NFS-Root mini-HOWTO

not maintained

V9, 20 September 2002
This mini-HOWTO tries explains how to set up a ``diskless'' Linux workstation, which mounts its root filesystems via NFS. The newest version of this mini-HOWTO can always be found at http://www.tldp.org/HOWTO/mini/NFS-Root.html or a Linux Documentation Project mirror NEAR YOU.

1. Copyright

(c) 1996 Andreas Kostyrka (e9207884@student.tuwien.ac.at or andreas@ag.or.at)

Unless otherwise stated, Linux HOWTO documents are copyrighted by their respective authors. Linux HOWTO documents may be reproduced and distributed in whole or in part, in any medium physical or electronic, as long as this copyright notice is retained on all copies. Commercial redistribution is allowed and encouraged; however, the author would like to be notified of any such distributions.

All translations, derivative works, or aggregate works incorporating any Linux HOWTO documents must be covered under this copyright notice. That is, you may not produce a derivative work from a HOWTO and impose additional restrictions on its distribution. Exceptions to these rules may be granted under certain conditions; please contact the Linux HOWTO coordinator at the address given below.

In short, we wish to promote dissemination of this information through as many channels as possible. However, we do wish to retain copyright on the HOWTO documents, and would like to be notified of any plans to redistribute the HOWTOs.

If you have questions, please contact Andreas Kostyrka <mailto:andreas@ag.or.at>, the author of this mini-HOWTO, or the Linux HOWTO coordinator, at <mailto:linux-howto@sunsite.unc.edu> via email.

1.1 Contributors

2. General Overview

An NFS-mounted root filesystem is typically most useful in two situations:

(In this document we'll use the terms client and workstation interchangeably.)

However, there are two small problems from the client's perspective:

The current implementation of NFSROOT in the Linux kernel (as of 2.4.x) allows for several approaches, including:

Since the most common dynamic-address protocol these days is DHCP, its addition as an option in kernels 2.2.19 and 2.4.x (3 < x <= 14) is particularly welcome.

Before starting to set up a diskless environment, you should decide if you will be booting via LILO, LOADLIN, or a custom, embedded bootloader. The advantage of using something like LILO is flexibility; the disadvantage is speed--booting a Linux kernel without LILO is faster. This may or may not be a consideration.

3. Setup on the server

3.1 Compiling the kernels

On the server side, if you don't plan to use the old, user-mode NFS daemon, you'll need to compile NFS server support into the kernel (``NFS server support,'' a.k.a. knfsd or CONFIG_NFSD). If you plan to use the older RARP protocol to assign the client an IP address, RARP support in the kernel of the server is probably a good idea. (You must have it if you will boot via RARP without kernel parameters.) On the other hand, it doesn't help you if the client isn't on the same subnet as the server.

The kernel for the workstation needs the following settings, as a minimum:

Where there is an option to compile something in as a module, do not do so; modules only work after the kernel is booted, and these things are needed during boot.

For dynamically assigned IP numbers, you'll also need to select one or more of these kernel options:

If the workstation will be booted without kernel parameters, you need also to set the root device to 0:255. Do this by creating a dummy device file with mknod /dev/nfsroot b 0 255. After having created such a device file, you can set root device of the kernel image with rdev <kernel-image> /dev/nfsroot. [NOTE: Modern kernels recognize root=/dev/nfs as a command-line argument; for consistency and/or compatibility, it may be better to use /dev/nfs as the device name instead of /dev/nfsroot.]

3.2 Creation of the root filesystem

Copying the filesystem

Warning: while these instruction might work for you, they are by no means sensefull in a production environment. For a better way to set up a root filesystem for the clients, see the NFS-Root-Client mini-HOWTO by Ofer Maor <ofer@hadar.co.il>.

After having decided where to place the root tree, create it with (e.g.) mkdir -p <directory> and tar cClf / - | tar xpCf <directory> -.

If you boot your kernel without LILO, then the rootdir has to be /tftpboot/<IP-address>. If you don't like it, you can change it in the top Makefile in the kernel sources, look for a line like: NFS_ROOT = -DNFS_ROOT="\"/tftpboot/%s\"" If you change this, you have to recompile the kernel.

Changes to the root filesystem

Now trim the unneeded files, and check the /etc/rc.d scripts. Some important points:

Exporting the filesystem

Export the root dir to the workstation. The basic idea is to edit /etc/exports to include a line similar to one of the following:

For example, a DHCP client receiving an IP address on a class C subnet would need an exports entry similar to this:

The no_root_squash parameter allows the superuser (root) to be treated as such by the NFS server; otherwise root will be remapped to nobody and will generally be unable to do anything useful with the filesystem. The no_all_squash parameter is similar but applies to non-root users. See the exports(5) man page for details.

