xref: /freebsd/sys/compat/linux/linux_event.c (revision 4ab7403bbd76f466d482cd69035091c04e409c09)
1 /*-
2  * Copyright (c) 2007 Roman Divacky
3  * Copyright (c) 2014 Dmitry Chagin
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_compat.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/imgact.h>
36 #include <sys/kernel.h>
37 #include <sys/limits.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/capability.h>
41 #include <sys/types.h>
42 #include <sys/user.h>
43 #include <sys/file.h>
44 #include <sys/filedesc.h>
45 #include <sys/errno.h>
46 #include <sys/event.h>
47 #include <sys/poll.h>
48 #include <sys/proc.h>
49 #include <sys/selinfo.h>
50 #include <sys/sx.h>
51 #include <sys/syscallsubr.h>
52 #include <sys/timespec.h>
53 
54 #ifdef COMPAT_LINUX32
55 #include <machine/../linux32/linux.h>
56 #include <machine/../linux32/linux32_proto.h>
57 #else
58 #include <machine/../linux/linux.h>
59 #include <machine/../linux/linux_proto.h>
60 #endif
61 
62 #include <compat/linux/linux_emul.h>
63 #include <compat/linux/linux_event.h>
64 #include <compat/linux/linux_file.h>
65 #include <compat/linux/linux_util.h>
66 
67 /*
68  * epoll defines 'struct epoll_event' with the field 'data' as 64 bits
69  * on all architectures. But on 32 bit architectures BSD 'struct kevent' only
70  * has 32 bit opaque pointer as 'udata' field. So we can't pass epoll supplied
71  * data verbatuim. Therefore we allocate 64-bit memory block to pass
72  * user supplied data for every file descriptor.
73  */
74 
75 typedef uint64_t	epoll_udata_t;
76 
77 struct epoll_emuldata {
78 	uint32_t	fdc;		/* epoll udata max index */
79 	epoll_udata_t	udata[1];	/* epoll user data vector */
80 };
81 
82 #define	EPOLL_DEF_SZ		16
83 #define	EPOLL_SIZE(fdn)			\
84 	(sizeof(struct epoll_emuldata)+(fdn) * sizeof(epoll_udata_t))
85 
86 struct epoll_event {
87 	uint32_t	events;
88 	epoll_udata_t	data;
89 }
90 #if defined(__amd64__)
91 __attribute__((packed))
92 #endif
93 ;
94 
95 #define	LINUX_MAX_EVENTS	(INT_MAX / sizeof(struct epoll_event))
96 
97 static void	epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata);
98 static int	epoll_to_kevent(struct thread *td, struct file *epfp,
99 		    int fd, struct epoll_event *l_event, int *kev_flags,
100 		    struct kevent *kevent, int *nkevents);
101 static void	kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event);
102 static int	epoll_kev_copyout(void *arg, struct kevent *kevp, int count);
103 static int	epoll_kev_copyin(void *arg, struct kevent *kevp, int count);
104 static int	epoll_delete_event(struct thread *td, struct file *epfp,
105 		    int fd, int filter);
106 static int	epoll_delete_all_events(struct thread *td, struct file *epfp,
107 		    int fd);
108 
109 struct epoll_copyin_args {
110 	struct kevent	*changelist;
111 };
112 
113 struct epoll_copyout_args {
