xref: /freebsd/sys/kern/kern_event.c (revision 61ba55bcf70f2340f9c943c9571113b3fd8eda69)
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_fill_kinfo = kqueue_fill_kinfo,
143 };
144 
145 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
146 static void 	knote_drop(struct knote *kn, struct thread *td);
147 static void 	knote_drop_detached(struct knote *kn, struct thread *td);
148 static void 	knote_enqueue(struct knote *kn);
149 static void 	knote_dequeue(struct knote *kn);
150 static void 	knote_init(void);
151 static struct 	knote *knote_alloc(int mflag);
152 static void 	knote_free(struct knote *kn);
153 
154 static void	filt_kqdetach(struct knote *kn);
155 static int	filt_kqueue(struct knote *kn, long hint);
156 static int	filt_procattach(struct knote *kn);
157 static void	filt_procdetach(struct knote *kn);
158 static int	filt_proc(struct knote *kn, long hint);
159 static int	filt_fileattach(struct knote *kn);
160 static void	filt_timerexpire(void *knx);
161 static void	filt_timerexpire_l(struct knote *kn, bool proc_locked);
162 static int	filt_timerattach(struct knote *kn);
163 static void	filt_timerdetach(struct knote *kn);
164 static void	filt_timerstart(struct knote *kn, sbintime_t to);
165 static void	filt_timertouch(struct knote *kn, struct kevent *kev,
166 		    u_long type);
167 static int	filt_timervalidate(struct knote *kn, sbintime_t *to);
168 static int	filt_timer(struct knote *kn, long hint);
169 static int	filt_userattach(struct knote *kn);
170 static void	filt_userdetach(struct knote *kn);
171 static int	filt_user(struct knote *kn, long hint);
172 static void	filt_usertouch(struct knote *kn, struct kevent *kev,
173 		    u_long type);
174 
175 static struct filterops file_filtops = {
176 	.f_isfd = 1,
177 	.f_attach = filt_fileattach,
178 };
179 static struct filterops kqread_filtops = {
180 	.f_isfd = 1,
181 	.f_detach = filt_kqdetach,
182 	.f_event = filt_kqueue,
183 };
184 /* XXX - move to kern_proc.c?  */
185 static struct filterops proc_filtops = {
186 	.f_isfd = 0,
187 	.f_attach = filt_procattach,
188 	.f_detach = filt_procdetach,
189 	.f_event = filt_proc,
190 };
191 static struct filterops timer_filtops = {
192 	.f_isfd = 0,
193 	.f_attach = filt_timerattach,
194 	.f_detach = filt_timerdetach,
195 	.f_event = filt_timer,
196 	.f_touch = filt_timertouch,
197 };
198 static struct filterops user_filtops = {
199 	.f_attach = filt_userattach,
200 	.f_detach = filt_userdetach,
201 	.f_event = filt_user,
202 	.f_touch = filt_usertouch,
203 };
204 
205 static uma_zone_t	knote_zone;
206 static unsigned int __exclusive_cache_line	kq_ncallouts;
207 static unsigned int 	kq_calloutmax = 4 * 1024;
208 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
209     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
210 
211 /* XXX - ensure not influx ? */
212 #define KNOTE_ACTIVATE(kn, islock) do { 				\
213 	if ((islock))							\
214 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
215 	else								\
216 		KQ_LOCK((kn)->kn_kq);					\
217 	(kn)->kn_status |= KN_ACTIVE;					\
218 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
219 		knote_enqueue((kn));					\
220 	if (!(islock))							\
221 		KQ_UNLOCK((kn)->kn_kq);					\
222 } while (0)
223 #define KQ_LOCK(kq) do {						\
224 	mtx_lock(&(kq)->kq_lock);					\
225 } while (0)
226 #define KQ_FLUX_WAKEUP(kq) do {						\
227 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
228 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
229 		wakeup((kq));						\
230 	}								\
231 } while (0)
232 #define KQ_UNLOCK_FLUX(kq) do {						\
233 	KQ_FLUX_WAKEUP(kq);						\
234 	mtx_unlock(&(kq)->kq_lock);					\
235 } while (0)
236 #define KQ_UNLOCK(kq) do {						\
237 	mtx_unlock(&(kq)->kq_lock);					\
238 } while (0)
239 #define KQ_OWNED(kq) do {						\
240 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
241 } while (0)
242 #define KQ_NOTOWNED(kq) do {						\
243 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
244 } while (0)
245 
246 static struct knlist *
247 kn_list_lock(struct knote *kn)
248 {
249 	struct knlist *knl;
250 
251 	knl = kn->kn_knlist;
252 	if (knl != NULL)
253 		knl->kl_lock(knl->kl_lockarg);
254 	return (knl);
255 }
256 
257 static void
258 kn_list_unlock(struct knlist *knl)
259 {
260 	bool do_free;
261 
262 	if (knl == NULL)
263 		return;
264 	do_free = knl->kl_autodestroy && knlist_empty(knl);
265 	knl->kl_unlock(knl->kl_lockarg);
266 	if (do_free) {
267 		knlist_destroy(knl);
268 		free(knl, M_KQUEUE);
269 	}
270 }
271 
272 static bool
273 kn_in_flux(struct knote *kn)
274 {
275 
276 	return (kn->kn_influx > 0);
277 }
278 
279 static void
280 kn_enter_flux(struct knote *kn)
281 {
282 
283 	KQ_OWNED(kn->kn_kq);
284 	MPASS(kn->kn_influx < INT_MAX);
285 	kn->kn_influx++;
286 }
287 
288 static bool
289 kn_leave_flux(struct knote *kn)
290 {
291 
292 	KQ_OWNED(kn->kn_kq);
293 	MPASS(kn->kn_influx > 0);
294 	kn->kn_influx--;
295 	return (kn->kn_influx == 0);
296 }
297 
298 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
299 	if (islocked)							\
300 		KNL_ASSERT_LOCKED(knl);				\
301 	else								\
302 		KNL_ASSERT_UNLOCKED(knl);				\
303 } while (0)
304 #ifdef INVARIANTS
305 #define	KNL_ASSERT_LOCKED(knl) do {					\
306 	knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED);		\
307 } while (0)
308 #define	KNL_ASSERT_UNLOCKED(knl) do {					\
309 	knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED);		\
310 } while (0)
311 #else /* !INVARIANTS */
312 #define	KNL_ASSERT_LOCKED(knl) do {} while (0)
313 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
314 #endif /* INVARIANTS */
315 
316 #ifndef	KN_HASHSIZE
317 #define	KN_HASHSIZE		64		/* XXX should be tunable */
318 #endif
319 
320 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
321 
322 static int
323 filt_nullattach(struct knote *kn)
324 {
325 
326 	return (ENXIO);
327 };
328 
329 struct filterops null_filtops = {
330 	.f_isfd = 0,
331 	.f_attach = filt_nullattach,
332 };
333 
334 /* XXX - make SYSINIT to add these, and move into respective modules. */
335 extern struct filterops sig_filtops;
336 extern struct filterops fs_filtops;
337 
338 /*
339  * Table for all system-defined filters.
340  */
341 static struct mtx	filterops_lock;
342 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF);
343 static struct {
344 	const struct filterops *for_fop;
345 	int for_nolock;
346 	int for_refcnt;
347 } sysfilt_ops[EVFILT_SYSCOUNT] = {
348 	{ &file_filtops, 1 },			/* EVFILT_READ */
349 	{ &file_filtops, 1 },			/* EVFILT_WRITE */
350 	{ &null_filtops },			/* EVFILT_AIO */
351 	{ &file_filtops, 1 },			/* EVFILT_VNODE */
352 	{ &proc_filtops, 1 },			/* EVFILT_PROC */
353 	{ &sig_filtops, 1 },			/* EVFILT_SIGNAL */
354 	{ &timer_filtops, 1 },			/* EVFILT_TIMER */
355 	{ &file_filtops, 1 },			/* EVFILT_PROCDESC */
356 	{ &fs_filtops, 1 },			/* EVFILT_FS */
357 	{ &null_filtops },			/* EVFILT_LIO */
358 	{ &user_filtops, 1 },			/* EVFILT_USER */
359 	{ &null_filtops },			/* EVFILT_SENDFILE */
360 	{ &file_filtops, 1 },                   /* EVFILT_EMPTY */
361 };
362 
363 /*
364  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
365  * method.
366  */
367 static int
368 filt_fileattach(struct knote *kn)
369 {
370 
371 	return (fo_kqfilter(kn->kn_fp, kn));
372 }
373 
374 /*ARGSUSED*/
375 static int
376 kqueue_kqfilter(struct file *fp, struct knote *kn)
377 {
378 	struct kqueue *kq = kn->kn_fp->f_data;
379 
380 	if (kn->kn_filter != EVFILT_READ)
381 		return (EINVAL);
382 
383 	kn->kn_status |= KN_KQUEUE;
384 	kn->kn_fop = &kqread_filtops;
385 	knlist_add(&kq->kq_sel.si_note, kn, 0);
386 
387 	return (0);
388 }
389 
390 static void
391 filt_kqdetach(struct knote *kn)
392 {
393 	struct kqueue *kq = kn->kn_fp->f_data;
394 
395 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
396 }
397 
398 /*ARGSUSED*/
399 static int
400 filt_kqueue(struct knote *kn, long hint)
401 {
402 	struct kqueue *kq = kn->kn_fp->f_data;
403 
404 	kn->kn_data = kq->kq_count;
405 	return (kn->kn_data > 0);
406 }
407 
408 /* XXX - move to kern_proc.c?  */
409 static int
410 filt_procattach(struct knote *kn)
411 {
412 	struct proc *p;
413 	int error;
414 	bool exiting, immediate;
415 
416 	exiting = immediate = false;
417 	if (kn->kn_sfflags & NOTE_EXIT)
418 		p = pfind_any(kn->kn_id);
419 	else
420 		p = pfind(kn->kn_id);
421 	if (p == NULL)
422 		return (ESRCH);
423 	if (p->p_flag & P_WEXIT)
424 		exiting = true;
425 
426 	if ((error = p_cansee(curthread, p))) {
427 		PROC_UNLOCK(p);
428 		return (error);
429 	}
430 
431 	kn->kn_ptr.p_proc = p;
432 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
433 
434 	/*
435 	 * Internal flag indicating registration done by kernel for the
436 	 * purposes of getting a NOTE_CHILD notification.
437 	 */
438 	if (kn->kn_flags & EV_FLAG2) {
439 		kn->kn_flags &= ~EV_FLAG2;
440 		kn->kn_data = kn->kn_sdata;		/* ppid */
441 		kn->kn_fflags = NOTE_CHILD;
442 		kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
443 		immediate = true; /* Force immediate activation of child note. */
444 	}
445 	/*
446 	 * Internal flag indicating registration done by kernel (for other than
447 	 * NOTE_CHILD).
