1 /* $NetBSD: sys_eventfd.c,v 1.9 2022/02/17 16:28:29 thorpej Exp $ */
2
3 /*-
4 * Copyright (c) 2020 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: sys_eventfd.c,v 1.9 2022/02/17 16:28:29 thorpej Exp $");
34
35 /*
36 * eventfd
37 *
38 * Eventfd objects present a simple counting object associated with a
39 * file descriptor. Writes and reads to this file descriptor increment
40 * and decrement the count, respectively. When the count is non-zero,
41 * the descriptor is considered "readable", and when less than the max
42 * value (EVENTFD_MAXVAL), is considered "writable".
43 *
44 * This implementation is API compatible with the Linux eventfd(2)
45 * interface.
46 */
47
48 #include <sys/param.h>
49 #include <sys/types.h>
50 #include <sys/condvar.h>
51 #include <sys/eventfd.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/kauth.h>
55 #include <sys/mutex.h>
56 #include <sys/poll.h>
57 #include <sys/proc.h>
58 #include <sys/select.h>
59 #include <sys/stat.h>
60 #include <sys/syscallargs.h>
61 #include <sys/uio.h>
62
63 struct eventfd {
64 kmutex_t efd_lock;
65 kcondvar_t efd_read_wait;
66 kcondvar_t efd_write_wait;
67 struct selinfo efd_read_sel;
68 struct selinfo efd_write_sel;
69 eventfd_t efd_val;
70 int64_t efd_nwaiters;
71 bool efd_restarting;
72 bool efd_has_read_waiters;
73 bool efd_has_write_waiters;
74 bool efd_is_semaphore;
75
76 /*
77 * Information kept for stat(2).
78 */
79 struct timespec efd_btime; /* time created */
80 struct timespec efd_mtime; /* last write */
81 struct timespec efd_atime; /* last read */
82 };
83
84 #define EVENTFD_MAXVAL (UINT64_MAX - 1)
85
86 /*
87 * eventfd_create:
88 *
89 * Create an eventfd object.
90 */
91 static struct eventfd *
92 eventfd_create(unsigned int const val, int const flags)
93 {
94 struct eventfd * const efd = kmem_zalloc(sizeof(*efd), KM_SLEEP);
95
96 mutex_init(&efd->efd_lock, MUTEX_DEFAULT, IPL_NONE);
97 cv_init(&efd->efd_read_wait, "efdread");
98 cv_init(&efd->efd_write_wait, "efdwrite");
99 selinit(&efd->efd_read_sel);
100 selinit(&efd->efd_write_sel);
101 efd->efd_val = val;
102 efd->efd_is_semaphore = !!(flags & EFD_SEMAPHORE);
103 getnanotime(&efd->efd_btime);
104
105 /* Caller deals with EFD_CLOEXEC and EFD_NONBLOCK. */
106
107 return efd;
108 }
109
110 /*
111 * eventfd_destroy:
112 *
113 * Destroy an eventfd object.
114 */
115 static void
116 eventfd_destroy(struct eventfd * const efd)
117 {
118
119 KASSERT(efd->efd_nwaiters == 0);
120 KASSERT(efd->efd_has_read_waiters == false);
121 KASSERT(efd->efd_has_write_waiters == false);
122
123 cv_destroy(&efd->efd_read_wait);
124 cv_destroy(&efd->efd_write_wait);
125
126 seldestroy(&efd->efd_read_sel);
127 seldestroy(&efd->efd_write_sel);
128
129 mutex_destroy(&efd->efd_lock);
130
131 kmem_free(efd, sizeof(*efd));
132 }
133
134 /*
135 * eventfd_wait:
136 *
137 * Block on an eventfd. Handles non-blocking, as well as
138 * the restart cases.
139 */
140 static int
141 eventfd_wait(struct eventfd * const efd, int const fflag, bool const is_write)
142 {
143 kcondvar_t *waitcv;
144 int error;
145
146 if (fflag & FNONBLOCK) {
147 return EAGAIN;
148 }
149
150 /*
151 * We're going to block. Check if we need to return ERESTART.
152 */
153 if (efd->efd_restarting) {
154 return ERESTART;
155 }
156
157 if (is_write) {
158 efd->efd_has_write_waiters = true;
159 waitcv = &efd->efd_write_wait;
160 } else {
161 efd->efd_has_read_waiters = true;
162 waitcv = &efd->efd_read_wait;
163 }
164
165 efd->efd_nwaiters++;
166 KASSERT(efd->efd_nwaiters > 0);
167 error = cv_wait_sig(waitcv, &efd->efd_lock);
168 efd->efd_nwaiters--;
169 KASSERT(efd->efd_nwaiters >= 0);
170
171 /*
172 * If a restart was triggered while we were asleep, we need
173 * to return ERESTART if no other error was returned.
174 */
175 if (efd->efd_restarting) {
176 if (error == 0) {
177 error = ERESTART;
178 }
179 }
180
181 return error;
182 }
183
184 /*
185 * eventfd_wake:
186 *
187 * Wake LWPs block on an eventfd.
188 */
189 static void
190 eventfd_wake(struct eventfd * const efd, bool const is_write)
191 {
192 kcondvar_t *waitcv = NULL;
193 struct selinfo *sel;
194 int pollev;
195
196 if (is_write) {
197 if (efd->efd_has_read_waiters) {
198 waitcv = &efd->efd_read_wait;
199 efd->efd_has_read_waiters = false;
200 }
201 sel = &efd->efd_read_sel;
202 pollev = POLLIN | POLLRDNORM;
203 } else {
204 if (efd->efd_has_write_waiters) {
205 waitcv = &efd->efd_write_wait;
206 efd->efd_has_write_waiters = false;
207 }
208 sel = &efd->efd_write_sel;
209 pollev = POLLOUT | POLLWRNORM;
210 }
211 if (waitcv != NULL) {
212 cv_broadcast(waitcv);
213 }
214 selnotify(sel, pollev, NOTE_SUBMIT);
215 }
216
217 /*
218 * eventfd file operations
219 */
220
221 static int
222 eventfd_fop_read(file_t * const fp, off_t * const offset,
223 struct uio * const uio, kauth_cred_t const cred, int const flags)
224 {
225 struct eventfd * const efd = fp->f_eventfd;
226 int const fflag = fp->f_flag;
227 eventfd_t return_value;
228 int error;
229
230 if (uio->uio_resid < sizeof(eventfd_t)) {
231 return EINVAL;
232 }
233
234 mutex_enter(&efd->efd_lock);
235
236 while (efd->efd_val == 0) {
237 if ((error = eventfd_wait(efd, fflag, false)) != 0) {
238 mutex_exit(&efd->efd_lock);
239 return error;
240 }
241 }
242
243 if (efd->efd_is_semaphore) {
244 return_value = 1;
245 efd->efd_val--;
246 } else {
247 return_value = efd->efd_val;
248 efd->efd_val = 0;
249 }
250
251 getnanotime(&efd->efd_atime);
252 eventfd_wake(efd, false);
253
254 mutex_exit(&efd->efd_lock);
255
256 error = uiomove(&return_value, sizeof(return_value), uio);
257
258 return error;
259 }
260
261 static int
262 eventfd_fop_write(file_t * const fp, off_t * const offset,
263 struct uio * const uio, kauth_cred_t const cred, int const flags)
264 {
265 struct eventfd * const efd = fp->f_eventfd;
266 int const fflag = fp->f_flag;
267 eventfd_t write_value;
268 int error;
269
270 if (uio->uio_resid < sizeof(eventfd_t)) {
271 return EINVAL;
272 }
273
274 if ((error = uiomove(&write_value, sizeof(write_value), uio)) != 0) {
275 return error;
276 }
277
278 if (write_value > EVENTFD_MAXVAL) {
279 error = EINVAL;
280 goto out;
281 }
282
283 mutex_enter(&efd->efd_lock);
284
285 KASSERT(efd->efd_val <= EVENTFD_MAXVAL);
286 while ((EVENTFD_MAXVAL - efd->efd_val) < write_value) {
287 if ((error = eventfd_wait(efd, fflag, true)) != 0) {
288 mutex_exit(&efd->efd_lock);
289 goto out;
290 }
291 }
292
293 efd->efd_val += write_value;
294 KASSERT(efd->efd_val <= EVENTFD_MAXVAL);
295
296 getnanotime(&efd->efd_mtime);
297 eventfd_wake(efd, true);
298
299 mutex_exit(&efd->efd_lock);
300
301 out:
302 if (error) {
303 /*
304 * Undo the effect of uiomove() so that the error
305 * gets reported correctly; see dofilewrite().
306 */
307 uio->uio_resid += sizeof(write_value);
308 }
309 return error;
310 }
311
312 static int
313 eventfd_ioctl(file_t * const fp, u_long const cmd, void * const data)
314 {
315 struct eventfd * const efd = fp->f_eventfd;
316
317 switch (cmd) {
318 case FIONBIO:
319 return 0;
320
321 case FIONREAD:
322 mutex_enter(&efd->efd_lock);
323 *(int *)data = efd->efd_val != 0 ? sizeof(eventfd_t) : 0;
324 mutex_exit(&efd->efd_lock);
325 return 0;
326
327 case FIONWRITE:
328 *(int *)data = 0;
329 return 0;
330
331 case FIONSPACE:
332 /*
333 * FIONSPACE doesn't really work for eventfd, because the
334 * writability depends on the contents (value) being written.
