xref: /freebsd/sys/kern/kern_event.c (revision d8a0fe102c0cfdfcd5b818f850eff09d8536c9bc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
5  * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org>
6  * Copyright (c) 2009 Apple, Inc.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_compat.h"
35 #include "opt_ktrace.h"
36 #include "opt_kqueue.h"
37 
38 #ifdef COMPAT_FREEBSD11
39 #define	_WANT_FREEBSD11_KEVENT
40 #endif
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/capsicum.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/rwlock.h>
49 #include <sys/proc.h>
50 #include <sys/malloc.h>
51 #include <sys/unistd.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/fcntl.h>
56 #include <sys/kthread.h>
57 #include <sys/selinfo.h>
58 #include <sys/queue.h>
59 #include <sys/event.h>
60 #include <sys/eventvar.h>
61 #include <sys/poll.h>
62 #include <sys/protosw.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sigio.h>
65 #include <sys/signalvar.h>
66 #include <sys/socket.h>
67 #include <sys/socketvar.h>
68 #include <sys/stat.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysproto.h>
71 #include <sys/syscallsubr.h>
72 #include <sys/taskqueue.h>
73 #include <sys/uio.h>
74 #include <sys/user.h>
75 #ifdef KTRACE
76 #include <sys/ktrace.h>
77 #endif
78 #include <machine/atomic.h>
79 
80 #include <vm/uma.h>
81 
82 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
83 
84 /*
85  * This lock is used if multiple kq locks are required.  This possibly
86  * should be made into a per proc lock.
87  */
88 static struct mtx	kq_global;
89 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
90 #define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
91 	if (!haslck)				\
92 		mtx_lock(lck);			\
93 	haslck = 1;				\
94 } while (0)
95 #define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
96 	if (haslck)				\
97 		mtx_unlock(lck);			\
98 	haslck = 0;				\
99 } while (0)
100 
101 TASKQUEUE_DEFINE_THREAD(kqueue_ctx);
102 
103 static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
104 static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
105 static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
106 		    struct thread *td, int waitok);
107 static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
108 static void	kqueue_release(struct kqueue *kq, int locked);
109 static void	kqueue_destroy(struct kqueue *kq);
110 static void	kqueue_drain(struct kqueue *kq, struct thread *td);
111 static int	kqueue_expand(struct kqueue *kq, struct filterops *fops,
112 		    uintptr_t ident, int waitok);
113 static void	kqueue_task(void *arg, int pending);
114 static int	kqueue_scan(struct kqueue *kq, int maxevents,
115 		    struct kevent_copyops *k_ops,
116 		    const struct timespec *timeout,
117 		    struct kevent *keva, struct thread *td);
118 static void 	kqueue_wakeup(struct kqueue *kq);
119 static struct filterops *kqueue_fo_find(int filt);
120 static void	kqueue_fo_release(int filt);
121 struct g_kevent_args;
122 static int	kern_kevent_generic(struct thread *td,
123 		    struct g_kevent_args *uap,
124 		    struct kevent_copyops *k_ops, const char *struct_name);
125 
126 static fo_ioctl_t	kqueue_ioctl;
127 static fo_poll_t	kqueue_poll;
128 static fo_kqfilter_t	kqueue_kqfilter;
129 static fo_stat_t	kqueue_stat;
130 static fo_close_t	kqueue_close;
131 static fo_fill_kinfo_t	kqueue_fill_kinfo;
132 
133 static struct fileops kqueueops = {
134 	.fo_read = invfo_rdwr,
135 	.fo_write = invfo_rdwr,
136 	.fo_truncate = invfo_truncate,
137 	.fo_ioctl = kqueue_ioctl,
138 	.fo_poll = kqueue_poll,
139 	.fo_kqfilter = kqueue_kqfilter,
140 	.fo_stat = kqueue_stat,
141 	.fo_close = kqueue_close,
142 	.fo_chmod = invfo_chmod,
143 	.fo_chown = invfo_chown,
144 	.fo_sendfile = invfo_sendfile,
145 	.fo_fill_kinfo = kqueue_fill_kinfo,
146 };
147 
148 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
149 static void 	knote_drop(struct knote *kn, struct thread *td);
150 static void 	knote_drop_detached(struct knote *kn, struct thread *td);
151 static void 	knote_enqueue(struct knote *kn);
152 static void 	knote_dequeue(struct knote *kn);
153 static void 	knote_init(void);
154 static struct 	knote *knote_alloc(int waitok);
155 static void 	knote_free(struct knote *kn);
156 
157 static void	filt_kqdetach(struct knote *kn);
158 static int	filt_kqueue(struct knote *kn, long hint);
159 static int	filt_procattach(struct knote *kn);
160 static void	filt_procdetach(struct knote *kn);
161 static int	filt_proc(struct knote *kn, long hint);
162 static int	filt_fileattach(struct knote *kn);
163 static void	filt_timerexpire(void *knx);
164 static int	filt_timerattach(struct knote *kn);
165 static void	filt_timerdetach(struct knote *kn);
166 static int	filt_timer(struct knote *kn, long hint);
167 static int	filt_userattach(struct knote *kn);
168 static void	filt_userdetach(struct knote *kn);
169 static int	filt_user(struct knote *kn, long hint);
170 static void	filt_usertouch(struct knote *kn, struct kevent *kev,
171 		    u_long type);
172 
173 static struct filterops file_filtops = {
174 	.f_isfd = 1,
175 	.f_attach = filt_fileattach,
176 };
177 static struct filterops kqread_filtops = {
178 	.f_isfd = 1,
179 	.f_detach = filt_kqdetach,
180 	.f_event = filt_kqueue,
181 };
182 /* XXX - move to kern_proc.c?  */
183 static struct filterops proc_filtops = {
184 	.f_isfd = 0,
185 	.f_attach = filt_procattach,
186 	.f_detach = filt_procdetach,
187 	.f_event = filt_proc,
188 };
189 static struct filterops timer_filtops = {
190 	.f_isfd = 0,
191 	.f_attach = filt_timerattach,
192 	.f_detach = filt_timerdetach,
193 	.f_event = filt_timer,
194 };
195 static struct filterops user_filtops = {
196 	.f_attach = filt_userattach,
197 	.f_detach = filt_userdetach,
198 	.f_event = filt_user,
199 	.f_touch = filt_usertouch,
200 };
201 
202 static uma_zone_t	knote_zone;
203 static unsigned int	kq_ncallouts = 0;
204 static unsigned int 	kq_calloutmax = 4 * 1024;
205 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
206     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
207 
208 /* XXX - ensure not influx ? */
209 #define KNOTE_ACTIVATE(kn, islock) do { 				\
210 	if ((islock))							\
211 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
212 	else								\
213 		KQ_LOCK((kn)->kn_kq);					\
214 	(kn)->kn_status |= KN_ACTIVE;					\
215 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
216 		knote_enqueue((kn));					\
217 	if (!(islock))							\
218 		KQ_UNLOCK((kn)->kn_kq);					\
219 } while(0)
220 #define KQ_LOCK(kq) do {						\
221 	mtx_lock(&(kq)->kq_lock);					\
222 } while (0)
223 #define KQ_FLUX_WAKEUP(kq) do {						\
224 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
225 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
226 		wakeup((kq));						\
227 	}								\
228 } while (0)
229 #define KQ_UNLOCK_FLUX(kq) do {						\
230 	KQ_FLUX_WAKEUP(kq);						\
231 	mtx_unlock(&(kq)->kq_lock);					\
232 } while (0)
233 #define KQ_UNLOCK(kq) do {						\
234 	mtx_unlock(&(kq)->kq_lock);					\
235 } while (0)
236 #define KQ_OWNED(kq) do {						\
237 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
238 } while (0)
239 #define KQ_NOTOWNED(kq) do {						\
240 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
241 } while (0)
242 
243 static struct knlist *
244 kn_list_lock(struct knote *kn)
245 {
246 	struct knlist *knl;
247 
248 	knl = kn->kn_knlist;
249 	if (knl != NULL)
250 		knl->kl_lock(knl->kl_lockarg);
251 	return (knl);
252 }
253 
254 static void
255 kn_list_unlock(struct knlist *knl)
256 {
257 	bool do_free;
258 
259 	if (knl == NULL)
260 		return;
261 	do_free = knl->kl_autodestroy && knlist_empty(knl);
262 	knl->kl_unlock(knl->kl_lockarg);
263 	if (do_free) {
264 		knlist_destroy(knl);
265 		free(knl, M_KQUEUE);
266 	}
267 }
268 
269 static bool
270 kn_in_flux(struct knote *kn)
271 {
272 
273 	return (kn->kn_influx > 0);
274 }
275 
276 static void
277 kn_enter_flux(struct knote *kn)
278 {
279 
280 	KQ_OWNED(kn->kn_kq);
281 	MPASS(kn->kn_influx < INT_MAX);
282 	kn->kn_influx++;
283 }
284 
285 static bool
286 kn_leave_flux(struct knote *kn)
287 {
288 
289 	KQ_OWNED(kn->kn_kq);
290 	MPASS(kn->kn_influx > 0);
291 	kn->kn_influx--;
292 	return (kn->kn_influx == 0);
293 }
294 
295 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
296 	if (islocked)							\
297 		KNL_ASSERT_LOCKED(knl);				\
298 	else								\
299 		KNL_ASSERT_UNLOCKED(knl);				\
300 } while (0)
301 #ifdef INVARIANTS
302 #define	KNL_ASSERT_LOCKED(knl) do {					\
303 	knl->kl_assert_locked((knl)->kl_lockarg);			\
304 } while (0)
305 #define	KNL_ASSERT_UNLOCKED(knl) do {					\
306 	knl->kl_assert_unlocked((knl)->kl_lockarg);			\
307 } while (0)
308 #else /* !INVARIANTS */
309 #define	KNL_ASSERT_LOCKED(knl) do {} while(0)
310 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
311 #endif /* INVARIANTS */
312 
313 #ifndef	KN_HASHSIZE
314 #define	KN_HASHSIZE		64		/* XXX should be tunable */
315 #endif
316 
317 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
318 
319 static int
320 filt_nullattach(struct knote *kn)
321 {
322 
323 	return (ENXIO);
324 };
325 
326 struct filterops null_filtops = {
327 	.f_isfd = 0,
328 	.f_attach = filt_nullattach,
329 };
330 
331 /* XXX - make SYSINIT to add these, and move into respective modules. */
332 extern struct filterops sig_filtops;
333 extern struct filterops fs_filtops;
334 
335 /*
336  * Table for for all system-defined filters.
