217

I want to communicate between several computers on my network (static Ethernet), through SSH. In order to do that I need to run ssh-add every time I log in on a specific machine, how can I do it so that it's set up once and it doesn't ask me for the passphrase every time I log in or reboot my machine?

I know that there is a way that you should add some lines to the bash_profile file, but I still need to type the password every time I reboot/log in to a specific machine.

if [ -z "$SSH_AUTH_SOCK" ] ; then
    eval `ssh-agent -s`
    ssh-add
fi

14 Answers 14

307

This is a typical example of a trade-off between security and convenience. Luckily there are a number of options. The most appropriate solution depends on the usage scenario and desired level of security.

ssh-key with passphrase, no ssh-agent

Now the passphrase has to be entered every time the key is used for authentication. While this is the best option from a security standpoint, it offers the worst usability. This may also lead to a weak passphrase being chosen in-order-to lessen the burden of entering it repeatedly.

ssh-key with passphrase, with ssh-agent

Adding the following to ~/.bash_profile will automatically start ssh-agent and load the ssh-key(s) on login:

if [ -z "$SSH_AUTH_SOCK" ] ; then
  eval `ssh-agent -s`
  ssh-add
fi

Now the passphrase must be entered upon every login. While slightly better from a usability perspective, this has the drawback that ssh-agent prompts for the passphrase regardless of if the key is to be used or not during the login session. Each new login also spawns a distinct ssh-agent instance which remains running with the added keys in memory even after logout, unless explicitly killed.

To kill ssh_agent on logout, add the following to ~/.bash_logout

if [ -n "$SSH_AUTH_SOCK" ] ; then
  eval `/usr/bin/ssh-agent -k`
fi

or the following to ~/.bash_profile

trap 'test -n "$SSH_AUTH_SOCK" && eval `/usr/bin/ssh-agent -k`' 0

Creating multiple ssh-agent instances can be avoided by creating a persistent communication socket to the agent at a fixed location in the file system, such as in Collin Anderson's answer. This is an improvement over spawning multiple agents instances, however, unless explicitly killed the decrypted key still remains in memory after logout.

On desktops, ssh-agents included with the desktop environment, such as the Gnome Keyring SSH Agent, can be a better approach as they typically can be made to prompt for the passphrase the first time the ssh-key is used during a login session and store the decrypted private key in memory until the end of the session.

ssh-key with passphrase, with ssh-ident

ssh-ident is a utility that can manage ssh-agent on your behalf and load identities as necessary. It adds keys only once as they are needed, regardless of how many terminals, ssh or login sessions that require access to an ssh-agent. It can also add and use a different agent and different set of keys depending on the host being connected to, or the directory ssh is invoked from. This allows for isolating keys when using agent forwarding with different hosts. It also allows to use multiple accounts on sites like GitHub.

To enable ssh-ident, install it and add the following alias to your ~/bash_profile:

alias ssh='/path/to/ssh-ident'

ssh-key with passphrase, with keychain

keychain is a small utility which manages ssh-agent on your behalf and allows the ssh-agent to remain running when the login session ends. On subsequent logins, keychain will connect to the existing ssh-agent instance. In practice, this means that the passphrase must be be entered only during the first login after a reboot. On subsequent logins, the unencrypted key from the existing ssh-agent instance is used. This can also be useful for allowing passwordless RSA/DSA authentication in cron jobs without passwordless ssh-keys.

To enable keychain, install it and add something like the following to ~/.bash_profile:

eval `keychain --agents ssh --eval id_rsa`

From a security point of view, ssh-ident and keychain are worse than ssh-agent instances limited to the lifetime of a particular session, but they offer a high level of convenience. To improve the security of keychain, some people add the --clear option to their ~/.bash_profile keychain invocation. By doing this passphrases must be re-entered on login as above, but cron jobs will still have access to the unencrypted keys after the user logs out. The keychain wiki page has more information and examples.

ssh-key without passphrase

From a security standpoint, this is the worst option since the private key is entirely unprotected in case it is exposed. This is, however, the only way to make sure that the passphrase need not be re-entered after a reboot.

ssh-key with passphrase, with ssh-agent, passing passphrase to ssh-add from script

While it might seem like a straightforward idea to pass the passphrase to ssh-add from a script, e.g. echo "passphrase\n" | ssh-add, this is not as straighforward as it seems as ssh-add does not read the passphrase from stdin, but opens /dev/tty directly for reading.