114 	struct epoll_event	*leventlist;
115 	struct proc		*p;
116 	uint32_t		count;
117 	int			error;
118 };
119 
120 /* eventfd */
121 typedef uint64_t	eventfd_t;
122 
123 static fo_rdwr_t	eventfd_read;
124 static fo_rdwr_t	eventfd_write;
125 static fo_truncate_t	eventfd_truncate;
126 static fo_ioctl_t	eventfd_ioctl;
127 static fo_poll_t	eventfd_poll;
128 static fo_kqfilter_t	eventfd_kqfilter;
129 static fo_stat_t	eventfd_stat;
130 static fo_close_t	eventfd_close;
131 static fo_fill_kinfo_t	eventfd_fill_kinfo;
132 
133 static struct fileops eventfdops = {
134 	.fo_read = eventfd_read,
135 	.fo_write = eventfd_write,
136 	.fo_truncate = eventfd_truncate,
137 	.fo_ioctl = eventfd_ioctl,
138 	.fo_poll = eventfd_poll,
139 	.fo_kqfilter = eventfd_kqfilter,
140 	.fo_stat = eventfd_stat,
141 	.fo_close = eventfd_close,
142 	.fo_chmod = invfo_chmod,
143 	.fo_chown = invfo_chown,
144 	.fo_sendfile = invfo_sendfile,
145 	.fo_fill_kinfo = eventfd_fill_kinfo,
146 	.fo_flags = DFLAG_PASSABLE
147 };
148 
149 static void	filt_eventfddetach(struct knote *kn);
150 static int	filt_eventfdread(struct knote *kn, long hint);
151 static int	filt_eventfdwrite(struct knote *kn, long hint);
152 
153 static struct filterops eventfd_rfiltops = {
154 	.f_isfd = 1,
155 	.f_detach = filt_eventfddetach,
156 	.f_event = filt_eventfdread
157 };
158 static struct filterops eventfd_wfiltops = {
159 	.f_isfd = 1,
160 	.f_detach = filt_eventfddetach,
161 	.f_event = filt_eventfdwrite
162 };
163 
164 struct eventfd {
165 	eventfd_t	efd_count;
166 	uint32_t	efd_flags;
167 	struct selinfo	efd_sel;
168 	struct mtx	efd_lock;
169 };
170 
171 static int	eventfd_create(struct thread *td, uint32_t initval, int flags);
172 
173 
174 static void
175 epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata)
176 {
177 	struct linux_pemuldata *pem;
178 	struct epoll_emuldata *emd;
179 	struct proc *p;
180 
181 	p = td->td_proc;
182 
183 	pem = pem_find(p);
184 	KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
185 
186 	LINUX_PEM_XLOCK(pem);
187 	if (pem->epoll == NULL) {
188 		emd = malloc(EPOLL_SIZE(fd), M_EPOLL, M_WAITOK);
189 		emd->fdc = fd;
190 		pem->epoll = emd;
191 	} else {
192 		emd = pem->epoll;
193 		if (fd > emd->fdc) {
194 			emd = realloc(emd, EPOLL_SIZE(fd), M_EPOLL, M_WAITOK);
195 			emd->fdc = fd;
196 			pem->epoll = emd;
197 		}
198 	}
199 	emd->udata[fd] = udata;
200 	LINUX_PEM_XUNLOCK(pem);
201 }
202 
203 static int
204 epoll_create_common(struct thread *td, int flags)
205 {
206 	int error;
207 
208 	error = kern_kqueue(td, flags);
209 	if (error)
210 		return (error);
211 
212 	epoll_fd_install(td, EPOLL_DEF_SZ, 0);
213 
214 	return (0);
215 }
216 
217 int
218 linux_epoll_create(struct thread *td, struct linux_epoll_create_args *args)
219 {
220 
221 	/*
222 	 * args->size is unused. Linux just tests it
223 	 * and then forgets it as well.
