xref: /netbsd-src/sys/kern/kern_event.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: kern_event.c,v 1.13 2003/03/21 21:13:51 dsl Exp $	*/
2 /*-
3  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp $
28  */
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/proc.h>
34 #include <sys/malloc.h>
35 #include <sys/unistd.h>
36 #include <sys/file.h>
37 #include <sys/fcntl.h>
38 #include <sys/select.h>
39 #include <sys/queue.h>
40 #include <sys/event.h>
41 #include <sys/eventvar.h>
42 #include <sys/poll.h>
43 #include <sys/pool.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/stat.h>
48 #include <sys/uio.h>
49 #include <sys/mount.h>
50 #include <sys/filedesc.h>
51 #include <sys/sa.h>
52 #include <sys/syscallargs.h>
53 
54 static int	kqueue_scan(struct file *fp, size_t maxevents,
55 		    struct kevent *ulistp, const struct timespec *timeout,
56 		    struct proc *p, register_t *retval);
57 static void	kqueue_wakeup(struct kqueue *kq);
58 
59 static int	kqueue_read(struct file *fp, off_t *offset, struct uio *uio,
60 		    struct ucred *cred, int flags);
61 static int	kqueue_write(struct file *fp, off_t *offset, struct uio *uio,
62 		    struct ucred *cred, int flags);
63 static int	kqueue_ioctl(struct file *fp, u_long com, void *data,
64 		    struct proc *p);
65 static int	kqueue_fcntl(struct file *fp, u_int com, void *data,
66 		    struct proc *p);
67 static int	kqueue_poll(struct file *fp, int events, struct proc *p);
68 static int	kqueue_kqfilter(struct file *fp, struct knote *kn);
69 static int	kqueue_stat(struct file *fp, struct stat *sp, struct proc *p);
70 static int	kqueue_close(struct file *fp, struct proc *p);
71 
72 static struct fileops kqueueops = {
73 	kqueue_read, kqueue_write, kqueue_ioctl, kqueue_fcntl, kqueue_poll,
74 	kqueue_stat, kqueue_close, kqueue_kqfilter
75 };
76 
77 static void	knote_attach(struct knote *kn, struct filedesc *fdp);
78 static void	knote_drop(struct knote *kn, struct proc *p,
79 		    struct filedesc *fdp);
80 static void	knote_enqueue(struct knote *kn);
81 static void	knote_dequeue(struct knote *kn);
82 
83 static void	filt_kqdetach(struct knote *kn);
84 static int	filt_kqueue(struct knote *kn, long hint);
85 static int	filt_procattach(struct knote *kn);
86 static void	filt_procdetach(struct knote *kn);
87 static int	filt_proc(struct knote *kn, long hint);
88 static int	filt_fileattach(struct knote *kn);
89 static void	filt_timerexpire(void *knx);
90 static int	filt_timerattach(struct knote *kn);
91 static void	filt_timerdetach(struct knote *kn);
92 static int	filt_timer(struct knote *kn, long hint);
93 
94 static const struct filterops kqread_filtops =
95 	{ 1, NULL, filt_kqdetach, filt_kqueue };
96 static const struct filterops proc_filtops =
97 	{ 0, filt_procattach, filt_procdetach, filt_proc };
98 static const struct filterops file_filtops =
99 	{ 1, filt_fileattach, NULL, NULL };
100 static struct filterops timer_filtops =
101 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
102 
103 struct pool	kqueue_pool;
104 struct pool	knote_pool;
105 static int	kq_ncallouts = 0;
106 static int	kq_calloutmax = (4 * 1024);
107 
108 MALLOC_DEFINE(M_KEVENT, "kevent", "kevents/knotes");
109 
110 #define	KNOTE_ACTIVATE(kn)						\
111 do {									\
112 	kn->kn_status |= KN_ACTIVE;					\
113 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
114 		knote_enqueue(kn);					\
115 } while(0)
116 
117 #define	KN_HASHSIZE		64		/* XXX should be tunable */
118 #define	KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
119 
120 extern const struct filterops sig_filtops;
121 
122 /*
123  * Table for for all system-defined filters.
124  * These should be listed in the numeric order of the EVFILT_* defines.
125  * If filtops is NULL, the filter isn't implemented in NetBSD.
