xref: /freebsd/sys/kern/kern_event.c (revision b28624fde638caadd4a89f50c9b7e7da0f98c4d2)
1 /*-
2  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
3  * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org>
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_ktrace.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/proc.h>
39 #include <sys/malloc.h>
40 #include <sys/unistd.h>
41 #include <sys/file.h>
42 #include <sys/filedesc.h>
43 #include <sys/filio.h>
44 #include <sys/fcntl.h>
45 #include <sys/kthread.h>
46 #include <sys/selinfo.h>
47 #include <sys/queue.h>
48 #include <sys/event.h>
49 #include <sys/eventvar.h>
50 #include <sys/poll.h>
51 #include <sys/protosw.h>
52 #include <sys/sigio.h>
53 #include <sys/signalvar.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/stat.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysproto.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/taskqueue.h>
61 #include <sys/uio.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65 
66 #include <vm/uma.h>
67 
68 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
69 
70 /*
71  * This lock is used if multiple kq locks are required.  This possibly
72  * should be made into a per proc lock.
73  */
74 static struct mtx	kq_global;
75 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
76 #define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
77 	if (!haslck)				\
78 		mtx_lock(lck);			\
79 	haslck = 1;				\
80 } while (0)
81 #define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
82 	if (haslck)				\
83 		mtx_unlock(lck);			\
84 	haslck = 0;				\
85 } while (0)
86 
87 TASKQUEUE_DEFINE_THREAD(kqueue);
88 
89 static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
90 static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
91 static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
92 		    struct thread *td, int waitok);
93 static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
94 static void	kqueue_release(struct kqueue *kq, int locked);
95 static int	kqueue_expand(struct kqueue *kq, struct filterops *fops,
96 		    uintptr_t ident, int waitok);
97 static void	kqueue_task(void *arg, int pending);
98 static int	kqueue_scan(struct kqueue *kq, int maxevents,
99 		    struct kevent_copyops *k_ops,
100 		    const struct timespec *timeout,
101 		    struct kevent *keva, struct thread *td);
102 static void 	kqueue_wakeup(struct kqueue *kq);
103 static struct filterops *kqueue_fo_find(int filt);
104 static void	kqueue_fo_release(int filt);
105 
106 static fo_rdwr_t	kqueue_read;
107 static fo_rdwr_t	kqueue_write;
108 static fo_ioctl_t	kqueue_ioctl;
109 static fo_poll_t	kqueue_poll;
110 static fo_kqfilter_t	kqueue_kqfilter;
111 static fo_stat_t	kqueue_stat;
112 static fo_close_t	kqueue_close;
113 
114 static struct fileops kqueueops = {
115 	.fo_read = kqueue_read,
116 	.fo_write = kqueue_write,
117 	.fo_ioctl = kqueue_ioctl,
118 	.fo_poll = kqueue_poll,
119 	.fo_kqfilter = kqueue_kqfilter,
120 	.fo_stat = kqueue_stat,
121 	.fo_close = kqueue_close,
122 };
123 
124 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
125 static void 	knote_drop(struct knote *kn, struct thread *td);
126 static void 	knote_enqueue(struct knote *kn);
127 static void 	knote_dequeue(struct knote *kn);
128 static void 	knote_init(void);
129 static struct 	knote *knote_alloc(int waitok);
130 static void 	knote_free(struct knote *kn);
131 
132 static void	filt_kqdetach(struct knote *kn);
133 static int	filt_kqueue(struct knote *kn, long hint);
134 static int	filt_procattach(struct knote *kn);
135 static void	filt_procdetach(struct knote *kn);
136 static int	filt_proc(struct knote *kn, long hint);
137 static int	filt_fileattach(struct knote *kn);
138 static void	filt_timerexpire(void *knx);
139 static int	filt_timerattach(struct knote *kn);
140 static void	filt_timerdetach(struct knote *kn);
141 static int	filt_timer(struct knote *kn, long hint);
142 
143 static struct filterops file_filtops =
144 	{ 1, filt_fileattach, NULL, NULL };
145 static struct filterops kqread_filtops =
146 	{ 1, NULL, filt_kqdetach, filt_kqueue };
147 /* XXX - move to kern_proc.c?  */
148 static struct filterops proc_filtops =
149 	{ 0, filt_procattach, filt_procdetach, filt_proc };
150 static struct filterops timer_filtops =
151 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
152 
153 static uma_zone_t	knote_zone;
154 static int 		kq_ncallouts = 0;
155 static int 		kq_calloutmax = (4 * 1024);
156 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
157     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
158 
159 /* XXX - ensure not KN_INFLUX?? */
160 #define KNOTE_ACTIVATE(kn, islock) do { 				\
161 	if ((islock))							\
162 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
163 	else								\
164 		KQ_LOCK((kn)->kn_kq);					\
165 	(kn)->kn_status |= KN_ACTIVE;					\
166 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
167 		knote_enqueue((kn));					\
168 	if (!(islock))							\
169 		KQ_UNLOCK((kn)->kn_kq);					\
170 } while(0)
171 #define KQ_LOCK(kq) do {						\
172 	mtx_lock(&(kq)->kq_lock);					\
173 } while (0)
174 #define KQ_FLUX_WAKEUP(kq) do {						\
175 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
176 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
177 		wakeup((kq));						\
178 	}								\
179 } while (0)
180 #define KQ_UNLOCK_FLUX(kq) do {						\
181 	KQ_FLUX_WAKEUP(kq);						\
182 	mtx_unlock(&(kq)->kq_lock);					\
183 } while (0)
184 #define KQ_UNLOCK(kq) do {						\
185 	mtx_unlock(&(kq)->kq_lock);					\
186 } while (0)
187 #define KQ_OWNED(kq) do {						\
188 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
189 } while (0)
190 #define KQ_NOTOWNED(kq) do {						\
191 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
192 } while (0)
193 #define KN_LIST_LOCK(kn) do {						\
194 	if (kn->kn_knlist != NULL)					\
195 		kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg);	\
196 } while (0)
197 #define KN_LIST_UNLOCK(kn) do {						\
198 	if (kn->kn_knlist != NULL) 					\
199 		kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg);	\
200 } while (0)
201 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
202 	if (islocked)							\
203 		KNL_ASSERT_LOCKED(knl);				\
204 	else								\
205 		KNL_ASSERT_UNLOCKED(knl);				\
206 } while (0)
207 #ifdef INVARIANTS
208 #define	KNL_ASSERT_LOCKED(knl) do {					\
209 	if (!knl->kl_locked((knl)->kl_lockarg))				\
210 			panic("knlist not locked, but should be");	\
211 } while (0)
212 #define	KNL_ASSERT_UNLOCKED(knl) do {				\
213 	if (knl->kl_locked((knl)->kl_lockarg))				\
214 		panic("knlist locked, but should not be");		\
215 } while (0)
216 #else /* !INVARIANTS */
217 #define	KNL_ASSERT_LOCKED(knl) do {} while(0)
218 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
219 #endif /* INVARIANTS */
220 
221 #define	KN_HASHSIZE		64		/* XXX should be tunable */
222 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
223 
224 static int
225 filt_nullattach(struct knote *kn)
226 {
227 
228 	return (ENXIO);
229 };
230 
231 struct filterops null_filtops =
232 	{ 0, filt_nullattach, NULL, NULL };
233 
234 /* XXX - make SYSINIT to add these, and move into respective modules. */
235 extern struct filterops sig_filtops;
236 extern struct filterops fs_filtops;
237 
238 /*
239  * Table for for all system-defined filters.
