1 """Metadata read/write support for bup."""
3 # Copyright (C) 2010 Rob Browning
5 # This code is covered under the terms of the GNU Library General
6 # Public License as described in the bup LICENSE file.
8 import errno, os, sys, stat, pwd, grp, struct, xattr, posix1e, re
10 from cStringIO import StringIO
12 from bup.helpers import mkdirp, log
13 from bup._helpers import get_linux_file_attr, set_linux_file_attr, lutimes
15 # WARNING: the metadata encoding is *not* stable yet. Caveat emptor!
17 # Q: Consider hardlink support?
18 # Q: Is it OK to store raw linux attr (chattr) flags?
19 # Q: Can anything other than S_ISREG(x) or S_ISDIR(x) support posix1e ACLs?
20 # Q: Is the application of posix1e has_extended() correct?
21 # Q: Is one global --numeric-ids argument sufficient?
22 # Q: Do nfsv4 acls trump posix1e acls? (seems likely)
23 # Q: Add support for crtime -- ntfs, and (only internally?) ext*?
25 # FIXME: Fix relative/abs path detection/stripping wrt other platforms.
26 # FIXME: Add nfsv4 acl handling - see nfs4-acl-tools.
27 # FIXME: Consider other entries mentioned in stat(2) (S_IFDOOR, etc.).
28 # FIXME: Consider pack('vvvvsss', ...) optimization.
29 # FIXME: Consider caching users/groups.
33 # osx (varies between hfs and hfs+):
34 # type - regular dir char block fifo socket ...
35 # perms - rwxrwxrwxsgt
36 # times - ctime atime mtime
39 # hard-link-info (hfs+ only)
42 # attributes-osx see chflags
48 # type - regular dir ...
49 # times - creation, modification, posix change, access
52 # attributes - see attrib
54 # forks (alternate data streams)
58 # type - regular dir ...
59 # perms - rwxrwxrwx (maybe - see wikipedia)
60 # times - creation, modification, access
61 # attributes - see attrib
65 _have_lchmod = hasattr(os, 'lchmod')
68 def _clean_up_path_for_archive(p):
69 # Not the most efficient approach.
72 # Take everything after any '/../'.
73 pos = result.rfind('/../')
75 result = result[result.rfind('/../') + 4:]
77 # Take everything after any remaining '../'.
78 if result.startswith("../"):
81 # Remove any '/./' sequences.
82 pos = result.find('/./')
84 result = result[0:pos] + '/' + result[pos + 3:]
85 pos = result.find('/./')
87 # Remove any leading '/'s.
88 result = result.lstrip('/')
90 # Replace '//' with '/' everywhere.
91 pos = result.find('//')
93 result = result[0:pos] + '/' + result[pos + 2:]
94 pos = result.find('//')
96 # Take everything after any remaining './'.
97 if result.startswith('./'):
100 # Take everything before any remaining '/.'.
101 if result.endswith('/.'):
104 if result == '' or result.endswith('/..'):
111 if p.startswith('/'):
113 if p.find('/../') != -1:
115 if p.startswith('../'):
117 if p.endswith('/..'):
122 def _clean_up_extract_path(p):
123 result = p.lstrip('/')
126 elif _risky_path(result):
132 # These tags are currently conceptually private to Metadata, and they
133 # must be unique, and must *never* be changed.
136 _rec_tag_common = 2 # times, owner, group, type, perms, etc.
137 _rec_tag_symlink_target = 3
138 _rec_tag_posix1e_acl = 4 # getfacl(1), setfacl(1), etc.
139 _rec_tag_nfsv4_acl = 5 # intended to supplant posix1e acls?
140 _rec_tag_linux_attr = 6 # lsattr(1) chattr(1)
141 _rec_tag_linux_xattr = 7 # getfattr(1) setfattr(1)
145 # Metadata is stored as a sequence of tagged binary records. Each
146 # record will have some subset of add, encode, load, create, and
147 # apply methods, i.e. _add_foo...
