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, nullptr);
114
115  return Child(Parent, NextLoc);
116}
117
118ErrorOr<StringRef> Archive::Child::getName() 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      return name;
124    if (name.size() == 2 && name[1] == '/') // String table.
125      return name;
126    // It's a long name.
127    // Get the offset.
128    std::size_t offset;
129    if (name.substr(1).rtrim(" ").getAsInteger(10, offset))
130      llvm_unreachable("Long name offset is not an integer");
131    const char *addr = Parent->StringTable->Data.begin()
132                       + sizeof(ArchiveMemberHeader)
133                       + offset;
134    // Verify it.
135    if (Parent->StringTable == Parent->child_end()
136        || addr < (Parent->StringTable->Data.begin()
137                   + sizeof(ArchiveMemberHeader))
138        || addr > (Parent->StringTable->Data.begin()
139                   + sizeof(ArchiveMemberHeader)
140                   + Parent->StringTable->getSize()))
141      return object_error::parse_failed;
142
143    // GNU long file names end with a /.
144    if (Parent->kind() == K_GNU) {
145      StringRef::size_type End = StringRef(addr).find('/');
146      return StringRef(addr, End);
147    }
148    return StringRef(addr);
149  } else if (name.startswith("#1/")) {
150    uint64_t name_size;
151    if (name.substr(3).rtrim(" ").getAsInteger(10, name_size))
152      llvm_unreachable("Long name length is not an ingeter");
153    return Data.substr(sizeof(ArchiveMemberHeader), name_size)
154        .rtrim(StringRef("\0", 1));
155  }
156  // It's a simple name.
157  if (name[name.size() - 1] == '/')
158    return name.substr(0, name.size() - 1);
159  return name;
160}
161
162ErrorOr<std::unique_ptr<MemoryBuffer>>
163Archive::Child::getMemoryBuffer(bool FullPath) const {
164  ErrorOr<StringRef> NameOrErr = getName();
165  if (std::error_code EC = NameOrErr.getError())
166    return EC;
167  StringRef Name = NameOrErr.get();
168  SmallString<128> Path;
169  std::unique_ptr<MemoryBuffer> Ret(MemoryBuffer::getMemBuffer(
170      getBuffer(),
171      FullPath
172          ? (Twine(Parent->getFileName()) + "(" + Name + ")").toStringRef(Path)
173          : Name,
174      false));
175  return std::move(Ret);
176}
177
178ErrorOr<std::unique_ptr<Binary>>
179Archive::Child::getAsBinary(LLVMContext *Context) const {
180  std::unique_ptr<Binary> ret;
181  ErrorOr<std::unique_ptr<MemoryBuffer>> BuffOrErr = getMemoryBuffer();
182  if (std::error_code EC = BuffOrErr.getError())
183    return EC;
184
185  std::unique_ptr<MemoryBuffer> Buff(BuffOrErr.get().release());
186  return createBinary(Buff, Context);
187}
188
189ErrorOr<Archive *> Archive::create(std::unique_ptr<MemoryBuffer> Source) {
190  std::error_code EC;
191  std::unique_ptr<Archive> Ret(new Archive(std::move(Source), EC));
192  if (EC)
193    return EC;
194  return Ret.release();
195}
196
197Archive::Archive(std::unique_ptr<MemoryBuffer> Source, std::error_code &ec)
198    : Binary(Binary::ID_Archive, std::move(Source)), SymbolTable(child_end()) {
199  // Check for sufficient magic.
200  if (Data->getBufferSize() < 8 ||
201      StringRef(Data->getBufferStart(), 8) != Magic) {
202    ec = object_error::invalid_file_type;
203    return;
204  }
205
206  // Get the special members.
