1 .\" This man page is Copyright (C) 1999 Andi Kleen <ak@muc.de>,
2 .\" Copyright (C) 2008-2014, Michael Kerrisk <mtk.manpages@gmail.com>,
3 .\" and Copyright (C) 2016, Heinrich Schuchardt <xypron.glpk@gmx.de>
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 .\" Modified, 2003-12-02, Michael Kerrisk, <mtk.manpages@gmail.com>
13 .\" Modified, 2003-09-23, Adam Langley
14 .\" Modified, 2004-05-27, Michael Kerrisk, <mtk.manpages@gmail.com>
15 .\" Added SOCK_SEQPACKET
16 .\" 2008-05-27, mtk, Provide a clear description of the three types of
17 .\" address that can appear in the sockaddr_un structure: pathname,
18 .\" unnamed, and abstract.
20 .TH UNIX 7 2021-03-22 "Linux man-pages (unreleased)"
22 unix \- sockets for local interprocess communication
25 .B #include <sys/socket.h>
26 .B #include <sys/un.h>
28 .IB unix_socket " = socket(AF_UNIX, type, 0);"
29 .IB error " = socketpair(AF_UNIX, type, 0, int *" sv ");"
36 socket family is used to communicate between processes on the same machine
38 Traditionally, UNIX domain sockets can be either unnamed,
39 or bound to a filesystem pathname (marked as being of type socket).
40 Linux also supports an abstract namespace which is independent of the
43 Valid socket types in the UNIX domain are:
45 for a stream-oriented socket;
47 for a datagram-oriented socket that preserves message boundaries
48 (as on most UNIX implementations, UNIX domain datagram
49 sockets are always reliable and don't reorder datagrams);
50 and (since Linux 2.6.4)
52 for a sequenced-packet socket that is connection-oriented,
53 preserves message boundaries,
54 and delivers messages in the order that they were sent.
56 UNIX domain sockets support passing file descriptors or process credentials
57 to other processes using ancillary data.
59 A UNIX domain socket address is represented in the following structure:
63 .\" #define UNIX_PATH_MAX 108
66 sa_family_t sun_family; /* AF_UNIX */
67 char sun_path[108]; /* Pathname */
78 is 108 bytes in size; see also NOTES, below.
80 Various systems calls (for example,
88 Some other system calls (for example,
94 return an argument of this type.
96 Three types of address are distinguished in the
101 a UNIX domain socket can be bound to a null-terminated
102 filesystem pathname using
104 When the address of a pathname socket is returned
105 (by one of the system calls noted above),
110 offsetof(struct sockaddr_un, sun_path) + strlen(sun_path) + 1
116 contains the null-terminated pathname.
119 expression equates to the same value as
120 .IR sizeof(sa_family_t) ,
121 but some other implementations include other fields before
125 expression more portably describes the size of the address structure.)
127 For further details of pathname sockets, see below.
130 A stream socket that has not been bound to a pathname using
133 Likewise, the two sockets created by
136 When the address of an unnamed socket is returned,
138 .IR "sizeof(sa_family_t)" ,
141 should not be inspected.
142 .\" There is quite some variation across implementations: FreeBSD
143 .\" says the length is 16 bytes, HP-UX 11 says it's zero bytes.
146 an abstract socket address is distinguished (from a pathname socket)
149 is a null byte (\(aq\e0\(aq).
150 The socket's address in this namespace is given by the additional
153 that are covered by the specified length of the address structure.
154 (Null bytes in the name have no special significance.)
155 The name has no connection with filesystem pathnames.
156 When the address of an abstract socket is returned,
160 .I sizeof(sa_family_t)
161 (i.e., greater than 2), and the name of the socket is contained in
163 .I (addrlen \- sizeof(sa_family_t))
167 When binding a socket to a pathname, a few rules should be observed
168 for maximum portability and ease of coding:
172 should be null-terminated.
