Below is a collection of notes regarding unikernels. I had originally prepared this stuff to submit to EkoParty’s CFP, but ended up not wanting to devote time to stabilizing PHP7’s heap structures and I lost interest in the rest of the project before it was complete. However, there are still some cool takeaways I figured I could write down. Maybe they’ll come in handy? If so, please let let me know.
Unikernels are a continuation of turning everything into a container or VM. Basically, as many VMs currently just run one userland application, the idea is that we can simplify our entire software stack by removing the userland/kernelland barrier and essentially compiling our usermode process into the kernel. This is, in the implementation I looked at, done with a NetBSD kernel and a variety of either native or lightly-patched POSIX applications (bonus: there is significant lag time between upstream fixes and rump package fixes, just like every other containerized solution).
While I don’t necessarily think that conceptually unikernels are a good idea (attack surface reduction vs mitigation removal), I do think people will start more widely deploying them shortly and I was curious what memory corruption exploitation would look like inside of them, and more generally what your payload options are like.
All of the following is based off of two unikernel programs, nginx and php5 and only makes use of public vulnerabilities. I am happy to provide all referenced code (in varying states of incompleteness), on request.more ...