This can be worked around with expect, a tool for automating interactive applications. Below is an example of a script which adds a ssh-key using a passphrase stored in the script:

#!/usr/bin/expect -f
spawn ssh-add /home/user/.ssh/id_rsa
expect "Enter passphrase for /home/user/.ssh/id_rsa:"
send "passphrase\n";
expect "Identity added: /home/user/.ssh/id_rsa (/home/user/.ssh/id_rsa)"
interact

Note that as the passphrase is stored in plaintext in the script, from a security perspective, this is hardly better than having a passwordless ssh-key. If this approach is to be used, it is important to make sure that the expect script containing the passphrase has proper permissions set to it, making it readable, writable and runnable only by the key owner.

  • 1
    Okay, but when I put your code to ~/.bash_profile I have to type in password every time I login, I don't want that either. I am not concerned about security at all. echo "pass\n" | ssh-add doesn't work – zdun8 Sep 17 '13 at 12:59
  • 3
    @user1607072 Yeah, that is how the ssh-agent snippet in ~/.bash_profile behaves as explained in the answer. You might want to look at the keychain utility. With keychain you need to enter the password on first login after reboot, but on subsequent logins keychain will connect to an existing ssh-agent instance with the decrypted key in memory. Apart from that there's the option of generating a ssh-key without a passphrase, but this is of course not recommended. – Thomas Nyman Sep 17 '13 at 13:35
  • 2
    @user1607072 While I would strongly suggest one of the more secure approaches, there is a way to pass the passphrase to ssh-add from a script. The reason echo "pass\n" | ssh-add does not work is that ssh-add does not read the password from stdin, but opens /dev/tty directly for reading. Updated the answer to include a workaround for this, using an utility called expect. – Thomas Nyman Sep 17 '13 at 18:03
  • 1
    @user1607072 It might be a bit overkill for your use case, but Kerberos in combination with ssh GSSAPI support can also be used for passwordless ssh logins. The corresponding authentication method in ssh is called gssapi-with-mic. This is usually used in larger networks, but of course if you have interest in this it might be worth looking into. – Thomas Nyman Sep 18 '13 at 11:54
  • 1
    @ErickBrown: Already answered here. The SSH Agent unit should be stopped on logout if you have user lingering disabled in the systemd login manager. If user lingering is enabled, the systemd user instance and the SSH Agent unit are kept running even after the last login session is closed. – Thomas Nyman Jul 25 '18 at 8:00
79

Add this to your ~/.bashrc:

if [ ! -S ~/.ssh/ssh_auth_sock ]; then
  eval `ssh-agent`
  ln -sf "$SSH_AUTH_SOCK" ~/.ssh/ssh_auth_sock
fi
export SSH_AUTH_SOCK=~/.ssh/ssh_auth_sock
ssh-add -l > /dev/null || ssh-add

This should only prompt for a password the first time you login after each reboot. It will keep reusing the same ssh-agent as long as it stays running.

  • 1
    very neat, this way you only have one ssh-agent running (: Multiple agents as in @thomasNyman's second solution seems a security risk to me... – drevicko Feb 10 '16 at 11:54
  • 1
    After researching in various sites and reading various solutions, this one here seems to be the clearest, straight to the point. Very nice. +1 – Dr Beco Jul 18 '16 at 0:40
  • 1
    better to do this: `alias ssh=ssh-check-agent", and have the check-agent version do the above. that way: a) you only get one agent and b) you only get the agent if you need it – Erik Aronesty Jul 26 '16 at 20:31
  • 2
    I think -s is the default, so we're already doing that. – Collin Anderson Nov 22 '16 at 16:25
  • 1
    ssh-add -l returns an exit code of 0 when the agent has identities and 1 when it does not so you can cut grep out of the last command and use ssh-add -l > '/dev/null' || ssh-add – Grant Humphries Dec 20 '16 at 2:21
11

