btrfs-backup

This project supports incremental backups for btrfs using snapshots and send/receive between filesystems. Think of it as a basic version of Time Machine.

Backups can be stored locally and/or remotely (e.g. via SSH). Multi-target setups are supported as well as dealing with transmission failures (e.g. due to network outage).

Its main goals are to be reliable and functional while maintaining user-friendliness. It should be easy to get started in just a few minutes without detailled knowledge on how btrfs send/receive works. However, you should have a basic understanding of snapshots and subvolumes.

btrfs-backup has almost no dependencies and hence is well suited for many kinds of setups with only minimal maintenance effort.

Originally, it started as a fork of a project with the same name, written by Chris Lawrence. Since then, most of the code has been refactored and many new features were added before this repository has been transferred to me. Many thanks to Chris for his work. The old code base has been tagged with legacy. If, for any reason, you want to continue using it and miss the new features, you can check that out.

Features

• Initial creation of full backups
• Incremental backups on subsequent runs
• Different backup storage engines:
  • Local storage
  • Remote storage via SSH
  • Custom storage: Alternatively, the output of btrfs send may be piped to a custom shell command.
• Multi-target support with tracking of which snapshots are missing at each location.
• Retransmission on errors (e.g. due to network outage).
• Simple and configurable retention policy for local and remote snapshots
• Optionally, create snapshots without transferring them anywhere and vice versa.
• Creation of backups without root privileges, if some special conditions are met
• Detailled logging output with configurable log level

Installation

Requirements

• Python 3.3 or later
• Appropriate btrfs-progs; typically you'll want at least 3.12 with Linux 3.12/3.13
• (optional) OpenSSH's ssh command - needed for remote backup pulling and pushing via SSH
• (optional) sshfs - only needed for pulling via SSH
• (optional) pv command for displaying progress during backups

Install via PIP

The easiest way to get up and running with the latest stable version is via PIP. If pip3 is missing on your system and you run a Debian-based distribution, simply install it via:

$ sudo apt-get install python3-pip python3-wheel

Then, you can fetch the latest version of btrfs-backup:

$ sudo pip3 install btrfs_backup

Pre-built packages

There are pre-built packages available for the following distributions.

• Arch Linux (thanks to XenGi for maintaining this)

Manual installation

Alternatively, clone this git repository

$ git clone https://github.com/efficiosoft/btrfs-backup
$ cd btrfs-backup
$ git checkout tags/v0.3.0  # optionally checkout a specific version
$ sudo ./setup.py install

Sample usage

Not every feature of btrfs-backup is explained in this README, since there is a detailled and descriptive help included with the command.

However, there are some sections about the general concepts and different sample usages to get started as quick as possible.

For reference, a copy of the output of btrfs-backup --help is attached below.

As root:

$ btrfs-backup /home /backup

This will create a read-only snapshot of /home in /home/snapshot/YYMMDD-HHMMSS, and then send it to /backup/YYMMDD-HHMMSS. On future runs, it will take a new read-only snapshot and send the difference between the previous snapshot and the new one.

Note: Both source and destination need to be on btrfs filesystems. Additionally, the source has to be either the root or any other subvolume, but not just an ordinary directory because snapshots can only be created of subvolumes.

For the backup to be sensible, source and destination shouldn't be the same filesystem. Otherwise you could just snapshot and save the hassle.

You can backup multiple subvolumes to multiple subfolders or subvolumes at the destination. For example, you might want to backup both / and /home. The main caveat is you'll want to put the backups in separate folders on the destination drive to avoid confusion.

$ btrfs-backup / /backup/root
$ btrfs-backup /home /backup/home

If you really want to store backups of different subvolumes at the same location, you have to specify a prefix using the -p/--snapshot-prefix option. Without that, btrfs-backup can't distinguish between your different backup chains and will mix them up. Using the example from above, it could look like the following:

$ btrfs-backup --snapshot-prefix root / /backup
$ btrfs-backup --snapshot-prefix home /home /backup

You can specify -N/--num-snapshots <num> to only keep the latest <num> number of snapshots on the source filesystem. -n/--num-backups <num> does the same thing for the backup location.

Remote backups

Backing up to a remote server via SSH is as easy as:

$ btrfs-backup /home ssh://server/mnt/backups

btrfs-backup doesn't need to be installed on the remote side for this to work. It is recommended to set up public key authentication to eliminate the need for entering passwords. A full description of how to customize the ssh call can be found in the help text.

Pulling backups from a remote side is now supported as well! Simply use the ssh:// scheme as source.

You could even do something like:

$ btrfs-backup ssh://source_server/home ssh://dest_server/mnt/backups

to pull backups from source_server and store them at dest_server. This might be used if you can't install btrfs-backup on either remote host for any reason. But keep in mind that this procedure will generate double traffic (from source_server to you and from you to dest_server).

Okay, just one last example, because I really like that one:

$ btrfs-backup ssh://source_server/home \
               /mnt/backups \
               ssh://dest_server/mnt/backups

Can you guess what it does? Well, it does the same as the command before + an extra sending to your local /mnt/backups folder. Please note that btrfs-backup is not smart enough to prevent the same data from being pulled from source_server twice. But that wouldn't be easy to implement with the current design.

Help text

This is the output of btrfs-backup --help. Taking a look at it, you should get a good insight in what it can and can't do (yet).

Cooming at the release.

