Hi Florian, I already told you, your patchset works in my testbed here. My only doubt still here is the need for the extra bit. I don't find the scenario that will trigger the problem yet. Some comments: On Thu, Jun 05, 2014 at 04:56:40PM +0200, Florian Westphal wrote: > Yes, its recycling. > IPS_DYING_BIT unset would either mean: > > a) 'This conntrack is dead and redelivery failed. Resend event, then > destroy this conntrack'. OK. In this case the conntrack in located in the dying list. > OR it can mean > > b) 'This conntrack is being allocated/setup as new connection, the > flag field was already cleared'. In this case, the conntrack is placed in the unconfirmed list or the hashes. > In the 2nd case, the conntrack_put would be fatal since the work queue > doesn't own the conntrack (plus the tuple is not dying after all...). The workqueue operates with conntracks that are placed in the dying list. If another CPU holds a reference, the use counter is 2, one for the dying list and another for the reference. The conntrack_put will either a) release the entry whose event was already delivered or b) decrement the use counter. > I've found no way to tell these two conditions apart except via new bit. I believe the rule: "all dead conntracks have the dying bit set" always fulfills. I must be overlooking something... let me know, thanks! -- To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
