Re: reencryption with multi keyslots in batch mode

[Ondrej Kozina <okozina@redhat.com>]

Hi,

On 03/06/2026 14:43, Frederic Hoerni wrote:
> Hello,
> 
> If we want to reencrypt a disk that has several keyslots (main and
> recovery keyslots), I believe the only way is to do it in interactive mode.

Correct.

> 
> But in some cases, we would like to do this in batch mode. We could
> imagine a scenario like this:
> 
> 1. Setup of the use case (no modification of cryptsetup needed so far)
> 
> # Initialize LUKS container
> dd if=/dev/zero count=200 of=disk.img bs=1M
> echo 0000 | sudo cryptsetup luksFormat --keyfile-size 4 \
>                                          --key-file - \
>                                          disk.img
> # Add two keyslots
> echo -n 0000x111111 | cryptsetup luksAddKey --key-file - \
>                                               --keyfile-size 4 \
>                                               disk.img
> echo -n 0000222     | cryptsetup luksAddKey --key-file - \
>                                               --keyfile-size 4 \
>                                               disk.img
> 
> 
> 2. Reencryption (this step needs modification of cryptsetup)
> 
> echo -n 0000x111111222 | \
>       sudo cryptsetup reencrypt --key-file - \
>                                 --keypipe-sizes "4,7,3" \
>                                 --batch-mode \
>                                 --force-offline-reencrypt \
>                                 disk.img
> 
> 3. Verify all keyslots
> 
> for passphrase in 0000 x111111 222; do
>       echo -n $passphrase | cryptsetup open disk.img --test-passphrase \
>                                                      --key-file -
> done
> 
> Do you have any on-going work in this direction?

No, currently I do not have any plans.

> Are you open to contribution for that? (I already have a proof of
> concept for the above scenario)

Of course! As with any other open source project feel free to contribute 
via a merge request in https://gitlab.com/cryptsetup/cryptsetup

> Do you have any particular caveat about developing this feature?

Me personally, I'd probably go with dropping the additional keyslots for 
the time of reencryption. The reencryption performance is more or less 
dependent on a free space in LUKS2 header keyslots area. The only 
exception from this rule is reencryption with datashift (reencrypt 
--encrypt --reduce-device-size XXX command for example).

And you can recreate additional (or recovery?) keyslots after the 
reencryption operation is completed.

With kind regards
O. Kozina