Monday, March 23rd, 2015
Last week, ICS-CERT released an advisory on a set of Johnson Control MetaSys vulnerabilities I reported. You can find the advisory here: https://ics-cert.us-cert.gov/advisories/ICSA-14-350-02
It’s interesting to note that my initial email describing the vulnerabilities was sent on November 22nd, 2013. So, 1 year, 3 months, and 23 days later… we finally get a public advisory. The vulnerabilities are extremely simple to exploit (as we’ll see in the post below), but they are also extremely easy to defend/detect. If you are a owner of one of these devices, the facility for which this device supports has been exposed for over a year. If you find these timelines unacceptable, you should call or email your Johnson Controls representative. Let them know that this is unacceptable and their security engineering practices need to be more agile. They obviously don’t care about security researchers, but maybe they’ll listen to their customers.
The Johnson Control NAE is a typical embedded device that you’ll find supporting Building Automation Systems. Here’s a picture of a NAE55.
The NAE55 runs Microsoft embedded standard on an Intel x86 Atom processor. Typically there is a flashdisk which can be accessed at /flashdisk. On the flashdisk, there is a large amount of web related code. Given that web interfaces are almost always remotely accessible, this is an excellent place to find impactful security issues. Some components created by Johnson Controls are written in .NET. While I didn’t find much raw source code, MetaSys makes use of precompiled binaries (dlls) on the NAE55. We can find these precompiled binaries on the flashdisk at: /flashdisk/Storage/Metasys/wwwroot/metasysIII/WS/bin
A quick note to device vendors, shipping .NET code, even in precompiled dlls, is the same as shipping raw source. But your code is tight… so you have nothing to worry about… right?.
There are a large number of binaries here, so where should we start? We’ll break out .NET reflector and we’ll look into a DLL named, “WebServices.Common.dll”. A screenshot of the WebService.Common.dll metadata is shown below.
WebServices.common refers to a number of interesting namespaces. The screenshot below shows the various namespaces.
“Security” is always interesting, so looking at the JohnsonControls.MetasysIII.Security namespace, we see the following:
“AdministationService” exposes several methods including a public method called, “getUserProperty()” getUserProperty accepts an INT value representing the userID and returns a XmlNode.
The implementation of “getUserProperty” is located in the Subsystems.Common assembly:
The Subsystems.Common code can be found in the “Subsystems.Common.dll”. Inside Subsystems.Common is a variety of namespaces including another namespace reference of “JohnsonControls.MetasysIII.Security”
Inside the “JohnsonControls.MetasysIII.Security” namespace is a class named “PrincipalStore”. A screenshot for “PrincipalStore” is shown below:
Inside the “PrincipalStore” class is the actual code for “getUserProperty()”
As you can see, the “getUserProperty()” method exposes user passwords (password hashes). If we can call this method remotely, we should be able to retrieve the password hash for any user on the device (along with a bunch of other user data).
