sshd with a ListenAddress fails at boot before DHCP
I rebooted my home server onto a new kernel and it came back without SSH. sshd_config binds one LAN address with ListenAddress, and stock ssh.service only waits for network.target, which doesn’t mean the address exists yet. on that boot sshd started 23 seconds before the interface got its address from DHCP, failed with Cannot bind any address and stayed down until it was fixed at the console.
the red herring: sshd -t run by hand while the service is down says Missing privilege separation directory: /run/sshd. that’s only because the unit creates /run/sshd when it starts. it says nothing about the config. journalctl -b -u ssh had the real error.
the drop-in
# /etc/systemd/system/ssh.service.d/override.conf
[Unit]
Wants=network-online.target
After=network-online.target
[Service]
Restart=on-failure
RestartSec=5
network-online covers the normal boot. the restart covers the boot where the router is off after a power cut and wait-online gives up after 2 minutes. every 5 seconds stays under systemd’s default start limit (5 in 10 s), so it keeps retrying until the address shows up.
proven on the next reboot the same day: sshd started 0.04 s after network-online, 0 restarts.
while I was in the journal
over the 54 days of uptime before that reboot, the link lost carrier 341 times in 88 bursts. 54 bursts started at 05:00:05, give or take a second. that’s the router’s own reboot schedule, set to every day at 05:00. most of the rest are likely power cuts. none of it matters to a box that’s already up.
the reboot also showed that unattended-upgrades installs kernels and never reboots: 54 days on 6.8.0-139 with 6.8.0-149 installed. /var/run/reboot-required is the signal.
see also: timers that assumed an always-on box