[lfs-dev] Systemd - KillUserProcesses=yes (Was: Re: [blfs-dev] Two hints about building firefox-47.0.1 and using screen with systemd-230)

Tim Tassonis stuff at decentral.ch
Tue Aug 2 09:43:13 PDT 2016


On 08/02/16 04:53, DJ Lucas wrote:
>
>
> On 07/31/2016 01:06 PM, Tim Tassonis wrote:
>> On 07/31/16 17:48, DJ Lucas wrote:
>>>
>>>
>>> On July 30, 2016 8:55:40 PM CDT, "Douglas R. Reno"
>>> <renodr at linuxfromscratch.org> wrote:
>>>
>>>>
>>>> I'll make the change to /etc/systemd/logind.conf with the update to
>>>> systemd-231 that I am working on.
>>>
>>>
>>> Let's not do that by default. Rather, put text containing an
>>> explanation, and the command with a nodump attribute in LFS, and add a
>>> note in BLFS on the screen page (only real consumer we have).
>>
>>
>> As far as I understand, also nohup from coreutils is concerned.
>>
>
> Michael and Tim BCC'd (not sure if you are sub'd to LFS-Dev).
>
> So I wanted to give this a bit more of a fair shake than simply not
> going against upstream. I spent more than a few hours sorting through
> systemd sucks/rules flame fests to find the occasional intelligent
> argument for or against the 230 change in default.
>
> I haven't changed my request to leave it at default. In short, the
> default is now a restrictive model vs. the previous permissive. The well
> written arguments both for, and against, are all essentially the same.
> To paraphrase: It fixed (or broke) 30 some-odd years of
> nohup/daemon()/setsid() being virtually unrestricted.
>
> So there are four different ways to turn it off in various contexts.
> Compile time (break/fix semi-permanently), config file (break/fix
> temporarily), user session (loginctl enable-linger [user] (break/fix for
> one user)), or at runtime via systemd-run. The systemd-run functionality
> was added as the replacement method to corral the former free for all.
> This essentially leaves it up to the user or system admin rather than
> allowing random binaries to run forever.
>
> As Tim mentioned, nohup is broken (by design) when enabled (the
> default). Other threads have suggested removing nohup and/or symlinking
> nohup->systemd-run, but I also don't think we should go so far as to
> disable installation of nohup (it still works when the new functionality
> is disabled using three of the methods above).
>
> https://www.freedesktop.org/software/systemd/man/loginctl.html
> https://www.freedesktop.org/software/systemd/man/systemd-run.html
>
> Instead, I'd like to see another sect2 added to 7.10 discussing the new
> functionality, meeting the official primary goal of LFS. I'll be happy
> to write it if agreed upon.
>
> Thoughts?

My thought is that generally disabling installation of nohup would be a 
bad choice, as there might be some users (apart from me) that choose the 
non-systemd approach for LFS and frequently like to use nohup, e.g. for 
long-running compiles on machines connected with ssh, allowing to 
disconnect and re-connect in the meantime.

If however the long-term goal of lfs is to only support systemd, then 
the removal/symlinking would make sense. Else, there might be a section 
in coreutils that makes this optional for systemd users.

Cheers
Tim


-- 
decentral.ch - IT Stuff
Tim Tassonis
Dennlerstasse 36
8047 Z├╝rich

stuff at decentral.ch
+41 79 229 36 17


More information about the lfs-dev mailing list