man7/ip.7

   1 '\" t
   2 .\" This man page is Copyright (C) 1999 Andi Kleen <ak@muc.de>.
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM_ONE_PARA)
   5 .\" Permission is granted to distribute possibly modified copies
   6 .\" of this page provided the header is included verbatim,
   7 .\" and in case of nontrivial modification author and date
   8 .\" of the modification is added to the header.
   9 .\" %%%LICENSE_END
  10 .\"
  11 .\" $Id: ip.7,v 1.19 2000/12/20 18:10:31 ak Exp $
  12 .\"
  13 .\" FIXME The following socket options are yet to be documented
  14 .\"     IP_XFRM_POLICY (2.5.48)
  15 .\"         Needs CAP_NET_ADMIN
  16 .\"     IP_IPSEC_POLICY (2.5.47)
  17 .\"         Needs CAP_NET_ADMIN
  18 .\"     IP_PASSSEC (2.6.17)
  19 .\"         Boolean
  20 .\"         commit 2c7946a7bf45ae86736ab3b43d0085e43947945c
  21 .\"         Author: Catherine Zhang <cxzhang@watson.ibm.com>
  22 .\"     IP_MINTTL (2.6.34)
  23 .\"         commit d218d11133d888f9745802146a50255a4781d37a
  24 .\"         Author: Stephen Hemminger <shemminger@vyatta.com>
  25 .\"     MCAST_JOIN_GROUP (2.4.22 / 2.6)
  26 .\"     MCAST_BLOCK_SOURCE (2.4.22 / 2.6)
  27 .\"     MCAST_UNBLOCK_SOURCE (2.4.22 / 2.6)
  28 .\"     MCAST_LEAVE_GROUP (2.4.22 / 2.6)
  29 .\"     MCAST_JOIN_SOURCE_GROUP (2.4.22 / 2.6)
  30 .\"     MCAST_LEAVE_SOURCE_GROUP (2.4.22 / 2.6)
  31 .\"     MCAST_MSFILTER (2.4.22 / 2.6)
  32 .\"     IP_UNICAST_IF (3.4)
  33 .\"         commit 76e21053b5bf33a07c76f99d27a74238310e3c71
  34 .\"         Author: Erich E. Hoover <ehoover@mines.edu>
  35 .\"
  36 .TH IP  7 2016-10-08 "Linux" "Linux Programmer's Manual"
  37 .SH NAME
  38 ip \- Linux IPv4 protocol implementation
  39 .SH SYNOPSIS
  40 .B #include <sys/socket.h>
  41 .br
  42 .\" .B #include <net/netinet.h> -- does not exist anymore
  43 .\" .B #include <linux/errqueue.h> -- never include <linux/foo.h>
  44 .B #include <netinet/in.h>
  45 .br
  46 .B #include <netinet/ip.h>        \fR/* superset of previous */
  47 .sp
  48 .IB tcp_socket " = socket(AF_INET, SOCK_STREAM, 0);"
  49 .br
  50 .IB udp_socket " = socket(AF_INET, SOCK_DGRAM, 0);"
  51 .br
  52 .IB raw_socket " = socket(AF_INET, SOCK_RAW, " protocol ");"
  53 .SH DESCRIPTION
  54 Linux implements the Internet Protocol, version 4,
  55 described in RFC\ 791 and RFC\ 1122.
  56 .B ip
  57 contains a level 2 multicasting implementation conforming to RFC\ 1112.
  58 It also contains an IP router including a packet filter.
  59 .\" FIXME . has someone verified that 2.1 is really 1812 compliant?
  60 .PP
  61 The programming interface is BSD-sockets compatible.
  62 For more information on sockets, see
  63 .BR socket (7).
  64 .PP
  65 An IP socket is created using
  66 .BR socket (2):
  67
  68     socket(AF_INET, socket_type, protocol);
  69
  70 Valid socket types are
  71 .B SOCK_STREAM
  72 to open a
  73 .BR tcp (7)
  74 socket,
  75 .B SOCK_DGRAM
  76 to open a
  77 .BR udp (7)
  78 socket, or
  79 .B SOCK_RAW
  80 to open a
  81 .BR raw (7)
  82 socket to access the IP protocol directly.
  83 .I protocol
  84 is the IP protocol in the IP header to be received or sent.
  85 The only valid values for
  86 .I protocol
  87 are 0 and
  88 .B IPPROTO_TCP
  89 for TCP sockets, and 0 and
  90 .B IPPROTO_UDP
  91 for UDP sockets.
  92 For
  93 .B SOCK_RAW
  94 you may specify a valid IANA IP protocol defined in
  95 RFC\ 1700 assigned numbers.
  96 .PP
  97 When a process wants to receive new incoming packets or connections, it
  98 should bind a socket to a local interface address using
  99 .BR bind (2).
 100 In this case, only one IP socket may be bound to any given local
 101 (address, port) pair.
 102 When
 103 .B INADDR_ANY
 104 is specified in the bind call, the socket will be bound to
 105 .I all
 106 local interfaces.
 107 When
 108 .BR listen (2)
 109 is called on an unbound socket, the socket is automatically bound
 110 to a random free port with the local address set to
 111 .BR INADDR_ANY .
 112 When
 113 .BR connect (2)
 114 is called on an unbound socket, the socket is automatically bound
 115 to a random free port or to a usable shared port with the local address
 116 set to
 117 .BR INADDR_ANY .
 118
 119 A TCP local socket address that has been bound is unavailable for
 120 some time after closing, unless the
 121 .B SO_REUSEADDR
 122 flag has been set.
 123 Care should be taken when using this flag as it makes TCP less reliable.
 124 .SS Address format
 125 An IP socket address is defined as a combination of an IP interface
 126 address and a 16-bit port number.
 127 The basic IP protocol does not supply port numbers, they
 128 are implemented by higher level protocols like
 129 .BR udp (7)
 130 and
 131 .BR tcp (7).
 132 On raw sockets
 133 .I sin_port
 134 is set to the IP protocol.
 135 .PP
 136 .in +4n
 137 .nf
 138 struct sockaddr_in {
 139     sa_family_t    sin_family; /* address family: AF_INET */
 140     in_port_t      sin_port;   /* port in network byte order */
 141     struct in_addr sin_addr;   /* internet address */
 142 };
 143
 144 /* Internet address. */
 145 struct in_addr {
 146     uint32_t       s_addr;     /* address in network byte order */
 147 };
 148 .fi
 149 .in
 150 .PP
 151 .I sin_family
 152 is always set to
 153 .BR AF_INET .
