xref: /openbsd-src/sys/kern/kern_event.c (revision eadad9f1af94df863ba43435828df1c86eadd88d)
1 /*	$OpenBSD: kern_event.c,v 1.194 2022/11/09 22:25:36 claudio Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD: src/sys/kern/kern_event.c,v 1.22 2001/02/23 20:32:42 jlemon Exp $
29  */
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/proc.h>
34 #include <sys/pledge.h>
35 #include <sys/malloc.h>
36 #include <sys/file.h>
37 #include <sys/filedesc.h>
38 #include <sys/fcntl.h>
39 #include <sys/queue.h>
40 #include <sys/event.h>
41 #include <sys/eventvar.h>
42 #include <sys/ktrace.h>
43 #include <sys/pool.h>
44 #include <sys/stat.h>
45 #include <sys/mount.h>
46 #include <sys/syscallargs.h>
47 #include <sys/time.h>
48 #include <sys/timeout.h>
49 #include <sys/vnode.h>
50 #include <sys/wait.h>
51 
52 #ifdef DIAGNOSTIC
53 #define KLIST_ASSERT_LOCKED(kl) do {					\
54 	if ((kl)->kl_ops != NULL)					\
55 		(kl)->kl_ops->klo_assertlk((kl)->kl_arg);		\
56 	else								\
57 		KERNEL_ASSERT_LOCKED();					\
58 } while (0)
59 #else
60 #define KLIST_ASSERT_LOCKED(kl)	((void)(kl))
61 #endif
62 
63 struct	kqueue *kqueue_alloc(struct filedesc *);
64 void	kqueue_terminate(struct proc *p, struct kqueue *);
65 void	KQREF(struct kqueue *);
66 void	KQRELE(struct kqueue *);
67 
68 void	kqueue_purge(struct proc *, struct kqueue *);
69 int	kqueue_sleep(struct kqueue *, struct timespec *);
70 
71 int	kqueue_read(struct file *, struct uio *, int);
72 int	kqueue_write(struct file *, struct uio *, int);
73 int	kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
74 		    struct proc *p);
75 int	kqueue_kqfilter(struct file *fp, struct knote *kn);
76 int	kqueue_stat(struct file *fp, struct stat *st, struct proc *p);
77 int	kqueue_close(struct file *fp, struct proc *p);
78 void	kqueue_wakeup(struct kqueue *kq);
79 
80 #ifdef KQUEUE_DEBUG
81 void	kqueue_do_check(struct kqueue *kq, const char *func, int line);
82 #define kqueue_check(kq)	kqueue_do_check((kq), __func__, __LINE__)
83 #else
84 #define kqueue_check(kq)	do {} while (0)
85 #endif
86 
87 static int	filter_attach(struct knote *kn);
88 static void	filter_detach(struct knote *kn);
89 static int	filter_event(struct knote *kn, long hint);
90 static int	filter_modify(struct kevent *kev, struct knote *kn);
91 static int	filter_process(struct knote *kn, struct kevent *kev);
92 static void	kqueue_expand_hash(struct kqueue *kq);
93 static void	kqueue_expand_list(struct kqueue *kq, int fd);
94 static void	kqueue_task(void *);
95 static int	klist_lock(struct klist *);
96 static void	klist_unlock(struct klist *, int);
97 
98 const struct fileops kqueueops = {
99 	.fo_read	= kqueue_read,
100 	.fo_write	= kqueue_write,
101 	.fo_ioctl	= kqueue_ioctl,
102 	.fo_kqfilter	= kqueue_kqfilter,
103 	.fo_stat	= kqueue_stat,
104 	.fo_close	= kqueue_close
105 };
106 
107 void	knote_attach(struct knote *kn);
108 void	knote_detach(struct knote *kn);
109 void	knote_drop(struct knote *kn, struct proc *p);
110 void	knote_enqueue(struct knote *kn);
111 void	knote_dequeue(struct knote *kn);
112 int	knote_acquire(struct knote *kn, struct klist *, int);
113 void	knote_release(struct knote *kn);
114 void	knote_activate(struct knote *kn);
115 void	knote_remove(struct proc *p, struct kqueue *kq, struct knlist **plist,
116 	    int idx, int purge);
117 
118 void	filt_kqdetach(struct knote *kn);
119 int	filt_kqueue(struct knote *kn, long hint);
120 int	filt_kqueuemodify(struct kevent *kev, struct knote *kn);
121 int	filt_kqueueprocess(struct knote *kn, struct kevent *kev);
122 int	filt_kqueue_common(struct knote *kn, struct kqueue *kq);
123 int	filt_procattach(struct knote *kn);
124 void	filt_procdetach(struct knote *kn);
125 int	filt_proc(struct knote *kn, long hint);
126 int	filt_fileattach(struct knote *kn);
127 void	filt_timerexpire(void *knx);
128 int	filt_timerattach(struct knote *kn);
129 void	filt_timerdetach(struct knote *kn);
130 int	filt_timermodify(struct kevent *kev, struct knote *kn);
131 int	filt_timerprocess(struct knote *kn, struct kevent *kev);
132 void	filt_seltruedetach(struct knote *kn);
133 
134 const struct filterops kqread_filtops = {
135 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
136 	.f_attach	= NULL,
137 	.f_detach	= filt_kqdetach,
138 	.f_event	= filt_kqueue,
139 	.f_modify	= filt_kqueuemodify,
140 	.f_process	= filt_kqueueprocess,
141 };
142 
143 const struct filterops proc_filtops = {
144 	.f_flags	= 0,
145 	.f_attach	= filt_procattach,
146 	.f_detach	= filt_procdetach,
147 	.f_event	= filt_proc,
148 };
149 
150 const struct filterops file_filtops = {
151 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
152 	.f_attach	= filt_fileattach,
153 	.f_detach	= NULL,
154 	.f_event	= NULL,
155 };
156 
157 const struct filterops timer_filtops = {
158 	.f_flags	= 0,
159 	.f_attach	= filt_timerattach,
160 	.f_detach	= filt_timerdetach,
161 	.f_event	= NULL,
162 	.f_modify	= filt_timermodify,
163 	.f_process	= filt_timerprocess,
164 };
165 
166 struct	pool knote_pool;
167 struct	pool kqueue_pool;
168 struct	mutex kqueue_klist_lock = MUTEX_INITIALIZER(IPL_MPFLOOR);
169 int kq_ntimeouts = 0;
170 int kq_timeoutmax = (4 * 1024);
171 
172 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
173 
174 /*
175  * Table for for all system-defined filters.
176  */
177 const struct filterops *const sysfilt_ops[] = {
178 	&file_filtops,			/* EVFILT_READ */
179 	&file_filtops,			/* EVFILT_WRITE */
180 	NULL, /*&aio_filtops,*/		/* EVFILT_AIO */
181 	&file_filtops,			/* EVFILT_VNODE */
182 	&proc_filtops,			/* EVFILT_PROC */
183 	&sig_filtops,			/* EVFILT_SIGNAL */
184 	&timer_filtops,			/* EVFILT_TIMER */
185 	&file_filtops,			/* EVFILT_DEVICE */
186 	&file_filtops,			/* EVFILT_EXCEPT */
187 };
188 
189 void
190 KQREF(struct kqueue *kq)
191 {
192 	refcnt_take(&kq->kq_refcnt);
193 }
194 
195 void
196 KQRELE(struct kqueue *kq)
197 {
198 	struct filedesc *fdp;
199 
200 	if (refcnt_rele(&kq->kq_refcnt) == 0)
201 		return;
202 
203 	fdp = kq->kq_fdp;
204 	if (rw_status(&fdp->fd_lock) == RW_WRITE) {
205 		LIST_REMOVE(kq, kq_next);
206 	} else {
207 		fdplock(fdp);
208 		LIST_REMOVE(kq, kq_next);
209 		fdpunlock(fdp);
210 	}
211 
212 	KASSERT(TAILQ_EMPTY(&kq->kq_head));
213 	KASSERT(kq->kq_nknotes == 0);
214 
215 	free(kq->kq_knlist, M_KEVENT, kq->kq_knlistsize *
216 	    sizeof(struct knlist));
217 	hashfree(kq->kq_knhash, KN_HASHSIZE, M_KEVENT);
218 	klist_free(&kq->kq_klist);
219 	pool_put(&kqueue_pool, kq);
220 }
221 
222 void
223 kqueue_init(void)
224 {
225 	pool_init(&kqueue_pool, sizeof(struct kqueue), 0, IPL_MPFLOOR,
226 	    PR_WAITOK, "kqueuepl", NULL);
227 	pool_init(&knote_pool, sizeof(struct knote), 0, IPL_MPFLOOR,
228 	    PR_WAITOK, "knotepl", NULL);
229 }
230 
231 void
232 kqueue_init_percpu(void)
233 {
234 	pool_cache_init(&knote_pool);
235 }
236 
237 int
238 filt_fileattach(struct knote *kn)
239 {
240 	struct file *fp = kn->kn_fp;
241 
242 	return fp->f_ops->fo_kqfilter(fp, kn);
243 }
244 
245 int
246 kqueue_kqfilter(struct file *fp, struct knote *kn)
247 {
248 	struct kqueue *kq = kn->kn_fp->f_data;
249 
250 	if (kn->kn_filter != EVFILT_READ)
251 		return (EINVAL);
252 
253 	kn->kn_fop = &kqread_filtops;
254 	klist_insert(&kq->kq_klist, kn);
255 	return (0);
256 }
257 
258 void
259 filt_kqdetach(struct knote *kn)
260 {
261 	struct kqueue *kq = kn->kn_fp->f_data;
262 
263 	klist_remove(&kq->kq_klist, kn);
264 }
265 
266 int
