xref: /openbsd/sys/kern/kern_event.c (revision 4cfece93)
1 /*	$OpenBSD: kern_event.c,v 1.141 2020/07/04 08:33:43 visa 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/atomic.h>
34 #include <sys/kernel.h>
35 #include <sys/proc.h>
36 #include <sys/pledge.h>
37 #include <sys/malloc.h>
38 #include <sys/unistd.h>
39 #include <sys/file.h>
40 #include <sys/filedesc.h>
41 #include <sys/fcntl.h>
42 #include <sys/selinfo.h>
43 #include <sys/queue.h>
44 #include <sys/event.h>
45 #include <sys/eventvar.h>
46 #include <sys/ktrace.h>
47 #include <sys/pool.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/stat.h>
52 #include <sys/uio.h>
53 #include <sys/mount.h>
54 #include <sys/poll.h>
55 #include <sys/syscallargs.h>
56 #include <sys/time.h>
57 #include <sys/timeout.h>
58 #include <sys/wait.h>
59 
60 void	kqueue_terminate(struct proc *p, struct kqueue *);
61 void	kqueue_free(struct kqueue *);
62 void	kqueue_init(void);
63 void	KQREF(struct kqueue *);
64 void	KQRELE(struct kqueue *);
65 
66 int	kqueue_sleep(struct kqueue *, struct timespec *);
67 int	kqueue_scan(struct kqueue *kq, int maxevents,
68 		    struct kevent *ulistp, struct timespec *timeout,
69 		    struct proc *p, int *retval);
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_poll(struct file *fp, int events, struct proc *p);
76 int	kqueue_kqfilter(struct file *fp, struct knote *kn);
77 int	kqueue_stat(struct file *fp, struct stat *st, struct proc *p);
78 int	kqueue_close(struct file *fp, struct proc *p);
79 void	kqueue_wakeup(struct kqueue *kq);
80 
81 static void	kqueue_expand_hash(struct kqueue *kq);
82 static void	kqueue_expand_list(struct kqueue *kq, int fd);
83 static void	kqueue_task(void *);
84 
85 const struct fileops kqueueops = {
86 	.fo_read	= kqueue_read,
87 	.fo_write	= kqueue_write,
88 	.fo_ioctl	= kqueue_ioctl,
89 	.fo_poll	= kqueue_poll,
90 	.fo_kqfilter	= kqueue_kqfilter,
91 	.fo_stat	= kqueue_stat,
92 	.fo_close	= kqueue_close
93 };
94 
95 void	knote_attach(struct knote *kn);
96 void	knote_drop(struct knote *kn, struct proc *p);
97 void	knote_enqueue(struct knote *kn);
98 void	knote_dequeue(struct knote *kn);
99 int	knote_acquire(struct knote *kn);
100 void	knote_release(struct knote *kn);
101 
102 void	filt_kqdetach(struct knote *kn);
103 int	filt_kqueue(struct knote *kn, long hint);
104 int	filt_procattach(struct knote *kn);
105 void	filt_procdetach(struct knote *kn);
106 int	filt_proc(struct knote *kn, long hint);
107 int	filt_fileattach(struct knote *kn);
108 void	filt_timerexpire(void *knx);
109 int	filt_timerattach(struct knote *kn);
110 void	filt_timerdetach(struct knote *kn);
111 int	filt_timer(struct knote *kn, long hint);
112 void	filt_seltruedetach(struct knote *kn);
113 
114 const struct filterops kqread_filtops = {
115 	.f_flags	= FILTEROP_ISFD,
116 	.f_attach	= NULL,
117 	.f_detach	= filt_kqdetach,
118 	.f_event	= filt_kqueue,
119 };
120 
121 const struct filterops proc_filtops = {
122 	.f_flags	= 0,
123 	.f_attach	= filt_procattach,
124 	.f_detach	= filt_procdetach,
125 	.f_event	= filt_proc,
126 };
127 
128 const struct filterops file_filtops = {
129 	.f_flags	= FILTEROP_ISFD,
130 	.f_attach	= filt_fileattach,
131 	.f_detach	= NULL,
132 	.f_event	= NULL,
133 };
134 
135 const struct filterops timer_filtops = {
136 	.f_flags	= 0,
137 	.f_attach	= filt_timerattach,
138 	.f_detach	= filt_timerdetach,
139 	.f_event	= filt_timer,
140 };
141 
142 struct	pool knote_pool;
143 struct	pool kqueue_pool;
144 int kq_ntimeouts = 0;
145 int kq_timeoutmax = (4 * 1024);
146 
147 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
148 
149 /*
150  * Table for for all system-defined filters.
151  */
152 const struct filterops *const sysfilt_ops[] = {
153 	&file_filtops,			/* EVFILT_READ */
154 	&file_filtops,			/* EVFILT_WRITE */
155 	NULL, /*&aio_filtops,*/		/* EVFILT_AIO */
156 	&file_filtops,			/* EVFILT_VNODE */
157 	&proc_filtops,			/* EVFILT_PROC */
158 	&sig_filtops,			/* EVFILT_SIGNAL */
159 	&timer_filtops,			/* EVFILT_TIMER */
160 	&file_filtops,			/* EVFILT_DEVICE */
161 	&file_filtops,			/* EVFILT_EXCEPT */
162 };
163 
164 void
165 KQREF(struct kqueue *kq)
166 {
167 	atomic_inc_int(&kq->kq_refs);
168 }
169 
170 void
171 KQRELE(struct kqueue *kq)
172 {
173 	struct filedesc *fdp;
174 
175 	if (atomic_dec_int_nv(&kq->kq_refs) > 0)
176 		return;
177 
178 	fdp = kq->kq_fdp;
179 	if (rw_status(&fdp->fd_lock) == RW_WRITE) {
180 		LIST_REMOVE(kq, kq_next);
181 	} else {
182 		fdplock(fdp);
183 		LIST_REMOVE(kq, kq_next);
184 		fdpunlock(fdp);
185 	}
186 
187 	kqueue_free(kq);
188 }
189 
190 void
191 kqueue_free(struct kqueue *kq)
192 {
193 	free(kq->kq_knlist, M_KEVENT, kq->kq_knlistsize *
194 	    sizeof(struct knlist));
195 	hashfree(kq->kq_knhash, KN_HASHSIZE, M_KEVENT);
196 	pool_put(&kqueue_pool, kq);
197 }
198 
199 void
200 kqueue_init(void)
201 {
202 	pool_init(&kqueue_pool, sizeof(struct kqueue), 0, IPL_MPFLOOR,
203 	    PR_WAITOK, "kqueuepl", NULL);
204 	pool_init(&knote_pool, sizeof(struct knote), 0, IPL_MPFLOOR,
205 	    PR_WAITOK, "knotepl", NULL);
206 }
207 
208 int
209 filt_fileattach(struct knote *kn)
210 {
211 	struct file *fp = kn->kn_fp;
212 
213 	return fp->f_ops->fo_kqfilter(fp, kn);
214 }
215 
216 int
217 kqueue_kqfilter(struct file *fp, struct knote *kn)
218 {
219 	struct kqueue *kq = kn->kn_fp->f_data;
220 
221 	if (kn->kn_filter != EVFILT_READ)
222 		return (EINVAL);
223 
224 	kn->kn_fop = &kqread_filtops;
225 	klist_insert(&kq->kq_sel.si_note, kn);
226 	return (0);
227 }
228 
229 void
230 filt_kqdetach(struct knote *kn)
231 {
232 	struct kqueue *kq = kn->kn_fp->f_data;
233 
234 	klist_remove(&kq->kq_sel.si_note, kn);
235 }
236 
237 int
238 filt_kqueue(struct knote *kn, long hint)
239 {
240 	struct kqueue *kq = kn->kn_fp->f_data;
241 
242 	kn->kn_data = kq->kq_count;
243 	return (kn->kn_data > 0);
244 }
245 
246 int
247 filt_procattach(struct knote *kn)
248 {
249 	struct process *pr;
250 	int s;
251 
252 	if ((curproc->p_p->ps_flags & PS_PLEDGE) &&
253 	    (curproc->p_p->ps_pledge & PLEDGE_PROC) == 0)
254 		return pledge_fail(curproc, EPERM, PLEDGE_PROC);
255 
256 	if (kn->kn_id > PID_MAX)
257 		return ESRCH;
258 
259 	pr = prfind(kn->kn_id);
260 	if (pr == NULL)
261 		return (ESRCH);
262 
263 	/* exiting processes can't be specified */
264 	if (pr->ps_flags & PS_EXITING)
265 		return (ESRCH);
266 
267 	kn->kn_ptr.p_process = pr;
268 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
269 
270 	/*
271 	 * internal flag indicating registration done by kernel
272 	 */
273 	if (kn->kn_flags & EV_FLAG1) {
274 		kn->kn_data = kn->kn_sdata;		/* ppid */
275 		kn->kn_fflags = NOTE_CHILD;
276 		kn->kn_flags &= ~EV_FLAG1;
277 	}
278 
279 	s = splhigh();
280 	klist_insert(&pr->ps_klist, kn);
281 	splx(s);
282 
283 	return (0);
284 }
285 
286 /*
287  * The knote may be attached to a different process, which may exit,
288  * leaving nothing for the knote to be attached to.  So when the process
289  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
290  * it will be deleted when read out.  However, as part of the knote deletion,
291  * this routine is called, so a check is needed to avoid actually performing
292  * a detach, because the original process does not exist any more.
293  */
294 void
295 filt_procdetach(struct knote *kn)
296 {
297 	struct process *pr = kn->kn_ptr.p_process;
298 	int s;
299 
300 	if (kn->kn_status & KN_DETACHED)
301 		return;
302 
303 	s = splhigh();
304 	klist_remove(&pr->ps_klist, kn);
305 	splx(s);
306 }
307 
308 int
309 filt_proc(struct knote *kn, long hint)
310 {
311 	u_int event;
312 
313 	/*
314 	 * mask off extra data
315 	 */
316 	event = (u_int)hint & NOTE_PCTRLMASK;
317 
318 	/*
319 	 * if the user is interested in this event, record it.
320 	 */
321 	if (kn->kn_sfflags & event)
322 		kn->kn_fflags |= event;
323 
324 	/*
325 	 * process is gone, so flag the event as finished and remove it
326 	 * from the process's klist
327 	 */
328 	if (event == NOTE_EXIT) {
329 		struct process *pr = kn->kn_ptr.p_process;
330 		int s;
331 
332 		s = splhigh();
333 		kn->kn_status |= KN_DETACHED;
334 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
335 		kn->kn_data = W_EXITCODE(pr->ps_xexit, pr->ps_xsig);
336 		klist_remove(&pr->ps_klist, kn);
337 		splx(s);
338 		return (1);
339 	}
340 
341 	/*
342 	 * process forked, and user wants to track the new process,
343 	 * so attach a new knote to it, and immediately report an
344 	 * event with the parent's pid.
345 	 */
346 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
347 		struct kevent kev;
348 		int error;
349 
350 		/*
351 		 * register knote with new process.
