1SELECT(2) Linux Programmer's Manual SELECT(2)
2
6 select, pselect, FD_CLR, FD_ISSET, FD_SET, FD_ZERO - synchronous I/O
7 multiplexing
8
10 #include <sys/select.h>
11 int select(int nfds, fd_set *readfds, fd_set *writefds,
13 fd_set *exceptfds, struct timeval *timeout);
14 void FD_CLR(int fd, fd_set *set);
16 int FD_ISSET(int fd, fd_set *set);
17 void FD_SET(int fd, fd_set *set);
18 void FD_ZERO(fd_set *set);
19 int pselect(int nfds, fd_set *readfds, fd_set *writefds,
21 fd_set *exceptfds, const struct timespec *timeout,
22 const sigset_t *sigmask);
23 Feature Test Macro Requirements for glibc (see feature_test_macros(7)):
25 pselect(): _POSIX_C_SOURCE >= 200112L
27
29 select() allows a program to monitor multiple file descriptors, waiting
30 until one or more of the file descriptors become "ready" for some class
31 of I/O operation (e.g., input possible). A file descriptor is consid‐
32 ered ready if it is possible to perform a corresponding I/O operation
33 (e.g., read(2), or a sufficiently small write(2)) without blocking.
34 select() can monitor only file descriptors numbers that are less than
36 FD_SETSIZE; poll(2) and epoll(7) do not have this limitation. See
37 BUGS.
38 File descriptor sets
40 The principal arguments of select() are three "sets" of file descrip‐
41 tors (declared with the type fd_set), which allow the caller to wait
42 for three classes of events on the specified set of file descriptors.
43 Each of the fd_set arguments may be specified as NULL if no file de‐
44 scriptors are to be watched for the corresponding class of events.
45 Note well: Upon return, each of the file descriptor sets is modified in
47 place to indicate which file descriptors are currently "ready". Thus,
48 if using select() within a loop, the sets must be reinitialized before
49 each call. The implementation of the fd_set arguments as value-result
50 arguments is a design error that is avoided in poll(2) and epoll(7).
51 The contents of a file descriptor set can be manipulated using the fol‐
53 lowing macros:
54 FD_ZERO()
56 This macro clears (removes all file descriptors from) set. It
57 should be employed as the first step in initializing a file de‐
58 scriptor set.
59 FD_SET()
61 This macro adds the file descriptor fd to set. Adding a file
62 descriptor that is already present in the set is a no-op, and
63 does not produce an error.
64 FD_CLR()
66 This macro removes the file descriptor fd from set. Removing a
67 file descriptor that is not present in the set is a no-op, and
68 does not produce an error.
69 FD_ISSET()
71 select() modifies the contents of the sets according to the
72 rules described below. After calling select(), the FD_ISSET()
73 macro can be used to test if a file descriptor is still present
74 in a set. FD_ISSET() returns nonzero if the file descriptor fd
75 is present in set, and zero if it is not.
76 Arguments
78 The arguments of select() are as follows:
79 readfds
81 The file descriptors in this set are watched to see if they are
82 ready for reading. A file descriptor is ready for reading if a
83 read operation will not block; in particular, a file descriptor
84 is also ready on end-of-file.
85 After select() has returned, readfds will be cleared of all file
87 descriptors except for those that are ready for reading.
88 writefds
90 The file descriptors in this set are watched to see if they are
91 ready for writing. A file descriptor is ready for writing if a
92 write operation will not block. However, even if a file de‐
93 scriptor indicates as writable, a large write may still block.
94 After select() has returned, writefds will be cleared of all
96 file descriptors except for those that are ready for writing.
97 exceptfds
99 The file descriptors in this set are watched for "exceptional
100 conditions". For examples of some exceptional conditions, see
101 the discussion of POLLPRI in poll(2).
102 After select() has returned, exceptfds will be cleared of all
104 file descriptors except for those for which an exceptional con‐
105 dition has occurred.
106 nfds This argument should be set to the highest-numbered file de‐
108 scriptor in any of the three sets, plus 1. The indicated file
109 descriptors in each set are checked, up to this limit (but see
110 BUGS).
111 timeout
113 The timeout argument is a timeval structure (shown below) that
114 specifies the interval that select() should block waiting for a
115 file descriptor to become ready. The call will block until ei‐
116 ther:
117 • a file descriptor becomes ready;
119 • the call is interrupted by a signal handler; or
121 • the timeout expires.
123 Note that the timeout interval will be rounded up to the system
125 clock granularity, and kernel scheduling delays mean that the
126 blocking interval may overrun by a small amount.
127 If both fields of the timeval structure are zero, then select()
129 returns immediately. (This is useful for polling.)
130 If timeout is specified as NULL, select() blocks indefinitely
132 waiting for a file descriptor to become ready.
133 pselect()
135 The pselect() system call allows an application to safely wait until
136 either a file descriptor becomes ready or until a signal is caught.
137 The operation of select() and pselect() is identical, other than these
139 three differences:
140 • select() uses a timeout that is a struct timeval (with seconds and
142 microseconds), while pselect() uses a struct timespec (with seconds
143 and nanoseconds).
144 • select() may update the timeout argument to indicate how much time
146 was left. pselect() does not change this argument.
147 • select() has no sigmask argument, and behaves as pselect() called
149 with NULL sigmask.
150 sigmask is a pointer to a signal mask (see sigprocmask(2)); if it is
152 not NULL, then pselect() first replaces the current signal mask by the
153 one pointed to by sigmask, then does the "select" function, and then
154 restores the original signal mask. (If sigmask is NULL, the signal
155 mask is not modified during the pselect() call.)
156 Other than the difference in the precision of the timeout argument, the
158 following pselect() call:
159 ready = pselect(nfds, &readfds, &writefds, &exceptfds,
161 timeout, &sigmask);
162 is equivalent to atomically executing the following calls:
164 sigset_t origmask;
166 pthread_sigmask(SIG_SETMASK, &sigmask, &origmask);
168 ready = select(nfds, &readfds, &writefds, &exceptfds, timeout);
169 pthread_sigmask(SIG_SETMASK, &origmask, NULL);
170 The reason that pselect() is needed is that if one wants to wait for
172 either a signal or for a file descriptor to become ready, then an
173 atomic test is needed to prevent race conditions. (Suppose the signal
174 handler sets a global flag and returns. Then a test of this global
175 flag followed by a call of select() could hang indefinitely if the sig‐
176 nal arrived just after the test but just before the call. By contrast,
177 pselect() allows one to first block signals, handle the signals that
178 have come in, then call pselect() with the desired sigmask, avoiding
179 the race.)
180 The timeout
182 The timeout argument for select() is a structure of the following type:
183 struct timeval {
185 time_t tv_sec; /* seconds */
186 suseconds_t tv_usec; /* microseconds */
187 };
188 The corresponding argument for pselect() has the following type:
190 struct timespec {
192 time_t tv_sec; /* seconds */
193 long tv_nsec; /* nanoseconds */
194 };
195 On Linux, select() modifies timeout to reflect the amount of time not
197 slept; most other implementations do not do this. (POSIX.1 permits ei‐
198 ther behavior.) This causes problems both when Linux code which reads
199 timeout is ported to other operating systems, and when code is ported
200 to Linux that reuses a struct timeval for multiple select()s in a loop
201 without reinitializing it. Consider timeout to be undefined after se‐
202 lect() returns.
203
205 On success, select() and pselect() return the number of file descrip‐
206 tors contained in the three returned descriptor sets (that is, the to‐
207 tal number of bits that are set in readfds, writefds, exceptfds). The
208 return value may be zero if the timeout expired before any file de‐
209 scriptors became ready.
210 On error, -1 is returned, and errno is set to indicate the error; the
212 file descriptor sets are unmodified, and timeout becomes undefined.
213
215 EBADF An invalid file descriptor was given in one of the sets. (Per‐
216 haps a file descriptor that was already closed, or one on which
217 an error has occurred.) However, see BUGS.
