Re: [PATCH] Fix shrink_dcache_parent() against shrink_dcache_memory() race (updated patch)

[Jan Blunck <jblunck@suse.de>]

On Thu, Mar 09, Balbir Singh wrote:

> > +/*
> > + * If we slept on waiting for other prunes to finish, there maybe are
> > + * some dentries the d_lru list that we have "overlooked" the last
> > + * time we called select_parent(). Therefor lets restart in this case.
> > + */
> >  void shrink_dcache_parent(struct dentry * parent)
> >  {
> >  	int found;
> > +	struct super_block *sb = parent->d_sb;
> >  
> > + again:
> >  	while ((found = select_parent(parent)) != 0)
> >  		prune_dcache(found);
> > +
> > +	/* If we are called from generic_shutdown_super() during
> > +	 * umount of a filesystem, we want to check for other prunes */
> > +	if (!sb->s_root && wait_on_prunes(sb))
> > +		goto again;
> >  }
> 
> cosmetic - could we do this with a do { } while() loop instead of a goto?
> 
> I think though that after select_parent(), wait_on_prunes() can sleep just
> once, so we do not need a goto. Just calling wait_on_prunes() should
> fix the race. For all the dentries missed in the race, wait_on_parent()
> will ensure that they are dput() by prune_one_dentry() before wait_on_parent()
> returns.
> 
> But, I do not have anything against the goto, so this patch should be just
> fine.
> 

No. Think about a dentry which parent isn't unhashed. This parent will end up
on the lru-list after the wait_on_prunes(). Therefore we have to do the
select_parent()/prune_dcache() again to get rid of all dentries.

And I missed the "goto vs. do...while()" against my colleagues here, too ;)
I'll send a followup.

> >  	if (root) {
> >  		sb->s_root = NULL;
> > -		shrink_dcache_parent(root);
> >  		shrink_dcache_anon(&sb->s_anon);
> > +		shrink_dcache_parent(root);
> >  		dput(root);
> 
> This change might conflict with the NFS patches in -mm.
> 

Hmm, right. Andrew, if you want a rediff against -mm just tell me. I'm
actually diff'ing against lates linux-2.6.git.

Regards,
	Jan

-- 
Jan Blunck                                               jblunck@suse.de
SuSE LINUX AG - A Novell company
Maxfeldstr. 5                                          +49-911-74053-608
D-90409 Nürnberg                                      http://www.suse.de