240  */
241 static struct mtx	filterops_lock;
242 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
243 	MTX_DEF);
244 static struct {
245 	struct filterops *for_fop;
246 	int for_refcnt;
247 } sysfilt_ops[EVFILT_SYSCOUNT] = {
248 	{ &file_filtops },			/* EVFILT_READ */
249 	{ &file_filtops },			/* EVFILT_WRITE */
250 	{ &null_filtops },			/* EVFILT_AIO */
251 	{ &file_filtops },			/* EVFILT_VNODE */
252 	{ &proc_filtops },			/* EVFILT_PROC */
253 	{ &sig_filtops },			/* EVFILT_SIGNAL */
254 	{ &timer_filtops },			/* EVFILT_TIMER */
255 	{ &file_filtops },			/* EVFILT_NETDEV */
256 	{ &fs_filtops },			/* EVFILT_FS */
257 	{ &null_filtops },			/* EVFILT_LIO */
258 };
259 
260 /*
261  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
262  * method.
263  */
264 static int
265 filt_fileattach(struct knote *kn)
266 {
267 
268 	return (fo_kqfilter(kn->kn_fp, kn));
269 }
270 
271 /*ARGSUSED*/
272 static int
273 kqueue_kqfilter(struct file *fp, struct knote *kn)
274 {
275 	struct kqueue *kq = kn->kn_fp->f_data;
276 
277 	if (kn->kn_filter != EVFILT_READ)
278 		return (EINVAL);
279 
280 	kn->kn_status |= KN_KQUEUE;
281 	kn->kn_fop = &kqread_filtops;
282 	knlist_add(&kq->kq_sel.si_note, kn, 0);
283 
284 	return (0);
285 }
286 
287 static void
288 filt_kqdetach(struct knote *kn)
289 {
290 	struct kqueue *kq = kn->kn_fp->f_data;
291 
292 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
293 }
294 
295 /*ARGSUSED*/
296 static int
297 filt_kqueue(struct knote *kn, long hint)
298 {
299 	struct kqueue *kq = kn->kn_fp->f_data;
300 
301 	kn->kn_data = kq->kq_count;
302 	return (kn->kn_data > 0);
303 }
304 
305 /* XXX - move to kern_proc.c?  */
306 static int
307 filt_procattach(struct knote *kn)
308 {
309 	struct proc *p;
310 	int immediate;
311 	int error;
312 
313 	immediate = 0;
314 	p = pfind(kn->kn_id);
315 	if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
316 		p = zpfind(kn->kn_id);
317 		immediate = 1;
318 	} else if (p != NULL && (p->p_flag & P_WEXIT)) {
319 		immediate = 1;
320 	}
321 
322 	if (p == NULL)
323 		return (ESRCH);
324 	if ((error = p_cansee(curthread, p)))
325 		return (error);
326 
327 	kn->kn_ptr.p_proc = p;
328 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
329 
330 	/*
331 	 * internal flag indicating registration done by kernel
332 	 */
333 	if (kn->kn_flags & EV_FLAG1) {
334 		kn->kn_data = kn->kn_sdata;		/* ppid */
335 		kn->kn_fflags = NOTE_CHILD;
336 		kn->kn_flags &= ~EV_FLAG1;
337 	}
338 
339 	if (immediate == 0)
340 		knlist_add(&p->p_klist, kn, 1);
341 
342 	/*
343 	 * Immediately activate any exit notes if the target process is a
344 	 * zombie.  This is necessary to handle the case where the target
345 	 * process, e.g. a child, dies before the kevent is registered.
346 	 */
347 	if (immediate && filt_proc(kn, NOTE_EXIT))
348 		KNOTE_ACTIVATE(kn, 0);
349 
350 	PROC_UNLOCK(p);
351 
352 	return (0);
353 }
354 
355 /*
356  * The knote may be attached to a different process, which may exit,
357  * leaving nothing for the knote to be attached to.  So when the process
358  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
359  * it will be deleted when read out.  However, as part of the knote deletion,
360  * this routine is called, so a check is needed to avoid actually performing
361  * a detach, because the original process does not exist any more.
362  */
363 /* XXX - move to kern_proc.c?  */
364 static void
365 filt_procdetach(struct knote *kn)
366 {
367 	struct proc *p;
368 
369 	p = kn->kn_ptr.p_proc;
370 	knlist_remove(&p->p_klist, kn, 0);
371 	kn->kn_ptr.p_proc = NULL;
372 }
373 
374 /* XXX - move to kern_proc.c?  */
375 static int
376 filt_proc(struct knote *kn, long hint)
377 {
378 	struct proc *p = kn->kn_ptr.p_proc;
379 	u_int event;
380 
381 	/*
382 	 * mask off extra data
383 	 */
384 	event = (u_int)hint & NOTE_PCTRLMASK;
385 
386 	/*
387 	 * if the user is interested in this event, record it.
388 	 */
389 	if (kn->kn_sfflags & event)
390 		kn->kn_fflags |= event;
391 
392 	/*
393 	 * process is gone, so flag the event as finished.
394 	 */
395 	if (event == NOTE_EXIT) {
396 		if (!(kn->kn_status & KN_DETACHED))
397 			knlist_remove_inevent(&p->p_klist, kn);
398 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
399 		kn->kn_data = p->p_xstat;
400 		kn->kn_ptr.p_proc = NULL;
401 		return (1);
402 	}
403 
404 	/*
405 	 * process forked, and user wants to track the new process,
406 	 * so attach a new knote to it, and immediately report an
407 	 * event with the parent's pid.
408 	 */
409 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
410 		struct kevent kev;
411 		int error;
412 
413 		/*
414 		 * register knote with new process.
415 		 */
416 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
417 		kev.filter = kn->kn_filter;
418 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
419 		kev.fflags = kn->kn_sfflags;
420 		kev.data = kn->kn_id;			/* parent */
421 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
422 		error = kqueue_register(kn->kn_kq, &kev, NULL, 0);
423 		if (error)
424 			kn->kn_fflags |= NOTE_TRACKERR;
425 	}
426 
427 	return (kn->kn_fflags != 0);
428 }
429 
430 static int
431 timertoticks(intptr_t data)
432 {
433 	struct timeval tv;
434 	int tticks;
435 
436 	tv.tv_sec = data / 1000;
437 	tv.tv_usec = (data % 1000) * 1000;
438 	tticks = tvtohz(&tv);
439 
440 	return tticks;
441 }
442 
443 /* XXX - move to kern_timeout.c? */
444 static void
445 filt_timerexpire(void *knx)
446 {
447 	struct knote *kn = knx;
448 	struct callout *calloutp;
449 
450 	kn->kn_data++;
451 	KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
452 
453 	if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
454 		calloutp = (struct callout *)kn->kn_hook;
455 		callout_reset(calloutp, timertoticks(kn->kn_sdata),
456 		    filt_timerexpire, kn);
457 	}
458 }
459 
460 /*
461  * data contains amount of time to sleep, in milliseconds
462  */
463 /* XXX - move to kern_timeout.c? */
464 static int
465 filt_timerattach(struct knote *kn)
466 {
467 	struct callout *calloutp;
468 
469 	atomic_add_int(&kq_ncallouts, 1);
470 
471 	if (kq_ncallouts >= kq_calloutmax) {
472 		atomic_add_int(&kq_ncallouts, -1);
473 		return (ENOMEM);
474 	}
475 
476 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
477 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add usually sets it */
478 	MALLOC(calloutp, struct callout *, sizeof(*calloutp),
479 	    M_KQUEUE, M_WAITOK);
480 	callout_init(calloutp, CALLOUT_MPSAFE);
481 	kn->kn_hook = calloutp;
482 	callout_reset(calloutp, timertoticks(kn->kn_sdata), filt_timerexpire,
483 	    kn);
484 
485 	return (0);
486 }
487 
488 /* XXX - move to kern_timeout.c? */
489 static void
490 filt_timerdetach(struct knote *kn)
491 {
492 	struct callout *calloutp;
493 
494 	calloutp = (struct callout *)kn->kn_hook;
495 	callout_drain(calloutp);
496 	FREE(calloutp, M_KQUEUE);
497 	atomic_add_int(&kq_ncallouts, -1);
498 	kn->kn_status |= KN_DETACHED;	/* knlist_remove usually clears it */
499 }
500 
501 /* XXX - move to kern_timeout.c? */
502 static int
503 filt_timer(struct knote *kn, long hint)
504 {
505 
506 	return (kn->kn_data != 0);
507 }
508 
509 int
510 kqueue(struct thread *td, struct kqueue_args *uap)
