On most architectures other than 32-bit x86, struct flow_tnl ends with 4
padding bytes. Until now, tnl_xlate_init() set those bytes to nonzero
values in the wildcard mask. When the wildcard mask passed through Netlink
attributes and back to userspace, the padding bytes of course became zero
again, which caused a wildcard mask mismatch and premature deletion of the
flow in revalidation. This cmomit fixes the problem.
Bug #1192516.
Reported-by: Krishna Miriyala <miriya...@vmware.com>
Signed-off-by: Ben Pfaff <b...@nicira.com>
---
ofproto/tunnel.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/ofproto/tunnel.c b/ofproto/tunnel.c
index 38b782f..001045a 100644
--- a/ofproto/tunnel.c
+++ b/ofproto/tunnel.c
@@ -324,10 +324,15 @@ tnl_xlate_init(const struct flow *base_flow, struct flow
*flow,
struct flow_wildcards *wc)
{
if (tnl_port_should_receive(flow)) {
- memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
+ wc->masks.tunnel.tun_id = OVS_BE64_MAX;
+ wc->masks.tunnel.ip_src = OVS_BE32_MAX;
+ wc->masks.tunnel.ip_dst = OVS_BE32_MAX;
wc->masks.tunnel.flags = (FLOW_TNL_F_DONT_FRAGMENT |
FLOW_TNL_F_CSUM |
FLOW_TNL_F_KEY);
+ wc->masks.tunnel.ip_tos = UINT8_MAX;
+ wc->masks.tunnel.ip_ttl = UINT8_MAX;
+
memset(&wc->masks.pkt_mark, 0xff, sizeof wc->masks.pkt_mark);
if (!tnl_ecn_ok(base_flow, flow)) {
--
1.8.5.3
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
