rsbackup

The systems you want to back up are called clients. The system that has the backup hard disk(s) attached to it is called the server, and it is on this system that the rsbackup program runs. The server can itself be a client.

Each client must have an SSH server and rsync installed. For Debian and Ubuntu systems it should be sufficient to install them as follows (if you don’t have them already):

apt-get install openssh-server rsync

The server requires rsync and an SSH client. Again for Debian:

apt-get install openssh-client rsync

For other platforms, you must consult their documentation, or install them from source:

• http://rsync.samba.org/
• http://www.openssh.com/

2.1.1 SSH Setup

The server’s root login needs to be able to SSH to each of the clients’ root logins without having to enter a password or confirm a key hash. You should consult the SSH documentation to set this up, but the general procedure, assuming you use RSA keys and OpenSSH, is as follows. If you are sufficiently familiar with SSH to do this without further documentation, skip to the next section.

On the server create an SSH key with:

sudo ssh-keygen

When asked for a passphrase, just hit return (but see below). Then copy ~root/.ssh/id_rsa.pub to each of the clients and append it to their ~root/.ssh/authorized_keys. At the same time, retrieve the clients’ host key hashes with:

ssh-keygen -l -f /etc/ssh/ssh_host_rsa_key.pub

As root on the server, ssh to each of the clients and verify their host keys hashes.

2.1.2 Installing rsbackup

To install rsbackup, go to www.greenend.org.uk/rjk/rsbackup and download the source code:

wget http://www.greenend.org.uk/rjk/rsbackup/rsbackup-1.1.tar.gz
tar xf rsbackup-1.1.tar.gz
cd rsbackup-1.1
make
sudo make install

(You will probably need to change the version number.)

On Debian systems you can use the pre-built .deb files:

wget http://www.greenend.org.uk/rjk/rsbackup/rsbackup_1.1_amd64.deb
sudo dpkg -i rsbackup_1.1_amd64.deb

(You will probably need to change the version number and perhaps the architecture.)

At this point it should be possible to read the man page, which contains reference information:

man rsbackup

2.1.3 Variations

• What If The Server Is Also A Client?

  If you want to backup the backup server itself then you don’t need to set up the server to be able to SSH to itself. See below for how to configure this.

• What If I’m Not The Superuser?

  rsbackup does not actually depend on being the superuser, although of course its functionality will be limited if it isn’t. However you could for instance use it to back up your home directory to your portable USB disk. The setup is the same except that you do it for your personal login rather than root and therefore don’t use sudo.

• What If I Don’t Like Empty Passphrases?

  In this case you will have to find some other way of making the server’s private SSH key available when backups run. This is outside the scope of this document.

rsbackup reads a configuration file from /etc/rsbackup/config. (Use the --config option to override this, if you prefer another location.) You will need to enter some global settings and then describe the backup clients.

2.2.1 Backup Storage

First you should define where backups will be stored. This guide will assume you use removable hard disks, but you can use permanently online backups too.

For each distinct backup device you need to define two things. The first is the mount point that the device will appear at. For example, if you have two backup disks and the mount points are /backup1 and /backup2 you would write the following:

store /backup1
store /backup2

The second is to define device names. Device names correspond to the contents of a single-line file called device-id in the root of the backup device. For instance, if you called your devices backup1 and backup2 you would write the following:

device backup1
device backup2

Of course, you must also create these files! For example:

echo backup1 > /backup1/device-id
echo backup2 > /backup2/device-id

rsbackup does not mind if the devices share a mount point (and only one is present at a time); any device may use any mount point as far as it’s concerned. You will probably find it more convenient to give them separate mount points though. If more than one device is mounted when you make a backup, backups will be made to all of them. You can use the --store option to select just one.

Although it would of course be convenient for users to be able to access backups of their files directly, it would also mean that they could go “back in time” past permission changes or deletions of private files belonging to one another. Therefore, the top directory of your backup devices should (usually) be owned by root and mode 0700 (i.e. -rwx------). By default, rsbackup will insist on this, although you can use the public option to change this behaviour.

chmod 700 /backup1 /backup2

Remember to update /etc/updatedb.conf to exclude your backup devices. Otherwise updatedb will spend ever-increasing amounts of time indexing your backups.

2.2.2 Global Parameters

Next you may want to define some global ageing parameters. These can be overridden for each volume you back up, so by defining global ones you are only setting defaults.

The first one you might want to set is the maximum age of the most recent backup. If any volume’s most recent backup is older than this many days then it will show up as red in the backup status report. The default is 3 days. To reduce it to (for instance) 1 day:

max-age 1

The second parameter to choose is the age at which backups are automatically pruned, i.e. deleted. The default is 366 days, ensuring that you will be able to “go back” up to a year. If you only wanted to go back a month you could reduce it as follows:

prune-age 31

Remember, these are defaults and can be overridden on a per-host or per-volume basis.

There are a few other global settings described in the man page. They will not be covered here.

2.2.3 Defining What To Back Up

The rest of the configuration file will define what to back up (and what to exclude).

There are two ways to organise your configuration file.

1. You can put all the configuration for all the hosts in the main configuration file.

2. You can put each host’s definitions in a file of its own and include them all. To do this put a line at the end of your configuration file as follows

   include /etc/rsbackup/hosts.d

   Then for each host create a file named after the host in this directory and use it to store the host’s configuration, as described below.

   The the Debian packaging of rsbackup uses this approach.

For each host to back up, you should write a host stanza. This will contain some host level settings and then a volume stanza for each part of the host’s filesystem back up.

Here is an example host stanza:

host sfere
  volume root /
  volume boot /boot
  volume home /home
    prune-age 366
    exclude /*/Desktop/Trash/*
    exclude /*/.local/share/Trash/*
    exclude /*/.mozilla/*/Cache/*
    exclude /*/.ccache
  volume var /var
    exclude /tmp/*

The meaning of this is as follows:

• The first line contains the name of the host. This would normally be its DNS name (see below for an example of where it is not).

• Each of the volume lines contains the name of a volume on the host and the path to that volume. By default, rsbackup will assume that each volume corresponds to a (mounted) filesystem and therefore not backup files from other filesystems.

• In this case there are four volumes. root and boot are quite simple: all the files in them will be backed up.

• home, however, is more complex. Firstly, it has a prune-age setting to ensure that it is kept for longer than the default lifetime. Secondly, it excludes various trash and cache directories.

  In the first three cases, it does this by backing up the directory but not its contents; they will be empty on the backup device. In the fourth case, it does not even backup the directory. Note that the exclusion patterns are rooted at the path to the volume - they are not absolute path names. (Consult the rsync documentation for --exclude for more information about these patterns.)

• /var is similar to home in that a temporary directory is excluded.

An important note: the indentation is not significant to rsbackup - only to the reader. Anything that comes after a host directive and before the next host or volume directive is considered part of that host. Similarly anything that comes after a volume directive and before the next host or volume directive is considered part of that volume.

This example shows how to back up a host where the name differs from the DNS name. The important part is the hostname directive:

host lith
  hostname chymax
  volume lith /Volumes/Lith
    exclude "Temporary Internet Files"
    exclude /RECYCLER
    exclude /pagefile.sys
    exclude "/Documents and Settings/*/Local Settings/Temp"
    exclude "/System Volume Information/_restore*"

What is actually going on here is that lith is really the Windows partition on chymax. The computer usually runs Unix, with the Windows partition mounted for convenience. So to get lith’s files, it is necessary to ssh to chymax.

This example shows how to back up without using SSH at all:

host araminta
  hostname localhost
  volume root /
  volume boot /boot
  volume home /home
    prune-age 366
    exclude /*/Desktop/Trash/*
    exclude /*/.local/share/Trash/*
    exclude /*/.mozilla/*/Cache/*
    exclude /*/.ccache
  volume var /var
    exclude /tmp/*
  volume news /var/spool/news
    prune-age 14

In this case, araminta is actually the backup server, so using SSH would mean SSHing to itself. The hostname localhost is special-cased to avoid using SSH at all.

Here is example of backing up a laptop:

host kakajou
  max-age 7
  volume users /Users
    prune-age 366
    exclude /*/.Trash/*
    exclude /*/Library/Caches
    exclude /*/Library/VirtualBox/Machines/*/Snapshots
    exclude /*/Library/VirtualBox/**.vdi
  volume local /local
    prune-age 366
  volume etc /etc
  volume sw /sw

This host is usually asleep or not even in the house, so opportunities to back it up are rare. Therefore it has a host-wide max-age setting.

Before you actually make a backup, you should do a “dry run” to verify that rsbackup does what you expect.

rsbackup --backup --dry-run

This will print out the commands that it would run without actually executing them. It’s also a good way of verifying that the syntax of the configuration is correct.

Once you’re happy with the output, you can try making an initial backup:

rsbackup --backup --verbose

The --verbose option makes rsbackup report what it is doing as it progresses. In normal use you would omit it but it’s useful when setting up and crucial when debugging.