511 {
512 	struct filedesc *fdp;
513 	struct kqueue *kq;
514 	struct file *fp;
515 	int fd, error;
516 
517 	fdp = td->td_proc->p_fd;
518 	error = falloc(td, &fp, &fd);
519 	if (error)
520 		goto done2;
521 
522 	/* An extra reference on `nfp' has been held for us by falloc(). */
523 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
524 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
525 	TAILQ_INIT(&kq->kq_head);
526 	kq->kq_fdp = fdp;
527 	knlist_init(&kq->kq_sel.si_note, &kq->kq_lock, NULL, NULL, NULL);
528 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
529 
530 	FILEDESC_XLOCK(fdp);
531 	SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
532 	FILEDESC_XUNLOCK(fdp);
533 
534 	FILE_LOCK(fp);
535 	fp->f_flag = FREAD | FWRITE;
536 	fp->f_type = DTYPE_KQUEUE;
537 	fp->f_data = kq;
538 	fp->f_ops = &kqueueops;
539 	FILE_UNLOCK(fp);
540 	fdrop(fp, td);
541 
542 	td->td_retval[0] = fd;
543 done2:
544 	return (error);
545 }
546 
547 #ifndef _SYS_SYSPROTO_H_
548 struct kevent_args {
549 	int	fd;
550 	const struct kevent *changelist;
551 	int	nchanges;
552 	struct	kevent *eventlist;
553 	int	nevents;
554 	const struct timespec *timeout;
555 };
556 #endif
557 int
558 kevent(struct thread *td, struct kevent_args *uap)
559 {
560 	struct timespec ts, *tsp;
561 	struct kevent_copyops k_ops = { uap,
562 					kevent_copyout,
563 					kevent_copyin};
564 	int error;
565 #ifdef KTRACE
566 	struct uio ktruio;
567 	struct iovec ktriov;
568 	struct uio *ktruioin = NULL;
569 	struct uio *ktruioout = NULL;
570 #endif
571 
572 	if (uap->timeout != NULL) {
573 		error = copyin(uap->timeout, &ts, sizeof(ts));
574 		if (error)
575 			return (error);
576 		tsp = &ts;
577 	} else
578 		tsp = NULL;
579 
580 #ifdef KTRACE
581 	if (KTRPOINT(td, KTR_GENIO)) {
582 		ktriov.iov_base = uap->changelist;
583 		ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
584 		ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
585 		    .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
586 		    .uio_td = td };
587 		ktruioin = cloneuio(&ktruio);
588 		ktriov.iov_base = uap->eventlist;
589 		ktriov.iov_len = uap->nevents * sizeof(struct kevent);
590 		ktruioout = cloneuio(&ktruio);
591 	}
592 #endif
593 
594 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
595 	    &k_ops, tsp);
596 
597 #ifdef KTRACE
598 	if (ktruioin != NULL) {
599 		ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
600 		ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
601 		ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
602 		ktrgenio(uap->fd, UIO_READ, ktruioout, error);
603 	}
604 #endif
605 
606 	return (error);
607 }
608 
609 /*
610  * Copy 'count' items into the destination list pointed to by uap->eventlist.
611  */
612 static int
613 kevent_copyout(void *arg, struct kevent *kevp, int count)
614 {
615 	struct kevent_args *uap;
616 	int error;
617 
618 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
619 	uap = (struct kevent_args *)arg;
620 
621 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
622 	if (error == 0)
623 		uap->eventlist += count;
624 	return (error);
625 }
626 
627 /*
628  * Copy 'count' items from the list pointed to by uap->changelist.
629  */
630 static int
631 kevent_copyin(void *arg, struct kevent *kevp, int count)
632 {
633 	struct kevent_args *uap;
634 	int error;
635 
636 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
637 	uap = (struct kevent_args *)arg;
638 
639 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
640 	if (error == 0)
641 		uap->changelist += count;
642 	return (error);
643 }
644 
645 int
646 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
647     struct kevent_copyops *k_ops, const struct timespec *timeout)
648 {
649 	struct kevent keva[KQ_NEVENTS];
650 	struct kevent *kevp, *changes;
651 	struct kqueue *kq;
652 	struct file *fp;
653 	int i, n, nerrors, error;
654 
655 	if ((error = fget(td, fd, &fp)) != 0)
656 		return (error);
657 	if ((error = kqueue_acquire(fp, &kq)) != 0)
658 		goto done_norel;
659 
660 	nerrors = 0;
661 
662 	while (nchanges > 0) {
663 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
664 		error = k_ops->k_copyin(k_ops->arg, keva, n);
665 		if (error)
666 			goto done;
667 		changes = keva;
668 		for (i = 0; i < n; i++) {
669 			kevp = &changes[i];
670 			if (!kevp->filter)
671 				continue;
672 			kevp->flags &= ~EV_SYSFLAGS;
673 			error = kqueue_register(kq, kevp, td, 1);
674 			if (error) {
675 				if (nevents != 0) {
676 					kevp->flags = EV_ERROR;
677 					kevp->data = error;
678 					(void) k_ops->k_copyout(k_ops->arg,
679 					    kevp, 1);
680 					nevents--;
681 					nerrors++;
682 				} else {
683 					goto done;
684 				}
685 			}
686 		}
687 		nchanges -= n;
688 	}
689 	if (nerrors) {
690 		td->td_retval[0] = nerrors;
691 		error = 0;
692 		goto done;
693 	}
694 
695 	error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
696 done:
697 	kqueue_release(kq, 0);
698 done_norel:
699 	fdrop(fp, td);
700 	return (error);
701 }
702 
703 int
704 kqueue_add_filteropts(int filt, struct filterops *filtops)
705 {
706 	int error;
707 
708 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
709 		printf(
710 "trying to add a filterop that is out of range: %d is beyond %d\n",
711 		    ~filt, EVFILT_SYSCOUNT);
712 		return EINVAL;
713 	}
714 	mtx_lock(&filterops_lock);
715 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
716 	    sysfilt_ops[~filt].for_fop != NULL)
717 		error = EEXIST;
718 	else {
719 		sysfilt_ops[~filt].for_fop = filtops;
720 		sysfilt_ops[~filt].for_refcnt = 0;
721 	}
722 	mtx_unlock(&filterops_lock);
723 
724 	return (0);
725 }
726 
727 int
728 kqueue_del_filteropts(int filt)
729 {
730 	int error;
731 
732 	error = 0;
733 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
734 		return EINVAL;
735 
736 	mtx_lock(&filterops_lock);
737 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
738 	    sysfilt_ops[~filt].for_fop == NULL)
739 		error = EINVAL;
740 	else if (sysfilt_ops[~filt].for_refcnt != 0)
741 		error = EBUSY;
742 	else {
743 		sysfilt_ops[~filt].for_fop = &null_filtops;
744 		sysfilt_ops[~filt].for_refcnt = 0;
745 	}
746 	mtx_unlock(&filterops_lock);
747 
748 	return error;
749 }
750 
751 static struct filterops *
752 kqueue_fo_find(int filt)
753 {
754 
755 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
756 		return NULL;
757 
758 	mtx_lock(&filterops_lock);
759 	sysfilt_ops[~filt].for_refcnt++;
760 	if (sysfilt_ops[~filt].for_fop == NULL)
761 		sysfilt_ops[~filt].for_fop = &null_filtops;
762 	mtx_unlock(&filterops_lock);
763 
764 	return sysfilt_ops[~filt].for_fop;
765 }
766 
767 static void
768 kqueue_fo_release(int filt)
769 {
770 
771 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
772 		return;
773 
774 	mtx_lock(&filterops_lock);
775 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
776 	    ("filter object refcount not valid on release"));
777 	sysfilt_ops[~filt].for_refcnt--;
778 	mtx_unlock(&filterops_lock);
779 }
780 
781 /*
782  * A ref to kq (obtained via kqueue_acquire) must be held.  waitok will
783  * influence if memory allocation should wait.  Make sure it is 0 if you
784  * hold any mutexes.
