openssh-2.5.2p2-1.7
initscripts-5.49-1

After upgrading from 2.3.0p4, I get to the syslog:

Mar 28 20:08:57 xxx sshd[761]: Received signal 15; terminating.
Mar 28 20:08:57 xxx sshd: sshd shutdown succeeded
Mar 28 20:08:59 xxx sshd: RSA key generation succeeded

At this point, master sshd daemon has been killed and /var/run/sshd.pid
deleted.

Now, 'service sshd start' won't start the daemon; no error messages are
printed.  This
is caused by some bug in 'daemon' code of initscripts (misdetecting current
ssh connections
as the master process?), as '/usr/sbin/sshd' works fine.

There are two ways the next step happens:
 1) if you're on console, 'service sshd restart' kills _ONE_ connected ssh
client and otherwise
restarts gracefully.
 2) if you're connected with ssh, 'service sshd restart' kills your
connection but won't start a new
one.  Time to walk to the console..

The fact that all systems AFAIR had at least one incoming sshd client
process running at the time sshd was restarted might have been crucial.

This has been replicated on 5+ systems.