[support] Re: [Dsmip] Mobile IPv6 packet fragmentation when packet > P-MTU

[Romain KUNTZ]

Hello,

Your question would be more suitable on the support ML (http://ml.nautilus6.org/mailman/listinfo/support 
), the DSMIP ML being used for issues specific to the DSMIP  
implementation.

I've CC the support ML, please continue the discussion on that list.  
More below:

On 2008/08/23, at 3:44, Shah Mahmood wrote:
> -------                         --------                              
> ------------------
> |CN |----------------------->|HA |---------------------------->|MN  
> at CoA|
> -------                         -------                               
> ------------------
>
> CN sends a packet > P-MTU (1280 bytes) to the HA and gets a ICMPv6  
> Packet Too Big message.
>
> ==========================================================
> In the Linux implementation of IPv6 the function ip6_output.c got  
> the following logic:
[snip]

I believe this logic is only for packets generated by the host, and  
not for forwarded packets. IPv6 does not allow fragmentation when  
forwarding a packet, and instead proposes to inform the original  
sender that fragmentation is needed with an "ICMPv6 Packet Too Big"  
message.

> I want my HA to fragment every packet larger than the PMTU and then  
> send it to the MN.

Some people think that tunnelling is in some way creating a new packet  
whose source is (in that case) the HA, so that such tunnelled packet  
could be fragmented. I'm not sure if the IPv6 specification says  
something about this though.

> It is not doing fragmentation by default when using UMIP; the linux  
> implementation of MIPv6 used by Nautilus.

It is not really a problem of UMIP, but instead if fragmenting a  
tunnelled packet is compliant with the IPv6 spec, and if it is,  
whether Linux can perform such fragmentation. While I'm not sure for  
the former, I don't think the latter is possible.

You may want to ask on a mailing list dedicated to the ipv6 stack on  
linux (e.g. the USAGI mailing list).

Regards,
-- 
Romain KUNTZ
kuntz at lsiit.u-strasbg.fr
LSIIT - Networks and Protocols Team
http://clarinet.u-strasbg.fr/~kuntz/