151 def _add_common(self, path, st):
152 self.mode = st.st_mode
155 self.atime = st.st_atime
156 self.mtime = st.st_mtime
157 self.ctime = st.st_ctime
158 self.rdev = st.st_rdev
159 self.owner = pwd.getpwuid(st.st_uid)[0]
160 self.group = grp.getgrgid(st.st_gid)[0]
162 def _encode_common(self):
163 result = vint.pack('VVsVsVVVVVVV',
172 int(self.atime * 1e9) % 1000000000,
173 int(self.mtime * 1e9) % 1000000000,
174 int(self.ctime * 1e9) % 1000000000,
178 def _load_common_rec(self, port):
179 data = vint.read_bvec(port)
191 self.rdev) = vint.unpack('VVsVsVVVVVVV', data)
192 self.atime = atime_s + (atime_ns / 1e9)
193 self.mtime = mtime_s + (mtime_ns / 1e9)
194 self.ctime = ctime_s + (ctime_ns / 1e9)
196 def _create_via_common_rec(self, path, create_symlinks=True):
197 if stat.S_ISREG(self.mode):
198 os.mknod(path, 0600 | stat.S_IFREG)
199 elif stat.S_ISDIR(self.mode):
201 elif stat.S_ISCHR(self.mode):
202 os.mknod(path, 0600 | stat.S_IFCHR, self.rdev)
203 elif stat.S_ISBLK(self.mode):
204 os.mknod(path, 0600 | stat.S_IFBLK, self.rdev)
205 elif stat.S_ISFIFO(self.mode):
206 os.mknod(path, 0600 | stat.S_IFIFO)
207 elif stat.S_ISLNK(self.mode):
208 if(self.symlink_target and create_symlinks):
209 os.symlink(self.symlink_target, path)
210 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
211 # Otherwise, do nothing.
213 def _apply_common_rec(self, path, restore_numeric_ids=False):
214 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
215 if stat.S_ISLNK(self.mode):
216 lutimes(path, (self.atime, self.mtime))
218 os.utime(path, (self.atime, self.mtime))
219 if stat.S_ISREG(self.mode) \
220 | stat.S_ISDIR(self.mode) \
221 | stat.S_ISCHR(self.mode) \
222 | stat.S_ISBLK(self.mode) \
223 | stat.S_ISLNK(self.mode) \
224 | stat.S_ISFIFO(self.mode):
228 elif not stat.S_ISLNK(self.mode):
233 if not restore_numeric_ids:
234 uid = pwd.getpwnam(self.owner)[2]
235 gid = grp.getgrnam(self.group)[2]
236 os.lchown(path, uid, gid)
239 os.lchmod(path, stat.S_IMODE(self.mode))
240 elif not stat.S_ISLNK(self.mode):
241 os.chmod(path, stat.S_IMODE(self.mode))
246 def _encode_path(self):
248 return vint.pack('s', self.path)
252 def _load_path_rec(self, port):
253 self.path = vint.unpack('s', vint.read_bvec(port))[0]
258 def _add_symlink_target(self, path, st):
259 if(stat.S_ISLNK(st.st_mode)):
260 self.symlink_target = os.readlink(path)
262 def _encode_symlink_target(self):
263 return self.symlink_target
265 def _load_symlink_target_rec(self, port):
266 self.symlink_target = vint.read_bvec(port)
269 ## POSIX1e ACL records
271 # Recorded as a list:
272 # [txt_id_acl, num_id_acl]
273 # or, if a directory:
274 # [txt_id_acl, num_id_acl, txt_id_default_acl, num_id_default_acl]
275 # The numeric/text distinction only matters when reading/restoring
277 def _add_posix1e_acl(self, path, st):
278 if not stat.S_ISLNK(st.st_mode):
280 if posix1e.has_extended(path):
281 acl = posix1e.ACL(file=path)
282 self.posix1e_acl = [acl, acl] # txt and num are the same
283 if stat.S_ISDIR(st.st_mode):
284 acl = posix1e.ACL(filedef=path)
285 self.posix1e_acl.extend([acl, acl])
286 except EnvironmentError, e:
287 if e.errno != errno.EOPNOTSUPP:
290 def _encode_posix1e_acl(self):
291 # Encode as two strings (w/default ACL string possibly empty).
