At Thu, 2 Oct 2008 07:33:20 -0400, "Hemant Singh (shemant)"
<[EMAIL PROTECTED]> wrote:

> FYI

Again, I'm sorry for my overly delayed response.  Now that it's open
(and I don't think there's no secret in the discussion), I'm going to
respond to some other points in the referred message:

> > If we incorporate this issue into the 'ipv6-subnet-model' draft,
> > I'd like to see a clear consensus on this in the working group
> > since it will significantly change the role of this document and
> > affect much more existing implementors.  We should probably even
> > change its title.  (I'd not necessarily object to this approach
> > once the wg can reach consensus on this point).
> 
> Sure. But I don't expect any real pushback from the WG. The on-link
> definition needs to be fixed somewhere. Who is going to argue
> against that? And this (the subnet model) document really needs to
> be able to state or point to the correct definition of on-link.

Frankly, I personally don't expect a pushback either.  But in my past
experience in the 2462bis work, even a most trivial (and reasonable to
me) change could get unexpected pushback once it turns out to require
modification of some existing implementation.  That's why I wanted to
see an explicit support from implementors (as I said in a separate
message, I understand we may still end up regarding no-response as a
"consensus").

> > For example, an attacking node can easily exploit forged redirect
> > messages for the same purpose (but only for hosts as the victim).
> > So, if we specifically want to change these particular points at
> > this timing while leaving other general issues, I believe we
> > should give sufficient explanation for that.  In my understand,
> > the main point is:
> 
> > - this can be used as a route injection attack from a malicious
> >   host to a router (not to another host)
> 
> Observation: the reason this attack works is because some
> implementations don't scope ND messages properly. That is, they
> allow an ND/NA received on one interface to impact data structures
> that are global rather than per-interface. Logically, ND/NA traffic
> should only apply to the interface/link on which it was received
> on. E.g, consider what happens with IPv6 LL addresses, especially if
> the same address (with a different MAC addr) is used on different
> links.

First of all, now that cert advisories have been issued, I have no
objection to handling this specific point only.

But, I don't agree with your observation above.  This is not because
an implementation don't scope ND messages.  Even if the node only has
a single interface (so no scoping is possible in the first place),
when it receives an NS whose source
- isn't link-local,
- doesn't match any on-link prefix advertised via RA, or
- hasn't been told as on-link via a Redirect,
the node could consider the NS source address as on-link if it
literally applies the definition of on-link as defined in RFC4861,
because it's an address from which a "Neighbor Discovery message is
received".

In fact, *BSDs correctly "scope" ND messages per-link, and it doesn't
confuse the same link-local addresses received on different links wrt
ND processing.  It still believes an unknown address as on-link if it
receives an ND message from that address, based on the literal RFC
definition of on-link.

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to