785  */
786 static int
787 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
788 {
789 	struct filterops *fops;
790 	struct file *fp;
791 	struct knote *kn, *tkn;
792 	int error, filt, event;
793 	int haskqglobal;
794 
795 	fp = NULL;
796 	kn = NULL;
797 	error = 0;
798 	haskqglobal = 0;
799 
800 	filt = kev->filter;
801 	fops = kqueue_fo_find(filt);
802 	if (fops == NULL)
803 		return EINVAL;
804 
805 	tkn = knote_alloc(waitok);		/* prevent waiting with locks */
806 
807 findkn:
808 	if (fops->f_isfd) {
809 		KASSERT(td != NULL, ("td is NULL"));
810 		error = fget(td, kev->ident, &fp);
811 		if (error)
812 			goto done;
813 
814 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
815 		    kev->ident, 0) != 0) {
816 			/* try again */
817 			fdrop(fp, td);
818 			fp = NULL;
819 			error = kqueue_expand(kq, fops, kev->ident, waitok);
820 			if (error)
821 				goto done;
822 			goto findkn;
823 		}
824 
825 		if (fp->f_type == DTYPE_KQUEUE) {
826 			/*
827 			 * if we add some inteligence about what we are doing,
828 			 * we should be able to support events on ourselves.
829 			 * We need to know when we are doing this to prevent
830 			 * getting both the knlist lock and the kq lock since
831 			 * they are the same thing.
832 			 */
833 			if (fp->f_data == kq) {
834 				error = EINVAL;
835 				goto done;
836 			}
837 
838 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
839 		}
840 
841 		KQ_LOCK(kq);
842 		if (kev->ident < kq->kq_knlistsize) {
843 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
844 				if (kev->filter == kn->kn_filter)
845 					break;
846 		}
847 	} else {
848 		if ((kev->flags & EV_ADD) == EV_ADD)
849 			kqueue_expand(kq, fops, kev->ident, waitok);
850 
851 		KQ_LOCK(kq);
852 		if (kq->kq_knhashmask != 0) {
853 			struct klist *list;
854 
855 			list = &kq->kq_knhash[
856 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
857 			SLIST_FOREACH(kn, list, kn_link)
858 				if (kev->ident == kn->kn_id &&
859 				    kev->filter == kn->kn_filter)
860 					break;
861 		}
862 	}
863 
864 	/* knote is in the process of changing, wait for it to stablize. */
865 	if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
866 		if (fp != NULL) {
867 			fdrop(fp, td);
868 			fp = NULL;
869 		}
870 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
871 		kq->kq_state |= KQ_FLUXWAIT;
872 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
873 		goto findkn;
874 	}
875 
876 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
877 		KQ_UNLOCK(kq);
878 		error = ENOENT;
879 		goto done;
880 	}
881 
882 	/*
883 	 * kn now contains the matching knote, or NULL if no match
884 	 */
885 	if (kev->flags & EV_ADD) {
886 		if (kn == NULL) {
887 			kn = tkn;
888 			tkn = NULL;
889 			if (kn == NULL) {
890 				KQ_UNLOCK(kq);
891 				error = ENOMEM;
892 				goto done;
893 			}
894 			kn->kn_fp = fp;
895 			kn->kn_kq = kq;
896 			kn->kn_fop = fops;
897 			/*
898 			 * apply reference counts to knote structure, and
899 			 * do not release it at the end of this routine.
900 			 */
901 			fops = NULL;
902 			fp = NULL;
903 
904 			kn->kn_sfflags = kev->fflags;
905 			kn->kn_sdata = kev->data;
906 			kev->fflags = 0;
907 			kev->data = 0;
908 			kn->kn_kevent = *kev;
909 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
910 			    EV_ENABLE | EV_DISABLE);
911 			kn->kn_status = KN_INFLUX|KN_DETACHED;
912 
913 			error = knote_attach(kn, kq);
914 			KQ_UNLOCK(kq);
915 			if (error != 0) {
916 				tkn = kn;
917 				goto done;
918 			}
919 
920 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
921 				knote_drop(kn, td);
922 				goto done;
923 			}
924 			KN_LIST_LOCK(kn);
925 		} else {
926 			/*
927 			 * The user may change some filter values after the
928 			 * initial EV_ADD, but doing so will not reset any
929 			 * filter which has already been triggered.
930 			 */
931 			kn->kn_status |= KN_INFLUX;
932 			KQ_UNLOCK(kq);
933 			KN_LIST_LOCK(kn);
934 			kn->kn_sfflags = kev->fflags;
935 			kn->kn_sdata = kev->data;
936 			kn->kn_kevent.udata = kev->udata;
937 		}
938 
939 		/*
940 		 * We can get here with kn->kn_knlist == NULL.
941 		 * This can happen when the initial attach event decides that
942 		 * the event is "completed" already.  i.e. filt_procattach
943 		 * is called on a zombie process.  It will call filt_proc
944 		 * which will remove it from the list, and NULL kn_knlist.
945 		 */
946 		event = kn->kn_fop->f_event(kn, 0);
947 		KQ_LOCK(kq);
948 		if (event)
949 			KNOTE_ACTIVATE(kn, 1);
950 		kn->kn_status &= ~KN_INFLUX;
951 		KN_LIST_UNLOCK(kn);
952 	} else if (kev->flags & EV_DELETE) {
953 		kn->kn_status |= KN_INFLUX;
954 		KQ_UNLOCK(kq);
955 		if (!(kn->kn_status & KN_DETACHED))
956 			kn->kn_fop->f_detach(kn);
957 		knote_drop(kn, td);
958 		goto done;
959 	}
960 
961 	if ((kev->flags & EV_DISABLE) &&
962 	    ((kn->kn_status & KN_DISABLED) == 0)) {
963 		kn->kn_status |= KN_DISABLED;
964 	}
965 
966 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
967 		kn->kn_status &= ~KN_DISABLED;
968 		if ((kn->kn_status & KN_ACTIVE) &&
969 		    ((kn->kn_status & KN_QUEUED) == 0))
970 			knote_enqueue(kn);
971 	}
972 	KQ_UNLOCK_FLUX(kq);
973 
974 done:
975 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
976 	if (fp != NULL)
977 		fdrop(fp, td);
978 	if (tkn != NULL)
979 		knote_free(tkn);
980 	if (fops != NULL)
981 		kqueue_fo_release(filt);
982 	return (error);
983 }
984 
985 static int
986 kqueue_acquire(struct file *fp, struct kqueue **kqp)
987 {
988 	int error;
989 	struct kqueue *kq;
990 
991 	error = 0;
992 
993 	FILE_LOCK(fp);
994 	do {
995 		kq = fp->f_data;
996 		if (fp->f_type != DTYPE_KQUEUE || kq == NULL) {
997 			error = EBADF;
998 			break;
999 		}
1000 		*kqp = kq;
1001 		KQ_LOCK(kq);
1002 		if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1003 			KQ_UNLOCK(kq);
1004 			error = EBADF;
1005 			break;
1006 		}
1007 		kq->kq_refcnt++;
1008 		KQ_UNLOCK(kq);
1009 	} while (0);
1010 	FILE_UNLOCK(fp);
1011 
1012 	return error;
1013 }
1014 
1015 static void
1016 kqueue_release(struct kqueue *kq, int locked)
1017 {
1018 	if (locked)
1019 		KQ_OWNED(kq);
1020 	else
1021 		KQ_LOCK(kq);
1022 	kq->kq_refcnt--;
1023 	if (kq->kq_refcnt == 1)
1024 		wakeup(&kq->kq_refcnt);
1025 	if (!locked)
1026 		KQ_UNLOCK(kq);
1027 }
1028 
1029 static void
1030 kqueue_schedtask(struct kqueue *kq)
1031 {
1032 
1033 	KQ_OWNED(kq);
1034 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1035 	    ("scheduling kqueue task while draining"));
1036 
1037 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1038 		taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1039 		kq->kq_state |= KQ_TASKSCHED;
1040 	}
1041 }
1042 
1043 /*
1044  * Expand the kq to make sure we have storage for fops/ident pair.