174 The length of the pathname, including the terminating null byte,
175 should not exceed the size of
180 argument that describes the enclosing
182 structure should have a value of at least:
186 offsetof(struct sockaddr_un, sun_path)+strlen(addr.sun_path)+1
193 .IR "sizeof(struct sockaddr_un)" .
195 There is some variation in how implementations handle UNIX domain
196 socket addresses that do not follow the above rules.
197 For example, some (but not all) implementations
198 .\" Linux does this, including for the case where the supplied path
200 append a null terminator if none is present in the supplied
203 When coding portable applications,
204 keep in mind that some implementations
208 as short as 92 bytes.
209 .\" Modern BSDs generally have 104, Tru64 and AIX have 104,
210 .\" Solaris and Irix have 108
217 return socket address structures.
218 When applied to UNIX domain sockets, the value-result
220 argument supplied to the call should be initialized as above.
221 Upon return, the argument is set to indicate the
223 size of the address structure.
224 The caller should check the value returned in this argument:
225 if the output value exceeds the input value,
226 then there is no guarantee that a null terminator is present in
230 .SS Pathname socket ownership and permissions
231 In the Linux implementation,
232 pathname sockets honor the permissions of the directory they are in.
233 Creation of a new socket fails if the process does not have write and
234 search (execute) permission on the directory in which the socket is created.
237 connecting to a stream socket object requires write permission on that socket;
238 sending a datagram to a datagram socket likewise
239 requires write permission on that socket.
240 POSIX does not make any statement about the effect of the permissions
241 on a socket file, and on some systems (e.g., older BSDs),
242 the socket permissions are ignored.
243 Portable programs should not rely on
244 this feature for security.
246 When creating a new socket, the owner and group of the socket file
247 are set according to the usual rules.
248 The socket file has all permissions enabled,
249 other than those that are turned off by the process
252 The owner, group, and permissions of a pathname socket can be changed (using
256 .\" However, fchown() and fchmod() do not seem to have an effect
259 Socket permissions have no meaning for abstract sockets:
262 has no effect when binding an abstract socket,
263 and changing the ownership and permissions of the object (via
267 has no effect on the accessibility of the socket.
269 Abstract sockets automatically disappear when all open references
270 to the socket are closed.
272 The abstract socket namespace is a nonportable Linux extension.
275 For historical reasons, these socket options are specified with a
277 type even though they are
286 as the socket family.
289 Enabling this socket option causes receipt of the credentials of
290 the sending process in an
291 .B SCM_CREDENTIALS ancillary
292 message in each subsequently received message.
293 The returned credentials are those specified by the sender using
294 .BR SCM_CREDENTIALS ,
295 or a default that includes the sender's PID, real user ID, and real group ID,
296 if the sender did not specify
300 When this option is set and the socket is not yet connected,
301 a unique name in the abstract namespace will be generated automatically.
303 The value given as an argument to
305 and returned as the result of
307 is an integer boolean flag.
310 Enables receiving of the SELinux security label of the peer socket
311 in an ancillary message of type
315 The value given as an argument to
317 and returned as the result of
319 is an integer boolean flag.
323 option is supported for UNIX domain datagram sockets
324 .\" commit 877ce7c1b3afd69a9b1caeb1b9964c992641f52a
326 support for UNIX domain stream sockets was added
327 .\" commit 37a9a8df8ce9de6ea73349c9ac8bdf6ba4ec4f70
335 This read-only socket option returns the
336 credentials of the peer process connected to this socket.
337 The returned credentials are those that were in effect at the time
347 structure; define the
349 feature test macro to obtain the definition of that structure from
352 The use of this option is possible only for connected
354 stream sockets and for
356 stream and datagram socket pairs created using
360 This read-only socket option returns the
361 security context of the peer socket connected to this socket.
362 By default, this will be the same as the security context of
363 the process that created the peer socket unless overridden
364 by the policy or by a process with the required permissions.
368 is a pointer to a buffer of the specified length in bytes
369 into which the security context string will be copied.
370 If the buffer length is less than the length of the security
377 and returns the required length via
379 The caller should allocate at least
381 bytes for the buffer initially, although this is not guaranteed
383 Resizing the buffer to the returned length
384 and retrying may be necessary.
