Possibility of non-destructive rollback to older snapshot similar to Btrfs?

[foxtrotcz]

Hi,
I am using TrueNAS (which uses ZFS) and recently I was reading about ZFS driver in Incus.
I am interested in its limitation of restoring older snapshots.
Described for example here: https://linuxcontainers.org/incus/docs/main/reference/storage_zfs/#limitations

In ZFS you can mount any snapshot as real-only, clone any snapshot to make changes or make destructive rollback.
If I understand correctly, in Btrfs you can the the same things but also make non-destructive rollback with set-default and even make changes to the older snapshot. You can also anytime return the the most recent snapshot.

So I want to ask if there is any chance to make snapshots little more flexible like in Btrfs.
Was there any discussion in this regard? Is it even technically feasible?
Thanks.

[amotin]

Snapshots are read-only by definition. Period. I don't know about BTRFS, but if you need to "make changes to the older snapshot", the writes must go somewhere, and that is what clones are for. They store all divergence from the original snapshot.

[foxtrotcz]

I understand that snapshots are read-only. You are right that clones solve this.

I was thinking what needs to be done to achieve similar result in ZFS:

1. Have dataset1 with snapshots mounted in /mnt/dataset1.
2. Clone old snapshot as dataset1_clone
3. Unmount dataset1
4. Mount dataset1_clone to /mnt/dataset1

This basically simulates what Btrfs does. It mounts old snapshot (as writable clone) to same path as original dataset. Basically making a switch.
And to go back to original dataset you just need to unmount the clone and remount the dataset1. Clone can be then removed.

I was just thinking that it would be nice if ZFS could do this in one go with single command. Something like zfs switch dataset1@snapshot1.

[amotin]

What worries me here most is the unmount operation. If even a single file from that mount point is open, Linux kernel won't let you to unmount it. And swapping the dataset beneath application without unmounting is a request for data corruption. If that is not a problem for you, just write a script of few lines doing that. But you need to keep also in mind possible interaction with other processes on a system, that will see the created clone and may get surprised by it in some way.

[foxtrotcz]

Thanks for answering. I thought about it some more and it really is not as simple as I thought at the beginning.