448 	 */
449 	if (kn->kn_flags & EV_FLAG1) {
450 		kn->kn_flags &= ~EV_FLAG1;
451 	}
452 
453 	knlist_add(p->p_klist, kn, 1);
454 
455 	/*
456 	 * Immediately activate any child notes or, in the case of a zombie
457 	 * target process, exit notes.  The latter is necessary to handle the
458 	 * case where the target process, e.g. a child, dies before the kevent
459 	 * is registered.
460 	 */
461 	if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
462 		KNOTE_ACTIVATE(kn, 0);
463 
464 	PROC_UNLOCK(p);
465 
466 	return (0);
467 }
468 
469 /*
470  * The knote may be attached to a different process, which may exit,
471  * leaving nothing for the knote to be attached to.  So when the process
472  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
473  * it will be deleted when read out.  However, as part of the knote deletion,
474  * this routine is called, so a check is needed to avoid actually performing
475  * a detach, because the original process does not exist any more.
476  */
477 /* XXX - move to kern_proc.c?  */
478 static void
479 filt_procdetach(struct knote *kn)
480 {
481 
482 	knlist_remove(kn->kn_knlist, kn, 0);
483 	kn->kn_ptr.p_proc = NULL;
484 }
485 
486 /* XXX - move to kern_proc.c?  */
487 static int
488 filt_proc(struct knote *kn, long hint)
489 {
490 	struct proc *p;
491 	u_int event;
492 
493 	p = kn->kn_ptr.p_proc;
494 	if (p == NULL) /* already activated, from attach filter */
495 		return (0);
496 
497 	/* Mask off extra data. */
498 	event = (u_int)hint & NOTE_PCTRLMASK;
499 
500 	/* If the user is interested in this event, record it. */
501 	if (kn->kn_sfflags & event)
502 		kn->kn_fflags |= event;
503 
504 	/* Process is gone, so flag the event as finished. */
505 	if (event == NOTE_EXIT) {
506 		kn->kn_flags |= EV_EOF | EV_ONESHOT;
507 		kn->kn_ptr.p_proc = NULL;
508 		if (kn->kn_fflags & NOTE_EXIT)
509 			kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
510 		if (kn->kn_fflags == 0)
511 			kn->kn_flags |= EV_DROP;
512 		return (1);
513 	}
514 
515 	return (kn->kn_fflags != 0);
516 }
517 
518 /*
519  * Called when the process forked. It mostly does the same as the
520  * knote(), activating all knotes registered to be activated when the
521  * process forked. Additionally, for each knote attached to the
522  * parent, check whether user wants to track the new process. If so
523  * attach a new knote to it, and immediately report an event with the
524  * child's pid.
525  */
526 void
527 knote_fork(struct knlist *list, int pid)
528 {
529 	struct kqueue *kq;
530 	struct knote *kn;
531 	struct kevent kev;
532 	int error;
533 
534 	MPASS(list != NULL);
535 	KNL_ASSERT_LOCKED(list);
536 	if (SLIST_EMPTY(&list->kl_list))
537 		return;
538 
539 	memset(&kev, 0, sizeof(kev));
540 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
541 		kq = kn->kn_kq;
542 		KQ_LOCK(kq);
543 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
544 			KQ_UNLOCK(kq);
545 			continue;
546 		}
547 
548 		/*
549 		 * The same as knote(), activate the event.
550 		 */
551 		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
552 			if (kn->kn_fop->f_event(kn, NOTE_FORK))
553 				KNOTE_ACTIVATE(kn, 1);
554 			KQ_UNLOCK(kq);
555 			continue;
556 		}
557 
558 		/*
559 		 * The NOTE_TRACK case. In addition to the activation
560 		 * of the event, we need to register new events to
561 		 * track the child. Drop the locks in preparation for
562 		 * the call to kqueue_register().
563 		 */
564 		kn_enter_flux(kn);
565 		KQ_UNLOCK(kq);
566 		list->kl_unlock(list->kl_lockarg);
567 
568 		/*
569 		 * Activate existing knote and register tracking knotes with
570 		 * new process.
571 		 *
572 		 * First register a knote to get just the child notice. This
573 		 * must be a separate note from a potential NOTE_EXIT
574 		 * notification since both NOTE_CHILD and NOTE_EXIT are defined
575 		 * to use the data field (in conflicting ways).
576 		 */
577 		kev.ident = pid;
578 		kev.filter = kn->kn_filter;
579 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
580 		    EV_FLAG2;
581 		kev.fflags = kn->kn_sfflags;
582 		kev.data = kn->kn_id;		/* parent */
583 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
584 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
585 		if (error)
586 			kn->kn_fflags |= NOTE_TRACKERR;
587 
588 		/*
589 		 * Then register another knote to track other potential events
590 		 * from the new process.
591 		 */
592 		kev.ident = pid;
593 		kev.filter = kn->kn_filter;
594 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
595 		kev.fflags = kn->kn_sfflags;
596 		kev.data = kn->kn_id;		/* parent */
597 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
598 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
599 		if (error)
600 			kn->kn_fflags |= NOTE_TRACKERR;
601 		if (kn->kn_fop->f_event(kn, NOTE_FORK))
602 			KNOTE_ACTIVATE(kn, 0);
603 		list->kl_lock(list->kl_lockarg);
604 		KQ_LOCK(kq);
605 		kn_leave_flux(kn);
606 		KQ_UNLOCK_FLUX(kq);
607 	}
608 }
609 
610 /*
611  * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
612  * interval timer support code.
613  */
614 
615 #define NOTE_TIMER_PRECMASK						\
616     (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)
617 
618 static sbintime_t
619 timer2sbintime(int64_t data, int flags)
620 {
621 	int64_t secs;
622 
623         /*
624          * Macros for converting to the fractional second portion of an
625          * sbintime_t using 64bit multiplication to improve precision.
626          */
627 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
628 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
629 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
630 	switch (flags & NOTE_TIMER_PRECMASK) {
631 	case NOTE_SECONDS:
632 #ifdef __LP64__
633 		if (data > (SBT_MAX / SBT_1S))
634 			return (SBT_MAX);
635 #endif
636 		return ((sbintime_t)data << 32);
637 	case NOTE_MSECONDS: /* FALLTHROUGH */
638 	case 0:
639 		if (data >= 1000) {
640 			secs = data / 1000;
641 #ifdef __LP64__
642 			if (secs > (SBT_MAX / SBT_1S))
643 				return (SBT_MAX);
644 #endif
645 			return (secs << 32 | MS_TO_SBT(data % 1000));
646 		}
647 		return (MS_TO_SBT(data));
648 	case NOTE_USECONDS:
649 		if (data >= 1000000) {
650 			secs = data / 1000000;
651 #ifdef __LP64__
652 			if (secs > (SBT_MAX / SBT_1S))
653 				return (SBT_MAX);
654 #endif
655 			return (secs << 32 | US_TO_SBT(data % 1000000));
656 		}
657 		return (US_TO_SBT(data));
658 	case NOTE_NSECONDS:
659 		if (data >= 1000000000) {
660 			secs = data / 1000000000;
661 #ifdef __LP64__
662 			if (secs > (SBT_MAX / SBT_1S))
663 				return (SBT_MAX);
664 #endif
665 			return (secs << 32 | NS_TO_SBT(data % 1000000000));
666 		}
667 		return (NS_TO_SBT(data));
668 	default:
669 		break;
670 	}
671 	return (-1);
672 }
673 
674 struct kq_timer_cb_data {
675 	struct callout c;
676 	struct proc *p;
677 	struct knote *kn;
678 	int cpuid;
679 	int flags;
680 	TAILQ_ENTRY(kq_timer_cb_data) link;
681 	sbintime_t next;	/* next timer event fires at */
682 	sbintime_t to;		/* precalculated timer period, 0 for abs */
683 };
684 
685 #define	KQ_TIMER_CB_ENQUEUED	0x01
686 
687 static void
688 kqtimer_sched_callout(struct kq_timer_cb_data *kc)
689 {
690 	callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn,
691 	    kc->cpuid, C_ABSOLUTE);
692 }
693 
694 void
695 kqtimer_proc_continue(struct proc *p)
696 {
697 	struct kq_timer_cb_data *kc, *kc1;
698 	struct bintime bt;
699 	sbintime_t now;
700 
701 	PROC_LOCK_ASSERT(p, MA_OWNED);
702 
703 	getboottimebin(&bt);
704 	now = bttosbt(bt);
705 
706 	TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) {
707 		TAILQ_REMOVE(&p->p_kqtim_stop, kc, link);
708 		kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
709 		if (kc->next <= now)
710 			filt_timerexpire_l(kc->kn, true);
711 		else
712 			kqtimer_sched_callout(kc);
713 	}
714 }
715 
716 static void
717 filt_timerexpire_l(struct knote *kn, bool proc_locked)
718 {
719 	struct kq_timer_cb_data *kc;
720 	struct proc *p;
721 	uint64_t delta;
722 	sbintime_t now;
723 
724 	kc = kn->kn_ptr.p_v;
725 
726 	if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) {
727 		kn->kn_data++;
728 		KNOTE_ACTIVATE(kn, 0);
729 		return;
730 	}
731 
732 	now = sbinuptime();
733 	if (now >= kc->next) {
734 		delta = (now - kc->next) / kc->to;
735 		if (delta == 0)
736 			delta = 1;
737 		kn->kn_data += delta;
738 		kc->next += delta * kc->to;
739 		if (now >= kc->next)	/* overflow */
740 			kc->next = now + kc->to;
741 		KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
742 	}
743 
744 	/*
745 	 * Initial check for stopped kc->p is racy.  It is fine to
746 	 * miss the set of the stop flags, at worst we would schedule
747 	 * one more callout.  On the other hand, it is not fine to not
748 	 * schedule when we we missed clearing of the flags, we
749 	 * recheck them under the lock and observe consistent state.
750 	 */
751 	p = kc->p;
752 	if (P_SHOULDSTOP(p) || P_KILLED(p)) {
753 		if (!proc_locked)
754 			PROC_LOCK(p);
755 		if (P_SHOULDSTOP(p) || P_KILLED(p)) {
756 			if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) {
757 				kc->flags |= KQ_TIMER_CB_ENQUEUED;
758 				TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link);
759 			}
760 			if (!proc_locked)
761 				PROC_UNLOCK(p);
762 			return;
763 		}
764 		if (!proc_locked)
765 			PROC_UNLOCK(p);
766 	}
767 	kqtimer_sched_callout(kc);
768 }
769 
770 static void
771 filt_timerexpire(void *knx)
772 {
773 	filt_timerexpire_l(knx, false);
774 }
775 
776 /*
777  * data contains amount of time to sleep
778  */
779 static int
780 filt_timervalidate(struct knote *kn, sbintime_t *to)
781 {
782 	struct bintime bt;
783 	sbintime_t sbt;
784 
785 	if (kn->kn_sdata < 0)
786 		return (EINVAL);
787 	if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
788 		kn->kn_sdata = 1;
789 	/*
790 	 * The only fflags values supported are the timer unit
791 	 * (precision) and the absolute time indicator.
