Network Working Group                   Bill Croft (Stanford University)
Request for Comments: 951                John Gilmore (Sun Microsystems)
                                                         September 1985

                      BOOTSTRAP PROTOCOL (BOOTP)


1. Status of this Memo

  This RFC suggests a proposed protocol for the ARPA-Internet
  community, and requests discussion and suggestions for improvements.
  Distribution of this memo is unlimited.

2. Overview

  This RFC describes an IP/UDP bootstrap protocol (BOOTP) which allows
  a diskless client machine to discover its own IP address, the address
  of a server host, and the name of a file to be loaded into memory and
  executed.  The bootstrap operation can be thought of as consisting of
  TWO PHASES.  This RFC describes the first phase, which could be
  labeled 'address determination and bootfile selection'.  After this
  address and filename information is obtained, control passes to the
  second phase of the bootstrap where a file transfer occurs.  The file
  transfer will typically use the TFTP protocol [9], since it is
  intended that both phases reside in PROM on the client.  However
  BOOTP could also work with other protocols such as SFTP [3] or
  FTP [6].

  We suggest that the client's PROM software provide a way to do a
  complete bootstrap without 'user' interaction.  This is the type of
  boot that would occur during an unattended power-up.  A mechanism
  should be provided for the user to manually supply the necessary
  address and filename information to bypass the BOOTP protocol and
  enter the file transfer phase directly.  If non-volatile storage is
  available, we suggest keeping default settings there and bypassing
  the BOOTP protocol unless these settings cause the file transfer
  phase to fail.  If the cached information fails, the bootstrap should
  fall back to phase 1 and use BOOTP.

  Here is a brief outline of the protocol:

     1. A single packet exchange is performed.  Timeouts are used to
     retransmit until a reply is received.  The same packet field
     layout is used in both directions.  Fixed length fields of maximum
     reasonable length are used to simplify structure definition and
     parsing.

     2. An 'opcode' field exists with two values.  The client
     broadcasts a 'bootrequest' packet.  The server then answers with a
     'bootreply' packet.  The bootrequest contains the client's
     hardware address and its IP address, if known.


Croft & Gilmore                                                 [Page 1]



RFC 951                                                   September 1985
Bootstrap Protocol


     3. The request can optionally contain the name of the server the
     client wishes to respond.  This is so the client can force the
     boot to occur from a specific host (e.g. if multiple versions of
     the same bootfile exist or if the server is in a far distant
     net/domain).  The client does not have to deal with name / domain
     services; instead this function is pushed off to the BOOTP server.

     4. The request can optionally contain the 'generic' filename to be
     booted.  For example 'unix' or 'ethertip'.  When the server sends
     the bootreply, it replaces this field with the fully qualified
     path name of the appropriate boot file.  In determining this name,
     the server may consult his own database correlating the client's
     address and filename request, with a particular boot file
     customized for that client.  If the bootrequest filename is a null
     string, then the server returns a filename field indicating the
     'default' file to be loaded for that client.

     5. In the case of clients who do not know their IP addresses, the
     server must also have a database relating hardware address to IP
     address.  This client IP address is then placed into a field in
     the bootreply.

     6. Certain network topologies (such as Stanford's) may be such
     that a given physical cable does not have a TFTP server directly
     attached to it (e.g. all the gateways and hosts on a certain cable
     may be diskless).  With the cooperation of neighboring gateways,
     BOOTP can allow clients to boot off of servers several hops away,
     through these gateways.  See the section 'Booting Through
     Gateways' below.  This part of the protocol requires no special
     action on the part of the client.  Implementation is optional and
     requires a small amount of additional code in gateways and
     servers.

3. Packet Format

  All numbers shown are decimal, unless indicated otherwise.  The BOOTP
  packet is enclosed in a standard IP [8] UDP [7] datagram.  For
  simplicity it is assumed that the BOOTP packet is never fragmented.
  Any numeric fields shown are packed in 'standard network byte order',
  i.e. high order bits are sent first.