218 EINTR A signal was caught; see signal(7).
220 EINVAL nfds is negative or exceeds the RLIMIT_NOFILE resource limit
222 (see getrlimit(2)).
223 EINVAL The value contained within timeout is invalid.
225 ENOMEM Unable to allocate memory for internal tables.
227
229 pselect() was added to Linux in kernel 2.6.16. Prior to this, pse‐
230 lect() was emulated in glibc (but see BUGS).
231
233 select() conforms to POSIX.1-2001, POSIX.1-2008, and 4.4BSD (select()
234 first appeared in 4.2BSD). Generally portable to/from non-BSD systems
235 supporting clones of the BSD socket layer (including System V vari‐
236 ants). However, note that the System V variant typically sets the
237 timeout variable before returning, but the BSD variant does not.
238 pselect() is defined in POSIX.1g, and in POSIX.1-2001 and POSIX.1-2008.
240
242 An fd_set is a fixed size buffer. Executing FD_CLR() or FD_SET() with
243 a value of fd that is negative or is equal to or larger than FD_SETSIZE
244 will result in undefined behavior. Moreover, POSIX requires fd to be a
245 valid file descriptor.
246 The operation of select() and pselect() is not affected by the O_NON‐
248 BLOCK flag.
249 On some other UNIX systems, select() can fail with the error EAGAIN if
251 the system fails to allocate kernel-internal resources, rather than
252 ENOMEM as Linux does. POSIX specifies this error for poll(2), but not
253 for select(). Portable programs may wish to check for EAGAIN and loop,
254 just as with EINTR.
255 The self-pipe trick
257 On systems that lack pselect(), reliable (and more portable) signal
258 trapping can be achieved using the self-pipe trick. In this technique,
259 a signal handler writes a byte to a pipe whose other end is monitored
260 by select() in the main program. (To avoid possibly blocking when
261 writing to a pipe that may be full or reading from a pipe that may be
262 empty, nonblocking I/O is used when reading from and writing to the
263 pipe.)
264 Emulating usleep(3)
266 Before the advent of usleep(3), some code employed a call to select()
267 with all three sets empty, nfds zero, and a non-NULL timeout as a
268 fairly portable way to sleep with subsecond precision.
269 Correspondence between select() and poll() notifications
271 Within the Linux kernel source, we find the following definitions which
272 show the correspondence between the readable, writable, and exceptional
273 condition notifications of select() and the event notifications pro‐
274 vided by poll(2) and epoll(7):
275 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN |
277 EPOLLHUP | EPOLLERR)
278 /* Ready for reading */
279 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT |
280 EPOLLERR)
281 /* Ready for writing */
282 #define POLLEX_SET (EPOLLPRI)
283 /* Exceptional condition */
284 Multithreaded applications
286 If a file descriptor being monitored by select() is closed in another
287 thread, the result is unspecified. On some UNIX systems, select() un‐
288 blocks and returns, with an indication that the file descriptor is
289 ready (a subsequent I/O operation will likely fail with an error, un‐
290 less another process reopens file descriptor between the time select()
291 returned and the I/O operation is performed). On Linux (and some other
292 systems), closing the file descriptor in another thread has no effect
293 on select(). In summary, any application that relies on a particular
294 behavior in this scenario must be considered buggy.
295 C library/kernel differences
297 The Linux kernel allows file descriptor sets of arbitrary size, deter‐
298 mining the length of the sets to be checked from the value of nfds.
299 However, in the glibc implementation, the fd_set type is fixed in size.
300 See also BUGS.
301 The pselect() interface described in this page is implemented by glibc.
303 The underlying Linux system call is named pselect6(). This system call
304 has somewhat different behavior from the glibc wrapper function.
305 The Linux pselect6() system call modifies its timeout argument. How‐
307 ever, the glibc wrapper function hides this behavior by using a local
308 variable for the timeout argument that is passed to the system call.
309 Thus, the glibc pselect() function does not modify its timeout argu‐
310 ment; this is the behavior required by POSIX.1-2001.
