Bug#773765: pbuilder issue which is related to systemd

[Thorsten Glaser]

Martin Pitt dixit:

>Usually the init system sets this up, but as init doesn't run in
>schroot, pbuilder, and friends, it's the job of those *chroot
>construction packages to set it up. schroot does, maybe pbuilder
>doesn't always set this up?

More or less.

With cowbuilder (which is how people normally use pbuilder),
CHROOT/run is a regular directory and a copy of what was in
the base.cow/tgz, but that doesn’t matter as it’s ⓐ fresh
and ⓑ deconstructed after the build. CHROOT/run/shm gets
bindmounted from the outside (used to be CHROOT/dev/shm
until it broke).

It’s not strictly necessary for throwaway chroots to actually
have CHROOT/run a tmpfs, plus there are setup costs involved
that are tricky to get right in a generic way (after mounting
the tmpfs, you have permissions, files, subdirectories, etc.
to take care of; normally, init scripts would do that), which
is why I’d not object to this setup.

>what package sets up which directory. I think initscripts should
>either create a /dev/shm directory and all init systems and
>schroot/pbuilder etc. mount tmpfs there, or we stick to the weird
>debianism of /run/shm and then initscripts sets up /dev/shm b /run/shm
					    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
eh what?

>and schroot/pbuilder etc. mout a tmpfs there.

I agree that we definitely fucked up bad (two releases after
the fact, still!) and should decide on one thing and stick
with that “period”. The release goal page mentions that /dev/shm
is going to eventually go away so AIUI a tmpfs on the /run/shm
directory (bindmounted by pbuilder) it is. If so, no changes to
pbuilder are required currently.

I think the other option is to agree that Debian fucked up,
revert /run/shm, and do stable uploads to the same effect,
so that chroot tools can continue to work the same independent
of the distribution in the chroot. That would not so be welcome.

Thanks for digging in deeper. I did not know that initscripts
broke this setup (my apologies towards the systemd camp), and
I wonder *why* there is a different setup for chroots vs. the
main system in the first place? A system could just have the
chroot setup, with early userspace mounting a tmpfs “over” it,
so we have the same thing everywhere. Less code paths.

bye,
//mirabilos
-- 
<hecker> cool ein Ada Lovelace Google-Doodle. aber zum 197. Geburtstag? Hätten
die nicht noch 3 Jahre warten können? <mirabilos> bis dahin gibts google nicht
mehr <hecker> ja, könnte man meinen. wahrscheinlich ist der angekündigte welt-
untergang aus dem maya-kalender die globale abschaltung von google ☺ und darum
müssen die die doodles vorher noch raushauen