1//===- Archive.cpp - ar File Format implementation --------------*- C++ -*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file defines the ArchiveObjectFile class.
11//
12//===----------------------------------------------------------------------===//
13
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
21using namespace llvm;
22using namespace object;
23
24static const char *const Magic = "!<arch>\n";
25
26void Archive::anchor() { }
27
28StringRef ArchiveMemberHeader::getName() const {
29  char EndCond;
30  if (Name[0] == '/' || Name[0] == '#')
31    EndCond = ' ';
32  else
33    EndCond = '/';
34  llvm::StringRef::size_type end =
35      llvm::StringRef(Name, sizeof(Name)).find(EndCond);
36  if (end == llvm::StringRef::npos)
37    end = sizeof(Name);
38  assert(end <= sizeof(Name) && end > 0);
39  // Don't include the EndCond if there is one.
40  return llvm::StringRef(Name, end);
41}
42
43uint32_t ArchiveMemberHeader::getSize() const {
44  uint32_t Ret;
45  if (llvm::StringRef(Size, sizeof(Size)).rtrim(" ").getAsInteger(10, Ret))
46    llvm_unreachable("Size is not a decimal number.");
47  return Ret;
48}
49
50sys::fs::perms ArchiveMemberHeader::getAccessMode() const {
51  unsigned Ret;
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);
55}
56
57sys::TimeValue ArchiveMemberHeader::getLastModified() const {
58  unsigned Seconds;
59  if (StringRef(LastModified, sizeof(LastModified)).rtrim(" ")
60          .getAsInteger(10, Seconds))
61    llvm_unreachable("Last modified time not a decimal number.");
62
63  sys::TimeValue Ret;
64  Ret.fromEpochTime(Seconds);
65  return Ret;
66}
67
68unsigned ArchiveMemberHeader::getUID() const {
69  unsigned Ret;
70  if (StringRef(UID, sizeof(UID)).rtrim(" ").getAsInteger(10, Ret))
71    llvm_unreachable("UID time not a decimal number.");
72  return Ret;
73}
74
75unsigned ArchiveMemberHeader::getGID() const {
76  unsigned Ret;
77  if (StringRef(GID, sizeof(GID)).rtrim(" ").getAsInteger(10, Ret))
78    llvm_unreachable("GID time not a decimal number.");
79  return Ret;
80}
81
82Archive::Child::Child(const Archive *Parent, const char *Start)
83    : Parent(Parent) {
84  if (!Start)
85    return;
86
87  const ArchiveMemberHeader *Header =
88      reinterpret_cast<const ArchiveMemberHeader *>(Start);
89  Data = StringRef(Start, sizeof(ArchiveMemberHeader) + Header->getSize());
90
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/")) {
96    uint64_t NameSize;
97    if (Name.substr(3).rtrim(" ").getAsInteger(10, NameSize))
98      llvm_unreachable("Long name length is not an integer");
99    StartOfFile += NameSize;
100  }
101}
102
103Archive::Child Archive::Child::getNext() const {
104  size_t SpaceToSkip = Data.size();
105  // If it's odd, add 1 to make it even.
106  if (SpaceToSkip & 1)
107    ++SpaceToSkip;
108
109  const char *NextLoc = Data.data() + SpaceToSkip;
110
111  // Check to see if this is past the end of the archive.
112  if (NextLoc >= Parent->Data->getBufferEnd())
113    return Child(Parent, NULL);
114
115  return Child(Parent, NextLoc);
116}
117
118error_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.
123      Result = name;
124      return object_error::success;
125    }
126    if (name.size() == 2 && name[1] == '/') { // String table.
127      Result = name;
128      return object_error::success;
129    }
130    // It's a long name.
131    // Get the offset.
132    std::size_t offset;
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)
137                       + offset;
138    // Verify it.
139    if (Parent->StringTable == Parent->end_children()
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;
146
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);
151    } else {
152      Result = addr;
153    }
154    return object_error::success;
155  } else if (name.startswith("#1/")) {
156    uint64_t name_size;
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;
162  }
163  // It's a simple name.
