About this document

This is the user manual for the bakonf project, version 0.7.0; the homepage is at http://github.com/iustin/bakonf/. You can also get new versions of this document there.

Introduction

Making backup is an important aspect of system administration. The techniques of backing up data are explained in any good document about system administration, and they won't be explained here again.

bakonf comes into play into a particular part of the backups: minimal backups of the system's configuration and state, for the case where a standard (full) backup is unfeasible.

The basic idea is that on a standard installation of a Unix-like system you have a lot of data which can be very easily restored from the original media, thus there is no point in archiving it. For example, after a fresh install of a RedHat Linux 8.0, you have ~4.5GB of space used. However, only a very small part of this amount is holding important information, the other part being binaries, libraries and other kind of data which will never modify in normal usage. Only the configuration files are changing (of course, also the user data is changing, but we are talking about an empty system).

If we classify the files existing on a Unix system, we have:

configuration data

These are the target of bakonf; they are usually small text files, partly coming from the system installation, maybe edited by the administrator, partly created by him. Size is (on the workstation I write this) around 15MB (true in 2002, still true in 2018).

binaries, libraries, other system files

These are mostly read-only; in a package based distribution, they came from the packages and are replaced when the package is upgraded. In classical systems, they come from the install archives. Size is (in our hypothetical rh8.0 full install) ~4.5GB.

system and user data

These are emails, web pages, documents, etc. - this is important data, and needs to be backed up regularly. They also don't come from the installation media, and are not touched by the system. Size is undetermined, but is guaranteed to be exactly the amount of free space on the system :). This is the main target of a regular backup.

system maintenance data

These are the files created and managed by the system, usually from the configuration files and other external variables. Examples: /var/lib/logrotate.status, /var/lib/slocate/slocate.db. These are not all critical files, some are needed to be included in a backup only for analysis purposes, others should not be included in backups (e.g. if you reinstall your system or restore from backup, some files will have for sure other contents, generated from the new installation).

non-file data

This is data that still lives on storage, but not directly as files. For example, partition tables, logical volume configuration.

From all these, only the configuration data and the system/user data are absolutely required to recreate the system. Depending on the setup, the non-file data might be required as well, but (in case of partitions, for example) is not required exactly as it was before. The binaries can come from the installation source. The system maintenance data will be recreated by the system. And since the difference in size between the configuration and user data is so great in a typical system, that I believe it deserves another backup method than a regular, full-backup - which, as said before, definitely has its place.

Quick start

  • run bakonf with the -L0 option to archive all config files and create its database:

    root@test:~$ bakonf -L0

    If everything went well, bakonf has created an archive under /var/lib/bakonf/archives named after your host. Look into that directory to find it. If any errors have occurred, bakonf will tell you:

    user@test:~$ bakonf -L0
Error: cannot read '/nfs/README': 'Permission denied'. Not archived.
Warning: '/sbin/lsusb -vv' exited with status 1.
Warning: '/sbin/sfdisk -d /dev/hda' exited with status 1.
[user@test user]$

  • run daily (or more often) bakonf with the -L1 option to archive only the changed files since the previous step. This archive should be much smaller. It will be easy, after encryption, to email it:

    root@test:~ bakonf -L1

  • every week, go back to the first step.

Generated archive

bakonf's output is a tar archive (optionally compressed) that contains some metadata and (if both enabled) two sections: file backup and command output.

Filename Description Created when
README A file which contains information about the archive: when it was generated, with which options and on what host Always
unarchived_files.lst A file which contains details about which files couldn't be backed up; this can happen when bakonf is not run as root, or for example when it scans NFS directories When file system backup has been performed
commands_with_errors.lst A file which contains details about which commands have exited with non-zero status. Their output is still stored in the archive, though. When command execution has been performed
filesystem/ Files backed up are stored under this path. When file system backup has been performed
commands/ Outputs from the command execution are stored under this path. When command execution has been performed

File system backup

Bakonf supports a simple incremental backup, by the use of two level: 0 and 1. In level 0, it archives all of the specified configuration files and registers the state of those configuration files in a database (called state database), of type Berkeley DB. In level 1, it archives only the files modified since the last level 0 backup; it does this by comparing the state database with the current state of the file system.

