OISF IDS/IPS engine prototype intro

Posted on Jan 7, 2009

For over a year I’ve been working on a prototype implementation of a new IDS/IPS engine for the Open Infosec Foundation. This is not necessarily going to be the engine we’ll be using in OISF, although it’s likely that at least some of the code will be used. Discussions about features for the engine are still ongoing ( wiki, list), once that settles down we’ll see whats usable and whats not. In the worst case I still think many parts like hashing functions, pattern matcher implementations, protocol decoders, etc can be used.

So what is there so far? It’s all new code written in the C language and has about 30k lines of code in 150+ files so far. It’s fully threaded in a way that should make it very scalable on many cores/cpu’s. More about the threading in a future post. The code is heavily unit tested, which really helps a lot in preventing and tracing bugs.

Right now it’s limited to being an inline IDS/IPS, using the libnetfilter_queue interface in Linux to acquire and verdict packets. The packet input and verdict subsystem is very modular (I learned a lot from the mess we created in Snort_inline, where we supported 3 types of inline packet capture methods, creating a true #ifdef hell). It has working protocol decoders for IPv4 and IPv6, TCP and UDP. It has a flow engine, a detection engine and output plugins.

For rules/signatures it currently only supports the Snort signature syntax, and loads about 70% of the current VRT and Emerging Threats signatures out there. The biggest thing missing is support for the flowbits option, which is used in a lot of the sigs. It has basic HTTP parsing, enabling at least uri matching.

A lot of things are missing too. For example there is fragment handling, TCP stream state tracking, TCP stream reassembly, a pcap mode, portscan detection, a flowbits like function, normalization, etc, etc.

There are a lot of plans and ideas, for example having output pipes for configurable captured network data. It’s already possible to capture for example a user agent in a rule and match on that captured data. I think it would be very useful to be able to have some pipe to an external program that receives just the user agents and does something with them. Many many more ideas and usecases exist and I hope to write about that more at a later stage.

The most interesting about writing this code is that every time I’m working on some part, I’m getting more and more ideas about possibilities for improvements, optimizations and such. I intent to share those here on my blog from now. Also, I intent to write about the various parts of the code I wrote already. So stay tuned!