It turns out, we can call this method as an unauthenticated user by making the following web service request (in the example below, we retrieve the details for the all powerful METASYSAGENT account, which is always userID = 1):
POST /MetasysIII/WS/Security/AdminService.asmx HTTP/1.1
<?xml version="1.0" encoding="utf-8"?>
<soap12:Envelope xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema" xmlns:soap12="http://www.w3.org/2003/05/soap-envelope">
This POST request should return something like this:
HTTP/1.1 200 OK
<?xml version="1.0" encoding="utf-8"?><soap:Envelope xmlns:soap="http://www.w3.org/2003/05/soap-envelope" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema"><soap:Body><GetUserPropertyResponse xmlns="http://johnsoncontrols.com/MetasysIII/WebServices/Security/"><GetUserPropertyResult><Stuff><MetasysUser id="1"><userName>MetasysSysAgent</userName><password>MD5-PASSWORDHASH</password><fullName>Metasys System Agent</fullName><emailAddress /><description>Metasys System Administrator</description><singleAccessUser>false</singleAccessUser><temporaryUser>false</temporaryUser><userExpiresDate>2099-02-01</userExpiresDate><passwordExpiresDate>ExpirationDate</passwordExpiresDate><mustChangePassword>false</mustChangePassword><cannotChangePassword>false</cannotChangePassword><accountDisabled>false</accountDisabled><accountLockedOut>false</accountLockedOut><modifyOwnProfile>true</modifyOwnProfile><canViewNavTree>true</canViewNavTree><userDefined>false</userDefined><Roles><Role id="4" /><Role id="1" /></Roles></MetasysUser></Stuff></GetUserPropertyResult></GetUserPropertyResponse></soap:Body></soap:Envelope>
We can increment the “userID” value and retrieve the password hashes for all users on the device. It’s a good thing that integrators always use really strong passwords, otherwise these would be easily cracked
Here is an interesting exercise… if you have MetaSys in your facilities, ask your integrator if you have any devices that are affected by this bug. If you do have vulnerable devices, ask your integrator to install the latest patches. I would also suggest reviewing your web logs to see if anyone suspicious has called the “AdminService.asmx” web service. Given that this bug allows someone to capture password hashes, it’s a good idea to ask your integrator to reset all the user passwords on the device after you’ve installed the security patches… unless of course changing one of the device passwords breaks inter-operability your integrator setup with other devices, in which case your integrator will be stuck with a pretty major project on their hands… but that would never happen
Probably more relevant for the next post, but for those trying to do last mile supply chain verification or forensics on the NAE, I’ve uploaded my MetaSys bins to WhiteScope.
Thursday, February 12th, 2015
Last week, someone told me that my blog was on the “LovelyHorse” list. I’ve always thought that I was the only person who cared about this blog, but I guess there is a lonely analyst out there that also cares… lonely analyst, this one is for you
I reported an issue affecting Visual Studio 2012 (which I had installed on one of my dev machines at the time). The issue was a blast from the past and reminded me of simpler times when I had the privilege of doing vulnerability research with Nate McFeters and Rob Carter :). Microsoft has addressed the issue, but determined that this issue did not warrant a bulletin, so you’ll have download the Visual Studio 2012 update 4 if you want to “patch” this issue. I have not verified whether other versions of Visual Studio were/are affected. Tested on Win7 with IE10 and VS2012.
Visual Studio 2012 registers the “vstfs” protocol handler during the installation process. This protocol handler calls devenv.exe in the following manner:
“C:\Program Files (x86)\Microsoft Visual Studio 11.0\Common7\IDE\devenv.exe” /TfsLink “%1″
As you know, protocol handlers can be instantiated remotely, most commonly via web pages. The “%1″ value can be attacker controlled and contains the values supplied when the attacker calls the protocol handler:
will result in the following being passed to the shell:
“C:\Program Files (x86)\Microsoft Visual Studio 11.0\Common7\IDE\devenv.exe” /TfsLink “vstfs:test“
Some time ago, I discovered that it was possible to escape out of double quotes when passing argument values to protocol handlers via Internet Explorer (and other browsers on Windows like Chrome). Sadly, this behavior is “by-design” and will not be fixed anytime soon. Knowing this, we can inject additional command line switches which will be passed to devenv.exe. If we know of suitable command line switch for devenv.exe we can repurpose devenv.exe and the vstfs protocol handler to do our bidding. For example, if we pass the following the Internet Explorer (copy/paste into the address bar or serve from a webpage):
vstfs:test” /command “Tools.Shell /c c:\windows\system32\calc.exe
We’ll end up with the following being passed to the remote system:
“C:\Program Files (x86)\Microsoft Visual Studio 11.0\Common7\IDE\devenv.exe” /TfsLink “vstfs:test” /command “Tools.Shell /c c:\windows\system32\calc.exe”
The line above, launches devenv.exe and also shells an executable of our choice (with the ability to pass arbitrary command line arguments to that arbitrary executable).
Fortunately, modern browsers (with the exception of Safari… but who uses Safari on Windows?!?!) have warnings when launching protocol handlers. This means users will likely encounter two prompts from IE (protocol handler warning and an elevation warning), you can witness this if you copy/paste the vstfs:test” /command “Tools.Shell /c c:\windows\system32\calc.exe” string into the IE address bar (tested against IE10 with a system that has VS2012).