  In the IP header of a bootrequest, the client fills in its own IP
  source address if known, otherwise zero.  When the server address is
  unknown, the IP destination address will be the 'broadcast address'
  255.255.255.255.  This address means 'broadcast on the local cable,
  (I don't know my net number)' [4].



Croft & Gilmore                                                 [Page 2]



RFC 951                                                   September 1985
Bootstrap Protocol


  The UDP header contains source and destination port numbers.  The
  BOOTP protocol uses two reserved port numbers, 'BOOTP client' (68)
  and 'BOOTP server' (67).  The client sends requests using 'BOOTP
  server' as the destination port; this is usually a broadcast.  The
  server sends replies using 'BOOTP client' as the destination port;
  depending on the kernel or driver facilities in the server, this may
  or may not be a broadcast (this is explained further in the section
  titled 'Chicken/Egg issues' below).  The reason TWO reserved ports
  are used, is to avoid 'waking up' and scheduling the BOOTP server
  daemons, when a bootreply must be broadcast to a client.  Since the
  server and other hosts won't be listening on the 'BOOTP client' port,
  any such incoming broadcasts will be filtered out at the kernel
  level.  We could not simply allow the client to pick a 'random' port
  number for the UDP source port field; since the server reply may be
  broadcast, a randomly chosen port number could confuse other hosts
  that happened to be listening on that port.

  The UDP length field is set to the length of the UDP plus BOOTP
  portions of the packet.  The UDP checksum field can be set to zero by
  the client (or server) if desired, to avoid this extra overhead in a
  PROM implementation.  In the 'Packet Processing' section below the
  phrase '[UDP checksum.]' is used whenever the checksum might be
  verified/computed.

     FIELD   BYTES   DESCRIPTION
     -----   -----   -----------

        op      1       packet op code / message type.
                        1 = BOOTREQUEST, 2 = BOOTREPLY

        htype   1       hardware address type,
                        see ARP section in "Assigned Numbers" RFC.
                        '1' = 10mb ethernet

        hlen    1       hardware address length
                        (eg '6' for 10mb ethernet).

        hops    1       client sets to zero,
                        optionally used by gateways
                        in cross-gateway booting.

        xid     4       transaction ID, a random number,
                        used to match this boot request with the
                        responses it generates.

        secs    2       filled in by client, seconds elapsed since
                        client started trying to boot.


Croft & Gilmore                                                 [Page 3]



RFC 951                                                   September 1985
Bootstrap Protocol


        --      2       unused

        ciaddr  4       client IP address;
                        filled in by client in bootrequest if known.

        yiaddr  4       'your' (client) IP address;
                        filled by server if client doesn't
                        know its own address (ciaddr was 0).

        siaddr  4       server IP address;
                        returned in bootreply by server.

        giaddr  4       gateway IP address,
                        used in optional cross-gateway booting.

        chaddr  16      client hardware address,
                        filled in by client.

        sname   64      optional server host name,
                        null terminated string.

        file    128     boot file name, null terminated string;
                        'generic' name or null in bootrequest,
                        fully qualified directory-path
                        name in bootreply.

        vend    64      optional vendor-specific area,
                        e.g. could be hardware type/serial on request,
                        or 'capability' / remote file system handle
                        on reply.  This info may be set aside for use
                        by a third phase bootstrap or kernel.

4. Chicken / Egg Issues

  How can the server send an IP datagram to the client, if the client
  doesnt know its own IP address (yet)?  Whenever a bootreply is being
  sent, the transmitting machine performs the following operations:

     1. If the client knows its own IP address ('ciaddr' field is
     nonzero), then the IP can be sent 'as normal', since the client
     will respond to ARPs [5].