File types and states

directories

Directories won't be archived if they don't contain files to be backed up. On the other hand, for each file to be backed up, bakonf will also backup (non-recursively) its parent directories (except root) so that you have the user, group, modification time and permissions of each directory. For example, if /usr/local/etc/myconfig has been selected for archiving, bakonf will actually archive this list of items: /usr, /usr/local, /usr/local/etc, /usr/local/etc/myconfig.

regular files

Regular files will be archived by bakonf if they aren't excluded by the noscan configuration directive. In case this is a partial backup (as opposed to a full backup), bakonf will make the following tests:

  • does the size of file saved in database differ from the current file size? if so, include;
  • do the saved hashes (md5 and sha1) differ from the current hashes? if so, include;
  • otherwise, file will not be included in the incremental backup.

bakonf doesn't follow symbolic links; it treats a symbolic link like a configuration file (its configuration data resides in its name and its target). For an incremental backup, the tests made by bakonf are, in order:

  • link target must be equal, or the file is backed up
  • user and group ownership must be equal, or the file is backed up
  • permission bits must be equal, or the file is backed up
  • the file is not backed up
block devices, character devices, fifos, sockets

bakonf always selects these to backed up. Of course, some of them won't be backed up by tar, but regarding bakonf, it will select those for backup.

changed file types

In case the file type has changed between the level 0 and level 1 backup, bakonf will always include this file.

Command execution/output

This section allows you to save more information about a system than is available in the file-system. The current implementation allows you to store output of shell commands. Suggestions about other items are welcome.

Examples

Partition table

One of the most important items about a system (that is not stored in a file) is partition table about your disks, in the eventuality that you have a data error in partition table. The command to back this up varies, for example sfdisk, fdisk, etc.

Further, one could consider the volume group configuration similar to the partition table, although this usually has backups.

Device list

Having the device list is and their hardware configuration is useful in order to have a quick overview if you want to clone the configuration from one system to another (to see correspondence between config files and hardware config). Examples of scanning the configuration are lspci -vv, lsusb -vv, pciconf -lv, etc.

Installed package list

While the installed package list can be recovered from a package database, sometimes this database is binary, or much more verbose. So the output of, for example, dpkg -l or rpm -ql is much easier to read and feed back to apt-get or rpm.

What can I use bakonf for?

Potential use cases:

  • Configuration rollback. Since the archives are small, you can keep many versions, but unlike in differential backup, here one archive contains all the needed data.
  • Configuration cloning. You can take a bakonf-generated archive from one system to another and 'clone' as much of the settings as you want.
  • Quick restore of a server in case of catastrophic hard-disk failure. Just reinstall the OS and put the config files back.

Requirements

To use bakonf, you must have the following:

  • a Unix-like operating system
  • Python version 2.4 or higher
  • the ElementTree library for Python, for parsing the configuration file(s)
  • the pybsddb library, if not bundled with your Python distribution

Configuration

Note: Older bakonf versions (before 0.5) had an entirely different config file, and version 0.5 had a different schema for the configuration files. If you upgraded, be sure to forward you changes to the new config files.

bakonf uses a main configuration file (by default /etc/bakonf/bakonf.yaml), which does some standard settings and tells bakonf what other files to include. These additional files are usually located in /etc/bakonf/sources and tell bakonf how to handle some special cases.