Depending on how much data you have (and how fast your disks and network are) the initial backup may take a very long time. I did my initial backups inside an instance of screen so that they couldn’t be affected by logging out, etc.

For each backup of each volume, a file will be created in /var/log/backup detailing the command executed and any error output.

Pruning refers to deleting a volume’s old backups. During pruning, a backup will be deleted if it is older than the volume’s max-age setting, with the proviso that the most recent min-backups backups on each store will not be deleted.

To prune any backups that are due to be deleted:

rsbackup --prune --verbose

The details of what is pruned are logged to files in /var/log/backup.

If a backup fails then it will be left in an incomplete state. You can tell rsbackup to pick up where it left off simply by running it again on the same day; if however you leave it until another day then that backup will never be completed. To delete any incomplete backups, therefore:

rsbackup --prune-incomplete --verbose

Backups are not pruned immediately because even if they are incomplete, the portion that succeeded may be useful to reduce the amount of data copied when you retry.

When you take a volume or host out of service, you need to tell rsbackup about this. The first part of this is to remove the corresponding volume or host sections in the configuration file. If you don’t do this then rsbackup will keep on trying to backup the obsolete volume or host.

The second part is to delete old backups for the volume and their corresponding logfiles. If you don’t do this then the report will complain about them. This can be done with the --retire option:

rsbackup --retire HOST:VOLUME

...or:

rsbackup --retire HOST

All backups on available devices and their corresponding logfiles will be deleted (possibly a slow process).

If you want to keep (say) the last backup for the volume then you should at least rename it aside, otherwise --retire will delete it; you may prefer to tar it up and compress it.

You can generate an HTML status report to a disk file:

rsbackup --html status.html

This will show:

• Any backups to unknown devices. (See below.)
• For each volume, the dates of the oldest backup, how many backups exist. The latter is red if there are any devices with no backups of the volume.
• For each volume’s backup to a each device, the date of the newest backup on the device and the total number of backups on the device. The former is red if it violates the max-age setting for the volume and the latter is red if there are no backups of the volume on the device.
• The logfile for the last backup, if it failed. You can use the --logs option to make the logfile section more verbose.
• Logfiles for pruning.

If you prefer plain text you can use --text instead of --html.

You can also request that the report by sent by email, with the --email option. This is intended for use when automating backups. The email contains both the plain text and the HTML versions, most mail clients should be able to pick whichever they are best able to display.

Hard disks get lost or stolen, and fail. In this case rsbackup needs to be told that one of its devices has gone away. The first part of this is to delete the corresponding device directive in the configuration file (and the store directive, if that’s unique to the device).

The second part is to delete logfiles for backups to the device. If you don’t do this then the report will complain about them. This can be done with the --retire-device option:

rsbackup --retire-device NAME

You can do these steps in either order, but if you delete the logfiles first, you will be ask if you are sure. You can override this with the --force option.

--retire-device will never delete actual backups, only logfiles.

If a backup device gets full you have several options:

• Reduce the number of backups to be kept on it by lowering prune-age. But sooner or later you may reach the point where you just cannot keep backups as long as you like.

• Introduce an entirely new, bigger device and take the old device out of service, either keeping it against a rainy day or destroying it as described above.

• Copy all its contents to a new, bigger device, keeping the same device name. Remember to delete the old device-id file, or confusion may follow!

The backup has the same layout, permissions etc as the original system, so it’s perfectly possible to simply copy files from a backup directory to their proper location. This is the most convenient way if you want to rescue only a small number of files.

Be careful to get file ownership right. The backup is stored with the same numeric user and group ID as the original system used. Put another way, the relationship between usernames and user IDs (and group names and group IDs) on the backup disk reflects the client, not the server (or any other machine the disk might be attached to).

Until a backup is completed, a corresponding .incomplete file will exist. Check for such a file before restoring any given backup.

You can do bulk restores with rsync. For example, supposing that host chymax has a volume called users in which user home directories are backed up, and user rjk wants their entire home directory to be restored:

rsync -aSHz --numeric-ids /store/chymax/users/2010-04-01/rjk/. chymax:~rjk/.

-a means recursive into directories; preserve symlinks, permissions, modification times, groups, owners, device files and special files. -S means handle sparse files efficiently. -H means preserve hard links. -z means compress data for transfer; you might want to omit this if your CPU is slow.

--numeric-ids is important as backups are stored with the same numeric user and group IDs as the original system; no translation via name is performed.