xref: /freebsd/sys/compat/linux/linux_event.c (revision 5f0216bd883edee71bf81051e3c20505e4820903)
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 static int
475 linux_epoll_wait_common(struct thread *td, int epfd, struct epoll_event *events,
476     int maxevents, int timeout, sigset_t *uset)
477 {
478 	struct file *epfp;
479 	struct timespec ts, *tsp;
480 	cap_rights_t rights;
481 	struct epoll_copyout_args coargs;
482 	struct kevent_copyops k_ops = { &coargs,
483 					epoll_kev_copyout,
484 					NULL};
485 	int error;
486 
487 	if (maxevents <= 0 || maxevents > LINUX_MAX_EVENTS)
488 		return (EINVAL);
489 
490 	if (uset != NULL) {
491 		error = kern_sigprocmask(td, SIG_SETMASK, uset,
492 		    &td->td_oldsigmask, 0);
493 		if (error != 0)
494 			return (error);
495 		td->td_pflags |= TDP_OLDMASK;
496 		/*
497 		 * Make sure that ast() is called on return to
498 		 * usermode and TDP_OLDMASK is cleared, restoring old
499 		 * sigmask.
500 		 */
501 		thread_lock(td);
502 		td->td_flags |= TDF_ASTPENDING;
503 		thread_unlock(td);
504 	}
505 
506 	error = fget(td, epfd,
507 	    cap_rights_init(&rights, CAP_KQUEUE_EVENT), &epfp);
508 	if (error != 0)
509 		return (error);
510 
511 	coargs.leventlist = events;
512 	coargs.p = td->td_proc;
513 	coargs.count = 0;
514 	coargs.error = 0;
515 
516 	if (timeout != -1) {
517 		if (timeout < 0) {
518 			error = EINVAL;
519 			goto leave;
520 		}
521 		/* Convert from milliseconds to timespec. */
522 		ts.tv_sec = timeout / 1000;
523 		ts.tv_nsec = (timeout % 1000) * 1000000;
524 		tsp = &ts;
525 	} else {
526 		tsp = NULL;
527 	}
528 
529 	error = kern_kevent_fp(td, epfp, 0, maxevents, &k_ops, tsp);
530 	if (error == 0 && coargs.error != 0)
531 		error = coargs.error;
532 
533 	/*
534 	 * kern_kevent might return ENOMEM which is not expected from epoll_wait.
535 	 * Maybe we should translate that but I don't think it matters at all.
536 	 */
537 	if (error == 0)
538 		td->td_retval[0] = coargs.count;
539 leave:
540 	fdrop(epfp, td);
541 	return (error);
542 }
543 
544 int
545 linux_epoll_wait(struct thread *td, struct linux_epoll_wait_args *args)
546 {
547 
548 	return (linux_epoll_wait_common(td, args->epfd, args->events,
549 	    args->maxevents, args->timeout, NULL));
550 }
551 
552 int
553 linux_epoll_pwait(struct thread *td, struct linux_epoll_pwait_args *args)
554 {
555 	sigset_t mask, *pmask;
556 	l_sigset_t lmask;
557 	int error;
558 
559 	if (args->mask != NULL) {
560 		error = copyin(args->mask, &lmask, sizeof(l_sigset_t));
561 		if (error != 0)
562 			return (error);
563 		linux_to_bsd_sigset(&lmask, &mask);
564 		pmask = &mask;
565 	} else
566 		pmask = NULL;
567 	return (linux_epoll_wait_common(td, args->epfd, args->events,
568 	    args->maxevents, args->timeout, pmask));
569 }
570 
571 static int
572 epoll_delete_event(struct thread *td, struct file *epfp, int fd, int filter)
573 {
574 	struct epoll_copyin_args ciargs;
575 	struct kevent kev;
576 	struct kevent_copyops k_ops = { &ciargs,
577 					NULL,
578 					epoll_kev_copyin};
579 	int error;
580 
581 	ciargs.changelist = &kev;
582 	EV_SET(&kev, fd, filter, EV_DELETE | EV_DISABLE, 0, 0, 0);
583 
584 	error = kern_kevent_fp(td, epfp, 1, 0, &k_ops, NULL);
585 
586 	/*
587 	 * here we ignore ENONT, because we don't keep track of events here
588 	 */
589 	if (error == ENOENT)
590 		error = 0;
591 	return (error);
592 }
593 
594 static int
595 epoll_delete_all_events(struct thread *td, struct file *epfp, int fd)
596 {
597 	int error1, error2;
598 