267 filt_kqueue_common(struct knote *kn, struct kqueue *kq)
268 {
269 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
270 
271 	kn->kn_data = kq->kq_count;
272 
273 	return (kn->kn_data > 0);
274 }
275 
276 int
277 filt_kqueue(struct knote *kn, long hint)
278 {
279 	struct kqueue *kq = kn->kn_fp->f_data;
280 	int active;
281 
282 	mtx_enter(&kq->kq_lock);
283 	active = filt_kqueue_common(kn, kq);
284 	mtx_leave(&kq->kq_lock);
285 
286 	return (active);
287 }
288 
289 int
290 filt_kqueuemodify(struct kevent *kev, struct knote *kn)
291 {
292 	struct kqueue *kq = kn->kn_fp->f_data;
293 	int active;
294 
295 	mtx_enter(&kq->kq_lock);
296 	knote_assign(kev, kn);
297 	active = filt_kqueue_common(kn, kq);
298 	mtx_leave(&kq->kq_lock);
299 
300 	return (active);
301 }
302 
303 int
304 filt_kqueueprocess(struct knote *kn, struct kevent *kev)
305 {
306 	struct kqueue *kq = kn->kn_fp->f_data;
307 	int active;
308 
309 	mtx_enter(&kq->kq_lock);
310 	if (kev != NULL && (kn->kn_flags & EV_ONESHOT))
311 		active = 1;
312 	else
313 		active = filt_kqueue_common(kn, kq);
314 	if (active)
315 		knote_submit(kn, kev);
316 	mtx_leave(&kq->kq_lock);
317 
318 	return (active);
319 }
320 
321 int
322 filt_procattach(struct knote *kn)
323 {
324 	struct process *pr;
325 	int s;
326 
327 	if ((curproc->p_p->ps_flags & PS_PLEDGE) &&
328 	    (curproc->p_p->ps_pledge & PLEDGE_PROC) == 0)
329 		return pledge_fail(curproc, EPERM, PLEDGE_PROC);
330 
331 	if (kn->kn_id > PID_MAX)
332 		return ESRCH;
333 
334 	pr = prfind(kn->kn_id);
335 	if (pr == NULL)
336 		return (ESRCH);
337 
338 	/* exiting processes can't be specified */
339 	if (pr->ps_flags & PS_EXITING)
340 		return (ESRCH);
341 
342 	kn->kn_ptr.p_process = pr;
343 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
344 
345 	/*
346 	 * internal flag indicating registration done by kernel
347 	 */
348 	if (kn->kn_flags & EV_FLAG1) {
349 		kn->kn_data = kn->kn_sdata;		/* ppid */
350 		kn->kn_fflags = NOTE_CHILD;
351 		kn->kn_flags &= ~EV_FLAG1;
352 	}
353 
354 	s = splhigh();
355 	klist_insert_locked(&pr->ps_klist, kn);
356 	splx(s);
357 
358 	return (0);
359 }
360 
361 /*
362  * The knote may be attached to a different process, which may exit,
363  * leaving nothing for the knote to be attached to.  So when the process
364  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
365  * it will be deleted when read out.  However, as part of the knote deletion,
366  * this routine is called, so a check is needed to avoid actually performing
367  * a detach, because the original process does not exist any more.
368  */
369 void
370 filt_procdetach(struct knote *kn)
371 {
372 	struct kqueue *kq = kn->kn_kq;
373 	struct process *pr = kn->kn_ptr.p_process;
374 	int s, status;
375 
376 	mtx_enter(&kq->kq_lock);
377 	status = kn->kn_status;
378 	mtx_leave(&kq->kq_lock);
379 
380 	if (status & KN_DETACHED)
381 		return;
382 
383 	s = splhigh();
384 	klist_remove_locked(&pr->ps_klist, kn);
385 	splx(s);
386 }
387 
388 int
389 filt_proc(struct knote *kn, long hint)
390 {
391 	struct kqueue *kq = kn->kn_kq;
392 	u_int event;
393 
394 	/*
395 	 * mask off extra data
396 	 */
397 	event = (u_int)hint & NOTE_PCTRLMASK;
398 
399 	/*
400 	 * if the user is interested in this event, record it.
401 	 */
402 	if (kn->kn_sfflags & event)
403 		kn->kn_fflags |= event;
404 
405 	/*
406 	 * process is gone, so flag the event as finished and remove it
407 	 * from the process's klist
408 	 */
409 	if (event == NOTE_EXIT) {
410 		struct process *pr = kn->kn_ptr.p_process;
411 		int s;
412 
413 		mtx_enter(&kq->kq_lock);
414 		kn->kn_status |= KN_DETACHED;
415 		mtx_leave(&kq->kq_lock);
416 
417 		s = splhigh();
418 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
419 		kn->kn_data = W_EXITCODE(pr->ps_xexit, pr->ps_xsig);
420 		klist_remove_locked(&pr->ps_klist, kn);
421 		splx(s);
422 		return (1);
423 	}
424 
425 	/*
426 	 * process forked, and user wants to track the new process,
427 	 * so attach a new knote to it, and immediately report an
428 	 * event with the parent's pid.
429 	 */
430 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
431 		struct kevent kev;
432 		int error;
433 
434 		/*
435 		 * register knote with new process.
436 		 */
437 		memset(&kev, 0, sizeof(kev));
438 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
439 		kev.filter = kn->kn_filter;
440 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
441 		kev.fflags = kn->kn_sfflags;
442 		kev.data = kn->kn_id;			/* parent */
443 		kev.udata = kn->kn_udata;		/* preserve udata */
444 		error = kqueue_register(kq, &kev, 0, NULL);
445 		if (error)
446 			kn->kn_fflags |= NOTE_TRACKERR;
447 	}
448 
449 	return (kn->kn_fflags != 0);
450 }
451 
452 static void
453 filt_timer_timeout_add(struct knote *kn)
454 {
455 	struct timeval tv;
456 	struct timeout *to = kn->kn_hook;
457 	int tticks;
458 
459 	tv.tv_sec = kn->kn_sdata / 1000;
460 	tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
461 	tticks = tvtohz(&tv);
462 	/* Remove extra tick from tvtohz() if timeout has fired before. */
463 	if (timeout_triggered(to))
464 		tticks--;
465 	timeout_add(to, (tticks > 0) ? tticks : 1);
466 }
467 
468 void
469 filt_timerexpire(void *knx)
470 {
471 	struct knote *kn = knx;
472 	struct kqueue *kq = kn->kn_kq;
473 
474 	kn->kn_data++;
475 	mtx_enter(&kq->kq_lock);
476 	knote_activate(kn);
477 	mtx_leave(&kq->kq_lock);
478 
479 	if ((kn->kn_flags & EV_ONESHOT) == 0)
480 		filt_timer_timeout_add(kn);
481 }
482 
483 
484 /*
485  * data contains amount of time to sleep, in milliseconds
486  */
487 int
488 filt_timerattach(struct knote *kn)
489 {
490 	struct timeout *to;
491 
492 	if (kq_ntimeouts > kq_timeoutmax)
493 		return (ENOMEM);
494 	kq_ntimeouts++;
495 
496 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
497 	to = malloc(sizeof(*to), M_KEVENT, M_WAITOK);
498 	timeout_set(to, filt_timerexpire, kn);
499 	kn->kn_hook = to;
500 	filt_timer_timeout_add(kn);
501 
502 	return (0);
503 }
504 
505 void
506 filt_timerdetach(struct knote *kn)
507 {
508 	struct timeout *to;
509 
510 	to = (struct timeout *)kn->kn_hook;
511 	timeout_del_barrier(to);
512 	free(to, M_KEVENT, sizeof(*to));
513 	kq_ntimeouts--;
514 }
515 
516 int
517 filt_timermodify(struct kevent *kev, struct knote *kn)
518 {
519 	struct kqueue *kq = kn->kn_kq;
520 	struct timeout *to = kn->kn_hook;
521 
522 	/* Reset the timer. Any pending events are discarded. */
523 
524 	timeout_del_barrier(to);
525 
526 	mtx_enter(&kq->kq_lock);
527 	if (kn->kn_status & KN_QUEUED)
528 		knote_dequeue(kn);
529 	kn->kn_status &= ~KN_ACTIVE;
530 	mtx_leave(&kq->kq_lock);
531 
532 	kn->kn_data = 0;
533 	knote_assign(kev, kn);
534 	/* Reinit timeout to invoke tick adjustment again. */
535 	timeout_set(to, filt_timerexpire, kn);
536 	filt_timer_timeout_add(kn);
537 
538 	return (0);
539 }
540 
541 int
542 filt_timerprocess(struct knote *kn, struct kevent *kev)
543 {
544 	int active, s;
545 
546 	s = splsoftclock();
547 	active = (kn->kn_data != 0);
548 	if (active)
549 		knote_submit(kn, kev);
550 	splx(s);
551 
552 	return (active);
553 }
554 
555 
556 /*
557  * filt_seltrue:
558  *
559  *	This filter "event" routine simulates seltrue().
560  */
561 int
562 filt_seltrue(struct knote *kn, long hint)
563 {
564 
565 	/*
566 	 * We don't know how much data can be read/written,
567 	 * but we know that it *can* be.  This is about as
568 	 * good as select/poll does as well.
569 	 */
570 	kn->kn_data = 0;
571 	return (1);
572 }
573 
574 int
575 filt_seltruemodify(struct kevent *kev, struct knote *kn)
576 {
577 	knote_assign(kev, kn);
578 	return (kn->kn_fop->f_event(kn, 0));
579 }
580 
581 int
582 filt_seltrueprocess(struct knote *kn, struct kevent *kev)
583 {
584 	int active;
585 
586 	active = kn->kn_fop->f_event(kn, 0);
587 	if (active)
588 		knote_submit(kn, kev);
589 	return (active);
590 }
591 
592 /*
593  * This provides full kqfilter entry for device switch tables, which
594  * has same effect as filter using filt_seltrue() as filter method.
595  */
596 void
597 filt_seltruedetach(struct knote *kn)
598 {
599 	/* Nothing to do */
600 }
601 
602 const struct filterops seltrue_filtops = {
603 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
604 	.f_attach	= NULL,
605 	.f_detach	= filt_seltruedetach,
606 	.f_event	= filt_seltrue,
607 	.f_modify	= filt_seltruemodify,
608 	.f_process	= filt_seltrueprocess,
609 };
610 
611 int
612 seltrue_kqfilter(dev_t dev, struct knote *kn)
613 {
614 	switch (kn->kn_filter) {
615 	case EVFILT_READ:
616 	case EVFILT_WRITE:
617 		kn->kn_fop = &seltrue_filtops;
618 		break;
619 	default:
620 		return (EINVAL);
621 	}
622 
623 	/* Nothing more to do */
624 	return (0);
625 }
626 
627 static int
628 filt_dead(struct knote *kn, long hint)
629 {
630 	if (kn->kn_filter == EVFILT_EXCEPT) {
631 		/*
632 		 * Do not deliver event because there is no out-of-band data.
633 		 * However, let HUP condition pass for poll(2).
634 		 */
635 		if ((kn->kn_flags & __EV_POLL) == 0) {
636 			kn->kn_flags |= EV_DISABLE;
637 			return (0);
638 		}
639 	}
640 
641 	kn->kn_flags |= (EV_EOF | EV_ONESHOT);
642 	if (kn->kn_flags & __EV_POLL)
643 		kn->kn_flags |= __EV_HUP;
644 	kn->kn_data = 0;
645 	return (1);
646 }
647 
648 static void
649 filt_deaddetach(struct knote *kn)
650 {
651 	/* Nothing to do */
652 }
653 
654 const struct filterops dead_filtops = {
655 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
656 	.f_attach	= NULL,
657 	.f_detach	= filt_deaddetach,
658 	.f_event	= filt_dead,
659 	.f_modify	= filt_seltruemodify,
660 	.f_process	= filt_seltrueprocess,
661 };
662 
663 static int
664 filt_badfd(struct knote *kn, long hint)
665 {
666 	kn->kn_flags |= (EV_ERROR | EV_ONESHOT);
667 	kn->kn_data = EBADF;
668 	return (1);
669 }
670 
671 /* For use with kqpoll. */
672 const struct filterops badfd_filtops = {
673 	.f_flags	= FILTEROP_ISFD | FILTEROP_MPSAFE,
674 	.f_attach	= NULL,
675 	.f_detach	= filt_deaddetach,
676 	.f_event	= filt_badfd,
677 	.f_modify	= filt_seltruemodify,
678 	.f_process	= filt_seltrueprocess,
679 };
680 
681 static int
682 filter_attach(struct knote *kn)
683 {
684 	int error;
685 
686 	if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
687 		error = kn->kn_fop->f_attach(kn);
688 	} else {
689 		KERNEL_LOCK();
690 		error = kn->kn_fop->f_attach(kn);
691 		KERNEL_UNLOCK();
692 	}
693 	return (error);
694 }
695 
696 static void
697 filter_detach(struct knote *kn)
698 {
699 	if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
700 		kn->kn_fop->f_detach(kn);
701 	} else {
702 		KERNEL_LOCK();
703 		kn->kn_fop->f_detach(kn);
704 		KERNEL_UNLOCK();
705 	}
706 }
707 
708 static int
709 filter_event(struct knote *kn, long hint)
710 {
711 	if ((kn->kn_fop->f_flags & FILTEROP_MPSAFE) == 0)
712 		KERNEL_ASSERT_LOCKED();
713 
714 	return (kn->kn_fop->f_event(kn, hint));
715 }
716 
717 static int
718 filter_modify(struct kevent *kev, struct knote *kn)
719 {
720 	int active, s;
721 
722 	if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
723 		active = kn->kn_fop->f_modify(kev, kn);
724 	} else {
725 		KERNEL_LOCK();
726 		if (kn->kn_fop->f_modify != NULL) {
727 			active = kn->kn_fop->f_modify(kev, kn);
728 		} else {
729 			s = splhigh();
730 			active = knote_modify(kev, kn);
731 			splx(s);
732 		}
733 		KERNEL_UNLOCK();
734 	}
735 	return (active);
736 }
737 
738 static int
739 filter_process(struct knote *kn, struct kevent *kev)
740 {
741 	int active, s;
742 
743 	if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
744 		active = kn->kn_fop->f_process(kn, kev);
745 	} else {
746 		KERNEL_LOCK();
747 		if (kn->kn_fop->f_process != NULL) {
748 			active = kn->kn_fop->f_process(kn, kev);
749 		} else {
750 			s = splhigh();
751 			active = knote_process(kn, kev);
752 			splx(s);
753 		}
754 		KERNEL_UNLOCK();
755 	}
756 	return (active);
757 }
758 
759 /*
760  * Initialize the current thread for poll/select system call.
761  * num indicates the number of serials that the system call may utilize.
762  * After this function, the valid range of serials is
763  * p_kq_serial <= x < p_kq_serial + num.
764  */
765 void
766 kqpoll_init(unsigned int num)
767 {
768 	struct proc *p = curproc;
769 	struct filedesc *fdp;
770 
771 	if (p->p_kq == NULL) {
772 		p->p_kq = kqueue_alloc(p->p_fd);
773 		p->p_kq_serial = arc4random();
774 		fdp = p->p_fd;
775 		fdplock(fdp);
776 		LIST_INSERT_HEAD(&fdp->fd_kqlist, p->p_kq, kq_next);
777 		fdpunlock(fdp);
778 	}
779 
780 	if (p->p_kq_serial + num < p->p_kq_serial) {
781 		/* Serial is about to wrap. Clear all attached knotes. */
782 		kqueue_purge(p, p->p_kq);
783 		p->p_kq_serial = 0;
784 	}
785 }
786 
787 /*
788  * Finish poll/select system call.
789  * num must have the same value that was used with kqpoll_init().
790  */
791 void
792 kqpoll_done(unsigned int num)
793 {
794 	struct proc *p = curproc;
795 	struct kqueue *kq = p->p_kq;
796 
797 	KASSERT(p->p_kq != NULL);
798 	KASSERT(p->p_kq_serial + num >= p->p_kq_serial);
799 
800 	p->p_kq_serial += num;
801 
802 	/*
803 	 * Because of kn_pollid key, a thread can in principle allocate
804 	 * up to O(maxfiles^2) knotes by calling poll(2) repeatedly
805 	 * with suitably varying pollfd arrays.
806 	 * Prevent such a large allocation by clearing knotes eagerly
807 	 * if there are too many of them.
808 	 *
809 	 * A small multiple of kq_knlistsize should give enough margin
810 	 * that eager clearing is infrequent, or does not happen at all,
811 	 * with normal programs.
812 	 * A single pollfd entry can use up to three knotes.
813 	 * Typically there is no significant overlap of fd and events
814 	 * between different entries in the pollfd array.
815 	 */
816 	if (kq->kq_nknotes > 4 * kq->kq_knlistsize)
817 		kqueue_purge(p, kq);
818 }
819 
820 void
821 kqpoll_exit(void)
822 {
823 	struct proc *p = curproc;
824 
825 	if (p->p_kq == NULL)
826 		return;
827 
828 	kqueue_purge(p, p->p_kq);
829 	kqueue_terminate(p, p->p_kq);
830 	KASSERT(p->p_kq->kq_refcnt.r_refs == 1);
831 	KQRELE(p->p_kq);
832 	p->p_kq = NULL;
833 }
834 
835 struct kqueue *
836 kqueue_alloc(struct filedesc *fdp)
837 {
838 	struct kqueue *kq;
839 
840 	kq = pool_get(&kqueue_pool, PR_WAITOK | PR_ZERO);
841 	refcnt_init(&kq->kq_refcnt);
842 	kq->kq_fdp = fdp;
843 	TAILQ_INIT(&kq->kq_head);
844 	mtx_init(&kq->kq_lock, IPL_HIGH);
845 	task_set(&kq->kq_task, kqueue_task, kq);
846 	klist_init_mutex(&kq->kq_klist, &kqueue_klist_lock);
847 
848 	return (kq);
849 }
850 
851 int
852 sys_kqueue(struct proc *p, void *v, register_t *retval)
853 {
854 	struct filedesc *fdp = p->p_fd;
855 	struct kqueue *kq;
856 	struct file *fp;
857 	int fd, error;
858 
859 	kq = kqueue_alloc(fdp);
860 
861 	fdplock(fdp);
862 	error = falloc(p, &fp, &fd);
863 	if (error)
864 		goto out;
865 	fp->f_flag = FREAD | FWRITE;
866 	fp->f_type = DTYPE_KQUEUE;
867 	fp->f_ops = &kqueueops;
868 	fp->f_data = kq;
869 	*retval = fd;
870 	LIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_next);
871 	kq = NULL;
872 	fdinsert(fdp, fd, 0, fp);
873 	FRELE(fp, p);
874 out:
875 	fdpunlock(fdp);
876 	if (kq != NULL)
877 		pool_put(&kqueue_pool, kq);
878 	return (error);
879 }
880 
881 int
882 sys_kevent(struct proc *p, void *v, register_t *retval)
883 {
884 	struct kqueue_scan_state scan;
885 	struct filedesc* fdp = p->p_fd;
886 	struct sys_kevent_args /* {
887 		syscallarg(int)	fd;
888 		syscallarg(const struct kevent *) changelist;
889 		syscallarg(int)	nchanges;
890 		syscallarg(struct kevent *) eventlist;
891 		syscallarg(int)	nevents;
892 		syscallarg(const struct timespec *) timeout;
893 	} */ *uap = v;
894 	struct kevent *kevp;
895 	struct kqueue *kq;
896 	struct file *fp;
897 	struct timespec ts;
898 	struct timespec *tsp = NULL;
899 	int i, n, nerrors, error;
900 	int ready, total;
901 	struct kevent kev[KQ_NEVENTS];
902 
903 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
904 		return (EBADF);
905 
906 	if (fp->f_type != DTYPE_KQUEUE) {
907 		error = EBADF;
908 		goto done;
909 	}
910 
911 	if (SCARG(uap, timeout) != NULL) {
912 		error = copyin(SCARG(uap, timeout), &ts, sizeof(ts));
913 		if (error)
914 			goto done;
915 #ifdef KTRACE
916 		if (KTRPOINT(p, KTR_STRUCT))
917 			ktrreltimespec(p, &ts);
918 #endif
919 		if (ts.tv_sec < 0 || !timespecisvalid(&ts)) {
920 			error = EINVAL;
921 			goto done;
922 		}
923 		tsp = &ts;
924 	}
925 
926 	kq = fp->f_data;
927 	nerrors = 0;
928 
929 	while ((n = SCARG(uap, nchanges)) > 0) {
930 		if (n > nitems(kev))
931 			n = nitems(kev);
932 		error = copyin(SCARG(uap, changelist), kev,
933 		    n * sizeof(struct kevent));
934 		if (error)
935 			goto done;
936 #ifdef KTRACE
937 		if (KTRPOINT(p, KTR_STRUCT))
938 			ktrevent(p, kev, n);
939 #endif
940 		for (i = 0; i < n; i++) {
941 			kevp = &kev[i];
942 			kevp->flags &= ~EV_SYSFLAGS;
943 			error = kqueue_register(kq, kevp, 0, p);
944 			if (error || (kevp->flags & EV_RECEIPT)) {
945 				if (SCARG(uap, nevents) != 0) {
946 					kevp->flags = EV_ERROR;
947 					kevp->data = error;
948 					copyout(kevp, SCARG(uap, eventlist),
949 					    sizeof(*kevp));
950 					SCARG(uap, eventlist)++;
951 					SCARG(uap, nevents)--;
952 					nerrors++;
953 				} else {
954 					goto done;
955 				}
956 			}
957 		}
958 		SCARG(uap, nchanges) -= n;
959 		SCARG(uap, changelist) += n;
960 	}
961 	if (nerrors) {
962 		*retval = nerrors;
963 		error = 0;
964 		goto done;
965 	}
966 
967 	kqueue_scan_setup(&scan, kq);
968 	FRELE(fp, p);
969 	/*
970 	 * Collect as many events as we can.  The timeout on successive
971 	 * loops is disabled (kqueue_scan() becomes non-blocking).
972 	 */
973 	total = 0;
974 	error = 0;
975 	while ((n = SCARG(uap, nevents) - total) > 0) {
976 		if (n > nitems(kev))
977 			n = nitems(kev);
978 		ready = kqueue_scan(&scan, n, kev, tsp, p, &error);
979 		if (ready == 0)
980 			break;
981 		error = copyout(kev, SCARG(uap, eventlist) + total,
982 		    sizeof(struct kevent) * ready);
983 #ifdef KTRACE
984 		if (KTRPOINT(p, KTR_STRUCT))
985 			ktrevent(p, kev, ready);
986 #endif
987 		total += ready;
988 		if (error || ready < n)
989 			break;
990 	}
991 	kqueue_scan_finish(&scan);
992 	*retval = total;
993 	return (error);
994 
995  done:
996 	FRELE(fp, p);
997 	return (error);
998 }
999 
1000 #ifdef KQUEUE_DEBUG
1001 void
1002 kqueue_do_check(struct kqueue *kq, const char *func, int line)
1003 {
1004 	struct knote *kn;
1005 	int count = 0, nmarker = 0;
1006 
1007 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1008 
1009 	TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) {
1010 		if (kn->kn_filter == EVFILT_MARKER) {
1011 			if ((kn->kn_status & KN_QUEUED) != 0)
1012 				panic("%s:%d: kq=%p kn=%p marker QUEUED",
1013 				    func, line, kq, kn);
1014 			nmarker++;
1015 		} else {
1016 			if ((kn->kn_status & KN_ACTIVE) == 0)
1017 				panic("%s:%d: kq=%p kn=%p knote !ACTIVE",
1018 				    func, line, kq, kn);
1019 			if ((kn->kn_status & KN_QUEUED) == 0)
1020 				panic("%s:%d: kq=%p kn=%p knote !QUEUED",
1021 				    func, line, kq, kn);
1022 			if (kn->kn_kq != kq)
1023 				panic("%s:%d: kq=%p kn=%p kn_kq=%p != kq",
1024 				    func, line, kq, kn, kn->kn_kq);
1025 			count++;
1026 			if (count > kq->kq_count)
1027 				goto bad;
1028 		}
1029 	}
1030 	if (count != kq->kq_count) {
1031 bad:
1032 		panic("%s:%d: kq=%p kq_count=%d count=%d nmarker=%d",
1033 		    func, line, kq, kq->kq_count, count, nmarker);
1034 	}
1035 }
1036 #endif
1037 
1038 int
1039 kqueue_register(struct kqueue *kq, struct kevent *kev, unsigned int pollid,
1040     struct proc *p)
1041 {
1042 	struct filedesc *fdp = kq->kq_fdp;
1043 	const struct filterops *fops = NULL;
1044 	struct file *fp = NULL;
1045 	struct knote *kn = NULL, *newkn = NULL;
1046 	struct knlist *list = NULL;
1047 	int active, error = 0;
1048 
1049 	KASSERT(pollid == 0 || (p != NULL && p->p_kq == kq));
1050 
1051 	if (kev->filter < 0) {
1052 		if (kev->filter + EVFILT_SYSCOUNT < 0)
1053 			return (EINVAL);
1054 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
1055 	}
1056 
1057 	if (fops == NULL) {
1058 		/*
1059 		 * XXX
1060 		 * filter attach routine is responsible for ensuring that
1061 		 * the identifier can be attached to it.
1062 		 */
1063 		return (EINVAL);
1064 	}
1065 
1066 	if (fops->f_flags & FILTEROP_ISFD) {
1067 		/* validate descriptor */
1068 		if (kev->ident > INT_MAX)
1069 			return (EBADF);
1070 	}
1071 
1072 	if (kev->flags & EV_ADD)
1073 		newkn = pool_get(&knote_pool, PR_WAITOK | PR_ZERO);
1074 
1075 again:
1076 	if (fops->f_flags & FILTEROP_ISFD) {
1077 		if ((fp = fd_getfile(fdp, kev->ident)) == NULL) {
1078 			error = EBADF;
1079 			goto done;
1080 		}
1081 		mtx_enter(&kq->kq_lock);
1082 		if (kev->flags & EV_ADD)
1083 			kqueue_expand_list(kq, kev->ident);
1084 		if (kev->ident < kq->kq_knlistsize)
1085 			list = &kq->kq_knlist[kev->ident];
1086 	} else {
1087 		mtx_enter(&kq->kq_lock);
1088 		if (kev->flags & EV_ADD)
1089 			kqueue_expand_hash(kq);
1090 		if (kq->kq_knhashmask != 0) {
1091 			list = &kq->kq_knhash[
1092 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1093 		}
1094 	}
1095 	if (list != NULL) {
1096 		SLIST_FOREACH(kn, list, kn_link) {
1097 			if (kev->filter == kn->kn_filter &&
1098 			    kev->ident == kn->kn_id &&
1099 			    pollid == kn->kn_pollid) {
1100 				if (!knote_acquire(kn, NULL, 0)) {
1101 					/* knote_acquire() has released
1102 					 * kq_lock. */
1103 					if (fp != NULL) {
1104 						FRELE(fp, p);
1105 						fp = NULL;
1106 					}
1107 					goto again;
1108 				}
1109 				break;
1110 			}
1111 		}
1112 	}
1113 	KASSERT(kn == NULL || (kn->kn_status & KN_PROCESSING) != 0);
1114 
1115 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
1116 		mtx_leave(&kq->kq_lock);
1117 		error = ENOENT;
1118 		goto done;
1119 	}
1120 
1121 	/*
1122 	 * kn now contains the matching knote, or NULL if no match.
1123 	 */
1124 	if (kev->flags & EV_ADD) {
1125 		if (kn == NULL) {
1126 			kn = newkn;
1127 			newkn = NULL;
1128 			kn->kn_status = KN_PROCESSING;
1129 			kn->kn_fp = fp;
1130 			kn->kn_kq = kq;
1131 			kn->kn_fop = fops;
1132 
1133 			/*
1134 			 * apply reference count to knote structure, and
1135 			 * do not release it at the end of this routine.
