xref: /netbsd-src/sys/kern/kern_event.c (revision da9817918ec7e88db2912a2882967c7570a83f47)
1 /*	$NetBSD: kern_event.c,v 1.65 2009/05/24 21:41:26 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
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 /*-
33  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  *
57  * FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp
58  */
59 
60 #include <sys/cdefs.h>
61 __KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.65 2009/05/24 21:41:26 ad Exp $");
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/kernel.h>
66 #include <sys/proc.h>
67 #include <sys/file.h>
68 #include <sys/select.h>
69 #include <sys/queue.h>
70 #include <sys/event.h>
71 #include <sys/eventvar.h>
72 #include <sys/poll.h>
73 #include <sys/kmem.h>
74 #include <sys/stat.h>
75 #include <sys/filedesc.h>
76 #include <sys/syscallargs.h>
77 #include <sys/kauth.h>
78 #include <sys/conf.h>
79 #include <sys/atomic.h>
80 
81 static int	kqueue_scan(file_t *, size_t, struct kevent *,
82 			    const struct timespec *, register_t *,
83 			    const struct kevent_ops *, struct kevent *,
84 			    size_t);
85 static int	kqueue_ioctl(file_t *, u_long, void *);
86 static int	kqueue_fcntl(file_t *, u_int, void *);
87 static int	kqueue_poll(file_t *, int);
88 static int	kqueue_kqfilter(file_t *, struct knote *);
89 static int	kqueue_stat(file_t *, struct stat *);
90 static int	kqueue_close(file_t *);
91 static int	kqueue_register(struct kqueue *, struct kevent *);
92 static void	kqueue_doclose(struct kqueue *, struct klist *, int);
93 
94 static void	knote_detach(struct knote *, filedesc_t *fdp, bool);
95 static void	knote_enqueue(struct knote *);
96 static void	knote_activate(struct knote *);
97 
98 static void	filt_kqdetach(struct knote *);
99 static int	filt_kqueue(struct knote *, long hint);
100 static int	filt_procattach(struct knote *);
101 static void	filt_procdetach(struct knote *);
102 static int	filt_proc(struct knote *, long hint);
103 static int	filt_fileattach(struct knote *);
104 static void	filt_timerexpire(void *x);
105 static int	filt_timerattach(struct knote *);
106 static void	filt_timerdetach(struct knote *);
107 static int	filt_timer(struct knote *, long hint);
108 
109 static const struct fileops kqueueops = {
110 	.fo_read = (void *)enxio,
111 	.fo_write = (void *)enxio,
112 	.fo_ioctl = kqueue_ioctl,
113 	.fo_fcntl = kqueue_fcntl,
114 	.fo_poll = kqueue_poll,
115 	.fo_stat = kqueue_stat,
116 	.fo_close = kqueue_close,
117 	.fo_kqfilter = kqueue_kqfilter,
118 	.fo_drain = fnullop_drain,
119 };
120 
121 static const struct filterops kqread_filtops =
122 	{ 1, NULL, filt_kqdetach, filt_kqueue };
123 static const struct filterops proc_filtops =
124 	{ 0, filt_procattach, filt_procdetach, filt_proc };
125 static const struct filterops file_filtops =
126 	{ 1, filt_fileattach, NULL, NULL };
127 static const struct filterops timer_filtops =
128 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
129 
130 static u_int	kq_ncallouts = 0;
131 static int	kq_calloutmax = (4 * 1024);
132 
133 #define	KN_HASHSIZE		64		/* XXX should be tunable */
134 #define	KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
135 
136 extern const struct filterops sig_filtops;
137 
138 /*
139  * Table for for all system-defined filters.
140  * These should be listed in the numeric order of the EVFILT_* defines.
141  * If filtops is NULL, the filter isn't implemented in NetBSD.
142  * End of list is when name is NULL.
143  *
144  * Note that 'refcnt' is meaningless for built-in filters.
145  */
146 struct kfilter {
147 	const char	*name;		/* name of filter */
148 	uint32_t	filter;		/* id of filter */
149 	unsigned	refcnt;		/* reference count */
150 	const struct filterops *filtops;/* operations for filter */
151 	size_t		namelen;	/* length of name string */
152 };
153 
154 /* System defined filters */
155 static struct kfilter sys_kfilters[] = {
156 	{ "EVFILT_READ",	EVFILT_READ,	0, &file_filtops, 0 },
157 	{ "EVFILT_WRITE",	EVFILT_WRITE,	0, &file_filtops, 0, },
158 	{ "EVFILT_AIO",		EVFILT_AIO,	0, NULL, 0 },
159 	{ "EVFILT_VNODE",	EVFILT_VNODE,	0, &file_filtops, 0 },
160 	{ "EVFILT_PROC",	EVFILT_PROC,	0, &proc_filtops, 0 },
161 	{ "EVFILT_SIGNAL",	EVFILT_SIGNAL,	0, &sig_filtops, 0 },
162 	{ "EVFILT_TIMER",	EVFILT_TIMER,	0, &timer_filtops, 0 },
163 	{ NULL,			0,		0, NULL, 0 },
164 };
165 
166 /* User defined kfilters */
167 static struct kfilter	*user_kfilters;		/* array */
168 static int		user_kfilterc;		/* current offset */
169 static int		user_kfiltermaxc;	/* max size so far */
170 static size_t		user_kfiltersz;		/* size of allocated memory */
171 
172 /* Locks */
173 static krwlock_t	kqueue_filter_lock;	/* lock on filter lists */
174 static kmutex_t		kqueue_misc_lock;	/* miscellaneous */
175 
176 /*
177  * Initialize the kqueue subsystem.
178  */
179 void
180 kqueue_init(void)
181 {
182 
183 	rw_init(&kqueue_filter_lock);
184 	mutex_init(&kqueue_misc_lock, MUTEX_DEFAULT, IPL_NONE);
185 }
186 
187 /*
188  * Find kfilter entry by name, or NULL if not found.
189  */
190 static struct kfilter *
191 kfilter_byname_sys(const char *name)
192 {
193 	int i;
194 
195 	KASSERT(rw_lock_held(&kqueue_filter_lock));
196 
197 	for (i = 0; sys_kfilters[i].name != NULL; i++) {
198 		if (strcmp(name, sys_kfilters[i].name) == 0)
199 			return &sys_kfilters[i];
200 	}
201 	return NULL;
202 }
203 
204 static struct kfilter *
205 kfilter_byname_user(const char *name)
206 {
207 	int i;
208 
209 	KASSERT(rw_lock_held(&kqueue_filter_lock));
210 
211 	/* user filter slots have a NULL name if previously deregistered */
212 	for (i = 0; i < user_kfilterc ; i++) {
213 		if (user_kfilters[i].name != NULL &&
214 		    strcmp(name, user_kfilters[i].name) == 0)
215 			return &user_kfilters[i];
216 	}
217 	return NULL;
218 }
219 
220 static struct kfilter *
221 kfilter_byname(const char *name)
222 {
223 	struct kfilter *kfilter;
224 
225 	KASSERT(rw_lock_held(&kqueue_filter_lock));
226 
227 	if ((kfilter = kfilter_byname_sys(name)) != NULL)
228 		return kfilter;
229 
230 	return kfilter_byname_user(name);
231 }
232 
233 /*
234  * Find kfilter entry by filter id, or NULL if not found.
