2 .\" This man page is Copyright (C) 1999 Andi Kleen <ak@muc.de>.
3 .\" and copyright (c) 1999 Matthew Wilcox.
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.
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.
21 .\" The following are not yet documented:
22 .\" SO_PEERNAME (2.4?)
24 .\" Seems to do something similar to getpeername(), but then
25 .\" why is it necessary / how does it differ?
26 .\" SO_TIMESTAMPNS (2.6.22)
27 .\" Documentation/networking/timestamping.txt
28 .\" commit 92f37fd2ee805aa77925c1e64fd56088b46094fc
29 .\" Author: Eric Dumazet <dada1@cosmosbay.com>
30 .\" SO_TIMESTAMPING (2.6.30)
31 .\" Documentation/networking/timestamping.txt
32 .\" commit cb9eff097831007afb30d64373f29d99825d0068
33 .\" Author: Patrick Ohly <patrick.ohly@intel.com>
34 .\" SO_WIFI_STATUS (3.3)
35 .\" commit 6e3e939f3b1bf8534b32ad09ff199d88800835a0
36 .\" Author: Johannes Berg <johannes.berg@intel.com>
37 .\" Also: SCM_WIFI_STATUS
39 .\" commit 3bdc0eba0b8b47797f4a76e377dd8360f317450f
40 .\" Author: Ben Greear <greearb@candelatech.com>
41 .\" SO_GET_FILTER (3.8)
42 .\" commit a8fc92778080c845eaadc369a0ecf5699a03bef0
43 .\" Author: Pavel Emelyanov <xemul@parallels.com>
44 .\" SO_REUSEPORT (3.9)
45 .\" commit c617f398edd4db2b8567a28e899a88f8f574798d
46 .\" https://lwn.net/Articles/542629/
47 .\" SO_LOCK_FILTER (3.9)
48 .\" commit d59577b6ffd313d0ab3be39cb1ab47e29bdc9182
49 .\" Author: Vincent Bernat <bernat@luffy.cx>
50 .\" SO_SELECT_ERR_QUEUE (3.10)
51 .\" commit 7d4c04fc170087119727119074e72445f2bb192b
52 .\" Author: Keller, Jacob E <jacob.e.keller@intel.com>
53 .\" SO_MAX_PACING_RATE (3.13)
54 .\" commit 62748f32d501f5d3712a7c372bbb92abc7c62bc7
56 .TH SOCKET 7 2014-02-21 Linux "Linux Programmer's Manual"
58 socket \- Linux socket interface
60 .B #include <sys/socket.h>
62 .IB sockfd " = socket(int " socket_family ", int " socket_type ", int " protocol );
64 This manual page describes the Linux networking socket layer user
66 The BSD compatible sockets
67 are the uniform interface
68 between the user process and the network protocol stacks in the kernel.
69 The protocol modules are grouped into
72 .BR AF_INET ", " AF_IPX ", and " AF_PACKET ,
81 for more information on families and types.
82 .SS Socket-layer functions
83 These functions are used by the user process to send or receive packets
84 and to do other socket operations.
85 For more information see their respective manual pages.
90 connects a socket to a remote socket address,
93 function binds a socket to a local socket address,
95 tells the socket that new connections shall be accepted, and
97 is used to get a new socket with a new incoming connection.
99 returns two connected anonymous sockets (implemented only for a few
107 send data over a socket, and
111 receive data from a socket.
115 wait for arriving data or a readiness to send data.
116 In addition, the standard I/O operations like
123 can be used to read and write data.
126 returns the local socket address and
128 returns the remote socket address.
132 are used to set or get socket layer or protocol options.
134 can be used to set or read some other options.
137 is used to close a socket.
139 closes parts of a full-duplex socket connection.
145 with a nonzero position is not supported on sockets.
