xref: /freebsd/sys/kern/kern_event.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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_ktrace.h"
35 #include "opt_kqueue.h"
36 
37 #ifdef COMPAT_FREEBSD11
38 #define	_WANT_FREEBSD11_KEVENT
39 #endif
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/capsicum.h>
44 #include <sys/kernel.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/proc.h>
49 #include <sys/malloc.h>
50 #include <sys/unistd.h>
51 #include <sys/file.h>
52 #include <sys/filedesc.h>
53 #include <sys/filio.h>
54 #include <sys/fcntl.h>
55 #include <sys/kthread.h>
56 #include <sys/selinfo.h>
57 #include <sys/queue.h>
58 #include <sys/event.h>
59 #include <sys/eventvar.h>
60 #include <sys/poll.h>
61 #include <sys/protosw.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sigio.h>
64 #include <sys/signalvar.h>
65 #include <sys/socket.h>
66 #include <sys/socketvar.h>
67 #include <sys/stat.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysproto.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/taskqueue.h>
72 #include <sys/uio.h>
73 #include <sys/user.h>
74 #ifdef KTRACE
75 #include <sys/ktrace.h>
76 #endif
77 #include <machine/atomic.h>
78 
79 #include <vm/uma.h>
80 
81 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
82 
83 /*
84  * This lock is used if multiple kq locks are required.  This possibly
85  * should be made into a per proc lock.
86  */
87 static struct mtx	kq_global;
88 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
89 #define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
90 	if (!haslck)				\
91 		mtx_lock(lck);			\
92 	haslck = 1;				\
93 } while (0)
94 #define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
95 	if (haslck)				\
96 		mtx_unlock(lck);			\
97 	haslck = 0;				\
98 } while (0)
99 
100 TASKQUEUE_DEFINE_THREAD(kqueue_ctx);
101 
102 static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
103 static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
104 static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
105 		    struct thread *td, int mflag);
106 static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
107 static void	kqueue_release(struct kqueue *kq, int locked);
108 static void	kqueue_destroy(struct kqueue *kq);
109 static void	kqueue_drain(struct kqueue *kq, struct thread *td);
110 static int	kqueue_expand(struct kqueue *kq, const struct filterops *fops,
111 		    uintptr_t ident, int mflag);
112 static void	kqueue_task(void *arg, int pending);
113 static int	kqueue_scan(struct kqueue *kq, int maxevents,
114 		    struct kevent_copyops *k_ops,
115 		    const struct timespec *timeout,
116 		    struct kevent *keva, struct thread *td);
117 static void 	kqueue_wakeup(struct kqueue *kq);
118 static const struct filterops *kqueue_fo_find(int filt);
119 static void	kqueue_fo_release(int filt);
120 struct g_kevent_args;
121 static int	kern_kevent_generic(struct thread *td,
122 		    struct g_kevent_args *uap,
123 		    struct kevent_copyops *k_ops, const char *struct_name);
124 
125 static fo_ioctl_t	kqueue_ioctl;
126 static fo_poll_t	kqueue_poll;
127 static fo_kqfilter_t	kqueue_kqfilter;
128 static fo_stat_t	kqueue_stat;
129 static fo_close_t	kqueue_close;
130 static fo_fill_kinfo_t	kqueue_fill_kinfo;
131 
132 static struct fileops kqueueops = {
133 	.fo_read = invfo_rdwr,
134 	.fo_write = invfo_rdwr,
135 	.fo_truncate = invfo_truncate,
136 	.fo_ioctl = kqueue_ioctl,
137 	.fo_poll = kqueue_poll,
138 	.fo_kqfilter = kqueue_kqfilter,
139 	.fo_stat = kqueue_stat,
140 	.fo_close = kqueue_close,
141 	.fo_chmod = invfo_chmod,
142 	.fo_chown = invfo_chown,
143 	.fo_sendfile = invfo_sendfile,
144 	.fo_fill_kinfo = kqueue_fill_kinfo,
145 };
146 
147 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
148 static void 	knote_drop(struct knote *kn, struct thread *td);
149 static void 	knote_drop_detached(struct knote *kn, struct thread *td);
150 static void 	knote_enqueue(struct knote *kn);
151 static void 	knote_dequeue(struct knote *kn);
152 static void 	knote_init(void);
153 static struct 	knote *knote_alloc(int mflag);
154 static void 	knote_free(struct knote *kn);
155 
156 static void	filt_kqdetach(struct knote *kn);
157 static int	filt_kqueue(struct knote *kn, long hint);
158 static int	filt_procattach(struct knote *kn);
159 static void	filt_procdetach(struct knote *kn);
160 static int	filt_proc(struct knote *kn, long hint);
161 static int	filt_fileattach(struct knote *kn);
162 static void	filt_timerexpire(void *knx);
163 static void	filt_timerexpire_l(struct knote *kn, bool proc_locked);
164 static int	filt_timerattach(struct knote *kn);
165 static void	filt_timerdetach(struct knote *kn);
166 static void	filt_timerstart(struct knote *kn, sbintime_t to);
167 static void	filt_timertouch(struct knote *kn, struct kevent *kev,
168 		    u_long type);
169 static int	filt_timervalidate(struct knote *kn, sbintime_t *to);
170 static int	filt_timer(struct knote *kn, long hint);
171 static int	filt_userattach(struct knote *kn);
172 static void	filt_userdetach(struct knote *kn);
173 static int	filt_user(struct knote *kn, long hint);
174 static void	filt_usertouch(struct knote *kn, struct kevent *kev,
175 		    u_long type);
176 
177 static struct filterops file_filtops = {
178 	.f_isfd = 1,
179 	.f_attach = filt_fileattach,
180 };
181 static struct filterops kqread_filtops = {
182 	.f_isfd = 1,
183 	.f_detach = filt_kqdetach,
184 	.f_event = filt_kqueue,
185 };
186 /* XXX - move to kern_proc.c?  */
187 static struct filterops proc_filtops = {
188 	.f_isfd = 0,
189 	.f_attach = filt_procattach,
190 	.f_detach = filt_procdetach,
191 	.f_event = filt_proc,
192 };
193 static struct filterops timer_filtops = {
194 	.f_isfd = 0,
195 	.f_attach = filt_timerattach,
196 	.f_detach = filt_timerdetach,
197 	.f_event = filt_timer,
198 	.f_touch = filt_timertouch,
199 };
200 static struct filterops user_filtops = {
201 	.f_attach = filt_userattach,
202 	.f_detach = filt_userdetach,
203 	.f_event = filt_user,
204 	.f_touch = filt_usertouch,
205 };
206 
207 static uma_zone_t	knote_zone;
208 static unsigned int __exclusive_cache_line	kq_ncallouts;
209 static unsigned int 	kq_calloutmax = 4 * 1024;
210 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
211     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
212 
213 /* XXX - ensure not influx ? */
214 #define KNOTE_ACTIVATE(kn, islock) do { 				\
215 	if ((islock))							\
216 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
217 	else								\
218 		KQ_LOCK((kn)->kn_kq);					\
219 	(kn)->kn_status |= KN_ACTIVE;					\
220 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
221 		knote_enqueue((kn));					\
222 	if (!(islock))							\
223 		KQ_UNLOCK((kn)->kn_kq);					\
224 } while (0)
225 #define KQ_LOCK(kq) do {						\
226 	mtx_lock(&(kq)->kq_lock);					\
227 } while (0)
228 #define KQ_FLUX_WAKEUP(kq) do {						\
229 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
230 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
231 		wakeup((kq));						\
232 	}								\
233 } while (0)
234 #define KQ_UNLOCK_FLUX(kq) do {						\
235 	KQ_FLUX_WAKEUP(kq);						\
236 	mtx_unlock(&(kq)->kq_lock);					\
237 } while (0)
238 #define KQ_UNLOCK(kq) do {						\
239 	mtx_unlock(&(kq)->kq_lock);					\
240 } while (0)
241 #define KQ_OWNED(kq) do {						\
242 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
243 } while (0)
244 #define KQ_NOTOWNED(kq) do {						\
245 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
246 } while (0)
247 
248 static struct knlist *
249 kn_list_lock(struct knote *kn)
250 {
251 	struct knlist *knl;
252 
253 	knl = kn->kn_knlist;
254 	if (knl != NULL)
255 		knl->kl_lock(knl->kl_lockarg);
256 	return (knl);
257 }
258 
259 static void
260 kn_list_unlock(struct knlist *knl)
261 {
262 	bool do_free;
263 
264 	if (knl == NULL)
265 		return;
266 	do_free = knl->kl_autodestroy && knlist_empty(knl);
267 	knl->kl_unlock(knl->kl_lockarg);
268 	if (do_free) {
269 		knlist_destroy(knl);
270 		free(knl, M_KQUEUE);
271 	}
272 }
273 
274 static bool
275 kn_in_flux(struct knote *kn)
276 {
277 
278 	return (kn->kn_influx > 0);
279 }
280 
281 static void
282 kn_enter_flux(struct knote *kn)
283 {
284 
285 	KQ_OWNED(kn->kn_kq);
286 	MPASS(kn->kn_influx < INT_MAX);
287 	kn->kn_influx++;
288 }
289 
290 static bool
291 kn_leave_flux(struct knote *kn)
292 {
293 
294 	KQ_OWNED(kn->kn_kq);
295 	MPASS(kn->kn_influx > 0);
296 	kn->kn_influx--;
297 	return (kn->kn_influx == 0);
298 }
299 
300 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
301 	if (islocked)							\
302 		KNL_ASSERT_LOCKED(knl);				\
303 	else								\
304 		KNL_ASSERT_UNLOCKED(knl);				\
305 } while (0)
306 #ifdef INVARIANTS
307 #define	KNL_ASSERT_LOCKED(knl) do {					\
308 	knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED);		\
309 } while (0)
310 #define	KNL_ASSERT_UNLOCKED(knl) do {					\
311 	knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED);		\
312 } while (0)
313 #else /* !INVARIANTS */
314 #define	KNL_ASSERT_LOCKED(knl) do {} while (0)
315 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
316 #endif /* INVARIANTS */
317 
318 #ifndef	KN_HASHSIZE
319 #define	KN_HASHSIZE		64		/* XXX should be tunable */
320 #endif
321 
322 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
323 
324 static int
325 filt_nullattach(struct knote *kn)
326 {
327 
328 	return (ENXIO);
329 };
330 
331 struct filterops null_filtops = {
332 	.f_isfd = 0,
333 	.f_attach = filt_nullattach,
334 };
335 
336 /* XXX - make SYSINIT to add these, and move into respective modules. */
337 extern struct filterops sig_filtops;
338 extern struct filterops fs_filtops;
339 
340 /*
341  * Table for all system-defined filters.
342  */
343 static struct mtx	filterops_lock;
344 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF);
345 static struct {
346 	const struct filterops *for_fop;
347 	int for_nolock;
348 	int for_refcnt;
349 } sysfilt_ops[EVFILT_SYSCOUNT] = {
350 	{ &file_filtops, 1 },			/* EVFILT_READ */
351 	{ &file_filtops, 1 },			/* EVFILT_WRITE */
352 	{ &null_filtops },			/* EVFILT_AIO */
353 	{ &file_filtops, 1 },			/* EVFILT_VNODE */
354 	{ &proc_filtops, 1 },			/* EVFILT_PROC */
355 	{ &sig_filtops, 1 },			/* EVFILT_SIGNAL */
356 	{ &timer_filtops, 1 },			/* EVFILT_TIMER */
357 	{ &file_filtops, 1 },			/* EVFILT_PROCDESC */
358 	{ &fs_filtops, 1 },			/* EVFILT_FS */
359 	{ &null_filtops },			/* EVFILT_LIO */
360 	{ &user_filtops, 1 },			/* EVFILT_USER */
361 	{ &null_filtops },			/* EVFILT_SENDFILE */
362 	{ &file_filtops, 1 },                   /* EVFILT_EMPTY */
363 };
364 
365 /*
366  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
367  * method.
368  */
369 static int
370 filt_fileattach(struct knote *kn)
371 {
372 
373 	return (fo_kqfilter(kn->kn_fp, kn));
374 }
375 
376 /*ARGSUSED*/
377 static int
378 kqueue_kqfilter(struct file *fp, struct knote *kn)
379 {
380 	struct kqueue *kq = kn->kn_fp->f_data;
381 
382 	if (kn->kn_filter != EVFILT_READ)
383 		return (EINVAL);
384 
385 	kn->kn_status |= KN_KQUEUE;
386 	kn->kn_fop = &kqread_filtops;
387 	knlist_add(&kq->kq_sel.si_note, kn, 0);
388 
389 	return (0);
390 }
391 
392 static void
393 filt_kqdetach(struct knote *kn)
394 {
395 	struct kqueue *kq = kn->kn_fp->f_data;
396 
397 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
398 }
399 
400 /*ARGSUSED*/
401 static int
402 filt_kqueue(struct knote *kn, long hint)
403 {
404 	struct kqueue *kq = kn->kn_fp->f_data;
405 
406 	kn->kn_data = kq->kq_count;
407 	return (kn->kn_data > 0);
408 }
409 
410 /* XXX - move to kern_proc.c?  */
411 static int
412 filt_procattach(struct knote *kn)
413 {
414 	struct proc *p;
415 	int error;
416 	bool exiting, immediate;
417 
418 	exiting = immediate = false;
419 	if (kn->kn_sfflags & NOTE_EXIT)
420 		p = pfind_any(kn->kn_id);
421 	else
422 		p = pfind(kn->kn_id);
423 	if (p == NULL)
424 		return (ESRCH);
425 	if (p->p_flag & P_WEXIT)
426 		exiting = true;
427 
428 	if ((error = p_cansee(curthread, p))) {
429 		PROC_UNLOCK(p);
430 		return (error);
431 	}
432 
433 	kn->kn_ptr.p_proc = p;
434 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
435 
436 	/*
437 	 * Internal flag indicating registration done by kernel for the
438 	 * purposes of getting a NOTE_CHILD notification.
439 	 */
440 	if (kn->kn_flags & EV_FLAG2) {
441 		kn->kn_flags &= ~EV_FLAG2;
442 		kn->kn_data = kn->kn_sdata;		/* ppid */
443 		kn->kn_fflags = NOTE_CHILD;
444 		kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
445 		immediate = true; /* Force immediate activation of child note. */
446 	}
447 	/*
448 	 * Internal flag indicating registration done by kernel (for other than
449 	 * NOTE_CHILD).
450 	 */
451 	if (kn->kn_flags & EV_FLAG1) {
452 		kn->kn_flags &= ~EV_FLAG1;
453 	}
454 
455 	knlist_add(p->p_klist, kn, 1);
456 
457 	/*
458 	 * Immediately activate any child notes or, in the case of a zombie
459 	 * target process, exit notes.  The latter is necessary to handle the
460 	 * case where the target process, e.g. a child, dies before the kevent
461 	 * is registered.
462 	 */
463 	if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
464 		KNOTE_ACTIVATE(kn, 0);
465 
466 	PROC_UNLOCK(p);
467 
468 	return (0);
469 }
470 
471 /*
472  * The knote may be attached to a different process, which may exit,
473  * leaving nothing for the knote to be attached to.  So when the process
474  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
475  * it will be deleted when read out.  However, as part of the knote deletion,
476  * this routine is called, so a check is needed to avoid actually performing
477  * a detach, because the original process does not exist any more.
478  */
479 /* XXX - move to kern_proc.c?  */
480 static void
481 filt_procdetach(struct knote *kn)
482 {
483 
484 	knlist_remove(kn->kn_knlist, kn, 0);
485 	kn->kn_ptr.p_proc = NULL;
486 }
487 
488 /* XXX - move to kern_proc.c?  */
489 static int
490 filt_proc(struct knote *kn, long hint)
491 {
492 	struct proc *p;
493 	u_int event;
494 
495 	p = kn->kn_ptr.p_proc;
496 	if (p == NULL) /* already activated, from attach filter */
497 		return (0);
498 
499 	/* Mask off extra data. */
500 	event = (u_int)hint & NOTE_PCTRLMASK;
501 
502 	/* If the user is interested in this event, record it. */
503 	if (kn->kn_sfflags & event)
504 		kn->kn_fflags |= event;
505 
506 	/* Process is gone, so flag the event as finished. */
507 	if (event == NOTE_EXIT) {
508 		kn->kn_flags |= EV_EOF | EV_ONESHOT;
509 		kn->kn_ptr.p_proc = NULL;
510 		if (kn->kn_fflags & NOTE_EXIT)
511 			kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
512 		if (kn->kn_fflags == 0)
513 			kn->kn_flags |= EV_DROP;
514 		return (1);
515 	}
516 
517 	return (kn->kn_fflags != 0);
518 }
519 
520 /*
521  * Called when the process forked. It mostly does the same as the
522  * knote(), activating all knotes registered to be activated when the
523  * process forked. Additionally, for each knote attached to the
524  * parent, check whether user wants to track the new process. If so
525  * attach a new knote to it, and immediately report an event with the
526  * child's pid.
527  */
528 void
529 knote_fork(struct knlist *list, int pid)
530 {
531 	struct kqueue *kq;
532 	struct knote *kn;
533 	struct kevent kev;
534 	int error;
535 
536 	MPASS(list != NULL);
537 	KNL_ASSERT_LOCKED(list);
538 	if (SLIST_EMPTY(&list->kl_list))
539 		return;
540 
541 	memset(&kev, 0, sizeof(kev));
542 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
543 		kq = kn->kn_kq;
544 		KQ_LOCK(kq);
545 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
546 			KQ_UNLOCK(kq);
547 			continue;
548 		}
549 
550 		/*
551 		 * The same as knote(), activate the event.
552 		 */
553 		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
554 			if (kn->kn_fop->f_event(kn, NOTE_FORK))
555 				KNOTE_ACTIVATE(kn, 1);
556 			KQ_UNLOCK(kq);
557 			continue;
558 		}
559 
560 		/*
561 		 * The NOTE_TRACK case. In addition to the activation
562 		 * of the event, we need to register new events to
563 		 * track the child. Drop the locks in preparation for
564 		 * the call to kqueue_register().
565 		 */
566 		kn_enter_flux(kn);
567 		KQ_UNLOCK(kq);
568 		list->kl_unlock(list->kl_lockarg);
569 
570 		/*
571 		 * Activate existing knote and register tracking knotes with
572 		 * new process.
573 		 *
574 		 * First register a knote to get just the child notice. This
575 		 * must be a separate note from a potential NOTE_EXIT
576 		 * notification since both NOTE_CHILD and NOTE_EXIT are defined
577 		 * to use the data field (in conflicting ways).
578 		 */
579 		kev.ident = pid;
580 		kev.filter = kn->kn_filter;
581 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
582 		    EV_FLAG2;
583 		kev.fflags = kn->kn_sfflags;
584 		kev.data = kn->kn_id;		/* parent */
585 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
586 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
587 		if (error)
588 			kn->kn_fflags |= NOTE_TRACKERR;
589 
590 		/*
591 		 * Then register another knote to track other potential events
592 		 * from the new process.
593 		 */
594 		kev.ident = pid;
595 		kev.filter = kn->kn_filter;
596 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
597 		kev.fflags = kn->kn_sfflags;
598 		kev.data = kn->kn_id;		/* parent */
599 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
600 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
601 		if (error)
602 			kn->kn_fflags |= NOTE_TRACKERR;
603 		if (kn->kn_fop->f_event(kn, NOTE_FORK))
604 			KNOTE_ACTIVATE(kn, 0);
605 		list->kl_lock(list->kl_lockarg);
606 		KQ_LOCK(kq);
607 		kn_leave_flux(kn);
608 		KQ_UNLOCK_FLUX(kq);
609 	}
610 }
611 
612 /*
613  * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
614  * interval timer support code.
615  */
616 
617 #define NOTE_TIMER_PRECMASK						\
618     (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)
619 
620 static sbintime_t
621 timer2sbintime(int64_t data, int flags)
622 {
623 	int64_t secs;
624 
625         /*
626          * Macros for converting to the fractional second portion of an
627          * sbintime_t using 64bit multiplication to improve precision.
628          */
629 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
630 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
631 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
632 	switch (flags & NOTE_TIMER_PRECMASK) {
633 	case NOTE_SECONDS:
634 #ifdef __LP64__
635 		if (data > (SBT_MAX / SBT_1S))
636 			return (SBT_MAX);
637 #endif
638 		return ((sbintime_t)data << 32);
639 	case NOTE_MSECONDS: /* FALLTHROUGH */
640 	case 0:
641 		if (data >= 1000) {
642 			secs = data / 1000;
643 #ifdef __LP64__
644 			if (secs > (SBT_MAX / SBT_1S))
645 				return (SBT_MAX);
646 #endif
647 			return (secs << 32 | MS_TO_SBT(data % 1000));
648 		}
649 		return (MS_TO_SBT(data));
650 	case NOTE_USECONDS:
651 		if (data >= 1000000) {
652 			secs = data / 1000000;
653 #ifdef __LP64__
654 			if (secs > (SBT_MAX / SBT_1S))
655 				return (SBT_MAX);
656 #endif
657 			return (secs << 32 | US_TO_SBT(data % 1000000));
658 		}
659 		return (US_TO_SBT(data));
660 	case NOTE_NSECONDS:
661 		if (data >= 1000000000) {
662 			secs = data / 1000000000;
663 #ifdef __LP64__
664 			if (secs > (SBT_MAX / SBT_1S))
665 				return (SBT_MAX);
666 #endif
667 			return (secs << 32 | NS_TO_SBT(data % 1000000000));
668 		}
669 		return (NS_TO_SBT(data));
670 	default:
671 		break;
672 	}
673 	return (-1);
674 }
675 
676 struct kq_timer_cb_data {
677 	struct callout c;
678 	struct proc *p;
679 	struct knote *kn;
680 	int cpuid;
681 	int flags;
682 	TAILQ_ENTRY(kq_timer_cb_data) link;
683 	sbintime_t next;	/* next timer event fires at */
684 	sbintime_t to;		/* precalculated timer period, 0 for abs */
685 };
686 
687 #define	KQ_TIMER_CB_ENQUEUED	0x01
688 
689 static void
690 kqtimer_sched_callout(struct kq_timer_cb_data *kc)
691 {
692 	callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn,
693 	    kc->cpuid, C_ABSOLUTE);
694 }
695 
696 void
697 kqtimer_proc_continue(struct proc *p)
698 {
699 	struct kq_timer_cb_data *kc, *kc1;
700 	struct bintime bt;
701 	sbintime_t now;
702 
703 	PROC_LOCK_ASSERT(p, MA_OWNED);
704 
705 	getboottimebin(&bt);
706 	now = bttosbt(bt);
707 
708 	TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) {
709 		TAILQ_REMOVE(&p->p_kqtim_stop, kc, link);
710 		kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
711 		if (kc->next <= now)
712 			filt_timerexpire_l(kc->kn, true);
713 		else
714 			kqtimer_sched_callout(kc);
715 	}
716 }
717 
718 static void
719 filt_timerexpire_l(struct knote *kn, bool proc_locked)
720 {
721 	struct kq_timer_cb_data *kc;
722 	struct proc *p;
723 	uint64_t delta;
724 	sbintime_t now;
725 
726 	kc = kn->kn_ptr.p_v;
727 
728 	if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) {
729 		kn->kn_data++;
730 		KNOTE_ACTIVATE(kn, 0);
731 		return;
732 	}
733 
734 	now = sbinuptime();
735 	if (now >= kc->next) {
736 		delta = (now - kc->next) / kc->to;
737 		if (delta == 0)
738 			delta = 1;
739 		kn->kn_data += delta;
740 		kc->next += delta * kc->to;
741 		if (now >= kc->next)	/* overflow */
742 			kc->next = now + kc->to;
743 		KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
744 	}
745 
746 	/*
747 	 * Initial check for stopped kc->p is racy.  It is fine to
748 	 * miss the set of the stop flags, at worst we would schedule
749 	 * one more callout.  On the other hand, it is not fine to not
750 	 * schedule when we we missed clearing of the flags, we
751 	 * recheck them under the lock and observe consistent state.
752 	 */
753 	p = kc->p;
754 	if (P_SHOULDSTOP(p) || P_KILLED(p)) {
755 		if (!proc_locked)
756 			PROC_LOCK(p);
757 		if (P_SHOULDSTOP(p) || P_KILLED(p)) {
758 			if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) {
759 				kc->flags |= KQ_TIMER_CB_ENQUEUED;
760 				TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link);
761 			}
762 			if (!proc_locked)
763 				PROC_UNLOCK(p);
764 			return;
765 		}
766 		if (!proc_locked)
767 			PROC_UNLOCK(p);
768 	}
769 	kqtimer_sched_callout(kc);
770 }
771 
772 static void
773 filt_timerexpire(void *knx)
774 {
775 	filt_timerexpire_l(knx, false);
776 }
777 
778 /*
779  * data contains amount of time to sleep
780  */
781 static int
782 filt_timervalidate(struct knote *kn, sbintime_t *to)
783 {
784 	struct bintime bt;
785 	sbintime_t sbt;
786 
787 	if (kn->kn_sdata < 0)
788 		return (EINVAL);
789 	if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
790 		kn->kn_sdata = 1;
791 	/*
792 	 * The only fflags values supported are the timer unit
793 	 * (precision) and the absolute time indicator.
794 	 */
795 	if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0)
796 		return (EINVAL);
797 
798 	*to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
799 	if (*to < 0)
800 		return (EINVAL);
801 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
802 		getboottimebin(&bt);
803 		sbt = bttosbt(bt);
804 		*to = MAX(0, *to - sbt);
805 	}
806 	return (0);
807 }
808 
809 static int
810 filt_timerattach(struct knote *kn)
811 {
812 	struct kq_timer_cb_data *kc;
813 	sbintime_t to;
814 	int error;
815 
816 	to = -1;
817 	error = filt_timervalidate(kn, &to);
818 	if (error != 0)
819 		return (error);
820 	KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 ||
821 	    (kn->kn_sfflags & NOTE_ABSTIME) != 0,
822 	    ("%s: periodic timer has a calculated zero timeout", __func__));
823 	KASSERT(to >= 0,
824 	    ("%s: timer has a calculated negative timeout", __func__));
825 
826 	if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) {
827 		atomic_subtract_int(&kq_ncallouts, 1);
828 		return (ENOMEM);
829 	}
830 
831 	if ((kn->kn_sfflags & NOTE_ABSTIME) == 0)
832 		kn->kn_flags |= EV_CLEAR;	/* automatically set */
833 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
834 	kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
835 	kc->kn = kn;
836 	kc->p = curproc;
837 	kc->cpuid = PCPU_GET(cpuid);
838 	kc->flags = 0;
839 	callout_init(&kc->c, 1);
840 	filt_timerstart(kn, to);
841 
842 	return (0);
843 }
844 
845 static void
846 filt_timerstart(struct knote *kn, sbintime_t to)
847 {
848 	struct kq_timer_cb_data *kc;
849 
850 	kc = kn->kn_ptr.p_v;
851 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
852 		kc->next = to;
853 		kc->to = 0;
854 	} else {
855 		kc->next = to + sbinuptime();
856 		kc->to = to;
857 	}
858 	kqtimer_sched_callout(kc);
859 }
860 
861 static void
862 filt_timerdetach(struct knote *kn)
863 {
864 	struct kq_timer_cb_data *kc;
865 	unsigned int old __unused;
866 	bool pending;
867 
868 	kc = kn->kn_ptr.p_v;
869 	do {
870 		callout_drain(&kc->c);
871 
872 		/*
873 		 * kqtimer_proc_continue() might have rescheduled this callout.
874 		 * Double-check, using the process mutex as an interlock.
875 		 */
876 		PROC_LOCK(kc->p);
877 		if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) {
878 			kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
879 			TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link);
880 		}
881 		pending = callout_pending(&kc->c);
882 		PROC_UNLOCK(kc->p);
883 	} while (pending);
884 	free(kc, M_KQUEUE);
885 	old = atomic_fetchadd_int(&kq_ncallouts, -1);
886 	KASSERT(old > 0, ("Number of callouts cannot become negative"));
887 	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
888 }
889 
890 static void
891 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type)
892 {
893 	struct kq_timer_cb_data *kc;
894 	struct kqueue *kq;
895 	sbintime_t to;
896 	int error;
897 
898 	switch (type) {
899 	case EVENT_REGISTER:
900 		/* Handle re-added timers that update data/fflags */
901 		if (kev->flags & EV_ADD) {
902 			kc = kn->kn_ptr.p_v;
903 
904 			/* Drain any existing callout. */
905 			callout_drain(&kc->c);
906 
907 			/* Throw away any existing undelivered record
908 			 * of the timer expiration. This is done under
909 			 * the presumption that if a process is
910 			 * re-adding this timer with new parameters,
911 			 * it is no longer interested in what may have
912 			 * happened under the old parameters. If it is
913 			 * interested, it can wait for the expiration,
914 			 * delete the old timer definition, and then
915 			 * add the new one.
916 			 *
917 			 * This has to be done while the kq is locked:
918 			 *   - if enqueued, dequeue
919 			 *   - make it no longer active
920 			 *   - clear the count of expiration events
921 			 */
922 			kq = kn->kn_kq;
923 			KQ_LOCK(kq);
924 			if (kn->kn_status & KN_QUEUED)
925 				knote_dequeue(kn);
926 
927 			kn->kn_status &= ~KN_ACTIVE;
928 			kn->kn_data = 0;
929 			KQ_UNLOCK(kq);
930 
931 			/* Reschedule timer based on new data/fflags */
932 			kn->kn_sfflags = kev->fflags;
933 			kn->kn_sdata = kev->data;
934 			error = filt_timervalidate(kn, &to);
935 			if (error != 0) {
936 			  	kn->kn_flags |= EV_ERROR;
937 				kn->kn_data = error;
938 			} else
939 			  	filt_timerstart(kn, to);
940 		}
941 		break;
942 
943         case EVENT_PROCESS:
944 		*kev = kn->kn_kevent;
945 		if (kn->kn_flags & EV_CLEAR) {
946 			kn->kn_data = 0;
947 			kn->kn_fflags = 0;
948 		}
949 		break;
950 
951 	default:
952 		panic("filt_timertouch() - invalid type (%ld)", type);
953 		break;
954 	}
955 }
956 
957 static int
958 filt_timer(struct knote *kn, long hint)
959 {
960 
961 	return (kn->kn_data != 0);
962 }
963 
964 static int
965 filt_userattach(struct knote *kn)
966 {
967 
968 	/*
969 	 * EVFILT_USER knotes are not attached to anything in the kernel.
970 	 */
971 	kn->kn_hook = NULL;
972 	if (kn->kn_fflags & NOTE_TRIGGER)
973 		kn->kn_hookid = 1;
974 	else
975 		kn->kn_hookid = 0;
976 	return (0);
977 }
978 
979 static void
980 filt_userdetach(__unused struct knote *kn)
981 {
982 
983 	/*
984 	 * EVFILT_USER knotes are not attached to anything in the kernel.
985 	 */
986 }
987 
988 static int
989 filt_user(struct knote *kn, __unused long hint)
990 {
991 
992 	return (kn->kn_hookid);
993 }
994 
995 static void
996 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
997 {
998 	u_int ffctrl;
999 
1000 	switch (type) {
1001 	case EVENT_REGISTER:
1002 		if (kev->fflags & NOTE_TRIGGER)
1003 			kn->kn_hookid = 1;
1004 
1005 		ffctrl = kev->fflags & NOTE_FFCTRLMASK;
1006 		kev->fflags &= NOTE_FFLAGSMASK;
1007 		switch (ffctrl) {
1008 		case NOTE_FFNOP:
1009 			break;
1010 
1011 		case NOTE_FFAND:
1012 			kn->kn_sfflags &= kev->fflags;
1013 			break;
1014 
1015 		case NOTE_FFOR:
1016 			kn->kn_sfflags |= kev->fflags;
1017 			break;
1018 
1019 		case NOTE_FFCOPY:
1020 			kn->kn_sfflags = kev->fflags;
1021 			break;
1022 
1023 		default:
1024 			/* XXX Return error? */
1025 			break;
1026 		}
1027 		kn->kn_sdata = kev->data;
1028 		if (kev->flags & EV_CLEAR) {
1029 			kn->kn_hookid = 0;
1030 			kn->kn_data = 0;
1031 			kn->kn_fflags = 0;
1032 		}
1033 		break;
1034 
1035         case EVENT_PROCESS:
1036 		*kev = kn->kn_kevent;
1037 		kev->fflags = kn->kn_sfflags;
1038 		kev->data = kn->kn_sdata;
1039 		if (kn->kn_flags & EV_CLEAR) {
1040 			kn->kn_hookid = 0;
1041 			kn->kn_data = 0;
1042 			kn->kn_fflags = 0;
1043 		}
1044 		break;
1045 
1046 	default:
1047 		panic("filt_usertouch() - invalid type (%ld)", type);
1048 		break;
1049 	}
1050 }
1051 
1052 int
1053 sys_kqueue(struct thread *td, struct kqueue_args *uap)
1054 {
1055 
1056 	return (kern_kqueue(td, 0, NULL));
1057 }
1058 
1059 int
1060 sys_kqueuex(struct thread *td, struct kqueuex_args *uap)
1061 {
1062 	int flags;
1063 
1064 	if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0)
1065 		return (EINVAL);
1066 	flags = 0;
1067 	if ((uap->flags & KQUEUE_CLOEXEC) != 0)
1068 		flags |= O_CLOEXEC;
1069 	return (kern_kqueue(td, flags, NULL));
1070 }
1071 
1072 static void
1073 kqueue_init(struct kqueue *kq)
1074 {
1075 
1076 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
1077 	TAILQ_INIT(&kq->kq_head);
1078 	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
1079 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
1080 }
1081 
1082 int
1083 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps)
1084 {
1085 	struct filedesc *fdp;
1086 	struct kqueue *kq;
1087 	struct file *fp;
1088 	struct ucred *cred;
1089 	int fd, error;
1090 
1091 	fdp = td->td_proc->p_fd;
1092 	cred = td->td_ucred;
1093 	if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
1094 		return (ENOMEM);
1095 
1096 	error = falloc_caps(td, &fp, &fd, flags, fcaps);
1097 	if (error != 0) {
1098 		chgkqcnt(cred->cr_ruidinfo, -1, 0);
1099 		return (error);
1100 	}
1101 
1102 	/* An extra reference on `fp' has been held for us by falloc(). */
1103 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
1104 	kqueue_init(kq);
1105 	kq->kq_fdp = fdp;
1106 	kq->kq_cred = crhold(cred);
1107 
1108 	FILEDESC_XLOCK(fdp);
1109 	TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
1110 	FILEDESC_XUNLOCK(fdp);
1111 
1112 	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
1113 	fdrop(fp, td);
1114 
1115 	td->td_retval[0] = fd;
1116 	return (0);
1117 }
1118 
1119 struct g_kevent_args {
1120 	int	fd;
1121 	const void *changelist;
1122 	int	nchanges;
1123 	void	*eventlist;
1124 	int	nevents;
1125 	const struct timespec *timeout;
1126 };
1127 
1128 int
1129 sys_kevent(struct thread *td, struct kevent_args *uap)
1130 {
1131 	struct kevent_copyops k_ops = {
1132 		.arg = uap,
1133 		.k_copyout = kevent_copyout,
1134 		.k_copyin = kevent_copyin,
1135 		.kevent_size = sizeof(struct kevent),
1136 	};
1137 	struct g_kevent_args gk_args = {
1138 		.fd = uap->fd,
1139 		.changelist = uap->changelist,
1140 		.nchanges = uap->nchanges,
1141 		.eventlist = uap->eventlist,
1142 		.nevents = uap->nevents,
1143 		.timeout = uap->timeout,
1144 	};
1145 
1146 	return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent"));
1147 }
1148 
1149 static int
1150 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap,
1151     struct kevent_copyops *k_ops, const char *struct_name)
1152 {
1153 	struct timespec ts, *tsp;
1154 #ifdef KTRACE
1155 	struct kevent *eventlist = uap->eventlist;
1156 #endif
1157 	int error;
1158 
1159 	if (uap->timeout != NULL) {
1160 		error = copyin(uap->timeout, &ts, sizeof(ts));
1161 		if (error)
1162 			return (error);
1163 		tsp = &ts;
1164 	} else
1165 		tsp = NULL;
1166 
1167 #ifdef KTRACE
1168 	if (KTRPOINT(td, KTR_STRUCT_ARRAY))
1169 		ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist,
1170 		    uap->nchanges, k_ops->kevent_size);
1171 #endif
1172 
1173 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
1174 	    k_ops, tsp);
1175 
1176 #ifdef KTRACE
1177 	if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
1178 		ktrstructarray(struct_name, UIO_USERSPACE, eventlist,
1179 		    td->td_retval[0], k_ops->kevent_size);
1180 #endif
1181 
1182 	return (error);
1183 }
1184 
1185 /*
1186  * Copy 'count' items into the destination list pointed to by uap->eventlist.
1187  */
1188 static int
1189 kevent_copyout(void *arg, struct kevent *kevp, int count)
1190 {
1191 	struct kevent_args *uap;
1192 	int error;
1193 
1194 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1195 	uap = (struct kevent_args *)arg;
1196 
1197 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
1198 	if (error == 0)
1199 		uap->eventlist += count;
1200 	return (error);
1201 }
1202 
1203 /*
1204  * Copy 'count' items from the list pointed to by uap->changelist.
1205  */
1206 static int
1207 kevent_copyin(void *arg, struct kevent *kevp, int count)
1208 {
1209 	struct kevent_args *uap;
1210 	int error;
1211 
1212 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1213 	uap = (struct kevent_args *)arg;
1214 
1215 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
1216 	if (error == 0)
1217 		uap->changelist += count;
1218 	return (error);
1219 }
1220 
1221 #ifdef COMPAT_FREEBSD11
1222 static int
1223 kevent11_copyout(void *arg, struct kevent *kevp, int count)
1224 {
1225 	struct freebsd11_kevent_args *uap;
1226 	struct freebsd11_kevent kev11;
1227 	int error, i;
1228 
1229 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1230 	uap = (struct freebsd11_kevent_args *)arg;
1231 
1232 	for (i = 0; i < count; i++) {
1233 		kev11.ident = kevp->ident;
1234 		kev11.filter = kevp->filter;
1235 		kev11.flags = kevp->flags;
1236 		kev11.fflags = kevp->fflags;
1237 		kev11.data = kevp->data;
1238 		kev11.udata = kevp->udata;
1239 		error = copyout(&kev11, uap->eventlist, sizeof(kev11));
1240 		if (error != 0)
1241 			break;
1242 		uap->eventlist++;
1243 		kevp++;
1244 	}
1245 	return (error);
1246 }
1247 
1248 /*
1249  * Copy 'count' items from the list pointed to by uap->changelist.
1250  */
1251 static int
1252 kevent11_copyin(void *arg, struct kevent *kevp, int count)
1253 {
1254 	struct freebsd11_kevent_args *uap;
1255 	struct freebsd11_kevent kev11;
1256 	int error, i;
1257 
1258 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1259 	uap = (struct freebsd11_kevent_args *)arg;
1260 
1261 	for (i = 0; i < count; i++) {
1262 		error = copyin(uap->changelist, &kev11, sizeof(kev11));
1263 		if (error != 0)
1264 			break;
1265 		kevp->ident = kev11.ident;
1266 		kevp->filter = kev11.filter;
1267 		kevp->flags = kev11.flags;
1268 		kevp->fflags = kev11.fflags;
1269 		kevp->data = (uintptr_t)kev11.data;
1270 		kevp->udata = kev11.udata;
1271 		bzero(&kevp->ext, sizeof(kevp->ext));
1272 		uap->changelist++;
1273 		kevp++;
1274 	}
1275 	return (error);
1276 }
1277 
1278 int
1279 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap)
1280 {
1281 	struct kevent_copyops k_ops = {
1282 		.arg = uap,
1283 		.k_copyout = kevent11_copyout,
1284 		.k_copyin = kevent11_copyin,
1285 		.kevent_size = sizeof(struct freebsd11_kevent),
1286 	};
1287 	struct g_kevent_args gk_args = {
1288 		.fd = uap->fd,
1289 		.changelist = uap->changelist,
1290 		.nchanges = uap->nchanges,
1291 		.eventlist = uap->eventlist,
1292 		.nevents = uap->nevents,
1293 		.timeout = uap->timeout,
1294 	};
1295 
1296 	return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent"));
1297 }
1298 #endif
1299 
1300 int
1301 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
1302     struct kevent_copyops *k_ops, const struct timespec *timeout)
1303 {
1304 	cap_rights_t rights;
1305 	struct file *fp;
1306 	int error;
1307 
1308 	cap_rights_init_zero(&rights);
1309 	if (nchanges > 0)
1310 		cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE);
1311 	if (nevents > 0)
1312 		cap_rights_set_one(&rights, CAP_KQUEUE_EVENT);
1313 	error = fget(td, fd, &rights, &fp);
1314 	if (error != 0)
1315 		return (error);
1316 
1317 	error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
1318 	fdrop(fp, td);
1319 
1320 	return (error);
1321 }
1322 
1323 static int
1324 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
1325     struct kevent_copyops *k_ops, const struct timespec *timeout)
1326 {
1327 	struct kevent keva[KQ_NEVENTS];
1328 	struct kevent *kevp, *changes;
1329 	int i, n, nerrors, error;
1330 
1331 	if (nchanges < 0)
1332 		return (EINVAL);
1333 
1334 	nerrors = 0;
1335 	while (nchanges > 0) {
1336 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
1337 		error = k_ops->k_copyin(k_ops->arg, keva, n);
1338 		if (error)
1339 			return (error);
1340 		changes = keva;
1341 		for (i = 0; i < n; i++) {
1342 			kevp = &changes[i];
1343 			if (!kevp->filter)
1344 				continue;
1345 			kevp->flags &= ~EV_SYSFLAGS;
1346 			error = kqueue_register(kq, kevp, td, M_WAITOK);
1347 			if (error || (kevp->flags & EV_RECEIPT)) {
1348 				if (nevents == 0)
1349 					return (error);
1350 				kevp->flags = EV_ERROR;
1351 				kevp->data = error;
1352 				(void)k_ops->k_copyout(k_ops->arg, kevp, 1);
1353 				nevents--;
1354 				nerrors++;
1355 			}
1356 		}
1357 		nchanges -= n;
1358 	}
1359 	if (nerrors) {
1360 		td->td_retval[0] = nerrors;
1361 		return (0);
1362 	}
1363 
1364 	return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
1365 }
1366 
1367 int
1368 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
1369     struct kevent_copyops *k_ops, const struct timespec *timeout)
1370 {
1371 	struct kqueue *kq;
1372 	int error;
1373 
1374 	error = kqueue_acquire(fp, &kq);
1375 	if (error != 0)
1376 		return (error);
1377 	error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
1378 	kqueue_release(kq, 0);
1379 	return (error);
1380 }
1381 
1382 /*
1383  * Performs a kevent() call on a temporarily created kqueue. This can be
1384  * used to perform one-shot polling, similar to poll() and select().
1385  */
1386 int
1387 kern_kevent_anonymous(struct thread *td, int nevents,
1388     struct kevent_copyops *k_ops)
1389 {
1390 	struct kqueue kq = {};
1391 	int error;
1392 
1393 	kqueue_init(&kq);
1394 	kq.kq_refcnt = 1;
1395 	error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
1396 	kqueue_drain(&kq, td);
1397 	kqueue_destroy(&kq);
1398 	return (error);
1399 }
1400 
1401 int
1402 kqueue_add_filteropts(int filt, const struct filterops *filtops)
1403 {
1404 	int error;
1405 
1406 	error = 0;
1407 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
1408 		printf(
1409 "trying to add a filterop that is out of range: %d is beyond %d\n",
1410 		    ~filt, EVFILT_SYSCOUNT);
1411 		return EINVAL;
1412 	}
1413 	mtx_lock(&filterops_lock);
1414 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
1415 	    sysfilt_ops[~filt].for_fop != NULL)
1416 		error = EEXIST;
1417 	else {
1418 		sysfilt_ops[~filt].for_fop = filtops;
1419 		sysfilt_ops[~filt].for_refcnt = 0;
1420 	}
1421 	mtx_unlock(&filterops_lock);
1422 
1423 	return (error);
1424 }
1425 
1426 int
1427 kqueue_del_filteropts(int filt)
1428 {
1429 	int error;
1430 
1431 	error = 0;
1432 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1433 		return EINVAL;
1434 
1435 	mtx_lock(&filterops_lock);
1436 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1437 	    sysfilt_ops[~filt].for_fop == NULL)
1438 		error = EINVAL;
1439 	else if (sysfilt_ops[~filt].for_refcnt != 0)
1440 		error = EBUSY;
1441 	else {
1442 		sysfilt_ops[~filt].for_fop = &null_filtops;
1443 		sysfilt_ops[~filt].for_refcnt = 0;
1444 	}
1445 	mtx_unlock(&filterops_lock);
1446 
1447 	return error;
1448 }
1449 
1450 static const struct filterops *
1451 kqueue_fo_find(int filt)
1452 {
1453 
1454 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1455 		return NULL;
1456 
1457 	if (sysfilt_ops[~filt].for_nolock)
1458 		return sysfilt_ops[~filt].for_fop;
1459 
1460 	mtx_lock(&filterops_lock);
1461 	sysfilt_ops[~filt].for_refcnt++;
1462 	if (sysfilt_ops[~filt].for_fop == NULL)
1463 		sysfilt_ops[~filt].for_fop = &null_filtops;
1464 	mtx_unlock(&filterops_lock);
1465 
1466 	return sysfilt_ops[~filt].for_fop;
1467 }
1468 
1469 static void
1470 kqueue_fo_release(int filt)
1471 {
1472 
1473 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1474 		return;
1475 
1476 	if (sysfilt_ops[~filt].for_nolock)
1477 		return;
1478 
1479 	mtx_lock(&filterops_lock);
1480 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1481 	    ("filter object refcount not valid on release"));
1482 	sysfilt_ops[~filt].for_refcnt--;
1483 	mtx_unlock(&filterops_lock);
1484 }
1485 
1486 /*
1487  * A ref to kq (obtained via kqueue_acquire) must be held.
1488  */
1489 static int
1490 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td,
1491     int mflag)
1492 {
1493 	const struct filterops *fops;
1494 	struct file *fp;
1495 	struct knote *kn, *tkn;
1496 	struct knlist *knl;
1497 	int error, filt, event;
1498 	int haskqglobal, filedesc_unlock;
1499 
1500 	if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
1501 		return (EINVAL);
1502 
1503 	fp = NULL;
1504 	kn = NULL;
1505 	knl = NULL;
1506 	error = 0;
1507 	haskqglobal = 0;
1508 	filedesc_unlock = 0;
1509 
1510 	filt = kev->filter;
1511 	fops = kqueue_fo_find(filt);
1512 	if (fops == NULL)
1513 		return EINVAL;
1514 
1515 	if (kev->flags & EV_ADD) {
1516 		/* Reject an invalid flag pair early */
1517 		if (kev->flags & EV_KEEPUDATA) {
1518 			tkn = NULL;
1519 			error = EINVAL;
1520 			goto done;
1521 		}
1522 
1523 		/*
1524 		 * Prevent waiting with locks.  Non-sleepable
1525 		 * allocation failures are handled in the loop, only
1526 		 * if the spare knote appears to be actually required.
1527 		 */
1528 		tkn = knote_alloc(mflag);
1529 	} else {
1530 		tkn = NULL;
1531 	}
1532 
1533 findkn:
1534 	if (fops->f_isfd) {
1535 		KASSERT(td != NULL, ("td is NULL"));
1536 		if (kev->ident > INT_MAX)
1537 			error = EBADF;
1538 		else
1539 			error = fget(td, kev->ident, &cap_event_rights, &fp);
1540 		if (error)
1541 			goto done;
1542 
1543 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1544 		    kev->ident, M_NOWAIT) != 0) {
1545 			/* try again */
1546 			fdrop(fp, td);
1547 			fp = NULL;
1548 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1549 			if (error)
1550 				goto done;
1551 			goto findkn;
1552 		}
1553 
1554 		if (fp->f_type == DTYPE_KQUEUE) {
1555 			/*
1556 			 * If we add some intelligence about what we are doing,
1557 			 * we should be able to support events on ourselves.
1558 			 * We need to know when we are doing this to prevent
1559 			 * getting both the knlist lock and the kq lock since
1560 			 * they are the same thing.
1561 			 */
1562 			if (fp->f_data == kq) {
1563 				error = EINVAL;
1564 				goto done;
1565 			}
1566 
1567 			/*
1568 			 * Pre-lock the filedesc before the global
1569 			 * lock mutex, see the comment in
1570 			 * kqueue_close().
1571 			 */
1572 			FILEDESC_XLOCK(td->td_proc->p_fd);
1573 			filedesc_unlock = 1;
1574 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1575 		}
1576 
1577 		KQ_LOCK(kq);
1578 		if (kev->ident < kq->kq_knlistsize) {
1579 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1580 				if (kev->filter == kn->kn_filter)
1581 					break;
1582 		}
1583 	} else {
1584 		if ((kev->flags & EV_ADD) == EV_ADD) {
1585 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1586 			if (error != 0)
1587 				goto done;
1588 		}
1589 
1590 		KQ_LOCK(kq);
1591 
1592 		/*
1593 		 * If possible, find an existing knote to use for this kevent.
1594 		 */
1595 		if (kev->filter == EVFILT_PROC &&
1596 		    (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
1597 			/* This is an internal creation of a process tracking
1598 			 * note. Don't attempt to coalesce this with an
1599 			 * existing note.
1600 			 */
1601 			;
1602 		} else if (kq->kq_knhashmask != 0) {
1603 			struct klist *list;
1604 
1605 			list = &kq->kq_knhash[
1606 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1607 			SLIST_FOREACH(kn, list, kn_link)
1608 				if (kev->ident == kn->kn_id &&
1609 				    kev->filter == kn->kn_filter)
1610 					break;
1611 		}
1612 	}
1613 
1614 	/* knote is in the process of changing, wait for it to stabilize. */
1615 	if (kn != NULL && kn_in_flux(kn)) {
1616 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1617 		if (filedesc_unlock) {
1618 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1619 			filedesc_unlock = 0;
1620 		}
1621 		kq->kq_state |= KQ_FLUXWAIT;
1622 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1623 		if (fp != NULL) {
1624 			fdrop(fp, td);
1625 			fp = NULL;
1626 		}
1627 		goto findkn;
1628 	}
1629 
1630 	/*
1631 	 * kn now contains the matching knote, or NULL if no match
1632 	 */
1633 	if (kn == NULL) {
1634 		if (kev->flags & EV_ADD) {
1635 			kn = tkn;
1636 			tkn = NULL;
1637 			if (kn == NULL) {
1638 				KQ_UNLOCK(kq);
1639 				error = ENOMEM;
1640 				goto done;
1641 			}
1642 			kn->kn_fp = fp;
1643 			kn->kn_kq = kq;
1644 			kn->kn_fop = fops;
1645 			/*
1646 			 * apply reference counts to knote structure, and
1647 			 * do not release it at the end of this routine.
1648 			 */
1649 			fops = NULL;
1650 			fp = NULL;
1651 
1652 			kn->kn_sfflags = kev->fflags;
1653 			kn->kn_sdata = kev->data;
1654 			kev->fflags = 0;
1655 			kev->data = 0;
1656 			kn->kn_kevent = *kev;
1657 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1658 			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1659 			kn->kn_status = KN_DETACHED;
1660 			if ((kev->flags & EV_DISABLE) != 0)
1661 				kn->kn_status |= KN_DISABLED;
1662 			kn_enter_flux(kn);
1663 
1664 			error = knote_attach(kn, kq);
1665 			KQ_UNLOCK(kq);
1666 			if (error != 0) {
1667 				tkn = kn;
1668 				goto done;
1669 			}
1670 
1671 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1672 				knote_drop_detached(kn, td);
1673 				goto done;
1674 			}
1675 			knl = kn_list_lock(kn);
1676 			goto done_ev_add;
1677 		} else {
1678 			/* No matching knote and the EV_ADD flag is not set. */
1679 			KQ_UNLOCK(kq);
1680 			error = ENOENT;
1681 			goto done;
1682 		}
1683 	}
1684 
1685 	if (kev->flags & EV_DELETE) {
1686 		kn_enter_flux(kn);
1687 		KQ_UNLOCK(kq);
1688 		knote_drop(kn, td);
1689 		goto done;
1690 	}
1691 
1692 	if (kev->flags & EV_FORCEONESHOT) {
1693 		kn->kn_flags |= EV_ONESHOT;
1694 		KNOTE_ACTIVATE(kn, 1);
1695 	}
1696 
1697 	if ((kev->flags & EV_ENABLE) != 0)
1698 		kn->kn_status &= ~KN_DISABLED;
1699 	else if ((kev->flags & EV_DISABLE) != 0)
1700 		kn->kn_status |= KN_DISABLED;
1701 
1702 	/*
1703 	 * The user may change some filter values after the initial EV_ADD,
1704 	 * but doing so will not reset any filter which has already been
1705 	 * triggered.
1706 	 */
1707 	kn->kn_status |= KN_SCAN;
1708 	kn_enter_flux(kn);
1709 	KQ_UNLOCK(kq);
1710 	knl = kn_list_lock(kn);
1711 	if ((kev->flags & EV_KEEPUDATA) == 0)
1712 		kn->kn_kevent.udata = kev->udata;
1713 	if (!fops->f_isfd && fops->f_touch != NULL) {
1714 		fops->f_touch(kn, kev, EVENT_REGISTER);
1715 	} else {
1716 		kn->kn_sfflags = kev->fflags;
1717 		kn->kn_sdata = kev->data;
1718 	}
1719 
1720 done_ev_add:
1721 	/*
1722 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1723 	 * the initial attach event decides that the event is "completed"
1724 	 * already, e.g., filt_procattach() is called on a zombie process.  It
1725 	 * will call filt_proc() which will remove it from the list, and NULL
1726 	 * kn_knlist.
1727 	 *
1728 	 * KN_DISABLED will be stable while the knote is in flux, so the
1729 	 * unlocked read will not race with an update.
1730 	 */
1731 	if ((kn->kn_status & KN_DISABLED) == 0)
1732 		event = kn->kn_fop->f_event(kn, 0);
1733 	else
1734 		event = 0;
1735 
1736 	KQ_LOCK(kq);
1737 	if (event)
1738 		kn->kn_status |= KN_ACTIVE;
1739 	if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
1740 	    KN_ACTIVE)
1741 		knote_enqueue(kn);
1742 	kn->kn_status &= ~KN_SCAN;
1743 	kn_leave_flux(kn);
1744 	kn_list_unlock(knl);
1745 	KQ_UNLOCK_FLUX(kq);
1746 
1747 done:
1748 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1749 	if (filedesc_unlock)
1750 		FILEDESC_XUNLOCK(td->td_proc->p_fd);
1751 	if (fp != NULL)
1752 		fdrop(fp, td);
1753 	knote_free(tkn);
1754 	if (fops != NULL)
1755 		kqueue_fo_release(filt);
1756 	return (error);
1757 }
1758 
1759 static int
1760 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1761 {
1762 	int error;
1763 	struct kqueue *kq;
1764 
1765 	error = 0;
1766 
1767 	kq = fp->f_data;
1768 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1769 		return (EBADF);
1770 	*kqp = kq;
1771 	KQ_LOCK(kq);
1772 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1773 		KQ_UNLOCK(kq);
1774 		return (EBADF);
1775 	}
1776 	kq->kq_refcnt++;
1777 	KQ_UNLOCK(kq);
1778 
1779 	return error;
1780 }
1781 
1782 static void
1783 kqueue_release(struct kqueue *kq, int locked)
1784 {
1785 	if (locked)
1786 		KQ_OWNED(kq);
1787 	else
1788 		KQ_LOCK(kq);
1789 	kq->kq_refcnt--;
1790 	if (kq->kq_refcnt == 1)
1791 		wakeup(&kq->kq_refcnt);
1792 	if (!locked)
1793 		KQ_UNLOCK(kq);
1794 }
1795 
1796 static void
1797 ast_kqueue(struct thread *td, int tda __unused)
1798 {
1799 	taskqueue_quiesce(taskqueue_kqueue_ctx);
1800 }
1801 
1802 static void
1803 kqueue_schedtask(struct kqueue *kq)
1804 {
1805 	KQ_OWNED(kq);
1806 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1807 	    ("scheduling kqueue task while draining"));
1808 
1809 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1810 		taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
1811 		kq->kq_state |= KQ_TASKSCHED;
1812 		ast_sched(curthread, TDA_KQUEUE);
1813 	}
1814 }
1815 
1816 /*
1817  * Expand the kq to make sure we have storage for fops/ident pair.
1818  *
1819  * Return 0 on success (or no work necessary), return errno on failure.
1820  */
1821 static int
1822 kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident,
1823     int mflag)
1824 {
1825 	struct klist *list, *tmp_knhash, *to_free;
1826 	u_long tmp_knhashmask;
1827 	int error, fd, size;
1828 
1829 	KQ_NOTOWNED(kq);
1830 
1831 	error = 0;
1832 	to_free = NULL;
1833 	if (fops->f_isfd) {
1834 		fd = ident;
1835 		if (kq->kq_knlistsize <= fd) {
1836 			size = kq->kq_knlistsize;
1837 			while (size <= fd)
1838 				size += KQEXTENT;
1839 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1840 			if (list == NULL)
1841 				return ENOMEM;
1842 			KQ_LOCK(kq);
1843 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1844 				to_free = list;
1845 				error = EBADF;
1846 			} else if (kq->kq_knlistsize > fd) {
1847 				to_free = list;
1848 			} else {
1849 				if (kq->kq_knlist != NULL) {
1850 					bcopy(kq->kq_knlist, list,
1851 					    kq->kq_knlistsize * sizeof(*list));
1852 					to_free = kq->kq_knlist;
1853 					kq->kq_knlist = NULL;
1854 				}
1855 				bzero((caddr_t)list +
1856 				    kq->kq_knlistsize * sizeof(*list),
1857 				    (size - kq->kq_knlistsize) * sizeof(*list));
1858 				kq->kq_knlistsize = size;
1859 				kq->kq_knlist = list;
1860 			}
1861 			KQ_UNLOCK(kq);
1862 		}
1863 	} else {
1864 		if (kq->kq_knhashmask == 0) {
1865 			tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE,
1866 			    &tmp_knhashmask, (mflag & M_WAITOK) != 0 ?
1867 			    HASH_WAITOK : HASH_NOWAIT);
1868 			if (tmp_knhash == NULL)
1869 				return (ENOMEM);
1870 			KQ_LOCK(kq);
1871 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1872 				to_free = tmp_knhash;
1873 				error = EBADF;
1874 			} else if (kq->kq_knhashmask == 0) {
1875 				kq->kq_knhash = tmp_knhash;
1876 				kq->kq_knhashmask = tmp_knhashmask;
1877 			} else {
1878 				to_free = tmp_knhash;
1879 			}
1880 			KQ_UNLOCK(kq);
1881 		}
1882 	}
1883 	free(to_free, M_KQUEUE);
1884 
1885 	KQ_NOTOWNED(kq);
1886 	return (error);
1887 }
1888 
1889 static void
1890 kqueue_task(void *arg, int pending)
1891 {
1892 	struct kqueue *kq;
1893 	int haskqglobal;
1894 
1895 	haskqglobal = 0;
1896 	kq = arg;
1897 
1898 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1899 	KQ_LOCK(kq);
1900 
1901 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1902 
1903 	kq->kq_state &= ~KQ_TASKSCHED;
1904 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1905 		wakeup(&kq->kq_state);
1906 	}
1907 	KQ_UNLOCK(kq);
1908 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1909 }
1910 
1911 /*
1912  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1913  * We treat KN_MARKER knotes as if they are in flux.
1914  */
1915 static int
1916 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1917     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1918 {
1919 	struct kevent *kevp;
1920 	struct knote *kn, *marker;
1921 	struct knlist *knl;
1922 	sbintime_t asbt, rsbt;
1923 	int count, error, haskqglobal, influx, nkev, touch;
1924 
1925 	count = maxevents;
1926 	nkev = 0;
1927 	error = 0;
1928 	haskqglobal = 0;
1929 
1930 	if (maxevents == 0)
1931 		goto done_nl;
1932 	if (maxevents < 0) {
1933 		error = EINVAL;
1934 		goto done_nl;
1935 	}
1936 
1937 	rsbt = 0;
1938 	if (tsp != NULL) {
1939 		if (!timespecvalid_interval(tsp)) {
1940 			error = EINVAL;
1941 			goto done_nl;
1942 		}
1943 		if (timespecisset(tsp)) {
1944 			if (tsp->tv_sec <= INT32_MAX) {
1945 				rsbt = tstosbt(*tsp);
1946 				if (TIMESEL(&asbt, rsbt))
1947 					asbt += tc_tick_sbt;
1948 				if (asbt <= SBT_MAX - rsbt)
1949 					asbt += rsbt;
1950 				else
1951 					asbt = 0;
1952 				rsbt >>= tc_precexp;
1953 			} else
1954 				asbt = 0;
1955 		} else
1956 			asbt = -1;
1957 	} else
1958 		asbt = 0;
1959 	marker = knote_alloc(M_WAITOK);
1960 	marker->kn_status = KN_MARKER;
1961 	KQ_LOCK(kq);
1962 
1963 retry:
1964 	kevp = keva;
1965 	if (kq->kq_count == 0) {
1966 		if (asbt == -1) {
1967 			error = EWOULDBLOCK;
1968 		} else {
1969 			kq->kq_state |= KQ_SLEEP;
1970 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1971 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1972 		}
1973 		if (error == 0)
1974 			goto retry;
1975 		/* don't restart after signals... */
1976 		if (error == ERESTART)
1977 			error = EINTR;
1978 		else if (error == EWOULDBLOCK)
1979 			error = 0;
1980 		goto done;
1981 	}
1982 
1983 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1984 	influx = 0;
1985 	while (count) {
1986 		KQ_OWNED(kq);
1987 		kn = TAILQ_FIRST(&kq->kq_head);
1988 
1989 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1990 		    kn_in_flux(kn)) {
1991 			if (influx) {
1992 				influx = 0;
1993 				KQ_FLUX_WAKEUP(kq);
1994 			}
1995 			kq->kq_state |= KQ_FLUXWAIT;
1996 			error = msleep(kq, &kq->kq_lock, PSOCK,
1997 			    "kqflxwt", 0);
1998 			continue;
1999 		}
2000 
2001 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2002 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
2003 			kn->kn_status &= ~KN_QUEUED;
2004 			kq->kq_count--;
2005 			continue;
2006 		}
2007 		if (kn == marker) {
2008 			KQ_FLUX_WAKEUP(kq);
2009 			if (count == maxevents)
2010 				goto retry;
2011 			goto done;
2012 		}
2013 		KASSERT(!kn_in_flux(kn),
2014 		    ("knote %p is unexpectedly in flux", kn));
2015 
2016 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
2017 			kn->kn_status &= ~KN_QUEUED;
2018 			kn_enter_flux(kn);
2019 			kq->kq_count--;
2020 			KQ_UNLOCK(kq);
2021 			/*
2022 			 * We don't need to lock the list since we've
2023 			 * marked it as in flux.
2024 			 */
2025 			knote_drop(kn, td);
2026 			KQ_LOCK(kq);
2027 			continue;
2028 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
2029 			kn->kn_status &= ~KN_QUEUED;
2030 			kn_enter_flux(kn);
2031 			kq->kq_count--;
2032 			KQ_UNLOCK(kq);
2033 			/*
2034 			 * We don't need to lock the list since we've
2035 			 * marked the knote as being in flux.
2036 			 */
2037 			*kevp = kn->kn_kevent;
2038 			knote_drop(kn, td);
2039 			KQ_LOCK(kq);
2040 			kn = NULL;
2041 		} else {
2042 			kn->kn_status |= KN_SCAN;
2043 			kn_enter_flux(kn);
2044 			KQ_UNLOCK(kq);
2045 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
2046 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
2047 			knl = kn_list_lock(kn);
2048 			if (kn->kn_fop->f_event(kn, 0) == 0) {
2049 				KQ_LOCK(kq);
2050 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2051 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
2052 				    KN_SCAN);
2053 				kn_leave_flux(kn);
2054 				kq->kq_count--;
2055 				kn_list_unlock(knl);
2056 				influx = 1;
2057 				continue;
2058 			}
2059 			touch = (!kn->kn_fop->f_isfd &&
2060 			    kn->kn_fop->f_touch != NULL);
2061 			if (touch)
2062 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
2063 			else
2064 				*kevp = kn->kn_kevent;
2065 			KQ_LOCK(kq);
2066 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2067 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
2068 				/*
2069 				 * Manually clear knotes who weren't
2070 				 * 'touch'ed.
2071 				 */
2072 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
2073 					kn->kn_data = 0;
2074 					kn->kn_fflags = 0;
2075 				}
2076 				if (kn->kn_flags & EV_DISPATCH)
2077 					kn->kn_status |= KN_DISABLED;
2078 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
2079 				kq->kq_count--;
2080 			} else
2081 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2082 
2083 			kn->kn_status &= ~KN_SCAN;
2084 			kn_leave_flux(kn);
2085 			kn_list_unlock(knl);
2086 			influx = 1;
2087 		}
2088 
2089 		/* we are returning a copy to the user */
2090 		kevp++;
2091 		nkev++;
2092 		count--;
2093 
2094 		if (nkev == KQ_NEVENTS) {
2095 			influx = 0;
2096 			KQ_UNLOCK_FLUX(kq);
2097 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2098 			nkev = 0;
2099 			kevp = keva;
2100 			KQ_LOCK(kq);
2101 			if (error)
2102 				break;
2103 		}
2104 	}
2105 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
2106 done:
2107 	KQ_OWNED(kq);
2108 	KQ_UNLOCK_FLUX(kq);
2109 	knote_free(marker);
2110 done_nl:
2111 	KQ_NOTOWNED(kq);
2112 	if (nkev != 0)
2113 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2114 	td->td_retval[0] = maxevents - count;
2115 	return (error);
2116 }
2117 
2118 /*ARGSUSED*/
2119 static int
2120 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
2121 	struct ucred *active_cred, struct thread *td)
2122 {
2123 	/*
2124 	 * Enabling sigio causes two major problems:
2125 	 * 1) infinite recursion:
2126 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
2127 	 * set.  On receipt of a signal this will cause a kqueue to recurse
2128 	 * into itself over and over.  Sending the sigio causes the kqueue
2129 	 * to become ready, which in turn posts sigio again, forever.
2130 	 * Solution: this can be solved by setting a flag in the kqueue that
2131 	 * we have a SIGIO in progress.
2132 	 * 2) locking problems:
2133 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
2134 	 * us above the proc and pgrp locks.
2135 	 * Solution: Post a signal using an async mechanism, being sure to
2136 	 * record a generation count in the delivery so that we do not deliver
2137 	 * a signal to the wrong process.
2138 	 *
2139 	 * Note, these two mechanisms are somewhat mutually exclusive!
2140 	 */
2141 #if 0
2142 	struct kqueue *kq;
2143 
2144 	kq = fp->f_data;
2145 	switch (cmd) {
2146 	case FIOASYNC:
2147 		if (*(int *)data) {
2148 			kq->kq_state |= KQ_ASYNC;
2149 		} else {
2150 			kq->kq_state &= ~KQ_ASYNC;
2151 		}
2152 		return (0);
2153 
2154 	case FIOSETOWN:
2155 		return (fsetown(*(int *)data, &kq->kq_sigio));
2156 
2157 	case FIOGETOWN:
2158 		*(int *)data = fgetown(&kq->kq_sigio);
2159 		return (0);
2160 	}
2161 #endif
2162 
2163 	return (ENOTTY);
2164 }
2165 
2166 /*ARGSUSED*/
2167 static int
2168 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
2169 	struct thread *td)
2170 {
2171 	struct kqueue *kq;
2172 	int revents = 0;
2173 	int error;
2174 
2175 	if ((error = kqueue_acquire(fp, &kq)))
2176 		return POLLERR;
2177 
2178 	KQ_LOCK(kq);
2179 	if (events & (POLLIN | POLLRDNORM)) {
2180 		if (kq->kq_count) {
2181 			revents |= events & (POLLIN | POLLRDNORM);
2182 		} else {
2183 			selrecord(td, &kq->kq_sel);
2184 			if (SEL_WAITING(&kq->kq_sel))
2185 				kq->kq_state |= KQ_SEL;
2186 		}
2187 	}
2188 	kqueue_release(kq, 1);
2189 	KQ_UNLOCK(kq);
2190 	return (revents);
2191 }
2192 
2193 /*ARGSUSED*/
2194 static int
2195 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
2196 {
2197 
2198 	bzero((void *)st, sizeof *st);
2199 	/*
2200 	 * We no longer return kq_count because the unlocked value is useless.
2201 	 * If you spent all this time getting the count, why not spend your
2202 	 * syscall better by calling kevent?
2203 	 *
2204 	 * XXX - This is needed for libc_r.
2205 	 */
2206 	st->st_mode = S_IFIFO;
2207 	return (0);
2208 }
2209 
2210 static void
2211 kqueue_drain(struct kqueue *kq, struct thread *td)
2212 {
2213 	struct knote *kn;
2214 	int i;
2215 
2216 	KQ_LOCK(kq);
2217 
2218 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
2219 	    ("kqueue already closing"));
2220 	kq->kq_state |= KQ_CLOSING;
2221 	if (kq->kq_refcnt > 1)
2222 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
2223 
2224 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
2225 
2226 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
2227 	    ("kqueue's knlist not empty"));
2228 
2229 	for (i = 0; i < kq->kq_knlistsize; i++) {
2230 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
2231 			if (kn_in_flux(kn)) {
2232 				kq->kq_state |= KQ_FLUXWAIT;
2233 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
2234 				continue;
2235 			}
2236 			kn_enter_flux(kn);
2237 			KQ_UNLOCK(kq);
2238 			knote_drop(kn, td);
2239 			KQ_LOCK(kq);
2240 		}
2241 	}
2242 	if (kq->kq_knhashmask != 0) {
2243 		for (i = 0; i <= kq->kq_knhashmask; i++) {
2244 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
2245 				if (kn_in_flux(kn)) {
2246 					kq->kq_state |= KQ_FLUXWAIT;
2247 					msleep(kq, &kq->kq_lock, PSOCK,
2248 					       "kqclo2", 0);
2249 					continue;
2250 				}
2251 				kn_enter_flux(kn);
2252 				KQ_UNLOCK(kq);
2253 				knote_drop(kn, td);
2254 				KQ_LOCK(kq);
2255 			}
2256 		}
2257 	}
2258 
2259 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
2260 		kq->kq_state |= KQ_TASKDRAIN;
2261 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
2262 	}
2263 
2264 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2265 		selwakeuppri(&kq->kq_sel, PSOCK);
2266 		if (!SEL_WAITING(&kq->kq_sel))
2267 			kq->kq_state &= ~KQ_SEL;
2268 	}
2269 
2270 	KQ_UNLOCK(kq);
2271 }
2272 
2273 static void
2274 kqueue_destroy(struct kqueue *kq)
2275 {
2276 
2277 	KASSERT(kq->kq_fdp == NULL,
2278 	    ("kqueue still attached to a file descriptor"));
2279 	seldrain(&kq->kq_sel);
2280 	knlist_destroy(&kq->kq_sel.si_note);
2281 	mtx_destroy(&kq->kq_lock);
2282 
2283 	if (kq->kq_knhash != NULL)
2284 		free(kq->kq_knhash, M_KQUEUE);
2285 	if (kq->kq_knlist != NULL)
2286 		free(kq->kq_knlist, M_KQUEUE);
2287 
2288 	funsetown(&kq->kq_sigio);
2289 }
2290 
2291 /*ARGSUSED*/
2292 static int
2293 kqueue_close(struct file *fp, struct thread *td)
2294 {
2295 	struct kqueue *kq = fp->f_data;
2296 	struct filedesc *fdp;
2297 	int error;
2298 	int filedesc_unlock;
2299 
2300 	if ((error = kqueue_acquire(fp, &kq)))
2301 		return error;
2302 	kqueue_drain(kq, td);
2303 
2304 	/*
2305 	 * We could be called due to the knote_drop() doing fdrop(),
2306 	 * called from kqueue_register().  In this case the global
2307 	 * lock is owned, and filedesc sx is locked before, to not
2308 	 * take the sleepable lock after non-sleepable.
2309 	 */
2310 	fdp = kq->kq_fdp;
2311 	kq->kq_fdp = NULL;
2312 	if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
2313 		FILEDESC_XLOCK(fdp);
2314 		filedesc_unlock = 1;
2315 	} else
2316 		filedesc_unlock = 0;
2317 	TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
2318 	if (filedesc_unlock)
2319 		FILEDESC_XUNLOCK(fdp);
2320 
2321 	kqueue_destroy(kq);
2322 	chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
2323 	crfree(kq->kq_cred);
2324 	free(kq, M_KQUEUE);
2325 	fp->f_data = NULL;
2326 
2327 	return (0);
2328 }
2329 
2330 static int
2331 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2332 {
2333 	struct kqueue *kq = fp->f_data;
2334 
2335 	kif->kf_type = KF_TYPE_KQUEUE;
2336 	kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq;
2337 	kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count;
2338 	kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state;
2339 	return (0);
2340 }
2341 
2342 static void
2343 kqueue_wakeup(struct kqueue *kq)
2344 {
2345 	KQ_OWNED(kq);
2346 
2347 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
2348 		kq->kq_state &= ~KQ_SLEEP;
2349 		wakeup(kq);
2350 	}
2351 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2352 		selwakeuppri(&kq->kq_sel, PSOCK);
2353 		if (!SEL_WAITING(&kq->kq_sel))
2354 			kq->kq_state &= ~KQ_SEL;
2355 	}
2356 	if (!knlist_empty(&kq->kq_sel.si_note))
2357 		kqueue_schedtask(kq);
2358 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
2359 		pgsigio(&kq->kq_sigio, SIGIO, 0);
2360 	}
2361 }
2362 
2363 /*
2364  * Walk down a list of knotes, activating them if their event has triggered.
2365  *
2366  * There is a possibility to optimize in the case of one kq watching another.
2367  * Instead of scheduling a task to wake it up, you could pass enough state
2368  * down the chain to make up the parent kqueue.  Make this code functional
2369  * first.
2370  */
2371 void
2372 knote(struct knlist *list, long hint, int lockflags)
2373 {
2374 	struct kqueue *kq;
2375 	struct knote *kn, *tkn;
2376 	int error;
2377 
2378 	if (list == NULL)
2379 		return;
2380 
2381 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
2382 
2383 	if ((lockflags & KNF_LISTLOCKED) == 0)
2384 		list->kl_lock(list->kl_lockarg);
2385 
2386 	/*
2387 	 * If we unlock the list lock (and enter influx), we can
2388 	 * eliminate the kqueue scheduling, but this will introduce
2389 	 * four lock/unlock's for each knote to test.  Also, marker
2390 	 * would be needed to keep iteration position, since filters
2391 	 * or other threads could remove events.
2392 	 */
2393 	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
2394 		kq = kn->kn_kq;
2395 		KQ_LOCK(kq);
2396 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
2397 			/*
2398 			 * Do not process the influx notes, except for
2399 			 * the influx coming from the kq unlock in the
2400 			 * kqueue_scan().  In the later case, we do
2401 			 * not interfere with the scan, since the code
2402 			 * fragment in kqueue_scan() locks the knlist,
2403 			 * and cannot proceed until we finished.
2404 			 */
2405 			KQ_UNLOCK(kq);
2406 		} else if ((lockflags & KNF_NOKQLOCK) != 0) {
2407 			kn_enter_flux(kn);
2408 			KQ_UNLOCK(kq);
2409 			error = kn->kn_fop->f_event(kn, hint);
2410 			KQ_LOCK(kq);
2411 			kn_leave_flux(kn);
2412 			if (error)
2413 				KNOTE_ACTIVATE(kn, 1);
2414 			KQ_UNLOCK_FLUX(kq);
2415 		} else {
2416 			if (kn->kn_fop->f_event(kn, hint))
2417 				KNOTE_ACTIVATE(kn, 1);
2418 			KQ_UNLOCK(kq);
2419 		}
2420 	}
2421 	if ((lockflags & KNF_LISTLOCKED) == 0)
2422 		list->kl_unlock(list->kl_lockarg);
2423 }
2424 
2425 /*
2426  * add a knote to a knlist
2427  */
2428 void
2429 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
2430 {
2431 
2432 	KNL_ASSERT_LOCK(knl, islocked);
2433 	KQ_NOTOWNED(kn->kn_kq);
2434 	KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn));
2435 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2436 	    ("knote %p was not detached", kn));
2437 	if (!islocked)
2438 		knl->kl_lock(knl->kl_lockarg);
2439 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
2440 	if (!islocked)
2441 		knl->kl_unlock(knl->kl_lockarg);
2442 	KQ_LOCK(kn->kn_kq);
2443 	kn->kn_knlist = knl;
2444 	kn->kn_status &= ~KN_DETACHED;
2445 	KQ_UNLOCK(kn->kn_kq);
2446 }
2447 
2448 static void
2449 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
2450     int kqislocked)
2451 {
2452 
2453 	KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked"));
2454 	KNL_ASSERT_LOCK(knl, knlislocked);
2455 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
2456 	KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn));
2457 	KASSERT((kn->kn_status & KN_DETACHED) == 0,
2458 	    ("knote %p was already detached", kn));
2459 	if (!knlislocked)
2460 		knl->kl_lock(knl->kl_lockarg);
2461 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
2462 	kn->kn_knlist = NULL;
2463 	if (!knlislocked)
2464 		kn_list_unlock(knl);
2465 	if (!kqislocked)
2466 		KQ_LOCK(kn->kn_kq);
2467 	kn->kn_status |= KN_DETACHED;
2468 	if (!kqislocked)
2469 		KQ_UNLOCK(kn->kn_kq);
2470 }
2471 
2472 /*
2473  * remove knote from the specified knlist
2474  */
2475 void
2476 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
2477 {
2478 
2479 	knlist_remove_kq(knl, kn, islocked, 0);
2480 }
2481 
2482 int
2483 knlist_empty(struct knlist *knl)
2484 {
2485 
2486 	KNL_ASSERT_LOCKED(knl);
2487 	return (SLIST_EMPTY(&knl->kl_list));
2488 }
2489 
2490 static struct mtx knlist_lock;
2491 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2492     MTX_DEF);
2493 static void knlist_mtx_lock(void *arg);
2494 static void knlist_mtx_unlock(void *arg);
2495 
2496 static void
2497 knlist_mtx_lock(void *arg)
2498 {
2499 
2500 	mtx_lock((struct mtx *)arg);
2501 }
2502 
2503 static void
2504 knlist_mtx_unlock(void *arg)
2505 {
2506 
2507 	mtx_unlock((struct mtx *)arg);
2508 }
2509 
2510 static void
2511 knlist_mtx_assert_lock(void *arg, int what)
2512 {
2513 
2514 	if (what == LA_LOCKED)
2515 		mtx_assert((struct mtx *)arg, MA_OWNED);
2516 	else
2517 		mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2518 }
2519 
2520 void
2521 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2522     void (*kl_unlock)(void *),
2523     void (*kl_assert_lock)(void *, int))
2524 {
2525 
2526 	if (lock == NULL)
2527 		knl->kl_lockarg = &knlist_lock;
2528 	else
2529 		knl->kl_lockarg = lock;
2530 
2531 	if (kl_lock == NULL)
2532 		knl->kl_lock = knlist_mtx_lock;
2533 	else
2534 		knl->kl_lock = kl_lock;
2535 	if (kl_unlock == NULL)
2536 		knl->kl_unlock = knlist_mtx_unlock;
2537 	else
2538 		knl->kl_unlock = kl_unlock;
2539 	if (kl_assert_lock == NULL)
2540 		knl->kl_assert_lock = knlist_mtx_assert_lock;
2541 	else
2542 		knl->kl_assert_lock = kl_assert_lock;
2543 
2544 	knl->kl_autodestroy = 0;
2545 	SLIST_INIT(&knl->kl_list);
2546 }
2547 
2548 void
2549 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2550 {
2551 
2552 	knlist_init(knl, lock, NULL, NULL, NULL);
2553 }
2554 
2555 struct knlist *
2556 knlist_alloc(struct mtx *lock)
2557 {
2558 	struct knlist *knl;
2559 
2560 	knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
2561 	knlist_init_mtx(knl, lock);
2562 	return (knl);
2563 }
2564 
2565 void
2566 knlist_destroy(struct knlist *knl)
2567 {
2568 
2569 	KASSERT(KNLIST_EMPTY(knl),
2570 	    ("destroying knlist %p with knotes on it", knl));
2571 }
2572 
2573 void
2574 knlist_detach(struct knlist *knl)
2575 {
2576 
2577 	KNL_ASSERT_LOCKED(knl);
2578 	knl->kl_autodestroy = 1;
2579 	if (knlist_empty(knl)) {
2580 		knlist_destroy(knl);
2581 		free(knl, M_KQUEUE);
2582 	}
2583 }
2584 
2585 /*
2586  * Even if we are locked, we may need to drop the lock to allow any influx
2587  * knotes time to "settle".
2588  */
2589 void
2590 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2591 {
2592 	struct knote *kn, *kn2;
2593 	struct kqueue *kq;
2594 
2595 	KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
2596 	if (islocked)
2597 		KNL_ASSERT_LOCKED(knl);
2598 	else {
2599 		KNL_ASSERT_UNLOCKED(knl);
2600 again:		/* need to reacquire lock since we have dropped it */
2601 		knl->kl_lock(knl->kl_lockarg);
2602 	}
2603 
2604 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2605 		kq = kn->kn_kq;
2606 		KQ_LOCK(kq);
2607 		if (kn_in_flux(kn)) {
2608 			KQ_UNLOCK(kq);
2609 			continue;
2610 		}
2611 		knlist_remove_kq(knl, kn, 1, 1);
2612 		if (killkn) {
2613 			kn_enter_flux(kn);
2614 			KQ_UNLOCK(kq);
2615 			knote_drop_detached(kn, td);
2616 		} else {
2617 			/* Make sure cleared knotes disappear soon */
2618 			kn->kn_flags |= EV_EOF | EV_ONESHOT;
2619 			KQ_UNLOCK(kq);
2620 		}
2621 		kq = NULL;
2622 	}
2623 
2624 	if (!SLIST_EMPTY(&knl->kl_list)) {
2625 		/* there are still in flux knotes remaining */
2626 		kn = SLIST_FIRST(&knl->kl_list);
2627 		kq = kn->kn_kq;
2628 		KQ_LOCK(kq);
2629 		KASSERT(kn_in_flux(kn), ("knote removed w/o list lock"));
2630 		knl->kl_unlock(knl->kl_lockarg);
2631 		kq->kq_state |= KQ_FLUXWAIT;
2632 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2633 		kq = NULL;
2634 		goto again;
2635 	}
2636 
2637 	if (islocked)
2638 		KNL_ASSERT_LOCKED(knl);
2639 	else {
2640 		knl->kl_unlock(knl->kl_lockarg);
2641 		KNL_ASSERT_UNLOCKED(knl);
2642 	}
2643 }
2644 
2645 /*
2646  * Remove all knotes referencing a specified fd must be called with FILEDESC
2647  * lock.  This prevents a race where a new fd comes along and occupies the
2648  * entry and we attach a knote to the fd.
2649  */
2650 void
2651 knote_fdclose(struct thread *td, int fd)
2652 {
2653 	struct filedesc *fdp = td->td_proc->p_fd;
2654 	struct kqueue *kq;
2655 	struct knote *kn;
2656 	int influx;
2657 
2658 	FILEDESC_XLOCK_ASSERT(fdp);
2659 
2660 	/*
2661 	 * We shouldn't have to worry about new kevents appearing on fd
2662 	 * since filedesc is locked.
2663 	 */
2664 	TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2665 		KQ_LOCK(kq);
2666 
2667 again:
2668 		influx = 0;
2669 		while (kq->kq_knlistsize > fd &&
2670 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2671 			if (kn_in_flux(kn)) {
2672 				/* someone else might be waiting on our knote */
2673 				if (influx)
2674 					wakeup(kq);
2675 				kq->kq_state |= KQ_FLUXWAIT;
2676 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2677 				goto again;
2678 			}
2679 			kn_enter_flux(kn);
2680 			KQ_UNLOCK(kq);
2681 			influx = 1;
2682 			knote_drop(kn, td);
2683 			KQ_LOCK(kq);
2684 		}
2685 		KQ_UNLOCK_FLUX(kq);
2686 	}
2687 }
2688 
2689 static int
2690 knote_attach(struct knote *kn, struct kqueue *kq)
2691 {
2692 	struct klist *list;
2693 
2694 	KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn));
2695 	KQ_OWNED(kq);
2696 
2697 	if ((kq->kq_state & KQ_CLOSING) != 0)
2698 		return (EBADF);
2699 	if (kn->kn_fop->f_isfd) {
2700 		if (kn->kn_id >= kq->kq_knlistsize)
2701 			return (ENOMEM);
2702 		list = &kq->kq_knlist[kn->kn_id];
2703 	} else {
2704 		if (kq->kq_knhash == NULL)
2705 			return (ENOMEM);
2706 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2707 	}
2708 	SLIST_INSERT_HEAD(list, kn, kn_link);
2709 	return (0);
2710 }
2711 
2712 static void
2713 knote_drop(struct knote *kn, struct thread *td)
2714 {
2715 
2716 	if ((kn->kn_status & KN_DETACHED) == 0)
2717 		kn->kn_fop->f_detach(kn);
2718 	knote_drop_detached(kn, td);
2719 }
2720 
2721 static void
2722 knote_drop_detached(struct knote *kn, struct thread *td)
2723 {
2724 	struct kqueue *kq;
2725 	struct klist *list;
2726 
2727 	kq = kn->kn_kq;
2728 
2729 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2730 	    ("knote %p still attached", kn));
2731 	KQ_NOTOWNED(kq);
2732 
2733 	KQ_LOCK(kq);
2734 	KASSERT(kn->kn_influx == 1,
2735 	    ("knote_drop called on %p with influx %d", kn, kn->kn_influx));
2736 
2737 	if (kn->kn_fop->f_isfd)
2738 		list = &kq->kq_knlist[kn->kn_id];
2739 	else
2740 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2741 
2742 	if (!SLIST_EMPTY(list))
2743 		SLIST_REMOVE(list, kn, knote, kn_link);
2744 	if (kn->kn_status & KN_QUEUED)
2745 		knote_dequeue(kn);
2746 	KQ_UNLOCK_FLUX(kq);
2747 
2748 	if (kn->kn_fop->f_isfd) {
2749 		fdrop(kn->kn_fp, td);
2750 		kn->kn_fp = NULL;
2751 	}
2752 	kqueue_fo_release(kn->kn_kevent.filter);
2753 	kn->kn_fop = NULL;
2754 	knote_free(kn);
2755 }
2756 
2757 static void
2758 knote_enqueue(struct knote *kn)
2759 {
2760 	struct kqueue *kq = kn->kn_kq;
2761 
2762 	KQ_OWNED(kn->kn_kq);
2763 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2764 
2765 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2766 	kn->kn_status |= KN_QUEUED;
2767 	kq->kq_count++;
2768 	kqueue_wakeup(kq);
2769 }
2770 
2771 static void
2772 knote_dequeue(struct knote *kn)
2773 {
2774 	struct kqueue *kq = kn->kn_kq;
2775 
2776 	KQ_OWNED(kn->kn_kq);
2777 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2778 
2779 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2780 	kn->kn_status &= ~KN_QUEUED;
2781 	kq->kq_count--;
2782 }
2783 
2784 static void
2785 knote_init(void)
2786 {
2787 
2788 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2789 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2790 	ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue);
2791 }
2792 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2793 
2794 static struct knote *
2795 knote_alloc(int mflag)
2796 {
2797 
2798 	return (uma_zalloc(knote_zone, mflag | M_ZERO));
2799 }
2800 
2801 static void
2802 knote_free(struct knote *kn)
2803 {
2804 
2805 	uma_zfree(knote_zone, kn);
2806 }
2807 
2808 /*
2809  * Register the kev w/ the kq specified by fd.
2810  */
2811 int
2812 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag)
2813 {
2814 	struct kqueue *kq;
2815 	struct file *fp;
2816 	cap_rights_t rights;
2817 	int error;
2818 
2819 	error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE),
2820 	    &fp);
2821 	if (error != 0)
2822 		return (error);
2823 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2824 		goto noacquire;
2825 
2826 	error = kqueue_register(kq, kev, td, mflag);
2827 	kqueue_release(kq, 0);
2828 
2829 noacquire:
2830 	fdrop(fp, td);
2831 	return (error);
2832 }
2833