Skip site navigation (1)Skip section navigation (2)

FreeBSD Manual Pages


home | help
DISKLESS(8)		  BSD System Manager's Manual		   DISKLESS(8)

     diskless -- booting a system over the network

     The ability to boot a machine over	the network is useful for diskless or
     dataless machines,	or as a	temporary measure while	repairing or re-in-
     stalling file systems on a	local disk.  This file provides	a general de-
     scription of the interactions between a client and	its server when	a
     client is booting over the	network.

     When booting a system over	the network, there are three phases of inter-
     action between client and server:

     1.	  The stage-1 bootstrap, typically PXE built into your Ethernet	card,
	  loads	a second-stage boot program.

     2.	  The second-stage boot	program, typically pxeboot(8), loads modules
	  and the kernel, and boots the	kernel.

     3.	  The kernel NFS mounts	the root directory and continues from there.

     Each of these phases are described	in further detail below.

     First, the	stage-1	bootstrap loads	the stage-2 boot program over the net-
     work.  The	stage-1	bootstrap typically uses BOOTP or DHCP to obtain the
     filename to load, then uses TFTP to load the file.	 This file is typi-
     cally called pxeboot, and should be copied	from /boot/pxeboot into	the
     TFTP directory on the server, which is typically /tftpdir.

     The stage-2 boot program then loads additional modules and	the kernel.
     These files may not exist on the DHCP or BOOTP server.  You can use the
     next-server option	available in DHCP configurations to specify the	server
     holding the second	stage boot files and kernel.  The stage-2 program uses
     NFS or TFTP to obtain these files.	 By default, NFS is used.  If you are
     using pxeboot(8), you can install a version that uses TFTP	by setting
     LOADER_TFTP_SUPPORT=YES in	your /etc/make.conf, then recompiling and re-
     installing	pxeboot(8) via the command listed below.  It is	often neces-
     sary to use TFTP here so you can place a custom kernel in /tftpdir/.  If
     you use NFS and do	not have a custom root file system for the diskless
     client, the stage-2 boot will load	your server's kernel as	the kernel for
     the diskless machine, which may not be what you want to have happen.

	   cd /usr/src/sys/i386/boot
	   make	clean; make; make install
	   cp /boot/pxeboot /tftpdir/

     In	phase 3, the kernel again uses DHCP or BOOTP to	acquire	configuration
     information, and proceeds to mount	the root file system and start opera-
     tion.  The	boot scripts recognize a diskless startup and perform the ac-
     tions found in /etc/rc.d/initdiskless and /etc/rc.d/diskless.  On older
     systems, the scripts are located in /etc/rc.diskless1 and

     In	order to run a diskless	client,	you need the following:

     +o	 An NFS	server which exports a root and	/usr partitions	with appropri-
	 ate permissions.  The diskless	scripts	work with read-only parti-
	 tions,	as long	as root	is exported with -maproot=0 so that some sys-
	 tem files can be accessed.  As	an example, /etc/exports can contain
	 the following lines:

	       <ROOT> -ro -maproot=0 -alldirs <list of diskless	clients>
	       /usr -ro	-alldirs <list of diskless clients>

	 where <ROOT> is the mount point on the	server of the root partition.
	 The script /usr/share/examples/diskless/clone_root can	be used	to
	 create	a shared read-only root	partition, but in many cases you may
	 decide	to export (again as read-only) the root	directory used by the
	 server	itself.

     +o	 A BOOTP or DHCP server.  bootpd(8) can	be enabled by uncommenting the
	 "bootps" line in /etc/inetd.conf.  A sample /etc/bootptab can be the



	 where <SERVER>, <GATEWAY> and <ROOT> have the obvious meanings.

     +o	 A properly initialized	root partition.	 The script
	 /usr/share/examples/diskless/clone_root can help in creating it, us-
	 ing the server's root partition as a reference.  If you are just
	 starting out, you should simply use the server's own root directory,
	 /, and	not try	to clone it.

