xref: /freebsd/sys/compat/linux/linux_event.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
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, NULL);
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 	if ((levents & LINUX_EPOLLRDHUP) != 0)
264 		*kev_flags |= EV_EOF;
265 
266 	/* flags related to what event is registered */
267 	if ((levents & LINUX_EPOLL_EVRD) != 0) {
268 		EV_SET(kevent++, fd, EVFILT_READ, *kev_flags, 0, 0, 0);
269 		++(*nkevents);
270 	}
271 	if ((levents & LINUX_EPOLL_EVWR) != 0) {
272 		EV_SET(kevent++, fd, EVFILT_WRITE, *kev_flags, 0, 0, 0);
273 		++(*nkevents);
274 	}
275 
276 	if ((levents & ~(LINUX_EPOLL_EVSUP)) != 0) {
277 		p = td->td_proc;
278 
279 		pem = pem_find(p);
280 		KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
281 		KASSERT(pem->epoll != NULL, ("epoll proc epolldata not found.\n"));
282 
283 		LINUX_PEM_XLOCK(pem);
284 		if ((pem->flags & LINUX_XUNSUP_EPOLL) == 0) {
285 			pem->flags |= LINUX_XUNSUP_EPOLL;
286 			LINUX_PEM_XUNLOCK(pem);
287 			linux_msg(td, "epoll_ctl unsupported flags: 0x%x\n",
288 			    levents);
289 		} else
290 			LINUX_PEM_XUNLOCK(pem);
291 		return (EINVAL);
292 	}
293 
294 	return (0);
295 }
296 
297 /*
298  * Structure converting function from kevent to epoll. In a case
299  * this is called on error in registration we store the error in
300  * event->data and pick it up later in linux_epoll_ctl().
301  */
302 static void
303 kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event)
304 {
305 
306 	if ((kevent->flags & EV_ERROR) != 0) {
307 		l_event->events = LINUX_EPOLLERR;
308 		return;
309 	}
310 
311 	switch (kevent->filter) {
312 	case EVFILT_READ:
313 		l_event->events = LINUX_EPOLLIN|LINUX_EPOLLRDNORM|LINUX_EPOLLPRI;
314 		if ((kevent->flags & EV_EOF) != 0)
315 			l_event->events |= LINUX_EPOLLRDHUP;
316 	break;
317 	case EVFILT_WRITE:
318 		l_event->events = LINUX_EPOLLOUT|LINUX_EPOLLWRNORM;
319 	break;
320 	}
321 }
322 
323 /*
324  * Copyout callback used by kevent. This converts kevent
325  * events to epoll events and copies them back to the
326  * userspace. This is also called on error on registering
327  * of the filter.
328  */
329 static int
330 epoll_kev_copyout(void *arg, struct kevent *kevp, int count)
331 {
332 	struct epoll_copyout_args *args;
333 	struct linux_pemuldata *pem;
334 	struct epoll_emuldata *emd;
335 	struct epoll_event *eep;
336 	int error, fd, i;
337 
338 	args = (struct epoll_copyout_args*) arg;
339 	eep = malloc(sizeof(*eep) * count, M_EPOLL, M_WAITOK | M_ZERO);
340 
341 	pem = pem_find(args->p);
342 	KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
343 	LINUX_PEM_SLOCK(pem);
344 	emd = pem->epoll;
345 	KASSERT(emd != NULL, ("epoll proc epolldata not found.\n"));
346 
347 	for (i = 0; i < count; i++) {
348 		kevent_to_epoll(&kevp[i], &eep[i]);
349 
350 		fd = kevp[i].ident;
351 		KASSERT(fd <= emd->fdc, ("epoll user data vector"
352 						    " is too small.\n"));
353 		eep[i].data = emd->udata[fd];
354 	}
355 	LINUX_PEM_SUNLOCK(pem);
356 
357 	error = copyout(eep, args->leventlist, count * sizeof(*eep));
358 	if (error == 0) {
359 		args->leventlist += count;
360 		args->count += count;
361 	} else if (args->error == 0)
362 		args->error = error;
363 
364 	free(eep, M_EPOLL);
365 	return (error);
366 }
367 
368 /*
369  * Copyin callback used by kevent. This copies already
370  * converted filters from kernel memory to the kevent
371  * internal kernel memory. Hence the memcpy instead of
372  * copyin.