352 		 */
353 		memset(&kev, 0, sizeof(kev));
354 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
355 		kev.filter = kn->kn_filter;
356 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
357 		kev.fflags = kn->kn_sfflags;
358 		kev.data = kn->kn_id;			/* parent */
359 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
360 		error = kqueue_register(kn->kn_kq, &kev, NULL);
361 		if (error)
362 			kn->kn_fflags |= NOTE_TRACKERR;
363 	}
364 
365 	return (kn->kn_fflags != 0);
366 }
367 
368 static void
369 filt_timer_timeout_add(struct knote *kn)
370 {
371 	struct timeval tv;
372 	struct timeout *to = kn->kn_hook;
373 	int tticks;
374 
375 	tv.tv_sec = kn->kn_sdata / 1000;
376 	tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
377 	tticks = tvtohz(&tv);
378 	/* Remove extra tick from tvtohz() if timeout has fired before. */
379 	if (timeout_triggered(to))
380 		tticks--;
381 	timeout_add(to, (tticks > 0) ? tticks : 1);
382 }
383 
384 void
385 filt_timerexpire(void *knx)
386 {
387 	struct knote *kn = knx;
388 
389 	kn->kn_data++;
390 	knote_activate(kn);
391 
392 	if ((kn->kn_flags & EV_ONESHOT) == 0)
393 		filt_timer_timeout_add(kn);
394 }
395 
396 
397 /*
398  * data contains amount of time to sleep, in milliseconds
399  */
400 int
401 filt_timerattach(struct knote *kn)
402 {
403 	struct timeout *to;
404 
405 	if (kq_ntimeouts > kq_timeoutmax)
406 		return (ENOMEM);
407 	kq_ntimeouts++;
408 
409 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
410 	to = malloc(sizeof(*to), M_KEVENT, M_WAITOK);
411 	timeout_set(to, filt_timerexpire, kn);
412 	kn->kn_hook = to;
413 	filt_timer_timeout_add(kn);
414 
415 	return (0);
416 }
417 
418 void
419 filt_timerdetach(struct knote *kn)
420 {
421 	struct timeout *to;
422 
423 	to = (struct timeout *)kn->kn_hook;
424 	timeout_del(to);
425 	free(to, M_KEVENT, sizeof(*to));
426 	kq_ntimeouts--;
427 }
428 
429 int
430 filt_timer(struct knote *kn, long hint)
431 {
432 	return (kn->kn_data != 0);
433 }
434 
435 
436 /*
437  * filt_seltrue:
438  *
439  *	This filter "event" routine simulates seltrue().
440  */
441 int
442 filt_seltrue(struct knote *kn, long hint)
443 {
444 
445 	/*
446 	 * We don't know how much data can be read/written,
447 	 * but we know that it *can* be.  This is about as
448 	 * good as select/poll does as well.
449 	 */
450 	kn->kn_data = 0;
451 	return (1);
452 }
453 
454 /*
455  * This provides full kqfilter entry for device switch tables, which
456  * has same effect as filter using filt_seltrue() as filter method.
457  */
458 void
459 filt_seltruedetach(struct knote *kn)
460 {
461 	/* Nothing to do */
462 }
463 
464 const struct filterops seltrue_filtops = {
465 	.f_flags	= FILTEROP_ISFD,
466 	.f_attach	= NULL,
467 	.f_detach	= filt_seltruedetach,
468 	.f_event	= filt_seltrue,
469 };
470 
471 int
472 seltrue_kqfilter(dev_t dev, struct knote *kn)
473 {
474 	switch (kn->kn_filter) {
475 	case EVFILT_READ:
476 	case EVFILT_WRITE:
477 		kn->kn_fop = &seltrue_filtops;
478 		break;
479 	default:
480 		return (EINVAL);
481 	}
482 
483 	/* Nothing more to do */
484 	return (0);
485 }
486 
487 static int
488 filt_dead(struct knote *kn, long hint)
489 {
490 	kn->kn_flags |= (EV_EOF | EV_ONESHOT);
491 	if (kn->kn_flags & __EV_POLL)
492 		kn->kn_flags |= __EV_HUP;
493 	kn->kn_data = 0;
494 	return (1);
495 }
496 
497 static void
498 filt_deaddetach(struct knote *kn)
499 {
500 	/* Nothing to do */
501 }
502 
503 const struct filterops dead_filtops = {
504 	.f_flags	= FILTEROP_ISFD,
505 	.f_attach	= NULL,
506 	.f_detach	= filt_deaddetach,
507 	.f_event	= filt_dead,
508 };
509 
510 struct kqueue *
511 kqueue_alloc(struct filedesc *fdp)
512 {
513 	struct kqueue *kq;
514 
515 	kq = pool_get(&kqueue_pool, PR_WAITOK | PR_ZERO);
516 	kq->kq_refs = 1;
517 	kq->kq_fdp = fdp;
518 	TAILQ_INIT(&kq->kq_head);
519 	task_set(&kq->kq_task, kqueue_task, kq);
520 
521 	return (kq);
522 }
523 
524 int
525 sys_kqueue(struct proc *p, void *v, register_t *retval)
526 {
527 	struct filedesc *fdp = p->p_fd;
528 	struct kqueue *kq;
529 	struct file *fp;
530 	int fd, error;
531 
532 	kq = kqueue_alloc(fdp);
533 
534 	fdplock(fdp);
535 	error = falloc(p, &fp, &fd);
536 	if (error)
537 		goto out;
538 	fp->f_flag = FREAD | FWRITE;
539 	fp->f_type = DTYPE_KQUEUE;
540 	fp->f_ops = &kqueueops;
541 	fp->f_data = kq;
542 	*retval = fd;
543 	LIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_next);
544 	kq = NULL;
545 	fdinsert(fdp, fd, 0, fp);
546 	FRELE(fp, p);
547 out:
548 	fdpunlock(fdp);
549 	if (kq != NULL)
550 		pool_put(&kqueue_pool, kq);
551 	return (error);
552 }
553 
554 int
555 sys_kevent(struct proc *p, void *v, register_t *retval)
556 {
557 	struct filedesc* fdp = p->p_fd;
558 	struct sys_kevent_args /* {
559 		syscallarg(int)	fd;
560 		syscallarg(const struct kevent *) changelist;
561 		syscallarg(int)	nchanges;
562 		syscallarg(struct kevent *) eventlist;
563 		syscallarg(int)	nevents;
564 		syscallarg(const struct timespec *) timeout;
565 	} */ *uap = v;
566 	struct kevent *kevp;
567 	struct kqueue *kq;
568 	struct file *fp;
569 	struct timespec ts;
570 	struct timespec *tsp = NULL;
571 	int i, n, nerrors, error;
572 	struct kevent kev[KQ_NEVENTS];
573 
574 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
575 		return (EBADF);
576 
577 	if (fp->f_type != DTYPE_KQUEUE) {
578 		error = EBADF;
579 		goto done;
580 	}
581 
582 	if (SCARG(uap, timeout) != NULL) {
583 		error = copyin(SCARG(uap, timeout), &ts, sizeof(ts));
584 		if (error)
585 			goto done;
586 		if (ts.tv_sec < 0 || !timespecisvalid(&ts)) {
587 			error = EINVAL;
588 			goto done;
589 		}
590 #ifdef KTRACE
591 		if (KTRPOINT(p, KTR_STRUCT))
592 			ktrreltimespec(p, &ts);
593 #endif
594 		tsp = &ts;
595 	}
596 
597 	kq = fp->f_data;
598 	nerrors = 0;
599 
600 	while (SCARG(uap, nchanges) > 0) {
601 		n = SCARG(uap, nchanges) > KQ_NEVENTS ?