599 	error1 = epoll_delete_event(td, epfp, fd, EVFILT_READ);
600 	error2 = epoll_delete_event(td, epfp, fd, EVFILT_WRITE);
601 
602 	/* report any errors we got */
603 	return (error1 == 0 ? error2 : error1);
604 }
605 
606 static int
607 eventfd_create(struct thread *td, uint32_t initval, int flags)
608 {
609 	struct filedesc *fdp;
610 	struct eventfd *efd;
611 	struct file *fp;
612 	int fflags, fd, error;
613 
614 	fflags = 0;
615 	if ((flags & LINUX_O_CLOEXEC) != 0)
616 		fflags |= O_CLOEXEC;
617 
618 	fdp = td->td_proc->p_fd;
619 	error = falloc(td, &fp, &fd, fflags);
620 	if (error)
621 		return (error);
622 
623 	efd = malloc(sizeof(*efd), M_EPOLL, M_WAITOK | M_ZERO);
624 	efd->efd_flags = flags;
625 	efd->efd_count = initval;
626 	mtx_init(&efd->efd_lock, "eventfd", NULL, MTX_DEF);
627 
628 	knlist_init_mtx(&efd->efd_sel.si_note, &efd->efd_lock);
629 
630 	fflags = FREAD | FWRITE;
631 	if ((flags & LINUX_O_NONBLOCK) != 0)
632 		fflags |= FNONBLOCK;
633 
634 	finit(fp, fflags, DTYPE_LINUXEFD, efd, &eventfdops);
635 	fdrop(fp, td);
636 
637 	td->td_retval[0] = fd;
638 	return (error);
639 }
640 
641 int
642 linux_eventfd(struct thread *td, struct linux_eventfd_args *args)
643 {
644 
645 	return (eventfd_create(td, args->initval, 0));
646 }
647 
648 int
649 linux_eventfd2(struct thread *td, struct linux_eventfd2_args *args)
650 {
651 
652 	if ((args->flags & ~(LINUX_O_CLOEXEC|LINUX_O_NONBLOCK|LINUX_EFD_SEMAPHORE)) != 0)
653 		return (EINVAL);
654 
655 	return (eventfd_create(td, args->initval, args->flags));
656 }
657 
658 static int
659 eventfd_close(struct file *fp, struct thread *td)
660 {
661 	struct eventfd *efd;
662 
663 	efd = fp->f_data;
664 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
665 		return (EBADF);
666 
667 	seldrain(&efd->efd_sel);
668 	knlist_destroy(&efd->efd_sel.si_note);
669 
670 	fp->f_ops = &badfileops;
671 	mtx_destroy(&efd->efd_lock);
672 	free(efd, M_EPOLL);
673 
674 	return (0);
675 }
676 
677 static int
678 eventfd_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
679 	int flags, struct thread *td)
680 {
681 	struct eventfd *efd;
682 	eventfd_t count;
683 	int error;
684 
685 	efd = fp->f_data;
686 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
687 		return (EBADF);
688 
689 	if (uio->uio_resid < sizeof(eventfd_t))
690 		return (EINVAL);
691 
692 	error = 0;
693 	mtx_lock(&efd->efd_lock);
694 retry:
695 	if (efd->efd_count == 0) {
696 		if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
697 			mtx_unlock(&efd->efd_lock);
698 			return (EAGAIN);
699 		}
700 		error = mtx_sleep(&efd->efd_count, &efd->efd_lock, PCATCH, "lefdrd", 0);
701 		if (error == 0)
702 			goto retry;
703 	}
704 	if (error == 0) {
705 		if ((efd->efd_flags & LINUX_EFD_SEMAPHORE) != 0) {
706 			count = 1;
707 			--efd->efd_count;
708 		} else {
709 			count = efd->efd_count;
710 			efd->efd_count = 0;
711 		}
712 		KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
713 		selwakeup(&efd->efd_sel);
714 		wakeup(&efd->efd_count);
715 		mtx_unlock(&efd->efd_lock);
716 		error = uiomove(&count, sizeof(eventfd_t), uio);
717 	} else
718 		mtx_unlock(&efd->efd_lock);
719 
720 	return (error);
721 }
722 
723 static int
724 eventfd_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
725 	 int flags, struct thread *td)
726 {
727 	struct eventfd *efd;
728 	eventfd_t count;
729 	int error;
730 
731 	efd = fp->f_data;
732 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
733 		return (EBADF);
734 
735 	if (uio->uio_resid < sizeof(eventfd_t))
736 		return (EINVAL);
737 
