Skip to content

controller: Flood local IGMP/MLD queries in L2 domain only. - #324

Closed
DeokarT wants to merge 1 commit into
ovn-org:mainfrom
DeokarT:ip-mcast-querier-flood-l2
Closed

controller: Flood local IGMP/MLD queries in L2 domain only.#324
DeokarT wants to merge 1 commit into
ovn-org:mainfrom
DeokarT:ip-mcast-querier-flood-l2

Conversation

@DeokarT

@DeokarT DeokarT commented Aug 26, 2026

Copy link
Copy Markdown

Summary

Locally generated IGMP/MLD membership queries were injected into OFTABLE_LOCAL_OUTPUT with outport set to MC_FLOOD. That floods the query to every local port, including patch ports towards logical routers.

In scaled topologies (many VIFs and/or many connected routers, e.g. OpenShift UDN gateway routers) the nested resubmits exceed OVS's 4096 resubmit limit. ovs-vswitchd then drops the packet-out:

WARN|over 4096 resubmit actions on bridge br-int while processing
ip,in_port=CONTROLLER,...,nw_dst=224.0.0.1,nw_proto=2,...
sending OFPFMFC_UNKNOWN error reply to OFPT_PACKET_OUT message

Snooped queries are already re-injected with MC_FLOOD_L2 via ip_mcast_forward_query(). This change uses the same key for locally generated querier queries so they stay in the L2 broadcast domain (RFC 4541) and skip router ports that would drop the traffic anyway.

This is the remaining IGMP path after ARP/GARP fan-out was reduced; those packets still used MC_FLOOD and skipped northd's eth.mcast && ip_MC_flood_l2 split.

Test plan

  • Existing IGMP snoop/querier/relay and MLD querier tests updated to assert queries reach the local switch VIFs and do not appear on other switches' ports
  • GitHub Actions make check (IGMP/MLD querier tests)
  • Backport to the OVN stream consumed by OCP 4.18.z after FDP review

Reported-at: https://issues.redhat.com/browse/OCPBUGS-114016
Signed-off-by: Trushna Deokar tdeokar@redhat.com

Locally generated IGMP/MLD membership queries were injected into
OFTABLE_LOCAL_OUTPUT with outport set to MC_FLOOD.  That floods the
query to every local port, including patch ports towards logical
routers.  In scaled topologies with many VIFs and/or many connected
routers, the nested resubmits exceed OVS's 4096 resubmit limit and
the packet-out is dropped.

Snooped queries are already re-injected with MC_FLOOD_L2.  Use the
same key for locally generated queries so they stay in the L2
broadcast domain (RFC 4541) and skip router ports that would drop
the traffic anyway.

Reported-at: https://issues.redhat.com/browse/OCPBUGS-114016
Signed-off-by: Trushna Deokar <tdeokar@redhat.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
@dceara

dceara commented Aug 26, 2026

Copy link
Copy Markdown
Collaborator

Thanks for the contribution, @DeokarT! As discussed privately, it's probably better if you post this as a patch to our development mailing list, as documented in our guidelines: https://github.com/ovn-org/ovn/blob/main/Documentation/internals/contributing/submitting-patches.rst

Feel free to CC me on the patch or to reach out if you face any issues while doing that.

Regards,
Dumitru

@DeokarT

DeokarT commented Aug 26, 2026

Copy link
Copy Markdown
Author

Closing in favor of a mailing-list submission to dev@openvswitch.org per OVN contributing guidelines. GitHub PRs are not the review path for this project.

The OpenShift-side change for OCPBUGS-114016 is openshift/ovn-kubernetes#3424

@DeokarT DeokarT closed this Aug 26, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants