I have a cluster where the /home directories are nfs mounted. In some of the accounts the SSH keys have stopped working.

SSH works fine with all accounts. I just can't get passwordless login to work on *some* accounts.

The SSH keys are the same in all accounts so there's not corruption. md5sum confirms this. I've even copied them from working accounts to faulty ones and it made no difference. I've deleted the .ssh directory and created a new one. No effect.

The permissions of /home/ or /home/user1 are OK. I've changed them from wrxwrxwrx to wrx------ and it makes not difference.


Any ideas?

I think there are some problems with permission of ~/.ssh. Compare two dirs, working and not working.
Also you can check secure.log.

The permissions are fine. I can do a chmod 755 -R /home and it makes no impact on any of the accounts. I'll check secure.log. Can't believe I forgot to do that.

Presumably they're using the right keys and the authorized_keys files are properly set up. Also, check your other SSH directives -- are you using AllowGroups or similar? As mentioned above, secure.log and other log files will probably lead to a sudden revelation.

I'll also mention that permissions and ownership are important if the sshd has strict mode enabled (which I think it usually is by default). The home directory needs to be owned by the user, and should be 700. .ssh should also be 700, and finally the authorized_keys file should be 600. If you're still having trouble figuring out the problem, as other posters mentioned, check /var/log/secure. And of course, you can use the "-v" flag (or even "-vv") with your ssh client to get more debugging info about why the connection is failing.

ssh -v chris@remotehost