xref: /freebsd/sys/kern/kern_event.c (revision ebacd8013fe5f7fdf9f6a5b286f6680dd2891036)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
5  * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org>
6  * Copyright (c) 2009 Apple, Inc.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_ktrace.h"
35 #include "opt_kqueue.h"
36 
37 #ifdef COMPAT_FREEBSD11
38 #define	_WANT_FREEBSD11_KEVENT
39 #endif
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/capsicum.h>
44 #include <sys/kernel.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/mutex.h>
48 #include <sys/proc.h>
49 #include <sys/malloc.h>
50 #include <sys/unistd.h>
51 #include <sys/file.h>
52 #include <sys/filedesc.h>
53 #include <sys/filio.h>
54 #include <sys/fcntl.h>
55 #include <sys/kthread.h>
56 #include <sys/selinfo.h>
57 #include <sys/queue.h>
58 #include <sys/event.h>
59 #include <sys/eventvar.h>
60 #include <sys/poll.h>
61 #include <sys/protosw.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sigio.h>
64 #include <sys/signalvar.h>
65 #include <sys/socket.h>
66 #include <sys/socketvar.h>
67 #include <sys/stat.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysproto.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/taskqueue.h>
72 #include <sys/uio.h>
73 #include <sys/user.h>
74 #ifdef KTRACE
75 #include <sys/ktrace.h>
76 #endif
77 #include <machine/atomic.h>
78 
79 #include <vm/uma.h>
80 
81 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
82 
83 /*
84  * This lock is used if multiple kq locks are required.  This possibly
85  * should be made into a per proc lock.
86  */
87 static struct mtx	kq_global;
88 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
89 #define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
90 	if (!haslck)				\
91 		mtx_lock(lck);			\
92 	haslck = 1;				\
93 } while (0)
94 #define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
95 	if (haslck)				\
96 		mtx_unlock(lck);			\
97 	haslck = 0;				\
98 } while (0)
99 
100 TASKQUEUE_DEFINE_THREAD(kqueue_ctx);
101 
102 static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
103 static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
104 static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
105 		    struct thread *td, int mflag);
106 static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
107 static void	kqueue_release(struct kqueue *kq, int locked);
108 static void	kqueue_destroy(struct kqueue *kq);
109 static void	kqueue_drain(struct kqueue *kq, struct thread *td);
110 static int	kqueue_expand(struct kqueue *kq, struct filterops *fops,
111 		    uintptr_t ident, int mflag);
112 static void	kqueue_task(void *arg, int pending);
113 static int	kqueue_scan(struct kqueue *kq, int maxevents,
114 		    struct kevent_copyops *k_ops,
115 		    const struct timespec *timeout,
116 		    struct kevent *keva, struct thread *td);
117 static void 	kqueue_wakeup(struct kqueue *kq);
118 static struct filterops *kqueue_fo_find(int filt);
119 static void	kqueue_fo_release(int filt);
120 struct g_kevent_args;
121 static int	kern_kevent_generic(struct thread *td,
122 		    struct g_kevent_args *uap,
123 		    struct kevent_copyops *k_ops, const char *struct_name);
124 
125 static fo_ioctl_t	kqueue_ioctl;
126 static fo_poll_t	kqueue_poll;
127 static fo_kqfilter_t	kqueue_kqfilter;
128 static fo_stat_t	kqueue_stat;
129 static fo_close_t	kqueue_close;
130 static fo_fill_kinfo_t	kqueue_fill_kinfo;
131 
132 static struct fileops kqueueops = {
133 	.fo_read = invfo_rdwr,
134 	.fo_write = invfo_rdwr,
135 	.fo_truncate = invfo_truncate,
136 	.fo_ioctl = kqueue_ioctl,
137 	.fo_poll = kqueue_poll,
138 	.fo_kqfilter = kqueue_kqfilter,
139 	.fo_stat = kqueue_stat,
140 	.fo_close = kqueue_close,
141 	.fo_chmod = invfo_chmod,
142 	.fo_chown = invfo_chown,
143 	.fo_sendfile = invfo_sendfile,
144 	.fo_fill_kinfo = kqueue_fill_kinfo,
145 };
146 
147 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
148 static void 	knote_drop(struct knote *kn, struct thread *td);
149 static void 	knote_drop_detached(struct knote *kn, struct thread *td);
150 static void 	knote_enqueue(struct knote *kn);
151 static void 	knote_dequeue(struct knote *kn);
152 static void 	knote_init(void);
153 static struct 	knote *knote_alloc(int mflag);
154 static void 	knote_free(struct knote *kn);
155 
156 static void	filt_kqdetach(struct knote *kn);
157 static int	filt_kqueue(struct knote *kn, long hint);
158 static int	filt_procattach(struct knote *kn);
159 static void	filt_procdetach(struct knote *kn);
160 static int	filt_proc(struct knote *kn, long hint);
161 static int	filt_fileattach(struct knote *kn);
162 static void	filt_timerexpire(void *knx);
163 static void	filt_timerexpire_l(struct knote *kn, bool proc_locked);
164 static int	filt_timerattach(struct knote *kn);
165 static void	filt_timerdetach(struct knote *kn);
166 static void	filt_timerstart(struct knote *kn, sbintime_t to);
167 static void	filt_timertouch(struct knote *kn, struct kevent *kev,
168 		    u_long type);
169 static int	filt_timervalidate(struct knote *kn, sbintime_t *to);
170 static int	filt_timer(struct knote *kn, long hint);
171 static int	filt_userattach(struct knote *kn);
172 static void	filt_userdetach(struct knote *kn);
173 static int	filt_user(struct knote *kn, long hint);
174 static void	filt_usertouch(struct knote *kn, struct kevent *kev,
175 		    u_long type);
176 
177 static struct filterops file_filtops = {
178 	.f_isfd = 1,
179 	.f_attach = filt_fileattach,
180 };
181 static struct filterops kqread_filtops = {
182 	.f_isfd = 1,
183 	.f_detach = filt_kqdetach,
184 	.f_event = filt_kqueue,
185 };
186 /* XXX - move to kern_proc.c?  */
187 static struct filterops proc_filtops = {
188 	.f_isfd = 0,
189 	.f_attach = filt_procattach,
190 	.f_detach = filt_procdetach,
191 	.f_event = filt_proc,
192 };
193 static struct filterops timer_filtops = {
194 	.f_isfd = 0,
195 	.f_attach = filt_timerattach,
196 	.f_detach = filt_timerdetach,
197 	.f_event = filt_timer,
198 	.f_touch = filt_timertouch,
199 };
200 static struct filterops user_filtops = {
201 	.f_attach = filt_userattach,
202 	.f_detach = filt_userdetach,
203 	.f_event = filt_user,
204 	.f_touch = filt_usertouch,
205 };
206 
207 static uma_zone_t	knote_zone;
208 static unsigned int __exclusive_cache_line	kq_ncallouts;
209 static unsigned int 	kq_calloutmax = 4 * 1024;
210 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
211     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
212 
213 /* XXX - ensure not influx ? */
214 #define KNOTE_ACTIVATE(kn, islock) do { 				\
215 	if ((islock))							\
216 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
217 	else								\
218 		KQ_LOCK((kn)->kn_kq);					\
219 	(kn)->kn_status |= KN_ACTIVE;					\
220 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
221 		knote_enqueue((kn));					\
222 	if (!(islock))							\
223 		KQ_UNLOCK((kn)->kn_kq);					\
224 } while (0)
225 #define KQ_LOCK(kq) do {						\
226 	mtx_lock(&(kq)->kq_lock);					\
227 } while (0)
228 #define KQ_FLUX_WAKEUP(kq) do {						\
229 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
230 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
231 		wakeup((kq));						\
232 	}								\
233 } while (0)
234 #define KQ_UNLOCK_FLUX(kq) do {						\
235 	KQ_FLUX_WAKEUP(kq);						\
236 	mtx_unlock(&(kq)->kq_lock);					\
237 } while (0)
238 #define KQ_UNLOCK(kq) do {						\
239 	mtx_unlock(&(kq)->kq_lock);					\
240 } while (0)
241 #define KQ_OWNED(kq) do {						\
242 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
243 } while (0)
244 #define KQ_NOTOWNED(kq) do {						\
245 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
246 } while (0)
247 
248 static struct knlist *
249 kn_list_lock(struct knote *kn)
250 {
251 	struct knlist *knl;
252 
253 	knl = kn->kn_knlist;
254 	if (knl != NULL)
255 		knl->kl_lock(knl->kl_lockarg);
256 	return (knl);
257 }
258 
259 static void
260 kn_list_unlock(struct knlist *knl)
261 {
262 	bool do_free;
263 
264 	if (knl == NULL)
265 		return;
266 	do_free = knl->kl_autodestroy && knlist_empty(knl);
267 	knl->kl_unlock(knl->kl_lockarg);
268 	if (do_free) {
269 		knlist_destroy(knl);
270 		free(knl, M_KQUEUE);
271 	}
272 }
273 
274 static bool
275 kn_in_flux(struct knote *kn)
276 {
277 
278 	return (kn->kn_influx > 0);
279 }
280 
281 static void
282 kn_enter_flux(struct knote *kn)
283 {
284 
285 	KQ_OWNED(kn->kn_kq);
286 	MPASS(kn->kn_influx < INT_MAX);
287 	kn->kn_influx++;
288 }
289 
290 static bool
291 kn_leave_flux(struct knote *kn)
292 {
293 
294 	KQ_OWNED(kn->kn_kq);
295 	MPASS(kn->kn_influx > 0);
296 	kn->kn_influx--;
297 	return (kn->kn_influx == 0);
298 }
299 
300 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
301 	if (islocked)							\
302 		KNL_ASSERT_LOCKED(knl);				\
303 	else								\
304 		KNL_ASSERT_UNLOCKED(knl);				\
305 } while (0)
306 #ifdef INVARIANTS
307 #define	KNL_ASSERT_LOCKED(knl) do {					\
308 	knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED);		\
309 } while (0)
310 #define	KNL_ASSERT_UNLOCKED(knl) do {					\
311 	knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED);		\
312 } while (0)
313 #else /* !INVARIANTS */
314 #define	KNL_ASSERT_LOCKED(knl) do {} while (0)
315 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
316 #endif /* INVARIANTS */
317 
318 #ifndef	KN_HASHSIZE
319 #define	KN_HASHSIZE		64		/* XXX should be tunable */
320 #endif
321 
322 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
323 
324 static int
325 filt_nullattach(struct knote *kn)
326 {
327 
328 	return (ENXIO);
329 };
330 
331 struct filterops null_filtops = {
332 	.f_isfd = 0,
333 	.f_attach = filt_nullattach,
334 };
335 
336 /* XXX - make SYSINIT to add these, and move into respective modules. */
337 extern struct filterops sig_filtops;
338 extern struct filterops fs_filtops;
339 
340 /*
341  * Table for all system-defined filters.
