[openstack-dev] Change in openstack/neutron[master]: Permit ICMPv6 RAs only from known routers

Robert Li (baoli) baoli at cisco.com
Wed Mar 5 14:48:46 UTC 2014

Hi Sean,

See embedded commentsŠ


On 3/4/14 3:25 PM, "Collins, Sean" <Sean_Collins2 at cable.comcast.com> wrote:

>On Tue, Mar 04, 2014 at 02:08:03PM EST, Robert Li (baoli) wrote:
>> Hi Xu Han & Sean,
>> Is this code going to be committed as it is? Based on this morning's
>> discussion, I thought that the IP address used to install the RA rule
>> comes from the qr-xxx interface's LLA address. I think that I'm
>Xu Han has a better grasp on the query than I do, but I'm going to try
>and take a crack at explaining the code as I read through it. Here's
>some sample data from the Neutron database - built using
>I don't have V6 addresses working in vagrant_devstack just yet, but for
>the sake of discourse I'm going to use it as an example.
>If you look at the queries he's building in 72252 - he's querying all
>the ports on the network, that are q_const.DEVICE_OWNER_ROUTER_INTF
>("network:router_interface"). The IP of those ports are added to the list
>of IPs.
>Then a second query is done to find the port connected from the router
>to the gateway, q_const.DEVICE_OWNER_ROUTER_GW
>('network:router_gateway'). Those IPs are then appended to the list of
>Finally, the last query adds the IPs of the gateway for each subnet
>in the network.
>So, ICMPv6 traffic from ports that are either:
>A) A gateway device
>B) A router
>C) The subnet's gateway

My understanding is that the RA (if enabled) will be sent to the router
interface (the qr interface). Therefore, the RA's source IP will be an LLA
from the qr interface

>Will be passed through to an instance.
>Now, please take note that I have *not* discussed what *kind* of IP
>address will be picked up. We intend for it to be a Link Local address,
>but that will be/is addressed in other patch sets.
>> Also this bug: Allow LLA as router interface of IPv6 subnet
>> https://review.openstack.org/76125 was created due to comments to 72252.
>> If We don't need to create a new LLA for the gateway IP, is the fix
>> needed? 
>Yes - we still need this patch - because that code path is how we are
>able to create ports on routers that are a link local address.

As a result of this change, it will end up having two LLA addresses in the
router's qr interface. It would have made more sense if the LLA will be
replacing the qr interface's automatically generated LLA address.

>This is at least my understanding of our progress so far, but I'm not
>perfect - Xu Han will probably have the last word.
>Sean M. Collins

More information about the OpenStack-dev mailing list