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.
7 import errno, os, sys, stat, pwd, grp, struct, re
8 from cStringIO import StringIO
10 from bup.drecurse import recursive_dirlist
11 from bup.helpers import add_error, mkdirp, log
12 from bup.xstat import utime, lutime, lstat, FSTime
13 import bup._helpers as _helpers
18 log('Warning: Linux xattr support missing; install python-pyxattr.\n')
23 except AttributeError:
24 log('Warning: python-xattr module is too old; '
25 'install python-pyxattr instead.\n')
30 log('Warning: POSIX ACL support missing; install python-pylibacl.\n')
33 from bup._helpers import get_linux_file_attr, set_linux_file_attr
35 # No need for a warning here; the only reason they won't exist is that we're
36 # not on Linux, in which case files don't have any linux attrs anyway, so
37 # lacking the functions isn't a problem.
38 get_linux_file_attr = set_linux_file_attr = None
41 # WARNING: the metadata encoding is *not* stable yet. Caveat emptor!
43 # Q: Consider hardlink support?
44 # Q: Is it OK to store raw linux attr (chattr) flags?
45 # Q: Can anything other than S_ISREG(x) or S_ISDIR(x) support posix1e ACLs?
46 # Q: Is the application of posix1e has_extended() correct?
47 # Q: Is one global --numeric-ids argument sufficient?
48 # Q: Do nfsv4 acls trump posix1e acls? (seems likely)
49 # Q: Add support for crtime -- ntfs, and (only internally?) ext*?
51 # FIXME: Fix relative/abs path detection/stripping wrt other platforms.
52 # FIXME: Add nfsv4 acl handling - see nfs4-acl-tools.
53 # FIXME: Consider other entries mentioned in stat(2) (S_IFDOOR, etc.).
54 # FIXME: Consider pack('vvvvsss', ...) optimization.
55 # FIXME: Consider caching users/groups.
59 # osx (varies between hfs and hfs+):
60 # type - regular dir char block fifo socket ...
61 # perms - rwxrwxrwxsgt
62 # times - ctime atime mtime
65 # hard-link-info (hfs+ only)
68 # attributes-osx see chflags
74 # type - regular dir ...
75 # times - creation, modification, posix change, access
78 # attributes - see attrib
80 # forks (alternate data streams)
84 # type - regular dir ...
85 # perms - rwxrwxrwx (maybe - see wikipedia)
86 # times - creation, modification, access
87 # attributes - see attrib
91 _have_lchmod = hasattr(os, 'lchmod')
94 def _clean_up_path_for_archive(p):
95 # Not the most efficient approach.
98 # Take everything after any '/../'.
99 pos = result.rfind('/../')
101 result = result[result.rfind('/../') + 4:]
103 # Take everything after any remaining '../'.
104 if result.startswith("../"):
107 # Remove any '/./' sequences.
108 pos = result.find('/./')
110 result = result[0:pos] + '/' + result[pos + 3:]
111 pos = result.find('/./')
113 # Remove any leading '/'s.
114 result = result.lstrip('/')
116 # Replace '//' with '/' everywhere.
117 pos = result.find('//')
119 result = result[0:pos] + '/' + result[pos + 2:]
120 pos = result.find('//')
122 # Take everything after any remaining './'.
123 if result.startswith('./'):
126 # Take everything before any remaining '/.'.
127 if result.endswith('/.'):
130 if result == '' or result.endswith('/..'):
137 if p.startswith('/'):
139 if p.find('/../') != -1:
141 if p.startswith('../'):
143 if p.endswith('/..'):
148 def _clean_up_extract_path(p):
149 result = p.lstrip('/')
152 elif _risky_path(result):
158 # These tags are currently conceptually private to Metadata, and they
159 # must be unique, and must *never* be changed.
162 _rec_tag_common = 2 # times, owner, group, type, perms, etc.
163 _rec_tag_symlink_target = 3
164 _rec_tag_posix1e_acl = 4 # getfacl(1), setfacl(1), etc.
165 _rec_tag_nfsv4_acl = 5 # intended to supplant posix1e acls?
166 _rec_tag_linux_attr = 6 # lsattr(1) chattr(1)
167 _rec_tag_linux_xattr = 7 # getfattr(1) setfattr(1)
170 class ApplyError(Exception):
171 # Thrown when unable to apply any given bit of metadata to a path.
176 # Metadata is stored as a sequence of tagged binary records. Each
177 # record will have some subset of add, encode, load, create, and
178 # apply methods, i.e. _add_foo...
182 # Timestamps are (sec, ns), relative to 1970-01-01 00:00:00, ns
183 # must be non-negative and < 10**9.
185 def _add_common(self, path, st):
186 self.mode = st.st_mode
189 self.rdev = st.st_rdev
190 self.atime = st.st_atime
191 self.mtime = st.st_mtime
192 self.ctime = st.st_ctime
193 self.owner = self.group = ''
195 self.owner = pwd.getpwuid(st.st_uid)[0]
197 add_error("no user name for id %s '%s'" % (st.st_gid, path))
199 self.group = grp.getgrgid(st.st_gid)[0]
201 add_error("no group name for id %s '%s'" % (st.st_gid, path))
203 def _encode_common(self):
204 atime = self.atime.to_timespec()
205 mtime = self.mtime.to_timespec()
206 ctime = self.ctime.to_timespec()
207 result = vint.pack('VVsVsVvVvVvV',
222 def _load_common_rec(self, port):
223 data = vint.read_bvec(port)
235 ctime_ns) = vint.unpack('VVsVsVvVvVvV', data)
236 self.atime = FSTime.from_timespec((self.atime, atime_ns))
237 self.mtime = FSTime.from_timespec((self.mtime, mtime_ns))
238 self.ctime = FSTime.from_timespec((self.ctime, ctime_ns))
240 def _create_via_common_rec(self, path, create_symlinks=True):
241 # If the path already exists and is a dir, try rmdir.
242 # If the path already exists and is anything else, try unlink.
247 if e.errno != errno.ENOENT:
250 if stat.S_ISDIR(st.st_mode):
254 if e.errno == errno.ENOTEMPTY:
255 msg = 'refusing to overwrite non-empty dir' + path
261 if stat.S_ISREG(self.mode):
262 fd = os.open(path, os.O_CREAT|os.O_WRONLY|os.O_EXCL, 0600)
264 elif stat.S_ISDIR(self.mode):
266 elif stat.S_ISCHR(self.mode):
267 os.mknod(path, 0600 | stat.S_IFCHR, self.rdev)
268 elif stat.S_ISBLK(self.mode):
269 os.mknod(path, 0600 | stat.S_IFBLK, self.rdev)
270 elif stat.S_ISFIFO(self.mode):
271 os.mknod(path, 0600 | stat.S_IFIFO)
272 elif stat.S_ISLNK(self.mode):
273 if self.symlink_target and create_symlinks:
274 os.symlink(self.symlink_target, path)
275 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
276 # Otherwise, do nothing.
278 def _apply_common_rec(self, path, restore_numeric_ids=False):
279 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
280 # EACCES errors at this stage are fatal for the current path.
281 if stat.S_ISLNK(self.mode):
283 lutime(path, (self.atime, self.mtime))
285 if e.errno == errno.EACCES:
286 raise ApplyError('lutime: %s' % e)
291 utime(path, (self.atime, self.mtime))
293 if e.errno == errno.EACCES:
294 raise ApplyError('utime: %s' % e)
298 # Don't try to restore owner unless we're root, and even
299 # if asked, don't try to restore the owner or group if
300 # it doesn't exist in the system db.
303 if not restore_numeric_ids:
306 add_error('ignoring missing owner for "%s"\n' % path)
308 if os.geteuid() != 0:
309 uid = -1 # Not root; assume we can't change owner.