207  child_iterator i = child_begin(false);
208  child_iterator e = child_end();
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    ErrorOr<StringRef> NameOrErr = i->getName();
249    ec = NameOrErr.getError();
250    if (ec)
251      return;
252    Name = NameOrErr.get();
253    if (Name == "__.SYMDEF SORTED") {
254      SymbolTable = i;
255      ++i;
256    }
257    FirstRegular = i;
258    return;
259  }
260
261  if (Name == "/") {
262    SymbolTable = i;
263
264    ++i;
265    if (i == e) {
266      ec = object_error::parse_failed;
267      return;
268    }
269    Name = i->getRawName();
270  }
271
272  if (Name == "//") {
273    Format = K_GNU;
274    StringTable = i;
275    ++i;
276    FirstRegular = i;
277    ec = object_error::success;
278    return;
279  }
280
281  if (Name[0] != '/') {
282    Format = K_GNU;
283    FirstRegular = i;
284    ec = object_error::success;
285    return;
286  }
287
288  if (Name != "/") {
289    ec = object_error::parse_failed;
290    return;
291  }
292
293  Format = K_COFF;
294  SymbolTable = i;
295
296  ++i;
297  if (i == e) {
298    FirstRegular = i;
299    ec = object_error::success;
300    return;
301  }
302
303  Name = i->getRawName();
304
305  if (Name == "//") {
306    StringTable = i;
307    ++i;
308  }
309
310  FirstRegular = i;
311  ec = object_error::success;
312}
313
314Archive::child_iterator Archive::child_begin(bool SkipInternal) const {
315  if (Data->getBufferSize() == 8) // empty archive.
316    return child_end();
317
318  if (SkipInternal)
319    return FirstRegular;
320
321  const char *Loc = Data->getBufferStart() + strlen(Magic);
322  Child c(this, Loc);
323  return c;
324}
325
326Archive::child_iterator Archive::child_end() const {
327  return Child(this, nullptr);
328}
329
330StringRef Archive::Symbol::getName() const {
331  return Parent->SymbolTable->getBuffer().begin() + StringIndex;
332}
333
334ErrorOr<Archive::child_iterator> Archive::Symbol::getMember() const {
335  const char *Buf = Parent->SymbolTable->getBuffer().begin();
336  const char *Offsets = Buf + 4;
337  uint32_t Offset = 0;
338  if (Parent->kind() == K_GNU) {
339    Offset = *(reinterpret_cast<const support::ubig32_t*>(Offsets)
340               + SymbolIndex);
341  } else if (Parent->kind() == K_BSD) {
342    // The SymbolIndex is an index into the ranlib structs that start at
343    // Offsets (the first uint32_t is the number of bytes of the ranlib
344    // structs).  The ranlib structs are a pair of uint32_t's the first
345    // being a string table offset and the second being the offset into
346    // the archive of the member that defines the symbol.  Which is what
347    // is needed here.
348    Offset = *(reinterpret_cast<const support::ulittle32_t *>(Offsets) +
349               (SymbolIndex * 2) + 1);
350  } else {
351    uint32_t MemberCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
352
353    // Skip offsets.
354    Buf += sizeof(support::ulittle32_t)
355           + (MemberCount * sizeof(support::ulittle32_t));
356
357    uint32_t SymbolCount = *reinterpret_cast<const support::ulittle32_t*>(Buf);
358
359    if (SymbolIndex >= SymbolCount)
360      return object_error::parse_failed;
361
362    // Skip SymbolCount to get to the indices table.
363    const char *Indices = Buf + sizeof(support::ulittle32_t);
364
365    // Get the index of the offset in the file member offset table for this
366    // symbol.
367    uint16_t OffsetIndex =
368      *(reinterpret_cast<const support::ulittle16_t*>(Indices)
369        + SymbolIndex);
370    // Subtract 1 since OffsetIndex is 1 based.
371    --OffsetIndex;
372
373    if (OffsetIndex >= MemberCount)
374      return object_error::parse_failed;
375
376    Offset = *(reinterpret_cast<const support::ulittle32_t*>(Offsets)
377               + OffsetIndex);
378  }
379
380  const char *Loc = Parent->getData().begin() + Offset;
381  child_iterator Iter(Child(Parent, Loc));
382  return Iter;
383}
384
385Archive::Symbol Archive::Symbol::getNext() const {
386  Symbol t(*this);
387  if (Parent->kind() == K_BSD) {
388    // t.StringIndex is an offset from the start of the __.SYMDEF or
389    // "__.SYMDEF SORTED" member into the string table for the ranlib
390    // struct indexed by t.SymbolIndex .  To change t.StringIndex to the
391    // offset in the string table for t.SymbolIndex+1 we subtract the
392    // its offset from the start of the string table for t.SymbolIndex
393    // and add the offset of the string table for t.SymbolIndex+1.
394
395    // The __.SYMDEF or "__.SYMDEF SORTED" member starts with a uint32_t
396    // which is the number of bytes of ranlib structs that follow.  The ranlib
397    // structs are a pair of uint32_t's the first being a string table offset
398    // and the second being the offset into the archive of the member that
399    // define the symbol. After that the next uint32_t is the byte count of
400    // the string table followed by the string table.
