Dne 3.7.2018 v 20:24 Mikulas Patocka napsal(a):
On Tue, 3 Jul 2018, Milan Broz wrote:
On 07/03/2018 03:21 PM, Mike Snitzer wrote:
Should we even have the possibility to stop recalculating?
If the device was recalculating and we activate it without the
"recalculate" flag, what should it do? If the user reads data that haven't
been recalculated yet, we could:
1. return an error
2. ignore the non-recalculated checksum and return data
3. continue with recalculating even if the flag "recalculate" is not present
In my opinion, the option 3 is the best. Option 1 makes the device
unreliable and option 2 makes the device silently lose the integrity
protection.
Yes, option 3. But you could even error out on table load; though that
may be too unforgiving.. reality is userspace needs to account for
staying in recalculating mode until it is complete.
I do not think table load error is the way to go.
Can we activate dmintegrity read-only (for example with read-only backend device)
that is not yet fully recalculated?
There's no read-only activation because the journal must be replayed. Most
journaled filesystems also don't provide read-only activation because they
need to replay the journal.
There would always an option to use some temporary space with old snapshot
COW target - and use read-only device on top of that.
Zdenek
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