738 	error = uiomove(&count, sizeof(eventfd_t), uio);
739 	if (error)
740 		return (error);
741 	if (count == UINT64_MAX)
742 		return (EINVAL);
743 
744 	mtx_lock(&efd->efd_lock);
745 retry:
746 	if (UINT64_MAX - efd->efd_count <= count) {
747 		if ((efd->efd_flags & LINUX_O_NONBLOCK) != 0) {
748 			mtx_unlock(&efd->efd_lock);
749 			return (EAGAIN);
750 		}
751 		error = mtx_sleep(&efd->efd_count, &efd->efd_lock,
752 		    PCATCH, "lefdwr", 0);
753 		if (error == 0)
754 			goto retry;
755 	}
756 	if (error == 0) {
757 		efd->efd_count += count;
758 		KNOTE_LOCKED(&efd->efd_sel.si_note, 0);
759 		selwakeup(&efd->efd_sel);
760 		wakeup(&efd->efd_count);
761 	}
762 	mtx_unlock(&efd->efd_lock);
763 
764 	return (error);
765 }
766 
767 static int
768 eventfd_poll(struct file *fp, int events, struct ucred *active_cred,
769 	struct thread *td)
770 {
771 	struct eventfd *efd;
772 	int revents = 0;
773 
774 	efd = fp->f_data;
775 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
776 		return (POLLERR);
777 
778 	mtx_lock(&efd->efd_lock);
779 	if ((events & (POLLIN|POLLRDNORM)) && efd->efd_count > 0)
780 		revents |= events & (POLLIN|POLLRDNORM);
781 	if ((events & (POLLOUT|POLLWRNORM)) && UINT64_MAX - 1 > efd->efd_count)
782 		revents |= events & (POLLOUT|POLLWRNORM);
783 	if (revents == 0)
784 		selrecord(td, &efd->efd_sel);
785 	mtx_unlock(&efd->efd_lock);
786 
787 	return (revents);
788 }
789 
790 /*ARGSUSED*/
791 static int
792 eventfd_kqfilter(struct file *fp, struct knote *kn)
793 {
794 	struct eventfd *efd;
795 
796 	efd = fp->f_data;
797 	if (fp->f_type != DTYPE_LINUXEFD || efd == NULL)
798 		return (EINVAL);
799 
800 	mtx_lock(&efd->efd_lock);
801 	switch (kn->kn_filter) {
802 	case EVFILT_READ:
803 		kn->kn_fop = &eventfd_rfiltops;
804 		break;
805 	case EVFILT_WRITE:
806 		kn->kn_fop = &eventfd_wfiltops;
807 		break;
808 	default:
809 		mtx_unlock(&efd->efd_lock);
810 		return (EINVAL);
811 	}
812 
813 	kn->kn_hook = efd;
814 	knlist_add(&efd->efd_sel.si_note, kn, 1);
815 	mtx_unlock(&efd->efd_lock);
816 
817 	return (0);
818 }
819 
820 static void
821 filt_eventfddetach(struct knote *kn)
822 {
823 	struct eventfd *efd = kn->kn_hook;
824 
825 	mtx_lock(&efd->efd_lock);
826 	knlist_remove(&efd->efd_sel.si_note, kn, 1);
827 	mtx_unlock(&efd->efd_lock);
828 }
829 
830 /*ARGSUSED*/
831 static int
832 filt_eventfdread(struct knote *kn, long hint)
833 {
834 	struct eventfd *efd = kn->kn_hook;
835 	int ret;
836 
837 	mtx_assert(&efd->efd_lock, MA_OWNED);
838 	ret = (efd->efd_count > 0);
839 
840 	return (ret);
841 }
842 
843 /*ARGSUSED*/
844 static int
845 filt_eventfdwrite(struct knote *kn, long hint)
846 {
847 	struct eventfd *efd = kn->kn_hook;
848 	int ret;
849 
850 	mtx_assert(&efd->efd_lock, MA_OWNED);
851 	ret = (UINT64_MAX - 1 > efd->efd_count);
852 
853 	return (ret);
854 }
855 
856 /*ARGSUSED*/
857 static int
858 eventfd_truncate(struct file *fp, off_t length, struct ucred *active_cred,
859 	struct thread *td)
860 {
861 
862 	return (ENXIO);
863 }
864 
865 /*ARGSUSED*/
866 static int
867 eventfd_ioctl(struct file *fp, u_long cmd, void *data,
868 	struct ucred *active_cred, struct thread *td)
869 {
870 
871 	return (ENXIO);
872 }
873 
874 /*ARGSUSED*/
875 static int
876 eventfd_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
877 	struct thread *td)
878 {
879 
880 	return (ENXIO);
881 }
882 
883 /*ARGSUSED*/
884 static int
885 eventfd_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
886 {
887 
888 	kif->kf_type = KF_TYPE_UNKNOWN;
889 	return (0);
890 }
891