t.SymbolIndex+1.
443 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
444 // which is the number of bytes of ranlib structs that follow. The ranlib
445 // structs are a pair of uint32_t's the first being a string table offset
446 // and the second being the offset into the archive of the member that
447 // define the symbol. After that the next uint32_t is the byte count of
448 // the string table followed by the string table.
449 const char *Buf = Parent->getSymbolTable().begin();
450 uint32_t RanlibCount = 0;
451 RanlibCount = read32le(Buf) / 8;
452 // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount)
453 // don't change the t.StringIndex as we don't want to reference a ranlib
455 if (t.SymbolIndex + 1 < RanlibCount) {
456 const char *Ranlibs = Buf + 4;
457 uint32_t CurRanStrx = 0;
458 uint32_t NextRanStrx = 0;
459 CurRanStrx = read32le(Ranlibs + t.SymbolIndex * 8);
460 NextRanStrx = read32le(Ranlibs + (t.SymbolIndex + 1) * 8);
461 t.StringIndex -= CurRanStrx;
462 t.StringIndex += NextRanStrx;
465 // Go to one past next null.
466 t.StringIndex = Parent->getSymbolTable().find('\0', t.StringIndex) + 1;
472 Archive::symbol_iterator Archive::symbol_begin() const {
473 if (!hasSymbolTable())
474 return symbol_iterator(Symbol(this, 0, 0));
476 const char *buf = getSymbolTable().begin();
477 if (kind() == K_GNU) {
478 uint32_t symbol_count = 0;
479 symbol_count = read32be(buf);
480 buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
481 } else if (kind() == K_MIPS64) {
482 uint64_t symbol_count = read64be(buf);
483 buf += sizeof(uint64_t) + (symbol_count * (sizeof(uint64_t)));
484 } else if (kind() == K_BSD) {
485 // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
486 // which is the number of bytes of ranlib structs that follow. The ranlib
487 // structs are a pair of uint32_t's the first being a string table offset
488 // and the second being the offset into the archive of the member that
489 // define the symbol. After that the next uint32_t is the byte count of
490 // the string table followed by the string table.
491 uint32_t ranlib_count = 0;
492 ranlib_count = read32le(buf) / 8;
493 const char *ranlibs = buf + 4;
494 uint32_t ran_strx = 0;
495 ran_strx = read32le(ranlibs);
496 buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t))));
497 // Skip the byte count of the string table.
498 buf += sizeof(uint32_t);
501 uint32_t member_count = 0;
502 uint32_t symbol_count = 0;
503 member_count = read32le(buf);
504 buf += 4 + (member_count * 4); // Skip offsets.
505 symbol_count = read32le(buf);
506 buf += 4 + (symbol_count * 2); // Skip indices.
508 uint32_t string_start_offset = buf - getSymbolTable().begin();
509 return symbol_iterator(Symbol(this, 0, string_start_offset));
512 Archive::symbol_iterator Archive::symbol_end() const {
513 if (!hasSymbolTable())
514 return symbol_iterator(Symbol(this, 0, 0));
515 return symbol_iterator(Symbol(this, getNumberOfSymbols(), 0));
518 uint32_t Archive::getNumberOfSymbols() const {
519 const char *buf = getSymbolTable().begin();
521 return read32be(buf);
522 if (kind() == K_MIPS64)
523 return read64be(buf);
525 return read32le(buf) / 8;
526 uint32_t member_count = 0;
527 member_count = read32le(buf);
528 buf += 4 + (member_count * 4); // Skip offsets.
529 return read32le(buf);
532 Archive::child_iterator Archive::findSym(StringRef name) const {
533 Archive::symbol_iterator bs = symbol_begin();
534 Archive::symbol_iterator es = symbol_end();
536 for (; bs != es; ++bs) {
537 StringRef SymName = bs->getName();
538 if (SymName == name) {
539 ErrorOr<Archive::child_iterator> ResultOrErr = bs->getMember();
540 // FIXME: Should we really eat the error?
541 if (ResultOrErr.getError())
543 return ResultOrErr.get();
549 bool Archive::hasSymbolTable() const {
550 return SymbolTable != child_end();