Traceability of configuration changes using Git¶
When seeking a solution to keep full traceability of configuration changes made by (various) users on your firewall, the git-backup plugin might be a useful addition to your setup.
In order to use this feature, one has to install the git-backup plugin first (insearch for os-git-backup).
Since git backup is a little bit different than the standard backup options available, we will explain briefly how it works using the diagram below.
config.xml changes happen due to user or api interaction, an event is triggered to which handlers can subscribe
Our git-backup plugin subscribes to these events in order to add the received backups and commits these with
information extracted from the received xml file. To prevent the system to lock during backups,
we choose this loosely coupled method. Events which are yet unprocessed are being left in the (existing) backup directory.
Events are processed from the moment the initial backup is configured, when disabling backups, the (local) changelog itself remains active.
Git backup will push collected commits to the upstream repository nightly. To shorten this default interval, a custom cronjob (see Settings) can be set up, selecting Remote Backup as the Command. The regular backup procedure (which is also being triggered using the test button in the user interface) is responsible for initialising the empty local repository and configuring the upstream target.
One can always change the upstream target, as long as the newly configured one is either “bare” (empty) or containing the exact same content (/change history) as the one used on this firewall.
The configuration part of this plugin is quite basic and offers two types of transport modes, https using a username and password combination or ssh using public key infrastructure.
Enable backup to the upstream target
The branch to push your commits to on the configured url
SSH private key
When using ssh, make sure to add a private key here
Username, when using gitlab and ssh, the default is
When using https authentication, choose a password here.
Make sure to push to a “bare” upstream repository, when pressing “Setup/Test Git” the initial commits should be send to your git server.
If you use GitHub, then your only option for git-backup, is to configure it for SSH access since GitHub has removed the ability for external applications to log into your account via your username and password.
The fields in OPNSense under
System / Configuration / Backups / Git should contain the following:
URL absolutely MUST follow this format when using GitHub and GitLab:
ssh://github.com/user_name/repo_name.git. Any URL string that does not follow this pattern will not work.
SSH Private key (discussed below)
User Name should ONLY contain the word
password: leave this field empty
You need to create your repository BEFORE enabling git-backup. Do not add any files or READMEs to the repository. In other words, create a BLANK repository.
Next, create a new SSH key specifically for git-backup (only generate the private / public keys per that document and skip the rest). It is imperative that you do not add a password to your key, or your backups will fail with authentication errors.
You should set up SSH access to just your repository by assigning your SSH public key to the repository instead of assigning it to your GitHub / GitLab account. Doing this ensures that you don’t arbitrarily expose more of your git resources to OPNSense than is absolutely necessary for git-backup to work properly.
If you use GitHub, you can add your SSH public key by going to your repository, then click on
Deploy keys. Or you can go straight to the URL using this format:
Check the box
Allow write Access.
Make sure the fields are populated as stated above and that the Enable box is checked, then click on
Setup / Test Git and you should see a message come back at the top of the page indicating that the first backup was successful.
When choosing an non-existing remote branch one will be created automatically. An existing one needs to be empty (blank) on order to use it, which in practice is only possible when creating the repository for the first time.
From the user interface no conflict resolution is offered, you need to configure an upstream repository and stick to it for the lifetime of the firewall. When for some reason a backup needs to be restored and one would like to stick to the same git repository, manual conflict resolution might be an option. Support on these scenario’s is not offered.
The repository is available on the OPNsense machine in the following directory
Conflict resolution can complicate the solution a lot (merging, fast-forward, ….), for this reason we will not accept feature requests trying to push to existing (used) repositories.
When errors occur these will be written to the normal system logging, search for
git-backup in the general
system logging ( ).
Some standard errors might be returned via the test button, which should provide a clear direction, known ones are:
authentication failure -> username/password combination is not valid or the provided ssh key doesn’t match the expected one
ssh hostkey changed -> it looks like a man-in-the-middle attack is happening, if that’s not the case and the remote identification changed for valid reasons, manual intervention is required (remove the offensive key from
git out of sync -> unable to synchronize, see “Conflict resolution” for additional info.
The repository is saved locally on the firewall in
/conf/backup/git, if for some reason one would like to remove the
collected history and start over from scratch, one can safetly remove this directory.
Login using a (ssh) console and remove the git directory in that case (
rm -rf /conf/backup/git)
As long as the plugin is installed and /conf/backup/git contains a git repository, the changes will be captured (also without an upstream). One could use this knowledge as well to keep a local (only) repository by creating a repository without assigning an upstream and leave the backup option disabled.
The firewall contains a local backup of the most recent changes (configured in
/conf/event_config_changed.json could be removed
to “forget” about the already handled config events (in which case all backups will be signaled again to all config syshook handlers)