Gemfire is an in-memory data grid.It pools memory across multiple processes to manage application objects and behavior. Its written in Java and has a key-value structure storage. This data is stored in something called as a ‘region’ which can be queried using Object query language much like how one would have SQL for RDBMS.

I came to know about this some time back and since the opportunity of abusing an OQL is rare, I googled a bit about it. The only journal which references gemfire OQL and a remote command injection is below.

http://blog.emaze.net/2014/11/gemfire-from-oqli-to-rce-through.html

While doing penetration testing, I tried following the examples cited in this blog. The application I was testing was doing for black list filtering and hence all attack vectors weren’t going through properly but the first attack vector that was successful was something like below.

  1. select * from /region1 limit 10.

Comments: My query returned exactly 10 rows and I knew then that I would be able to pass any data in the parameter and  it is getting appended to code also.

2. select p.getclass.forName(‘java.lang.Runtime’).getDeclaredMethods()[7].getName() from /region1 p.

This query is similar to what is explained in the emaze blog and using the above construct, you would be able to list out all the methods of the run time. Getting to this stage was a little tough as I needed to tweak and find out which parameter was getting accepted and which wasn’t. For some reason, the invoke method wasn’t working at all. Before calling it a day, I had passed on this query to my colleague Prashanth working in a different timezone who cracked the shell.

3.select p.getclass.forName(‘java.lang.System’).getDeclaredMethods()[5].invoke(null, ‘os.version’.split(‘asd’)) from /region1 p

This gave out the version number and like wise, one can get all the System properties from similar queries. So, in your testing even if you don’t get a RCE, try an information leakage like above.

 

 

 

 

 

 

Advertisements