path: root/rdiff-backup/rdiff_backup/regress.py

# Copyright 2002, 2005 Ben Escoto
#
# This file is part of rdiff-backup.
#
# rdiff-backup is free software; you can redistribute it and/or modify
# under the terms of the GNU General Public License as published by the
# Free Software Foundation; either version 2 of the License, or (at your
# option) any later version.
#
# rdiff-backup is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with rdiff-backup; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
# USA

"""Code for reverting the rdiff-backup directory to prev state

This module is used after an aborted session, and the rdiff-backup
destination directory may be in-between states.  In this situation we
need to bring back the directory as it was after the last successful
backup.  The basic strategy is to restore all the attributes from the
metadata file (which we assume is intact) and delete the extra
increments.  For regular files we examine the mirror file and use the
increment file to get the old data if the mirror file is out of date.

Currently this does not recover hard links.  This may make the
regressed directory take up more disk space, but hard links can still
be recovered.

"""

from __future__ import generators
import signal, errno, re, os
import Globals, restore, log, rorpiter, TempFile, metadata, rpath, C, \
	   Time, backup, robust, longname

# regress_time should be set to the time we want to regress back to
# (usually the time of the last successful backup)
regress_time = None

# This should be set to the latest unsuccessful backup time
unsuccessful_backup_time = None


class RegressException(Exception):
	"""Raised on any exception in regress process"""
	pass


def Regress(mirror_rp):
	"""Bring mirror and inc directory back to regress_to_time

	Also affects the rdiff-backup-data directory, so Globals.rbdir
	should be set.  Regress should only work one step at a time
	(i.e. don't "regress" through two separate backup sets.  This
	function should be run locally to the rdiff-backup-data directory.

	"""
	inc_rpath = Globals.rbdir.append_path("increments")
	assert mirror_rp.index == () and inc_rpath.index == ()
	assert mirror_rp.isdir() and inc_rpath.isdir()
	assert mirror_rp.conn is inc_rpath.conn is Globals.local_connection
	manager, former_current_mirror_rp = set_regress_time()
	set_restore_times()
	regress_rbdir(manager)
	ITR = rorpiter.IterTreeReducer(RegressITRB, [])
	for rf in iterate_meta_rfs(mirror_rp, inc_rpath): ITR(rf.index, rf)
	ITR.Finish()
	if former_current_mirror_rp:
		C.sync() # Sync first, since we are marking dest dir as good now
		former_current_mirror_rp.delete()

def set_regress_time():
	"""Set global regress_time to previous sucessful backup

	If there are two current_mirror increments, then the last one
	corresponds to a backup session that failed.

	"""
	global regress_time, unsuccessful_backup_time
	manager = metadata.SetManager()
	curmir_incs = manager.sorted_prefix_inclist('current_mirror')
	assert len(curmir_incs) == 2, \
		   "Found %s current_mirror flags, expected 2" % len(curmir_incs)
	mirror_rp_to_delete = curmir_incs[0]
	regress_time = curmir_incs[1].getinctime()
	unsuccessful_backup_time = mirror_rp_to_delete.getinctime()
	log.Log("Regressing to " + Time.timetopretty(regress_time), 4)
	return manager, mirror_rp_to_delete

def set_restore_times():
	"""Set _rest_time and _mirror_time in the restore module

	_rest_time (restore time) corresponds to the last successful
	backup time.  _mirror_time is the unsuccessful backup time.

	"""
	restore.MirrorStruct._mirror_time = unsuccessful_backup_time
	restore.MirrorStruct._rest_time = regress_time

def regress_rbdir(meta_manager):
	"""Delete the increments in the rdiff-backup-data directory

	Returns the former current mirror rp so we can delete it later.
	All of the other rp's should be deleted before the actual regress,
	to clear up disk space the rest of the procedure may need.

	Also, in case the previous session failed while diffing the
	metadata file, either recreate the mirror_metadata snapshot, or
	delete the extra regress_time diff.

	"""
	has_meta_diff, has_meta_snap = 0, 0
	for old_rp in meta_manager.timerpmap[regress_time]:
		if old_rp.getincbase_str() == 'mirror_metadata':
			if old_rp.getinctype() == 'snapshot': has_meta_snap = 1
			else:
				assert old_rp.getinctype() == 'diff', old_rp
				has_meta_diff = 1
	if has_meta_diff and not has_meta_snap: recreate_meta(meta_manager)

	for new_rp in meta_manager.timerpmap[unsuccessful_backup_time]:
		if new_rp.getincbase_str() != 'current_mirror':
			log.Log("Deleting old diff at " + new_rp.path, 5)
			new_rp.delete()
	for rp in meta_manager.timerpmap[regress_time]:
		if (rp.getincbase_str() == 'mirror_metadata' and
			rp.getinctype() == 'diff'):
			rp.delete()
			break

def recreate_meta(meta_manager):
	"""Make regress_time mirror_metadata snapshot by patching

	We write to a tempfile first.  Otherwise, in case of a crash, it
	would seem we would have an intact snapshot and partial diff, not
	the reverse.

