1TIMERFD_CREATE(2) Linux Programmer's Manual TIMERFD_CREATE(2)
2
6 timerfd_create, timerfd_settime, timerfd_gettime - timers that notify
7 via file descriptors
8
10 #include <sys/timerfd.h>
11 int timerfd_create(int clockid, int flags);
13 int timerfd_settime(int fd, int flags,
15 const struct itimerspec *new_value,
16 struct itimerspec *old_value);
17 int timerfd_gettime(int fd, struct itimerspec *curr_value);
19
21 These system calls create and operate on a timer that delivers timer
22 expiration notifications via a file descriptor. They provide an alter‐
23 native to the use of setitimer(2) or timer_create(2), with the advan‐
24 tage that the file descriptor may be monitored by select(2), poll(2),
25 and epoll(7).
26 The use of these three system calls is analogous to the use of
28 timer_create(2), timer_settime(2), and timer_gettime(2). (There is no
29 analog of timer_getoverrun(2), since that functionality is provided by
30 read(2), as described below.)
31 timerfd_create()
33 timerfd_create() creates a new timer object, and returns a file de‐
34 scriptor that refers to that timer. The clockid argument specifies the
35 clock that is used to mark the progress of the timer, and must be one
36 of the following:
37 CLOCK_REALTIME
39 A settable system-wide real-time clock.
40 CLOCK_MONOTONIC
42 A nonsettable monotonically increasing clock that measures time
43 from some unspecified point in the past that does not change af‐
44 ter system startup.
45 CLOCK_BOOTTIME (Since Linux 3.15)
47 Like CLOCK_MONOTONIC, this is a monotonically increasing clock.
48 However, whereas the CLOCK_MONOTONIC clock does not measure the
49 time while a system is suspended, the CLOCK_BOOTTIME clock does
50 include the time during which the system is suspended. This is
51 useful for applications that need to be suspend-aware.
52 CLOCK_REALTIME is not suitable for such applications, since that
53 clock is affected by discontinuous changes to the system clock.
54 CLOCK_REALTIME_ALARM (since Linux 3.11)
56 This clock is like CLOCK_REALTIME, but will wake the system if
57 it is suspended. The caller must have the CAP_WAKE_ALARM capa‐
58 bility in order to set a timer against this clock.
59 CLOCK_BOOTTIME_ALARM (since Linux 3.11)
61 This clock is like CLOCK_BOOTTIME, but will wake the system if
62 it is suspended. The caller must have the CAP_WAKE_ALARM capa‐
63 bility in order to set a timer against this clock.
64 See clock_getres(2) for some further details on the above clocks.
66 The current value of each of these clocks can be retrieved using
68 clock_gettime(2).
69 Starting with Linux 2.6.27, the following values may be bitwise ORed in
71 flags to change the behavior of timerfd_create():
72 TFD_NONBLOCK Set the O_NONBLOCK file status flag on the open file de‐
74 scription (see open(2)) referred to by the new file de‐
75 scriptor. Using this flag saves extra calls to fcntl(2)
76 to achieve the same result.
77 TFD_CLOEXEC Set the close-on-exec (FD_CLOEXEC) flag on the new file
79 descriptor. See the description of the O_CLOEXEC flag in
80 open(2) for reasons why this may be useful.
81 In Linux versions up to and including 2.6.26, flags must be specified
83 as zero.
84 timerfd_settime()
86 timerfd_settime() arms (starts) or disarms (stops) the timer referred
87 to by the file descriptor fd.
88 The new_value argument specifies the initial expiration and interval
90 for the timer. The itimerspec structure used for this argument con‐
91 tains two fields, each of which is in turn a structure of type time‐
92 spec:
93 struct timespec {
95 time_t tv_sec; /* Seconds */
96 long tv_nsec; /* Nanoseconds */
97 };
98 struct itimerspec {
100 struct timespec it_interval; /* Interval for periodic timer */
101 struct timespec it_value; /* Initial expiration */
102 };
103 new_value.it_value specifies the initial expiration of the timer, in
105 seconds and nanoseconds. Setting either field of new_value.it_value to
106 a nonzero value arms the timer. Setting both fields of
107 new_value.it_value to zero disarms the timer.
