Bypassing CSRF Protections With Dynamic CSRF
At our presentation at Black Hat and Defcon this year, Shawn Moyer and I gave a talk called Weaponizing the Web: More Attacks on User Generated Content. During the talk we discussed something we are calling Dynamic Cross-Site Request Forgery (CSRF). Dynamic CSRF methods can be used to create new attack vectors and compromise CSRF protection mechanisms. CSRF is often difficult to explain in the first place, so I just wanted to do a post about it and link our Dynamic CSRF White Paper.
Static CSRF Attacks
The current way people are exploiting CSRF issues is pretty static. A link may be crafted and people are tricked in to clicking on it or viewing a page with this link embedded. They may even do something such as create an auto-submitting POST that the user executes when visiting certain content. The moral is that someone crafts this attack for a specific location and chooses a vector for delivering it to users. On it’s own, this static CSRF attack method can be pretty devastating or perhaps totally innocuous. That’s just the nature of CSRF. There aren’t too many vulnerabilities that fall in to this category, which may be why it can sometimes be hard to understand.
Let’s just think about some things that would make CSRF even more dangerous. What if we had visibility in to CSRF tokens or even session identifiers? What if we knew for sure a user was on the site that we wanted to attack? What if we could use a central point to dynamically attack multiple sites? What if we could create a custom payload on the fly to include tokens and session data? What if we could request data in another process and fixate it on to a request that the user executes? CSRF would become much more dangerous. Well, it just did….
Dynamic CSRF Attacks
With Dynamic CSRF attacks the game changes a bit. Attacks can be adapted based on where requests are originating from. Sensitive information sent in a cross-domain fashion can be packaged in to a customized payload and sent to a location. Disjoined or bolted on components can be queried and fixated on to a user’s request. This opens up new opportunities for attacks. Dynamic CSRF allows you to run one central system and run attacks for multiple sites and applications with the ability to remove and repackage any leaked session information.
Using dynamic methods may allow for the bypass of CSRF protections. Whether these protections are leaked in the referer or if they can be requested by other means and fixated on to the request, bypassing may be possible. This makes requests that were previously not vulnerable to CSRF vulnerable again.
When not limited by static requests a few new attack vectors open up. I just wanted to highlight a couple here.
In the next couple of sections I am going to point out some dynamic methods for bypassing CSRF protection mechanisms.
A simple redirect attack can be made more devastating with the addition of session related data added to the request. So a user hits a piece of HTML on a web page that sends a GET request containing session data (Session IDs, CSRF Tokens, etc) in the referer. This session data can be taken from the referer and repackaged in to a payload that is sent to the users browser as a location for a 302 redirect.
A scenario for this type of attack might be embedding a hidden image tag that links to a location offsite. This image tag may leak session data through referer allowing a tool to repackage that session data and get the user’s browser to execute it as a payload. This could be used to bypass CSRF protections by using the CSRF tokens from the referer.
For this scenario a hyperlink could point to a page that is created dynamically per-each request. This page contains a hidden GET or POST that could contain session data taken from referer as mentioned previously in the HTTP Redirect. The page would then refresh to a legitimate destination. To a user there would only appear to be a short pause prior to their landing on the legitimate destination, but in the background an attack has happened.
This attack could be made more interesting with URL shortening services. People have become accustomed to URL shorteners due to their use of Twitter and other social networks. Using a URL shortener would mask the location of the dynamic page and may not raise further suspicion.
A scenario for this would be posting a link that is shortened with a URL shortener on a web site that supposedly links to a video on YouTube. When the user clicks the link a hidden action (GET or POST) is executed. After the hidden action takes place, the browser does a refresh to the expected destination (the YouTube video). The victim may get little or no indication that the hidden attack has happened.
Fixation Based Attacks
Here is where things get a bit interesting. This attack hinges on the ability to have a valid CSRF token and a valid session identifier, but those two values aren’t associated together from a session perspective. Say you are posting on a message board. This message board issues you a Session ID in a cookie and also issues you a token for use when posting. If these two values aren’t linked from a session perspective, it would allow for any valid token to be added to the request. Since the Session ID is in the cookie value, if a forged request had a valid token appended this would bypass CSRF protections.
What makes this fixation attack an interesting scenario is it requires no leaking of any session data in a cross domain fashion. As long as you have some visibility to the token, it can be requested and fixated to the request.
There are other scenarios as well. It really depends on how much session data gets leaked in a cross-domain fashion. For example, if the full session ID is leaked, there could be some real problems. It could allow for a tool to construct any GET or any POST to the target application.
MonkeyFist is a tool I wrote to PoC Dynamic CSRF attacks. The tool runs as a small service on a system listening for all incoming input. When it receives a request it looks in the referer to determine where the request is coming from. It then uses this information to parse a payloads file and look for the specific attack to construct for that domain. You can have multiple entries in the payloads file performing multiple types of attacks. Redirect, POST, Page (GET and POST) and an Experimental Fixation handler are all included in the application. You can learn more about MonkeyFist on the the Hexsec Labs page. It is not exactly where I would like it to be, but it works. More streamlining and modifications will come soon. I am still recovering from Black Hat and Defcon
I have some posts on MonkeyFist that are coming soon as well.
That’s it for a quick intro to Dynamic CSRF. If you would like more information you can read our white paper which is available here.