Thanks for the fast answer.
So the following scenarios should be ok:
1) Normal boot (no crash, previous mount was ro as well, journal empty)
-> mount ro from the beginning => partition hash stays the same
2) Mount rw at any point -> make changes -> remount ro -> sync -> create
new hash -> next boot with ro mount has this new hash.
Christoph Hellwig wrote:
> On Thu, Aug 13, 2009 at 02:37:30PM +0200, Bgs wrote:
>> I don't know xfs' superblock handling well enough, so I'm asking for
>> advice here:
>> Does xfs write anything on the disk when mounting read-only? Is it
>> possible to use a partitions hash (or some well defined portion of the
>> partition) for integrity checks?
> It should not write anything to disk , but older versions did so due to
> bugs. Doing a hash of a read-only filesystem should be fine.
>> The partitions are used read-only and hash would be re-generated if any
>> rw action was done (after remount ro or full reboot of course). Can this
>> be done? I'm concerned about 'mount count' like writes...
> There is no mount-count like write in XFS.
>  the only exception is that log reocvery is performed when we attempt
> a read-only mount with an unclean log, that is the box crashed while
> writing the filesystems. To make sure you never do this always mark
> the underlying device readonly, using blkdev --ro.