342  */
343 static struct mtx	filterops_lock;
344 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
345 	MTX_DEF);
346 static struct {
347 	struct filterops *for_fop;
348 	int for_nolock;
349 	int for_refcnt;
350 } sysfilt_ops[EVFILT_SYSCOUNT] = {
351 	{ &file_filtops, 1 },			/* EVFILT_READ */
352 	{ &file_filtops, 1 },			/* EVFILT_WRITE */
353 	{ &null_filtops },			/* EVFILT_AIO */
354 	{ &file_filtops, 1 },			/* EVFILT_VNODE */
355 	{ &proc_filtops, 1 },			/* EVFILT_PROC */
356 	{ &sig_filtops, 1 },			/* EVFILT_SIGNAL */
357 	{ &timer_filtops, 1 },			/* EVFILT_TIMER */
358 	{ &file_filtops, 1 },			/* EVFILT_PROCDESC */
359 	{ &fs_filtops, 1 },			/* EVFILT_FS */
360 	{ &null_filtops },			/* EVFILT_LIO */
361 	{ &user_filtops, 1 },			/* EVFILT_USER */
362 	{ &null_filtops },			/* EVFILT_SENDFILE */
363 	{ &file_filtops, 1 },                   /* EVFILT_EMPTY */
364 };
365 
366 /*
367  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
368  * method.
369  */
370 static int
371 filt_fileattach(struct knote *kn)
372 {
373 
374 	return (fo_kqfilter(kn->kn_fp, kn));
375 }
376 
377 /*ARGSUSED*/
378 static int
379 kqueue_kqfilter(struct file *fp, struct knote *kn)
380 {
381 	struct kqueue *kq = kn->kn_fp->f_data;
382 
383 	if (kn->kn_filter != EVFILT_READ)
384 		return (EINVAL);
385 
386 	kn->kn_status |= KN_KQUEUE;
387 	kn->kn_fop = &kqread_filtops;
388 	knlist_add(&kq->kq_sel.si_note, kn, 0);
389 
390 	return (0);
391 }
392 
393 static void
394 filt_kqdetach(struct knote *kn)
395 {
396 	struct kqueue *kq = kn->kn_fp->f_data;
397 
398 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
399 }
400 
401 /*ARGSUSED*/
402 static int
403 filt_kqueue(struct knote *kn, long hint)
404 {
405 	struct kqueue *kq = kn->kn_fp->f_data;
406 
407 	kn->kn_data = kq->kq_count;
408 	return (kn->kn_data > 0);
409 }
410 
411 /* XXX - move to kern_proc.c?  */
412 static int
413 filt_procattach(struct knote *kn)
414 {
415 	struct proc *p;
416 	int error;
417 	bool exiting, immediate;
418 
419 	exiting = immediate = false;
420 	if (kn->kn_sfflags & NOTE_EXIT)
421 		p = pfind_any(kn->kn_id);
422 	else
423 		p = pfind(kn->kn_id);
424 	if (p == NULL)
425 		return (ESRCH);
426 	if (p->p_flag & P_WEXIT)
427 		exiting = true;
428 
429 	if ((error = p_cansee(curthread, p))) {
430 		PROC_UNLOCK(p);
431 		return (error);
432 	}
433 
434 	kn->kn_ptr.p_proc = p;
435 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
436 
437 	/*
438 	 * Internal flag indicating registration done by kernel for the
439 	 * purposes of getting a NOTE_CHILD notification.
440 	 */
441 	if (kn->kn_flags & EV_FLAG2) {
442 		kn->kn_flags &= ~EV_FLAG2;
443 		kn->kn_data = kn->kn_sdata;		/* ppid */
444 		kn->kn_fflags = NOTE_CHILD;
445 		kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
446 		immediate = true; /* Force immediate activation of child note. */
447 	}
448 	/*
449 	 * Internal flag indicating registration done by kernel (for other than
450 	 * NOTE_CHILD).
451 	 */
452 	if (kn->kn_flags & EV_FLAG1) {
453 		kn->kn_flags &= ~EV_FLAG1;
454 	}
455 
456 	knlist_add(p->p_klist, kn, 1);
457 
458 	/*
459 	 * Immediately activate any child notes or, in the case of a zombie
460 	 * target process, exit notes.  The latter is necessary to handle the
461 	 * case where the target process, e.g. a child, dies before the kevent
462 	 * is registered.
463 	 */
464 	if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
465 		KNOTE_ACTIVATE(kn, 0);
466 
467 	PROC_UNLOCK(p);
468 
469 	return (0);
470 }
471 
472 /*
473  * The knote may be attached to a different process, which may exit,
474  * leaving nothing for the knote to be attached to.  So when the process
475  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
476  * it will be deleted when read out.  However, as part of the knote deletion,
477  * this routine is called, so a check is needed to avoid actually performing
478  * a detach, because the original process does not exist any more.
479  */
480 /* XXX - move to kern_proc.c?  */
481 static void
482 filt_procdetach(struct knote *kn)
483 {
484 
485 	knlist_remove(kn->kn_knlist, kn, 0);
486 	kn->kn_ptr.p_proc = NULL;
487 }
488 
489 /* XXX - move to kern_proc.c?  */
490 static int
491 filt_proc(struct knote *kn, long hint)
492 {
493 	struct proc *p;
494 	u_int event;
495 
496 	p = kn->kn_ptr.p_proc;
497 	if (p == NULL) /* already activated, from attach filter */
498 		return (0);
499 
500 	/* Mask off extra data. */
501 	event = (u_int)hint & NOTE_PCTRLMASK;
502 
503 	/* If the user is interested in this event, record it. */
504 	if (kn->kn_sfflags & event)
505 		kn->kn_fflags |= event;
506 
507 	/* Process is gone, so flag the event as finished. */
508 	if (event == NOTE_EXIT) {
509 		kn->kn_flags |= EV_EOF | EV_ONESHOT;
510 		kn->kn_ptr.p_proc = NULL;
511 		if (kn->kn_fflags & NOTE_EXIT)
512 			kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
513 		if (kn->kn_fflags == 0)
514 			kn->kn_flags |= EV_DROP;
515 		return (1);
516 	}
517 
518 	return (kn->kn_fflags != 0);
519 }
520 
521 /*
522  * Called when the process forked. It mostly does the same as the
523  * knote(), activating all knotes registered to be activated when the
524  * process forked. Additionally, for each knote attached to the
525  * parent, check whether user wants to track the new process. If so
526  * attach a new knote to it, and immediately report an event with the
527  * child's pid.
528  */
529 void
530 knote_fork(struct knlist *list, int pid)
531 {
532 	struct kqueue *kq;
533 	struct knote *kn;
534 	struct kevent kev;
535 	int error;
536 
537 	MPASS(list != NULL);
538 	KNL_ASSERT_LOCKED(list);
539 	if (SLIST_EMPTY(&list->kl_list))
540 		return;
541 
542 	memset(&kev, 0, sizeof(kev));
543 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
544 		kq = kn->kn_kq;
545 		KQ_LOCK(kq);
546 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
547 			KQ_UNLOCK(kq);
548 			continue;
549 		}
550 
551 		/*
552 		 * The same as knote(), activate the event.
553 		 */
554 		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
555 			if (kn->kn_fop->f_event(kn, NOTE_FORK))
556 				KNOTE_ACTIVATE(kn, 1);
557 			KQ_UNLOCK(kq);
558 			continue;
559 		}
560 
561 		/*
562 		 * The NOTE_TRACK case. In addition to the activation
563 		 * of the event, we need to register new events to
564 		 * track the child. Drop the locks in preparation for
565 		 * the call to kqueue_register().
566 		 */
567 		kn_enter_flux(kn);
568 		KQ_UNLOCK(kq);
569 		list->kl_unlock(list->kl_lockarg);
570 
571 		/*
572 		 * Activate existing knote and register tracking knotes with
573 		 * new process.
574 		 *
575 		 * First register a knote to get just the child notice. This
576 		 * must be a separate note from a potential NOTE_EXIT
577 		 * notification since both NOTE_CHILD and NOTE_EXIT are defined
578 		 * to use the data field (in conflicting ways).
579 		 */
580 		kev.ident = pid;
581 		kev.filter = kn->kn_filter;
582 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
583 		    EV_FLAG2;
584 		kev.fflags = kn->kn_sfflags;
585 		kev.data = kn->kn_id;		/* parent */
586 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
587 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
588 		if (error)
589 			kn->kn_fflags |= NOTE_TRACKERR;
590 
591 		/*
592 		 * Then register another knote to track other potential events
593 		 * from the new process.
594 		 */
595 		kev.ident = pid;
596 		kev.filter = kn->kn_filter;
597 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
598 		kev.fflags = kn->kn_sfflags;
599 		kev.data = kn->kn_id;		/* parent */
600 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
601 		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
602 		if (error)
603 			kn->kn_fflags |= NOTE_TRACKERR;
604 		if (kn->kn_fop->f_event(kn, NOTE_FORK))
605 			KNOTE_ACTIVATE(kn, 0);
606 		list->kl_lock(list->kl_lockarg);
607 		KQ_LOCK(kq);
608 		kn_leave_flux(kn);
609 		KQ_UNLOCK_FLUX(kq);
610 	}
611 }
612 
613 /*
614  * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
615  * interval timer support code.
616  */
617 
618 #define NOTE_TIMER_PRECMASK						\
619     (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)
620 
621 static sbintime_t
622 timer2sbintime(int64_t data, int flags)
623 {
624 	int64_t secs;
625 
626         /*
627          * Macros for converting to the fractional second portion of an
628          * sbintime_t using 64bit multiplication to improve precision.
629          */
630 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
631 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
632 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
633 	switch (flags & NOTE_TIMER_PRECMASK) {
634 	case NOTE_SECONDS:
635 #ifdef __LP64__
636 		if (data > (SBT_MAX / SBT_1S))
637 			return (SBT_MAX);
638 #endif
639 		return ((sbintime_t)data << 32);
640 	case NOTE_MSECONDS: /* FALLTHROUGH */
641 	case 0:
642 		if (data >= 1000) {
643 			secs = data / 1000;
644 #ifdef __LP64__
645 			if (secs > (SBT_MAX / SBT_1S))
646 				return (SBT_MAX);
647 #endif
648 			return (secs << 32 | MS_TO_SBT(data % 1000));
649 		}
650 		return (MS_TO_SBT(data));
651 	case NOTE_USECONDS:
652 		if (data >= 1000000) {
653 			secs = data / 1000000;
654 #ifdef __LP64__
655 			if (secs > (SBT_MAX / SBT_1S))
656 				return (SBT_MAX);
657 #endif
658 			return (secs << 32 | US_TO_SBT(data % 1000000));
659 		}
660 		return (US_TO_SBT(data));
661 	case NOTE_NSECONDS:
662 		if (data >= 1000000000) {
663 			secs = data / 1000000000;
664 #ifdef __LP64__
665 			if (secs > (SBT_MAX / SBT_1S))
666 				return (SBT_MAX);
667 #endif
668 			return (secs << 32 | NS_TO_SBT(data % 1000000000));
669 		}
670 		return (NS_TO_SBT(data));
671 	default:
672 		break;
673 	}
674 	return (-1);
675 }
676 
677 struct kq_timer_cb_data {
678 	struct callout c;
679 	struct proc *p;
680 	struct knote *kn;
681 	int cpuid;
682 	int flags;
683 	TAILQ_ENTRY(kq_timer_cb_data) link;
684 	sbintime_t next;	/* next timer event fires at */
685 	sbintime_t to;		/* precalculated timer period, 0 for abs */
686 };
687 
688 #define	KQ_TIMER_CB_ENQUEUED	0x01
689 
690 static void
691 kqtimer_sched_callout(struct kq_timer_cb_data *kc)
692 {
693 	callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn,
694 	    kc->cpuid, C_ABSOLUTE);
695 }
696 
697 void
698 kqtimer_proc_continue(struct proc *p)
699 {
700 	struct kq_timer_cb_data *kc, *kc1;
701 	struct bintime bt;
702 	sbintime_t now;
703 
704 	PROC_LOCK_ASSERT(p, MA_OWNED);
705 
706 	getboottimebin(&bt);
707 	now = bttosbt(bt);
708 
709 	TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) {
710 		TAILQ_REMOVE(&p->p_kqtim_stop, kc, link);
711 		kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
712 		if (kc->next <= now)
713 			filt_timerexpire_l(kc->kn, true);
714 		else
715 			kqtimer_sched_callout(kc);
716 	}
717 }
718 
719 static void
720 filt_timerexpire_l(struct knote *kn, bool proc_locked)
721 {
722 	struct kq_timer_cb_data *kc;
723 	struct proc *p;
724 	uint64_t delta;
725 	sbintime_t now;
726 
727 	kc = kn->kn_ptr.p_v;
728 
729 	if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) {
730 		kn->kn_data++;
731 		KNOTE_ACTIVATE(kn, 0);
732 		return;
733 	}
734 
735 	now = sbinuptime();
736 	if (now >= kc->next) {
737 		delta = (now - kc->next) / kc->to;
738 		if (delta == 0)
739 			delta = 1;
740 		kn->kn_data += delta;
741 		kc->next += delta * kc->to;
742 		if (now >= kc->next)	/* overflow */
743 			kc->next = now + kc->to;
744 		KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
745 	}
746 
747 	/*
748 	 * Initial check for stopped kc->p is racy.  It is fine to
749 	 * miss the set of the stop flags, at worst we would schedule
750 	 * one more callout.  On the other hand, it is not fine to not
751 	 * schedule when we we missed clearing of the flags, we
752 	 * recheck them under the lock and observe consistent state.
