[Bug 2020641] Re: Installing or removing apps through snap-store launches another gdm session

[Daniel van Vugt]

Thanks for the patch. Please propose it upstream at:
https://gitlab.gnome.org/GNOME/gdm/-/merge_requests

** Changed in: gdm3 (Ubuntu)
       Status: New => In Progress

-- 
You received this bug notification because you are a member of Ubuntu
Sponsors Team, which is subscribed to the bug report.
https://bugs.launchpad.net/bugs/2020641

Title:
  Installing or removing apps through snap-store launches another gdm
  session

Status in gdm3 package in Ubuntu:
  In Progress

Bug description:
  gdm3 version: 42.0-1ubuntu7.22.04.2
  Ubuntu version: Ubuntu 22.04.2 LTS

  [Description]
  Installing or removing snap packages through NICE DCV on AWS will cause the user to be kicked out of their session. The issue happens on machines running nvidia GPUs (passed through to a VM) with the GRID driver installed as well as the normal nvidia driver.

  [Steps to reproduce]
  Simply install or remove any snap through the snap store will trigger the issue, for example:
  $ snap install skype

  also running any of the following commands will also trigger the issue:
  $ snap connect skype:opengl :opengl
  $ snap disconnect skype:opengl :opengl
  $ snap connect skype:camera :camera
  $ snap disconnect skype:camera :camera

  After further investigation I was able to pin down the issue to udev and could reproduce the issue by running the following command:
  $ sudo udevadm trigger --parent-match=/sys/devices/pci0000:00/0000:00:1e.0/drm/card0

  where "/sys/devices/pci0000:00/0000:00:1e.0/drm/card0" corresponds to
  the nvidia GPU of my instance.

  A more generic way of triggering the issue would be running:
  $ sudo udevadm trigger

  [Solution]

  I have investigated the issue and discovered that it lies within GDM3 in the "udev_is_settled" function (daemon/gdm-local-display-factory.c).
  In the case where the udev is settled the line "g_clear_signal_handler (&factory->uevent_handler_id, factory->gudev_client);" at the end of the function is triggered however this is not the case when the function returns early and will lead to the user being logged out. In its current implementation there are three different return points before "g_clear_signal_handler" is executed where the udev devices would already have settled.

  I have written a patch that fixes this issue by making sure the
  function "g_clear_signal_handler" is executed in all cases for which
  the udev is settled.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/gdm3/+bug/2020641/+subscriptions