Let me weigh in late to offer an answer to the core question: Where would I
use a site-local address?

In a network with a statically allocated global IPv6 address, site-locals
are indeed redundant.  But:

- IPv6 is built on the assumption that networks may renumber

- Leaf networks currently are, and in future may well be, constructed with
transient prefixes.

- These leaf networks may not always have global connectivity.


Consider a network using a dialup IPv4 connection to the router.  While this
connection is up, the network has a valid 2002:....:..../48 prefix.  If the
network goes down, or changes IPv4 address (and yes, this does happen),
suddenly all our prefixes change.

Prefix renumbering isn't too bad, since our IPv6 apps are supposed to be
able to handle this.  But while there is no prefix, there's no addresses. 
Connectivity over the site is lost pending availability of a new global
prefix.

Enter site locals.


A solution is to allocate a site local prefix for the leaf network, as well
as any global prefix/es that happen to exist.  All these prefixes are given
to the routers and hosts within the site.  The border routers don't route
site local addresses outside the site, and should also drop any packets with
a site local source address.

Sensible host source address selection rules are required.  eg: prefer the
source address that matches most closely with the destination address, and
prefer globally scoped to site-local addresses (unless perhaps the target is
also site local).


Naming also can be site scoped.  See:
  draft-williams-dnsext-private-namespace-01.txt
for a site-scoped naming proposal.


Under this model, there really is no need for multiple site-local scopes. 
The bits between /16 and /48 can be used to subdivide the site local space
for manually partitioning, or else the full fec0::/48 space can be used and
include whatever it can reach.

As to router defaults, I'm inclined to suggest that backbone routers should
perform site-local filtering unless explicitly configured to allow it
between particular interfaces.  Routers used within a site should freely
route, but it's probably safer to require site-local forwarding to be off by
default.  People wanting more sophisticated configurations (with multiple
different interfaces in multiple different sites) can manually configure how
packets are to be routed.


Site scoping boundaries can be bypassed through either tunnels or
encapsulating addresses in higher level protocols.  However, in this case
things will simply fail to work, which is acceptable.  With some effort,
spoofing attacks are possible, but it is well known that hosts should not
trust simple address based authentication for serious security.

-- 
Andrew White                [EMAIL PROTECTED]
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to