753 	 */
754 	p = kc->p;
755 	if (P_SHOULDSTOP(p) || P_KILLED(p)) {
756 		if (!proc_locked)
757 			PROC_LOCK(p);
758 		if (P_SHOULDSTOP(p) || P_KILLED(p)) {
759 			if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) {
760 				kc->flags |= KQ_TIMER_CB_ENQUEUED;
761 				TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link);
762 			}
763 			if (!proc_locked)
764 				PROC_UNLOCK(p);
765 			return;
766 		}
767 		if (!proc_locked)
768 			PROC_UNLOCK(p);
769 	}
770 	kqtimer_sched_callout(kc);
771 }
772 
773 static void
774 filt_timerexpire(void *knx)
775 {
776 	filt_timerexpire_l(knx, false);
777 }
778 
779 /*
780  * data contains amount of time to sleep
781  */
782 static int
783 filt_timervalidate(struct knote *kn, sbintime_t *to)
784 {
785 	struct bintime bt;
786 	sbintime_t sbt;
787 
788 	if (kn->kn_sdata < 0)
789 		return (EINVAL);
790 	if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
791 		kn->kn_sdata = 1;
792 	/*
793 	 * The only fflags values supported are the timer unit
794 	 * (precision) and the absolute time indicator.
795 	 */
796 	if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0)
797 		return (EINVAL);
798 
799 	*to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
800 	if (*to < 0)
801 		return (EINVAL);
802 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
803 		getboottimebin(&bt);
804 		sbt = bttosbt(bt);
805 		*to = MAX(0, *to - sbt);
806 	}
807 	return (0);
808 }
809 
810 static int
811 filt_timerattach(struct knote *kn)
812 {
813 	struct kq_timer_cb_data *kc;
814 	sbintime_t to;
815 	int error;
816 
817 	to = -1;
818 	error = filt_timervalidate(kn, &to);
819 	if (error != 0)
820 		return (error);
821 	KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 ||
822 	    (kn->kn_sfflags & NOTE_ABSTIME) != 0,
823 	    ("%s: periodic timer has a calculated zero timeout", __func__));
824 	KASSERT(to >= 0,
825 	    ("%s: timer has a calculated negative timeout", __func__));
826 
827 	if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) {
828 		atomic_subtract_int(&kq_ncallouts, 1);
829 		return (ENOMEM);
830 	}
831 
832 	if ((kn->kn_sfflags & NOTE_ABSTIME) == 0)
833 		kn->kn_flags |= EV_CLEAR;	/* automatically set */
834 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
835 	kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
836 	kc->kn = kn;
837 	kc->p = curproc;
838 	kc->cpuid = PCPU_GET(cpuid);
839 	kc->flags = 0;
840 	callout_init(&kc->c, 1);
841 	filt_timerstart(kn, to);
842 
843 	return (0);
844 }
845 
846 static void
847 filt_timerstart(struct knote *kn, sbintime_t to)
848 {
849 	struct kq_timer_cb_data *kc;
850 
851 	kc = kn->kn_ptr.p_v;
852 	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
853 		kc->next = to;
854 		kc->to = 0;
855 	} else {
856 		kc->next = to + sbinuptime();
857 		kc->to = to;
858 	}
859 	kqtimer_sched_callout(kc);
860 }
861 
862 static void
863 filt_timerdetach(struct knote *kn)
864 {
865 	struct kq_timer_cb_data *kc;
866 	unsigned int old __unused;
867 	bool pending;
868 
869 	kc = kn->kn_ptr.p_v;
870 	do {
871 		callout_drain(&kc->c);
872 
873 		/*
874 		 * kqtimer_proc_continue() might have rescheduled this callout.
875 		 * Double-check, using the process mutex as an interlock.
876 		 */
877 		PROC_LOCK(kc->p);
878 		if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) {
879 			kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
880 			TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link);
881 		}
882 		pending = callout_pending(&kc->c);
883 		PROC_UNLOCK(kc->p);
884 	} while (pending);
885 	free(kc, M_KQUEUE);
886 	old = atomic_fetchadd_int(&kq_ncallouts, -1);
887 	KASSERT(old > 0, ("Number of callouts cannot become negative"));
888 	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
889 }
890 
891 static void
892 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type)
893 {
894 	struct kq_timer_cb_data *kc;
895 	struct kqueue *kq;
896 	sbintime_t to;
897 	int error;
898 
899 	switch (type) {
900 	case EVENT_REGISTER:
901 		/* Handle re-added timers that update data/fflags */
902 		if (kev->flags & EV_ADD) {
903 			kc = kn->kn_ptr.p_v;
904 
905 			/* Drain any existing callout. */
906 			callout_drain(&kc->c);
907 
908 			/* Throw away any existing undelivered record
909 			 * of the timer expiration. This is done under
910 			 * the presumption that if a process is
911 			 * re-adding this timer with new parameters,
912 			 * it is no longer interested in what may have
913 			 * happened under the old parameters. If it is
914 			 * interested, it can wait for the expiration,
915 			 * delete the old timer definition, and then
916 			 * add the new one.
917 			 *
918 			 * This has to be done while the kq is locked:
919 			 *   - if enqueued, dequeue
920 			 *   - make it no longer active
921 			 *   - clear the count of expiration events
922 			 */
923 			kq = kn->kn_kq;
924 			KQ_LOCK(kq);
925 			if (kn->kn_status & KN_QUEUED)
926 				knote_dequeue(kn);
927 
928 			kn->kn_status &= ~KN_ACTIVE;
929 			kn->kn_data = 0;
930 			KQ_UNLOCK(kq);
931 
932 			/* Reschedule timer based on new data/fflags */
933 			kn->kn_sfflags = kev->fflags;
934 			kn->kn_sdata = kev->data;
935 			error = filt_timervalidate(kn, &to);
936 			if (error != 0) {
937 			  	kn->kn_flags |= EV_ERROR;
938 				kn->kn_data = error;
939 			} else
940 			  	filt_timerstart(kn, to);
941 		}
942 		break;
943 
944         case EVENT_PROCESS:
945 		*kev = kn->kn_kevent;
946 		if (kn->kn_flags & EV_CLEAR) {
947 			kn->kn_data = 0;
948 			kn->kn_fflags = 0;
949 		}
950 		break;
951 
952 	default:
953 		panic("filt_timertouch() - invalid type (%ld)", type);
954 		break;
955 	}
956 }
957 
958 static int
959 filt_timer(struct knote *kn, long hint)
960 {
961 
962 	return (kn->kn_data != 0);
963 }
964 
965 static int
966 filt_userattach(struct knote *kn)
967 {
968 
969 	/*
970 	 * EVFILT_USER knotes are not attached to anything in the kernel.
971 	 */
972 	kn->kn_hook = NULL;
973 	if (kn->kn_fflags & NOTE_TRIGGER)
974 		kn->kn_hookid = 1;
975 	else
976 		kn->kn_hookid = 0;
977 	return (0);
978 }
979 
980 static void
981 filt_userdetach(__unused struct knote *kn)
982 {
983 
984 	/*
985 	 * EVFILT_USER knotes are not attached to anything in the kernel.
