Kaya is a Bash front end for restic, a modern incremental backup solution written in Go. Kaya provides centralized backup functionality via SSH, similar to "pull" mode backups, but with most of the heavy lifting being done by each client. Kaya makes use of restic's rest-server in append-only mode, so in theory, backed up machines can't delete their past backups, nor see backups for other machines.

Kaya is ALPHA software, and may contain security bugs or other flaws. Kaya's API is still under development, and may change in future releases.

GPL-3.0-or-later

Install dependencies:

apt install pwgen flock apache2-utils

Copy the binaries into place:

cp  kaya            /usr/local/bin/
scp kaya-client     [email protected]:/usr/local/bin/

cp kaya.conf        /etc/
chown root:root     /etc/kaya.conf
vim                 /etc/kaya.conf

mkdir -p    /srv/backups/kaya/
touch       /srv/backups/kaya/.htpasswd

Also install rest-server on the backup host, and restic on your backup targets and backup host. Both can be complied into a statically-linked binary, so it's just a matter of copying the appropriate binaries into the PATH of each machine.

First you need to start restic's rest-server on the backup host:

rest-server --path /srv/backups/kaya/ --append-only --private-repos

Now you can initialize or update a backup like so:

kaya www1.example.com backup -- --one-file-system / /srv /home --exclude /foo

kaya www1.example.com backup -- / --exclude /proc --exclude /sys

The parameters after -- are passed to restic on the backup target. (man restic-backup)

Note that the when using --one-file-system, you must explicitly include the path for each mounted file system you wish to backup. If you forget to add a mount point like /home, you could miss important data. The alternative is to not use that option, but make heavy use of statements like --exclude /proc, etc.

mkdir ~/mount/

cd /srv/backups/kaya/www1.example.com/

restic -p repo-password-keep -r . mount ~/mount/

When you're done, type ^C or run fusermount -u ~/mount/

⚠️ Security limitation: The remote client backed up via Kaya runs the restic command, so it has control over setting time stamp metadata for new backups. If malicious time stamps are set by the client, and you then prune your backups, legitimate backups you want to keep may be automatically removed, leaving illegitimate ones.

The main reasons for using restic is that it is easy to deploy, even on older systems, and it offers the rest-server mode for interaction.

Kaya trusts the central backup server, so it stores the restic repo password in plain text in each repo's data directory. That way, it doesn't have to be synced from the backup server after initializaing the repository, copied to each backup target, saved in your password list for recoveries, nor quickly changed after cloning a production server (see below).

With that in mind, the file system you back up to should be encrypted with LUKS, and it should require a password during the boot process, or when mounting the backup partition manually.

One of the reasons why the central server initiates backups is so client machines won't try to send their backups all at the same time, or in a randomized way.

The other main reason for central coordination is because you may clone a production machine, and run it as a dev instance, with a different IP address and domain name. If both machines have the same credentials for pushing their backup, they would both end up in the same repo. If you then prune backups from time to time, you would occasionally delete a production backup and leave a dev backup in place. If you export snapshots to tape archives, then in that scenario, you would possibly archive a dev backup to tape.

Kaya avoids this issue by connecting via SSH from a backup server to machines with persistent domain names. It also sends credentials to the target machine as part of the connection, so the target machine doesn't have to keep track of them. Authentication and authorization are mediated by SSH keys.

Kaya uses SSH tunneling, with reverse port forwarding, so the client can talk to the rest-server on the backup server without the need to rely upon CA certificates. This is especially useful for older machines that may have an outdated root certificate store.

The rest-server's --append-only mode is meant to prevent infected machines from deleting their own past backups. Target machines are still able to push new backups, and to read archived data.

If you'd like to contribute to Kaya, feel free to open an issue or pull request, or email me with a link to your branch.