Re: Reproducible Containers Paper

[Ludovic Courtès]

Hi Carl,

Carl Dong <contact@carldong.me> skribis:

> The paper is titled "Reproducible Containers”: 
> https://dl.acm.org/doi/pdf/10.1145/3373376.3378519, featured by Meta in their 
> post: 
> https://developers.facebook.com/blog/post/2022/11/22/hermit-deterministic-linux-testing/

Interesting!  Probably a sledgehammer for reproducible builds… except
perhaps for timestamp issues, as you write.

> I was thinking about this in terms of how to have timestamp reproducibility 
> inside Guix build containers which has been a problem in the past for 
> building from source, their approach in 5.3 is interesting (but perhaps a bit 
> invasive).

Yes, this bit is interesting:

  To ensure airtight interception of vDSO calls, DetTrace instead, just
  after each execve system call, replaces the vDSO library code with our
  implementation where each vDSO function makes a direct system
  call—which is duly intercepted via ptrace. We furthermore make the
  vvar page unreadable to prohibit any access to the raw
  nondeterministic data that vDSO timing calls use.

I wonder what this entails exactly, and whether one needs to ptrace
every single process to intercept exec calls in child processes, which
would be prohibitive.

Thanks for sharing!

Ludo’.