986 	 */
987 }
988 
989 static int
990 filt_user(struct knote *kn, __unused long hint)
991 {
992 
993 	return (kn->kn_hookid);
994 }
995 
996 static void
997 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
998 {
999 	u_int ffctrl;
1000 
1001 	switch (type) {
1002 	case EVENT_REGISTER:
1003 		if (kev->fflags & NOTE_TRIGGER)
1004 			kn->kn_hookid = 1;
1005 
1006 		ffctrl = kev->fflags & NOTE_FFCTRLMASK;
1007 		kev->fflags &= NOTE_FFLAGSMASK;
1008 		switch (ffctrl) {
1009 		case NOTE_FFNOP:
1010 			break;
1011 
1012 		case NOTE_FFAND:
1013 			kn->kn_sfflags &= kev->fflags;
1014 			break;
1015 
1016 		case NOTE_FFOR:
1017 			kn->kn_sfflags |= kev->fflags;
1018 			break;
1019 
1020 		case NOTE_FFCOPY:
1021 			kn->kn_sfflags = kev->fflags;
1022 			break;
1023 
1024 		default:
1025 			/* XXX Return error? */
1026 			break;
1027 		}
1028 		kn->kn_sdata = kev->data;
1029 		if (kev->flags & EV_CLEAR) {
1030 			kn->kn_hookid = 0;
1031 			kn->kn_data = 0;
1032 			kn->kn_fflags = 0;
1033 		}
1034 		break;
1035 
1036         case EVENT_PROCESS:
1037 		*kev = kn->kn_kevent;
1038 		kev->fflags = kn->kn_sfflags;
1039 		kev->data = kn->kn_sdata;
1040 		if (kn->kn_flags & EV_CLEAR) {
1041 			kn->kn_hookid = 0;
1042 			kn->kn_data = 0;
1043 			kn->kn_fflags = 0;
1044 		}
1045 		break;
1046 
1047 	default:
1048 		panic("filt_usertouch() - invalid type (%ld)", type);
1049 		break;
1050 	}
1051 }
1052 
1053 int
1054 sys_kqueue(struct thread *td, struct kqueue_args *uap)
1055 {
1056 
1057 	return (kern_kqueue(td, 0, NULL));
1058 }
1059 
1060 int
1061 sys_kqueuex(struct thread *td, struct kqueuex_args *uap)
1062 {
1063 	int flags;
1064 
1065 	if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0)
1066 		return (EINVAL);
1067 	flags = 0;
1068 	if ((uap->flags & KQUEUE_CLOEXEC) != 0)
1069 		flags |= O_CLOEXEC;
1070 	return (kern_kqueue(td, flags, NULL));
1071 }
1072 
1073 static void
1074 kqueue_init(struct kqueue *kq)
1075 {
1076 
1077 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
1078 	TAILQ_INIT(&kq->kq_head);
1079 	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
1080 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
1081 }
1082 
1083 int
1084 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps)
1085 {
1086 	struct filedesc *fdp;
1087 	struct kqueue *kq;
1088 	struct file *fp;
1089 	struct ucred *cred;
1090 	int fd, error;
1091 
1092 	fdp = td->td_proc->p_fd;
1093 	cred = td->td_ucred;
1094 	if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
1095 		return (ENOMEM);
1096 
1097 	error = falloc_caps(td, &fp, &fd, flags, fcaps);
1098 	if (error != 0) {
1099 		chgkqcnt(cred->cr_ruidinfo, -1, 0);
1100 		return (error);
1101 	}
1102 
1103 	/* An extra reference on `fp' has been held for us by falloc(). */
1104 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
1105 	kqueue_init(kq);
1106 	kq->kq_fdp = fdp;
1107 	kq->kq_cred = crhold(cred);
1108 
1109 	FILEDESC_XLOCK(fdp);
1110 	TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
1111 	FILEDESC_XUNLOCK(fdp);
1112 
1113 	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
1114 	fdrop(fp, td);
1115 
1116 	td->td_retval[0] = fd;
1117 	return (0);
1118 }
1119 
1120 struct g_kevent_args {
1121 	int	fd;
1122 	const void *changelist;
1123 	int	nchanges;
1124 	void	*eventlist;
1125 	int	nevents;
1126 	const struct timespec *timeout;
1127 };
1128 
1129 int
1130 sys_kevent(struct thread *td, struct kevent_args *uap)
1131 {
1132 	struct kevent_copyops k_ops = {
1133 		.arg = uap,
1134 		.k_copyout = kevent_copyout,
1135 		.k_copyin = kevent_copyin,
1136 		.kevent_size = sizeof(struct kevent),
1137 	};
1138 	struct g_kevent_args gk_args = {
1139 		.fd = uap->fd,
1140 		.changelist = uap->changelist,
1141 		.nchanges = uap->nchanges,
1142 		.eventlist = uap->eventlist,
1143 		.nevents = uap->nevents,
1144 		.timeout = uap->timeout,
1145 	};
1146 
1147 	return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent"));
1148 }
1149 
1150 static int
1151 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap,
1152     struct kevent_copyops *k_ops, const char *struct_name)
1153 {
1154 	struct timespec ts, *tsp;
1155 #ifdef KTRACE
1156 	struct kevent *eventlist = uap->eventlist;
1157 #endif
1158 	int error;
1159 
1160 	if (uap->timeout != NULL) {
1161 		error = copyin(uap->timeout, &ts, sizeof(ts));
1162 		if (error)
1163 			return (error);
1164 		tsp = &ts;
1165 	} else
1166 		tsp = NULL;
1167 
1168 #ifdef KTRACE
1169 	if (KTRPOINT(td, KTR_STRUCT_ARRAY))
1170 		ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist,
1171 		    uap->nchanges, k_ops->kevent_size);
1172 #endif
1173 
1174 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
1175 	    k_ops, tsp);
1176 
1177 #ifdef KTRACE
1178 	if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
1179 		ktrstructarray(struct_name, UIO_USERSPACE, eventlist,
1180 		    td->td_retval[0], k_ops->kevent_size);
1181 #endif
1182 
1183 	return (error);
1184 }
1185 
1186 /*
1187  * Copy 'count' items into the destination list pointed to by uap->eventlist.
1188  */
1189 static int
1190 kevent_copyout(void *arg, struct kevent *kevp, int count)
1191 {
1192 	struct kevent_args *uap;
1193 	int error;
1194 
1195 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1196 	uap = (struct kevent_args *)arg;
1197 
1198 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
1199 	if (error == 0)
1200 		uap->eventlist += count;
1201 	return (error);
1202 }
1203 
1204 /*
1205  * Copy 'count' items from the list pointed to by uap->changelist.
1206  */
1207 static int
1208 kevent_copyin(void *arg, struct kevent *kevp, int count)
1209 {
1210 	struct kevent_args *uap;
1211 	int error;
1212 
1213 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1214 	uap = (struct kevent_args *)arg;
1215 
1216 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
1217 	if (error == 0)
1218 		uap->changelist += count;
1219 	return (error);
1220 }
1221 
1222 #ifdef COMPAT_FREEBSD11
1223 static int
1224 kevent11_copyout(void *arg, struct kevent *kevp, int count)
1225 {
1226 	struct freebsd11_kevent_args *uap;
1227 	struct freebsd11_kevent kev11;
1228 	int error, i;
1229 
1230 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1231 	uap = (struct freebsd11_kevent_args *)arg;
1232 
1233 	for (i = 0; i < count; i++) {
1234 		kev11.ident = kevp->ident;
1235 		kev11.filter = kevp->filter;
1236 		kev11.flags = kevp->flags;
1237 		kev11.fflags = kevp->fflags;
1238 		kev11.data = kevp->data;
1239 		kev11.udata = kevp->udata;
1240 		error = copyout(&kev11, uap->eventlist, sizeof(kev11));
1241 		if (error != 0)
1242 			break;
1243 		uap->eventlist++;
1244 		kevp++;
1245 	}
1246 	return (error);
1247 }
1248 
1249 /*
1250  * Copy 'count' items from the list pointed to by uap->changelist.
1251  */
1252 static int
1253 kevent11_copyin(void *arg, struct kevent *kevp, int count)
1254 {
1255 	struct freebsd11_kevent_args *uap;
1256 	struct freebsd11_kevent kev11;
1257 	int error, i;
1258 
1259 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1260 	uap = (struct freebsd11_kevent_args *)arg;
1261 
1262 	for (i = 0; i < count; i++) {
1263 		error = copyin(uap->changelist, &kev11, sizeof(kev11));
1264 		if (error != 0)
1265 			break;
1266 		kevp->ident = kev11.ident;
1267 		kevp->filter = kev11.filter;
1268 		kevp->flags = kev11.flags;
1269 		kevp->fflags = kev11.fflags;
1270 		kevp->data = (uintptr_t)kev11.data;
1271 		kevp->udata = kev11.udata;
1272 		bzero(&kevp->ext, sizeof(kevp->ext));
1273 		uap->changelist++;
1274 		kevp++;
1275 	}
1276 	return (error);
1277 }
1278 
1279 int
1280 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap)
1281 {
1282 	struct kevent_copyops k_ops = {
1283 		.arg = uap,
1284 		.k_copyout = kevent11_copyout,
1285 		.k_copyin = kevent11_copyin,
1286 		.kevent_size = sizeof(struct freebsd11_kevent),
1287 	};
1288 	struct g_kevent_args gk_args = {
1289 		.fd = uap->fd,
1290 		.changelist = uap->changelist,
1291 		.nchanges = uap->nchanges,
1292 		.eventlist = uap->eventlist,
1293 		.nevents = uap->nevents,
1294 		.timeout = uap->timeout,
1295 	};
1296 
1297 	return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent"));
1298 }
1299 #endif
1300 
1301 int
1302 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
1303     struct kevent_copyops *k_ops, const struct timespec *timeout)
1304 {
1305 	cap_rights_t rights;
1306 	struct file *fp;
1307 	int error;
1308 
1309 	cap_rights_init_zero(&rights);
1310 	if (nchanges > 0)
1311 		cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE);
1312 	if (nevents > 0)
1313 		cap_rights_set_one(&rights, CAP_KQUEUE_EVENT);
1314 	error = fget(td, fd, &rights, &fp);
1315 	if (error != 0)
1316 		return (error);
1317 
1318 	error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
1319 	fdrop(fp, td);
1320 
1321 	return (error);
1322 }
1323 
1324 static int
1325 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
1326     struct kevent_copyops *k_ops, const struct timespec *timeout)
1327 {
1328 	struct kevent keva[KQ_NEVENTS];
1329 	struct kevent *kevp, *changes;
1330 	int i, n, nerrors, error;
1331 
1332 	if (nchanges < 0)
1333 		return (EINVAL);
1334 
1335 	nerrors = 0;
1336 	while (nchanges > 0) {
1337 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
1338 		error = k_ops->k_copyin(k_ops->arg, keva, n);
1339 		if (error)
1340 			return (error);
1341 		changes = keva;
1342 		for (i = 0; i < n; i++) {
1343 			kevp = &changes[i];
1344 			if (!kevp->filter)
1345 				continue;
1346 			kevp->flags &= ~EV_SYSFLAGS;
1347 			error = kqueue_register(kq, kevp, td, M_WAITOK);
1348 			if (error || (kevp->flags & EV_RECEIPT)) {
1349 				if (nevents == 0)
1350 					return (error);
1351 				kevp->flags = EV_ERROR;
1352 				kevp->data = error;
1353 				(void)k_ops->k_copyout(k_ops->arg, kevp, 1);
1354 				nevents--;
1355 				nerrors++;
1356 			}
1357 		}
1358 		nchanges -= n;
1359 	}
1360 	if (nerrors) {
1361 		td->td_retval[0] = nerrors;
1362 		return (0);
1363 	}
1364 
1365 	return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
1366 }
1367 
1368 int
1369 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
1370     struct kevent_copyops *k_ops, const struct timespec *timeout)
1371 {
1372 	struct kqueue *kq;
1373 	int error;
1374 
1375 	error = kqueue_acquire(fp, &kq);
1376 	if (error != 0)
1377 		return (error);
1378 	error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
1379 	kqueue_release(kq, 0);
1380 	return (error);
1381 }
1382 
1383 /*
1384  * Performs a kevent() call on a temporarily created kqueue. This can be
1385  * used to perform one-shot polling, similar to poll() and select().
1386  */
1387 int
1388 kern_kevent_anonymous(struct thread *td, int nevents,
1389     struct kevent_copyops *k_ops)
1390 {
1391 	struct kqueue kq = {};
1392 	int error;
1393 
1394 	kqueue_init(&kq);
1395 	kq.kq_refcnt = 1;
1396 	error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
1397 	kqueue_drain(&kq, td);
1398 	kqueue_destroy(&kq);
1399 	return (error);
1400 }
1401 
1402 int
1403 kqueue_add_filteropts(int filt, struct filterops *filtops)
1404 {
1405 	int error;
1406 
1407 	error = 0;
1408 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
1409 		printf(
1410 "trying to add a filterop that is out of range: %d is beyond %d\n",
1411 		    ~filt, EVFILT_SYSCOUNT);
1412 		return EINVAL;
1413 	}
1414 	mtx_lock(&filterops_lock);
1415 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
1416 	    sysfilt_ops[~filt].for_fop != NULL)
1417 		error = EEXIST;
1418 	else {
1419 		sysfilt_ops[~filt].for_fop = filtops;
1420 		sysfilt_ops[~filt].for_refcnt = 0;
1421 	}
1422 	mtx_unlock(&filterops_lock);
1423 
1424 	return (error);
1425 }
1426 
1427 int
1428 kqueue_del_filteropts(int filt)
1429 {
1430 	int error;
1431 
1432 	error = 0;
1433 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1434 		return EINVAL;
1435 
1436 	mtx_lock(&filterops_lock);
1437 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1438 	    sysfilt_ops[~filt].for_fop == NULL)
1439 		error = EINVAL;
1440 	else if (sysfilt_ops[~filt].for_refcnt != 0)
1441 		error = EBUSY;
1442 	else {
1443 		sysfilt_ops[~filt].for_fop = &null_filtops;
1444 		sysfilt_ops[~filt].for_refcnt = 0;
1445 	}
1446 	mtx_unlock(&filterops_lock);
1447 
1448 	return error;
1449 }
1450 
1451 static struct filterops *
1452 kqueue_fo_find(int filt)
1453 {
1454 
1455 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1456 		return NULL;
1457 
1458 	if (sysfilt_ops[~filt].for_nolock)
1459 		return sysfilt_ops[~filt].for_fop;
1460 
1461 	mtx_lock(&filterops_lock);
1462 	sysfilt_ops[~filt].for_refcnt++;
1463 	if (sysfilt_ops[~filt].for_fop == NULL)
1464 		sysfilt_ops[~filt].for_fop = &null_filtops;
1465 	mtx_unlock(&filterops_lock);
1466 
1467 	return sysfilt_ops[~filt].for_fop;
1468 }
1469 
1470 static void
1471 kqueue_fo_release(int filt)
1472 {
1473 
1474 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1475 		return;
1476 
1477 	if (sysfilt_ops[~filt].for_nolock)
1478 		return;
1479 
1480 	mtx_lock(&filterops_lock);
1481 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1482 	    ("filter object refcount not valid on release"));
1483 	sysfilt_ops[~filt].for_refcnt--;
1484 	mtx_unlock(&filterops_lock);
1485 }
1486 
1487 /*
1488  * A ref to kq (obtained via kqueue_acquire) must be held.
1489  */
1490 static int
1491 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td,
1492     int mflag)
1493 {
1494 	struct filterops *fops;
1495 	struct file *fp;
1496 	struct knote *kn, *tkn;
1497 	struct knlist *knl;
1498 	int error, filt, event;
1499 	int haskqglobal, filedesc_unlock;
1500 
1501 	if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
1502 		return (EINVAL);
1503 
1504 	fp = NULL;
1505 	kn = NULL;
1506 	knl = NULL;
1507 	error = 0;
1508 	haskqglobal = 0;
1509 	filedesc_unlock = 0;
1510 
1511 	filt = kev->filter;
1512 	fops = kqueue_fo_find(filt);
1513 	if (fops == NULL)
1514 		return EINVAL;
1515 
1516 	if (kev->flags & EV_ADD) {
1517 		/* Reject an invalid flag pair early */
1518 		if (kev->flags & EV_KEEPUDATA) {
1519 			tkn = NULL;
1520 			error = EINVAL;
1521 			goto done;
1522 		}
1523 
1524 		/*
1525 		 * Prevent waiting with locks.  Non-sleepable
1526 		 * allocation failures are handled in the loop, only
1527 		 * if the spare knote appears to be actually required.
1528 		 */
1529 		tkn = knote_alloc(mflag);
1530 	} else {
1531 		tkn = NULL;
1532 	}
1533 
1534 findkn:
1535 	if (fops->f_isfd) {
1536 		KASSERT(td != NULL, ("td is NULL"));
1537 		if (kev->ident > INT_MAX)
1538 			error = EBADF;
1539 		else
1540 			error = fget(td, kev->ident, &cap_event_rights, &fp);
1541 		if (error)
1542 			goto done;
1543 
1544 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1545 		    kev->ident, M_NOWAIT) != 0) {
1546 			/* try again */
1547 			fdrop(fp, td);
1548 			fp = NULL;
1549 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1550 			if (error)
1551 				goto done;
1552 			goto findkn;
1553 		}
1554 
1555 		if (fp->f_type == DTYPE_KQUEUE) {
1556 			/*
1557 			 * If we add some intelligence about what we are doing,
1558 			 * we should be able to support events on ourselves.
1559 			 * We need to know when we are doing this to prevent
1560 			 * getting both the knlist lock and the kq lock since
1561 			 * they are the same thing.
1562 			 */
1563 			if (fp->f_data == kq) {
1564 				error = EINVAL;
1565 				goto done;
1566 			}
1567 
1568 			/*
1569 			 * Pre-lock the filedesc before the global
1570 			 * lock mutex, see the comment in
1571 			 * kqueue_close().
1572 			 */
1573 			FILEDESC_XLOCK(td->td_proc->p_fd);
1574 			filedesc_unlock = 1;
1575 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1576 		}
1577 
1578 		KQ_LOCK(kq);
1579 		if (kev->ident < kq->kq_knlistsize) {
1580 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1581 				if (kev->filter == kn->kn_filter)
1582 					break;
1583 		}
1584 	} else {
1585 		if ((kev->flags & EV_ADD) == EV_ADD) {
1586 			error = kqueue_expand(kq, fops, kev->ident, mflag);
1587 			if (error != 0)
1588 				goto done;
1589 		}
1590 
1591 		KQ_LOCK(kq);
1592 
1593 		/*
1594 		 * If possible, find an existing knote to use for this kevent.
1595 		 */
1596 		if (kev->filter == EVFILT_PROC &&
1597 		    (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
1598 			/* This is an internal creation of a process tracking
1599 			 * note. Don't attempt to coalesce this with an
1600 			 * existing note.
1601 			 */
1602 			;
1603 		} else if (kq->kq_knhashmask != 0) {
1604 			struct klist *list;
1605 
1606 			list = &kq->kq_knhash[
1607 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1608 			SLIST_FOREACH(kn, list, kn_link)
1609 				if (kev->ident == kn->kn_id &&
1610 				    kev->filter == kn->kn_filter)
1611 					break;
1612 		}
1613 	}
1614 
1615 	/* knote is in the process of changing, wait for it to stabilize. */
1616 	if (kn != NULL && kn_in_flux(kn)) {
1617 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1618 		if (filedesc_unlock) {
1619 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1620 			filedesc_unlock = 0;
1621 		}
1622 		kq->kq_state |= KQ_FLUXWAIT;
1623 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1624 		if (fp != NULL) {
1625 			fdrop(fp, td);
1626 			fp = NULL;
1627 		}
1628 		goto findkn;
1629 	}
1630 
1631 	/*
1632 	 * kn now contains the matching knote, or NULL if no match
1633 	 */
1634 	if (kn == NULL) {
1635 		if (kev->flags & EV_ADD) {
1636 			kn = tkn;
1637 			tkn = NULL;
1638 			if (kn == NULL) {
1639 				KQ_UNLOCK(kq);
1640 				error = ENOMEM;
1641 				goto done;
1642 			}
1643 			kn->kn_fp = fp;
1644 			kn->kn_kq = kq;
1645 			kn->kn_fop = fops;
1646 			/*
1647 			 * apply reference counts to knote structure, and
1648 			 * do not release it at the end of this routine.
1649 			 */
1650 			fops = NULL;
1651 			fp = NULL;
1652 
1653 			kn->kn_sfflags = kev->fflags;
1654 			kn->kn_sdata = kev->data;
1655 			kev->fflags = 0;
1656 			kev->data = 0;
1657 			kn->kn_kevent = *kev;
1658 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1659 			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1660 			kn->kn_status = KN_DETACHED;
1661 			if ((kev->flags & EV_DISABLE) != 0)
1662 				kn->kn_status |= KN_DISABLED;
1663 			kn_enter_flux(kn);
1664 
1665 			error = knote_attach(kn, kq);
1666 			KQ_UNLOCK(kq);
1667 			if (error != 0) {
1668 				tkn = kn;
1669 				goto done;
1670 			}
1671 
1672 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1673 				knote_drop_detached(kn, td);
1674 				goto done;
1675 			}
1676 			knl = kn_list_lock(kn);
1677 			goto done_ev_add;
1678 		} else {
1679 			/* No matching knote and the EV_ADD flag is not set. */
1680 			KQ_UNLOCK(kq);
1681 			error = ENOENT;
1682 			goto done;
1683 		}
1684 	}
1685 
1686 	if (kev->flags & EV_DELETE) {
1687 		kn_enter_flux(kn);
1688 		KQ_UNLOCK(kq);
1689 		knote_drop(kn, td);
1690 		goto done;
1691 	}
1692 
1693 	if (kev->flags & EV_FORCEONESHOT) {
1694 		kn->kn_flags |= EV_ONESHOT;
1695 		KNOTE_ACTIVATE(kn, 1);
1696 	}
1697 
1698 	if ((kev->flags & EV_ENABLE) != 0)
1699 		kn->kn_status &= ~KN_DISABLED;
1700 	else if ((kev->flags & EV_DISABLE) != 0)
1701 		kn->kn_status |= KN_DISABLED;
1702 
1703 	/*
1704 	 * The user may change some filter values after the initial EV_ADD,
1705 	 * but doing so will not reset any filter which has already been
1706 	 * triggered.
1707 	 */
1708 	kn->kn_status |= KN_SCAN;
1709 	kn_enter_flux(kn);
1710 	KQ_UNLOCK(kq);
1711 	knl = kn_list_lock(kn);
1712 	if ((kev->flags & EV_KEEPUDATA) == 0)
1713 		kn->kn_kevent.udata = kev->udata;
1714 	if (!fops->f_isfd && fops->f_touch != NULL) {
1715 		fops->f_touch(kn, kev, EVENT_REGISTER);
1716 	} else {
1717 		kn->kn_sfflags = kev->fflags;
1718 		kn->kn_sdata = kev->data;
1719 	}
1720 
1721 done_ev_add:
1722 	/*
1723 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1724 	 * the initial attach event decides that the event is "completed"
1725 	 * already, e.g., filt_procattach() is called on a zombie process.  It
1726 	 * will call filt_proc() which will remove it from the list, and NULL
1727 	 * kn_knlist.
1728 	 *
1729 	 * KN_DISABLED will be stable while the knote is in flux, so the
1730 	 * unlocked read will not race with an update.
1731 	 */
1732 	if ((kn->kn_status & KN_DISABLED) == 0)
1733 		event = kn->kn_fop->f_event(kn, 0);
1734 	else
1735 		event = 0;
1736 
1737 	KQ_LOCK(kq);
1738 	if (event)
1739 		kn->kn_status |= KN_ACTIVE;
1740 	if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
1741 	    KN_ACTIVE)
1742 		knote_enqueue(kn);
1743 	kn->kn_status &= ~KN_SCAN;
1744 	kn_leave_flux(kn);
1745 	kn_list_unlock(knl);
1746 	KQ_UNLOCK_FLUX(kq);
1747 
1748 done:
1749 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1750 	if (filedesc_unlock)
1751 		FILEDESC_XUNLOCK(td->td_proc->p_fd);
1752 	if (fp != NULL)
1753 		fdrop(fp, td);
1754 	knote_free(tkn);
1755 	if (fops != NULL)
1756 		kqueue_fo_release(filt);
1757 	return (error);
1758 }
1759 
1760 static int
1761 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1762 {
1763 	int error;
1764 	struct kqueue *kq;
1765 
1766 	error = 0;
1767 
1768 	kq = fp->f_data;
1769 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1770 		return (EBADF);
1771 	*kqp = kq;
1772 	KQ_LOCK(kq);
1773 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1774 		KQ_UNLOCK(kq);
1775 		return (EBADF);
1776 	}
1777 	kq->kq_refcnt++;
1778 	KQ_UNLOCK(kq);
1779 
1780 	return error;
1781 }
1782 
1783 static void
1784 kqueue_release(struct kqueue *kq, int locked)
1785 {
1786 	if (locked)
1787 		KQ_OWNED(kq);
1788 	else
1789 		KQ_LOCK(kq);
1790 	kq->kq_refcnt--;
1791 	if (kq->kq_refcnt == 1)
1792 		wakeup(&kq->kq_refcnt);
1793 	if (!locked)
1794 		KQ_UNLOCK(kq);
1795 }
1796 
1797 static void
1798 ast_kqueue(struct thread *td, int tda __unused)
1799 {
1800 	taskqueue_quiesce(taskqueue_kqueue_ctx);
1801 }
1802 
1803 static void
1804 kqueue_schedtask(struct kqueue *kq)
1805 {
1806 	KQ_OWNED(kq);
1807 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1808 	    ("scheduling kqueue task while draining"));
1809 
1810 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1811 		taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
1812 		kq->kq_state |= KQ_TASKSCHED;
1813 		ast_sched(curthread, TDA_KQUEUE);
1814 	}
1815 }
1816 
1817 /*
1818  * Expand the kq to make sure we have storage for fops/ident pair.