164  if (name[name.size() - 1] == '/')
165    Result = name.substr(0, name.size() - 1);
166  else
167    Result = name;
168  return object_error::success;
169}
170
171error_code Archive::Child::getMemoryBuffer(OwningPtr<MemoryBuffer> &Result,
172                                           bool FullPath) const {
173  StringRef Name;
174  if (error_code ec = getName(Name))
175    return ec;
176  SmallString<128> Path;
177  Result.reset(MemoryBuffer::getMemBuffer(
178      getBuffer(), FullPath ? (Twine(Parent->getFileName()) + "(" + Name + ")")
179                                  .toStringRef(Path)
180                            : Name,
181      false));
182  return error_code::success();
183}
184
185error_code Archive::Child::getAsBinary(OwningPtr<Binary> &Result) const {
186  OwningPtr<Binary> ret;
187  OwningPtr<MemoryBuffer> Buff;
188  if (error_code ec = getMemoryBuffer(Buff))
189    return ec;
190  if (error_code ec = createBinary(Buff.take(), ret))
191    return ec;
192  Result.swap(ret);
193  return object_error::success;
194}
195
196Archive::Archive(MemoryBuffer *source, error_code &ec)
197  : Binary(Binary::ID_Archive, source), SymbolTable(end_children()) {
198  // Check for sufficient magic.
199  assert(source);
200  if (source->getBufferSize() < 8 ||
201      StringRef(source->getBufferStart(), 8) != Magic) {
202    ec = object_error::invalid_file_type;
203    return;
204  }
205
206  // Get the special members.
207  child_iterator i = begin_children(false);
208  child_iterator e = end_children();
209
210  if (i == e) {
211    ec = object_error::success;
212    return;
213  }
214
215  StringRef Name = i->getRawName();
216
217  // Below is the pattern that is used to figure out the archive format
218  // GNU archive format
219  //  First member : / (may exist, if it exists, points to the symbol table )
220  //  Second member : // (may exist, if it exists, points to the string table)
221  //  Note : The string table is used if the filename exceeds 15 characters
222  // BSD archive format
223  //  First member : __.SYMDEF or "__.SYMDEF SORTED" (the symbol table)
224  //  There is no string table, if the filename exceeds 15 characters or has a
225  //  embedded space, the filename has #1/<size>, The size represents the size
226  //  of the filename that needs to be read after the archive header
227  // COFF archive format
228  //  First member : /
229  //  Second member : / (provides a directory of symbols)
230  //  Third member : // (may exist, if it exists, contains the string table)
231  //  Note: Microsoft PE/COFF Spec 8.3 says that the third member is present
232  //  even if the string table is empty. However, lib.exe does not in fact
233  //  seem to create the third member if there's no member whose filename
234  //  exceeds 15 characters. So the third member is optional.
235
236  if (Name == "__.SYMDEF") {
237    Format = K_BSD;
238    SymbolTable = i;
239    ++i;
240    FirstRegular = i;
241    ec = object_error::success;
242    return;
243  }
244
245  if (Name.startswith("#1/")) {
246    Format = K_BSD;
247    // We know this is BSD, so getName will work since there is no string table.
248    ec = i->getName(Name);
249    if (ec)
250      return;
251    if (Name == "__.SYMDEF SORTED") {
252      SymbolTable = i;
253      ++i;
254    }
255    FirstRegular = i;
256    return;
257  }
258
259  if (Name == "/") {
260    SymbolTable = i;
261
262    ++i;
263    if (i == e) {
264      ec = object_error::parse_failed;
265      return;
266    }
267    Name = i->getRawName();
268  }
269
270  if (Name == "//") {
271    Format = K_GNU;
272    StringTable = i;
273    ++i;
274    FirstRegular = i;
275    ec = object_error::success;
276    return;
277  }
278
279  if (Name[0] != '/') {
280    Format = K_GNU;
281    FirstRegular = i;
282    ec = object_error::success;
283    return;
284  }
285
286  if (Name != "/") {
287    ec = object_error::parse_failed;
288    return;
289  }
290
291  Format = K_COFF;
292  SymbolTable = i;
293
294  ++i;
295  if (i == e) {
296    FirstRegular = i;
297    ec = object_error::success;
298    return;
299  }
300
301  Name = i->getRawName();
302
303  if (Name == "//") {
304    StringTable = i;
305    ++i;
306  }
307
308  FirstRegular = i;
309  ec = object_error::success;
310}
311
312Archive::child_iterator Archive::begin_children(bool SkipInternal) const {
313  if (Data->getBufferSize() == 8) // empty archive.
