Debian Patches

Status for ovn/26.03.3-4

Patch Description Author Forwarded Bugs Origin Last update
lp-2066194-tests-Ignore-ovs-vswitchd-received-packet-on-unknown.patch tests: Ignore ovs-vswitchd received packet on unknown port.
The ovs-vswitchd daemon may in some situations, typically during
teardown, log a warning level message 'received packet on unknown
port ...'.

Ignore this message.
Frode Nordahl <fnordahl@ubuntu.com> no 2024-05-20
fix-ovn-vif-test-depends.patch Fix ovn-vif test depends
===================================================================
Thomas Goirand <zigo@debian.org> no debian 2026-05-19
fix-build-with-openvswitch-4.0.patch Remove calls to service_start() et al. to build against OVS 4.0 Upstream commit 609c91314 "treewide: Remove Windows service handling"
dropped service_start(), service_stop() and should_service_stop() from
the OVS daemon API, released in OVS 4.0. OVN 26.03.x still calls them,
so the build against openvswitch-source 4.0.0-2 fails with
-Werror=implicit-function-declaration in utilities/ovn-dbctl.c, and the
other daemons would fail at link time. On Linux these calls were
no-ops. This is the ovn part of that commit, already shipped in OVN
26.09.

diff --git a/br-controller/ovn-br-controller.c b/br-controller/ovn-br-controller.c
index 93526a2f6..ca6db7db7 100644
Ilya Maximets <i.maximets@ovn.org> no upstream, https://github.com/ovn-org/ovn/commit/609c91314f1da28964817d0cc1251bdb9da44877 2026-11-10
fix-build-with-openvswitch-4.0-route-table.patch Adapt to OVS 4.0 route_table_dump_one_table() API Upstream commit fdc6056b7 "ovs: Bump to include support for filtering
IDL monitor requests." adapts to OVS commit 73fa3e72172c
("route-table: Only dump IPv4 and IPv6 routes."), which added a
sa_family_t argument to route_table_dump_one_table() and moved the
IPv4/IPv6 filtering from the dump callback to the caller. OVN 26.03.x
still uses the 3-argument form, so building against
openvswitch-source 4.0.0-2 fails with -Wint-conversion and
-Wincompatible-pointer-types in controller/route-exchange-netlink.c.
Dump both families explicitly, as done upstream. The pinctrl.c part
of that commit is already covered by the 26.03.x tree (its
compose_nd_ns() call sites already use the OVS 4.0 form); the ovs
submodule bump and the test changes are not applicable here.

diff --git a/controller/route-exchange-netlink.c b/controller/route-exchange-netlink.c
index a9fd1f58a..8f1615c4e 100644
Dumitru Ceara <dceara@redhat.com> no upstream, https://github.com/ovn-org/ovn/commit/fdc6056b76e7f472e98c9e4304f0fc71b2f24410 2026-11-10
skip-tls-deletion-test-step-for-openssl-4.patch Skip TLS deletion test step for OpenSSL 4 Deleting the certificate files from disk no longer drops active
connections under OpenSSL 4, causing the test suite to time out.
This patch comments out the deletion test sequence to unblock the
build.
.
No upstream fix exists: the tested sequence is unchanged in OVN
upstream, and the SSL file-watch logic in OVS' lib/stream-ssl.c has
no OpenSSL 4 handling either. The behavior only resurfaces when
building against OpenSSL 4.0 (experimental).
MJ Ponsonby <mj.ponsonby@canonical.com> no 2026-11-10
northd-learn-arp-nd-inner-mac-addresses-in-fdb.patch northd: Learn ARP/ND inner MAC addresses in FDB. Currently, build_lswitch_learn_fdb_op() only generates
put_fdb(inport, eth.src) flows, so only eth.src is ever learned
into the FDB. In VRRP gratuitous ARP packets, eth.src is the
physical MAC while arp.sha is the virtual MAC (VMAC). Since
put_fdb only records eth.src, the VMAC is never learned and
traffic destined to it falls through to MC_UNKNOWN flooding.
.
The same gap exists for IPv6 Neighbor Discovery: unsolicited
NAs carry the VMAC in nd.tll, and NSes carry it in nd.sll.
.
Add higher-priority (110) flows in both the lookup_fdb and
put_fdb stages for ARP, ND NA, and ND NS packets. The lookup
flows check both eth.src and the inner MAC field (arp.sha,
nd.tll, nd.sll) and store the inner result in a new regbit
REGBIT_LKUP_FDB_DL_INNER (reg0[22]). The put flows fire when
either MAC is unknown and learn both via two put_fdb calls.
When both MACs are already known, no put_fdb flow matches and
no packet-in is generated.
.
ARP/ND packets may lack the arp.sha/SLL/TLL, in which case
arp.sha/nd.sll/nd.tll is all-zeros. Higher priority (120)
flows for ARP/ND flows prevent learning a zero MAC into the
FDB.
.
Backport of branch-26.03 commit ae9f59629 (master commit 988dc20fe), the
direct parent of a61f891f9 ("northd: Allow FDB learning on ports with
port security") whose VRRP VMAC system test depends on the ARP/ND inner
MAC learning flows added here; without them put_fdb only records
eth.src and the test's expected VMAC FDB entry never appears.
===================================================================
Dumitru Ceara <dceara@redhat.com> no upstream, https://github.com/ovn-org/ovn/commit/ae9f596292b59a8c029637db59790ae97014ff2a 2026-11-19
northd-allow-fdb-learning-on-ports-with-port-security.patch northd: Allow FDB learning on ports with port security. When a logical switch port has both port_security and "unknown" in its
addresses, build_lswitch_learn_fdb_op() skipped FDB learning flow
generation because of an early-return guard that checked
op->lsp_has_port_sec.
.
When multiple logical switch ports share the same MAC address (e.g., a
VRRP virtual MAC) in their LSP.addresses, northd generates duplicate L2
lookup flows at the same priority with the same match but different
outport actions. Since ovn-controller can only install one, traffic to
the shared MAC is nondeterministically sent to only one of the ports.
.
The recommended configuration to avoid this is to omit the shared MAC
from LSP.addresses (placing it only in port_security) and include
"unknown" in addresses, so the shared MAC is resolved dynamically via
FDB learning. However, the early-return guard blocked FDB learning on
ports with port_security, preventing this configuration from working.
.
The original guard was added in commit dd94f1266 ("northd: MAC
learning: Add logical flows for fdb") under the assumption that ports
with port security should not participate in FDB learning. However,
ingress port security (ls_in_check_port_sec) validates source MACs
before the FDB learning stages (ls_in_lookup_fdb, ls_in_put_fdb), so
only MACs that pass port security are recorded. Egress port security
also validates packets after L2 lookup. Removing the lsp_has_port_sec
check is therefore safe.
.
Backport of branch-26.03 commit a61f891f9 (master commit 873ac74eb),
needed by neutron's VRRP shared-VMAC handling which puts "unknown" in
LSP addresses (Gerrit 1016458).
===================================================================
Dumitru Ceara <dceara@redhat.com> no upstream, https://github.com/ovn-org/ovn/commit/a61f891f922aa08fbe03220b65574ee8aba68870 2026-11-19

All known versions for source package 'ovn'

Links