> On Fri, Sep 14, 2001 at 11:14:30AM +1100, Nathan Scott wrote:
> > In your case, it seems the driver has marked /dev/vg0/masano1
> > as readonly (I don't know why that would be - looks like LVM).
> LVM snapshots are always read-only unless you add special patches. LVM
> only does block COW for writes to the original block device, not for the
> snapshot. While it is possible to patch it to do COW for snapshots also
> it is probably unsafe (LVM data structures are not journaled etc.)
> To handle such an unpatched LVM xfs_freeze would need to flush the log so
> that no recovery is needed. I actually though it did so already, but it seems
> that was mistaken.
I think a merge of the Irix version of this code has changed the situation,
and it is now attempting to replay the log. It should be fairly simple
to back out this part of the change.