On Fri, 12 Apr 2002, Mikael Olsson wrote:

> > [on reassembly DoS]
> > I wonder if it's really better under 2k+patches, or if the 
> > threading is just better?
> 
> It isn't a total system DoS. It just DoSes fragment reassembly.
> The CPU load of the target system won't even increase a single
> percent.

Right, but I was wondering if the frag reassembly code is now threaded 
where it wasn't before.

> > Ah, but since in my case, the software wasn't proxy software, it 
> > doesn't count as part of the architecture, where as in your case 
> > it does! :)
> 
> The barbiesoft URL filter wasn't a proxy? *boggle* :)

Nope, it was a content filter sitting behind a proxy :-P

> 
> > > in these layers, there is no "real" difference in the
> > > protection that a proxy firewall affords compared to an SPF. [3]
> > 
> > The proxy is the only client that needs patching in client-side issues,
> > and the only place that needs content filtering in content issues.  *Some*
> > content issues can be negated with an SPF, but not all and certainly the
> > range of client issues that can be fixed is significantly less.
> 
> I was talking about L3 and L4 only in this paragraph.
> And even when/if a problem is found in L3/L4 filtering/logic
> on an SPF, the SPF is the only box that needs to be patched.

Ah, but a firewall isn't just a L3/4 protection device, and clients aren't 
just L3/4 devices.  The essence of "more secure" is "less places 
exposed to high risk."

> Assume two boxes want to speak NetBIOS to eachother. (Yes, I know, 
> horrid. Let's assume that the server is a very stripped-down samba.)
> 
> Assume box 1 behind if1 has IP 1.2.3.9, and wants to communicate
> with hosts behind if2 with IPs 1.2.3.1--254 (sans .9 of course).
> Tell me how a host route on _an available proxy firewall package_ 
> solves this.

Absent the broadcast stuff, proxy ARP for the target victim and 
something plug-gw-ish should work just fine.  I'm pretty sure I could write a 
"transparent" proxy that would include the broadcast stuff (and MAC target the 
broadcasts to the specific victim on the other end (SOCK_RAW is your friend.)

I have zero experience with any available firewalls which cliam NetBIOS 
proxy support, so I can't say how/if they'd make anything possible- 
there's no way in hell I'd ever let it in/out through a firewall.  If they 
work with subnets though, I can't see a reason they wouldn't work in a 
host-specific scenerio unless there's a broadcast issue- and most of that 
should be solvable with WINS or LMHOSTS unless there's something 
broadcasty other than name service about NetBIOS (it's been years since I 
had to support NetBIOS.)

> Yes, I know, you can always get more routable space. Or rearrange
> your IP allocations. But I've seen enough cases where people
> DON'T segment their servers as much as they'd really like, just
> because firewall boxes with rigid subnetting rules and poor
> segmentation support functions puts the effort bar too high.

Subnetting support is an OS function- ever since Solaris 2.6 the Variable 
Length Subnet Mask problem hasn't been an issue for anything I can think 
of, and before 2.6 you could use a VLSM as long as it was for a local 
interface (which was your scenerio,) the route command simply didn't 
support them before 2.6.   Once again, it's also possible to 
hyper-address a physical network with multiple logical networks.  Since I 
don't know anyone sane who'd still be running a firewall on Solaris 2.54, 
I think you lose this one. :)

> > > - And.. important.. speed. When you segment dozens of servers, they
> > >   very often still need to be able to talk to each other at high
> > >   speeds. (Think web server -> backend DB. Think backups.)
> > 
> > Speed is the only point I'll readily concede.
> 
> Hey. I just thought of another one:
> - For many protocols, a thorough ALG is a very complex beast. Enough
>   so that at least _I_ don't want to trust it to be immune to 
>   exploits that give you (partial) control of the machine it is
>   running on. 
>   (Very good case in point: a content and virus scanning HTTP proxy.)
>   If I center my firewalling environment around an SPF with good
>   segmentation support functions, I can (easily) put such a box on a
>   separate segment and only allow required communication to and 
>   from that box. With the (not poorly written) SPF sitting on 
>   a separate box, I can be reasonably assured that said proxy box
>   won't gain control over the entire packet flow, or be able to 
>   connect anywhere it damn pleases.

If you're going to add additional boxes, I can proxy-to-proxy my ALG to a 
content inspecting thing on a different subnet as well.

> > > [on the protection that internal boxes need]
> > > 2) Analyzed on the application layer.
> > >    I believe that this is best done by specialized relays/proxies
> > >    (NOT! a proxy firewall with everything from a single vendor or
> > >    organization!).
> > 
> > A proxy firewall doesn't have to be a single box, and a single box doesn't
> > have to run everything from the vendor.  For instance, ever since I
> > started using Postfix in production (2nd Alpha version), for instance,
> > I've ripped out the native SMTP proxy and replaced it with Postfix
> 
> I still think that you're talking about the same thing as I. :)

Yes, but the *important* point there is that with an ALG you can pick the 
BEST per-protocol gateway available.  With an SPF, you're stuck with 
whatever protocol support the vendor chose (e.g. I could put Raptor's SMTP 
gateway on a Guantlet, while you couldn't put Checkpoint's SMTP stuff on a 
PIX.)

> > you can't mix and match pieces of an SPF on a per-protocol basis.
> 
> I believe this is a good quality in SPFs. I want my "traffic control 
> filtering box" to be compact enough that I can assume that it does 
> what it is told and nothing else. If equipped with login shells,
> multiuser environment, compilers, etc etc etc, that you'd need for 
> installing postfix on it, this is no longer true. Not by a long shot.

I don't need a compiler, or a multi-user environment to install an ALG on 
a target host.  I could use the same sort of maintenance and 
configurations an SPF uses, but the point is moot- I don't *need* remote 
management on either type of firewall, so the exposure is able to be 
limited to be pretty-much nil (even staying away from B2ish ALGs, which 
are a whole new set of fun, and provide an interesting level of 
segmentation that doesn't happen in SPFs!)

> > Sure, and my firewall implementations always include some sort of 
> > packet filter (preferably with state)- generally IPFilter/NetBSD 
> 
> See? :)

What, that hybrids are a solution?  That I don't rely soley on SPFs? 

> > [0] Really, check again[2]!
> > [2] There was no [0].
> 
> *bzzt* brain stalled.. *bzzt* infinite loop.. *bzzkktt* *boom*

's watcha get for including footnote references without including the 
footnotes :-P

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