On Mon, Jul 29, 2019 at 01:05:50PM +0000, Paul Blakey wrote: > Hi, > > I reproduced both traces. > > The first rcu trace is because we deference the qdisc block without rtnl > locked (flower allows it unlocked) > > and the second trace is because res->goto_tp can be null if the last > prio of a chain we goto was just deleted, and we try and deference it > unconditionally when we set the skb extension in tcf_classify. > > This can happen when a filter is deleted while there is still traffic, I > have a test for it and will try on the fixes. > > I'll post fixed patches (Patch by VladB) to the tc skb extension to fix > both of the above, and ovs userspace RFC v2 to fix the nat issue.
Cool, thanks Paul! Marcelo _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev