Re: Issue: Should MOBIKE react to problems not known locally (#NEW)

Greg Daley writes:
> Hi Jari,
> 
> Jari Arkko wrote:
> > 
> > Should MOBIKE only react to problems that are known
> > to the nodes locally? For instance, if an interface
> > in a host goes down, or it is assigned a new address,
> > then the node knows about this and can tell its peer.
> > 
> > Or should MOBIKE also deal with other problems, typically
> > based on the symptom of not getting packets through?
> > For instance, should MOBIKE attempt to do a failover
> > upon seeing
> > 
> > - DPD failure
> > - ICMP error
> > - No response to the MOBIKE protocol messages
> > - No incoming payload packets
> 
> I think that there's obviously an issue with
> using a mechanism which isn't in itself secured
> to indicate changes.

To indicate that there might be change, and we should start checks and
perhaps a recovery. I.e. if we do not get any replies for out IKE
packets then we should try to use other address for the other end,
even when it have not send notification asking us to move to that
address. The reason that address might be broken could be some router
alogn the path which is dropping all packets. The other end cannot
know that, thus it cannot send notifications to us, asking us to move
to another address.

ICMP host unreachable, can also used to trigger a dead peer detection
check for the IKE sa, just to verify whether the IP-addresses still
work. 
-- 
[email protected]