792 	 */
793 	if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0)
794 		return (EINVAL);
795 
796 	*to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
797 	if (*to < 0)
798 		return (EINVAL);
799 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
800 		getboottimebin(&bt);
801 		sbt = bttosbt(bt);
802 		*to = MAX(0, *to - sbt);
803 	}
804 	return (0);
805 }
806 
807 static int
808 filt_timerattach(struct knote *kn)
809 {
810 	struct kq_timer_cb_data *kc;
811 	sbintime_t to;
812 	int error;
813 
814 	to = -1;
815 	error = filt_timervalidate(kn, &to);
816 	if (error != 0)
817 		return (error);
818 	KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 ||
819 	    (kn->kn_sfflags & NOTE_ABSTIME) != 0,
820 	    ("%s: periodic timer has a calculated zero timeout", __func__));
821 	KASSERT(to >= 0,
822 	    ("%s: timer has a calculated negative timeout", __func__));
823 
824 	if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) {
825 		atomic_subtract_int(&kq_ncallouts, 1);
826 		return (ENOMEM);
827 	}
828 
829 	if ((kn->kn_sfflags & NOTE_ABSTIME) == 0)
830 		kn->kn_flags |= EV_CLEAR;	/* automatically set */
831 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
832 	kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
833 	kc->kn = kn;
834 	kc->p = curproc;
835 	kc->cpuid = PCPU_GET(cpuid);
836 	kc->flags = 0;
837 	callout_init(&kc->c, 1);
838 	filt_timerstart(kn, to);
839 
840 	return (0);
841 }
842 
843 static void
844 filt_timerstart(struct knote *kn, sbintime_t to)
845 {
846 	struct kq_timer_cb_data *kc;
847 
848 	kc = kn->kn_ptr.p_v;
849 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
850 		kc->next = to;
851 		kc->to = 0;
852 	} else {
853 		kc->next = to + sbinuptime();
854 		kc->to = to;
855 	}
856 	kqtimer_sched_callout(kc);
857 }
858 
859 static void
860 filt_timerdetach(struct knote *kn)
861 {
862 	struct kq_timer_cb_data *kc;
863 	unsigned int old __unused;
864 	bool pending;
865 
866 	kc = kn->kn_ptr.p_v;
867 	do {
868 		callout_drain(&kc->c);
869 
870 		/*
871 		 * kqtimer_proc_continue() might have rescheduled this callout.
872 		 * Double-check, using the process mutex as an interlock.
873 		 */
874 		PROC_LOCK(kc->p);
875 		if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) {
876 			kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
877 			TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link);
878 		}
879 		pending = callout_pending(&kc->c);
880 		PROC_UNLOCK(kc->p);
881 	} while (pending);
882 	free(kc, M_KQUEUE);
883 	old = atomic_fetchadd_int(&kq_ncallouts, -1);
884 	KASSERT(old > 0, ("Number of callouts cannot become negative"));
885 	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
886 }
887 
888 static void
889 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type)
890 {
891 	struct kq_timer_cb_data *kc;
892 	struct kqueue *kq;
893 	sbintime_t to;
894 	int error;
895 
896 	switch (type) {
897 	case EVENT_REGISTER:
898 		/* Handle re-added timers that update data/fflags */
899 		if (kev->flags & EV_ADD) {
900 			kc = kn->kn_ptr.p_v;
901 
902 			/* Drain any existing callout. */
903 			callout_drain(&kc->c);
904 
905 			/* Throw away any existing undelivered record
906 			 * of the timer expiration. This is done under
907 			 * the presumption that if a process is
908 			 * re-adding this timer with new parameters,
909 			 * it is no longer interested in what may have
910 			 * happened under the old parameters. If it is
911 			 * interested, it can wait for the expiration,
912 			 * delete the old timer definition, and then
913 			 * add the new one.
914 			 *
915 			 * This has to be done while the kq is locked:
916 			 *   - if enqueued, dequeue
917 			 *   - make it no longer active
918 			 *   - clear the count of expiration events
919 			 */
920 			kq = kn->kn_kq;
921 			KQ_LOCK(kq);
922 			if (kn->kn_status & KN_QUEUED)
923 				knote_dequeue(kn);
924 
925 			kn->kn_status &= ~KN_ACTIVE;
926 			kn->kn_data = 0;
927 			KQ_UNLOCK(kq);
928 
929 			/* Reschedule timer based on new data/fflags */
930 			kn->kn_sfflags = kev->fflags;
931 			kn->kn_sdata = kev->data;
932 			error = filt_timervalidate(kn, &to);
933 			if (error != 0) {
934 			  	kn->kn_flags |= EV_ERROR;
935 				kn->kn_data = error;
936 			} else
937 			  	filt_timerstart(kn, to);
938 		}
939 		break;
940 
941         case EVENT_PROCESS:
942 		*kev = kn->kn_kevent;
943 		if (kn->kn_flags & EV_CLEAR) {
944 			kn->kn_data = 0;
945 			kn->kn_fflags = 0;
946 		}
947 		break;
948 
949 	default:
950 		panic("filt_timertouch() - invalid type (%ld)", type);
951 		break;
952 	}
953 }
954 
955 static int
956 filt_timer(struct knote *kn, long hint)
957 {
958 
959 	return (kn->kn_data != 0);
960 }
961 
962 static int
963 filt_userattach(struct knote *kn)
964 {
965 
966 	/*
967 	 * EVFILT_USER knotes are not attached to anything in the kernel.
968 	 */
969 	kn->kn_hook = NULL;
970 	if (kn->kn_fflags & NOTE_TRIGGER)
971 		kn->kn_hookid = 1;
972 	else
973 		kn->kn_hookid = 0;
974 	return (0);
975 }
976 
977 static void
978 filt_userdetach(__unused struct knote *kn)
979 {
980 
981 	/*
982 	 * EVFILT_USER knotes are not attached to anything in the kernel.
983 	 */
984 }
985 
986 static int
987 filt_user(struct knote *kn, __unused long hint)
988 {
989 
990 	return (kn->kn_hookid);
991 }
992 
993 static void
994 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
995 {
996 	u_int ffctrl;
997 
998 	switch (type) {
999 	case EVENT_REGISTER:
1000 		if (kev->fflags & NOTE_TRIGGER)
1001 			kn->kn_hookid = 1;
1002 
1003 		ffctrl = kev->fflags & NOTE_FFCTRLMASK;
1004 		kev->fflags &= NOTE_FFLAGSMASK;
1005 		switch (ffctrl) {
1006 		case NOTE_FFNOP:
1007 			break;
1008 
1009 		case NOTE_FFAND:
1010 			kn->kn_sfflags &= kev->fflags;
1011 			break;
1012 
1013 		case NOTE_FFOR:
1014 			kn->kn_sfflags |= kev->fflags;
1015 			break;
1016 
1017 		case NOTE_FFCOPY:
1018 			kn->kn_sfflags = kev->fflags;
1019 			break;
1020 
1021 		default:
1022 			/* XXX Return error? */
1023 			break;
1024 		}
1025 		kn->kn_sdata = kev->data;
1026 		if (kev->flags & EV_CLEAR) {
1027 			kn->kn_hookid = 0;
1028 			kn->kn_data = 0;
1029 			kn->kn_fflags = 0;
1030 		}
1031 		break;
1032 
1033         case EVENT_PROCESS:
1034 		*kev = kn->kn_kevent;
1035 		kev->fflags = kn->kn_sfflags;
1036 		kev->data = kn->kn_sdata;
1037 		if (kn->kn_flags & EV_CLEAR) {
1038 			kn->kn_hookid = 0;
1039 			kn->kn_data = 0;
1040 			kn->kn_fflags = 0;
1041 		}
1042 		break;
1043 
1044 	default:
1045 		panic("filt_usertouch() - invalid type (%ld)", type);
1046 		break;
1047 	}
1048 }
1049 
1050 int
1051 sys_kqueue(struct thread *td, struct kqueue_args *uap)
1052 {
1053 
1054 	return (kern_kqueue(td, 0, NULL));
1055 }
1056 
1057 int
1058 sys_kqueuex(struct thread *td, struct kqueuex_args *uap)
1059 {
1060 	int flags;
1061 
1062 	if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0)
1063 		return (EINVAL);
1064 	flags = 0;
1065 	if ((uap->flags & KQUEUE_CLOEXEC) != 0)
1066 		flags |= O_CLOEXEC;
1067 	return (kern_kqueue(td, flags, NULL));
1068 }
1069 
1070 static void
1071 kqueue_init(struct kqueue *kq)
1072 {
1073 
1074 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
1075 	TAILQ_INIT(&kq->kq_head);
1076 	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
1077 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
1078 }
1079 
1080 int
1081 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps)
1082 {
1083 	struct filedesc *fdp;
1084 	struct kqueue *kq;
1085 	struct file *fp;
1086 	struct ucred *cred;
1087 	int fd, error;
1088 
1089 	fdp = td->td_proc->p_fd;
1090 	cred = td->td_ucred;
1091 	if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
1092 		return (ENOMEM);
1093 
1094 	error = falloc_caps(td, &fp, &fd, flags, fcaps);
1095 	if (error != 0) {
1096 		chgkqcnt(cred->cr_ruidinfo, -1, 0);
1097 		return (error);
1098 	}
1099 
1100 	/* An extra reference on `fp' has been held for us by falloc(). */
1101 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
1102 	kqueue_init(kq);
1103 	kq->kq_fdp = fdp;
1104 	kq->kq_cred = crhold(cred);
1105 
1106 	FILEDESC_XLOCK(fdp);
1107 	TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
1108 	FILEDESC_XUNLOCK(fdp);
1109 
1110 	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
1111 	fdrop(fp, td);
1112 
1113 	td->td_retval[0] = fd;
1114 	return (0);
1115 }
1116 
1117 struct g_kevent_args {
1118 	int	fd;
1119 	const void *changelist;
1120 	int	nchanges;
1121 	void	*eventlist;
1122 	int	nevents;
1123 	const struct timespec *timeout;
1124 };
1125 
1126 int
1127 sys_kevent(struct thread *td, struct kevent_args *uap)
1128 {
1129 	struct kevent_copyops k_ops = {
1130 		.arg = uap,
1131 		.k_copyout = kevent_copyout,
1132 		.k_copyin = kevent_copyin,
1133 		.kevent_size = sizeof(struct kevent),
1134 	};
1135 	struct g_kevent_args gk_args = {
1136 		.fd = uap->fd,
1137 		.changelist = uap->changelist,
1138 		.nchanges = uap->nchanges,
1139 		.eventlist = uap->eventlist,
1140 		.nevents = uap->nevents,
1141 		.timeout = uap->timeout,
1142 	};
1143 
1144 	return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent"));
1145 }
1146 
1147 static int
1148 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap,
1149     struct kevent_copyops *k_ops, const char *struct_name)
1150 {
1151 	struct timespec ts, *tsp;
1152 #ifdef KTRACE
1153 	struct kevent *eventlist = uap->eventlist;
1154 #endif
1155 	int error;
1156 
1157 	if (uap->timeout != NULL) {
1158 		error = copyin(uap->timeout, &ts, sizeof(ts));
1159 		if (error)
1160 			return (error);
1161 		tsp = &ts;
1162 	} else
1163 		tsp = NULL;
1164 
1165 #ifdef KTRACE
1166 	if (KTRPOINT(td, KTR_STRUCT_ARRAY))
1167 		ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist,
1168 		    uap->nchanges, k_ops->kevent_size);
1169 #endif
1170 
1171 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
1172 	    k_ops, tsp);
1173 
1174 #ifdef KTRACE
1175 	if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
1176 		ktrstructarray(struct_name, UIO_USERSPACE, eventlist,
1177 		    td->td_retval[0], k_ops->kevent_size);
1178 #endif
1179 
1180 	return (error);
1181 }
1182 
1183 /*
1184  * Copy 'count' items into the destination list pointed to by uap->eventlist.
