Documentation/filesystems/nfs/nfsroot.txt

   1 Mounting the root filesystem via NFS (nfsroot)
   2 ===============================================
   3
   4 Written 1996 by Gero Kuhlmann <gero@gkminix.han.de>
   5 Updated 1997 by Martin Mares <mj@atrey.karlin.mff.cuni.cz>
   6 Updated 2006 by Nico Schottelius <nico-kernel-nfsroot@schottelius.org>
   7 Updated 2006 by Horms <horms@verge.net.au>
   8 Updated 2018 by Chris Novakovic <chris@chrisn.me.uk>
   9
  10
  11
  12 In order to use a diskless system, such as an X-terminal or printer server
  13 for example, it is necessary for the root filesystem to be present on a
  14 non-disk device. This may be an initramfs (see Documentation/filesystems/
  15 ramfs-rootfs-initramfs.txt), a ramdisk (see Documentation/admin-guide/initrd.rst) or a
  16 filesystem mounted via NFS. The following text describes on how to use NFS
  17 for the root filesystem. For the rest of this text 'client' means the
  18 diskless system, and 'server' means the NFS server.
  19
  20
  21
  22
  23 1.) Enabling nfsroot capabilities
  24     -----------------------------
  25
  26 In order to use nfsroot, NFS client support needs to be selected as
  27 built-in during configuration. Once this has been selected, the nfsroot
  28 option will become available, which should also be selected.
  29
  30 In the networking options, kernel level autoconfiguration can be selected,
  31 along with the types of autoconfiguration to support. Selecting all of
  32 DHCP, BOOTP and RARP is safe.
  33
  34
  35
  36
  37 2.) Kernel command line
  38     -------------------
  39
  40 When the kernel has been loaded by a boot loader (see below) it needs to be
  41 told what root fs device to use. And in the case of nfsroot, where to find
  42 both the server and the name of the directory on the server to mount as root.
  43 This can be established using the following kernel command line parameters:
  44
  45
  46 root=/dev/nfs
  47
  48   This is necessary to enable the pseudo-NFS-device. Note that it's not a
  49   real device but just a synonym to tell the kernel to use NFS instead of
  50   a real device.
  51
  52
  53 nfsroot=[<server-ip>:]<root-dir>[,<nfs-options>]
  54
  55   If the `nfsroot' parameter is NOT given on the command line,
  56   the default "/tftpboot/%s" will be used.
  57
  58   <server-ip>   Specifies the IP address of the NFS server.
  59                 The default address is determined by the `ip' parameter
  60                 (see below). This parameter allows the use of different
  61                 servers for IP autoconfiguration and NFS.
  62
  63   <root-dir>    Name of the directory on the server to mount as root.
  64                 If there is a "%s" token in the string, it will be
  65                 replaced by the ASCII-representation of the client's
  66                 IP address.
  67
  68   <nfs-options> Standard NFS options. All options are separated by commas.
  69                 The following defaults are used:
  70                         port            = as given by server portmap daemon
  71                         rsize           = 4096
  72                         wsize           = 4096
  73                         timeo           = 7
  74                         retrans         = 3
  75                         acregmin        = 3
  76                         acregmax        = 60
  77                         acdirmin        = 30
  78                         acdirmax        = 60
  79                         flags           = hard, nointr, noposix, cto, ac
  80
  81
  82 ip=<client-ip>:<server-ip>:<gw-ip>:<netmask>:<hostname>:<device>:<autoconf>:
  83    <dns0-ip>:<dns1-ip>:<ntp0-ip>
  84
  85   This parameter tells the kernel how to configure IP addresses of devices
  86   and also how to set up the IP routing table. It was originally called
  87   `nfsaddrs', but now the boot-time IP configuration works independently of
  88   NFS, so it was renamed to `ip' and the old name remained as an alias for
  89   compatibility reasons.
  90
  91   If this parameter is missing from the kernel command line, all fields are
  92   assumed to be empty, and the defaults mentioned below apply. In general
  93   this means that the kernel tries to configure everything using
  94   autoconfiguration.
  95
  96   The <autoconf> parameter can appear alone as the value to the `ip'
  97   parameter (without all the ':' characters before).  If the value is
  98   "ip=off" or "ip=none", no autoconfiguration will take place, otherwise
  99   autoconfiguration will take place.  The most common way to use this
 100   is "ip=dhcp".
 101
 102   <client-ip>   IP address of the client.
 103
 104                 Default:  Determined using autoconfiguration.
 105
 106   <server-ip>   IP address of the NFS server. If RARP is used to determine
 107                 the client address and this parameter is NOT empty only
 108                 replies from the specified server are accepted.
 109
 110                 Only required for NFS root. That is autoconfiguration
 111                 will not be triggered if it is missing and NFS root is not
 112                 in operation.
 113
 114                 Value is exported to /proc/net/pnp with the prefix "bootserver "
 115                 (see below).
