-
Bug
-
Resolution: Done-Errata
-
Critical
-
rhos-17.1.z, rhos-18.0.z
-
None
-
8
-
False
-
-
False
-
?
-
openstack-neutron-22.2.2-18.0.20250707134700.51e19bb.el9ost
-
None
-
-
-
Neutron Sprint 12, Neutron Sprint 13, Neutron Sprint 14
-
3
-
Critical
To close this bug, you should
Cherry pick upstream patch(which ever branch is appropriate) https://review.opendev.org/c/openstack/neutron/+/948783 once merged into 18-trunk-patches
Update "fix in build" version and move it to "review"
-------------------------- Original description ----------------
When you create a network log resource to enable OVN logging:
https://docs.openstack.org/neutron/latest/contributor/internals/ovn/ovn_network_logging.html
It seems that will add a an ACL rule similar to this:
_uuid : bd6eaad4-939d-4624-b2f9-9701d1ec402e action : drop direction : to-lport external_ids : {} label : 1753594327 log : true match : "outport == @neutron_pg_drop && ip" meter : acl_log_meter name : neutron-4054c65c-9a05-4bf4-8abe-f31959dbd56f options : {log-related="true"} priority : 1001 severity : info tier : 0
However, once that ACL rule comes in, Neutron starts to think that the rule is not supposed to be there, so when a sync runs, it'll "create" new ACLs and "delete" these ones
To Reproduce Steps to reproduce the behavior:
1. sudo ovn-nbctl find ACL direction=to-lport action=drop # Check for default drop ACL
2. openstack network log create --resource-type security_group --resource <SG_ID> --event DROP logmedrop -f value -c ID
3. sudo ovn-nbctl find ACL direction=to-lport action=drop # Once network log created default port group ACL get's updated
4. /opt/stack/data/venv/bin/neutron-ovn-db-sync-util --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/plugins/ml2/ml2_conf.ini --ovn-neutron_sync_mode repair # First repair removes the above ACL
5. rerun repair /opt/stack/data/venv/bin/neutron-ovn-db-sync-util --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/plugins/ml2/ml2_conf.ini --ovn-neutron_sync_mode repair # Second repair run get's back the original ACL from step 1
Expected behavior
DBSync should expect pg_drop ACL with SGL options set.
Bug impact
- Very critical. PG_drop ACL should never be destroyed since it is the basis of the blacklisting policy for Security groups
- 18, 17.1.
- Could also happen in 16.2 if someone enables SGL as TP.
- links to
-
RHBA-2025:152056 Release of components for RHOSO 18.0