From: qv@iol.unh.edu (Quaizar Vohra)
Message-Id: <9609091816.AA09219@iol.unh.edu>
Subject: (IPng 2091) Re:  Some questions
To: ipng@sunroof.Eng.Sun.COM (ipng)
Date: Mon, 9 Sep 1996 14:16:49 -0400 (EDT)
Content-Type: text
Sender: owner-ipng@sunroof.Eng.Sun.COM
Precedence: bulk

Hi Peter,
 > Hi Quaizar,
 > 
 > > 3)Is there a document on source address selection in IPv6. I have 
 > > some concerns regarding this issue. I think we should have atleast a
 > > set of guidelines regarding this. For example :
 > > 
 > > Consider the scenario of a dual node A which has a tunneling interface
 > > with a v4-compatible address ::A and a global IPv6 address G. Now
 > > consider an isolated host B with a v4-compatible ::B. Now A has a UDP
 > > server application running which is bound to a well known port and the
 > > wildcard address i.e. all zeros IPv6 addresss. Now a client
 > > application on B sends a packet to the server appl. on A using A's
 > > global address G (which it got via DNS) as destination. When the
 > > server application replies, it will use ::A as source address if
 > > source address selection is done in the same fashion as is being done
 > > for v4, as I can see on BSD based code, i.e use the address of the
 > > outgoing interface address as source unless specified by the user
 > > explicitly.
 > 
 > I don't see anything wrong with this. It seems like a reasonable source 
 > address selection algorithm to pick an IPv4-compatible source address when 
the 
 > destination address is IPv4-compatible. There is no such thing as a "reply" 
in 
 > UDP, so if you want a request/reply protocol, you need to do that on top of 
 > UDP. So I think in this case the application should have provided the 
source 
 > address.
	
	I agree with you here. My purpose was to find out what is the
consensus on correct behaviour.


 > 
 > I am, however, not comfortable with the view that a tunnel interface is a 
real 
 > interface with an IPv6 address. To me, a tunnel interface (or 
 > *pseudo*-interface, really) is just an implementation trick, and I don't 
see 
 > why it would need any particular kind of address. If you don't want node A 
to 
 > use ::A as source address, then A shouldn't have an interface with that 
 > address! (A could still tunnel to B, since it has an IPv4 address.)
 > 
 > Peter
 > 

Well a while ago I argued that we don't need v4-compatible addresses
(though now I agree that they are useful on isolated nodes) for
similar reasons.

My real worry is that do we move away from the IPv4 BSDish behaviour
of using an address on the outgoing interface and how ?

Obviously in IPv4 there was no notion of scopes and uncomfortable
addresses like v4-compatible, but now since such notions exists
we need to have a different algorithm for source address selection.

If we still follow the old way, we will end up choosing link-local
address as source in most cases.

Now my concern is should we use source address, other than 
any address on the outgoing interface if forced. For example,
if I have a node with 2 interfaces X & Y. X is assigned a global
address G while Y doesn't have a global address but a site-local
address S. But my default route is thru Y. If I am sending a packet
to an off-link global address which matches the default route, I
should be using G rather than S. Though common-sense suggests that
such cases should not occur, but still I want input from people with
more experience. 

It makes one feel that these global addresses were more anycastish 
with one per node as a whole ;-). 

Also I feel that nodes with global addresses e.g. ISP based should
keep away form using v4-compatible address as source. I think it is
time that someone defines clearly the role of v4-comptible addresses.

Quaizar


------------------------------------------------------------------------------
IETF IPng Mailing List		      FTP archive: ftp.parc.xerox.com:/pub/ipng
IPng Home Page:          	      http://playground.sun.com/ipng
Direct all administrative requests to majordomo@sunroof.eng.sun.com