MTU Option should be included in ICMPv6 Router Advertisements
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| neutron |
Fix Released
|
Undecided
|
Sridhar Gaddam | ||
Bug Description
When using an overlay network on a physical network with standard Ethernet MTU (1500 octets), the instances' effective MTU is reduced.
The Neutron Router should inform the nodes about this fact, by including the MTU Option in the ICMPv6 Router Advertisements it sends. The current situation leads to blackholing of traffic, as the absence of the MTU Option causes the instance to believe it will be able to successfully transmit 1500 octets large frames to the network. However, these will be silently discarded. The symptom of is usually that the TCP three-way handshake succeeds, but that the connection appears to hang the moment payload starts being transmitted.
The MTU Option is documented here: https:/
I observe this problem on OpenStack Kilo.
| Changed in neutron: | |
| assignee: | nobody → Sridhar Gaddam (sridhargaddam) |
| Changed in neutron: | |
| status: | New → In Progress |
| tags: | added: ipv6 |

I was mistaken regarding the Neutron router being aware of the reduced effective MTU. This had been hard-coded in /etc/neutron/ dnsmasq- neutron. conf, in accordance with the documentation in http:// docs.openstack. org/kilo/ install- guide/install/ apt/content/ neutron- network- node.html.
Nevertheless, a similar method for configuring/ hard-coding the MTU option in ICMPv6 RAs is required in order to prevent blackholing of IPv6 traffic in overlay networks with reduced MTU.