602 		    KQ_NEVENTS : SCARG(uap, nchanges);
603 		error = copyin(SCARG(uap, changelist), kev,
604 		    n * sizeof(struct kevent));
605 		if (error)
606 			goto done;
607 #ifdef KTRACE
608 		if (KTRPOINT(p, KTR_STRUCT))
609 			ktrevent(p, kev, n);
610 #endif
611 		for (i = 0; i < n; i++) {
612 			kevp = &kev[i];
613 			kevp->flags &= ~EV_SYSFLAGS;
614 			error = kqueue_register(kq, kevp, p);
615 			if (error || (kevp->flags & EV_RECEIPT)) {
616 				if (SCARG(uap, nevents) != 0) {
617 					kevp->flags = EV_ERROR;
618 					kevp->data = error;
619 					copyout(kevp, SCARG(uap, eventlist),
620 					    sizeof(*kevp));
621 					SCARG(uap, eventlist)++;
622 					SCARG(uap, nevents)--;
623 					nerrors++;
624 				} else {
625 					goto done;
626 				}
627 			}
628 		}
629 		SCARG(uap, nchanges) -= n;
630 		SCARG(uap, changelist) += n;
631 	}
632 	if (nerrors) {
633 		*retval = nerrors;
634 		error = 0;
635 		goto done;
636 	}
637 
638 	KQREF(kq);
639 	FRELE(fp, p);
640 	error = kqueue_scan(kq, SCARG(uap, nevents), SCARG(uap, eventlist),
641 	    tsp, p, &n);
642 	KQRELE(kq);
643 	*retval = n;
644 	return (error);
645 
646  done:
647 	FRELE(fp, p);
648 	return (error);
649 }
650 
651 #ifdef KQUEUE_DEBUG
652 void
653 kqueue_do_check(struct kqueue *kq, const char *func, int line)
654 {
655 	struct knote *kn;
656 	int count = 0, nmarker = 0;
657 
658 	KERNEL_ASSERT_LOCKED();
659 	splassert(IPL_HIGH);
660 
661 	TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) {
662 		if (kn->kn_filter == EVFILT_MARKER) {
663 			if ((kn->kn_status & KN_QUEUED) != 0)
664 				panic("%s:%d: kq=%p kn=%p marker QUEUED",
665 				    func, line, kq, kn);
666 			nmarker++;
667 		} else {
668 			if ((kn->kn_status & KN_ACTIVE) == 0)
669 				panic("%s:%d: kq=%p kn=%p knote !ACTIVE",
670 				    func, line, kq, kn);
671 			if ((kn->kn_status & KN_QUEUED) == 0)
672 				panic("%s:%d: kq=%p kn=%p knote !QUEUED",
673 				    func, line, kq, kn);
674 			if (kn->kn_kq != kq)
675 				panic("%s:%d: kq=%p kn=%p kn_kq=%p != kq",
676 				    func, line, kq, kn, kn->kn_kq);
677 			count++;
678 			if (count > kq->kq_count)
679 				goto bad;
680 		}
681 	}
682 	if (count != kq->kq_count) {
683 bad:
684 		panic("%s:%d: kq=%p kq_count=%d count=%d nmarker=%d",
685 		    func, line, kq, kq->kq_count, count, nmarker);
686 	}
687 }
688 #define kqueue_check(kq)	kqueue_do_check((kq), __func__, __LINE__)
689 #else
690 #define kqueue_check(kq)	do {} while (0)
691 #endif
692 
693 int
694 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p)
695 {
696 	struct filedesc *fdp = kq->kq_fdp;
697 	const struct filterops *fops = NULL;
698 	struct file *fp = NULL;
699 	struct knote *kn = NULL, *newkn = NULL;
700 	struct knlist *list = NULL;
701 	int s, error = 0;
702 
703 	if (kev->filter < 0) {
704 		if (kev->filter + EVFILT_SYSCOUNT < 0)
705 			return (EINVAL);
706 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
707 	}
708 
709 	if (fops == NULL) {
710 		/*
711 		 * XXX
712 		 * filter attach routine is responsible for ensuring that
713 		 * the identifier can be attached to it.
714 		 */
715 		return (EINVAL);
716 	}
717 
718 	if (fops->f_flags & FILTEROP_ISFD) {
719 		/* validate descriptor */
720 		if (kev->ident > INT_MAX)
721 			return (EBADF);
722 	}
723 
724 	if (kev->flags & EV_ADD)
725 		newkn = pool_get(&knote_pool, PR_WAITOK | PR_ZERO);
726 
727 again:
728 	if (fops->f_flags & FILTEROP_ISFD) {
729 		if ((fp = fd_getfile(fdp, kev->ident)) == NULL) {
730 			error = EBADF;
731 			goto done;
732 		}
733 		if (kev->flags & EV_ADD)
734 			kqueue_expand_list(kq, kev->ident);
735 		if (kev->ident < kq->kq_knlistsize)
736 			list = &kq->kq_knlist[kev->ident];
737 	} else {
738 		if (kev->flags & EV_ADD)
739 			kqueue_expand_hash(kq);
740 		if (kq->kq_knhashmask != 0) {
741 			list = &kq->kq_knhash[
742 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
743 		}
744 	}
745 	if (list != NULL) {
746 		SLIST_FOREACH(kn, list, kn_link) {
747 			if (kev->filter == kn->kn_filter &&
748 			    kev->ident == kn->kn_id) {
749 				s = splhigh();
750 				if (!knote_acquire(kn)) {
751 					splx(s);
752 					if (fp != NULL) {
753 						FRELE(fp, p);
754 						fp = NULL;
755 					}
756 					goto again;
757 				}
758 				splx(s);
759 				break;
760 			}
761 		}
762 	}
763 	KASSERT(kn == NULL || (kn->kn_status & KN_PROCESSING) != 0);
764 
765 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
766 		error = ENOENT;
767 		goto done;
768 	}
769 
770 	/*
771 	 * kn now contains the matching knote, or NULL if no match.
