On Thu, Jul 15, 2021 at 06:55:03AM +0100, Christoph Hellwig wrote: > On Wed, Jul 14, 2021 at 07:45:33PM +1000, Dave Chinner wrote: > > That's what happens later in the patchset. The XFS_FEAT_ATTR2 is > > set when either the mount option or the on-disk sb flag is set, and > > it is overridden after log recovery (which can set the SB flag) if > > the XFS_FEAT_NOATTR2 feature has been specified. > > I see where this is going. I still think keeping the logic changes > together and killing XFS_MOUNT_ATTR2 here would be preferably, > especially with all the documentation you have for the attr2 > situation in the commit log here. The problem is that I can't kill XFS_MOUNT_ATTR2 until all the m_features infrastructure is in place to replace XFS_MOUNT_ATTR2 being set in m_flags. We still have to check that the mount options are valid (i.e. attr2 && noattr2 is not allowed) before we read in the superblock and determine if XFS_SB_VERSION_ATTR2BIT is set on disk... It's a catch-22 situation, and this was the simplest way I could come up with to enable a relatively clean change-over... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