     2. If the client does not yet know its IP address (ciaddr zero),
     then the client cannot respond to ARPs sent by the transmitter of
     the bootreply.  There are two options:

        a. If the transmitter has the necessary kernel or driver hooks


Croft & Gilmore                                                 [Page 4]



RFC 951                                                   September 1985
Bootstrap Protocol


        to 'manually' construct an ARP address cache entry, then it can
        fill in an entry using the 'chaddr' and 'yiaddr' fields.  Of
        course, this entry should have a timeout on it, just like any
        other entry made by the normal ARP code itself.  The
        transmitter of the bootreply can then simply send the bootreply
        to the client's IP address.  UNIX (4.2 BSD) has this
        capability.

        b. If the transmitter lacks these kernel hooks, it can simply
        send the bootreply to the IP broadcast address on the
        appropriate interface.  This is only one additional broadcast
        over the previous case.

5. Client Use of ARP

  The client PROM must contain a simple implementation of ARP, e.g. the
  address cache could be just one entry in size.  This will allow a
  second-phase-only boot (TFTP) to be performed when the client knows
  the IP addresses and bootfile name.

  Any time the client is expecting to receive a TFTP or BOOTP reply, it
  should be prepared to answer an ARP request for its own IP to
  hardware address mapping (if known).

  Since the bootreply will contain (in the hardware encapsulation) the
  hardware source address of the server/gateway, the client MAY be able
  to avoid sending an ARP request for the server/gateway IP address to
  be used in the following TFTP phase.  However this should be treated
  only as a special case, since it is desirable to still allow a
  second-phase-only boot as described above.

6. Comparison to RARP

  An earlier protocol, Reverse Address Resolution Protocol (RARP) [1]
  was proposed to allow a client to determine its IP address, given
  that it knew its hardware address.  However RARP had the disadvantage
  that it was a hardware link level protocol (not IP/UDP based).  This
  means that RARP could only be implemented on hosts containing special
  kernel or driver modifications to access these 'raw' packets.  Since
  there are many network kernels existent now, with each source
  maintained by different organizations, a boot protocol that does not
  require kernel modifications is a decided advantage.

  BOOTP provides this hardware to IP address lookup function, in
  addition to the other useful features described in the sections
  above.



Croft & Gilmore                                                 [Page 5]



RFC 951                                                   September 1985
Bootstrap Protocol


7. Packet Processing

  7.1. Client Transmission

     Before setting up the packet for the first time, it is a good idea
     to clear the entire packet buffer to all zeros; this will place
     all fields in their default state.  The client then creates a
     packet with the following fields.

     The IP destination address is set to 255.255.255.255.  (the
     broadcast address) or to the server's IP address (if known).  The
     IP source address and 'ciaddr' are set to the client's IP address
     if known, else 0.  The UDP header is set with the proper length;
     source port = 'BOOTP client' port destination port = 'BOOTP
     server' port.

     'op' is set to '1', BOOTREQUEST.  'htype' is set to the hardware
     address type as assigned in the ARP section of the "Assigned
     Numbers" RFC. 'hlen' is set to the length of the hardware address,
     e.g. '6' for 10mb ethernet.

     'xid' is set to a 'random' transaction id.  'secs' is set to the
     number of seconds that have elapsed since the client has started
     booting.  This will let the servers know how long a client has
     been trying.  As the number gets larger, certain servers may feel
     more 'sympathetic' towards a client they don't normally service.
     If a client lacks a suitable clock, it could construct a rough
     estimate using a loop timer.  Or it could choose to simply send
     this field as always a fixed value, say 100 seconds.

     If the client knows its IP address, 'ciaddr' (and the IP source
     address) are set to this value.  'chaddr' is filled in with the
     client's hardware address.

     If the client wishes to restrict booting to a particular server
     name, it may place a null-terminated string in 'sname'.  The name
     used should be any of the allowable names or nicknames of the
     desired host.

     The client has several options for filling the 'file' name field.
     If left null, the meaning is 'I want to boot the default file for
     my machine'.  A null file name can also mean 'I am only interested
     in finding out client/server/gateway IP addresses, I dont care
     about file names'.

