Comment 2 for bug 931584

I agree with Steve here: Upstart is careful to set CLOEXEC on all the fds it has control over, so I would much prefer this be fixed at source.

A temporary work-around for jobs that are affected by this issue is to add the following code at the top of all appropriate script sections - remember that all script and exec sections run as different processes so if you have a pre-start and a script section, the code below should be added to the top of *both* those sections:

script
  # close unexpectedly open fds
  # The 'readlink' ensures we don't try to close the fd we're using to read the /proc/self/fd directory.
  for fd in /proc/self/fd/*
  do
    fd=$(basename $fd)
    case "$fd" in
      0|1|2) ;;
      *) readlink $fd && eval "exec $fd>&-" || :;;
    esac
  done

  # << do something useful here >>

end script