Configuration language

The configuration file is written in YAML, and should represent an object (map) with the following keys:

configs

(list) tells bakonf to also parse any files which match the shell patterns passed. These are files which modify bakonf's own behaviour, and are usually located in /etc/bakonf/sources. These are not directories to be backed up! (Although, if modified and included by the filesystem/include entries, they will be). Note that the sub-files can contain only filesystem and commands keys.

include

(list) Tells bakonf to add any file or directory which matches shell pattern given by the contents of the tag to its include list. These can be files or directories. Bakonf will descend directories, but will not follow symbolic links! The symbolic links are considered configuration items also, so they will be backed up themselves.

exclude

(list) Tells bakonf to ignore any file or directory which matches the contents of the elements (interpreted as a regular expression) from the archive; it won't even open or stat these files. Note that the patterns given are prefix matches, so for example an expression /path/to will match (and thus exclude) a file /path/to/file.

commands

(list) Contains declarations about command output to be included in the archive. Each element in the list is a dictionary with the following keys:

cmd:

Defines the command line to be executed (as a shell command).

dest:

Defines the destination archive member to be used for the output under the commands subtree in the created archive. If this key is not given, then the destination is takes as the command line with slashes replaced by underscores. For example, the element:

    cmd: cat /proc/version
    path: proc/version

will create a file in the archive with the name commands/proc/version which will contain the /proc/version file. A shortened entry of:

    cmd: /usr/bin/uptime

will create a file commands/usr_bin_uptime.

database

This elements contains the filename of the state database.

maxsize

This element denotes the maximum size of files to be backed up.

The order of precedence for include/exclude is:

  • bakonf will start scanning all items defined with 'include'.
  • if at any point in the file system scan, the current file matches any one of noscan regexps, the scan will ignore it. For directories, it will prevent recursion into them, which means ignoring all the files they contain, so please be careful about it.

Using these, you can select where you want bakonf to look for files for archiving. The default config file includes /etc, /usr/etc, /usr/local/etc and some others (look in /etc/bakonf after installing).

Example main configuration file (the file included in the distribution):

database: /var/lib/bakonf/state.db
configs:
- /etc/bakonf/sources/*.yaml
include:
- /etc
- /usr/etc
- /usr/local/etc
- /var/lib/alternatives
exclude:
- /etc/ssl/private

Example configuration file for saving system information:

commands:
- cmd: lsblk
- cmd: sfdisk -d /dev/sda
  path: partitions/sda
- cmd: dpkg -l
  path: dpkg.list

File list

bakonf is composed of:

/etc/bakonf/bakonf.yml

Main configuration file.

/etc/bakonf/conf.d/*.yml

Configuration files for special cases (config files outside of etc dirs).

/usr/bin/bakonf

Main program

/etc/cron.d/bakonf

Cron file, by default it does not run bakonf, you must un-comment a line to run it.

/var/lib/bakonf/archives

Default directory for configuration file archives.

Note: You must decide yourself what to do with the configuration archives after bakonf creates them!

Using bakonf

Backup phase

For details about the actual parameters to bakonf, see the man page.

To use bakonf, choose to either:

  • run it manually, when you want, either always with -L0 or with a combination of multiple backup levels
  • use the provided cron script to run it automatically

In any case, you have to do something with the generated archives. Write the to tape, CD, other machine, but don't just ignore them, you defeat the purpose of bakonf.

Restore phase

Configuration rollback

In this case, just make sure you have the bakonf-generated archive near the date in the past you are interested in. If so:

  1. if your system uses packages/ports, compare the actual package list with the one recorded by bakonf when it created the archive

  2. install/remove software as needed

  3. copy the configuration files for the services you want to rollback over the current files

Complete system restoration

If you had a catastrophic system failure, you should follow these steps:

  1. Reinstall the operating system on a clean machine. Use the given information in the /commands directory in the archive to achieve an as close as possible configuration as the old system (e.g. partition layout, packages installed, etc.)
  2. Copy all the files in the archive in the file system, overwriting the defaults from the packages.

Glossary

statefile

The database that is used for saving file states for the multi-level backup method