[GENIUS-20] Cannot have same vtep both on vxlan TZ and vxlan-gpe TZ

Tue, 27 Feb 2018 10:01:35 +0000

On a context of:

ODL (Boron)
netvirt enabled as neutron provider
SFC enabled
genius transport zones being used to configure the tunnel meshes both for netvirt and SFC
Yi Yang OVS patches applied (https://github.com/yyang13/ovs_nsh_patches)

consider the following transport zone configuration:

{
"transport-zones": {
"transport-zone": [
{
"zone-name": "dovs-tz",
"tunnel-type": "odl-interface:tunnel-type-vxlan-gpe",
"subnets": [
{
"prefix": "172.17.0.0/16",
"vteps": [

{ "dpn-id": 263936438496426, "portname": "dovs-vtep-1", "ip-address": "172.17.0.2" }

{ "dpn-id": 5274128771036, "portname": "dovs-vtep-2", "ip-address": "172.17.0.3" }

],
"vlan-id": 0,
"gateway-ip": "0.0.0.0"
}
]
},
{
"zone-name": "49e000d3-2812-4c81-bc7a-3c113e98e350",
"tunnel-type": "odl-interface:tunnel-type-vxlan",
"subnets": [
{
"prefix": "10.0.0.0/24",
"vteps": [

{ "dpn-id": 263936438496426, "portname": "tunnel_port", "ip-address": "172.17.0.2" }

{ "dpn-id": 5274128771036, "portname": "tunnel_port", "ip-address": "172.17.0.3" }

],
"gateway-ip": "0.0.0.0",
"vlan-id": 0
}
]
}
]
}
}

Zone "49e000d3-2812-4c81-bc7a-3c113e98e350" is created by netvirt. Zone "dovs-tz" is created for SFC.

We look into the resulting OVS configuration of DPN 263936438496426:

> ovs-vsctl show

fbeb6060-1a19-4a35-a37f-3a6913dee417
Manager "tcp:10.0.2.2:6640"
is_connected: true
Bridge br-int
Controller "tcp:10.0.2.2:6653"
is_connected: true
fail_mode: secure
Port "tapbfc5baaa-a1"
Interface "tapbfc5baaa-a1"
Port "tunb32255ba66c"
Interface "tunb32255ba66c"
type: vxlan
options:

{key=flow, local_ip="172.17.0.2", remote_ip="172.17.0.3"}

Port "tun445ad3536e6"
Interface "tun445ad3536e6"
type: vxlan
options:

{exts=gpe, key=flow, local_ip="172.17.0.2", "nshc1"=flow, "nshc2"=flow, "nshc3"=flow, "nshc4"=flow, nsi=flow, nsp=flow, remote_ip="172.17.0.3"}

Port br-int
Interface br-int
type: internal
ovs_version: "2.5.90"

As you can see we have two vxlan ports, with equal configuration regarding source IP, destination IP and key. Main difference being one of them is vxlan while the other is vxlan-gpe.

But then when we check how things are at openflow level:

—
> ovs-ofctl show -O OpenFlow13 br-int

OFPT_FEATURES_REPLY (OF1.3) (xid=0x2): dpid:0000ac2034e48141
n_tables:254, n_buffers:256
capabilities: FLOW_STATS TABLE_STATS PORT_STATS GROUP_STATS QUEUE_STATS
OFPST_PORT_DESC reply (OF1.3) (xid=0x3):
1(tapbfc5baaa-a1): addr:7e:1e:2a:b4:8c:09
config: 0
state: 0
current: 10GB-FD COPPER
speed: 10000 Mbps now, 0 Mbps max
2(tun445ad3536e6): addr:a2:42:54:79:62:3e
config: 0
state: 0
speed: 0 Mbps now, 0 Mbps max
LOCAL(br-int): addr:ac:20:34:e4:81:41
config: PORT_DOWN
state: LINK_DOWN
speed: 0 Mbps now, 0 Mbps max
OFPT_GET_CONFIG_REPLY (OF1.3) (xid=0x5): frags=normal miss_send_len=0
—

One of the tunnel ports, "tunb32255ba66c", is missing it's openflow definition. This cause various problems down the line.

It looks like OVS forms a tuple with some of the tunnel ports configuration parameters that uses to distinguish one tunnel port from another, and that in our case the basic difference of gpe vs non gpe is not part of this tuple; and results in the problem observed.

There are some alternatives:

TZ definition could allow to define a port number. OVS would presumably be able to tell apart both tunnel port definitions if they have a different port number.
Genius could add the gpe extension to an existing vxlan tunnel port, instead of creating a new tunnel port, if the vxlan port of the same characteristics already exists. Other way around, i would not need to create a vxlan port if there is already a vxlan-gpe one of the same characteristics. For this to be considered, a vxlan-gpe endpoint should be able to exchange standard vxlan traffic with a vxlan endpoint. Own tests indicate that this might indeed be the case.
Maybe this could be an OVS bug and it should be solved there.

Also, it looks like genius is using the port-name of each vtep inside the transport zone definition (i.e. dovs-vtep-1 in the example above) to generate the tunnel port name (i.e. tunb32255ba66c). Probably that is not the best approach as that is not a meaningful parameter at OVS level. Genius also uses the tunnel type for the tunnel port name and in this case vxlan and vxlan-gpe should not be considered different tunnel types for this purpose either. Basically the data used to generate the port name should be the same data that OVS uses to tell tunnel ports apart.

OpenDaylight JIRA

[GENIUS-20] Cannot have same vtep both on vxlan TZ and vxlan-gpe TZ