337  */
338 static struct mtx	filterops_lock;
339 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
340 	MTX_DEF);
341 static struct {
342 	struct filterops *for_fop;
343 	int for_nolock;
344 	int for_refcnt;
345 } sysfilt_ops[EVFILT_SYSCOUNT] = {
346 	{ &file_filtops, 1 },			/* EVFILT_READ */
347 	{ &file_filtops, 1 },			/* EVFILT_WRITE */
348 	{ &null_filtops },			/* EVFILT_AIO */
349 	{ &file_filtops, 1 },			/* EVFILT_VNODE */
350 	{ &proc_filtops, 1 },			/* EVFILT_PROC */
351 	{ &sig_filtops, 1 },			/* EVFILT_SIGNAL */
352 	{ &timer_filtops, 1 },			/* EVFILT_TIMER */
353 	{ &file_filtops, 1 },			/* EVFILT_PROCDESC */
354 	{ &fs_filtops, 1 },			/* EVFILT_FS */
355 	{ &null_filtops },			/* EVFILT_LIO */
356 	{ &user_filtops, 1 },			/* EVFILT_USER */
357 	{ &null_filtops },			/* EVFILT_SENDFILE */
358 	{ &file_filtops, 1 },                   /* EVFILT_EMPTY */
359 };
360 
361 /*
362  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
363  * method.
364  */
365 static int
366 filt_fileattach(struct knote *kn)
367 {
368 
369 	return (fo_kqfilter(kn->kn_fp, kn));
370 }
371 
372 /*ARGSUSED*/
373 static int
374 kqueue_kqfilter(struct file *fp, struct knote *kn)
375 {
376 	struct kqueue *kq = kn->kn_fp->f_data;
377 
378 	if (kn->kn_filter != EVFILT_READ)
379 		return (EINVAL);
380 
381 	kn->kn_status |= KN_KQUEUE;
382 	kn->kn_fop = &kqread_filtops;
383 	knlist_add(&kq->kq_sel.si_note, kn, 0);
384 
385 	return (0);
386 }
387 
388 static void
389 filt_kqdetach(struct knote *kn)
390 {
391 	struct kqueue *kq = kn->kn_fp->f_data;
392 
393 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
394 }
395 
396 /*ARGSUSED*/
397 static int
398 filt_kqueue(struct knote *kn, long hint)
399 {
400 	struct kqueue *kq = kn->kn_fp->f_data;
401 
402 	kn->kn_data = kq->kq_count;
403 	return (kn->kn_data > 0);
404 }
405 
406 /* XXX - move to kern_proc.c?  */
407 static int
408 filt_procattach(struct knote *kn)
409 {
410 	struct proc *p;
411 	int error;
412 	bool exiting, immediate;
413 
414 	exiting = immediate = false;
415 	if (kn->kn_sfflags & NOTE_EXIT)
416 		p = pfind_any(kn->kn_id);
417 	else
418 		p = pfind(kn->kn_id);
419 	if (p == NULL)
420 		return (ESRCH);
421 	if (p->p_flag & P_WEXIT)
422 		exiting = true;
423 
424 	if ((error = p_cansee(curthread, p))) {
425 		PROC_UNLOCK(p);
426 		return (error);
427 	}
428 
429 	kn->kn_ptr.p_proc = p;
430 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
431 
432 	/*
433 	 * Internal flag indicating registration done by kernel for the
434 	 * purposes of getting a NOTE_CHILD notification.
435 	 */
436 	if (kn->kn_flags & EV_FLAG2) {
437 		kn->kn_flags &= ~EV_FLAG2;
438 		kn->kn_data = kn->kn_sdata;		/* ppid */
439 		kn->kn_fflags = NOTE_CHILD;
440 		kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
441 		immediate = true; /* Force immediate activation of child note. */
442 	}
443 	/*
444 	 * Internal flag indicating registration done by kernel (for other than
445 	 * NOTE_CHILD).
446 	 */
447 	if (kn->kn_flags & EV_FLAG1) {
448 		kn->kn_flags &= ~EV_FLAG1;
449 	}
450 
451 	knlist_add(p->p_klist, kn, 1);
452 
453 	/*
454 	 * Immediately activate any child notes or, in the case of a zombie
455 	 * target process, exit notes.  The latter is necessary to handle the
456 	 * case where the target process, e.g. a child, dies before the kevent
457 	 * is registered.
458 	 */
459 	if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
460 		KNOTE_ACTIVATE(kn, 0);
461 
462 	PROC_UNLOCK(p);
463 
464 	return (0);
465 }
466 
467 /*
468  * The knote may be attached to a different process, which may exit,
469  * leaving nothing for the knote to be attached to.  So when the process
470  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
471  * it will be deleted when read out.  However, as part of the knote deletion,
472  * this routine is called, so a check is needed to avoid actually performing
473  * a detach, because the original process does not exist any more.
474  */
475 /* XXX - move to kern_proc.c?  */
476 static void
477 filt_procdetach(struct knote *kn)
478 {
479 
480 	knlist_remove(kn->kn_knlist, kn, 0);
481 	kn->kn_ptr.p_proc = NULL;
482 }
483 
484 /* XXX - move to kern_proc.c?  */
485 static int
486 filt_proc(struct knote *kn, long hint)
487 {
488 	struct proc *p;
489 	u_int event;
490 
491 	p = kn->kn_ptr.p_proc;
492 	if (p == NULL) /* already activated, from attach filter */
493 		return (0);
494 
495 	/* Mask off extra data. */
496 	event = (u_int)hint & NOTE_PCTRLMASK;
497 
498 	/* If the user is interested in this event, record it. */
499 	if (kn->kn_sfflags & event)
500 		kn->kn_fflags |= event;
501 
502 	/* Process is gone, so flag the event as finished. */
503 	if (event == NOTE_EXIT) {
504 		kn->kn_flags |= EV_EOF | EV_ONESHOT;
505 		kn->kn_ptr.p_proc = NULL;
506 		if (kn->kn_fflags & NOTE_EXIT)
507 			kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
508 		if (kn->kn_fflags == 0)
509 			kn->kn_flags |= EV_DROP;
510 		return (1);
511 	}
512 
513 	return (kn->kn_fflags != 0);
514 }
515 
516 /*
517  * Called when the process forked. It mostly does the same as the
518  * knote(), activating all knotes registered to be activated when the
519  * process forked. Additionally, for each knote attached to the
520  * parent, check whether user wants to track the new process. If so
521  * attach a new knote to it, and immediately report an event with the
522  * child's pid.
523  */
524 void
525 knote_fork(struct knlist *list, int pid)
526 {
527 	struct kqueue *kq;
528 	struct knote *kn;
529 	struct kevent kev;
530 	int error;
531 
532 	if (list == NULL)
533 		return;
534 	list->kl_lock(list->kl_lockarg);
535 
536 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
537 		kq = kn->kn_kq;
538 		KQ_LOCK(kq);
539 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
540 			KQ_UNLOCK(kq);
541 			continue;
542 		}
543 
544 		/*
545 		 * The same as knote(), activate the event.
546 		 */
547 		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
548 			kn->kn_status |= KN_HASKQLOCK;
549 			if (kn->kn_fop->f_event(kn, NOTE_FORK))
550 				KNOTE_ACTIVATE(kn, 1);
551 			kn->kn_status &= ~KN_HASKQLOCK;
552 			KQ_UNLOCK(kq);
553 			continue;
554 		}
555 
556 		/*
557 		 * The NOTE_TRACK case. In addition to the activation
558 		 * of the event, we need to register new events to
559 		 * track the child. Drop the locks in preparation for
560 		 * the call to kqueue_register().
561 		 */
562 		kn_enter_flux(kn);
563 		KQ_UNLOCK(kq);
564 		list->kl_unlock(list->kl_lockarg);
565 
566 		/*
567 		 * Activate existing knote and register tracking knotes with
568 		 * new process.
569 		 *
570 		 * First register a knote to get just the child notice. This
571 		 * must be a separate note from a potential NOTE_EXIT
572 		 * notification since both NOTE_CHILD and NOTE_EXIT are defined
573 		 * to use the data field (in conflicting ways).
574 		 */
575 		kev.ident = pid;
576 		kev.filter = kn->kn_filter;
577 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
578 		    EV_FLAG2;
579 		kev.fflags = kn->kn_sfflags;
580 		kev.data = kn->kn_id;		/* parent */
581 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
582 		error = kqueue_register(kq, &kev, NULL, 0);
583 		if (error)
584 			kn->kn_fflags |= NOTE_TRACKERR;
585 
586 		/*
587 		 * Then register another knote to track other potential events
588 		 * from the new process.
589 		 */
590 		kev.ident = pid;
591 		kev.filter = kn->kn_filter;
592 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
593 		kev.fflags = kn->kn_sfflags;
594 		kev.data = kn->kn_id;		/* parent */
595 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
596 		error = kqueue_register(kq, &kev, NULL, 0);
597 		if (error)
598 			kn->kn_fflags |= NOTE_TRACKERR;
599 		if (kn->kn_fop->f_event(kn, NOTE_FORK))
600 			KNOTE_ACTIVATE(kn, 0);
601 		KQ_LOCK(kq);
602 		kn_leave_flux(kn);
603 		KQ_UNLOCK_FLUX(kq);
604 		list->kl_lock(list->kl_lockarg);
605 	}
606 	list->kl_unlock(list->kl_lockarg);
607 }
608 
609 /*
610  * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
611  * interval timer support code.
612  */
613 
614 #define NOTE_TIMER_PRECMASK						\
615     (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)
616 
617 static sbintime_t
618 timer2sbintime(intptr_t data, int flags)
619 {
620 	int64_t secs;
621 
622         /*
623          * Macros for converting to the fractional second portion of an
624          * sbintime_t using 64bit multiplication to improve precision.
625          */
626 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
627 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
628 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
629 	switch (flags & NOTE_TIMER_PRECMASK) {
630 	case NOTE_SECONDS:
631 #ifdef __LP64__
632 		if (data > (SBT_MAX / SBT_1S))
633 			return (SBT_MAX);
634 #endif
635 		return ((sbintime_t)data << 32);
636 	case NOTE_MSECONDS: /* FALLTHROUGH */
637 	case 0:
638 		if (data >= 1000) {
639 			secs = data / 1000;
640 #ifdef __LP64__
641 			if (secs > (SBT_MAX / SBT_1S))
642 				return (SBT_MAX);
643 #endif
644 			return (secs << 32 | MS_TO_SBT(data % 1000));
645 		}
646 		return (MS_TO_SBT(data));
647 	case NOTE_USECONDS:
648 		if (data >= 1000000) {
649 			secs = data / 1000000;
650 #ifdef __LP64__
651 			if (secs > (SBT_MAX / SBT_1S))
652 				return (SBT_MAX);
653 #endif
654 			return (secs << 32 | US_TO_SBT(data % 1000000));
655 		}
656 		return (US_TO_SBT(data));
657 	case NOTE_NSECONDS:
658 		if (data >= 1000000000) {
659 			secs = data / 1000000000;
660 #ifdef __LP64__
661 			if (secs > (SBT_MAX / SBT_1S))
662 				return (SBT_MAX);
663 #endif
664 			return (secs << 32 | US_TO_SBT(data % 1000000000));
665 		}
666 		return (NS_TO_SBT(data));
667 	default:
668 		break;
669 	}
670 	return (-1);
671 }
672 
673 struct kq_timer_cb_data {
674 	struct callout c;
675 	sbintime_t next;	/* next timer event fires at */
676 	sbintime_t to;		/* precalculated timer period, 0 for abs */
677 };
678 
679 static void
680 filt_timerexpire(void *knx)
681 {
682 	struct knote *kn;
683 	struct kq_timer_cb_data *kc;
684 
685 	kn = knx;
686 	kn->kn_data++;
687 	KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
688 
689 	if ((kn->kn_flags & EV_ONESHOT) != 0)
690 		return;
691 	kc = kn->kn_ptr.p_v;
692 	if (kc->to == 0)
693 		return;
694 	kc->next += kc->to;
695 	callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn,
696 	    PCPU_GET(cpuid), C_ABSOLUTE);
697 }
698 
699 /*
700  * data contains amount of time to sleep
701  */
702 static int
703 filt_timerattach(struct knote *kn)
704 {
705 	struct kq_timer_cb_data *kc;
706 	struct bintime bt;
707 	sbintime_t to, sbt;
708 	unsigned int ncallouts;
709 
710 	if (kn->kn_sdata < 0)
711 		return (EINVAL);
712 	if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
713 		kn->kn_sdata = 1;
714 	/* Only precision unit are supported in flags so far */
715 	if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0)
716 		return (EINVAL);
717 
718 	to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
719 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
720 		getboottimebin(&bt);
721 		sbt = bttosbt(bt);
722 		to -= sbt;
723 	}
724 	if (to < 0)
725 		return (EINVAL);
726 
727 	do {
728 		ncallouts = kq_ncallouts;
729 		if (ncallouts >= kq_calloutmax)
730 			return (ENOMEM);
731 	} while (!atomic_cmpset_int(&kq_ncallouts, ncallouts, ncallouts + 1));
732 
733 	if ((kn->kn_sfflags & NOTE_ABSTIME) == 0)
734 		kn->kn_flags |= EV_CLEAR;	/* automatically set */
735 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
736 	kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
737 	callout_init(&kc->c, 1);
738 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
739 		kc->next = to;
740 		kc->to = 0;
741 	} else {
742 		kc->next = to + sbinuptime();
743 		kc->to = to;
744 	}
745 	callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn,
746 	    PCPU_GET(cpuid), C_ABSOLUTE);
747 
748 	return (0);
749 }
750 
751 static void
752 filt_timerdetach(struct knote *kn)
753 {
754 	struct kq_timer_cb_data *kc;
755 	unsigned int old;
756 
757 	kc = kn->kn_ptr.p_v;
758 	callout_drain(&kc->c);
759 	free(kc, M_KQUEUE);
760 	old = atomic_fetchadd_int(&kq_ncallouts, -1);
761 	KASSERT(old > 0, ("Number of callouts cannot become negative"));
762 	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
763 }
764 
765 static int
766 filt_timer(struct knote *kn, long hint)
767 {
768 
769 	return (kn->kn_data != 0);
770 }
771 
772 static int
773 filt_userattach(struct knote *kn)
774 {
775 
776 	/*
777 	 * EVFILT_USER knotes are not attached to anything in the kernel.
778 	 */
779 	kn->kn_hook = NULL;
780 	if (kn->kn_fflags & NOTE_TRIGGER)
781 		kn->kn_hookid = 1;
782 	else
783 		kn->kn_hookid = 0;
784 	return (0);
785 }
786 
787 static void
788 filt_userdetach(__unused struct knote *kn)
789 {
790 
791 	/*
792 	 * EVFILT_USER knotes are not attached to anything in the kernel.
793 	 */
794 }
795 
796 static int
797 filt_user(struct knote *kn, __unused long hint)
798 {
799 
800 	return (kn->kn_hookid);
801 }
802 
803 static void
804 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
805 {
806 	u_int ffctrl;
807 
808 	switch (type) {
809 	case EVENT_REGISTER:
810 		if (kev->fflags & NOTE_TRIGGER)
811 			kn->kn_hookid = 1;
812 
813 		ffctrl = kev->fflags & NOTE_FFCTRLMASK;
814 		kev->fflags &= NOTE_FFLAGSMASK;
815 		switch (ffctrl) {
816 		case NOTE_FFNOP:
817 			break;
818 
819 		case NOTE_FFAND:
820 			kn->kn_sfflags &= kev->fflags;
821 			break;
822 
823 		case NOTE_FFOR:
824 			kn->kn_sfflags |= kev->fflags;
825 			break;
826 
827 		case NOTE_FFCOPY:
828 			kn->kn_sfflags = kev->fflags;
829 			break;
830 
831 		default:
832 			/* XXX Return error? */
833 			break;
834 		}
835 		kn->kn_sdata = kev->data;
836 		if (kev->flags & EV_CLEAR) {
837 			kn->kn_hookid = 0;
838 			kn->kn_data = 0;
839 			kn->kn_fflags = 0;
840 		}
841 		break;
842 
843         case EVENT_PROCESS:
844 		*kev = kn->kn_kevent;
845 		kev->fflags = kn->kn_sfflags;
846 		kev->data = kn->kn_sdata;
847 		if (kn->kn_flags & EV_CLEAR) {
848 			kn->kn_hookid = 0;
849 			kn->kn_data = 0;
850 			kn->kn_fflags = 0;
851 		}
852 		break;
853 
854 	default:
855 		panic("filt_usertouch() - invalid type (%ld)", type);
856 		break;
857 	}
858 }
859 
860 int
861 sys_kqueue(struct thread *td, struct kqueue_args *uap)
862 {
863 
864 	return (kern_kqueue(td, 0, NULL));
865 }
866 
867 static void
868 kqueue_init(struct kqueue *kq)
869 {
870 
871 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
872 	TAILQ_INIT(&kq->kq_head);
873 	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
874 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
875 }
876 
877 int
878 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps)
879 {
880 	struct filedesc *fdp;
881 	struct kqueue *kq;
882 	struct file *fp;
883 	struct ucred *cred;
884 	int fd, error;
885 
886 	fdp = td->td_proc->p_fd;
887 	cred = td->td_ucred;
888 	if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
889 		return (ENOMEM);
890 
891 	error = falloc_caps(td, &fp, &fd, flags, fcaps);
892 	if (error != 0) {
893 		chgkqcnt(cred->cr_ruidinfo, -1, 0);
894 		return (error);
895 	}
896 
897 	/* An extra reference on `fp' has been held for us by falloc(). */
898 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
899 	kqueue_init(kq);
900 	kq->kq_fdp = fdp;
901 	kq->kq_cred = crhold(cred);
902 
903 	FILEDESC_XLOCK(fdp);
904 	TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
905 	FILEDESC_XUNLOCK(fdp);
906 
907 	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
908 	fdrop(fp, td);
909 
910 	td->td_retval[0] = fd;
911 	return (0);
912 }
913 
914 struct g_kevent_args {
915 	int	fd;
916 	void	*changelist;
917 	int	nchanges;
918 	void	*eventlist;
919 	int	nevents;
920 	const struct timespec *timeout;
921 };
922 
923 int
924 sys_kevent(struct thread *td, struct kevent_args *uap)
925 {
926 	struct kevent_copyops k_ops = {
927 		.arg = uap,
928 		.k_copyout = kevent_copyout,
929 		.k_copyin = kevent_copyin,
930 		.kevent_size = sizeof(struct kevent),
931 	};
932 	struct g_kevent_args gk_args = {
933 		.fd = uap->fd,
934 		.changelist = uap->changelist,
935 		.nchanges = uap->nchanges,
936 		.eventlist = uap->eventlist,
937 		.nevents = uap->nevents,
938 		.timeout = uap->timeout,
939 	};
940 
941 	return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent"));
942 }
943 
944 static int
945 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap,
946     struct kevent_copyops *k_ops, const char *struct_name)
947 {
948 	struct timespec ts, *tsp;
949 #ifdef KTRACE
950 	struct kevent *eventlist = uap->eventlist;
951 #endif
952 	int error;
953 
954 	if (uap->timeout != NULL) {
955 		error = copyin(uap->timeout, &ts, sizeof(ts));
956 		if (error)
957 			return (error);
958 		tsp = &ts;
959 	} else
960 		tsp = NULL;
961 
962 #ifdef KTRACE
963 	if (KTRPOINT(td, KTR_STRUCT_ARRAY))
964 		ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist,
965 		    uap->nchanges, k_ops->kevent_size);
966 #endif
967 
968 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
969 	    k_ops, tsp);
970 
971 #ifdef KTRACE
972 	if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
973 		ktrstructarray(struct_name, UIO_USERSPACE, eventlist,
974 		    td->td_retval[0], k_ops->kevent_size);
975 #endif
976 
977 	return (error);
978 }
979 
980 /*
981  * Copy 'count' items into the destination list pointed to by uap->eventlist.
982  */
983 static int
984 kevent_copyout(void *arg, struct kevent *kevp, int count)
985 {
986 	struct kevent_args *uap;
987 	int error;
988 
989 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
990 	uap = (struct kevent_args *)arg;
991 
992 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
993 	if (error == 0)
994 		uap->eventlist += count;
995 	return (error);
996 }
997 
998 /*
999  * Copy 'count' items from the list pointed to by uap->changelist.
1000  */
1001 static int
1002 kevent_copyin(void *arg, struct kevent *kevp, int count)
1003 {
1004 	struct kevent_args *uap;
1005 	int error;
1006 
1007 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1008 	uap = (struct kevent_args *)arg;
1009 
1010 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
1011 	if (error == 0)
1012 		uap->changelist += count;
1013 	return (error);
1014 }
1015 
1016 #ifdef COMPAT_FREEBSD11
1017 static int
1018 kevent11_copyout(void *arg, struct kevent *kevp, int count)
1019 {
1020 	struct freebsd11_kevent_args *uap;
1021 	struct kevent_freebsd11 kev11;
1022 	int error, i;
1023 
1024 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1025 	uap = (struct freebsd11_kevent_args *)arg;
1026 
1027 	for (i = 0; i < count; i++) {
1028 		kev11.ident = kevp->ident;
1029 		kev11.filter = kevp->filter;
1030 		kev11.flags = kevp->flags;
1031 		kev11.fflags = kevp->fflags;
1032 		kev11.data = kevp->data;
1033 		kev11.udata = kevp->udata;
1034 		error = copyout(&kev11, uap->eventlist, sizeof(kev11));
1035 		if (error != 0)
1036 			break;
1037 		uap->eventlist++;
1038 		kevp++;
1039 	}
1040 	return (error);
1041 }
1042 
1043 /*
1044  * Copy 'count' items from the list pointed to by uap->changelist.
1045  */
1046 static int
1047 kevent11_copyin(void *arg, struct kevent *kevp, int count)
1048 {
1049 	struct freebsd11_kevent_args *uap;
1050 	struct kevent_freebsd11 kev11;
1051 	int error, i;
1052 
1053 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1054 	uap = (struct freebsd11_kevent_args *)arg;
1055 
1056 	for (i = 0; i < count; i++) {
1057 		error = copyin(uap->changelist, &kev11, sizeof(kev11));
1058 		if (error != 0)
1059 			break;
1060 		kevp->ident = kev11.ident;
1061 		kevp->filter = kev11.filter;
1062 		kevp->flags = kev11.flags;
1063 		kevp->fflags = kev11.fflags;
1064 		kevp->data = (uintptr_t)kev11.data;
1065 		kevp->udata = kev11.udata;
1066 		bzero(&kevp->ext, sizeof(kevp->ext));
1067 		uap->changelist++;
1068 		kevp++;
1069 	}
1070 	return (error);
1071 }
1072 
1073 int
1074 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap)
1075 {
1076 	struct kevent_copyops k_ops = {
1077 		.arg = uap,
1078 		.k_copyout = kevent11_copyout,
1079 		.k_copyin = kevent11_copyin,
1080 		.kevent_size = sizeof(struct kevent_freebsd11),
1081 	};
1082 	struct g_kevent_args gk_args = {
1083 		.fd = uap->fd,
1084 		.changelist = uap->changelist,
1085 		.nchanges = uap->nchanges,
1086 		.eventlist = uap->eventlist,
1087 		.nevents = uap->nevents,
1088 		.timeout = uap->timeout,
1089 	};
1090 
1091 	return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent_freebsd11"));
1092 }
1093 #endif
1094 
1095 int
1096 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
1097     struct kevent_copyops *k_ops, const struct timespec *timeout)
1098 {
1099 	cap_rights_t rights;
1100 	struct file *fp;
1101 	int error;
1102 
1103 	cap_rights_init(&rights);
1104 	if (nchanges > 0)
1105 		cap_rights_set(&rights, CAP_KQUEUE_CHANGE);
1106 	if (nevents > 0)
1107 		cap_rights_set(&rights, CAP_KQUEUE_EVENT);
1108 	error = fget(td, fd, &rights, &fp);
1109 	if (error != 0)
1110 		return (error);
1111 
1112 	error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
1113 	fdrop(fp, td);
1114 
1115 	return (error);
1116 }
1117 
1118 static int
1119 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
1120     struct kevent_copyops *k_ops, const struct timespec *timeout)
1121 {
1122 	struct kevent keva[KQ_NEVENTS];
1123 	struct kevent *kevp, *changes;
1124 	int i, n, nerrors, error;
1125 
1126 	nerrors = 0;
1127 	while (nchanges > 0) {
1128 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
1129 		error = k_ops->k_copyin(k_ops->arg, keva, n);
1130 		if (error)
1131 			return (error);
1132 		changes = keva;
1133 		for (i = 0; i < n; i++) {
1134 			kevp = &changes[i];
1135 			if (!kevp->filter)
1136 				continue;
1137 			kevp->flags &= ~EV_SYSFLAGS;
1138 			error = kqueue_register(kq, kevp, td, 1);
1139 			if (error || (kevp->flags & EV_RECEIPT)) {
1140 				if (nevents == 0)
1141 					return (error);
1142 				kevp->flags = EV_ERROR;
1143 				kevp->data = error;
1144 				(void)k_ops->k_copyout(k_ops->arg, kevp, 1);
1145 				nevents--;
1146 				nerrors++;
1147 			}
1148 		}
1149 		nchanges -= n;
1150 	}
1151 	if (nerrors) {
1152 		td->td_retval[0] = nerrors;
1153 		return (0);
1154 	}
1155 
1156 	return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
1157 }
1158 
1159 int
1160 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
1161     struct kevent_copyops *k_ops, const struct timespec *timeout)
1162 {
1163 	struct kqueue *kq;
1164 	int error;
1165 
1166 	error = kqueue_acquire(fp, &kq);
1167 	if (error != 0)
1168 		return (error);
1169 	error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
1170 	kqueue_release(kq, 0);
1171 	return (error);
1172 }
1173 
1174 /*
1175  * Performs a kevent() call on a temporarily created kqueue. This can be
1176  * used to perform one-shot polling, similar to poll() and select().
1177  */
1178 int
1179 kern_kevent_anonymous(struct thread *td, int nevents,
1180     struct kevent_copyops *k_ops)
1181 {
1182 	struct kqueue kq = {};
1183 	int error;
1184 
1185 	kqueue_init(&kq);
1186 	kq.kq_refcnt = 1;
1187 	error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
1188 	kqueue_drain(&kq, td);
1189 	kqueue_destroy(&kq);
1190 	return (error);
1191 }
1192 
1193 int
1194 kqueue_add_filteropts(int filt, struct filterops *filtops)
1195 {
1196 	int error;
1197 
1198 	error = 0;
1199 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
1200 		printf(
1201 "trying to add a filterop that is out of range: %d is beyond %d\n",
1202 		    ~filt, EVFILT_SYSCOUNT);
1203 		return EINVAL;
1204 	}
1205 	mtx_lock(&filterops_lock);
1206 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
1207 	    sysfilt_ops[~filt].for_fop != NULL)
1208 		error = EEXIST;
1209 	else {
1210 		sysfilt_ops[~filt].for_fop = filtops;
1211 		sysfilt_ops[~filt].for_refcnt = 0;
1212 	}
1213 	mtx_unlock(&filterops_lock);
1214 
1215 	return (error);
1216 }
1217 
1218 int
1219 kqueue_del_filteropts(int filt)
1220 {
1221 	int error;
1222 
1223 	error = 0;
1224 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1225 		return EINVAL;
1226 
1227 	mtx_lock(&filterops_lock);
1228 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1229 	    sysfilt_ops[~filt].for_fop == NULL)
1230 		error = EINVAL;
1231 	else if (sysfilt_ops[~filt].for_refcnt != 0)
1232 		error = EBUSY;
1233 	else {
1234 		sysfilt_ops[~filt].for_fop = &null_filtops;
1235 		sysfilt_ops[~filt].for_refcnt = 0;
1236 	}
1237 	mtx_unlock(&filterops_lock);
1238 
1239 	return error;
1240 }
1241 
1242 static struct filterops *
1243 kqueue_fo_find(int filt)
1244 {
1245 
1246 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1247 		return NULL;
1248 
1249 	if (sysfilt_ops[~filt].for_nolock)
1250 		return sysfilt_ops[~filt].for_fop;
1251 
1252 	mtx_lock(&filterops_lock);
1253 	sysfilt_ops[~filt].for_refcnt++;
1254 	if (sysfilt_ops[~filt].for_fop == NULL)
1255 		sysfilt_ops[~filt].for_fop = &null_filtops;
1256 	mtx_unlock(&filterops_lock);
1257 
1258 	return sysfilt_ops[~filt].for_fop;
1259 }
1260 
1261 static void
1262 kqueue_fo_release(int filt)
1263 {
1264 
1265 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1266 		return;
1267 
1268 	if (sysfilt_ops[~filt].for_nolock)
1269 		return;
1270 
1271 	mtx_lock(&filterops_lock);
1272 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1273 	    ("filter object refcount not valid on release"));
1274 	sysfilt_ops[~filt].for_refcnt--;
1275 	mtx_unlock(&filterops_lock);
1276 }
1277 
1278 /*
1279  * A ref to kq (obtained via kqueue_acquire) must be held.  waitok will
1280  * influence if memory allocation should wait.  Make sure it is 0 if you
1281  * hold any mutexes.
1282  */
1283 static int
1284 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
1285 {
1286 	struct filterops *fops;
1287 	struct file *fp;
1288 	struct knote *kn, *tkn;
1289 	struct knlist *knl;
1290 	cap_rights_t rights;
1291 	int error, filt, event;
1292 	int haskqglobal, filedesc_unlock;
1293 
1294 	if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
1295 		return (EINVAL);
1296 
1297 	fp = NULL;
1298 	kn = NULL;
1299 	knl = NULL;
1300 	error = 0;
1301 	haskqglobal = 0;
1302 	filedesc_unlock = 0;
1303 
1304 	filt = kev->filter;
1305 	fops = kqueue_fo_find(filt);
1306 	if (fops == NULL)
1307 		return EINVAL;
1308 
1309 	if (kev->flags & EV_ADD) {
1310 		/*
1311 		 * Prevent waiting with locks.  Non-sleepable
1312 		 * allocation failures are handled in the loop, only
1313 		 * if the spare knote appears to be actually required.
1314 		 */
1315 		tkn = knote_alloc(waitok);
1316 	} else {
1317 		tkn = NULL;
1318 	}
1319 
1320 findkn:
1321 	if (fops->f_isfd) {
1322 		KASSERT(td != NULL, ("td is NULL"));
1323 		if (kev->ident > INT_MAX)
1324 			error = EBADF;
1325 		else
1326 			error = fget(td, kev->ident,
1327 			    cap_rights_init(&rights, CAP_EVENT), &fp);
1328 		if (error)
1329 			goto done;
1330 
1331 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1332 		    kev->ident, 0) != 0) {
1333 			/* try again */
1334 			fdrop(fp, td);
1335 			fp = NULL;
1336 			error = kqueue_expand(kq, fops, kev->ident, waitok);
1337 			if (error)
1338 				goto done;
1339 			goto findkn;
1340 		}
1341 
1342 		if (fp->f_type == DTYPE_KQUEUE) {
1343 			/*
1344 			 * If we add some intelligence about what we are doing,
1345 			 * we should be able to support events on ourselves.
1346 			 * We need to know when we are doing this to prevent
1347 			 * getting both the knlist lock and the kq lock since
1348 			 * they are the same thing.
1349 			 */
1350 			if (fp->f_data == kq) {
1351 				error = EINVAL;
1352 				goto done;
1353 			}
1354 
1355 			/*
1356 			 * Pre-lock the filedesc before the global
1357 			 * lock mutex, see the comment in
1358 			 * kqueue_close().
1359 			 */
1360 			FILEDESC_XLOCK(td->td_proc->p_fd);
1361 			filedesc_unlock = 1;
1362 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1363 		}
1364 
1365 		KQ_LOCK(kq);
1366 		if (kev->ident < kq->kq_knlistsize) {
1367 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1368 				if (kev->filter == kn->kn_filter)
1369 					break;
1370 		}
1371 	} else {
1372 		if ((kev->flags & EV_ADD) == EV_ADD)
1373 			kqueue_expand(kq, fops, kev->ident, waitok);
1374 
1375 		KQ_LOCK(kq);
1376 
1377 		/*
1378 		 * If possible, find an existing knote to use for this kevent.
1379 		 */
1380 		if (kev->filter == EVFILT_PROC &&
1381 		    (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
1382 			/* This is an internal creation of a process tracking
1383 			 * note. Don't attempt to coalesce this with an
1384 			 * existing note.
1385 			 */
1386 			;
1387 		} else if (kq->kq_knhashmask != 0) {
1388 			struct klist *list;
1389 
1390 			list = &kq->kq_knhash[
1391 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1392 			SLIST_FOREACH(kn, list, kn_link)
1393 				if (kev->ident == kn->kn_id &&
1394 				    kev->filter == kn->kn_filter)
1395 					break;
1396 		}
1397 	}
1398 
1399 	/* knote is in the process of changing, wait for it to stabilize. */
1400 	if (kn != NULL && kn_in_flux(kn)) {
1401 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1402 		if (filedesc_unlock) {
1403 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1404 			filedesc_unlock = 0;
1405 		}
1406 		kq->kq_state |= KQ_FLUXWAIT;
1407 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1408 		if (fp != NULL) {
1409 			fdrop(fp, td);
1410 			fp = NULL;
1411 		}
1412 		goto findkn;
1413 	}
1414 
1415 	/*
1416 	 * kn now contains the matching knote, or NULL if no match
1417 	 */
1418 	if (kn == NULL) {
1419 		if (kev->flags & EV_ADD) {
1420 			kn = tkn;
1421 			tkn = NULL;
1422 			if (kn == NULL) {
1423 				KQ_UNLOCK(kq);
1424 				error = ENOMEM;
1425 				goto done;
1426 			}
1427 			kn->kn_fp = fp;
1428 			kn->kn_kq = kq;
1429 			kn->kn_fop = fops;
1430 			/*
1431 			 * apply reference counts to knote structure, and
1432 			 * do not release it at the end of this routine.
1433 			 */
1434 			fops = NULL;
1435 			fp = NULL;
1436 
1437 			kn->kn_sfflags = kev->fflags;
1438 			kn->kn_sdata = kev->data;
1439 			kev->fflags = 0;
1440 			kev->data = 0;
1441 			kn->kn_kevent = *kev;
1442 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1443 			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1444 			kn->kn_status = KN_DETACHED;
1445 			kn_enter_flux(kn);
1446 
1447 			error = knote_attach(kn, kq);
1448 			KQ_UNLOCK(kq);
1449 			if (error != 0) {
1450 				tkn = kn;
1451 				goto done;
1452 			}
1453 
1454 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1455 				knote_drop_detached(kn, td);
1456 				goto done;
1457 			}
1458 			knl = kn_list_lock(kn);
1459 			goto done_ev_add;
1460 		} else {
1461 			/* No matching knote and the EV_ADD flag is not set. */
1462 			KQ_UNLOCK(kq);
1463 			error = ENOENT;
1464 			goto done;
1465 		}
1466 	}
1467 
1468 	if (kev->flags & EV_DELETE) {
1469 		kn_enter_flux(kn);
1470 		KQ_UNLOCK(kq);
1471 		knote_drop(kn, td);
1472 		goto done;
1473 	}
1474 
1475 	if (kev->flags & EV_FORCEONESHOT) {
1476 		kn->kn_flags |= EV_ONESHOT;
1477 		KNOTE_ACTIVATE(kn, 1);
1478 	}
1479 
1480 	/*
1481 	 * The user may change some filter values after the initial EV_ADD,
1482 	 * but doing so will not reset any filter which has already been
1483 	 * triggered.
1484 	 */
1485 	kn->kn_status |= KN_SCAN;
1486 	kn_enter_flux(kn);
1487 	KQ_UNLOCK(kq);
1488 	knl = kn_list_lock(kn);
1489 	kn->kn_kevent.udata = kev->udata;
1490 	if (!fops->f_isfd && fops->f_touch != NULL) {
1491 		fops->f_touch(kn, kev, EVENT_REGISTER);
1492 	} else {
1493 		kn->kn_sfflags = kev->fflags;
1494 		kn->kn_sdata = kev->data;
1495 	}
1496 
1497 	/*
1498 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1499 	 * the initial attach event decides that the event is "completed"
1500 	 * already.  i.e. filt_procattach is called on a zombie process.  It
1501 	 * will call filt_proc which will remove it from the list, and NULL
1502 	 * kn_knlist.
1503 	 */
1504 done_ev_add:
1505 	if ((kev->flags & EV_ENABLE) != 0)
1506 		kn->kn_status &= ~KN_DISABLED;
1507 	else if ((kev->flags & EV_DISABLE) != 0)
1508 		kn->kn_status |= KN_DISABLED;
1509 
1510 	if ((kn->kn_status & KN_DISABLED) == 0)
1511 		event = kn->kn_fop->f_event(kn, 0);
1512 	else
1513 		event = 0;
1514 
1515 	KQ_LOCK(kq);
1516 	if (event)
1517 		kn->kn_status |= KN_ACTIVE;
1518 	if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
1519 	    KN_ACTIVE)
1520 		knote_enqueue(kn);
1521 	kn->kn_status &= ~KN_SCAN;
1522 	kn_leave_flux(kn);
1523 	kn_list_unlock(knl);
1524 	KQ_UNLOCK_FLUX(kq);
1525 
1526 done:
1527 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1528 	if (filedesc_unlock)
1529 		FILEDESC_XUNLOCK(td->td_proc->p_fd);
1530 	if (fp != NULL)
1531 		fdrop(fp, td);
1532 	knote_free(tkn);
1533 	if (fops != NULL)
1534 		kqueue_fo_release(filt);
1535 	return (error);
1536 }
1537 
1538 static int
1539 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1540 {
1541 	int error;
1542 	struct kqueue *kq;
1543 
1544 	error = 0;
1545 
1546 	kq = fp->f_data;
1547 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1548 		return (EBADF);
1549 	*kqp = kq;
1550 	KQ_LOCK(kq);
1551 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1552 		KQ_UNLOCK(kq);
1553 		return (EBADF);
1554 	}
1555 	kq->kq_refcnt++;
1556 	KQ_UNLOCK(kq);
1557 
1558 	return error;
1559 }
1560 
1561 static void
1562 kqueue_release(struct kqueue *kq, int locked)
1563 {
1564 	if (locked)
1565 		KQ_OWNED(kq);
1566 	else
1567 		KQ_LOCK(kq);
1568 	kq->kq_refcnt--;
1569 	if (kq->kq_refcnt == 1)
1570 		wakeup(&kq->kq_refcnt);
1571 	if (!locked)
1572 		KQ_UNLOCK(kq);
1573 }
1574 
1575 static void
1576 kqueue_schedtask(struct kqueue *kq)
1577 {
1578 
1579 	KQ_OWNED(kq);
1580 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1581 	    ("scheduling kqueue task while draining"));
1582 
1583 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1584 		taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
1585 		kq->kq_state |= KQ_TASKSCHED;
1586 	}
1587 }
1588 
1589 /*
1590  * Expand the kq to make sure we have storage for fops/ident pair.
1591  *
1592  * Return 0 on success (or no work necessary), return errno on failure.
1593  *
1594  * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1595  * If kqueue_register is called from a non-fd context, there usually/should
1596  * be no locks held.
1597  */
1598 static int
1599 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1600 	int waitok)
1601 {
1602 	struct klist *list, *tmp_knhash, *to_free;
1603 	u_long tmp_knhashmask;
1604 	int size;
1605 	int fd;
1606 	int mflag = waitok ? M_WAITOK : M_NOWAIT;
1607 
1608 	KQ_NOTOWNED(kq);
1609 
1610 	to_free = NULL;
1611 	if (fops->f_isfd) {
1612 		fd = ident;
1613 		if (kq->kq_knlistsize <= fd) {
1614 			size = kq->kq_knlistsize;
1615 			while (size <= fd)
1616 				size += KQEXTENT;
1617 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1618 			if (list == NULL)
1619 				return ENOMEM;
1620 			KQ_LOCK(kq);
1621 			if (kq->kq_knlistsize > fd) {
1622 				to_free = list;
1623 				list = NULL;
1624 			} else {
1625 				if (kq->kq_knlist != NULL) {
1626 					bcopy(kq->kq_knlist, list,
1627 					    kq->kq_knlistsize * sizeof(*list));
1628 					to_free = kq->kq_knlist;
1629 					kq->kq_knlist = NULL;
1630 				}
1631 				bzero((caddr_t)list +
1632 				    kq->kq_knlistsize * sizeof(*list),
1633 				    (size - kq->kq_knlistsize) * sizeof(*list));
1634 				kq->kq_knlistsize = size;
1635 				kq->kq_knlist = list;
1636 			}
1637 			KQ_UNLOCK(kq);
1638 		}
1639 	} else {
1640 		if (kq->kq_knhashmask == 0) {
1641 			tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1642 			    &tmp_knhashmask);
1643 			if (tmp_knhash == NULL)
1644 				return ENOMEM;
1645 			KQ_LOCK(kq);
1646 			if (kq->kq_knhashmask == 0) {
1647 				kq->kq_knhash = tmp_knhash;
1648 				kq->kq_knhashmask = tmp_knhashmask;
1649 			} else {
1650 				to_free = tmp_knhash;
1651 			}
1652 			KQ_UNLOCK(kq);
1653 		}
1654 	}
1655 	free(to_free, M_KQUEUE);
1656 
1657 	KQ_NOTOWNED(kq);
1658 	return 0;
1659 }
1660 
1661 static void
1662 kqueue_task(void *arg, int pending)
1663 {
1664 	struct kqueue *kq;
1665 	int haskqglobal;
1666 
1667 	haskqglobal = 0;
1668 	kq = arg;
1669 
1670 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1671 	KQ_LOCK(kq);
1672 
1673 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1674 
1675 	kq->kq_state &= ~KQ_TASKSCHED;
1676 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1677 		wakeup(&kq->kq_state);
1678 	}
1679 	KQ_UNLOCK(kq);
1680 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1681 }
1682 
1683 /*
1684  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1685  * We treat KN_MARKER knotes as if they are in flux.
1686  */
1687 static int
1688 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1689     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1690 {
1691 	struct kevent *kevp;
1692 	struct knote *kn, *marker;
1693 	struct knlist *knl;
1694 	sbintime_t asbt, rsbt;
1695 	int count, error, haskqglobal, influx, nkev, touch;
1696 
1697 	count = maxevents;
1698 	nkev = 0;
1699 	error = 0;
1700 	haskqglobal = 0;
1701 
1702 	if (maxevents == 0)
1703 		goto done_nl;
1704 
1705 	rsbt = 0;
1706 	if (tsp != NULL) {
1707 		if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 ||
1708 		    tsp->tv_nsec >= 1000000000) {
1709 			error = EINVAL;
1710 			goto done_nl;
1711 		}
1712 		if (timespecisset(tsp)) {
1713 			if (tsp->tv_sec <= INT32_MAX) {
1714 				rsbt = tstosbt(*tsp);
1715 				if (TIMESEL(&asbt, rsbt))
1716 					asbt += tc_tick_sbt;
1717 				if (asbt <= SBT_MAX - rsbt)
1718 					asbt += rsbt;
1719 				else
1720 					asbt = 0;
1721 				rsbt >>= tc_precexp;
1722 			} else
1723 				asbt = 0;
1724 		} else
1725 			asbt = -1;
1726 	} else
1727 		asbt = 0;
1728 	marker = knote_alloc(1);
1729 	marker->kn_status = KN_MARKER;
1730 	KQ_LOCK(kq);
1731 
1732 retry:
1733 	kevp = keva;
1734 	if (kq->kq_count == 0) {
1735 		if (asbt == -1) {
1736 			error = EWOULDBLOCK;
1737 		} else {
1738 			kq->kq_state |= KQ_SLEEP;
1739 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1740 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1741 		}
1742 		if (error == 0)
1743 			goto retry;
1744 		/* don't restart after signals... */
1745 		if (error == ERESTART)
1746 			error = EINTR;
1747 		else if (error == EWOULDBLOCK)
1748 			error = 0;
1749 		goto done;
1750 	}
1751 
1752 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1753 	influx = 0;
1754 	while (count) {
1755 		KQ_OWNED(kq);
1756 		kn = TAILQ_FIRST(&kq->kq_head);
1757 
1758 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1759 		    kn_in_flux(kn)) {
1760 			if (influx) {
1761 				influx = 0;
1762 				KQ_FLUX_WAKEUP(kq);
1763 			}
1764 			kq->kq_state |= KQ_FLUXWAIT;
1765 			error = msleep(kq, &kq->kq_lock, PSOCK,
1766 			    "kqflxwt", 0);
1767 			continue;
1768 		}
1769 
1770 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1771 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1772 			kn->kn_status &= ~KN_QUEUED;
1773 			kq->kq_count--;
1774 			continue;
1775 		}
1776 		if (kn == marker) {
1777 			KQ_FLUX_WAKEUP(kq);
1778 			if (count == maxevents)
1779 				goto retry;
1780 			goto done;
1781 		}
1782 		KASSERT(!kn_in_flux(kn),
1783 		    ("knote %p is unexpectedly in flux", kn));
1784 
1785 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1786 			kn->kn_status &= ~KN_QUEUED;
1787 			kn_enter_flux(kn);
1788 			kq->kq_count--;
1789 			KQ_UNLOCK(kq);
1790 			/*
1791 			 * We don't need to lock the list since we've
1792 			 * marked it as in flux.
1793 			 */
1794 			knote_drop(kn, td);
1795 			KQ_LOCK(kq);
1796 			continue;
1797 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1798 			kn->kn_status &= ~KN_QUEUED;
1799 			kn_enter_flux(kn);
1800 			kq->kq_count--;
1801 			KQ_UNLOCK(kq);
1802 			/*
1803 			 * We don't need to lock the list since we've
1804 			 * marked the knote as being in flux.
1805 			 */
1806 			*kevp = kn->kn_kevent;
1807 			knote_drop(kn, td);
1808 			KQ_LOCK(kq);
1809 			kn = NULL;
1810 		} else {
1811 			kn->kn_status |= KN_SCAN;
1812 			kn_enter_flux(kn);
1813 			KQ_UNLOCK(kq);
1814 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1815 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1816 			knl = kn_list_lock(kn);
1817 			if (kn->kn_fop->f_event(kn, 0) == 0) {
1818 				KQ_LOCK(kq);
1819 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1820 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
1821 				    KN_SCAN);
1822 				kn_leave_flux(kn);
1823 				kq->kq_count--;
1824 				kn_list_unlock(knl);
1825 				influx = 1;
1826 				continue;
1827 			}
1828 			touch = (!kn->kn_fop->f_isfd &&
1829 			    kn->kn_fop->f_touch != NULL);
1830 			if (touch)
1831 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1832 			else
1833 				*kevp = kn->kn_kevent;
1834 			KQ_LOCK(kq);
1835 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1836 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1837 				/*
1838 				 * Manually clear knotes who weren't
1839 				 * 'touch'ed.
1840 				 */
1841 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1842 					kn->kn_data = 0;
1843 					kn->kn_fflags = 0;
1844 				}
1845 				if (kn->kn_flags & EV_DISPATCH)
1846 					kn->kn_status |= KN_DISABLED;
1847 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1848 				kq->kq_count--;
1849 			} else
1850 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1851 
1852 			kn->kn_status &= ~KN_SCAN;
1853 			kn_leave_flux(kn);
1854 			kn_list_unlock(knl);
1855 			influx = 1;
1856 		}
1857 
1858 		/* we are returning a copy to the user */
1859 		kevp++;
1860 		nkev++;
1861 		count--;
1862 
1863 		if (nkev == KQ_NEVENTS) {
1864 			influx = 0;
1865 			KQ_UNLOCK_FLUX(kq);
1866 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1867 			nkev = 0;
1868 			kevp = keva;
1869 			KQ_LOCK(kq);
1870 			if (error)
1871 				break;
1872 		}
1873 	}
1874 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1875 done:
1876 	KQ_OWNED(kq);
1877 	KQ_UNLOCK_FLUX(kq);
1878 	knote_free(marker);
1879 done_nl:
1880 	KQ_NOTOWNED(kq);
1881 	if (nkev != 0)
1882 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1883 	td->td_retval[0] = maxevents - count;
1884 	return (error);
1885 }
1886 
1887 /*ARGSUSED*/
1888 static int
1889 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1890 	struct ucred *active_cred, struct thread *td)
1891 {
1892 	/*
1893 	 * Enabling sigio causes two major problems:
1894 	 * 1) infinite recursion:
1895 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
1896 	 * set.  On receipt of a signal this will cause a kqueue to recurse
1897 	 * into itself over and over.  Sending the sigio causes the kqueue
1898 	 * to become ready, which in turn posts sigio again, forever.
1899 	 * Solution: this can be solved by setting a flag in the kqueue that
1900 	 * we have a SIGIO in progress.
1901 	 * 2) locking problems:
1902 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1903 	 * us above the proc and pgrp locks.
1904 	 * Solution: Post a signal using an async mechanism, being sure to
1905 	 * record a generation count in the delivery so that we do not deliver
1906 	 * a signal to the wrong process.
1907 	 *
1908 	 * Note, these two mechanisms are somewhat mutually exclusive!
1909 	 */
1910 #if 0
1911 	struct kqueue *kq;
1912 
1913 	kq = fp->f_data;
1914 	switch (cmd) {
1915 	case FIOASYNC:
1916 		if (*(int *)data) {
1917 			kq->kq_state |= KQ_ASYNC;
1918 		} else {
1919 			kq->kq_state &= ~KQ_ASYNC;
1920 		}
1921 		return (0);
1922 
1923 	case FIOSETOWN:
1924 		return (fsetown(*(int *)data, &kq->kq_sigio));
1925 
1926 	case FIOGETOWN:
1927 		*(int *)data = fgetown(&kq->kq_sigio);
1928 		return (0);
1929 	}
1930 #endif
1931 
1932 	return (ENOTTY);
1933 }
1934 
1935 /*ARGSUSED*/
1936 static int
1937 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1938 	struct thread *td)
1939 {
1940 	struct kqueue *kq;
1941 	int revents = 0;
1942 	int error;
1943 
1944 	if ((error = kqueue_acquire(fp, &kq)))
1945 		return POLLERR;
1946 
1947 	KQ_LOCK(kq);
1948 	if (events & (POLLIN | POLLRDNORM)) {
1949 		if (kq->kq_count) {
1950 			revents |= events & (POLLIN | POLLRDNORM);
1951 		} else {
1952 			selrecord(td, &kq->kq_sel);
1953 			if (SEL_WAITING(&kq->kq_sel))
1954 				kq->kq_state |= KQ_SEL;
1955 		}
1956 	}
1957 	kqueue_release(kq, 1);
1958 	KQ_UNLOCK(kq);
1959 	return (revents);
1960 }
1961 
1962 /*ARGSUSED*/
1963 static int
1964 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1965 	struct thread *td)
1966 {
1967 
1968 	bzero((void *)st, sizeof *st);
1969 	/*
1970 	 * We no longer return kq_count because the unlocked value is useless.
1971 	 * If you spent all this time getting the count, why not spend your
1972 	 * syscall better by calling kevent?
1973 	 *
1974 	 * XXX - This is needed for libc_r.
1975 	 */
1976 	st->st_mode = S_IFIFO;
1977 	return (0);
1978 }
1979 
1980 static void
1981 kqueue_drain(struct kqueue *kq, struct thread *td)
1982 {
1983 	struct knote *kn;
1984 	int i;
1985 
1986 	KQ_LOCK(kq);
1987 
1988 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1989 	    ("kqueue already closing"));
1990 	kq->kq_state |= KQ_CLOSING;
1991 	if (kq->kq_refcnt > 1)
1992 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1993 
1994 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1995 
1996 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
1997 	    ("kqueue's knlist not empty"));
1998 
1999 	for (i = 0; i < kq->kq_knlistsize; i++) {
2000 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
2001 			if (kn_in_flux(kn)) {
2002 				kq->kq_state |= KQ_FLUXWAIT;
2003 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
2004 				continue;
2005 			}
2006 			kn_enter_flux(kn);
2007 			KQ_UNLOCK(kq);
2008 			knote_drop(kn, td);
2009 			KQ_LOCK(kq);
2010 		}
2011 	}
2012 	if (kq->kq_knhashmask != 0) {
2013 		for (i = 0; i <= kq->kq_knhashmask; i++) {
2014 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
2015 				if (kn_in_flux(kn)) {
2016 					kq->kq_state |= KQ_FLUXWAIT;
2017 					msleep(kq, &kq->kq_lock, PSOCK,
2018 					       "kqclo2", 0);
2019 					continue;
2020 				}
2021 				kn_enter_flux(kn);
2022 				KQ_UNLOCK(kq);
2023 				knote_drop(kn, td);
2024 				KQ_LOCK(kq);
2025 			}
2026 		}
2027 	}
2028 
2029 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
2030 		kq->kq_state |= KQ_TASKDRAIN;
2031 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
2032 	}
2033 
2034 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2035 		selwakeuppri(&kq->kq_sel, PSOCK);
2036 		if (!SEL_WAITING(&kq->kq_sel))
2037 			kq->kq_state &= ~KQ_SEL;
2038 	}
2039 
2040 	KQ_UNLOCK(kq);
2041 }
2042 
2043 static void
2044 kqueue_destroy(struct kqueue *kq)
2045 {
2046 
2047 	KASSERT(kq->kq_fdp == NULL,
2048 	    ("kqueue still attached to a file descriptor"));
2049 	seldrain(&kq->kq_sel);
2050 	knlist_destroy(&kq->kq_sel.si_note);
2051 	mtx_destroy(&kq->kq_lock);
2052 
2053 	if (kq->kq_knhash != NULL)
2054 		free(kq->kq_knhash, M_KQUEUE);
2055 	if (kq->kq_knlist != NULL)
2056 		free(kq->kq_knlist, M_KQUEUE);
2057 
2058 	funsetown(&kq->kq_sigio);
2059 }
2060 
2061 /*ARGSUSED*/
2062 static int
2063 kqueue_close(struct file *fp, struct thread *td)
2064 {
2065 	struct kqueue *kq = fp->f_data;
2066 	struct filedesc *fdp;
2067 	int error;
2068 	int filedesc_unlock;
2069 
2070 	if ((error = kqueue_acquire(fp, &kq)))
2071 		return error;
2072 	kqueue_drain(kq, td);
2073 
2074 	/*
2075 	 * We could be called due to the knote_drop() doing fdrop(),
2076 	 * called from kqueue_register().  In this case the global
2077 	 * lock is owned, and filedesc sx is locked before, to not
2078 	 * take the sleepable lock after non-sleepable.
2079 	 */
2080 	fdp = kq->kq_fdp;
2081 	kq->kq_fdp = NULL;
2082 	if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
2083 		FILEDESC_XLOCK(fdp);
2084 		filedesc_unlock = 1;
2085 	} else
2086 		filedesc_unlock = 0;
2087 	TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
2088 	if (filedesc_unlock)
2089 		FILEDESC_XUNLOCK(fdp);
2090 
2091 	kqueue_destroy(kq);
2092 	chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
2093 	crfree(kq->kq_cred);
2094 	free(kq, M_KQUEUE);
2095 	fp->f_data = NULL;
2096 
2097 	return (0);
2098 }
2099 
2100 static int
2101 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2102 {
2103 
2104 	kif->kf_type = KF_TYPE_KQUEUE;
2105 	return (0);
2106 }
2107 
2108 static void
2109 kqueue_wakeup(struct kqueue *kq)
2110 {
2111 	KQ_OWNED(kq);
2112 
2113 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
2114 		kq->kq_state &= ~KQ_SLEEP;
2115 		wakeup(kq);
2116 	}
2117 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2118 		selwakeuppri(&kq->kq_sel, PSOCK);
2119 		if (!SEL_WAITING(&kq->kq_sel))
2120 			kq->kq_state &= ~KQ_SEL;
2121 	}
2122 	if (!knlist_empty(&kq->kq_sel.si_note))
2123 		kqueue_schedtask(kq);
2124 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
2125 		pgsigio(&kq->kq_sigio, SIGIO, 0);
2126 	}
2127 }
2128 
2129 /*
2130  * Walk down a list of knotes, activating them if their event has triggered.
2131  *
2132  * There is a possibility to optimize in the case of one kq watching another.
2133  * Instead of scheduling a task to wake it up, you could pass enough state
2134  * down the chain to make up the parent kqueue.  Make this code functional
2135  * first.
2136  */
2137 void
2138 knote(struct knlist *list, long hint, int lockflags)
2139 {
2140 	struct kqueue *kq;
2141 	struct knote *kn, *tkn;
2142 	int error;
2143 
2144 	if (list == NULL)
2145 		return;
2146 
2147 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
2148 
2149 	if ((lockflags & KNF_LISTLOCKED) == 0)
2150 		list->kl_lock(list->kl_lockarg);
2151 
2152 	/*
2153 	 * If we unlock the list lock (and enter influx), we can
2154 	 * eliminate the kqueue scheduling, but this will introduce
2155 	 * four lock/unlock's for each knote to test.  Also, marker
2156 	 * would be needed to keep iteration position, since filters
2157 	 * or other threads could remove events.
2158 	 */
2159 	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
2160 		kq = kn->kn_kq;
2161 		KQ_LOCK(kq);
2162 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
2163 			/*
2164 			 * Do not process the influx notes, except for
2165 			 * the influx coming from the kq unlock in the
2166 			 * kqueue_scan().  In the later case, we do
2167 			 * not interfere with the scan, since the code
2168 			 * fragment in kqueue_scan() locks the knlist,
2169 			 * and cannot proceed until we finished.
2170 			 */
2171 			KQ_UNLOCK(kq);
2172 		} else if ((lockflags & KNF_NOKQLOCK) != 0) {
2173 			kn_enter_flux(kn);
2174 			KQ_UNLOCK(kq);
2175 			error = kn->kn_fop->f_event(kn, hint);
2176 			KQ_LOCK(kq);
2177 			kn_leave_flux(kn);
2178 			if (error)
2179 				KNOTE_ACTIVATE(kn, 1);
2180 			KQ_UNLOCK_FLUX(kq);
2181 		} else {
2182 			kn->kn_status |= KN_HASKQLOCK;
2183 			if (kn->kn_fop->f_event(kn, hint))
2184 				KNOTE_ACTIVATE(kn, 1);
2185 			kn->kn_status &= ~KN_HASKQLOCK;
2186 			KQ_UNLOCK(kq);
2187 		}
2188 	}
2189 	if ((lockflags & KNF_LISTLOCKED) == 0)
2190 		list->kl_unlock(list->kl_lockarg);
2191 }
2192 
2193 /*
2194  * add a knote to a knlist
2195  */
2196 void
2197 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
2198 {
2199 
2200 	KNL_ASSERT_LOCK(knl, islocked);
2201 	KQ_NOTOWNED(kn->kn_kq);
2202 	KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn));
2203 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2204 	    ("knote %p was not detached", kn));
2205 	if (!islocked)
2206 		knl->kl_lock(knl->kl_lockarg);
2207 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
2208 	if (!islocked)
2209 		knl->kl_unlock(knl->kl_lockarg);
2210 	KQ_LOCK(kn->kn_kq);
2211 	kn->kn_knlist = knl;
2212 	kn->kn_status &= ~KN_DETACHED;
2213 	KQ_UNLOCK(kn->kn_kq);
2214 }
2215 
2216 static void
2217 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
2218     int kqislocked)
2219 {
2220 
2221 	KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked"));
2222 	KNL_ASSERT_LOCK(knl, knlislocked);
2223 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
2224 	KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn));
2225 	KASSERT((kn->kn_status & KN_DETACHED) == 0,
2226 	    ("knote %p was already detached", kn));
2227 	if (!knlislocked)
2228 		knl->kl_lock(knl->kl_lockarg);
2229 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
2230 	kn->kn_knlist = NULL;
2231 	if (!knlislocked)
2232 		kn_list_unlock(knl);
2233 	if (!kqislocked)
2234 		KQ_LOCK(kn->kn_kq);
2235 	kn->kn_status |= KN_DETACHED;
2236 	if (!kqislocked)
2237 		KQ_UNLOCK(kn->kn_kq);
2238 }
2239 
2240 /*
2241  * remove knote from the specified knlist
2242  */
2243 void
2244 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
2245 {
2246 
2247 	knlist_remove_kq(knl, kn, islocked, 0);
2248 }
2249 
2250 int
2251 knlist_empty(struct knlist *knl)
2252 {
2253 
2254 	KNL_ASSERT_LOCKED(knl);
2255 	return (SLIST_EMPTY(&knl->kl_list));
2256 }
2257 
2258 static struct mtx knlist_lock;
2259 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2260     MTX_DEF);
2261 static void knlist_mtx_lock(void *arg);
2262 static void knlist_mtx_unlock(void *arg);
2263 
2264 static void
2265 knlist_mtx_lock(void *arg)
2266 {
2267 
2268 	mtx_lock((struct mtx *)arg);
2269 }
2270 
2271 static void
2272 knlist_mtx_unlock(void *arg)
2273 {
2274 
2275 	mtx_unlock((struct mtx *)arg);
2276 }
2277 
2278 static void
2279 knlist_mtx_assert_locked(void *arg)
2280 {
2281 
2282 	mtx_assert((struct mtx *)arg, MA_OWNED);
2283 }
2284 
2285 static void
2286 knlist_mtx_assert_unlocked(void *arg)
2287 {
2288 
2289 	mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2290 }
2291 
2292 static void
2293 knlist_rw_rlock(void *arg)
2294 {
2295 
2296 	rw_rlock((struct rwlock *)arg);
2297 }
2298 
2299 static void
2300 knlist_rw_runlock(void *arg)
2301 {
2302 
2303 	rw_runlock((struct rwlock *)arg);
2304 }
2305 
2306 static void
2307 knlist_rw_assert_locked(void *arg)
2308 {
2309 
2310 	rw_assert((struct rwlock *)arg, RA_LOCKED);
2311 }
2312 
2313 static void
2314 knlist_rw_assert_unlocked(void *arg)
2315 {
2316 
2317 	rw_assert((struct rwlock *)arg, RA_UNLOCKED);
2318 }
2319 
2320 void
2321 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2322     void (*kl_unlock)(void *),
2323     void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
2324 {
2325 
2326 	if (lock == NULL)
2327 		knl->kl_lockarg = &knlist_lock;
2328 	else
2329 		knl->kl_lockarg = lock;
2330 
2331 	if (kl_lock == NULL)
2332 		knl->kl_lock = knlist_mtx_lock;
2333 	else
2334 		knl->kl_lock = kl_lock;
2335 	if (kl_unlock == NULL)
2336 		knl->kl_unlock = knlist_mtx_unlock;
2337 	else
2338 		knl->kl_unlock = kl_unlock;
2339 	if (kl_assert_locked == NULL)
2340 		knl->kl_assert_locked = knlist_mtx_assert_locked;
2341 	else
2342 		knl->kl_assert_locked = kl_assert_locked;
2343 	if (kl_assert_unlocked == NULL)
2344 		knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
2345 	else
2346 		knl->kl_assert_unlocked = kl_assert_unlocked;
2347 
2348 	knl->kl_autodestroy = 0;
2349 	SLIST_INIT(&knl->kl_list);
2350 }
2351 
2352 void
2353 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2354 {
2355 
2356 	knlist_init(knl, lock, NULL, NULL, NULL, NULL);
2357 }
2358 
2359 struct knlist *
2360 knlist_alloc(struct mtx *lock)
2361 {
2362 	struct knlist *knl;
2363 
2364 	knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
2365 	knlist_init_mtx(knl, lock);
2366 	return (knl);
2367 }
2368 
2369 void
2370 knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock)
2371 {
2372 
2373 	knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock,
2374 	    knlist_rw_assert_locked, knlist_rw_assert_unlocked);
2375 }
2376 
2377 void
2378 knlist_destroy(struct knlist *knl)
2379 {
2380 
2381 	KASSERT(KNLIST_EMPTY(knl),
2382 	    ("destroying knlist %p with knotes on it", knl));
2383 }
2384 
2385 void
2386 knlist_detach(struct knlist *knl)
2387 {
2388 
2389 	KNL_ASSERT_LOCKED(knl);
2390 	knl->kl_autodestroy = 1;
2391 	if (knlist_empty(knl)) {
2392 		knlist_destroy(knl);
2393 		free(knl, M_KQUEUE);
2394 	}
2395 }
2396 
2397 /*
2398  * Even if we are locked, we may need to drop the lock to allow any influx
2399  * knotes time to "settle".
2400  */
2401 void
2402 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2403 {
2404 	struct knote *kn, *kn2;
2405 	struct kqueue *kq;
2406 
2407 	KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
2408 	if (islocked)
2409 		KNL_ASSERT_LOCKED(knl);
2410 	else {
2411 		KNL_ASSERT_UNLOCKED(knl);
2412 again:		/* need to reacquire lock since we have dropped it */
2413 		knl->kl_lock(knl->kl_lockarg);
2414 	}
2415 
2416 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2417 		kq = kn->kn_kq;
2418 		KQ_LOCK(kq);
2419 		if (kn_in_flux(kn)) {
2420 			KQ_UNLOCK(kq);
2421 			continue;
2422 		}
2423 		knlist_remove_kq(knl, kn, 1, 1);
2424 		if (killkn) {
2425 			kn_enter_flux(kn);
2426 			KQ_UNLOCK(kq);
2427 			knote_drop_detached(kn, td);
2428 		} else {
2429 			/* Make sure cleared knotes disappear soon */
2430 			kn->kn_flags |= EV_EOF | EV_ONESHOT;
2431 			KQ_UNLOCK(kq);
2432 		}
2433 		kq = NULL;
2434 	}
2435 
2436 	if (!SLIST_EMPTY(&knl->kl_list)) {
2437 		/* there are still in flux knotes remaining */
2438 		kn = SLIST_FIRST(&knl->kl_list);
2439 		kq = kn->kn_kq;
2440 		KQ_LOCK(kq);
2441 		KASSERT(kn_in_flux(kn), ("knote removed w/o list lock"));
2442 		knl->kl_unlock(knl->kl_lockarg);
2443 		kq->kq_state |= KQ_FLUXWAIT;
2444 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2445 		kq = NULL;
2446 		goto again;
2447 	}
2448 
2449 	if (islocked)
2450 		KNL_ASSERT_LOCKED(knl);
2451 	else {
2452 		knl->kl_unlock(knl->kl_lockarg);
2453 		KNL_ASSERT_UNLOCKED(knl);
2454 	}
2455 }
2456 
2457 /*
2458  * Remove all knotes referencing a specified fd must be called with FILEDESC
2459  * lock.  This prevents a race where a new fd comes along and occupies the
2460  * entry and we attach a knote to the fd.
2461  */
2462 void
2463 knote_fdclose(struct thread *td, int fd)
2464 {
2465 	struct filedesc *fdp = td->td_proc->p_fd;
2466 	struct kqueue *kq;
2467 	struct knote *kn;
2468 	int influx;
2469 
2470 	FILEDESC_XLOCK_ASSERT(fdp);
2471 
2472 	/*
2473 	 * We shouldn't have to worry about new kevents appearing on fd
2474 	 * since filedesc is locked.
2475 	 */
2476 	TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2477 		KQ_LOCK(kq);
2478 
2479 again:
2480 		influx = 0;
2481 		while (kq->kq_knlistsize > fd &&
2482 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2483 			if (kn_in_flux(kn)) {
2484 				/* someone else might be waiting on our knote */
2485 				if (influx)
2486 					wakeup(kq);
2487 				kq->kq_state |= KQ_FLUXWAIT;
2488 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2489 				goto again;
2490 			}
2491 			kn_enter_flux(kn);
2492 			KQ_UNLOCK(kq);
2493 			influx = 1;
2494 			knote_drop(kn, td);
2495 			KQ_LOCK(kq);
2496 		}
2497 		KQ_UNLOCK_FLUX(kq);
2498 	}
2499 }
2500 
2501 static int
2502 knote_attach(struct knote *kn, struct kqueue *kq)
2503 {
2504 	struct klist *list;
2505 
2506 	KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn));
2507 	KQ_OWNED(kq);
2508 
2509 	if (kn->kn_fop->f_isfd) {
2510 		if (kn->kn_id >= kq->kq_knlistsize)
2511 			return (ENOMEM);
2512 		list = &kq->kq_knlist[kn->kn_id];
2513 	} else {
2514 		if (kq->kq_knhash == NULL)
2515 			return (ENOMEM);
2516 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2517 	}
2518 	SLIST_INSERT_HEAD(list, kn, kn_link);
2519 	return (0);
2520 }
2521 
2522 static void
2523 knote_drop(struct knote *kn, struct thread *td)
2524 {
2525 
2526 	if ((kn->kn_status & KN_DETACHED) == 0)
2527 		kn->kn_fop->f_detach(kn);
2528 	knote_drop_detached(kn, td);
2529 }
2530 
2531 static void
2532 knote_drop_detached(struct knote *kn, struct thread *td)
2533 {
2534 	struct kqueue *kq;
2535 	struct klist *list;
2536 
2537 	kq = kn->kn_kq;
2538 
2539 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2540 	    ("knote %p still attached", kn));
2541 	KQ_NOTOWNED(kq);
2542 
2543 	KQ_LOCK(kq);
2544 	KASSERT(kn->kn_influx == 1,
2545 	    ("knote_drop called on %p with influx %d", kn, kn->kn_influx));
2546 
2547 	if (kn->kn_fop->f_isfd)
2548 		list = &kq->kq_knlist[kn->kn_id];
2549 	else
2550 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2551 
2552 	if (!SLIST_EMPTY(list))
2553 		SLIST_REMOVE(list, kn, knote, kn_link);
2554 	if (kn->kn_status & KN_QUEUED)
2555 		knote_dequeue(kn);
2556 	KQ_UNLOCK_FLUX(kq);
2557 
2558 	if (kn->kn_fop->f_isfd) {
2559 		fdrop(kn->kn_fp, td);
2560 		kn->kn_fp = NULL;
2561 	}
2562 	kqueue_fo_release(kn->kn_kevent.filter);
2563 	kn->kn_fop = NULL;
2564 	knote_free(kn);
2565 }
2566 
2567 static void
2568 knote_enqueue(struct knote *kn)
2569 {
2570 	struct kqueue *kq = kn->kn_kq;
2571 
2572 	KQ_OWNED(kn->kn_kq);
2573 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2574 
2575 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2576 	kn->kn_status |= KN_QUEUED;
2577 	kq->kq_count++;
2578 	kqueue_wakeup(kq);
2579 }
2580 
2581 static void
2582 knote_dequeue(struct knote *kn)
2583 {
2584 	struct kqueue *kq = kn->kn_kq;
2585 
2586 	KQ_OWNED(kn->kn_kq);
2587 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2588 
2589 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2590 	kn->kn_status &= ~KN_QUEUED;
2591 	kq->kq_count--;
2592 }
2593 
2594 static void
2595 knote_init(void)
2596 {
2597 
2598 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2599 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2600 }
2601 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2602 
2603 static struct knote *
2604 knote_alloc(int waitok)
2605 {
2606 
2607 	return (uma_zalloc(knote_zone, (waitok ? M_WAITOK : M_NOWAIT) |
2608 	    M_ZERO));
2609 }
2610 
2611 static void
2612 knote_free(struct knote *kn)
2613 {
2614 
2615 	uma_zfree(knote_zone, kn);
2616 }
2617 
2618 /*
2619  * Register the kev w/ the kq specified by fd.
2620  */
2621 int
2622 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2623 {
2624 	struct kqueue *kq;
2625 	struct file *fp;
2626 	cap_rights_t rights;
2627 	int error;
2628 
2629 	error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp);
2630 	if (error != 0)
2631 		return (error);
2632 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2633 		goto noacquire;
2634 
2635 	error = kqueue_register(kq, kev, td, waitok);
2636 	kqueue_release(kq, 0);
2637 
2638 noacquire:
2639 	fdrop(fp, td);
2640 	return (error);
2641 }
2642