[support] home agent on dummy interface

[Romain KUNTZ]

Hi Jeff,

On 2009/02/25, at 10:25, Jeff Napper wrote:
>> On 2009/02/23, at 17:43, Jeff Napper wrote:
>>> Greetings, all. I have a newbie question on setting up a virtual  
>>> home agent.
>>
>> Do you mean a Home Agent with a virtual home link?
>
> Er, good question. My setup will appear strange. I have the HA on  
> the dummy interface of the mobile node (yes, I realize this doesn't  
> help much with mobility <grin>), and the MN on a normal eth  
> interface. The idea is to have a local "spoofed" HA on the dummy  
> interface. I also started an radvd on the dummy interface.
>
> The dummy interface is started with a "modprobe dummy" command and  
> is configured with the HA address. Perhaps this is the problem?  
> Should the dummy0 interface have a different address on the same  
> network prefix instead?

Ok if I understood correctly you want to co-locate the HA on the MN,  
and you want that MN to register on this co-located HA. I'm not sure I  
understand the reason behind this but let's assume I do :)

More below.

>>> According to the NEMO Howto, the Home Agent can be placed on a  
>>> dummy interface (or simply an alias on the lo interface I'd guess).
>>
>> Yes, the home link can be emulated with a dummy interface. Note  
>> however that it does not work with the loopback interface.

Ok I realize there was maybe a misunderstanding here: what I meant  
here is that in a "normal" configuration, the home link can be  
emulated with a dummy interface *on the HA* (and not on the MN). So  
this is different from what you would like to achieve.

>>> However, when I set this up, the HA cannot send packets because  
>>> the dummy intf. doesn't allow outgoing packets not addressed to  
>>> itself. In fact, the error is EPERM for the sendmsg() call for  
>>> sending the BA. Is there something I'm missing in the routing  
>>> somewhere or is this just not possible anymore?
>>
>> First, BA are not supposed to be sent through the dummy interface,  
>> they should be sent via the interface connected to the Internet, in  
>> order to reach your mobile node located somewhere in your network.
>
> Indeed, this appears to be the problem (assuming you don't think the  
> setup is inherently flawed). The BA cannot be sent through the dummy  
> interface although it is destined for the localhost because the HA  
> is running on the MN! In my strange configuration, the mobile node  
> isn't located out somewhere in the network, but is co-located with  
> the HA (for reasons having to do with the research).

UMIP was not designed to work this way, so I don't think it will work  
without a few hacks. Furthermore, both the HA and the MN plays with  
XFRM rules and routing, so you may have some conflicts there.

Alternatively, you may try to use virtual machines and networks (e.g.  
vmware) on your MN?

Cheers,
romain