Re: CERT vulnerability note VU# 539363

["Paul D. Robertson" <proberts () patriot net>]

On Wed, 16 Oct 2002, Mikael Olsson wrote:

> Although this is something that people need to keep in mind when 
> picking / designing a firewall, I'd argue that anything north of
> a stateless packet filter is going to be vulnerable to these sort
> of attacks.  

So will anything south of a firewall- hosts aren't immune to flooding 
attacks either, with our without state, nor are routers...

> If you keep state, you will be vulnerable to state table overflows. 

I don't know that "overflow" is the right word here, "exhaustion" seems 
more fitting.

When I first looked at this, I kind of shrugged and said "So what?"  the 
firewall is doing its job- stopping packets when there's an attack-

The issue is more of how easy that is, than the fact that you can do it.
So the real message here is know the failure mode and capacity limits of 
the firewalls you use.  

If you're being attacked, the firewall not allowing new traffic is 
probably not a bad thing.  For most folks, the ability to create a new 
state table entry is an "outbound traffic only" issue for firewalls that 
aren't "protecting" external services like Web servers.

If you're hosting public resources behind the same firewall that's 
protecting everything else in your enterprise, you've probably made a 
questionable architectural decision.  If you're keeping state on say 
inbound SMTP traffic, the question is "Why?"  If the 'Net as a whole can 
connect to something, the state itself isn't going to do much good.  If 
you're trying to rewrite sequence numbers because of a host that talks to 
the public with high predeictability, again you're probably made a 
questionable architectural decision.

Public-talking hosts should be protectable with simple non-stateful packet 
filtering rules- *especially* those which allow the untrusted side to 
initiate connections.

> Period.  The only real question is: how much work does the attacker 
> need to put in before it becomes painful for the networks that the 
> firewall is protecting?  Is being able to resist a  1 Mbps stream 
> (~4500 pps) "Not vulnerable"?  Is being able resist a 34 Mbps stream
> (~150 kpps) "Not vulnerable"?  Or should every single firewall
> vendor report in and say "Vulnerable", and describe what the limit is?

That's also a scale thing- if a firewall is in front of 10 hosts, the 
effort to protect them from floods might be scalable to 10x, but if it's 
1000 hosts, the ammount of protection is amost certainly going be be less 
than 1000x[1].

> And, yes, ALG-only firewalls can also be overloaded. It's just a 
> different type of 'state'.

Anything without some sort of artificial rate limiting can be DoS'ed.

What this really says to me is "Don't keep state on stuff that doesn't 
*need* it."  Possibly combined with "if you have a large number of 
untrused users, make sure your policies let you disconnect them if they 
cause trouble, and have enough diagnostic infrastructure to be able to 
figure out where an internal attacker is (personally, I prefer 
losts of routers.)"  

As far as ALG state goes, at least the ALG is the final determination 
point for the traffic, so it can deal with many issues (such as the CRC 
one detailed in the vulnerability note) immediately, rather than having to 
rely on a network reply from a host, or in some cases, just not knowing.

Paul
[1] There are products which scale higher, but they tend to be the kinds 
of things you put in front of DSLAMs and cost several hundred thousand 
dollars each.
-----------------------------------------------------------------------------
Paul D. Robertson      "My statements in this message are personal opinions
proberts () patriot net      which may have no basis whatsoever in fact."
probertson () trusecure com Director of Risk Assessment TruSecure Corporation

_______________________________________________
firewall-wizards mailing list
firewall-wizards () honor icsalabs com
http://honor.icsalabs.com/mailman/listinfo/firewall-wizards