	"""
	temprp = [TempFile.new_in_dir(Globals.rbdir)]
	def callback(rp): temprp[0] = rp
	writer = metadata.MetadataFile(temprp[0], 'w', check_path = 0, callback = callback)
	for rorp in meta_manager.get_meta_at_time(regress_time, None):
		writer.write_object(rorp)
	writer.close()

	finalrp = Globals.rbdir.append("mirror_metadata.%s.snapshot.gz" %
								   Time.timetostring(regress_time))
	assert not finalrp.lstat(), finalrp
	rpath.rename(temprp[0], finalrp)
	if Globals.fsync_directories: Globals.rbdir.fsync()

def iterate_raw_rfs(mirror_rp, inc_rp):
	"""Iterate all RegressFile objects in mirror/inc directory

	Also changes permissions of unreadable files.  We don't have to
	change them back later because regress will do that for us.

	"""
	root_rf = RegressFile(mirror_rp, inc_rp, restore.get_inclist(inc_rp))
	def helper(rf):
		mirror_rp = rf.mirror_rp
		if Globals.process_uid != 0:
			if mirror_rp.isreg() and not mirror_rp.readable():
				mirror_rp.chmod(0400 | mirror_rp.getperms())
			elif mirror_rp.isdir() and not mirror_rp.hasfullperms():
				mirror_rp.chmod(0700 | mirror_rp.getperms())
		yield rf
		if rf.mirror_rp.isdir() or rf.inc_rp.isdir():
			for sub_rf in rf.yield_sub_rfs():
				for sub_sub_rf in helper(sub_rf):
					yield sub_sub_rf
	return helper(root_rf)

def yield_metadata():
	"""Iterate rorps from metadata file, if any are available"""
	metadata.SetManager()
	metadata_iter = metadata.ManagerObj.GetAtTime(regress_time)
	if metadata_iter: return metadata_iter
	log.Log.FatalError("No metadata for time %s (%s) found,\ncannot regress"
					   % (Time.timetopretty(regress_time), regress_time))

def iterate_meta_rfs(mirror_rp, inc_rp):
	"""Yield RegressFile objects with extra metadata information added

	Each RegressFile will have an extra object variable .metadata_rorp
	which will contain the metadata attributes of the mirror file at
	regress_time.

	"""
	raw_rfs = iterate_raw_rfs(mirror_rp, inc_rp)
	collated = rorpiter.Collate2Iters(raw_rfs, yield_metadata())
	for raw_rf, metadata_rorp in collated:
		raw_rf = longname.update_regressfile(raw_rf, metadata_rorp, mirror_rp)
		if not raw_rf:
			log.Log("Warning, metadata file has entry for %s,\n"
					"but there are no associated files." %
					(metadata_rorp.get_indexpath(),), 2)
			continue
		raw_rf.set_metadata_rorp(metadata_rorp)
		yield raw_rf


class RegressFile(restore.RestoreFile):
	"""Like RestoreFile but with metadata

	Hold mirror_rp and related incs, but also put metadata info for
	the mirror file at regress time in self.metadata_rorp.
	self.metadata_rorp is not set in this class.

	"""
	def __init__(self, mirror_rp, inc_rp, inc_list):
		restore.RestoreFile.__init__(self, mirror_rp, inc_rp, inc_list)
		self.set_regress_inc()
		
	def set_metadata_rorp(self, metadata_rorp):
		"""Set self.metadata_rorp, creating empty if given None"""
		if metadata_rorp: self.metadata_rorp = metadata_rorp
		else: self.metadata_rorp = rpath.RORPath(self.index)

	def isdir(self):
		"""Return true if regress needs before/after processing"""
		return ((self.metadata_rorp and self.metadata_rorp.isdir()) or
				(self.mirror_rp and self.mirror_rp.isdir()))

	def set_regress_inc(self):
		"""Set self.regress_inc to increment to be removed (or None)"""
		newer_incs = self.get_newer_incs()
		assert len(newer_incs) <= 1, "Too many recent increments"
		if newer_incs: self.regress_inc = newer_incs[0] # first is mirror_rp
		else: self.regress_inc = None


class RegressITRB(rorpiter.ITRBranch):
	"""Turn back state of dest directory (use with IterTreeReducer)

	The arguments to the ITR will be RegressFiles.  There are two main
	assumptions this procedure makes (besides those mentioned above):

	1.  The mirror_rp and the metadata_rorp equal_loose correctly iff
	    they contain the same data.  If this is the case, then the inc
	    file is unnecessary and we can delete it.

	2.  If the don't match, then applying the inc file will
	    successfully get us back to the previous state.

	Since the metadata file is required, the two above really only
	matter for regular files.