311 The final argument of the pselect6() system call is not a sigset_t *
313 pointer, but is instead a structure of the form:
314 struct {
316 const kernel_sigset_t *ss; /* Pointer to signal set */
317 size_t ss_len; /* Size (in bytes) of object
318 pointed to by 'ss' */
319 };
320 This allows the system call to obtain both a pointer to the signal set
322 and its size, while allowing for the fact that most architectures sup‐
323 port a maximum of 6 arguments to a system call. See sigprocmask(2) for
324 a discussion of the difference between the kernel and libc notion of
325 the signal set.
326 Historical glibc details
328 Glibc 2.0 provided an incorrect version of pselect() that did not take
329 a sigmask argument.
330 In glibc versions 2.1 to 2.2.1, one must define _GNU_SOURCE in order to
332 obtain the declaration of pselect() from <sys/select.h>.
333
335 POSIX allows an implementation to define an upper limit, advertised via
336 the constant FD_SETSIZE, on the range of file descriptors that can be
337 specified in a file descriptor set. The Linux kernel imposes no fixed
338 limit, but the glibc implementation makes fd_set a fixed-size type,
339 with FD_SETSIZE defined as 1024, and the FD_*() macros operating ac‐
340 cording to that limit. To monitor file descriptors greater than 1023,
341 use poll(2) or epoll(7) instead.
342 According to POSIX, select() should check all specified file descrip‐
344 tors in the three file descriptor sets, up to the limit nfds-1. How‐
345 ever, the current implementation ignores any file descriptor in these
346 sets that is greater than the maximum file descriptor number that the
347 process currently has open. According to POSIX, any such file descrip‐
348 tor that is specified in one of the sets should result in the error
349 EBADF.
350 Starting with version 2.1, glibc provided an emulation of pselect()
352 that was implemented using sigprocmask(2) and select(). This implemen‐
353 tation remained vulnerable to the very race condition that pselect()
354 was designed to prevent. Modern versions of glibc use the (race-free)
355 pselect() system call on kernels where it is provided.
356 On Linux, select() may report a socket file descriptor as "ready for
358 reading", while nevertheless a subsequent read blocks. This could for
359 example happen when data has arrived but upon examination has the wrong
360 checksum and is discarded. There may be other circumstances in which a
361 file descriptor is spuriously reported as ready. Thus it may be safer
362 to use O_NONBLOCK on sockets that should not block.
363 On Linux, select() also modifies timeout if the call is interrupted by
365 a signal handler (i.e., the EINTR error return). This is not permitted
366 by POSIX.1. The Linux pselect() system call has the same behavior, but
367 the glibc wrapper hides this behavior by internally copying the timeout
368 to a local variable and passing that variable to the system call.
369
371 #include <stdio.h>
372 #include <stdlib.h>
373 #include <sys/select.h>
374 int
376 main(void)
377 {
378 fd_set rfds;
379 struct timeval tv;
380 int retval;
381 /* Watch stdin (fd 0) to see when it has input. */
383 FD_ZERO(&rfds);
385 FD_SET(0, &rfds);
386 /* Wait up to five seconds. */
388 tv.tv_sec = 5;
390 tv.tv_usec = 0;
391 retval = select(1, &rfds, NULL, NULL, &tv);
393 /* Don't rely on the value of tv now! */
394 if (retval == -1)
396 perror("select()");
397 else if (retval)
398 printf("Data is available now.\n");
399 /* FD_ISSET(0, &rfds) will be true. */
400 else
401 printf("No data within five seconds.\n");
402 exit(EXIT_SUCCESS);
404 }
405
407 accept(2), connect(2), poll(2), read(2), recv(2), restart_syscall(2),
408 send(2), sigprocmask(2), write(2), epoll(7), time(7)
409 For a tutorial with discussion and examples, see select_tut(2).
411
413 This page is part of release 5.10 of the Linux man-pages project. A
414 description of the project, information about reporting bugs, and the
415 latest version of this page, can be found at
416 https://www.kernel.org/doc/man-pages/.
417Linux 2020-11-01 SELECT(2)