373  */
374 static int
375 epoll_kev_copyin(void *arg, struct kevent *kevp, int count)
376 {
377 	struct epoll_copyin_args *args;
378 
379 	args = (struct epoll_copyin_args*) arg;
380 
381 	memcpy(kevp, args->changelist, count * sizeof(*kevp));
382 	args->changelist += count;
383 
384 	return (0);
385 }
386 
387 /*
388  * Load epoll filter, convert it to kevent filter
389  * and load it into kevent subsystem.
390  */
391 int
392 linux_epoll_ctl(struct thread *td, struct linux_epoll_ctl_args *args)
393 {
394 	struct file *epfp, *fp;
395 	struct epoll_copyin_args ciargs;
396 	struct kevent kev[2];
397 	struct kevent_copyops k_ops = { &ciargs,
398 					NULL,
399 					epoll_kev_copyin};
400 	struct epoll_event le;
401 	cap_rights_t rights;
402 	int kev_flags;
403 	int nchanges = 0;
404 	int error;
405 
406 	if (args->op != LINUX_EPOLL_CTL_DEL) {
407 		error = copyin(args->event, &le, sizeof(le));
408 		if (error != 0)
409 			return (error);
410 	}
411 
412 	error = fget(td, args->epfd,
413 	    cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &epfp);
414 	if (error != 0)
415 		return (error);
416 	if (epfp->f_type != DTYPE_KQUEUE)
417 		goto leave1;
418 
419 	 /* Protect user data vector from incorrectly supplied fd. */
420 	error = fget(td, args->fd, cap_rights_init(&rights, CAP_POLL_EVENT), &fp);
421 	if (error != 0)
422 		goto leave1;
423 
424 	/* Linux disallows spying on himself */
425 	if (epfp == fp) {
426 		error = EINVAL;
427 		goto leave0;
428 	}
429 
430 	ciargs.changelist = kev;
431 
432 	switch (args->op) {
433 	case LINUX_EPOLL_CTL_MOD:
434 		/*
435 		 * We don't memorize which events were set for this FD
436 		 * on this level, so just delete all we could have set:
437 		 * EVFILT_READ and EVFILT_WRITE, ignoring any errors
438 		 */
439 		error = epoll_delete_all_events(td, epfp, args->fd);
440 		if (error)
441 			goto leave0;
442 		/* FALLTHROUGH */
443 
444 	case LINUX_EPOLL_CTL_ADD:
445 			kev_flags = EV_ADD | EV_ENABLE;
446 		break;
447 
448 	case LINUX_EPOLL_CTL_DEL:
449 		/* CTL_DEL means unregister this fd with this epoll */
450 		error = epoll_delete_all_events(td, epfp, args->fd);
451 		goto leave0;
452 
453 	default:
454 		error = EINVAL;
455 		goto leave0;
456 	}
457 
458 	error = epoll_to_kevent(td, epfp, args->fd, &le, &kev_flags,
459 	    kev, &nchanges);
460 	if (error)
461 		goto leave0;
462 
463 	epoll_fd_install(td, args->fd, le.data);
464 
465 	error = kern_kevent_fp(td, epfp, nchanges, 0, &k_ops, NULL);
466 
467 leave0:
468 	fdrop(fp, td);
469 
470 leave1:
471 	fdrop(epfp, td);
472 	return (error);
473 }
474 
475 /*
476  * Wait for a filter to be triggered on the epoll file descriptor.
477  */
478 static int
479 linux_epoll_wait_common(struct thread *td, int epfd, struct epoll_event *events,
480     int maxevents, int timeout, sigset_t *uset)
481 {
482 	struct file *epfp;
483 	struct timespec ts, *tsp;
484 	cap_rights_t rights;
485 	struct epoll_copyout_args coargs;
486 	struct kevent_copyops k_ops = { &coargs,
487 					epoll_kev_copyout,
488 					NULL};
489 	int error;
490 
491 	if (maxevents <= 0 || maxevents > LINUX_MAX_EVENTS)
492 		return (EINVAL);
493 
494 	if (uset != NULL) {
495 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
496 		    &td->td_oldsigmask, 0);
497 		if (error != 0)
498 			return (error);
499 		td->td_pflags |= TDP_OLDMASK;
500 		/*
501 		 * Make sure that ast() is called on return to
502 		 * usermode and TDP_OLDMASK is cleared, restoring old
503 		 * sigmask.