108 Setting one or both fields of new_value.it_interval to nonzero values
110 specifies the period, in seconds and nanoseconds, for repeated timer
111 expirations after the initial expiration. If both fields of
112 new_value.it_interval are zero, the timer expires just once, at the
113 time specified by new_value.it_value.
114 By default, the initial expiration time specified in new_value is in‐
116 terpreted relative to the current time on the timer's clock at the time
117 of the call (i.e., new_value.it_value specifies a time relative to the
118 current value of the clock specified by clockid). An absolute timeout
119 can be selected via the flags argument.
120 The flags argument is a bit mask that can include the following values:
122 TFD_TIMER_ABSTIME
124 Interpret new_value.it_value as an absolute value on the timer's
125 clock. The timer will expire when the value of the timer's
126 clock reaches the value specified in new_value.it_value.
127 TFD_TIMER_CANCEL_ON_SET
129 If this flag is specified along with TFD_TIMER_ABSTIME and the
130 clock for this timer is CLOCK_REALTIME or CLOCK_REALTIME_ALARM,
131 then mark this timer as cancelable if the real-time clock under‐
132 goes a discontinuous change (settimeofday(2), clock_settime(2),
133 or similar). When such changes occur, a current or future
134 read(2) from the file descriptor will fail with the error ECAN‐
135 CELED.
136 If the old_value argument is not NULL, then the itimerspec structure
138 that it points to is used to return the setting of the timer that was
139 current at the time of the call; see the description of timerfd_get‐
140 time() following.
141 timerfd_gettime()
143 timerfd_gettime() returns, in curr_value, an itimerspec structure that
144 contains the current setting of the timer referred to by the file de‐
145 scriptor fd.
146 The it_value field returns the amount of time until the timer will next
148 expire. If both fields of this structure are zero, then the timer is
149 currently disarmed. This field always contains a relative value, re‐
150 gardless of whether the TFD_TIMER_ABSTIME flag was specified when set‐
151 ting the timer.
152 The it_interval field returns the interval of the timer. If both
154 fields of this structure are zero, then the timer is set to expire just
155 once, at the time specified by curr_value.it_value.
156 Operating on a timer file descriptor
158 The file descriptor returned by timerfd_create() supports the following
159 additional operations:
160 read(2)
162 If the timer has already expired one or more times since its
163 settings were last modified using timerfd_settime(), or since
164 the last successful read(2), then the buffer given to read(2)
165 returns an unsigned 8-byte integer (uint64_t) containing the
166 number of expirations that have occurred. (The returned value
167 is in host byte order—that is, the native byte order for inte‐
168 gers on the host machine.)
169 If no timer expirations have occurred at the time of the
171 read(2), then the call either blocks until the next timer expi‐
172 ration, or fails with the error EAGAIN if the file descriptor
173 has been made nonblocking (via the use of the fcntl(2) F_SETFL
174 operation to set the O_NONBLOCK flag).
175 A read(2) fails with the error EINVAL if the size of the sup‐
177 plied buffer is less than 8 bytes.
178 If the associated clock is either CLOCK_REALTIME or CLOCK_REAL‐
180 TIME_ALARM, the timer is absolute (TFD_TIMER_ABSTIME), and the
181 flag TFD_TIMER_CANCEL_ON_SET was specified when calling
182 timerfd_settime(), then read(2) fails with the error ECANCELED
183 if the real-time clock undergoes a discontinuous change. (This
184 allows the reading application to discover such discontinuous
185 changes to the clock.)
186 If the associated clock is either CLOCK_REALTIME or CLOCK_REAL‐
188 TIME_ALARM, the timer is absolute (TFD_TIMER_ABSTIME), and the
189 flag TFD_TIMER_CANCEL_ON_SET was not specified when calling
190 timerfd_settime(), then a discontinuous negative change to the
191 clock (e.g., clock_settime(2)) may cause read(2) to unblock, but
192 return a value of 0 (i.e., no bytes read), if the clock change
193 occurs after the time expired, but before the read(2) on the
194 file descriptor.
195 poll(2), select(2) (and similar)
197 The file descriptor is readable (the select(2) readfds argument;
198 the poll(2) POLLIN flag) if one or more timer expirations have
199 occurred.
200 The file descriptor also supports the other file-descriptor mul‐
202 tiplexing APIs: pselect(2), ppoll(2), and epoll(7).
