ETOOBUSY 🚀 minimal blogging for the impatient
Storing secrets in a Git repository is still not fully solved in my opinion.
I know, I know. Secret stuff do not belong to code, but to configuration, so they have no place in a source code versioning system.
But Git can be much more, including helping out track configurations in a system to allow for controlled evolution of… configurations. When this is the case, secrets might be one of the things that should belong into the repository.
Locally this makes sense: if the secret is there, why shouldn’t the
.git sub-directory contain it? But at this point we would lose the
capability of Git of replicating stuff in a central location for
ease of backup, evolution, whatever.
Now there can be a phylosophical approach where the stuff in this central repository belong to the same security perimeter as the one for the system we’re storing the configurations and secrets for, or even something more core.
Another case is that we don’t or can’t, which pushes to find ways to store these secrets securely in the repository, i.e. encrypting them before storing them.
Project git-secret aims at solving this problem, and it’s a good help. In my opinion, though, it still feels like a half-baked solution, because of a couple of shortcomings:
- some sub-commands only work by decrypting stuff, which is fair of course but also require that the encryption includes a full keypair in the place that does the encryption. This assumption might not be true in some setups
- It’s not easy to track changes to the plaintext files, because they’re explicitly ignored. This is correct from a security point of view, but somehow defeats the use of a SCM.
- Connected to be bullet above, it’s not clear what a best practice for refreshing the encrypted data should be. There is a vague suggestion to always do this at every commit, but there’s no clear example and I think this is on purpose.
So OK, there’s some software and it can use some enhancements. It’s open source baby, why don’t you shut up and show us some code?
Right, I don’t have any to be honest! But I’m thinking about an approach
where the digests of the plaintext files are tracked somehow, and can be
used to figure out the status of secret stuff - including the need to
commit it. I hope the ideas will reach a good point.
Incidentally, I also miss a lot not having Git hooks for
read-only stuff like - you know -
status. I understand that I can
wrap the command somehow, but this seems a bit too far-reaching and
having a hook might help in my opinion.
Well, it is not the case, anyway. Stay safe folks!