[H-GEN] Systemd killing processes after logout

[Tomas Marko Miljenović]

On Wed, 8 Jun 2016 05:58:13 PM David Seikel wrote:
> On Wed, 08 Jun 2016 17:48:06 +1000 Russell Stuart
> <russell-humbug at stuart.id.au> wrote:
> > On Wed, 2016-06-08 at 15:28 +1000, Clinton Roy wrote:
> > > Keep your nappy on
> >
> > But, but ... it's my nappy and I've just found someone else's poo in
> > it.  (That change has already made it's way to my laptop.)
> >
> > And while it's true it won't get into Jessie, it's not Jessie I'm
> > worried about.  I'll be keeping a close eye on what's going on "down
> > there" for some time.  It's a job best done with the nappy off.
> 
> Why am I suddenly glad I don't go to meetings any more?  lol

Dave, how can you say that?

I haven't been able to make it to a HUMBUG meeting for a while myself. All the 
quality technical commentary, I suppose I can get that elsewhere. But this - 
the nappy references and the poo analogies - is exactly what I miss.

--

For the curious, this appears to be where it[1] all started:
* https://bugs.freedesktop.org/show_bug.cgi?id=94508
* https://github.com/systemd/systemd/commit/97e5530cf20

My basic understanding of what transpired:-
1. systemd maintainers were hit with bug reports. Services are lingering on.
2. Maintainers blame the way certain distros configured dbus-daemon[2].
3. Guess what? These distros configured DBUS contrary to upstream defaults.
4. The fix: new upstream defaults in logind.
  * soon to be followed by *
5. Guess what? Certain distros configure logind contrary to upstream defaults.

The best part is that screen and tmux issues were explicitly mentioned, both 
in the referenced bug report and commit message. As upstream said, they can 
easily be worked around by running
$ systemd-run --black-magic-which-doesn't-quite-work-as-intended-yet screen
(or, hopefully, something better which will be out well before any of this 
hits production systems).

--

For what it's worth, my personal take is this:
Once they implement some DSO or whatever for 'special' processes that are 
supposed to (and have been explicitly allowed to) linger around after user 
logout, then all (other) user processes should be nuked on logout for security 
reasons. By default. Trusted systems should have the option of changing the 
default, if that's what the maintainer(s) explicitly require.


Wish I was there live for the start of this chat!
- Tomas

________
1. Upstream changing the default value for KillUserProcesses, not the cool 
HUMBUG commentary.
2. There are reasons for using user sessions, and there are processes not 
using DBUS at all which were (or now are, after the logind change) affected.