1185  */
1186 static int
1187 kevent_copyout(void *arg, struct kevent *kevp, int count)
1188 {
1189 	struct kevent_args *uap;
1190 	int error;
1191 
1192 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1193 	uap = (struct kevent_args *)arg;
1194 
1195 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
1196 	if (error == 0)
1197 		uap->eventlist += count;
1198 	return (error);
1199 }
1200 
1201 /*
1202  * Copy 'count' items from the list pointed to by uap->changelist.
1203  */
1204 static int
1205 kevent_copyin(void *arg, struct kevent *kevp, int count)
1206 {
1207 	struct kevent_args *uap;
1208 	int error;
1209 
1210 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1211 	uap = (struct kevent_args *)arg;
1212 
1213 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
1214 	if (error == 0)
1215 		uap->changelist += count;
1216 	return (error);
1217 }
1218 
1219 #ifdef COMPAT_FREEBSD11
1220 static int
1221 kevent11_copyout(void *arg, struct kevent *kevp, int count)
1222 {
1223 	struct freebsd11_kevent_args *uap;
1224 	struct freebsd11_kevent kev11;
1225 	int error, i;
1226 
1227 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1228 	uap = (struct freebsd11_kevent_args *)arg;
1229 
1230 	for (i = 0; i < count; i++) {
1231 		kev11.ident = kevp->ident;
1232 		kev11.filter = kevp->filter;
1233 		kev11.flags = kevp->flags;
1234 		kev11.fflags = kevp->fflags;
1235 		kev11.data = kevp->data;
1236 		kev11.udata = kevp->udata;
1237 		error = copyout(&kev11, uap->eventlist, sizeof(kev11));
1238 		if (error != 0)
1239 			break;
1240 		uap->eventlist++;
1241 		kevp++;
1242 	}
1243 	return (error);
1244 }
1245 
1246 /*
1247  * Copy 'count' items from the list pointed to by uap->changelist.
1248  */
1249 static int
1250 kevent11_copyin(void *arg, struct kevent *kevp, int count)
1251 {
1252 	struct freebsd11_kevent_args *uap;
1253 	struct freebsd11_kevent kev11;
1254 	int error, i;
1255 
1256 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1257 	uap = (struct freebsd11_kevent_args *)arg;
1258 
1259 	for (i = 0; i < count; i++) {
1260 		error = copyin(uap->changelist, &kev11, sizeof(kev11));
1261 		if (error != 0)
1262 			break;
1263 		kevp->ident = kev11.ident;
1264 		kevp->filter = kev11.filter;
1265 		kevp->flags = kev11.flags;
1266 		kevp->fflags = kev11.fflags;
1267 		kevp->data = (uintptr_t)kev11.data;
1268 		kevp->udata = kev11.udata;
1269 		bzero(&kevp->ext, sizeof(kevp->ext));
1270 		uap->changelist++;
1271 		kevp++;
1272 	}
1273 	return (error);
1274 }
1275 
1276 int
1277 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap)
1278 {
1279 	struct kevent_copyops k_ops = {
1280 		.arg = uap,
1281 		.k_copyout = kevent11_copyout,
1282 		.k_copyin = kevent11_copyin,
1283 		.kevent_size = sizeof(struct freebsd11_kevent),
1284 	};
1285 	struct g_kevent_args gk_args = {
1286 		.fd = uap->fd,
1287 		.changelist = uap->changelist,
1288 		.nchanges = uap->nchanges,
1289 		.eventlist = uap->eventlist,
1290 		.nevents = uap->nevents,
1291 		.timeout = uap->timeout,
1292 	};
1293 
1294 	return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent"));
1295 }
1296 #endif
1297 
1298 int
1299 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
1300     struct kevent_copyops *k_ops, const struct timespec *timeout)
1301 {
1302 	cap_rights_t rights;
1303 	struct file *fp;
1304 	int error;
1305 
1306 	cap_rights_init_zero(&rights);
1307 	if (nchanges > 0)
1308 		cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE);
1309 	if (nevents > 0)
1310 		cap_rights_set_one(&rights, CAP_KQUEUE_EVENT);
1311 	error = fget(td, fd, &rights, &fp);
1312 	if (error != 0)
1313 		return (error);
1314 
1315 	error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
1316 	fdrop(fp, td);
1317 
1318 	return (error);
1319 }
1320 
1321 static int
1322 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
1323     struct kevent_copyops *k_ops, const struct timespec *timeout)
1324 {
1325 	struct kevent keva[KQ_NEVENTS];
1326 	struct kevent *kevp, *changes;
1327 	int i, n, nerrors, error;
1328 
1329 	if (nchanges < 0)
1330 		return (EINVAL);
1331 
1332 	nerrors = 0;
1333 	while (nchanges > 0) {
1334 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
1335 		error = k_ops->k_copyin(k_ops->arg, keva, n);
1336 		if (error)
1337 			return (error);
1338 		changes = keva;
1339 		for (i = 0; i < n; i++) {
1340 			kevp = &changes[i];
1341 			if (!kevp->filter)
1342 				continue;
1343 			kevp->flags &= ~EV_SYSFLAGS;
1344 			error = kqueue_register(kq, kevp, td, M_WAITOK);
1345 			if (error || (kevp->flags & EV_RECEIPT)) {
1346 				if (nevents == 0)
1347 					return (error);
1348 				kevp->flags = EV_ERROR;
1349 				kevp->data = error;
1350 				(void)k_ops->k_copyout(k_ops->arg, kevp, 1);
1351 				nevents--;
1352 				nerrors++;
1353 			}
1354 		}
1355 		nchanges -= n;
1356 	}
1357 	if (nerrors) {
1358 		td->td_retval[0] = nerrors;
1359 		return (0);
1360 	}
1361 
1362 	return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
1363 }
1364 
1365 int
1366 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
1367     struct kevent_copyops *k_ops, const struct timespec *timeout)
1368 {
1369 	struct kqueue *kq;
1370 	int error;
1371 
1372 	error = kqueue_acquire(fp, &kq);
1373 	if (error != 0)
1374 		return (error);
1375 	error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
1376 	kqueue_release(kq, 0);
1377 	return (error);
1378 }
1379 
1380 /*
1381  * Performs a kevent() call on a temporarily created kqueue. This can be
1382  * used to perform one-shot polling, similar to poll() and select().
1383  */
1384 int
1385 kern_kevent_anonymous(struct thread *td, int nevents,
1386     struct kevent_copyops *k_ops)
1387 {
1388 	struct kqueue kq = {};
1389 	int error;
1390 
1391 	kqueue_init(&kq);
1392 	kq.kq_refcnt = 1;
1393 	error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
1394 	kqueue_drain(&kq, td);
1395 	kqueue_destroy(&kq);
1396 	return (error);
1397 }
1398 
1399 int
1400 kqueue_add_filteropts(int filt, const struct filterops *filtops)
1401 {
1402 	int error;
1403 
1404 	error = 0;
1405 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
1406 		printf(
1407 "trying to add a filterop that is out of range: %d is beyond %d\n",
1408 		    ~filt, EVFILT_SYSCOUNT);
1409 		return EINVAL;
1410 	}
1411 	mtx_lock(&filterops_lock);
1412 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
1413 	    sysfilt_ops[~filt].for_fop != NULL)
1414 		error = EEXIST;
1415 	else {
1416 		sysfilt_ops[~filt].for_fop = filtops;
1417 		sysfilt_ops[~filt].for_refcnt = 0;
1418 	}
1419 	mtx_unlock(&filterops_lock);
1420 
1421 	return (error);
1422 }
1423 
1424 int
1425 kqueue_del_filteropts(int filt)
1426 {
1427 	int error;
1428 
1429 	error = 0;
1430 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1431 		return EINVAL;
1432 
1433 	mtx_lock(&filterops_lock);
1434 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1435 	    sysfilt_ops[~filt].for_fop == NULL)
1436 		error = EINVAL;
1437 	else if (sysfilt_ops[~filt].for_refcnt != 0)
1438 		error = EBUSY;
1439 	else {
1440 		sysfilt_ops[~filt].for_fop = &null_filtops;
1441 		sysfilt_ops[~filt].for_refcnt = 0;
1442 	}
1443 	mtx_unlock(&filterops_lock);
1444 
1445 	return error;
1446 }
1447 
1448 static const struct filterops *
1449 kqueue_fo_find(int filt)
1450 {
1451 
1452 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1453 		return NULL;
1454 
1455 	if (sysfilt_ops[~filt].for_nolock)
1456 		return sysfilt_ops[~filt].for_fop;
1457 
1458 	mtx_lock(&filterops_lock);
1459 	sysfilt_ops[~filt].for_refcnt++;
1460 	if (sysfilt_ops[~filt].for_fop == NULL)
1461 		sysfilt_ops[~filt].for_fop = &null_filtops;
1462 	mtx_unlock(&filterops_lock);
1463 
1464 	return sysfilt_ops[~filt].for_fop;
1465 }
1466 
1467 static void
1468 kqueue_fo_release(int filt)
1469 {
1470 
1471 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1472 		return;
1473 
1474 	if (sysfilt_ops[~filt].for_nolock)
1475 		return;
1476 
1477 	mtx_lock(&filterops_lock);
1478 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1479 	    ("filter object refcount not valid on release"));
1480 	sysfilt_ops[~filt].for_refcnt--;
1481 	mtx_unlock(&filterops_lock);
1482 }
1483 
1484 /*
1485  * A ref to kq (obtained via kqueue_acquire) must be held.