224 	 */
225 	if (args->size <= 0)
226 		return (EINVAL);
227 
228 	return (epoll_create_common(td, 0));
229 }
230 
231 int
232 linux_epoll_create1(struct thread *td, struct linux_epoll_create1_args *args)
233 {
234 	int flags;
235 
236 	if ((args->flags & ~(LINUX_O_CLOEXEC)) != 0)
237 		return (EINVAL);
238 
239 	flags = 0;
240 	if ((args->flags & LINUX_O_CLOEXEC) != 0)
241 		flags |= O_CLOEXEC;
242 
243 	return (epoll_create_common(td, flags));
244 }
245 
246 /* Structure converting function from epoll to kevent. */
247 static int
248 epoll_to_kevent(struct thread *td, struct file *epfp,
249     int fd, struct epoll_event *l_event, int *kev_flags,
250     struct kevent *kevent, int *nkevents)
251 {
252 	uint32_t levents = l_event->events;
253 	struct linux_pemuldata *pem;
254 	struct proc *p;
255 
256 	/* flags related to how event is registered */
257 	if ((levents & LINUX_EPOLLONESHOT) != 0)
258 		*kev_flags |= EV_ONESHOT;
259 	if ((levents & LINUX_EPOLLET) != 0)
260 		*kev_flags |= EV_CLEAR;
261 	if ((levents & LINUX_EPOLLERR) != 0)
262 		*kev_flags |= EV_ERROR;
263 
264 	/* flags related to what event is registered */
265 	if ((levents & LINUX_EPOLL_EVRD) != 0) {
266 		EV_SET(kevent++, fd, EVFILT_READ, *kev_flags, 0, 0, 0);
267 		++(*nkevents);
268 	}
269 	if ((levents & LINUX_EPOLL_EVWR) != 0) {
270 		EV_SET(kevent++, fd, EVFILT_WRITE, *kev_flags, 0, 0, 0);
271 		++(*nkevents);
272 	}
273 
274 	if ((levents & ~(LINUX_EPOLL_EVSUP)) != 0) {
275 		p = td->td_proc;
276 
277 		pem = pem_find(p);
278 		KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
279 		KASSERT(pem->epoll != NULL, ("epoll proc epolldata not found.\n"));
280 
281 		LINUX_PEM_XLOCK(pem);
282 		if ((pem->flags & LINUX_XUNSUP_EPOLL) == 0) {
283 			pem->flags |= LINUX_XUNSUP_EPOLL;
284 			LINUX_PEM_XUNLOCK(pem);
285 			linux_msg(td, "epoll_ctl unsupported flags: 0x%x\n",
286 			    levents);
287 		} else
288 			LINUX_PEM_XUNLOCK(pem);
289 		return (EINVAL);
290 	}
291 
292 	return (0);
293 }
294 
295 /*
296  * Structure converting function from kevent to epoll. In a case
297  * this is called on error in registration we store the error in
298  * event->data and pick it up later in linux_epoll_ctl().
299  */
300 static void
301 kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event)
302 {
303 
304 	if ((kevent->flags & EV_ERROR) != 0) {
305 		l_event->events = LINUX_EPOLLERR;
306 		return;
307 	}
308 
309 	switch (kevent->filter) {
310 	case EVFILT_READ:
311 		l_event->events = LINUX_EPOLLIN|LINUX_EPOLLRDNORM|LINUX_EPOLLPRI;
312 	break;
313 	case EVFILT_WRITE:
314 		l_event->events = LINUX_EPOLLOUT|LINUX_EPOLLWRNORM;
315 	break;
316 	}
317 }
318 
319 /*
320  * Copyout callback used by kevent. This converts kevent
321  * events to epoll events and copies them back to the
322  * userspace. This is also called on error on registering
323  * of the filter.
324  */
325 static int
326 epoll_kev_copyout(void *arg, struct kevent *kevp, int count)
327 {
328 	struct epoll_copyout_args *args;
329 	struct linux_pemuldata *pem;
330 	struct epoll_emuldata *emd;
331 	struct epoll_event *eep;
332 	int error, fd, i;
333 
334 	args = (struct epoll_copyout_args*) arg;
335 	eep = malloc(sizeof(*eep) * count, M_EPOLL, M_WAITOK | M_ZERO);
336 
337 	pem = pem_find(args->p);
338 	KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
339 	LINUX_PEM_SLOCK(pem);
340 	emd = pem->epoll;
341 	KASSERT(emd != NULL, ("epoll proc epolldata not found.\n"));
342 
343 	for (i = 0; i < count; i++) {
344 		kevent_to_epoll(&kevp[i], &eep[i]);
345 
346 		fd = kevp[i].ident;
347 		KASSERT(fd <= emd->fdc, ("epoll user data vector"
348 						    " is too small.\n"));
349 		eep[i].data = emd->udata[fd];
350 	}
351 	LINUX_PEM_SUNLOCK(pem);
352 
353 	error = copyout(eep, args->leventlist, count * sizeof(*eep));
354 	if (error == 0) {
355 		args->leventlist += count;
356 		args->count += count;
357 	} else if (args->error == 0)
358 		args->error = error;
359 
360 	free(eep, M_EPOLL);
361 	return (error);
362 }
363 
364 /*
365  * Copyin callback used by kevent. This copies already
366  * converted filters from kernel memory to the kevent
367  * internal kernel memory. Hence the memcpy instead of
368  * copyin.
