socket unit stuck forever in final-sigterm after daemon-reload #44457

systemd version the issue has been seen with

=260.4

Used distribution

centos9/10

Linux kernel version used

6.16.1

CPU architectures issue was seen on

None

Component

systemd

Expected behaviour you didn't see

After TimeoutSec=, the socket unit goes from final-sigterm to final-sigkill and then to failed (timeout), as it does when there's no reload.

Unexpected behaviour you saw

If PID 1 is reloaded while a .socket unit is in final-sigterm and the unit has no control process, the unit stays in deactivating (final-sigterm) forever. Its cgroup is empty and its stop job never finishes. stop, start, daemon-reload and daemon-reexec don't recover it, so only a reboot does.

socket_coldplug() only re-arms the timer when control_pid is set, and socket units have no notify_cgroup_empty handler. After the reload, nothing moves the unit forward. mount.c and swap.c use the same coldplug pattern.

Steps to reproduce the problem

# repro.socket
[Socket]
ListenStream=/run/repro.sock
TimeoutSec=10
ExecStartPost=/bin/sh -c 'setsid sh -c "trap \"sleep 3; exit\" TERM; while :; do sleep 0.2; done" </dev/null >/dev/null 2>&1 &'

systemctl start repro.socket
systemctl stop --no-block repro.socket   # -> final-sigterm, timer armed
systemctl daemon-reload                  # within TimeoutSec
sleep 30
systemctl status repro.socket            # still "deactivating (final-sigterm)", empty cgroup

Without the daemon-reload, the unit goes to failed (timeout) after 10 s as expected.

Additional program output to the terminal or log subsystem illustrating the issue

● repro.socket
     Loaded: loaded (/run/systemd/system/repro.socket; static)
     Active: deactivating (final-sigterm) since Mon 2026-10-26 12:37:36 PDT; 3min 20s ago
        Job: 2318
      Tasks: 0
     CGroup: /system.slice/repro.socket

systemd[1]: Starting repro.socket...
systemd[1]: Listening on repro.socket.
systemd[1]: Stopping repro.socket...

$ cat /sys/fs/cgroup/system.slice/repro.socket/cgroup.events
populated 0


Without the second `daemon-reload`:


systemd[1]: Stopping repro.socket...
systemd[1]: repro.socket: Stopping timed out (2). Killing.
systemd[1]: repro.socket: Failed with result 'timeout'.