1045  *
1046  * Return 0 on success (or no work necessary), return errno on failure.
1047  *
1048  * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1049  * If kqueue_register is called from a non-fd context, there usually/should
1050  * be no locks held.
1051  */
1052 static int
1053 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1054 	int waitok)
1055 {
1056 	struct klist *list, *tmp_knhash;
1057 	u_long tmp_knhashmask;
1058 	int size;
1059 	int fd;
1060 	int mflag = waitok ? M_WAITOK : M_NOWAIT;
1061 
1062 	KQ_NOTOWNED(kq);
1063 
1064 	if (fops->f_isfd) {
1065 		fd = ident;
1066 		if (kq->kq_knlistsize <= fd) {
1067 			size = kq->kq_knlistsize;
1068 			while (size <= fd)
1069 				size += KQEXTENT;
1070 			MALLOC(list, struct klist *,
1071 			    size * sizeof list, M_KQUEUE, mflag);
1072 			if (list == NULL)
1073 				return ENOMEM;
1074 			KQ_LOCK(kq);
1075 			if (kq->kq_knlistsize > fd) {
1076 				FREE(list, M_KQUEUE);
1077 				list = NULL;
1078 			} else {
1079 				if (kq->kq_knlist != NULL) {
1080 					bcopy(kq->kq_knlist, list,
1081 					    kq->kq_knlistsize * sizeof list);
1082 					FREE(kq->kq_knlist, M_KQUEUE);
1083 					kq->kq_knlist = NULL;
1084 				}
1085 				bzero((caddr_t)list +
1086 				    kq->kq_knlistsize * sizeof list,
1087 				    (size - kq->kq_knlistsize) * sizeof list);
1088 				kq->kq_knlistsize = size;
1089 				kq->kq_knlist = list;
1090 			}
1091 			KQ_UNLOCK(kq);
1092 		}
1093 	} else {
1094 		if (kq->kq_knhashmask == 0) {
1095 			tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1096 			    &tmp_knhashmask);
1097 			if (tmp_knhash == NULL)
1098 				return ENOMEM;
1099 			KQ_LOCK(kq);
1100 			if (kq->kq_knhashmask == 0) {
1101 				kq->kq_knhash = tmp_knhash;
1102 				kq->kq_knhashmask = tmp_knhashmask;
1103 			} else {
1104 				free(tmp_knhash, M_KQUEUE);
1105 			}
1106 			KQ_UNLOCK(kq);
1107 		}
1108 	}
1109 
1110 	KQ_NOTOWNED(kq);
1111 	return 0;
1112 }
1113 
1114 static void
1115 kqueue_task(void *arg, int pending)
1116 {
1117 	struct kqueue *kq;
1118 	int haskqglobal;
1119 
1120 	haskqglobal = 0;
1121 	kq = arg;
1122 
1123 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1124 	KQ_LOCK(kq);
1125 
1126 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1127 
1128 	kq->kq_state &= ~KQ_TASKSCHED;
1129 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1130 		wakeup(&kq->kq_state);
1131 	}
1132 	KQ_UNLOCK(kq);
1133 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1134 }
1135 
1136 /*
1137  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1138  * We treat KN_MARKER knotes as if they are INFLUX.
1139  */
1140 static int
1141 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1142     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1143 {
1144 	struct kevent *kevp;
1145 	struct timeval atv, rtv, ttv;
1146 	struct knote *kn, *marker;
1147 	int count, timeout, nkev, error;
1148 	int haskqglobal;
1149 
1150 	count = maxevents;
1151 	nkev = 0;
1152 	error = 0;
1153 	haskqglobal = 0;
1154 
1155 	if (maxevents == 0)
1156 		goto done_nl;
1157 
1158 	if (tsp != NULL) {
1159 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
1160 		if (itimerfix(&atv)) {
1161 			error = EINVAL;
1162 			goto done_nl;
1163 		}
1164 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1165 			timeout = -1;
1166 		else
1167 			timeout = atv.tv_sec > 24 * 60 * 60 ?
1168 			    24 * 60 * 60 * hz : tvtohz(&atv);
1169 		getmicrouptime(&rtv);
1170 		timevaladd(&atv, &rtv);
1171 	} else {
1172 		atv.tv_sec = 0;
1173 		atv.tv_usec = 0;
1174 		timeout = 0;
1175 	}
1176 	marker = knote_alloc(1);
1177 	if (marker == NULL) {
1178 		error = ENOMEM;
1179 		goto done_nl;
1180 	}
1181 	marker->kn_status = KN_MARKER;
1182 	KQ_LOCK(kq);
1183 	goto start;
1184 
1185 retry:
1186 	if (atv.tv_sec || atv.tv_usec) {
1187 		getmicrouptime(&rtv);
1188 		if (timevalcmp(&rtv, &atv, >=))
1189 			goto done;
1190 		ttv = atv;
1191 		timevalsub(&ttv, &rtv);
1192 		timeout = ttv.tv_sec > 24 * 60 * 60 ?
1193 			24 * 60 * 60 * hz : tvtohz(&ttv);
1194 	}
1195 
1196 start:
1197 	kevp = keva;
1198 	if (kq->kq_count == 0) {
1199 		if (timeout < 0) {
1200 			error = EWOULDBLOCK;
1201 		} else {
1202 			kq->kq_state |= KQ_SLEEP;
1203 			error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH,
1204 			    "kqread", timeout);
1205 		}
1206 		if (error == 0)
1207 			goto retry;
1208 		/* don't restart after signals... */
1209 		if (error == ERESTART)
1210 			error = EINTR;
1211 		else if (error == EWOULDBLOCK)
1212 			error = 0;
1213 		goto done;
1214 	}
1215 
1216 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1217 	while (count) {
1218 		KQ_OWNED(kq);
1219 		kn = TAILQ_FIRST(&kq->kq_head);
1220 
1221 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1222 		    (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1223 			kq->kq_state |= KQ_FLUXWAIT;
1224 			error = msleep(kq, &kq->kq_lock, PSOCK,
1225 			    "kqflxwt", 0);
1226 			continue;
1227 		}
1228 
1229 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1230 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1231 			kn->kn_status &= ~KN_QUEUED;
1232 			kq->kq_count--;
1233 			continue;
1234 		}
1235 		if (kn == marker) {
1236 			KQ_FLUX_WAKEUP(kq);
1237 			if (count == maxevents)
1238 				goto retry;
1239 			goto done;
1240 		}
1241 		KASSERT((kn->kn_status & KN_INFLUX) == 0,
1242 		    ("KN_INFLUX set when not suppose to be"));
1243 
1244 		if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1245 			kn->kn_status &= ~KN_QUEUED;
1246 			kn->kn_status |= KN_INFLUX;
1247 			kq->kq_count--;
1248 			KQ_UNLOCK(kq);
1249 			/*
1250 			 * We don't need to lock the list since we've marked
1251 			 * it _INFLUX.