369  */
370 static int
371 epoll_kev_copyin(void *arg, struct kevent *kevp, int count)
372 {
373 	struct epoll_copyin_args *args;
374 
375 	args = (struct epoll_copyin_args*) arg;
376 
377 	memcpy(kevp, args->changelist, count * sizeof(*kevp));
378 	args->changelist += count;
379 
380 	return (0);
381 }
382 
383 /*
384  * Load epoll filter, convert it to kevent filter
385  * and load it into kevent subsystem.
386  */
387 int
388 linux_epoll_ctl(struct thread *td, struct linux_epoll_ctl_args *args)
389 {
390 	struct file *epfp, *fp;
391 	struct epoll_copyin_args ciargs;
392 	struct kevent kev[2];
393 	struct kevent_copyops k_ops = { &ciargs,
394 					NULL,
395 					epoll_kev_copyin};
396 	struct epoll_event le;
397 	cap_rights_t rights;
398 	int kev_flags;
399 	int nchanges = 0;
400 	int error;
401 
402 	if (args->op != LINUX_EPOLL_CTL_DEL) {
403 		error = copyin(args->event, &le, sizeof(le));
404 		if (error != 0)
405 			return (error);
406 	}
407 
408 	error = fget(td, args->epfd,
409 	    cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &epfp);
410 	if (error != 0)
411 		return (error);
412 	if (epfp->f_type != DTYPE_KQUEUE)
413 		goto leave1;
414 
415 	 /* Protect user data vector from incorrectly supplied fd. */
416 	error = fget(td, args->fd, cap_rights_init(&rights, CAP_POLL_EVENT), &fp);
417 	if (error != 0)
418 		goto leave1;
419 
420 	/* Linux disallows spying on himself */
421 	if (epfp == fp) {
422 		error = EINVAL;
423 		goto leave0;
424 	}
425 
426 	ciargs.changelist = kev;
427 
428 	switch (args->op) {
429 	case LINUX_EPOLL_CTL_MOD:
430 		/*
431 		 * We don't memorize which events were set for this FD
432 		 * on this level, so just delete all we could have set:
433 		 * EVFILT_READ and EVFILT_WRITE, ignoring any errors
434 		 */
435 		error = epoll_delete_all_events(td, epfp, args->fd);
436 		if (error)
437 			goto leave0;
438 		/* FALLTHROUGH */
439 
440 	case LINUX_EPOLL_CTL_ADD:
441 			kev_flags = EV_ADD | EV_ENABLE;
442 		break;
443 
444 	case LINUX_EPOLL_CTL_DEL:
445 		/* CTL_DEL means unregister this fd with this epoll */
446 		error = epoll_delete_all_events(td, epfp, args->fd);
447 		goto leave0;
448 
449 	default:
450 		error = EINVAL;
451 		goto leave0;
452 	}
453 
454 	error = epoll_to_kevent(td, epfp, args->fd, &le, &kev_flags,
455 	    kev, &nchanges);
456 	if (error)
457 		goto leave0;
458 
459 	epoll_fd_install(td, args->fd, le.data);
460 
461 	error = kern_kevent_fp(td, epfp, nchanges, 0, &k_ops, NULL);
462 
463 leave0:
464 	fdrop(fp, td);
465 
466 leave1:
467 	fdrop(epfp, td);
468 	return (error);
469 }
470 
471 /*
472  * Wait for a filter to be triggered on the epoll file descriptor.
