# Copyright 2002 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 """Store and retrieve metadata in destination directory The plan is to store metadata information for all files in the destination directory in a special metadata file. There are two reasons for this: 1) The filesystem of the mirror directory may not be able to handle types of metadata that the source filesystem can. For instance, rdiff-backup may not have root access on the destination side, so cannot set uid/gid. Or the source side may have ACLs and the destination side doesn't. Hopefully every file system can store binary data. Storing metadata separately allows us to back up anything (ok, maybe strange filenames are still a problem). 2) Metadata can be more quickly read from a file than it can by traversing the mirror directory over and over again. In many cases most of rdiff-backup's time is spent compaing metadata (like file size and modtime), trying to find differences. Reading this data sequentially from a file is significantly less taxing than listing directories and statting files all over the mirror directory. The metadata is stored in a text file, which is a bunch of records concatenated together. Each record has the format: File ... Where the lines are separated by newlines. See the code below for the field names and values. """ from __future__ import generators import re, gzip, os, binascii, codecs import log, Globals, rpath, Time, robust, increment, static, rorpiter class ParsingError(Exception): """This is raised when bad or unparsable data is received""" pass def carbonfile2string(cfile): """Convert CarbonFile data to a string suitable for storing.""" if not cfile: return "None" retvalparts = [] retvalparts.append('creator:%s' % binascii.hexlify(cfile['creator'])) retvalparts.append('type:%s' % binascii.hexlify(cfile['type'])) retvalparts.append('location:%d,%d' % cfile['location']) retvalparts.append('flags:%d' % cfile['flags']) try: retvalparts.append('createDate:%d' % cfile['createDate']) except KeyError: log.Log("Writing pre-1.1.6 style metadata, without creation date", 9) return '|'.join(retvalparts) def string2carbonfile(data): """Re-constitute CarbonFile data from a string stored by carbonfile2string.""" retval = {} for component in data.split('|'): key, value = component.split(':') if key == 'creator': retval['creator'] = binascii.unhexlify(value) elif key == 'type': retval['type'] = binascii.unhexlify(value) elif key == 'location': a, b = value.split(',') retval['location'] = (int(a), int(b)) elif key == 'flags': retval['flags'] = int(value) elif key == 'createDate': retval['createDate'] = int(value) return retval def RORP2Record(rorpath): """From RORPath, return text record of file's metadata""" str_list = ["File %s\n" % quote_path(rorpath.get_indexpath())] # Store file type, e.g. "dev", "reg", or "sym", and type-specific data type = rorpath.gettype() if type is None: type = "None" str_list.append(" Type %s\n" % type) if type == "reg": str_list.append(" Size %s\n" % rorpath.getsize()) # If there is a resource fork, save it. if rorpath.has_resource_fork(): if not rorpath.get_resource_fork(): rf = "None" else: rf = binascii.hexlify(rorpath.get_resource_fork()) str_list.append(" ResourceFork %s\n" % (rf,)) # If there is Carbon data, save it. if rorpath.has_carbonfile(): cfile = carbonfile2string(rorpath.get_carbonfile()) str_list.append(" CarbonFile %s\n" % (cfile,)) # If file is hardlinked, add that information if Globals.preserve_hardlinks != 0: numlinks = rorpath.getnumlinks() if numlinks > 1: str_list.append(" NumHardLinks %s\n" % numlinks) str_list.append(" Inode %s\n" % rorpath.getinode()) str_list.append(" DeviceLoc %s\n" % rorpath.getdevloc()) # Save any hashes, if available if rorpath.has_sha1(): str_list.append(' SHA1Digest %s\n' % rorpath.get_sha1()) elif type == "None": return "".join(str_list) elif type == "dir" or type == "sock" or type == "fifo": pass elif type == "sym": str_list.append(" SymData %s\n" % quote_path(rorpath.readlink())) elif type == "dev": major, minor = rorpath.getdevnums() if rorpath.isblkdev(): devchar = "b" else: assert rorpath.ischardev() devchar = "c" str_list.append(" DeviceNum %s %s %s\n" % (devchar, major, minor)) # Store time information if type != 'sym' and type != 'dev': str_list.append(" ModTime %s\n" % rorpath.getmtime()) # Add user, group, and permission information uid, gid = rorpath.getuidgid() str_list.append(" Uid %s\n" % uid) str_list.append(" Uname %s\n" % (rorpath.getuname() or ":")) str_list.append(" Gid %s\n" % gid) str_list.append(" Gname %s\n" % (rorpath.getgname() or ":")) str_list.append(" Permissions %s\n" % rorpath.getperms()) # Add long filename information if rorpath.has_alt_mirror_name(): str_list.append(" AlternateMirrorName %s\n" % (rorpath.get_alt_mirror_name(),)) elif rorpath.has_alt_inc_name(): str_list.append(" AlternateIncrementName %s\n" % (rorpath.get_alt_inc_name(),)) return "".join(str_list) line_parsing_regexp = re.compile("^ *([A-Za-z0-9]+) (.+)$", re.M) def Record2RORP(record_string): """Given record_string, return RORPath For speed reasons, write the RORPath data dictionary directly instead of calling rorpath functions. Profiling has shown this to be a time critical function. """ data_dict = {} for field, data in line_parsing_regexp.findall(record_string): if field == "File": index = quoted_filename_to_index(data) elif field == "Type": if data == "None": data_dict['type'] = None else: data_dict['type'] = data elif field == "Size": data_dict['size'] = long(data) elif field == "ResourceFork": if data == "None": data_dict['resourcefork'] = "" else: data_dict['resourcefork'] = binascii.unhexlify(data) elif field == "CarbonFile": if data == "None": data_dict['carbonfile'] = None else: data_dict['carbonfile'] = string2carbonfile(data) elif field == "SHA1Digest": data_dict['sha1'] = data elif field == "NumHardLinks": data_dict['nlink'] = int(data) elif field == "Inode": data_dict['inode'] = long(data) elif field == "DeviceLoc": data_dict['devloc'] = long(data) elif field == "SymData": data_dict['linkname'] = unquote_path(data) elif field == "DeviceNum": devchar, major_str, minor_str = data.split(" ") data_dict['devnums'] = (devchar, int(major_str), int(minor_str)) elif field == "ModTime": data_dict['mtime'] = long(data) elif field == "Uid": data_dict['uid'] = int(data) elif field == "Gid": data_dict['gid'] = int(data) elif field == "Uname": if data == ":" or data == 'None': data_dict['uname'] = None else: data_dict['uname'] = data elif field == "Gname": if data == ':' or data == 'None': data_dict['gname'] = None else: data_dict['gname'] = data elif field == "Permissions": data_dict['perms'] = int(data) elif field == "AlternateMirrorName": data_dict['mirrorname'] = data elif field == "AlternateIncrementName": data_dict['incname'] = data else: log.Log("Unknown field in line '%s %s'" % (field, data), 2) return rpath.RORPath(index, data_dict) chars_to_quote = re.compile("\\n|\\\\") def quote_path(path_string): """Return quoted verson of path_string Because newlines are used to separate fields in a record, they are replaced with \n. Backslashes become \\ and everything else is left the way it is. """ def replacement_func(match_obj): """This is called on the match obj of any char that needs quoting""" char = match_obj.group(0) if char == "\n": return "\\n" elif char == "\\": return "\\\\" assert 0, "Bad char %s needs quoting" % char return chars_to_quote.sub(replacement_func, path_string) def unquote_path(quoted_string): """Reverse what was done by quote_path""" def replacement_func(match_obj): """Unquote match obj of two character sequence""" two_chars = match_obj.group(0) if two_chars == "\\n": return "\n" elif two_chars == "\\\\": return "\\" log.Log("Warning, unknown quoted sequence %s found" % two_chars, 2) return two_chars