Public bug reported:

I am using Ubuntu 24.04.5 LTS with GNOME on Wayland. The installed versions at 
the captured event were gnome-session-bin 46.0-1ubuntu4 and systemd 
255.4-1ubuntu8.17 (kernel 6.17.0-1032-oem).
The effective logind InhibitDelayMaxSec on this system is 30 seconds, set by 
Ubuntu's unattended-upgrades drop-in 
/usr/lib/systemd/logind.conf.d/unattended-upgrades-logind-maxdelay.conf.

When I choose Restart from the GNOME system menu, it intermittently
takes about 30 seconds before logind proceeds with the reboot. This has
happened repeatedly over several days; it does not happen on every
attempt.

In a captured delayed reboot on 2026-09-23, GNOME Shell sent one Reboot
request to the session-manager D-Bus API. All 18 queried session clients
answered within about 10 ms. The same gnome-session-b process acquired a
logind shutdown delay-inhibitor FD, then sent Reboot(true) to logind at
12:34:14.518368. At 12:34:14.518437—69 microseconds later, before logind
replied—it sent a second equivalent delay-inhibitor request. Logind
accepted the reboot request at 12:34:14.533568 and rejected the second
inhibitor request at 12:34:14.533574 with
org.freedesktop.login1.OperationInProgress. GNOME then sent
CancelEndSession, but the first accepted delay-inhibitor FD remained
held. Logind reported the timeout at 12:34:44.608107; an automatic
snapshot still listed the first inhibitor at 12:34:44.751, and logind
announced reboot at 12:34:44.899355. The trace therefore shows the first
inhibitor remained held through the timeout; it does not establish
exactly when it was released afterward. The separate file-migration
block inhibitor had already been removed before the reboot request.

Interpretation: this is consistent with duplicate or re-entrant
shutdown-preparation progression inside gnome-session. Source inspection
found two end-session phase-progress paths that can reach shutdown
preparation, but the D-Bus trace cannot identify the exact callback
ordering. The failure path appears to cancel the session end sequence
without releasing the delay FD acquired by the first successful
inhibitor request.

Expected: one shutdown-preparation attempt for a reboot, or safe
handling of a duplicate attempt, without cancelling a reboot already
accepted by logind or leaving the first delay inhibitor held through the
full timeout.

Related upstream context:
- GNOME gnome-session work item 162 (manager phase-transition edge cases): 
https://gitlab.gnome.org/GNOME/gnome-session/-/work_items/162
- GNOME gnome-session merge request 170 (inhibitor fixes): 
https://gitlab.gnome.org/GNOME/gnome-session/-/merge_requests/170
- Ubuntu gnome-session bug 1841286 (older shutdown-phase assertion failure; 
related context, not an exact match): https://bugs.launchpad.net/bugs/1841286

The raw D-Bus traces are withheld because they contain session/process
metadata. A sanitized excerpt of the relevant D-Bus and journal sequence
is attached; it preserves the method sequence, outcomes, and timestamps
while omitting personal/session identifiers. See shutdown-delay-gnome-
inhibitor-redacted-excerpt-2026-09-23.txt.

** Affects: gnome-session (Ubuntu)
     Importance: Undecided
         Status: New

** Attachment added: "Sanitized D-Bus and journal excerpt for the 2026-09-23 
delayed reboot capture."
   
https://bugs.launchpad.net/bugs/2168273/+attachment/6002044/+files/shutdown-delay-gnome-inhibitor-redacted-excerpt-2026-09-23.txt

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2168273

Title:
  gnome-session duplicate delay-inhibitor request is rejected; first
  delay inhibitor remains held through timeout

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/gnome-session/+bug/2168273/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to