1819  *
1820  * Return 0 on success (or no work necessary), return errno on failure.
1821  */
1822 static int
1823 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1824     int mflag)
1825 {
1826 	struct klist *list, *tmp_knhash, *to_free;
1827 	u_long tmp_knhashmask;
1828 	int error, fd, size;
1829 
1830 	KQ_NOTOWNED(kq);
1831 
1832 	error = 0;
1833 	to_free = NULL;
1834 	if (fops->f_isfd) {
1835 		fd = ident;
1836 		if (kq->kq_knlistsize <= fd) {
1837 			size = kq->kq_knlistsize;
1838 			while (size <= fd)
1839 				size += KQEXTENT;
1840 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1841 			if (list == NULL)
1842 				return ENOMEM;
1843 			KQ_LOCK(kq);
1844 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1845 				to_free = list;
1846 				error = EBADF;
1847 			} else if (kq->kq_knlistsize > fd) {
1848 				to_free = list;
1849 			} else {
1850 				if (kq->kq_knlist != NULL) {
1851 					bcopy(kq->kq_knlist, list,
1852 					    kq->kq_knlistsize * sizeof(*list));
1853 					to_free = kq->kq_knlist;
1854 					kq->kq_knlist = NULL;
1855 				}
1856 				bzero((caddr_t)list +
1857 				    kq->kq_knlistsize * sizeof(*list),
1858 				    (size - kq->kq_knlistsize) * sizeof(*list));
1859 				kq->kq_knlistsize = size;
1860 				kq->kq_knlist = list;
1861 			}
1862 			KQ_UNLOCK(kq);
1863 		}
1864 	} else {
1865 		if (kq->kq_knhashmask == 0) {
1866 			tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE,
1867 			    &tmp_knhashmask, (mflag & M_WAITOK) != 0 ?
1868 			    HASH_WAITOK : HASH_NOWAIT);
1869 			if (tmp_knhash == NULL)
1870 				return (ENOMEM);
1871 			KQ_LOCK(kq);
1872 			if ((kq->kq_state & KQ_CLOSING) != 0) {
1873 				to_free = tmp_knhash;
1874 				error = EBADF;
1875 			} else if (kq->kq_knhashmask == 0) {
1876 				kq->kq_knhash = tmp_knhash;
1877 				kq->kq_knhashmask = tmp_knhashmask;
1878 			} else {
1879 				to_free = tmp_knhash;
1880 			}
1881 			KQ_UNLOCK(kq);
1882 		}
1883 	}
1884 	free(to_free, M_KQUEUE);
1885 
1886 	KQ_NOTOWNED(kq);
1887 	return (error);
1888 }
1889 
1890 static void
1891 kqueue_task(void *arg, int pending)
1892 {
1893 	struct kqueue *kq;
1894 	int haskqglobal;
1895 
1896 	haskqglobal = 0;
1897 	kq = arg;
1898 
1899 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1900 	KQ_LOCK(kq);
1901 
1902 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1903 
1904 	kq->kq_state &= ~KQ_TASKSCHED;
1905 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1906 		wakeup(&kq->kq_state);
1907 	}
1908 	KQ_UNLOCK(kq);
1909 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1910 }
1911 
1912 /*
1913  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1914  * We treat KN_MARKER knotes as if they are in flux.
1915  */
1916 static int
1917 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1918     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1919 {
1920 	struct kevent *kevp;
1921 	struct knote *kn, *marker;
1922 	struct knlist *knl;
1923 	sbintime_t asbt, rsbt;
1924 	int count, error, haskqglobal, influx, nkev, touch;
1925 
1926 	count = maxevents;
1927 	nkev = 0;
1928 	error = 0;
1929 	haskqglobal = 0;
1930 
1931 	if (maxevents == 0)
1932 		goto done_nl;
1933 	if (maxevents < 0) {
1934 		error = EINVAL;
1935 		goto done_nl;
1936 	}
1937 
1938 	rsbt = 0;
1939 	if (tsp != NULL) {
1940 		if (!timespecvalid_interval(tsp)) {
1941 			error = EINVAL;
1942 			goto done_nl;
1943 		}
1944 		if (timespecisset(tsp)) {
1945 			if (tsp->tv_sec <= INT32_MAX) {
1946 				rsbt = tstosbt(*tsp);
1947 				if (TIMESEL(&asbt, rsbt))
1948 					asbt += tc_tick_sbt;
1949 				if (asbt <= SBT_MAX - rsbt)
1950 					asbt += rsbt;
1951 				else
1952 					asbt = 0;
1953 				rsbt >>= tc_precexp;
1954 			} else
1955 				asbt = 0;
1956 		} else
1957 			asbt = -1;
1958 	} else
1959 		asbt = 0;
1960 	marker = knote_alloc(M_WAITOK);
1961 	marker->kn_status = KN_MARKER;
1962 	KQ_LOCK(kq);
1963 
1964 retry:
1965 	kevp = keva;
1966 	if (kq->kq_count == 0) {
1967 		if (asbt == -1) {
1968 			error = EWOULDBLOCK;
1969 		} else {
1970 			kq->kq_state |= KQ_SLEEP;
1971 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1972 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1973 		}
1974 		if (error == 0)
1975 			goto retry;
1976 		/* don't restart after signals... */
1977 		if (error == ERESTART)
1978 			error = EINTR;
1979 		else if (error == EWOULDBLOCK)
1980 			error = 0;
1981 		goto done;
1982 	}
1983 
1984 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1985 	influx = 0;
1986 	while (count) {
1987 		KQ_OWNED(kq);
1988 		kn = TAILQ_FIRST(&kq->kq_head);
1989 
1990 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1991 		    kn_in_flux(kn)) {
1992 			if (influx) {
1993 				influx = 0;
1994 				KQ_FLUX_WAKEUP(kq);
1995 			}
1996 			kq->kq_state |= KQ_FLUXWAIT;
1997 			error = msleep(kq, &kq->kq_lock, PSOCK,
1998 			    "kqflxwt", 0);
1999 			continue;
2000 		}
2001 
2002 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2003 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
2004 			kn->kn_status &= ~KN_QUEUED;
2005 			kq->kq_count--;
2006 			continue;
2007 		}
2008 		if (kn == marker) {
2009 			KQ_FLUX_WAKEUP(kq);
2010 			if (count == maxevents)
2011 				goto retry;
2012 			goto done;
2013 		}
2014 		KASSERT(!kn_in_flux(kn),
2015 		    ("knote %p is unexpectedly in flux", kn));
2016 
2017 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
2018 			kn->kn_status &= ~KN_QUEUED;
2019 			kn_enter_flux(kn);
2020 			kq->kq_count--;
2021 			KQ_UNLOCK(kq);
2022 			/*
2023 			 * We don't need to lock the list since we've
2024 			 * marked it as in flux.
2025 			 */
2026 			knote_drop(kn, td);
2027 			KQ_LOCK(kq);
2028 			continue;
2029 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
2030 			kn->kn_status &= ~KN_QUEUED;
2031 			kn_enter_flux(kn);
2032 			kq->kq_count--;
2033 			KQ_UNLOCK(kq);
2034 			/*
2035 			 * We don't need to lock the list since we've
2036 			 * marked the knote as being in flux.
2037 			 */
2038 			*kevp = kn->kn_kevent;
2039 			knote_drop(kn, td);
2040 			KQ_LOCK(kq);
2041 			kn = NULL;
2042 		} else {
2043 			kn->kn_status |= KN_SCAN;
2044 			kn_enter_flux(kn);
2045 			KQ_UNLOCK(kq);
2046 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
2047 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
2048 			knl = kn_list_lock(kn);
2049 			if (kn->kn_fop->f_event(kn, 0) == 0) {
2050 				KQ_LOCK(kq);
2051 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2052 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
2053 				    KN_SCAN);
2054 				kn_leave_flux(kn);
2055 				kq->kq_count--;
2056 				kn_list_unlock(knl);
2057 				influx = 1;
2058 				continue;
2059 			}
2060 			touch = (!kn->kn_fop->f_isfd &&
2061 			    kn->kn_fop->f_touch != NULL);
2062 			if (touch)
2063 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
2064 			else
2065 				*kevp = kn->kn_kevent;
2066 			KQ_LOCK(kq);
2067 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2068 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
2069 				/*
2070 				 * Manually clear knotes who weren't
2071 				 * 'touch'ed.
2072 				 */
2073 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
2074 					kn->kn_data = 0;
2075 					kn->kn_fflags = 0;
2076 				}
2077 				if (kn->kn_flags & EV_DISPATCH)
2078 					kn->kn_status |= KN_DISABLED;
2079 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
2080 				kq->kq_count--;
2081 			} else
2082 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2083 
2084 			kn->kn_status &= ~KN_SCAN;
2085 			kn_leave_flux(kn);
2086 			kn_list_unlock(knl);
2087 			influx = 1;
2088 		}
2089 
2090 		/* we are returning a copy to the user */
2091 		kevp++;
2092 		nkev++;
2093 		count--;
2094 
2095 		if (nkev == KQ_NEVENTS) {
2096 			influx = 0;
2097 			KQ_UNLOCK_FLUX(kq);
2098 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2099 			nkev = 0;
2100 			kevp = keva;
2101 			KQ_LOCK(kq);
2102 			if (error)
2103 				break;
2104 		}
2105 	}
2106 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
2107 done:
2108 	KQ_OWNED(kq);
2109 	KQ_UNLOCK_FLUX(kq);
2110 	knote_free(marker);
2111 done_nl:
2112 	KQ_NOTOWNED(kq);
2113 	if (nkev != 0)
2114 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2115 	td->td_retval[0] = maxevents - count;
2116 	return (error);
2117 }
2118 
2119 /*ARGSUSED*/
2120 static int
2121 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
2122 	struct ucred *active_cred, struct thread *td)
2123 {
2124 	/*
2125 	 * Enabling sigio causes two major problems:
2126 	 * 1) infinite recursion:
2127 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
2128 	 * set.  On receipt of a signal this will cause a kqueue to recurse
2129 	 * into itself over and over.  Sending the sigio causes the kqueue
2130 	 * to become ready, which in turn posts sigio again, forever.
2131 	 * Solution: this can be solved by setting a flag in the kqueue that
2132 	 * we have a SIGIO in progress.