1486  */
1487 static int
1488 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td,
1489     int mflag)
1490 {
1491 	const struct filterops *fops;
1492 	struct file *fp;
1493 	struct knote *kn, *tkn;
1494 	struct knlist *knl;
1495 	int error, filt, event;
1496 	int haskqglobal, filedesc_unlock;
1497 
1498 	if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
1499 		return (EINVAL);
1500 
1501 	fp = NULL;
1502 	kn = NULL;
1503 	knl = NULL;
1504 	error = 0;
1505 	haskqglobal = 0;
1506 	filedesc_unlock = 0;
1507 
1508 	filt = kev->filter;
1509 	fops = kqueue_fo_find(filt);
1510 	if (fops == NULL)
1511 		return EINVAL;
1512 
1513 	if (kev->flags & EV_ADD) {
1514 		/* Reject an invalid flag pair early */
1515 		if (kev->flags & EV_KEEPUDATA) {
1516 			tkn = NULL;
1517 			error = EINVAL;
1518 			goto done;
1519 		}
1520 
1521 		/*
1522 		 * Prevent waiting with locks.  Non-sleepable
1523 		 * allocation failures are handled in the loop, only
1524 		 * if the spare knote appears to be actually required.
1525 		 */
1526 		tkn = knote_alloc(mflag);
1527 	} else {
1528 		tkn = NULL;
1529 	}
1530 
1531 findkn:
1532 	if (fops->f_isfd) {
1533 		KASSERT(td != NULL, ("td is NULL"));
1534 		if (kev->ident > INT_MAX)
1535 			error = EBADF;
1536 		else
1537 			error = fget(td, kev->ident, &cap_event_rights, &fp);
1538 		if (error)
1539 			goto done;
1540 
1541 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1542 		    kev->ident, M_NOWAIT) != 0) {
1543 			/* try again */
1544 			fdrop(fp, td);
1545 			fp = NULL;
1546 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1547 			if (error)
1548 				goto done;
1549 			goto findkn;
1550 		}
1551 
1552 		if (fp->f_type == DTYPE_KQUEUE) {
1553 			/*
1554 			 * If we add some intelligence about what we are doing,
1555 			 * we should be able to support events on ourselves.
1556 			 * We need to know when we are doing this to prevent
1557 			 * getting both the knlist lock and the kq lock since
1558 			 * they are the same thing.
1559 			 */
1560 			if (fp->f_data == kq) {
1561 				error = EINVAL;
1562 				goto done;
1563 			}
1564 
1565 			/*
1566 			 * Pre-lock the filedesc before the global
1567 			 * lock mutex, see the comment in
1568 			 * kqueue_close().
1569 			 */
1570 			FILEDESC_XLOCK(td->td_proc->p_fd);
1571 			filedesc_unlock = 1;
1572 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1573 		}
1574 
1575 		KQ_LOCK(kq);
1576 		if (kev->ident < kq->kq_knlistsize) {
1577 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1578 				if (kev->filter == kn->kn_filter)
1579 					break;
1580 		}
1581 	} else {
1582 		if ((kev->flags & EV_ADD) == EV_ADD) {
1583 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1584 			if (error != 0)
1585 				goto done;
1586 		}
1587 
1588 		KQ_LOCK(kq);
1589 
1590 		/*
1591 		 * If possible, find an existing knote to use for this kevent.
1592 		 */
1593 		if (kev->filter == EVFILT_PROC &&
1594 		    (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
1595 			/* This is an internal creation of a process tracking
1596 			 * note. Don't attempt to coalesce this with an
1597 			 * existing note.
1598 			 */
1599 			;
1600 		} else if (kq->kq_knhashmask != 0) {
1601 			struct klist *list;
1602 
1603 			list = &kq->kq_knhash[
1604 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1605 			SLIST_FOREACH(kn, list, kn_link)
1606 				if (kev->ident == kn->kn_id &&
1607 				    kev->filter == kn->kn_filter)
1608 					break;
1609 		}
1610 	}
1611 
1612 	/* knote is in the process of changing, wait for it to stabilize. */
1613 	if (kn != NULL && kn_in_flux(kn)) {
1614 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1615 		if (filedesc_unlock) {
1616 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1617 			filedesc_unlock = 0;
1618 		}
1619 		kq->kq_state |= KQ_FLUXWAIT;
1620 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1621 		if (fp != NULL) {
1622 			fdrop(fp, td);
1623 			fp = NULL;
1624 		}
1625 		goto findkn;
1626 	}
1627 
1628 	/*
1629 	 * kn now contains the matching knote, or NULL if no match
1630 	 */
1631 	if (kn == NULL) {
1632 		if (kev->flags & EV_ADD) {
1633 			kn = tkn;
1634 			tkn = NULL;
1635 			if (kn == NULL) {
1636 				KQ_UNLOCK(kq);
1637 				error = ENOMEM;
1638 				goto done;
1639 			}
1640 			kn->kn_fp = fp;
1641 			kn->kn_kq = kq;
1642 			kn->kn_fop = fops;
1643 			/*
1644 			 * apply reference counts to knote structure, and
1645 			 * do not release it at the end of this routine.
1646 			 */
1647 			fops = NULL;
1648 			fp = NULL;
1649 
1650 			kn->kn_sfflags = kev->fflags;
1651 			kn->kn_sdata = kev->data;
1652 			kev->fflags = 0;
1653 			kev->data = 0;
1654 			kn->kn_kevent = *kev;
1655 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1656 			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1657 			kn->kn_status = KN_DETACHED;
1658 			if ((kev->flags & EV_DISABLE) != 0)
1659 				kn->kn_status |= KN_DISABLED;
1660 			kn_enter_flux(kn);
1661 
1662 			error = knote_attach(kn, kq);
1663 			KQ_UNLOCK(kq);
1664 			if (error != 0) {
1665 				tkn = kn;
1666 				goto done;
1667 			}
1668 
1669 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1670 				knote_drop_detached(kn, td);
1671 				goto done;
1672 			}
1673 			knl = kn_list_lock(kn);
1674 			goto done_ev_add;
1675 		} else {
1676 			/* No matching knote and the EV_ADD flag is not set. */
1677 			KQ_UNLOCK(kq);
1678 			error = ENOENT;
1679 			goto done;
1680 		}
1681 	}
1682 
1683 	if (kev->flags & EV_DELETE) {
1684 		kn_enter_flux(kn);
1685 		KQ_UNLOCK(kq);
1686 		knote_drop(kn, td);
1687 		goto done;
1688 	}
1689 
1690 	if (kev->flags & EV_FORCEONESHOT) {
1691 		kn->kn_flags |= EV_ONESHOT;
1692 		KNOTE_ACTIVATE(kn, 1);
1693 	}
1694 
1695 	if ((kev->flags & EV_ENABLE) != 0)
1696 		kn->kn_status &= ~KN_DISABLED;
1697 	else if ((kev->flags & EV_DISABLE) != 0)
1698 		kn->kn_status |= KN_DISABLED;
1699 
1700 	/*
1701 	 * The user may change some filter values after the initial EV_ADD,
1702 	 * but doing so will not reset any filter which has already been
1703 	 * triggered.
1704 	 */
1705 	kn->kn_status |= KN_SCAN;
1706 	kn_enter_flux(kn);
1707 	KQ_UNLOCK(kq);
1708 	knl = kn_list_lock(kn);
1709 	if ((kev->flags & EV_KEEPUDATA) == 0)
1710 		kn->kn_kevent.udata = kev->udata;
1711 	if (!fops->f_isfd && fops->f_touch != NULL) {
1712 		fops->f_touch(kn, kev, EVENT_REGISTER);
1713 	} else {
1714 		kn->kn_sfflags = kev->fflags;
1715 		kn->kn_sdata = kev->data;
1716 	}
1717 
1718 done_ev_add:
1719 	/*
1720 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1721 	 * the initial attach event decides that the event is "completed"
1722 	 * already, e.g., filt_procattach() is called on a zombie process.  It
1723 	 * will call filt_proc() which will remove it from the list, and NULL
1724 	 * kn_knlist.
1725 	 *
1726 	 * KN_DISABLED will be stable while the knote is in flux, so the
1727 	 * unlocked read will not race with an update.
1728 	 */
1729 	if ((kn->kn_status & KN_DISABLED) == 0)
1730 		event = kn->kn_fop->f_event(kn, 0);
1731 	else
1732 		event = 0;
1733 
1734 	KQ_LOCK(kq);
1735 	if (event)
1736 		kn->kn_status |= KN_ACTIVE;
1737 	if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
1738 	    KN_ACTIVE)
1739 		knote_enqueue(kn);
1740 	kn->kn_status &= ~KN_SCAN;
1741 	kn_leave_flux(kn);
1742 	kn_list_unlock(knl);
1743 	KQ_UNLOCK_FLUX(kq);
1744 
1745 done:
1746 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1747 	if (filedesc_unlock)
1748 		FILEDESC_XUNLOCK(td->td_proc->p_fd);
1749 	if (fp != NULL)
1750 		fdrop(fp, td);
1751 	knote_free(tkn);
1752 	if (fops != NULL)
1753 		kqueue_fo_release(filt);
1754 	return (error);
1755 }
1756 
1757 static int
1758 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1759 {
1760 	int error;
1761 	struct kqueue *kq;
1762 
1763 	error = 0;
1764 
1765 	kq = fp->f_data;
1766 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1767 		return (EBADF);
1768 	*kqp = kq;
1769 	KQ_LOCK(kq);
1770 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1771 		KQ_UNLOCK(kq);
1772 		return (EBADF);
1773 	}
1774 	kq->kq_refcnt++;
1775 	KQ_UNLOCK(kq);
1776 
1777 	return error;
1778 }
1779 
1780 static void
1781 kqueue_release(struct kqueue *kq, int locked)
1782 {
1783 	if (locked)
1784 		KQ_OWNED(kq);
1785 	else
1786 		KQ_LOCK(kq);
1787 	kq->kq_refcnt--;
1788 	if (kq->kq_refcnt == 1)
1789 		wakeup(&kq->kq_refcnt);
1790 	if (!locked)
1791 		KQ_UNLOCK(kq);
1792 }
1793 
1794 static void
1795 ast_kqueue(struct thread *td, int tda __unused)
1796 {
1797 	taskqueue_quiesce(taskqueue_kqueue_ctx);
1798 }
1799 
1800 static void
1801 kqueue_schedtask(struct kqueue *kq)
1802 {
1803 	KQ_OWNED(kq);
1804 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1805 	    ("scheduling kqueue task while draining"));
1806 
1807 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1808 		taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
1809 		kq->kq_state |= KQ_TASKSCHED;
1810 		ast_sched(curthread, TDA_KQUEUE);
1811 	}
1812 }
1813 
1814 /*
1815  * Expand the kq to make sure we have storage for fops/ident pair.
1816  *
1817  * Return 0 on success (or no work necessary), return errno on failure.
