1SELECT(2) Linux Programmer's Manual SELECT(2)
2
3
4
6 select, pselect, FD_CLR, FD_ISSET, FD_SET, FD_ZERO - synchronous I/O
7 multiplexing
8
10 #include <sys/select.h>
11
12 int select(int nfds, fd_set *restrict readfds,
13 fd_set *restrict writefds, fd_set *restrict exceptfds,
14 struct timeval *restrict timeout);
15
16 void FD_CLR(int fd, fd_set *set);
17 int FD_ISSET(int fd, fd_set *set);
18 void FD_SET(int fd, fd_set *set);
19 void FD_ZERO(fd_set *set);
20
21 int pselect(int nfds, fd_set *restrict readfds,
22 fd_set *restrict writefds, fd_set *restrict exceptfds,
23 const struct timespec *restrict timeout,
24 const sigset_t *restrict sigmask);
25
26 Feature Test Macro Requirements for glibc (see feature_test_macros(7)):
27
28 pselect():
29 _POSIX_C_SOURCE >= 200112L
30
32 WARNING: select() can monitor only file descriptors numbers that are
33 less than FD_SETSIZE (1024)—an unreasonably low limit for many modern
34 applications—and this limitation will not change. All modern applica‐
35 tions should instead use poll(2) or epoll(7), which do not suffer this
36 limitation.
37
38 select() allows a program to monitor multiple file descriptors, waiting
39 until one or more of the file descriptors become "ready" for some class
40 of I/O operation (e.g., input possible). A file descriptor is consid‐
41 ered ready if it is possible to perform a corresponding I/O operation
42 (e.g., read(2), or a sufficiently small write(2)) without blocking.
43
44 File descriptor sets
45 The principal arguments of select() are three "sets" of file descrip‐
46 tors (declared with the type fd_set), which allow the caller to wait
47 for three classes of events on the specified set of file descriptors.
48 Each of the fd_set arguments may be specified as NULL if no file de‐
49 scriptors are to be watched for the corresponding class of events.
50
51 Note well: Upon return, each of the file descriptor sets is modified in
52 place to indicate which file descriptors are currently "ready". Thus,
53 if using select() within a loop, the sets must be reinitialized before
54 each call.
55
56 The contents of a file descriptor set can be manipulated using the fol‐
57 lowing macros:
58
59 FD_ZERO()
60 This macro clears (removes all file descriptors from) set. It
61 should be employed as the first step in initializing a file de‐
62 scriptor set.
63
64 FD_SET()
65 This macro adds the file descriptor fd to set. Adding a file
66 descriptor that is already present in the set is a no-op, and
67 does not produce an error.
68
69 FD_CLR()
70 This macro removes the file descriptor fd from set. Removing a
71 file descriptor that is not present in the set is a no-op, and
72 does not produce an error.
73
74 FD_ISSET()
75 select() modifies the contents of the sets according to the
76 rules described below. After calling select(), the FD_ISSET()
77 macro can be used to test if a file descriptor is still present
78 in a set. FD_ISSET() returns nonzero if the file descriptor fd
79 is present in set, and zero if it is not.
80
81 Arguments
82 The arguments of select() are as follows:
83
84 readfds
85 The file descriptors in this set are watched to see if they are
86 ready for reading. A file descriptor is ready for reading if a
87 read operation will not block; in particular, a file descriptor
88 is also ready on end-of-file.
89
90 After select() has returned, readfds will be cleared of all file
91 descriptors except for those that are ready for reading.
92
93 writefds
94 The file descriptors in this set are watched to see if they are
95 ready for writing. A file descriptor is ready for writing if a
96 write operation will not block. However, even if a file de‐
97 scriptor indicates as writable, a large write may still block.
98
99 After select() has returned, writefds will be cleared of all
100 file descriptors except for those that are ready for writing.
101
102 exceptfds
103 The file descriptors in this set are watched for "exceptional
104 conditions". For examples of some exceptional conditions, see
105 the discussion of POLLPRI in poll(2).
106
107 After select() has returned, exceptfds will be cleared of all
108 file descriptors except for those for which an exceptional con‐
109 dition has occurred.