     The field can also be a 'generic' name such as 'unix' or



Croft & Gilmore                                                 [Page 6]



RFC 951                                                   September 1985
Bootstrap Protocol


     'gateway'; this means 'boot the named program configured for my
     machine'.  Finally the field can be a fully directory qualified
     path name.

     The 'vend' field can be filled in by the client with
     vendor-specific strings or structures.  For example the machine
     hardware type or serial number may be placed here.  However the
     operation of the BOOTP server should not DEPEND on this
     information existing.

     If the 'vend' field is used, it is recommended that a 4 byte
     'magic number' be the first item within 'vend'.  This lets a
     server determine what kind of information it is seeing in this
     field.  Numbers can be assigned by the usual 'magic number'
     process --you pick one and it's magic.  A different magic number
     could be used for bootreply's than bootrequest's to allow the
     client to take special action with the reply information.

     [UDP checksum.]

  7.2. Client Retransmission Strategy

     If no reply is received for a certain length of time, the client
     should retransmit the request.  The time interval must be chosen
     carefully so as not to flood the network.  Consider the case of a
     cable containing 100 machines that are just coming up after a
     power failure.  Simply retransmitting the request every four
     seconds will inundate the net.

     As a possible strategy, you might consider backing off
     exponentially, similar to the way ethernet backs off on a
     collision.  So for example if the first packet is at time 0:00,
     the second would be at :04, then :08, then :16, then :32, then
     :64.  You should also randomize each time; this would be done
     similar to the ethernet specification by starting with a mask and
     'and'ing that with with a random number to get the first backoff.
     On each succeeding backoff, the mask is increased in length by one
     bit.  This doubles the average delay on each backoff.

     After the 'average' backoff reaches about 60 seconds, it should be
     increased no further, but still randomized.

     Before each retransmission, the client should update the 'secs'
     field. [UDP checksum.]





Croft & Gilmore                                                 [Page 7]



RFC 951                                                   September 1985
Bootstrap Protocol


  7.3. Server Receives BOOTREQUEST

     [UDP checksum.]  If the UDP destination port does not match the
     'BOOTP server' port, discard the packet.

     If the server name field (sname) is null (no particular server
     specified), or sname is specified and matches our name or
     nickname, then continue with packet processing.

     If the sname field is specified, but does not match 'us', then
     there are several options:

        1. You may choose to simply discard this packet.

        2. If a name lookup on sname shows it to be on this same cable,
        discard the packet.

        3. If sname is on a different net, you may choose to forward
        the packet to that address.  If so, check the 'giaddr' (gateway
        address) field.  If 'giaddr' is zero, fill it in with my
        address or the address of a gateway that can be used to get to
        that net.  Then forward the packet.

     If the client IP address (ciaddr) is zero, then the client does
     not know its own IP address.  Attempt to lookup the client
     hardware address (chaddr, hlen, htype) in our database.  If no
     match is found, discard the packet.  Otherwise we now have an IP
     address for this client; fill it into the 'yiaddr' (your IP
     address) field.

     We now check the boot file name field (file).  The field will be
     null if the client is not interested in filenames, or wants the
     default bootfile.  If the field is non-null, it is used as a
     lookup key in a database, along with the client's IP address.  If
     there is a default file or generic file (possibly indexed by the
     client address) or a fully-specified path name that matches, then
     replace the 'file' field with the fully-specified path name of the
     selected boot file.  If the field is non-null and no match was
     found, then the client is asking for a file we dont have; discard
     the packet, perhaps some other BOOTP server will have it.

     The 'vend' vendor-specific data field should now be checked and if
     a recognized type of data is provided, client-specific actions
     should be taken, and a response placed in the 'vend' data field of
     the reply packet.  For example, a workstation client could provide




Croft & Gilmore                                                 [Page 8]



RFC 951                                                   September 1985
Bootstrap Protocol


     an authentication key and receive from the server a capability for
     remote file access, or a set of configuration options, which can
     be passed to the operating system that will shortly be booted in.

