Package: ntp Version: 1:4.2.6.p5+dfsg-7+deb8u2 Severity: important Dear Maintainer,
Under systemd restarting the NTP daemon is racy, which can lead to a silent failure to start a new instance of the daemon during upgrades. The problem is that systemd has no separate restart action, just does a stop and a start. But the stop action does not wait for the daemon to terminate, thus the start action races with the exiting instance. If the start action is fast enough, the new instance may fail to bind because the exiting one still holds the port, thus exit. The result is no daemon running. There is an error message in syslog, but the restart operation exits successfully nevertheless. The problem could be sidestepped by providing a native systemd unit for the NTP daemon, because the sleep 2 in the init script restart action mostly covers the problem under SysV init. The init script would still need to be fixed, though. Easiest is to use the --retry option of start-stop-daemon (see #814881 for an example), which would obsolete the restart sleep as well. -- Regards, Feri.