 116
 117                 Default: Determined using autoconfiguration.
 118                          The address of the autoconfiguration server is used.
 119
 120   <gw-ip>       IP address of a gateway if the server is on a different subnet.
 121
 122                 Default: Determined using autoconfiguration.
 123
 124   <netmask>     Netmask for local network interface. If unspecified
 125                 the netmask is derived from the client IP address assuming
 126                 classful addressing.
 127
 128                 Default:  Determined using autoconfiguration.
 129
 130   <hostname>    Name of the client. If a '.' character is present, anything
 131                 before the first '.' is used as the client's hostname, and anything
 132                 after it is used as its NIS domain name. May be supplied by
 133                 autoconfiguration, but its absence will not trigger autoconfiguration.
 134                 If specified and DHCP is used, the user-provided hostname (and NIS
 135                 domain name, if present) will be carried in the DHCP request; this
 136                 may cause a DNS record to be created or updated for the client.
 137
 138                 Default: Client IP address is used in ASCII notation.
 139
 140   <device>      Name of network device to use.
 141
 142                 Default: If the host only has one device, it is used.
 143                          Otherwise the device is determined using
 144                          autoconfiguration. This is done by sending
 145                          autoconfiguration requests out of all devices,
 146                          and using the device that received the first reply.
 147
 148   <autoconf>    Method to use for autoconfiguration. In the case of options
 149                 which specify multiple autoconfiguration protocols,
 150                 requests are sent using all protocols, and the first one
 151                 to reply is used.
 152
 153                 Only autoconfiguration protocols that have been compiled
 154                 into the kernel will be used, regardless of the value of
 155                 this option.
 156
 157                   off or none: don't use autoconfiguration
 158                                 (do static IP assignment instead)
 159                   on or any:   use any protocol available in the kernel
 160                                (default)
 161                   dhcp:        use DHCP
 162                   bootp:       use BOOTP
 163                   rarp:        use RARP
 164                   both:        use both BOOTP and RARP but not DHCP
 165                                (old option kept for backwards compatibility)
 166
 167                 if dhcp is used, the client identifier can be used by following
 168                 format "ip=dhcp,client-id-type,client-id-value"
 169
 170                 Default: any
 171
 172   <dns0-ip>     IP address of primary nameserver.
 173                 Value is exported to /proc/net/pnp with the prefix "nameserver "
 174                 (see below).
 175
 176                 Default: None if not using autoconfiguration; determined
 177                 automatically if using autoconfiguration.
 178
 179   <dns1-ip>     IP address of secondary nameserver.
 180                 See <dns0-ip>.
 181
 182   <ntp0-ip>     IP address of a Network Time Protocol (NTP) server.
 183                 Value is exported to /proc/net/ipconfig/ntp_servers, but is
 184                 otherwise unused (see below).
 185
 186                 Default: None if not using autoconfiguration; determined
 187                 automatically if using autoconfiguration.
 188
 189   After configuration (whether manual or automatic) is complete, two files
 190   are created in the following format; lines are omitted if their respective
 191   value is empty following configuration:
 192
 193   - /proc/net/pnp:
 194
 195         #PROTO: <DHCP|BOOTP|RARP|MANUAL>        (depending on configuration method)
 196         domain <dns-domain>                     (if autoconfigured, the DNS domain)
 197         nameserver <dns0-ip>                    (primary name server IP)
 198         nameserver <dns1-ip>                    (secondary name server IP)
 199         nameserver <dns2-ip>                    (tertiary name server IP)
 200         bootserver <server-ip>                  (NFS server IP)
 201
 202   - /proc/net/ipconfig/ntp_servers:
 203
 204         <ntp0-ip>                               (NTP server IP)
 205         <ntp1-ip>                               (NTP server IP)
 206         <ntp2-ip>                               (NTP server IP)
 207
 208   <dns-domain> and <dns2-ip> (in /proc/net/pnp) and <ntp1-ip> and <ntp2-ip>
 209   (in /proc/net/ipconfig/ntp_servers) are requested during autoconfiguration;
 210   they cannot be specified as part of the "ip=" kernel command line parameter.
 211
 212   Because the "domain" and "nameserver" options are recognised by DNS
 213   resolvers, /etc/resolv.conf is often linked to /proc/net/pnp on systems
 214   that use an NFS root filesystem.
 215
 216   Note that the kernel will not synchronise the system time with any NTP
 217   servers it discovers; this is the responsibility of a user space process
 218   (e.g. an initrd/initramfs script that passes the IP addresses listed in
 219   /proc/net/ipconfig/ntp_servers to an NTP client before mounting the real
 220   root filesystem if it is on NFS).
 221
 222
 223 nfsrootdebug
 224
 225   This parameter enables debugging messages to appear in the kernel
 226   log at boot time so that administrators can verify that the correct
 227   NFS mount options, server address, and root path are passed to the
 228   NFS client.
 229
 230
 231 rdinit=<executable file>
 232
 233   To specify which file contains the program that starts system
 234   initialization, administrators can use this command line parameter.
