Supporting network boot via PXE required a few changes to various configurations.

Server side changes

On my Gentoo system, I needed to

emerge tftp-hpa
rc-update add in.tftpd default

The configuration change for in.tftpd was in /etc/conf.d/in.tftpd:

/etc/conf.d/in.tftpd

# Config file for /etc/init.d/in.tftpd
# Remove the -l if you use [x]inetd

INTFTPD_PATH="/export/tftpboot/"
INTFTPD_OPTS="-l -s ${INTFTPD_PATH}"

I already had a DHCPd running on my server, so the setup here was fairly trivial for me. I can remember, though, that the DHCPd setup was mostly easy in the first place. There were some global settings that needed to be provided, as well as settings specific to the MythTV host I was installing.

/etc/dhcpd.conf

• Global changes:

  .
  .
  .
  allow booting;
  allow bootp;
  
  option subnet-mask              255.255.255.224;
  option broadcast-address        172.16.0.31;
  option routers                  172.16.0.3;
  option domain-name-servers      172.16.0.3;
  option log-servers              172.16.0.3;
  option domain-name              "house";
  .
  .
  .
  

• Host specific:

          .
          .
          .
          # The MythTV box.
          host myth {
               next-server        172.16.0.3;
               hardware ethernet  00:0c:6e:25:cd:e0;
               fixed-address      172.16.0.9;
               filename           "172.16.0.9/pxelinux.0";
          }
          .
          .
          .
  

Now that all the pre-requisites are setup, the PXE boot step is next. PXE does most of the heavy lifting for network boot in my case. The first step is to install the pxelinux.0 file in the proper place. This is the file that the remote client asks for as a "kernel" image when netbooting.

pxelinux.0 seems to load enough brains into the machine to figure out to go ask for a host specific configuration file which will tell it what Linux kernel file to ask for. From that point, the kernel boots, mounts the root filesystem via NFS and the boot process is the same as a normal boot.

When the PXE ROM begins booting, it does the following:

1. DHCP for an IP address. This address will include information about what kernel filename to download. In the case of my setup, it's 172.16.0.9/pxelinux.0.
2. TFTP the kernel file specified by DHCP.
3. pxelinux.0 tries to TFTP the configuration file for PXELinux. This file is located, on my system, in /export/tftpboot/172.16.0.9/pxelinux.cfg. Naming the file default is sufficient.
4. PXELinux follows the instructions contained in the configuration file.

pxelinux.cfg/default

label linux
kernel root/boot/bzImage-2.4.26
   append nfsroot=172.16.0.3:/export/tftpboot/172.16.0.9/root,wsize=8192, \
               rsize=8192,nfsvers=3 video=vc:64-64
   ipappend 1

This file is pretty similar to what you'd see in a lilo.conf or grub.conf. A kernel image is specified, and then any necessary boot paramters are listed. It's also possible to specify an initrd image to be used, but I've found that I really don't need that at this time.

Client side changes

This was simple: Simply specify in the BIOS to use the Broadcom network bood ROM to boot the system. Et voila!

• Stale NFS File Handles

  For some time, I was receiving a large number of "Stale NFS file handle" errors when accessing the NFS mounted file systems. This was especially true for the root partition, which was mounted with the nfsroot= option in my PXE boot setup. I have not seen the problem since I changed my mount options for all NFS mounts to include the following:

  rsize=8192,wsize=8192,nfsvers=3,tcp

  I believe it's the tcp and nfsvers=3 options that are eliminating the problem. It reappears occasionally, but it's not nearly the troublemaker it once was.