203 ioctl(2)
205 The following timerfd-specific command is supported:
206 TFD_IOC_SET_TICKS (since Linux 3.17)
208 Adjust the number of timer expirations that have oc‐
209 curred. The argument is a pointer to a nonzero 8-byte
210 integer (uint64_t*) containing the new number of expira‐
211 tions. Once the number is set, any waiter on the timer
212 is woken up. The only purpose of this command is to re‐
213 store the expirations for the purpose of checkpoint/re‐
214 store. This operation is available only if the kernel
215 was configured with the CONFIG_CHECKPOINT_RESTORE option.
216 close(2)
218 When the file descriptor is no longer required it should be
219 closed. When all file descriptors associated with the same
220 timer object have been closed, the timer is disarmed and its re‐
221 sources are freed by the kernel.
222 fork(2) semantics
224 After a fork(2), the child inherits a copy of the file descriptor cre‐
225 ated by timerfd_create(). The file descriptor refers to the same un‐
226 derlying timer object as the corresponding file descriptor in the par‐
227 ent, and read(2)s in the child will return information about expira‐
228 tions of the timer.
229 execve(2) semantics
231 A file descriptor created by timerfd_create() is preserved across ex‐
232 ecve(2), and continues to generate timer expirations if the timer was
233 armed.
234
236 On success, timerfd_create() returns a new file descriptor. On error,
237 -1 is returned and errno is set to indicate the error.
238 timerfd_settime() and timerfd_gettime() return 0 on success; on error
240 they return -1, and set errno to indicate the error.
241
243 timerfd_create() can fail with the following errors:
244 EINVAL The clockid is not valid.
246 EINVAL flags is invalid; or, in Linux 2.6.26 or earlier, flags is non‐
248 zero.
249 EMFILE The per-process limit on the number of open file descriptors has
251 been reached.
252 ENFILE The system-wide limit on the total number of open files has been
254 reached.
255 ENODEV Could not mount (internal) anonymous inode device.
257 ENOMEM There was insufficient kernel memory to create the timer.
259 EPERM clockid was CLOCK_REALTIME_ALARM or CLOCK_BOOTTIME_ALARM but the
261 caller did not have the CAP_WAKE_ALARM capability.
262 timerfd_settime() and timerfd_gettime() can fail with the following er‐
264 rors:
265 EBADF fd is not a valid file descriptor.
267 EFAULT new_value, old_value, or curr_value is not valid a pointer.
269 EINVAL fd is not a valid timerfd file descriptor.
271 timerfd_settime() can also fail with the following errors:
273 ECANCELED
275 See NOTES.
276 EINVAL new_value is not properly initialized (one of the tv_nsec falls
278 outside the range zero to 999,999,999).
279 EINVAL flags is invalid.
281
283 These system calls are available on Linux since kernel 2.6.25. Library
284 support is provided by glibc since version 2.8.
285
287 These system calls are Linux-specific.
288
290 Suppose the following scenario for CLOCK_REALTIME or CLOCK_REAL‐
291 TIME_ALARM timer that was created with timerfd_create():
292 (a) The timer has been started (timerfd_settime()) with the
294 TFD_TIMER_ABSTIME and TFD_TIMER_CANCEL_ON_SET flags;
295 (b) A discontinuous change (e.g., settimeofday(2)) is subsequently made
297 to the CLOCK_REALTIME clock; and
298 (c) the caller once more calls timerfd_settime() to rearm the timer
300 (without first doing a read(2) on the file descriptor).
301 In this case the following occurs:
303 • The timerfd_settime() returns -1 with errno set to ECANCELED. (This
305 enables the caller to know that the previous timer was affected by a
306 discontinuous change to the clock.)
307 • The timer is successfully rearmed with the settings provided in the
309 second timerfd_settime() call. (This was probably an implementation
310 accident, but won't be fixed now, in case there are applications that
311 depend on this behaviour.)
312
314 Currently, timerfd_create() supports fewer types of clock IDs than
315 timer_create(2).
316
318 The following program creates a timer and then monitors its progress.
319 The program accepts up to three command-line arguments. The first ar‐
320 gument specifies the number of seconds for the initial expiration of
321 the timer. The second argument specifies the interval for the timer,
322 in seconds. The third argument specifies the number of times the pro‐
323 gram should allow the timer to expire before terminating. The second
324 and third command-line arguments are optional.