Configuration files

By default, btrfs-backup doesn't read any configuration file. However, you can create one or more and specify them at the command line:

$ btrfs-backup @path/to/backup_home.conf

Any argument prefixed by a @ is treated as file name of a configuration file.

The format of these files is simple. On every line, there may be one flag, option or argument you would normally specify at the command line. Valid configuration files might look like the following.

backup_home.conf:

# This is a comment and thus ignored, as well as blank lines.

# Include another configuration file here.
@global.conf

        # Indentation has no effect.
        -p home

# This is the source.
/home

# Back up to both local and remote storage.
/mnt/backups/home
ssh://server/mnt/btrfs_storage/backups/home

global.conf:

# This file gets included by the other one.
--quiet

--num-snapshots 1
--num-backups 3

A more detailled explanation about the format can be found in the help text.

What are locks?

btrfs-backup uses so called "locks" to keep track of failed snapshot transfers. There is a file called .outstanding_transfers created in the snapshot folder. This file is in JSON format and thus human-readable, if necessary.

Locking works as follows:

1. When a snapshot transfer is started, an entry is created in that file, telling that a snapshot transfer of a specific snapshot to a specific destination has begun. We call this entry a lock.
2. If the snapshot transfer used another snapshot as parent, that one gets an entry as well, but no lock, just the note that it's a parent for something that failed to transfer.
3. When the transfer
   1. finishes without errors, the locks for the snapshot (and its parent) are removed.
   2. aborts (e.g. due to network outage or a full disk), the locks are kept.

Now, there are multiple options for dealing with those failed transfers.

When you run btrfs-backup the next time, it finds the corrupt snapshot at the destination and deletes it, together with the corresponding lock and parent notes. Afterwards, the way is free for a new transfer. You may also use --no-snapshot to only do the transfers without creating new snapshots.

There is a special flag called --locked-dests available. If supplied, it automatically adds all destinations which locks exist for as if they were specified at the command line. You might do something like:

$ btrfs-backup --no-snapshot --locked-dests /home

to retry all failed backup transfers of snapshots of /home. This could be executed periodically because it just does nothing if there are no locks.

Snapshots for which locks or parent notes exist are excluded from the retention policy and won't be purged until the locks are removed either automatically (because the partially transferred snapshots could be deleted from the destination) or manually (see below).

As a last resort for removing locks for transfers you don't want to retry anymore, there is a flag called --remove-locks. Use it with caution and only if you can assure that there are no corrupt snapshots at the destinations you apply the flag on.

$ btrfs-backup --no-snapshot --no-transfer --remove-locks /home ssh://nas/backups

will remove all locks for the destination ssh://nas/backups from /home/snapshot/.outstanding_transfers. Of course, using --locked-dests instead of specifying the destination explicitly is possible as well.

Backing up regularly

Note that there is no locking included with btrfs-backup. If you back up too often (i.e. more quickly than it takes the first call to finish, which can take several minutes, hours or even days on a filesystem with lots of files), you might end up with a new backup starting while an old one is still in progress.

You can workaround the lack of locking using the flock(1) command, as suggested at https://github.com/efficiosoft/btrfs-backup/issues/4.

With anacron on Debian, you could simply add a file /etc/cron.daily/local-backup:

#!/bin/sh
flock -n /tmp/btrfs-backup-home.lock \
    ionice -c 3 btrfs-backup --quiet --num-snapshots 1 --num-backups 3 \
                /home /backup/home

You may omit the -n flag if you want to wait rather than fail in case a backup is already running.

More or less frequent backups could be made using other cron.* scripts.

Restoring a snapshot

If necessary, you can restore a whole snapshot by using e.g.

$ mkdir /home/snapshot
$ btrfs send /backup/YYMMDD-HHMMSS | btrfs receive /home/snapshot

Then you need to take the read-only snapshot and turn it back into a root filesystem:

$ cp -aR --reflink /home/snapshot/YYMMDD-HHMMSS /home

You might instead have some luck taking the restored snapshot and turning it into a read-write snapshot, and then re-pivoting your mounted subvolume to the read-write snapshot.

Alternative workflow

An alternative structure is to keep all subvolumes in the root directory

/
/active
/active/root
/active/home
/inactive
/snapshot/root/YYMMDD-HHMMSS
/snapshot/home/YYMMDD-HHMMSS

and have corresponding entries in /etc/fstab to mount the subvolumes from /active/*. One benefit of this approach is that restoring a snapshot can be done entirely with btrfs tools:

$ btrfs send /backup/root/YYMMDD-HHMMSS | btrfs receive /snapshot/home
$ btrfs send /backup/home/YYMMDD-HHMMSS | btrfs receive /snapshot/root
$ mv /active/root /inactive
$ mv /active/home /inactive
$ btrfs subvolume snapshot /snapshot/root/YYMMDD-HHMMSS /active/root
$ btrfs subvolume snapshot /snapshot/home/YYMMDD-HHMMSS /active/home

The snapshots from btrfs-backup may be placed in /snapshots/ by using the --snapshot-folder option.

Issues and Contribution

As in every piece of software, there likely are bugs. When you find one, please open an issue on GitHub. If you do so, please include the output with debug log level (-v debug) and provide steps to reproduce the problem. Thank you!

If you want to contribute, that's great! You can create issues (even for feature requests), send pull requests or contact me via email at r.schindler@efficiosoft.com.

Copyright