man7/socket.7

   1 '\" t
   2 .\" This man page is Copyright (C) 1999 Andi Kleen <ak@muc.de>.
   3 .\" and copyright (c) 1999 Matthew Wilcox.
   4 .\"
   5 .\" %%%LICENSE_START(VERBATIM_ONE_PARA)
   6 .\" Permission is granted to distribute possibly modified copies
   7 .\" of this page provided the header is included verbatim,
   8 .\" and in case of nontrivial modification author and date
   9 .\" of the modification is added to the header.
  10 .\" %%%LICENSE_END
  11 .\"
  12 .\" 2002-10-30, Michael Kerrisk, <mtk.manpages@gmail.com>
  13 .\"     Added description of SO_ACCEPTCONN
  14 .\" 2004-05-20, aeb, added SO_RCVTIMEO/SO_SNDTIMEO text.
  15 .\" Modified, 27 May 2004, Michael Kerrisk <mtk.manpages@gmail.com>
  16 .\"     Added notes on capability requirements
  17 .\"     A few small grammar fixes
  18 .\" 2010-06-13 Jan Engelhardt <jengelh@medozas.de>
  19 .\"     Documented SO_DOMAIN and SO_PROTOCOL.
  20 .\"
  21 .\" FIXME
  22 .\" The following are not yet documented:
  23 .\"
  24 .\" SO_PEERNAME (2.4?)
  25 .\"     get only
  26 .\"     Seems to do something similar to getpeername(), but then
  27 .\"     why is it necessary / how does it differ?
  28 .\"
  29 .\" SO_TIMESTAMPNS (2.6.22)
  30 .\"     Documentation/networking/timestamping.txt
  31 .\"     commit 92f37fd2ee805aa77925c1e64fd56088b46094fc
  32 .\"     Author: Eric Dumazet <dada1@cosmosbay.com>
  33 .\"
  34 .\" SO_TIMESTAMPING (2.6.30)
  35 .\"     Documentation/networking/timestamping.txt
  36 .\"     commit cb9eff097831007afb30d64373f29d99825d0068
  37 .\"     Author: Patrick Ohly <patrick.ohly@intel.com>
  38 .\"
  39 .\"  SO_WIFI_STATUS (3.3)
  40 .\"     commit 6e3e939f3b1bf8534b32ad09ff199d88800835a0
  41 .\"     Author: Johannes Berg <johannes.berg@intel.com>
  42 .\"     Also: SCM_WIFI_STATUS
  43 .\"
  44 .\" SO_NOFCS (3.4)
  45 .\"     commit 3bdc0eba0b8b47797f4a76e377dd8360f317450f
  46 .\"     Author: Ben Greear <greearb@candelatech.com>
  47 .\"
  48 .\"  SO_GET_FILTER (3.8)
  49 .\"     commit a8fc92778080c845eaadc369a0ecf5699a03bef0
  50 .\"     Author: Pavel Emelyanov <xemul@parallels.com>
  51 .\"
  52 .\" SO_SELECT_ERR_QUEUE (3.10)
  53 .\"     commit 7d4c04fc170087119727119074e72445f2bb192b
  54 .\"     Author: Keller, Jacob E <jacob.e.keller@intel.com>
  55 .\"
  56 .\" SO_MAX_PACING_RATE (3.13)
  57 .\"     commit 62748f32d501f5d3712a7c372bbb92abc7c62bc7
  58 .\"     Author: Eric Dumazet <edumazet@google.com>
  59 .\"
  60 .\" SO_BPF_EXTENSIONS (3.14)
  61 .\"     commit ea02f9411d9faa3553ed09ce0ec9f00ceae9885e
  62 .\"     Author: Michal Sekletar <msekleta@redhat.com>
  63 .\"
  64 .TH SOCKET 7 2019-03-06 Linux "Linux Programmer's Manual"
  65 .SH NAME
  66 socket \- Linux socket interface
  67 .SH SYNOPSIS
  68 .B #include <sys/socket.h>
  69 .PP
  70 .IB sockfd " = socket(int " socket_family ", int " socket_type ", int " protocol );
  71 .SH DESCRIPTION
  72 This manual page describes the Linux networking socket layer user
  73 interface.
  74 The BSD compatible sockets
  75 are the uniform interface
  76 between the user process and the network protocol stacks in the kernel.
  77 The protocol modules are grouped into
  78 .I protocol families
  79 such as
  80 .BR AF_INET ", " AF_IPX ", and " AF_PACKET ,
  81 and
  82 .I socket types
  83 such as
  84 .B SOCK_STREAM
  85 or
  86 .BR SOCK_DGRAM .
  87 See
  88 .BR socket (2)
  89 for more information on families and types.
  90 .SS Socket-layer functions
  91 These functions are used by the user process to send or receive packets
  92 and to do other socket operations.
  93 For more information see their respective manual pages.
  94 .PP
  95 .BR socket (2)
  96 creates a socket,
  97 .BR connect (2)
  98 connects a socket to a remote socket address,
  99 the
 100 .BR bind (2)
 101 function binds a socket to a local socket address,
 102 .BR listen (2)
 103 tells the socket that new connections shall be accepted, and
 104 .BR accept (2)
 105 is used to get a new socket with a new incoming connection.
 106 .BR socketpair (2)
 107 returns two connected anonymous sockets (implemented only for a few
 108 local families like
 109 .BR AF_UNIX )
 110 .PP
 111 .BR send (2),
 112 .BR sendto (2),
 113 and
 114 .BR sendmsg (2)
 115 send data over a socket, and
 116 .BR recv (2),
 117 .BR recvfrom (2),
 118 .BR recvmsg (2)
 119 receive data from a socket.
 120 .BR poll (2)
 121 and
 122 .BR select (2)
 123 wait for arriving data or a readiness to send data.
 124 In addition, the standard I/O operations like
 125 .BR write (2),
 126 .BR writev (2),
 127 .BR sendfile (2),
 128 .BR read (2),
 129 and
 130 .BR readv (2)
 131 can be used to read and write data.
 132 .PP
 133 .BR getsockname (2)
 134 returns the local socket address and
 135 .BR getpeername (2)
 136 returns the remote socket address.
 137 .BR getsockopt (2)
 138 and
 139 .BR setsockopt (2)
 140 are used to set or get socket layer or protocol options.
 141 .BR ioctl (2)
 142 can be used to set or read some other options.
 143 .PP
 144 .BR close (2)
 145 is used to close a socket.
 146 .BR shutdown (2)
 147 closes parts of a full-duplex socket connection.
 148 .PP
 149 Seeking, or calling
 150 .BR pread (2)
 151 or
 152 .BR pwrite (2)
 153 with a nonzero position is not supported on sockets.
 154 .PP
 155 It is possible to do nonblocking I/O on sockets by setting the
 156 .B O_NONBLOCK
 157 flag on a socket file descriptor using
 158 .BR fcntl (2).
 159 Then all operations that would block will (usually)
 160 return with
 161 .B EAGAIN
 162 (operation should be retried later);
 163 .BR connect (2)
 164 will return
 165 .B EINPROGRESS
 166 error.
 167 The user can then wait for various events via
 168 .BR poll (2)
 169 or
 170 .BR select (2).
 171 .TS
 172 tab(:) allbox;
 173 c s s
 174 l l l.
 175 I/O events
 176 Event:Poll flag:Occurrence
 177 Read:POLLIN:T{
 178 New data arrived.
 179 T}
 180 Read:POLLIN:T{
 181 A connection setup has been completed
 182 (for connection-oriented sockets)
 183 T}
 184 Read:POLLHUP:T{
 185 A disconnection request has been initiated by the other end.
 186 T}
 187 Read:POLLHUP:T{
 188 A connection is broken (only for connection-oriented protocols).
 189 When the socket is written
 190 .B SIGPIPE
 191 is also sent.
 192 T}
 193 Write:POLLOUT:T{
 194 Socket has enough send buffer space for writing new data.
 195 T}
 196 Read/Write:T{
 197 POLLIN |
 198 .br
 199 POLLOUT
 200 T}:T{
 201 An outgoing
 202 .BR connect (2)
 203 finished.
 204 T}
 205 Read/Write:POLLERR:An asynchronous error occurred.
 206 Read/Write:POLLHUP:The other end has shut down one direction.
 207 Exception:POLLPRI:T{
 208 Urgent data arrived.
 209 .B SIGURG
 210 is sent then.
 211 T}
 212 .\" FIXME . The following is not true currently:
 213 .\" It is no I/O event when the connection
 214 .\" is broken from the local end using
 215 .\" .BR shutdown (2)
 216 .\" or
 217 .\" .BR close (2).
 218 .TE
 219 .PP
 220 An alternative to
 221 .BR poll (2)
 222 and
 223 .BR select (2)
 224 is to let the kernel inform the application about events
 225 via a
 226 .B SIGIO
 227 signal.
 228 For that the
 229 .B O_ASYNC
 230 flag must be set on a socket file descriptor via
 231 .BR fcntl (2)
 232 and a valid signal handler for
 233 .B SIGIO
 234 must be installed via
 235 .BR sigaction (2).
 236 See the
 237 .I Signals
 238 discussion below.
 239 .SS Socket address structures
 240 Each socket domain has its own format for socket addresses,
 241 with a domain-specific address structure.
 242 Each of these structures begins with an
 243 integer "family" field (typed as
 244 .IR sa_family_t )
 245 that indicates the type of the address structure.
 246 This allows
 247 the various system calls (e.g.,
 248 .BR connect (2),
 249 .BR bind (2),
 250 .BR accept (2),
 251 .BR getsockname (2),
 252 .BR getpeername (2)),
 253 which are generic to all socket domains,
 254 to determine the domain of a particular socket address.
 255 .PP
 256 To allow any type of socket address to be passed to
 257 interfaces in the sockets API,
 258 the type
 259 .IR "struct sockaddr"
 260 is defined.
 261 The purpose of this type is purely to allow casting of
 262 domain-specific socket address types to a "generic" type,
 263 so as to avoid compiler warnings about type mismatches in
 264 calls to the sockets API.
 265 .PP
 266 In addition, the sockets API provides the data type
 267 .IR "struct sockaddr_storage".
 268 This type
 269 is suitable to accommodate all supported domain-specific socket
 270 address structures; it is large enough and is aligned properly.
 271 (In particular, it is large enough to hold
 272 IPv6 socket addresses.)
 273 The structure includes the following field, which can be used to identify
 274 the type of socket address actually stored in the structure:
 275 .PP
 276 .in +4n
 277 .EX
 278     sa_family_t ss_family;
 279 .EE
 280 .in
 281 .PP
 282 The
 283 .I sockaddr_storage
 284 structure is useful in programs that must handle socket addresses
 285 in a generic way
 286 (e.g., programs that must deal with both IPv4 and IPv6 socket addresses).
 287 .SS Socket options
 288 The socket options listed below can be set by using
 289 .BR setsockopt (2)
 290 and read with
 291 .BR getsockopt (2)
 292 with the socket level set to
 293 .B SOL_SOCKET
 294 for all sockets.
 295 Unless otherwise noted,
 296 .I optval
 297 is a pointer to an
 298 .IR int .
 299 .\" FIXME .
 300 .\" In the list below, the text used to describe argument types
 301 .\" for each socket option should be more consistent
 302 .\"
 303 .\" SO_ACCEPTCONN is in POSIX.1-2001, and its origin is explained in
 304 .\" W R Stevens, UNPv1
 305 .TP
 306 .B SO_ACCEPTCONN
 307 Returns a value indicating whether or not this socket has been marked
 308 to accept connections with
 309 .BR listen (2).
 310 The value 0 indicates that this is not a listening socket,
 311 the value 1 indicates that this is a listening socket.
 312 This socket option is read-only.
 313 .TP
 314 .BR SO_ATTACH_FILTER " (since Linux 2.2), " SO_ATTACH_BPF " (since Linux 3.19)"
 315 Attach a classic BPF
 316 .RB ( SO_ATTACH_FILTER )
 317 or an extended BPF
 318 .RB ( SO_ATTACH_BPF )
 319 program to the socket for use as a filter of incoming packets.
 320 A packet will be dropped if the filter program returns zero.
 321 If the filter program returns a
 322 nonzero value which is less than the packet's data length,
 323 the packet will be truncated to the length returned.
 324 If the value returned by the filter is greater than or equal to the
 325 packet's data length, the packet is allowed to proceed unmodified.
 326 .IP
 327 The argument for
 328 .BR SO_ATTACH_FILTER
 329 is a
 330 .I sock_fprog
 331 structure, defined in
 332 .IR <linux/filter.h> :
 333 .IP
 334 .in +4n
 335 .EX
 336 struct sock_fprog {
 337     unsigned short      len;
 338     struct sock_filter *filter;
 339 };
 340 .EE
 341 .in
 342 .IP
 343 The argument for
 344 .BR SO_ATTACH_BPF
 345 is a file descriptor returned by the
 346 .BR bpf (2)
 347 system call and must refer to a program of type
 348 .BR BPF_PROG_TYPE_SOCKET_FILTER .
 349 .IP
 350 These options may be set multiple times for a given socket,
 351 each time replacing the previous filter program.
 352 The classic and extended versions may be called on the same socket,
 353 but the previous filter will always be replaced such that a socket
 354 never has more than one filter defined.
 355 .IP
 356 Both classic and extended BPF are explained in the kernel source file
 357 .I Documentation/networking/filter.txt
 358 .TP
 359 .BR SO_ATTACH_REUSEPORT_CBPF ", " SO_ATTACH_REUSEPORT_EBPF
 360 For use with the
 361 .BR SO_REUSEPORT
 362 option, these options allow the user to set a classic BPF
 363 .RB ( SO_ATTACH_REUSEPORT_CBPF )
 364 or an extended BPF
 365 .RB ( SO_ATTACH_REUSEPORT_EBPF )
 366 program which defines how packets are assigned to
 367 the sockets in the reuseport group (that is, all sockets which have
 368 .BR SO_REUSEPORT
 369 set and are using the same local address to receive packets).
 370 .IP
 371 The BPF program must return an index between 0 and N\-1 representing
 372 the socket which should receive the packet
 373 (where N is the number of sockets in the group).
 374 If the BPF program returns an invalid index,
 375 socket selection will fall back to the plain
 376 .BR SO_REUSEPORT
 377 mechanism.
 378 .IP
 379 Sockets are numbered in the order in which they are added to the group
 380 (that is, the order of
 381 .BR bind (2)
 382 calls for UDP sockets or the order of
 383 .BR listen (2)
 384 calls for TCP sockets).
 385 New sockets added to a reuseport group will inherit the BPF program.
 386 When a socket is removed from a reuseport group (via
 387 .BR close (2)),
 388 the last socket in the group will be moved into the closed socket's
 389 position.
 390 .IP
 391 These options may be set repeatedly at any time on any socket in the group
 392 to replace the current BPF program used by all sockets in the group.
 393 .IP
 394 .BR SO_ATTACH_REUSEPORT_CBPF
 395 takes the same argument type as
 396 .BR SO_ATTACH_FILTER
 397 and
 398 .BR SO_ATTACH_REUSEPORT_EBPF
 399 takes the same argument type as
 400 .BR SO_ATTACH_BPF .
 401 .IP
 402 UDP support for this feature is available since Linux 4.5;
 403 TCP support is available since Linux 4.6.
 404 .TP
 405 .B SO_BINDTODEVICE
 406 Bind this socket to a particular device like \(lqeth0\(rq,
 407 as specified in the passed interface name.
 408 If the
 409 name is an empty string or the option length is zero, the socket device
 410 binding is removed.
 411 The passed option is a variable-length null-terminated
 412 interface name string with the maximum size of
 413 .BR IFNAMSIZ .
 414 If a socket is bound to an interface,
 415 only packets received from that particular interface are processed by the
 416 socket.
 417 Note that this works only for some socket types, particularly
 418 .B AF_INET
 419 sockets.
 420 It is not supported for packet sockets (use normal
 421 .BR bind (2)
 422 there).
 423 .IP
 424 Before Linux 3.8,
 425 this socket option could be set, but could not retrieved with
 426 .BR getsockopt (2).
 427 Since Linux 3.8, it is readable.
 428 The
 429 .I optlen
 430 argument should contain the buffer size available
 431 to receive the device name and is recommended to be
 432 .BR IFNAMSIZ
 433 bytes.
 434 The real device name length is reported back in the
 435 .I optlen
 436 argument.
 437 .TP
 438 .B SO_BROADCAST
 439 Set or get the broadcast flag.
 440 When enabled, datagram sockets are allowed to send
 441 packets to a broadcast address.
 442 This option has no effect on stream-oriented sockets.
 443 .TP
 444 .B SO_BSDCOMPAT
 445 Enable BSD bug-to-bug compatibility.
 446 This is used by the UDP protocol module in Linux 2.0 and 2.2.
 447 If enabled, ICMP errors received for a UDP socket will not be passed
 448 to the user program.
 449 In later kernel versions, support for this option has been phased out:
 450 Linux 2.4 silently ignores it, and Linux 2.6 generates a kernel warning
 451 (printk()) if a program uses this option.
 452 Linux 2.0 also enabled BSD bug-to-bug compatibility
 453 options (random header changing, skipping of the broadcast flag) for raw
 454 sockets with this option, but that was removed in Linux 2.2.
 455 .TP
 456 .B SO_DEBUG
 457 Enable socket debugging.
 458 Allowed only for processes with the
 459 .B CAP_NET_ADMIN
 460 capability or an effective user ID of 0.
 461 .TP
 462 .BR SO_DETACH_FILTER " (since Linux 2.2), " SO_DETACH_BPF " (since Linux 3.19)"
 463 These two options, which are synonyms,
 464 may be used to remove the classic or extended BPF
 465 program attached to a socket with either
 466 .BR SO_ATTACH_FILTER
 467 or
 468 .BR SO_ATTACH_BPF .
 469 The option value is ignored.
 470 .TP
 471 .BR SO_DOMAIN " (since Linux 2.6.32)"
 472 Retrieves the socket domain as an integer, returning a value such as
 473 .BR AF_INET6 .
 474 See
 475 .BR socket (2)
 476 for details.
 477 This socket option is read-only.
 478 .TP
 479 .B SO_ERROR
 480 Get and clear the pending socket error.
 481 This socket option is read-only.
 482 Expects an integer.
 483 .TP
 484 .B SO_DONTROUTE
 485 Don't send via a gateway, send only to directly connected hosts.
 486 The same effect can be achieved by setting the
 487 .B MSG_DONTROUTE
 488 flag on a socket
 489 .BR send (2)
 490 operation.
 491 Expects an integer boolean flag.
 492 .TP
 493 .BR SO_INCOMING_CPU " (gettable since Linux 3.19, settable since Linux 4.4)"
 494 .\" getsockopt 2c8c56e15df3d4c2af3d656e44feb18789f75837
 495 .\" setsockopt 70da268b569d32a9fddeea85dc18043de9d89f89
 496 Sets or gets the CPU affinity of a socket.
 497 Expects an integer flag.
 498 .IP
 499 .in +4n
 500 .EX
 501 int cpu = 1;
 502 setsockopt(fd, SOL_SOCKET, SO_INCOMING_CPU, &cpu, sizeof(cpu));
 503 .EE
 504 .in
 505 .IP
 506 Because all of the packets for a single stream
 507 (i.e., all packets for the same 4-tuple)
 508 arrive on the single RX queue that is associated with a particular CPU,
 509 the typical use case is to employ one listening process per RX queue,
 510 with the incoming flow being handled by a listener
 511 on the same CPU that is handling the RX queue.
 512 This provides optimal NUMA behavior and keeps CPU caches hot.
 513 .\"
 514 .\" From an email conversation with Eric Dumazet:
 515 .\" >> Note that setting the option is not supported if SO_REUSEPORT is used.
 516 .\" >
 517 .\" > Please define "not supported". Does this yield an API diagnostic?
 518 .\" > If so, what is it?
 519 .\" >
 520 .\" >> Socket will be selected from an array, either by a hash or BPF program
 521 .\" >> that has no access to this information.
 522 .\" >
 523 .\" > Sorry -- I'm lost here. How does this comment relate to the proposed
 524 .\" > man page text above?
 525 .\"
 526 .\" Simply that :
 527 .\"
 528 .\" If an application uses both SO_INCOMING_CPU and SO_REUSEPORT, then
 529 .\" SO_REUSEPORT logic, selecting the socket to receive the packet, ignores
 530 .\" SO_INCOMING_CPU setting.
 531 .TP
 532 .B SO_KEEPALIVE
 533 Enable sending of keep-alive messages on connection-oriented sockets.
 534 Expects an integer boolean flag.
 535 .TP
 536 .B SO_LINGER
 537 Sets or gets the
 538 .B SO_LINGER
 539 option.
 540 The argument is a
 541 .I linger
 542 structure.
 543 .IP
 544 .in +4n
 545 .EX
 546 struct linger {
 547     int l_onoff;    /* linger active */
 548     int l_linger;   /* how many seconds to linger for */
 549 };
 550 .EE
 551 .in
 552 .IP
 553 When enabled, a
 554 .BR close (2)
 555 or
 556 .BR shutdown (2)
 557 will not return until all queued messages for the socket have been
 558 successfully sent or the linger timeout has been reached.
 559 Otherwise,
 560 the call returns immediately and the closing is done in the background.
 561 When the socket is closed as part of
 562 .BR exit (2),
 563 it always lingers in the background.
 564 .TP
 565 .B SO_LOCK_FILTER
 566 .\" commit d59577b6ffd313d0ab3be39cb1ab47e29bdc9182
 567 When set, this option will prevent
 568 changing the filters associated with the socket.
 569 These filters include any set using the socket options
 570 .BR SO_ATTACH_FILTER ,
 571 .BR SO_ATTACH_BPF ,
 572 .BR SO_ATTACH_REUSEPORT_CBPF ,
 573 and
 574 .BR SO_ATTACH_REUSEPORT_EBPF .
 575 .IP
 576 The typical use case is for a privileged process to set up a raw socket
 577 (an operation that requires the
 578 .BR CAP_NET_RAW
 579 capability), apply a restrictive filter, set the
 580 .BR SO_LOCK_FILTER
 581 option,
 582 and then either drop its privileges or pass the socket file descriptor
 583 to an unprivileged process via a UNIX domain socket.
 584 .IP
 585 Once the
 586 .BR SO_LOCK_FILTER
 587 option has been enabled, attempts to change or remove the filter
 588 attached to a socket, or to disable the
 589 .BR SO_LOCK_FILTER
 590 option will fail with the error
 591 .BR EPERM .
 592 .TP
 593 .BR SO_MARK " (since Linux 2.6.25)"
 594 .\" commit 4a19ec5800fc3bb64e2d87c4d9fdd9e636086fe0
 595 .\" and    914a9ab386a288d0f22252fc268ecbc048cdcbd5
 596 Set the mark for each packet sent through this socket
 597 (similar to the netfilter MARK target but socket-based).
 598 Changing the mark can be used for mark-based
 599 routing without netfilter or for packet filtering.
 600 Setting this option requires the
 601 .B CAP_NET_ADMIN
 602 capability.
 603 .TP
 604 .B SO_OOBINLINE
 605 If this option is enabled,
 606 out-of-band data is directly placed into the receive data stream.
 607 Otherwise, out-of-band data is passed only when the
 608 .B MSG_OOB
 609 flag is set during receiving.
 610 .\" don't document it because it can do too much harm.
 611 .\".B SO_NO_CHECK
 612 .\"     The kernel has support for the SO_NO_CHECK socket
 613 .\"     option (boolean: 0 == default, calculate checksum on xmit,
 614 .\"     1 == do not calculate checksum on xmit).
 615 .\" Additional note from Andi Kleen on SO_NO_CHECK (2010-08-30)
 616 .\"     On Linux UDP checksums are essentially free and there's no reason
 617 .\"     to turn them off and it would disable another safety line.
 618 .\"     That is why I didn't document the option.
 619 .TP
 620 .B SO_PASSCRED
 621 Enable or disable the receiving of the
 622 .B SCM_CREDENTIALS
 623 control message.
 624 For more information see
 625 .BR unix (7).
 626 .TP
 627 .B SO_PASSSEC
 628 Enable or disable the receiving of the
 629 .B SCM_SECURITY
 630 control message.
 631 For more information see
 632 .BR unix (7).
 633 .TP
 634 .BR SO_PEEK_OFF " (since Linux 3.4)"
 635 .\" commit ef64a54f6e558155b4f149bb10666b9e914b6c54
 636 This option, which is currently supported only for
 637 .BR unix (7)
 638 sockets, sets the value of the "peek offset" for the
 639 .BR recv (2)
 640 system call when used with
 641 .BR MSG_PEEK
 642 flag.
 643 .IP
 644 When this option is set to a negative value
 645 (it is set to \-1 for all new sockets),
 646 traditional behavior is provided:
 647 .BR recv (2)
 648 with the
 649 .BR MSG_PEEK
 650 flag will peek data from the front of the queue.
 651 .IP
 652 When the option is set to a value greater than or equal to zero,
 653 then the next peek at data queued in the socket will occur at
 654 the byte offset specified by the option value.
 655 At the same time, the "peek offset" will be
 656 incremented by the number of bytes that were peeked from the queue,
 657 so that a subsequent peek will return the next data in the queue.
 658 .IP
 659 If data is removed from the front of the queue via a call to
 660 .BR recv (2)
 661 (or similar) without the
 662 .BR MSG_PEEK
 663 flag, the "peek offset" will be decreased by the number of bytes removed.
 664 In other words, receiving data without the
 665 .B MSG_PEEK
 666 flag will cause the "peek offset" to be adjusted to maintain
 667 the correct relative position in the queued data,
 668 so that a subsequent peek will retrieve the data that would have been
 669 retrieved had the data not been removed.
 670 .IP
 671 For datagram sockets, if the "peek offset" points to the middle of a packet,
 672 the data returned will be marked with the
 673 .BR MSG_TRUNC
 674 flag.
 675 .IP
 676 The following example serves to illustrate the use of
 677 .BR SO_PEEK_OFF .
 678 Suppose a stream socket has the following queued input data:
 679 .IP
 680     aabbccddeeff
 681 .IP
 682 The following sequence of
 683 .BR recv (2)
 684 calls would have the effect noted in the comments:
 685 .IP
 686 .in +4n
 687 .EX
 688 int ov = 4;                  // Set peek offset to 4
 689 setsockopt(fd, SOL_SOCKET, SO_PEEK_OFF, &ov, sizeof(ov));
 690
 691 recv(fd, buf, 2, MSG_PEEK);  // Peeks "cc"; offset set to 6
 692 recv(fd, buf, 2, MSG_PEEK);  // Peeks "dd"; offset set to 8
 693 recv(fd, buf, 2, 0);         // Reads "aa"; offset set to 6
 694 recv(fd, buf, 2, MSG_PEEK);  // Peeks "ee"; offset set to 8
 695 .EE
 696 .in
 697 .TP
 698 .B SO_PEERCRED
 699 Return the credentials of the peer process connected to this socket.
 700 For further details, see
 701 .BR unix (7).
 702 .TP
 703 .B SO_PRIORITY
 704 Set the protocol-defined priority for all packets to be sent on
 705 this socket.
 706 Linux uses this value to order the networking queues:
 707 packets with a higher priority may be processed first depending
 708 on the selected device queueing discipline.
 709 .\" For
 710 .\" .BR ip (7),
 711 .\" this also sets the IP type-of-service (TOS) field for outgoing packets.
 712 Setting a priority outside the range 0 to 6 requires the
 713 .B CAP_NET_ADMIN
 714 capability.
 715 .TP
 716 .BR SO_PROTOCOL " (since Linux 2.6.32)"
 717 Retrieves the socket protocol as an integer, returning a value such as
 718 .BR IPPROTO_SCTP .
 719 See
 720 .BR socket (2)
 721 for details.
 722 This socket option is read-only.
 723 .TP
 724 .B SO_RCVBUF
 725 Sets or gets the maximum socket receive buffer in bytes.
 726 The kernel doubles this value (to allow space for bookkeeping overhead)
 727 when it is set using
 728 .\" Most (all?) other implementations do not do this -- MTK, Dec 05
 729 .BR setsockopt (2),
 730 and this doubled value is returned by
 731 .BR getsockopt (2).
 732 .\" The following thread on LMKL is quite informative:
 733 .\" getsockopt/setsockopt with SO_RCVBUF and SO_SNDBUF "non-standard" behavior
 734 .\" 17 July 2012
 735 .\" http://thread.gmane.org/gmane.linux.kernel/1328935
 736 The default value is set by the
 737 .I /proc/sys/net/core/rmem_default
 738 file, and the maximum allowed value is set by the
 739 .I /proc/sys/net/core/rmem_max
 740 file.
 741 The minimum (doubled) value for this option is 256.
 742 .TP
 743 .BR SO_RCVBUFFORCE " (since Linux 2.6.14)"
 744 Using this socket option, a privileged
 745 .RB ( CAP_NET_ADMIN )
 746 process can perform the same task as
 747 .BR SO_RCVBUF ,
 748 but the
 749 .I rmem_max
 750 limit can be overridden.
 751 .TP
 752 .BR SO_RCVLOWAT " and " SO_SNDLOWAT
 753 Specify the minimum number of bytes in the buffer until the socket layer
 754 will pass the data to the protocol
 755 .RB ( SO_SNDLOWAT )
 756 or the user on receiving
 757 .RB ( SO_RCVLOWAT ).
 758 These two values are initialized to 1.
 759 .B SO_SNDLOWAT
 760 is not changeable on Linux
 761 .RB ( setsockopt (2)
 762 fails with the error
 763 .BR ENOPROTOOPT ).
 764 .B SO_RCVLOWAT
 765 is changeable
 766 only since Linux 2.4.
 767 The
 768 .BR select (2)
 769 and
 770 .BR poll (2)
 771 system calls currently do not respect the
 772 .B SO_RCVLOWAT
 773 setting on Linux,
 774 and mark a socket readable when even a single byte of data is available.
 775 A subsequent read from the socket will block until
 776 .B SO_RCVLOWAT
 777 bytes are available.
 778 .\" See http://marc.theaimsgroup.com/?l=linux-kernel&m=111049368106984&w=2
 779 .\" Tested on kernel 2.6.14 -- mtk, 30 Nov 05
 780 .TP
 781 .BR SO_RCVTIMEO " and " SO_SNDTIMEO
 782 .\" Not implemented in 2.0.
 783 .\" Implemented in 2.1.11 for getsockopt: always return a zero struct.
 784 .\" Implemented in 2.3.41 for setsockopt, and actually used.
 785 Specify the receiving or sending timeouts until reporting an error.
 786 The argument is a
 787 .IR "struct timeval" .
 788 If an input or output function blocks for this period of time, and
 789 data has been sent or received, the return value of that function
 790 will be the amount of data transferred; if no data has been transferred
 791 and the timeout has been reached, then \-1 is returned with
 792 .I errno
 793 set to
 794 .BR EAGAIN
 795 or
 796 .BR EWOULDBLOCK ,
 797 .\" in fact to EAGAIN
 798 or
 799 .B EINPROGRESS
 800 (for
 801 .BR connect (2))
 802 just as if the socket was specified to be nonblocking.
 803 If the timeout is set to zero (the default),
 804 then the operation will never timeout.
 805 Timeouts only have effect for system calls that perform socket I/O (e.g.,
 806 .BR read (2),
 807 .BR recvmsg (2),
 808 .BR send (2),
 809 .BR sendmsg (2));
 810 timeouts have no effect for
 811 .BR select (2),
 812 .BR poll (2),
 813 .BR epoll_wait (2),
 814 and so on.
 815 .TP
 816 .B SO_REUSEADDR
 817 .\"     commit c617f398edd4db2b8567a28e899a88f8f574798d
 818 .\"     https://lwn.net/Articles/542629/
 819 Indicates that the rules used in validating addresses supplied in a
 820 .BR bind (2)
 821 call should allow reuse of local addresses.
 822 For
 823 .B AF_INET
 824 sockets this
 825 means that a socket may bind, except when there
 826 is an active listening socket bound to the address.
 827 When the listening socket is bound to
 828 .B INADDR_ANY
 829 with a specific port then it is not possible
 830 to bind to this port for any local address.
 831 Argument is an integer boolean flag.
 832 .TP
 833 .BR SO_REUSEPORT " (since Linux 3.9)"
 834 Permits multiple
 835 .B AF_INET
 836 or
 837 .B AF_INET6
 838 sockets to be bound to an identical socket address.
 839 This option must be set on each socket (including the first socket)
 840 prior to calling
 841 .BR bind (2)
 842 on the socket.
 843 To prevent port hijacking,
 844 all of the processes binding to the same address must have the same
 845 effective UID.
 846 This option can be employed with both TCP and UDP sockets.
 847 .IP
 848 For TCP sockets, this option allows
 849 .BR accept (2)
 850 load distribution in a multi-threaded server to be improved by
 851 using a distinct listener socket for each thread.
 852 This provides improved load distribution as compared
 853 to traditional techniques such using a single
 854 .BR accept (2)ing
 855 thread that distributes connections,
 856 or having multiple threads that compete to
 857 .BR accept (2)
 858 from the same socket.
 859 .IP
 860 For UDP sockets,
 861 the use of this option can provide better distribution
 862 of incoming datagrams to multiple processes (or threads) as compared
 863 to the traditional technique of having multiple processes
 864 compete to receive datagrams on the same socket.
 865 .TP
 866 .BR SO_RXQ_OVFL " (since Linux 2.6.33)"
 867 .\" commit 3b885787ea4112eaa80945999ea0901bf742707f
 868 Indicates that an unsigned 32-bit value ancillary message (cmsg)
 869 should be attached to received skbs indicating
 870 the number of packets dropped by the socket since its creation.
 871 .TP
 872 .B SO_SNDBUF
 873 Sets or gets the maximum socket send buffer in bytes.
 874 The kernel doubles this value (to allow space for bookkeeping overhead)
 875 when it is set using
 876 .\" Most (all?) other implementations do not do this -- MTK, Dec 05
 877 .\" See also the comment to SO_RCVBUF (17 Jul 2012 LKML mail)
 878 .BR setsockopt (2),
 879 and this doubled value is returned by
 880 .BR getsockopt (2).
 881 The default value is set by the
 882 .I /proc/sys/net/core/wmem_default
 883 file and the maximum allowed value is set by the
 884 .I /proc/sys/net/core/wmem_max
 885 file.
 886 The minimum (doubled) value for this option is 2048.
 887 .TP
 888 .BR SO_SNDBUFFORCE " (since Linux 2.6.14)"
 889 Using this socket option, a privileged
 890 .RB ( CAP_NET_ADMIN )
 891 process can perform the same task as
 892 .BR SO_SNDBUF ,
 893 but the
 894 .I wmem_max
 895 limit can be overridden.
 896 .TP
 897 .B SO_TIMESTAMP
 898 Enable or disable the receiving of the
 899 .B SO_TIMESTAMP
 900 control message.
 901 The timestamp control message is sent with level
 902 .B SOL_SOCKET
 903 and the
 904 .I cmsg_data
 905 field is a
 906 .I "struct timeval"
 907 indicating the
 908 reception time of the last packet passed to the user in this call.
 909 See
 910 .BR cmsg (3)
 911 for details on control messages.
 912 .TP
 913 .B SO_TYPE
 914 Gets the socket type as an integer (e.g.,
 915 .BR SOCK_STREAM ).
 916 This socket option is read-only.
 917 .TP
 918 .BR SO_BUSY_POLL " (since Linux 3.11)"
 919 Sets the approximate time in microseconds to busy poll on a blocking receive
 920 when there is no data.
 921 Increasing this value requires
 922 .BR CAP_NET_ADMIN .
 923 The default for this option is controlled by the
 924 .I /proc/sys/net/core/busy_read
 925 file.
 926 .IP
 927 The value in the
 928 .I /proc/sys/net/core/busy_poll
 929 file determines how long
 930 .BR select (2)
 931 and
 932 .BR poll (2)
 933 will busy poll when they operate on sockets with
 934 .BR SO_BUSY_POLL
 935 set and no events to report are found.
 936 .IP
 937 In both cases,
 938 busy polling will only be done when the socket last received data
 939 from a network device that supports this option.
 940 .IP
 941 While busy polling may improve latency of some applications,
 942 care must be taken when using it since this will increase
 943 both CPU utilization and power usage.
 944 .SS Signals
 945 When writing onto a connection-oriented socket that has been shut down
 946 (by the local or the remote end)
 947 .B SIGPIPE
 948 is sent to the writing process and
 949 .B EPIPE
 950 is returned.
 951 The signal is not sent when the write call
 952 specified the
 953 .B MSG_NOSIGNAL
 954 flag.
 955 .PP
 956 When requested with the
 957 .B FIOSETOWN
 958 .BR fcntl (2)
 959 or
 960 .B SIOCSPGRP
 961 .BR ioctl (2),
 962 .B SIGIO
 963 is sent when an I/O event occurs.
 964 It is possible to use
 965 .BR poll (2)
 966 or
 967 .BR select (2)
 968 in the signal handler to find out which socket the event occurred on.
 969 An alternative (in Linux 2.2) is to set a real-time signal using the
 970 .B F_SETSIG
 971 .BR fcntl (2);
 972 the handler of the real time signal will be called with
 973 the file descriptor in the
 974 .I si_fd
 975 field of its
 976 .IR siginfo_t .
 977 See
 978 .BR fcntl (2)
 979 for more information.
 980 .PP
 981 Under some circumstances (e.g., multiple processes accessing a
 982 single socket), the condition that caused the
 983 .B SIGIO
 984 may have already disappeared when the process reacts to the signal.
 985 If this happens, the process should wait again because Linux
 986 will resend the signal later.
 987 .\" .SS Ancillary messages
 988 .SS /proc interfaces
 989 The core socket networking parameters can be accessed
 990 via files in the directory
 991 .IR /proc/sys/net/core/ .
 992 .TP
 993 .I rmem_default
 994 contains the default setting in bytes of the socket receive buffer.
 995 .TP
 996 .I rmem_max
 997 contains the maximum socket receive buffer size in bytes which a user may
 998 set by using the
 999 .B SO_RCVBUF
1000 socket option.
1001 .TP
1002 .I wmem_default
1003 contains the default setting in bytes of the socket send buffer.
1004 .TP
1005 .I wmem_max
1006 contains the maximum socket send buffer size in bytes which a user may
1007 set by using the
1008 .B SO_SNDBUF
1009 socket option.
1010 .TP
1011 .IR message_cost " and " message_burst
1012 configure the token bucket filter used to load limit warning messages
1013 caused by external network events.
1014 .TP
1015 .I netdev_max_backlog
1016 Maximum number of packets in the global input queue.
1017 .TP
1018 .I optmem_max
1019 Maximum length of ancillary data and user control data like the iovecs
1020 per socket.
1021 .\" netdev_fastroute is not documented because it is experimental
1022 .SS Ioctls
1023 These operations can be accessed using
1024 .BR ioctl (2):
1025 .PP
1026 .in +4n
1027 .EX
1028 .IB error " = ioctl(" ip_socket ", " ioctl_type ", " &value_result ");"
1029 .EE
1030 .in
1031 .TP
1032 .B SIOCGSTAMP
1033 Return a
1034 .I struct timeval
1035 with the receive timestamp of the last packet passed to the user.
1036 This is useful for accurate round trip time measurements.
1037 See
1038 .BR setitimer (2)
1039 for a description of
1040 .IR "struct timeval" .
1041 .\"
1042 This ioctl should be used only if the socket option
1043 .B SO_TIMESTAMP
1044 is not set on the socket.
1045 Otherwise, it returns the timestamp of the
1046 last packet that was received while
1047 .B SO_TIMESTAMP
1048 was not set, or it fails if no such packet has been received,
1049 (i.e.,
1050 .BR ioctl (2)
1051 returns \-1 with
1052 .I errno
1053 set to
1054 .BR ENOENT ).
1055 .TP
1056 .B SIOCSPGRP
1057 Set the process or process group that is to receive
1058 .B SIGIO
1059 or
1060 .B SIGURG
1061 signals when I/O becomes possible or urgent data is available.
1062 The argument is a pointer to a
1063 .IR pid_t .
1064 For further details, see the description of
1065 .BR F_SETOWN
1066 in
1067 .BR fcntl (2).
1068 .TP
1069 .B FIOASYNC
1070 Change the
1071 .B O_ASYNC
1072 flag to enable or disable asynchronous I/O mode of the socket.
1073 Asynchronous I/O mode means that the
1074 .B SIGIO
1075 signal or the signal set with
1076 .B F_SETSIG
1077 is raised when a new I/O event occurs.
1078 .IP
1079 Argument is an integer boolean flag.
1080 (This operation is synonymous with the use of
1081 .BR fcntl (2)
1082 to set the
1083 .B O_ASYNC
1084 flag.)
1085 .\"
1086 .TP
1087 .B SIOCGPGRP
1088 Get the current process or process group that receives
1089 .B SIGIO
1090 or
1091 .B SIGURG
1092 signals,
1093 or 0
1094 when none is set.
1095 .PP
1096 Valid
1097 .BR fcntl (2)
1098 operations:
1099 .TP
1100 .B FIOGETOWN
1101 The same as the
1102 .B SIOCGPGRP
1103 .BR ioctl (2).
1104 .TP
1105 .B FIOSETOWN
1106 The same as the
1107 .B SIOCSPGRP
1108 .BR ioctl (2).
1109 .SH VERSIONS
1110 .B SO_BINDTODEVICE
1111 was introduced in Linux 2.0.30.
1112 .B SO_PASSCRED
1113 is new in Linux 2.2.
1114 The
1115 .I /proc
1116 interfaces were introduced in Linux 2.2.
1117 .B SO_RCVTIMEO
1118 and
1119 .B SO_SNDTIMEO
1120 are supported since Linux 2.3.41.
1121 Earlier, timeouts were fixed to
1122 a protocol-specific setting, and could not be read or written.
1123 .SH NOTES
1124 Linux assumes that half of the send/receive buffer is used for internal
1125 kernel structures; thus the values in the corresponding
1126 .I /proc
1127 files are twice what can be observed on the wire.
1128 .PP
1129 Linux will allow port reuse only with the
1130 .B SO_REUSEADDR
1131 option
1132 when this option was set both in the previous program that performed a
1133 .BR bind (2)
1134 to the port and in the program that wants to reuse the port.
1135 This differs from some implementations (e.g., FreeBSD)
1136 where only the later program needs to set the
1137 .B SO_REUSEADDR
1138 option.
1139 Typically this difference is invisible, since, for example, a server
1140 program is designed to always set this option.
1141 .\" .SH AUTHORS
1142 .\" This man page was written by Andi Kleen.
1143 .SH SEE ALSO
1144 .BR wireshark (1),
1145 .BR bpf (2),
1146 .BR connect (2),
1147 .BR getsockopt (2),
1148 .BR setsockopt (2),
1149 .BR socket (2),
1150 .BR pcap (3),
1151 .BR address_families (7),
1152 .BR capabilities (7),
1153 .BR ddp (7),
1154 .BR ip (7),
1155 .BR packet (7),
1156 .BR tcp (7),
1157 .BR udp (7),
1158 .BR unix (7),
1159 .BR tcpdump (8)