473  */
474 int
475 linux_epoll_wait(struct thread *td, struct linux_epoll_wait_args *args)
476 {
477 	struct file *epfp;
478 	struct timespec ts, *tsp;
479 	cap_rights_t rights;
480 	struct epoll_copyout_args coargs;
481 	struct kevent_copyops k_ops = { &coargs,
482 					epoll_kev_copyout,
483 					NULL};
484 	int error;
485 
486 	if (args->maxevents <= 0 || args->maxevents > LINUX_MAX_EVENTS)
487 		return (EINVAL);
488 
489 	error = fget(td, args->epfd,
490 	    cap_rights_init(&rights, CAP_KQUEUE_EVENT), &epfp);
491 	if (error != 0)
492 		return (error);
493 
494 	coargs.leventlist = args->events;
495 	coargs.p = td->td_proc;
496 	coargs.count = 0;
497 	coargs.error = 0;
498 
499 	if (args->timeout != -1) {
500 		if (args->timeout < 0) {
501 			error = EINVAL;
502 			goto leave;
503 		}
504 		/* Convert from milliseconds to timespec. */
505 		ts.tv_sec = args->timeout / 1000;
506 		ts.tv_nsec = (args->timeout % 1000) * 1000000;
507 		tsp = &ts;
508 	} else {
509 		tsp = NULL;
510 	}
511 
512 	error = kern_kevent_fp(td, epfp, 0, args->maxevents, &k_ops, tsp);
513 	if (error == 0 && coargs.error != 0)
514 		error = coargs.error;
515 
516 	/*
517 	 * kern_kevent might return ENOMEM which is not expected from epoll_wait.
518 	 * Maybe we should translate that but I don't think it matters at all.
519 	 */
520 	if (error == 0)
521 		td->td_retval[0] = coargs.count;
522 leave:
523 	fdrop(epfp, td);
524 	return (error);
525 }
526 
527 static int
528 epoll_delete_event(struct thread *td, struct file *epfp, int fd, int filter)
529 {
530 	struct epoll_copyin_args ciargs;
531 	struct kevent kev;
532 	struct kevent_copyops k_ops = { &ciargs,
533 					NULL,
534 					epoll_kev_copyin};
535 	int error;
536 
537 	ciargs.changelist = &kev;
538 	EV_SET(&kev, fd, filter, EV_DELETE | EV_DISABLE, 0, 0, 0);
539 
540 	error = kern_kevent_fp(td, epfp, 1, 0, &k_ops, NULL);
541 
542 	/*
543 	 * here we ignore ENONT, because we don't keep track of events here
544 	 */
545 	if (error == ENOENT)
546 		error = 0;
547 	return (error);
548 }
549 
550 static int
551 epoll_delete_all_events(struct thread *td, struct file *epfp, int fd)
552 {
553 	int error1, error2;
554 
555 	error1 = epoll_delete_event(td, epfp, fd, EVFILT_READ);
556 	error2 = epoll_delete_event(td, epfp, fd, EVFILT_WRITE);
557 
558 	/* report any errors we got */
559 	return (error1 == 0 ? error2 : error1);
560 }
561 
562 static int
563 eventfd_create(struct thread *td, uint32_t initval, int flags)
564 {
565 	struct filedesc *fdp;
566 	struct eventfd *efd;
567 	struct file *fp;
568 	int fflags, fd, error;
569 
570 	fflags = 0;
571 	if ((flags & LINUX_O_CLOEXEC) != 0)
572 		fflags |= O_CLOEXEC;
573 
574 	fdp = td->td_proc->p_fd;
575 	error = falloc(td, &fp, &fd, fflags);
576 	if (error)
577 		return (error);
578 
579 	efd = malloc(sizeof(*efd), M_EPOLL, M_WAITOK | M_ZERO);
580 	efd->efd_flags = flags;
581 	efd->efd_count = initval;
582 	mtx_init(&efd->efd_lock, "eventfd", NULL, MTX_DEF);
583 
584 	knlist_init_mtx(&efd->efd_sel.si_note, &efd->efd_lock);
585 
586 	fflags = FREAD | FWRITE;