312 uid = pwd.getpwnam(self.owner)[2]
315 fmt = 'ignoring unknown owner %s for "%s"\n'
316 add_error(fmt % (self.owner, path))
319 add_error('ignoring missing group for "%s"\n' % path)
322 gid = grp.getgrnam(self.group)[2]
325 add_error('ignoring unknown group %s for "%s"\n'
326 % (self.group, path))
329 os.lchown(path, uid, gid)
331 if e.errno == errno.EPERM:
332 add_error('lchown: %s' % e)
337 os.lchmod(path, stat.S_IMODE(self.mode))
338 elif not stat.S_ISLNK(self.mode):
339 os.chmod(path, stat.S_IMODE(self.mode))
344 def _encode_path(self):
346 return vint.pack('s', self.path)
350 def _load_path_rec(self, port):
351 self.path = vint.unpack('s', vint.read_bvec(port))[0]
356 def _add_symlink_target(self, path, st):
358 if stat.S_ISLNK(st.st_mode):
359 self.symlink_target = os.readlink(path)
361 add_error('readlink: %s', e)
363 def _encode_symlink_target(self):
364 return self.symlink_target
366 def _load_symlink_target_rec(self, port):
367 self.symlink_target = vint.read_bvec(port)
370 ## POSIX1e ACL records
372 # Recorded as a list:
373 # [txt_id_acl, num_id_acl]
374 # or, if a directory:
375 # [txt_id_acl, num_id_acl, txt_id_default_acl, num_id_default_acl]
376 # The numeric/text distinction only matters when reading/restoring
378 def _add_posix1e_acl(self, path, st):
379 if not posix1e: return
380 if not stat.S_ISLNK(st.st_mode):
382 if posix1e.has_extended(path):
383 acl = posix1e.ACL(file=path)
384 self.posix1e_acl = [acl, acl] # txt and num are the same
385 if stat.S_ISDIR(st.st_mode):
386 acl = posix1e.ACL(filedef=path)
387 self.posix1e_acl.extend([acl, acl])
388 except EnvironmentError, e:
389 if e.errno != errno.EOPNOTSUPP:
392 def _encode_posix1e_acl(self):
393 # Encode as two strings (w/default ACL string possibly empty).
395 acls = self.posix1e_acl
396 txt_flags = posix1e.TEXT_ABBREVIATE
397 num_flags = posix1e.TEXT_ABBREVIATE | posix1e.TEXT_NUMERIC_IDS
398 acl_reps = [acls[0].to_any_text('', '\n', txt_flags),
399 acls[1].to_any_text('', '\n', num_flags)]
403 acl_reps.append(acls[2].to_any_text('', '\n', txt_flags))
404 acl_reps.append(acls[3].to_any_text('', '\n', num_flags))
405 return vint.pack('ssss',
406 acl_reps[0], acl_reps[1], acl_reps[2], acl_reps[3])
410 def _load_posix1e_acl_rec(self, port):
411 data = vint.read_bvec(port)
412 acl_reps = vint.unpack('ssss', data)
413 if acl_reps[2] == '':
414 acl_reps = acl_reps[:2]
415 self.posix1e_acl = [posix1e.ACL(text=x) for x in acl_reps]
417 def _apply_posix1e_acl_rec(self, path, restore_numeric_ids=False):
420 add_error("%s: can't restore ACLs; posix1e support missing.\n"
424 acls = self.posix1e_acl
426 if restore_numeric_ids:
427 acls[3].applyto(path, posix1e.ACL_TYPE_DEFAULT)
429 acls[2].applyto(path, posix1e.ACL_TYPE_DEFAULT)
430 if restore_numeric_ids:
431 acls[1].applyto(path, posix1e.ACL_TYPE_ACCESS)
433 acls[0].applyto(path, posix1e.ACL_TYPE_ACCESS)
436 ## Linux attributes (lsattr(1), chattr(1))
438 def _add_linux_attr(self, path, st):
439 if not get_linux_file_attr: return
440 if stat.S_ISREG(st.st_mode) or stat.S_ISDIR(st.st_mode):
442 attr = get_linux_file_attr(path)
444 self.linux_attr = attr
446 if e.errno == errno.EACCES:
447 add_error('read Linux attr: %s' % e)
448 elif e.errno == errno.ENOTTY: # Inappropriate ioctl for device.
