On Wed, Dec 18, 2024 at 07:01:35PM +0000, Mark Hindley wrote:
> Maybe. I am at (or probably beyond) my area of expertise. At the moment I am
> unsure if elogind ought to be using its own cgroups namespace or if docker.io 
> is
> behaving badly here. Time for some reading....

My reading of systemd's cg_pid_get_owner_uid() is that it calls
cg_pid_get_path_shifted() with root=NULL and hardcodes
SYSTEMD_CGROUP_CONTROLLER, so consulting alternative namespaces are not
possible.  Therefore, for elogind to use libsystemd0 compatible cgroups,
/sys/fs/cgroup/user.slice is the only possible option. But I am happy to be
corrected here if I have misread the code.

Looking at docker.io's initscript, it checks if /sys/fs/cgroup is a mount point
before mounting cgroups there. So, adding elogind to LSB header 'Should-Start'
ought to ensure elogind is started first to mount /sys/fs/cgroup. Does that
work? Does it break docker?

Mark

_______________________________________________
Pkg-utopia-maintainers mailing list
[email protected]
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/pkg-utopia-maintainers

Reply via email to