On Tue, 2012-03-20 at 16:41 +0200, Artem Bityutskiy wrote:
> +       INIT_DELAYED_WORK(&sbwork->dwork, write_super);
> +       sbwork->sb = sb;
> +       delay = msecs_to_jiffies(dirty_writeback_interval * 10);
> +       queue_delayed_work(sbi->dio_unwritten_wq, &sbwork->dwork, delay); 

I've just realized that the side-effect of using DIO workqueue is that
'syncfs()' will also synchronize the superblock because it flushes the
workqueue:

static int ext4_sync_fs(struct super_block *sb, int wait)
{
 	...
        flush_workqueue(sbi->dio_unwritten_wq);
	...
}

But before my change, it seems the superblock was not flushed on
'syncfs()', at least I do no see how this would be done. However, I
think it is OK because I think it is correct to write the dirty
superblock out on 'syncfs()', right?

-- 
Best Regards,
Artem Bityutskiy