xref: /freebsd/sys/kern/kern_event.c (revision f5463265955b829775bbb32e1fd0bc11dafc36ce)
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 #include "opt_ktrace.h"
33 #include "opt_kqueue.h"
34 
35 #ifdef COMPAT_FREEBSD11
36 #define	_WANT_FREEBSD11_KEVENT
37 #endif
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/capsicum.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/proc.h>
47 #include <sys/malloc.h>
48 #include <sys/unistd.h>
49 #include <sys/file.h>
50 #include <sys/filedesc.h>
51 #include <sys/filio.h>
52 #include <sys/fcntl.h>
53 #include <sys/kthread.h>
54 #include <sys/selinfo.h>
55 #include <sys/queue.h>
56 #include <sys/event.h>
57 #include <sys/eventvar.h>
58 #include <sys/poll.h>
59 #include <sys/protosw.h>
60 #include <sys/resourcevar.h>
61 #include <sys/sigio.h>
62 #include <sys/signalvar.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysproto.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/taskqueue.h>
70 #include <sys/uio.h>
71 #include <sys/user.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75 #include <machine/atomic.h>
76 
77 #include <vm/uma.h>
78 
79 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
80 
81 /*
82  * This lock is used if multiple kq locks are required.  This possibly
83  * should be made into a per proc lock.
84  */
85 static struct mtx	kq_global;
86 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
87 #define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
88 	if (!haslck)				\
89 		mtx_lock(lck);			\
90 	haslck = 1;				\
91 } while (0)
92 #define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
93 	if (haslck)				\
94 		mtx_unlock(lck);			\
95 	haslck = 0;				\
96 } while (0)
97 
98 TASKQUEUE_DEFINE_THREAD(kqueue_ctx);
99 
100 static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
101 static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
102 static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
103 		    struct thread *td, int mflag);
104 static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
105 static void	kqueue_release(struct kqueue *kq, int locked);
106 static void	kqueue_destroy(struct kqueue *kq);
107 static void	kqueue_drain(struct kqueue *kq, struct thread *td);
108 static int	kqueue_expand(struct kqueue *kq, const struct filterops *fops,
109 		    uintptr_t ident, int mflag);
110 static void	kqueue_task(void *arg, int pending);
111 static int	kqueue_scan(struct kqueue *kq, int maxevents,
112 		    struct kevent_copyops *k_ops,
113 		    const struct timespec *timeout,
114 		    struct kevent *keva, struct thread *td);
115 static void 	kqueue_wakeup(struct kqueue *kq);
116 static const struct filterops *kqueue_fo_find(int filt);
117 static void	kqueue_fo_release(int filt);
118 struct g_kevent_args;
119 static int	kern_kevent_generic(struct thread *td,
120 		    struct g_kevent_args *uap,
121 		    struct kevent_copyops *k_ops, const char *struct_name);
122 
123 static fo_ioctl_t	kqueue_ioctl;
124 static fo_poll_t	kqueue_poll;
125 static fo_kqfilter_t	kqueue_kqfilter;
126 static fo_stat_t	kqueue_stat;
127 static fo_close_t	kqueue_close;
128 static fo_fill_kinfo_t	kqueue_fill_kinfo;
129 
130 static struct fileops kqueueops = {
131 	.fo_read = invfo_rdwr,
132 	.fo_write = invfo_rdwr,
133 	.fo_truncate = invfo_truncate,
134 	.fo_ioctl = kqueue_ioctl,
135 	.fo_poll = kqueue_poll,
136 	.fo_kqfilter = kqueue_kqfilter,
137 	.fo_stat = kqueue_stat,
138 	.fo_close = kqueue_close,
139 	.fo_chmod = invfo_chmod,
140 	.fo_chown = invfo_chown,
141 	.fo_sendfile = invfo_sendfile,
142 	.fo_cmp = file_kcmp_generic,
143 	.fo_fill_kinfo = kqueue_fill_kinfo,
144 };
145 
146 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
147 static void 	knote_drop(struct knote *kn, struct thread *td);
148 static void 	knote_drop_detached(struct knote *kn, struct thread *td);
149 static void 	knote_enqueue(struct knote *kn);
150 static void 	knote_dequeue(struct knote *kn);
151 static void 	knote_init(void);
152 static struct 	knote *knote_alloc(int mflag);
153 static void 	knote_free(struct knote *kn);
154 
155 static void	filt_kqdetach(struct knote *kn);
156 static int	filt_kqueue(struct knote *kn, long hint);
157 static int	filt_procattach(struct knote *kn);
158 static void	filt_procdetach(struct knote *kn);
159 static int	filt_proc(struct knote *kn, long hint);
160 static int	filt_fileattach(struct knote *kn);
161 static void	filt_timerexpire(void *knx);
162 static void	filt_timerexpire_l(struct knote *kn, bool proc_locked);
163 static int	filt_timerattach(struct knote *kn);
164 static void	filt_timerdetach(struct knote *kn);
165 static void	filt_timerstart(struct knote *kn, sbintime_t to);
166 static void	filt_timertouch(struct knote *kn, struct kevent *kev,
167 		    u_long type);
168 static int	filt_timervalidate(struct knote *kn, sbintime_t *to);
169 static int	filt_timer(struct knote *kn, long hint);
170 static int	filt_userattach(struct knote *kn);
171 static void	filt_userdetach(struct knote *kn);
172 static int	filt_user(struct knote *kn, long hint);
173 static void	filt_usertouch(struct knote *kn, struct kevent *kev,
174 		    u_long type);
175 
176 static struct filterops file_filtops = {
177 	.f_isfd = 1,
178 	.f_attach = filt_fileattach,
179 };
180 static struct filterops kqread_filtops = {
181 	.f_isfd = 1,
182 	.f_detach = filt_kqdetach,
183 	.f_event = filt_kqueue,
184 };
185 /* XXX - move to kern_proc.c?  */
186 static struct filterops proc_filtops = {
187 	.f_isfd = 0,
188 	.f_attach = filt_procattach,
189 	.f_detach = filt_procdetach,
190 	.f_event = filt_proc,
191 };
192 static struct filterops timer_filtops = {
193 	.f_isfd = 0,
194 	.f_attach = filt_timerattach,
195 	.f_detach = filt_timerdetach,
196 	.f_event = filt_timer,
197 	.f_touch = filt_timertouch,
198 };
199 static struct filterops user_filtops = {
200 	.f_attach = filt_userattach,
201 	.f_detach = filt_userdetach,
202 	.f_event = filt_user,
203 	.f_touch = filt_usertouch,
204 };
205 
206 static uma_zone_t	knote_zone;
207 static unsigned int __exclusive_cache_line	kq_ncallouts;
208 static unsigned int 	kq_calloutmax = 4 * 1024;
209 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
210     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
211 
212 /* XXX - ensure not influx ? */
213 #define KNOTE_ACTIVATE(kn, islock) do { 				\
214 	if ((islock))							\
215 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
216 	else								\
217 		KQ_LOCK((kn)->kn_kq);					\
218 	(kn)->kn_status |= KN_ACTIVE;					\
219 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
220 		knote_enqueue((kn));					\
221 	if (!(islock))							\
222 		KQ_UNLOCK((kn)->kn_kq);					\
223 } while (0)
224 #define KQ_LOCK(kq) do {						\
225 	mtx_lock(&(kq)->kq_lock);					\
226 } while (0)
227 #define KQ_FLUX_WAKEUP(kq) do {						\
228 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
229 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
230 		wakeup((kq));						\
231 	}								\
232 } while (0)
233 #define KQ_UNLOCK_FLUX(kq) do {						\
234 	KQ_FLUX_WAKEUP(kq);						\
235 	mtx_unlock(&(kq)->kq_lock);					\
236 } while (0)
237 #define KQ_UNLOCK(kq) do {						\
238 	mtx_unlock(&(kq)->kq_lock);					\
239 } while (0)
240 #define KQ_OWNED(kq) do {						\
241 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
242 } while (0)
243 #define KQ_NOTOWNED(kq) do {						\
244 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
245 } while (0)
246 
247 static struct knlist *
248 kn_list_lock(struct knote *kn)
249 {
250 	struct knlist *knl;
251 
252 	knl = kn->kn_knlist;
253 	if (knl != NULL)
254 		knl->kl_lock(knl->kl_lockarg);
255 	return (knl);
256 }
257 
258 static void
259 kn_list_unlock(struct knlist *knl)
260 {
261 	bool do_free;
262 
263 	if (knl == NULL)
264 		return;
265 	do_free = knl->kl_autodestroy && knlist_empty(knl);
266 	knl->kl_unlock(knl->kl_lockarg);
267 	if (do_free) {
268 		knlist_destroy(knl);
269 		free(knl, M_KQUEUE);
270 	}
271 }
272 
273 static bool
274 kn_in_flux(struct knote *kn)
275 {
276 
277 	return (kn->kn_influx > 0);
278 }
279 
280 static void
281 kn_enter_flux(struct knote *kn)
282 {
283 
284 	KQ_OWNED(kn->kn_kq);
285 	MPASS(kn->kn_influx < INT_MAX);
286 	kn->kn_influx++;
287 }
288 
289 static bool
290 kn_leave_flux(struct knote *kn)
291 {
292 
293 	KQ_OWNED(kn->kn_kq);
294 	MPASS(kn->kn_influx > 0);
295 	kn->kn_influx--;
296 	return (kn->kn_influx == 0);
297 }
298 
299 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
300 	if (islocked)							\
301 		KNL_ASSERT_LOCKED(knl);				\
302 	else								\
303 		KNL_ASSERT_UNLOCKED(knl);				\
304 } while (0)
305 #ifdef INVARIANTS
306 #define	KNL_ASSERT_LOCKED(knl) do {					\
307 	knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED);		\
308 } while (0)
309 #define	KNL_ASSERT_UNLOCKED(knl) do {					\
310 	knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED);		\
311 } while (0)
312 #else /* !INVARIANTS */
313 #define	KNL_ASSERT_LOCKED(knl) do {} while (0)
314 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
315 #endif /* INVARIANTS */
316 
317 #ifndef	KN_HASHSIZE
318 #define	KN_HASHSIZE		64		/* XXX should be tunable */
319 #endif
320 
321 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
322 
323 static int
324 filt_nullattach(struct knote *kn)
325 {
326 
327 	return (ENXIO);
328 };
329 
330 struct filterops null_filtops = {
331 	.f_isfd = 0,
332 	.f_attach = filt_nullattach,
333 };
334 
335 /* XXX - make SYSINIT to add these, and move into respective modules. */
336 extern struct filterops sig_filtops;
337 extern struct filterops fs_filtops;
338 
339 /*
340  * Table for all system-defined filters.
341  */
342 static struct mtx	filterops_lock;
343 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF);
344 static struct {
345 	const struct filterops *for_fop;
346 	int for_nolock;
347 	int for_refcnt;
348 } sysfilt_ops[EVFILT_SYSCOUNT] = {
349 	{ &file_filtops, 1 },			/* EVFILT_READ */
350 	{ &file_filtops, 1 },			/* EVFILT_WRITE */
351 	{ &null_filtops },			/* EVFILT_AIO */
352 	{ &file_filtops, 1 },			/* EVFILT_VNODE */
353 	{ &proc_filtops, 1 },			/* EVFILT_PROC */
354 	{ &sig_filtops, 1 },			/* EVFILT_SIGNAL */
355 	{ &timer_filtops, 1 },			/* EVFILT_TIMER */
356 	{ &file_filtops, 1 },			/* EVFILT_PROCDESC */
357 	{ &fs_filtops, 1 },			/* EVFILT_FS */
358 	{ &null_filtops },			/* EVFILT_LIO */
359 	{ &user_filtops, 1 },			/* EVFILT_USER */
360 	{ &null_filtops },			/* EVFILT_SENDFILE */
361 	{ &file_filtops, 1 },                   /* EVFILT_EMPTY */
362 };
363 
364 /*
365  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
366  * method.
367  */
368 static int
369 filt_fileattach(struct knote *kn)
370 {
371 
372 	return (fo_kqfilter(kn->kn_fp, kn));
373 }
374 
375 /*ARGSUSED*/
376 static int
377 kqueue_kqfilter(struct file *fp, struct knote *kn)
378 {
379 	struct kqueue *kq = kn->kn_fp->f_data;
380 
381 	if (kn->kn_filter != EVFILT_READ)
382 		return (EINVAL);
383 
384 	kn->kn_status |= KN_KQUEUE;
385 	kn->kn_fop = &kqread_filtops;
386 	knlist_add(&kq->kq_sel.si_note, kn, 0);
387 
388 	return (0);
389 }
390 
391 static void
392 filt_kqdetach(struct knote *kn)
393 {
394 	struct kqueue *kq = kn->kn_fp->f_data;
395 
396 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
397 }
398 
399 /*ARGSUSED*/
400 static int
401 filt_kqueue(struct knote *kn, long hint)
402 {
403 	struct kqueue *kq = kn->kn_fp->f_data;
404 
405 	kn->kn_data = kq->kq_count;
406 	return (kn->kn_data > 0);
407 }
408 
409 /* XXX - move to kern_proc.c?  */
410 static int
411 filt_procattach(struct knote *kn)
412 {
413 	struct proc *p;
414 	int error;
415 	bool exiting, immediate;
416 
417 	exiting = immediate = false;
418 	if (kn->kn_sfflags & NOTE_EXIT)
419 		p = pfind_any(kn->kn_id);
420 	else
421 		p = pfind(kn->kn_id);
422 	if (p == NULL)
423 		return (ESRCH);
424 	if (p->p_flag & P_WEXIT)
425 		exiting = true;
426 
427 	if ((error = p_cansee(curthread, p))) {
428 		PROC_UNLOCK(p);
429 		return (error);
430 	}
431 
432 	kn->kn_ptr.p_proc = p;
433 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
434 
435 	/*
436 	 * Internal flag indicating registration done by kernel for the
437 	 * purposes of getting a NOTE_CHILD notification.
438 	 */
439 	if (kn->kn_flags & EV_FLAG2) {
440 		kn->kn_flags &= ~EV_FLAG2;
441 		kn->kn_data = kn->kn_sdata;		/* ppid */
442 		kn->kn_fflags = NOTE_CHILD;
443 		kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
444 		immediate = true; /* Force immediate activation of child note. */
445 	}
446 	/*
447 	 * Internal flag indicating registration done by kernel (for other than
448 	 * NOTE_CHILD).
449 	 */
450 	if (kn->kn_flags & EV_FLAG1) {
451 		kn->kn_flags &= ~EV_FLAG1;
452 	}
453 
454 	knlist_add(p->p_klist, kn, 1);
455 
456 	/*
457 	 * Immediately activate any child notes or, in the case of a zombie
458 	 * target process, exit notes.  The latter is necessary to handle the
459 	 * case where the target process, e.g. a child, dies before the kevent
460 	 * is registered.
461 	 */
462 	if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
463 		KNOTE_ACTIVATE(kn, 0);
464 
465 	PROC_UNLOCK(p);
466 
467 	return (0);
468 }
469 
470 /*
471  * The knote may be attached to a different process, which may exit,
472  * leaving nothing for the knote to be attached to.  So when the process
473  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
474  * it will be deleted when read out.  However, as part of the knote deletion,
475  * this routine is called, so a check is needed to avoid actually performing
476  * a detach, because the original process does not exist any more.
477  */
478 /* XXX - move to kern_proc.c?  */
479 static void
480 filt_procdetach(struct knote *kn)
481 {
482 
483 	knlist_remove(kn->kn_knlist, kn, 0);
484 	kn->kn_ptr.p_proc = NULL;
485 }
486 
487 /* XXX - move to kern_proc.c?  */
488 static int
489 filt_proc(struct knote *kn, long hint)
490 {
491 	struct proc *p;
492 	u_int event;
493 
494 	p = kn->kn_ptr.p_proc;
495 	if (p == NULL) /* already activated, from attach filter */
496 		return (0);
497 
498 	/* Mask off extra data. */
499 	event = (u_int)hint & NOTE_PCTRLMASK;
500 
501 	/* If the user is interested in this event, record it. */
502 	if (kn->kn_sfflags & event)
503 		kn->kn_fflags |= event;
504 
505 	/* Process is gone, so flag the event as finished. */
506 	if (event == NOTE_EXIT) {
507 		kn->kn_flags |= EV_EOF | EV_ONESHOT;
508 		kn->kn_ptr.p_proc = NULL;
509 		if (kn->kn_fflags & NOTE_EXIT)
510 			kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
511 		if (kn->kn_fflags == 0)
512 			kn->kn_flags |= EV_DROP;
513 		return (1);
514 	}
515 
516 	return (kn->kn_fflags != 0);
517 }
518 
519 /*
520  * Called when the process forked. It mostly does the same as the
521  * knote(), activating all knotes registered to be activated when the
522  * process forked. Additionally, for each knote attached to the
523  * parent, check whether user wants to track the new process. If so
524  * attach a new knote to it, and immediately report an event with the
525  * child's pid.
526  */
527 void
528 knote_fork(struct knlist *list, int pid)
529 {
530 	struct kqueue *kq;
531 	struct knote *kn;
532 	struct kevent kev;
533 	int error;
534 
535 	MPASS(list != NULL);
536 	KNL_ASSERT_LOCKED(list);
537 	if (SLIST_EMPTY(&list->kl_list))
538 		return;
539 
540 	memset(&kev, 0, sizeof(kev));
541 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
542 		kq = kn->kn_kq;
543 		KQ_LOCK(kq);
544 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
545 			KQ_UNLOCK(kq);
546 			continue;
547 		}
548 
549 		/*
550 		 * The same as knote(), activate the event.
551 		 */
552 		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
553 			if (kn->kn_fop->f_event(kn, NOTE_FORK))
554 				KNOTE_ACTIVATE(kn, 1);
555 			KQ_UNLOCK(kq);
556 			continue;
557 		}
558 
559 		/*
560 		 * The NOTE_TRACK case. In addition to the activation
561 		 * of the event, we need to register new events to
562 		 * track the child. Drop the locks in preparation for
563 		 * the call to kqueue_register().
564 		 */
565 		kn_enter_flux(kn);
566 		KQ_UNLOCK(kq);
567 		list->kl_unlock(list->kl_lockarg);
568 
569 		/*
570 		 * Activate existing knote and register tracking knotes with
571 		 * new process.
572 		 *
573 		 * First register a knote to get just the child notice. This
574 		 * must be a separate note from a potential NOTE_EXIT
575 		 * notification since both NOTE_CHILD and NOTE_EXIT are defined
576 		 * to use the data field (in conflicting ways).
577 		 */
578 		kev.ident = pid;
579 		kev.filter = kn->kn_filter;
580 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
581 		    EV_FLAG2;
582 		kev.fflags = kn->kn_sfflags;
583 		kev.data = kn->kn_id;		/* parent */
584 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
585 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
586 		if (error)
587 			kn->kn_fflags |= NOTE_TRACKERR;
588 
589 		/*
590 		 * Then register another knote to track other potential events
591 		 * from the new process.
592 		 */
593 		kev.ident = pid;
594 		kev.filter = kn->kn_filter;
595 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
596 		kev.fflags = kn->kn_sfflags;
597 		kev.data = kn->kn_id;		/* parent */
598 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
599 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
600 		if (error)
601 			kn->kn_fflags |= NOTE_TRACKERR;
602 		if (kn->kn_fop->f_event(kn, NOTE_FORK))
603 			KNOTE_ACTIVATE(kn, 0);
604 		list->kl_lock(list->kl_lockarg);
605 		KQ_LOCK(kq);
606 		kn_leave_flux(kn);
607 		KQ_UNLOCK_FLUX(kq);
608 	}
609 }
610 
611 /*
612  * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
613  * interval timer support code.
614  */
615 
616 #define NOTE_TIMER_PRECMASK						\
617     (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)
618 
619 static sbintime_t
620 timer2sbintime(int64_t data, int flags)
621 {
622 	int64_t secs;
623 
624         /*
625          * Macros for converting to the fractional second portion of an
626          * sbintime_t using 64bit multiplication to improve precision.
627          */
628 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
629 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
630 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
631 	switch (flags & NOTE_TIMER_PRECMASK) {
632 	case NOTE_SECONDS:
633 #ifdef __LP64__
634 		if (data > (SBT_MAX / SBT_1S))
635 			return (SBT_MAX);
636 #endif
637 		return ((sbintime_t)data << 32);
638 	case NOTE_MSECONDS: /* FALLTHROUGH */
639 	case 0:
640 		if (data >= 1000) {
641 			secs = data / 1000;
642 #ifdef __LP64__
643 			if (secs > (SBT_MAX / SBT_1S))
644 				return (SBT_MAX);
645 #endif
646 			return (secs << 32 | MS_TO_SBT(data % 1000));
647 		}
648 		return (MS_TO_SBT(data));
649 	case NOTE_USECONDS:
650 		if (data >= 1000000) {
651 			secs = data / 1000000;
652 #ifdef __LP64__
653 			if (secs > (SBT_MAX / SBT_1S))
654 				return (SBT_MAX);
655 #endif
656 			return (secs << 32 | US_TO_SBT(data % 1000000));
657 		}
658 		return (US_TO_SBT(data));
659 	case NOTE_NSECONDS:
660 		if (data >= 1000000000) {
661 			secs = data / 1000000000;
662 #ifdef __LP64__
663 			if (secs > (SBT_MAX / SBT_1S))
664 				return (SBT_MAX);
665 #endif
666 			return (secs << 32 | NS_TO_SBT(data % 1000000000));
667 		}
668 		return (NS_TO_SBT(data));
669 	default:
670 		break;
671 	}
672 	return (-1);
673 }
674 
675 struct kq_timer_cb_data {
676 	struct callout c;
677 	struct proc *p;
678 	struct knote *kn;
679 	int cpuid;
680 	int flags;
681 	TAILQ_ENTRY(kq_timer_cb_data) link;
682 	sbintime_t next;	/* next timer event fires at */
683 	sbintime_t to;		/* precalculated timer period, 0 for abs */
684 };
685 
686 #define	KQ_TIMER_CB_ENQUEUED	0x01
687 
688 static void
689 kqtimer_sched_callout(struct kq_timer_cb_data *kc)
690 {
691 	callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn,
692 	    kc->cpuid, C_ABSOLUTE);
693 }
694 
695 void
696 kqtimer_proc_continue(struct proc *p)
697 {
698 	struct kq_timer_cb_data *kc, *kc1;
699 	struct bintime bt;
700 	sbintime_t now;
701 
702 	PROC_LOCK_ASSERT(p, MA_OWNED);
703 
704 	getboottimebin(&bt);
705 	now = bttosbt(bt);
706 
707 	TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) {
708 		TAILQ_REMOVE(&p->p_kqtim_stop, kc, link);
709 		kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
710 		if (kc->next <= now)
711 			filt_timerexpire_l(kc->kn, true);
712 		else
713 			kqtimer_sched_callout(kc);
714 	}
715 }
716 
717 static void
718 filt_timerexpire_l(struct knote *kn, bool proc_locked)
719 {
720 	struct kq_timer_cb_data *kc;
721 	struct proc *p;
722 	uint64_t delta;
723 	sbintime_t now;
724 
725 	kc = kn->kn_ptr.p_v;
726 
727 	if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) {
728 		kn->kn_data++;
729 		KNOTE_ACTIVATE(kn, 0);
730 		return;
731 	}
732 
733 	now = sbinuptime();
734 	if (now >= kc->next) {
735 		delta = (now - kc->next) / kc->to;
736 		if (delta == 0)
737 			delta = 1;
738 		kn->kn_data += delta;
739 		kc->next += delta * kc->to;
740 		if (now >= kc->next)	/* overflow */
741 			kc->next = now + kc->to;
742 		KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
743 	}
744 
745 	/*
746 	 * Initial check for stopped kc->p is racy.  It is fine to
747 	 * miss the set of the stop flags, at worst we would schedule
748 	 * one more callout.  On the other hand, it is not fine to not
749 	 * schedule when we we missed clearing of the flags, we
750 	 * recheck them under the lock and observe consistent state.
751 	 */
752 	p = kc->p;
753 	if (P_SHOULDSTOP(p) || P_KILLED(p)) {
754 		if (!proc_locked)
755 			PROC_LOCK(p);
756 		if (P_SHOULDSTOP(p) || P_KILLED(p)) {
757 			if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) {
758 				kc->flags |= KQ_TIMER_CB_ENQUEUED;
759 				TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link);
760 			}
761 			if (!proc_locked)
762 				PROC_UNLOCK(p);
763 			return;
764 		}
765 		if (!proc_locked)
766 			PROC_UNLOCK(p);
767 	}
768 	kqtimer_sched_callout(kc);
769 }
770 
771 static void
772 filt_timerexpire(void *knx)
773 {
774 	filt_timerexpire_l(knx, false);
775 }
776 
777 /*
778  * data contains amount of time to sleep
779  */
780 static int
781 filt_timervalidate(struct knote *kn, sbintime_t *to)
782 {
783 	struct bintime bt;
784 	sbintime_t sbt;
785 
786 	if (kn->kn_sdata < 0)
787 		return (EINVAL);
788 	if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
789 		kn->kn_sdata = 1;
790 	/*
791 	 * The only fflags values supported are the timer unit
792 	 * (precision) and the absolute time indicator.
793 	 */
794 	if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0)
795 		return (EINVAL);
796 
797 	*to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
798 	if (*to < 0)
799 		return (EINVAL);
800 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
801 		getboottimebin(&bt);
802 		sbt = bttosbt(bt);
803 		*to = MAX(0, *to - sbt);
804 	}
805 	return (0);
806 }
807 
808 static int
809 filt_timerattach(struct knote *kn)
810 {
811 	struct kq_timer_cb_data *kc;
812 	sbintime_t to;
813 	int error;
814 
815 	to = -1;
816 	error = filt_timervalidate(kn, &to);
817 	if (error != 0)
818 		return (error);
819 	KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 ||
820 	    (kn->kn_sfflags & NOTE_ABSTIME) != 0,
821 	    ("%s: periodic timer has a calculated zero timeout", __func__));
822 	KASSERT(to >= 0,
823 	    ("%s: timer has a calculated negative timeout", __func__));
824 
825 	if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) {
826 		atomic_subtract_int(&kq_ncallouts, 1);
827 		return (ENOMEM);
828 	}
829 
830 	if ((kn->kn_sfflags & NOTE_ABSTIME) == 0)
831 		kn->kn_flags |= EV_CLEAR;	/* automatically set */
832 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
833 	kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
834 	kc->kn = kn;
835 	kc->p = curproc;
836 	kc->cpuid = PCPU_GET(cpuid);
837 	kc->flags = 0;
838 	callout_init(&kc->c, 1);
839 	filt_timerstart(kn, to);
840 
841 	return (0);
842 }
843 
844 static void
845 filt_timerstart(struct knote *kn, sbintime_t to)
846 {
847 	struct kq_timer_cb_data *kc;
848 
849 	kc = kn->kn_ptr.p_v;
850 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
851 		kc->next = to;
852 		kc->to = 0;
853 	} else {
854 		kc->next = to + sbinuptime();
855 		kc->to = to;
856 	}
857 	kqtimer_sched_callout(kc);
858 }
859 
860 static void
861 filt_timerdetach(struct knote *kn)
862 {
863 	struct kq_timer_cb_data *kc;
864 	unsigned int old __unused;
865 	bool pending;
866 
867 	kc = kn->kn_ptr.p_v;
868 	do {
869 		callout_drain(&kc->c);
870 
871 		/*
872 		 * kqtimer_proc_continue() might have rescheduled this callout.
873 		 * Double-check, using the process mutex as an interlock.
874 		 */
875 		PROC_LOCK(kc->p);
876 		if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) {
877 			kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
878 			TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link);
879 		}
880 		pending = callout_pending(&kc->c);
881 		PROC_UNLOCK(kc->p);
882 	} while (pending);
883 	free(kc, M_KQUEUE);
884 	old = atomic_fetchadd_int(&kq_ncallouts, -1);
885 	KASSERT(old > 0, ("Number of callouts cannot become negative"));
886 	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
887 }
888 
889 static void
890 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type)
891 {
892 	struct kq_timer_cb_data *kc;
893 	struct kqueue *kq;
894 	sbintime_t to;
895 	int error;
896 
897 	switch (type) {
898 	case EVENT_REGISTER:
899 		/* Handle re-added timers that update data/fflags */
900 		if (kev->flags & EV_ADD) {
901 			kc = kn->kn_ptr.p_v;
902 
903 			/* Drain any existing callout. */
904 			callout_drain(&kc->c);
905 
906 			/* Throw away any existing undelivered record
907 			 * of the timer expiration. This is done under
908 			 * the presumption that if a process is
909 			 * re-adding this timer with new parameters,
910 			 * it is no longer interested in what may have
911 			 * happened under the old parameters. If it is
912 			 * interested, it can wait for the expiration,
913 			 * delete the old timer definition, and then
914 			 * add the new one.
915 			 *
916 			 * This has to be done while the kq is locked:
917 			 *   - if enqueued, dequeue
918 			 *   - make it no longer active
919 			 *   - clear the count of expiration events
920 			 */
921 			kq = kn->kn_kq;
922 			KQ_LOCK(kq);
923 			if (kn->kn_status & KN_QUEUED)
924 				knote_dequeue(kn);
925 
926 			kn->kn_status &= ~KN_ACTIVE;
927 			kn->kn_data = 0;
928 			KQ_UNLOCK(kq);
929 
930 			/* Reschedule timer based on new data/fflags */
931 			kn->kn_sfflags = kev->fflags;
932 			kn->kn_sdata = kev->data;
933 			error = filt_timervalidate(kn, &to);
934 			if (error != 0) {
935 			  	kn->kn_flags |= EV_ERROR;
936 				kn->kn_data = error;
937 			} else
938 			  	filt_timerstart(kn, to);
939 		}
940 		break;
941 
942         case EVENT_PROCESS:
943 		*kev = kn->kn_kevent;
944 		if (kn->kn_flags & EV_CLEAR) {
945 			kn->kn_data = 0;
946 			kn->kn_fflags = 0;
947 		}
948 		break;
949 
950 	default:
951 		panic("filt_timertouch() - invalid type (%ld)", type);
952 		break;
953 	}
954 }
955 
956 static int
957 filt_timer(struct knote *kn, long hint)
958 {
959 
960 	return (kn->kn_data != 0);
961 }
962 
963 static int
964 filt_userattach(struct knote *kn)
965 {
966 
967 	/*
968 	 * EVFILT_USER knotes are not attached to anything in the kernel.
969 	 */
970 	kn->kn_hook = NULL;
971 	if (kn->kn_fflags & NOTE_TRIGGER)
972 		kn->kn_hookid = 1;
973 	else
974 		kn->kn_hookid = 0;
975 	return (0);
976 }
977 
978 static void
979 filt_userdetach(__unused struct knote *kn)
980 {
981 
982 	/*
983 	 * EVFILT_USER knotes are not attached to anything in the kernel.