1818  */
1819 static int
1820 kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident,
1821     int mflag)
1822 {
1823 	struct klist *list, *tmp_knhash, *to_free;
1824 	u_long tmp_knhashmask;
1825 	int error, fd, size;
1826 
1827 	KQ_NOTOWNED(kq);
1828 
1829 	error = 0;
1830 	to_free = NULL;
1831 	if (fops->f_isfd) {
1832 		fd = ident;
1833 		if (kq->kq_knlistsize <= fd) {
1834 			size = kq->kq_knlistsize;
1835 			while (size <= fd)
1836 				size += KQEXTENT;
1837 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1838 			if (list == NULL)
1839 				return ENOMEM;
1840 			KQ_LOCK(kq);
1841 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1842 				to_free = list;
1843 				error = EBADF;
1844 			} else if (kq->kq_knlistsize > fd) {
1845 				to_free = list;
1846 			} else {
1847 				if (kq->kq_knlist != NULL) {
1848 					bcopy(kq->kq_knlist, list,
1849 					    kq->kq_knlistsize * sizeof(*list));
1850 					to_free = kq->kq_knlist;
1851 					kq->kq_knlist = NULL;
1852 				}
1853 				bzero((caddr_t)list +
1854 				    kq->kq_knlistsize * sizeof(*list),
1855 				    (size - kq->kq_knlistsize) * sizeof(*list));
1856 				kq->kq_knlistsize = size;
1857 				kq->kq_knlist = list;
1858 			}
1859 			KQ_UNLOCK(kq);
1860 		}
1861 	} else {
1862 		if (kq->kq_knhashmask == 0) {
1863 			tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE,
1864 			    &tmp_knhashmask, (mflag & M_WAITOK) != 0 ?
1865 			    HASH_WAITOK : HASH_NOWAIT);
1866 			if (tmp_knhash == NULL)
1867 				return (ENOMEM);
1868 			KQ_LOCK(kq);
1869 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1870 				to_free = tmp_knhash;
1871 				error = EBADF;
1872 			} else if (kq->kq_knhashmask == 0) {
1873 				kq->kq_knhash = tmp_knhash;
1874 				kq->kq_knhashmask = tmp_knhashmask;
1875 			} else {
1876 				to_free = tmp_knhash;
1877 			}
1878 			KQ_UNLOCK(kq);
1879 		}
1880 	}
1881 	free(to_free, M_KQUEUE);
1882 
1883 	KQ_NOTOWNED(kq);
1884 	return (error);
1885 }
1886 
1887 static void
1888 kqueue_task(void *arg, int pending)
1889 {
1890 	struct kqueue *kq;
1891 	int haskqglobal;
1892 
1893 	haskqglobal = 0;
1894 	kq = arg;
1895 
1896 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1897 	KQ_LOCK(kq);
1898 
1899 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1900 
1901 	kq->kq_state &= ~KQ_TASKSCHED;
1902 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1903 		wakeup(&kq->kq_state);
1904 	}
1905 	KQ_UNLOCK(kq);
1906 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1907 }
1908 
1909 /*
1910  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1911  * We treat KN_MARKER knotes as if they are in flux.
1912  */
1913 static int
1914 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1915     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1916 {
1917 	struct kevent *kevp;
1918 	struct knote *kn, *marker;
1919 	struct knlist *knl;
1920 	sbintime_t asbt, rsbt;
1921 	int count, error, haskqglobal, influx, nkev, touch;
1922 
1923 	count = maxevents;
1924 	nkev = 0;
1925 	error = 0;
1926 	haskqglobal = 0;
1927 
1928 	if (maxevents == 0)
1929 		goto done_nl;
1930 	if (maxevents < 0) {
1931 		error = EINVAL;
1932 		goto done_nl;
1933 	}
1934 
1935 	rsbt = 0;
1936 	if (tsp != NULL) {
1937 		if (!timespecvalid_interval(tsp)) {
1938 			error = EINVAL;
1939 			goto done_nl;
1940 		}
1941 		if (timespecisset(tsp)) {
1942 			if (tsp->tv_sec <= INT32_MAX) {
1943 				rsbt = tstosbt(*tsp);
1944 				if (TIMESEL(&asbt, rsbt))
1945 					asbt += tc_tick_sbt;
1946 				if (asbt <= SBT_MAX - rsbt)
1947 					asbt += rsbt;
1948 				else
1949 					asbt = 0;
1950 				rsbt >>= tc_precexp;
1951 			} else
1952 				asbt = 0;
1953 		} else
1954 			asbt = -1;
1955 	} else
1956 		asbt = 0;
1957 	marker = knote_alloc(M_WAITOK);
1958 	marker->kn_status = KN_MARKER;
1959 	KQ_LOCK(kq);
1960 
1961 retry:
1962 	kevp = keva;
1963 	if (kq->kq_count == 0) {
1964 		if (asbt == -1) {
1965 			error = EWOULDBLOCK;
1966 		} else {
1967 			kq->kq_state |= KQ_SLEEP;
1968 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1969 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1970 		}
1971 		if (error == 0)
1972 			goto retry;
1973 		/* don't restart after signals... */
1974 		if (error == ERESTART)
1975 			error = EINTR;
1976 		else if (error == EWOULDBLOCK)
1977 			error = 0;
1978 		goto done;
1979 	}
1980 
1981 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1982 	influx = 0;
1983 	while (count) {
1984 		KQ_OWNED(kq);
1985 		kn = TAILQ_FIRST(&kq->kq_head);
1986 
1987 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1988 		    kn_in_flux(kn)) {
1989 			if (influx) {
1990 				influx = 0;
1991 				KQ_FLUX_WAKEUP(kq);
1992 			}
1993 			kq->kq_state |= KQ_FLUXWAIT;
1994 			error = msleep(kq, &kq->kq_lock, PSOCK,
1995 			    "kqflxwt", 0);
1996 			continue;
1997 		}
1998 
1999 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2000 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
2001 			kn->kn_status &= ~KN_QUEUED;
2002 			kq->kq_count--;
2003 			continue;
2004 		}
2005 		if (kn == marker) {
2006 			KQ_FLUX_WAKEUP(kq);
2007 			if (count == maxevents)
2008 				goto retry;
2009 			goto done;
2010 		}
2011 		KASSERT(!kn_in_flux(kn),
2012 		    ("knote %p is unexpectedly in flux", kn));
2013 
2014 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
2015 			kn->kn_status &= ~KN_QUEUED;
2016 			kn_enter_flux(kn);
2017 			kq->kq_count--;
2018 			KQ_UNLOCK(kq);
2019 			/*
2020 			 * We don't need to lock the list since we've
2021 			 * marked it as in flux.
2022 			 */
2023 			knote_drop(kn, td);
2024 			KQ_LOCK(kq);
2025 			continue;
2026 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
2027 			kn->kn_status &= ~KN_QUEUED;
2028 			kn_enter_flux(kn);
2029 			kq->kq_count--;
2030 			KQ_UNLOCK(kq);
2031 			/*
2032 			 * We don't need to lock the list since we've
2033 			 * marked the knote as being in flux.
2034 			 */
2035 			*kevp = kn->kn_kevent;
2036 			knote_drop(kn, td);
2037 			KQ_LOCK(kq);
2038 			kn = NULL;
2039 		} else {
2040 			kn->kn_status |= KN_SCAN;
2041 			kn_enter_flux(kn);
2042 			KQ_UNLOCK(kq);
2043 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
2044 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
2045 			knl = kn_list_lock(kn);
2046 			if (kn->kn_fop->f_event(kn, 0) == 0) {
2047 				KQ_LOCK(kq);
2048 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2049 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
2050 				    KN_SCAN);
2051 				kn_leave_flux(kn);
2052 				kq->kq_count--;
2053 				kn_list_unlock(knl);
2054 				influx = 1;
2055 				continue;
2056 			}
2057 			touch = (!kn->kn_fop->f_isfd &&
2058 			    kn->kn_fop->f_touch != NULL);
2059 			if (touch)
2060 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
2061 			else
2062 				*kevp = kn->kn_kevent;
2063 			KQ_LOCK(kq);
2064 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2065 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
2066 				/*
2067 				 * Manually clear knotes who weren't
2068 				 * 'touch'ed.
2069 				 */
2070 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
2071 					kn->kn_data = 0;
2072 					kn->kn_fflags = 0;
2073 				}
2074 				if (kn->kn_flags & EV_DISPATCH)
2075 					kn->kn_status |= KN_DISABLED;
2076 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
2077 				kq->kq_count--;
2078 			} else
2079 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2080 
2081 			kn->kn_status &= ~KN_SCAN;
2082 			kn_leave_flux(kn);
2083 			kn_list_unlock(knl);
2084 			influx = 1;
2085 		}
2086 
2087 		/* we are returning a copy to the user */
2088 		kevp++;
2089 		nkev++;
2090 		count--;
2091 
2092 		if (nkev == KQ_NEVENTS) {
2093 			influx = 0;
2094 			KQ_UNLOCK_FLUX(kq);
2095 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2096 			nkev = 0;
2097 			kevp = keva;
2098 			KQ_LOCK(kq);
2099 			if (error)
2100 				break;
2101 		}
2102 	}
2103 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
2104 done:
2105 	KQ_OWNED(kq);
2106 	KQ_UNLOCK_FLUX(kq);
2107 	knote_free(marker);
2108 done_nl:
2109 	KQ_NOTOWNED(kq);
2110 	if (nkev != 0)
2111 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2112 	td->td_retval[0] = maxevents - count;
2113 	return (error);
2114 }
2115 
2116 /*ARGSUSED*/
2117 static int
2118 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
2119 	struct ucred *active_cred, struct thread *td)
2120 {
2121 	/*
2122 	 * Enabling sigio causes two major problems:
2123 	 * 1) infinite recursion:
2124 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
2125 	 * set.  On receipt of a signal this will cause a kqueue to recurse
2126 	 * into itself over and over.  Sending the sigio causes the kqueue
2127 	 * to become ready, which in turn posts sigio again, forever.
2128 	 * Solution: this can be solved by setting a flag in the kqueue that
2129 	 * we have a SIGIO in progress.
2130 	 * 2) locking problems:
2131 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
2132 	 * us above the proc and pgrp locks.
2133 	 * Solution: Post a signal using an async mechanism, being sure to
2134 	 * record a generation count in the delivery so that we do not deliver
2135 	 * a signal to the wrong process.
2136 	 *
2137 	 * Note, these two mechanisms are somewhat mutually exclusive!
