1 .\" Copyright (C) 2005, 2008, Michael Kerrisk <mtk.manpages@gmail.com>
2 .\" (A few fragments remain from an earlier (1992) version by
3 .\" Drew Eckhardt <drew@cs.colorado.edu>.)
5 .\" SPDX-License-Identifier: Linux-man-pages-copyleft
7 .\" Modified by Michael Haardt <michael@moria.de>
8 .\" Modified 1993-07-23 by Rik Faith <faith@cs.unc.edu>
9 .\" Modified 1996-10-22 by Eric S. Raymond <esr@thyrsus.com>
10 .\" Modified 2004-06-17 by Michael Kerrisk <mtk.manpages@gmail.com>
11 .\" Modified 2005, mtk: added an example program
12 .\" Modified 2008-01-09, mtk: rewrote DESCRIPTION; minor additions
14 .\" 2008-10-10, mtk: add description of pipe2()
16 .TH PIPE 2 2021-03-22 "Linux" "Linux Programmer's Manual"
18 pipe, pipe2 \- create pipe
21 .RI ( libc ", " \-lc )
24 .B #include <unistd.h>
26 .BI "int pipe(int " pipefd [2]);
28 .BR "#define _GNU_SOURCE" " /* See feature_test_macros(7) */"
29 .BR "#include <fcntl.h>" " /* Definition of " O_* " constants */"
30 .B #include <unistd.h>
32 .BI "int pipe2(int " pipefd "[2], int " flags );
34 /* On Alpha, IA-64, MIPS, SuperH, and SPARC/SPARC64, pipe() has the
35 following prototype; see NOTES */
37 .B #include <unistd.h>
42 .B struct fd_pair pipe(void);
46 creates a pipe, a unidirectional data channel that
47 can be used for interprocess communication.
50 is used to return two file descriptors referring to the ends of the pipe.
52 refers to the read end of the pipe.
54 refers to the write end of the pipe.
55 Data written to the write end of the pipe is buffered by the kernel
56 until it is read from the read end of the pipe.
57 For further details, see
66 The following values can be bitwise ORed in
68 to obtain different behavior:
73 flag on the two new file descriptors.
74 See the description of the same flag in
76 for reasons why this may be useful.
78 .BR O_DIRECT " (since Linux 3.4)"
79 .\" commit 9883035ae7edef3ec62ad215611cb8e17d6a1a5d
80 Create a pipe that performs I/O in "packet" mode.
83 to the pipe is dealt with as a separate packet, and
85 from the pipe will read one packet at a time.
86 Note the following points:
89 Writes of greater than
93 will be split into multiple packets.
101 specifies a buffer size that is smaller than the next packet,
102 then the requested number of bytes are read,
103 and the excess bytes in the packet are discarded.
104 Specifying a buffer size of
106 will be sufficient to read the largest possible packets
107 (see the previous point).
109 Zero-length packets are not supported.
112 that specifies a buffer size of zero is a no-op, and returns 0.)
115 Older kernels that do not support this flag will indicate this via an
120 .\" commit 0dbf5f20652108106cb822ad7662c786baaa03ff
121 .\" FIXME . But, it is not possible to specify O_DIRECT when opening a FIFO
122 it is possible to change the
124 setting of a pipe file descriptor using
130 file status flag on the open file descriptions
131 referred to by the new file descriptors.
132 Using this flag saves extra calls to
134 to achieve the same result.
136 .B O_NOTIFICATION_PIPE
138 .\" commit c73be61cede5882f9605a852414db559c0ebedfd
139 general notification mechanism is built on the top of the pipe where
140 kernel splices notification messages into pipes opened by user space.
141 The owner of the pipe has to tell the kernel which sources of events to watch
142 and filters can also be applied to select
143 which subevents should be placed into the pipe.
145 On success, zero is returned.
146 On error, \-1 is returned,
148 is set to indicate the error, and
152 On Linux (and other systems),
157 A requirement standardizing this behavior was added in POSIX.1-2008 TC2.
158 .\" http://austingroupbugs.net/view.php?id=467
162 likewise does not modify
177 The per-process limit on the number of open file descriptors has been reached.
180 The system-wide limit on the total number of open files has been reached.
183 The user hard limit on memory that can be allocated for pipes
184 has been reached and the caller is not privileged; see
189 .B O_NOTIFICATION_PIPE
192 and support for notifications
193 .RB ( CONFIG_WATCH_QUEUE )
194 is not compiled into the kernel.
197 was added to Linux in version 2.6.27;
198 glibc support is available starting with
202 POSIX.1-2001, POSIX.1-2008.
207 .\" See http://math-atlas.sourceforge.net/devel/assembly/64.psabi.1.33.ps.Z
208 .\" for example, section 3.2.1 "Registers and the Stack Frame".
209 The System V ABI on some architectures allows the use of more than one register
210 for returning multiple values; several architectures
211 (namely, Alpha, IA-64, MIPS, SuperH, and SPARC/SPARC64)
212 (ab)use this feature in order to implement the
214 system call in a functional manner:
215 the call doesn't take any arguments and returns
216 a pair of file descriptors as the return value on success.
219 wrapper function transparently deals with this.
222 for information regarding registers used for storing second file descriptor.
224 .\" fork.2 refers to this example program.
225 The following program creates a pipe, and then
227 to create a child process;
228 the child inherits a duplicate set of file
229 descriptors that refer to the same pipe.
232 each process closes the file descriptors that it doesn't need for the pipe
235 The parent then writes the string contained in the program's
236 command-line argument to the pipe,
237 and the child reads this string a byte at a time from the pipe
238 and echoes it on standard output.
240 .\" SRC BEGIN (pipe.c)
245 #include <sys/wait.h>
249 main(int argc, char *argv[])
256 fprintf(stderr, "Usage: %s <string>\en", argv[0]);
260 if (pipe(pipefd) == \-1) {
271 if (cpid == 0) { /* Child reads from pipe */
272 close(pipefd[1]); /* Close unused write end */
274 while (read(pipefd[0], &buf, 1) > 0)
275 write(STDOUT_FILENO, &buf, 1);
277 write(STDOUT_FILENO, "\en", 1);
281 } else { /* Parent writes argv[1] to pipe */
282 close(pipefd[0]); /* Close unused read end */
283 write(pipefd[1], argv[1], strlen(argv[1]));
284 close(pipefd[1]); /* Reader will see EOF */
285 wait(NULL); /* Wait for child */