293 acls = self.posix1e_acl
294 txt_flags = posix1e.TEXT_ABBREVIATE
295 num_flags = posix1e.TEXT_ABBREVIATE | posix1e.TEXT_NUMERIC_IDS
296 acl_reps = [acls[0].to_any_text('', '\n', txt_flags),
297 acls[1].to_any_text('', '\n', num_flags)]
301 acl_reps.append(acls[2].to_any_text('', '\n', txt_flags))
302 acl_reps.append(acls[3].to_any_text('', '\n', num_flags))
303 return vint.pack('ssss',
304 acl_reps[0], acl_reps[1], acl_reps[2], acl_reps[3])
308 def _load_posix1e_acl_rec(self, port):
309 data = vint.read_bvec(port)
310 acl_reps = vint.unpack('ssss', data)
311 if(acl_reps[2] == ''):
312 acl_reps = acl_reps[:2]
313 self.posix1e_acl = [posix1e.ACL(x) for x in acl_reps]
315 def _apply_posix1e_acl_rec(self, path, restore_numeric_ids=False):
316 if(self.posix1e_acl):
317 acls = self.posix1e_acl
319 if restore_numeric_ids:
320 acls[3].applyto(path, posix1e.ACL_TYPE_DEFAULT)
322 acls[2].applyto(path, posix1e.ACL_TYPE_DEFAULT)
323 if restore_numeric_ids:
324 acls[1].applyto(path, posix1e.ACL_TYPE_ACCESS)
326 acls[0].applyto(path, posix1e.ACL_TYPE_ACCESS)
329 ## Linux attributes (lsattr(1), chattr(1))
331 def _add_linux_attr(self, path, st):
332 if stat.S_ISREG(st.st_mode) or stat.S_ISDIR(st.st_mode):
333 attr = get_linux_file_attr(path)
335 self.linux_attr = get_linux_file_attr(path)
337 def _encode_linux_attr(self):
339 return vint.pack('V', self.linux_attr)
343 def _load_linux_attr_rec(self, port):
344 data = vint.read_bvec(port)
345 self.linux_attr = vint.unpack('V', data)[0]
347 def _apply_linux_attr_rec(self, path, restore_numeric_ids=False):
349 set_linux_file_attr(path, self.linux_attr)
352 ## Linux extended attributes (getfattr(1), setfattr(1))
354 def _add_linux_xattr(self, path, st):
356 self.linux_xattr = xattr.get_all(path, nofollow=True)
357 except EnvironmentError, e:
358 if e.errno != errno.EOPNOTSUPP:
361 def _encode_linux_xattr(self):
363 result = vint.pack('V', len(items))
364 for name, value in self.attrs:
365 result += vint.pack('ss', name, value)
370 def _load_linux_xattr_rec(self, file):
371 data = vint.read_bvec(file)
372 memfile = StringIO(data)
374 for i in range(vint.read_vuint(memfile)):
375 key = vint.read_bvec(memfile)
376 value = vint.read_bvec(memfile)
377 result.append((key, value))
378 self.linux_xattr = result
380 def _apply_linux_xattr_rec(self, path, restore_numeric_ids=False):
381 if(self.linux_xattr):
382 for k, v in self.linux_xattr:
383 xattr.set(path, k, v, nofollow=True)
388 self.symlink_target = None
389 self.linux_attr = None
390 self.linux_xattr = None
391 self.posix1e_acl = None
392 self.posix1e_acl_default = None
394 def write(self, port, include_path=True):
395 records = [(_rec_tag_path, self._encode_path())] if include_path else []
396 records.extend([(_rec_tag_common, self._encode_common()),
397 (_rec_tag_symlink_target, self._encode_symlink_target()),
398 (_rec_tag_posix1e_acl, self._encode_posix1e_acl()),
399 (_rec_tag_linux_attr, self._encode_linux_attr()),
400 (_rec_tag_linux_xattr, self._encode_linux_xattr())])
401 for tag, data in records:
403 vint.write_vuint(port, tag)
404 vint.write_bvec(port, data)
405 vint.write_vuint(port, _rec_tag_end)
409 # This method should either: return a valid Metadata object;
410 # throw EOFError if there was nothing at all to read; throw an
411 # Exception if a valid object could not be read completely.
412 tag = vint.read_vuint(port)
413 try: # From here on, EOF is an error.
