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 os.mknod(path, 0600 | stat.S_IFREG)
263 elif stat.S_ISDIR(self.mode):
265 elif stat.S_ISCHR(self.mode):
266 os.mknod(path, 0600 | stat.S_IFCHR, self.rdev)
267 elif stat.S_ISBLK(self.mode):
268 os.mknod(path, 0600 | stat.S_IFBLK, self.rdev)
269 elif stat.S_ISFIFO(self.mode):
270 os.mknod(path, 0600 | stat.S_IFIFO)
271 elif stat.S_ISLNK(self.mode):
272 if self.symlink_target and create_symlinks:
273 os.symlink(self.symlink_target, path)
274 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
275 # Otherwise, do nothing.
277 def _apply_common_rec(self, path, restore_numeric_ids=False):
278 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
279 # EACCES errors at this stage are fatal for the current path.
280 if stat.S_ISLNK(self.mode):
282 lutime(path, (self.atime, self.mtime))
284 if e.errno == errno.EACCES:
285 raise ApplyError('lutime: %s' % e)
290 utime(path, (self.atime, self.mtime))
292 if e.errno == errno.EACCES:
293 raise ApplyError('utime: %s' % e)
297 # Don't try to restore owner unless we're root, and even
298 # if asked, don't try to restore the owner or group if
299 # it doesn't exist in the system db.
302 if not restore_numeric_ids:
305 add_error('ignoring missing owner for "%s"\n' % path)
307 if os.geteuid() != 0:
308 uid = -1 # Not root; assume we can't change owner.
311 uid = pwd.getpwnam(self.owner)[2]
314 fmt = 'ignoring unknown owner %s for "%s"\n'
315 add_error(fmt % (self.owner, path))
318 add_error('ignoring missing group for "%s"\n' % path)
321 gid = grp.getgrnam(self.group)[2]
324 add_error('ignoring unknown group %s for "%s"\n'
325 % (self.group, path))
328 os.lchown(path, uid, gid)
330 if e.errno == errno.EPERM:
331 add_error('lchown: %s' % e)
336 os.lchmod(path, stat.S_IMODE(self.mode))
337 elif not stat.S_ISLNK(self.mode):
338 os.chmod(path, stat.S_IMODE(self.mode))
343 def _encode_path(self):
345 return vint.pack('s', self.path)
349 def _load_path_rec(self, port):
350 self.path = vint.unpack('s', vint.read_bvec(port))[0]
355 def _add_symlink_target(self, path, st):
357 if stat.S_ISLNK(st.st_mode):
358 self.symlink_target = os.readlink(path)
360 add_error('readlink: %s', e)
362 def _encode_symlink_target(self):
363 return self.symlink_target
365 def _load_symlink_target_rec(self, port):
366 self.symlink_target = vint.read_bvec(port)
369 ## POSIX1e ACL records
371 # Recorded as a list:
372 # [txt_id_acl, num_id_acl]
373 # or, if a directory:
374 # [txt_id_acl, num_id_acl, txt_id_default_acl, num_id_default_acl]
375 # The numeric/text distinction only matters when reading/restoring
377 def _add_posix1e_acl(self, path, st):
378 if not posix1e: return
379 if not stat.S_ISLNK(st.st_mode):
381 if posix1e.has_extended(path):
382 acl = posix1e.ACL(file=path)
383 self.posix1e_acl = [acl, acl] # txt and num are the same
384 if stat.S_ISDIR(st.st_mode):
385 acl = posix1e.ACL(filedef=path)
386 self.posix1e_acl.extend([acl, acl])
387 except EnvironmentError, e:
388 if e.errno != errno.EOPNOTSUPP:
391 def _encode_posix1e_acl(self):
392 # Encode as two strings (w/default ACL string possibly empty).