You will have to notify the NFS server after making any changes to the exports file. Under Red Hat this can easily be done by typing /etc/rc.d/init.d/nfs stop; /etc/rc.d/init.d/nfs start. On other systems, a simple /etc/rc.d/init.d/nfs restart or even exportfs -a may suffice, while on older machines running the user-mode NFS daemon you may actually need to killall -HUP rpc.mountd; killall -HUP rpc.nfsd. (Do not killall -HUP rpc.portmap, however!)

You may also need to edit /etc/hosts.allow and/or /etc/hosts.deny if tcp_wrappers are installed. In particular, if the remote system (client) gets RPC: connection refused errors, /etc/hosts.deny probably contains portmap: ALL or ALL: ALL. To enable the client to use the server's portmapper, add a corresponding line to /etc/hosts.allow:

portmap: <client-IP-number>
portmap: <2nd-client-IP-number>
portmap: <client-IP-network>/<client-IP-netmask>

There is no need to restart anything in this case. You can check by running rpcinfo -p on the NFS server and rpcinfo -p NFS-server on a Linux client within the allowed range; the RPC services listed by both should match.

In case of problems, check /var/log/messages and /var/log/syslog for errors (for example, run tail -f /var/log/messages /var/log/syslog and then try booting the client), and check your man pages (exports, exportfs, portmap, etc.). As a last resort, a reboot of the NFS server may help, but that's a borderline Microsoftism...

RARP setup

Set up the RARP somewhere on the net. If you boot without a nfsroot parameter, the RARP server has to be the NFS server. Usually this will be the NFS server. To do this, you will need to run a kernel with RARP support.

To do this, execute (and install it somewhere in /etc/rc.d of the server!):

/sbin/rarp -s <ip-addr> <hardware-addr>

where

ip-addr

is the IP address of the workstation, and

hardware-addr

is the ethernet address of the network card of the workstation.

example: /sbin/rarp -s 131.131.90.200 00:00:c0:47:10:12

You can also use a symbolic name instead of the IP address, as long the server is able to find out the IP address. (/etc/hosts or DNS lookups)

BOOTP setup

For BOOTP setup you need to edit /etc/bootptab. Please consult the bootpd(8) and bootptab(5) man pages.

DHCP setup

There is no need for the DHCP server to be the same as the NFS server, and in most cases, a DHCP server will already be set up. If one is not, however, consult the DHCP mini-HOWTO for further help.

Finding out hardware addresses

I don't know the hardware address! How can I find it out?

4. Booting the workstation

4.1 Using a boot ROM

As I have not used such a beast myself yet, I can give you only the following tips (courtesy of Christian Leutloff <leutloff@sundancer.tng.oche.de>):

4.2 Using a raw kernel disk

If you have exported the root filesystem with the correct name for the default naming and your NFS server is also the RARP server (which implies that the boxes are on the same subnet.), than you can just boot the kernel by cating it to a disk. (You have to set the root device in the kernel to 0:255.) This assumes, that the root directory on the server is /tftpboot/IP Address (this value can be changed when compiling the kernel.)

4.3 Using a bootloader & RARP

Give the kernel all needed parameters when booting, and add nfsroot=<server-ip-addr>:</path/to/mount> where server-ip-addr is the IP address of your NFS-server, and /path/to/mount is the path to the root directory.

Tips:

4.4 Using a bootloader without RARP

The ip and nfsroot kernel parameters (which can be hardcoded into the kernel, interactively entered at some bootloader prompts, or included in lilo.conf via the append= parameter; see the next subsection) provide all of the information necessary for the client to set up its ethernet interface and to contact the NFS server, respectively. The parameters are fully documented in Documentation/nfsroot.txt, which is included in the kernel sources (usually found under /usr/src/linux). Here's the format for a machine with a static (pre-assigned) IP address:

DHCP is much simpler:

4.5 Sample kernel command lines

Here's an example of a complete kernel command line such as you might include in lilo.conf or equivalent; only the IP numbers and NFS path are bogus:

That uses DHCP to assign an IP address to the machine and puts its boot messages (console) on the second serial port. The following is the corresponding example using a static IP address; it also explicitly specifies Busybox's (non-standard) location for init:

5. Known problems

5.1 /sbin/init doesn't start.

A common problem with /sbin/init is that some distributions (e.g., Red Hat Linux) come with /sbin/init dynamically linked. So you have to provide a correct /lib setup to the client. An easy thing one could try is replacing /sbin/init (for the client) with a statically linked ``Hello World'' program. This way you know if it is something more basic, or ``just'' a problem with dynamic linking.

Also note that Busybox by default installs its init symlink in /bin rather than /sbin. You may need to move it or pass an explicit init= parameter on the kernel command line, as shown in the final example of the previous section.

5.2 /dev troubles.

If you get some garbled messages about ttys when booting, then you should run a MAKEDEV from the client in the /dev directory. There are rumors that this doesn't work with certain server OSes that use 64-bit device numbers; should you run into this, please consider updating this section! A potential solution would be to create a small /dev ram disk early in the boot process and reinstall the device nodes each time, or simply embed directly into the kernel a suitably initialized ramdisk.

6. Other resources