Notes from a Messy Desk

Hi,

We have a problem with ssh-agent and ssh-add.

We have our program which starts the ssh-agent and stores its
environment variables in a hash dictionary. Another function in the
same program would read these environment variables, set the same
(using putenv()), start a pty and then add the key to the agent using
ssh-add (pty is started in order to give the passphrase as the input).

But, while executing ssh-add, we are getting the error "Could not open
a connection to your authentication agent.". We have ensured that
environment variables are set by executing getenv().

Within a process, we are starting the agent like this,
code snippet:

fp = popen("ssh-agent -s","r");
while(fgets(temp_buff,100,fp) !=NULL)
{
if(line_count ==2){
break;}
line_count++;

char *buff;
buff= new char[100];
bzero(buff,100);

int i =0, buff_len =0, rc1;
for(int j=0;jagentEnvs->env1 = new char[buff_len+1];
strncpy(env_value->agentEnvs->env1,buff,buff_len);
env_value->agentEnvs->env1[buff_len] = '';
env_value->len1 = buff_len+1;
rc1= putenv(env_value->agentEnvs->env1); //Here !!
bzero(buff,100);
break; //Out of the for loop
}
else if(line_count == 2)
{
env_value->agentEnvs->env2 = new char[buff_len+1];
strncpy(env_value->agentEnvs->env2,buff,buff_len);
env_value->agentEnvs->env2[buff_len] = '';
env_value->len2 = buff_len+1;
rc1= putenv(env_value->agentEnvs->env2); //Here !!
bzero(buff,100);
break; //Out of the for loop
}
}
buff[i++] = temp_buff[j];
}
}
pclose(fp);
system("ssh-add -l"); //testing purpose

Both ssh-agent and ssh-add are run by the same user.

While we start the agent, we are capturing the ssh-agent environment
variables (SSH_AUTH_SOCK and SSH_AGENT_PID) and storing it.
i.e., the cmd "ssh-agent -s" would result in an output like this,
SSH_AUTH_SOCK=/tmp/ssh-QDL22304/agent.22304; export SSH_AUTH_SOCK;
SSH_AGENT_PID=22306; export SSH_AGENT_PID;
echo Agent pid 22306;
Here, we are cutting the first env SSH_AUTH_SOCK until we find the
delimiter ";" and then we are appending a null character "" for the
cut string. Similarly, two env variables are extracted and stored.
Well, then this string is given as input for putenv().

It would be very helpful and greatly appreciated if you can help us
in solving this problem.

Thanks
Guru
sgprasad@gmail.com

Thanks! This is the most elegant solution I've found yet. I've slightly modified it so that I can ssh into another host (with the same bashrc) and then ssh from that host onwards; the terminal type of the remote shell will be 'screen', but it won't have access to the ${screen_ssh_agent} socket. So I just avoid changing the socket if the permanent one doesn't exist.

----------------
# slightly modified from http://woss.name/2005/08/17/using-ssh-agent-and-screen-together/
if [ -n "$SSH_AUTH_SOCK" ]; then
user=`whoami`
screen_ssh_agent="/tmp/${user}-screen-ssh-agent.sock"
if [ "$TERM" = "screen" ]; then
# shell in a screen. If we're on the host with the permanent name
# for the ssh agent, use that instead. if this shell is
# in an ssh to a different host (which happens to be in a screen),
# though, we'll keep the variable name (which will be
# persistent as long as that ssh remains up, exactly what we want).
if [ -e ${screen_ssh_agent} ]; then
export SSH_AUTH_SOCK=${screen_ssh_agent}
fi
else
ln -snf ${SSH_AUTH_SOCK} ${screen_ssh_agent}
fi
fi
----------------

Great solution but using STY is more reliable then matching against the TERM env.

if [ -n "$SSH_AUTH_SOCK" ]; then
screen_ssh_agent="/tmp/${USER}-screen-ssh-agent.sock"
if [ ${STY} ]; then
if [ -e ${screen_ssh_agent} ]; then
export SSH_AUTH_SOCK=${screen_ssh_agent}
fi
else
ln -snf ${SSH_AUTH_SOCK} ${screen_ssh_agent}
fi
fi