Margaret's draft highlights three failures of site local addressing...

(1) The addresses are unreachable outside of their original context.

(2) The addresses may overlap other private address spaces, creating
ambiguity.

However, it seems to me that #1 is not specific to site local addresses. 
Instead, it applies to ANY address that is not practically globally
routable.

What do I mean by "practically globally routable?"  That not only is it
allowed to be globally routeable, but that there's not a filter somewhere on
the way to the default free zone that will kill it.  This is an address's
'practical scope' - the area of the internet that it could reasonably apply
to.

Once we start handing out addresses whose practical scope is less than
global, we either need proxies (eg NAT) or multi-homing.  Few to none of the
limited scope or multi-homing problems are specific to site locals.  In
fact, site locals have a slight advantage in a multi-homing situation in
that they promise that they do not have global scope, and thus may be
singled out by address selection algorithms.


To my eye, ambiguity problems manifest themselves in three main situations.

- Addresses used in the wrong context may arrive at an unintended network or
machine.

- Reverse name lookups don't work globally scoped ambiguous addresses.

- Site border nodes.

The IPv6 machine ID helps reduce the likelihood of a misdirected connection
actually arriving on an interface (rather than just timing out).  Beyond
that, it's up to an application to decide whether it wants to consider the
failure case where TCP/IP connection is successful but then the machine
fails to authenticate itself.  Note that this same problem could occur for a
variety of other reasons that ambiguous addresses (deliberate spoofing, bad
DNS entries, other network configuration problems).

Proper reverse name lookups rely on sites using site locals correctly
configuring their internal DNS.  Forward lookup issues can be trivially
solved by putting internal addresses in a localised namespace.

Site border nodes are always going to be problematic.  Logical partitioning,
as in Bob Hinden's draft or Hiroki Ishibashi's implementation, and refusing
to allow sites to 'overlap' seems to be an effective way of avoiding these
problems.


I feel the correct conclusion from Margaret's draft is that there are a
number of issues that people using addresses with "site" based practical
scope should be aware of.  However, that people WILL use firewalls and
address filters as one part of security management seems inevitable.  Thus
it's a useful summary of "these things may break, understand the issues".

But despite the title, site-local addresses are only one example of a
broader category of addresses that these problems apply to.  Site locals as
currently specified do add additional issues for architecture (DNS and SBRs)
if not correctly managed.  However, at the client and application level,
most site-local alternatives suggested here (everything except fully
provider independent routed addresses, without address-based filtering) fall
foul of the same problems as site local addresses themselves.

-- 
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