2138 	 */
2139 #if 0
2140 	struct kqueue *kq;
2141 
2142 	kq = fp->f_data;
2143 	switch (cmd) {
2144 	case FIOASYNC:
2145 		if (*(int *)data) {
2146 			kq->kq_state |= KQ_ASYNC;
2147 		} else {
2148 			kq->kq_state &= ~KQ_ASYNC;
2149 		}
2150 		return (0);
2151 
2152 	case FIOSETOWN:
2153 		return (fsetown(*(int *)data, &kq->kq_sigio));
2154 
2155 	case FIOGETOWN:
2156 		*(int *)data = fgetown(&kq->kq_sigio);
2157 		return (0);
2158 	}
2159 #endif
2160 
2161 	return (ENOTTY);
2162 }
2163 
2164 /*ARGSUSED*/
2165 static int
2166 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
2167 	struct thread *td)
2168 {
2169 	struct kqueue *kq;
2170 	int revents = 0;
2171 	int error;
2172 
2173 	if ((error = kqueue_acquire(fp, &kq)))
2174 		return POLLERR;
2175 
2176 	KQ_LOCK(kq);
2177 	if (events & (POLLIN | POLLRDNORM)) {
2178 		if (kq->kq_count) {
2179 			revents |= events & (POLLIN | POLLRDNORM);
2180 		} else {
2181 			selrecord(td, &kq->kq_sel);
2182 			if (SEL_WAITING(&kq->kq_sel))
2183 				kq->kq_state |= KQ_SEL;
2184 		}
2185 	}
2186 	kqueue_release(kq, 1);
2187 	KQ_UNLOCK(kq);
2188 	return (revents);
2189 }
2190 
2191 /*ARGSUSED*/
2192 static int
2193 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
2194 {
2195 
2196 	bzero((void *)st, sizeof *st);
2197 	/*
2198 	 * We no longer return kq_count because the unlocked value is useless.
2199 	 * If you spent all this time getting the count, why not spend your
2200 	 * syscall better by calling kevent?
2201 	 *
2202 	 * XXX - This is needed for libc_r.
2203 	 */
2204 	st->st_mode = S_IFIFO;
2205 	return (0);
2206 }
2207 
2208 static void
2209 kqueue_drain(struct kqueue *kq, struct thread *td)
2210 {
2211 	struct knote *kn;
2212 	int i;
2213 
2214 	KQ_LOCK(kq);
2215 
2216 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
2217 	    ("kqueue already closing"));
2218 	kq->kq_state |= KQ_CLOSING;
2219 	if (kq->kq_refcnt > 1)
2220 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
2221 
2222 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
2223 
2224 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
2225 	    ("kqueue's knlist not empty"));
2226 
2227 	for (i = 0; i < kq->kq_knlistsize; i++) {
2228 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
2229 			if (kn_in_flux(kn)) {
2230 				kq->kq_state |= KQ_FLUXWAIT;
2231 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
2232 				continue;
2233 			}
2234 			kn_enter_flux(kn);
2235 			KQ_UNLOCK(kq);
2236 			knote_drop(kn, td);
2237 			KQ_LOCK(kq);
2238 		}
2239 	}
2240 	if (kq->kq_knhashmask != 0) {
2241 		for (i = 0; i <= kq->kq_knhashmask; i++) {
2242 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
2243 				if (kn_in_flux(kn)) {
2244 					kq->kq_state |= KQ_FLUXWAIT;
2245 					msleep(kq, &kq->kq_lock, PSOCK,
2246 					       "kqclo2", 0);
2247 					continue;
2248 				}
2249 				kn_enter_flux(kn);
2250 				KQ_UNLOCK(kq);
2251 				knote_drop(kn, td);
2252 				KQ_LOCK(kq);
2253 			}
2254 		}
2255 	}
2256 
2257 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
2258 		kq->kq_state |= KQ_TASKDRAIN;
2259 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
2260 	}
2261 
2262 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2263 		selwakeuppri(&kq->kq_sel, PSOCK);
2264 		if (!SEL_WAITING(&kq->kq_sel))
2265 			kq->kq_state &= ~KQ_SEL;
2266 	}
2267 
2268 	KQ_UNLOCK(kq);
2269 }
2270 
2271 static void
2272 kqueue_destroy(struct kqueue *kq)
2273 {
2274 
2275 	KASSERT(kq->kq_fdp == NULL,
2276 	    ("kqueue still attached to a file descriptor"));
2277 	seldrain(&kq->kq_sel);
2278 	knlist_destroy(&kq->kq_sel.si_note);
2279 	mtx_destroy(&kq->kq_lock);
2280 
2281 	if (kq->kq_knhash != NULL)
2282 		free(kq->kq_knhash, M_KQUEUE);
2283 	if (kq->kq_knlist != NULL)
2284 		free(kq->kq_knlist, M_KQUEUE);
2285 
2286 	funsetown(&kq->kq_sigio);
2287 }
2288 
2289 /*ARGSUSED*/
2290 static int
2291 kqueue_close(struct file *fp, struct thread *td)
2292 {
2293 	struct kqueue *kq = fp->f_data;
2294 	struct filedesc *fdp;
2295 	int error;
2296 	int filedesc_unlock;
2297 
2298 	if ((error = kqueue_acquire(fp, &kq)))
2299 		return error;
2300 	kqueue_drain(kq, td);
2301 
2302 	/*
2303 	 * We could be called due to the knote_drop() doing fdrop(),
2304 	 * called from kqueue_register().  In this case the global
2305 	 * lock is owned, and filedesc sx is locked before, to not
2306 	 * take the sleepable lock after non-sleepable.
2307 	 */
2308 	fdp = kq->kq_fdp;
2309 	kq->kq_fdp = NULL;
2310 	if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
2311 		FILEDESC_XLOCK(fdp);
2312 		filedesc_unlock = 1;
2313 	} else
2314 		filedesc_unlock = 0;
2315 	TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
2316 	if (filedesc_unlock)
2317 		FILEDESC_XUNLOCK(fdp);
2318 
2319 	kqueue_destroy(kq);
2320 	chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
2321 	crfree(kq->kq_cred);
2322 	free(kq, M_KQUEUE);
2323 	fp->f_data = NULL;
2324 
2325 	return (0);
2326 }
2327 
2328 static int
2329 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2330 {
2331 	struct kqueue *kq = fp->f_data;
2332 
2333 	kif->kf_type = KF_TYPE_KQUEUE;
2334 	kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq;
2335 	kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count;
2336 	kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state;
2337 	return (0);
2338 }
2339 
2340 static void
2341 kqueue_wakeup(struct kqueue *kq)
2342 {
2343 	KQ_OWNED(kq);
2344 
2345 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
2346 		kq->kq_state &= ~KQ_SLEEP;
2347 		wakeup(kq);
2348 	}
2349 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2350 		selwakeuppri(&kq->kq_sel, PSOCK);
2351 		if (!SEL_WAITING(&kq->kq_sel))
2352 			kq->kq_state &= ~KQ_SEL;
2353 	}
2354 	if (!knlist_empty(&kq->kq_sel.si_note))
2355 		kqueue_schedtask(kq);
2356 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
2357 		pgsigio(&kq->kq_sigio, SIGIO, 0);
2358 	}
2359 }
2360 
2361 /*
2362  * Walk down a list of knotes, activating them if their event has triggered.
2363  *
2364  * There is a possibility to optimize in the case of one kq watching another.
2365  * Instead of scheduling a task to wake it up, you could pass enough state
2366  * down the chain to make up the parent kqueue.  Make this code functional
2367  * first.
2368  */
2369 void
2370 knote(struct knlist *list, long hint, int lockflags)
2371 {
2372 	struct kqueue *kq;
2373 	struct knote *kn, *tkn;
2374 	int error;
2375 
2376 	if (list == NULL)
2377 		return;
2378 
2379 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
2380 
2381 	if ((lockflags & KNF_LISTLOCKED) == 0)
2382 		list->kl_lock(list->kl_lockarg);
2383 
2384 	/*
2385 	 * If we unlock the list lock (and enter influx), we can
2386 	 * eliminate the kqueue scheduling, but this will introduce
2387 	 * four lock/unlock's for each knote to test.  Also, marker
2388 	 * would be needed to keep iteration position, since filters
2389 	 * or other threads could remove events.
2390 	 */
2391 	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
2392 		kq = kn->kn_kq;
2393 		KQ_LOCK(kq);
2394 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
2395 			/*
2396 			 * Do not process the influx notes, except for
2397 			 * the influx coming from the kq unlock in the
2398 			 * kqueue_scan().  In the later case, we do
2399 			 * not interfere with the scan, since the code
2400 			 * fragment in kqueue_scan() locks the knlist,
2401 			 * and cannot proceed until we finished.
