I wanted to append [idea] but it became too long. I think there should be balance "musings" and actual work. I learned very valuable lessons in my wakegp research. Even though I haven't completed it. If the one doing this research would be me, I would go for no actual work as long as I can. And when I start to do the work, I should do as little as I can and move very slowly. Like the saying "When you move slowly, you can see the path ahead clear". After all, we should know if the next step we go forward, the ground under us is solid.
This is a difficult project. The blog post seems to hint at reasonable feasibility, this stuff is hard! We build a less ambitious tool in the university lab: "ASTANA: Practical String Deobfuscation for Android Applications Using Program Slicing" [0].
Would advise to first read the reverse engineering related work. Genetic programming is just a technique best used when everything else has failed :-)
Thanks for the hints. Of course, it's very very difficult. But one thing I think you missed, is that I'm proposing a "byte equivalent decompilation". And after that, we should go into reading the code readable and understandable.
If we could create a program doing all this, automatically or semi-automatically, it will be great-great because then not releasing the kernel code doesn't matter. I believe if enough effort and time is put into it, there is a good chance we could see such a thing in like 5-7 years.
After that, we might be able to target the binary blobs, the propriety firmwares. Those might have some legal issue, of course. But as long as it is used only to write a FOSS alternative, that probably won't be an issue, I think.
I've read the abstract of your article. I am not much in the field of decompilers. Let alone deobfuscation. It's even hard for me to type it :))
I think it is probably a safe assumption that the kernel binary found on Android devices is not obfuscated. Tho I probably need more research to confirm this.
> Theoretically, we could also go for finding the semantically equivalent C code. However, last time I researched, checking semantic equivalency is a very complex problem. I think it was NP hard.
Already deciding whether two finite automata decide the same (regular) language is PSPACE-complete; it's undecidable for anything that can decide arbitrary context-free languages (which C programs can clearly do).
While you might be able to produce code that would re-compile into a binary resembling the original binary, how would you be able to generate the same code that was compiled to produce the original binary? Seems like there would be a significant difference between the actual original code and the decompiled version - enough that it could not be used to prove license violation.
10 comments
[ 2.8 ms ] story [ 27.2 ms ] threadhttps://www.fsfla.org/ikiwiki/selibre/linux-libre/
Would advise to first read the reverse engineering related work. Genetic programming is just a technique best used when everything else has failed :-)
[0] https://arxiv.org/pdf/2104.02612
If we could create a program doing all this, automatically or semi-automatically, it will be great-great because then not releasing the kernel code doesn't matter. I believe if enough effort and time is put into it, there is a good chance we could see such a thing in like 5-7 years.
After that, we might be able to target the binary blobs, the propriety firmwares. Those might have some legal issue, of course. But as long as it is used only to write a FOSS alternative, that probably won't be an issue, I think.
I think it is probably a safe assumption that the kernel binary found on Android devices is not obfuscated. Tho I probably need more research to confirm this.
Already deciding whether two finite automata decide the same (regular) language is PSPACE-complete; it's undecidable for anything that can decide arbitrary context-free languages (which C programs can clearly do).