Not closely related to the OP's question, but it might be useful to others: since 7.2.0 ssh(1) has an option that allows adding a key to ssh-agent upon first authentication; the option is AddKeysToAgent and can be set to yes, no, ask, or confirm, systemwide or on your personal .ssh/config file.

Reference: https://www.openssh.com/txt/release-7.2

  • 2
    Applicable to those who are new to the .ssh/config file: this applies to ssh and anything that uses ssh behind it, for example scp, and can be done on a per-host basis. – SEoF Sep 25 '17 at 14:32
5

ssh-agent caches various unlocked ssh-keys, so you can have ssh-keys protected by passwords, but without having to type them every single time.

In order to cache unlocked keys, it obviously needs to unlock those keys. For unlocking keys that are locked with a passphrase, it obviously needs to know these passphrases.

Any method that does not require authorization from a human being (e.g. "typing in a password") will not only make your system insecure; it will also render the entire purpose of the ssh-agent meaningless.

Having said all this, you can simply use ssh-keys that are not password protected (hit Enter when asked for a password during key-generation). Since there isn't any password, ssh-agent doesn't need to ask you for one in order to (not) cache it.

  • I agree, as long as your keys are properly user-only-permissioned, there is little advantage to ssh-agent over permissionless keys. i like to ssh into a login server, and then, that server has a bunch of permssionless keys, each of which can only be used to unlock one other server. the login server does nothing else, so it's much harder to hack/spoof, etc... the other servers have no password access, are key-only. – Erik Aronesty Jul 26 '16 at 20:34
4

I won't recommend you ssh-add (which need to open a ssh-agent) at login. This is because you can't control when the ssh-agent section ends, and can create security risk when you need not use the keyfiles at one login section.

Rather, I recommend to write a script which opens a ssh-agent's section sub-shell, with all keyfiles auto added, and be called when needed to use ssh. If you could adopt so, read on.

You would have two choices:

  1. Remove all passphrases for your keys, which have weak security if your key files are stolen. (thus not recommended)

  2. Use the same passphrase for your keys. Then when you ssh-add keyfile1 keyfile2 ..., you will only need to type the passphrase once, per section.

In both cases, you could write such script file "ssh_keys_section.sh" as below:

#!/bin/bash
# This script run a ssh-agent on a sub-shell and automatically ssh-add all keyfiles at once.
# This agent ends when you type `exit` to close the sub-shell.
exec ssh-agent bash -c "ssh-add /path/to/keyfile1 /path/to/keyfile2 ...; exec bash"

Remarks:

  • Command to change or delete passphrase: ssh-keygen -p -f keyfile
  • Within the sub-shell, you might even fork more terminals which share the same unlocked keys, by using maybe a command like /path/to/yourterminal & (depends on OS)
3

Here is a workaround to automate your SSH passphrase.

  1. Create a one-liner script which prints your passphrase to standard output, e.g.:

     echo 'echo MY_SSH_PASSWORD' > ~/.print_ssh_password && chmod 700 ~/.print_ssh_password
    

    Important: Ensure you copy the leading space to prevent storing your password to your history.

And use one of the below methods.

  • using a standard input approach:

    cat ~/.ssh/id_rsa | SSH_ASKPASS=~/.print_ssh_password ssh-add -
    
  • or named pipe approach:

    1. Create a named pipe (you could also try a process substitution):

      mkfifo --mode 0600 ~/.ssh_fifo
      
    2. Run ssh-add by specifying the program used for the authentication:

      cat ~/.ssh/id_rsa >~/.ssh_fifo | SSH_ASKPASS=~/.print_ssh_password ssh-add ~/.ssh_fifo
      

    See: man ssh-add to read more about SSH_ASKPASS.