335 */
336 break;
337
338 default:
339 break;
340 }
341
342 return EPASSTHROUGH;
343 }
344
345 static int
346 eventfd_fop_poll(file_t * const fp, int const events)
347 {
348 struct eventfd * const efd = fp->f_eventfd;
349 int revents = 0;
350
351 /*
352 * Note that Linux will return POLLERR if the eventfd count
353 * overflows, but that is not possible in the normal read/write
354 * API, only with Linux kernel-internal interfaces. So, this
355 * implementation never returns POLLERR.
356 *
357 * Also note that the Linux eventfd(2) man page does not
358 * specifically discuss returning POLLRDNORM, but we check
359 * for that event in addition to POLLIN.
360 */
361
362 mutex_enter(&efd->efd_lock);
363
364 if (events & (POLLIN | POLLRDNORM)) {
365 if (efd->efd_val != 0) {
366 revents |= events & (POLLIN | POLLRDNORM);
367 } else {
368 selrecord(curlwp, &efd->efd_read_sel);
369 }
370 }
371
372 if (events & (POLLOUT | POLLWRNORM)) {
373 if (efd->efd_val < EVENTFD_MAXVAL) {
374 revents |= events & (POLLOUT | POLLWRNORM);
375 } else {
376 selrecord(curlwp, &efd->efd_write_sel);
377 }
378 }
379
380 mutex_exit(&efd->efd_lock);
381
382 return revents;
383 }
384
385 static int
386 eventfd_fop_stat(file_t * const fp, struct stat * const st)
387 {
388 struct eventfd * const efd = fp->f_eventfd;
389
390 memset(st, 0, sizeof(*st));
391
392 mutex_enter(&efd->efd_lock);
393 st->st_size = (off_t)efd->efd_val;
394 st->st_blksize = sizeof(eventfd_t);
395 st->st_mode = S_IFIFO | S_IRUSR | S_IWUSR;
396 st->st_blocks = 1;
397 st->st_birthtimespec = st->st_ctimespec = efd->efd_btime;
398 st->st_atimespec = efd->efd_atime;
399 st->st_mtimespec = efd->efd_mtime;
400 st->st_uid = kauth_cred_geteuid(fp->f_cred);
401 st->st_gid = kauth_cred_getegid(fp->f_cred);
402 mutex_exit(&efd->efd_lock);
403
404 return 0;
405 }
406
407 static int
408 eventfd_fop_close(file_t * const fp)
409 {
410 struct eventfd * const efd = fp->f_eventfd;
411
412 fp->f_eventfd = NULL;
413 eventfd_destroy(efd);
414
415 return 0;
416 }
417
418 static void
419 eventfd_filt_read_detach(struct knote * const kn)
420 {
421 struct eventfd * const efd = ((file_t *)kn->kn_obj)->f_eventfd;
422
423 mutex_enter(&efd->efd_lock);
424 KASSERT(kn->kn_hook == efd);
425 selremove_knote(&efd->efd_read_sel, kn);
426 mutex_exit(&efd->efd_lock);
427 }
428
429 static int
430 eventfd_filt_read(struct knote * const kn, long const hint)
431 {
432 struct eventfd * const efd = ((file_t *)kn->kn_obj)->f_eventfd;
433 int rv;
434
435 if (hint & NOTE_SUBMIT) {
436 KASSERT(mutex_owned(&efd->efd_lock));
437 } else {
438 mutex_enter(&efd->efd_lock);
439 }
440
441 kn->kn_data = (int64_t)efd->efd_val;
442 rv = (eventfd_t)kn->kn_data > 0;
443
444 if ((hint & NOTE_SUBMIT) == 0) {
445 mutex_exit(&efd->efd_lock);
446 }
447
448 return rv;
449 }
450
451 static const struct filterops eventfd_read_filterops = {
452 .f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE,
453 .f_detach = eventfd_filt_read_detach,
454 .f_event = eventfd_filt_read,
455 };
456
457 static void
458 eventfd_filt_write_detach(struct knote * const kn)
459 {
460 struct eventfd * const efd = ((file_t *)kn->kn_obj)->f_eventfd;
461
462 mutex_enter(&efd->efd_lock);
463 KASSERT(kn->kn_hook == efd);
464 selremove_knote(&efd->efd_write_sel, kn);
465 mutex_exit(&efd->efd_lock);
466 }
467
468 static int
469 eventfd_filt_write(struct knote * const kn, long const hint)