147 It is possible to do nonblocking I/O on sockets by setting the
149 flag on a socket file descriptor using
151 Then all operations that would block will (usually)
154 (operation should be retried later);
159 The user can then wait for various events via
168 Event:Poll flag:Occurrence
173 A connection setup has been completed
174 (for connection-oriented sockets)
177 A disconnection request has been initiated by the other end.
180 A connection is broken (only for connection-oriented protocols).
181 When the socket is written
186 Socket has enough send buffer space for writing new data.
197 Read/Write:POLLERR:An asynchronous error occurred.
198 Read/Write:POLLHUP:The other end has shut down one direction.
204 .\" FIXME . The following is not true currently:
205 .\" It is no I/O event when the connection
206 .\" is broken from the local end using
216 is to let the kernel inform the application about events
222 flag must be set on a socket file descriptor via
224 and a valid signal handler for
226 must be installed via
231 .SS Socket address structures
232 Each socket domain has its own format for socket addresses,
233 with a domain-specific address structure.
234 Each of these structures begins with an
235 integer "family" field (typed as
237 that indicates the type of the address structure.
239 the various system calls (e.g.,
244 .BR getpeername (2)),
245 which are generic to all socket domains,
246 to determine the domain of a particular socket address.
248 To allow any type of socket address to be passed to
249 interfaces in the sockets API,
251 .IR "struct sockaddr"
253 The purpose of this type is purely to allow casting of
254 domain-specific socket address types to a "generic" type,
255 so as to avoid compiler warnings about type mismatches in
256 calls to the sockets API.
258 In addition, the sockets API provides the data type
259 .IR "struct sockaddr_storage".
261 is suitable to accommodate all supported domain-specific socket
262 address structures; it is large enough and is aligned properly.
263 (In particular, it is large enough to hold
264 IPv6 socket addresses.)
265 The structure includes the following field, which can be used to identify
266 the type of socket address actually stored in the structure:
270 sa_family_t ss_family;
276 structure is useful in programs that must handle socket addresses
278 (e.g., programs that must deal with both IPv4 and IPv6 socket addresses).
280 The socket options listed below can be set by using
284 with the socket level set to
287 Unless otherwise noted,
292 .\" In the list below, the text used to describe argument types
293 .\" for each socket option should be more consistent
295 .\" SO_ACCEPTCONN is in POSIX.1-2001, and its origin is explained in
296 .\" W R Stevens, UNPv1
299 Returns a value indicating whether or not this socket has been marked
300 to accept connections with
302 The value 0 indicates that this is not a listening socket,
303 the value 1 indicates that this is a listening socket.
304 This socket option is read-only.
307 Bind this socket to a particular device like \(lqeth0\(rq,
308 as specified in the passed interface name.
310 name is an empty string or the option length is zero, the socket device
312 The passed option is a variable-length null-terminated
313 interface name string with the maximum size of
315 If a socket is bound to an interface,
316 only packets received from that particular interface are processed by the
318 Note that this works only for some socket types, particularly
321 It is not supported for packet sockets (use normal
326 this socket option could be set, but could not retrieved with
328 Since Linux 3.8, it is readable.
331 argument should contain the buffer size available
332 to receive the device name and is recommended to be
335 The real device name length is reported back in the
340 Set or get the broadcast flag.
341 When enabled, datagram sockets are allowed to send
342 packets to a broadcast address.
343 This option has no effect on stream-oriented sockets.
346 Enable BSD bug-to-bug compatibility.
347 This is used by the UDP protocol module in Linux 2.0 and 2.2.
348 If enabled, ICMP errors received for a UDP socket will not be passed
350 In later kernel versions, support for this option has been phased out:
351 Linux 2.4 silently ignores it, and Linux 2.6 generates a kernel warning
352 (printk()) if a program uses this option.
353 Linux 2.0 also enabled BSD bug-to-bug compatibility
354 options (random header changing, skipping of the broadcast flag) for raw
355 sockets with this option, but that was removed in Linux 2.2.
358 Enable socket debugging.
359 Only allowed for processes with the
361 capability or an effective user ID of 0.
363 .BR SO_DOMAIN " (since Linux 2.6.32)"
364 Retrieves the socket domain as an integer, returning a value such as
369 This socket option is read-only.
372 Get and clear the pending socket error.
373 This socket option is read-only.
377 Don't send via a gateway, send only to directly connected hosts.
378 The same effect can be achieved by setting the
383 Expects an integer boolean flag.
386 Enable sending of keep-alive messages on connection-oriented sockets.
387 Expects an integer boolean flag.
400 int l_onoff; /* linger active */
401 int l_linger; /* how many seconds to linger for */
410 will not return until all queued messages for the socket have been
411 successfully sent or the linger timeout has been reached.
413 the call returns immediately and the closing is done in the background.
414 When the socket is closed as part of
416 it always lingers in the background.
418 .BR SO_MARK " (since Linux 2.6.25)"
419 .\" commit 4a19ec5800fc3bb64e2d87c4d9fdd9e636086fe0
420 .\" and 914a9ab386a288d0f22252fc268ecbc048cdcbd5
421 Set the mark for each packet sent through this socket
422 (similar to the netfilter MARK target but socket-based).
423 Changing the mark can be used for mark-based
424 routing without netfilter or for packet filtering.
425 Setting this option requires the
430 If this option is enabled,
431 out-of-band data is directly placed into the receive data stream.
432 Otherwise, out-of-band data is passed only when the
434 flag is set during receiving.
435 .\" don't document it because it can do too much harm.
439 Enable or disable the receiving of the
442 For more information see
444 .\" FIXME Document SO_PASSSEC, added in 2.6.18; there is some info
445 .\" in the 2.6.18 ChangeLog
447 .BR SO_PEEK_OFF " (since Linux 3.4)"
448 .\" commit ef64a54f6e558155b4f149bb10666b9e914b6c54
449 This option, which is currently supported only for
451 sockets, sets the value of the "peek offset" for the
453 system call when used with
457 When this option is set to a negative value
458 (it is set to \-1 for all new sockets),
459 traditional behavior is provided:
463 flag will peek data from the front of the queue.
465 When the option is set to a value greater than or equal to zero,
466 then the next peek at data queued in the socket will occur at
467 the byte offset specified by the option value.
468 At the same time, the "peek offset" will be
469 incremented by the number of bytes that were peeked from the queue,
470 so that a subsequent peek will return the next data in the queue.
472 If data is removed from the front of the queue via a call to
474 (or similar) without the
476 flag, the "peek offset" will be decreased by the number of bytes removed.
477 In other words, receiving data without the
479 flag will cause the "peek offset" to be adjusted to maintain
480 the correct relative position in the queued data,
481 so that a subsequent peek will retrieve the data that would have been
482 retrieved had the data not been removed.
484 For datagram sockets, if the "peek offset" points to the middle of a packet,
485 the data returned will be marked with the
489 The following example serves to illustrate the use of
491 Suppose a stream socket has the following queued input data:
496 The following sequence of
498 calls would have the effect noted in the comments:
502 int ov = 4; // Set peek offset to 4
503 setsockopt(fd, SOL_SOCKET, SO_PEEK_OFF, &ov, sizeof(ov));
505 recv(fd, buf, 2, MSG_PEEK); // Peeks "cc"; offset set to 6
506 recv(fd, buf, 2, MSG_PEEK); // Peeks "dd"; offset set to 8
507 recv(fd, buf, 2, 0); // Reads "aa"; offset set to 6
508 recv(fd, buf, 2, MSG_PEEK); // Peeks "ee"; offset set to 8
513 Return the credentials of the foreign process connected to this socket.
514 This is possible only for connected
518 stream and datagram socket pairs created using
522 The returned credentials are those that were in effect at the time
529 structure; define the
531 feature test macro to obtain the definition of that structure from
533 This socket option is read-only.
536 Set the protocol-defined priority for all packets to be sent on
538 Linux uses this value to order the networking queues:
539 packets with a higher priority may be processed first depending
540 on the selected device queueing discipline.
543 this also sets the IP type-of-service (TOS) field for outgoing packets.
544 Setting a priority outside the range 0 to 6 requires the
548 .BR SO_PROTOCOL " (since Linux 2.6.32)"
549 Retrieves the socket protocol as an integer, returning a value such as
554 This socket option is read-only.
557 Sets or gets the maximum socket receive buffer in bytes.
558 The kernel doubles this value (to allow space for bookkeeping overhead)
560 .\" Most (all?) other implementations do not do this -- MTK, Dec 05
562 and this doubled value is returned by
564 .\" The following thread on LMKL is quite informative:
565 .\" getsockopt/setsockopt with SO_RCVBUF and SO_SNDBUF "non-standard" behavior
567 .\" http://thread.gmane.org/gmane.linux.kernel/1328935
568 The default value is set by the
569 .I /proc/sys/net/core/rmem_default
570 file, and the maximum allowed value is set by the
571 .I /proc/sys/net/core/rmem_max
573 The minimum (doubled) value for this option is 256.
575 .BR SO_RCVBUFFORCE " (since Linux 2.6.14)"
576 Using this socket option, a privileged
577 .RB ( CAP_NET_ADMIN )
578 process can perform the same task as
582 limit can be overridden.
584 .BR SO_RCVLOWAT " and " SO_SNDLOWAT
585 Specify the minimum number of bytes in the buffer until the socket layer
586 will pass the data to the protocol
588 or the user on receiving
590 These two values are initialized to 1.
592 is not changeable on Linux
598 only since Linux 2.4.
603 system calls currently do not respect the
606 and mark a socket readable when even a single byte of data is available.
607 A subsequent read from the socket will block until
610 .\" See http://marc.theaimsgroup.com/?l=linux-kernel&m=111049368106984&w=2
611 .\" Tested on kernel 2.6.14 -- mtk, 30 Nov 05
613 .BR SO_RCVTIMEO " and " SO_SNDTIMEO
614 .\" Not implemented in 2.0.
615 .\" Implemented in 2.1.11 for getsockopt: always return a zero struct.
616 .\" Implemented in 2.3.41 for setsockopt, and actually used.
617 Specify the receiving or sending timeouts until reporting an error.
619 .IR "struct timeval" .
620 If an input or output function blocks for this period of time, and
621 data has been sent or received, the return value of that function
622 will be the amount of data transferred; if no data has been transferred
623 and the timeout has been reached, then \-1 is returned with
629 .\" in fact to EAGAIN
634 just as if the socket was specified to be nonblocking.
635 If the timeout is set to zero (the default),
636 then the operation will never timeout.
637 Timeouts only have effect for system calls that perform socket I/O (e.g.,
642 timeouts have no effect for
649 Indicates that the rules used in validating addresses supplied in a
651 call should allow reuse of local addresses.
655 means that a socket may bind, except when there
656 is an active listening socket bound to the address.
657 When the listening socket is bound to
659 with a specific port then it is not possible
660 to bind to this port for any local address.
661 Argument is an integer boolean flag.
663 .BR SO_RXQ_OVFL " (since Linux 2.6.33)"
664 .\" commit 3b885787ea4112eaa80945999ea0901bf742707f
665 Indicates that an unsigned 32-bit value ancillary message (cmsg)
666 should be attached to received skbs indicating
667 the number of packets dropped by the socket between
668 the last received packet and this received packet.
671 Sets or gets the maximum socket send buffer in bytes.
672 The kernel doubles this value (to allow space for bookkeeping overhead)
674 .\" Most (all?) other implementations do not do this -- MTK, Dec 05
675 .\" See also the comment to SO_RCVBUF (17 Jul 2012 LKML mail)
677 and this doubled value is returned by
679 The default value is set by the
680 .I /proc/sys/net/core/wmem_default
681 file and the maximum allowed value is set by the
682 .I /proc/sys/net/core/wmem_max
684 The minimum (doubled) value for this option is 2048.
686 .BR SO_SNDBUFFORCE " (since Linux 2.6.14)"
687 Using this socket option, a privileged
688 .RB ( CAP_NET_ADMIN )
689 process can perform the same task as
693 limit can be overridden.
696 Enable or disable the receiving of the
699 The timestamp control message is sent with level
706 reception time of the last packet passed to the user in this call.
709 for details on control messages.
712 Gets the socket type as an integer (e.g.,
714 This socket option is read-only.
716 .BR SO_BUSY_POLL " (since Linux 3.11)"
717 Sets the approximate time in microseconds to busy poll on a blocking receive
718 when there is no data.
719 Increasing this value requires
721 The default for this option is controlled by the
722 .I /proc/sys/net/core/busy_read
726 .I /proc/sys/net/core/busy_poll
727 file determines how long
731 will busy poll when they operate on sockets with
733 set and no events to report are found.
736 busy polling will only be done when the socket last received data
737 from a network device that supports this option.
739 While busy polling may improve latency of some applications,
740 care must be taken when using it since this will increase
741 both CPU utilization and power usage.
743 When writing onto a connection-oriented socket that has been shut down
744 (by the local or the remote end)
746 is sent to the writing process and
749 The signal is not sent when the write call
754 When requested with the
761 is sent when an I/O event occurs.
762 It is possible to use
766 in the signal handler to find out which socket the event occurred on.
767 An alternative (in Linux 2.2) is to set a real-time signal using the
770 the handler of the real time signal will be called with
771 the file descriptor in the
777 for more information.
779 Under some circumstances (e.g., multiple processes accessing a
780 single socket), the condition that caused the
782 may have already disappeared when the process reacts to the signal.
783 If this happens, the process should wait again because Linux
784 will resend the signal later.
785 .\" .SS Ancillary messages
787 The core socket networking parameters can be accessed
788 via files in the directory
789 .IR /proc/sys/net/core/ .
792 contains the default setting in bytes of the socket receive buffer.
795 contains the maximum socket receive buffer size in bytes which a user may
801 contains the default setting in bytes of the socket send buffer.
804 contains the maximum socket send buffer size in bytes which a user may
809 .IR message_cost " and " message_burst
810 configure the token bucket filter used to load limit warning messages
811 caused by external network events.
813 .I netdev_max_backlog
814 Maximum number of packets in the global input queue.
817 Maximum length of ancillary data and user control data like the iovecs
819 .\" netdev_fastroute is not documented because it is experimental
821 These operations can be accessed using
826 .IB error " = ioctl(" ip_socket ", " ioctl_type ", " &value_result ");"
833 with the receive timestamp of the last packet passed to the user.
834 This is useful for accurate round trip time measurements.
838 .IR "struct timeval" .
840 This ioctl should be used only if the socket option
842 is not set on the socket.
843 Otherwise, it returns the timestamp of the
844 last packet that was received while
846 was not set, or it fails if no such packet has been received,
855 Set the process or process group to send
861 asynchronous I/O operation has finished or urgent data is available.
862 The argument is a pointer to a
864 If the argument is positive, send the signals to that process.
866 argument is negative, send the signals to the process group with the ID
867 of the absolute value of the argument.
868 The process may only choose itself or its own process group to receive
869 signals unless it has the
871 capability or an effective UID of 0.
876 flag to enable or disable asynchronous I/O mode of the socket.
877 Asynchronous I/O mode means that the
879 signal or the signal set with
881 is raised when a new I/O event occurs.
883 Argument is an integer boolean flag.
884 (This operation is synonymous with the use of
892 Get the current process or process group that receives
915 was introduced in Linux 2.0.30.
920 interfaces was introduced in Linux 2.2.
924 are supported since Linux 2.3.41.
925 Earlier, timeouts were fixed to
926 a protocol-specific setting, and could not be read or written.
928 Linux assumes that half of the send/receive buffer is used for internal
929 kernel structures; thus the values in the corresponding
931 files are twice what can be observed on the wire.
933 Linux will only allow port reuse with the
936 when this option was set both in the previous program that performed a
938 to the port and in the program that wants to reuse the port.
939 This differs from some implementations (e.g., FreeBSD)
940 where only the later program needs to set the
943 Typically this difference is invisible, since, for example, a server
944 program is designed to always set this option.
952 .\" FIXME Document SO_ATTACH_FILTER and SO_DETACH_FILTER
954 The suggested interface to use them is via the libpcap
957 .\" This man page was written by Andi Kleen.
963 .BR capabilities (7),