1136 			 */
1137 			fp = NULL;
1138 
1139 			kn->kn_sfflags = kev->fflags;
1140 			kn->kn_sdata = kev->data;
1141 			kev->fflags = 0;
1142 			kev->data = 0;
1143 			kn->kn_kevent = *kev;
1144 			kn->kn_pollid = pollid;
1145 
1146 			knote_attach(kn);
1147 			mtx_leave(&kq->kq_lock);
1148 
1149 			error = filter_attach(kn);
1150 			if (error != 0) {
1151 				knote_drop(kn, p);
1152 				goto done;
1153 			}
1154 
1155 			/*
1156 			 * If this is a file descriptor filter, check if
1157 			 * fd was closed while the knote was being added.
1158 			 * knote_fdclose() has missed kn if the function
1159 			 * ran before kn appeared in kq_knlist.
1160 			 */
1161 			if ((fops->f_flags & FILTEROP_ISFD) &&
1162 			    fd_checkclosed(fdp, kev->ident, kn->kn_fp)) {
1163 				/*
1164 				 * Drop the knote silently without error
1165 				 * because another thread might already have
1166 				 * seen it. This corresponds to the insert
1167 				 * happening in full before the close.
1168 				 */
1169 				filter_detach(kn);
1170 				knote_drop(kn, p);
1171 				goto done;
1172 			}
1173 
1174 			/* Check if there is a pending event. */
1175 			active = filter_process(kn, NULL);
1176 			mtx_enter(&kq->kq_lock);
1177 			if (active)
1178 				knote_activate(kn);
1179 		} else if (kn->kn_fop == &badfd_filtops) {
1180 			/*
1181 			 * Nothing expects this badfd knote any longer.
1182 			 * Drop it to make room for the new knote and retry.
1183 			 */
1184 			KASSERT(kq == p->p_kq);
1185 			mtx_leave(&kq->kq_lock);
1186 			filter_detach(kn);
1187 			knote_drop(kn, p);
1188 
1189 			KASSERT(fp != NULL);
1190 			FRELE(fp, p);
1191 			fp = NULL;
1192 
1193 			goto again;
1194 		} else {
1195 			/*
1196 			 * The user may change some filter values after the
1197 			 * initial EV_ADD, but doing so will not reset any
1198 			 * filters which have already been triggered.
1199 			 */
1200 			mtx_leave(&kq->kq_lock);
1201 			active = filter_modify(kev, kn);
1202 			mtx_enter(&kq->kq_lock);
1203 			if (active)
1204 				knote_activate(kn);
1205 			if (kev->flags & EV_ERROR) {
1206 				error = kev->data;
1207 				goto release;
1208 			}
1209 		}
1210 	} else if (kev->flags & EV_DELETE) {
1211 		mtx_leave(&kq->kq_lock);
1212 		filter_detach(kn);
1213 		knote_drop(kn, p);
1214 		goto done;
1215 	}
1216 
1217 	if ((kev->flags & EV_DISABLE) && ((kn->kn_status & KN_DISABLED) == 0))
1218 		kn->kn_status |= KN_DISABLED;
1219 
1220 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1221 		kn->kn_status &= ~KN_DISABLED;
1222 		mtx_leave(&kq->kq_lock);
1223 		/* Check if there is a pending event. */
1224 		active = filter_process(kn, NULL);
1225 		mtx_enter(&kq->kq_lock);
1226 		if (active)
1227 			knote_activate(kn);
1228 	}
1229 
1230 release:
1231 	knote_release(kn);
1232 	mtx_leave(&kq->kq_lock);
1233 done:
1234 	if (fp != NULL)
1235 		FRELE(fp, p);
1236 	if (newkn != NULL)
1237 		pool_put(&knote_pool, newkn);
1238 	return (error);
1239 }
1240 
1241 int
1242 kqueue_sleep(struct kqueue *kq, struct timespec *tsp)
1243 {
1244 	struct timespec elapsed, start, stop;
1245 	uint64_t nsecs;
1246 	int error;
1247 
1248 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1249 
1250 	if (tsp != NULL) {
1251 		getnanouptime(&start);
1252 		nsecs = MIN(TIMESPEC_TO_NSEC(tsp), MAXTSLP);
1253 	} else
1254 		nsecs = INFSLP;
1255 	error = msleep_nsec(kq, &kq->kq_lock, PSOCK | PCATCH | PNORELOCK,
1256 	    "kqread", nsecs);
1257 	if (tsp != NULL) {
1258 		getnanouptime(&stop);
1259 		timespecsub(&stop, &start, &elapsed);
1260 		timespecsub(tsp, &elapsed, tsp);
1261 		if (tsp->tv_sec < 0)
1262 			timespecclear(tsp);
1263 	}
1264 
1265 	return (error);
1266 }
1267 
1268 /*
1269  * Scan the kqueue, blocking if necessary until the target time is reached.
1270  * If tsp is NULL we block indefinitely.  If tsp->ts_secs/nsecs are both
1271  * 0 we do not block at all.
1272  */
1273 int
1274 kqueue_scan(struct kqueue_scan_state *scan, int maxevents,
1275     struct kevent *kevp, struct timespec *tsp, struct proc *p, int *errorp)
1276 {
1277 	struct kqueue *kq = scan->kqs_kq;
1278 	struct knote *kn;
1279 	int error = 0, nkev = 0;
1280 	int reinserted;
1281 
1282 	if (maxevents == 0)
1283 		goto done;
1284 retry:
1285 	KASSERT(nkev == 0);
1286 
1287 	error = 0;
1288 	reinserted = 0;
1289 
1290 	mtx_enter(&kq->kq_lock);
1291 
1292 	if (kq->kq_state & KQ_DYING) {
1293 		mtx_leave(&kq->kq_lock);
1294 		error = EBADF;
1295 		goto done;
1296 	}
1297 
1298 	if (kq->kq_count == 0) {
1299 		/*
1300 		 * Successive loops are only necessary if there are more
1301 		 * ready events to gather, so they don't need to block.
1302 		 */
1303 		if ((tsp != NULL && !timespecisset(tsp)) ||
1304 		    scan->kqs_nevent != 0) {
1305 			mtx_leave(&kq->kq_lock);
1306 			error = 0;
1307 			goto done;
1308 		}
1309 		kq->kq_state |= KQ_SLEEP;
1310 		error = kqueue_sleep(kq, tsp);
1311 		/* kqueue_sleep() has released kq_lock. */
1312 		if (error == 0 || error == EWOULDBLOCK)
1313 			goto retry;
1314 		/* don't restart after signals... */
1315 		if (error == ERESTART)
1316 			error = EINTR;
1317 		goto done;
1318 	}
1319 
1320 	/*
1321 	 * Put the end marker in the queue to limit the scan to the events
1322 	 * that are currently active.  This prevents events from being
1323 	 * recollected if they reactivate during scan.
1324 	 *
1325 	 * If a partial scan has been performed already but no events have
1326 	 * been collected, reposition the end marker to make any new events
1327 	 * reachable.
1328 	 */
1329 	if (!scan->kqs_queued) {
1330 		TAILQ_INSERT_TAIL(&kq->kq_head, &scan->kqs_end, kn_tqe);
1331 		scan->kqs_queued = 1;
1332 	} else if (scan->kqs_nevent == 0) {
1333 		TAILQ_REMOVE(&kq->kq_head, &scan->kqs_end, kn_tqe);
1334 		TAILQ_INSERT_TAIL(&kq->kq_head, &scan->kqs_end, kn_tqe);
1335 	}
1336 
1337 	TAILQ_INSERT_HEAD(&kq->kq_head, &scan->kqs_start, kn_tqe);
1338 	while (nkev < maxevents) {
1339 		kn = TAILQ_NEXT(&scan->kqs_start, kn_tqe);
1340 		if (kn->kn_filter == EVFILT_MARKER) {
1341 			if (kn == &scan->kqs_end)
1342 				break;
1343 
1344 			/* Move start marker past another thread's marker. */
1345 			TAILQ_REMOVE(&kq->kq_head, &scan->kqs_start, kn_tqe);
1346 			TAILQ_INSERT_AFTER(&kq->kq_head, kn, &scan->kqs_start,
1347 			    kn_tqe);
1348 			continue;
1349 		}
1350 
1351 		if (!knote_acquire(kn, NULL, 0)) {
1352 			/* knote_acquire() has released kq_lock. */
1353 			mtx_enter(&kq->kq_lock);
1354 			continue;
1355 		}
1356 
1357 		kqueue_check(kq);
1358 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1359 		kn->kn_status &= ~KN_QUEUED;
1360 		kq->kq_count--;
1361 		kqueue_check(kq);
1362 
1363 		if (kn->kn_status & KN_DISABLED) {
1364 			knote_release(kn);
1365 			continue;
1366 		}
1367 
1368 		mtx_leave(&kq->kq_lock);
1369 
1370 		/* Drop expired kqpoll knotes. */
1371 		if (p->p_kq == kq &&
1372 		    p->p_kq_serial > (unsigned long)kn->kn_udata) {
1373 			filter_detach(kn);
1374 			knote_drop(kn, p);
1375 			mtx_enter(&kq->kq_lock);
1376 			continue;
1377 		}
1378 
1379 		/*
1380 		 * Invalidate knotes whose vnodes have been revoked.
