On 5/26/26 3:41 PM, Naveen Yerramneni wrote: > > >> On 26 May 2026, at 6:37 PM, Ilya Maximets <[email protected]> wrote: >> >> !-------------------------------------------------------------------| >> CAUTION: External Email >> >> |-------------------------------------------------------------------! >> >> On 5/26/26 5:53 AM, Naveen Yerramneni wrote: >>> >>> Hi Ilya, >>> >>> Just checking whether you got a chance to look at this patch? >> >> Hi, Naveen. Yes, I did. The problem however, is that the entire logic >> behind the reversible_actions() call is broken by design. The issue you >> described is valid, but the solution should not be specific to conntrack >> or ct_clear. Just looking at the OpenFlow actions we can't really tell >> what kind of datapath actions will be emitted and if they are reversible >> or not. For these pipeline actions, there can be anything in there, >> so the right solution here will be to always unconditionally clone. >> But that is not a good solution, as it will trigger a lot of unnecessary >> clones on the datapath level. And the previous patch for ct_clear >> already cause that problem here: >> >> https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.openvswitch.org_pipermail_ovs-2Ddev_2026-2DMay_432488.html&d=DwIDaQ&c=s883GpUCOChKOHiocYtGcg&r=2PQjSDR7A28z1kXE1ptSm6X36oL_nCq1XxeEt7FkLmA&m=oXA2IV6TKkiyx4caK-10xaGaX-sNubwSQ7NCCYwn34OqyiklsnQcA0g_arahwVoK&s=AHRsco8GLjLT41m1iLxmu8ws54BXnFX8tsIP8C_PfJM&e= >> >> >> We should be checking the actual datapath actions, and not OpenFlow. >> There was an old thread 4 years ago where I raised this same point: >> >> https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.openvswitch.org_pipermail_ovs-2Ddev_2022-2DJuly_396578.html&d=DwIDaQ&c=s883GpUCOChKOHiocYtGcg&r=2PQjSDR7A28z1kXE1ptSm6X36oL_nCq1XxeEt7FkLmA&m=oXA2IV6TKkiyx4caK-10xaGaX-sNubwSQ7NCCYwn34OqyiklsnQcA0g_arahwVoK&s=ROURudnNSvanPQpv3Ba3MHYE4GKWS4j2gUwbCitkTYw&e= >> >> >> I do have some code for it and I'll be posting it today or a bit later >> this week. It will be a noticeably large change though and it has >> its own side effects, but should be much more correct. > > Thanks Ilya for sharing the details!
FWIW, I posted the patches here: https://patchwork.ozlabs.org/project/openvswitch/list/?series=506140&state=%2A&archive=both Will mark this one as superseded for now. Best regards, Ilya Maximets. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
