Re: new revision of rfc2462bis is available

[JINMEI Tatuya / 神明達哉 <[e-mail]>]

>>>>> On Tue, 10 Feb 2004 20:56:12 +0100, 
>>>>> Felipe Alfaro Solana <[e-mail]> said:

> I don't understand how is it possible for two hosts to communicate when
> no stateful server or router advertising a prefix is attached to the
> link. In this case, both hosts are configured exclusively with
> link-local addresses, but since those addresses have a fe80::/10 prefix,
> how can the IPv6 stack know what interface should the packets be sent
> on?

Just because link-local addresses has a well-known prefix fe80::/10
cannot be the source of the problem.  The problem is that link-local
addresses are only unique within a single link and are ambiguous for
a multi-linked host.  If the host in question has only one physical
interface to which it is connected to the link and the interface is
specified as "the default" (see draft-ietf-ipv6-scoping-arch-00.txt
for more details on this), link-local addresses work just like global
addresses.  Even if the default interface is not specified, link-local
addresses will still work when the user specified the correct link
(interface) (again, see the scoping-arch draft for details).

> For instance, on Linux 2.6 kernels, when trying to ping another host
> using its link-local IPv6, an invalid argument error is generated since
> the kernel is not able to guess which interface the packet should be
> sent to, since any local interface has exactly the same prefix and
> length. This gets complicated even more if the machine has multiple
> interfaces, for example, two Ethernet interfaces.

This is just an implementation.  For example, *BSDs have the notion of
the "default interface" (and the default link automatically derived
from the interface) to implement the scope-arch draft, with which the
above scenario works perfectly.

So,

> I thought that two IPv6-enabled hosts could only communicate if both
> have a site-local or global IPv6 address configured.

this is not really correct.  Also, site-local addresses even have the
same problem of ambiguity.  And besides, the IETF is now going to kill
site-local addresses.  Let's just concentrate on link-local and global.

					JINMEI, Tatuya
					Communication Platform Lab.
					Corporate R&D Center, Toshiba Corp.
					[e-mail]

--------------------------------------------------------------------
IETF IPv6 working group mailing list
[e-mail]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------