401    const char *Buf = Parent->SymbolTable->getBuffer().begin();
402    uint32_t RanlibCount = 0;
403    RanlibCount = (*reinterpret_cast<const support::ulittle32_t *>(Buf)) /
404                  (sizeof(uint32_t) * 2);
405    // If t.SymbolIndex + 1 will be past the count of symbols (the RanlibCount)
406    // don't change the t.StringIndex as we don't want to reference a ranlib
407    // past RanlibCount.
408    if (t.SymbolIndex + 1 < RanlibCount) {
409      const char *Ranlibs = Buf + 4;
410      uint32_t CurRanStrx = 0;
411      uint32_t NextRanStrx = 0;
412      CurRanStrx = *(reinterpret_cast<const support::ulittle32_t *>(Ranlibs) +
413                     (t.SymbolIndex * 2));
414      NextRanStrx = *(reinterpret_cast<const support::ulittle32_t *>(Ranlibs) +
415                      ((t.SymbolIndex + 1) * 2));
416      t.StringIndex -= CurRanStrx;
417      t.StringIndex += NextRanStrx;
418    }
419  } else {
420    // Go to one past next null.
421    t.StringIndex =
422        Parent->SymbolTable->getBuffer().find('\0', t.StringIndex) + 1;
423  }
424  ++t.SymbolIndex;
425  return t;
426}
427
428Archive::symbol_iterator Archive::symbol_begin() const {
429  if (!hasSymbolTable())
430    return symbol_iterator(Symbol(this, 0, 0));
431
432  const char *buf = SymbolTable->getBuffer().begin();
433  if (kind() == K_GNU) {
434    uint32_t symbol_count = 0;
435    symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
436    buf += sizeof(uint32_t) + (symbol_count * (sizeof(uint32_t)));
437  } else if (kind() == K_BSD) {
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    uint32_t ranlib_count = 0;
445    ranlib_count = (*reinterpret_cast<const support::ulittle32_t *>(buf)) /
446                   (sizeof(uint32_t) * 2);
447    const char *ranlibs = buf + 4;
448    uint32_t ran_strx = 0;
449    ran_strx = *(reinterpret_cast<const support::ulittle32_t *>(ranlibs));
450    buf += sizeof(uint32_t) + (ranlib_count * (2 * (sizeof(uint32_t))));
451    // Skip the byte count of the string table.
452    buf += sizeof(uint32_t);
453    buf += ran_strx;
454  } else {
455    uint32_t member_count = 0;
456    uint32_t symbol_count = 0;
457    member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
458    buf += 4 + (member_count * 4); // Skip offsets.
459    symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
460    buf += 4 + (symbol_count * 2); // Skip indices.
461  }
462  uint32_t string_start_offset = buf - SymbolTable->getBuffer().begin();
463  return symbol_iterator(Symbol(this, 0, string_start_offset));
464}
465
466Archive::symbol_iterator Archive::symbol_end() const {
467  if (!hasSymbolTable())
468    return symbol_iterator(Symbol(this, 0, 0));
469
470  const char *buf = SymbolTable->getBuffer().begin();
471  uint32_t symbol_count = 0;
472  if (kind() == K_GNU) {
473    symbol_count = *reinterpret_cast<const support::ubig32_t*>(buf);
474  } else if (kind() == K_BSD) {
475    symbol_count = (*reinterpret_cast<const support::ulittle32_t *>(buf)) /
476                   (sizeof(uint32_t) * 2);
477  } else {
478    uint32_t member_count = 0;
479    member_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
480    buf += 4 + (member_count * 4); // Skip offsets.
481    symbol_count = *reinterpret_cast<const support::ulittle32_t*>(buf);
482  }
483  return symbol_iterator(
484    Symbol(this, symbol_count, 0));
485}
486
487Archive::child_iterator Archive::findSym(StringRef name) const {
488  Archive::symbol_iterator bs = symbol_begin();
489  Archive::symbol_iterator es = symbol_end();
490
491  for (; bs != es; ++bs) {
492    StringRef SymName = bs->getName();
493    if (SymName == name) {
494      ErrorOr<Archive::child_iterator> ResultOrErr = bs->getMember();
495      // FIXME: Should we really eat the error?
496      if (ResultOrErr.getError())
497        return child_end();
498      return ResultOrErr.get();
499    }
500  }
501  return child_end();
502}
503
504bool Archive::hasSymbolTable() const {
505  return SymbolTable != child_end();
506}
507