394 acls = self.posix1e_acl
395 txt_flags = posix1e.TEXT_ABBREVIATE
396 num_flags = posix1e.TEXT_ABBREVIATE | posix1e.TEXT_NUMERIC_IDS
397 acl_reps = [acls[0].to_any_text('', '\n', txt_flags),
398 acls[1].to_any_text('', '\n', num_flags)]
402 acl_reps.append(acls[2].to_any_text('', '\n', txt_flags))
403 acl_reps.append(acls[3].to_any_text('', '\n', num_flags))
404 return vint.pack('ssss',
405 acl_reps[0], acl_reps[1], acl_reps[2], acl_reps[3])
409 def _load_posix1e_acl_rec(self, port):
410 data = vint.read_bvec(port)
411 acl_reps = vint.unpack('ssss', data)
412 if acl_reps[2] == '':
413 acl_reps = acl_reps[:2]
414 self.posix1e_acl = [posix1e.ACL(text=x) for x in acl_reps]
416 def _apply_posix1e_acl_rec(self, path, restore_numeric_ids=False):
419 add_error("%s: can't restore ACLs; posix1e support missing.\n"
423 acls = self.posix1e_acl
425 if restore_numeric_ids:
426 acls[3].applyto(path, posix1e.ACL_TYPE_DEFAULT)
428 acls[2].applyto(path, posix1e.ACL_TYPE_DEFAULT)
429 if restore_numeric_ids:
430 acls[1].applyto(path, posix1e.ACL_TYPE_ACCESS)
432 acls[0].applyto(path, posix1e.ACL_TYPE_ACCESS)
435 ## Linux attributes (lsattr(1), chattr(1))
437 def _add_linux_attr(self, path, st):
438 if not get_linux_file_attr: return
439 if stat.S_ISREG(st.st_mode) or stat.S_ISDIR(st.st_mode):
441 attr = get_linux_file_attr(path)
443 self.linux_attr = attr
445 if e.errno == errno.EACCES:
446 add_error('read Linux attr: %s' % e)
447 elif e.errno == errno.ENOTTY: # Inappropriate ioctl for device.
448 add_error('read Linux attr: %s' % e)
452 def _encode_linux_attr(self):
454 return vint.pack('V', self.linux_attr)
458 def _load_linux_attr_rec(self, port):
459 data = vint.read_bvec(port)
460 self.linux_attr = vint.unpack('V', data)[0]
462 def _apply_linux_attr_rec(self, path, restore_numeric_ids=False):
464 if not set_linux_file_attr:
465 add_error("%s: can't restore linuxattrs: "
466 "linuxattr support missing.\n" % path)
468 set_linux_file_attr(path, self.linux_attr)
471 ## Linux extended attributes (getfattr(1), setfattr(1))
473 def _add_linux_xattr(self, path, st):
476 self.linux_xattr = xattr.get_all(path, nofollow=True)
477 except EnvironmentError, e:
478 if e.errno != errno.EOPNOTSUPP:
481 def _encode_linux_xattr(self):
483 result = vint.pack('V', len(self.linux_xattr))
484 for name, value in self.linux_xattr:
485 result += vint.pack('ss', name, value)
490 def _load_linux_xattr_rec(self, file):
491 data = vint.read_bvec(file)
492 memfile = StringIO(data)
494 for i in range(vint.read_vuint(memfile)):
495 key = vint.read_bvec(memfile)
496 value = vint.read_bvec(memfile)
497 result.append((key, value))
498 self.linux_xattr = result
500 def _apply_linux_xattr_rec(self, path, restore_numeric_ids=False):
503 add_error("%s: can't restore xattr; xattr support missing.\n"
506 existing_xattrs = set(xattr.list(path, nofollow=True))
508 for k, v in self.linux_xattr:
509 if k not in existing_xattrs \
510 or v != xattr.get(path, k, nofollow=True):
512 xattr.set(path, k, v, nofollow=True)
514 if e.errno == errno.EPERM:
515 raise ApplyError('xattr.set: %s' % e)
518 existing_xattrs -= frozenset([k])
519 for k in existing_xattrs:
521 xattr.remove(path, k, nofollow=True)
523 if e.errno == errno.EPERM:
524 raise ApplyError('xattr.remove: %s' % e)
531 self.symlink_target = None
532 self.linux_attr = None
533 self.linux_xattr = None
534 self.posix1e_acl = None
535 self.posix1e_acl_default = None
537 def write(self, port, include_path=True):
538 records = [(_rec_tag_path, self._encode_path())] if include_path else []
539 records.extend([(_rec_tag_common, self._encode_common()),
540 (_rec_tag_symlink_target, self._encode_symlink_target()),
541 (_rec_tag_posix1e_acl, self._encode_posix1e_acl()),
542 (_rec_tag_linux_attr, self._encode_linux_attr()),
543 (_rec_tag_linux_xattr, self._encode_linux_xattr())])
544 for tag, data in records:
546 vint.write_vuint(port, tag)
547 vint.write_bvec(port, data)
548 vint.write_vuint(port, _rec_tag_end)
552 # This method should either: return a valid Metadata object;
553 # throw EOFError if there was nothing at all to read; throw an
554 # Exception if a valid object could not be read completely.
555 tag = vint.read_vuint(port)
556 try: # From here on, EOF is an error.
