What follows is a review of the SANS FOR610: reverse engineering malware class taken at the SANS Prague 2013 event. What follows are rough notes, feelings and impressions from the class as it was taking place… Take it as you will, and I hope it proves useful for you in evaluating the class.
Note: Prior to the class I’d never done reverse engineering of malware… however I have done limited exploit dev and other tasks that dovetail well with some of the concepts.
Day 1 started off slowly walking through some simple malware using some behavioral and code analysis, before moving into the setup of our lab environments. Although the first half of the day was a little slow (and instructor led, so no hands-on), it brought everybody up to speed on what we’d be doing in the coming days and helped to lay the foundation for some of the more fun topics I’m sure are headed our way in the later part of the class. After a quick bite to eat we moved into the first hands-on section of the class.
The afternoon was spent slowly walking through the step by step process of performing a behavioral analysis of brbbot in our labs. The pace for the hands-on was a little slower than I’d have liked, but the step by step methodology helped really push home the process required. It was also nice to see procDOT (from the Austrian CERT team) mentioned as well, as an easy way to perform some of the steps.
The day finished off by walking through debugging brbbot in OllyDBG to perform code analysis and dive into some of the malware’s inner workings. Again that walk-through with the class was slow-paced with a few students struggling. With a class like this it’s hard to find a suitable pace to ensure everyone can follow along. That said, even though I’ve never done malware reversing before day 1 seemed slower than it really needed to be in a lot of places.
As is standard with most SANS classes, we kicked-off day 2 with a quick recap of the topics covered in day 1 of the class, finishing off the day 1 material with a brief discussion of what should be included in a malware analysis report.
Diving into the real day 2 content we starting looking at commonly seen packers and how to deal with them. Aside from the simple UPX cases (which are apparently more common than I thought in malware) we looked at dumping the unpacked code from memory using a number of tools. Unfortunately we didn’t dive more into the process and how to do it without the tool. I know this isn’t a simple process, but I’ve always been more of the “do it the hard way and learn, before you use the point and click tool” guy. Still you can’t have everything, and the walk-through on doing the same kind of dump using OllyDbg and the OllyDump plugin was an interesting diversion.
After a quick bite to eat we ventured into some patching of executables using OllyDbg (in particular patching to avoid debugger checks). Moving away from Windows executables, we began looking at analyzing browser-based malware.
No reversing class would be complete without some serious time spent crawling around in assembly. The morning was filled with the usual fare, compilers, linkers, commonly seen assembly codes and the like. Nothing that most wouldn’t have already seen and read if they were interested in this area of security. Where I think the topics really made things clear was breaking down recursion and jumps and looking at the raw C and compiled assembly versions (or one possible assembly version). It really helped to show that you can tell if the original code was a loop or an If Else. Probably nothing earth shattering for most, but I found it interesting to go down to the level. We also covered jump logic extensively looking at the way the registers are used to store the results used for common jumps. Good stuff to know if you’re reversing unknown code.
The afternoon covered some anti-disassembling examples that malware authors use to cause headaches for people reversing their work, before moving into the final real topic of the day, user-mode rootkits. The afternoon session seemed to be a little rushed, maybe due to some delays earlier on in the day. Day 3 certainly tries to cram in more than previous days have and it showed at the end with a few sections covered in a little less detail than I’d have liked. Pity really, as I’d rather a few day 1 or 2 topics were given short shrift and replaced with more time to really dig in and look at sniffers, keyloggers and the like in Olly/IDA. Overall though, the best day so far… very good stuff!
After a sleepless night (gotta love hotels) we dove into self defending malware and started to look at some of the more commonly seen anti-debugging techniques used by modern malware. Kicking things off we manually walked through a malware sample using SEH to fool analysts, and a simple packer to prevent easy analysis of the real malware. The same was interesting in the way it used multiple techniques to fool people trying to understand the executable, but was pretty simple to bypass the protections given a few minutes and a few hints on how it works. The SEH trick was an interesting one and not one that I’d thought of before (I’m not a malware analyst after all).
The afternoon continued on from the morning session dealing with more anti-debugging tricks used commonly by packers and malware writers (including timing based RDTSC and TLS Callback techniques, amongst others). Some of these techniques are probably well-known and documented amongst analysts, but as I’ve said before, as a non-malware analyst the techniques were new to me and made for an interesting afternoon in OllyDbg and shellcode.
Despite only 1.5 hours sleep (don’t ask…) day 5 started quickly diving into malicious Office and PDF documents. After covering the basics, we moved into some quick analysis of malicious samples using a range of tools and scripts. As expected, the PDF section of the class was a relative what’s what of Didier Stevens research. Some of it is things I’ve already read or seen based on Didier’s wonderful training and workshops. Still, there were some interesting extensions to this using tools like shellcode2exe to create a debuggable PE in order to walk through injected shellcode in a more controllable way. I particularly liked the technique of using libemu’s sctest to create an overview of what the shellcode is doing. Something that will definitely come in handy…
Moving away from malicious files we focused on the uses of memory forensics and how malware analysis can be performed on forensic images. As expected, volatility played a big part of the afternoon, using it to check connections, find hidden processes and dump out the unpacked malware from memory into a PE file to be analysed in the normal way. Using some of the more run modules in volatility we looked at detection of kernel and user-level rootkits and how to extract information from memory dumps that help us as malware analysts understand what malware is doing on a system.
CTF style challenges to reinforce the material covered in days 1-5… I won’t go further into this, but the experience was a good one for sure! I managed to hover around 10th position before having to head off early to catch a train. Not stellar, but for a first time malware analyst, I guess it could have been worse 😉
The pace of the class was a common theme for me throughout the first half of the week, but quickly picked up pace midway through day 3. There’s a number of reasons for this, and it was hard to tell if the class was just paced for a more beginner crowd (disclaimer: I’ve never done malware reversing, but I’ve used a majority of the tools before for various other tasks/purposes), or if the pace was effected by a few people in the class that seemed to get lost pretty easily. I won’t launch into a tirade here about people doing technical classes and not having the chops for it, so consider the issue of pace discussed 😉
The tips/tricks and general processes for working with malware were very clearly discussed and worked very well during the course. How they work in real life with modern malware that wanted to stop you at every turn is to be seen. Although we covered some of the more common methods used by malware to cause problems for analysts, there’s no way you can cover them all. Things like that can only be learnt from experience and expose to different techniques and methods after all. The processes helped me quite a lot with a few debugger questions I had. Not just when it comes to malware, but also for more offense-based reversing (thick clients etc…)
The class is very tools based. This isn’t necessarily a bad thing, as most of what we do on a day-to-day basis is tools based… but the first half in particular lacked a certain amount of “rolling up your sleeves and getting into the dirt” that I personally like to see from classes like this. Although it’s a hard way to learn, I find that learning the hard way before you just load a tool makes you respect and understand the issue and the solution. This was dealt with better in the final days with much more hands on stuff. Also, as much as people like to say the day 6 CTF style challenge is a wasted day, I found that some parts of it helped bring back points I’d forgotten from the earlier parts of the class. This really helped reinforce things for me.
Overall the class was good for a beginner just looking to get a grounding in the topic. That said however the pace made it feel like the class could have achieved so much more in the first 2 days in particular. Missing were more automated and scriptable debugging of malware. Things like pydbg, vdb/vivisect and a number of other script based debuggers would have been nice to cover, even if only in passing. A number of the examples really fell into the category of easily scriptable, and it would have been nice to get some hands on with that stuff.
I think the class really came into its own in the later days… much more focused, technical and designed to really stimulate your thinking.