386 The security context string may include a terminating null character
387 in the returned length, but is not guaranteed to do so: a security
388 context "foo" might be represented as either {'f','o','o'} of length 3
389 or {'f','o','o','\\0'} of length 4, which are considered to be
391 The string is printable, does not contain non-terminating null characters,
392 and is in an unspecified encoding (in particular, it
393 is not guaranteed to be ASCII or UTF-8).
395 The use of this option for sockets in the
397 address family is supported since Linux 2.6.2 for connected stream sockets,
399 .\" commit 0b811db2cb2aabc910e53d34ebb95a15997c33e7
400 also for stream and datagram socket pairs created using
409 .IR sizeof(sa_family_t) ,
410 .\" i.e., sizeof(short)
413 socket option was specified for a socket that was
414 not explicitly bound to an address,
415 then the socket is autobound to an abstract address.
416 The address consists of a null byte
417 followed by 5 bytes in the character set
419 Thus, there is a limit of 2^20 autobind addresses.
420 (From Linux 2.1.15, when the autobind feature was added,
421 8 bytes were used, and the limit was thus 2^32 autobind addresses.
422 The change to 5 bytes came in Linux 2.3.15.)
424 The following paragraphs describe domain-specific details and
425 unsupported features of the sockets API for UNIX domain sockets on Linux.
427 UNIX domain sockets do not support the transmission of
428 out-of-band data (the
438 flag is not supported by UNIX domain sockets.
441 .\" commit 9f6f9af7694ede6314bed281eec74d588ba9474f
448 was not supported by UNIX domain sockets.
452 socket option does have an effect for UNIX domain sockets, but the
455 For datagram sockets, the
457 value imposes an upper limit on the size of outgoing datagrams.
458 This limit is calculated as the doubled (see
460 option value less 32 bytes used for overhead.
461 .SS Ancillary messages
462 Ancillary data is sent and received using
466 For historical reasons, the ancillary message types listed below
469 type even though they are
472 To send them, set the
481 For more information, see
485 Send or receive a set of open file descriptors from another process.
486 The data portion contains an integer array of the file descriptors.
488 Commonly, this operation is referred to as "passing a file descriptor"
490 However, more accurately,
491 what is being passed is a reference to an open file description (see
493 and in the receiving process it is likely that a different
494 file descriptor number will be used.
495 Semantically, this operation is equivalent to duplicating
497 a file descriptor into the file descriptor table of another process.
499 If the buffer used to receive the ancillary data containing
500 file descriptors is too small (or is absent),
501 then the ancillary data is truncated (or discarded)
502 and the excess file descriptors are automatically closed
503 in the receiving process.
505 If the number of file descriptors received in the ancillary data would
506 cause the process to exceed its
510 the excess file descriptors are automatically closed
511 in the receiving process.
515 defines a limit on the number of file descriptors in the array.
516 Attempting to send an array larger than this limit causes
518 to fail with the error
523 .\" commit bba14de98753cb6599a2dae0e520714b2153522d
527 Send or receive UNIX credentials.
528 This can be used for authentication.
529 The credentials are passed as a
532 This structure is defined in
539 pid_t pid; /* Process ID of the sending process */
540 uid_t uid; /* User ID of the sending process */
541 gid_t gid; /* Group ID of the sending process */
548 feature test macro must be defined (before including
550 header files) in order to obtain the definition
553 The credentials which the sender specifies are checked by the kernel.
554 A privileged process is allowed to specify values that do not match its own.
555 The sender must specify its own process ID (unless it has the capability
557 in which case the PID of any existing process may be specified),
558 its real user ID, effective user ID, or saved set-user-ID (unless it has
560 and its real group ID, effective group ID, or saved set-group-ID
568 option must be enabled on the socket.
571 Receive the SELinux security context (the security label)
573 The received ancillary data is a null-terminated string containing
574 the security context.
575 The receiver should allocate at least
577 bytes in the data portion of the ancillary message for this data.