504 		 */
505 		thread_lock(td);
506 		td->td_flags |= TDF_ASTPENDING;
507 		thread_unlock(td);
508 	}
509 
510 	error = fget(td, epfd,
511 	    cap_rights_init(&rights, CAP_KQUEUE_EVENT), &epfp);
512 	if (error != 0)
513 		return (error);
514 
515 	coargs.leventlist = events;
516 	coargs.p = td->td_proc;
517 	coargs.count = 0;
518 	coargs.error = 0;
519 
520 	if (timeout != -1) {
521 		if (timeout < 0) {
522 			error = EINVAL;
523 			goto leave;
524 		}
525 		/* Convert from milliseconds to timespec. */
526 		ts.tv_sec = timeout / 1000;
527 		ts.tv_nsec = (timeout % 1000) * 1000000;
528 		tsp = &ts;
529 	} else {
530 		tsp = NULL;
531 	}
532 
533 	error = kern_kevent_fp(td, epfp, 0, maxevents, &k_ops, tsp);
534 	if (error == 0 && coargs.error != 0)
535 		error = coargs.error;
536 
537 	/*
538 	 * kern_kevent might return ENOMEM which is not expected from epoll_wait.
539 	 * Maybe we should translate that but I don't think it matters at all.
540 	 */
541 	if (error == 0)
542 		td->td_retval[0] = coargs.count;
543 leave:
544 	fdrop(epfp, td);
545 	return (error);
546 }
547 
548 int
549 linux_epoll_wait(struct thread *td, struct linux_epoll_wait_args *args)
550 {
551 
552 	return (linux_epoll_wait_common(td, args->epfd, args->events,
553 	    args->maxevents, args->timeout, NULL));
554 }
555 
556 int
557 linux_epoll_pwait(struct thread *td, struct linux_epoll_pwait_args *args)
558 {
559 	sigset_t mask, *pmask;
560 	l_sigset_t lmask;
561 	int error;
562 
563 	if (args->mask != NULL) {
564 		error = copyin(args->mask, &lmask, sizeof(l_sigset_t));
565 		if (error != 0)
566 			return (error);
567 		linux_to_bsd_sigset(&lmask, &mask);
568 		pmask = &mask;
569 	} else
570 		pmask = NULL;
571 	return (linux_epoll_wait_common(td, args->epfd, args->events,
572 	    args->maxevents, args->timeout, pmask));
573 }
574 
575 static int
576 epoll_delete_event(struct thread *td, struct file *epfp, int fd, int filter)
577 {
578 	struct epoll_copyin_args ciargs;
579 	struct kevent kev;
580 	struct kevent_copyops k_ops = { &ciargs,
581 					NULL,
582 					epoll_kev_copyin};
583 	int error;
584 
585 	ciargs.changelist = &kev;
586 	EV_SET(&kev, fd, filter, EV_DELETE | EV_DISABLE, 0, 0, 0);
587 
588 	error = kern_kevent_fp(td, epfp, 1, 0, &k_ops, NULL);
589 
590 	/*
591 	 * here we ignore ENONT, because we don't keep track of events here
592 	 */
593 	if (error == ENOENT)
594 		error = 0;
595 	return (error);
596 }
597 
598 static int
599 epoll_delete_all_events(struct thread *td, struct file *epfp, int fd)
600 {
601 	int error1, error2;
602 
603 	error1 = epoll_delete_event(td, epfp, fd, EVFILT_READ);
604 	error2 = epoll_delete_event(td, epfp, fd, EVFILT_WRITE);
605 
606 	/* report any errors we got */
607 	return (error1 == 0 ? error2 : error1);
608 }
609 
610 static int
611 eventfd_create(struct thread *td, uint32_t initval, int flags)
612 {
613 	struct filedesc *fdp;
614 	struct eventfd *efd;
615 	struct file *fp;
616 	int fflags, fd, error;
617 
618 	fflags = 0;
619 	if ((flags & LINUX_O_CLOEXEC) != 0)
620 		fflags |= O_CLOEXEC;
621 
622 	fdp = td->td_proc->p_fd;
623 	error = falloc(td, &fp, &fd, fflags);
624 	if (error)
625 		return (error);
626 
627 	efd = malloc(sizeof(*efd), M_EPOLL, M_WAITOK | M_ZERO);
