ssh-agent not getting cleaned on unclean exit

[MagikEh]

Issue:

When you run a command that invokes the sr3_ssh script, it sets up a ssh-agent but that agent's socket is not cleaned up if the script unexpectedly exits. This leads to an absolute metric truckload of $TMPDIR/ssh-* dirs being created on the host.

Steps to reproduce

ls -dl  $TMPDIR/ssh-*
> # nothing returned

sr3l 'echo blah'
> host1: blah
> host2: blah
> host3: blah

ls -dl  $TMPDIR/ssh-*
> # nothing returned because the agent was cleaned up properly


sr3l 'echo blah'
> host1: blah
^C  #<--- Force the run to stop before it completes.

ls -dl  $TMPDIR/ssh-*
drwx------ 2 user      user      26 Feb  6 19:58 /tmp/ssh-XXXXXX2j1w9S/

Suggested Fix:

Add a trap to the bash script for on-exit signals:

trap 'ssh-agent -k' EXIT

[MagikEh]

Attempted to run with

if [ -e "${ssh_user_config}" ] && { [ -z "${SSH_AUTH_SOCK}" ] || [ -z "${SSH_AGENT_PID}" ]; } && grep -q "ssh-add" "${ssh_user_config}"; then
    eval "$(ssh-agent -s)" > /dev/null
    trap 'ssh-agent -k' EXIT
    trap 'ssh-agent -k' ERR
fi

And commented out the Clean up ssh-agent code at the bottom and it worked for normal scenarios, but not the ctrl+c..

I wonder if it's because the ssh-agent is being executed within a child shell..

[reidsunderland]

This is interesting, and I think you got 98% of the way there. There might have also be some confusion when you were testing, because sr3l doesn't use sr3_ssh. I had to modify sr3r (which calls dsh), sr3_ssh (which calls regular ssh) and sr3_scp. The ssh-agent stuff is duplicated in all 3. I got confused myself when testing.

I think also trapping SIGINT fixes it. Can you try the issue23 branch?

[MagikEh]

Reid has tested, I have confirmed it to be worky.