Towards better tools: Part 1
TL; DR: Not enough effort goes into how tools might be used, so its hard to integrate into all workflows. What would you change, and how?
There are various classes of tools I use during a computer forensic investigation. Some tools produce data, such as list-mft or RegRipper. Others, like sed, awk, grep, and pyp, transform data into a useful format. Finally, software like Microsoft Word or LaTeX display data in reports that I provide to clients. Tools in the latter classes are well established with comprehensive documentation and use cases. However, tools like parsers and forensic frameworks continue to spring up daily as the field develops. In this post, I’d like to describe some issues I have with many new tools, with an ultimate goal of potentially improving their quality.
Some of my greatest technical gripes with many forensic tools are that, (1) they are difficult to integrate with other tools and workflows, and (2) the output formats are inconsistent. Fortunately, (1) and (2) are related, and I think can be tackled together. For instance, the RegRipper ecosystem is a thorough body of knowledge on Registry artifact parsing that suffers from these issues: as currently designed, their component parts are not easily reused, and their output formats unique. I say this not as a diss or comment against the capacity of the developers, but as a call to action! Let’s make tools better and more flexible.
Here’s what I’d like
First, let me enumerate what I think I’d want in a perfect world. In a followup post, I’ll describe a potential approach for implementing these dreams.
In today’s example, let’s consider a new browser history parsing script. What are the ways in which I’d like to use it, and why does this make the script difficult to develop?
Sane defaults for one-offs
When I first see I can parse browser histories with the script, I’ll run it on an example user profile and see what it outputs. I’ll probably expect that you (well, the developer of the plugin, whoever that is) know at least a little bit about the artifact, and picked a sane default output format. Probably, I’ll hope that I can understand the output as a human without resorting to post processing. This is how I’d use the tool on a one-off basis, and to validate results of other processes.
Machine format for scaling
Subsequently, I’d probably use the browser history parser during an investigation and hopefully many large scale investigations. It’s likely that at some point I’d attempt to automate the history parser to pull tons of history entries and slice-n-dice the results. For this, I’d want an output format that can be easily tokenized and processed by common text-oriented tools (sed/awk/grep). Unfortunately, it’s likely that the human-friendly format is not the same as the computer-friendly format.
I might also feel lazy one day and email you, “I need CSV for my report, but your tool outputs TSV, so can you add a new output format for that?” Surely sed can do this, but why not build it right in? And BSV, and Bodyfile, and XML, too. It’s unreasonable to support all of these, but I’m not quite sure which subset is the best choice. The developer scratches their itch, but I need mine scratched, as well. How should we organize this?
Finally, as this browser history parser becomes well tested and respected, I might like to include it in a comprehensive forensic toolkit. Though the toolkit could call down to the system shell to execute the parser and tokenize the textual results, this is fairly hacky and unmaintainable. I’d really like to just import the script and use native the native Python (or Perl, or C) interface instead. So, it would be awesome if the script could be reused programmatically as well. But, this requires forethought and planning — things which may not be prioritized during the critical PoC stage of a new tool.
Do your tools do all these?
To summarize, there are three common uses of forensic tools:
- once off
- batched together
So what am I getting at? There are many ways every tool may be used, and often developers fail to think critically about these cases. Then, it becomes a chore for users to run tools, and less effort goes to what really matters: analysis.
To alleviate these issues, I recommend thinking about, designing, and implementing a common style of tool interaction. Not a new file format (we have enough of those), but a consistent means for producing useful tool results. How should a tool’s code be best structured? How can a tool’s output best satisfy a user? Although it strikes me as difficult to design something generic enough to satisfy all possible tools, we might as well try.
Take a minute now to think about how you like to interact with forensic tools, and what you’d want in an ideal world. Can you think of a better way to structure things? Tomorrow, I’ll post my proposal, and I already look forward to a lively discussion.