110
111 nfds This argument should be set to the highest-numbered file de‐
112 scriptor in any of the three sets, plus 1. The indicated file
113 descriptors in each set are checked, up to this limit (but see
114 BUGS).
115
116 timeout
117 The timeout argument is a timeval structure (shown below) that
118 specifies the interval that select() should block waiting for a
119 file descriptor to become ready. The call will block until ei‐
120 ther:
121
122 • a file descriptor becomes ready;
123
124 • the call is interrupted by a signal handler; or
125
126 • the timeout expires.
127
128 Note that the timeout interval will be rounded up to the system
129 clock granularity, and kernel scheduling delays mean that the
130 blocking interval may overrun by a small amount.
131
132 If both fields of the timeval structure are zero, then select()
133 returns immediately. (This is useful for polling.)
134
135 If timeout is specified as NULL, select() blocks indefinitely
136 waiting for a file descriptor to become ready.
137
138 pselect()
139 The pselect() system call allows an application to safely wait until
140 either a file descriptor becomes ready or until a signal is caught.
141
142 The operation of select() and pselect() is identical, other than these
143 three differences:
144
145 • select() uses a timeout that is a struct timeval (with seconds and
146 microseconds), while pselect() uses a struct timespec (with seconds
147 and nanoseconds).
148
149 • select() may update the timeout argument to indicate how much time
150 was left. pselect() does not change this argument.
151
152 • select() has no sigmask argument, and behaves as pselect() called
153 with NULL sigmask.
154
155 sigmask is a pointer to a signal mask (see sigprocmask(2)); if it is
156 not NULL, then pselect() first replaces the current signal mask by the
157 one pointed to by sigmask, then does the "select" function, and then
158 restores the original signal mask. (If sigmask is NULL, the signal
159 mask is not modified during the pselect() call.)
160
161 Other than the difference in the precision of the timeout argument, the
162 following pselect() call:
163
164 ready = pselect(nfds, &readfds, &writefds, &exceptfds,
165 timeout, &sigmask);
166
167 is equivalent to atomically executing the following calls:
168
169 sigset_t origmask;
170
171 pthread_sigmask(SIG_SETMASK, &sigmask, &origmask);
172 ready = select(nfds, &readfds, &writefds, &exceptfds, timeout);
173 pthread_sigmask(SIG_SETMASK, &origmask, NULL);
174
175 The reason that pselect() is needed is that if one wants to wait for
176 either a signal or for a file descriptor to become ready, then an
177 atomic test is needed to prevent race conditions. (Suppose the signal
178 handler sets a global flag and returns. Then a test of this global
179 flag followed by a call of select() could hang indefinitely if the sig‐
180 nal arrived just after the test but just before the call. By contrast,
181 pselect() allows one to first block signals, handle the signals that
182 have come in, then call pselect() with the desired sigmask, avoiding
183 the race.)
184
185 The timeout
186 The timeout argument for select() is a structure of the following type:
187
188 struct timeval {
189 time_t tv_sec; /* seconds */
190 suseconds_t tv_usec; /* microseconds */
191 };
192
193 The corresponding argument for pselect() has the following type:
194
195 struct timespec {
196 time_t tv_sec; /* seconds */
197 long tv_nsec; /* nanoseconds */
198 };
199
200 On Linux, select() modifies timeout to reflect the amount of time not
201 slept; most other implementations do not do this. (POSIX.1 permits ei‐
202 ther behavior.) This causes problems both when Linux code which reads
203 timeout is ported to other operating systems, and when code is ported
204 to Linux that reuses a struct timeval for multiple select()s in a loop
205 without reinitializing it. Consider timeout to be undefined after se‐
206 lect() returns.
207
209 On success, select() and pselect() return the number of file descrip‐
210 tors contained in the three returned descriptor sets (that is, the to‐
211 tal number of bits that are set in readfds, writefds, exceptfds). The
212 return value may be zero if the timeout expired before any file de‐
213 scriptors became ready.
214
215 On error, -1 is returned, and errno is set to indicate the error; the
216 file descriptor sets are unmodified, and timeout becomes undefined.
217
219 EBADF An invalid file descriptor was given in one of the sets. (Per‐
220 haps a file descriptor that was already closed, or one on which
221 an error has occurred.) However, see BUGS.