235  * Assumes entries are indexed in filter id order, for speed.
236  */
237 static struct kfilter *
238 kfilter_byfilter(uint32_t filter)
239 {
240 	struct kfilter *kfilter;
241 
242 	KASSERT(rw_lock_held(&kqueue_filter_lock));
243 
244 	if (filter < EVFILT_SYSCOUNT)	/* it's a system filter */
245 		kfilter = &sys_kfilters[filter];
246 	else if (user_kfilters != NULL &&
247 	    filter < EVFILT_SYSCOUNT + user_kfilterc)
248 					/* it's a user filter */
249 		kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT];
250 	else
251 		return (NULL);		/* out of range */
252 	KASSERT(kfilter->filter == filter);	/* sanity check! */
253 	return (kfilter);
254 }
255 
256 /*
257  * Register a new kfilter. Stores the entry in user_kfilters.
258  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
259  * If retfilter != NULL, the new filterid is returned in it.
260  */
261 int
262 kfilter_register(const char *name, const struct filterops *filtops,
263 		 int *retfilter)
264 {
265 	struct kfilter *kfilter;
266 	size_t len;
267 	int i;
268 
269 	if (name == NULL || name[0] == '\0' || filtops == NULL)
270 		return (EINVAL);	/* invalid args */
271 
272 	rw_enter(&kqueue_filter_lock, RW_WRITER);
273 	if (kfilter_byname(name) != NULL) {
274 		rw_exit(&kqueue_filter_lock);
275 		return (EEXIST);	/* already exists */
276 	}
277 	if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT) {
278 		rw_exit(&kqueue_filter_lock);
279 		return (EINVAL);	/* too many */
280 	}
281 
282 	for (i = 0; i < user_kfilterc; i++) {
283 		kfilter = &user_kfilters[i];
284 		if (kfilter->name == NULL) {
285 			/* Previously deregistered slot.  Reuse. */
286 			goto reuse;
287 		}
288 	}
289 
290 	/* check if need to grow user_kfilters */
291 	if (user_kfilterc + 1 > user_kfiltermaxc) {
292 		/* Grow in KFILTER_EXTENT chunks. */
293 		user_kfiltermaxc += KFILTER_EXTENT;
294 		len = user_kfiltermaxc * sizeof(struct filter *);
295 		kfilter = kmem_alloc(len, KM_SLEEP);
296 		memset((char *)kfilter + user_kfiltersz, 0, len - user_kfiltersz);
297 		if (user_kfilters != NULL) {
298 			memcpy(kfilter, user_kfilters, user_kfiltersz);
299 			kmem_free(user_kfilters, user_kfiltersz);
300 		}
301 		user_kfiltersz = len;
302 		user_kfilters = kfilter;
303 	}
304 	/* Adding new slot */
305 	kfilter = &user_kfilters[user_kfilterc++];
306 reuse:
307 	kfilter->namelen = strlen(name) + 1;
308 	kfilter->name = kmem_alloc(kfilter->namelen, KM_SLEEP);
309 	memcpy(__UNCONST(kfilter->name), name, kfilter->namelen);
310 
311 	kfilter->filter = (kfilter - user_kfilters) + EVFILT_SYSCOUNT;
312 
313 	kfilter->filtops = kmem_alloc(sizeof(*filtops), KM_SLEEP);
314 	memcpy(__UNCONST(kfilter->filtops), filtops, sizeof(*filtops));
315 
316 	if (retfilter != NULL)
317 		*retfilter = kfilter->filter;
318 	rw_exit(&kqueue_filter_lock);
319 
320 	return (0);
321 }
322 
323 /*
324  * Unregister a kfilter previously registered with kfilter_register.
325  * This retains the filter id, but clears the name and frees filtops (filter
326  * operations), so that the number isn't reused during a boot.
327  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
328  */
329 int
330 kfilter_unregister(const char *name)
331 {
332 	struct kfilter *kfilter;
333 
334 	if (name == NULL || name[0] == '\0')
335 		return (EINVAL);	/* invalid name */
336 
337 	rw_enter(&kqueue_filter_lock, RW_WRITER);
338 	if (kfilter_byname_sys(name) != NULL) {
339 		rw_exit(&kqueue_filter_lock);
340 		return (EINVAL);	/* can't detach system filters */
341 	}
342 
343 	kfilter = kfilter_byname_user(name);
344 	if (kfilter == NULL) {
345 		rw_exit(&kqueue_filter_lock);
346 		return (ENOENT);
347 	}
348 	if (kfilter->refcnt != 0) {
349 		rw_exit(&kqueue_filter_lock);
350 		return (EBUSY);
351 	}
352 
353 	/* Cast away const (but we know it's safe. */
354 	kmem_free(__UNCONST(kfilter->name), kfilter->namelen);
355 	kfilter->name = NULL;	/* mark as `not implemented' */
356 
357 	if (kfilter->filtops != NULL) {
358 		/* Cast away const (but we know it's safe. */
359 		kmem_free(__UNCONST(kfilter->filtops),
360 		    sizeof(*kfilter->filtops));
361 		kfilter->filtops = NULL; /* mark as `not implemented' */
362 	}
363 	rw_exit(&kqueue_filter_lock);
364 
365 	return (0);
366 }
367 
368 
369 /*
370  * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file
371  * descriptors. Calls fileops kqfilter method for given file descriptor.
372  */
373 static int
374 filt_fileattach(struct knote *kn)
375 {
376 	file_t *fp;
377 
378 	fp = kn->kn_obj;
379 
380 	return (*fp->f_ops->fo_kqfilter)(fp, kn);
381 }
382 
383 /*
384  * Filter detach method for EVFILT_READ on kqueue descriptor.
385  */
386 static void
387 filt_kqdetach(struct knote *kn)
388 {
389 	struct kqueue *kq;
390 
391 	kq = ((file_t *)kn->kn_obj)->f_data;
392 
393 	mutex_spin_enter(&kq->kq_lock);
394 	SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext);
395 	mutex_spin_exit(&kq->kq_lock);
396 }
397 
398 /*
399  * Filter event method for EVFILT_READ on kqueue descriptor.
400  */
401 /*ARGSUSED*/
402 static int
403 filt_kqueue(struct knote *kn, long hint)
404 {
405 	struct kqueue *kq;
406 	int rv;
407 
408 	kq = ((file_t *)kn->kn_obj)->f_data;
409 
410 	if (hint != NOTE_SUBMIT)
411 		mutex_spin_enter(&kq->kq_lock);
412 	kn->kn_data = kq->kq_count;
413 	rv = (kn->kn_data > 0);
414 	if (hint != NOTE_SUBMIT)
415 		mutex_spin_exit(&kq->kq_lock);
416 
417 	return rv;
418 }
419 
420 /*
421  * Filter attach method for EVFILT_PROC.
422  */
423 static int
424 filt_procattach(struct knote *kn)
425 {
426 	struct proc *p, *curp;
427 	struct lwp *curl;
428 
429 	curl = curlwp;
430 	curp = curl->l_proc;
431 
432 	mutex_enter(proc_lock);
433 	p = p_find(kn->kn_id, PFIND_LOCKED);
434 	if (p == NULL) {
435 		mutex_exit(proc_lock);
436 		return ESRCH;
437 	}
438 
439 	/*
440 	 * Fail if it's not owned by you, or the last exec gave us
441 	 * setuid/setgid privs (unless you're root).
442 	 */
443 	mutex_enter(p->p_lock);
444 	mutex_exit(proc_lock);
445 	if (kauth_authorize_process(curl->l_cred, KAUTH_PROCESS_KEVENT_FILTER,
446 	    p, NULL, NULL, NULL) != 0) {
447 	    	mutex_exit(p->p_lock);
448 		return EACCES;
449 	}
450 
451 	kn->kn_obj = p;
452 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
453 
454 	/*
455 	 * internal flag indicating registration done by kernel
456 	 */
457 	if (kn->kn_flags & EV_FLAG1) {
458 		kn->kn_data = kn->kn_sdata;	/* ppid */
459 		kn->kn_fflags = NOTE_CHILD;
460 		kn->kn_flags &= ~EV_FLAG1;
461 	}
462 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
463     	mutex_exit(p->p_lock);
464 
465 	return 0;
466 }
467 
468 /*
469  * Filter detach method for EVFILT_PROC.
470  *
471  * The knote may be attached to a different process, which may exit,
472  * leaving nothing for the knote to be attached to.  So when the process
473  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
474  * it will be deleted when read out.  However, as part of the knote deletion,
475  * this routine is called, so a check is needed to avoid actually performing
476  * a detach, because the original process might not exist any more.
477  */
478 static void
479 filt_procdetach(struct knote *kn)
480 {
481 	struct proc *p;
482 
483 	if (kn->kn_status & KN_DETACHED)
484 		return;
485 
486 	p = kn->kn_obj;
487 
488 	mutex_enter(p->p_lock);
489 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
490 	mutex_exit(p->p_lock);
491 }
492 
493 /*
494  * Filter event method for EVFILT_PROC.
495  */
496 static int
497 filt_proc(struct knote *kn, long hint)
498 {
499 	u_int event, fflag;
500 	struct kevent kev;
501 	struct kqueue *kq;
502 	int error;
503 
504 	event = (u_int)hint & NOTE_PCTRLMASK;
505 	kq = kn->kn_kq;
506 	fflag = 0;
507 
508 	/* If the user is interested in this event, record it. */
509 	if (kn->kn_sfflags & event)
510 		fflag |= event;
511 
512 	if (event == NOTE_EXIT) {
513 		/*
514 		 * Process is gone, so flag the event as finished.
515 		 *
516 		 * Detach the knote from watched process and mark
517 		 * it as such. We can't leave this to kqueue_scan(),
518 		 * since the process might not exist by then. And we
519 		 * have to do this now, since psignal KNOTE() is called
520 		 * also for zombies and we might end up reading freed
521 		 * memory if the kevent would already be picked up
522 		 * and knote g/c'ed.
523 		 */
524 		filt_procdetach(kn);
525 
526 		mutex_spin_enter(&kq->kq_lock);
527 		kn->kn_status |= KN_DETACHED;
528 		/* Mark as ONESHOT, so that the knote it g/c'ed when read */
529 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
530 		kn->kn_fflags |= fflag;
531 		mutex_spin_exit(&kq->kq_lock);
532 
533 		return 1;
534 	}
535 
536 	mutex_spin_enter(&kq->kq_lock);
537 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
538 		/*
539 		 * Process forked, and user wants to track the new process,
540 		 * so attach a new knote to it, and immediately report an
541 		 * event with the parent's pid.  Register knote with new
542 		 * process.
543 		 */
544 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
545 		kev.filter = kn->kn_filter;
546 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
547 		kev.fflags = kn->kn_sfflags;
548 		kev.data = kn->kn_id;			/* parent */
549 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
550 		mutex_spin_exit(&kq->kq_lock);
551 		error = kqueue_register(kq, &kev);
552 		mutex_spin_enter(&kq->kq_lock);
553 		if (error != 0)
554 			kn->kn_fflags |= NOTE_TRACKERR;
555 	}
556 	kn->kn_fflags |= fflag;
557 	fflag = kn->kn_fflags;
558 	mutex_spin_exit(&kq->kq_lock);
559 
560 	return fflag != 0;
561 }
562 
563 static void
564 filt_timerexpire(void *knx)
565 {
566 	struct knote *kn = knx;
567 	int tticks;
568 
569 	mutex_enter(&kqueue_misc_lock);
570 	kn->kn_data++;
571 	knote_activate(kn);
572 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
573 		tticks = mstohz(kn->kn_sdata);
574 		callout_schedule((callout_t *)kn->kn_hook, tticks);
575 	}
576 	mutex_exit(&kqueue_misc_lock);
577 }
578 
579 /*
580  * data contains amount of time to sleep, in milliseconds
581  */
582 static int
583 filt_timerattach(struct knote *kn)
584 {
585 	callout_t *calloutp;
586 	struct kqueue *kq;
587 	int tticks;
588 
589 	tticks = mstohz(kn->kn_sdata);
590 
591 	/* if the supplied value is under our resolution, use 1 tick */
592 	if (tticks == 0) {
593 		if (kn->kn_sdata == 0)
594 			return EINVAL;
595 		tticks = 1;
596 	}
597 
598 	if (atomic_inc_uint_nv(&kq_ncallouts) >= kq_calloutmax ||
599 	    (calloutp = kmem_alloc(sizeof(*calloutp), KM_NOSLEEP)) == NULL) {
600 		atomic_dec_uint(&kq_ncallouts);
601 		return ENOMEM;
602 	}
603 	callout_init(calloutp, CALLOUT_MPSAFE);
604 
605 	kq = kn->kn_kq;
606 	mutex_spin_enter(&kq->kq_lock);
607 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
608 	kn->kn_hook = calloutp;
609 	mutex_spin_exit(&kq->kq_lock);
610 
611 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
612 
613 	return (0);
614 }
615 
616 static void
617 filt_timerdetach(struct knote *kn)
618 {
619 	callout_t *calloutp;
620 
621 	calloutp = (callout_t *)kn->kn_hook;
622 	callout_halt(calloutp, NULL);
623 	callout_destroy(calloutp);
624 	kmem_free(calloutp, sizeof(*calloutp));
625 	atomic_dec_uint(&kq_ncallouts);
626 }
627 
628 static int
629 filt_timer(struct knote *kn, long hint)
630 {
631 	int rv;
632 
633 	mutex_enter(&kqueue_misc_lock);
634 	rv = (kn->kn_data != 0);
635 	mutex_exit(&kqueue_misc_lock);
636 
637 	return rv;
638 }
639 
640 /*
641  * filt_seltrue:
642  *
643  *	This filter "event" routine simulates seltrue().