449 add_error('read Linux attr: %s' % e)
453 def _encode_linux_attr(self):
455 return vint.pack('V', self.linux_attr)
459 def _load_linux_attr_rec(self, port):
460 data = vint.read_bvec(port)
461 self.linux_attr = vint.unpack('V', data)[0]
463 def _apply_linux_attr_rec(self, path, restore_numeric_ids=False):
465 if not set_linux_file_attr:
466 add_error("%s: can't restore linuxattrs: "
467 "linuxattr support missing.\n" % path)
469 set_linux_file_attr(path, self.linux_attr)
472 ## Linux extended attributes (getfattr(1), setfattr(1))
474 def _add_linux_xattr(self, path, st):
477 self.linux_xattr = xattr.get_all(path, nofollow=True)
478 except EnvironmentError, e:
479 if e.errno != errno.EOPNOTSUPP:
482 def _encode_linux_xattr(self):
484 result = vint.pack('V', len(self.linux_xattr))
485 for name, value in self.linux_xattr:
486 result += vint.pack('ss', name, value)
491 def _load_linux_xattr_rec(self, file):
492 data = vint.read_bvec(file)
493 memfile = StringIO(data)
495 for i in range(vint.read_vuint(memfile)):
496 key = vint.read_bvec(memfile)
497 value = vint.read_bvec(memfile)
498 result.append((key, value))
499 self.linux_xattr = result
501 def _apply_linux_xattr_rec(self, path, restore_numeric_ids=False):
504 add_error("%s: can't restore xattr; xattr support missing.\n"
507 existing_xattrs = set(xattr.list(path, nofollow=True))
509 for k, v in self.linux_xattr:
510 if k not in existing_xattrs \
511 or v != xattr.get(path, k, nofollow=True):
513 xattr.set(path, k, v, nofollow=True)
515 if e.errno == errno.EPERM:
516 raise ApplyError('xattr.set: %s' % e)
519 existing_xattrs -= frozenset([k])
520 for k in existing_xattrs:
522 xattr.remove(path, k, nofollow=True)
524 if e.errno == errno.EPERM:
525 raise ApplyError('xattr.remove: %s' % e)
532 self.symlink_target = None
533 self.linux_attr = None
534 self.linux_xattr = None
535 self.posix1e_acl = None
536 self.posix1e_acl_default = None
538 def write(self, port, include_path=True):
539 records = include_path and [(_rec_tag_path, self._encode_path())] or []
540 records.extend([(_rec_tag_common, self._encode_common()),
541 (_rec_tag_symlink_target, self._encode_symlink_target()),
542 (_rec_tag_posix1e_acl, self._encode_posix1e_acl()),
543 (_rec_tag_linux_attr, self._encode_linux_attr()),
544 (_rec_tag_linux_xattr, self._encode_linux_xattr())])
545 for tag, data in records:
547 vint.write_vuint(port, tag)
548 vint.write_bvec(port, data)
549 vint.write_vuint(port, _rec_tag_end)
553 # This method should either: return a valid Metadata object;
554 # throw EOFError if there was nothing at all to read; throw an
555 # Exception if a valid object could not be read completely.
556 tag = vint.read_vuint(port)
557 try: # From here on, EOF is an error.
559 while True: # only exit is error (exception) or _rec_tag_end
560 if tag == _rec_tag_path:
561 result._load_path_rec(port)
562 elif tag == _rec_tag_common:
563 result._load_common_rec(port)
564 elif tag == _rec_tag_symlink_target:
565 result._load_symlink_target_rec(port)
566 elif tag == _rec_tag_posix1e_acl:
567 result._load_posix1e_acl_rec(port)
568 elif tag ==_rec_tag_nfsv4_acl:
569 result._load_nfsv4_acl_rec(port)
570 elif tag == _rec_tag_linux_attr:
571 result._load_linux_attr_rec(port)
572 elif tag == _rec_tag_linux_xattr:
573 result._load_linux_xattr_rec(port)
574 elif tag == _rec_tag_end:
576 else: # unknown record
578 tag = vint.read_vuint(port)
580 raise Exception("EOF while reading Metadata")
583 return stat.S_ISDIR(self.mode)
585 def create_path(self, path, create_symlinks=True):
586 self._create_via_common_rec(path, create_symlinks=create_symlinks)
588 def apply_to_path(self, path=None, restore_numeric_ids=False):
589 # apply metadata to path -- file must exist
593 raise Exception('Metadata.apply_to_path() called with no path');
594 num_ids = restore_numeric_ids
596 self._apply_common_rec(path, restore_numeric_ids=num_ids)
597 self._apply_posix1e_acl_rec(path, restore_numeric_ids=num_ids)
598 self._apply_linux_attr_rec(path, restore_numeric_ids=num_ids)
599 self._apply_linux_xattr_rec(path, restore_numeric_ids=num_ids)
600 except ApplyError, e:
604 def from_path(path, statinfo=None, archive_path=None, save_symlinks=True):
606 result.path = archive_path
607 st = statinfo or lstat(path)
608 result._add_common(path, st)
610 result._add_symlink_target(path, st)
611 result._add_posix1e_acl(path, st)
612 result._add_linux_attr(path, st)
613 result._add_linux_xattr(path, st)
617 def save_tree(output_file, paths,
623 # Issue top-level rewrite warnings.
