[Dshield] FW: [Full-Disclosure] DCOM RPC exploit (dcom.c)

Paul Marsh pmarsh at nmefdn.org
Mon Jul 28 20:29:52 GMT 2003

I don't know how many of the Dshielders also watch FD but the kids are hard at work trying to get this one to work.

Thanx, Paul

On Mon, 28 Jul 2003, Schmehl, Paul L wrote:

> > 2) For this DCOM RPC problem in particular, everyone's
> > talking about worms.  How would the worm know what return
> > address to use?  Remote OS fingerprinting would mean it would
> > be relatively large, slow, and unreliable (compared with
> > Slammer), and sticking with one would cause more machines to
> > just crash than to spread the worm.  I haven't looked into
> > this very closely yet to see if it can be generalized.
> What fingerprinting?  If you've got 135/UDP open to the Internet, you're
> screwed.  Slammer didn't fingerprint.  It simply hit every box it could
> find on port 1434/UDP, and the exploit either worked or it didn't.  Most
> worms do the same.  They attack indiscriminately, and infect those Oses
> that are susceptible.  And with Windows, that's enough boxes to cause a
> real problem.

Thanks for responding.  I realize that having 135 open on any Windows
machine makes you vulnerable, and that you wouldn't need to differentiate
Windows/OtherOSes.  My question is about different Windows versions.  The
version (NT/2000/XP), service pack, and language at least have to be known
to get the return address right.  If it's "guessed" wrong, the system goes
down with no shell executed.

Any worm using this would need to know the return address before
attempting to exploit If a worm were to stick to targetting one return
address (say, English XP  SP1), everytime it ran across something slightly
different (SP0, german, win2k, etc) it would simply crash it and not
spread.  One of three things would happen in the case of this worm :

1) Sticks with one return address, makes a spectacular DoS against all
other languages/versions/SPs.  This could limit how quickly it spreads.

2) Somehow finds out ahead of time what the remote language/version/SP is.
Could be very unreliable and slow.

3) There is some way of generalizing the return address in a way that
would work on at least a large portion of installs.  This is what would
bring it into the league of Very Scary Worms.

Has anyone seen any indication in the private exploits or in their
research that there's a way to get it to work reliably on systems without
having to know version/SP/etc?

Full-Disclosure - We believe in it.
Charter: http://lists.netsys.com/full-disclosure-charter.html

More information about the list mailing list