return re.sub("\\\\n|\\\\\\\\", replacement_func, quoted_string) def quoted_filename_to_index(quoted_filename): """Return tuple index given quoted filename""" if quoted_filename == '.': return () else: return tuple(unquote_path(quoted_filename).split('/')) class FlatExtractor: """Controls iterating objects from flat file""" # Set this in subclass. record_boundary_regexp should match # beginning of next record. The first group should start at the # beginning of the record. The second group should contain the # (possibly quoted) filename. record_boundary_regexp = None # Set in subclass to function that converts text record to object record_to_object = None def __init__(self, fileobj): self.fileobj = fileobj # holds file object we are reading from self.buf = "" # holds the next part of the file self.at_end = 0 # True if we are at the end of the file self.blocksize = 32 * 1024 def get_next_pos(self): """Return position of next record in buffer, or end pos if none""" while 1: m = self.record_boundary_regexp.search(self.buf, 1) if m: return m.start(1) else: # add next block to the buffer, loop again newbuf = self.fileobj.read(self.blocksize) if not newbuf: self.at_end = 1 return len(self.buf) else: self.buf += newbuf def iterate(self): """Return iterator that yields all objects with records""" for record in self.iterate_records(): try: yield self.record_to_object(record) except (ParsingError, ValueError), e: if self.at_end: break # Ignore whitespace/bad records at end log.Log("Error parsing flat file: %s" % (e,), 2) def iterate_records(self): """Yield all text records in order""" while 1: next_pos = self.get_next_pos() if self.at_end: if next_pos: yield self.buf[:next_pos] break yield self.buf[:next_pos] self.buf = self.buf[next_pos:] assert not self.fileobj.close() def skip_to_index(self, index): """Scan through the file, set buffer to beginning of index record Here we make sure that the buffer always ends in a newline, so we will not be splitting lines in half. """ assert not self.buf or self.buf.endswith("\n") while 1: self.buf = self.fileobj.read(self.blocksize) self.buf += self.fileobj.readline() if not self.buf: self.at_end = 1 return while 1: m = self.record_boundary_regexp.search(self.buf) if not m: break cur_index = self.filename_to_index(m.group(2)) if cur_index >= index: self.buf = self.buf[m.start(1):] return else: self.buf = self.buf[m.end(1):] def iterate_starting_with(self, index): """Iterate objects whose index starts with given index""" self.skip_to_index(index) if self.at_end: return while 1: next_pos = self.get_next_pos() try: obj = self.record_to_object(self.buf[:next_pos]) except (ParsingError, ValueError), e: log.Log("Error parsing metadata file: %s" % (e,), 2) else: if obj.index[:len(index)] != index: break yield obj if self.at_end: break self.buf = self.buf[next_pos:] assert not self.fileobj.close() def filename_to_index(self, filename): """Translate filename, possibly quoted, into an index tuple The filename is the first group matched by regexp_boundary_regexp. """ assert 0 # subclass class RorpExtractor(FlatExtractor): """Iterate rorps from metadata file""" record_boundary_regexp = re.compile("(?:\\n|^)(File (.*?))\\n") record_to_object = staticmethod(Record2RORP) filename_to_index = staticmethod(quoted_filename_to_index) class FlatFile: """Manage a flat (probably text) file containing info on various files This is used for metadata information, and possibly EAs and ACLs. The main read interface is as an iterator. The storage format is a flat, probably compressed file, so random access is not recommended. """ rp, fileobj, mode = None, None, None _buffering_on = 1 # Buffering may be useful because gzip writes are slow _record_buffer, _max_buffer_size = None, 100 _extractor = FlatExtractor # Override to class that iterates objects _object_to_record = None # Set to function converting object to record _prefix = None # Set to