I wanted to have the backend startup just like any other service under Gentoo. I wrote the following mythbackend script that lives in /etc/init.d/mythbackend:

/etc/init.d/mythbackend

#!/sbin/runscript

MYTHTVDIR="/home/mythtv"
LOGFILE="/home/mythtv/logs/be.log"
PIDFILE="/var/run/mythtv/mythbe.pid"
MYTHBE="/usr/local/bin/mythbackend"
NFSLOCKFILE="/mnt/store/nfslockfile.lock"
#OPTIONS="--daemon --logfile $LOGFILE --pidfile $PIDFILE --verbose all"
OPTIONS="--daemon --logfile $LOGFILE --pidfile $PIDFILE"

# Start up the MythTV backend server and lircd.

depend() {
        need net netmount lircd
}

start() {
        local retval=0

        ebegin "Starting MythTV backend process"
        export HOME=$MYTHTVDIR

        # Clean the NFS lock file
        [ -f $NFSLOCKFILE ] && rm -f $NFSLOCKFILE

        # Start the backend service
        echo "==============================" >> $LOGFILE
        echo "Starting mythbackend at `date`" >> $LOGFILE
        echo "==============================" >> $LOGFILE
        start-stop-daemon --start --quiet --chuid mythtv --exec $MYTHBE -- $OPTIONS

        eend $? "Failed to start mythbackend"
}

stop() {
        ebegin "Stopping mythbackend"

        start-stop-daemon --stop --quiet --pidfile $PIDFILE
        echo "==============================" >> $LOGFILE
        echo "Stopped  mythbackend at `date`" >> $LOGFILE
        echo "==============================" >> $LOGFILE
        eend $? "Failed to stop mythbackend"

        # Clean the pidfile
        [ -f $PIDFILE ] && rm -f $PIDFILE
}

That process was pretty simple to setup. The echo commands just insert markers into the backend log file so that I know when things start and stop.

Setting things up to run automatically at boot is easy:

rc-update add mythbackend default

Starting the frontend is more tricky. It turns out that starting X automatically isn't the most straight-forward task on a Unix system. Because I wanted to have control over it like any other service, I wrote another script for /etc/init.d:

/etc/inid.d/mythfrontend

#!/sbin/runscript

# program to exec
MYTH_HOME=/home/mythtv
FRONTEND=$MYTH_HOME/init-frontend.sh
PIDFILE="/var/run/mythtv/mythfe.pid"
LOGFILE="/home/mythtv/logs/fe.log"
#OPTIONS="--logfile $LOGFILE"
OPTIONS="--verbose playback --logfile $LOGFILE"

# Ensure that MythTV backend has started.  This wouldn't work so
# well on a system that's a remote frontend.  In that case, some sort
# of last nonsense would be best.
depend() {
        need mythbackend
}

start() {
        ebegin "Starting MythTV frontend process"

        /usr/sbin/alsactl restore 0

        # Need to run xinit /usr/local/bin/mythfrontend as mythtv
        echo "===============================" >> $LOGFILE
        echo "Starting mythfrontend at `date`" >> $LOGFILE
        echo "===============================" >> $LOGFILE
        start-stop-daemon --start --background --quiet --exec $FRONTEND

        eend 0
}

stop() {
        ebegin "Stopping MythTV frontend process"

        # This doesn't stop the frontend process by pid,
        # rather it goes postal on the xinit process
        # which everything is hanging on.
        start-stop-daemon --stop --quiet --name xinit

        eend $? "Failed to stop MythTV frontend process"

        echo "==============================" >> $LOGFILE
        echo "Stopped mythfrontend at `date`" >> $LOGFILE
        echo "==============================" >> $LOGFILE
}

Again, the echo commands are just markers in the log files.

As you can see, this script calls init-frontend.sh:

init-frontend.sh

#!/bin/bash

MYTH_HOME=/home/mythtv
MYTH_FE_LOG=/home/mythtv/logs/fe.log
XINITRC=/home/mythtv/.xinitrc
PATH=$PATH:/usr/bin:/usr/X11R6/bin
export PATH
cd $MYTH_HOME

# Take a nap for a while to wait for the
# backend process to get started.
sleep 10

sudo -H -u mythtv xinit $XINITRC

This simply runs xinit as user mythtv. The .xinitrc contains the following:

.xinitrc

LOGFILE="/home/mythtv/logs/fe.log"
xsetroot -solid black &

while [ 1 ]; do
    /usr/local/bin/mythfrontend  >> $LOGFILE 2>&1
done

The while loop ensures that if mythfrontend exits, it starts right back up again.

Because of the way this is all setup, I can now simply restart mythfrontend remotely by issuing commands via SSH:

/etc/init.d/mythfrontend restart

Getting this to start automatically at boot is easy too:

rc-update add mythfrontend default

It works for me, and is simple for my poor brain.

This setup also allows me to not have to worry about window managers under X, and logging users in automatically via that. In my opinion, there's no reason to have a window manager, since mythfrontend is the only application running under X. (I've never been able to figure out why anyone actually bothers with a window manager for myth. If there's only one application running, why do you need something that manages multiple windows?)

This is the old method I used for starting myth up properly. It wasn't easy to start/stop this remotely, so I ditched it.

1. Create an autologin program. This is done as root:

   autologin.c

   #include 
   
   int main (void) {
      execlp ("login", "login", "-f", "mythtv", 0);
   }
   

   Build with gcc -o mythtv-autologin autologin.c

2. Call the resulting program from inittab:

   Excerpt from /etc/inittab (on a Gentoo system):

   # TERMINALS
   c1:12345:respawn:/sbin/agetty 38400 tty1 linux
   c2:12345:respawn:/sbin/agetty 38400 tty2 linux
   c3:12345:respawn:/sbin/agetty 38400 tty3 linux
   c4:12345:respawn:/sbin/agetty 38400 tty4 linux
   c5:12345:respawn:/sbin/agetty -n -l /usr/local/sbin/mythtv-autologin 38400 tty5 linux
   

   This logs mythtv in automatically on virtual console 5.

3. Create a .xinitrc to start mythfrontend:

   /home/mythtv/.xinitrc

   /usr/local/bin/mythfrontend >> myth.log 2>&1
   

4. Make bash call startx if appropriate:

   /home/mythtv/.bash_profile

   # /etc/skel/.bash_profile:
   # $Header: /home/cvsroot/gentoo-src/rc-scripts/etc/skel/.bash_profile,v 1.10 200
   2/11/18 19:39:22 azarah Exp $
   
   #This file is sourced by bash when you log in interactively.
   [ -f ~/.bashrc ] && . ~/.bashrc
   
   if [ -z "$DISPLAY" ] && [ $(tty) == /dev/vc/5 ]; then
       startx
   fi
   

I decided that it was probably easiest to use Apple's DVD Player application to play DVDs on the mini. I wrote a simple little shell script that MythDVD calls when it want's to play a DVD. The script is used to check (poorly) if a DVD is mounted and run DVD Player if it is. Otherwise it does nothing. This was needed because DVD Player would start if no disk was inserted, and that looked goofy.

dvd.sh

#!/bin/sh

DISK1=`mount | grep -i disk1`
DISK2=`mount | grep -i disk2`

if [ "" != "$DISK1" ]; then
   /Applications/DVD\ Player.app/Contents/MacOS/DVD\ Player
else
   echo "No /dev/disk1."
fi

if [ "" != "$DISK2" ]; then
   /Applications/DVD\ Player.app/Contents/MacOS/DVD\ Player
else
   echo "No /dev/disk2."
fi

On my mini, it seems that the DVD device is /dev/disk1 or sometimes /dev/disk2, so that's why I do the grep for "disk1" and "disk2".

I've got the DVD Player software setup to start in full screen mode, have the controls disappear in 1 second, use the "Mean" de-interlacing setting, and start playing where the disk left off last. This makes it start up nicely with only a couple seconds worth of non-Myth OSX GUI elements showing on screen.

I believe that I need to write a simple AppleScript to make DVD Player exit properly and spit out the DVD. There's not a lot of buttons on the KeySpan remote, so I don't want to dedicate one to "Eject". Basically, I need to map a key in DVD Player such that it will "Eject" then "Quit". Not hard, but I've been lazy.