644  */
645 int
646 filt_seltrue(struct knote *kn, long hint)
647 {
648 
649 	/*
650 	 * We don't know how much data can be read/written,
651 	 * but we know that it *can* be.  This is about as
652 	 * good as select/poll does as well.
653 	 */
654 	kn->kn_data = 0;
655 	return (1);
656 }
657 
658 /*
659  * This provides full kqfilter entry for device switch tables, which
660  * has same effect as filter using filt_seltrue() as filter method.
661  */
662 static void
663 filt_seltruedetach(struct knote *kn)
664 {
665 	/* Nothing to do */
666 }
667 
668 const struct filterops seltrue_filtops =
669 	{ 1, NULL, filt_seltruedetach, filt_seltrue };
670 
671 int
672 seltrue_kqfilter(dev_t dev, struct knote *kn)
673 {
674 	switch (kn->kn_filter) {
675 	case EVFILT_READ:
676 	case EVFILT_WRITE:
677 		kn->kn_fop = &seltrue_filtops;
678 		break;
679 	default:
680 		return (EINVAL);
681 	}
682 
683 	/* Nothing more to do */
684 	return (0);
685 }
686 
687 /*
688  * kqueue(2) system call.
689  */
690 int
691 sys_kqueue(struct lwp *l, const void *v, register_t *retval)
692 {
693 	struct kqueue *kq;
694 	file_t *fp;
695 	int fd, error;
696 
697 	if ((error = fd_allocfile(&fp, &fd)) != 0)
698 		return error;
699 	fp->f_flag = FREAD | FWRITE;
700 	fp->f_type = DTYPE_KQUEUE;
701 	fp->f_ops = &kqueueops;
702 	kq = kmem_zalloc(sizeof(*kq), KM_SLEEP);
703 	mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED);
704 	cv_init(&kq->kq_cv, "kqueue");
705 	selinit(&kq->kq_sel);
706 	TAILQ_INIT(&kq->kq_head);
707 	fp->f_data = kq;
708 	*retval = fd;
709 	kq->kq_fdp = curlwp->l_fd;
710 	fd_affix(curproc, fp, fd);
711 	return error;
712 }
713 
714 /*
715  * kevent(2) system call.
716  */
717 int
718 kevent_fetch_changes(void *private, const struct kevent *changelist,
719     struct kevent *changes, size_t index, int n)
720 {
721 
722 	return copyin(changelist + index, changes, n * sizeof(*changes));
723 }
724 
725 int
726 kevent_put_events(void *private, struct kevent *events,
727     struct kevent *eventlist, size_t index, int n)
728 {
729 
730 	return copyout(events, eventlist + index, n * sizeof(*events));
731 }
732 
733 static const struct kevent_ops kevent_native_ops = {
734 	.keo_private = NULL,
735 	.keo_fetch_timeout = copyin,
736 	.keo_fetch_changes = kevent_fetch_changes,
737 	.keo_put_events = kevent_put_events,
738 };
739 
740 int
741 sys___kevent50(struct lwp *l, const struct sys___kevent50_args *uap,
742     register_t *retval)
743 {
744 	/* {
745 		syscallarg(int) fd;
746 		syscallarg(const struct kevent *) changelist;
747 		syscallarg(size_t) nchanges;
748 		syscallarg(struct kevent *) eventlist;
749 		syscallarg(size_t) nevents;
750 		syscallarg(const struct timespec *) timeout;
751 	} */
752 
753 	return kevent1(retval, SCARG(uap, fd), SCARG(uap, changelist),
754 	    SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents),
755 	    SCARG(uap, timeout), &kevent_native_ops);
756 }
757 
758 int
759 kevent1(register_t *retval, int fd,
760 	const struct kevent *changelist, size_t nchanges,
761 	struct kevent *eventlist, size_t nevents,
762 	const struct timespec *timeout,
763 	const struct kevent_ops *keops)
764 {
765 	struct kevent *kevp;
766 	struct kqueue *kq;
767 	struct timespec	ts;
768 	size_t i, n, ichange;
769 	int nerrors, error;
770 	struct kevent kevbuf[8];	/* approx 300 bytes on 64-bit */
771 	file_t *fp;
772 
773 	/* check that we're dealing with a kq */
774 	fp = fd_getfile(fd);
775 	if (fp == NULL)
776 		return (EBADF);
777 
778 	if (fp->f_type != DTYPE_KQUEUE) {
779 		fd_putfile(fd);
780 		return (EBADF);
781 	}
782 
783 	if (timeout != NULL) {
784 		error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts));
785 		if (error)
786 			goto done;
787 		timeout = &ts;
788 	}
789 
790 	kq = (struct kqueue *)fp->f_data;
791 	nerrors = 0;
792 	ichange = 0;
793 
794 	/* traverse list of events to register */
795 	while (nchanges > 0) {
796 		n = MIN(nchanges, __arraycount(kevbuf));
797 		error = (*keops->keo_fetch_changes)(keops->keo_private,
798 		    changelist, kevbuf, ichange, n);
799 		if (error)
800 			goto done;
801 		for (i = 0; i < n; i++) {
802 			kevp = &kevbuf[i];
803 			kevp->flags &= ~EV_SYSFLAGS;
804 			/* register each knote */
805 			error = kqueue_register(kq, kevp);
806 			if (error) {
807 				if (nevents != 0) {
808 					kevp->flags = EV_ERROR;
809 					kevp->data = error;