1252 			 */
1253 			*kevp = kn->kn_kevent;
1254 			if (!(kn->kn_status & KN_DETACHED))
1255 				kn->kn_fop->f_detach(kn);
1256 			knote_drop(kn, td);
1257 			KQ_LOCK(kq);
1258 			kn = NULL;
1259 		} else {
1260 			kn->kn_status |= KN_INFLUX;
1261 			KQ_UNLOCK(kq);
1262 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1263 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1264 			KN_LIST_LOCK(kn);
1265 			if (kn->kn_fop->f_event(kn, 0) == 0) {
1266 				KQ_LOCK(kq);
1267 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1268 				kn->kn_status &=
1269 				    ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX);
1270 				kq->kq_count--;
1271 				KN_LIST_UNLOCK(kn);
1272 				continue;
1273 			}
1274 			*kevp = kn->kn_kevent;
1275 			KQ_LOCK(kq);
1276 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1277 			if (kn->kn_flags & EV_CLEAR) {
1278 				kn->kn_data = 0;
1279 				kn->kn_fflags = 0;
1280 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1281 				kq->kq_count--;
1282 			} else
1283 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1284 
1285 			kn->kn_status &= ~(KN_INFLUX);
1286 			KN_LIST_UNLOCK(kn);
1287 		}
1288 
1289 		/* we are returning a copy to the user */
1290 		kevp++;
1291 		nkev++;
1292 		count--;
1293 
1294 		if (nkev == KQ_NEVENTS) {
1295 			KQ_UNLOCK_FLUX(kq);
1296 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1297 			nkev = 0;
1298 			kevp = keva;
1299 			KQ_LOCK(kq);
1300 			if (error)
1301 				break;
1302 		}
1303 	}
1304 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1305 done:
1306 	KQ_OWNED(kq);
1307 	KQ_UNLOCK_FLUX(kq);
1308 	knote_free(marker);
1309 done_nl:
1310 	KQ_NOTOWNED(kq);
1311 	if (nkev != 0)
1312 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1313 	td->td_retval[0] = maxevents - count;
1314 	return (error);
1315 }
1316 
1317 /*
1318  * XXX
1319  * This could be expanded to call kqueue_scan, if desired.
1320  */
1321 /*ARGSUSED*/
1322 static int
1323 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1324 	int flags, struct thread *td)
1325 {
1326 	return (ENXIO);
1327 }
1328 
1329 /*ARGSUSED*/
1330 static int
1331 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1332 	 int flags, struct thread *td)
1333 {
1334 	return (ENXIO);
1335 }
1336 
1337 /*ARGSUSED*/
1338 static int
1339 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1340 	struct ucred *active_cred, struct thread *td)
1341 {
1342 	/*
1343 	 * Enabling sigio causes two major problems:
1344 	 * 1) infinite recursion:
1345 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
1346 	 * set.  On receipt of a signal this will cause a kqueue to recurse
1347 	 * into itself over and over.  Sending the sigio causes the kqueue
1348 	 * to become ready, which in turn posts sigio again, forever.
1349 	 * Solution: this can be solved by setting a flag in the kqueue that
1350 	 * we have a SIGIO in progress.
1351 	 * 2) locking problems:
1352 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1353 	 * us above the proc and pgrp locks.
1354 	 * Solution: Post a signal using an async mechanism, being sure to
1355 	 * record a generation count in the delivery so that we do not deliver
1356 	 * a signal to the wrong process.
1357 	 *
1358 	 * Note, these two mechanisms are somewhat mutually exclusive!
1359 	 */
1360 #if 0
1361 	struct kqueue *kq;
1362 
1363 	kq = fp->f_data;
1364 	switch (cmd) {
1365 	case FIOASYNC:
1366 		if (*(int *)data) {
1367 			kq->kq_state |= KQ_ASYNC;
1368 		} else {
1369 			kq->kq_state &= ~KQ_ASYNC;
1370 		}
1371 		return (0);
1372 
1373 	case FIOSETOWN:
1374 		return (fsetown(*(int *)data, &kq->kq_sigio));
1375 
1376 	case FIOGETOWN:
1377 		*(int *)data = fgetown(&kq->kq_sigio);
1378 		return (0);
1379 	}
1380 #endif
1381 
1382 	return (ENOTTY);
1383 }
1384 
1385 /*ARGSUSED*/
1386 static int
1387 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1388 	struct thread *td)
1389 {
1390 	struct kqueue *kq;
1391 	int revents = 0;
1392 	int error;
1393 
1394 	if ((error = kqueue_acquire(fp, &kq)))
1395 		return POLLERR;
1396 
1397 	KQ_LOCK(kq);
1398 	if (events & (POLLIN | POLLRDNORM)) {
1399 		if (kq->kq_count) {
1400 			revents |= events & (POLLIN | POLLRDNORM);
1401 		} else {
1402 			selrecord(td, &kq->kq_sel);
1403 			kq->kq_state |= KQ_SEL;
1404 		}
1405 	}
1406 	kqueue_release(kq, 1);
1407 	KQ_UNLOCK(kq);
1408 	return (revents);
1409 }
1410 
1411 /*ARGSUSED*/
1412 static int
1413 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1414 	struct thread *td)
1415 {
1416 
1417 	bzero((void *)st, sizeof *st);
1418 	/*
1419 	 * We no longer return kq_count because the unlocked value is useless.
1420 	 * If you spent all this time getting the count, why not spend your
1421 	 * syscall better by calling kevent?
1422 	 *
1423 	 * XXX - This is needed for libc_r.
1424 	 */
1425 	st->st_mode = S_IFIFO;
1426 	return (0);
1427 }
1428 
1429 /*ARGSUSED*/
1430 static int
1431 kqueue_close(struct file *fp, struct thread *td)
1432 {
1433 	struct kqueue *kq = fp->f_data;
1434 	struct filedesc *fdp;
1435 	struct knote *kn;
1436 	int i;
1437 	int error;
1438 
1439 	if ((error = kqueue_acquire(fp, &kq)))
1440 		return error;
1441 
1442 	KQ_LOCK(kq);
1443 
1444 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1445 	    ("kqueue already closing"));
1446 	kq->kq_state |= KQ_CLOSING;
1447 	if (kq->kq_refcnt > 1)
1448 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1449 
1450 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1451 	fdp = kq->kq_fdp;
1452 
1453 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
1454 	    ("kqueue's knlist not empty"));
1455 
1456 	for (i = 0; i < kq->kq_knlistsize; i++) {
1457 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1458 			KASSERT((kn->kn_status & KN_INFLUX) == 0,
1459 			    ("KN_INFLUX set when not suppose to be"));
1460 			kn->kn_status |= KN_INFLUX;
1461 			KQ_UNLOCK(kq);
1462 			if (!(kn->kn_status & KN_DETACHED))
1463 				kn->kn_fop->f_detach(kn);
1464 			knote_drop(kn, td);
1465 			KQ_LOCK(kq);
1466 		}
1467 	}
1468 	if (kq->kq_knhashmask != 0) {
1469 		for (i = 0; i <= kq->kq_knhashmask; i++) {
1470 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1471 				KASSERT((kn->kn_status & KN_INFLUX) == 0,
1472 				    ("KN_INFLUX set when not suppose to be"));
1473 				kn->kn_status |= KN_INFLUX;
1474 				KQ_UNLOCK(kq);
1475 				if (!(kn->kn_status & KN_DETACHED))
1476 					kn->kn_fop->f_detach(kn);
1477 				knote_drop(kn, td);
1478 				KQ_LOCK(kq);
1479 			}
1480 		}
1481 	}
1482 
1483 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1484 		kq->kq_state |= KQ_TASKDRAIN;
1485 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1486 	}
1487 
1488 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1489 		kq->kq_state &= ~KQ_SEL;
1490 		selwakeuppri(&kq->kq_sel, PSOCK);
1491 	}
1492 
1493 	KQ_UNLOCK(kq);
1494 
1495 	FILEDESC_XLOCK(fdp);
1496 	SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list);
1497 	FILEDESC_XUNLOCK(fdp);
1498 
1499 	knlist_destroy(&kq->kq_sel.si_note);
1500 	mtx_destroy(&kq->kq_lock);
1501 	kq->kq_fdp = NULL;
1502 
1503 	if (kq->kq_knhash != NULL)
1504 		free(kq->kq_knhash, M_KQUEUE);
1505 	if (kq->kq_knlist != NULL)
1506 		free(kq->kq_knlist, M_KQUEUE);
1507 
1508 	funsetown(&kq->kq_sigio);
1509 	free(kq, M_KQUEUE);
1510 	fp->f_data = NULL;
1511 
1512 	return (0);
1513 }
1514 
1515 static void
1516 kqueue_wakeup(struct kqueue *kq)
1517 {
1518 	KQ_OWNED(kq);
1519 
1520 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1521 		kq->kq_state &= ~KQ_SLEEP;
1522 		wakeup(kq);
1523 	}
1524 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1525 		kq->kq_state &= ~KQ_SEL;
1526 		selwakeuppri(&kq->kq_sel, PSOCK);
1527 	}
1528 	if (!knlist_empty(&kq->kq_sel.si_note))
1529 		kqueue_schedtask(kq);
1530 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1531 		pgsigio(&kq->kq_sigio, SIGIO, 0);
1532 	}
1533 }
1534 
1535 /*
1536  * Walk down a list of knotes, activating them if their event has triggered.
1537  *
1538  * There is a possibility to optimize in the case of one kq watching another.
1539  * Instead of scheduling a task to wake it up, you could pass enough state
1540  * down the chain to make up the parent kqueue.  Make this code functional
1541  * first.
1542  */
1543 void
1544 knote(struct knlist *list, long hint, int islocked)
1545 {
1546 	struct kqueue *kq;
1547 	struct knote *kn;
1548 
1549 	if (list == NULL)
1550 		return;
1551 
1552 	KNL_ASSERT_LOCK(list, islocked);
1553 
1554 	if (!islocked)
1555 		list->kl_lock(list->kl_lockarg);
1556 
1557 	/*
1558 	 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1559 	 * the kqueue scheduling, but this will introduce four
1560 	 * lock/unlock's for each knote to test.  If we do, continue to use
1561 	 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1562 	 * only safe if you want to remove the current item, which we are
1563 	 * not doing.
1564 	 */
1565 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1566 		kq = kn->kn_kq;
1567 		if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1568 			KQ_LOCK(kq);
1569 			if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1570 				kn->kn_status |= KN_HASKQLOCK;
1571 				if (kn->kn_fop->f_event(kn, hint))
1572 					KNOTE_ACTIVATE(kn, 1);
1573 				kn->kn_status &= ~KN_HASKQLOCK;
1574 			}
1575 			KQ_UNLOCK(kq);
1576 		}
1577 		kq = NULL;
1578 	}
1579 	if (!islocked)
1580 		list->kl_unlock(list->kl_lockarg);
1581 }
1582 
1583 /*
1584  * add a knote to a knlist
1585  */
1586 void
1587 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1588 {
1589 	KNL_ASSERT_LOCK(knl, islocked);
1590 	KQ_NOTOWNED(kn->kn_kq);
1591 	KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1592 	    (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1593 	if (!islocked)
1594 		knl->kl_lock(knl->kl_lockarg);
1595 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1596 	if (!islocked)
1597 		knl->kl_unlock(knl->kl_lockarg);
1598 	KQ_LOCK(kn->kn_kq);
1599 	kn->kn_knlist = knl;
1600 	kn->kn_status &= ~KN_DETACHED;
1601 	KQ_UNLOCK(kn->kn_kq);
1602 }
1603 
1604 static void
1605 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1606 {
1607 	KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1608 	KNL_ASSERT_LOCK(knl, knlislocked);
1609 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1610 	if (!kqislocked)
1611 		KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1612     ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1613 	if (!knlislocked)
1614 		knl->kl_lock(knl->kl_lockarg);
1615 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1616 	kn->kn_knlist = NULL;
1617 	if (!knlislocked)
1618 		knl->kl_unlock(knl->kl_lockarg);
1619 	if (!kqislocked)
1620 		KQ_LOCK(kn->kn_kq);
1621 	kn->kn_status |= KN_DETACHED;
1622 	if (!kqislocked)
1623 		KQ_UNLOCK(kn->kn_kq);
1624 }
1625 
1626 /*
1627  * remove all knotes from a specified klist
1628  */
1629 void
1630 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1631 {
1632 
1633 	knlist_remove_kq(knl, kn, islocked, 0);
1634 }
1635 
1636 /*
1637  * remove knote from a specified klist while in f_event handler.
1638  */
1639 void
1640 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1641 {
1642 
1643 	knlist_remove_kq(knl, kn, 1,
1644 	    (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1645 }
1646 
1647 int
1648 knlist_empty(struct knlist *knl)
1649 {
1650 	KNL_ASSERT_LOCKED(knl);
1651 	return SLIST_EMPTY(&knl->kl_list);
1652 }
1653 
1654 static struct mtx	knlist_lock;
1655 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1656 	MTX_DEF);
1657 static void knlist_mtx_lock(void *arg);
1658 static void knlist_mtx_unlock(void *arg);
1659 static int knlist_mtx_locked(void *arg);
1660 
1661 static void
1662 knlist_mtx_lock(void *arg)
1663 {
1664 	mtx_lock((struct mtx *)arg);
1665 }
1666 
1667 static void
1668 knlist_mtx_unlock(void *arg)
1669 {
1670 	mtx_unlock((struct mtx *)arg);
1671 }
1672 
1673 static int
1674 knlist_mtx_locked(void *arg)
1675 {
1676 	return (mtx_owned((struct mtx *)arg));
1677 }
1678 
1679 void
1680 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1681     void (*kl_unlock)(void *), int (*kl_locked)(void *))
1682 {
1683 
1684 	if (lock == NULL)
1685 		knl->kl_lockarg = &knlist_lock;
1686 	else
1687 		knl->kl_lockarg = lock;
1688 
1689 	if (kl_lock == NULL)
1690 		knl->kl_lock = knlist_mtx_lock;
1691 	else
1692 		knl->kl_lock = kl_lock;
1693 	if (kl_unlock == NULL)
1694 		knl->kl_unlock = knlist_mtx_unlock;
1695 	else
1696 		knl->kl_unlock = kl_unlock;
1697 	if (kl_locked == NULL)
1698 		knl->kl_locked = knlist_mtx_locked;
1699 	else
1700 		knl->kl_locked = kl_locked;
1701 
1702 	SLIST_INIT(&knl->kl_list);
1703 }
1704 
1705 void
1706 knlist_destroy(struct knlist *knl)
1707 {
1708 
1709 #ifdef INVARIANTS
1710 	/*
1711 	 * if we run across this error, we need to find the offending
1712 	 * driver and have it call knlist_clear.
1713 	 */
1714 	if (!SLIST_EMPTY(&knl->kl_list))
1715 		printf("WARNING: destroying knlist w/ knotes on it!\n");
1716 #endif
1717 
1718 	knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
1719 	SLIST_INIT(&knl->kl_list);
1720 }
1721 
1722 /*
1723  * Even if we are locked, we may need to drop the lock to allow any influx
1724  * knotes time to "settle".