1381 		 * This is a workaround; it is tricky to clear existing
1382 		 * knotes and prevent new ones from being registered
1383 		 * with the current revocation mechanism.
1384 		 */
1385 		if ((kn->kn_fop->f_flags & FILTEROP_ISFD) &&
1386 		    kn->kn_fp != NULL &&
1387 		    kn->kn_fp->f_type == DTYPE_VNODE) {
1388 			struct vnode *vp = kn->kn_fp->f_data;
1389 
1390 			if (__predict_false(vp->v_op == &dead_vops &&
1391 			    kn->kn_fop != &dead_filtops)) {
1392 				filter_detach(kn);
1393 				kn->kn_fop = &dead_filtops;
1394 
1395 				/*
1396 				 * Check if the event should be delivered.
1397 				 * Use f_event directly because this is
1398 				 * a special situation.
1399 				 */
1400 				if (kn->kn_fop->f_event(kn, 0) == 0) {
1401 					filter_detach(kn);
1402 					knote_drop(kn, p);
1403 					mtx_enter(&kq->kq_lock);
1404 					continue;
1405 				}
1406 			}
1407 		}
1408 
1409 		memset(kevp, 0, sizeof(*kevp));
1410 		if (filter_process(kn, kevp) == 0) {
1411 			mtx_enter(&kq->kq_lock);
1412 			if ((kn->kn_status & KN_QUEUED) == 0)
1413 				kn->kn_status &= ~KN_ACTIVE;
1414 			knote_release(kn);
1415 			kqueue_check(kq);
1416 			continue;
1417 		}
1418 
1419 		/*
1420 		 * Post-event action on the note
1421 		 */
1422 		if (kevp->flags & EV_ONESHOT) {
1423 			filter_detach(kn);
1424 			knote_drop(kn, p);
1425 			mtx_enter(&kq->kq_lock);
1426 		} else if (kevp->flags & (EV_CLEAR | EV_DISPATCH)) {
1427 			mtx_enter(&kq->kq_lock);
1428 			if (kevp->flags & EV_DISPATCH)
1429 				kn->kn_status |= KN_DISABLED;
1430 			if ((kn->kn_status & KN_QUEUED) == 0)
1431 				kn->kn_status &= ~KN_ACTIVE;
1432 			knote_release(kn);
1433 		} else {
1434 			mtx_enter(&kq->kq_lock);
1435 			if ((kn->kn_status & KN_QUEUED) == 0) {
1436 				kqueue_check(kq);
1437 				kq->kq_count++;
1438 				kn->kn_status |= KN_QUEUED;
1439 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1440 				/* Wakeup is done after loop. */
1441 				reinserted = 1;
1442 			}
1443 			knote_release(kn);
1444 		}
1445 		kqueue_check(kq);
1446 
1447 		kevp++;
1448 		nkev++;
1449 		scan->kqs_nevent++;
1450 	}
1451 	TAILQ_REMOVE(&kq->kq_head, &scan->kqs_start, kn_tqe);
1452 	if (reinserted && kq->kq_count != 0)
1453 		kqueue_wakeup(kq);
1454 	mtx_leave(&kq->kq_lock);
1455 	if (scan->kqs_nevent == 0)
1456 		goto retry;
1457 done:
1458 	*errorp = error;
1459 	return (nkev);
1460 }
1461 
1462 void
1463 kqueue_scan_setup(struct kqueue_scan_state *scan, struct kqueue *kq)
1464 {
1465 	memset(scan, 0, sizeof(*scan));
1466 
1467 	KQREF(kq);
1468 	scan->kqs_kq = kq;
1469 	scan->kqs_start.kn_filter = EVFILT_MARKER;
1470 	scan->kqs_start.kn_status = KN_PROCESSING;
1471 	scan->kqs_end.kn_filter = EVFILT_MARKER;
1472 	scan->kqs_end.kn_status = KN_PROCESSING;
1473 }
1474 
1475 void
1476 kqueue_scan_finish(struct kqueue_scan_state *scan)
1477 {
1478 	struct kqueue *kq = scan->kqs_kq;
1479 
1480 	KASSERT(scan->kqs_start.kn_filter == EVFILT_MARKER);
1481 	KASSERT(scan->kqs_start.kn_status == KN_PROCESSING);
1482 	KASSERT(scan->kqs_end.kn_filter == EVFILT_MARKER);
1483 	KASSERT(scan->kqs_end.kn_status == KN_PROCESSING);
1484 
1485 	if (scan->kqs_queued) {
1486 		scan->kqs_queued = 0;
1487 		mtx_enter(&kq->kq_lock);
1488 		TAILQ_REMOVE(&kq->kq_head, &scan->kqs_end, kn_tqe);
1489 		mtx_leave(&kq->kq_lock);
1490 	}
1491 	KQRELE(kq);
1492 }
1493 
1494 /*
1495  * XXX
1496  * This could be expanded to call kqueue_scan, if desired.
1497  */
1498 int
1499 kqueue_read(struct file *fp, struct uio *uio, int fflags)
1500 {
1501 	return (ENXIO);
1502 }
1503 
1504 int
1505 kqueue_write(struct file *fp, struct uio *uio, int fflags)
1506 {
1507 	return (ENXIO);
1508 }
1509 
1510 int
1511 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
1512 {
1513 	return (ENOTTY);
1514 }
1515 
1516 int
1517 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
1518 {
1519 	struct kqueue *kq = fp->f_data;
1520 
1521 	memset(st, 0, sizeof(*st));
1522 	st->st_size = kq->kq_count;	/* unlocked read */
1523 	st->st_blksize = sizeof(struct kevent);
1524 	st->st_mode = S_IFIFO;
1525 	return (0);
1526 }
1527 
1528 void
1529 kqueue_purge(struct proc *p, struct kqueue *kq)
1530 {
1531 	int i;
1532 
1533 	mtx_enter(&kq->kq_lock);
1534 	for (i = 0; i < kq->kq_knlistsize; i++)
1535 		knote_remove(p, kq, &kq->kq_knlist, i, 1);
1536 	if (kq->kq_knhashmask != 0) {
1537 		for (i = 0; i < kq->kq_knhashmask + 1; i++)
1538 			knote_remove(p, kq, &kq->kq_knhash, i, 1);
1539 	}
1540 	mtx_leave(&kq->kq_lock);
1541 }
1542 
1543 void
1544 kqueue_terminate(struct proc *p, struct kqueue *kq)
1545 {
1546 	struct knote *kn;
1547 	int state;
1548 
1549 	mtx_enter(&kq->kq_lock);
1550 
1551 	/*
1552 	 * Any remaining entries should be scan markers.
1553 	 * They are removed when the ongoing scans finish.
1554 	 */
1555 	KASSERT(kq->kq_count == 0);
1556 	TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe)
1557 		KASSERT(kn->kn_filter == EVFILT_MARKER);
1558 
1559 	kq->kq_state |= KQ_DYING;
1560 	state = kq->kq_state;
1561 	kqueue_wakeup(kq);
1562 	mtx_leave(&kq->kq_lock);
1563 
1564 	/*
1565 	 * Any knotes that were attached to this kqueue were deleted
1566 	 * by knote_fdclose() when this kqueue's file descriptor was closed.
1567 	 */
1568 	KASSERT(klist_empty(&kq->kq_klist));
1569 	if (state & KQ_TASK)
1570 		taskq_del_barrier(systqmp, &kq->kq_task);
1571 }
1572 
1573 int
1574 kqueue_close(struct file *fp, struct proc *p)
1575 {
1576 	struct kqueue *kq = fp->f_data;
1577 
1578 	fp->f_data = NULL;
1579 
1580 	kqueue_purge(p, kq);
1581 	kqueue_terminate(p, kq);
1582 
1583 	KQRELE(kq);
1584 
1585 	return (0);
1586 }
1587 
1588 static void
1589 kqueue_task(void *arg)
1590 {
1591 	struct kqueue *kq = arg;
1592 
1593 	mtx_enter(&kqueue_klist_lock);
1594 	KNOTE(&kq->kq_klist, 0);
1595 	mtx_leave(&kqueue_klist_lock);
1596 }
1597 
1598 void
1599 kqueue_wakeup(struct kqueue *kq)
1600 {
1601 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1602 
1603 	if (kq->kq_state & KQ_SLEEP) {
1604 		kq->kq_state &= ~KQ_SLEEP;
1605 		wakeup(kq);
1606 	}
1607 	if (!klist_empty(&kq->kq_klist)) {
1608 		/* Defer activation to avoid recursion. */
1609 		kq->kq_state |= KQ_TASK;
1610 		task_add(systqmp, &kq->kq_task);
1611 	}
1612 }
1613 
1614 static void
1615 kqueue_expand_hash(struct kqueue *kq)
1616 {
1617 	struct knlist *hash;
1618 	u_long hashmask;
1619 
1620 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1621 
1622 	if (kq->kq_knhashmask == 0) {
1623 		mtx_leave(&kq->kq_lock);
1624 		hash = hashinit(KN_HASHSIZE, M_KEVENT, M_WAITOK, &hashmask);
1625 		mtx_enter(&kq->kq_lock);
1626 		if (kq->kq_knhashmask == 0) {
1627 			kq->kq_knhash = hash;
1628 			kq->kq_knhashmask = hashmask;
1629 		} else {
1630 			/* Another thread has allocated the hash. */
1631 			mtx_leave(&kq->kq_lock);
1632 			hashfree(hash, KN_HASHSIZE, M_KEVENT);
1633 			mtx_enter(&kq->kq_lock);
1634 		}
1635 	}
1636 }
1637 
1638 static void
1639 kqueue_expand_list(struct kqueue *kq, int fd)
1640 {
1641 	struct knlist *list, *olist;
1642 	int size, osize;
1643 
1644 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1645 
1646 	if (kq->kq_knlistsize <= fd) {
1647 		size = kq->kq_knlistsize;
1648 		mtx_leave(&kq->kq_lock);
1649 		while (size <= fd)
1650 			size += KQEXTENT;
1651 		list = mallocarray(size, sizeof(*list), M_KEVENT, M_WAITOK);
1652 		mtx_enter(&kq->kq_lock);
1653 		if (kq->kq_knlistsize <= fd) {
1654 			memcpy(list, kq->kq_knlist,
1655 			    kq->kq_knlistsize * sizeof(*list));
1656 			memset(&list[kq->kq_knlistsize], 0,
1657 			    (size - kq->kq_knlistsize) * sizeof(*list));
1658 			olist = kq->kq_knlist;
1659 			osize = kq->kq_knlistsize;
1660 			kq->kq_knlist = list;
1661 			kq->kq_knlistsize = size;
1662 			mtx_leave(&kq->kq_lock);
1663 			free(olist, M_KEVENT, osize * sizeof(*list));
1664 			mtx_enter(&kq->kq_lock);
1665 		} else {
1666 			/* Another thread has expanded the list. */
1667 			mtx_leave(&kq->kq_lock);
1668 			free(list, M_KEVENT, size * sizeof(*list));
1669 			mtx_enter(&kq->kq_lock);
1670 		}
1671 	}
1672 }
1673 
1674 /*
1675  * Acquire a knote, return non-zero on success, 0 on failure.