325 The following shell session demonstrates the use of the program:
327 $ a.out 3 1 100
329 0.000: timer started
330 3.000: read: 1; total=1
331 4.000: read: 1; total=2
332 ^Z # type control-Z to suspend the program
333 [1]+ Stopped ./timerfd3_demo 3 1 100
334 $ fg # Resume execution after a few seconds
335 a.out 3 1 100
336 9.660: read: 5; total=7
337 10.000: read: 1; total=8
338 11.000: read: 1; total=9
339 ^C # type control-C to suspend the program
340 Program source
342 #include <sys/timerfd.h>
344 #include <time.h>
345 #include <unistd.h>
346 #include <inttypes.h> /* Definition of PRIu64 */
347 #include <stdlib.h>
348 #include <stdio.h>
349 #include <stdint.h> /* Definition of uint64_t */
350 #define handle_error(msg) \
352 do { perror(msg); exit(EXIT_FAILURE); } while (0)
353 static void
355 print_elapsed_time(void)
356 {
357 static struct timespec start;
358 struct timespec curr;
359 static int first_call = 1;
360 int secs, nsecs;
361 if (first_call) {
363 first_call = 0;
364 if (clock_gettime(CLOCK_MONOTONIC, &start) == -1)
365 handle_error("clock_gettime");
366 }
367 if (clock_gettime(CLOCK_MONOTONIC, &curr) == -1)
369 handle_error("clock_gettime");
370 secs = curr.tv_sec - start.tv_sec;
372 nsecs = curr.tv_nsec - start.tv_nsec;
373 if (nsecs < 0) {
374 secs--;
375 nsecs += 1000000000;
376 }
377 printf("%d.%03d: ", secs, (nsecs + 500000) / 1000000);
378 }
379 int
381 main(int argc, char *argv[])
382 {
383 struct itimerspec new_value;
384 int max_exp, fd;
385 struct timespec now;
386 uint64_t exp, tot_exp;
387 ssize_t s;
388 if ((argc != 2) && (argc != 4)) {
390 fprintf(stderr, "%s init-secs [interval-secs max-exp]\n",
391 argv[0]);
392 exit(EXIT_FAILURE);
393 }
394 if (clock_gettime(CLOCK_REALTIME, &now) == -1)
396 handle_error("clock_gettime");
397 /* Create a CLOCK_REALTIME absolute timer with initial
399 expiration and interval as specified in command line */
400 new_value.it_value.tv_sec = now.tv_sec + atoi(argv[1]);
402 new_value.it_value.tv_nsec = now.tv_nsec;
403 if (argc == 2) {
404 new_value.it_interval.tv_sec = 0;
405 max_exp = 1;
406 } else {
407 new_value.it_interval.tv_sec = atoi(argv[2]);
408 max_exp = atoi(argv[3]);
409 }
410 new_value.it_interval.tv_nsec = 0;
411 fd = timerfd_create(CLOCK_REALTIME, 0);
413 if (fd == -1)
414 handle_error("timerfd_create");
415 if (timerfd_settime(fd, TFD_TIMER_ABSTIME, &new_value, NULL) == -1)
417 handle_error("timerfd_settime");
418 print_elapsed_time();
420 printf("timer started\n");
421 for (tot_exp = 0; tot_exp < max_exp;) {
423 s = read(fd, &exp, sizeof(uint64_t));
424 if (s != sizeof(uint64_t))
425 handle_error("read");
426 tot_exp += exp;
428 print_elapsed_time();
429 printf("read: %" PRIu64 "; total=%" PRIu64 "\n", exp, tot_exp);
430 }
431 exit(EXIT_SUCCESS);
433 }
434
436 eventfd(2), poll(2), read(2), select(2), setitimer(2), signalfd(2),
437 timer_create(2), timer_gettime(2), timer_settime(2), epoll(7), time(7)
438
440 This page is part of release 5.10 of the Linux man-pages project. A
441 description of the project, information about reporting bugs, and the
442 latest version of this page, can be found at
443 https://www.kernel.org/doc/man-pages/.
444Linux 2020-08-13 TIMERFD_CREATE(2)