[systemd-devel] RFC: removing initctl support

[Daniel P. Berrange]

On Tue, Sep 22, 2015 at 02:31:25AM +0200, Lennart Poettering wrote:
> Heya!
> 
> Since a long time systemd has been shipping with two-way compat
> support for /dev/initctl, and I am tempted to remove it. Before I do
> so, I'd like some input on the relevance of this interface:
> 
> a) there's support in systemctl to reboot the system by sending the
>    right bytes to /dev/initctl as fallback, so that you can reboot a
>    sysvinit system with "systemctl reboot".
> 
> b) There's a mini-daemon "systemd-initctl.service" that is
>    fifo-activated on /dev/initctl, and forwards reboot requests from
>    old sysvinit clients to systemd.
> 
> Both of this was supposed to help transition between sysvinit and
> systemd systems: if you mix sysvinit clients with a systemd init
> system and vice versa, you can still use the the tools to reboot the
> other system.
> 
> I'd claim the interface is borderline useless: the only operation you
> can actually readlly properly dispatch with it is rebooting, and
> reloading PID1. And that's pretty much it.
> 
> We never even really used this stuff on Fedora properly (since we
> actually transitioned from Upstart, not sysvinit, and we never had the
> same level of compat for that...).
> 
> This code has been bitrotting for a while, and nobody really cared.
> 
> And most importantly: the entire protocol use by sysvinit via
> /dev/initctl is deeply flawed, since it sends messages over
> /dev/initctl that are not a divisor of PIPE_SIZE in length. Thus, if
> PID 1 didn't read messages quick enough the messages queued could be
> half-written and be partially interleaved with another client's
> messages, and there is no way the system can ever recover from that.
> 
> Thus, I'd really like to kill this. Does anybody care about it, and
> can give me a strong enough reason to keep this anyway?

The libvirt virDomainShutdown|Reboot APIs for triggering controlled
shutdown/reboots of guest OS have support for using /dev/initctl with
containers, as it was the lowest common denominator that easily worked
across systemd, sysvinit & upstart.

We could add further code to use a systemd specific interface if
needed, so it wouldn't be the end of the world of /dev/initctl was
removed, but it'd be nice to not have todo that.

Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|