Bug#504044: rng-tools-debian: systemd complains about missing unit file

[Thorsten Glaser]

On Sat, 26 Oct 2024, Sven Geuer wrote:

>Feel free to integrate my additions into your repo. You might want to
>test the sysv-init functionality still works. According to my tests
>sysv-init and systemd unit work flawlessly at least when running
>autopkgtest from a Qemu VM.

I’ve integrated that, but I don’t have a sid system on which to test
this, unfortunately. I did both read *and* diff so I have reasonable
trust that it won’t break anything. Thank you again for your effort,
and if you could test what I uploaded, that would also be great.

I do have some questions regarding debian/systemd-helper that also
affect debian/rng-tools-debian.service:

• “/usr/sbin/rngd -p /run/rngd.pid […] && sleep 0.1”
 ⇒ should this not be “exec /usr/sbin/rngd -p /run/rngd.pid […]”
   instead so the wrapper script is replaced by the actual dæmon?

• this is Type=forking, should it better be Type=exec (with rngd -f)
  or is there anything (use of a wrapper script, pidfiles allowing
  tracking the actual dæmon, …) that indicates no?

• if no rng device is configured nor found, systemd-helper exits 1;
  is this a case for RestartPreventExitStatus in the unit perhaps?
 ⇒ with a different exit status than 1 of course, ideally one that
   rngd itself won’t throw, in combination with the exec above
  (note that there are two cases now: no device manually configured
  and autodetection found none, in which case the startup should
  probably be marked as failed until the user fixes the situation
  or disables the unit, as first case, and for second case, again
  no device manually configured but autodetection found /dev/hwrng,
  in which case the dæmon should just silently not start and not
  show up as failed; this is new (also for the sysvinit variant)
  in 2.5 because I dug through kernel source and found that the
  active device from the hwrng subsystem is already auto-polled by
  a kernel thread)

Perhaps the systemd people have an idea regarding this. As all of
this is no regression to the old state (the sysvinit→unit generator
would have resulted in something very much like this), it’s not
something of dire need, though, so I’m closing the bugreport with
the upload, as discussed.

bye,
//mirabilos
-- 
[...] if maybe ext3fs wasn't a better pick, or jfs, or maybe reiserfs, oh but
what about xfs, and if only i had waited until reiser4 was ready... in the be-
ginning, there was ffs, and in the middle, there was ffs, and at the end, there
was still ffs, and the sys admins knew it was good. :)  -- Ted Unangst über *fs