314    return end_children();
315
316  if (SkipInternal)
317    return FirstRegular;
318
319  const char *Loc = Data->getBufferStart() + strlen(Magic);
320  Child c(this, Loc);
321  return c;
322}
323
324Archive::child_iterator Archive::end_children() const {
325  return Child(this, NULL);
326}
327
328error_code Archive::Symbol::getName(StringRef &Result) const {
329  Result = StringRef(Parent->SymbolTable->getBuffer().begin() + StringIndex);
330  return object_error::success;
331}
332
333error_code Archive::Symbol::getMember(child_iterator &Result) const {
334  const char *Buf = Parent->SymbolTable->getBuffer().begin();
335  const char *Offsets = Buf + 4;
336  uint32_t Offset = 0;
337  if (Parent->kind() == K_GNU) {
338    Offset = *(reinterpret_cast<const support::ubig32_t*>(Offsets)
339               + SymbolIndex);
340  } else if (Parent->kind() == K_BSD) {
341    llvm_unreachable("BSD format is not supported");
342  } else {
343    uint32_t MemberCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
344
345    // Skip offsets.
346    Buf += sizeof(support::ulittle32_t)
347           + (MemberCount * sizeof(support::ulittle32_t));
348
349    uint32_t SymbolCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
350
351    if (SymbolIndex >= SymbolCount)
352      return object_error::parse_failed;
353
354    // Skip SymbolCount to get to the indices table.
355    const char *Indices = Buf + sizeof(support::ulittle32_t);
356
357    // Get the index of the offset in the file member offset table for this
358    // symbol.
359    uint16_t OffsetIndex =
360      *(reinterpret_cast<const support::ulittle16_t*>(Indices)
361        + SymbolIndex);
362    // Subtract 1 since OffsetIndex is 1 based.
363    --OffsetIndex;
364
365    if (OffsetIndex >= MemberCount)
366      return object_error::parse_failed;
367
368    Offset = *(reinterpret_cast<const support::ulittle32_t*>(Offsets)
369               + OffsetIndex);
370  }
371
372  const char *Loc = Parent->getData().begin() + Offset;
373  Result = Child(Parent, Loc);
374
375  return object_error::success;
376}
377
378Archive::Symbol Archive::Symbol::getNext() const {
379  Symbol t(*this);
380  // Go to one past next null.
381  t.StringIndex =
382      Parent->SymbolTable->getBuffer().find('\0', t.StringIndex) + 1;
383  ++t.SymbolIndex;
384  return t;
385}
386
387Archive::symbol_iterator Archive::begin_symbols() const {
388  if (!hasSymbolTable())
389    return symbol_iterator(Symbol(this, 0, 0));
390
391  const char *buf = SymbolTable->getBuffer().begin();
392  if (kind() == K_GNU) {
393    uint32_t symbol_count = 0;
394    symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
395    buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
396  } else if (kind() == K_BSD) {
397    llvm_unreachable("BSD archive format is not supported");
398  } else {
399    uint32_t member_count = 0;
400    uint32_t symbol_count = 0;
401    member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
402    buf += 4 + (member_count * 4); // Skip offsets.
403    symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
404    buf += 4 + (symbol_count * 2); // Skip indices.
405  }
406  uint32_t string_start_offset = buf - SymbolTable->getBuffer().begin();
407  return symbol_iterator(Symbol(this, 0, string_start_offset));
408}
409
410Archive::symbol_iterator Archive::end_symbols() const {
411  if (!hasSymbolTable())
412    return symbol_iterator(Symbol(this, 0, 0));
413
414  const char *buf = SymbolTable->getBuffer().begin();
415  uint32_t symbol_count = 0;
416  if (kind() == K_GNU) {
417    symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
418  } else if (kind() == K_BSD) {
419    llvm_unreachable("BSD archive format is not supported");
420  } else {
421    uint32_t member_count = 0;
422    member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
423    buf += 4 + (member_count * 4); // Skip offsets.
424    symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
425  }
426  return symbol_iterator(
427    Symbol(this, symbol_count, 0));
428}
429
430Archive::child_iterator Archive::findSym(StringRef name) const {
431  Archive::symbol_iterator bs = begin_symbols();
432  Archive::symbol_iterator es = end_symbols();
433  Archive::child_iterator result;
434
435  StringRef symname;
436  for (; bs != es; ++bs) {
437    if (bs->getName(symname))
438        return end_children();
439    if (symname == name) {
440      if (bs->getMember(result))
441        return end_children();
442      return result;
443    }
444  }
445  return end_children();
446}
447
448bool Archive::hasSymbolTable() const {
449  return SymbolTable != end_children();
450}
451