Pekka Nikander: With regard to the DCS source address, the current
spec says the following:

         An IP address assigned to the sending interface, or the
         unspecified address if no address is assigned to the sending
         interface.

Should we be more specific, and say that the source address SHOULD be
the link local address?  Right now I don't see any specific reason,
other than the old saying of being conservative when sending and
liberal when receiving.

I didn't find anything in the outstanding issues, but maybe I just
missed it?

---------------

Jari Arkko: In issue 21 we agreed that DCA source should be
link-local.  It seems logical to require the same from DCS?

---------------

Greg Daley: This is a holdover from RFC-2461 where the RS has as
source any address on the host's interface, and the RA is only
link-local.

The original motivation may indeed be as Pekka mentioned.

I'm not fussed now about where we the group
goes with this, but I guess that the same issues which
affect RS/RA will affect DCS/DCA.

I'm currently concerned with RS/DCS messages coming from an
address which is not on an adjacent link.  I don't know
how this is handled in the case where RS's are sent from
a global address today (rfc-2461 doesn't seem to specify
this).

Should we even listen to such DCS's?

I'm pretty certain that we don't want to create
erroneous NC state on the router in this case, so
unicast responses are out.

---------------

Jari Arkko:

> This is a holdover from RFC-2461 where the RS has as source any
> address on the host's interface, and the RA is only
> link-local.

Indeed.

> The original motivation may indeed be as Pekka mentioned.
>
> I'm not fussed now about where we the group
> goes with this, but I guess that the same issues which
> affect RS/RA will affect DCS/DCA.
>
> I'm currently concerned with RS/DCS messages coming from an
> address which is not on an adjacent link.  I don't know
> how this is handled in the case where RS's are sent from
> a global address today (rfc-2461 doesn't seem to specify
> this).
>
> Should we even listen to such DCS's?

Maybe not. But as you say, I'm not sure this is documented
even in the ND case.

> I'm pretty certain that we don't want to create
> erroneous NC state on the router in this case, so
> unicast responses are out.

It still seems safest to me that we should
use link-local addresses. 

---------------

Jari Arkko: This issue is about the type of the source address in a
DCS message. There has not been that much discussion on this issue,
but I think the feeling was that we should restrict the addresses to
be link-local.

At the same time, I tried to make the keyword rules for the different
DCS & DCA fields consistent. Right now (a leftover from 2461) some of
the fields contain a keyword while others do not. I think it is better
to not include a keyword if the definition of a field is just <some
value>, e.g., a value assigned by IANA for ICMPv6 type, 0 for Code, or
an address of some sort for the source field.

I have also checked the current draft with regards to issue 21, which
had to do with the source address of the DCA message. It seems that
the resolution of that issue did not get reflected in the draft, so I
am changing that as well.

Finally, I edited the behaviour rules to not repeat the format
rules.

The changes can be seen in:

http://www.arkko.com/publications/send/issues/issue28diff.html

---------------

Greg Daley: This looks good. 

---------------

---------------

---------------