558 while True: # only exit is error (exception) or _rec_tag_end
559 if tag == _rec_tag_path:
560 result._load_path_rec(port)
561 elif tag == _rec_tag_common:
562 result._load_common_rec(port)
563 elif tag == _rec_tag_symlink_target:
564 result._load_symlink_target_rec(port)
565 elif tag == _rec_tag_posix1e_acl:
566 result._load_posix1e_acl_rec(port)
567 elif tag ==_rec_tag_nfsv4_acl:
568 result._load_nfsv4_acl_rec(port)
569 elif tag == _rec_tag_linux_attr:
570 result._load_linux_attr_rec(port)
571 elif tag == _rec_tag_linux_xattr:
572 result._load_linux_xattr_rec(port)
573 elif tag == _rec_tag_end:
575 else: # unknown record
577 tag = vint.read_vuint(port)
579 raise Exception("EOF while reading Metadata")
582 return stat.S_ISDIR(self.mode)
584 def create_path(self, path, create_symlinks=True):
585 self._create_via_common_rec(path, create_symlinks=create_symlinks)
587 def apply_to_path(self, path=None, restore_numeric_ids=False):
588 # apply metadata to path -- file must exist
592 raise Exception('Metadata.apply_to_path() called with no path');
593 num_ids = restore_numeric_ids
595 self._apply_common_rec(path, restore_numeric_ids=num_ids)
596 self._apply_posix1e_acl_rec(path, restore_numeric_ids=num_ids)
597 self._apply_linux_attr_rec(path, restore_numeric_ids=num_ids)
598 self._apply_linux_xattr_rec(path, restore_numeric_ids=num_ids)
599 except ApplyError, e:
603 def from_path(path, statinfo=None, archive_path=None, save_symlinks=True):
605 result.path = archive_path
606 st = statinfo if statinfo else lstat(path)
607 result._add_common(path, st)
609 result._add_symlink_target(path, st)
610 result._add_posix1e_acl(path, st)
611 result._add_linux_attr(path, st)
612 result._add_linux_xattr(path, st)
616 def save_tree(output_file, paths,
622 # Issue top-level rewrite warnings.
624 safe_path = _clean_up_path_for_archive(path)
625 if safe_path != path:
626 log('archiving "%s" as "%s"\n' % (path, safe_path))
628 start_dir = os.getcwd()
630 for (p, st) in recursive_dirlist(paths, xdev=xdev):
631 dirlist_dir = os.getcwd()
633 safe_path = _clean_up_path_for_archive(p)
634 m = from_path(p, statinfo=st, archive_path=safe_path,
635 save_symlinks=save_symlinks)
637 print >> sys.stderr, m.path
638 m.write(output_file, include_path=write_paths)
639 os.chdir(dirlist_dir)
644 def _set_up_path(meta, create_symlinks=True):
645 # Allow directories to exist as a special case -- might have
646 # been created by an earlier longer path.
650 parent = os.path.dirname(meta.path)
653 meta.create_path(meta.path, create_symlinks=create_symlinks)
656 class _ArchiveIterator:
659 return Metadata.read(self._file)
661 raise StopIteration()
666 def __init__(self, file):
670 def display_archive(file):
671 for meta in _ArchiveIterator(file):
673 print meta.path # FIXME
678 def start_extract(file, create_symlinks=True):
679 for meta in _ArchiveIterator(file):
681 print >> sys.stderr, meta.path
682 xpath = _clean_up_extract_path(meta.path)
684 add_error(Exception('skipping risky path "%s"' % meta.path))
687 _set_up_path(meta, create_symlinks=create_symlinks)
690 def finish_extract(file, restore_numeric_ids=False):
692 for meta in _ArchiveIterator(file):
693 xpath = _clean_up_extract_path(meta.path)
695 add_error(Exception('skipping risky path "%s"' % dir.path))
697 if os.path.isdir(meta.path):
698 all_dirs.append(meta)
701 print >> sys.stderr, meta.path
702 meta.apply_to_path(path=xpath,
703 restore_numeric_ids=restore_numeric_ids)
704 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
706 # Don't need to check xpath -- won't be in all_dirs if not OK.
707 xpath = _clean_up_extract_path(dir.path)
709 print >> sys.stderr, dir.path
710 dir.apply_to_path(path=xpath, restore_numeric_ids=restore_numeric_ids)
713 def extract(file, restore_numeric_ids=False, create_symlinks=True):
714 # For now, just store all the directories and handle them last,
717 for meta in _ArchiveIterator(file):
718 xpath = _clean_up_extract_path(meta.path)
720 add_error(Exception('skipping risky path "%s"' % meta.path))
724 print >> sys.stderr, '+', meta.path
725 _set_up_path(meta, create_symlinks=create_symlinks)
726 if os.path.isdir(meta.path):
727 all_dirs.append(meta)
730 print >> sys.stderr, '=', meta.path
731 meta.apply_to_path(restore_numeric_ids=restore_numeric_ids)
732 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
734 # Don't need to check xpath -- won't be in all_dirs if not OK.
735 xpath = _clean_up_extract_path(dir.path)
737 print >> sys.stderr, '=', xpath
738 # Shouldn't have to check for risky paths here (omitted above).
739 dir.apply_to_path(path=dir.path,
740 restore_numeric_ids=restore_numeric_ids)