587 	if ((flags & LINUX_O_NONBLOCK) != 0)
588 		fflags |= FNONBLOCK;
589 
590 	finit(fp, fflags, DTYPE_LINUXEFD, efd, &eventfdops);
591 	fdrop(fp, td);
592 
593 	td->td_retval[0] = fd;
594 	return (error);
595 }
596 
597 int
598 linux_eventfd(struct thread *td, struct linux_eventfd_args *args)
599 {
600 
601 	return (eventfd_create(td, args->initval, 0));
602 }
603 
604 int
605 linux_eventfd2(struct thread *td, struct linux_eventfd2_args *args)
606 {
607 
608 	if ((args->flags & ~(LINUX_O_CLOEXEC|LINUX_O_NONBLOCK|LINUX_EFD_SEMAPHORE)) != 0)
609 		return (EINVAL);
610 
611 	return (eventfd_create(td, args->initval, args->flags));
612 }
613 
614 static int
615 eventfd_close(struct file *fp, struct thread *td)
616 {
617 	struct eventfd *efd;
618 
619 	efd = fp->f_data;
620 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
621 		return (EBADF);
622 
623 	seldrain(&efd->efd_sel);
624 	knlist_destroy(&efd->efd_sel.si_note);
625 
626 	fp->f_ops = &badfileops;
627 	mtx_destroy(&efd->efd_lock);
628 	free(efd, M_EPOLL);
629 
630 	return (0);
631 }
632 
633 static int
634 eventfd_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
635 	int flags, struct thread *td)
636 {
637 	struct eventfd *efd;
638 	eventfd_t count;
639 	int error;
640 
641 	efd = fp->f_data;
642 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
643 		return (EBADF);
644 
645 	if (uio->uio_resid < sizeof(eventfd_t))
646 		return (EINVAL);
647 
648 	error = 0;
649 	mtx_lock(&efd->efd_lock);
650 retry:
651 	if (efd->efd_count == 0) {
652 		if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
653 			mtx_unlock(&efd->efd_lock);
654 			return (EAGAIN);
655 		}
656 		error = mtx_sleep(&efd->efd_count, &efd->efd_lock, PCATCH, "lefdrd", 0);
657 		if (error == 0)
658 			goto retry;
659 	}
660 	if (error == 0) {
661 		if ((efd->efd_flags & LINUX_EFD_SEMAPHORE) != 0) {
662 			count = 1;
663 			--efd->efd_count;
664 		} else {
665 			count = efd->efd_count;
666 			efd->efd_count = 0;
667 		}
668 		KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
669 		selwakeup(&efd->efd_sel);
670 		wakeup(&efd->efd_count);
671 		mtx_unlock(&efd->efd_lock);
672 		error = uiomove(&count, sizeof(eventfd_t), uio);
673 	} else
674 		mtx_unlock(&efd->efd_lock);
675 
676 	return (error);
677 }
678 
679 static int
680 eventfd_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
681 	 int flags, struct thread *td)
682 {
683 	struct eventfd *efd;
684 	eventfd_t count;
685 	int error;
686 
687 	efd = fp->f_data;
688 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
689 		return (EBADF);
690 
691 	if (uio->uio_resid < sizeof(eventfd_t))
692 		return (EINVAL);
693 
694 	error = uiomove(&count, sizeof(eventfd_t), uio);
695 	if (error)
696 		return (error);
697 	if (count == UINT64_MAX)
698 		return (EINVAL);
699 
700 	mtx_lock(&efd->efd_lock);
701 retry:
702 	if (UINT64_MAX - efd->efd_count <= count) {
703 		if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
704 			mtx_unlock(&efd->efd_lock);
705 			return (EAGAIN);
706 		}
707 		error = mtx_sleep(&efd->efd_count, &efd->efd_lock,
708 		    PCATCH, "lefdwr", 0);
