Hi everyone,

For those not on IRC, I have identified the last remaining F18 blocker. You can read the gory details on my G+ and I'll write a blog, but briefly the summary is that systemd-logind is used to manage user sessions[0]. It talks over D-BUS (what doesn't these days) and is collaborating with a custom systemd pam plugin which runs during session setup. During a pkexec (polkit1) invocation of a binary, a call is made to systemd-logind to setup a new session, which it does. But when it does this, on systems without CONFIG_AUDITSYSCALL enabled[1], it cannot see /proc/pid/sessionid, and decides that this means there cannot be an existing session (the one you used to invoke pkexec).

When a new session is created, systemd opens a special fifo for the session and registers to poll on this so that it can see if the session died mysterously and needs cleaning up. In the case of a sudo-like command, it needs to clean this fifo up ahead of time so that it won't later (right as the new processing being driven by pkexec is spawned) kill what you just tried to run. But if you do not have auditing setup, systemd will never do any of these things and its "login manager" will kill whatever you are trying to run with a SIGTERM. That's why you see those "Terminated" messages and errors when trying to run e.g. yumex.

I'll write this up using better and more accurate language. I just wanted to throw something out there to say that we know what the problem is and that a fixed kernel package will be built soon.

Jon.

[0] see the FreeDesktop docs on heads and sessions, also see loginctl.

[1] This is set in the base config but is not inherited on all of the 3.6 kernel variants, and a similar though slightly different situation exists on 3.7. Peter is fixing the kernel config with a new 3.6 kernel.