Question on address configuration

[Michel Py]

Has someone try to spin that idea in the ipv6 WG?

Michel

-----Original Message-----
From: Pekka Savola [mailto:pekkas@netcore.fi]
Sent: Monday, January 07, 2002 12:23 AM
To: Nathan Lutchansky
Cc: Edward Lewis; 6BONE List
Subject: Re: Question on address configuration

On Sun, 6 Jan 2002, Nathan Lutchansky wrote:
> On Sat, Jan 05, 2002 at 07:45:59PM +0200, Pekka Savola wrote:
> > On Fri, 4 Jan 2002, Edward Lewis wrote:
> > > E.g., Could (linux) ifcfg-eth0 have a line IPV6ADDR="::3" which tells the
> > > router solicitor to append that to the router's prefix?
> >
> > No, that isn't possible.
> >
> > I think this would be an interesting development; basically a simple
> > implementation of user-space router solicitation client would be required.
>
> I should think the better solution would be to modify the IPv6 stack to
> generate global addresses by combining advertised prefixes with the lower
> 64 from each link-local address.  Then you can add all your well-known
> interface IDs as link-local addresses and your globals will just appear.
>
> Thus, if I have eth0 with
>
>   - autoconfigured address       fe80::2a0:56ff:fe94:4444 and
>   - manually configured address  fe80::1,
>
> my global addresses would be generated from the advertised prefix of
> 3ffe:2900:f10a:701::/64 to give
>
>   - 3ffe:2900:f10a:701:2a0:56ff:fe94:4444 and
>   - 3ffe:2900:f10a:701::1.
>
> Now I have a well-known interface ID used in an autoconfigured prefix.
>
> I haven't thought about this enough to find any problems, and I haven't
> checked to see if RFC 2462 would even allow this approach.  But I can
> certainly see benefits to doing autoconfiguration this way.  -Nathan

This is an interesting idea, and IMO, might be nice to have as a on/off
togglable feature at least.

The problem is that if someone adds a manual address to the link-locals,
he might not want it in globals.

One additional (but perhaps minor) problem here is DAD.  If the node has
fe80::1 as its address, it has already performed duplicate address
detection for it.  By the spec, the DAD check can be omitted if the suffix
is the same, so no DAD would be performed if one configured 3ffe:ffff::1.
For global addresses, this might be worse than for link-locals.

But then again, DAD isn't (error/fool)proof anyway, so I don't think this
is a too big issue.

--
Pekka Savola                 "Tell me of difficulties surmounted,
Netcore Oy                   not those you stumble over and fall"
Systems. Networks. Security.  -- Robert Jordan: A Crown of Swords