  • 2
    The echo my_passphrase is a big security hole. First after you have typed it, the password is in clear text in the history file of what ever shell you use. And second command line arguments are world readable on Unix (ps -ef). Never put passwords in command line arguments! – ceving Aug 24 '16 at 9:03
  • 1
    @ceving Adding extra leading space solves the problem with the history file. Added extra info. – kenorb Aug 24 '16 at 9:25
  • @kenorb: That doesn't solve the bigger problem of the password visible in ps output. The history file is typically only readable by the owning user anyway, but the command lines are readable by all users on a system. – Thomas Nyman Nov 23 '16 at 14:37
2
if [ ! -S ${HOME}/.ssh/ssh_auth_sock ]; then
  eval $(ssh-agent)
  ln -sf "${SSH_AUTH_SOCK}" ${HOME}/.ssh/ssh_auth_sock
fi
export SSH_AUTH_SOCK=${HOME}/.ssh/ssh_auth_sock

ssh_keys=$(find -E ~/.ssh -type f -regex '.*(rsa$|pem)')
ssh_agent_keys=$(ssh-add -l | awk '{key=NF-1; print $key}')

for k in "${ssh_keys}"; do
    for l in "${ssh_agent_keys}"; do
        if [[ ! "${k}" = "${l}" ]]; then
            ssh-add "${k}" > /dev/null 2>&1
        fi
    done
done
1
SSH_ENV="$HOME/.ssh/environment"

function start_agent {
     echo "Initialising new SSH agent..."
     /usr/bin/ssh-agent | sed 's/^echo/#echo/' > "${SSH_ENV}"
     echo succeeded
     chmod 600 "${SSH_ENV}"
     . "${SSH_ENV}" > /dev/null
     /usr/bin/ssh-add;
}

# Source SSH settings, if applicable

if [ -f "${SSH_ENV}" ]; then
     . "${SSH_ENV}" > /dev/null
     #ps ${SSH_AGENT_PID} doesn't work under cywgin
     ps -ef | grep ${SSH_AGENT_PID} | grep ssh-agent$ > /dev/null || {
         start_agent;
     }
else
     start_agent;
fi

Giving credit here: https://www.cygwin.com/ml/cygwin/2001-06/msg00537.html

This solution is also endorsed here: http://mah.everybody.org/docs/ssh

1

Single sign on solution for SSH could lead me to pam_ssh.

According to this article, the concept is:

If you work with multiple *nix-based machines via ssh, you are probably tired of constantly having to enter your password every time you want to access another box. There is a secure way to allow you to access every machine, that you have ssh access to, without having to enter another password (other than the one you signed on with originally.)


This is actually quite simple to do, you basically just create a public/private key pair to authenticate yourself to your other machines, then have PAM spawn an agent to load your keys after you logon, providing a single signon solution to accessing all your remote machines. This guide will walk you through setting this up.

I have not verified this would actually work.

1

I used to use the script mentioned by steampowered, I've made the below one now, because it doesn't leave files lying around.

Working on zsh shell only.

#!/bin/sh

AGENT_BIN=`which ssh-agent`
AGENT_ADD_BIN=`which ssh-add`
AGENT_PID=`ps -fe | grep ${AGENT_BIN} | awk -vuser=$USER -vcmd="$AGENT_BIN" '$1==user && $8==cmd{print $2;exit;}'`
if [ -z "$AGENT_BIN" ]; then
    echo "no ssh agent found!";
    return
fi
if [ "" -eq "$AGENT_PID" ]; then
    if read -sq "YN?Do you want to unlock your ssh keys?"; then
        echo ""
        output=`$AGENT_BIN | sed 's/echo/#echo/g'`
        eval $output
        $AGENT_ADD_BIN
    fi
else
    for f in "/proc/"*
    do
        cmdline=`cat "$f/cmdline"`
        if [ "${AGENT_BIN}" -ef "${cmdline}" ]; then
            export SSH_AUTH_SOCK=`cat $f/net/unix | grep --binary-file=text -oP '((/[^/]*?)+/ssh-[^/]+/agent\.\d+$)'`
            export SSH_AGENT_PID=${f##*/}
            break;
        fi
    done
fi
0