2133 	 * 2) locking problems:
2134 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
2135 	 * us above the proc and pgrp locks.
2136 	 * Solution: Post a signal using an async mechanism, being sure to
2137 	 * record a generation count in the delivery so that we do not deliver
2138 	 * a signal to the wrong process.
2139 	 *
2140 	 * Note, these two mechanisms are somewhat mutually exclusive!
2141 	 */
2142 #if 0
2143 	struct kqueue *kq;
2144 
2145 	kq = fp->f_data;
2146 	switch (cmd) {
2147 	case FIOASYNC:
2148 		if (*(int *)data) {
2149 			kq->kq_state |= KQ_ASYNC;
2150 		} else {
2151 			kq->kq_state &= ~KQ_ASYNC;
2152 		}
2153 		return (0);
2154 
2155 	case FIOSETOWN:
2156 		return (fsetown(*(int *)data, &kq->kq_sigio));
2157 
2158 	case FIOGETOWN:
2159 		*(int *)data = fgetown(&kq->kq_sigio);
2160 		return (0);
2161 	}
2162 #endif
2163 
2164 	return (ENOTTY);
2165 }
2166 
2167 /*ARGSUSED*/
2168 static int
2169 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
2170 	struct thread *td)
2171 {
2172 	struct kqueue *kq;
2173 	int revents = 0;
2174 	int error;
2175 
2176 	if ((error = kqueue_acquire(fp, &kq)))
2177 		return POLLERR;
2178 
2179 	KQ_LOCK(kq);
2180 	if (events & (POLLIN | POLLRDNORM)) {
2181 		if (kq->kq_count) {
2182 			revents |= events & (POLLIN | POLLRDNORM);
2183 		} else {
2184 			selrecord(td, &kq->kq_sel);
2185 			if (SEL_WAITING(&kq->kq_sel))
2186 				kq->kq_state |= KQ_SEL;
2187 		}
2188 	}
2189 	kqueue_release(kq, 1);
2190 	KQ_UNLOCK(kq);
2191 	return (revents);
2192 }
2193 
2194 /*ARGSUSED*/
2195 static int
2196 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
2197 {
2198 
2199 	bzero((void *)st, sizeof *st);
2200 	/*
2201 	 * We no longer return kq_count because the unlocked value is useless.
2202 	 * If you spent all this time getting the count, why not spend your
2203 	 * syscall better by calling kevent?
2204 	 *
2205 	 * XXX - This is needed for libc_r.
2206 	 */
2207 	st->st_mode = S_IFIFO;
2208 	return (0);
2209 }
2210 
2211 static void
2212 kqueue_drain(struct kqueue *kq, struct thread *td)
2213 {
2214 	struct knote *kn;
2215 	int i;
2216 
2217 	KQ_LOCK(kq);
2218 
2219 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
2220 	    ("kqueue already closing"));
2221 	kq->kq_state |= KQ_CLOSING;
2222 	if (kq->kq_refcnt > 1)
2223 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
2224 
2225 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
2226 
2227 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
2228 	    ("kqueue's knlist not empty"));
2229 
2230 	for (i = 0; i < kq->kq_knlistsize; i++) {
2231 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
2232 			if (kn_in_flux(kn)) {
2233 				kq->kq_state |= KQ_FLUXWAIT;
2234 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
2235 				continue;
2236 			}
2237 			kn_enter_flux(kn);
2238 			KQ_UNLOCK(kq);
2239 			knote_drop(kn, td);
2240 			KQ_LOCK(kq);
2241 		}
2242 	}
2243 	if (kq->kq_knhashmask != 0) {
2244 		for (i = 0; i <= kq->kq_knhashmask; i++) {
2245 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
2246 				if (kn_in_flux(kn)) {
2247 					kq->kq_state |= KQ_FLUXWAIT;
2248 					msleep(kq, &kq->kq_lock, PSOCK,
2249 					       "kqclo2", 0);
2250 					continue;
2251 				}
2252 				kn_enter_flux(kn);
2253 				KQ_UNLOCK(kq);
2254 				knote_drop(kn, td);
2255 				KQ_LOCK(kq);
2256 			}
2257 		}
2258 	}
2259 
2260 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
2261 		kq->kq_state |= KQ_TASKDRAIN;
2262 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
2263 	}
2264 
2265 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2266 		selwakeuppri(&kq->kq_sel, PSOCK);
2267 		if (!SEL_WAITING(&kq->kq_sel))
2268 			kq->kq_state &= ~KQ_SEL;
2269 	}
2270 
2271 	KQ_UNLOCK(kq);
2272 }
2273 
2274 static void
2275 kqueue_destroy(struct kqueue *kq)
2276 {
2277 
2278 	KASSERT(kq->kq_fdp == NULL,
2279 	    ("kqueue still attached to a file descriptor"));
2280 	seldrain(&kq->kq_sel);
2281 	knlist_destroy(&kq->kq_sel.si_note);
2282 	mtx_destroy(&kq->kq_lock);
2283 
2284 	if (kq->kq_knhash != NULL)
2285 		free(kq->kq_knhash, M_KQUEUE);
2286 	if (kq->kq_knlist != NULL)
2287 		free(kq->kq_knlist, M_KQUEUE);
2288 
2289 	funsetown(&kq->kq_sigio);
2290 }
2291 
2292 /*ARGSUSED*/
2293 static int
2294 kqueue_close(struct file *fp, struct thread *td)
2295 {
2296 	struct kqueue *kq = fp->f_data;
2297 	struct filedesc *fdp;
2298 	int error;
2299 	int filedesc_unlock;
2300 
2301 	if ((error = kqueue_acquire(fp, &kq)))
2302 		return error;
2303 	kqueue_drain(kq, td);
2304 
2305 	/*
2306 	 * We could be called due to the knote_drop() doing fdrop(),
2307 	 * called from kqueue_register().  In this case the global
2308 	 * lock is owned, and filedesc sx is locked before, to not
2309 	 * take the sleepable lock after non-sleepable.
2310 	 */
2311 	fdp = kq->kq_fdp;
2312 	kq->kq_fdp = NULL;
2313 	if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
2314 		FILEDESC_XLOCK(fdp);
2315 		filedesc_unlock = 1;
2316 	} else
2317 		filedesc_unlock = 0;
2318 	TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
2319 	if (filedesc_unlock)
2320 		FILEDESC_XUNLOCK(fdp);
2321 
2322 	kqueue_destroy(kq);
2323 	chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
2324 	crfree(kq->kq_cred);
2325 	free(kq, M_KQUEUE);
2326 	fp->f_data = NULL;
2327 
2328 	return (0);
2329 }
2330 
2331 static int
2332 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2333 {
2334 	struct kqueue *kq = fp->f_data;
2335 
2336 	kif->kf_type = KF_TYPE_KQUEUE;
2337 	kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq;
2338 	kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count;
2339 	kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state;
2340 	return (0);
2341 }
2342 
2343 static void
2344 kqueue_wakeup(struct kqueue *kq)
2345 {
2346 	KQ_OWNED(kq);
2347 
2348 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
2349 		kq->kq_state &= ~KQ_SLEEP;
2350 		wakeup(kq);
2351 	}
2352 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2353 		selwakeuppri(&kq->kq_sel, PSOCK);
2354 		if (!SEL_WAITING(&kq->kq_sel))
2355 			kq->kq_state &= ~KQ_SEL;
2356 	}
2357 	if (!knlist_empty(&kq->kq_sel.si_note))
2358 		kqueue_schedtask(kq);
2359 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
2360 		pgsigio(&kq->kq_sigio, SIGIO, 0);
2361 	}
2362 }
2363 
2364 /*
2365  * Walk down a list of knotes, activating them if their event has triggered.
2366  *
2367  * There is a possibility to optimize in the case of one kq watching another.
2368  * Instead of scheduling a task to wake it up, you could pass enough state
2369  * down the chain to make up the parent kqueue.  Make this code functional
2370  * first.
2371  */
2372 void
2373 knote(struct knlist *list, long hint, int lockflags)
2374 {
2375 	struct kqueue *kq;
2376 	struct knote *kn, *tkn;
2377 	int error;
2378 
2379 	if (list == NULL)
2380 		return;
2381 
2382 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
2383 
2384 	if ((lockflags & KNF_LISTLOCKED) == 0)
2385 		list->kl_lock(list->kl_lockarg);
2386 
2387 	/*
2388 	 * If we unlock the list lock (and enter influx), we can
2389 	 * eliminate the kqueue scheduling, but this will introduce
2390 	 * four lock/unlock's for each knote to test.  Also, marker
2391 	 * would be needed to keep iteration position, since filters
2392 	 * or other threads could remove events.
2393 	 */
2394 	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
2395 		kq = kn->kn_kq;
2396 		KQ_LOCK(kq);
2397 		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
2398 			/*
2399 			 * Do not process the influx notes, except for
2400 			 * the influx coming from the kq unlock in the
2401 			 * kqueue_scan().  In the later case, we do
2402 			 * not interfere with the scan, since the code
2403 			 * fragment in kqueue_scan() locks the knlist,
2404 			 * and cannot proceed until we finished.