579 To receive the security context, the
581 option must be enabled on the socket (see above).
583 When sending ancillary data with
585 only one item of each of the above types may be included in the sent message.
587 At least one byte of real data should be sent when sending ancillary data.
588 On Linux, this is required to successfully send ancillary data over
589 a UNIX domain stream socket.
590 When sending ancillary data over a UNIX domain datagram socket,
591 it is not necessary on Linux to send any accompanying real data.
592 However, portable applications should also include at least one byte
593 of real data when sending ancillary data over a datagram socket.
595 When receiving from a stream socket,
596 ancillary data forms a kind of barrier for the received data.
597 For example, suppose that the sender transmits as follows:
603 of four bytes, with no ancillary data.
606 of one byte, with ancillary data.
609 of four bytes, with no ancillary data.
613 Suppose that the receiver now performs
615 calls each with a buffer size of 20 bytes.
616 The first call will receive five bytes of data,
617 along with the ancillary data sent by the second
620 The next call will receive the remaining four bytes of data.
622 If the space allocated for receiving incoming ancillary data is too small
623 then the ancillary data is truncated to the number of headers
624 that will fit in the supplied buffer (or, in the case of an
626 file descriptor list, the list of file descriptors may be truncated).
627 If no buffer is provided for incoming ancillary data (i.e., the
631 structure supplied to
634 then the incoming ancillary data is discarded.
635 In both of these cases, the
637 flag will be set in the
645 calls return information in
647 The correct syntax is:
652 .IB error " = ioctl(" unix_socket ", " ioctl_type ", &" value ");"
662 sockets, this call returns the number of unread bytes in the receive buffer.
663 The socket must not be in LISTEN state, otherwise an error
668 .IR <linux/sockios.h> .
669 .\" FIXME . http://sources.redhat.com/bugzilla/show_bug.cgi?id=12002,
670 .\" filed 2010-09-10, may cause SIOCINQ to be defined in glibc headers
672 you can use the synonymous
676 .\" SIOCOUTQ also has an effect for UNIX domain sockets, but not
677 .\" quite what userland might expect. It seems to return the number
678 .\" of bytes allocated for buffers containing pending output.
679 .\" That number is normally larger than the number of bytes of pending
680 .\" output. Since this info is, from userland's point of view, imprecise,
681 .\" and it may well change, probably best not to document this now.
685 the returned value is the same as
686 for Internet domain datagram sockets;
692 The specified local address is already in use or the filesystem socket
693 object already exists.
696 This error can occur for
698 when sending a file descriptor as ancillary data over
699 a UNIX domain socket (see the description of
701 above), and indicates that the file descriptor number that
702 is being sent is not valid (e.g., it is not an open file descriptor).
705 The remote address specified by
707 was not a listening socket.
708 This error can also occur if the target pathname is not a socket.
711 Remote socket was unexpectedly closed.
714 User memory address was not valid.
717 Invalid argument passed.
718 A common cause is that the value
720 was not specified in the
722 field of passed addresses, or the socket was in an
723 invalid state for the applied operation.
727 called on an already connected socket or a target address was
728 specified on a connected socket.
731 The system-wide limit on the total number of open files has been reached.
734 The pathname in the remote address specified to
742 Socket operation needs a target address, but the socket is not connected.
745 Stream operation called on non-stream oriented socket or tried to
746 use the out-of-band data option.
749 The sender passed invalid credentials in the
753 Remote socket was closed on a stream socket.
757 This can be avoided by passing the
765 Passed protocol is not
769 Remote socket does not match the local socket type
778 While sending an ancillary message containing credentials
779 .RB ( SCM_CREDENTIALS ),
780 the caller specified a PID that does not match any existing process.
783 This error can occur for
785 when sending a file descriptor as ancillary data over
786 a UNIX domain socket (see the description of
789 It occurs if the number of "in-flight" file descriptors exceeds the
791 resource limit and the caller does not have the
794 An in-flight file descriptor is one that has been sent using
796 but has not yet been accepted in the recipient process using
799 This error is diagnosed since mainline Linux 4.5
800 (and in some earlier kernel versions where the fix has been backported).
801 .\" commit 712f4aad406bb1ed67f3f98d04c044191f0ff593
802 In earlier kernel versions,
803 it was possible to place an unlimited number of file descriptors in flight,
804 by sending each file descriptor with
806 and then closing the file descriptor so that it was not accounted against the
810 Other errors can be generated by the generic socket layer or
811 by the filesystem while generating a filesystem socket object.
812 See the appropriate manual pages for more information.
815 and the abstract namespace were introduced with Linux 2.2 and should not
816 be used in portable programs.
817 (Some BSD-derived systems also support credential passing,
818 but the implementation details differ.)
820 Binding to a socket with a filename creates a socket
821 in the filesystem that must be deleted by the caller when it is no
824 The usual UNIX close-behind semantics apply; the socket can be unlinked
825 at any time and will be finally removed from the filesystem when the last
826 reference to it is closed.
828 To pass file descriptors or credentials over a
831 send or receive at least one byte of nonancillary data in the same
837 UNIX domain stream sockets do not support the notion of out-of-band data.
840 When binding a socket to an address,
841 Linux is one of the implementations that appends a null terminator
842 if none is supplied in
844 In most cases this is unproblematic:
845 when the socket address is retrieved,
846 it will be one byte longer than that supplied when the socket was bound.
847 However, there is one case where confusing behavior can result:
848 if 108 non-null bytes are supplied when a socket is bound,
849 then the addition of the null terminator takes the length of
851 .IR sizeof(sun_path) .
852 Consequently, when retrieving the socket address
855 .\" The behavior on Solaris is quite similar.
858 argument for the retrieving call is specified as
859 .IR "sizeof(struct sockaddr_un)" ,
860 then the returned address structure
862 have a null terminator in
865 In addition, some implementations
866 .\" i.e., traditional BSD
867 don't require a null terminator when binding a socket (the
869 argument is used to determine the length of
871 and when the socket address is retrieved on these implementations,
872 there is no null terminator in
875 Applications that retrieve socket addresses can (portably) code
876 to handle the possibility that there is no null terminator in
878 by respecting the fact that the number of valid bytes in the pathname is:
882 strnlen(addr.sun_path, addrlen \- offsetof(sockaddr_un, sun_path))
885 .\" The following patch to amend kernel behavior was rejected:
886 .\" http://thread.gmane.org/gmane.linux.kernel.api/2437
887 .\" Subject: [patch] Fix handling of overlength pathname in AF_UNIX sun_path
889 .\" And there was a related discussion in the Austin list:
890 .\" http://thread.gmane.org/gmane.comp.standards.posix.austin.general/5735
891 .\" Subject: Having a sun_path with no null terminator
894 .\" FIXME . Track http://austingroupbugs.net/view.php?id=561
896 Alternatively, an application can retrieve
897 the socket address by allocating a buffer of size
898 .I "sizeof(struct sockaddr_un)+1"
899 that is zeroed out before the retrieval.
900 The retrieving call can specify
903 .IR "sizeof(struct sockaddr_un)" ,
904 and the extra zero byte ensures that there will be
905 a null terminator for the string returned in
912 addrlen = sizeof(struct sockaddr_un);
913 addrp = malloc(addrlen + 1);
916 memset(addrp, 0, addrlen + 1);
918 if (getsockname(sfd, (struct sockaddr *) addrp, &addrlen)) == \-1)
921 printf("sun_path = %s\en", ((struct sockaddr_un *) addrp)\->sun_path);
925 This sort of messiness can be avoided if it is guaranteed
926 that the applications that
928 pathname sockets follow the rules outlined above under
929 .IR "Pathname sockets" .
931 The following code demonstrates the use of sequenced-packet
932 sockets for local interprocess communication.
933 It consists of two programs.
934 The server program waits for a connection from the client program.
935 The client sends each of its command-line arguments in separate messages.
936 The server treats the incoming messages as integers and adds them up.
937 The client sends the command string "END".
938 The server sends back a message containing the sum of the client's integers.
939 The client prints the sum and exits.
940 The server waits for the next client to connect.
941 To stop the server, the client is called with the command-line argument "DOWN".
943 The following output was recorded while running the server in the background
944 and repeatedly executing the client.
945 Execution of the server program ends when it receives the "DOWN" command.
953 $ \fB./client 11 \-5\fP
955 $ \fB./client DOWN\fP
968 #define SOCKET_NAME "/tmp/9Lq7BNBnBycd6nxy.socket"
969 #define BUFFER_SIZE 12
978 #include <sys/socket.h>
981 #include "connection.h"
984 main(int argc, char *argv[])
986 struct sockaddr_un name;
989 int connection_socket;
992 char buffer[BUFFER_SIZE];
994 /* Create local socket. */
996 connection_socket = socket(AF_UNIX, SOCK_SEQPACKET, 0);
997 if (connection_socket == \-1) {
1003 * For portability clear the whole structure, since some
1004 * implementations have additional (nonstandard) fields in
1008 memset(&name, 0, sizeof(name));
1010 /* Bind socket to socket name. */
1012 name.sun_family = AF_UNIX;
1013 strncpy(name.sun_path, SOCKET_NAME, sizeof(name.sun_path) \- 1);
1015 ret = bind(connection_socket, (const struct sockaddr *) &name,
1023 * Prepare for accepting connections. The backlog size is set
1024 * to 20. So while one request is being processed other requests
1028 ret = listen(connection_socket, 20);
1034 /* This is the main loop for handling connections. */
1038 /* Wait for incoming connection. */
1040 data_socket = accept(connection_socket, NULL, NULL);
1041 if (data_socket == \-1) {
1049 /* Wait for next data packet. */
1051 ret = read(data_socket, buffer, sizeof(buffer));
1057 /* Ensure buffer is 0\-terminated. */
1059 buffer[sizeof(buffer) \- 1] = 0;
1061 /* Handle commands. */
1063 if (!strncmp(buffer, "DOWN", sizeof(buffer))) {
1068 if (!strncmp(buffer, "END", sizeof(buffer))) {
1072 /* Add received summand. */
1074 result += atoi(buffer);
1079 sprintf(buffer, "%d", result);
1080 ret = write(data_socket, buffer, sizeof(buffer));
1090 /* Quit on DOWN command. */
1097 close(connection_socket);
1099 /* Unlink the socket. */
1101 unlink(SOCKET_NAME);
1114 #include <sys/socket.h>
1117 #include "connection.h"
1120 main(int argc, char *argv[])
1122 struct sockaddr_un addr;
1125 char buffer[BUFFER_SIZE];
1127 /* Create local socket. */
1129 data_socket = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1130 if (data_socket == \-1) {
1136 * For portability clear the whole structure, since some
1137 * implementations have additional (nonstandard) fields in
1141 memset(&addr, 0, sizeof(addr));
1143 /* Connect socket to socket address. */
1145 addr.sun_family = AF_UNIX;
1146 strncpy(addr.sun_path, SOCKET_NAME, sizeof(addr.sun_path) \- 1);
1148 ret = connect(data_socket, (const struct sockaddr *) &addr,
1151 fprintf(stderr, "The server is down.\en");
1155 /* Send arguments. */
1157 for (int i = 1; i < argc; ++i) {
1158 ret = write(data_socket, argv[i], strlen(argv[i]) + 1);
1165 /* Request result. */
1167 strcpy(buffer, "END");
1168 ret = write(data_socket, buffer, strlen(buffer) + 1);
1174 /* Receive result. */
1176 ret = read(data_socket, buffer, sizeof(buffer));
1182 /* Ensure buffer is 0\-terminated. */
1184 buffer[sizeof(buffer) \- 1] = 0;
1186 printf("Result = %s\en", buffer);
1196 For examples of the use of
1201 .BR seccomp_unotify (2).
1208 .BR capabilities (7),
1209 .BR credentials (7),