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.drecurse import recursive_dirlist
13 from bup.helpers import add_error, mkdirp, log
14 from bup.xstat import utime, lutime, lstat, FSTime
15 import bup._helpers as _helpers
17 if _helpers.get_linux_file_attr:
18 from bup._helpers import get_linux_file_attr, set_linux_file_attr
20 # WARNING: the metadata encoding is *not* stable yet. Caveat emptor!
22 # Q: Consider hardlink support?
23 # Q: Is it OK to store raw linux attr (chattr) flags?
24 # Q: Can anything other than S_ISREG(x) or S_ISDIR(x) support posix1e ACLs?
25 # Q: Is the application of posix1e has_extended() correct?
26 # Q: Is one global --numeric-ids argument sufficient?
27 # Q: Do nfsv4 acls trump posix1e acls? (seems likely)
28 # Q: Add support for crtime -- ntfs, and (only internally?) ext*?
30 # FIXME: Fix relative/abs path detection/stripping wrt other platforms.
31 # FIXME: Add nfsv4 acl handling - see nfs4-acl-tools.
32 # FIXME: Consider other entries mentioned in stat(2) (S_IFDOOR, etc.).
33 # FIXME: Consider pack('vvvvsss', ...) optimization.
34 # FIXME: Consider caching users/groups.
38 # osx (varies between hfs and hfs+):
39 # type - regular dir char block fifo socket ...
40 # perms - rwxrwxrwxsgt
41 # times - ctime atime mtime
44 # hard-link-info (hfs+ only)
47 # attributes-osx see chflags
53 # type - regular dir ...
54 # times - creation, modification, posix change, access
57 # attributes - see attrib
59 # forks (alternate data streams)
63 # type - regular dir ...
64 # perms - rwxrwxrwx (maybe - see wikipedia)
65 # times - creation, modification, access
66 # attributes - see attrib
70 _have_lchmod = hasattr(os, 'lchmod')
73 def _clean_up_path_for_archive(p):
74 # Not the most efficient approach.
77 # Take everything after any '/../'.
78 pos = result.rfind('/../')
80 result = result[result.rfind('/../') + 4:]
82 # Take everything after any remaining '../'.
83 if result.startswith("../"):
86 # Remove any '/./' sequences.
87 pos = result.find('/./')
89 result = result[0:pos] + '/' + result[pos + 3:]
90 pos = result.find('/./')
92 # Remove any leading '/'s.
93 result = result.lstrip('/')
95 # Replace '//' with '/' everywhere.
96 pos = result.find('//')
98 result = result[0:pos] + '/' + result[pos + 2:]
99 pos = result.find('//')
101 # Take everything after any remaining './'.
102 if result.startswith('./'):
105 # Take everything before any remaining '/.'.
106 if result.endswith('/.'):
109 if result == '' or result.endswith('/..'):
116 if p.startswith('/'):
118 if p.find('/../') != -1:
120 if p.startswith('../'):
122 if p.endswith('/..'):
127 def _clean_up_extract_path(p):
128 result = p.lstrip('/')
131 elif _risky_path(result):
137 # These tags are currently conceptually private to Metadata, and they
138 # must be unique, and must *never* be changed.
141 _rec_tag_common = 2 # times, owner, group, type, perms, etc.
142 _rec_tag_symlink_target = 3
143 _rec_tag_posix1e_acl = 4 # getfacl(1), setfacl(1), etc.
144 _rec_tag_nfsv4_acl = 5 # intended to supplant posix1e acls?
145 _rec_tag_linux_attr = 6 # lsattr(1) chattr(1)
146 _rec_tag_linux_xattr = 7 # getfattr(1) setfattr(1)
149 class MetadataAcquisitionError(Exception):
150 # Thrown when unable to extract any given bit of metadata from a path.
154 class MetadataApplicationError(Exception):
155 # Thrown when unable to apply any given bit of metadata to a path.
160 # Metadata is stored as a sequence of tagged binary records. Each
161 # record will have some subset of add, encode, load, create, and
162 # apply methods, i.e. _add_foo...
166 # Timestamps are (sec, ns), relative to 1970-01-01 00:00:00, ns
167 # must be non-negative and < 10**9.
169 def _add_common(self, path, st):
170 self.mode = st.st_mode
173 self.rdev = st.st_rdev
174 self.atime = st.st_atime
175 self.mtime = st.st_mtime
176 self.ctime = st.st_ctime
177 self.owner = pwd.getpwuid(st.st_uid)[0]
178 self.group = grp.getgrgid(st.st_gid)[0]
180 def _encode_common(self):
181 atime = self.atime.to_timespec()
182 mtime = self.mtime.to_timespec()
183 ctime = self.ctime.to_timespec()
184 result = vint.pack('VVsVsVvVvVvV',
199 def _load_common_rec(self, port):
200 data = vint.read_bvec(port)
212 ctime_ns) = vint.unpack('VVsVsVvVvVvV', data)
213 self.atime = FSTime.from_timespec((self.atime, atime_ns))
214 self.mtime = FSTime.from_timespec((self.mtime, mtime_ns))
215 self.ctime = FSTime.from_timespec((self.ctime, ctime_ns))
217 def _create_via_common_rec(self, path, create_symlinks=True):
218 # If the path already exists and is a dir, try rmdir.
219 # If the path already exists and is anything else, try unlink.
224 if e.errno != errno.ENOENT:
227 if stat.S_ISDIR(st.st_mode):
231 if e.errno == errno.ENOTEMPTY:
232 msg = 'refusing to overwrite non-empty dir' + path
238 if stat.S_ISREG(self.mode):
239 os.mknod(path, 0600 | stat.S_IFREG)
240 elif stat.S_ISDIR(self.mode):
242 elif stat.S_ISCHR(self.mode):
243 os.mknod(path, 0600 | stat.S_IFCHR, self.rdev)
244 elif stat.S_ISBLK(self.mode):
245 os.mknod(path, 0600 | stat.S_IFBLK, self.rdev)
246 elif stat.S_ISFIFO(self.mode):
247 os.mknod(path, 0600 | stat.S_IFIFO)
248 elif stat.S_ISLNK(self.mode):
249 if(self.symlink_target and create_symlinks):
250 os.symlink(self.symlink_target, path)
251 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
252 # Otherwise, do nothing.
254 def _apply_common_rec(self, path, restore_numeric_ids=False):
255 # FIXME: S_ISDOOR, S_IFMPB, S_IFCMP, S_IFNWK, ... see stat(2).
256 if stat.S_ISLNK(self.mode):
257 lutime(path, (self.atime, self.mtime))
259 utime(path, (self.atime, self.mtime))
260 if stat.S_ISREG(self.mode) \
261 | stat.S_ISDIR(self.mode) \
262 | stat.S_ISCHR(self.mode) \
263 | stat.S_ISBLK(self.mode) \
264 | stat.S_ISLNK(self.mode) \
265 | stat.S_ISFIFO(self.mode):
269 elif not stat.S_ISLNK(self.mode):
272 # Don't try to restore owner unless we're root, and even
273 # if asked, don't try to restore the owner or group if
274 # it doesn't exist in the system db.
277 if not restore_numeric_ids:
278 if os.geteuid() == 0:
280 uid = pwd.getpwnam(self.owner)[2]
283 log('bup: ignoring unknown owner %s for "%s"\n'
284 % (self.owner, path))
286 uid = -1 # Not root; assume we can't change owner.