628 	efd->efd_flags = flags;
629 	efd->efd_count = initval;
630 	mtx_init(&efd->efd_lock, "eventfd", NULL, MTX_DEF);
631 
632 	knlist_init_mtx(&efd->efd_sel.si_note, &efd->efd_lock);
633 
634 	fflags = FREAD | FWRITE;
635 	if ((flags & LINUX_O_NONBLOCK) != 0)
636 		fflags |= FNONBLOCK;
637 
638 	finit(fp, fflags, DTYPE_LINUXEFD, efd, &eventfdops);
639 	fdrop(fp, td);
640 
641 	td->td_retval[0] = fd;
642 	return (error);
643 }
644 
645 int
646 linux_eventfd(struct thread *td, struct linux_eventfd_args *args)
647 {
648 
649 	return (eventfd_create(td, args->initval, 0));
650 }
651 
652 int
653 linux_eventfd2(struct thread *td, struct linux_eventfd2_args *args)
654 {
655 
656 	if ((args->flags & ~(LINUX_O_CLOEXEC|LINUX_O_NONBLOCK|LINUX_EFD_SEMAPHORE)) != 0)
657 		return (EINVAL);
658 
659 	return (eventfd_create(td, args->initval, args->flags));
660 }
661 
662 static int
663 eventfd_close(struct file *fp, struct thread *td)
664 {
665 	struct eventfd *efd;
666 
667 	efd = fp->f_data;
668 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
669 		return (EBADF);
670 
671 	seldrain(&efd->efd_sel);
672 	knlist_destroy(&efd->efd_sel.si_note);
673 
674 	fp->f_ops = &badfileops;
675 	mtx_destroy(&efd->efd_lock);
676 	free(efd, M_EPOLL);
677 
678 	return (0);
679 }
680 
681 static int
682 eventfd_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
683 	int flags, struct thread *td)
684 {
685 	struct eventfd *efd;
686 	eventfd_t count;
687 	int error;
688 
689 	efd = fp->f_data;
690 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
691 		return (EBADF);
692 
693 	if (uio->uio_resid < sizeof(eventfd_t))
694 		return (EINVAL);
695 
696 	error = 0;
697 	mtx_lock(&efd->efd_lock);
698 retry:
699 	if (efd->efd_count == 0) {
700 		if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
701 			mtx_unlock(&efd->efd_lock);
702 			return (EAGAIN);
703 		}
704 		error = mtx_sleep(&efd->efd_count, &efd->efd_lock, PCATCH, "lefdrd", 0);
705 		if (error == 0)
706 			goto retry;
707 	}
708 	if (error == 0) {
709 		if ((efd->efd_flags & LINUX_EFD_SEMAPHORE) != 0) {
710 			count = 1;
711 			--efd->efd_count;
712 		} else {
713 			count = efd->efd_count;
714 			efd->efd_count = 0;
715 		}
716 		KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
717 		selwakeup(&efd->efd_sel);
718 		wakeup(&efd->efd_count);
719 		mtx_unlock(&efd->efd_lock);
720 		error = uiomove(&count, sizeof(eventfd_t), uio);
721 	} else
722 		mtx_unlock(&efd->efd_lock);
723 
724 	return (error);
725 }
726 
727 static int
728 eventfd_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
729 	 int flags, struct thread *td)
730 {
731 	struct eventfd *efd;
732 	eventfd_t count;
733 	int error;
734 
735 	efd = fp->f_data;
736 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
737 		return (EBADF);
738 
739 	if (uio->uio_resid < sizeof(eventfd_t))
740 		return (EINVAL);
741 
742 	error = uiomove(&count, sizeof(eventfd_t), uio);
743 	if (error)
744 		return (error);
745 	if (count == UINT64_MAX)
746 		return (EINVAL);
747 
748 	mtx_lock(&efd->efd_lock);
749 retry:
750 	if (UINT64_MAX - efd->efd_count <= count) {
751 		if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
752 			mtx_unlock(&efd->efd_lock);