470 {
471 struct eventfd * const efd = ((file_t *)kn->kn_obj)->f_eventfd;
472 int rv;
473
474 if (hint & NOTE_SUBMIT) {
475 KASSERT(mutex_owned(&efd->efd_lock));
476 } else {
477 mutex_enter(&efd->efd_lock);
478 }
479
480 kn->kn_data = (int64_t)efd->efd_val;
481 rv = (eventfd_t)kn->kn_data < EVENTFD_MAXVAL;
482
483 if ((hint & NOTE_SUBMIT) == 0) {
484 mutex_exit(&efd->efd_lock);
485 }
486
487 return rv;
488 }
489
490 static const struct filterops eventfd_write_filterops = {
491 .f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE,
492 .f_detach = eventfd_filt_write_detach,
493 .f_event = eventfd_filt_write,
494 };
495
496 static int
497 eventfd_fop_kqfilter(file_t * const fp, struct knote * const kn)
498 {
499 struct eventfd * const efd = ((file_t *)kn->kn_obj)->f_eventfd;
500 struct selinfo *sel;
501
502 switch (kn->kn_filter) {
503 case EVFILT_READ:
504 sel = &efd->efd_read_sel;
505 kn->kn_fop = &eventfd_read_filterops;
506 break;
507
508 case EVFILT_WRITE:
509 sel = &efd->efd_write_sel;
510 kn->kn_fop = &eventfd_write_filterops;
511 break;
512
513 default:
514 return EINVAL;
515 }
516
517 kn->kn_hook = efd;
518
519 mutex_enter(&efd->efd_lock);
520 selrecord_knote(sel, kn);
521 mutex_exit(&efd->efd_lock);
522
523 return 0;
524 }
525
526 static void
527 eventfd_fop_restart(file_t * const fp)
528 {
529 struct eventfd * const efd = fp->f_eventfd;
530
531 /*
532 * Unblock blocked reads/writes in order to allow close() to complete.
533 * System calls return ERESTART so that the fd is revalidated.
534 */
535
536 mutex_enter(&efd->efd_lock);
537
538 if (efd->efd_nwaiters != 0) {
539 efd->efd_restarting = true;
540 if (efd->efd_has_read_waiters) {
541 cv_broadcast(&efd->efd_read_wait);
542 efd->efd_has_read_waiters = false;
543 }
544 if (efd->efd_has_write_waiters) {
545 cv_broadcast(&efd->efd_write_wait);
546 efd->efd_has_write_waiters = false;
547 }
548 }
549
550 mutex_exit(&efd->efd_lock);
551 }
552
553 static const struct fileops eventfd_fileops = {
554 .fo_name = "eventfd",
555 .fo_read = eventfd_fop_read,
556 .fo_write = eventfd_fop_write,
557 .fo_ioctl = eventfd_ioctl,
558 .fo_fcntl = fnullop_fcntl,
559 .fo_poll = eventfd_fop_poll,
560 .fo_stat = eventfd_fop_stat,
561 .fo_close = eventfd_fop_close,
562 .fo_kqfilter = eventfd_fop_kqfilter,
563 .fo_restart = eventfd_fop_restart,
564 };
565
566 /*
567 * eventfd(2) system call
568 */
569 int
570 do_eventfd(struct lwp * const l, unsigned int const val, int const flags,
571 register_t *retval)
572 {
573 file_t *fp;
574 int fd, error;
575
576 if (flags & ~(EFD_CLOEXEC | EFD_NONBLOCK | EFD_SEMAPHORE)) {
577 return EINVAL;
578 }
579
580 if ((error = fd_allocfile(&fp, &fd)) != 0) {
581 return error;
582 }
583
584 fp->f_flag = FREAD | FWRITE;
585 if (flags & EFD_NONBLOCK) {
586 fp->f_flag |= FNONBLOCK;
587 }
588 fp->f_type = DTYPE_EVENTFD;
589 fp->f_ops = &eventfd_fileops;
590 fp->f_eventfd = eventfd_create(val, flags);
591 fd_set_exclose(l, fd, !!(flags & EFD_CLOEXEC));
592 fd_affix(curproc, fp, fd);
593
594 *retval = fd;
595 return 0;
596 }
597
598 int
599 sys_eventfd(struct lwp *l, const struct sys_eventfd_args *uap,
600 register_t *retval)
601 {
602 /* {
603 syscallarg(unsigned int) val;
604 syscallarg(int) flags;
605 } */
606
607 return do_eventfd(l, SCARG(uap, val), SCARG(uap, flags), retval);
608 }
Cache object: 40292b39a95ef333ae3390ba7c396520
|