288 gid = grp.getgrnam(self.group)[2]
291 log('bup: ignoring unknown group %s for "%s"\n'
292 % (self.group, path))
293 os.lchown(path, uid, gid)
296 os.lchmod(path, stat.S_IMODE(self.mode))
297 elif not stat.S_ISLNK(self.mode):
298 os.chmod(path, stat.S_IMODE(self.mode))
303 def _encode_path(self):
305 return vint.pack('s', self.path)
309 def _load_path_rec(self, port):
310 self.path = vint.unpack('s', vint.read_bvec(port))[0]
315 def _add_symlink_target(self, path, st):
316 if(stat.S_ISLNK(st.st_mode)):
317 self.symlink_target = os.readlink(path)
319 def _encode_symlink_target(self):
320 return self.symlink_target
322 def _load_symlink_target_rec(self, port):
323 self.symlink_target = vint.read_bvec(port)
326 ## POSIX1e ACL records
328 # Recorded as a list:
329 # [txt_id_acl, num_id_acl]
330 # or, if a directory:
331 # [txt_id_acl, num_id_acl, txt_id_default_acl, num_id_default_acl]
332 # The numeric/text distinction only matters when reading/restoring
334 def _add_posix1e_acl(self, path, st):
335 if not stat.S_ISLNK(st.st_mode):
337 if posix1e.has_extended(path):
338 acl = posix1e.ACL(file=path)
339 self.posix1e_acl = [acl, acl] # txt and num are the same
340 if stat.S_ISDIR(st.st_mode):
341 acl = posix1e.ACL(filedef=path)
342 self.posix1e_acl.extend([acl, acl])
343 except EnvironmentError, e:
344 if e.errno != errno.EOPNOTSUPP:
347 def _encode_posix1e_acl(self):
348 # Encode as two strings (w/default ACL string possibly empty).
350 acls = self.posix1e_acl
351 txt_flags = posix1e.TEXT_ABBREVIATE
352 num_flags = posix1e.TEXT_ABBREVIATE | posix1e.TEXT_NUMERIC_IDS
353 acl_reps = [acls[0].to_any_text('', '\n', txt_flags),
354 acls[1].to_any_text('', '\n', num_flags)]
358 acl_reps.append(acls[2].to_any_text('', '\n', txt_flags))
359 acl_reps.append(acls[3].to_any_text('', '\n', num_flags))
360 return vint.pack('ssss',
361 acl_reps[0], acl_reps[1], acl_reps[2], acl_reps[3])
365 def _load_posix1e_acl_rec(self, port):
366 data = vint.read_bvec(port)
367 acl_reps = vint.unpack('ssss', data)
368 if(acl_reps[2] == ''):
369 acl_reps = acl_reps[:2]
370 self.posix1e_acl = [posix1e.ACL(x) for x in acl_reps]
372 def _apply_posix1e_acl_rec(self, path, restore_numeric_ids=False):
373 if(self.posix1e_acl):
374 acls = self.posix1e_acl
376 if restore_numeric_ids:
377 acls[3].applyto(path, posix1e.ACL_TYPE_DEFAULT)
379 acls[2].applyto(path, posix1e.ACL_TYPE_DEFAULT)
380 if restore_numeric_ids:
381 acls[1].applyto(path, posix1e.ACL_TYPE_ACCESS)
383 acls[0].applyto(path, posix1e.ACL_TYPE_ACCESS)
386 ## Linux attributes (lsattr(1), chattr(1))
388 def _add_linux_attr(self, path, st):
389 if stat.S_ISREG(st.st_mode) or stat.S_ISDIR(st.st_mode):
390 attr = get_linux_file_attr(path)
392 self.linux_attr = get_linux_file_attr(path)
394 def _encode_linux_attr(self):
396 return vint.pack('V', self.linux_attr)
400 def _load_linux_attr_rec(self, port):
401 data = vint.read_bvec(port)
402 self.linux_attr = vint.unpack('V', data)[0]
404 def _apply_linux_attr_rec(self, path, restore_numeric_ids=False):
406 set_linux_file_attr(path, self.linux_attr)
409 ## Linux extended attributes (getfattr(1), setfattr(1))
411 def _add_linux_xattr(self, path, st):
413 self.linux_xattr = xattr.get_all(path, nofollow=True)
414 except EnvironmentError, e:
415 if e.errno != errno.EOPNOTSUPP:
418 def _encode_linux_xattr(self):
420 result = vint.pack('V', len(self.linux_xattr))
421 for name, value in self.linux_xattr:
422 result += vint.pack('ss', name, value)
427 def _load_linux_xattr_rec(self, file):
428 data = vint.read_bvec(file)
429 memfile = StringIO(data)
431 for i in range(vint.read_vuint(memfile)):
432 key = vint.read_bvec(memfile)
433 value = vint.read_bvec(memfile)
434 result.append((key, value))
435 self.linux_xattr = result
437 def _apply_linux_xattr_rec(self, path, restore_numeric_ids=False):
438 if(self.linux_xattr):
439 for k, v in self.linux_xattr:
440 xattr.set(path, k, v, nofollow=True)
445 self.symlink_target = None
446 self.linux_attr = None
447 self.linux_xattr = None
448 self.posix1e_acl = None
449 self.posix1e_acl_default = None
451 def write(self, port, include_path=True):
452 records = [(_rec_tag_path, self._encode_path())] if include_path else []
453 records.extend([(_rec_tag_common, self._encode_common()),
454 (_rec_tag_symlink_target, self._encode_symlink_target()),
455 (_rec_tag_posix1e_acl, self._encode_posix1e_acl()),
456 (_rec_tag_linux_attr, self._encode_linux_attr()),
457 (_rec_tag_linux_xattr, self._encode_linux_xattr())])
458 for tag, data in records:
460 vint.write_vuint(port, tag)
461 vint.write_bvec(port, data)
462 vint.write_vuint(port, _rec_tag_end)
466 # This method should either: return a valid Metadata object;
467 # throw EOFError if there was nothing at all to read; throw an
468 # Exception if a valid object could not be read completely.
469 tag = vint.read_vuint(port)
470 try: # From here on, EOF is an error.
472 while(True): # only exit is error (exception) or _rec_tag_end
473 if tag == _rec_tag_path:
474 result._load_path_rec(port)
475 elif tag == _rec_tag_common:
476 result._load_common_rec(port)
477 elif tag == _rec_tag_symlink_target:
478 result._load_symlink_target_rec(port)
479 elif tag == _rec_tag_posix1e_acl:
480 result._load_posix1e_acl(port)
481 elif tag ==_rec_tag_nfsv4_acl:
482 result._load_nfsv4_acl_rec(port)
483 elif tag == _rec_tag_linux_attr:
484 result._load_linux_attr_rec(port)
485 elif tag == _rec_tag_linux_xattr:
486 result._load_linux_xattr_rec(port)
487 elif tag == _rec_tag_end:
489 else: # unknown record
491 tag = vint.read_vuint(port)
493 raise Exception("EOF while reading Metadata")
496 return stat.S_ISDIR(self.mode)
498 def create_path(self, path, create_symlinks=True):
499 self._create_via_common_rec(path, create_symlinks=create_symlinks)
501 def apply_to_path(self, path=None, restore_numeric_ids=False):
502 # apply metadata to path -- file must exist
506 raise Exception('Metadata.apply_to_path() called with no path');
507 num_ids = restore_numeric_ids
508 try: # Later we may want to push this down and make it finer grained.
509 self._apply_common_rec(path, restore_numeric_ids=num_ids)
510 self._apply_posix1e_acl_rec(path, restore_numeric_ids=num_ids)
511 self._apply_linux_attr_rec(path, restore_numeric_ids=num_ids)
512 self._apply_linux_xattr_rec(path, restore_numeric_ids=num_ids)
514 raise MetadataApplicationError(e)
517 def from_path(path, archive_path=None, save_symlinks=True):
519 result.path = archive_path
521 try: # Later we may want to push this down and make it finer grained.