Add this to your ~/.bashrc file:

ssh-add -L|grep identities > /dev/null && ssh-add /path/to/ssh/private/key
  • I don't see how this relates to the question, which is about not being prompted for the password on subsequent logins. – Chris Down Jan 6 '14 at 13:23
0

In order to add a (possibly passwordless) key and ensure that ssh-add will not prompt for a password, no matter what, even when running under X:

DISPLAY= ssh-add -k /path/to/key </dev/null &>/dev/null

Exit status indicates success or failure.

0

If you are running seahorse as your password manager ... Which you probably are ;D

Another solution that achieves the goal you are looking for is simply adding the ssh keys to seahorse for automatic unlock upon login. The major benefit to this is that you never have to enter a password for the keys after you login through gdm, or whatever your loging in with even if the keys have a password. This REQUIRES both the private key and the public key. They also MUST follow a naming convention for seahorse. The default is acceptable (id_rsa for private key and id_rsa.pub for public key... Really anything that is privatekeyname and privatekeyname.pub)

To add you ssh key to seahorse for automatic unlock upon login; (on fedora25, I'm not sure where the path is on other distros though its most likely very similar)

/lib64/seahorse/seahorse-ssh-askpass /path/to/keys/here

For me, it was

/lib64/seahorse/seahorse-ssh-askpass ~/.ssh/id_rsa

(seahorse will automatically assume that the public key in my case was id_rsa.pub)

After executing the command, seahorse will pop open a cute little gtk password field to enter the password for the private key into. or just leave it blank if you generated the key without a password.

Seahorse won't prompt you if everything went okay. You will need to attempt to ssh into the target machine. Then seahorse will prompt you to unlock the key with a password graphically (THIS WILL ONLY HAPPEN ONCE) again but it should look a little different this time ;P (this is also the part where seahorse does some seahorse to ssh-add magic I believe), and offer the OPTION to unlock the key upon login, you must check this option to achieve your goal.

Just because I didn't read all the answers, I would recommend undoing what everyone told you to do with ssh-add before attempting this answer. Doing so otherwise might result in something bad happening to your keys, idk.

0

Here is the definitive script.

Update $PASSW, then copy-paste it in your Terminal

# <sshpass> via typinator
# Updated: 2017-01-18_21h36
#
# apt-get update -y; apt-get install expect -qy

# Pass this value to ssh-add
PASSW="myfancypass123"

# Define a name for this script
THIS_SCRIPT="$(date +%Y-%m-%d_%H-%M-%S-%N)".sh

# Create a fresh directory to work from / Clean up
rm -rf ~/temp; mkdir -p ~/temp; cd ~/temp; ls -la


# Output our bash script file - BEGIN
cat <<< '
#!/bin/bash

set -u     # Stop if an unbound variable is referenced
set -e     # Stop on first error
export HISTIGNORE="expect*";

# Normal CMDs
echo && echo "The process should take about 10 seconds:" && echo
eval "$(ssh-agent -s)"; sleep 0.5;

# Define VAR passed when this bash-script was launched
password="$@"

# Launch the expect magic
expect -c "
    spawn ssh-add /root/.ssh/id_rsa
    expect "?assword:"
    send \"$password\r\"
    expect "?password:"
    send \"$password\r\"
    expect eof"

export HISTIGNORE="";
export password="";
' > $THIS_SCRIPT
# Output our bash script file - END


# Ensure we are in the right path
cd ~/temp; ls -la; sleep 1;

# Run the bash script
chmod +x ./$THIS_SCRIPT; ./$THIS_SCRIPT "$PASSW"; unset password;

# Clean up
rm -rf ~/temp; mkdir -p ~/temp; cd ~/temp; ls -la

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.