222
223 EINTR A signal was caught; see signal(7).
224
225 EINVAL nfds is negative or exceeds the RLIMIT_NOFILE resource limit
226 (see getrlimit(2)).
227
228 EINVAL The value contained within timeout is invalid.
229
230 ENOMEM Unable to allocate memory for internal tables.
231
233 pselect() was added to Linux in kernel 2.6.16. Prior to this, pse‐
234 lect() was emulated in glibc (but see BUGS).
235
237 select() conforms to POSIX.1-2001, POSIX.1-2008, and 4.4BSD (select()
238 first appeared in 4.2BSD). Generally portable to/from non-BSD systems
239 supporting clones of the BSD socket layer (including System V vari‐
240 ants). However, note that the System V variant typically sets the
241 timeout variable before returning, but the BSD variant does not.
242
243 pselect() is defined in POSIX.1g, and in POSIX.1-2001 and POSIX.1-2008.
244
246 An fd_set is a fixed size buffer. Executing FD_CLR() or FD_SET() with
247 a value of fd that is negative or is equal to or larger than FD_SETSIZE
248 will result in undefined behavior. Moreover, POSIX requires fd to be a
249 valid file descriptor.
250
251 The operation of select() and pselect() is not affected by the O_NON‐
252 BLOCK flag.
253
254 On some other UNIX systems, select() can fail with the error EAGAIN if
255 the system fails to allocate kernel-internal resources, rather than
256 ENOMEM as Linux does. POSIX specifies this error for poll(2), but not
257 for select(). Portable programs may wish to check for EAGAIN and loop,
258 just as with EINTR.
259
260 The self-pipe trick
261 On systems that lack pselect(), reliable (and more portable) signal
262 trapping can be achieved using the self-pipe trick. In this technique,
263 a signal handler writes a byte to a pipe whose other end is monitored
264 by select() in the main program. (To avoid possibly blocking when
265 writing to a pipe that may be full or reading from a pipe that may be
266 empty, nonblocking I/O is used when reading from and writing to the
267 pipe.)
268
269 Emulating usleep(3)
270 Before the advent of usleep(3), some code employed a call to select()
271 with all three sets empty, nfds zero, and a non-NULL timeout as a
272 fairly portable way to sleep with subsecond precision.
273
274 Correspondence between select() and poll() notifications
275 Within the Linux kernel source, we find the following definitions which
276 show the correspondence between the readable, writable, and exceptional
277 condition notifications of select() and the event notifications pro‐
278 vided by poll(2) and epoll(7):
279
280 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN |
281 EPOLLHUP | EPOLLERR)
282 /* Ready for reading */
283 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT |
284 EPOLLERR)
285 /* Ready for writing */
286 #define POLLEX_SET (EPOLLPRI)
287 /* Exceptional condition */
288
289 Multithreaded applications
290 If a file descriptor being monitored by select() is closed in another
291 thread, the result is unspecified. On some UNIX systems, select() un‐
292 blocks and returns, with an indication that the file descriptor is
293 ready (a subsequent I/O operation will likely fail with an error, un‐
294 less another process reopens file descriptor between the time select()
295 returned and the I/O operation is performed). On Linux (and some other
296 systems), closing the file descriptor in another thread has no effect
297 on select(). In summary, any application that relies on a particular
298 behavior in this scenario must be considered buggy.
299
300 C library/kernel differences
301 The Linux kernel allows file descriptor sets of arbitrary size, deter‐
302 mining the length of the sets to be checked from the value of nfds.
303 However, in the glibc implementation, the fd_set type is fixed in size.
304 See also BUGS.
305
306 The pselect() interface described in this page is implemented by glibc.
307 The underlying Linux system call is named pselect6(). This system call
308 has somewhat different behavior from the glibc wrapper function.
309
310 The Linux pselect6() system call modifies its timeout argument. How‐
311 ever, the glibc wrapper function hides this behavior by using a local
312 variable for the timeout argument that is passed to the system call.
313 Thus, the glibc pselect() function does not modify its timeout argu‐
314 ment; this is the behavior required by POSIX.1-2001.