     Place my (server) IP address in the 'siaddr' field.  Set the 'op'
     field to BOOTREPLY.  The UDP destination port is set to 'BOOTP
     client'.  If the client address 'ciaddr' is nonzero, send the
     packet there; else if the gateway address 'giaddr' is nonzero, set
     the UDP destination port to 'BOOTP server' and send the packet to
     'giaddr'; else the client is on one of our cables but it doesnt
     know its own IP address yet --use a method described in the 'Egg'
     section above to send it to the client. If 'Egg' is used and we
     have multiple interfaces on this host, use the 'yiaddr' (your IP
     address) field to figure out which net (cable/interface) to send
     the packet to.  [UDP checksum.]

  7.4. Server/Gateway Receives BOOTREPLY

     [UDP checksum.]  If 'yiaddr' (your [the client's] IP address)
     refers to one of our cables, use one of the 'Egg' methods above to
     forward it to the client.  Be sure to send it to the 'BOOTP
     client' UDP destination port.

  7.5. Client Reception

     Don't forget to process ARP requests for my own IP address (if I
     know it).  [UDP checksum.]  The client should discard incoming
     packets that: are not IP/UDPs addressed to the boot port; are not
     BOOTREPLYs; do not match my IP address (if I know it) or my
     hardware address; do not match my transaction id.  Otherwise we
     have received a successful reply. 'yiaddr' will contain my IP
     address, if I didnt know it before.  'file' is the name of the
     file name to TFTP 'read request'.  The server address is in
     'siaddr'.  If 'giaddr' (gateway address) is nonzero, then the
     packets should be forwarded there first, in order to get to the
     server.

8. Booting Through Gateways

  This part of the protocol is optional and requires some additional
  code in cooperating gateways and servers, but it allows cross-gateway
  booting.  This is mainly useful when gateways are diskless machines.
  Gateways containing disks (e.g. a UNIX machine acting as a gateway),
  might as well run their own BOOTP/TFTP servers.

  Gateways listening to broadcast BOOTREQUESTs may decide to forward or
  rebroadcast these requests 'when appropriate'.  For example, the


Croft & Gilmore                                                 [Page 9]



RFC 951                                                   September 1985
Bootstrap Protocol


  gateway could have, as part of his configuration tables, a list of
  other networks or hosts to receive a copy of any broadcast
  BOOTREQUESTs.  Even though a 'hops' field exists, it is a poor idea
  to simply globally rebroadcast the requests, since broadcast loops
  will almost certainly occur.

  The forwarding could begin immediately, or wait until the 'secs'
  (seconds client has been trying) field passes a certain threshold.

  If a gateway does decide to forward the request, it should look at
  the 'giaddr' (gateway IP address) field.  If zero, it should plug its
  own IP address (on the receiving cable) into this field.  It may also
  use the 'hops' field to optionally control how far the packet is
  reforwarded. Hops should be incremented on each forwarding.  For
  example, if hops passes '3', the packet should probably be discarded.
  [UDP checksum.]

  Here we have recommended placing this special forwarding function in
  the gateways.  But that does not have to be the case.  As long as
  some 'BOOTP forwarding agent' exists on the net with the booting
  client, the agent can do the forwarding when appropriate.  Thus this
  service may or may not be co-located with the gateway.

  In the case of a forwarding agent not located in the gateway, the
  agent could save himself some work by plugging the broadcast address
  of the interface receiving the bootrequest into the 'giaddr' field.
  Thus the reply would get forwarded using normal gateways, not
  involving the forwarding agent.  Of course the disadvantage here is
  that you lose the ability to use the 'Egg' non-broadcast method of
  sending the reply, causing extra overhead for every host on the
  client cable.

