Re: Issue 34: Separate path test, handling changes in NAT mappings

Mohan Parthasarathy writes:
> mohanp>  But there a case in the draft today for which the message
> mohanp>  is sent with the different address from that of the operational
> mohanp>  pair and the peer does not update the address i.e when  
> mohanp>  the operational pair does not work, you reransmit the existing
> mohanp> message in the window with the new address but the peer
> mohanp > does not update with the new address until the CHANGE_PATH
> mohanp> is sent (section 2.3 ).

Actually I think the responder will change the path, but the initiator
will simply issue new CHANGE_PATH when the first one finishes and the
window again has space. The second one will make sure they have same
information of the current operational address pair.

> mohanp> If there is a NAT in this path, are we not
> mohanp> changing the behavior already ? So, why can't this be extended
> mohanp> to path test message also i.e path test messages don't update
> mohanp> the addresses on the SA  unless explictly notified? Am i missing 
> mohanp> something....

In my opinion any of the messages should not update the address pair
unless there is explicit CHANGE_PATH notify in the packet. The problem
case can happen when when we are sending that CHANGE_PATH and then we
notice that the address pair we tried to use to send that one (i.e.
the one where we tried to change the address pair to) is broken, and
we need to retransmit it again with different addresses.
-- 
[email protected]