PIDFID breaks pkexec with ConsoleKit session tracking

[0strodamus]

Compiling polkit 124 with ConsoleKit 1.26 and polkit option session_tracking=ConsoleKit breaks launching any application with pkexec.
The following error is generated on the command line:

Error checking for authorization org.freedesktop.policykit.exec: GDBus.Error:org.freedesktop.PolicyKit1.Error.Failed: Process not found.

Reverting the PIDFID commits starting with polkitd: use PIDFDs if available to track processes and related subsequent commits fixes the issue.

Also commenting or removing the HAVE_PIDFD_OPEN line from meson.build will fix the issue.

Does this functionality require elogind or systemd? Can an option be added to meson_options.txt to toggle HAVE_PIDFD_OPEN off for systems that it isn't going to work on?

[bluca]

As you can see from the linked commit, there is already a build time check for the sd_pidfd_get_session function, and a fallback is used if not available, which would be your case. So the issue is somewhere else, and you'll need to debug further and find out what exactly is failing and where, as this is not really actionable as it is.

[0strodamus]

Thanks for the reply.

I'm not a developer, so maybe I'm missing the fallback you're referring to. I see an if/else statement in the linked commit where sd_pid_get_session appears to be the fallback to sd_pidfd_get_session, however they're wrapped in a HAVE_LIBSYSTEMD ifdef/endif so this whole section of code should not be included when I compile because my system does not have systemd or elogind installed. I'm using ConsoleKit 1.26 with OpenRC/sysV init.

I have spent a lot of time looking over the code and I don't know how to debug this any further. I'm content to keep hacking meson.build, although looking at the commits made to polkit-125, I may stay on 123 and backport any future changes that aren't targeting systems with elogind/systemd.

If this is still not actionable, please close my issue report. Thanks again for taking the time to reply and all the best to you and the polkit project.

[thesamesam]

We had a report of something similar with elogind in Gentoo (https://bugs.gentoo.org/934314), bisecting to the same commit, but I haven't dug into it more (yet). Interestingly, https://923614.bugs.gentoo.org/attachment.cgi?id=901896 didn't work even just as an experiment, which kind of mirrors #451 (comment).

[RomainNaour]

Hello,

We also noticed this regression with PIDFD using SYSV init.

The issue is reproduced with polkit 124 and 125 (and master)

https://gitlab.com/buildroot.org/buildroot/-/jobs/8199992596