810 					error = (*keops->keo_put_events)
811 					    (keops->keo_private, kevp,
812 					    eventlist, nerrors, 1);
813 					if (error)
814 						goto done;
815 					nevents--;
816 					nerrors++;
817 				} else {
818 					goto done;
819 				}
820 			}
821 		}
822 		nchanges -= n;	/* update the results */
823 		ichange += n;
824 	}
825 	if (nerrors) {
826 		*retval = nerrors;
827 		error = 0;
828 		goto done;
829 	}
830 
831 	/* actually scan through the events */
832 	error = kqueue_scan(fp, nevents, eventlist, timeout, retval, keops,
833 	    kevbuf, __arraycount(kevbuf));
834  done:
835 	fd_putfile(fd);
836 	return (error);
837 }
838 
839 /*
840  * Register a given kevent kev onto the kqueue
841  */
842 static int
843 kqueue_register(struct kqueue *kq, struct kevent *kev)
844 {
845 	struct kfilter *kfilter;
846 	filedesc_t *fdp;
847 	file_t *fp;
848 	fdfile_t *ff;
849 	struct knote *kn, *newkn;
850 	struct klist *list;
851 	int error, fd, rv;
852 
853 	fdp = kq->kq_fdp;
854 	fp = NULL;
855 	kn = NULL;
856 	error = 0;
857 	fd = 0;
858 
859 	newkn = kmem_zalloc(sizeof(*newkn), KM_SLEEP);
860 
861 	rw_enter(&kqueue_filter_lock, RW_READER);
862 	kfilter = kfilter_byfilter(kev->filter);
863 	if (kfilter == NULL || kfilter->filtops == NULL) {
864 		/* filter not found nor implemented */
865 		rw_exit(&kqueue_filter_lock);
866 		kmem_free(newkn, sizeof(*newkn));
867 		return (EINVAL);
868 	}
869 
870  	mutex_enter(&fdp->fd_lock);
871 
872 	/* search if knote already exists */
873 	if (kfilter->filtops->f_isfd) {
874 		/* monitoring a file descriptor */
875 		fd = kev->ident;
876 		if ((fp = fd_getfile(fd)) == NULL) {
877 		 	mutex_exit(&fdp->fd_lock);
878 			rw_exit(&kqueue_filter_lock);
879 			kmem_free(newkn, sizeof(*newkn));
880 			return EBADF;
881 		}
882 		ff = fdp->fd_dt->dt_ff[fd];
883 		if (fd <= fdp->fd_lastkqfile) {
884 			SLIST_FOREACH(kn, &ff->ff_knlist, kn_link) {
885 				if (kq == kn->kn_kq &&
886 				    kev->filter == kn->kn_filter)
887 					break;
888 			}
889 		}
890 	} else {
891 		/*
892 		 * not monitoring a file descriptor, so
893 		 * lookup knotes in internal hash table
894 		 */
895 		if (fdp->fd_knhashmask != 0) {
896 			list = &fdp->fd_knhash[
897 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
898 			SLIST_FOREACH(kn, list, kn_link) {
899 				if (kev->ident == kn->kn_id &&
900 				    kq == kn->kn_kq &&
901 				    kev->filter == kn->kn_filter)
902 					break;
903 			}
904 		}
905 	}
906 
907 	/*
908 	 * kn now contains the matching knote, or NULL if no match
909 	 */
910 	if (kev->flags & EV_ADD) {
911 		if (kn == NULL) {
912 			/* create new knote */
913 			kn = newkn;
914 			newkn = NULL;
915 			kn->kn_obj = fp;
916 			kn->kn_kq = kq;
917 			kn->kn_fop = kfilter->filtops;
918 			kn->kn_kfilter = kfilter;
919 			kn->kn_sfflags = kev->fflags;
920 			kn->kn_sdata = kev->data;
921 			kev->fflags = 0;
922 			kev->data = 0;
923 			kn->kn_kevent = *kev;
924 
925 			/*
926 			 * apply reference count to knote structure, and
927 			 * do not release it at the end of this routine.
928 			 */
929 			fp = NULL;
930 
931 			if (!kn->kn_fop->f_isfd) {
932 				/*
933 				 * If knote is not on an fd, store on
934 				 * internal hash table.
935 				 */
936 				if (fdp->fd_knhashmask == 0) {
937 					/* XXXAD can block with fd_lock held */
938 					fdp->fd_knhash = hashinit(KN_HASHSIZE,
939 					    HASH_LIST, true,
940 					    &fdp->fd_knhashmask);
941 				}
942 				list = &fdp->fd_knhash[KN_HASH(kn->kn_id,
943 				    fdp->fd_knhashmask)];
944 			} else {
945 				/* Otherwise, knote is on an fd. */
946 				list = (struct klist *)
947 				    &fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist;
948 				if ((int)kn->kn_id > fdp->fd_lastkqfile)
949 					fdp->fd_lastkqfile = kn->kn_id;
950 			}
951 			SLIST_INSERT_HEAD(list, kn, kn_link);
952 
953 			KERNEL_LOCK(1, NULL);		/* XXXSMP */
954 			error = (*kfilter->filtops->f_attach)(kn);
955 			KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
956 			if (error != 0) {
957 				/* knote_detach() drops fdp->fd_lock */
958 				knote_detach(kn, fdp, false);
959 				goto done;
960 			}
961 			atomic_inc_uint(&kfilter->refcnt);
962 		} else {
963 			/*
964 			 * The user may change some filter values after the
965 			 * initial EV_ADD, but doing so will not reset any
966 			 * filter which have already been triggered.