 154 This is required; in Linux 2.2 most networking functions return
 155 .B EINVAL
 156 when this setting is missing.
 157 .I sin_port
 158 contains the port in network byte order.
 159 The port numbers below 1024 are called
 160 .IR "privileged ports"
 161 (or sometimes:
 162 .IR "reserved ports" ).
 163 Only a privileged process
 164 (on Linux: a process that has the
 165 .B CAP_NET_BIND_SERVICE
 166 capability in the user namespace governing its network namespace) may
 167 .BR bind (2)
 168 to these sockets.
 169 Note that the raw IPv4 protocol as such has no concept of a
 170 port, they are implemented only by higher protocols like
 171 .BR tcp (7)
 172 and
 173 .BR udp (7).
 174 .PP
 175 .I sin_addr
 176 is the IP host address.
 177 The
 178 .I s_addr
 179 member of
 180 .I struct in_addr
 181 contains the host interface address in network byte order.
 182 .I in_addr
 183 should be assigned one of the
 184 .BR INADDR_*
 185 values (e.g.,
 186 .BR INADDR_ANY )
 187 or set using the
 188 .BR inet_aton (3),
 189 .BR inet_addr (3),
 190 .BR inet_makeaddr (3)
 191 library functions or directly with the name resolver (see
 192 .BR gethostbyname (3)).
 193
 194 IPv4 addresses are divided into unicast, broadcast,
 195 and multicast addresses.
 196 Unicast addresses specify a single interface of a host,
 197 broadcast addresses specify all hosts on a network, and multicast
 198 addresses address all hosts in a multicast group.
 199 Datagrams to broadcast addresses can be sent or received only when the
 200 .B SO_BROADCAST
 201 socket flag is set.
 202 In the current implementation, connection-oriented sockets are allowed
 203 to use only unicast addresses.
 204 .\" Leave a loophole for XTP @)
 205
 206 Note that the address and the port are always stored in
 207 network byte order.
 208 In particular, this means that you need to call
 209 .BR htons (3)
 210 on the number that is assigned to a port.
 211 All address/port manipulation
 212 functions in the standard library work in network byte order.
 213
 214 There are several special addresses:
 215 .B INADDR_LOOPBACK
 216 (127.0.0.1)
 217 always refers to the local host via the loopback device;
 218 .B INADDR_ANY
 219 (0.0.0.0)
 220 means any address for binding;
 221 .B INADDR_BROADCAST
 222 (255.255.255.255)
 223 means any host and has the same effect on bind as
 224 .B INADDR_ANY
 225 for historical reasons.
 226 .SS Socket options
 227 IP supports some protocol-specific socket options that can be set with
 228 .BR setsockopt (2)
 229 and read with
 230 .BR getsockopt (2).
 231 The socket option level for IP is
 232 .BR IPPROTO_IP .
 233 .\" or SOL_IP on Linux
 234 A boolean integer flag is zero when it is false, otherwise true.
 235
 236 When an invalid socket option is specified,
 237 .BR getsockopt (2)
 238 and
 239 .BR setsockopt (2)
 240 fail with the error
 241 .BR ENOPROTOOPT .
 242 .TP
 243 .BR IP_ADD_MEMBERSHIP " (since Linux 1.2)"
 244 Join a multicast group.
 245 Argument is an
 246 .I ip_mreqn
 247 structure.
 248 .sp
 249 .in +4n
 250 .nf
 251 struct ip_mreqn {
 252     struct in_addr imr_multiaddr; /* IP multicast group
 253                                      address */
 254     struct in_addr imr_address;   /* IP address of local
 255                                      interface */
 256     int            imr_ifindex;   /* interface index */
 257 };
 258 .fi
 259 .in
 260 .sp
 261 .I imr_multiaddr
 262 contains the address of the multicast group the application
 263 wants to join or leave.
 264 It must be a valid multicast address
 265 .\" (i.e., within the 224.0.0.0-239.255.255.255 range)
 266 (or
 267 .BR setsockopt (2)
 268 fails with the error
 269 .BR EINVAL ).
 270 .I imr_address
 271 is the address of the local interface with which the system
 272 should join the multicast group; if it is equal to
 273 .BR INADDR_ANY ,
 274 an appropriate interface is chosen by the system.
 275 .I imr_ifindex
 276 is the interface index of the interface that should join/leave the
 277 .I imr_multiaddr
 278 group, or 0 to indicate any interface.
 279 .IP
 280 The
 281 .I ip_mreqn
 282 structure is available only since Linux 2.2.
 283 For compatibility, the old
 284 .I ip_mreq
 285 structure (present since Linux 1.2) is still supported;
 286 it differs from
 287 .I ip_mreqn
 288 only by not including the
 289 .I imr_ifindex
 290 field.
 291 (The kernel determines which structure is being passed based
 292 on the size passed in
 293 .IR optlen .)
 294
 295 .B IP_ADD_MEMBERSHIP
 296 is valid only for
 297 .BR setsockopt (2).
 298 .\"
 299 .TP
 300 .BR IP_ADD_SOURCE_MEMBERSHIP " (since Linux 2.4.22 / 2.5.68)"
 301 Join a multicast group and allow receiving data only
 302 from a specified source.
 303 Argument is an
 304 .I ip_mreq_source
 305 structure.
 306 .sp
 307 .in +4n
 308 .nf
 309 struct ip_mreq_source {
 310     struct in_addr imr_multiaddr;  /* IP multicast group
 311                                       address */
 312     struct in_addr imr_interface;  /* IP address of local
 313                                       interface */
 314     struct in_addr imr_sourceaddr; /* IP address of
 315                                       multicast source */
 316 };
 317 .fi
 318 .in
 319 .sp
 320 The
 321 .I ip_mreq_source
 322 structure is similar to
 323 .I ip_mreqn
 324 described under
 325 .BR IP_ADD_MEMBERSIP .
 326 The
 327 .I imr_multiaddr
 328 field contains the address of the multicast group the application
 329 wants to join or leave.
 330 The
 331 .I imr_interface
 332 field is the address of the local interface with which
 333 the system should join the multicast group.
 334 Finally, the
 335 .I imr_sourceaddr
 336 field contains the address of the source the
 337 application wants to receive data from.
 338 .IP
 339 This option can be used multiple times to allow
 340 receiving data from more than one source.
 341 .TP
 342 .BR IP_BIND_ADDRESS_NO_PORT " (since Linux 4.2)"
 343 .\" commit 90c337da1524863838658078ec34241f45d8394d
 344 Inform the kernel to not reserve an ephemeral port when using
 345 .BR bind (2)
 346 with a port number of 0.
 347 The port will later be automatically chosen at
 348 .BR connect (2)
 349 time,
 350 in a way that allows sharing a source port as long as the 4-tuple is unique.
 351 .TP
 352 .BR IP_BLOCK_SOURCE " (since Linux 2.4.22 / 2.5.68)"
 353 Stop receiving multicast data from a specific source in a given group.
 354 This is valid only after the application has subscribed
 355 to the multicast group using either
 356 .BR IP_ADD_MEMBERSHIP
 357 or
 358 .BR IP_ADD_SOURCE_MEMBERSHIP .
 359 .IP
 360 Argument is an
 361 .I ip_mreq_source
 362 structure as described under
 363 .BR IP_ADD_SOURCE_MEMBERSHIP .
 364 .TP
 365 .BR IP_DROP_MEMBERSHIP " (since Linux 1.2)"
 366 Leave a multicast group.
 367 Argument is an
 368 .I ip_mreqn
 369 or
 370 .I ip_mreq
 371 structure similar to
 372 .BR IP_ADD_MEMBERSHIP .
 373 .TP
 374 .BR IP_DROP_SOURCE_MEMBERSHIP " (since Linux 2.4.22 / 2.5.68)"
 375 Leave a source-specific group\(emthat is, stop receiving data from
 376 a given multicast group that come from a given source.
 377 If the application has subscribed to multiple sources within
 378 the same group, data from the remaining sources will still be delivered.
 379 To stop receiving data from all sources at once, use
 380 .BR IP_DROP_MEMBERSHIP .
 381 .IP
 382 Argument is an
 383 .I ip_mreq_source
 384 structure as described under
 385 .BR IP_ADD_SOURCE_MEMBERSHIP .
 386 .TP
 387 .BR IP_FREEBIND " (since Linux 2.4)"
 388 .\" Precisely: 2.4.0-test10
 389 If enabled, this boolean option allows binding to an IP address
 390 that is nonlocal or does not (yet) exist.
 391 This permits listening on a socket,
 392 without requiring the underlying network interface or the
 393 specified dynamic IP address to be up at the time that
 394 the application is trying to bind to it.
 395 This option is the per-socket equivalent of the
 396 .IR ip_nonlocal_bind
 397 .I /proc
 398 interface described below.
 399 .TP
 400 .BR IP_HDRINCL " (since Linux 2.0)"
 401 If enabled,
 402 the user supplies an IP header in front of the user data.
 403 Valid only for
 404 .B SOCK_RAW
 405 sockets; see
 406 .BR raw (7)
 407 for more information.
 408 When this flag is enabled, the values set by
 409 .BR IP_OPTIONS ,
 410 .BR IP_TTL ,
 411 and
 412 .B IP_TOS
 413 are ignored.
 414 .TP
 415 .BR IP_MSFILTER " (since Linux 2.4.22 / 2.5.68)"
 416 This option provides access to the advanced full-state filtering API.
 417 Argument is an
 418 .I ip_msfilter
 419 structure.
 420 .sp
 421 .in +4n
 422 .nf
 423 struct ip_msfilter {
 424     struct in_addr imsf_multiaddr; /* IP multicast group
 425                                       address */
 426     struct in_addr imsf_interface; /* IP address of local
 427                                       interface */
 428     uint32_t       imsf_fmode;     /* Filter-mode */
 429
 430     uint32_t       imsf_numsrc;    /* Number of sources in
 431                                       the following array */
 432     struct in_addr imsf_slist[1];  /* Array of source
 433                                       addresses */
 434 };
 435 .fi
 436 .in
 437 .sp
 438 There are two macros,
 439 .BR MCAST_INCLUDE
 440 and
 441 .BR MCAST_EXCLUDE ,
 442 which can be used to specify the filtering mode.
 443 Additionally, the
 444 .BR IP_MSFILTER_SIZE (n)
 445 macro exists to determine how much memory is needed to store
 446 .I ip_msfilter
 447 structure with
 448 .I n
 449 sources in the source list.
 450 .IP
 451 For the full description of multicast source filtering
 452 refer to RFC 3376.
 453 .TP
 454 .BR IP_MTU " (since Linux 2.2)"
 455 .\" Precisely: 2.1.124
 456 Retrieve the current known path MTU of the current socket.
 457 Returns an integer.
 458
 459 .B IP_MTU
 460 is valid only for
 461 .BR getsockopt (2)
 462 and can be employed only when the socket has been connected.
 463 .TP
 464 .BR IP_MTU_DISCOVER " (since Linux 2.2)"
 465 .\" Precisely: 2.1.124
 466 Set or receive the Path MTU Discovery setting for a socket.
 467 When enabled, Linux will perform Path MTU Discovery
 468 as defined in RFC\ 1191 on
 469 .B SOCK_STREAM
 470 sockets.
 471 For
 472 .RB non- SOCK_STREAM
 473 sockets,
 474 .B IP_PMTUDISC_DO
 475 forces the don't-fragment flag to be set on all outgoing packets.
 476 It is the user's responsibility to packetize the data
 477 in MTU-sized chunks and to do the retransmits if necessary.
 478 The kernel will reject (with
 479 .BR EMSGSIZE )
 480 datagrams that are bigger than the known path MTU.
 481 .B IP_PMTUDISC_WANT
 482 will fragment a datagram if needed according to the path MTU,
 483 or will set the don't-fragment flag otherwise.
 484
 485 The system-wide default can be toggled between
 486 .B IP_PMTUDISC_WANT
 487 and
 488 .B IP_PMTUDISC_DONT
 489 by writing (respectively, zero and nonzero values) to the
 490 .I /proc/sys/net/ipv4/ip_no_pmtu_disc
 491 file.
 492 .TS
 493 tab(:);
 494 c l
 495 l l.
 496 Path MTU discovery value:Meaning
 497 IP_PMTUDISC_WANT:Use per-route settings.
 498 IP_PMTUDISC_DONT:Never do Path MTU Discovery.
 499 IP_PMTUDISC_DO:Always do Path MTU Discovery.
 500 IP_PMTUDISC_PROBE:Set DF but ignore Path MTU.
 501 .TE
 502
 503 When PMTU discovery is enabled, the kernel automatically keeps track of
 504 the path MTU per destination host.
 505 When it is connected to a specific peer with
 506 .BR connect (2),
 507 the currently known path MTU can be retrieved conveniently using the
 508 .B IP_MTU
 509 socket option (e.g., after an
 510 .B EMSGSIZE
 511 error occurred).
 512 The path MTU may change over time.
 513 For connectionless sockets with many destinations,
 514 the new MTU for a given destination can also be accessed using the
 515 error queue (see
 516 .BR IP_RECVERR ).
 517 A new error will be queued for every incoming MTU update.
 518
 519 While MTU discovery is in progress, initial packets from datagram sockets
 520 may be dropped.
 521 Applications using UDP should be aware of this and not
 522 take it into account for their packet retransmit strategy.
 523
 524 To bootstrap the path MTU discovery process on unconnected sockets, it
 525 is possible to start with a big datagram size
 526 (up to 64K-headers bytes long) and let it shrink by updates of the path MTU.
 527 .\" FIXME . this is an ugly hack
 528
 529 To get an initial estimate of the
 530 path MTU, connect a datagram socket to the destination address using
 531 .BR connect (2)
 532 and retrieve the MTU by calling
 533 .BR getsockopt (2)
 534 with the
 535 .B IP_MTU
 536 option.
 537
 538 It is possible to implement RFC 4821 MTU probing with
 539 .B SOCK_DGRAM
 540 or
 541 .B SOCK_RAW
 542 sockets by setting a value of
 543 .BR IP_PMTUDISC_PROBE
 544 (available since Linux 2.6.22).
 545 This is also particularly useful for diagnostic tools such as
 546 .BR tracepath (8)
 547 that wish to deliberately send probe packets larger than
 548 the observed Path MTU.
 549 .TP
 550 .BR IP_MULTICAST_ALL " (since Linux 2.6.31)"
 551 This option can be used to modify the delivery policy of multicast messages
 552 to sockets bound to the wildcard
 553 .B INADDR_ANY
 554 address.
 555 The argument is a boolean integer (defaults to 1).
 556 If set to 1,
 557 the socket will receive messages from all the groups that have been joined
 558 globally on the whole system.
 559 Otherwise, it will deliver messages only from
 560 the groups that have been explicitly joined (for example via the
 561 .B IP_ADD_MEMBERSHIP
 562 option) on this particular socket.
 563 .TP
 564 .BR IP_MULTICAST_IF " (since Linux 1.2)"
 565 Set the local device for a multicast socket.
 566 The argument for
 567 .BR setsockopt (2)
 568 is an
 569 .I ip_mreqn
 570 or
 571 .\" net: IP_MULTICAST_IF setsockopt now recognizes struct mreq
 572 .\" Commit: 3a084ddb4bf299a6e898a9a07c89f3917f0713f7
 573 (since Linux 3.5)
 574 .I ip_mreq
 575 structure similar to
 576 .BR IP_ADD_MEMBERSHIP ,
 577 or an
 578 .I in_addr
 579 structure.
 580 (The kernel determines which structure is being passed based
 581 on the size passed in
 582 .IR optlen .)
 583 For
 584 .BR getsockopt (2),
 585 the argument is an
 586 .I in_addr
 587 structure.
 588 .TP
 589 .BR IP_MULTICAST_LOOP " (since Linux 1.2)"
 590 Set or read a boolean integer argument that determines whether
 591 sent multicast packets should be looped back to the local sockets.
 592 .TP
 593 .BR IP_MULTICAST_TTL " (since Linux 1.2)"
 594 Set or read the time-to-live value of outgoing multicast packets for this
 595 socket.
 596 It is very important for multicast packets to set the smallest TTL possible.
 597 The default is 1 which means that multicast packets don't leave the local
 598 network unless the user program explicitly requests it.
 599 Argument is an integer.
 600 .TP
 601 .BR IP_NODEFRAG " (since Linux 2.6.36)"
 602 If enabled (argument is nonzero),
 603 the reassembly of outgoing packets is disabled in the netfilter layer.
 604 The argument is an integer.
 605
 606 This option is valid only for
 607 .B SOCK_RAW
 608 sockets.
 609 .TP
 610 .BR IP_OPTIONS " (since Linux 2.0)"
 611 .\" Precisely: 1.3.30
 612 Set or get the IP options to be sent with every packet from this socket.
 613 The arguments are a pointer to a memory buffer containing the options
 614 and the option length.
 615 The
 616 .BR setsockopt (2)
 617 call sets the IP options associated with a socket.
 618 The maximum option size for IPv4 is 40 bytes.
 619 See RFC\ 791 for the allowed options.
 620 When the initial connection request packet for a
 621 .B SOCK_STREAM
 622 socket contains IP options, the IP options will be set automatically
 623 to the options from the initial packet with routing headers reversed.
 624 Incoming packets are not allowed to change options after the connection
 625 is established.
 626 The processing of all incoming source routing options
 627 is disabled by default and can be enabled by using the
 628 .I accept_source_route
 629 .I /proc
 630 interface.
 631 Other options like timestamps are still handled.
 632 For datagram sockets, IP options can be only set by the local user.
 633 Calling
 634 .BR getsockopt (2)
 635 with
 636 .B IP_OPTIONS
 637 puts the current IP options used for sending into the supplied buffer.
 638 .TP
 639 .BR IP_PKTINFO " (since Linux 2.2)"
 640 .\" Precisely: 2.1.68
 641 Pass an
 642 .B IP_PKTINFO
 643 ancillary message that contains a
 644 .I pktinfo
 645 structure that supplies some information about the incoming packet.
 646 This only works for datagram oriented sockets.
 647 The argument is a flag that tells the socket whether the
 648 .B IP_PKTINFO
 649 message should be passed or not.
 650 The message itself can only be sent/retrieved
 651 as control message with a packet using
 652 .BR recvmsg (2)
 653 or
 654 .BR sendmsg (2).
 655 .IP
 656 .in +4n
 657 .nf
 658 struct in_pktinfo {
 659     unsigned int   ipi_ifindex;  /* Interface index */
 660     struct in_addr ipi_spec_dst; /* Local address */
 661     struct in_addr ipi_addr;     /* Header Destination
 662                                     address */
 663 };
 664 .fi
 665 .in
 666 .IP
 667 .\" FIXME . elaborate on that.
 668 .I ipi_ifindex
 669 is the unique index of the interface the packet was received on.
 670 .I ipi_spec_dst
 671 is the local address of the packet and
 672 .I ipi_addr
 673 is the destination address in the packet header.
 674 If
 675 .B IP_PKTINFO
 676 is passed to
 677 .BR sendmsg (2)
 678 and
 679 .\" This field is grossly misnamed
 680 .I ipi_spec_dst
 681 is not zero, then it is used as the local source address for the routing
 682 table lookup and for setting up IP source route options.
 683 When
 684 .I ipi_ifindex
 685 is not zero, the primary local address of the interface specified by the
 686 index overwrites
 687 .I ipi_spec_dst
 688 for the routing table lookup.
 689 .TP
 690 .BR IP_RECVERR " (since Linux 2.2)"
 691 .\" Precisely: 2.1.15
 692 Enable extended reliable error message passing.
 693 When enabled on a datagram socket, all
 694 generated errors will be queued in a per-socket error queue.
 695 When the user receives an error from a socket operation,
 696 the errors can be received by calling
 697 .BR recvmsg (2)
 698 with the
 699 .B MSG_ERRQUEUE
 700 flag set.
 701 The
 702 .I sock_extended_err
 703 structure describing the error will be passed in an ancillary message with
 704 the type
 705 .B IP_RECVERR
 706 and the level
 707 .BR IPPROTO_IP .
 708 .\" or SOL_IP on Linux
 709 This is useful for reliable error handling on unconnected sockets.
 710 The received data portion of the error queue contains the error packet.
 711 .IP
 712 The
 713 .B IP_RECVERR
 714 control message contains a
 715 .I sock_extended_err
 716 structure:
 717 .IP
 718 .in +4n
 719 .ne 18
 720 .nf
 721 #define SO_EE_ORIGIN_NONE    0
 722 #define SO_EE_ORIGIN_LOCAL   1
 723 #define SO_EE_ORIGIN_ICMP    2
 724 #define SO_EE_ORIGIN_ICMP6   3
 725
 726 struct sock_extended_err {
 727     uint32_t ee_errno;   /* error number */
 728     uint8_t  ee_origin;  /* where the error originated */
 729     uint8_t  ee_type;    /* type */
 730     uint8_t  ee_code;    /* code */
 731     uint8_t  ee_pad;
 732     uint32_t ee_info;    /* additional information */
 733     uint32_t ee_data;    /* other data */
 734     /* More data may follow */
 735 };
 736
 737 struct sockaddr *SO_EE_OFFENDER(struct sock_extended_err *);
 738 .fi
 739 .in
 740 .IP
 741 .I ee_errno
 742 contains the
 743 .I errno
 744 number of the queued error.
 745 .I ee_origin
 746 is the origin code of where the error originated.
 747 The other fields are protocol-specific.
 748 The macro
 749 .B SO_EE_OFFENDER
 750 returns a pointer to the address of the network object
 751 where the error originated from given a pointer to the ancillary message.
 752 If this address is not known, the
 753 .I sa_family
 754 member of the
 755 .I sockaddr
 756 contains
 757 .B AF_UNSPEC
 758 and the other fields of the
 759 .I sockaddr
 760 are undefined.
 761 .IP
 762 IP uses the
 763 .I sock_extended_err
 764 structure as follows:
 765 .I ee_origin
 766 is set to
 767 .B SO_EE_ORIGIN_ICMP
 768 for errors received as an ICMP packet, or
 769 .B SO_EE_ORIGIN_LOCAL
 770 for locally generated errors.
 771 Unknown values should be ignored.
 772 .I ee_type
 773 and
 774 .I ee_code
 775 are set from the type and code fields of the ICMP header.
 776 .I ee_info
 777 contains the discovered MTU for
 778 .B EMSGSIZE
 779 errors.
 780 The message also contains the
 781 .I sockaddr_in of the node
 782 caused the error, which can be accessed with the
 783 .B SO_EE_OFFENDER
 784 macro.
 785 The
 786 .I sin_family
 787 field of the
 788 .B SO_EE_OFFENDER
 789 address is
 790 .B AF_UNSPEC
 791 when the source was unknown.
 792 When the error originated from the network, all IP options
 793 .RB ( IP_OPTIONS ", " IP_TTL ", "
 794 etc.) enabled on the socket and contained in the
 795 error packet are passed as control messages.
 796 The payload of the packet causing the error is returned as normal payload.
 797 .\" FIXME . Is it a good idea to document that? It is a dubious feature.
 798 .\" On
 799 .\" .B SOCK_STREAM
 800 .\" sockets,
 801 .\" .B IP_RECVERR
 802 .\" has slightly different semantics. Instead of
 803 .\" saving the errors for the next timeout, it passes all incoming
 804 .\" errors immediately to the user.
 805 .\" This might be useful for very short-lived TCP connections which
 806 .\" need fast error handling. Use this option with care:
 807 .\" it makes TCP unreliable
 808 .\" by not allowing it to recover properly from routing
 809 .\" shifts and other normal
 810 .\" conditions and breaks the protocol specification.
 811 Note that TCP has no error queue;
 812 .B MSG_ERRQUEUE
 813 is not permitted on
 814 .B SOCK_STREAM
 815 sockets.
 816 .B IP_RECVERR
 817 is valid for TCP, but all errors are returned by socket function return or
 818 .B SO_ERROR
 819 only.
 820 .IP
 821 For raw sockets,
 822 .B IP_RECVERR
 823 enables passing of all received ICMP errors to the
 824 application, otherwise errors are only reported on connected sockets
 825 .IP
 826 It sets or retrieves an integer boolean flag.
 827 .B IP_RECVERR
 828 defaults to off.
 829 .TP
 830 .BR IP_RECVOPTS " (since Linux 2.2)"
 831 .\" Precisely: 2.1.15
 832 Pass all incoming IP options to the user in a
 833 .B IP_OPTIONS
 834 control message.
 835 The routing header and other options are already filled in
 836 for the local host.
 837 Not supported for
 838 .B SOCK_STREAM
 839 sockets.
 840 .TP
 841 .BR IP_RECVORIGDSTADDR " (since Linux 2.6.29)"
 842 .\" commit e8b2dfe9b4501ed0047459b2756ba26e5a940a69
 843 This boolean option enables the
 844 .B IP_ORIGDSTADDR
 845 ancillary message in
 846 .BR recvmsg (2),
 847 in which the kernel returns the original destination address
 848 of the datagram being received.
 849 The ancillary message contains a
 850 .IR "struct sockaddr_in" .
 851 .TP
 852 .BR IP_RECVTOS " (since Linux 2.2)"
 853 .\" Precisely: 2.1.68
 854 If enabled, the
 855 .B IP_TOS
 856 ancillary message is passed with incoming packets.
 857 It contains a byte which specifies the Type of Service/Precedence
 858 field of the packet header.
 859 Expects a boolean integer flag.
 860 .TP
 861 .BR IP_RECVTTL " (since Linux 2.2)"
 862 .\" Precisely: 2.1.68
 863 When this flag is set, pass a
 864 .B IP_TTL
 865 control message with the time-to-live
 866 field of the received packet as a byte.
 867 Not supported for
 868 .B SOCK_STREAM
 869 sockets.
 870 .TP
 871 .BR IP_RETOPTS " (since Linux 2.2)"
 872 .\" Precisely: 2.1.15
 873 Identical to
 874 .BR IP_RECVOPTS ,
 875 but returns raw unprocessed options with timestamp and route record
 876 options not filled in for this hop.
 877 .TP
 878 .BR IP_ROUTER_ALERT " (since Linux 2.2)"
 879 .\" Precisely: 2.1.68
 880 Pass all to-be forwarded packets with the
 881 IP Router Alert option set to this socket.
 882 Valid only for raw sockets.
 883 This is useful, for instance, for user-space RSVP daemons.
 884 The tapped packets are not forwarded by the kernel; it is
 885 the user's responsibility to send them out again.
 886 Socket binding is ignored,
 887 such packets are only filtered by protocol.
 888 Expects an integer flag.
 889 .TP
 890 .BR IP_TOS " (since Linux 1.0)"
 891 Set or receive the Type-Of-Service (TOS) field that is sent
 892 with every IP packet originating from this socket.
 893 It is used to prioritize packets on the network.
 894 TOS is a byte.
 895 There are some standard TOS flags defined:
 896 .B IPTOS_LOWDELAY
 897 to minimize delays for interactive traffic,
 898 .B IPTOS_THROUGHPUT
 899 to optimize throughput,
 900 .B IPTOS_RELIABILITY
 901 to optimize for reliability,
 902 .B IPTOS_MINCOST
 903 should be used for "filler data" where slow transmission doesn't matter.
 904 At most one of these TOS values can be specified.
 905 Other bits are invalid and shall be cleared.
 906 Linux sends
 907 .B IPTOS_LOWDELAY
 908 datagrams first by default,
 909 but the exact behavior depends on the configured queueing discipline.
 910 .\" FIXME elaborate on this
 911 Some high-priority levels may require superuser privileges (the
 912 .B CAP_NET_ADMIN
 913 capability).
 914 .\" The priority can also be set in a protocol-independent way by the
 915 .\" .RB ( SOL_SOCKET ", " SO_PRIORITY )
 916 .\" socket option (see
 917 .\" .BR socket (7)).
 918 .TP
 919 .BR IP_TRANSPARENT " (since Linux 2.6.24)"
 920 .\" commit f5715aea4564f233767ea1d944b2637a5fd7cd2e
 921 .\"     This patch introduces the IP_TRANSPARENT socket option: enabling that
 922 .\"     will make the IPv4 routing omit the non-local source address check on
 923 .\"     output. Setting IP_TRANSPARENT requires NET_ADMIN capability.
 924 .\" http://lwn.net/Articles/252545/
 925 Setting this boolean option enables transparent proxying on this socket.
 926 This socket option allows
 927 the calling application to bind to a nonlocal IP address and operate
 928 both as a client and a server with the foreign address as the local endpoint.
 929 NOTE: this requires that routing be set up in a way that
 930 packets going to the foreign address are routed through the TProxy box
 931 (i.e., the system hosting the application that employs the
 932 .B IP_TRANSPARENT
 933 socket option).
 934 Enabling this socket option requires superuser privileges
 935 (the
 936 .BR CAP_NET_ADMIN
 937 capability).
 938 .IP
 939 TProxy redirection with the iptables TPROXY target also requires that
 940 this option be set on the redirected socket.
 941 .TP
 942 .BR IP_TTL " (since Linux 1.0)"
 943 Set or retrieve the current time-to-live field that is used in every packet
 944 sent from this socket.
 945 .TP
 946 .BR IP_UNBLOCK_SOURCE " (since Linux 2.4.22 / 2.5.68)"
 947 Unblock previously blocked multicast source.
 948 Returns
 949 .BR EADDRNOTAVAIL
 950 when given source is not being blocked.
 951 .IP
 952 Argument is an
 953 .I ip_mreq_source
 954 structure as described under
 955 .BR IP_ADD_SOURCE_MEMBERSHIP .
 956 .SS /proc interfaces
 957 The IP protocol
 958 supports a set of
 959 .I /proc
 960 interfaces to configure some global parameters.
 961 The parameters can be accessed by reading or writing files in the directory
 962 .IR /proc/sys/net/ipv4/ .
 963 .\" FIXME As at 2.6.12, 14 Jun 2005, the following are undocumented:
 964 .\"     ip_queue_maxlen
 965 .\"     ip_conntrack_max
 966 Interfaces described as
 967 .I Boolean
 968 take an integer value, with a nonzero value ("true") meaning that
 969 the corresponding option is enabled, and a zero value ("false")
 970 meaning that the option is disabled.
 971 .\"
 972 .TP
 973 .IR ip_always_defrag " (Boolean; since Linux 2.2.13)"
 974 [New with kernel 2.2.13; in earlier kernel versions this feature
 975 was controlled at compile time by the
 976 .B CONFIG_IP_ALWAYS_DEFRAG
 977 option; this option is not present in 2.4.x and later]
 978
 979 When this boolean flag is enabled (not equal 0), incoming fragments
 980 (parts of IP packets
 981 that arose when some host between origin and destination decided
 982 that the packets were too large and cut them into pieces) will be
 983 reassembled (defragmented) before being processed, even if they are
 984 about to be forwarded.
 985
 986 Only enable if running either a firewall that is the sole link
 987 to your network or a transparent proxy; never ever use it for a
 988 normal router or host.
 989 Otherwise, fragmented communication can be disturbed
 990 if the fragments travel over different links.
 991 Defragmentation also has a large memory and CPU time cost.
 992
 993 This is automagically turned on when masquerading or transparent
 994 proxying are configured.
 995 .\"
 996 .TP
 997 .IR ip_autoconfig " (since Linux 2.2 to 2.6.17)"
 998 .\" Precisely: since 2.1.68
 999 .\" FIXME document ip_autoconfig
1000 Not documented.
1001 .\"
1002 .TP
1003 .IR ip_default_ttl " (integer; default: 64; since Linux 2.2)"
1004 .\" Precisely: 2.1.15
1005 Set the default time-to-live value of outgoing packets.
1006 This can be changed per socket with the
1007 .B IP_TTL
1008 option.
1009 .\"
1010 .TP
1011 .IR ip_dynaddr " (Boolean; default: disabled; since Linux 2.0.31)"
1012 Enable dynamic socket address and masquerading entry rewriting on interface
1013 address change.
1014 This is useful for dialup interface with changing IP addresses.
1015 0 means no rewriting, 1 turns it on and 2 enables verbose mode.
1016 .\"
1017 .TP
1018 .IR ip_forward " (Boolean; default: disabled; since Linux 1.2)"
1019 Enable IP forwarding with a boolean flag.
1020 IP forwarding can be also set on a per-interface basis.
1021 .\"
1022 .TP
1023 .IR ip_local_port_range " (since Linux 2.2)"
1024 .\" Precisely: since 2.1.68
1025 This file contains two integers that define the default local port range
1026 allocated to sockets that are not explicitly bound to a port number\(emthat
1027 is, the range used for
1028 .IR "ephemeral ports" .
1029 An ephemeral port is allocated to a socket in the following circumstances:
1030 .RS
1031 .IP * 3
1032 the port number in a socket address is specified as 0 when calling
1033 .BR bind (2);
1034 .IP *
1035 .BR listen (2)
1036 is called on a stream socket that was not previously bound;
1037 .IP *
1038 .BR connect (2)
1039 was called on a socket that was not previously bound;
1040 .IP *
1041 .BR sendto (2)
1042 is called on a datagram socket that was not previously bound.
1043 .RE
1044 .IP
1045 Allocation of ephemeral ports starts with the first number in
1046 .IR ip_local_port_range
1047 and ends with the second number.
1048 If the range of ephemeral ports is exhausted,
1049 then the relevant system call returns an error (but see BUGS).
1050 .IP
1051 Note that the port range in
1052 .IR ip_local_port_range
1053 should not conflict with the ports used by masquerading
1054 (although the case is handled).
1055 Also, arbitrary choices may cause problems with some firewall packet
1056 filters that make assumptions about the local ports in use.
1057 The first number should be at least greater than 1024,
1058 or better, greater than 4096, to avoid clashes
1059 with well known ports and to minimize firewall problems.
1060 .\"
1061 .TP
1062 .IR ip_no_pmtu_disc " (Boolean; default: disabled; since Linux 2.2)"
1063 .\" Precisely: 2.1.15
1064 If enabled, don't do Path MTU Discovery for TCP sockets by default.
1065 Path MTU discovery may fail if misconfigured firewalls (that drop
1066 all ICMP packets) or misconfigured interfaces (e.g., a point-to-point
1067 link where the both ends don't agree on the MTU) are on the path.
1068 It is better to fix the broken routers on the path than to turn off
1069 Path MTU Discovery globally, because not doing it incurs a high cost
1070 to the network.
1071 .\"
1072 .\" The following is from 2.6.12: Documentation/networking/ip-sysctl.txt
1073 .TP
1074 .IR ip_nonlocal_bind " (Boolean; default: disabled; since Linux 2.4)"
1075 .\" Precisely: patch-2.4.0-test10
1076 If set, allows processes to
1077 .BR bind (2)
1078 to nonlocal IP addresses,
1079 which can be quite useful, but may break some applications.
1080 .\"
1081 .\" The following is from 2.6.12: Documentation/networking/ip-sysctl.txt
1082 .TP
1083 .IR ip6frag_time " (integer; default: 30)"
1084 Time in seconds to keep an IPv6 fragment in memory.
1085 .\"
1086 .\" The following is from 2.6.12: Documentation/networking/ip-sysctl.txt
1087 .TP
1088 .IR ip6frag_secret_interval " (integer; default: 600)"
1089 Regeneration interval (in seconds) of the hash secret (or lifetime
1090 for the hash secret) for IPv6 fragments.
1091 .TP
1092 .IR ipfrag_high_thresh " (integer), " ipfrag_low_thresh " (integer)"
1093 If the amount of queued IP fragments reaches
1094 .IR ipfrag_high_thresh ,
1095 the queue is pruned down to
1096 .IR ipfrag_low_thresh .
1097 Contains an integer with the number of bytes.
1098 .TP
1099 .I neigh/*
1100 See
1101 .BR arp (7).
1102 .\" FIXME Document the conf/*/* interfaces
1103 .\"
1104 .\" FIXME Document the route/* interfaces
1105 .SS Ioctls
1106 All ioctls described in
1107 .BR socket (7)
1108 apply to
1109 .BR ip .
1110 .\" 2006-04-02, mtk
1111 .\" commented out the following because ipchains is obsolete
1112 .\" .PP
1113 .\" The ioctls to configure firewalling are documented in
1114 .\" .BR ipfw (4)
1115 .\" from the
1116 .\" .B ipchains
1117 .\" package.
1118 .PP
1119 Ioctls to configure generic device parameters are described in
1120 .BR netdevice (7).
1121 .\" FIXME Add a discussion of multicasting
1122 .SH ERRORS
1123 .\" FIXME document all errors.
1124 .\"     We should really fix the kernels to give more uniform
1125 .\"     error returns (ENOMEM vs ENOBUFS, EPERM vs EACCES etc.)
1126 .TP
1127 .B EACCES
1128 The user tried to execute an operation without the necessary permissions.
1129 These include:
1130 sending a packet to a broadcast address without having the
1131 .B SO_BROADCAST
1132 flag set;
1133 sending a packet via a
1134 .I prohibit
1135 route;
1136 modifying firewall settings without superuser privileges (the
1137 .B CAP_NET_ADMIN
1138 capability);
1139 binding to a privileged port without superuser privileges (the
1140 .B CAP_NET_BIND_SERVICE
1141 capability).
1142 .TP
1143 .B EADDRINUSE
1144 Tried to bind to an address already in use.
1145 .TP
1146 .B EADDRNOTAVAIL
1147 A nonexistent interface was requested or the requested source
1148 address was not local.
1149 .TP
1150 .B EAGAIN
1151 Operation on a nonblocking socket would block.
1152 .TP
1153 .B EALREADY
1154 A connection operation on a nonblocking socket is already in progress.
1155 .TP
1156 .B ECONNABORTED
1157 A connection was closed during an
1158 .BR accept (2).
1159 .TP
1160 .B EHOSTUNREACH
1161 No valid routing table entry matches the destination address.
1162 This error can be caused by a ICMP message from a remote router or
1163 for the local routing table.
1164 .TP
1165 .B EINVAL
1166 Invalid argument passed.
1167 For send operations this can be caused by sending to a
1168 .I blackhole
1169 route.
1170 .TP
1171 .B EISCONN
1172 .BR connect (2)
1173 was called on an already connected socket.
1174 .TP
1175 .B EMSGSIZE
1176 Datagram is bigger than an MTU on the path and it cannot be fragmented.
1177 .TP
1178 .BR ENOBUFS ", " ENOMEM
1179 Not enough free memory.
1180 This often means that the memory allocation is limited by the socket
1181 buffer limits, not by the system memory, but this is not 100% consistent.
1182 .TP
1183 .B ENOENT
1184 .B SIOCGSTAMP
1185 was called on a socket where no packet arrived.
1186 .TP
1187 .B ENOPKG
1188 A kernel subsystem was not configured.
1189 .TP
1190 .BR ENOPROTOOPT " and " EOPNOTSUPP
1191 Invalid socket option passed.
1192 .TP
1193 .B ENOTCONN
1194 The operation is defined only on a connected socket, but the socket wasn't
1195 connected.
1196 .TP
1197 .B EPERM
1198 User doesn't have permission to set high priority, change configuration,
1199 or send signals to the requested process or group.
1200 .TP
1201 .B EPIPE
1202 The connection was unexpectedly closed or shut down by the other end.
1203 .TP
1204 .B ESOCKTNOSUPPORT
1205 The socket is not configured or an unknown socket type was requested.
1206 .PP
1207 Other errors may be generated by the overlaying protocols; see
1208 .BR tcp (7),
1209 .BR raw (7),
1210 .BR udp (7),
1211 and
1212 .BR socket (7).
1213 .SH NOTES
1214 .BR IP_FREEBIND ,
1215 .BR IP_MSFILTER ,
1216 .BR IP_MTU ,
1217 .BR IP_MTU_DISCOVER ,
1218 .BR IP_RECVORIGDSTADDR ,
1219 .BR IP_PKTINFO ,
1220 .BR IP_RECVERR ,
1221 .BR IP_ROUTER_ALERT ,
1222 and
1223 .BR IP_TRANSPARENT
1224 are Linux-specific.
1225 .\" IP_PASSSEC is Linux-specific
1226 .\" IP_XFRM_POLICY is Linux-specific
1227 .\" IP_IPSEC_POLICY is a nonstandard extension, also present on some BSDs
1228
1229 Be very careful with the
1230 .B SO_BROADCAST
1231 option \- it is not privileged in Linux.
1232 It is easy to overload the network
1233 with careless broadcasts.
1234 For new application protocols
1235 it is better to use a multicast group instead of broadcasting.
1236 Broadcasting is discouraged.
1237 .PP
1238 Some other BSD sockets implementations provide
1239 .B IP_RCVDSTADDR
1240 and
1241 .B IP_RECVIF
1242 socket options to get the destination address and the interface of
1243 received datagrams.
1244 Linux has the more general
1245 .B IP_PKTINFO
1246 for the same task.
1247 .PP
1248 Some BSD sockets implementations also provide an
1249 .B IP_RECVTTL
1250 option, but an ancillary message with type
1251 .B IP_RECVTTL
1252 is passed with the incoming packet.
1253 This is different from the
1254 .B IP_TTL
1255 option used in Linux.
1256 .PP
1257 Using the
1258 .B SOL_IP
1259 socket options level isn't portable; BSD-based stacks use the
1260 .B IPPROTO_IP
1261 level.
1262 .SS Compatibility
1263 For compatibility with Linux 2.0, the obsolete
1264 .BI "socket(AF_INET, SOCK_PACKET, " protocol )
1265 syntax is still supported to open a
1266 .BR packet (7)
1267 socket.
1268 This is deprecated and should be replaced by
1269 .BI "socket(AF_PACKET, SOCK_RAW, " protocol )
1270 instead.
1271 The main difference is the new
1272 .I sockaddr_ll
1273 address structure for generic link layer information instead of the old
1274 .BR sockaddr_pkt .
1275 .SH BUGS
1276 There are too many inconsistent error values.
1277 .PP
1278 The error used to diagnose exhaustion of the ephemeral port range differs
1279 across the various system calls
1280 .RB ( connect (2),
1281 .BR bind (2),
1282 .BR listen (2),
1283 .BR sendto (2))
1284 that can assign ephemeral ports.
1285 .PP
1286 The ioctls to configure IP-specific interface options and ARP tables are
1287 not described.
1288 .\" .PP
1289 .\" Some versions of glibc forget to declare
1290 .\" .IR in_pktinfo .
1291 .\" Workaround currently is to copy it into your program from this man page.
1292 .PP
1293 Receiving the original destination address with
1294 .B MSG_ERRQUEUE
1295 in
1296 .I msg_name
1297 by
1298 .BR recvmsg (2)
1299 does not work in some 2.2 kernels.
1300 .\" .SH AUTHORS
1301 .\" This man page was written by Andi Kleen.
1302 .SH SEE ALSO
1303 .BR recvmsg (2),
1304 .BR sendmsg (2),
1305 .BR byteorder (3),
1306 .BR ipfw (4),
1307 .BR capabilities (7),
1308 .BR icmp (7),
1309 .BR ipv6 (7),
1310 .BR netlink (7),
1311 .BR raw (7),
1312 .BR socket (7),
1313 .BR tcp (7),
1314 .BR udp (7)
1315 .PP
1316 RFC\ 791 for the original IP specification.
1317 RFC\ 1122 for the IPv4 host requirements.
1318 RFC\ 1812 for the IPv4 router requirements.