required prefix def __init__(self, rp_base, mode, check_path = 1, compress = 1, callback = None): """Open rp (or rp+'.gz') for reading ('r') or writing ('w') If callback is available, it will be called on the rp upon closing (because the rp may not be known in advance). """ self.mode = mode self.callback = callback self._record_buffer = [] if check_path: assert (rp_base.isincfile() and rp_base.getincbase_str() == self._prefix), rp_base compress = 1 if mode == 'r': self.rp = rp_base self.fileobj = rpath.MaybeUnicode(self.rp.open("rb", compress)) else: assert mode == 'w' if compress and check_path and not rp_base.isinccompressed(): def callback(rp): self.rp = rp self.fileobj = rpath.MaybeUnicode(rpath.MaybeGzip(rp_base, callback)) else: self.rp = rp_base assert not self.rp.lstat(), self.rp self.fileobj = rpath.MaybeUnicode(self.rp.open("wb", compress = compress)) def write_record(self, record): """Write a (text) record into the file""" if self._buffering_on: self._record_buffer.append(record) if len(self._record_buffer) >= self._max_buffer_size: self.fileobj.write("".join(self._record_buffer)) self._record_buffer = [] else: self.fileobj.write(record) def write_object(self, object): """Convert one object to record and write to file""" self.write_record(self._object_to_record(object)) def get_objects(self, restrict_index = None): """Return iterator of objects records from file rp""" if not restrict_index: return self._extractor(self.fileobj).iterate() extractor = self._extractor(self.fileobj) return extractor.iterate_starting_with(restrict_index) def get_records(self): """Return iterator of text records""" return self._extractor(self.fileobj).iterate_records() def close(self): """Close file, for when any writing is done""" assert self.fileobj, "File already closed" if self._buffering_on and self._record_buffer: self.fileobj.write("".join(self._record_buffer)) self._record_buffer = [] result = self.fileobj.close() self.fileobj = None self.rp.fsync_with_dir() self.rp.setdata() if self.callback: self.callback(self.rp) return result class MetadataFile(FlatFile): """Store/retrieve metadata from mirror_metadata as rorps""" _prefix = "mirror_metadata" _extractor = RorpExtractor _object_to_record = staticmethod(RORP2Record) class CombinedWriter: """Used for simultaneously writting metadata, eas, and acls""" def __init__(self, metawriter, eawriter, aclwriter, winaclwriter): self.metawriter = metawriter self.eawriter, self.aclwriter, self.winaclwriter = \ eawriter, aclwriter, winaclwriter # these can be None def write_object(self, rorp): """Write information in rorp to all the writers""" self.metawriter.write_object(rorp) if self.eawriter and not rorp.get_ea().empty(): self.eawriter.write_object(rorp.get_ea()) if self.aclwriter and not rorp.get_acl().is_basic(): self.aclwriter.write_object(rorp.get_acl()) if self.winaclwriter: self.winaclwriter.write_object(rorp.get_win_acl()) def close(self): self.metawriter.close() if self.eawriter: self.eawriter.close() if self.aclwriter: self.aclwriter.close() if self.winaclwriter: self.winaclwriter.close() class Manager: """Read/Combine/Write metadata files by time""" meta_prefix = 'mirror_metadata' acl_prefix = 'access_control_lists' ea_prefix = 'extended_attributes' wacl_prefix = 'win_access_control_lists' def __init__(self): """Set listing of rdiff-backup-data dir""" self.rplist = [] self.timerpmap, self.prefixmap = {}, {} for filename in Globals.rbdir.listdir(): rp = Globals.rbdir.append(filename) if rp.isincfile(): self.add_incrp(rp) def add_incrp(self, rp): """Add rp to list of inc rps in the rbdir""" assert rp.isincfile(), rp self.rplist.append(rp) time = rp.getinctime() if self.timerpmap.has_key(time): self.timerpmap[time].append(rp) else: self.timerpmap[time] = [rp] incbase = rp.getincbase_str() if self.prefixmap.has_key(incbase): self.prefixmap[incbase].append(rp) else: self.prefixmap[incbase] = [rp] def _iter_helper(self, prefix, flatfileclass, time, restrict_index): """Used below to find the right kind of file by time""" if not self.timerpmap.has_key(time): return None for rp in self.timerpmap[time]: if rp.getincbase_str() == prefix: return flatfileclass(rp, 'r').get_objects(restrict_index) return None def get_meta_at_time(self, time, restrict_index): """Return iter of metadata rorps at given time (or None)""" return self._iter_helper(self.meta_prefix, MetadataFile, time, restrict_index) def get_eas_at_time(self, time, restrict_index): """Return Extended Attributes iter at given time (or None)""" return self._iter_helper(self.ea_prefix, eas_acls.ExtendedAttributesFile, time, restrict_index) def get_acls_at_time(self, time, restrict_index): """Return ACLs iter at given time from recordfile (or None)""" return self._iter_helper(self.acl_prefix, eas_acls.AccessControlListFile, time, restrict_index) def get_win_acls_at_time(self, time, restrict_index): """Return WACLs iter at given time from recordfile (or None)""" return self._iter_helper(self.wacl_prefix, win_acls.WinAccessControlListFile, time, restrict_index) def GetAtTime(self, time, restrict_index = None): """Return combined metadata iter with ea/acl info if necessary""" cur_iter = self.get_meta_at_time(time, restrict_index) if not cur_iter: log.Log("Warning, could not find mirror_metadata file.\n" "Metadata will be read from filesystem instead.", 2) return None if Globals.acls_active: acl_iter = self.get_acls_at_time(time, restrict_index) if not acl_iter: log.Log("Warning: Access Control List file not found", 2) acl_iter = iter([]) cur_iter = eas_acls.join_acl_iter(cur_iter, acl_iter) if Globals.eas_active: ea_iter = self.get_eas_at_time(time, restrict_index) if not ea_iter: log.Log("Warning: Extended Attributes file not found", 2) ea_iter = iter([]) cur_iter = eas_acls.join_ea_iter(cur_iter, ea_iter) if Globals.win_acls_active: wacl_iter = self.get_win_acls_at_time(time, restrict_index) if not wacl_iter: log.Log("Warning: Windows Access Control List file not" " found.", 2) wacl_iter = iter([]) cur_iter = win_acls.join_wacl_iter(cur_iter, wacl_iter) return cur_iter def _writer_helper(self, prefix, flatfileclass, typestr, time): """Used in the get_xx_writer functions, returns a writer class""" if time is None: timestr = Time.curtimestr else: timestr = Time.timetostring(time) filename = '%s.%s.%s' % (prefix, timestr, typestr) rp = Globals.rbdir.append(filename) assert not rp.lstat(), "File %s already exists!" % (rp.path,) assert rp.isincfile() return flatfileclass(rp, 'w', callback = self.add_incrp) def get_meta_writer(self, typestr, time): """Return MetadataFile object opened for writing at given time""" return self._writer_helper(self.meta_prefix, MetadataFile, typestr, time) def get_ea_writer(self, typestr, time): """Return ExtendedAttributesFile opened for writing""" return self._writer_helper(self.ea_prefix, eas_acls.ExtendedAttributesFile, typestr, time) def get_acl_writer(self, typestr, time): """Return AccessControlListFile opened for writing""" return self._writer_helper(self.acl_prefix, eas_acls.AccessControlListFile, typestr, time) def get_win_acl_writer(self, typestr, time): """Return WinAccessControlListFile opened for writing""" return self._writer_helper(self.wacl_prefix, win_acls.WinAccessControlListFile, typestr, time) def GetWriter(self, typestr = 'snapshot', time = None): """Get a writer object that can write meta and possibly acls/eas""" metawriter = self.get_meta_writer(typestr, time) if not Globals.eas_active and not Globals.acls_active and \ not Globals.win_acls_active: return metawriter # no need for a CombinedWriter if Globals.eas_active: ea_writer = self.get_ea_writer(typestr, time) else: ea_writer = None if Globals.acls_active: acl_writer = self.get_acl_writer(typestr, time) else: acl_writer = None if Globals.win_acls_active: win_acl_writer = \ self.get_win_acl_writer(typestr, time) else: win_acl_writer = None return CombinedWriter(metawriter, ea_writer, acl_writer, win_acl_writer) class PatchDiffMan(Manager): """Contains functions for patching and diffing metadata To save space, we can record a full list of only the most recent metadata, using the normal rdiff-backup reverse increment strategy. Instead of using librsync to compute diffs, though, we use our own technique so that the diff files are still hand-editable. A mirror_metadata diff has the same format as a mirror_metadata snapshot. If the record for an index is missing from the diff, it indicates no change from the original. If it is present it replaces the mirror_metadata entry, unless it has Type None, which indicates the record should be deleted from the original. """ max_diff_chain = 9 # After this many diffs, make a new snapshot def get_diffiter(self, new_iter, old_iter): """Iterate meta diffs of new_iter -> old_iter""" for new_rorp, old_rorp in rorpiter.Collate2Iters(new_iter, old_iter): if not old_rorp: yield rpath.RORPath(new_rorp.index) elif not new_rorp or new_rorp.data != old_rorp.data: # exact compare here, can't use == on rorps yield old_rorp def sorted_prefix_inclist(self, prefix, min_time = 0): """Return reverse sorted (by time) list of incs with given prefix""" if not self.prefixmap.has_key(prefix): return [] sortlist = [(rp.getinctime(), rp) for rp in self.prefixmap[prefix]] sortlist.sort() sortlist.reverse() return [rp for (time, rp) in sortlist if time >= min_time] def check_needs_diff(self): """Check if we should diff, returns (new, old) rps, or (None, None)""" inclist = self.sorted_prefix_inclist('mirror_metadata') assert len(inclist) >= 1 if len(inclist) == 1: return (None, None) newrp, oldrp = inclist[:2] assert newrp.getinctype() == oldrp.getinctype() == 'snapshot' chainlen = 1 for rp in inclist[2:]: if rp.getinctype() != 'diff': break chainlen += 1 if chainlen >= self.max_diff_chain: return (None, None) return (newrp, oldrp) def ConvertMetaToDiff(self): """Replace a mirror snapshot with a diff if it's appropriate""" newrp, oldrp = self.check_needs_diff() if not newrp: return log.Log("Writing mirror_metadata diff", 6) diff_writer = self.get_meta_writer('diff', oldrp.getinctime()) new_iter = MetadataFile(newrp, 'r').get_objects() old_iter = MetadataFile(oldrp, 'r').get_objects() for diff_rorp in self.get_diffiter(new_iter, old_iter): diff_writer.write_object(diff_rorp) diff_writer.close() # includes sync oldrp.delete() def get_meta_at_time(self, time, restrict_index): """Get metadata rorp iter, possibly by patching with diffs""" meta_iters = [MetadataFile(rp, 'r').get_objects(restrict_index) for rp in self.relevant_meta_incs(time)] if not meta_iters: return None if len(meta_iters) == 1: return meta_iters[0] return self.iterate_patched_meta(meta_iters) def relevant_meta_incs(self, time): """Return list [snapshotrp, diffrps ...] time sorted""" inclist = self.sorted_prefix_inclist('mirror_metadata', min_time=time) if not inclist: return inclist assert inclist[-1].getinctime() == time, inclist[-1] for i in range(len(inclist)-1, -1, -1): if inclist[i].getinctype() == 'snapshot': return inclist[i:] assert 0, "Inclist %s contains no snapshots" % (inclist,) def iterate_patched_meta(self, meta_iter_list): """Return an iter of metadata rorps by combining the given iters The iters should be given as a list/tuple in reverse chronological order. The earliest rorp in each iter will supercede all the later ones. """ for meta_tuple in rorpiter.CollateIterators(*meta_iter_list): for i in range(len(meta_tuple)-1, -1, -1): if meta_tuple[i]: if meta_tuple[i].lstat(): yield meta_tuple[i] break # move to next index else: assert 0, "No valid rorps" ManagerObj = None # Set this later to Manager instance def SetManager(): global ManagerObj ManagerObj = PatchDiffMan() return ManagerObj import eas_acls, win_acls # put at bottom to avoid python circularity bug