 235   The default value of this parameter is "/init".  If the specified
 236   file exists and the kernel can execute it, root filesystem related
 237   kernel command line parameters, including `nfsroot=', are ignored.
 238
 239   A description of the process of mounting the root file system can be
 240   found in:
 241
 242     Documentation/early-userspace/early_userspace_support.rst
 243
 244
 245
 246
 247 3.) Boot Loader
 248     ----------
 249
 250 To get the kernel into memory different approaches can be used.
 251 They depend on various facilities being available:
 252
 253
 254 3.1)  Booting from a floppy using syslinux
 255
 256         When building kernels, an easy way to create a boot floppy that uses
 257         syslinux is to use the zdisk or bzdisk make targets which use zimage
 258         and bzimage images respectively. Both targets accept the
 259         FDARGS parameter which can be used to set the kernel command line.
 260
 261         e.g.
 262            make bzdisk FDARGS="root=/dev/nfs"
 263
 264         Note that the user running this command will need to have
 265         access to the floppy drive device, /dev/fd0
 266
 267         For more information on syslinux, including how to create bootdisks
 268         for prebuilt kernels, see http://syslinux.zytor.com/
 269
 270         N.B: Previously it was possible to write a kernel directly to
 271              a floppy using dd, configure the boot device using rdev, and
 272              boot using the resulting floppy. Linux no longer supports this
 273              method of booting.
 274
 275 3.2) Booting from a cdrom using isolinux
 276
 277         When building kernels, an easy way to create a bootable cdrom that
 278         uses isolinux is to use the isoimage target which uses a bzimage
 279         image. Like zdisk and bzdisk, this target accepts the FDARGS
 280         parameter which can be used to set the kernel command line.
 281
 282         e.g.
 283           make isoimage FDARGS="root=/dev/nfs"
 284
 285         The resulting iso image will be arch/<ARCH>/boot/image.iso
 286         This can be written to a cdrom using a variety of tools including
 287         cdrecord.
 288
 289         e.g.
 290           cdrecord dev=ATAPI:1,0,0 arch/x86/boot/image.iso
 291
 292         For more information on isolinux, including how to create bootdisks
 293         for prebuilt kernels, see http://syslinux.zytor.com/
 294
 295 3.2) Using LILO
 296         When using LILO all the necessary command line parameters may be
 297         specified using the 'append=' directive in the LILO configuration
 298         file.
 299
 300         However, to use the 'root=' directive you also need to create
 301         a dummy root device, which may be removed after LILO is run.
 302
 303         mknod /dev/boot255 c 0 255
 304
 305         For information on configuring LILO, please refer to its documentation.
 306
 307 3.3) Using GRUB
 308         When using GRUB, kernel parameter are simply appended after the kernel
 309         specification: kernel <kernel> <parameters>
 310
 311 3.4) Using loadlin
 312         loadlin may be used to boot Linux from a DOS command prompt without
 313         requiring a local hard disk to mount as root. This has not been
 314         thoroughly tested by the authors of this document, but in general
 315         it should be possible configure the kernel command line similarly
 316         to the configuration of LILO.
 317
 318         Please refer to the loadlin documentation for further information.
 319
 320 3.5) Using a boot ROM
 321         This is probably the most elegant way of booting a diskless client.
 322         With a boot ROM the kernel is loaded using the TFTP protocol. The
 323         authors of this document are not aware of any no commercial boot
 324         ROMs that support booting Linux over the network. However, there
 325         are two free implementations of a boot ROM, netboot-nfs and
 326         etherboot, both of which are available on sunsite.unc.edu, and both
 327         of which contain everything you need to boot a diskless Linux client.
 328
 329 3.6) Using pxelinux
 330         Pxelinux may be used to boot linux using the PXE boot loader
 331         which is present on many modern network cards.
 332
 333         When using pxelinux, the kernel image is specified using
 334         "kernel <relative-path-below /tftpboot>". The nfsroot parameters
 335         are passed to the kernel by adding them to the "append" line.
 336         It is common to use serial console in conjunction with pxeliunx,
 337         see Documentation/admin-guide/serial-console.rst for more information.
 338
 339         For more information on isolinux, including how to create bootdisks
 340         for prebuilt kernels, see http://syslinux.zytor.com/
 341
 342
 343
 344
 345 4.) Credits
 346     -------
 347
 348   The nfsroot code in the kernel and the RARP support have been written
 349   by Gero Kuhlmann <gero@gkminix.han.de>.
 350
 351   The rest of the IP layer autoconfiguration code has been written
 352   by Martin Mares <mj@atrey.karlin.mff.cuni.cz>.
 353
 354   In order to write the initial version of nfsroot I would like to thank
 355   Jens-Uwe Mager <jum@anubis.han.de> for his help.