Re: [v6ops] privacy point re. unsolicited NA / router neighbor cache

[David Lamparter <[e-mail]>]

On Tue, Jul 23, 2019 at 09:01:41AM +0200, Gert Doering wrote:
> On Tue, Jul 23, 2019 at 02:00:49AM +0200, David Lamparter wrote:
> > => Yes: it seems someone configured multicast.  (You don't get this
> >    without config, it's a bit of a PITA honestly.)  Proceed to 2.
> > 
> > 2. is whoever configured multicast competent?
> 
> The general assumption for on-link multicast should be "it's broadcast
> or it is getting lost".
> 
> Switch vendor track record shows that this is a constantly recurring
> problem, and the *best* you can hope for operational robustness is
> "on-link multicast is handled like broadcast".  All the other snooping
> things just cause more work in figuring out this week's bugs and how
> to turn off the snopping bits.
> 
> The general idea that on-link multicast is a positive aspect of IPv6 
> is nice, but wrong.

I'm happy to agree that in most cases multicast will be broadcast, and
gleaning DAD on the router(s) to prime a neighbor cache entry works
perfectly fine for that situation.

I also think that we shouldn't break things by design, and the
address-dependent DAD multicast groups are a feature both in terms of
limiting flooding as well as improving privacy, even if neither of these
two materializes in common setups.

Which of these matters more is of no impact for DAD gleaning since it
happens to work out to a good solution for both.  It *does* break for
networks with badly configured multicast, but I believe this is an
acceptable caveat -- what we're trying to fix here is a delay in
establishing communication, so the failure case is that this delay
remains.  I'd argue this is an acceptable result of an incompatible
multicast configuration made out of either ignorance or error.

I'm also confident enough in saying I don't believe any relevant switch
or router default configurations include such broken multicast snooping
config.  If it's broken, it's something the operator broke.  I'd even
agree with you that operators are well-advised to not enable multicast
snooping in most cases, so it keeps being broadcast.  Exceptions prove
the rule.

Cheers,


-David