9. Sample BOOTP Server Database

  As a suggestion, we show a sample text file database that the BOOTP
  server program might use.  The database has two sections, delimited
  by a line containing an percent in column 1.  The first section
  contains a 'default directory' and mappings from generic names to
  directory/pathnames.  The first generic name in this section is the
  'default file' you get when the bootrequest contains a null 'file'
  string.

  The second section maps hardware addresstype/address into an
  ipaddress. Optionally you can also overide the default generic name
  by supplying a ipaddress specific genericname.  A 'suffix' item is
  also an option; if supplied, any generic names specified by the
  client will be accessed by first appending 'suffix' to the 'pathname'


Croft & Gilmore                                                [Page 10]



RFC 951                                                   September 1985
Bootstrap Protocol


  appropriate to that generic name.  If that file is not found, then
  the plain 'pathname' will be tried.  This 'suffix' option allows a
  whole set of custom generics to be setup without a lot of effort.
  Below is shown the general format; fields are delimited by one or
  more spaces or tabs; trailing empty fields may be omitted; blank
  lines and lines beginning with '#' are ignored.

     # comment line

     homedirectory
     genericname1    pathname1
     genericname2    pathname2
     ...

     % end of generic names, start of address mappings

     hostname1 hardwaretype hardwareaddr1 ipaddr1 genericname suffix
     hostname2 hardwaretype hardwareaddr2 ipaddr2 genericname suffix
     ...

  Here is a specific example.  Note the 'hardwaretype' number is the
  same as that shown in the ARP section of the 'Assigned Numbers' RFC.
  The 'hardwaretype' and 'ipaddr' numbers are in decimal;
  'hardwareaddr' is in hex.

     # last updated by smith

     /usr/boot
     vmunix          vmunix
     tip             ethertip
     watch           /usr/diag/etherwatch
     gate            gate.

     % end of generic names, start of address mappings

     hamilton        1 02.60.8c.06.34.98     36.19.0.5
     burr            1 02.60.8c.34.11.78     36.44.0.12
     101-gateway     1 02.60.8c.23.ab.35     36.44.0.32      gate 101
     mjh-gateway     1 02.60.8c.12.32.bc     36.42.0.64      gate mjh
     welch-tipa      1 02.60.8c.22.65.32     36.47.0.14      tip
     welch-tipb      1 02.60.8c.12.15.c8     36.46.0.12      tip

  In the example above, if 'mjh-gateway' does a default boot, it will
  get the file '/usr/boot/gate.mjh'.





Croft & Gilmore                                                [Page 11]



RFC 951                                                   September 1985
Bootstrap Protocol


10. Acknowledgements

  Ross Finlayson (et. al.) produced two earlier RFC's discussing TFTP
  bootstraping [2] using RARP [1].

  We would also like to acknowledge the previous work and comments of
  Noel Chiappa, Bob Lyon, Jeff Mogul, Mark Lewis, and David Plummer.

REFERENCES

  1.  Ross Finlayson, Timothy Mann, Jeffrey Mogul, Marvin Theimer.  A
      Reverse Address Resolution Protocol.  RFC 903, NIC, June, 1984.

  2.  Ross Finlayson.  Bootstrap Loading using TFTP.  RFC 906, NIC,
      June, 1984.

  3.  Mark Lottor.  Simple File Transfer Protocol.  RFC 913, NIC,
      September, 1984.

  4.  Jeffrey Mogul.  Broadcasting Internet Packets.  RFC 919, NIC,
      October, 1984.

  5.  David Plummer.  An Ethernet Address Resolution Protocol.  RFC
      826, NIC, September, 1982.

  6.  Jon Postel.  File Transfer Protocol.  RFC 765, NIC, June, 1980.

  7.  Jon Postel.  User Datagram Protocol.  RFC 768, NIC, August, 1980.

  8.  Jon Postel.  Internet Protocol.  RFC 791, NIC, September, 1981.

  9.  K. R. Sollins, Noel Chiappa.  The TFTP Protocol.  RFC 783, NIC,
      June, 1981.
















Croft & Gilmore                                                [Page 12]