SMB_RELAY

[natronicus at gmail.com (natronicus)]

DNS rebinding allows you to rebind the name attacker.com to the IP address
of the local host.  This works like so:

1. Victor loads malicious java applet from http://www.alice.is.evil.com
2. The applet code looks up the local ip address with a .getlocalhost() or
similar method and returns it to Alice
3. Alice rebinds the domain name Alice.is.evil.com from its internet address
(e.g. 64.233.167.99) to to Victor's internal IP address (e.g. 10.10.10.10)
4. Victor (through Alice's applet) opens an iframe to http://ALICE
5. The applet starts spamming NBNS answers for the hostname "ALICE" with IP
address 64.233.167.99.
Note: This is possible because Windows tries DNS first, which fails, then
tries NBNS/WINS.  NBNS does not randomize transaction IDs under Windows,
starting at 0x8000 on boot, and windows doesn't care that it received the
answer from its own IP address.
6. Victor maps ALICE to Alice's internet IP 64.233.167.99, but thinks ALICE
is local because it received the answer through NBNS and because there is no
TLD.
7. Since Victor thinks ALICE is local, it will auto authenticate via IWA
... and the attack proceeds as outlined in the previous email.

Variations: You don't have to use NBNS.  You can use dynamic SOA updates to
DNS if the client's network is allowing Active Directory to manage DNS.
Often times, if the name ends in the same domain name it will place it in a
higher trust zone.

Also, all the trickery above does is place alice's IP address in a higher
zone of trust.  This often allows you to do things like auto-load unsafe
activex controls.

natron

On Mon, Mar 10, 2008 at 10:49 AM, Kurt Grutzmacher <grutz at jingojango.net>
wrote:

> Your memory is correct. The classical SMB Relay attack assumes that the
> user is a local administrator of the workstation they're coming from. Since
> I have been in more environments lately where this is no longer the case
> I've shifted the attack to some of the foundational services that use NTLM
> authentication: IMAP, POP3, HTTP, etc.
>
> - Bob talks HTTP to Alice
> - Alice returns HTTP NTLM Authorization header, closes HTTP connection
> - Bob talks HTTP to Alice and sends Type 1 header
> - Alice initiates POP3/IMAP/SMTP/etc to Victor, passing Type 1 message
> - Victor responds with Type 2 message, including challenge
> - Alice sends Bob Type 2 message in HTTP header
> - Bob receives and processes request, sends credentials to Alice in Type 3
> message
> - Alice ends Type 3 message to Victor, checks for authorization
> - Alice closes HTTP session with Bob
> - Alice sends commands as Bob to Victor, wash, rinse, repeat
>
> http://grutz.jingojango.net/exploits/psyduck-pop3.rb
>
> When you and I last talked about the DNS Pinning issue I was on-board and
> did some research and talked with another researcher doing NTLM/HTTP work.
> We've come to the same conclusion that DNS pinning probably will not work
> for this attack. Pinning changes the IP address of an already-known DNS name
> and by the IE has already put it in its specific trust zone. If you've got
> some code or something that shows this changing because of pinning, I'd love
> to see it.
>
> Right now this attack, like SMB Relay, only affects an Enterprise.
>
> Enterprise may be 50,000+ users of course. :)
>
> Kurt
>
>
> On Mon, Mar 10, 2008 at 7:56 AM, natronicus <natronicus at gmail.com> wrote:
>
> > An NTLM-over-HTTP implementation would allow more than server attacks
> > that require authorization (although this is handy for that, too).  You can
> > perform the regular SMB_RELAY attack, but you're substituting the attacker's
> > SMB server for an HTTP server, thus getting around the port restriction
> > problems.
> >
> > It's been a few months since I've looked at the protocol and how this
> > attack works, but if my memory is correct, it is functionally:
> >
> > 1. (HTTP) Victor clicks link / previews email / etc and issues a GET to
> > http://alice, which requests Integrated Windows Authentication (IWA)
> > with NTLMv1.
> > 2. (SMB) Alice begins an NTLMv1 SMB connection to Victor
> > 3. (SMB) Victor replies to Alice with the challenge hash
> > 4. (HTTP) Alice replies to the IWA request with the challenge hash just
> > received
> > 5. (HTTP) Victor supplies the answer to the challenge request
> > 6. (SMB) Alice copies the answer from the HTTP stream to the SMB stream
> > and authentication occurs
> > 7. (SMB) Alice uploads the payload and executes
> >
> > From a functional perspective, the attack is exactly the same, but uses
> > port 80 on the attacker's box instead of the problematic 137-139 or 445.
> >
> > Additionally, with some java applets and DNS rebinding chicanery, you
> > can extend this attack over the internet.
> >
> > Natron
> >
> >
> > On Sun, Mar 9, 2008 at 3:08 AM, Kurt Grutzmacher <grutz at jingojango.net>
> > wrote:
> >
> > > I've done the work to get NTLM Type-message processing into MSF. At
> > > this tim there aren't any exploits within MSF that use the library, I just
> > > referenced it from some external ruby code I wrote but we should be able to
> > > integrate client-side NTLM-over-HTTP fairly easily for server attacks that
> > > may require authorization. I just haven't put it on the top of my list yet.
> > >
> > > http://grutz.jingojango.net/exploits/pokehashball.html has some of the
> > > information along with two exploits (hash grabber and HTTP-to-POP3 proxy
> > > exploit).
> > >
> > > If anyone wants to work on implementing any exploits, let me know and
> > > I'll work with you.
> > >
> > >
> > > 2008/3/7 natronicus <natronicus at gmail.com>:
> > >
> > > Is there a particular reason you're trying to use Windows for this
> > > > one?  I tried to mess with implementing NTLM-over-HTTP / Windows Integrated
> > > > Auth a few months back, but got frustrated learning Ruby and it hit the
> > > > projects-to-finish-later pile.  I recently saw HD mentioned in another
> > > > thread someone was working on this problem, but it sounded like they may be
> > > > focused on other items first (NTLMv2, for example).
> > > >
> > > > In any event, until the HTTP version is implemented, you're always
> > > > going to have problems getting it to work on Windows, because Windows is
> > > > incredibly greedy about those particular ports.  Why not use a Linux image
> > > > in VMWare instead?  If your network will allow 2 IPs for 1 MAC address,
> > > > there's no reason why you can't use it under (e.g.) Backtrack and
> > > > still have access to whatever you need Windows for.
> > > >
> > > > Just a though,
> > > > N
> > > >
> > > > 2008/3/7 Karlsson Anders <Anders.Karlsson at atea.com>:
> > > >
> > > > >  And it is realy hard to use port 445. I needed to disable almost
> > > > > every service and binding in my XP machine. After that I can not use the
> > > > > machine to connect to the server with old plain "net use", so I do not think
> > > > > using port 445 is the right way....
> > > > >
> > > > > /A
> > > > >
> > > > > _______________________________________________
> > > > > http://spool.metasploit.com/mailman/listinfo/framework
> > > >
> > > > _______________________________________________
> > > > http://spool.metasploit.com/mailman/listinfo/framework
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.metasploit.com/pipermail/framework/attachments/20080310/b5290141/attachment.htm>