ssh upstart job fails to actually bring up ssh

[Kees Cook]

When I reboot my router, ssh does not start.

Feb 15 22:51:32 router init: ssh main process (897) terminated with status 255
Feb 15 22:51:32 router init: ssh main process ended, respawning
Feb 15 22:51:32 router init: ssh main process (915) terminated with status 255
Feb 15 22:51:32 router init: ssh respawning too fast, stopped

I do not bind to 0.0.0.0, so I must have networking up fully before ssh can start. Because "filesystem" happens before "networking" has started, the job fails. We might want to consider changing the upstart job to:

start on filesystem and started networking

ProblemType: Bug
DistroRelease: Ubuntu 10.10
Package: openssh-server 1:5.5p1-4ubuntu5
ProcVersionSignature: Ubuntu 2.6.35-25.44-generic-pae 2.6.35.10
Uname: Linux 2.6.35-25-generic-pae i686
Architecture: i386
Date: Tue Feb 15 23:03:11 2011
ProcEnviron:
 PATH=(custom, no user)
 LANG=en_US.UTF-8
 SHELL=/bin/bash
SourcePackage: openssh

[John Seeger (pechspilz)]

Actually, this goes even further. All services that have the possibility not bind to 0.0.0.0 should only be started after the last network interface is up. I'm on a machine with 10 IP addresses and have to use something like

start on (filesystem
        and net-device-up IFACE=venet0:8)

in squid.conf and ssh.conf.

That's gonna be a tricky one.

[Robbie Williamson (robbiew)]

reversed a mistake at the request of John Seeger

[Mark Goris (mgoris)]

This sounds a lot like what happened to me when the "fix" for this defect was released: https://bugs.launchpad.net/ubuntu/+source/openssh/+bug/687535

I had to roll back to 5.3p1-3ubuntu4 to be able to restart Xen based Lucid instances and have sshd restart reliably.

[zwigno (zwigno)]

I'm getting something that sounds a lot like this bug in 12.04 LTS Server - I have a dual interface home server with 2 nics. ssh, xinetd, and isc-dhcp-server all fail to work correctly on boot and need to be restarted. All 3 also have custom configurations for binding to one nic or the other.
eth0 is internal. eth2 is public.

ssh - different Listenaddresses/ports based on IP(nic)
dhcp - bound to internal interface only
xinetd - tftpd is bound to internal interface only

Other daemons that don't have a interface-specific config seem to work fine (postfix, apache) or maybe they just don't care.

If it helps troubleshooting - I originally had an eth1 instead of eth2 but I swapped it out for a different model (3c905 for a intel e100) and it registered as eth2. eth0 has stayed the same (RTL8168e).

I tried setting ssh to run on "started network-manager" but that didn't seem to work. I'll try what Seeger suggests above and do the "net-device-up IFACE=eth2" to see if that helps.