On Fri, 2008-04-25 at 11:06 -0400, Sebastien Roy wrote:
> On Fri, 2008-04-25 at 04:05 -0400, Peter Memishian wrote:
> > Going further, it's unclear to
> > me if there are consumers that care about both DL_CURR_PHYS_ADDR and
> > DL_CURR_DEST_ADDR.  That is, does it make sense to have both of these
> > show up as DL_NOTE_PHYS_ADDR, or would it be better to have a new
> > DL_NOTE_DEST_ADDR message type?  (With such a change, the libdlpi change
> > would no longer be mandatory, but would still probably be worthwhile.)
> 
> My first reaction to this was that it feels wrong to design the DLPI
> protocol around libdlpi, but thinking about this further, I don't have a
> problem adding a DL_NOTE_DEST_ADDR given that I'm not at all enamored
> with the existing DL_NOTE_PHYS_ADDR design.  Let's proceed down this
> path.
> 
> As for DL_IPV6_LINK_LAYER_ADDR, I think it's an ATM specific thing (I
> wasn't privy to the design nor implementation of this thing), maybe
> libdlpi could just drop these on the floor?

I went down this path, and the fix didn't feel right at all.  We'd have
a DL_PHYS_ADDR_REQ/DL_PHYS_ADDR_ACK mechanism which supports multiple
address types just fine, but a DL_NOTE_PHYS_ADDR (which uses the same
data structure) which supports some of the same address types except for
one, which is handled by a different notification message,
DL_NOTE_DEST_ADDR.  It's not copasetic.

I've implemented what is IMO a cleaner fix:

http://cr.opensolaris.org/~seb/physaddr-fix/

DL_NOTE_PHYS_ADDR/DL_CURR_DEST_ADDR is only sent up for datalinks that
have a destination, and libdlpi drops notifications for address types it
doesn't know about (which is everything except for DL_CURR_PHYS_ADDR).
To me, this makes the most sense.

I'm currently in the process of testing this from the IP tunnel
perspective, and it's fine.  This is as expected, as the fix is quite
unobtrusive to the IP tunnel code.  I'd be interested in verifying that
this fix addresses your original problem (I'd bet money that it does),
but I don't think we can be completely sure of that until I putback and
you bringover to your child.  Any better ideas?

-Seb



Reply via email to