Extending your X11 Desktop using VNC

This is mostly for debugging purposes and will hold most of the relevant information for my SSH tunnel and the VNC server.

TMPD=/tmp/extra_screen
if ! [[ -d $TMPD ]]; then
    mkdir -p $TMPD
fi

The below part is still a little messy, but it gets the job done for now. I'm using the new monitor in a portrait orientation, so I'm creating a output of dimension 1024x1280. I'm also telling xrandr to put this output to the right of the output DP2.

# the local virtual xrandr device
DEVICE="VIRTUAL1"

# get the modeline from
# gtf 1024 1280 60 | sed -n 's/.*Modeline "\([^" ]\+\)" \(.*\)/\1 \2/p'
MODELINE="1024x1280_60.00 110.66 1024 1096 1208 1392 1280 1281 1284 1325 -HSync +Vsync"
NAME="1024x1280_60.00"

# FIXME get rid of error messages
xrandr --delmode "$DEVICE" "${NAME}"
xrandr --rmmode "${NAME}"
xrandr --newmode ${MODELINE}
xrandr --addmode "$DEVICE" "${NAME}"
xrandr --output $DEVICE --mode $NAME --right-of DP2

The above commands extend my desktop over a—as of right now—invisible area of 1024x1280 pixels. We could theoretically drag windows in this area. Next, I'm using VNC to make this area available to a client on my desktop machine.

From now on, any reference to $REMOTE should be replaced with SSH username and hostname, e.g. [email protected].

PW=$(openssl rand -hex 50 | vncpasswd -f | tee $TMPD/vncpw)
scp $TMPD/vncpw $REMOTE:~/.ava_vncpw

Since this is the 21st century, we want to encrypt any traffic that we send through a network. Here, we set up a reverse SSH Tunnel, which forwards port 5900 on my laptop to port 5900 on my desktop machine through an encrypted pipe. I'm also using the nohup command to send the process in the background and have it persist even after I close my terminal.

PORT=5900
nohup ssh -2tnNv -R $PORT:localhost:$PORT $REMOTE > $TMPD/tunnel_log 2>&1 &
echo -n $! > $TMPD/tunnel_pid # write the pid of the sshtunnel to kill later

x11vnc has a bunch of configuration options and I spent a significant amount of time reading tweaking its configuration to get the best performance.

CLIP=$(xrandr | grep "^$DEVICE.*$" | grep -o '[0-9]*x[0-9]*+[0-9]*+[0-9]*')
nohup x11vnc -clip $CLIP -noxinerama -noxrandr \
      -repeat -localhost -nevershared -forever \
      -rfbauth $TMPD/vncpw \
      -nowf -noncache -wait 1 -defer 1 > $TMPD/x11vnc_log.log 2>&1 &
echo -n $! > $TMPD/x11vnc_pid # write the pid of x11vnc to kill later

• -clip defines the area which is exported. This coincides with the area of the output created using the mode line above

• I had to set -noxinerama to get rid of some nasty letterboxing

• Without setting -repeat, I couldn't enter a letter more than once by long-pressing a key on the keyboard. That was annoying

• -localhost lets x11vnc only accept connections from localhost, which is fine since we're using a SSH tunnel.

• -forever makes x11vnc not quit after the first client quits

• without the -noncache switch, x11vnc exported a much larger area filled with mostly black pixels

This saves a few keystrokes during debugging.

KILL=$TMPD/kill.sh
echo "#!/bin/sh" > $KILL
echo -n "kill $(cat $TMPD/tunnel_pid) && " >> $KILL
echo -n "kill $(cat $TMPD/x11vnc_pid) && " >> $KILL
echo "echo \"Killed x11vnc and SSH.\"" >> $KILL
chmod +x $KILL

Since I'm not doing anything else on this machine other than running the VNC client, I have an infinite loop in my .xinitrc file which attempts to restart the client if it is not running.

# rotate the display
xrandr --output DVI-0 --rotate left
# disable screen sleep
xset s off
xset -dpms
# unclutter removes the local mouse pointer
unclutter &
force reloading of the vnc viewer to simulate plug & play
while true; do
    if ! ps aux | grep -v grep | grep -q vncviewer; then
        vncviewer localhost:0 -passwd ~/.ava_vncpw -viewonly -fullscreen -encodings "copyrect tight"
    else
      sleep 0.5
    fi
done

This was fun. I hope this inspires someone to do their own thrifty grad student multihead set-up.