Talk:SSH keys

From ArchWiki
Revision as of 16:22, 21 December 2011 by Ntwk (talk | contribs) (Alternative to manual key installation)
Jump to: navigation, search

Maybe the default 2048 bit rsa key is better?Vogt 01:54, 31 August 2008 (EDT)

I have just completed a tidyup, this including removing the section on connection control as I deemed it irrelivant. If needed, it is available in the history. Thelucster 13:51, 13 April 2009 (EDT)

I'm hesitant to make the edit myself, as I'm new the the whole 'wiki' thing, but I noticed that the Gnupg instructions here seem incomplete. After following them several times myself, I turned to the man pages and found the solution. Edit: Nevermind. My problem was something different, and the current /etc/profile.d script handles the proper exports. Ryeguy146 21:19, 22 March 2011 (PDT)


Sometimes the 'ssh-add' is not enough to log in without a password. It is possible that ssh is configured in such way that only a limited group of users is allowed to the machine. In this case - you need root-access to the server! - you have to change the configuration-file. Mostly you can find it as /etc/ssh/sshd_config. If the last line(s) of this file read(s): 'AllowUsers <username>', you will have to add a similar line with your own username. Don't forget to restart the ssh deamon: '/etc/init.d/sshd restart'.

Using pam_ssh module

I just want to add that one could also use the pam_ss module, available here or in the AUR to decrypt the ssh key on login and automatically start ssh-agent and add the keys. This way one would have a truely password less ssh session and in the same way not compromise security by using a passphrase less key.

I have opened a new section on using pam_ssh to descrypt a user's ssh keys upon login. My experience with PAM in general is limited, so the content currently consists of a description of pam_ssh, some basic configuration instructions, and some of the limitations of pam_ssh which I have personally encountered. — Ntwk 16:37, 18 December 2011 (EST)


The current wiki entries tells to "$ echo 'eval `ssh-agent`' >> ~/.bashrc" which will everytime spawn a new ssh-agent. I think a more elegant way is only to add the export commands of ssh-agent to the .bashrc, so one ssh-agent can be used from every shell. This could be put in a small wrapper script:

# check if ssh-agent is running
if [ -n "`ps -e|grep ssh-agent`" ];then
        echo "ssh-agent is already running" >&2
        exit 1
# get new sock and pid
agent=`ssh-agent |head -2`
# delete old sock, pid and comment
sed -i -e "/SSH_\(AUTH_SOCK\|AGENT_PID\)/d" ~/.bashrc
# insert new sock and pid for new shells
echo -e "# auto generated SSH_AUTH_SOCK and SSH_AGENT_PID" >> ~/.bashrc
echo $agent >> ~/.bashrc
# for evaluation in the current shell
echo $agent
$ eval `./` " 

this would make ssh-agent available on the current and all new shells.

Be sure you have added the key to your /etc/ssh/ssh_config:

IdentityFile path/to/key

Alternative to manual key installation

We might want to mention that there's a script called 'ssh-copy-id' which comes with OpenSSH that install your public key in a remote machine's authorized_keys. There's a few caveat with it (it changes permissions of the user home directory, which should be a no-op in most situations; see the man page of ssh-copy-id -- and it also tells the user to make sure the script hasn't added extra keys, which might be a bit confusing for some).

The article currently gives a description of how to use ssh-copy-id as well providing instructions on how to manually copy your pivate key to the remote server. I can't find any mention of ssh-copy-id altering file or directory permissions on the remote server. On the contrary, the ssh-copy-id man page dated 14 November 1999 currently included in the OpenSSH man page states that it "does not modify the permissions of any pre-existing files of directories." — Ntwk 11:22, 21 December 2011 (EST)