1676  *
1677  * If we cannot acquire the knote we sleep and return 0.  The knote
1678  * may be stale on return in this case and the caller must restart
1679  * whatever loop they are in.
1680  *
1681  * If we are about to sleep and klist is non-NULL, the list is unlocked
1682  * before sleep and remains unlocked on return.
1683  */
1684 int
1685 knote_acquire(struct knote *kn, struct klist *klist, int ls)
1686 {
1687 	struct kqueue *kq = kn->kn_kq;
1688 
1689 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1690 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1691 
1692 	if (kn->kn_status & KN_PROCESSING) {
1693 		kn->kn_status |= KN_WAITING;
1694 		if (klist != NULL) {
1695 			mtx_leave(&kq->kq_lock);
1696 			klist_unlock(klist, ls);
1697 			/* XXX Timeout resolves potential loss of wakeup. */
1698 			tsleep_nsec(kn, 0, "kqepts", SEC_TO_NSEC(1));
1699 		} else {
1700 			msleep_nsec(kn, &kq->kq_lock, PNORELOCK, "kqepts",
1701 			    SEC_TO_NSEC(1));
1702 		}
1703 		/* knote may be stale now */
1704 		return (0);
1705 	}
1706 	kn->kn_status |= KN_PROCESSING;
1707 	return (1);
1708 }
1709 
1710 /*
1711  * Release an acquired knote, clearing KN_PROCESSING.
1712  */
1713 void
1714 knote_release(struct knote *kn)
1715 {
1716 	MUTEX_ASSERT_LOCKED(&kn->kn_kq->kq_lock);
1717 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1718 	KASSERT(kn->kn_status & KN_PROCESSING);
1719 
1720 	if (kn->kn_status & KN_WAITING) {
1721 		kn->kn_status &= ~KN_WAITING;
1722 		wakeup(kn);
1723 	}
1724 	kn->kn_status &= ~KN_PROCESSING;
1725 	/* kn should not be accessed anymore */
1726 }
1727 
1728 /*
1729  * activate one knote.
1730  */
1731 void
1732 knote_activate(struct knote *kn)
1733 {
1734 	MUTEX_ASSERT_LOCKED(&kn->kn_kq->kq_lock);
1735 
1736 	kn->kn_status |= KN_ACTIVE;
1737 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)
1738 		knote_enqueue(kn);
1739 }
1740 
1741 /*
1742  * walk down a list of knotes, activating them if their event has triggered.
1743  */
1744 void
1745 knote(struct klist *list, long hint)
1746 {
1747 	struct knote *kn, *kn0;
1748 	struct kqueue *kq;
1749 
1750 	KLIST_ASSERT_LOCKED(list);
1751 
1752 	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, kn0) {
1753 		if (filter_event(kn, hint)) {
1754 			kq = kn->kn_kq;
1755 			mtx_enter(&kq->kq_lock);
1756 			knote_activate(kn);
1757 			mtx_leave(&kq->kq_lock);
1758 		}
1759 	}
1760 }
1761 
1762 /*
1763  * remove all knotes from a specified knlist
1764  */
1765 void
1766 knote_remove(struct proc *p, struct kqueue *kq, struct knlist **plist, int idx,
1767     int purge)
1768 {
1769 	struct knote *kn;
1770 
1771 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1772 
1773 	/* Always fetch array pointer as another thread can resize kq_knlist. */
1774 	while ((kn = SLIST_FIRST(*plist + idx)) != NULL) {
1775 		KASSERT(kn->kn_kq == kq);
1776 
1777 		if (!purge) {
1778 			/* Skip pending badfd knotes. */
1779 			while (kn->kn_fop == &badfd_filtops) {
1780 				kn = SLIST_NEXT(kn, kn_link);
1781 				if (kn == NULL)
1782 					return;
1783 				KASSERT(kn->kn_kq == kq);
1784 			}
1785 		}
1786 
1787 		if (!knote_acquire(kn, NULL, 0)) {
1788 			/* knote_acquire() has released kq_lock. */
1789 			mtx_enter(&kq->kq_lock);
1790 			continue;
1791 		}
1792 		mtx_leave(&kq->kq_lock);
1793 		filter_detach(kn);
1794 
1795 		/*
1796 		 * Notify poll(2) and select(2) when a monitored
1797 		 * file descriptor is closed.
1798 		 *
1799 		 * This reuses the original knote for delivering the
1800 		 * notification so as to avoid allocating memory.
1801 		 */
1802 		if (!purge && (kn->kn_flags & (__EV_POLL | __EV_SELECT)) &&
1803 		    !(p->p_kq == kq &&
1804 		      p->p_kq_serial > (unsigned long)kn->kn_udata) &&
1805 		    kn->kn_fop != &badfd_filtops) {
1806 			KASSERT(kn->kn_fop->f_flags & FILTEROP_ISFD);
1807 			FRELE(kn->kn_fp, p);
1808 			kn->kn_fp = NULL;
1809 
1810 			kn->kn_fop = &badfd_filtops;
1811 			filter_event(kn, 0);
1812 			mtx_enter(&kq->kq_lock);
1813 			knote_activate(kn);
1814 			knote_release(kn);
1815 			continue;
1816 		}
1817 
1818 		knote_drop(kn, p);
1819 		mtx_enter(&kq->kq_lock);
1820 	}
1821 }
1822 
1823 /*
1824  * remove all knotes referencing a specified fd
1825  */
1826 void
1827 knote_fdclose(struct proc *p, int fd)
1828 {
1829 	struct filedesc *fdp = p->p_p->ps_fd;
1830 	struct kqueue *kq;
1831 
1832 	/*
1833 	 * fdplock can be ignored if the file descriptor table is being freed
1834 	 * because no other thread can access the fdp.
1835 	 */
1836 	if (fdp->fd_refcnt != 0)
1837 		fdpassertlocked(fdp);
1838 
1839 	LIST_FOREACH(kq, &fdp->fd_kqlist, kq_next) {
1840 		mtx_enter(&kq->kq_lock);
1841 		if (fd < kq->kq_knlistsize)
1842 			knote_remove(p, kq, &kq->kq_knlist, fd, 0);
1843 		mtx_leave(&kq->kq_lock);
1844 	}
1845 }
1846 
1847 /*
1848  * handle a process exiting, including the triggering of NOTE_EXIT notes
1849  * XXX this could be more efficient, doing a single pass down the klist
1850  */
1851 void
1852 knote_processexit(struct process *pr)
1853 {
1854 	KERNEL_ASSERT_LOCKED();
1855 
1856 	KNOTE(&pr->ps_klist, NOTE_EXIT);
1857 
1858 	/* remove other knotes hanging off the process */
1859 	klist_invalidate(&pr->ps_klist);
1860 }
1861 
1862 void
1863 knote_attach(struct knote *kn)
1864 {
1865 	struct kqueue *kq = kn->kn_kq;
1866 	struct knlist *list;
1867 
1868 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1869 	KASSERT(kn->kn_status & KN_PROCESSING);
1870 
1871 	if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
1872 		KASSERT(kq->kq_knlistsize > kn->kn_id);
1873 		list = &kq->kq_knlist[kn->kn_id];
1874 	} else {
1875 		KASSERT(kq->kq_knhashmask != 0);
1876 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1877 	}
1878 	SLIST_INSERT_HEAD(list, kn, kn_link);
1879 	kq->kq_nknotes++;
1880 }
1881 
1882 void
1883 knote_detach(struct knote *kn)
1884 {
1885 	struct kqueue *kq = kn->kn_kq;
1886 	struct knlist *list;
1887 
1888 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1889 	KASSERT(kn->kn_status & KN_PROCESSING);
1890 
1891 	kq->kq_nknotes--;
1892 	if (kn->kn_fop->f_flags & FILTEROP_ISFD)
1893 		list = &kq->kq_knlist[kn->kn_id];
1894 	else
1895 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1896 	SLIST_REMOVE(list, kn, knote, kn_link);
1897 }
1898 
1899 /*
1900  * should be called at spl == 0, since we don't want to hold spl
1901  * while calling FRELE and pool_put.
