(usagi-users 03429) Re: Source address for neighbor solicitations

[Ignacy Gawedzki <ig@xxxxxx>]

On Fri, Jun 24, 2005 at 10:25:42PM +0200, thus spake Peder Chr. Norgaard:
> I don't think I really understand the context of your question.  (I assume
> you are talking IPv6, yes?).  What are the "routers" that you are writing
> about?  It cannot be linux 2.6.something kernels - they certainly have
> link-local addresses on all IPv6 interfaces.

They are linux 2.4 kernels in fact, but I suppose they could easily be 2.6
too.  An interface gets assigned a link-local address automatically as soon as
it's set up, but you can ever remove that address manually.  By "routers" I
mean linuces that have forwarding activated.

> There is no such case in IPv6 that allows a router (or a host) to not have
> a link-local address on any IPv6-capable interface.  If the router hasn't
> a link-local address is simply isn't IPv6, so it is quite irrelevant to
> ask "in a case a router doesn't".  In that case, you really should return
> the router to the vendor, require refund of your payment and don't worry
> any more.

I'm talking about the case when some misconfiguration has taken place.  Of
course we are out of what RFCs require, but this doesn't mean we can't
cope with this in a smart way.

> (practically, IPv6 routers are using their link-local addresses heavily in
> OSPFv3, so an IPv6 router with no link-local addresses wouldn't be able to
> run OSPFv3.....)

Of course, but I'm not talking about integrated boxes.

> On the basic level, the RFCs are truly a bit vague on the topic of source
> addresses in Neigbor Solicitations from a router, when the NSs are
> triggered by forwarding packages.  The RFCs really don't say what to use -
> but my general understanding of IPv6 tells me that that the link-local
> address is right thing to use in this case.  Mainly because it should
> work, and, in the forwarding situation, there really is no mechanism to
> choose between any of the potentially several globally routable addresses
> assigned to an egress interface.

Great, I agree, but my point is that in case that interface doesn't have any
link-local address assigned, common sense suggests that any other address
assigned to that interface should be used.  This doesn't break RFC conformance
and *may*, in some cases, improve things.  So why not to take the opportunity?

-- 
I have not lost my mind, it's backed up on disk somewhere.