There might be a way to bounce this protocol handler off a whitelisted protocol handler (one that doesn’t cause a protocol handler warning) running at medium. Alternatively, we could pass this protocol handler to a remote user via an application that doesn’t have protocol handler warnings (like a chat application that supports URLs). If we can find such a case there will be no warnings to the user and we’ll have a zero click or one click remote command injection exploit against Visual Studio users.
Monday, November 26th, 2012
In July of this year, I wrote about some of the frustrations I encountered when working with Tridium and trying to get them to fix various issues with their Niagara framework. The Niagara framework is the most prevalent Industrial Control System (ICS) in the world; it links together various ICS technologies and protocols. Looking at Eireann Leverett’s research on Internet accessible ICS, we see that Tridium Niagara is the prevalent ICS system available from the Internet. I didn’t talk much about the details of the issues I reported to DHS, but considering the patch has been out for nearly six months, I figure now is a good time.
The initial issue I reported to Tridum was a directory traversal issue that allowed remote, authenticated users to access files outside of the webroot. “Authenticated” includes demo and low privileged accounts. The directory traversal is very simple. Most web application security specialists know the classic “../../” style directory traversal, however Tridium was a bit different. Tridium makes use of “ordinals” to enable various functionalities within Niagara. For example, here is a URL for a Niagara deployment that uses the “station” ordinal:
The Niagara framework supports a large number of ordinals. One of these ordinals is the “file” ordinal, which is used to retrieve files from the Niagara framework server. The “file” ordinal followed by the path and filename to some file located within the webroot of the Niagara framework server. The Tridium Niagara framework isn’t susceptible to “../../” style traversal attacks, instead the Niagara framework uses the “^” character to traverse directories. Knowing this, we can specify a “^” character immediately following the file ordinal to traverse outside of the webroot. There are several files just outside of the webroot, however there is one file that is particularly interesting, a file named “config.bog”. The config.bog file is the configuration file for the entire Tridium Niagara deployment. The config.bog file contains all the configuration settings including username and encrapted (yes, I said encrapted… not encrypted) passwords for all accounts enabled on the system. Knowing this, we have a simple, reliable form of privilege escalation for any Tridium Niagara device. The exploit is very simple:
When you make the request above, you’ll download a compressed file. Unzip the compressed file and you’ll find the clear text config.bog for the Niagara server. Inside the config.bog, you’ll see the entire, detailed configuration for the Tridium device along with the username and passwords (protected with encraption). That’s it, it’s that simple. When I reported this issue to Tridium, I sent them a copy of their config.bog file for their marketing demo deployment.
Let’s ignore the fact that demo, default, and guest accounts are fairly common on these devices. In addition to the directory traversal, I also reported a weak session issue and insecure storage of user credentials issue. The Tridium Niagara framework generates sessionids that have about 9 bits of strength. This makes brute force attacks completely feasible and allows a remote attacker to quickly gain access to an authenticated state. Once authenticated, they are free to utilize the directory traversal to escalate privilege to Administrator. If that weren’t enough, the Tridium Niagara framework also stores a copy of the current username and password (base64’d) in the cookie giving any XSS bug the potential to divulge the clear text username and password.
There is a shining light to this story. When I first reported this issue to Tridium, the initial response was horrid. 6 months after the initial report, Tridium’s leadership attempted to pass these vulnerabilities off as “by-design”. Eventually, the folks at Honeywell (Tridium’s parent company) found about these issues and took over the response process. Three weeks later, they had a patch ready to go. Honeywell made the patch available to me a few days in advance of the release so I could take a look and verify the issues were fixed. They even gave me credentials to the new demo site so I could see the new features and security changes. It was welcoming to see an ICS vendor take such a stance towards security researchers, I hope other ICS vendors take note and follow suit. I’d like to personally thank Kevin Staggs for driving the renewed focus on security for Tridium Niagara, if you’re a Tridum customer, you should thank Kevin too. If you are a Tridium customer, you can learn more about the patch here: