There’s been a bunch of new information coming out the past few days about the potential big TCP denial of service flaw. The three most informative posts I’ve read are:
- Fyodor’s discussion of either the same, or a similar issue.
- Richard Bejtlich’s overview.
- Rob Graham’s take on the potential attack.
Here’s what I think you need to know:
- It is almost certainly real.
- Using this technique, and attacker with very few resources can lock up the TCP stack of the target system, potentially draining other resources, and maybe even forcing a reboot (which could trash the host OS).
- Anything that accepts TCP connections is vulnerable. I believe that means passive sniffing/routing is safe.
- The attack is obvious and traceable. Since we are using TCP and creating open connections (not UDP) it means spoofing/anonymous attacks don’t seem possible.
- Thus, I’d be more worried about a botnet that floods your upstream provider than this targeted attack.
- This is the kind of thing we should be able to filter, once our defenses are updated.
In other words- a bad attack, but a moderate risk. That said, there are aspects of this that still concern me and we need to keep in mind:
- We don’t know if our assumptions are correct. This could be a different, and much more serious, technique. Such as something spoofable. Unlikely, but possible.
- Nothing says a botnet can’t use this- and thus make filtering and tracing a real pain. Imagine a botnet rotating this attack technique around to different nodes. It could support more efficient botnet attacks, that could then drop to regular flood mode if it doesn’t think the more efficient direct mode is working.
- We don’t know it doesn’t do something to the router infrastructure or passive monitoring. Again, unlikely based on the information released, but I’d hate to dismiss this as concerning until we know more.
- Until there’s some sort of fix, the Defcon network and coffee shops near universities are really going to suck.
- Until this is fixed, small businesses and individuals are the most likely to suffer. An enterprise might be able to detect and drop an attack like this, but individual users and small business don’t have the resources. Get ready for vendor pouncing.