2402 			 */
2403 			KQ_UNLOCK(kq);
2404 		} else if ((lockflags & KNF_NOKQLOCK) != 0) {
2405 			kn_enter_flux(kn);
2406 			KQ_UNLOCK(kq);
2407 			error = kn->kn_fop->f_event(kn, hint);
2408 			KQ_LOCK(kq);
2409 			kn_leave_flux(kn);
2410 			if (error)
2411 				KNOTE_ACTIVATE(kn, 1);
2412 			KQ_UNLOCK_FLUX(kq);
2413 		} else {
2414 			if (kn->kn_fop->f_event(kn, hint))
2415 				KNOTE_ACTIVATE(kn, 1);
2416 			KQ_UNLOCK(kq);
2417 		}
2418 	}
2419 	if ((lockflags & KNF_LISTLOCKED) == 0)
2420 		list->kl_unlock(list->kl_lockarg);
2421 }
2422 
2423 /*
2424  * add a knote to a knlist
2425  */
2426 void
2427 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
2428 {
2429 
2430 	KNL_ASSERT_LOCK(knl, islocked);
2431 	KQ_NOTOWNED(kn->kn_kq);
2432 	KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn));
2433 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2434 	    ("knote %p was not detached", kn));
2435 	if (!islocked)
2436 		knl->kl_lock(knl->kl_lockarg);
2437 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
2438 	if (!islocked)
2439 		knl->kl_unlock(knl->kl_lockarg);
2440 	KQ_LOCK(kn->kn_kq);
2441 	kn->kn_knlist = knl;
2442 	kn->kn_status &= ~KN_DETACHED;
2443 	KQ_UNLOCK(kn->kn_kq);
2444 }
2445 
2446 static void
2447 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
2448     int kqislocked)
2449 {
2450 
2451 	KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked"));
2452 	KNL_ASSERT_LOCK(knl, knlislocked);
2453 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
2454 	KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn));
2455 	KASSERT((kn->kn_status & KN_DETACHED) == 0,
2456 	    ("knote %p was already detached", kn));
2457 	if (!knlislocked)
2458 		knl->kl_lock(knl->kl_lockarg);
2459 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
2460 	kn->kn_knlist = NULL;
2461 	if (!knlislocked)
2462 		kn_list_unlock(knl);
2463 	if (!kqislocked)
2464 		KQ_LOCK(kn->kn_kq);
2465 	kn->kn_status |= KN_DETACHED;
2466 	if (!kqislocked)
2467 		KQ_UNLOCK(kn->kn_kq);
2468 }
2469 
2470 /*
2471  * remove knote from the specified knlist
2472  */
2473 void
2474 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
2475 {
2476 
2477 	knlist_remove_kq(knl, kn, islocked, 0);
2478 }
2479 
2480 int
2481 knlist_empty(struct knlist *knl)
2482 {
2483 
2484 	KNL_ASSERT_LOCKED(knl);
2485 	return (SLIST_EMPTY(&knl->kl_list));
2486 }
2487 
2488 static struct mtx knlist_lock;
2489 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2490     MTX_DEF);
2491 static void knlist_mtx_lock(void *arg);
2492 static void knlist_mtx_unlock(void *arg);
2493 
2494 static void
2495 knlist_mtx_lock(void *arg)
2496 {
2497 
2498 	mtx_lock((struct mtx *)arg);
2499 }
2500 
2501 static void
2502 knlist_mtx_unlock(void *arg)
2503 {
2504 
2505 	mtx_unlock((struct mtx *)arg);
2506 }
2507 
2508 static void
2509 knlist_mtx_assert_lock(void *arg, int what)
2510 {
2511 
2512 	if (what == LA_LOCKED)
2513 		mtx_assert((struct mtx *)arg, MA_OWNED);
2514 	else
2515 		mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2516 }
2517 
2518 void
2519 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2520     void (*kl_unlock)(void *),
2521     void (*kl_assert_lock)(void *, int))
2522 {
2523 
2524 	if (lock == NULL)
2525 		knl->kl_lockarg = &knlist_lock;
2526 	else
2527 		knl->kl_lockarg = lock;
2528 
2529 	if (kl_lock == NULL)
2530 		knl->kl_lock = knlist_mtx_lock;
2531 	else
2532 		knl->kl_lock = kl_lock;
2533 	if (kl_unlock == NULL)
2534 		knl->kl_unlock = knlist_mtx_unlock;
2535 	else
2536 		knl->kl_unlock = kl_unlock;
2537 	if (kl_assert_lock == NULL)
2538 		knl->kl_assert_lock = knlist_mtx_assert_lock;
2539 	else
2540 		knl->kl_assert_lock = kl_assert_lock;
2541 
2542 	knl->kl_autodestroy = 0;
2543 	SLIST_INIT(&knl->kl_list);
2544 }
2545 
2546 void
2547 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2548 {
2549 
2550 	knlist_init(knl, lock, NULL, NULL, NULL);
2551 }
2552 
2553 struct knlist *
2554 knlist_alloc(struct mtx *lock)
2555 {
2556 	struct knlist *knl;
2557 
2558 	knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
2559 	knlist_init_mtx(knl, lock);
2560 	return (knl);
2561 }
2562 
2563 void
2564 knlist_destroy(struct knlist *knl)
2565 {
2566 
2567 	KASSERT(KNLIST_EMPTY(knl),
2568 	    ("destroying knlist %p with knotes on it", knl));
2569 }
2570 
2571 void
2572 knlist_detach(struct knlist *knl)
2573 {
2574 
2575 	KNL_ASSERT_LOCKED(knl);
2576 	knl->kl_autodestroy = 1;
2577 	if (knlist_empty(knl)) {
2578 		knlist_destroy(knl);
2579 		free(knl, M_KQUEUE);
2580 	}
2581 }
2582 
2583 /*
2584  * Even if we are locked, we may need to drop the lock to allow any influx
2585  * knotes time to "settle".
2586  */
2587 void
2588 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2589 {
2590 	struct knote *kn, *kn2;
2591 	struct kqueue *kq;
2592 
2593 	KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
2594 	if (islocked)
2595 		KNL_ASSERT_LOCKED(knl);
2596 	else {
2597 		KNL_ASSERT_UNLOCKED(knl);
2598 again:		/* need to reacquire lock since we have dropped it */
2599 		knl->kl_lock(knl->kl_lockarg);
2600 	}
2601 
2602 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2603 		kq = kn->kn_kq;
2604 		KQ_LOCK(kq);
2605 		if (kn_in_flux(kn)) {
2606 			KQ_UNLOCK(kq);
2607 			continue;
2608 		}
2609 		knlist_remove_kq(knl, kn, 1, 1);
2610 		if (killkn) {
2611 			kn_enter_flux(kn);
2612 			KQ_UNLOCK(kq);
2613 			knote_drop_detached(kn, td);
2614 		} else {
2615 			/* Make sure cleared knotes disappear soon */
2616 			kn->kn_flags |= EV_EOF | EV_ONESHOT;
2617 			KQ_UNLOCK(kq);
2618 		}
2619 		kq = NULL;
2620 	}
2621 
2622 	if (!SLIST_EMPTY(&knl->kl_list)) {
2623 		/* there are still in flux knotes remaining */
2624 		kn = SLIST_FIRST(&knl->kl_list);
2625 		kq = kn->kn_kq;
2626 		KQ_LOCK(kq);
2627 		KASSERT(kn_in_flux(kn), ("knote removed w/o list lock"));
2628 		knl->kl_unlock(knl->kl_lockarg);
2629 		kq->kq_state |= KQ_FLUXWAIT;
2630 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2631 		kq = NULL;
2632 		goto again;
2633 	}
2634 
2635 	if (islocked)
2636 		KNL_ASSERT_LOCKED(knl);
2637 	else {
2638 		knl->kl_unlock(knl->kl_lockarg);
2639 		KNL_ASSERT_UNLOCKED(knl);
2640 	}
2641 }
2642 
2643 /*
2644  * Remove all knotes referencing a specified fd must be called with FILEDESC
2645  * lock.  This prevents a race where a new fd comes along and occupies the
2646  * entry and we attach a knote to the fd.
2647  */
2648 void
2649 knote_fdclose(struct thread *td, int fd)
2650 {
2651 	struct filedesc *fdp = td->td_proc->p_fd;
2652 	struct kqueue *kq;
2653 	struct knote *kn;
2654 	int influx;
2655 
2656 	FILEDESC_XLOCK_ASSERT(fdp);
2657 
2658 	/*
2659 	 * We shouldn't have to worry about new kevents appearing on fd
2660 	 * since filedesc is locked.
2661 	 */
2662 	TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2663 		KQ_LOCK(kq);
2664 
2665 again:
2666 		influx = 0;
2667 		while (kq->kq_knlistsize > fd &&
2668 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2669 			if (kn_in_flux(kn)) {
2670 				/* someone else might be waiting on our knote */
2671 				if (influx)
2672 					wakeup(kq);
2673 				kq->kq_state |= KQ_FLUXWAIT;
2674 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2675 				goto again;
2676 			}
2677 			kn_enter_flux(kn);
2678 			KQ_UNLOCK(kq);
2679 			influx = 1;
2680 			knote_drop(kn, td);
2681 			KQ_LOCK(kq);
2682 		}
2683 		KQ_UNLOCK_FLUX(kq);
2684 	}
2685 }
2686 
2687 static int
2688 knote_attach(struct knote *kn, struct kqueue *kq)
2689 {
2690 	struct klist *list;
2691 
2692 	KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn));
2693 	KQ_OWNED(kq);
2694 
2695 	if ((kq->kq_state & KQ_CLOSING) != 0)
2696 		return (EBADF);
2697 	if (kn->kn_fop->f_isfd) {
2698 		if (kn->kn_id >= kq->kq_knlistsize)
2699 			return (ENOMEM);
2700 		list = &kq->kq_knlist[kn->kn_id];
2701 	} else {
2702 		if (kq->kq_knhash == NULL)
2703 			return (ENOMEM);
2704 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2705 	}
2706 	SLIST_INSERT_HEAD(list, kn, kn_link);
2707 	return (0);
2708 }
2709 
2710 static void
2711 knote_drop(struct knote *kn, struct thread *td)
2712 {
2713 
2714 	if ((kn->kn_status & KN_DETACHED) == 0)
2715 		kn->kn_fop->f_detach(kn);
2716 	knote_drop_detached(kn, td);
2717 }
2718 
2719 static void
2720 knote_drop_detached(struct knote *kn, struct thread *td)
2721 {
2722 	struct kqueue *kq;
2723 	struct klist *list;
2724 
2725 	kq = kn->kn_kq;
2726 
2727 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2728 	    ("knote %p still attached", kn));
2729 	KQ_NOTOWNED(kq);
2730 
2731 	KQ_LOCK(kq);
2732 	KASSERT(kn->kn_influx == 1,
2733 	    ("knote_drop called on %p with influx %d", kn, kn->kn_influx));
2734 
2735 	if (kn->kn_fop->f_isfd)
2736 		list = &kq->kq_knlist[kn->kn_id];
2737 	else
2738 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2739 
2740 	if (!SLIST_EMPTY(list))
2741 		SLIST_REMOVE(list, kn, knote, kn_link);
2742 	if (kn->kn_status & KN_QUEUED)
2743 		knote_dequeue(kn);
2744 	KQ_UNLOCK_FLUX(kq);
2745 
2746 	if (kn->kn_fop->f_isfd) {
2747 		fdrop(kn->kn_fp, td);
2748 		kn->kn_fp = NULL;
2749 	}
2750 	kqueue_fo_release(kn->kn_kevent.filter);
2751 	kn->kn_fop = NULL;
2752 	knote_free(kn);
2753 }
2754 
2755 static void
2756 knote_enqueue(struct knote *kn)
2757 {
2758 	struct kqueue *kq = kn->kn_kq;
2759 
2760 	KQ_OWNED(kn->kn_kq);
2761 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2762 
2763 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2764 	kn->kn_status |= KN_QUEUED;
2765 	kq->kq_count++;
2766 	kqueue_wakeup(kq);
2767 }
2768 
2769 static void
2770 knote_dequeue(struct knote *kn)
2771 {
2772 	struct kqueue *kq = kn->kn_kq;
2773 
2774 	KQ_OWNED(kn->kn_kq);
2775 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2776 
2777 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2778 	kn->kn_status &= ~KN_QUEUED;
2779 	kq->kq_count--;
2780 }
2781 
2782 static void
2783 knote_init(void)
2784 {
2785 
2786 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2787 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2788 	ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue);
2789 }
2790 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2791 
2792 static struct knote *
2793 knote_alloc(int mflag)
2794 {
2795 
2796 	return (uma_zalloc(knote_zone, mflag | M_ZERO));
2797 }
2798 
2799 static void
2800 knote_free(struct knote *kn)
2801 {
2802 
2803 	uma_zfree(knote_zone, kn);
2804 }
2805 
2806 /*
2807  * Register the kev w/ the kq specified by fd.
2808  */
2809 int
2810 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag)
2811 {
2812 	struct kqueue *kq;
2813 	struct file *fp;
2814 	cap_rights_t rights;
2815 	int error;
2816 
2817 	error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE),
2818 	    &fp);
2819 	if (error != 0)
2820 		return (error);
2821 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2822 		goto noacquire;
2823 
2824 	error = kqueue_register(kq, kev, td, mflag);
2825 	kqueue_release(kq, 0);
2826 
2827 noacquire:
2828 	fdrop(fp, td);
2829 	return (error);
2830 }
2831