753 			return (EAGAIN);
754 		}
755 		error = mtx_sleep(&efd->efd_count, &efd->efd_lock,
756 		    PCATCH, "lefdwr", 0);
757 		if (error == 0)
758 			goto retry;
759 	}
760 	if (error == 0) {
761 		efd->efd_count += count;
762 		KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
763 		selwakeup(&efd->efd_sel);
764 		wakeup(&efd->efd_count);
765 	}
766 	mtx_unlock(&efd->efd_lock);
767 
768 	return (error);
769 }
770 
771 static int
772 eventfd_poll(struct file *fp, int events, struct ucred *active_cred,
773 	struct thread *td)
774 {
775 	struct eventfd *efd;
776 	int revents = 0;
777 
778 	efd = fp->f_data;
779 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
780 		return (POLLERR);
781 
782 	mtx_lock(&efd->efd_lock);
783 	if ((events & (POLLIN|POLLRDNORM)) && efd->efd_count > 0)
784 		revents |= events & (POLLIN|POLLRDNORM);
785 	if ((events & (POLLOUT|POLLWRNORM)) && UINT64_MAX - 1 > efd->efd_count)
786 		revents |= events & (POLLOUT|POLLWRNORM);
787 	if (revents == 0)
788 		selrecord(td, &efd->efd_sel);
789 	mtx_unlock(&efd->efd_lock);
790 
791 	return (revents);
792 }
793 
794 /*ARGSUSED*/
795 static int
796 eventfd_kqfilter(struct file *fp, struct knote *kn)
797 {
798 	struct eventfd *efd;
799 
800 	efd = fp->f_data;
801 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
802 		return (EINVAL);
803 
804 	mtx_lock(&efd->efd_lock);
805 	switch (kn->kn_filter) {
806 	case EVFILT_READ:
807 		kn->kn_fop = &eventfd_rfiltops;
808 		break;
809 	case EVFILT_WRITE:
810 		kn->kn_fop = &eventfd_wfiltops;
811 		break;
812 	default:
813 		mtx_unlock(&efd->efd_lock);
814 		return (EINVAL);
815 	}
816 
817 	kn->kn_hook = efd;
818 	knlist_add(&efd->efd_sel.si_note, kn, 1);
819 	mtx_unlock(&efd->efd_lock);
820 
821 	return (0);
822 }
823 
824 static void
825 filt_eventfddetach(struct knote *kn)
826 {
827 	struct eventfd *efd = kn->kn_hook;
828 
829 	mtx_lock(&efd->efd_lock);
830 	knlist_remove(&efd->efd_sel.si_note, kn, 1);
831 	mtx_unlock(&efd->efd_lock);
832 }
833 
834 /*ARGSUSED*/
835 static int
836 filt_eventfdread(struct knote *kn, long hint)
837 {
838 	struct eventfd *efd = kn->kn_hook;
839 	int ret;
840 
841 	mtx_assert(&efd->efd_lock, MA_OWNED);
842 	ret = (efd->efd_count > 0);
843 
844 	return (ret);
845 }
846 
847 /*ARGSUSED*/
848 static int
849 filt_eventfdwrite(struct knote *kn, long hint)
850 {
851 	struct eventfd *efd = kn->kn_hook;
852 	int ret;
853 
854 	mtx_assert(&efd->efd_lock, MA_OWNED);
855 	ret = (UINT64_MAX - 1 > efd->efd_count);
856 
857 	return (ret);
858 }
859 
860 /*ARGSUSED*/
861 static int
862 eventfd_truncate(struct file *fp, off_t length, struct ucred *active_cred,
863 	struct thread *td)
864 {
865 
866 	return (ENXIO);
867 }
868 
869 /*ARGSUSED*/
870 static int
871 eventfd_ioctl(struct file *fp, u_long cmd, void *data,
872 	struct ucred *active_cred, struct thread *td)
873 {
874 
875 	return (ENXIO);
876 }
877 
878 /*ARGSUSED*/
879 static int
880 eventfd_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
881 	struct thread *td)
882 {
883 
884 	return (ENXIO);
885 }
886 
887 /*ARGSUSED*/
888 static int
889 eventfd_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
890 {
891 
892 	kif->kf_type = KF_TYPE_UNKNOWN;
893 	return (0);
894 }
895