967 			 */
968 			kn->kn_sfflags = kev->fflags;
969 			kn->kn_sdata = kev->data;
970 			kn->kn_kevent.udata = kev->udata;
971 		}
972 		KERNEL_LOCK(1, NULL);			/* XXXSMP */
973 		rv = (*kn->kn_fop->f_event)(kn, 0);
974 		KERNEL_UNLOCK_ONE(NULL);		/* XXXSMP */
975 		if (rv)
976 			knote_activate(kn);
977 	} else {
978 		if (kn == NULL) {
979 			error = ENOENT;
980 		 	mutex_exit(&fdp->fd_lock);
981 			goto done;
982 		}
983 		if (kev->flags & EV_DELETE) {
984 			/* knote_detach() drops fdp->fd_lock */
985 			knote_detach(kn, fdp, true);
986 			goto done;
987 		}
988 	}
989 
990 	/* disable knote */
991 	if ((kev->flags & EV_DISABLE)) {
992 		mutex_spin_enter(&kq->kq_lock);
993 		if ((kn->kn_status & KN_DISABLED) == 0)
994 			kn->kn_status |= KN_DISABLED;
995 		mutex_spin_exit(&kq->kq_lock);
996 	}
997 
998 	/* enable knote */
999 	if ((kev->flags & EV_ENABLE)) {
1000 		knote_enqueue(kn);
1001 	}
1002 	mutex_exit(&fdp->fd_lock);
1003  done:
1004 	rw_exit(&kqueue_filter_lock);
1005 	if (newkn != NULL)
1006 		kmem_free(newkn, sizeof(*newkn));
1007 	if (fp != NULL)
1008 		fd_putfile(fd);
1009 	return (error);
1010 }
1011 
1012 #if defined(DEBUG)
1013 static void
1014 kq_check(struct kqueue *kq)
1015 {
1016 	const struct knote *kn;
1017 	int count;
1018 	int nmarker;
1019 
1020 	KASSERT(mutex_owned(&kq->kq_lock));
1021 	KASSERT(kq->kq_count >= 0);
1022 
1023 	count = 0;
1024 	nmarker = 0;
1025 	TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) {
1026 		if ((kn->kn_status & (KN_MARKER | KN_QUEUED)) == 0) {
1027 			panic("%s: kq=%p kn=%p inconsist 1", __func__, kq, kn);
1028 		}
1029 		if ((kn->kn_status & KN_MARKER) == 0) {
1030 			if (kn->kn_kq != kq) {
1031 				panic("%s: kq=%p kn=%p inconsist 2",
1032 				    __func__, kq, kn);
1033 			}
1034 			if ((kn->kn_status & KN_ACTIVE) == 0) {
1035 				panic("%s: kq=%p kn=%p: not active",
1036 				    __func__, kq, kn);
1037 			}
1038 			count++;
1039 			if (count > kq->kq_count) {
1040 				goto bad;
1041 			}
1042 		} else {
1043 			nmarker++;
1044 #if 0
1045 			if (nmarker > 10000) {
1046 				panic("%s: kq=%p too many markers: %d != %d, "
1047 				    "nmarker=%d",
1048 				    __func__, kq, kq->kq_count, count, nmarker);
1049 			}
1050 #endif
1051 		}
1052 	}
1053 	if (kq->kq_count != count) {
1054 bad:
1055 		panic("%s: kq=%p inconsist 3: %d != %d, nmarker=%d",
1056 		    __func__, kq, kq->kq_count, count, nmarker);
1057 	}
1058 }
1059 #else /* defined(DEBUG) */
1060 #define	kq_check(a)	/* nothing */
1061 #endif /* defined(DEBUG) */
1062 
1063 /*
1064  * Scan through the list of events on fp (for a maximum of maxevents),
1065  * returning the results in to ulistp. Timeout is determined by tsp; if
1066  * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait
1067  * as appropriate.
1068  */
1069 static int
1070 kqueue_scan(file_t *fp, size_t maxevents, struct kevent *ulistp,
1071 	    const struct timespec *tsp, register_t *retval,
1072 	    const struct kevent_ops *keops, struct kevent *kevbuf,
1073 	    size_t kevcnt)
1074 {
1075 	struct kqueue	*kq;
1076 	struct kevent	*kevp;
1077 	struct timespec	ats, sleepts;
1078 	struct knote	*kn, *marker;
1079 	size_t		count, nkev, nevents;
1080 	int		timeout, error, rv;
1081 	filedesc_t	*fdp;
1082 
1083 	fdp = curlwp->l_fd;
1084 	kq = fp->f_data;
1085 	count = maxevents;
1086 	nkev = nevents = error = 0;
1087 	if (count == 0) {
1088 		*retval = 0;
1089 		return 0;
1090 	}
1091 
1092 	if (tsp) {				/* timeout supplied */
1093 		ats = *tsp;
1094 		if (inittimeleft(&ats, &sleepts) == -1) {
1095 			*retval = maxevents;
1096 			return EINVAL;
1097 		}
1098 		timeout = tstohz(&ats);
1099 		if (timeout <= 0)
1100 			timeout = -1;           /* do poll */
1101 	} else {
1102 		/* no timeout, wait forever */
1103 		timeout = 0;
1104 	}
1105 
1106 	marker = kmem_zalloc(sizeof(*marker), KM_SLEEP);
1107 	marker->kn_status = KN_MARKER;
1108 	mutex_spin_enter(&kq->kq_lock);
1109  retry:
1110 	kevp = kevbuf;
1111 	if (kq->kq_count == 0) {
1112 		if (timeout >= 0) {
1113 			error = cv_timedwait_sig(&kq->kq_cv,
1114 			    &kq->kq_lock, timeout);
1115 			if (error == 0) {
1116 				 if (tsp == NULL || (timeout =
1117 				     gettimeleft(&ats, &sleepts)) > 0)
1118 					goto retry;
1119 			} else {
1120 				/* don't restart after signals... */
1121 				if (error == ERESTART)
1122 					error = EINTR;
1123 				if (error == EWOULDBLOCK)
1124 					error = 0;
1125 			}
1126 		}
1127 	} else {
1128 		/* mark end of knote list */
1129 		TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1130 
1131 		while (count != 0) {
1132 			kn = TAILQ_FIRST(&kq->kq_head);	/* get next knote */
1133 			while ((kn->kn_status & KN_MARKER) != 0) {
1134 				if (kn == marker) {
1135 					/* it's our marker, stop */
1136 					TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1137 					if (count < maxevents || (tsp != NULL &&
1138 					    (timeout = gettimeleft(&ats,
1139 					    &sleepts)) <= 0))
1140 						goto done;
1141 					goto retry;
1142 				}
1143 				/* someone else's marker. */
1144 				kn = TAILQ_NEXT(kn, kn_tqe);
1145 			}
1146 			kq_check(kq);
1147 			TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1148 			kq->kq_count--;
1149 			kn->kn_status &= ~KN_QUEUED;
1150 			kq_check(kq);
1151 			if (kn->kn_status & KN_DISABLED) {
1152 				/* don't want disabled events */
1153 				continue;
1154 			}
1155 			if ((kn->kn_flags & EV_ONESHOT) == 0) {
1156 				mutex_spin_exit(&kq->kq_lock);
1157 				KERNEL_LOCK(1, NULL);		/* XXXSMP */
1158 				rv = (*kn->kn_fop->f_event)(kn, 0);
1159 				KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1160 				mutex_spin_enter(&kq->kq_lock);
1161 				/* Re-poll if note was re-enqueued. */
1162 				if ((kn->kn_status & KN_QUEUED) != 0)
1163 					continue;
1164 				if (rv == 0) {
1165 					/*
1166 					 * non-ONESHOT event that hasn't
1167 					 * triggered again, so de-queue.