522 result._add_common(path, st)
524 result._add_symlink_target(path, st)
525 result._add_posix1e_acl(path, st)
526 result._add_linux_attr(path, st)
527 result._add_linux_xattr(path, st)
529 raise MetadataAcquisitionError(e)
533 def save_tree(output_file, paths,
539 # Issue top-level rewrite warnings.
541 safe_path = _clean_up_path_for_archive(path)
542 if(safe_path != path):
543 log('bup: archiving "%s" as "%s"\n' % (path, safe_path))
545 start_dir = os.getcwd()
547 for (p, st) in recursive_dirlist(paths, xdev=xdev):
548 dirlist_dir = os.getcwd()
550 safe_path = _clean_up_path_for_archive(p)
552 m = from_path(p, archive_path=safe_path,
553 save_symlinks=save_symlinks)
554 except MetadataAcquisitionError, e:
558 print >> sys.stderr, m.path
559 m.write(output_file, include_path=write_paths)
560 os.chdir(dirlist_dir)
565 def _set_up_path(meta, create_symlinks=True):
566 # Allow directories to exist as a special case -- might have
567 # been created by an earlier longer path.
569 mkdirp(meta.path, 0700)
571 parent = os.path.dirname(meta.path)
574 meta.create_path(meta.path, create_symlinks=create_symlinks)
577 class _ArchiveIterator:
580 return Metadata.read(self._file)
582 raise StopIteration()
587 def __init__(self, file):
591 def display_archive(file):
592 for meta in _ArchiveIterator(file):
594 print meta.path # FIXME
599 def start_extract(file, create_symlinks=True):
600 for meta in _ArchiveIterator(file):
602 print >> sys.stderr, meta.path
603 xpath = _clean_up_extract_path(meta.path)
605 add_error(Exception('skipping risky path "%s"' % meta.path))
608 _set_up_path(meta, create_symlinks=create_symlinks)
611 def finish_extract(file, restore_numeric_ids=False):
613 for meta in _ArchiveIterator(file):
614 xpath = _clean_up_extract_path(meta.path)
616 add_error(Exception('skipping risky path "%s"' % dir.path))
618 if os.path.isdir(meta.path):
619 all_dirs.append(meta)
622 print >> sys.stderr, meta.path
624 meta.apply_to_path(path=xpath,
625 restore_numeric_ids=restore_numeric_ids)
626 except MetadataApplicationError, e:
629 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
631 # Don't need to check xpath -- won't be in all_dirs if not OK.
632 xpath = _clean_up_extract_path(dir.path)
634 print >> sys.stderr, dir.path
636 dir.apply_to_path(path=xpath,
637 restore_numeric_ids=restore_numeric_ids)
638 except MetadataApplicationError, e:
642 def extract(file, restore_numeric_ids=False, create_symlinks=True):
643 # For now, just store all the directories and handle them last,
646 for meta in _ArchiveIterator(file):
647 xpath = _clean_up_extract_path(meta.path)
649 add_error(Exception('skipping risky path "%s"' % meta.path))
653 print >> sys.stderr, '+', meta.path
654 _set_up_path(meta, create_symlinks=create_symlinks)
655 if os.path.isdir(meta.path):
656 all_dirs.append(meta)
659 print >> sys.stderr, '=', meta.path
661 meta.apply_to_path(restore_numeric_ids=restore_numeric_ids)
662 except MetadataApplicationError, e:
664 all_dirs.sort(key = lambda x : len(x.path), reverse=True)
666 # Don't need to check xpath -- won't be in all_dirs if not OK.
667 xpath = _clean_up_extract_path(meta.path)
669 print >> sys.stderr, '=', meta.path
670 # Shouldn't have to check for risky paths here (omitted above).
672 dir.apply_to_path(path=dir.path,
673 restore_numeric_ids=restore_numeric_ids)
674 except MetadataApplicationError, e: