Add a retain old fs option to replace?

[CorrosiveTruths]

Hi, would it feasible to have an option when using btrfs replace with a single fs device to leave the old filesystem intact on the replaced device?

That is, instead of removing the filesystem on the device being removed, generate a new UUID for it.

This would be useful as a backup / recovery in case something goes wrong on the new device (I had the idea when someone on reddit reported this very thing happened).

It would also be useful as a way to clone filesystems - so people don't use dd and end up with two devices with the same UUID.

[kdave]

So there are probably more things how to do tat:

• tweak replace - it could be possible with a minor change to the finalization, not deleting the superblock but updating it with a new metadata_uuid (that's only change in the superblock, otherwise the metadata blocks keep he old uuid, but it's effectively hidden)
• ioctl to clone a device on a mounted filesystem, probably with constraints like only for a single device, and I'm not sure how exactly the new device should be logically connected, as it would be a new device but unusable for writes
• add a safe way to "dd" the device with metadata_uuid change at the end, on an image it would be easy, on a mounted filesystem it would require quiescent state (like read-only mount or maybe turning the device itself to read-only)

[Forza-tng]

A btrfs clone could be a useful command. It save having to do the mkfs.btrfs, subvol create, and send|receive (hard to do with many or nested subvols and snaps) operations.