Re: [RFC PATCH 01/11] vfs: push __sync_blockdev calls down into sync_fs routines

[Jeff Layton]

On Fri, 2018-05-18 at 08:56 -0700, Christoph Hellwig wrote:
> > --- a/fs/xfs/xfs_super.c
> > +++ b/fs/xfs/xfs_super.c
> > @@ -1097,7 +1097,7 @@ xfs_fs_sync_fs(
> >  	 * Doing anything during the async pass would be counterproductive.
> >  	 */
> >  	if (!wait)
> > -		return 0;
> > +		goto out;
> >  
> >  	xfs_log_force(mp, XFS_LOG_SYNC);
> >  	if (laptop_mode) {
> > @@ -1108,8 +1108,8 @@ xfs_fs_sync_fs(
> >  		 */
> >  		flush_delayed_work(&mp->m_log->l_work);
> >  	}
> > -
> > -	return 0;
> > +out:
> > +	return __sync_blockdev(sb->s_bdev, wait);
> 
> XFS never uses the block device mapping for anything, so this is
> not needed.
> 

Thanks, I wasn't sure about xfs. I'll drop this hunk.

FWIW, I think pushing this call down into the sync_fs routines is still
probably the right thing to do, regardless of the state of the later
patches.

> > +/*
> > + * Many legacy filesystems don't have a sync_fs op. For them, we just flush
> > + * the block device (if there is one).
> > + */
> > +static inline int call_sync_fs(struct super_block *sb, int wait)
> > +{
> > +	if (sb->s_op->sync_fs)
> > +		return sb->s_op->sync_fs(sb, wait);
> > +	return __sync_blockdev(sb->s_bdev, wait);
> > +}
> 
> The proper name for this would be vfs_sync_fs.  And I don't think it
> warrants an inline.

I patterned the name after the call_mmap (and now-defunct call_fsync)
helpers. I'll rename it and change it to be non-inlined.

Thanks,
-- 
Jeff Layton <jlayton@xxxxxxxxxx>