Re: 6to4 default MTU [was Re: (ngtrans) 6to4 MRU - need WG consensus]

[Brian E Carpenter <brian@hursley.ibm.com>]

Tony Hain wrote:
> 
> kre wrote:
> > Actually, no it isn't, and this is the point.   Packets
> > exceeding the MRU
> > generally just get dropped (unless something implements the
> > idea that Christian suggested, which currently, I know of
> > nothing that does). No ICMP gets sent back.
> >
> > That is, PMTUD doesn't find this problem, what happens is a
> > black hole.
> 
> An apparently false assumption on my part was that implementations had a
> black-hole detector tied to their PMTU discovery process. At least for
> TCP the stack should know that the small packets through SYN-ACK worked,
> so if large ones are failing, it would seem logical to shrink them
> before trying again, then try to grow them over time.
> 
> Short of that, I agree it makes the most sense to require a 6to4
> receiver to send an IPv6 MTU exceeded message when reassembly fails due
> to size. It will have the necessary information, and as a router its job
> is to inform the upstream node when a link capacity is being exceded.
> The unusual aspect of this case is that the link is a logical one inside
> the router between IPv4 and IPv6, and the failure notice is coming from
> the inbound IPv6 interface rather than an outbound one. The more I think
> about it, this would be a reasonable requirement for all inbound IPv6
> interfaces, because we may end up with the same kinds of MTU/MRU
> mismatch issues we found with FDDI-Ethernet bridges as we move into gigE
> ...

Exactly. There is nothing special about a 6to4 interface in this respect.

  Brian