Re: compat_linux failures due to execve, openat, etc.

[Kevin Bloom <ktnb%posteo.net@localhost>]

Robert Swindells <rjs%fdy2.co.uk@localhost> wrote:

> 
> Kevin Bloom <ktnb%posteo.net@localhost> wrote:
> > Robert Swindells <rjs%fdy2.co.uk@localhost> wrote:
> >> 
> >> Kevin Bloom <ktnb%posteo.net@localhost> wrote:
> >> > I've tried settings LD_LIBRARY_PATH and LD_PRELOAD to no avail. I
> >> > can't seem to come up with any other hacks to try to convince nix-env
> >> > everything is okay. Any ideas?
> >> 
> >> Have you tried running ktrace(1) on a non-nix Linux binary?
> >> 
> >> One that I just tried also returns an error from execve but carries on
> >> after that and works correctly.
> >
> > Hmm, you're right. I ran nodeJS just fine and see the same errors.
> > I wonder _what_ exactly is causing the nix-env program to not run
> > all the way. When I run it, I get these errors in the STDOUT:
> > GC warning: pthread_setname_np failed
> > error: setting death signal: Function not implemented
> >
> > When I look through the ktrace of nix-env, I don't see anything
> > obvious. It might be that I just don't know what to look for too.
> > Do you have any general rules of thumb when looking at this sort
> > of thing?
> 
> UTSL

I've never heard this before so I looked it up. I love it!

> Look at the sources to whatever version of glibc is in nix and see what
> pthread_setname_np() is doing.

Seems to me that this likely isn't the cause of the binary not working
since I do get one other non-error/warnings in the stdout as well.
After those GC warnings I get one line:
installing 'nix-2.32.2'
Then I get those errors with the death signal. However, it may be
that there is still something valuable in the glibc source.

> You could compile tst-setgetname.c into a single executable on Linux,
> copy that onto your NetBSD system and try it out.
> 
> You do have an emulated Linux procfs mounted as well as native /proc?

I did as the compat_linux man page says, namely:
mount_procfs procfs /emul/linux/proc
making sure that the -o flag isn't set. Is there more to be done
here? nix seems to be convinced that /emul/linux/proc is fine
according to ktrace(1).

One other note -- readelf(1)'s "NEEDED" libs are all installed on the
system where they ought to be. Although ldd(1) points everything to
the libs that are in the nix store, not the system's shared libs.

> Maybe we need to extend the proc emulation to add some extra things.

As mentioned above, nix seems convinced that /proc is okay (from what
I've gathered)