(usagi-users 00659) Re: [mipl] Re: new suggestion for mobile node implementation

[Lars Henrik Petander <lpetande@xxxxxxxxxx>]

Hello Itojun,


On Thu, 2 Aug 2001 itojun@xxxxxxxxxx wrote:

> 
> >  This is Takamiya, a core member of USAGI.
> >  We are thinking about introducing new virtual device for Mobile
> >  Node in Mobile IPv6.
> 
> 	Bump-in-the wire mobile-ip6 implementation?  there are multiple
> 	problem with it.  I don't recommend this approach.

I talked with Andi Kleen at OLS and his opinion of using a virtual device
was similar for the same reasons.


> 	2. MTU issues.  TCP segment size computation should consider
> 	   mobile-ip6 extension headers, and with bump-in-the-wire
> 	   approach this is not possible.

The handling of MTU for tcp is currently done by reserving 128 bytes of
space for possible destination options in every packet in tcp_ipv6.c. In
any case this should be redesigned in a cleaner way.

One possibility would be to add a new field for MTU available to the
trasnport layer to the destination cache entries and possibly also
idevs. However, this would require changes to the MTU code in general.


> 	4. You also need to duplicate L3 fragmentation code, into mip6 virtual
> 	   device code, as well as IPv6 output logic.

This was the primary reason behind our decision to implement MIPL as a
part of the IPv6 stack. 

Henrik