415 while(True): # only exit is error (exception) or _rec_tag_end
416 if tag == _rec_tag_path:
417 result._load_path_rec(port)
418 elif tag == _rec_tag_common:
419 result._load_common_rec(port)
420 elif tag == _rec_tag_symlink_target:
421 result._load_symlink_target_rec(port)
422 elif tag == _rec_tag_posix1e_acl:
423 result._load_posix1e_acl(port)
424 elif tag ==_rec_tag_nfsv4_acl:
425 result._load_nfsv4_acl_rec(port)
426 elif tag == _rec_tag_linux_attr:
427 result._load_linux_attr_rec(port)
428 elif tag == _rec_tag_linux_xattr:
429 result._load_linux_xattr_rec(port)
430 elif tag == _rec_tag_end:
432 else: # unknown record
434 tag = vint.read_vuint(port)
436 raise Exception("EOF while reading Metadata")
439 return stat.S_ISDIR(self.mode)
441 def create_path(self, path, create_symlinks=True):
442 self._create_via_common_rec(path, create_symlinks=create_symlinks)
444 def apply_to_path(self, path=None, restore_numeric_ids=False):
445 # apply metadata to path -- file must exist
449 raise Exception('Metadata.apply_to_path() called with no path');
450 num_ids = restore_numeric_ids
451 self._apply_common_rec(path, restore_numeric_ids=num_ids)
452 self._apply_posix1e_acl_rec(path, restore_numeric_ids=num_ids)
453 self._apply_linux_attr_rec(path, restore_numeric_ids=num_ids)
454 self._apply_linux_xattr_rec(path, restore_numeric_ids=num_ids)
457 def from_path(path, archive_path=None, save_symlinks=True):
459 result.path = archive_path
461 result._add_common(path, st)
463 result._add_symlink_target(path, st)
464 result._add_posix1e_acl(path, st)
465 result._add_linux_attr(path, st)
466 result._add_linux_xattr(path, st)
470 def save_tree(output_file, paths,
475 safe_path = _clean_up_path_for_archive(p)
477 log('bup: archiving "%s" as "%s"\n' % (p, safe_path))
479 # Handle path itself.
480 m = from_path(p, archive_path=safe_path, save_symlinks=save_symlinks)
482 print >> sys.stderr, m.path
483 m.write(output_file, include_path=write_paths)
485 if recurse and os.path.isdir(p):
488 for root, dirs, files in os.walk(p, onerror=raise_error):
490 for sub_path in items:
491 full_path = os.path.join(root, sub_path)
492 safe_path = _clean_up_path_for_archive(full_path)
493 m = from_path(full_path,
494 archive_path=safe_path,
495 save_symlinks=save_symlinks)
497 print >> sys.stderr, m.path
498 m.write(output_file, include_path=write_paths)
501 def _set_up_path(meta, create_symlinks=True):
502 # Allow directories to exist as a special case -- might have
503 # been created by an earlier longer path.
505 mkdirp(meta.path, 0700)
507 parent = os.path.dirname(meta.path)
510 meta.create_path(meta.path, create_symlinks=create_symlinks)
513 class _ArchiveIterator:
516 return Metadata.read(self._file)
518 raise StopIteration()
523 def __init__(self, file):
527 def display_archive(file):
528 for meta in _ArchiveIterator(file):
530 print meta.path # FIXME
535 def start_extract(file, create_symlinks=True):
536 for meta in _ArchiveIterator(file):
538 print >> sys.stderr, meta.path
539 xpath = _clean_up_extract_path(meta.path)
541 add_error(Exception('skipping risky path "%s"' % meta.path))
544 _set_up_path(meta, create_symlinks=create_symlinks)
547 def finish_extract(file, restore_numeric_ids=False):
549 for meta in _ArchiveIterator(file):
550 xpath = _clean_up_extract_path(meta.path)
552 add_error(Exception('skipping risky path "%s"' % dir.path))
554 if os.path.isdir(meta.path):
555 all_dirs.append(meta)
558 print >> sys.stderr, meta.path
559 meta.apply_to_path(path=xpath,
560 restore_numeric_ids=restore_numeric_ids)
562 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
564 # Don't need to check xpath -- won't be in all_dirs if not OK.
565 xpath = _clean_up_extract_path(dir.path)
567 print >> sys.stderr, dir.path
568 dir.apply_to_path(path=xpath,
569 restore_numeric_ids=restore_numeric_ids)
572 def extract(file, restore_numeric_ids=False, create_symlinks=True):
573 # For now, just store all the directories and handle them last,
576 for meta in _ArchiveIterator(file):
577 xpath = _clean_up_extract_path(meta.path)
579 add_error(Exception('skipping risky path "%s"' % meta.path))
583 print >> sys.stderr, '+', meta.path
584 _set_up_path(meta, create_symlinks=create_symlinks)
585 if os.path.isdir(meta.path):
586 all_dirs.append(meta)
589 print >> sys.stderr, '=', meta.path
590 meta.apply_to_path(restore_numeric_ids=restore_numeric_ids)
591 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
593 # Don't need to check xpath -- won't be in all_dirs if not OK.
594 xpath = _clean_up_extract_path(meta.path)
596 print >> sys.stderr, '=', meta.path
597 # Shouldn't have to check for risky paths here (omitted above).
598 dir.apply_to_path(path=dir.path,
599 restore_numeric_ids=restore_numeric_ids)