NAT66 was Re: using "reserved" IPv6 space

NAT is good for getting the return traffic to the right firewall. How
else do you deal with multiple firewalls & asymmetric routing?

Yes, it's possible to get traffic back to the right place without NAT.
But is it as easy as just NATing the outbound traffic at the
firewall?

Lee

>
> Why would you want NAT66? ICK!!! One of the best benefits of IPv6 is being
> able to eliminate NAT. NAT was a necessary evil for IPv4 address
> conservation. It has no good use in IPv6.

NAT is good for getting the return traffic to the right firewall. How
else do you deal with multiple firewalls & asymmetric routing?

Traffic goes where the routing protocols direct it. NAT doesn't
help this and may actually hinder as the source address cannot be
used internally to direct traffic to the correct egress point.

Instead you need internal routers that have to try to track traffic
flows rather than making simple decisions based on source and
destination addresess.

Applications that use multiple connections may not always end up
with consistent external source addresses.

Yes, it's possible to get traffic back to the right place without NAT.
But is it as easy as just NATing the outbound traffic at the
firewall?

It can be and it can be easier to debug without NAT mangling
addresses.

The only thing helpful NAT66 does is delay the externally visible
source address selection until the packet passes the NAT66 box.

Mark

If you are running an HA pair, why would you care which box it went back
through?

-Grant

It still doesn't change the arguement. You still need to have flow
based routers or you may choose the wrong egress point and if you
need NAT66 you have 4+ upstream connections though two of them may
be tunnels. You also need a protocol to keep the HA pair state
tables in sync.

Mark

Why would you want NAT66? ICK!!! One of the best benefits of IPv6 is being
able to eliminate NAT. NAT was a necessary evil for IPv4 address
conservation. It has no good use in IPv6.

NAT is good for getting the return traffic to the right firewall. How
else do you deal with multiple firewalls & asymmetric routing?

1. Share state across the firewalls or go with stateless firewalls.
2. Move the firewalls close enough to the end hosts to avoid this problem,
  Keep the asymmetric routing outside the perimeter.
3. Very creative source address selection mechanisms.
4. LISP (if you must).

Yes, it's possible to get traffic back to the right place without NAT.
But is it as easy as just NATing the outbound traffic at the
firewall?

That depends on whose life you are trying to make easy. If you asked the
application developers or the people that have to build all the problematic
ALGs that creates a need for, I'd bet they would have a different opinion
than the guy configuring the firewall.

In terms of overall problems created, cost to the community, increased insecurity,
and the other costs associated with a NAT-based solution, I'd say that it is
a net loss to use NAT and a net gain to avoid it.

From the perspective of the firewall administrator alone without a broader
view of the total consequences, toxic pollution of the internet seems like
a good idea.

Owen

Think HA pairs in Pittsburgh, Dallas, and San Jose.

Now imagine each has different upstream connectivity and the backbone
network connecting all the corporate sites lives inside those firewalls.

The real solution to this is to move the backbone outside of the firewalls
and connect the internal networks via VPNS that ride the external backbone
and can be routed over the internet safely when a backbone link fails.

However, this still requires some interesting effort in terms of source address
selection, routing, etc. in order to avoid triangle routing out of the firewall
in Pittsburgh resulting in a return trying to come in via Dallas or San Jose.

I think in IPv6, as firewall vendors begin ot mature their products, we'll
either see a departure from stateful inspection, or, more likely an ability
to set up HA clusters across diverse geography where state tables are
kept in sync across the WAN.

Owen

Wouldn't this be even easier if you gave each machine involved multiple
addresses, one ULA and one external? This isn't IPv4 anymore, you can
stick multiple addresses on an interface. :slight_smile:

If you are running an HA pair, why would you care which box it went back
through?

Because it could be/is a stateful firewall and the backup will drop the traffic. (FreeBSD CARP)

Cheers,

Seth

Not really... Doesn't help with the situation where you go from
  host->Firewall A-> web server on the external internet
and the response goes
  web server->Firewall B-> X (Firewall B has no state table entry for the session).

Owen

In message
<CAD8GWsswFwnPKTfxt=squUmZofs3_-yriHY8o4Gt3W9+x6fVUQ@mail.gmail.com>, Lee
writes:

>
> Why would you want NAT66? ICK!!! One of the best benefits of IPv6 is
> being
> able to eliminate NAT. NAT was a necessary evil for IPv4 address
> conservation. It has no good use in IPv6.

NAT is good for getting the return traffic to the right firewall. How
else do you deal with multiple firewalls & asymmetric routing?

Traffic goes where the routing protocols direct it. NAT doesn't
help this and may actually hinder as the source address cannot be
used internally to direct traffic to the correct egress point.

_source_ address + 'used internally'?? I like policy based routing
about as much as the more opinionated members of this list like NAT :slight_smile:

Instead you need internal routers that have to try to track traffic
flows rather than making simple decisions based on source and
destination addresess.

Applications that use multiple connections may not always end up
with consistent external source addresses.

In the general case, sure. At work, the only time your external
source address changes is when something quits working and you're
automatically failed over to the working firewall (ha pair).

Yes, it's possible to get traffic back to the right place without NAT.
But is it as easy as just NATing the outbound traffic at the
firewall?

It can be and it can be easier to debug without NAT mangling
addresses.

Yes, there are times when NAT isn't the appropriate solution. I'm not
religious about it.. just saying there's times when NAT is the
simplest/easiest solution.

Regards,
Lee

If you are running an HA pair, why would you care which box it went back
through?

You wouldn't. But if you've got an HA pair at site A and another HA
pair at site B..

Lee

I have almost one hundred FWs. Some physical. Some virtual. Various vendors. Your point is spot on.

-Hammer-

"I was a normal American nerd"
-Jack Herer