On 8 Jun 2026, at 18:13, Adrian Moreno via dev wrote:
> The setup of this test includes, among other things, creating a pair of
> veth ports and adding both ends to the same ovs bridge.
> This is use as a loopback monitoring. Openflow flows are created to
> send (truncated) traffic through one end, and drop everything that comes
> from the other end.
>
> However, Openflow flows are created after wiring up the ports so there
> is some time during which the bridge (br0) has "action=NORMAL" with both
> ends of a veth pair connected to it. This could easily lead to a packet
> loop if the kernel automatically generates any multicast traffic
> (e.g: IPv6 NDP, IGMP MLD, etc) on those interfaces.
>
> I guess normally the test is quick enough to remove the default flow and
> install the new ones, but I was running the test suite on a debug kernel
> and reproduced the loop with considerable consistency.
>
> Fix it by removing the default flow before doing the non-conventional
> loopback wiring.
>
> Signed-off-by: Adrian Moreno <[email protected]>
Thanks for finding and fixing this Adrian. The change makes sense to me.
Acked-by: Eelco Chaudron <[email protected]>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
