1 .\" SPDX-License-Identifier: Linux-man-pages-1-para
3 .\" This man page is Copyright (C) 1999 Andi Kleen <ak@muc.de>,
4 .\" Copyright (C) 2008-2014, Michael Kerrisk <mtk.manpages@gmail.com>,
5 .\" and Copyright (C) 2016, Heinrich Schuchardt <xypron.glpk@gmx.de>
7 .\" Modified, 2003-12-02, Michael Kerrisk, <mtk.manpages@gmail.com>
8 .\" Modified, 2003-09-23, Adam Langley
9 .\" Modified, 2004-05-27, Michael Kerrisk, <mtk.manpages@gmail.com>
10 .\" Added SOCK_SEQPACKET
11 .\" 2008-05-27, mtk, Provide a clear description of the three types of
12 .\" address that can appear in the sockaddr_un structure: pathname,
13 .\" unnamed, and abstract.
15 .TH UNIX 7 (date) "Linux man-pages (unreleased)"
17 unix \- sockets for local interprocess communication
20 .B #include <sys/socket.h>
21 .B #include <sys/un.h>
23 .IB unix_socket " = socket(AF_UNIX, type, 0);"
24 .IB error " = socketpair(AF_UNIX, type, 0, int *" sv ");"
31 socket family is used to communicate between processes on the same machine
33 Traditionally, UNIX domain sockets can be either unnamed,
34 or bound to a filesystem pathname (marked as being of type socket).
35 Linux also supports an abstract namespace which is independent of the
38 Valid socket types in the UNIX domain are:
40 for a stream-oriented socket;
42 for a datagram-oriented socket that preserves message boundaries
43 (as on most UNIX implementations, UNIX domain datagram
44 sockets are always reliable and don't reorder datagrams);
45 and (since Linux 2.6.4)
47 for a sequenced-packet socket that is connection-oriented,
48 preserves message boundaries,
49 and delivers messages in the order that they were sent.
51 UNIX domain sockets support passing file descriptors or process credentials
52 to other processes using ancillary data.
54 A UNIX domain socket address is represented in the following structure:
58 .\" #define UNIX_PATH_MAX 108
61 sa_family_t sun_family; /* AF_UNIX */
62 char sun_path[108]; /* Pathname */
73 is 108 bytes in size; see also BUGS, below.
75 Various system calls (for example,
83 Some other system calls (for example,
89 return an argument of this type.
91 Three types of address are distinguished in the
96 a UNIX domain socket can be bound to a null-terminated
97 filesystem pathname using
99 When the address of a pathname socket is returned
100 (by one of the system calls noted above),
105 offsetof(struct sockaddr_un, sun_path) + strlen(sun_path) + 1
111 contains the null-terminated pathname.
114 expression equates to the same value as
115 .IR sizeof(sa_family_t) ,
116 but some other implementations include other fields before
120 expression more portably describes the size of the address structure.)
122 For further details of pathname sockets, see below.
125 A stream socket that has not been bound to a pathname using
128 Likewise, the two sockets created by
131 When the address of an unnamed socket is returned,
133 .IR "sizeof(sa_family_t)" ,
136 should not be inspected.
137 .\" There is quite some variation across implementations: FreeBSD
138 .\" says the length is 16 bytes, HP-UX 11 says it's zero bytes.
141 an abstract socket address is distinguished (from a pathname socket)
144 is a null byte (\[aq]\[rs]0\[aq]).
145 The socket's address in this namespace is given by the additional
148 that are covered by the specified length of the address structure.
149 (Null bytes in the name have no special significance.)
150 The name has no connection with filesystem pathnames.
151 When the address of an abstract socket is returned,
155 .I sizeof(sa_family_t)
156 (i.e., greater than 2), and the name of the socket is contained in
158 .I (addrlen \- sizeof(sa_family_t))
162 When binding a socket to a pathname, a few rules should be observed
163 for maximum portability and ease of coding:
167 should be null-terminated.
169 The length of the pathname, including the terminating null byte,
170 should not exceed the size of
175 argument that describes the enclosing
177 structure should have a value of at least:
181 offsetof(struct sockaddr_un, sun_path)+strlen(addr.sun_path)+1
188 .IR "sizeof(struct sockaddr_un)" .
190 There is some variation in how implementations handle UNIX domain
191 socket addresses that do not follow the above rules.
192 For example, some (but not all) implementations
193 .\" Linux does this, including for the case where the supplied path
195 append a null terminator if none is present in the supplied
198 When coding portable applications,
199 keep in mind that some implementations
203 as short as 92 bytes.
204 .\" Modern BSDs generally have 104, Tru64 and AIX have 104,
205 .\" Solaris and Irix have 108
212 return socket address structures.
213 When applied to UNIX domain sockets, the value-result
215 argument supplied to the call should be initialized as above.
216 Upon return, the argument is set to indicate the
218 size of the address structure.
219 The caller should check the value returned in this argument:
220 if the output value exceeds the input value,
221 then there is no guarantee that a null terminator is present in
225 .SS Pathname socket ownership and permissions
226 In the Linux implementation,
227 pathname sockets honor the permissions of the directory they are in.
228 Creation of a new socket fails if the process does not have write and
229 search (execute) permission on the directory in which the socket is created.
232 connecting to a stream socket object requires write permission on that socket;
233 sending a datagram to a datagram socket likewise
234 requires write permission on that socket.
235 POSIX does not make any statement about the effect of the permissions
236 on a socket file, and on some systems (e.g., older BSDs),
237 the socket permissions are ignored.
238 Portable programs should not rely on
239 this feature for security.
241 When creating a new socket, the owner and group of the socket file
242 are set according to the usual rules.
243 The socket file has all permissions enabled,
244 other than those that are turned off by the process
247 The owner, group, and permissions of a pathname socket can be changed (using
251 .\" However, fchown() and fchmod() do not seem to have an effect
254 Socket permissions have no meaning for abstract sockets:
257 has no effect when binding an abstract socket,
258 and changing the ownership and permissions of the object (via
262 has no effect on the accessibility of the socket.
264 Abstract sockets automatically disappear when all open references
265 to the socket are closed.
267 The abstract socket namespace is a nonportable Linux extension.
270 For historical reasons, these socket options are specified with a
272 type even though they are
281 as the socket family.
284 Enabling this socket option causes receipt of the credentials of
285 the sending process in an
286 .B SCM_CREDENTIALS ancillary
287 message in each subsequently received message.
288 The returned credentials are those specified by the sender using
289 .BR SCM_CREDENTIALS ,
290 or a default that includes the sender's PID, real user ID, and real group ID,
291 if the sender did not specify
295 When this option is set and the socket is not yet connected,
296 a unique name in the abstract namespace will be generated automatically.
298 The value given as an argument to
300 and returned as the result of
302 is an integer boolean flag.
305 Enables receiving of the SELinux security label of the peer socket
306 in an ancillary message of type
310 The value given as an argument to
312 and returned as the result of
314 is an integer boolean flag.
318 option is supported for UNIX domain datagram sockets
319 .\" commit 877ce7c1b3afd69a9b1caeb1b9964c992641f52a
321 support for UNIX domain stream sockets was added
322 .\" commit 37a9a8df8ce9de6ea73349c9ac8bdf6ba4ec4f70
330 This read-only socket option returns the
331 credentials of the peer process connected to this socket.
332 The returned credentials are those that were in effect at the time
343 structure; define the
345 feature test macro to obtain the definition of that structure from
348 The use of this option is possible only for connected
350 stream sockets and for
352 stream and datagram socket pairs created using
356 This read-only socket option returns the
357 security context of the peer socket connected to this socket.
358 By default, this will be the same as the security context of
359 the process that created the peer socket unless overridden
360 by the policy or by a process with the required permissions.
364 is a pointer to a buffer of the specified length in bytes
365 into which the security context string will be copied.
366 If the buffer length is less than the length of the security
373 and returns the required length via
375 The caller should allocate at least
377 bytes for the buffer initially, although this is not guaranteed
379 Resizing the buffer to the returned length
380 and retrying may be necessary.
382 The security context string may include a terminating null character
383 in the returned length, but is not guaranteed to do so: a security
384 context "foo" might be represented as either {'f','o','o'} of length 3
385 or {'f','o','o','\\0'} of length 4, which are considered to be
387 The string is printable, does not contain non-terminating null characters,
388 and is in an unspecified encoding (in particular, it
389 is not guaranteed to be ASCII or UTF-8).
391 The use of this option for sockets in the
393 address family is supported since Linux 2.6.2 for connected stream sockets,
395 .\" commit 0b811db2cb2aabc910e53d34ebb95a15997c33e7
396 also for stream and datagram socket pairs created using
405 .IR sizeof(sa_family_t) ,
406 .\" i.e., sizeof(short)
409 socket option was specified for a socket that was
410 not explicitly bound to an address,
411 then the socket is autobound to an abstract address.
412 The address consists of a null byte
413 followed by 5 bytes in the character set
415 Thus, there is a limit of 2\[ha]20 autobind addresses.
416 (From Linux 2.1.15, when the autobind feature was added,
417 8 bytes were used, and the limit was thus 2\[ha]32 autobind addresses.
418 The change to 5 bytes came in Linux 2.3.15.)
420 The following paragraphs describe domain-specific details and
421 unsupported features of the sockets API for UNIX domain sockets on Linux.
423 UNIX domain sockets do not support the transmission of
424 out-of-band data (the
434 flag is not supported by UNIX domain sockets.
437 .\" commit 9f6f9af7694ede6314bed281eec74d588ba9474f
444 was not supported by UNIX domain sockets.
448 socket option does have an effect for UNIX domain sockets, but the
451 For datagram sockets, the
453 value imposes an upper limit on the size of outgoing datagrams.
454 This limit is calculated as the doubled (see
456 option value less 32 bytes used for overhead.
457 .SS Ancillary messages
458 Ancillary data is sent and received using
462 For historical reasons, the ancillary message types listed below
465 type even though they are
468 To send them, set the
477 For more information, see
481 Send or receive a set of open file descriptors from another process.
482 The data portion contains an integer array of the file descriptors.
484 Commonly, this operation is referred to as "passing a file descriptor"
486 However, more accurately,
487 what is being passed is a reference to an open file description (see
489 and in the receiving process it is likely that a different
490 file descriptor number will be used.
491 Semantically, this operation is equivalent to duplicating
493 a file descriptor into the file descriptor table of another process.
495 If the buffer used to receive the ancillary data containing
496 file descriptors is too small (or is absent),
497 then the ancillary data is truncated (or discarded)
498 and the excess file descriptors are automatically closed
499 in the receiving process.
501 If the number of file descriptors received in the ancillary data would
502 cause the process to exceed its
506 the excess file descriptors are automatically closed
507 in the receiving process.
511 defines a limit on the number of file descriptors in the array.
512 Attempting to send an array larger than this limit causes
514 to fail with the error
518 .\" commit bba14de98753cb6599a2dae0e520714b2153522d
519 (or 255 before Linux 2.6.38).
522 Send or receive UNIX credentials.
523 This can be used for authentication.
524 The credentials are passed as a
527 This structure is defined in
534 pid_t pid; /* Process ID of the sending process */
535 uid_t uid; /* User ID of the sending process */
536 gid_t gid; /* Group ID of the sending process */
543 feature test macro must be defined (before including
545 header files) in order to obtain the definition
548 The credentials which the sender specifies are checked by the kernel.
549 A privileged process is allowed to specify values that do not match its own.
550 The sender must specify its own process ID (unless it has the capability
552 in which case the PID of any existing process may be specified),
553 its real user ID, effective user ID, or saved set-user-ID (unless it has
555 and its real group ID, effective group ID, or saved set-group-ID
563 option must be enabled on the socket.
566 Receive the SELinux security context (the security label)
568 The received ancillary data is a null-terminated string containing
569 the security context.
570 The receiver should allocate at least
572 bytes in the data portion of the ancillary message for this data.
574 To receive the security context, the
576 option must be enabled on the socket (see above).
578 When sending ancillary data with
580 only one item of each of the above types may be included in the sent message.
582 At least one byte of real data should be sent when sending ancillary data.
583 On Linux, this is required to successfully send ancillary data over
584 a UNIX domain stream socket.
585 When sending ancillary data over a UNIX domain datagram socket,
586 it is not necessary on Linux to send any accompanying real data.
587 However, portable applications should also include at least one byte
588 of real data when sending ancillary data over a datagram socket.
590 When receiving from a stream socket,
591 ancillary data forms a kind of barrier for the received data.
592 For example, suppose that the sender transmits as follows:
598 of four bytes, with no ancillary data.
601 of one byte, with ancillary data.
604 of four bytes, with no ancillary data.
608 Suppose that the receiver now performs
610 calls each with a buffer size of 20 bytes.
611 The first call will receive five bytes of data,
612 along with the ancillary data sent by the second
615 The next call will receive the remaining four bytes of data.
617 If the space allocated for receiving incoming ancillary data is too small
618 then the ancillary data is truncated to the number of headers
619 that will fit in the supplied buffer (or, in the case of an
621 file descriptor list, the list of file descriptors may be truncated).
622 If no buffer is provided for incoming ancillary data (i.e., the
626 structure supplied to
629 then the incoming ancillary data is discarded.
630 In both of these cases, the
632 flag will be set in the
640 calls return information in
642 The correct syntax is:
647 .IB error " = ioctl(" unix_socket ", " ioctl_type ", &" value ");"
657 sockets, this call returns the number of unread bytes in the receive buffer.
658 The socket must not be in LISTEN state, otherwise an error
663 .IR <linux/sockios.h> .
664 .\" FIXME . https://www.sourceware.org/bugzilla/show_bug.cgi?id=12002,
665 .\" filed 2010-09-10, may cause SIOCINQ to be defined in glibc headers
667 you can use the synonymous
671 .\" SIOCOUTQ also has an effect for UNIX domain sockets, but not
672 .\" quite what userland might expect. It seems to return the number
673 .\" of bytes allocated for buffers containing pending output.
674 .\" That number is normally larger than the number of bytes of pending
675 .\" output. Since this info is, from userland's point of view, imprecise,
676 .\" and it may well change, probably best not to document this now.
680 the returned value is the same as
681 for Internet domain datagram sockets;
687 The specified local address is already in use or the filesystem socket
688 object already exists.
691 This error can occur for
693 when sending a file descriptor as ancillary data over
694 a UNIX domain socket (see the description of
696 above), and indicates that the file descriptor number that
697 is being sent is not valid (e.g., it is not an open file descriptor).
700 The remote address specified by
702 was not a listening socket.
703 This error can also occur if the target pathname is not a socket.
706 Remote socket was unexpectedly closed.
709 User memory address was not valid.
712 Invalid argument passed.
713 A common cause is that the value
715 was not specified in the
717 field of passed addresses, or the socket was in an
718 invalid state for the applied operation.
722 called on an already connected socket or a target address was
723 specified on a connected socket.
726 The system-wide limit on the total number of open files has been reached.
729 The pathname in the remote address specified to
737 Socket operation needs a target address, but the socket is not connected.
740 Stream operation called on non-stream oriented socket or tried to
741 use the out-of-band data option.
744 The sender passed invalid credentials in the
748 Remote socket was closed on a stream socket.
752 This can be avoided by passing the
760 Passed protocol is not
764 Remote socket does not match the local socket type
773 While sending an ancillary message containing credentials
774 .RB ( SCM_CREDENTIALS ),
775 the caller specified a PID that does not match any existing process.
778 This error can occur for
780 when sending a file descriptor as ancillary data over
781 a UNIX domain socket (see the description of
784 It occurs if the number of "in-flight" file descriptors exceeds the
786 resource limit and the caller does not have the
789 An in-flight file descriptor is one that has been sent using
791 but has not yet been accepted in the recipient process using
794 This error is diagnosed since mainline Linux 4.5
795 (and in some earlier kernel versions where the fix has been backported).
796 .\" commit 712f4aad406bb1ed67f3f98d04c044191f0ff593
797 In earlier kernel versions,
798 it was possible to place an unlimited number of file descriptors in flight,
799 by sending each file descriptor with
801 and then closing the file descriptor so that it was not accounted against the
805 Other errors can be generated by the generic socket layer or
806 by the filesystem while generating a filesystem socket object.
807 See the appropriate manual pages for more information.
810 and the abstract namespace were introduced with Linux 2.2 and should not
811 be used in portable programs.
812 (Some BSD-derived systems also support credential passing,
813 but the implementation details differ.)
815 Binding to a socket with a filename creates a socket
816 in the filesystem that must be deleted by the caller when it is no
819 The usual UNIX close-behind semantics apply; the socket can be unlinked
820 at any time and will be finally removed from the filesystem when the last
821 reference to it is closed.
823 To pass file descriptors or credentials over a
826 send or receive at least one byte of nonancillary data in the same
832 UNIX domain stream sockets do not support the notion of out-of-band data.
835 When binding a socket to an address,
836 Linux is one of the implementations that append a null terminator
837 if none is supplied in
839 In most cases this is unproblematic:
840 when the socket address is retrieved,
841 it will be one byte longer than that supplied when the socket was bound.
842 However, there is one case where confusing behavior can result:
843 if 108 non-null bytes are supplied when a socket is bound,
844 then the addition of the null terminator takes the length of
846 .IR sizeof(sun_path) .
847 Consequently, when retrieving the socket address
850 .\" The behavior on Solaris is quite similar.
853 argument for the retrieving call is specified as
854 .IR "sizeof(struct sockaddr_un)" ,
855 then the returned address structure
857 have a null terminator in
860 In addition, some implementations
861 .\" i.e., traditional BSD
862 don't require a null terminator when binding a socket (the
864 argument is used to determine the length of
866 and when the socket address is retrieved on these implementations,
867 there is no null terminator in
870 Applications that retrieve socket addresses can (portably) code
871 to handle the possibility that there is no null terminator in
873 by respecting the fact that the number of valid bytes in the pathname is:
877 strnlen(addr.sun_path, addrlen \- offsetof(sockaddr_un, sun_path))
880 .\" The following patch to amend kernel behavior was rejected:
881 .\" http://thread.gmane.org/gmane.linux.kernel.api/2437
882 .\" Subject: [patch] Fix handling of overlength pathname in AF_UNIX sun_path
884 .\" And there was a related discussion in the Austin list:
885 .\" http://thread.gmane.org/gmane.comp.standards.posix.austin.general/5735
886 .\" Subject: Having a sun_path with no null terminator
889 .\" FIXME . Track http://austingroupbugs.net/view.php?id=561
891 Alternatively, an application can retrieve
892 the socket address by allocating a buffer of size
893 .I "sizeof(struct sockaddr_un)+1"
894 that is zeroed out before the retrieval.
895 The retrieving call can specify
898 .IR "sizeof(struct sockaddr_un)" ,
899 and the extra zero byte ensures that there will be
900 a null terminator for the string returned in
907 addrlen = sizeof(struct sockaddr_un);
908 addrp = malloc(addrlen + 1);
911 memset(addrp, 0, addrlen + 1);
913 if (getsockname(sfd, (struct sockaddr *) addrp, &addrlen)) == \-1)
916 printf("sun_path = %s\[rs]n", ((struct sockaddr_un *) addrp)\->sun_path);
920 This sort of messiness can be avoided if it is guaranteed
921 that the applications that
923 pathname sockets follow the rules outlined above under
924 .IR "Pathname sockets" .
926 The following code demonstrates the use of sequenced-packet
927 sockets for local interprocess communication.
928 It consists of two programs.
929 The server program waits for a connection from the client program.
930 The client sends each of its command-line arguments in separate messages.
931 The server treats the incoming messages as integers and adds them up.
932 The client sends the command string "END".
933 The server sends back a message containing the sum of the client's integers.
934 The client prints the sum and exits.
935 The server waits for the next client to connect.
936 To stop the server, the client is called with the command-line argument "DOWN".
938 The following output was recorded while running the server in the background
939 and repeatedly executing the client.
940 Execution of the server program ends when it receives the "DOWN" command.
948 $ \fB./client 11 \-5\fP
950 $ \fB./client DOWN\fP
958 .\" SRC BEGIN (connection.h)
966 #define SOCKET_NAME "/tmp/9Lq7BNBnBycd6nxy.socket"
967 #define BUFFER_SIZE 12
969 #endif // include guard
973 .\" SRC BEGIN (server.c)
982 #include <sys/socket.h>
983 #include <sys/types.h>
987 #include "connection.h"
994 int connection_socket;
998 struct sockaddr_un name;
999 char buffer[BUFFER_SIZE];
1001 /* Create local socket. */
1003 connection_socket = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1004 if (connection_socket == \-1) {
1010 * For portability clear the whole structure, since some
1011 * implementations have additional (nonstandard) fields in
1015 memset(&name, 0, sizeof(name));
1017 /* Bind socket to socket name. */
1019 name.sun_family = AF_UNIX;
1020 strncpy(name.sun_path, SOCKET_NAME, sizeof(name.sun_path) \- 1);
1022 ret = bind(connection_socket, (const struct sockaddr *) &name,
1030 * Prepare for accepting connections. The backlog size is set
1031 * to 20. So while one request is being processed other requests
1035 ret = listen(connection_socket, 20);
1041 /* This is the main loop for handling connections. */
1045 /* Wait for incoming connection. */
1047 data_socket = accept(connection_socket, NULL, NULL);
1048 if (data_socket == \-1) {
1056 /* Wait for next data packet. */
1058 r = read(data_socket, buffer, sizeof(buffer));
1064 /* Ensure buffer is 0\-terminated. */
1066 buffer[sizeof(buffer) \- 1] = 0;
1068 /* Handle commands. */
1070 if (!strncmp(buffer, "DOWN", sizeof(buffer))) {
1075 if (!strncmp(buffer, "END", sizeof(buffer))) {
1083 /* Add received summand. */
1085 result += atoi(buffer);
1090 sprintf(buffer, "%d", result);
1091 w = write(data_socket, buffer, sizeof(buffer));
1101 /* Quit on DOWN command. */
1108 close(connection_socket);
1110 /* Unlink the socket. */
1112 unlink(SOCKET_NAME);
1119 .\" SRC BEGIN (client.c)
1128 #include <sys/socket.h>
1129 #include <sys/types.h>
1133 #include "connection.h"
1136 main(int argc, char *argv[])
1141 struct sockaddr_un addr;
1142 char buffer[BUFFER_SIZE];
1144 /* Create local socket. */
1146 data_socket = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1147 if (data_socket == \-1) {
1153 * For portability clear the whole structure, since some
1154 * implementations have additional (nonstandard) fields in
1158 memset(&addr, 0, sizeof(addr));
1160 /* Connect socket to socket address. */
1162 addr.sun_family = AF_UNIX;
1163 strncpy(addr.sun_path, SOCKET_NAME, sizeof(addr.sun_path) \- 1);
1165 ret = connect(data_socket, (const struct sockaddr *) &addr,
1168 fprintf(stderr, "The server is down.\[rs]n");
1172 /* Send arguments. */
1174 for (int i = 1; i < argc; ++i) {
1175 w = write(data_socket, argv[i], strlen(argv[i]) + 1);
1182 /* Request result. */
1184 strcpy(buffer, "END");
1185 w = write(data_socket, buffer, strlen(buffer) + 1);
1191 /* Receive result. */
1193 r = read(data_socket, buffer, sizeof(buffer));
1199 /* Ensure buffer is 0\-terminated. */
1201 buffer[sizeof(buffer) \- 1] = 0;
1203 printf("Result = %s\[rs]n", buffer);
1214 For examples of the use of
1219 .BR seccomp_unotify (2).
1226 .BR capabilities (7),
1227 .BR credentials (7),