709 		if (error == 0)
710 			goto retry;
711 	}
712 	if (error == 0) {
713 		efd->efd_count += count;
714 		KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
715 		selwakeup(&efd->efd_sel);
716 		wakeup(&efd->efd_count);
717 	}
718 	mtx_unlock(&efd->efd_lock);
719 
720 	return (error);
721 }
722 
723 static int
724 eventfd_poll(struct file *fp, int events, struct ucred *active_cred,
725 	struct thread *td)
726 {
727 	struct eventfd *efd;
728 	int revents = 0;
729 
730 	efd = fp->f_data;
731 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
732 		return (POLLERR);
733 
734 	mtx_lock(&efd->efd_lock);
735 	if ((events & (POLLIN|POLLRDNORM)) && efd->efd_count > 0)
736 		revents |= events & (POLLIN|POLLRDNORM);
737 	if ((events & (POLLOUT|POLLWRNORM)) && UINT64_MAX - 1 > efd->efd_count)
738 		revents |= events & (POLLOUT|POLLWRNORM);
739 	if (revents == 0)
740 		selrecord(td, &efd->efd_sel);
741 	mtx_unlock(&efd->efd_lock);
742 
743 	return (revents);
744 }
745 
746 /*ARGSUSED*/
747 static int
748 eventfd_kqfilter(struct file *fp, struct knote *kn)
749 {
750 	struct eventfd *efd;
751 
752 	efd = fp->f_data;
753 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
754 		return (EINVAL);
755 
756 	mtx_lock(&efd->efd_lock);
757 	switch (kn->kn_filter) {
758 	case EVFILT_READ:
759 		kn->kn_fop = &eventfd_rfiltops;
760 		break;
761 	case EVFILT_WRITE:
762 		kn->kn_fop = &eventfd_wfiltops;
763 		break;
764 	default:
765 		mtx_unlock(&efd->efd_lock);
766 		return (EINVAL);
767 	}
768 
769 	kn->kn_hook = efd;
770 	knlist_add(&efd->efd_sel.si_note, kn, 1);
771 	mtx_unlock(&efd->efd_lock);
772 
773 	return (0);
774 }
775 
776 static void
777 filt_eventfddetach(struct knote *kn)
778 {
779 	struct eventfd *efd = kn->kn_hook;
780 
781 	mtx_lock(&efd->efd_lock);
782 	knlist_remove(&efd->efd_sel.si_note, kn, 1);
783 	mtx_unlock(&efd->efd_lock);
784 }
785 
786 /*ARGSUSED*/
787 static int
788 filt_eventfdread(struct knote *kn, long hint)
789 {
790 	struct eventfd *efd = kn->kn_hook;
791 	int ret;
792 
793 	mtx_assert(&efd->efd_lock, MA_OWNED);
794 	ret = (efd->efd_count > 0);
795 
796 	return (ret);
797 }
798 
799 /*ARGSUSED*/
800 static int
801 filt_eventfdwrite(struct knote *kn, long hint)
802 {
803 	struct eventfd *efd = kn->kn_hook;
804 	int ret;
805 
806 	mtx_assert(&efd->efd_lock, MA_OWNED);
807 	ret = (UINT64_MAX - 1 > efd->efd_count);
808 
809 	return (ret);
810 }
811 
812 /*ARGSUSED*/
813 static int
814 eventfd_truncate(struct file *fp, off_t length, struct ucred *active_cred,
815 	struct thread *td)
816 {
817 
818 	return (ENXIO);
819 }
820 
821 /*ARGSUSED*/
822 static int
823 eventfd_ioctl(struct file *fp, u_long cmd, void *data,
824 	struct ucred *active_cred, struct thread *td)
825 {
826 
827 	return (ENXIO);
828 }
829 
830 /*ARGSUSED*/
831 static int
832 eventfd_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
833 	struct thread *td)
834 {
835 
836 	return (ENXIO);
837 }
838 
839 /*ARGSUSED*/
840 static int
841 eventfd_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
842 {
843 
844 	kif->kf_type = KF_TYPE_UNKNOWN;
845 	return (0);
846 }
847