Sebastien Roy wrote:
> This is not an acceptable way to do this IMO, since LLA's are not unique
> across multiple data-links. You could ARP on one interface and get a
> response, but from a system that just happens to have the same address
> as another system you intended to communicate with on another interface.
>
> The only reliable way to communicate to a LLA when the system has more
> than one interface is to specify the output interface somehow.
I understand this. That is why I asked the question, do
we want to support routable <-> LLA? I think Jim suggests
that doing nothing is OK as an admin can always cheat and
assign a LLA to an interface. But I guess this does not fit
well with the "auto magic" part of the project.
Suppose we somehow assigned an interface to be THE interface
with routable address to be able communicate with LLA. And
the way we assign it is simple. If there is only one interface,
it is automatically THE one. If there are more than one
interfaces, either
1. if a user has specified a preference, use that.
2. just picked the first interface to be up and running.
We introduce a new flag, something like IFF_LLA_COMM, and
set it in THE interface. We also enforce that only one
interface in any stack instance can have this flag set.
Comments?
--
K. Poon.
[EMAIL PROTECTED]
_______________________________________________
networking-discuss mailing list
[email protected]