1168 					 */
1169 					kn->kn_status &= ~KN_ACTIVE;
1170 					continue;
1171 				}
1172 			}
1173 			/* XXXAD should be got from f_event if !oneshot. */
1174 			*kevp++ = kn->kn_kevent;
1175 			nkev++;
1176 			if (kn->kn_flags & EV_ONESHOT) {
1177 				/* delete ONESHOT events after retrieval */
1178 				mutex_spin_exit(&kq->kq_lock);
1179 				mutex_enter(&fdp->fd_lock);
1180 				knote_detach(kn, fdp, true);
1181 				mutex_spin_enter(&kq->kq_lock);
1182 			} else if (kn->kn_flags & EV_CLEAR) {
1183 				/* clear state after retrieval */
1184 				kn->kn_data = 0;
1185 				kn->kn_fflags = 0;
1186 				kn->kn_status &= ~KN_ACTIVE;
1187 			} else {
1188 				/* add event back on list */
1189 				kq_check(kq);
1190 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1191 				kq->kq_count++;
1192 				kn->kn_status |= KN_QUEUED;
1193 				kq_check(kq);
1194 			}
1195 			if (nkev == kevcnt) {
1196 				/* do copyouts in kevcnt chunks */
1197 				mutex_spin_exit(&kq->kq_lock);
1198 				error = (*keops->keo_put_events)
1199 				    (keops->keo_private,
1200 				    kevbuf, ulistp, nevents, nkev);
1201 				mutex_spin_enter(&kq->kq_lock);
1202 				nevents += nkev;
1203 				nkev = 0;
1204 				kevp = kevbuf;
1205 			}
1206 			count--;
1207 			if (error != 0 || count == 0) {
1208 				/* remove marker */
1209 				TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1210 				break;
1211 			}
1212 		}
1213 	}
1214  done:
1215  	mutex_spin_exit(&kq->kq_lock);
1216 	if (marker != NULL)
1217 		kmem_free(marker, sizeof(*marker));
1218 	if (nkev != 0) {
1219 		/* copyout remaining events */
1220 		error = (*keops->keo_put_events)(keops->keo_private,
1221 		    kevbuf, ulistp, nevents, nkev);
1222 	}
1223 	*retval = maxevents - count;
1224 
1225 	return error;
1226 }
1227 
1228 /*
1229  * fileops ioctl method for a kqueue descriptor.
1230  *
1231  * Two ioctls are currently supported. They both use struct kfilter_mapping:
1232  *	KFILTER_BYNAME		find name for filter, and return result in
1233  *				name, which is of size len.
1234  *	KFILTER_BYFILTER	find filter for name. len is ignored.
1235  */
1236 /*ARGSUSED*/
1237 static int
1238 kqueue_ioctl(file_t *fp, u_long com, void *data)
1239 {
1240 	struct kfilter_mapping	*km;
1241 	const struct kfilter	*kfilter;
1242 	char			*name;
1243 	int			error;
1244 
1245 	km = data;
1246 	error = 0;
1247 	name = kmem_alloc(KFILTER_MAXNAME, KM_SLEEP);
1248 
1249 	switch (com) {
1250 	case KFILTER_BYFILTER:	/* convert filter -> name */
1251 		rw_enter(&kqueue_filter_lock, RW_READER);
1252 		kfilter = kfilter_byfilter(km->filter);
1253 		if (kfilter != NULL) {
1254 			strlcpy(name, kfilter->name, KFILTER_MAXNAME);
1255 			rw_exit(&kqueue_filter_lock);
1256 			error = copyoutstr(name, km->name, km->len, NULL);
1257 		} else {
1258 			rw_exit(&kqueue_filter_lock);
1259 			error = ENOENT;
1260 		}
1261 		break;
1262 
1263 	case KFILTER_BYNAME:	/* convert name -> filter */
1264 		error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL);
1265 		if (error) {
1266 			break;
1267 		}
1268 		rw_enter(&kqueue_filter_lock, RW_READER);
1269 		kfilter = kfilter_byname(name);
1270 		if (kfilter != NULL)
1271 			km->filter = kfilter->filter;
1272 		else
1273 			error = ENOENT;
1274 		rw_exit(&kqueue_filter_lock);
1275 		break;
1276 
1277 	default:
1278 		error = ENOTTY;
1279 		break;
1280 
1281 	}
1282 	kmem_free(name, KFILTER_MAXNAME);
1283 	return (error);
1284 }
1285 
1286 /*
1287  * fileops fcntl method for a kqueue descriptor.
1288  */
1289 static int
1290 kqueue_fcntl(file_t *fp, u_int com, void *data)
1291 {
1292 
1293 	return (ENOTTY);
1294 }
1295 
1296 /*
1297  * fileops poll method for a kqueue descriptor.
1298  * Determine if kqueue has events pending.
1299  */
1300 static int
1301 kqueue_poll(file_t *fp, int events)
1302 {
1303 	struct kqueue	*kq;
1304 	int		revents;
1305 
1306 	kq = fp->f_data;
1307 
1308 	revents = 0;
1309 	if (events & (POLLIN | POLLRDNORM)) {
1310 		mutex_spin_enter(&kq->kq_lock);
1311 		if (kq->kq_count != 0) {
1312 			revents |= events & (POLLIN | POLLRDNORM);
1313 		} else {
1314 			selrecord(curlwp, &kq->kq_sel);
1315 		}
1316 		kq_check(kq);
1317 		mutex_spin_exit(&kq->kq_lock);
1318 	}
1319 
1320 	return revents;
1321 }
1322 
1323 /*
1324  * fileops stat method for a kqueue descriptor.
1325  * Returns dummy info, with st_size being number of events pending.
1326  */
1327 static int
1328 kqueue_stat(file_t *fp, struct stat *st)
1329 {
1330 	struct kqueue *kq;
1331 
1332 	kq = fp->f_data;
1333 
1334 	memset(st, 0, sizeof(*st));
1335 	st->st_size = kq->kq_count;
1336 	st->st_blksize = sizeof(struct kevent);
1337 	st->st_mode = S_IFIFO;
1338 
1339 	return 0;
1340 }
1341 
1342 static void
1343 kqueue_doclose(struct kqueue *kq, struct klist *list, int fd)
1344 {
1345 	struct knote *kn;
1346 	filedesc_t *fdp;
1347 
1348 	fdp = kq->kq_fdp;
1349 
1350 	KASSERT(mutex_owned(&fdp->fd_lock));
1351 
1352 	for (kn = SLIST_FIRST(list); kn != NULL;) {
1353 		if (kq != kn->kn_kq) {
1354 			kn = SLIST_NEXT(kn, kn_link);
1355 			continue;
1356 		}
1357 		knote_detach(kn, fdp, true);
1358 		mutex_enter(&fdp->fd_lock);
1359 		kn = SLIST_FIRST(list);
1360 	}
1361 }
1362 
1363 
1364 /*
1365  * fileops close method for a kqueue descriptor.