1725  */
1726 void
1727 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
1728 {
1729 	struct knote *kn, *kn2;
1730 	struct kqueue *kq;
1731 
1732 	if (islocked)
1733 		KNL_ASSERT_LOCKED(knl);
1734 	else {
1735 		KNL_ASSERT_UNLOCKED(knl);
1736 again:		/* need to reacquire lock since we have dropped it */
1737 		knl->kl_lock(knl->kl_lockarg);
1738 	}
1739 
1740 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
1741 		kq = kn->kn_kq;
1742 		KQ_LOCK(kq);
1743 		if ((kn->kn_status & KN_INFLUX)) {
1744 			KQ_UNLOCK(kq);
1745 			continue;
1746 		}
1747 		knlist_remove_kq(knl, kn, 1, 1);
1748 		if (killkn) {
1749 			kn->kn_status |= KN_INFLUX | KN_DETACHED;
1750 			KQ_UNLOCK(kq);
1751 			knote_drop(kn, td);
1752 		} else {
1753 			/* Make sure cleared knotes disappear soon */
1754 			kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1755 			KQ_UNLOCK(kq);
1756 		}
1757 		kq = NULL;
1758 	}
1759 
1760 	if (!SLIST_EMPTY(&knl->kl_list)) {
1761 		/* there are still KN_INFLUX remaining */
1762 		kn = SLIST_FIRST(&knl->kl_list);
1763 		kq = kn->kn_kq;
1764 		KQ_LOCK(kq);
1765 		KASSERT(kn->kn_status & KN_INFLUX,
1766 		    ("knote removed w/o list lock"));
1767 		knl->kl_unlock(knl->kl_lockarg);
1768 		kq->kq_state |= KQ_FLUXWAIT;
1769 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
1770 		kq = NULL;
1771 		goto again;
1772 	}
1773 
1774 	if (islocked)
1775 		KNL_ASSERT_LOCKED(knl);
1776 	else {
1777 		knl->kl_unlock(knl->kl_lockarg);
1778 		KNL_ASSERT_UNLOCKED(knl);
1779 	}
1780 }
1781 
1782 /*
1783  * Remove all knotes referencing a specified fd must be called with FILEDESC
1784  * lock.  This prevents a race where a new fd comes along and occupies the
1785  * entry and we attach a knote to the fd.
1786  */
1787 void
1788 knote_fdclose(struct thread *td, int fd)
1789 {
1790 	struct filedesc *fdp = td->td_proc->p_fd;
1791 	struct kqueue *kq;
1792 	struct knote *kn;
1793 	int influx;
1794 
1795 	FILEDESC_XLOCK_ASSERT(fdp);
1796 
1797 	/*
1798 	 * We shouldn't have to worry about new kevents appearing on fd
1799 	 * since filedesc is locked.
1800 	 */
1801 	SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
1802 		KQ_LOCK(kq);
1803 
1804 again:
1805 		influx = 0;
1806 		while (kq->kq_knlistsize > fd &&
1807 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
1808 			if (kn->kn_status & KN_INFLUX) {
1809 				/* someone else might be waiting on our knote */
1810 				if (influx)
1811 					wakeup(kq);
1812 				kq->kq_state |= KQ_FLUXWAIT;
1813 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
1814 				goto again;
1815 			}
1816 			kn->kn_status |= KN_INFLUX;
1817 			KQ_UNLOCK(kq);
1818 			if (!(kn->kn_status & KN_DETACHED))
1819 				kn->kn_fop->f_detach(kn);
1820 			knote_drop(kn, td);
1821 			influx = 1;
1822 			KQ_LOCK(kq);
1823 		}
1824 		KQ_UNLOCK_FLUX(kq);
1825 	}
1826 }
1827 
1828 static int
1829 knote_attach(struct knote *kn, struct kqueue *kq)
1830 {
1831 	struct klist *list;
1832 
1833 	KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
1834 	KQ_OWNED(kq);
1835 
1836 	if (kn->kn_fop->f_isfd) {
1837 		if (kn->kn_id >= kq->kq_knlistsize)
1838 			return ENOMEM;
1839 		list = &kq->kq_knlist[kn->kn_id];
1840 	} else {
1841 		if (kq->kq_knhash == NULL)
1842 			return ENOMEM;
1843 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1844 	}
1845 
1846 	SLIST_INSERT_HEAD(list, kn, kn_link);
1847 
1848 	return 0;
1849 }
1850 
1851 /*
1852  * knote must already have been detached using the f_detach method.
1853  * no lock need to be held, it is assumed that the KN_INFLUX flag is set
1854  * to prevent other removal.
1855  */
1856 static void
1857 knote_drop(struct knote *kn, struct thread *td)
1858 {
1859 	struct kqueue *kq;
1860 	struct klist *list;
1861 
1862 	kq = kn->kn_kq;
1863 
1864 	KQ_NOTOWNED(kq);
1865 	KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
1866 	    ("knote_drop called without KN_INFLUX set in kn_status"));
1867 
1868 	KQ_LOCK(kq);
1869 	if (kn->kn_fop->f_isfd)
1870 		list = &kq->kq_knlist[kn->kn_id];
1871 	else
1872 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1873 
1874 	if (!SLIST_EMPTY(list))
1875 		SLIST_REMOVE(list, kn, knote, kn_link);
1876 	if (kn->kn_status & KN_QUEUED)
1877 		knote_dequeue(kn);
1878 	KQ_UNLOCK_FLUX(kq);
1879 
1880 	if (kn->kn_fop->f_isfd) {
1881 		fdrop(kn->kn_fp, td);
1882 		kn->kn_fp = NULL;
1883 	}
1884 	kqueue_fo_release(kn->kn_kevent.filter);
1885 	kn->kn_fop = NULL;
1886 	knote_free(kn);
1887 }
1888 
1889 static void
1890 knote_enqueue(struct knote *kn)
1891 {
1892 	struct kqueue *kq = kn->kn_kq;
1893 
1894 	KQ_OWNED(kn->kn_kq);
1895 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1896 
1897 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1898 	kn->kn_status |= KN_QUEUED;
1899 	kq->kq_count++;
1900 	kqueue_wakeup(kq);
1901 }
1902 
1903 static void
1904 knote_dequeue(struct knote *kn)
1905 {
1906 	struct kqueue *kq = kn->kn_kq;
1907 
1908 	KQ_OWNED(kn->kn_kq);
1909 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1910 
1911 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1912 	kn->kn_status &= ~KN_QUEUED;
1913 	kq->kq_count--;
1914 }
1915 
1916 static void
1917 knote_init(void)
1918 {
1919 
1920 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
1921 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1922 }
1923 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1924 
1925 static struct knote *
1926 knote_alloc(int waitok)
1927 {
1928 	return ((struct knote *)uma_zalloc(knote_zone,
1929 	    (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
1930 }
1931 
1932 static void
1933 knote_free(struct knote *kn)
1934 {
1935 	if (kn != NULL)
1936 		uma_zfree(knote_zone, kn);
1937 }
1938 
1939 /*
1940  * Register the kev w/ the kq specified by fd.
1941  */
1942 int
1943 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
1944 {
1945 	struct kqueue *kq;
1946 	struct file *fp;
1947 	int error;
1948 
1949 	if ((error = fget(td, fd, &fp)) != 0)
1950 		return (error);
1951 	if ((error = kqueue_acquire(fp, &kq)) != 0)
1952 		goto noacquire;
1953 
1954 	error = kqueue_register(kq, kev, td, waitok);
1955 
1956 	kqueue_release(kq, 0);
1957 
1958 noacquire:
1959 	fdrop(fp, td);
1960 
1961 	return error;
1962 }
1963