Recipe 5.15 Killing Processes via sudo

5.15.1 Problem

Allow a user to kill a certain process but no others.

5.15.2 Solution

Create a script that kills the process by looking up its PID dynamically and safely. Add the script to /etc/sudoers.

5.15.3 Discussion

Because we don't know a process's PID until runtime, we cannot solve this problem with /etc/sudoers alone, which is written before runtime. You need a script to deduce the PID for killing.

For example, to let users restart sshd :

#!/bin/sh
pidfile=/var/run/sshd.pid
sshd=/usr/sbin/sshd

# sanity check that pid is numeric
pid=`/usr/bin/perl -ne 'print if /^\d+$/; last;' $pidfile`
if [ -z "$pid" ]
then
        echo "$0: error: non-numeric pid $pid found in $pidfile" 1>&2
        exit 1
fi

# sanity check that pid is a running process
if [ ! -d "/proc/$pid" ]
then
        echo "$0: no such process" 1>&2
        exit 1
fi

# sanity check that pid is sshd
if [ `readlink "/proc/$pid/exe"` != "$sshd" ]
then
        echo "$0: error: attempt to kill non-sshd process" 1>&2
        exit 1
fi

kill -HUP "$pid"

Call the script /usr/local/bin/sshd-restart and let users invoke it via sudo:

# /etc/sudoers:
smith ALL = /usr/local/bin/sshd-restart ""

The empty double-quotes prevent arguments from being passed to the script. [Recipe 5.9]

Our script carefully signals only the parent sshd process, not its child processes for SSH sessions already in progress. If you prefer to kill all processes with a given name, use the pidof command:

# kill -USR1 `pidof mycommand`

or the skill command:

# skill -USR1 mycommand

5.15.4 See Also

kill(1), proc(5), pidof(8), skill(1), readlink(1).



    Chapter 9. Testing and Monitoring