On Wed, May 20, 2020 at 05:13:39PM +1000, Dave Chinner wrote:
> > > @@ -883,9 +885,7 @@ xfs_quiesce_attr(
> > > {
> > > int error = 0;
> > >
> > > - /* wait for all modifications to complete */
> > > - while (atomic_read(&mp->m_active_trans) > 0)
> > > - delay(100);
> > > + cancel_delayed_work_sync(&mp->m_log->l_work);
> >
> > Shouldn't the cancel_delayed_work_sync for l_work in xfs_log_quiesce
> > be removed now given that we've already cancelled it here?
>
> No, because every other caller of xfs_log_quiesce() requires the
> work to be cancelled before the log is quiesced.
The only other caller is xfs_log_unmount, which could grow it. But
I guess an extra cancel isn't harmful in the end.
