When a flow_tnl is being translated to Netlink attributes, the tun_id field
was included only if the FLOW_TNL_F_KEY flag was set. This meant that for
a mask, where one would not necessarily expect that flag to be set even if
there were a key, the tun_id could be omitted even if it were nonzero.
This led to kernel flows that did not match on a field that was required
to be matched (possibly causing incorrect treatment of packets) and
premature deletion of kernel flows due to mask mismatch. This commit
fixes the problem.
Bug #1192516.
Reported-by: Krishna Miriyala <miriya...@vmware.com>
Signed-off-by: Ben Pfaff <b...@nicira.com>
---
lib/odp-util.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/lib/odp-util.c b/lib/odp-util.c
index d86dbbb..463f008 100644
--- a/lib/odp-util.c
+++ b/lib/odp-util.c
@@ -872,7 +872,8 @@ tun_key_to_attr(struct ofpbuf *a, const struct flow_tnl
*tun_key)
tun_key_ofs = nl_msg_start_nested(a, OVS_KEY_ATTR_TUNNEL);
- if (tun_key->flags & FLOW_TNL_F_KEY) {
+ /* tun_id != 0 without FLOW_TNL_F_KEY is valid if tun_key is a mask. */
+ if (tun_key->tun_id || tun_key->flags & FLOW_TNL_F_KEY) {
nl_msg_put_be64(a, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id);
}
if (tun_key->ip_src) {
--
1.8.5.3
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
