[systemd-devel] [PATCH 1/2] core: Refuse to run a user instance when the system hasn't been booted with systemd.

[Kok, Auke-jan H]

On Sat, Oct 6, 2012 at 1:40 AM, Thomas Bächler <thomas at archlinux.org> wrote:
> Am 06.10.2012 05:42, schrieb Kok, Auke-jan H:
>> On Fri, Oct 5, 2012 at 4:11 PM, Thomas Bächler <thomas at archlinux.org> wrote:
>>> Running as a user instance won't work at all if systemd isn't running as system
>>> manager, so refuse to start in that case.
>>
>> huh?
>>
>> I'd like to think this is a situation that should really work...
>
> It doesn't work, but it doesn't say so.
>
>> - to test systemd in a non-systemd environment
>> - Chroot/namespacesVM situations where there's no contact with the
>> real init process
>> - some calamity happened or IPC is temporarily offlined
>> - just because it makes no sense to enforce it
>>
>> what exactly breaks?
>
> In the namespace situation (you are likely talking about nspawn or lxc),
> we are PID 1 and should never hit that code path.
>
> I tried running systemd as a user on a non-systemd system because I
> thought it should work. It first complains that it cannot set up its
> cgroups - I don't know if that problem can be solved, but I think it
> can't (you need root to configure the main systemd cgroup, I think).
>
> Furthermore, systemctl was unable to do anything - not even 'systemctl
> --user exit' worked - it couldn't find systemd on the bus, as it refuses
> to look for the bus unless sd_booted() is true. You can send the signal
> manually, but that is not what you want.

you explicitly have to set DBUS_SESSION_BUS_ADDRESS no matter what.
Even with systemd running as pid 1 that is required. Perhaps that is a
bug in itself.

> The situation gets worse: You can only kill the systemd instance with
> kill -9 - both kill -15 and Ctrl+C result in an error message about dbus.

probably due to the above dbus URL still need. There's still some
problems with it as well, even with systemd running as pid 1, systemd
--user will complain first itself about the session bus missing (lol),
and then starts it properly anyway (if you have the
dbus.service/socket units).

> As systemctl refuses to any work (except enable/disable) when
> sd_booted() == false, so should systemd --user - this patch merely makes
> systemd's behaviour consistent.

I think it actually masks the DBus issues, so, we should resolve those instead.

Perhaps a large warning is in order - something I would appreciate
more than a plain error and exit.



Auke