Re: [PATCH nf-next v3 00/16] netfilter: conntrack: remove percpu lists

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Fri, Apr 08, 2022 at 12:05:38PM +0200, Pablo Neira Ayuso wrote:
> On Fri, Apr 08, 2022 at 11:59:59AM +0200, Pablo Neira Ayuso wrote:
> > On Fri, Apr 08, 2022 at 11:56:09AM +0200, Pablo Neira Ayuso wrote:
> > > On Wed, Mar 23, 2022 at 02:21:58PM +0100, Florian Westphal wrote:
> > > > This series removes the unconfirmed and dying percpu lists.
> > > > 
> > > > Dying list is replaced by pernet list, only used when reliable event
> > > > delivery mode was requested.
> > > > 
> > > > Unconfirmed list is replaced by a generation id for the conntrack
> > > > extesions, to detect when pointers to external objects (timeout policy,
> > > > helper, ...) has gone stale.
> > > > 
> > > > An alternative to the genid would be to always take references on
> > > > such external objects, let me know if that is the preferred solution.
> > > 
> > > Applied 1, 2, 3, 5, 6 and 8.
> > 
> > Not 6 actually, since it depends on 4.
> > 
> > So I'm taking the preparation patches of this batch.
> 
> Wait. Can we possibly set a dummy event handler instead?
> 
> void nf_conntrack_unregister_notifier(void)
> {
>         rcu_assign_pointer(nf_conntrack_event_cb, nfct_event_null_handler);
> }
> 
> which does nothing?
> 
> It also needs to be set on initially to this null event handler?
> 
> So we can avoid the stash trick in nfnetlink too?

Forget this idea, we can't, this event handler is again global.