2405 			 */
2406 			KQ_UNLOCK(kq);
2407 		} else if ((lockflags & KNF_NOKQLOCK) != 0) {
2408 			kn_enter_flux(kn);
2409 			KQ_UNLOCK(kq);
2410 			error = kn->kn_fop->f_event(kn, hint);
2411 			KQ_LOCK(kq);
2412 			kn_leave_flux(kn);
2413 			if (error)
2414 				KNOTE_ACTIVATE(kn, 1);
2415 			KQ_UNLOCK_FLUX(kq);
2416 		} else {
2417 			if (kn->kn_fop->f_event(kn, hint))
2418 				KNOTE_ACTIVATE(kn, 1);
2419 			KQ_UNLOCK(kq);
2420 		}
2421 	}
2422 	if ((lockflags & KNF_LISTLOCKED) == 0)
2423 		list->kl_unlock(list->kl_lockarg);
2424 }
2425 
2426 /*
2427  * add a knote to a knlist
2428  */
2429 void
2430 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
2431 {
2432 
2433 	KNL_ASSERT_LOCK(knl, islocked);
2434 	KQ_NOTOWNED(kn->kn_kq);
2435 	KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn));
2436 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2437 	    ("knote %p was not detached", kn));
2438 	if (!islocked)
2439 		knl->kl_lock(knl->kl_lockarg);
2440 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
2441 	if (!islocked)
2442 		knl->kl_unlock(knl->kl_lockarg);
2443 	KQ_LOCK(kn->kn_kq);
2444 	kn->kn_knlist = knl;
2445 	kn->kn_status &= ~KN_DETACHED;
2446 	KQ_UNLOCK(kn->kn_kq);
2447 }
2448 
2449 static void
2450 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
2451     int kqislocked)
2452 {
2453 
2454 	KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked"));
2455 	KNL_ASSERT_LOCK(knl, knlislocked);
2456 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
2457 	KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn));
2458 	KASSERT((kn->kn_status & KN_DETACHED) == 0,
2459 	    ("knote %p was already detached", kn));
2460 	if (!knlislocked)
2461 		knl->kl_lock(knl->kl_lockarg);
2462 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
2463 	kn->kn_knlist = NULL;
2464 	if (!knlislocked)
2465 		kn_list_unlock(knl);
2466 	if (!kqislocked)
2467 		KQ_LOCK(kn->kn_kq);
2468 	kn->kn_status |= KN_DETACHED;
2469 	if (!kqislocked)
2470 		KQ_UNLOCK(kn->kn_kq);
2471 }
2472 
2473 /*
2474  * remove knote from the specified knlist
2475  */
2476 void
2477 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
2478 {
2479 
2480 	knlist_remove_kq(knl, kn, islocked, 0);
2481 }
2482 
2483 int
2484 knlist_empty(struct knlist *knl)
2485 {
2486 
2487 	KNL_ASSERT_LOCKED(knl);
2488 	return (SLIST_EMPTY(&knl->kl_list));
2489 }
2490 
2491 static struct mtx knlist_lock;
2492 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2493     MTX_DEF);
2494 static void knlist_mtx_lock(void *arg);
2495 static void knlist_mtx_unlock(void *arg);
2496 
2497 static void
2498 knlist_mtx_lock(void *arg)
2499 {
2500 
2501 	mtx_lock((struct mtx *)arg);
2502 }
2503 
2504 static void
2505 knlist_mtx_unlock(void *arg)
2506 {
2507 
2508 	mtx_unlock((struct mtx *)arg);
2509 }
2510 
2511 static void
2512 knlist_mtx_assert_lock(void *arg, int what)
2513 {
2514 
2515 	if (what == LA_LOCKED)
2516 		mtx_assert((struct mtx *)arg, MA_OWNED);
2517 	else
2518 		mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2519 }
2520 
2521 void
2522 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2523     void (*kl_unlock)(void *),
2524     void (*kl_assert_lock)(void *, int))
2525 {
2526 
2527 	if (lock == NULL)
2528 		knl->kl_lockarg = &knlist_lock;
2529 	else
2530 		knl->kl_lockarg = lock;
2531 
2532 	if (kl_lock == NULL)
2533 		knl->kl_lock = knlist_mtx_lock;
2534 	else
2535 		knl->kl_lock = kl_lock;
2536 	if (kl_unlock == NULL)
2537 		knl->kl_unlock = knlist_mtx_unlock;
2538 	else
2539 		knl->kl_unlock = kl_unlock;
2540 	if (kl_assert_lock == NULL)
2541 		knl->kl_assert_lock = knlist_mtx_assert_lock;
2542 	else
2543 		knl->kl_assert_lock = kl_assert_lock;
2544 
2545 	knl->kl_autodestroy = 0;
2546 	SLIST_INIT(&knl->kl_list);
2547 }
2548 
2549 void
2550 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2551 {
2552 
2553 	knlist_init(knl, lock, NULL, NULL, NULL);
2554 }
2555 
2556 struct knlist *
2557 knlist_alloc(struct mtx *lock)
2558 {
2559 	struct knlist *knl;
2560 
2561 	knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
2562 	knlist_init_mtx(knl, lock);
2563 	return (knl);
2564 }
2565 
2566 void
2567 knlist_destroy(struct knlist *knl)
2568 {
2569 
2570 	KASSERT(KNLIST_EMPTY(knl),
2571 	    ("destroying knlist %p with knotes on it", knl));
2572 }
2573 
2574 void
2575 knlist_detach(struct knlist *knl)
2576 {
2577 
2578 	KNL_ASSERT_LOCKED(knl);
2579 	knl->kl_autodestroy = 1;
2580 	if (knlist_empty(knl)) {
2581 		knlist_destroy(knl);
2582 		free(knl, M_KQUEUE);
2583 	}
2584 }
2585 
2586 /*
2587  * Even if we are locked, we may need to drop the lock to allow any influx
2588  * knotes time to "settle".
2589  */
2590 void
2591 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2592 {
2593 	struct knote *kn, *kn2;
2594 	struct kqueue *kq;
2595 
2596 	KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
2597 	if (islocked)
2598 		KNL_ASSERT_LOCKED(knl);
2599 	else {
2600 		KNL_ASSERT_UNLOCKED(knl);
2601 again:		/* need to reacquire lock since we have dropped it */
2602 		knl->kl_lock(knl->kl_lockarg);
2603 	}
2604 
2605 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2606 		kq = kn->kn_kq;
2607 		KQ_LOCK(kq);
2608 		if (kn_in_flux(kn)) {
2609 			KQ_UNLOCK(kq);
2610 			continue;
2611 		}
2612 		knlist_remove_kq(knl, kn, 1, 1);
2613 		if (killkn) {
2614 			kn_enter_flux(kn);
2615 			KQ_UNLOCK(kq);
2616 			knote_drop_detached(kn, td);
2617 		} else {
2618 			/* Make sure cleared knotes disappear soon */
2619 			kn->kn_flags |= EV_EOF | EV_ONESHOT;
2620 			KQ_UNLOCK(kq);
2621 		}
2622 		kq = NULL;
2623 	}
2624 
2625 	if (!SLIST_EMPTY(&knl->kl_list)) {
2626 		/* there are still in flux knotes remaining */
2627 		kn = SLIST_FIRST(&knl->kl_list);
2628 		kq = kn->kn_kq;
2629 		KQ_LOCK(kq);
2630 		KASSERT(kn_in_flux(kn), ("knote removed w/o list lock"));
2631 		knl->kl_unlock(knl->kl_lockarg);
2632 		kq->kq_state |= KQ_FLUXWAIT;
2633 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2634 		kq = NULL;
2635 		goto again;
2636 	}
2637 
2638 	if (islocked)
2639 		KNL_ASSERT_LOCKED(knl);
2640 	else {
2641 		knl->kl_unlock(knl->kl_lockarg);
2642 		KNL_ASSERT_UNLOCKED(knl);
2643 	}
2644 }
2645 
2646 /*
2647  * Remove all knotes referencing a specified fd must be called with FILEDESC
2648  * lock.  This prevents a race where a new fd comes along and occupies the
2649  * entry and we attach a knote to the fd.
2650  */
2651 void
2652 knote_fdclose(struct thread *td, int fd)
2653 {
2654 	struct filedesc *fdp = td->td_proc->p_fd;
2655 	struct kqueue *kq;
2656 	struct knote *kn;
2657 	int influx;
2658 
2659 	FILEDESC_XLOCK_ASSERT(fdp);
2660 
2661 	/*
2662 	 * We shouldn't have to worry about new kevents appearing on fd
2663 	 * since filedesc is locked.
2664 	 */
2665 	TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2666 		KQ_LOCK(kq);
2667 
2668 again:
2669 		influx = 0;
2670 		while (kq->kq_knlistsize > fd &&
2671 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2672 			if (kn_in_flux(kn)) {
2673 				/* someone else might be waiting on our knote */
2674 				if (influx)
2675 					wakeup(kq);
2676 				kq->kq_state |= KQ_FLUXWAIT;
2677 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2678 				goto again;
2679 			}
2680 			kn_enter_flux(kn);
2681 			KQ_UNLOCK(kq);
2682 			influx = 1;
2683 			knote_drop(kn, td);
2684 			KQ_LOCK(kq);
2685 		}
2686 		KQ_UNLOCK_FLUX(kq);
2687 	}
2688 }
2689 
2690 static int
2691 knote_attach(struct knote *kn, struct kqueue *kq)
2692 {
2693 	struct klist *list;
2694 
2695 	KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn));
2696 	KQ_OWNED(kq);
2697 
2698 	if ((kq->kq_state & KQ_CLOSING) != 0)
2699 		return (EBADF);
2700 	if (kn->kn_fop->f_isfd) {
2701 		if (kn->kn_id >= kq->kq_knlistsize)
2702 			return (ENOMEM);
2703 		list = &kq->kq_knlist[kn->kn_id];
2704 	} else {
2705 		if (kq->kq_knhash == NULL)
2706 			return (ENOMEM);
2707 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2708 	}
2709 	SLIST_INSERT_HEAD(list, kn, kn_link);
2710 	return (0);
2711 }
2712 
2713 static void
2714 knote_drop(struct knote *kn, struct thread *td)
2715 {
2716 
2717 	if ((kn->kn_status & KN_DETACHED) == 0)
2718 		kn->kn_fop->f_detach(kn);
2719 	knote_drop_detached(kn, td);
2720 }
2721 
2722 static void
2723 knote_drop_detached(struct knote *kn, struct thread *td)
2724 {
2725 	struct kqueue *kq;
2726 	struct klist *list;
2727 
2728 	kq = kn->kn_kq;
2729 
2730 	KASSERT((kn->kn_status & KN_DETACHED) != 0,
2731 	    ("knote %p still attached", kn));
2732 	KQ_NOTOWNED(kq);
2733 
2734 	KQ_LOCK(kq);
2735 	KASSERT(kn->kn_influx == 1,
2736 	    ("knote_drop called on %p with influx %d", kn, kn->kn_influx));
2737 
2738 	if (kn->kn_fop->f_isfd)
2739 		list = &kq->kq_knlist[kn->kn_id];
2740 	else
2741 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2742 
2743 	if (!SLIST_EMPTY(list))
2744 		SLIST_REMOVE(list, kn, knote, kn_link);
2745 	if (kn->kn_status & KN_QUEUED)
2746 		knote_dequeue(kn);
2747 	KQ_UNLOCK_FLUX(kq);
2748 
2749 	if (kn->kn_fop->f_isfd) {
2750 		fdrop(kn->kn_fp, td);
2751 		kn->kn_fp = NULL;
2752 	}
2753 	kqueue_fo_release(kn->kn_kevent.filter);
2754 	kn->kn_fop = NULL;
2755 	knote_free(kn);
2756 }
2757 
2758 static void
2759 knote_enqueue(struct knote *kn)
2760 {
2761 	struct kqueue *kq = kn->kn_kq;
2762 
2763 	KQ_OWNED(kn->kn_kq);
2764 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2765 
2766 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2767 	kn->kn_status |= KN_QUEUED;
2768 	kq->kq_count++;
2769 	kqueue_wakeup(kq);
2770 }
2771 
2772 static void
2773 knote_dequeue(struct knote *kn)
2774 {
2775 	struct kqueue *kq = kn->kn_kq;
2776 
2777 	KQ_OWNED(kn->kn_kq);
2778 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2779 
2780 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2781 	kn->kn_status &= ~KN_QUEUED;
2782 	kq->kq_count--;
2783 }
2784 
2785 static void
2786 knote_init(void)
2787 {
2788 
2789 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2790 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2791 	ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue);
2792 }
2793 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2794 
2795 static struct knote *
2796 knote_alloc(int mflag)
2797 {
2798 
2799 	return (uma_zalloc(knote_zone, mflag | M_ZERO));
2800 }
2801 
2802 static void
2803 knote_free(struct knote *kn)
2804 {
2805 
2806 	uma_zfree(knote_zone, kn);
2807 }
2808 
2809 /*
2810  * Register the kev w/ the kq specified by fd.
2811  */
2812 int
2813 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag)
2814 {
2815 	struct kqueue *kq;
2816 	struct file *fp;
2817 	cap_rights_t rights;
2818 	int error;
2819 
2820 	error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE),
2821 	    &fp);
2822 	if (error != 0)
2823 		return (error);
2824 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2825 		goto noacquire;
2826 
2827 	error = kqueue_register(kq, kev, td, mflag);
2828 	kqueue_release(kq, 0);
2829 
2830 noacquire:
2831 	fdrop(fp, td);
2832 	return (error);
2833 }
2834