315
316 The final argument of the pselect6() system call is not a sigset_t *
317 pointer, but is instead a structure of the form:
318
319 struct {
320 const kernel_sigset_t *ss; /* Pointer to signal set */
321 size_t ss_len; /* Size (in bytes) of object
322 pointed to by 'ss' */
323 };
324
325 This allows the system call to obtain both a pointer to the signal set
326 and its size, while allowing for the fact that most architectures sup‐
327 port a maximum of 6 arguments to a system call. See sigprocmask(2) for
328 a discussion of the difference between the kernel and libc notion of
329 the signal set.
330
331 Historical glibc details
332 Glibc 2.0 provided an incorrect version of pselect() that did not take
333 a sigmask argument.
334
335 In glibc versions 2.1 to 2.2.1, one must define _GNU_SOURCE in order to
336 obtain the declaration of pselect() from <sys/select.h>.
337
339 POSIX allows an implementation to define an upper limit, advertised via
340 the constant FD_SETSIZE, on the range of file descriptors that can be
341 specified in a file descriptor set. The Linux kernel imposes no fixed
342 limit, but the glibc implementation makes fd_set a fixed-size type,
343 with FD_SETSIZE defined as 1024, and the FD_*() macros operating ac‐
344 cording to that limit. To monitor file descriptors greater than 1023,
345 use poll(2) or epoll(7) instead.
346
347 The implementation of the fd_set arguments as value-result arguments is
348 a design error that is avoided in poll(2) and epoll(7).
349
350 According to POSIX, select() should check all specified file descrip‐
351 tors in the three file descriptor sets, up to the limit nfds-1. How‐
352 ever, the current implementation ignores any file descriptor in these
353 sets that is greater than the maximum file descriptor number that the
354 process currently has open. According to POSIX, any such file descrip‐
355 tor that is specified in one of the sets should result in the error
356 EBADF.
357
358 Starting with version 2.1, glibc provided an emulation of pselect()
359 that was implemented using sigprocmask(2) and select(). This implemen‐
360 tation remained vulnerable to the very race condition that pselect()
361 was designed to prevent. Modern versions of glibc use the (race-free)
362 pselect() system call on kernels where it is provided.
363
364 On Linux, select() may report a socket file descriptor as "ready for
365 reading", while nevertheless a subsequent read blocks. This could for
366 example happen when data has arrived but upon examination has the wrong
367 checksum and is discarded. There may be other circumstances in which a
368 file descriptor is spuriously reported as ready. Thus it may be safer
369 to use O_NONBLOCK on sockets that should not block.
370
371 On Linux, select() also modifies timeout if the call is interrupted by
372 a signal handler (i.e., the EINTR error return). This is not permitted
373 by POSIX.1. The Linux pselect() system call has the same behavior, but
374 the glibc wrapper hides this behavior by internally copying the timeout
375 to a local variable and passing that variable to the system call.
376
378 #include <stdio.h>
379 #include <stdlib.h>
380 #include <sys/select.h>
381
382 int
383 main(void)
384 {
385 fd_set rfds;
386 struct timeval tv;
387 int retval;
388
389 /* Watch stdin (fd 0) to see when it has input. */
390
391 FD_ZERO(&rfds);
392 FD_SET(0, &rfds);
393
394 /* Wait up to five seconds. */
395
396 tv.tv_sec = 5;
397 tv.tv_usec = 0;
398
399 retval = select(1, &rfds, NULL, NULL, &tv);
400 /* Don't rely on the value of tv now! */
401
402 if (retval == -1)
403 perror("select()");
404 else if (retval)
405 printf("Data is available now.\n");
406 /* FD_ISSET(0, &rfds) will be true. */
407 else
408 printf("No data within five seconds.\n");
409
410 exit(EXIT_SUCCESS);
411 }
412
414 accept(2), connect(2), poll(2), read(2), recv(2), restart_syscall(2),
415 send(2), sigprocmask(2), write(2), epoll(7), time(7)
416
417 For a tutorial with discussion and examples, see select_tut(2).
418
420 This page is part of release 5.13 of the Linux man-pages project. A
421 description of the project, information about reporting bugs, and the
422 latest version of this page, can be found at
423 https://www.kernel.org/doc/man-pages/.
424
425
426
427Linux 2021-03-22 SELECT(2)