	 You often do not want to use the same rc.conf or rc.local files for
	 the diskless boot as you do on	the server.  The diskless boot scripts
	 provide a mechanism through which you can override various files in
	 /etc (as well as other	subdirectories of root).  The scripts provide
	 four overriding directories situated in /conf/base, /conf/default,
	 /conf/_broadcast-ip_, and /conf/_machine-ip_.	You should always cre-
	 ate /conf/base/etc, which will	entirely replace the server's /etc on
	 the diskless machine.	You can	clone the server's /etc	here or	you
	 can create a special file which tells the diskless boot scripts to
	 remount the server's /etc onto	/conf/base/etc.	 You do	this by	creat-
	 ing the file /conf/base/etc/diskless_remount containing the mount
	 point to use as a basis of the	diskless machine's /etc.  For example,
	 the file might	contain:

	 Alternativly, if the server contains several independent roots, the
	 file might contain:

	 This would work, but if you copied /usr/diskless/4.7-RELEASE to
	 /usr/diskless/4.8-RELEASE and upgraded	the installation, you would
	 need to modify	the diskless_remount files to reflect that move.  To
	 avoid that, paths in diskless_remount files begining with / have the
	 actual	path of	the client's root prepended to them so the file	could
	 instead contain:


	 The diskless scripts create memory file systems to hold the overriden
	 directories.  Only a 2MB partition is created by default, which may
	 not be	sufficient for your purposes.  To override this, you can cre-
	 ate the file /conf/base/etc/md_size containing	the size, in 512 byte
	 sectors, of the memory	disk to	create for that	directory.

	 You then typically provide file-by-file overrides in the
	 /conf/default/etc directory.  At a minimum, you must provide over-
	 rides for /etc/fstab, /etc/rc.conf, and /etc/rc.local via
	 /conf/default/etc/fstab, /conf/default/etc/rc.conf, and

	 Overrides are hierarchical.  You can supply network-specific defaults
	 in the	/conf/<BROADCASTIP>/etc	directory, where <BROADCASTIP> repre-
	 sents the broadcast IP	address	of the diskless	system as given	to it
	 via BOOTP.  The diskless_remount and md_size features work in any of
	 these directories.  The configuration feature works on	directories
	 other then /etc, you simply create the	directory you wish to replace
	 or override in	/conf/{base,default,_broadcast_,_ip_}/*	and work it in
	 the same way that you work /etc.

	 Since you normally clone the server's /etc using the
	 /conf/base/etc/diskless_remount, you might wish to remove unneeded
	 files from the	memory file system.  For example, if the server	has a
	 firewall but you do not, you might wish to remove /etc/ipfw.conf.
	 You can do this by creating a /conf/base/<DIRECTORY>.remove file.
	 For example, /conf/base/etc.remove, which contains a list of relative
	 paths that the	boot scripts should remove from	the memory file	sys-

	 As a minimum, you normally need to have the following in

	       <SERVER>:<ROOT> /     nfs    ro 0 0
	       <SERVER>:/usr   /usr  nfs    ro 0 0
	       proc	       /proc procfs rw 0 0

	 You also need to create a customized version of
	 /conf/default/etc/rc.conf which should	contain	the startup options
	 for the diskless client, and /conf/default/etc/rc.local which could
	 be empty but prevents the server's own	/etc/rc.local from leaking
	 onto the diskless system.

	 In rc.conf, most likely you will not need to set hostname and
	 ifconfig_* because these will be already set by the startup code.
	 Finally, it might be convenient to use	a case statement using
	 `hostname` as the switch variable to do machine-specific configura-
	 tion in case a	number of diskless clients share the same configura-
	 tion files.

     +o	 The kernel for	the diskless clients, which will be loaded using NFS
	 or TFTP, should be built with at least	the following options:

	       options BOOTP
	       options BOOTP_NFSROOT
	       options BOOTP_COMPAT

	 In the	devices	section	add:

	       device md

	 If you	use the	firewall, remember to default to "open", or your ker-
	 nel will not be able to send/receive the BOOTP	packets.

     Be	warned that using unencrypted NFS to mount root	and user partitions
     may expose	information such as encryption keys.

     This manpage is probably incomplete.

     FreeBSD sometimes requires	to write onto the root partition, so the
     startup scripts mount MFS file systems on some locations (e.g. /etc and
     /var), while trying to preserve the original content.  The	process	might
     not handle	all cases.

     ethers(5),	exports(5), bootpd(8), mountd(8), nfsd(8), pxeboot(8),
     reboot(8),	tftpd(8)


BSD			       December	23, 2002			   BSD


Want to link to this manual page? Use this URL:

home | help