772 	 * If adding a new knote, sleeping is not allowed until the knote
773 	 * has been inserted.
774 	 */
775 	if (kev->flags & EV_ADD) {
776 		if (kn == NULL) {
777 			kn = newkn;
778 			newkn = NULL;
779 			kn->kn_status = KN_PROCESSING;
780 			kn->kn_fp = fp;
781 			kn->kn_kq = kq;
782 			kn->kn_fop = fops;
783 
784 			/*
785 			 * apply reference count to knote structure, and
786 			 * do not release it at the end of this routine.
787 			 */
788 			fp = NULL;
789 
790 			kn->kn_sfflags = kev->fflags;
791 			kn->kn_sdata = kev->data;
792 			kev->fflags = 0;
793 			kev->data = 0;
794 			kn->kn_kevent = *kev;
795 
796 			knote_attach(kn);
797 			if ((error = fops->f_attach(kn)) != 0) {
798 				knote_drop(kn, p);
799 				goto done;
800 			}
801 
802 			/*
803 			 * If this is a file descriptor filter, check if
804 			 * fd was closed while the knote was being added.
805 			 * knote_fdclose() has missed kn if the function
806 			 * ran before kn appeared in kq_knlist.
807 			 */
808 			if ((fops->f_flags & FILTEROP_ISFD) &&
809 			    fd_checkclosed(fdp, kev->ident, kn->kn_fp)) {
810 				/*
811 				 * Drop the knote silently without error
812 				 * because another thread might already have
813 				 * seen it. This corresponds to the insert
814 				 * happening in full before the close.
815 				 */
816 				kn->kn_fop->f_detach(kn);
817 				knote_drop(kn, p);
818 				goto done;
819 			}
820 		} else {
821 			/*
822 			 * The user may change some filter values after the
823 			 * initial EV_ADD, but doing so will not reset any
824 			 * filters which have already been triggered.
825 			 */
826 			kn->kn_sfflags = kev->fflags;
827 			kn->kn_sdata = kev->data;
828 			kn->kn_kevent.udata = kev->udata;
829 		}
830 
831 		s = splhigh();
832 		if (kn->kn_fop->f_event(kn, 0))
833 			knote_activate(kn);
834 		splx(s);
835 
836 	} else if (kev->flags & EV_DELETE) {
837 		kn->kn_fop->f_detach(kn);
838 		knote_drop(kn, p);
839 		goto done;
840 	}
841 
842 	if ((kev->flags & EV_DISABLE) &&
843 	    ((kn->kn_status & KN_DISABLED) == 0)) {
844 		s = splhigh();
845 		kn->kn_status |= KN_DISABLED;
846 		splx(s);
847 	}
848 
849 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
850 		s = splhigh();
851 		kn->kn_status &= ~KN_DISABLED;
852 		if (kn->kn_fop->f_event(kn, 0))
853 			kn->kn_status |= KN_ACTIVE;
854 		if ((kn->kn_status & KN_ACTIVE) &&
855 		    ((kn->kn_status & KN_QUEUED) == 0))
856 			knote_enqueue(kn);
857 		splx(s);
858 	}
859 
860 	s = splhigh();
861 	knote_release(kn);
862 	splx(s);
863 done:
864 	if (fp != NULL)
865 		FRELE(fp, p);
866 	if (newkn != NULL)
867 		pool_put(&knote_pool, newkn);
868 	return (error);
869 }
870 
871 int
872 kqueue_sleep(struct kqueue *kq, struct timespec *tsp)
873 {
874 	struct timespec elapsed, start, stop;
875 	uint64_t nsecs;
876 	int error;
877 
878 	splassert(IPL_HIGH);
879 
880 	if (tsp != NULL) {
881 		getnanouptime(&start);
882 		nsecs = MIN(TIMESPEC_TO_NSEC(tsp), MAXTSLP);
883 	} else
884 		nsecs = INFSLP;
885 	error = tsleep_nsec(kq, PSOCK | PCATCH, "kqread", nsecs);
886 	if (tsp != NULL) {
887 		getnanouptime(&stop);
888 		timespecsub(&stop, &start, &elapsed);
889 		timespecsub(tsp, &elapsed, tsp);
890 		if (tsp->tv_sec < 0)
891 			timespecclear(tsp);
892 	}
893 
894 	return (error);
895 }
896 
897 int
898 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent *ulistp,
899     struct timespec *tsp, struct proc *p, int *retval)
900 {
901 	struct kevent *kevp;
902 	struct knote mend, mstart, *kn;
903 	int s, count, nkev = 0, error = 0;
904 	struct kevent kev[KQ_NEVENTS];
905 
906 	count = maxevents;
907 	if (count == 0)
908 		goto done;
909 
910 	memset(&mstart, 0, sizeof(mstart));
911 	memset(&mend, 0, sizeof(mend));
912 
913 retry:
914 	if (kq->kq_state & KQ_DYING) {
915 		error = EBADF;
916 		goto done;
917 	}
918 
919 	kevp = &kev[0];
920 	s = splhigh();
921 	if (kq->kq_count == 0) {
922 		if (tsp != NULL && !timespecisset(tsp)) {
923 			splx(s);
924 			error = 0;
925 			goto done;
926 		}
927 		kq->kq_state |= KQ_SLEEP;
928 		error = kqueue_sleep(kq, tsp);
929 		splx(s);
930 		if (error == 0 || error == EWOULDBLOCK)
931 			goto retry;
932 		/* don't restart after signals... */
933 		if (error == ERESTART)
934 			error = EINTR;
935 		goto done;
936 	}
937 
938 	mstart.kn_filter = EVFILT_MARKER;
939 	mstart.kn_status = KN_PROCESSING;
940 	TAILQ_INSERT_HEAD(&kq->kq_head, &mstart, kn_tqe);
941 	mend.kn_filter = EVFILT_MARKER;
942 	mend.kn_status = KN_PROCESSING;
943 	TAILQ_INSERT_TAIL(&kq->kq_head, &mend, kn_tqe);
944 	while (count) {
945 		kn = TAILQ_NEXT(&mstart, kn_tqe);
946 		if (kn->kn_filter == EVFILT_MARKER) {
947 			if (kn == &mend) {
948 				TAILQ_REMOVE(&kq->kq_head, &mend, kn_tqe);
949 				TAILQ_REMOVE(&kq->kq_head, &mstart, kn_tqe);
950 				splx(s);
951 				if (count == maxevents)
952 					goto retry;
953 				goto done;
954 			}
955 
956 			/* Move start marker past another thread's marker. */
957 			TAILQ_REMOVE(&kq->kq_head, &mstart, kn_tqe);
958 			TAILQ_INSERT_AFTER(&kq->kq_head, kn, &mstart, kn_tqe);
959 			continue;
960 		}
961 
962 		if (!knote_acquire(kn))
963 			continue;
964 
965 		kqueue_check(kq);
966 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
967 		kn->kn_status &= ~KN_QUEUED;
968 		kq->kq_count--;
969 		kqueue_check(kq);
970 
971 		if (kn->kn_status & KN_DISABLED) {
972 			knote_release(kn);
973 			continue;
974 		}
975 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
976 		    kn->kn_fop->f_event(kn, 0) == 0) {
977 			if ((kn->kn_status & KN_QUEUED) == 0)
978 				kn->kn_status &= ~KN_ACTIVE;
979 			knote_release(kn);
980 			kqueue_check(kq);
981 			continue;
982 		}
983 		*kevp = kn->kn_kevent;
984 		kevp++;
985 		nkev++;
986 		if (kn->kn_flags & EV_ONESHOT) {
987 			splx(s);
988 			kn->kn_fop->f_detach(kn);
989 			knote_drop(kn, p);
990 			s = splhigh();
991 		} else if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
992 			if (kn->kn_flags & EV_CLEAR) {
993 				kn->kn_data = 0;
994 				kn->kn_fflags = 0;
995 			}
996 			if (kn->kn_flags & EV_DISPATCH)
997 				kn->kn_status |= KN_DISABLED;
998 			if ((kn->kn_status & KN_QUEUED) == 0)
999 				kn->kn_status &= ~KN_ACTIVE;
1000 			knote_release(kn);
1001 		} else {
1002 			if ((kn->kn_status & KN_QUEUED) == 0) {
1003 				kqueue_check(kq);
1004 				kq->kq_count++;
1005 				kn->kn_status |= KN_QUEUED;
1006 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1007 			}
1008 			knote_release(kn);
1009 		}
1010 		kqueue_check(kq);
1011 		count--;
1012 		if (nkev == KQ_NEVENTS) {
1013 			splx(s);
1014 #ifdef KTRACE
1015 			if (KTRPOINT(p, KTR_STRUCT))
1016 				ktrevent(p, kev, nkev);
1017 #endif
1018 			error = copyout(kev, ulistp,
1019 			    sizeof(struct kevent) * nkev);
1020 			ulistp += nkev;
1021 			nkev = 0;
1022 			kevp = &kev[0];
1023 			s = splhigh();
1024 			if (error)
1025 				break;
1026 		}
1027 	}
1028 	TAILQ_REMOVE(&kq->kq_head, &mend, kn_tqe);
1029 	TAILQ_REMOVE(&kq->kq_head, &mstart, kn_tqe);
1030 	splx(s);
1031 done:
1032 	if (nkev != 0) {
1033 #ifdef KTRACE
1034 		if (KTRPOINT(p, KTR_STRUCT))
1035 			ktrevent(p, kev, nkev);
1036 #endif
1037 		error = copyout(kev, ulistp,
1038 		    sizeof(struct kevent) * nkev);
1039 	}
1040 	*retval = maxevents - count;
1041 	return (error);
1042 }
1043 
1044 /*
1045  * XXX
1046  * This could be expanded to call kqueue_scan, if desired.
1047  */
1048 int
1049 kqueue_read(struct file *fp, struct uio *uio, int fflags)
1050 {
1051 	return (ENXIO);
1052 }
1053 
1054 int
1055 kqueue_write(struct file *fp, struct uio *uio, int fflags)
1056 {
1057 	return (ENXIO);
1058 }
1059 
1060 int
1061 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
1062 {
1063 	return (ENOTTY);
1064 }
1065 
1066 int
1067 kqueue_poll(struct file *fp, int events, struct proc *p)
1068 {
1069 	struct kqueue *kq = (struct kqueue *)fp->f_data;
1070 	int revents = 0;
1071 	int s = splhigh();
1072 
1073 	if (events & (POLLIN | POLLRDNORM)) {
1074 		if (kq->kq_count) {
1075 			revents |= events & (POLLIN | POLLRDNORM);
1076 		} else {
1077 			selrecord(p, &kq->kq_sel);
1078 			kq->kq_state |= KQ_SEL;
1079 		}
1080 	}
1081 	splx(s);
1082 	return (revents);
1083 }
1084 
1085 int
1086 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
1087 {
1088 	struct kqueue *kq = fp->f_data;
1089 
1090 	memset(st, 0, sizeof(*st));
1091 	st->st_size = kq->kq_count;
1092 	st->st_blksize = sizeof(struct kevent);
1093 	st->st_mode = S_IFIFO;
1094 	return (0);
1095 }
1096 
1097 void
1098 kqueue_terminate(struct proc *p, struct kqueue *kq)
1099 {
1100 	int i;
1101 
1102 	KERNEL_ASSERT_LOCKED();
1103 
1104 	for (i = 0; i < kq->kq_knlistsize; i++)
1105 		knote_remove(p, &kq->kq_knlist[i]);
1106 	if (kq->kq_knhashmask != 0) {
1107 		for (i = 0; i < kq->kq_knhashmask + 1; i++)
1108 			knote_remove(p, &kq->kq_knhash[i]);
1109 	}
1110 	kq->kq_state |= KQ_DYING;
1111 	kqueue_wakeup(kq);
1112 
1113 	KASSERT(klist_empty(&kq->kq_sel.si_note));
1114 	task_del(systq, &kq->kq_task);
1115 
1116 }
1117 
1118 int
1119 kqueue_close(struct file *fp, struct proc *p)
1120 {
1121 	struct kqueue *kq = fp->f_data;
1122 
1123 	KERNEL_LOCK();
1124 	kqueue_terminate(p, kq);
1125 	fp->f_data = NULL;
1126 
1127 	KQRELE(kq);
1128 
1129 	KERNEL_UNLOCK();
1130 
1131 	return (0);
1132 }
1133 
1134 static void
1135 kqueue_task(void *arg)
1136 {
1137 	struct kqueue *kq = arg;
1138 
1139 	if (kq->kq_state & KQ_SEL) {
1140 		kq->kq_state &= ~KQ_SEL;
1141 		selwakeup(&kq->kq_sel);
1142 	} else {
1143 		KNOTE(&kq->kq_sel.si_note, 0);
1144 	}
1145 	KQRELE(kq);
1146 }
1147 
1148 void
1149 kqueue_wakeup(struct kqueue *kq)
1150 {
1151 
1152 	if (kq->kq_state & KQ_SLEEP) {
1153 		kq->kq_state &= ~KQ_SLEEP;
1154 		wakeup(kq);
1155 	}
1156 	if ((kq->kq_state & KQ_SEL) || !klist_empty(&kq->kq_sel.si_note)) {
1157 		/* Defer activation to avoid recursion. */
1158 		KQREF(kq);
1159 		if (!task_add(systq, &kq->kq_task))
1160 			KQRELE(kq);
1161 	}
1162 }
1163 
1164 static void
1165 kqueue_expand_hash(struct kqueue *kq)
1166 {
1167 	struct knlist *hash;
1168 	u_long hashmask;
1169 
1170 	if (kq->kq_knhashmask == 0) {
1171 		hash = hashinit(KN_HASHSIZE, M_KEVENT, M_WAITOK, &hashmask);
1172 		if (kq->kq_knhashmask == 0) {
1173 			kq->kq_knhash = hash;
1174 			kq->kq_knhashmask = hashmask;
1175 		} else {
1176 			/* Another thread has allocated the hash. */
1177 			hashfree(hash, KN_HASHSIZE, M_KEVENT);
1178 		}
1179 	}
1180 }
1181 
1182 static void
1183 kqueue_expand_list(struct kqueue *kq, int fd)
1184 {
1185 	struct knlist *list;
1186 	int size;
1187 
1188 	if (kq->kq_knlistsize <= fd) {
1189 		size = kq->kq_knlistsize;
1190 		while (size <= fd)
1191 			size += KQEXTENT;
1192 		list = mallocarray(size, sizeof(*list), M_KEVENT, M_WAITOK);
1193 		if (kq->kq_knlistsize <= fd) {
1194 			memcpy(list, kq->kq_knlist,
1195 			    kq->kq_knlistsize * sizeof(*list));
1196 			memset(&list[kq->kq_knlistsize], 0,
1197 			    (size - kq->kq_knlistsize) * sizeof(*list));
1198 			free(kq->kq_knlist, M_KEVENT,
1199 			    kq->kq_knlistsize * sizeof(*list));
1200 			kq->kq_knlist = list;
1201 			kq->kq_knlistsize = size;
1202 		} else {
1203 			/* Another thread has expanded the list. */
1204 			free(list, M_KEVENT, size * sizeof(*list));
1205 		}
1206 	}
1207 }
1208 
1209 /*
1210  * Acquire a knote, return non-zero on success, 0 on failure.
1211  *
1212  * If we cannot acquire the knote we sleep and return 0.  The knote
1213  * may be stale on return in this case and the caller must restart
1214  * whatever loop they are in.
1215  */
1216 int
1217 knote_acquire(struct knote *kn)
1218 {
1219 	splassert(IPL_HIGH);
1220 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1221 
1222 	if (kn->kn_status & KN_PROCESSING) {
1223 		kn->kn_status |= KN_WAITING;
1224 		tsleep_nsec(kn, 0, "kqepts", SEC_TO_NSEC(1));
1225 		/* knote may be stale now */
1226 		return (0);
1227 	}
1228 	kn->kn_status |= KN_PROCESSING;
1229 	return (1);
1230 }
1231 
1232 /*
1233  * Release an acquired knote, clearing KN_PROCESSING.
1234  */
1235 void
1236 knote_release(struct knote *kn)
1237 {
1238 	splassert(IPL_HIGH);
1239 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1240 	KASSERT(kn->kn_status & KN_PROCESSING);
1241 
1242 	if (kn->kn_status & KN_WAITING) {
1243 		kn->kn_status &= ~KN_WAITING;
1244 		wakeup(kn);
1245 	}
1246 	kn->kn_status &= ~KN_PROCESSING;
1247 	/* kn should not be accessed anymore */
1248 }
1249 
1250 /*
1251  * activate one knote.
1252  */
1253 void
1254 knote_activate(struct knote *kn)
1255 {
1256 	int s;
1257 
1258 	s = splhigh();
1259 	kn->kn_status |= KN_ACTIVE;
1260 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)
1261 		knote_enqueue(kn);
1262 	splx(s);
1263 }
1264 
1265 /*
1266  * walk down a list of knotes, activating them if their event has triggered.
1267  */
1268 void
1269 knote(struct klist *list, long hint)
1270 {
1271 	struct knote *kn, *kn0;
1272 
1273 	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, kn0)
1274 		if (kn->kn_fop->f_event(kn, hint))
1275 			knote_activate(kn);
1276 }
1277 
1278 /*
1279  * remove all knotes from a specified knlist
1280  */
1281 void
1282 knote_remove(struct proc *p, struct knlist *list)
1283 {
1284 	struct knote *kn;
1285 	int s;
1286 
1287 	while ((kn = SLIST_FIRST(list)) != NULL) {
1288 		s = splhigh();
1289 		if (!knote_acquire(kn)) {
1290 			splx(s);
1291 			continue;
1292 		}
1293 		splx(s);
1294 		kn->kn_fop->f_detach(kn);
1295 		knote_drop(kn, p);
1296 	}
1297 }
1298 
1299 /*
1300  * remove all knotes referencing a specified fd
1301  */
1302 void
1303 knote_fdclose(struct proc *p, int fd)
1304 {
1305 	struct filedesc *fdp = p->p_p->ps_fd;
1306 	struct kqueue *kq;
1307 	struct knlist *list;
1308 
1309 	/*
1310 	 * fdplock can be ignored if the file descriptor table is being freed
1311 	 * because no other thread can access the fdp.
1312 	 */
1313 	if (fdp->fd_refcnt != 0)
1314 		fdpassertlocked(fdp);
1315 
1316 	if (LIST_EMPTY(&fdp->fd_kqlist))
1317 		return;
1318 
1319 	KERNEL_LOCK();
1320 	LIST_FOREACH(kq, &fdp->fd_kqlist, kq_next) {
1321 		if (fd >= kq->kq_knlistsize)
1322 			continue;
1323 
1324 		list = &kq->kq_knlist[fd];
1325 		knote_remove(p, list);
1326 	}
1327 	KERNEL_UNLOCK();
1328 }
1329 
1330 /*
1331  * handle a process exiting, including the triggering of NOTE_EXIT notes
1332  * XXX this could be more efficient, doing a single pass down the klist
1333  */
1334 void
1335 knote_processexit(struct proc *p)
1336 {
1337 	struct process *pr = p->p_p;
1338 
1339 	KASSERT(p == curproc);
1340 
1341 	KNOTE(&pr->ps_klist, NOTE_EXIT);
1342 
1343 	/* remove other knotes hanging off the process */
1344 	klist_invalidate(&pr->ps_klist);
1345 }
1346 
1347 void
1348 knote_attach(struct knote *kn)
1349 {
1350 	struct kqueue *kq = kn->kn_kq;
1351 	struct knlist *list;
1352 
1353 	if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
1354 		KASSERT(kq->kq_knlistsize > kn->kn_id);
1355 		list = &kq->kq_knlist[kn->kn_id];
1356 	} else {
1357 		KASSERT(kq->kq_knhashmask != 0);
1358 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1359 	}
1360 	SLIST_INSERT_HEAD(list, kn, kn_link);
1361 }
1362 
1363 /*
1364  * should be called at spl == 0, since we don't want to hold spl
1365  * while calling FRELE and pool_put.
1366  */
1367 void
1368 knote_drop(struct knote *kn, struct proc *p)
1369 {
1370 	struct kqueue *kq = kn->kn_kq;
1371 	struct knlist *list;
1372 	int s;
1373 
1374 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1375 
1376 	if (kn->kn_fop->f_flags & FILTEROP_ISFD)
1377 		list = &kq->kq_knlist[kn->kn_id];
1378 	else
1379 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1380 
1381 	SLIST_REMOVE(list, kn, knote, kn_link);
1382 	s = splhigh();
1383 	if (kn->kn_status & KN_QUEUED)
1384 		knote_dequeue(kn);
1385 	if (kn->kn_status & KN_WAITING) {
1386 		kn->kn_status &= ~KN_WAITING;
1387 		wakeup(kn);
1388 	}
1389 	splx(s);
1390 	if (kn->kn_fop->f_flags & FILTEROP_ISFD)
1391 		FRELE(kn->kn_fp, p);
1392 	pool_put(&knote_pool, kn);
1393 }
1394 
1395 
1396 void
1397 knote_enqueue(struct knote *kn)
1398 {
1399 	struct kqueue *kq = kn->kn_kq;
1400 
1401 	splassert(IPL_HIGH);
1402 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1403 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
1404 
1405 	kqueue_check(kq);
1406 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1407 	kn->kn_status |= KN_QUEUED;
1408 	kq->kq_count++;
1409 	kqueue_check(kq);
1410 	kqueue_wakeup(kq);
1411 }
1412 
1413 void
1414 knote_dequeue(struct knote *kn)
1415 {
1416 	struct kqueue *kq = kn->kn_kq;
1417 
1418 	splassert(IPL_HIGH);
1419 	KASSERT(kn->kn_filter != EVFILT_MARKER);
1420 	KASSERT(kn->kn_status & KN_QUEUED);
1421 
1422 	kqueue_check(kq);
1423 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1424 	kn->kn_status &= ~KN_QUEUED;
1425 	kq->kq_count--;
1426 	kqueue_check(kq);
1427 }
1428 
1429 void
1430 klist_insert(struct klist *klist, struct knote *kn)
1431 {
1432 	SLIST_INSERT_HEAD(&klist->kl_list, kn, kn_selnext);
1433 }
1434 
1435 void
1436 klist_remove(struct klist *klist, struct knote *kn)
1437 {
1438 	SLIST_REMOVE(&klist->kl_list, kn, knote, kn_selnext);
1439 }
1440 
1441 int
1442 klist_empty(struct klist *klist)
1443 {
1444 	return (SLIST_EMPTY(&klist->kl_list));
1445 }
1446 
1447 void
1448 klist_invalidate(struct klist *list)
1449 {
1450 	struct knote *kn;
1451 	struct proc *p = curproc;
1452 	int s;
1453 
1454 	/*
1455 	 * NET_LOCK() must not be held because it can block another thread
1456 	 * in f_event with a knote acquired.
1457 	 */
1458 	NET_ASSERT_UNLOCKED();
1459 
1460 	s = splhigh();
1461 	while ((kn = SLIST_FIRST(&list->kl_list)) != NULL) {
1462 		if (!knote_acquire(kn))
1463 			continue;
1464 		splx(s);
1465 		kn->kn_fop->f_detach(kn);
1466 		if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
1467 			kn->kn_fop = &dead_filtops;
1468 			knote_activate(kn);
1469 			s = splhigh();
1470 			knote_release(kn);
1471 		} else {
1472 			knote_drop(kn, p);
1473 			s = splhigh();
1474 		}
1475 	}
1476 	splx(s);
1477 }
1478