	"""
	def __init__(self):
		"""Just initialize some variables to None"""
		self.rf = None # will hold RegressFile applying to a directory

	def can_fast_process(self, index, rf):
		"""True if none of the rps is a directory"""
		return not rf.mirror_rp.isdir() and not rf.metadata_rorp.isdir()

	def fast_process(self, index, rf):
		"""Process when nothing is a directory"""
		if not rf.metadata_rorp.equal_loose(rf.mirror_rp):
			log.Log("Regressing file %s" %
					(rf.metadata_rorp.get_indexpath()), 5)
			if rf.metadata_rorp.isreg(): self.restore_orig_regfile(rf)
			else:
				if rf.mirror_rp.lstat(): rf.mirror_rp.delete()
				if rf.metadata_rorp.isspecial():
					robust.check_common_error(None, rpath.copy_with_attribs,
											  (rf.metadata_rorp, rf.mirror_rp))
				else: rpath.copy_with_attribs(rf.metadata_rorp, rf.mirror_rp)
		if rf.regress_inc:
			log.Log("Deleting increment " + rf.regress_inc.path, 5)
			rf.regress_inc.delete()

	def restore_orig_regfile(self, rf):
		"""Restore original regular file

		This is the trickiest case for avoiding information loss,
		because we don't want to delete the increment before the
		mirror is fully written.

		"""
		assert rf.metadata_rorp.isreg()
		if rf.mirror_rp.isreg():
			tf = TempFile.new(rf.mirror_rp)
			tf.write_from_fileobj(rf.get_restore_fp())
			tf.fsync_with_dir() # make sure tf fully written before move
			rpath.copy_attribs(rf.metadata_rorp, tf)
			rpath.rename(tf, rf.mirror_rp) # move is atomic
		else:
			if rf.mirror_rp.lstat(): rf.mirror_rp.delete()
			rf.mirror_rp.write_from_fileobj(rf.get_restore_fp())
			rpath.copy_attribs(rf.metadata_rorp, rf.mirror_rp)
		if Globals.fsync_directories:
			rf.mirror_rp.get_parent_rp().fsync() # force move before inc delete

	def start_process(self, index, rf):
		"""Start processing directory"""
		if rf.metadata_rorp.isdir():
			# make sure mirror is a readable dir
			if not rf.mirror_rp.isdir():
				if rf.mirror_rp.lstat(): rf.mirror_rp.delete()
				rf.mirror_rp.mkdir()
			if not rf.mirror_rp.hasfullperms(): rf.mirror_rp.chmod(0700)
		self.rf = rf

	def end_process(self):
		"""Finish processing a directory"""
		rf = self.rf
		if rf.metadata_rorp.isdir():
			if rf.mirror_rp.isdir():
				rf.mirror_rp.setdata()
				if not rf.metadata_rorp.equal_loose(rf.mirror_rp):
					log.Log("Regressing attributes of " + rf.mirror_rp.path, 5)
					rpath.copy_attribs(rf.metadata_rorp, rf.mirror_rp)
			else:
				rf.mirror_rp.delete()
				log.Log("Regressing file " + rf.mirror_rp.path, 5)
				rpath.copy_with_attribs(rf.metadata_rorp, rf.mirror_rp)
		else: # replacing a dir with some other kind of file
			assert rf.mirror_rp.isdir()
			log.Log("Replacing directory " + rf.mirror_rp.path, 5)
			if rf.metadata_rorp.isreg(): self.restore_orig_regfile(rf)
			else:
				rf.mirror_rp.delete()
				rpath.copy_with_attribs(rf.metadata_rorp, rf.mirror_rp)
		if rf.regress_inc:
			log.Log("Deleting increment " + rf.regress_inc.path, 5)
			rf.regress_inc.delete()


def check_pids(curmir_incs):
	"""Check PIDs in curmir markers to make sure rdiff-backup not running"""
	pid_re = re.compile("^PID\s*([0-9]+)", re.I | re.M)
	def extract_pid(curmir_rp):
		"""Return process ID from a current mirror marker, if any"""
		match = pid_re.search(curmir_rp.get_data())
		if not match: return None
		else: return int(match.group(1))

	def pid_running(pid):
		"""True if we know if process with pid is currently running"""
		try: os.kill(pid, 0)
		except OSError, exc:
			if exc[0] == errno.ESRCH: return 0
			else: log.Log("Warning: unable to check if PID %d still running" % (pid,), 2)
		except AttributeError:
			assert os.name == 'nt'
			import win32api, win32con, pywintypes
			process = None
			try:
				process = win32api.OpenProcess(win32con.PROCESS_ALL_ACCESS, 
											0, pid)
			except pywintypes.error, error:
				if error[0] == 87: return 0
				else:
					msg = "Warning: unable to check if PID %d still running"
					log.Log(msg % pid, 2)
			if process:
				win32api.CloseHandle(process)
				return 1
			return 0
		return 1

	for curmir_rp in curmir_incs:
		assert Globals.local_connection is curmir_rp.conn
		pid = extract_pid(curmir_rp)
		if pid is not None and pid_running(pid):
			log.Log.FatalError(
"""It appears that a previous rdiff-backup session with process
id %d is still running.  If two different rdiff-backup processes write
the same repository simultaneously, data corruption will probably
result.  To proceed with regress anyway, rerun rdiff-backup with the
--force option.""" % (pid,))