On Mon, 8 Apr 2002, Mikael Olsson wrote:

> > Perhaps.
>
> "Perhaps"? "PERHAPS"?! How do you expect me to be able to roast
> you properly on a measly "perhaps"?! :)

After the wire brush quote, I figured you'd have the ammo ;)

> Question: I believe you've had the chance to go hands-on in real-world
> situations with more firewall makes and models than I have. In my
> opinion, network segmentation (physical interfaces) is easier to do
> on stateful boxen than proxies, generally speaking. Am I just way off
> base here? (Or is there simply a few choice exceptions to this rule?)

I think these days it's about the same degree of difficulty for proxy
products that support multiple interfaces well.  Back when I started with
commercial products, I had as many token ring interfaces as I could stuff
into an IBM 55L doing a bit of both (proxy and filtering) and never had a
problem with rulesets (but I had a pretty static policy.)

> Oh yeah, I just ran across this from you in another subthread
> that I've neglected. I'll just quote it right here:
> > 1. No need to pass DNS to all the clients, eliminating resolver attacks,
> > DNS tunneling, etc.
>
> Since when is passing raw DNS to all the clients a necessity in
> a filtered environment? (And doesn't DNS tunneling work just as
> well through a cacheing resolver?)

If you're filtering without a proxy, the client needs to look up
www.wirebrush.com to start its connection.  In a (non-transparent)
proxy situation, ONLY the proxy must resolve DNS, the client must simply
pass the request to the proxy.

> Unless, of course, you mean the (not so common any more?) case where
> all internal clients are configured to access the outside world
> through "real" proxy mechanisms rather than configuring your proxy
> firewall to be transparent to the inside.

Exactly.

> > 2. Ditto ICMP without breaking unreachables, etc.
>
> I'm not quite following you here.
> Unless of course you're talking about a transparent proxy
> doing some ICMP error magic that I'm not familiar with,
> in which case you just invalidated your previous point :)

ICMP unreachables to the external interface is much less dangerous than to
every client.

> > 3. IP transport layer issues such as the old URG pointer stuff in Windows
> > products that are unknown by the designer don't get to kill all the
> > "protected clients."
>
> Ooold! You need to give me something better! :)

Old, but still probably a default block in mainstream commercial products.
It's illustrative of the transport layer issues surrounding many clients
versus one client.

> > 4. Fragment overlap attacks and attempts aren't handled inconsistantly by
> > different "protected" machines since they're all handled by the proxy.
>
> (Warning: generalization based on what I've designed myself. Again,
> I should probably go do more testing with current makes and models.)
> Don't most of the "recognized" SPF vendors (pseudo-)reassemble and
> reorder properly by now?

Reordering "properly" is OS dependent.  That's why multiple overlapping
frags are "interesting" IDS attacks.

> > 5. No need to worry about sequence number preditiction attacks for
> > thousands of clients, just one host.
>
> Hmm.. I realize that the following has a very real potential of making
> me look like a marketroid. I apologize for this up front; I just don't
> know of another example.
>
> We offset all sequence numbers by random numbers generated by
> the Yarrow PRNG. I'd say that this is an order of magnitude better
> than what general-purpose TCP/IP stacks in general do.
> (OTOH, maybe this isn't a very common think for SPFs to do.)

This is indeed not very common.

> > 6. Potential to "fix" contetent and protocol issues at the proxy
> > instead of normally having to do so at each individual client.
>
> Score one for properly ALG-enabled firewalls (and I'll readily
> agree that grepping for strings in raw TCP isn't "properly").
> This all given a responsive enough vendor, of course. (Or the
> possibility to roll your own; score one for open source, for
> giving people the possibility to shoot themselves in the foot
> rather than having a vendor do it for them :))

Having had to do this once or twice, I put this one high on my list
(fixing, not shooting.)  That's why after having played with most of the
commercial products, I still lean toward source-available implementations.

Paul
-----------------------------------------------------------------------------
Paul D. Robertson      "My statements in this message are personal opinions
[EMAIL PROTECTED]      which may have no basis whatsoever in fact."

_______________________________________________
Firewalls mailing list
[EMAIL PROTECTED]
http://lists.gnac.net/mailman/listinfo/firewalls

Reply via email to