Killing the systemd user session on logout

[Bernie Innocenti]

On 13/07/2020 17.48, David Edmundson wrote:
> 
> 
> On Sun, Jul 12, 2020 at 5:29 PM Bernie Innocenti <bernie at codewiz.org 
> <mailto:bernie at codewiz.org>> wrote:
> 
>     (re-posting because most subscribers might have missed my previous post
>     due to an excessively strict DKIM policy applied by my domain).
> 
> 
>     I'm trying to fix this longstanding bug where there are dangling user
>     processes after the desktop session exists:
> 
>     https://bugs.kde.org/show_bug.cgi?id=359651
>     <https://bugs.kde.org/show_bug.cgi?id=359651>
> 
>     This is reproducible every time on multiple distributions and KDE
>     versions. It also causes subsequent logins to fail or behave
>     surprisingly due to the presence of extraneous dbus services,
>     pulseaudio
>     clients, etc.
> 
> Yes it's a problem.
> Pulseaudio deliberately tries to survive whilst a user is logged in, so 
> that's fine
> The DBus services is a bit weird.
> 
> Apparently gnome used to solve this problem by killing dbus-daemon 
> explicitly :/
> 
>     I looked into plasma-workspace/startkde, and there's no hint of it
>     starting "systemd --user" directly.
> 
> 
>  >So I'm starting to think that it might be done implicitly via PAM
> 
>     (there's a pam_systemd module).
> 
> That's exactly what happens.
> 
>     But, if a PAM module is starting "systemd
>     --user", shouldn't the same PAM module also do the cleanup?
> 
> 
> IMHO yes, what I've been trying to do to solve this is to make the 
> systemd user session aware of all our stuff.
> We've currently introduced cgroups round the apps, we're tackling the 
> services slowly as part of the plasma boot.
> 
> Then "all" we need to do is ship two drop-ins for app-*  and plasma-* 
> scopes and services.
> That include the line BindsTo=graphic-session.target
> 
> Once that ends, it'll tear down everything in a controlled manner just 
> like a system process.
> This is also what gnome is doing to do.

Do you have a work-in-progress patch I could see?


By the way, I've been looking into this other bug where starting a 
Plasma session will eat a VT even after logging out:

   https://github.com/sddm/sddm/issues/1200#issuecomment-657084636

I believe we're leaking the fd. Could you confirm that I'm on the right 
track? This is my first tentative fix:

   https://codewiz.org/pub/sddm-fd-leak.patch

I wasn't able to test it because I couldn't figure out how to run my own 
sddm build without overwriting the system binaries.


> Some notes about that topic: https://systemd.io/DESKTOP_ENVIRONMENTS/ 
> <https://systemd.io/DESKTOP_ENVIRONMENTS/>

Interesting read.


> David
> 
> 
>     -- 
>     _ // Bernie Innocenti
>     \X/ https://codewiz.org/ <https://codewiz.org/>

-- 
_ // Bernie Innocenti
\X/  https://codewiz.org/