1902  */
1903 void
1904 knote_drop(struct knote *kn, struct proc *p)
1905 {
1906 	struct kqueue *kq = kn->kn_kq;
1907 
1908 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1909 
1910 	mtx_enter(&kq->kq_lock);
1911 	knote_detach(kn);
1912 	if (kn->kn_status & KN_QUEUED)
1913 		knote_dequeue(kn);
1914 	if (kn->kn_status & KN_WAITING) {
1915 		kn->kn_status &= ~KN_WAITING;
1916 		wakeup(kn);
1917 	}
1918 	mtx_leave(&kq->kq_lock);
1919 
1920 	if ((kn->kn_fop->f_flags & FILTEROP_ISFD) && kn->kn_fp != NULL)
1921 		FRELE(kn->kn_fp, p);
1922 	pool_put(&knote_pool, kn);
1923 }
1924 
1925 
1926 void
1927 knote_enqueue(struct knote *kn)
1928 {
1929 	struct kqueue *kq = kn->kn_kq;
1930 
1931 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1932 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1933 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
1934 
1935 	kqueue_check(kq);
1936 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1937 	kn->kn_status |= KN_QUEUED;
1938 	kq->kq_count++;
1939 	kqueue_check(kq);
1940 	kqueue_wakeup(kq);
1941 }
1942 
1943 void
1944 knote_dequeue(struct knote *kn)
1945 {
1946 	struct kqueue *kq = kn->kn_kq;
1947 
1948 	MUTEX_ASSERT_LOCKED(&kq->kq_lock);
1949 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1950 	KASSERT(kn->kn_status & KN_QUEUED);
1951 
1952 	kqueue_check(kq);
1953 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1954 	kn->kn_status &= ~KN_QUEUED;
1955 	kq->kq_count--;
1956 	kqueue_check(kq);
1957 }
1958 
1959 /*
1960  * Assign parameters to the knote.
1961  *
1962  * The knote's object lock must be held.
1963  */
1964 void
1965 knote_assign(const struct kevent *kev, struct knote *kn)
1966 {
1967 	if ((kn->kn_fop->f_flags & FILTEROP_MPSAFE) == 0)
1968 		KERNEL_ASSERT_LOCKED();
1969 
1970 	kn->kn_sfflags = kev->fflags;
1971 	kn->kn_sdata = kev->data;
1972 	kn->kn_udata = kev->udata;
1973 }
1974 
1975 /*
1976  * Submit the knote's event for delivery.
1977  *
1978  * The knote's object lock must be held.
1979  */
1980 void
1981 knote_submit(struct knote *kn, struct kevent *kev)
1982 {
1983 	if ((kn->kn_fop->f_flags & FILTEROP_MPSAFE) == 0)
1984 		KERNEL_ASSERT_LOCKED();
1985 
1986 	if (kev != NULL) {
1987 		*kev = kn->kn_kevent;
1988 		if (kn->kn_flags & EV_CLEAR) {
1989 			kn->kn_fflags = 0;
1990 			kn->kn_data = 0;
1991 		}
1992 	}
1993 }
1994 
1995 void
1996 klist_init(struct klist *klist, const struct klistops *ops, void *arg)
1997 {
1998 	SLIST_INIT(&klist->kl_list);
1999 	klist->kl_ops = ops;
2000 	klist->kl_arg = arg;
2001 }
2002 
2003 void
2004 klist_free(struct klist *klist)
2005 {
2006 	KASSERT(SLIST_EMPTY(&klist->kl_list));
2007 }
2008 
2009 void
2010 klist_insert(struct klist *klist, struct knote *kn)
2011 {
2012 	int ls;
2013 
2014 	ls = klist_lock(klist);
2015 	SLIST_INSERT_HEAD(&klist->kl_list, kn, kn_selnext);
2016 	klist_unlock(klist, ls);
2017 }
2018 
2019 void
2020 klist_insert_locked(struct klist *klist, struct knote *kn)
2021 {
2022 	KLIST_ASSERT_LOCKED(klist);
2023 
2024 	SLIST_INSERT_HEAD(&klist->kl_list, kn, kn_selnext);
2025 }
2026 
2027 void
2028 klist_remove(struct klist *klist, struct knote *kn)
2029 {
2030 	int ls;
2031 
2032 	ls = klist_lock(klist);
2033 	SLIST_REMOVE(&klist->kl_list, kn, knote, kn_selnext);
2034 	klist_unlock(klist, ls);
2035 }
2036 
2037 void
2038 klist_remove_locked(struct klist *klist, struct knote *kn)
2039 {
2040 	KLIST_ASSERT_LOCKED(klist);
2041 
2042 	SLIST_REMOVE(&klist->kl_list, kn, knote, kn_selnext);
2043 }
2044 
2045 /*
2046  * Detach all knotes from klist. The knotes are rewired to indicate EOF.
2047  *
2048  * The caller of this function must not hold any locks that can block
2049  * filterops callbacks that run with KN_PROCESSING.
2050  * Otherwise this function might deadlock.
2051  */
2052 void
2053 klist_invalidate(struct klist *list)
2054 {
2055 	struct knote *kn;
2056 	struct kqueue *kq;
2057 	struct proc *p = curproc;
2058 	int ls;
2059 
2060 	NET_ASSERT_UNLOCKED();
2061 
2062 	ls = klist_lock(list);
2063 	while ((kn = SLIST_FIRST(&list->kl_list)) != NULL) {
2064 		kq = kn->kn_kq;
2065 		mtx_enter(&kq->kq_lock);
2066 		if (!knote_acquire(kn, list, ls)) {
2067 			/* knote_acquire() has released kq_lock
2068 			 * and klist lock. */
2069 			ls = klist_lock(list);
2070 			continue;
2071 		}
2072 		mtx_leave(&kq->kq_lock);
2073 		klist_unlock(list, ls);
2074 		filter_detach(kn);
2075 		if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
2076 			kn->kn_fop = &dead_filtops;
2077 			filter_event(kn, 0);
2078 			mtx_enter(&kq->kq_lock);
2079 			knote_activate(kn);
2080 			knote_release(kn);
2081 			mtx_leave(&kq->kq_lock);
2082 		} else {
2083 			knote_drop(kn, p);
2084 		}
2085 		ls = klist_lock(list);
2086 	}
2087 	klist_unlock(list, ls);
2088 }
2089 
2090 static int
2091 klist_lock(struct klist *list)
2092 {
2093 	int ls = 0;
2094 
2095 	if (list->kl_ops != NULL) {
2096 		ls = list->kl_ops->klo_lock(list->kl_arg);
2097 	} else {
2098 		KERNEL_LOCK();
2099 		ls = splhigh();
2100 	}
2101 	return ls;
2102 }
2103 
2104 static void
2105 klist_unlock(struct klist *list, int ls)
2106 {
2107 	if (list->kl_ops != NULL) {
2108 		list->kl_ops->klo_unlock(list->kl_arg, ls);
2109 	} else {
2110 		splx(ls);
2111 		KERNEL_UNLOCK();
2112 	}
2113 }
2114 
2115 static void
2116 klist_mutex_assertlk(void *arg)
2117 {
2118 	struct mutex *mtx = arg;
2119 
2120 	(void)mtx;
2121 
2122 	MUTEX_ASSERT_LOCKED(mtx);
2123 }
2124 
2125 static int
2126 klist_mutex_lock(void *arg)
2127 {
2128 	struct mutex *mtx = arg;
2129 
2130 	mtx_enter(mtx);
2131 	return 0;
2132 }
2133 
2134 static void
2135 klist_mutex_unlock(void *arg, int s)
2136 {
2137 	struct mutex *mtx = arg;
2138 
2139 	mtx_leave(mtx);
2140 }
2141 
2142 static const struct klistops mutex_klistops = {
2143 	.klo_assertlk	= klist_mutex_assertlk,
2144 	.klo_lock	= klist_mutex_lock,
2145 	.klo_unlock	= klist_mutex_unlock,
2146 };
2147 
2148 void
2149 klist_init_mutex(struct klist *klist, struct mutex *mtx)
2150 {
2151 	klist_init(klist, &mutex_klistops, mtx);
2152 }
2153 
2154 static void
2155 klist_rwlock_assertlk(void *arg)
2156 {
2157 	struct rwlock *rwl = arg;
2158 
2159 	(void)rwl;
2160 
2161 	rw_assert_wrlock(rwl);
2162 }
2163 
2164 static int
2165 klist_rwlock_lock(void *arg)
2166 {
2167 	struct rwlock *rwl = arg;
2168 
2169 	rw_enter_write(rwl);
2170 	return 0;
2171 }
2172 
2173 static void
2174 klist_rwlock_unlock(void *arg, int s)
2175 {
2176 	struct rwlock *rwl = arg;
2177 
2178 	rw_exit_write(rwl);
2179 }
2180 
2181 static const struct klistops rwlock_klistops = {
2182 	.klo_assertlk	= klist_rwlock_assertlk,
2183 	.klo_lock	= klist_rwlock_lock,
2184 	.klo_unlock	= klist_rwlock_unlock,
2185 };
2186 
2187 void
2188 klist_init_rwlock(struct klist *klist, struct rwlock *rwl)
2189 {
2190 	klist_init(klist, &rwlock_klistops, rwl);
2191 }
2192