625 safe_path = _clean_up_path_for_archive(path)
626 if safe_path != path:
627 log('archiving "%s" as "%s"\n' % (path, safe_path))
629 start_dir = os.getcwd()
631 for (p, st) in recursive_dirlist(paths, xdev=xdev):
632 dirlist_dir = os.getcwd()
634 safe_path = _clean_up_path_for_archive(p)
635 m = from_path(p, statinfo=st, archive_path=safe_path,
636 save_symlinks=save_symlinks)
638 print >> sys.stderr, m.path
639 m.write(output_file, include_path=write_paths)
640 os.chdir(dirlist_dir)
645 def _set_up_path(meta, create_symlinks=True):
646 # Allow directories to exist as a special case -- might have
647 # been created by an earlier longer path.
651 parent = os.path.dirname(meta.path)
654 meta.create_path(meta.path, create_symlinks=create_symlinks)
657 class _ArchiveIterator:
660 return Metadata.read(self._file)
662 raise StopIteration()
667 def __init__(self, file):
671 def display_archive(file):
672 for meta in _ArchiveIterator(file):
674 print meta.path # FIXME
679 def start_extract(file, create_symlinks=True):
680 for meta in _ArchiveIterator(file):
682 print >> sys.stderr, meta.path
683 xpath = _clean_up_extract_path(meta.path)
685 add_error(Exception('skipping risky path "%s"' % meta.path))
688 _set_up_path(meta, create_symlinks=create_symlinks)
691 def finish_extract(file, restore_numeric_ids=False):
693 for meta in _ArchiveIterator(file):
694 xpath = _clean_up_extract_path(meta.path)
696 add_error(Exception('skipping risky path "%s"' % dir.path))
698 if os.path.isdir(meta.path):
699 all_dirs.append(meta)
702 print >> sys.stderr, meta.path
703 meta.apply_to_path(path=xpath,
704 restore_numeric_ids=restore_numeric_ids)
705 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
707 # Don't need to check xpath -- won't be in all_dirs if not OK.
708 xpath = _clean_up_extract_path(dir.path)
710 print >> sys.stderr, dir.path
711 dir.apply_to_path(path=xpath, restore_numeric_ids=restore_numeric_ids)
714 def extract(file, restore_numeric_ids=False, create_symlinks=True):
715 # For now, just store all the directories and handle them last,
718 for meta in _ArchiveIterator(file):
719 xpath = _clean_up_extract_path(meta.path)
721 add_error(Exception('skipping risky path "%s"' % meta.path))
725 print >> sys.stderr, '+', meta.path
726 _set_up_path(meta, create_symlinks=create_symlinks)
727 if os.path.isdir(meta.path):
728 all_dirs.append(meta)
731 print >> sys.stderr, '=', meta.path
732 meta.apply_to_path(restore_numeric_ids=restore_numeric_ids)
733 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
735 # Don't need to check xpath -- won't be in all_dirs if not OK.
736 xpath = _clean_up_extract_path(dir.path)
738 print >> sys.stderr, '=', xpath
739 # Shouldn't have to check for risky paths here (omitted above).
740 dir.apply_to_path(path=dir.path,
741 restore_numeric_ids=restore_numeric_ids)