Re: [PATCH 11/19] xfs: Convert to new freezing code

[Jan Kara <jack@suse.cz>]

On Fri 09-03-12 10:20:41, Dave Chinner wrote:
> On Mon, Mar 05, 2012 at 05:01:09PM +0100, Jan Kara wrote:
> > Generic code now blocks all writers from standard write paths. So we block all
> > writers coming from ioctl and replace blocking of transactions on frozen
> > filesystem with a debugging check. As a bonus, we get a protection of ioctl
> > against racing remount read-only. We also convert xfs_file_aio_write() to a
> > non-racy freeze protection.
> ....
> > diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
> > index 329b06a..6468a2a 100644
> > --- a/fs/xfs/xfs_trans.c
> > +++ b/fs/xfs/xfs_trans.c
> > @@ -577,7 +577,6 @@ xfs_trans_alloc(
> >  	xfs_mount_t	*mp,
> >  	uint		type)
> >  {
> > -	xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
> >  	return _xfs_trans_alloc(mp, type, KM_SLEEP);
> >  }
> 
> So what is there to stop internal XFS threads from starting
> transactions when the filesystem is frozen? Previously this
> SB_FREEZE_TRANS would guarantee even internal fucntions would get
> stopped, but now there's nothing?
  I've checked XFS now and I didn't find any work that would be done on a
clean filesystem. I found:
_xfs_mru_cache_reap() - doesn't seem to do any IO
xfs_end_io(), xfs_buf_iodone_work() - shouldn't be a problem either since only
  reads can happen on a frozen filesystem
xfs_reclaim_worker() - everything is clean so no IO should happen
xfs_flush_worker() - everything is clean so no IO should happen
xfs_sync_worker() - again the same if I understand the code right

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR