OSX as PXEboot server

I have found myself doing a lot of Linux installs recently. I had to install a 10 node Rocks HPC cluster three weeks ago and this week I am working on a 20 node Eucalyptus cluster. In my quest to reduce time to install OS, I decided to install a PXE boot server. Now my primary laptop is a 15″ mac book pro with SSD, a very fast machine that I am pleased with. I run Parrells so it would have been a fairly trivial to quickly put together a CentOS (my favorite OS second to OSX) VM and used that to be my PXE boot server. But Since OSX is based off unix and has everything to put together a PXE boot server, I decided to use that instead.

Now, since 2009, Apple has been dumbing down the capabilities of OSX and recently launched a separate package that can be installed on top of OSX to give you GUI access to the tools that I was using. To put together a PXE boot server, i needed the following items in place.

1. DHCP server
2. TFTP boot server
3. HTTP server
4. OS ISOs etc etc

DHCP server.

Now this is fairly easy. If you have an external DHCP server, TFTP will listen to that server for its configuration. I install my machines in closed environments that do not tie to the outside world so I needed a DHCP server. OSX has a built in DHCP server that can be configured by making changes to this file.

/usr/local/etc/dhcpd.conf

In this file, you need to put in your IP subnet, and ip address range that you want to lease out. I also needed to specify the interface that would be used to lease out the IP addresses, OSX uses en0, en1, enx for the ethernet interfaces. I had to set my ip interface en0 to a static IP which can be done from the system preferences

en0: flags=8963; mtu 1500
 options=2b;
 ether 10:9a:dd:6e:b1:4e 
 inet6 fe80::129a:ddff:fe6e:b14e%en0 prefixlen 64 scopeid 0x5 
 inet 10.10.1.1 netmask 0xffff0000 broadcast 10.10.255.255
 media: autoselect

→ cat /usr/local/etc/dhcpd.conf #
# PXE Configuration
default-lease-time 600;
max-lease-time 7200;
subnet 10.10.0.0 netmask 255.255.0.0 {
range 10.10.1.2 10.10.1.10; option routers 10.10.1.1;  
option domain-name-servers 8.8.8.8, 8.8.4.4; 
filename "pxelinux.0"; }

In the DHCP config file, I have specified that 10.10.1.1(en0) is my router IP , I am leasing out IP addresses from 10.10.1.2-10.10.1.10

To start the DHCP server, you need to run the following command

sudo /usr/local/sbin/dhcpd -f -d -cf /usr/local/etc/dhcpd.conf

Also in the dhcp file, you would also have to to specify location of the pxelinux.0 file. this will be the absolute path from the root of the TFTP server. More on this on the section below

TFTP Boot server

The second piece of the puzzle is the TFTP boot server. Changes are made by modifying this file.

sudo vi /System/Library/LaunchDaemons/tftp.plist

The first thing you need is to enable the TFTP boot server. You can use VI or nano to edit the file and change

<key>Disabled</key>
<true/>

to

<key>Disabled</key>
<false/>

You also need to specify the location of your root TFTP folder that you will be dropping a couple of files and this is specified in this section

<string>/usr/libexec/tftpd</string>
		<string>-s&lt/string>
		<string>/Users/mymac/Downloads/install/</string>

To start the TFTP server, you run the following command

sudo launchctl load -F /System/Library/LaunchDaemons/tftp.plist

and to stop it

sudo launchctl unload -F /System/Library/LaunchDaemons/tftp.plist

Under the path

/Users/mymac/Downloads/install/

we will be dropping a couple of files that we will use to setup the PXE boot environment. You get some of this files from your linux distribution, in my case CentOS DVD, and from syslinux.

I created the following directory structure under the path and populated it with the directory structure and files below

/Users/mymac/Downloads/install/
-> Centos6.3/
-> Centos6.3/initrd.img
-> Centos6.3/vmlinuz
-> pxelinux.cfg/
-> pxelinux.cfg/default
-> menu.msg
-> pxelinux.0

The files Initrd.img and vmlinuz files can be obtained from your linux distribution. You are supposed to use a version as close to the version you are trying to deploy. I was deploying CentOS 6.3 and I found that I had the greatest success usings the versions under the CentOS 6.3 distro

The file menu.msg is used to display an announcement. You can put in any text you want in there. This is what I had in mine

########IT FREAKING WORKS ####################
########Centos 6.3 Installing#################
##############################################

The file pxelinux.0 is the lightweight boot loader that kicks off the PXE boot process. I obtained this from SYSLINUX org and only had success with version 4.06. Version 5.x seems to have a bug

Most of the PXEboot configuration was contained in file pxelinux.cfg/default.

→ cat default 
PROMPT 1
TIMEOUT 2
DISPLAY menu.msg
DEFAULT Centos6
LABEL linux
 localboot 0
LABEL Centos6
 MENU LABEL CentOS 6 x86_64 KS
 kernel Centos6.3/vmlinuz
 append initrd=Centos6.3/initrd.img ramdisk_size=100 ksdevice=eth0 ks=http://10.10.1.1:81/Centos6.3.ks

I had a very simple configuration, my default file had specifed my pxeboot options. Each Label in my case Label Linux and Label Centos6 represented one pxeboot boot option. The linux option was booting from local disk disk. I configured the Centos6 as the default boot option. CentOS 6 was configured to load the kernel vmlinuz and pointed to a kickstart file I was serving using http on the IP address 10.10.1.1. This was the ip address of my web server running on OSX.

OSX Web server

OSX uses apache no surprise there. Again apple has taken away the Gui niceties since Leopard. I wanted to serve up the CentOS 6.3 repo and the Kickstart file using HTTP. To do this I used Apache port separation to separate my HTTP sessions. HTTP is controlled by making modifications to this files.

sudo vi /etc/apache2/httpd.conf

In this file, i allowed for the use of virtual hosts by uncommenting this line

Include /private/etc/apache2/extra/httpd-vhosts.conf

Now alittle explanation on virtual hosts. Apache gives you the choice of name based or port based virtual web servers.Virtual web servers allow you to have different sites being served from the same web servers. I found port virtual hosts gave me much more flexibility for what I was trying to do. This are controlled by httpd-vhosts.conf file. I edited the file and added the following snippet a the bottom of the file; this pointed the location of my kickstart folder to port 81.

sudo vi /private/etc/apache2/extra/httpd-vhosts.conf

&ltVirtualHost *:81&gt
    DocumentRoot "/Users/mymac/Sites/ks"
    ServerName ks
&ltVirtualHost&gt

I then restarted apache using the following command.

sudo apachectl gracefull

Inside the folder /Users/mymac/Sites/ks I had the following items.

1. /Users/mymac/Sites/ks/Centos6.3.ks kickstart file referenced by my pxeboot/default file and
2. /Users/mymac/Sites/ks/x86_64 that contained files from my distro which I pulled a CentOS 6.3 DVD iso Image

One more thing

KICKSTART file

I created the following kickstart file.

→ cat /Users/mymac/Sites/ks/Centos6.3.ks 
install
url --url http://10.10.1.1:81/x86_64/
lang en_US.UTF-8
keyboard us
timezone --utc America/Los_Angeles
network --noipv6 --onboot=yes --bootproto dhcp
authconfig --enableshadow --enablemd5
rootpw --plaintext CHANGEME
firewall --enabled --port 22:tcp
selinux --disabled
bootloader --location=mbr --driveorder=sda --append="crashkernel=auth rhgb"

# Disk Partitioning
clearpart --all --initlabel --drives=sda
part /boot --fstype=ext4 --size=200 #create boot of size 200MB
part pv.1 --grow --size=1
volgroup vg1 --pesize=4096 pv.1

logvol / --fstype=ext4 --name=lv001 --vgname=vg1 --size=6000 
logvol /var --fstype=ext4 --name=lv002 --vgname=vg1 --grow --size=1 #create var rest of drive
logvol swap --name=lv003 --vgname=vg1 --size=2048 #create swap size 2G
# END of Disk Partitioning

# Make sure we reboot into the new system when we are finished
reboot

# Package Selection
%packages --nobase --excludedocs
@core
-*firmware
-iscsi*
-b43-openfwwf
kernel-firmware
-efibootmgr
wget
sudo
perl
ntp

%pre

%post --log=/root/install-post.log
(
PATH=/bin:/sbin:/usr/sbin:/usr/sbin
export PATH

# PLACE YOUR POST DIRECTIVES HERE
) 2>&1 >/root/install-post-sh.log
EOF

You can create your own custom custom configuration, but the most important bit was to point to the location of the distro files that you will be installing. I was using a web server so my configuration had the line highlighted in red that pointed it to the location being served by my http server. Notice the port 81 in the url.

Putting it all together

I launched the dhcp server using the command

sudo /usr/local/sbin/dhcpd -f -d -cf /usr/local/etc/dhcpd.conf

It runs actively on terminal and you can see whether you are actually leasing out ip address. I hope this helps out someone out there.