984 	 */
985 }
986 
987 static int
988 filt_user(struct knote *kn, __unused long hint)
989 {
990 
991 	return (kn->kn_hookid);
992 }
993 
994 static void
995 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
996 {
997 	u_int ffctrl;
998 
999 	switch (type) {
1000 	case EVENT_REGISTER:
1001 		if (kev->fflags & NOTE_TRIGGER)
1002 			kn->kn_hookid = 1;
1003 
1004 		ffctrl = kev->fflags & NOTE_FFCTRLMASK;
1005 		kev->fflags &= NOTE_FFLAGSMASK;
1006 		switch (ffctrl) {
1007 		case NOTE_FFNOP:
1008 			break;
1009 
1010 		case NOTE_FFAND:
1011 			kn->kn_sfflags &= kev->fflags;
1012 			break;
1013 
1014 		case NOTE_FFOR:
1015 			kn->kn_sfflags |= kev->fflags;
1016 			break;
1017 
1018 		case NOTE_FFCOPY:
1019 			kn->kn_sfflags = kev->fflags;
1020 			break;
1021 
1022 		default:
1023 			/* XXX Return error? */
1024 			break;
1025 		}
1026 		kn->kn_sdata = kev->data;
1027 		if (kev->flags & EV_CLEAR) {
1028 			kn->kn_hookid = 0;
1029 			kn->kn_data = 0;
1030 			kn->kn_fflags = 0;
1031 		}
1032 		break;
1033 
1034         case EVENT_PROCESS:
1035 		*kev = kn->kn_kevent;
1036 		kev->fflags = kn->kn_sfflags;
1037 		kev->data = kn->kn_sdata;
1038 		if (kn->kn_flags & EV_CLEAR) {
1039 			kn->kn_hookid = 0;
1040 			kn->kn_data = 0;
1041 			kn->kn_fflags = 0;
1042 		}
1043 		break;
1044 
1045 	default:
1046 		panic("filt_usertouch() - invalid type (%ld)", type);
1047 		break;
1048 	}
1049 }
1050 
1051 int
1052 sys_kqueue(struct thread *td, struct kqueue_args *uap)
1053 {
1054 
1055 	return (kern_kqueue(td, 0, NULL));
1056 }
1057 
1058 int
1059 sys_kqueuex(struct thread *td, struct kqueuex_args *uap)
1060 {
1061 	int flags;
1062 
1063 	if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0)
1064 		return (EINVAL);
1065 	flags = 0;
1066 	if ((uap->flags & KQUEUE_CLOEXEC) != 0)
1067 		flags |= O_CLOEXEC;
1068 	return (kern_kqueue(td, flags, NULL));
1069 }
1070 
1071 static void
1072 kqueue_init(struct kqueue *kq)
1073 {
1074 
1075 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
1076 	TAILQ_INIT(&kq->kq_head);
1077 	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
1078 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
1079 }
1080 
1081 int
1082 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps)
1083 {
1084 	struct filedesc *fdp;
1085 	struct kqueue *kq;
1086 	struct file *fp;
1087 	struct ucred *cred;
1088 	int fd, error;
1089 
1090 	fdp = td->td_proc->p_fd;
1091 	cred = td->td_ucred;
1092 	if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
1093 		return (ENOMEM);
1094 
1095 	error = falloc_caps(td, &fp, &fd, flags, fcaps);
1096 	if (error != 0) {
1097 		chgkqcnt(cred->cr_ruidinfo, -1, 0);
1098 		return (error);
1099 	}
1100 
1101 	/* An extra reference on `fp' has been held for us by falloc(). */
1102 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
1103 	kqueue_init(kq);
1104 	kq->kq_fdp = fdp;
1105 	kq->kq_cred = crhold(cred);
1106 
1107 	FILEDESC_XLOCK(fdp);
1108 	TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
1109 	FILEDESC_XUNLOCK(fdp);
1110 
1111 	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
1112 	fdrop(fp, td);
1113 
1114 	td->td_retval[0] = fd;
1115 	return (0);
1116 }
1117 
1118 struct g_kevent_args {
1119 	int	fd;
1120 	const void *changelist;
1121 	int	nchanges;
1122 	void	*eventlist;
1123 	int	nevents;
1124 	const struct timespec *timeout;
1125 };
1126 
1127 int
1128 sys_kevent(struct thread *td, struct kevent_args *uap)
1129 {
1130 	struct kevent_copyops k_ops = {
1131 		.arg = uap,
1132 		.k_copyout = kevent_copyout,
1133 		.k_copyin = kevent_copyin,
1134 		.kevent_size = sizeof(struct kevent),
1135 	};
1136 	struct g_kevent_args gk_args = {
1137 		.fd = uap->fd,
1138 		.changelist = uap->changelist,
1139 		.nchanges = uap->nchanges,
1140 		.eventlist = uap->eventlist,
1141 		.nevents = uap->nevents,
1142 		.timeout = uap->timeout,
1143 	};
1144 
1145 	return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent"));
1146 }
1147 
1148 static int
1149 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap,
1150     struct kevent_copyops *k_ops, const char *struct_name)
1151 {
1152 	struct timespec ts, *tsp;
1153 #ifdef KTRACE
1154 	struct kevent *eventlist = uap->eventlist;
1155 #endif
1156 	int error;
1157 
1158 	if (uap->timeout != NULL) {
1159 		error = copyin(uap->timeout, &ts, sizeof(ts));
1160 		if (error)
1161 			return (error);
1162 		tsp = &ts;
1163 	} else
1164 		tsp = NULL;
1165 
1166 #ifdef KTRACE
1167 	if (KTRPOINT(td, KTR_STRUCT_ARRAY))
1168 		ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist,
1169 		    uap->nchanges, k_ops->kevent_size);
1170 #endif
1171 
1172 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
1173 	    k_ops, tsp);
1174 
1175 #ifdef KTRACE
1176 	if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
1177 		ktrstructarray(struct_name, UIO_USERSPACE, eventlist,
1178 		    td->td_retval[0], k_ops->kevent_size);
1179 #endif
1180 
1181 	return (error);
1182 }
1183 
1184 /*
1185  * Copy 'count' items into the destination list pointed to by uap->eventlist.
1186  */
1187 static int
1188 kevent_copyout(void *arg, struct kevent *kevp, int count)
1189 {
1190 	struct kevent_args *uap;
1191 	int error;
1192 
1193 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1194 	uap = (struct kevent_args *)arg;
1195 
1196 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
1197 	if (error == 0)
1198 		uap->eventlist += count;
1199 	return (error);
1200 }
1201 
1202 /*
1203  * Copy 'count' items from the list pointed to by uap->changelist.
1204  */
1205 static int
1206 kevent_copyin(void *arg, struct kevent *kevp, int count)
1207 {
1208 	struct kevent_args *uap;
1209 	int error;
1210 
1211 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1212 	uap = (struct kevent_args *)arg;
1213 
1214 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
1215 	if (error == 0)
1216 		uap->changelist += count;
1217 	return (error);
1218 }
1219 
1220 #ifdef COMPAT_FREEBSD11
1221 static int
1222 kevent11_copyout(void *arg, struct kevent *kevp, int count)
1223 {
1224 	struct freebsd11_kevent_args *uap;
1225 	struct freebsd11_kevent kev11;
1226 	int error, i;
1227 
1228 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1229 	uap = (struct freebsd11_kevent_args *)arg;
1230 
1231 	for (i = 0; i < count; i++) {
1232 		kev11.ident = kevp->ident;
1233 		kev11.filter = kevp->filter;
1234 		kev11.flags = kevp->flags;
1235 		kev11.fflags = kevp->fflags;
1236 		kev11.data = kevp->data;
1237 		kev11.udata = kevp->udata;
1238 		error = copyout(&kev11, uap->eventlist, sizeof(kev11));
1239 		if (error != 0)
1240 			break;
1241 		uap->eventlist++;
1242 		kevp++;
1243 	}
1244 	return (error);
1245 }
1246 
1247 /*
1248  * Copy 'count' items from the list pointed to by uap->changelist.
1249  */
1250 static int
1251 kevent11_copyin(void *arg, struct kevent *kevp, int count)
1252 {
1253 	struct freebsd11_kevent_args *uap;
1254 	struct freebsd11_kevent kev11;
1255 	int error, i;
1256 
1257 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1258 	uap = (struct freebsd11_kevent_args *)arg;
1259 
1260 	for (i = 0; i < count; i++) {
1261 		error = copyin(uap->changelist, &kev11, sizeof(kev11));
1262 		if (error != 0)
1263 			break;
1264 		kevp->ident = kev11.ident;
1265 		kevp->filter = kev11.filter;
1266 		kevp->flags = kev11.flags;
1267 		kevp->fflags = kev11.fflags;
1268 		kevp->data = (uintptr_t)kev11.data;
1269 		kevp->udata = kev11.udata;
1270 		bzero(&kevp->ext, sizeof(kevp->ext));
1271 		uap->changelist++;
1272 		kevp++;
1273 	}
1274 	return (error);
1275 }
1276 
1277 int
1278 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap)
1279 {
1280 	struct kevent_copyops k_ops = {
1281 		.arg = uap,
1282 		.k_copyout = kevent11_copyout,
1283 		.k_copyin = kevent11_copyin,
1284 		.kevent_size = sizeof(struct freebsd11_kevent),
1285 	};
1286 	struct g_kevent_args gk_args = {
1287 		.fd = uap->fd,
1288 		.changelist = uap->changelist,
1289 		.nchanges = uap->nchanges,
1290 		.eventlist = uap->eventlist,
1291 		.nevents = uap->nevents,
1292 		.timeout = uap->timeout,
1293 	};
1294 
1295 	return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent"));
1296 }
1297 #endif
1298 
1299 int
1300 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
1301     struct kevent_copyops *k_ops, const struct timespec *timeout)
1302 {
1303 	cap_rights_t rights;
1304 	struct file *fp;
1305 	int error;
1306 
1307 	cap_rights_init_zero(&rights);
1308 	if (nchanges > 0)
1309 		cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE);
1310 	if (nevents > 0)
1311 		cap_rights_set_one(&rights, CAP_KQUEUE_EVENT);
1312 	error = fget(td, fd, &rights, &fp);
1313 	if (error != 0)
1314 		return (error);
1315 
1316 	error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
1317 	fdrop(fp, td);
1318 
1319 	return (error);
1320 }
1321 
1322 static int
1323 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
1324     struct kevent_copyops *k_ops, const struct timespec *timeout)
1325 {
1326 	struct kevent keva[KQ_NEVENTS];
1327 	struct kevent *kevp, *changes;
1328 	int i, n, nerrors, error;
1329 
1330 	if (nchanges < 0)
1331 		return (EINVAL);
1332 
1333 	nerrors = 0;
1334 	while (nchanges > 0) {
1335 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
1336 		error = k_ops->k_copyin(k_ops->arg, keva, n);
1337 		if (error)
1338 			return (error);
1339 		changes = keva;
1340 		for (i = 0; i < n; i++) {
1341 			kevp = &changes[i];
1342 			if (!kevp->filter)
1343 				continue;
1344 			kevp->flags &= ~EV_SYSFLAGS;
1345 			error = kqueue_register(kq, kevp, td, M_WAITOK);
1346 			if (error || (kevp->flags & EV_RECEIPT)) {
1347 				if (nevents == 0)
1348 					return (error);
1349 				kevp->flags = EV_ERROR;
1350 				kevp->data = error;
1351 				(void)k_ops->k_copyout(k_ops->arg, kevp, 1);
1352 				nevents--;
1353 				nerrors++;
1354 			}
1355 		}
1356 		nchanges -= n;
1357 	}
1358 	if (nerrors) {
1359 		td->td_retval[0] = nerrors;
1360 		return (0);
1361 	}
1362 
1363 	return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
1364 }
1365 
1366 int
1367 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
1368     struct kevent_copyops *k_ops, const struct timespec *timeout)
1369 {
1370 	struct kqueue *kq;
1371 	int error;
1372 
1373 	error = kqueue_acquire(fp, &kq);
1374 	if (error != 0)
1375 		return (error);
1376 	error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
1377 	kqueue_release(kq, 0);
1378 	return (error);
1379 }
1380 
1381 /*
1382  * Performs a kevent() call on a temporarily created kqueue. This can be
1383  * used to perform one-shot polling, similar to poll() and select().
1384  */
1385 int
1386 kern_kevent_anonymous(struct thread *td, int nevents,
1387     struct kevent_copyops *k_ops)
1388 {
1389 	struct kqueue kq = {};
1390 	int error;
1391 
1392 	kqueue_init(&kq);
1393 	kq.kq_refcnt = 1;
1394 	error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
1395 	kqueue_drain(&kq, td);
1396 	kqueue_destroy(&kq);
1397 	return (error);
1398 }
1399 
1400 int
1401 kqueue_add_filteropts(int filt, const struct filterops *filtops)
1402 {
1403 	int error;
1404 
1405 	error = 0;
1406 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
1407 		printf(
1408 "trying to add a filterop that is out of range: %d is beyond %d\n",
1409 		    ~filt, EVFILT_SYSCOUNT);
1410 		return EINVAL;
1411 	}
1412 	mtx_lock(&filterops_lock);
1413 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
1414 	    sysfilt_ops[~filt].for_fop != NULL)
1415 		error = EEXIST;
1416 	else {
1417 		sysfilt_ops[~filt].for_fop = filtops;
1418 		sysfilt_ops[~filt].for_refcnt = 0;
1419 	}
1420 	mtx_unlock(&filterops_lock);
1421 
1422 	return (error);
1423 }
1424 
1425 int
1426 kqueue_del_filteropts(int filt)
1427 {
1428 	int error;
1429 
1430 	error = 0;
1431 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1432 		return EINVAL;
1433 
1434 	mtx_lock(&filterops_lock);
1435 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1436 	    sysfilt_ops[~filt].for_fop == NULL)
1437 		error = EINVAL;
1438 	else if (sysfilt_ops[~filt].for_refcnt != 0)
1439 		error = EBUSY;
1440 	else {
1441 		sysfilt_ops[~filt].for_fop = &null_filtops;
1442 		sysfilt_ops[~filt].for_refcnt = 0;
1443 	}
1444 	mtx_unlock(&filterops_lock);
1445 
1446 	return error;
1447 }
1448 
1449 static const struct filterops *
1450 kqueue_fo_find(int filt)
1451 {
1452 
1453 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1454 		return NULL;
1455 
1456 	if (sysfilt_ops[~filt].for_nolock)
1457 		return sysfilt_ops[~filt].for_fop;
1458 
1459 	mtx_lock(&filterops_lock);
1460 	sysfilt_ops[~filt].for_refcnt++;
1461 	if (sysfilt_ops[~filt].for_fop == NULL)
1462 		sysfilt_ops[~filt].for_fop = &null_filtops;
1463 	mtx_unlock(&filterops_lock);
1464 
1465 	return sysfilt_ops[~filt].for_fop;
1466 }
1467 
1468 static void
1469 kqueue_fo_release(int filt)
1470 {
1471 
1472 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1473 		return;
1474 
1475 	if (sysfilt_ops[~filt].for_nolock)
1476 		return;
1477 
1478 	mtx_lock(&filterops_lock);
1479 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1480 	    ("filter object refcount not valid on release"));
1481 	sysfilt_ops[~filt].for_refcnt--;
1482 	mtx_unlock(&filterops_lock);
1483 }
1484 
1485 /*
1486  * A ref to kq (obtained via kqueue_acquire) must be held.
1487  */
1488 static int
1489 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td,
1490     int mflag)
1491 {
1492 	const struct filterops *fops;
1493 	struct file *fp;
1494 	struct knote *kn, *tkn;
1495 	struct knlist *knl;
1496 	int error, filt, event;
1497 	int haskqglobal, filedesc_unlock;
1498 
1499 	if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
1500 		return (EINVAL);
1501 
1502 	fp = NULL;
1503 	kn = NULL;
1504 	knl = NULL;
1505 	error = 0;
1506 	haskqglobal = 0;
1507 	filedesc_unlock = 0;
1508 
1509 	filt = kev->filter;
1510 	fops = kqueue_fo_find(filt);
1511 	if (fops == NULL)
1512 		return EINVAL;
1513 
1514 	if (kev->flags & EV_ADD) {
1515 		/* Reject an invalid flag pair early */
1516 		if (kev->flags & EV_KEEPUDATA) {
1517 			tkn = NULL;
1518 			error = EINVAL;
1519 			goto done;
1520 		}
1521 
1522 		/*
1523 		 * Prevent waiting with locks.  Non-sleepable
1524 		 * allocation failures are handled in the loop, only
1525 		 * if the spare knote appears to be actually required.
1526 		 */
1527 		tkn = knote_alloc(mflag);
1528 	} else {
1529 		tkn = NULL;
1530 	}
1531 
1532 findkn:
1533 	if (fops->f_isfd) {
1534 		KASSERT(td != NULL, ("td is NULL"));
1535 		if (kev->ident > INT_MAX)
1536 			error = EBADF;
1537 		else
1538 			error = fget(td, kev->ident, &cap_event_rights, &fp);
1539 		if (error)
1540 			goto done;
1541 
1542 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1543 		    kev->ident, M_NOWAIT) != 0) {
1544 			/* try again */
1545 			fdrop(fp, td);
1546 			fp = NULL;
1547 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1548 			if (error)
1549 				goto done;
1550 			goto findkn;
1551 		}
1552 
1553 		if (fp->f_type == DTYPE_KQUEUE) {
1554 			/*
1555 			 * If we add some intelligence about what we are doing,
1556 			 * we should be able to support events on ourselves.
1557 			 * We need to know when we are doing this to prevent
1558 			 * getting both the knlist lock and the kq lock since
1559 			 * they are the same thing.
1560 			 */
1561 			if (fp->f_data == kq) {
1562 				error = EINVAL;
1563 				goto done;
1564 			}
1565 
1566 			/*
1567 			 * Pre-lock the filedesc before the global
1568 			 * lock mutex, see the comment in
1569 			 * kqueue_close().
1570 			 */
1571 			FILEDESC_XLOCK(td->td_proc->p_fd);
1572 			filedesc_unlock = 1;
1573 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1574 		}
1575 
1576 		KQ_LOCK(kq);
1577 		if (kev->ident < kq->kq_knlistsize) {
1578 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1579 				if (kev->filter == kn->kn_filter)
1580 					break;
1581 		}
1582 	} else {
1583 		if ((kev->flags & EV_ADD) == EV_ADD) {
1584 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1585 			if (error != 0)
1586 				goto done;
1587 		}
1588 
1589 		KQ_LOCK(kq);
1590 
1591 		/*
1592 		 * If possible, find an existing knote to use for this kevent.
1593 		 */
1594 		if (kev->filter == EVFILT_PROC &&
1595 		    (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
1596 			/* This is an internal creation of a process tracking
1597 			 * note. Don't attempt to coalesce this with an
1598 			 * existing note.
1599 			 */
1600 			;
1601 		} else if (kq->kq_knhashmask != 0) {
1602 			struct klist *list;
1603 
1604 			list = &kq->kq_knhash[
1605 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1606 			SLIST_FOREACH(kn, list, kn_link)
1607 				if (kev->ident == kn->kn_id &&
1608 				    kev->filter == kn->kn_filter)
1609 					break;
1610 		}
1611 	}
1612 
1613 	/* knote is in the process of changing, wait for it to stabilize. */
1614 	if (kn != NULL && kn_in_flux(kn)) {
1615 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1616 		if (filedesc_unlock) {
1617 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1618 			filedesc_unlock = 0;
1619 		}
1620 		kq->kq_state |= KQ_FLUXWAIT;
1621 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1622 		if (fp != NULL) {
1623 			fdrop(fp, td);
1624 			fp = NULL;
1625 		}
1626 		goto findkn;
1627 	}
1628 
1629 	/*
1630 	 * kn now contains the matching knote, or NULL if no match
1631 	 */
1632 	if (kn == NULL) {
1633 		if (kev->flags & EV_ADD) {
1634 			kn = tkn;
1635 			tkn = NULL;
1636 			if (kn == NULL) {
1637 				KQ_UNLOCK(kq);
1638 				error = ENOMEM;
1639 				goto done;
1640 			}
1641 			kn->kn_fp = fp;
1642 			kn->kn_kq = kq;
1643 			kn->kn_fop = fops;
1644 			/*
1645 			 * apply reference counts to knote structure, and
1646 			 * do not release it at the end of this routine.
1647 			 */
1648 			fops = NULL;
1649 			fp = NULL;
1650 
1651 			kn->kn_sfflags = kev->fflags;
1652 			kn->kn_sdata = kev->data;
1653 			kev->fflags = 0;
1654 			kev->data = 0;
1655 			kn->kn_kevent = *kev;
1656 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1657 			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1658 			kn->kn_status = KN_DETACHED;
1659 			if ((kev->flags & EV_DISABLE) != 0)
1660 				kn->kn_status |= KN_DISABLED;
1661 			kn_enter_flux(kn);
1662 
1663 			error = knote_attach(kn, kq);
1664 			KQ_UNLOCK(kq);
1665 			if (error != 0) {
1666 				tkn = kn;
1667 				goto done;
1668 			}
1669 
1670 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1671 				knote_drop_detached(kn, td);
1672 				goto done;
1673 			}
1674 			knl = kn_list_lock(kn);
1675 			goto done_ev_add;
1676 		} else {
1677 			/* No matching knote and the EV_ADD flag is not set. */
1678 			KQ_UNLOCK(kq);
1679 			error = ENOENT;
1680 			goto done;
1681 		}
1682 	}
1683 
1684 	if (kev->flags & EV_DELETE) {
1685 		kn_enter_flux(kn);
1686 		KQ_UNLOCK(kq);
1687 		knote_drop(kn, td);
1688 		goto done;
1689 	}
1690 
1691 	if (kev->flags & EV_FORCEONESHOT) {
1692 		kn->kn_flags |= EV_ONESHOT;
1693 		KNOTE_ACTIVATE(kn, 1);
1694 	}
1695 
1696 	if ((kev->flags & EV_ENABLE) != 0)
1697 		kn->kn_status &= ~KN_DISABLED;
1698 	else if ((kev->flags & EV_DISABLE) != 0)
1699 		kn->kn_status |= KN_DISABLED;
1700 
1701 	/*
1702 	 * The user may change some filter values after the initial EV_ADD,
1703 	 * but doing so will not reset any filter which has already been
1704 	 * triggered.
1705 	 */
1706 	kn->kn_status |= KN_SCAN;
1707 	kn_enter_flux(kn);
1708 	KQ_UNLOCK(kq);
1709 	knl = kn_list_lock(kn);
1710 	if ((kev->flags & EV_KEEPUDATA) == 0)
1711 		kn->kn_kevent.udata = kev->udata;
1712 	if (!fops->f_isfd && fops->f_touch != NULL) {
1713 		fops->f_touch(kn, kev, EVENT_REGISTER);
1714 	} else {
1715 		kn->kn_sfflags = kev->fflags;
1716 		kn->kn_sdata = kev->data;
1717 	}
1718 
1719 done_ev_add:
1720 	/*
1721 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1722 	 * the initial attach event decides that the event is "completed"
1723 	 * already, e.g., filt_procattach() is called on a zombie process.  It
1724 	 * will call filt_proc() which will remove it from the list, and NULL
1725 	 * kn_knlist.
1726 	 *
1727 	 * KN_DISABLED will be stable while the knote is in flux, so the
1728 	 * unlocked read will not race with an update.
1729 	 */
1730 	if ((kn->kn_status & KN_DISABLED) == 0)
1731 		event = kn->kn_fop->f_event(kn, 0);
1732 	else
1733 		event = 0;
1734 
1735 	KQ_LOCK(kq);
1736 	if (event)
1737 		kn->kn_status |= KN_ACTIVE;
1738 	if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
1739 	    KN_ACTIVE)
1740 		knote_enqueue(kn);
1741 	kn->kn_status &= ~KN_SCAN;
1742 	kn_leave_flux(kn);
1743 	kn_list_unlock(knl);
1744 	KQ_UNLOCK_FLUX(kq);
1745 
1746 done:
1747 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1748 	if (filedesc_unlock)
1749 		FILEDESC_XUNLOCK(td->td_proc->p_fd);
1750 	if (fp != NULL)
1751 		fdrop(fp, td);
1752 	knote_free(tkn);
1753 	if (fops != NULL)
1754 		kqueue_fo_release(filt);
1755 	return (error);
1756 }
1757 
1758 static int
1759 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1760 {
1761 	int error;
1762 	struct kqueue *kq;
1763 
1764 	error = 0;
1765 
1766 	kq = fp->f_data;
1767 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1768 		return (EBADF);
1769 	*kqp = kq;
1770 	KQ_LOCK(kq);
1771 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1772 		KQ_UNLOCK(kq);
1773 		return (EBADF);
1774 	}
1775 	kq->kq_refcnt++;
1776 	KQ_UNLOCK(kq);
1777 
1778 	return error;
1779 }
1780 
1781 static void
1782 kqueue_release(struct kqueue *kq, int locked)
1783 {
1784 	if (locked)
1785 		KQ_OWNED(kq);
1786 	else
1787 		KQ_LOCK(kq);
1788 	kq->kq_refcnt--;
1789 	if (kq->kq_refcnt == 1)
1790 		wakeup(&kq->kq_refcnt);
1791 	if (!locked)
1792 		KQ_UNLOCK(kq);
1793 }
1794 
1795 static void
1796 ast_kqueue(struct thread *td, int tda __unused)
1797 {
1798 	taskqueue_quiesce(taskqueue_kqueue_ctx);
1799 }
1800 
1801 static void
1802 kqueue_schedtask(struct kqueue *kq)
1803 {
1804 	KQ_OWNED(kq);
1805 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1806 	    ("scheduling kqueue task while draining"));
1807 
1808 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1809 		taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
1810 		kq->kq_state |= KQ_TASKSCHED;
1811 		ast_sched(curthread, TDA_KQUEUE);
1812 	}
1813 }
1814 
1815 /*
1816  * Expand the kq to make sure we have storage for fops/ident pair.
1817  *
1818  * Return 0 on success (or no work necessary), return errno on failure.
1819  */
1820 static int
1821 kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident,
1822     int mflag)
1823 {
1824 	struct klist *list, *tmp_knhash, *to_free;
1825 	u_long tmp_knhashmask;
1826 	int error, fd, size;
1827 
1828 	KQ_NOTOWNED(kq);
1829 
1830 	error = 0;
1831 	to_free = NULL;
1832 	if (fops->f_isfd) {
1833 		fd = ident;
1834 		if (kq->kq_knlistsize <= fd) {
1835 			size = kq->kq_knlistsize;
1836 			while (size <= fd)
1837 				size += KQEXTENT;
1838 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1839 			if (list == NULL)
1840 				return ENOMEM;
1841 			KQ_LOCK(kq);
1842 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1843 				to_free = list;
1844 				error = EBADF;
1845 			} else if (kq->kq_knlistsize > fd) {
1846 				to_free = list;
1847 			} else {
1848 				if (kq->kq_knlist != NULL) {
1849 					bcopy(kq->kq_knlist, list,
1850 					    kq->kq_knlistsize * sizeof(*list));
1851 					to_free = kq->kq_knlist;
1852 					kq->kq_knlist = NULL;
1853 				}
1854 				bzero((caddr_t)list +
1855 				    kq->kq_knlistsize * sizeof(*list),
1856 				    (size - kq->kq_knlistsize) * sizeof(*list));
1857 				kq->kq_knlistsize = size;
1858 				kq->kq_knlist = list;
1859 			}
1860 			KQ_UNLOCK(kq);
1861 		}
1862 	} else {
1863 		if (kq->kq_knhashmask == 0) {
1864 			tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE,
1865 			    &tmp_knhashmask, (mflag & M_WAITOK) != 0 ?
1866 			    HASH_WAITOK : HASH_NOWAIT);
1867 			if (tmp_knhash == NULL)
1868 				return (ENOMEM);
1869 			KQ_LOCK(kq);
1870 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1871 				to_free = tmp_knhash;
1872 				error = EBADF;
1873 			} else if (kq->kq_knhashmask == 0) {
1874 				kq->kq_knhash = tmp_knhash;
1875 				kq->kq_knhashmask = tmp_knhashmask;
1876 			} else {
1877 				to_free = tmp_knhash;
1878 			}
1879 			KQ_UNLOCK(kq);
1880 		}
1881 	}
1882 	free(to_free, M_KQUEUE);
1883 
1884 	KQ_NOTOWNED(kq);
1885 	return (error);
1886 }
1887 
1888 static void
1889 kqueue_task(void *arg, int pending)
1890 {
1891 	struct kqueue *kq;
1892 	int haskqglobal;
1893 
1894 	haskqglobal = 0;
1895 	kq = arg;
1896 
1897 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1898 	KQ_LOCK(kq);
1899 
1900 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1901 
1902 	kq->kq_state &= ~KQ_TASKSCHED;
1903 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1904 		wakeup(&kq->kq_state);
1905 	}
1906 	KQ_UNLOCK(kq);
1907 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1908 }
1909 
1910 /*
1911  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1912  * We treat KN_MARKER knotes as if they are in flux.
1913  */
1914 static int
1915 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1916     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1917 {
1918 	struct kevent *kevp;
1919 	struct knote *kn, *marker;
1920 	struct knlist *knl;
1921 	sbintime_t asbt, rsbt;
1922 	int count, error, haskqglobal, influx, nkev, touch;
1923 
1924 	count = maxevents;
1925 	nkev = 0;
1926 	error = 0;
1927 	haskqglobal = 0;
1928 
1929 	if (maxevents == 0)
1930 		goto done_nl;
1931 	if (maxevents < 0) {
1932 		error = EINVAL;
1933 		goto done_nl;
1934 	}
1935 
1936 	rsbt = 0;
1937 	if (tsp != NULL) {
1938 		if (!timespecvalid_interval(tsp)) {
1939 			error = EINVAL;
1940 			goto done_nl;
1941 		}
1942 		if (timespecisset(tsp)) {
1943 			if (tsp->tv_sec <= INT32_MAX) {
1944 				rsbt = tstosbt(*tsp);
1945 				if (TIMESEL(&asbt, rsbt))
1946 					asbt += tc_tick_sbt;
1947 				if (asbt <= SBT_MAX - rsbt)
1948 					asbt += rsbt;
1949 				else
1950 					asbt = 0;
1951 				rsbt >>= tc_precexp;
1952 			} else
1953 				asbt = 0;
1954 		} else
1955 			asbt = -1;
1956 	} else
1957 		asbt = 0;
1958 	marker = knote_alloc(M_WAITOK);
1959 	marker->kn_status = KN_MARKER;
1960 	KQ_LOCK(kq);
1961 
1962 retry:
1963 	kevp = keva;
1964 	if (kq->kq_count == 0) {
1965 		if (asbt == -1) {
1966 			error = EWOULDBLOCK;
1967 		} else {
1968 			kq->kq_state |= KQ_SLEEP;
1969 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1970 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1971 		}
1972 		if (error == 0)
1973 			goto retry;
1974 		/* don't restart after signals... */
1975 		if (error == ERESTART)
1976 			error = EINTR;
1977 		else if (error == EWOULDBLOCK)
1978 			error = 0;
1979 		goto done;
1980 	}
1981 
1982 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1983 	influx = 0;
1984 	while (count) {
1985 		KQ_OWNED(kq);
1986 		kn = TAILQ_FIRST(&kq->kq_head);
1987 
1988 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1989 		    kn_in_flux(kn)) {
1990 			if (influx) {
1991 				influx = 0;
1992 				KQ_FLUX_WAKEUP(kq);
1993 			}
1994 			kq->kq_state |= KQ_FLUXWAIT;
1995 			error = msleep(kq, &kq->kq_lock, PSOCK,
1996 			    "kqflxwt", 0);
1997 			continue;
1998 		}
1999 
2000 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2001 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
2002 			kn->kn_status &= ~KN_QUEUED;
2003 			kq->kq_count--;
2004 			continue;
2005 		}
2006 		if (kn == marker) {
2007 			KQ_FLUX_WAKEUP(kq);
2008 			if (count == maxevents)
2009 				goto retry;
2010 			goto done;
2011 		}
2012 		KASSERT(!kn_in_flux(kn),
2013 		    ("knote %p is unexpectedly in flux", kn));
2014 
2015 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
2016 			kn->kn_status &= ~KN_QUEUED;
2017 			kn_enter_flux(kn);
2018 			kq->kq_count--;
2019 			KQ_UNLOCK(kq);
2020 			/*
2021 			 * We don't need to lock the list since we've
2022 			 * marked it as in flux.
2023 			 */
2024 			knote_drop(kn, td);
2025 			KQ_LOCK(kq);
2026 			continue;
2027 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
2028 			kn->kn_status &= ~KN_QUEUED;
2029 			kn_enter_flux(kn);
2030 			kq->kq_count--;
2031 			KQ_UNLOCK(kq);
2032 			/*
2033 			 * We don't need to lock the list since we've
2034 			 * marked the knote as being in flux.
2035 			 */
2036 			*kevp = kn->kn_kevent;
2037 			knote_drop(kn, td);
2038 			KQ_LOCK(kq);
2039 			kn = NULL;
2040 		} else {
2041 			kn->kn_status |= KN_SCAN;
2042 			kn_enter_flux(kn);
2043 			KQ_UNLOCK(kq);
2044 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
2045 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
2046 			knl = kn_list_lock(kn);
2047 			if (kn->kn_fop->f_event(kn, 0) == 0) {
2048 				KQ_LOCK(kq);
2049 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2050 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
2051 				    KN_SCAN);
2052 				kn_leave_flux(kn);
2053 				kq->kq_count--;
2054 				kn_list_unlock(knl);
2055 				influx = 1;
2056 				continue;
2057 			}
2058 			touch = (!kn->kn_fop->f_isfd &&
2059 			    kn->kn_fop->f_touch != NULL);
2060 			if (touch)
2061 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
2062 			else
2063 				*kevp = kn->kn_kevent;
2064 			KQ_LOCK(kq);
2065 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2066 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
2067 				/*
2068 				 * Manually clear knotes who weren't
2069 				 * 'touch'ed.
2070 				 */
2071 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
2072 					kn->kn_data = 0;
2073 					kn->kn_fflags = 0;
2074 				}
2075 				if (kn->kn_flags & EV_DISPATCH)
2076 					kn->kn_status |= KN_DISABLED;
2077 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
2078 				kq->kq_count--;
2079 			} else
2080 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2081 
2082 			kn->kn_status &= ~KN_SCAN;
2083 			kn_leave_flux(kn);
2084 			kn_list_unlock(knl);
2085 			influx = 1;
2086 		}
2087 
2088 		/* we are returning a copy to the user */
2089 		kevp++;
2090 		nkev++;
2091 		count--;
2092 
2093 		if (nkev == KQ_NEVENTS) {
2094 			influx = 0;
2095 			KQ_UNLOCK_FLUX(kq);
2096 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2097 			nkev = 0;
2098 			kevp = keva;
2099 			KQ_LOCK(kq);
2100 			if (error)
2101 				break;
2102 		}
2103 	}
2104 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
2105 done:
2106 	KQ_OWNED(kq);
2107 	KQ_UNLOCK_FLUX(kq);
2108 	knote_free(marker);
2109 done_nl:
2110 	KQ_NOTOWNED(kq);
2111 	if (nkev != 0)
2112 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2113 	td->td_retval[0] = maxevents - count;
2114 	return (error);
2115 }
2116 
2117 /*ARGSUSED*/
2118 static int
2119 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
2120 	struct ucred *active_cred, struct thread *td)
2121 {
2122 	/*
2123 	 * Enabling sigio causes two major problems:
2124 	 * 1) infinite recursion:
2125 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
2126 	 * set.  On receipt of a signal this will cause a kqueue to recurse
2127 	 * into itself over and over.  Sending the sigio causes the kqueue
2128 	 * to become ready, which in turn posts sigio again, forever.
2129 	 * Solution: this can be solved by setting a flag in the kqueue that
2130 	 * we have a SIGIO in progress.
2131 	 * 2) locking problems:
2132 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
2133 	 * us above the proc and pgrp locks.
2134 	 * Solution: Post a signal using an async mechanism, being sure to
2135 	 * record a generation count in the delivery so that we do not deliver
2136 	 * a signal to the wrong process.
2137 	 *
2138 	 * Note, these two mechanisms are somewhat mutually exclusive!
2139 	 */
2140 #if 0
2141 	struct kqueue *kq;
2142 
2143 	kq = fp->f_data;
2144 	switch (cmd) {
2145 	case FIOASYNC:
2146 		if (*(int *)data) {
2147 			kq->kq_state |= KQ_ASYNC;
2148 		} else {
2149 			kq->kq_state &= ~KQ_ASYNC;
2150 		}
2151 		return (0);
2152 
2153 	case FIOSETOWN:
2154 		return (fsetown(*(int *)data, &kq->kq_sigio));
2155 
2156 	case FIOGETOWN:
2157 		*(int *)data = fgetown(&kq->kq_sigio);
2158 		return (0);
2159 	}
2160 #endif
2161 
2162 	return (ENOTTY);
2163 }
2164 
2165 /*ARGSUSED*/
2166 static int
2167 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
2168 	struct thread *td)
2169 {
2170 	struct kqueue *kq;
2171 	int revents = 0;
2172 	int error;
2173 
2174 	if ((error = kqueue_acquire(fp, &kq)))
2175 		return POLLERR;
2176 
2177 	KQ_LOCK(kq);
2178 	if (events & (POLLIN | POLLRDNORM)) {
2179 		if (kq->kq_count) {
2180 			revents |= events & (POLLIN | POLLRDNORM);
2181 		} else {
2182 			selrecord(td, &kq->kq_sel);
2183 			if (SEL_WAITING(&kq->kq_sel))
2184 				kq->kq_state |= KQ_SEL;
2185 		}
2186 	}
2187 	kqueue_release(kq, 1);
2188 	KQ_UNLOCK(kq);
2189 	return (revents);
2190 }
2191 
2192 /*ARGSUSED*/
2193 static int
2194 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
2195 {
2196 
2197 	bzero((void *)st, sizeof *st);
2198 	/*
2199 	 * We no longer return kq_count because the unlocked value is useless.
2200 	 * If you spent all this time getting the count, why not spend your
2201 	 * syscall better by calling kevent?
2202 	 *
2203 	 * XXX - This is needed for libc_r.
2204 	 */
2205 	st->st_mode = S_IFIFO;
2206 	return (0);
2207 }
2208 
2209 static void
2210 kqueue_drain(struct kqueue *kq, struct thread *td)
2211 {
2212 	struct knote *kn;
2213 	int i;
2214 
2215 	KQ_LOCK(kq);
2216 
2217 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
2218 	    ("kqueue already closing"));
2219 	kq->kq_state |= KQ_CLOSING;
2220 	if (kq->kq_refcnt > 1)
2221 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
2222 
2223 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
2224 
2225 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
2226 	    ("kqueue's knlist not empty"));
2227 
2228 	for (i = 0; i < kq->kq_knlistsize; i++) {
2229 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
2230 			if (kn_in_flux(kn)) {
2231 				kq->kq_state |= KQ_FLUXWAIT;
2232 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
2233 				continue;
2234 			}
2235 			kn_enter_flux(kn);
2236 			KQ_UNLOCK(kq);
2237 			knote_drop(kn, td);
2238 			KQ_LOCK(kq);
2239 		}
2240 	}
2241 	if (kq->kq_knhashmask != 0) {
2242 		for (i = 0; i <= kq->kq_knhashmask; i++) {
2243 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
2244 				if (kn_in_flux(kn)) {
2245 					kq->kq_state |= KQ_FLUXWAIT;
2246 					msleep(kq, &kq->kq_lock, PSOCK,
2247 					       "kqclo2", 0);
2248 					continue;
2249 				}
2250 				kn_enter_flux(kn);
2251 				KQ_UNLOCK(kq);
2252 				knote_drop(kn, td);
2253 				KQ_LOCK(kq);
2254 			}
2255 		}
2256 	}
2257 
2258 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
2259 		kq->kq_state |= KQ_TASKDRAIN;
2260 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
2261 	}
2262 
2263 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2264 		selwakeuppri(&kq->kq_sel, PSOCK);
2265 		if (!SEL_WAITING(&kq->kq_sel))
2266 			kq->kq_state &= ~KQ_SEL;
2267 	}
2268 
2269 	KQ_UNLOCK(kq);
2270 }
2271 
2272 static void
2273 kqueue_destroy(struct kqueue *kq)
2274 {
2275 
2276 	KASSERT(kq->kq_fdp == NULL,
2277 	    ("kqueue still attached to a file descriptor"));
2278 	seldrain(&kq->kq_sel);
2279 	knlist_destroy(&kq->kq_sel.si_note);
2280 	mtx_destroy(&kq->kq_lock);
2281 
2282 	if (kq->kq_knhash != NULL)
2283 		free(kq->kq_knhash, M_KQUEUE);
2284 	if (kq->kq_knlist != NULL)
2285 		free(kq->kq_knlist, M_KQUEUE);
2286 
2287 	funsetown(&kq->kq_sigio);
2288 }
2289 
2290 /*ARGSUSED*/
2291 static int
2292 kqueue_close(struct file *fp, struct thread *td)
2293 {
2294 	struct kqueue *kq = fp->f_data;
2295 	struct filedesc *fdp;
2296 	int error;
2297 	int filedesc_unlock;
2298 
2299 	if ((error = kqueue_acquire(fp, &kq)))
2300 		return error;
2301 	kqueue_drain(kq, td);
2302 
2303 	/*
2304 	 * We could be called due to the knote_drop() doing fdrop(),
2305 	 * called from kqueue_register().  In this case the global
2306 	 * lock is owned, and filedesc sx is locked before, to not
2307 	 * take the sleepable lock after non-sleepable.
2308 	 */
2309 	fdp = kq->kq_fdp;
2310 	kq->kq_fdp = NULL;
2311 	if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
2312 		FILEDESC_XLOCK(fdp);
2313 		filedesc_unlock = 1;
2314 	} else
2315 		filedesc_unlock = 0;
2316 	TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
2317 	if (filedesc_unlock)
2318 		FILEDESC_XUNLOCK(fdp);
2319 
2320 	kqueue_destroy(kq);
2321 	chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
2322 	crfree(kq->kq_cred);
2323 	free(kq, M_KQUEUE);
2324 	fp->f_data = NULL;
2325 
2326 	return (0);
2327 }
2328 
2329 static int
2330 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2331 {
2332 	struct kqueue *kq = fp->f_data;
2333 
2334 	kif->kf_type = KF_TYPE_KQUEUE;
2335 	kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq;
2336 	kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count;
2337 	kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state;
2338 	return (0);
2339 }
2340 
2341 static void
2342 kqueue_wakeup(struct kqueue *kq)
2343 {
2344 	KQ_OWNED(kq);
2345 
2346 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
2347 		kq->kq_state &= ~KQ_SLEEP;
2348 		wakeup(kq);
2349 	}
2350 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2351 		selwakeuppri(&kq->kq_sel, PSOCK);
2352 		if (!SEL_WAITING(&kq->kq_sel))
2353 			kq->kq_state &= ~KQ_SEL;
2354 	}
2355 	if (!knlist_empty(&kq->kq_sel.si_note))
2356 		kqueue_schedtask(kq);
2357 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
2358 		pgsigio(&kq->kq_sigio, SIGIO, 0);
2359 	}
2360 }
2361 
2362 /*
2363  * Walk down a list of knotes, activating them if their event has triggered.
2364  *
2365  * There is a possibility to optimize in the case of one kq watching another.
2366  * Instead of scheduling a task to wake it up, you could pass enough state
2367  * down the chain to make up the parent kqueue.  Make this code functional
2368  * first.
2369  */
2370 void
2371 knote(struct knlist *list, long hint, int lockflags)
2372 {
2373 	struct kqueue *kq;
2374 	struct knote *kn, *tkn;
2375 	int error;
2376 
2377 	if (list == NULL)
2378 		return;
2379 
2380 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
2381 
2382 	if ((lockflags & KNF_LISTLOCKED) == 0)
2383 		list->kl_lock(list->kl_lockarg);
2384 
2385 	/*
2386 	 * If we unlock the list lock (and enter influx), we can
2387 	 * eliminate the kqueue scheduling, but this will introduce
2388 	 * four lock/unlock's for each knote to test.  Also, marker
2389 	 * would be needed to keep iteration position, since filters
2390 	 * or other threads could remove events.
2391 	 */
2392 	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
2393 		kq = kn->kn_kq;
2394 		KQ_LOCK(kq);
2395 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
2396 			/*
2397 			 * Do not process the influx notes, except for
2398 			 * the influx coming from the kq unlock in the
2399 			 * kqueue_scan().  In the later case, we do
2400 			 * not interfere with the scan, since the code
2401 			 * fragment in kqueue_scan() locks the knlist,
2402 			 * and cannot proceed until we finished.
2403 			 */
2404 			KQ_UNLOCK(kq);
2405 		} else if ((lockflags & KNF_NOKQLOCK) != 0) {
2406 			kn_enter_flux(kn);
2407 			KQ_UNLOCK(kq);
2408 			error = kn->kn_fop->f_event(kn, hint);
2409 			KQ_LOCK(kq);
2410 			kn_leave_flux(kn);
2411 			if (error)
2412 				KNOTE_ACTIVATE(kn, 1);
2413 			KQ_UNLOCK_FLUX(kq);
2414 		} else {
2415 			if (kn->kn_fop->f_event(kn, hint))
2416 				KNOTE_ACTIVATE(kn, 1);
2417 			KQ_UNLOCK(kq);
2418 		}
2419 	}
2420 	if ((lockflags & KNF_LISTLOCKED) == 0)
2421 		list->kl_unlock(list->kl_lockarg);
2422 }
2423 
2424 /*
2425  * add a knote to a knlist
2426  */
2427 void
2428 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
2429 {
2430 
2431 	KNL_ASSERT_LOCK(knl, islocked);
2432 	KQ_NOTOWNED(kn->kn_kq);
2433 	KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn));
2434 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2435 	    ("knote %p was not detached", kn));
2436 	if (!islocked)
2437 		knl->kl_lock(knl->kl_lockarg);
2438 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
2439 	if (!islocked)
2440 		knl->kl_unlock(knl->kl_lockarg);
2441 	KQ_LOCK(kn->kn_kq);
2442 	kn->kn_knlist = knl;
2443 	kn->kn_status &= ~KN_DETACHED;
2444 	KQ_UNLOCK(kn->kn_kq);
2445 }
2446 
2447 static void
2448 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
2449     int kqislocked)
2450 {
2451 
2452 	KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked"));
2453 	KNL_ASSERT_LOCK(knl, knlislocked);
2454 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
2455 	KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn));
2456 	KASSERT((kn->kn_status & KN_DETACHED) == 0,
2457 	    ("knote %p was already detached", kn));
2458 	if (!knlislocked)
2459 		knl->kl_lock(knl->kl_lockarg);
2460 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
2461 	kn->kn_knlist = NULL;
2462 	if (!knlislocked)
2463 		kn_list_unlock(knl);
2464 	if (!kqislocked)
2465 		KQ_LOCK(kn->kn_kq);
2466 	kn->kn_status |= KN_DETACHED;
2467 	if (!kqislocked)
2468 		KQ_UNLOCK(kn->kn_kq);
2469 }
2470 
2471 /*
2472  * remove knote from the specified knlist
2473  */
2474 void
2475 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
2476 {
2477 
2478 	knlist_remove_kq(knl, kn, islocked, 0);
2479 }
2480 
2481 int
2482 knlist_empty(struct knlist *knl)
2483 {
2484 
2485 	KNL_ASSERT_LOCKED(knl);
2486 	return (SLIST_EMPTY(&knl->kl_list));
2487 }
2488 
2489 static struct mtx knlist_lock;
2490 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2491     MTX_DEF);
2492 static void knlist_mtx_lock(void *arg);
2493 static void knlist_mtx_unlock(void *arg);
2494 
2495 static void
2496 knlist_mtx_lock(void *arg)
2497 {
2498 
2499 	mtx_lock((struct mtx *)arg);
2500 }
2501 
2502 static void
2503 knlist_mtx_unlock(void *arg)
2504 {
2505 
2506 	mtx_unlock((struct mtx *)arg);
2507 }
2508 
2509 static void
2510 knlist_mtx_assert_lock(void *arg, int what)
2511 {
2512 
2513 	if (what == LA_LOCKED)
2514 		mtx_assert((struct mtx *)arg, MA_OWNED);
2515 	else
2516 		mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2517 }
2518 
2519 void
2520 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2521     void (*kl_unlock)(void *),
2522     void (*kl_assert_lock)(void *, int))
2523 {
2524 
2525 	if (lock == NULL)
2526 		knl->kl_lockarg = &knlist_lock;
2527 	else
2528 		knl->kl_lockarg = lock;
2529 
2530 	if (kl_lock == NULL)
2531 		knl->kl_lock = knlist_mtx_lock;
2532 	else
2533 		knl->kl_lock = kl_lock;
2534 	if (kl_unlock == NULL)
2535 		knl->kl_unlock = knlist_mtx_unlock;
2536 	else
2537 		knl->kl_unlock = kl_unlock;
2538 	if (kl_assert_lock == NULL)
2539 		knl->kl_assert_lock = knlist_mtx_assert_lock;
2540 	else
2541 		knl->kl_assert_lock = kl_assert_lock;
2542 
2543 	knl->kl_autodestroy = 0;
2544 	SLIST_INIT(&knl->kl_list);
2545 }
2546 
2547 void
2548 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2549 {
2550 
2551 	knlist_init(knl, lock, NULL, NULL, NULL);
2552 }
2553 
2554 struct knlist *
2555 knlist_alloc(struct mtx *lock)
2556 {
2557 	struct knlist *knl;
2558 
2559 	knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
2560 	knlist_init_mtx(knl, lock);
2561 	return (knl);
2562 }
2563 
2564 void
2565 knlist_destroy(struct knlist *knl)
2566 {
2567 
2568 	KASSERT(KNLIST_EMPTY(knl),
2569 	    ("destroying knlist %p with knotes on it", knl));
2570 }
2571 
2572 void
2573 knlist_detach(struct knlist *knl)
2574 {
2575 
2576 	KNL_ASSERT_LOCKED(knl);
2577 	knl->kl_autodestroy = 1;
2578 	if (knlist_empty(knl)) {
2579 		knlist_destroy(knl);
2580 		free(knl, M_KQUEUE);
2581 	}
2582 }
2583 
2584 /*
2585  * Even if we are locked, we may need to drop the lock to allow any influx
2586  * knotes time to "settle".
2587  */
2588 void
2589 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2590 {
2591 	struct knote *kn, *kn2;
2592 	struct kqueue *kq;
2593 
2594 	KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
2595 	if (islocked)
2596 		KNL_ASSERT_LOCKED(knl);
2597 	else {
2598 		KNL_ASSERT_UNLOCKED(knl);
2599 again:		/* need to reacquire lock since we have dropped it */
2600 		knl->kl_lock(knl->kl_lockarg);
2601 	}
2602 
2603 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2604 		kq = kn->kn_kq;
2605 		KQ_LOCK(kq);
2606 		if (kn_in_flux(kn)) {
2607 			KQ_UNLOCK(kq);
2608 			continue;
2609 		}
2610 		knlist_remove_kq(knl, kn, 1, 1);
2611 		if (killkn) {
2612 			kn_enter_flux(kn);
2613 			KQ_UNLOCK(kq);
2614 			knote_drop_detached(kn, td);
2615 		} else {
2616 			/* Make sure cleared knotes disappear soon */
2617 			kn->kn_flags |= EV_EOF | EV_ONESHOT;
2618 			KQ_UNLOCK(kq);
2619 		}
2620 		kq = NULL;
2621 	}
2622 
2623 	if (!SLIST_EMPTY(&knl->kl_list)) {
2624 		/* there are still in flux knotes remaining */
2625 		kn = SLIST_FIRST(&knl->kl_list);
2626 		kq = kn->kn_kq;
2627 		KQ_LOCK(kq);
2628 		KASSERT(kn_in_flux(kn), ("knote removed w/o list lock"));
2629 		knl->kl_unlock(knl->kl_lockarg);
2630 		kq->kq_state |= KQ_FLUXWAIT;
2631 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2632 		kq = NULL;
2633 		goto again;
2634 	}
2635 
2636 	if (islocked)
2637 		KNL_ASSERT_LOCKED(knl);
2638 	else {
2639 		knl->kl_unlock(knl->kl_lockarg);
2640 		KNL_ASSERT_UNLOCKED(knl);
2641 	}
2642 }
2643 
2644 /*
2645  * Remove all knotes referencing a specified fd must be called with FILEDESC
2646  * lock.  This prevents a race where a new fd comes along and occupies the
2647  * entry and we attach a knote to the fd.
2648  */
2649 void
2650 knote_fdclose(struct thread *td, int fd)
2651 {
2652 	struct filedesc *fdp = td->td_proc->p_fd;
2653 	struct kqueue *kq;
2654 	struct knote *kn;
2655 	int influx;
2656 
2657 	FILEDESC_XLOCK_ASSERT(fdp);
2658 
2659 	/*
2660 	 * We shouldn't have to worry about new kevents appearing on fd
2661 	 * since filedesc is locked.
2662 	 */
2663 	TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2664 		KQ_LOCK(kq);
2665 
2666 again:
2667 		influx = 0;
2668 		while (kq->kq_knlistsize > fd &&
2669 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2670 			if (kn_in_flux(kn)) {
2671 				/* someone else might be waiting on our knote */
2672 				if (influx)
2673 					wakeup(kq);
2674 				kq->kq_state |= KQ_FLUXWAIT;
2675 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2676 				goto again;
2677 			}
2678 			kn_enter_flux(kn);
2679 			KQ_UNLOCK(kq);
2680 			influx = 1;
2681 			knote_drop(kn, td);
2682 			KQ_LOCK(kq);
2683 		}
2684 		KQ_UNLOCK_FLUX(kq);
2685 	}
2686 }
2687 
2688 static int
2689 knote_attach(struct knote *kn, struct kqueue *kq)
2690 {
2691 	struct klist *list;
2692 
2693 	KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn));
2694 	KQ_OWNED(kq);
2695 
2696 	if ((kq->kq_state & KQ_CLOSING) != 0)
2697 		return (EBADF);
2698 	if (kn->kn_fop->f_isfd) {
2699 		if (kn->kn_id >= kq->kq_knlistsize)
2700 			return (ENOMEM);
2701 		list = &kq->kq_knlist[kn->kn_id];
2702 	} else {
2703 		if (kq->kq_knhash == NULL)
2704 			return (ENOMEM);
2705 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2706 	}
2707 	SLIST_INSERT_HEAD(list, kn, kn_link);
2708 	return (0);
2709 }
2710 
2711 static void
2712 knote_drop(struct knote *kn, struct thread *td)
2713 {
2714 
2715 	if ((kn->kn_status & KN_DETACHED) == 0)
2716 		kn->kn_fop->f_detach(kn);
2717 	knote_drop_detached(kn, td);
2718 }
2719 
2720 static void
2721 knote_drop_detached(struct knote *kn, struct thread *td)
2722 {
2723 	struct kqueue *kq;
2724 	struct klist *list;
2725 
2726 	kq = kn->kn_kq;
2727 
2728 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2729 	    ("knote %p still attached", kn));
2730 	KQ_NOTOWNED(kq);
2731 
2732 	KQ_LOCK(kq);
2733 	KASSERT(kn->kn_influx == 1,
2734 	    ("knote_drop called on %p with influx %d", kn, kn->kn_influx));
2735 
2736 	if (kn->kn_fop->f_isfd)
2737 		list = &kq->kq_knlist[kn->kn_id];
2738 	else
2739 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2740 
2741 	if (!SLIST_EMPTY(list))
2742 		SLIST_REMOVE(list, kn, knote, kn_link);
2743 	if (kn->kn_status & KN_QUEUED)
2744 		knote_dequeue(kn);
2745 	KQ_UNLOCK_FLUX(kq);
2746 
2747 	if (kn->kn_fop->f_isfd) {
2748 		fdrop(kn->kn_fp, td);
2749 		kn->kn_fp = NULL;
2750 	}
2751 	kqueue_fo_release(kn->kn_kevent.filter);
2752 	kn->kn_fop = NULL;
2753 	knote_free(kn);
2754 }
2755 
2756 static void
2757 knote_enqueue(struct knote *kn)
2758 {
2759 	struct kqueue *kq = kn->kn_kq;
2760 
2761 	KQ_OWNED(kn->kn_kq);
2762 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2763 
2764 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2765 	kn->kn_status |= KN_QUEUED;
2766 	kq->kq_count++;
2767 	kqueue_wakeup(kq);
2768 }
2769 
2770 static void
2771 knote_dequeue(struct knote *kn)
2772 {
2773 	struct kqueue *kq = kn->kn_kq;
2774 
2775 	KQ_OWNED(kn->kn_kq);
2776 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2777 
2778 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2779 	kn->kn_status &= ~KN_QUEUED;
2780 	kq->kq_count--;
2781 }
2782 
2783 static void
2784 knote_init(void)
2785 {
2786 
2787 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2788 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2789 	ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue);
2790 }
2791 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2792 
2793 static struct knote *
2794 knote_alloc(int mflag)
2795 {
2796 
2797 	return (uma_zalloc(knote_zone, mflag | M_ZERO));
2798 }
2799 
2800 static void
2801 knote_free(struct knote *kn)
2802 {
2803 
2804 	uma_zfree(knote_zone, kn);
2805 }
2806 
2807 /*
2808  * Register the kev w/ the kq specified by fd.
2809  */
2810 int
2811 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag)
2812 {
2813 	struct kqueue *kq;
2814 	struct file *fp;
2815 	cap_rights_t rights;
2816 	int error;
2817 
2818 	error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE),
2819 	    &fp);
2820 	if (error != 0)
2821 		return (error);
2822 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2823 		goto noacquire;
2824 
2825 	error = kqueue_register(kq, kev, td, mflag);
2826 	kqueue_release(kq, 0);
2827 
2828 noacquire:
2829 	fdrop(fp, td);
2830 	return (error);
2831 }
2832