1366  */
1367 static int
1368 kqueue_close(file_t *fp)
1369 {
1370 	struct kqueue *kq;
1371 	filedesc_t *fdp;
1372 	fdfile_t *ff;
1373 	int i;
1374 
1375 	kq = fp->f_data;
1376 	fdp = curlwp->l_fd;
1377 
1378 	mutex_enter(&fdp->fd_lock);
1379 	for (i = 0; i <= fdp->fd_lastkqfile; i++) {
1380 		if ((ff = fdp->fd_dt->dt_ff[i]) == NULL)
1381 			continue;
1382 		kqueue_doclose(kq, (struct klist *)&ff->ff_knlist, i);
1383 	}
1384 	if (fdp->fd_knhashmask != 0) {
1385 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
1386 			kqueue_doclose(kq, &fdp->fd_knhash[i], -1);
1387 		}
1388 	}
1389 	mutex_exit(&fdp->fd_lock);
1390 
1391 	KASSERT(kq->kq_count == 0);
1392 	mutex_destroy(&kq->kq_lock);
1393 	cv_destroy(&kq->kq_cv);
1394 	seldestroy(&kq->kq_sel);
1395 	kmem_free(kq, sizeof(*kq));
1396 	fp->f_data = NULL;
1397 
1398 	return (0);
1399 }
1400 
1401 /*
1402  * struct fileops kqfilter method for a kqueue descriptor.
1403  * Event triggered when monitored kqueue changes.
1404  */
1405 static int
1406 kqueue_kqfilter(file_t *fp, struct knote *kn)
1407 {
1408 	struct kqueue *kq;
1409 	filedesc_t *fdp;
1410 
1411 	kq = ((file_t *)kn->kn_obj)->f_data;
1412 
1413 	KASSERT(fp == kn->kn_obj);
1414 
1415 	if (kn->kn_filter != EVFILT_READ)
1416 		return 1;
1417 
1418 	kn->kn_fop = &kqread_filtops;
1419 	fdp = curlwp->l_fd;
1420 	mutex_enter(&kq->kq_lock);
1421 	SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext);
1422 	mutex_exit(&kq->kq_lock);
1423 
1424 	return 0;
1425 }
1426 
1427 
1428 /*
1429  * Walk down a list of knotes, activating them if their event has
1430  * triggered.  The caller's object lock (e.g. device driver lock)
1431  * must be held.
1432  */
1433 void
1434 knote(struct klist *list, long hint)
1435 {
1436 	struct knote *kn;
1437 
1438 	SLIST_FOREACH(kn, list, kn_selnext) {
1439 		if ((*kn->kn_fop->f_event)(kn, hint))
1440 			knote_activate(kn);
1441 	}
1442 }
1443 
1444 /*
1445  * Remove all knotes referencing a specified fd
1446  */
1447 void
1448 knote_fdclose(int fd)
1449 {
1450 	struct klist *list;
1451 	struct knote *kn;
1452 	filedesc_t *fdp;
1453 
1454 	fdp = curlwp->l_fd;
1455 	list = (struct klist *)&fdp->fd_dt->dt_ff[fd]->ff_knlist;
1456 	mutex_enter(&fdp->fd_lock);
1457 	while ((kn = SLIST_FIRST(list)) != NULL) {
1458 		knote_detach(kn, fdp, true);
1459 		mutex_enter(&fdp->fd_lock);
1460 	}
1461 	mutex_exit(&fdp->fd_lock);
1462 }
1463 
1464 /*
1465  * Drop knote.  Called with fdp->fd_lock held, and will drop before
1466  * returning.
1467  */
1468 static void
1469 knote_detach(struct knote *kn, filedesc_t *fdp, bool dofop)
1470 {
1471 	struct klist *list;
1472 	struct kqueue *kq;
1473 
1474 	kq = kn->kn_kq;
1475 
1476 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1477 	KASSERT(mutex_owned(&fdp->fd_lock));
1478 
1479 	/* Remove from monitored object. */
1480 	if (dofop) {
1481 		KERNEL_LOCK(1, NULL);		/* XXXSMP */
1482 		(*kn->kn_fop->f_detach)(kn);
1483 		KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1484 	}
1485 
1486 	/* Remove from descriptor table. */
1487 	if (kn->kn_fop->f_isfd)
1488 		list = (struct klist *)&fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist;
1489 	else
1490 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1491 
1492 	SLIST_REMOVE(list, kn, knote, kn_link);
1493 
1494 	/* Remove from kqueue. */
1495 	/* XXXAD should verify not in use by kqueue_scan. */
1496 	mutex_spin_enter(&kq->kq_lock);
1497 	if ((kn->kn_status & KN_QUEUED) != 0) {
1498 		kq_check(kq);
1499 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1500 		kn->kn_status &= ~KN_QUEUED;
1501 		kq->kq_count--;
1502 		kq_check(kq);
1503 	}
1504 	mutex_spin_exit(&kq->kq_lock);
1505 
1506 	mutex_exit(&fdp->fd_lock);
1507 	if (kn->kn_fop->f_isfd)
1508 		fd_putfile(kn->kn_id);
1509 	atomic_dec_uint(&kn->kn_kfilter->refcnt);
1510 	kmem_free(kn, sizeof(*kn));
1511 }
1512 
1513 /*
1514  * Queue new event for knote.
1515  */
1516 static void
1517 knote_enqueue(struct knote *kn)
1518 {
1519 	struct kqueue *kq;
1520 
1521 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1522 
1523 	kq = kn->kn_kq;
1524 
1525 	mutex_spin_enter(&kq->kq_lock);
1526 	if ((kn->kn_status & KN_DISABLED) != 0) {
1527 		kn->kn_status &= ~KN_DISABLED;
1528 	}
1529 	if ((kn->kn_status & (KN_ACTIVE | KN_QUEUED)) == KN_ACTIVE) {
1530 		kq_check(kq);
1531 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1532 		kn->kn_status |= KN_QUEUED;
1533 		kq->kq_count++;
1534 		kq_check(kq);
1535 		cv_broadcast(&kq->kq_cv);
1536 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1537 	}
1538 	mutex_spin_exit(&kq->kq_lock);
1539 }
1540 /*
1541  * Queue new event for knote.
1542  */
1543 static void
1544 knote_activate(struct knote *kn)
1545 {
1546 	struct kqueue *kq;
1547 
1548 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1549 
1550 	kq = kn->kn_kq;
1551 
1552 	mutex_spin_enter(&kq->kq_lock);
1553 	kn->kn_status |= KN_ACTIVE;
1554 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) {
1555 		kq_check(kq);
1556 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1557 		kn->kn_status |= KN_QUEUED;
1558 		kq->kq_count++;
1559 		kq_check(kq);
1560 		cv_broadcast(&kq->kq_cv);
1561 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1562 	}
1563 	mutex_spin_exit(&kq->kq_lock);
1564 }
1565