126  * End of list is when name is NULL.
127  */
128 struct kfilter {
129 	const char	 *name;		/* name of filter */
130 	uint32_t	  filter;	/* id of filter */
131 	const struct filterops *filtops;/* operations for filter */
132 };
133 
134 		/* System defined filters */
135 static const struct kfilter sys_kfilters[] = {
136 	{ "EVFILT_READ",	EVFILT_READ,	&file_filtops },
137 	{ "EVFILT_WRITE",	EVFILT_WRITE,	&file_filtops },
138 	{ "EVFILT_AIO",		EVFILT_AIO,	NULL },
139 	{ "EVFILT_VNODE",	EVFILT_VNODE,	&file_filtops },
140 	{ "EVFILT_PROC",	EVFILT_PROC,	&proc_filtops },
141 	{ "EVFILT_SIGNAL",	EVFILT_SIGNAL,	&sig_filtops },
142 	{ "EVFILT_TIMER",	EVFILT_TIMER,	&timer_filtops },
143 	{ NULL,			0,		NULL },	/* end of list */
144 };
145 
146 		/* User defined kfilters */
147 static struct kfilter	*user_kfilters;		/* array */
148 static int		user_kfilterc;		/* current offset */
149 static int		user_kfiltermaxc;	/* max size so far */
150 
151 /*
152  * kqueue_init:
153  *
154  *	Initialize the kqueue/knote facility.
155  */
156 void
157 kqueue_init(void)
158 {
159 
160 	pool_init(&kqueue_pool, sizeof(struct kqueue), 0, 0, 0, "kqueuepl",
161 	    NULL);
162 	pool_init(&knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl",
163 	    NULL);
164 }
165 
166 /*
167  * Find kfilter entry by name, or NULL if not found.
168  */
169 static const struct kfilter *
170 kfilter_byname_sys(const char *name)
171 {
172 	int i;
173 
174 	for (i = 0; sys_kfilters[i].name != NULL; i++) {
175 		if (strcmp(name, sys_kfilters[i].name) == 0)
176 			return (&sys_kfilters[i]);
177 	}
178 	return (NULL);
179 }
180 
181 static struct kfilter *
182 kfilter_byname_user(const char *name)
183 {
184 	int i;
185 
186 	/* user_kfilters[] could be NULL if no filters were registered */
187 	if (!user_kfilters)
188 		return (NULL);
189 
190 	for (i = 0; user_kfilters[i].name != NULL; i++) {
191 		if (user_kfilters[i].name != '\0' &&
192 		    strcmp(name, user_kfilters[i].name) == 0)
193 			return (&user_kfilters[i]);
194 	}
195 	return (NULL);
196 }
197 
198 static const struct kfilter *
199 kfilter_byname(const char *name)
200 {
201 	const struct kfilter *kfilter;
202 
203 	if ((kfilter = kfilter_byname_sys(name)) != NULL)
204 		return (kfilter);
205 
206 	return (kfilter_byname_user(name));
207 }
208 
209 /*
210  * Find kfilter entry by filter id, or NULL if not found.
211  * Assumes entries are indexed in filter id order, for speed.
212  */
213 static const struct kfilter *
214 kfilter_byfilter(uint32_t filter)
215 {
216 	const struct kfilter *kfilter;
217 
218 	if (filter < EVFILT_SYSCOUNT)	/* it's a system filter */
219 		kfilter = &sys_kfilters[filter];
220 	else if (user_kfilters != NULL &&
221 	    filter < EVFILT_SYSCOUNT + user_kfilterc)
222 					/* it's a user filter */
223 		kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT];
224 	else
225 		return (NULL);		/* out of range */
226 	KASSERT(kfilter->filter == filter);	/* sanity check! */
227 	return (kfilter);
228 }
229 
230 /*
231  * Register a new kfilter. Stores the entry in user_kfilters.
232  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
233  * If retfilter != NULL, the new filterid is returned in it.
234  */
235 int
236 kfilter_register(const char *name, const struct filterops *filtops,
237     int *retfilter)
238 {
239 	struct kfilter *kfilter;
240 	void *space;
241 	int len;
242 
243 	if (name == NULL || name[0] == '\0' || filtops == NULL)
244 		return (EINVAL);	/* invalid args */
245 	if (kfilter_byname(name) != NULL)
246 		return (EEXIST);	/* already exists */
247 	if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT)
248 		return (EINVAL);	/* too many */
249 
250 	/* check if need to grow user_kfilters */
251 	if (user_kfilterc + 1 > user_kfiltermaxc) {
252 		/*
253 		 * Grow in KFILTER_EXTENT chunks. Use malloc(9), because we
254 		 * want to traverse user_kfilters as an array.
255 		 */
256 		user_kfiltermaxc += KFILTER_EXTENT;
257 		kfilter = malloc(user_kfiltermaxc * sizeof(struct filter *),
258 		    M_KEVENT, M_WAITOK);
259 
260 		/* copy existing user_kfilters */
261 		if (user_kfilters != NULL)
262 			memcpy((caddr_t)kfilter, (caddr_t)user_kfilters,
263 			    user_kfilterc * sizeof(struct kfilter *));
264 					/* zero new sections */
265 		memset((caddr_t)kfilter +
266 		    user_kfilterc * sizeof(struct kfilter *), 0,
267 		    (user_kfiltermaxc - user_kfilterc) *
268 		    sizeof(struct kfilter *));
269 					/* switch to new kfilter */
270 		if (user_kfilters != NULL)
271 			free(user_kfilters, M_KEVENT);
272 		user_kfilters = kfilter;
273 	}
274 	len = strlen(name) + 1;		/* copy name */
275 	space = malloc(len, M_KEVENT, M_WAITOK);
276 	memcpy(space, name, len);
277 	user_kfilters[user_kfilterc].name = space;
278 
279 	user_kfilters[user_kfilterc].filter = user_kfilterc + EVFILT_SYSCOUNT;
280 
281 	len = sizeof(struct filterops);	/* copy filtops */
282 	space = malloc(len, M_KEVENT, M_WAITOK);
283 	memcpy(space, filtops, len);
284 	user_kfilters[user_kfilterc].filtops = space;
285 
286 	if (retfilter != NULL)
287 		*retfilter = user_kfilters[user_kfilterc].filter;
288 	user_kfilterc++;		/* finally, increment count */
289 	return (0);
290 }
291 
292 /*
293  * Unregister a kfilter previously registered with kfilter_register.
294  * This retains the filter id, but clears the name and frees filtops (filter
295  * operations), so that the number isn't reused during a boot.
296  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
297  */
298 int
299 kfilter_unregister(const char *name)
300 {
301 	struct kfilter *kfilter;
302 
303 	if (name == NULL || name[0] == '\0')
304 		return (EINVAL);	/* invalid name */
305 
306 	if (kfilter_byname_sys(name) != NULL)
307 		return (EINVAL);	/* can't detach system filters */
308 
309 	kfilter = kfilter_byname_user(name);
310 	if (kfilter == NULL)		/* not found */
311 		return (ENOENT);
312 
313 	if (kfilter->name[0] != '\0') {
314 		/* XXX Cast away const (but we know it's safe. */
315 		free((void *) kfilter->name, M_KEVENT);
316 		kfilter->name = "";	/* mark as `not implemented' */
317 	}
318 	if (kfilter->filtops != NULL) {
319 		/* XXX Cast away const (but we know it's safe. */
320 		free((void *) kfilter->filtops, M_KEVENT);
321 		kfilter->filtops = NULL; /* mark as `not implemented' */
322 	}
323 	return (0);
324 }
325 
326 
327 /*
328  * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file
329  * descriptors. Calls struct fileops kqfilter method for given file descriptor.
330  */
331 static int
332 filt_fileattach(struct knote *kn)
333 {
334 	struct file *fp;
335 
336 	fp = kn->kn_fp;
337 	return ((*fp->f_ops->fo_kqfilter)(fp, kn));
338 }
339 
340 /*
341  * Filter detach method for EVFILT_READ on kqueue descriptor.
342  */
343 static void
344 filt_kqdetach(struct knote *kn)
345 {
346 	struct kqueue *kq;
347 
348 	kq = (struct kqueue *)kn->kn_fp->f_data;
349 	SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext);
350 }
351 
352 /*
353  * Filter event method for EVFILT_READ on kqueue descriptor.
354  */
355 /*ARGSUSED*/
356 static int
357 filt_kqueue(struct knote *kn, long hint)
358 {
359 	struct kqueue *kq;
360 
361 	kq = (struct kqueue *)kn->kn_fp->f_data;
362 	kn->kn_data = kq->kq_count;
363 	return (kn->kn_data > 0);
364 }
365 
366 /*
367  * Filter attach method for EVFILT_PROC.
368  */
369 static int
370 filt_procattach(struct knote *kn)
371 {
372 	struct proc *p;
373 
374 	p = pfind(kn->kn_id);
375 	if (p == NULL)
376 		return (ESRCH);
377 
378 	/*
379 	 * Fail if it's not owned by you, or the last exec gave us
380 	 * setuid/setgid privs (unless you're root).
381 	 */
382 	if ((p->p_cred->p_ruid != curproc->p_cred->p_ruid ||
383 		(p->p_flag & P_SUGID))
384 	    && suser(curproc->p_ucred, &curproc->p_acflag) != 0)
385 		return (EACCES);
386 
387 	kn->kn_ptr.p_proc = p;
388 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
389 
390 	/*
391 	 * internal flag indicating registration done by kernel
392 	 */
393 	if (kn->kn_flags & EV_FLAG1) {
394 		kn->kn_data = kn->kn_sdata;	/* ppid */
395 		kn->kn_fflags = NOTE_CHILD;
396 		kn->kn_flags &= ~EV_FLAG1;
397 	}
398 
399 	/* XXXSMP lock the process? */
400 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
401 
402 	return (0);
403 }
404 
405 /*
406  * Filter detach method for EVFILT_PROC.
407  *
408  * The knote may be attached to a different process, which may exit,
409  * leaving nothing for the knote to be attached to.  So when the process
410  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
411  * it will be deleted when read out.  However, as part of the knote deletion,
412  * this routine is called, so a check is needed to avoid actually performing
413  * a detach, because the original process might not exist any more.
414  */
415 static void
416 filt_procdetach(struct knote *kn)
417 {
418 	struct proc *p;
419 
420 	if (kn->kn_status & KN_DETACHED)
421 		return;
422 
423 	p = kn->kn_ptr.p_proc;
424 	KASSERT(p->p_stat == SDEAD || pfind(kn->kn_id) == p);
425 
426 	/* XXXSMP lock the process? */
427 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
428 }
429 
430 /*
431  * Filter event method for EVFILT_PROC.
432  */
433 static int
434 filt_proc(struct knote *kn, long hint)
435 {
436 	u_int event;
437 
438 	/*
439 	 * mask off extra data
440 	 */
441 	event = (u_int)hint & NOTE_PCTRLMASK;
442 
443 	/*
444 	 * if the user is interested in this event, record it.
445 	 */
446 	if (kn->kn_sfflags & event)
447 		kn->kn_fflags |= event;
448 
449 	/*
450 	 * process is gone, so flag the event as finished.
451 	 */
452 	if (event == NOTE_EXIT) {
453 		/*
454 		 * Detach the knote from watched process and mark
455 		 * it as such. We can't leave this to kqueue_scan(),
456 		 * since the process might not exist by then. And we
457 		 * have to do this now, since psignal KNOTE() is called
458 		 * also for zombies and we might end up reading freed
459 		 * memory if the kevent would already be picked up
460 		 * and knote g/c'ed.
461 		 */
462 		kn->kn_fop->f_detach(kn);
463 		kn->kn_status |= KN_DETACHED;
464 
465 		/* Mark as ONESHOT, so that the knote it g/c'ed when read */
466 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
467 		return (1);
468 	}
469 
470 	/*
471 	 * process forked, and user wants to track the new process,
472 	 * so attach a new knote to it, and immediately report an
473 	 * event with the parent's pid.
474 	 */
475 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
476 		struct kevent kev;
477 		int error;
478 
479 		/*
480 		 * register knote with new process.
481 		 */
482 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
483 		kev.filter = kn->kn_filter;
484 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
485 		kev.fflags = kn->kn_sfflags;
486 		kev.data = kn->kn_id;			/* parent */
487 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
488 		error = kqueue_register(kn->kn_kq, &kev, NULL);
489 		if (error)
490 			kn->kn_fflags |= NOTE_TRACKERR;
491 	}
492 
493 	return (kn->kn_fflags != 0);
494 }
495 
496 static void
497 filt_timerexpire(void *knx)
498 {
499 	struct knote *kn = knx;
500 	int tticks;
501 
502 	kn->kn_data++;
503 	KNOTE_ACTIVATE(kn);
504 
505 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
506 		tticks = mstohz(kn->kn_sdata);
507 		callout_schedule((struct callout *)kn->kn_hook, tticks);
508 	}
509 }
510 
511 /*
512  * data contains amount of time to sleep, in milliseconds
513  */
514 static int
515 filt_timerattach(struct knote *kn)
516 {
517 	struct callout *calloutp;
518 	int tticks;
519 
520 	if (kq_ncallouts >= kq_calloutmax)
521 		return (ENOMEM);
522 	kq_ncallouts++;
523 
524 	tticks = mstohz(kn->kn_sdata);
525 
526 	/* if the supplied value is under our resolution, use 1 tick */
527 	if (tticks == 0) {
528 		if (kn->kn_sdata == 0)
529 			return (EINVAL);
530 		tticks = 1;
531 	}
532 
533 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
534 	MALLOC(calloutp, struct callout *, sizeof(*calloutp),
535 	    M_KEVENT, 0);
536 	callout_init(calloutp);
537 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
538 	kn->kn_hook = calloutp;
539 
540 	return (0);
541 }
542 
543 static void
544 filt_timerdetach(struct knote *kn)
545 {
546 	struct callout *calloutp;
547 
548 	calloutp = (struct callout *)kn->kn_hook;
549 	callout_stop(calloutp);
550 	FREE(calloutp, M_KEVENT);
551 	kq_ncallouts--;
552 }
553 
554 static int
555 filt_timer(struct knote *kn, long hint)
556 {
557 	return (kn->kn_data != 0);
558 }
559 
560 /*
561  * filt_seltrue:
562  *
563  *	This filter "event" routine simulates seltrue().
564  */
565 int
566 filt_seltrue(struct knote *kn, long hint)
567 {
568 
569 	/*
570 	 * We don't know how much data can be read/written,
571 	 * but we know that it *can* be.  This is about as
572 	 * good as select/poll does as well.
573 	 */
574 	kn->kn_data = 0;
575 	return (1);
576 }
577 
578 /*
579  * This provides full kqfilter entry for device switch tables, which
580  * has same effect as filter using filt_seltrue() as filter method.
581  */
582 static void
583 filt_seltruedetach(struct knote *kn)
584 {
585 	/* Nothing to do */
586 }
587 
588 static const struct filterops seltrue_filtops =
589 	{ 1, NULL, filt_seltruedetach, filt_seltrue };
590 
591 int
592 seltrue_kqfilter(dev_t dev, struct knote *kn)
593 {
594 	switch (kn->kn_filter) {
595 	case EVFILT_READ:
596 	case EVFILT_WRITE:
597 		kn->kn_fop = &seltrue_filtops;
598 		break;
599 	default:
600 		return (1);
601 	}
602 
603 	/* Nothing more to do */
604 	return (0);
605 }
606 
607 /*
608  * kqueue(2) system call.
609  */
610 int
611 sys_kqueue(struct lwp *l, void *v, register_t *retval)
612 {
613 	struct filedesc	*fdp;
614 	struct kqueue	*kq;
615 	struct file	*fp;
616 	struct proc	*p;
617 	int		fd, error;
618 
619 	p = l->l_proc;
620 	fdp = p->p_fd;
621 	error = falloc(p, &fp, &fd);	/* setup a new file descriptor */
622 	if (error)
623 		return (error);
624 	fp->f_flag = FREAD | FWRITE;
625 	fp->f_type = DTYPE_KQUEUE;
626 	fp->f_ops = &kqueueops;
627 	kq = pool_get(&kqueue_pool, PR_WAITOK);
628 	memset((char *)kq, 0, sizeof(struct kqueue));
629 	simple_lock_init(&kq->kq_lock);
630 	TAILQ_INIT(&kq->kq_head);
631 	fp->f_data = (caddr_t)kq;	/* store the kqueue with the fp */
632 	*retval = fd;
633 	if (fdp->fd_knlistsize < 0)
634 		fdp->fd_knlistsize = 0;	/* this process has a kq */
635 	kq->kq_fdp = fdp;
636 	FILE_SET_MATURE(fp);
637 	FILE_UNUSE(fp, p);		/* falloc() does FILE_USE() */
638 	return (error);
639 }
640 
641 /*
642  * kevent(2) system call.
643  */
644 int
645 sys_kevent(struct lwp *l, void *v, register_t *retval)
646 {
647 	struct sys_kevent_args /* {
648 		syscallarg(int) fd;
649 		syscallarg(const struct kevent *) changelist;
650 		syscallarg(size_t) nchanges;
651 		syscallarg(struct kevent *) eventlist;
652 		syscallarg(size_t) nevents;
653 		syscallarg(const struct timespec *) timeout;
654 	} */ *uap = v;
655 	struct kevent	*kevp;
656 	struct kqueue	*kq;
657 	struct file	*fp;
658 	struct timespec	ts;
659 	struct proc	*p;
660 	size_t		i, n;
661 	int		nerrors, error;
662 
663 	p = l->l_proc;
664 	/* check that we're dealing with a kq */
665 	fp = fd_getfile(p->p_fd, SCARG(uap, fd));
666 	if (fp == NULL)
667 		return (EBADF);
668 
669 	if (fp->f_type != DTYPE_KQUEUE) {
670 		simple_unlock(&fp->f_slock);
671 		return (EBADF);
672 	}
673 
674 	FILE_USE(fp);
675 
676 	if (SCARG(uap, timeout) != NULL) {
677 		error = copyin(SCARG(uap, timeout), &ts, sizeof(ts));
678 		if (error)
679 			goto done;
680 		SCARG(uap, timeout) = &ts;
681 	}
682 
683 	kq = (struct kqueue *)fp->f_data;
684 	nerrors = 0;
685 
686 	/* traverse list of events to register */
687 	while (SCARG(uap, nchanges) > 0) {
688 		/* copyin a maximum of KQ_EVENTS at each pass */
689 		n = MIN(SCARG(uap, nchanges), KQ_NEVENTS);
690 		error = copyin(SCARG(uap, changelist), kq->kq_kev,
691 		    n * sizeof(struct kevent));
692 		if (error)
693 			goto done;
694 		for (i = 0; i < n; i++) {
695 			kevp = &kq->kq_kev[i];
696 			kevp->flags &= ~EV_SYSFLAGS;
697 			/* register each knote */
698 			error = kqueue_register(kq, kevp, p);
699 			if (error) {
700 				if (SCARG(uap, nevents) != 0) {
701 					kevp->flags = EV_ERROR;
702 					kevp->data = error;
703 					error = copyout((caddr_t)kevp,
704 					    (caddr_t)SCARG(uap, eventlist),
705 					    sizeof(*kevp));
706 					if (error)
707 						goto done;
708 					SCARG(uap, eventlist)++;
709 					SCARG(uap, nevents)--;
710 					nerrors++;
711 				} else {
712 					goto done;
713 				}
714 			}
715 		}
716 		SCARG(uap, nchanges) -= n;	/* update the results */
717 		SCARG(uap, changelist) += n;
718 	}
719 	if (nerrors) {
720 		*retval = nerrors;
721 		error = 0;
722 		goto done;
723 	}
724 
725 	/* actually scan through the events */
726 	error = kqueue_scan(fp, SCARG(uap, nevents), SCARG(uap, eventlist),
727 	    SCARG(uap, timeout), p, retval);
728  done:
729 	FILE_UNUSE(fp, p);
730 	return (error);
731 }
732 
733 /*
734  * Register a given kevent kev onto the kqueue
735  */
736 int
737 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p)
738 {
739 	const struct kfilter *kfilter;
740 	struct filedesc	*fdp;
741 	struct file	*fp;
742 	struct knote	*kn;
743 	int		s, error;
744 
745 	fdp = kq->kq_fdp;
746 	fp = NULL;
747 	kn = NULL;
748 	error = 0;
749 	kfilter = kfilter_byfilter(kev->filter);
750 	if (kfilter == NULL || kfilter->filtops == NULL) {
751 		/* filter not found nor implemented */
752 		return (EINVAL);
753 	}
754 
755 	/* search if knote already exists */
756 	if (kfilter->filtops->f_isfd) {
757 		/* monitoring a file descriptor */
758 		if ((fp = fd_getfile(fdp, kev->ident)) == NULL)
759 			return (EBADF);	/* validate descriptor */
760 		FILE_USE(fp);
761 
762 		if (kev->ident < fdp->fd_knlistsize) {
763 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
764 				if (kq == kn->kn_kq &&
765 				    kev->filter == kn->kn_filter)
766 					break;
767 		}
768 	} else {
769 		/*
770 		 * not monitoring a file descriptor, so
771 		 * lookup knotes in internal hash table
772 		 */
773 		if (fdp->fd_knhashmask != 0) {
774 			struct klist *list;
775 
776 			list = &fdp->fd_knhash[
777 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
778 			SLIST_FOREACH(kn, list, kn_link)
779 				if (kev->ident == kn->kn_id &&
780 				    kq == kn->kn_kq &&
781 				    kev->filter == kn->kn_filter)
782 					break;
783 		}
784 	}
785 
786 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
787 		error = ENOENT;		/* filter not found */
788 		goto done;
789 	}
790 
791 	/*
792 	 * kn now contains the matching knote, or NULL if no match
793 	 */
794 	if (kev->flags & EV_ADD) {
795 		/* add knote */
796 
797 		if (kn == NULL) {
798 			/* create new knote */
799 			kn = pool_get(&knote_pool, PR_WAITOK);
800 			if (kn == NULL) {
801 				error = ENOMEM;
802 				goto done;
803 			}
804 			kn->kn_fp = fp;
805 			kn->kn_kq = kq;
806 			kn->kn_fop = kfilter->filtops;
807 
808 			/*
809 			 * apply reference count to knote structure, and
810 			 * do not release it at the end of this routine.
811 			 */
812 			fp = NULL;
813 
814 			kn->kn_sfflags = kev->fflags;
815 			kn->kn_sdata = kev->data;
816 			kev->fflags = 0;
817 			kev->data = 0;
818 			kn->kn_kevent = *kev;
819 
820 			knote_attach(kn, fdp);
821 			if ((error = kfilter->filtops->f_attach(kn)) != 0) {
822 				knote_drop(kn, p, fdp);
823 				goto done;
824 			}
825 		} else {
826 			/* modify existing knote */
827 
828 			/*
829 			 * The user may change some filter values after the
830 			 * initial EV_ADD, but doing so will not reset any
831 			 * filter which have already been triggered.
832 			 */
833 			kn->kn_sfflags = kev->fflags;
834 			kn->kn_sdata = kev->data;
835 			kn->kn_kevent.udata = kev->udata;
836 		}
837 
838 		s = splsched();
839 		if (kn->kn_fop->f_event(kn, 0))
840 			KNOTE_ACTIVATE(kn);
841 		splx(s);
842 
843 	} else if (kev->flags & EV_DELETE) {	/* delete knote */
844 		kn->kn_fop->f_detach(kn);
845 		knote_drop(kn, p, fdp);
846 		goto done;
847 	}
848 
849 	/* disable knote */
850 	if ((kev->flags & EV_DISABLE) &&
851 	    ((kn->kn_status & KN_DISABLED) == 0)) {
852 		s = splsched();
853 		kn->kn_status |= KN_DISABLED;
854 		splx(s);
855 	}
856 
857 	/* enable knote */
858 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
859 		s = splsched();
860 		kn->kn_status &= ~KN_DISABLED;
861 		if ((kn->kn_status & KN_ACTIVE) &&
862 		    ((kn->kn_status & KN_QUEUED) == 0))
863 			knote_enqueue(kn);
864 		splx(s);
865 	}
866 
867  done:
868 	if (fp != NULL)
869 		FILE_UNUSE(fp, p);
870 	return (error);
871 }
872 
873 /*
874  * Scan through the list of events on fp (for a maximum of maxevents),
875  * returning the results in to ulistp. Timeout is determined by tsp; if
876  * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait
877  * as appropriate.
878  */
879 static int
880 kqueue_scan(struct file *fp, size_t maxevents, struct kevent *ulistp,
881 	const struct timespec *tsp, struct proc *p, register_t *retval)
882 {
883 	struct kqueue	*kq;
884 	struct kevent	*kevp;
885 	struct timeval	atv;
886 	struct knote	*kn, marker;
887 	size_t		count, nkev;
888 	int		s, timeout, error;
889 
890 	kq = (struct kqueue *)fp->f_data;
891 	count = maxevents;
892 	nkev = error = 0;
893 	if (count == 0)
894 		goto done;
895 
896 	if (tsp) {				/* timeout supplied */
897 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
898 		if (itimerfix(&atv)) {
899 			error = EINVAL;
900 			goto done;
901 		}
902 		s = splclock();
903 		timeradd(&atv, &time, &atv);	/* calc. time to wait until */
904 		splx(s);
905 		timeout = hzto(&atv);
906 		if (timeout <= 0)
907 			timeout = -1;		/* do poll */
908 	} else {
909 		/* no timeout, wait forever */
910 		timeout = 0;
911 	}
912 	goto start;
913 
914  retry:
915 	if (tsp) {
916 		/*
917 		 * We have to recalculate the timeout on every retry.
918 		 */
919 		timeout = hzto(&atv);
920 		if (timeout <= 0)
921 			goto done;
922 	}
923 
924  start:
925 	kevp = kq->kq_kev;
926 	s = splsched();
927 	simple_lock(&kq->kq_lock);
928 	if (kq->kq_count == 0) {
929 		if (timeout < 0) {
930 			error = EWOULDBLOCK;
931 		} else {
932 			kq->kq_state |= KQ_SLEEP;
933 			error = ltsleep(kq, PSOCK | PCATCH | PNORELOCK,
934 					"kqread", timeout, &kq->kq_lock);
935 		}
936 		splx(s);
937 		if (error == 0)
938 			goto retry;
939 		/* don't restart after signals... */
940 		if (error == ERESTART)
941 			error = EINTR;
942 		else if (error == EWOULDBLOCK)
943 			error = 0;
944 		goto done;
945 	}
946 
947 	/* mark end of knote list */
948 	TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
949 	simple_unlock(&kq->kq_lock);
950 
951 	while (count) {				/* while user wants data ... */
952 		simple_lock(&kq->kq_lock);
953 		kn = TAILQ_FIRST(&kq->kq_head);	/* get next knote */
954 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
955 		if (kn == &marker) {		/* if it's our marker, stop */
956 			/* What if it's some else's marker? */
957 			simple_unlock(&kq->kq_lock);
958 			splx(s);
959 			if (count == maxevents)
960 				goto retry;
961 			goto done;
962 		}
963 		kq->kq_count--;
964 		simple_unlock(&kq->kq_lock);
965 
966 		if (kn->kn_status & KN_DISABLED) {
967 			/* don't want disabled events */
968 			kn->kn_status &= ~KN_QUEUED;
969 			continue;
970 		}
971 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
972 		    kn->kn_fop->f_event(kn, 0) == 0) {
973 			/*
974 			 * non-ONESHOT event that hasn't
975 			 * triggered again, so de-queue.
976 			 */
977 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
978 			continue;
979 		}
980 		*kevp = kn->kn_kevent;
981 		kevp++;
982 		nkev++;
983 		if (kn->kn_flags & EV_ONESHOT) {
984 			/* delete ONESHOT events after retrieval */
985 			kn->kn_status &= ~KN_QUEUED;
986 			splx(s);
987 			kn->kn_fop->f_detach(kn);
988 			knote_drop(kn, p, p->p_fd);
989 			s = splsched();
990 		} else if (kn->kn_flags & EV_CLEAR) {
991 			/* clear state after retrieval */
992 			kn->kn_data = 0;
993 			kn->kn_fflags = 0;
994 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
995 		} else {
996 			/* add event back on list */
997 			simple_lock(&kq->kq_lock);
998 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
999 			kq->kq_count++;
1000 			simple_unlock(&kq->kq_lock);
1001 		}
1002 		count--;
1003 		if (nkev == KQ_NEVENTS) {
1004 			/* do copyouts in KQ_NEVENTS chunks */
1005 			splx(s);
1006 			error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
1007 			    sizeof(struct kevent) * nkev);
1008 			ulistp += nkev;
1009 			nkev = 0;
1010 			kevp = kq->kq_kev;
1011 			s = splsched();
1012 			if (error)
1013 				break;
1014 		}
1015 	}
1016 
1017 	/* remove marker */
1018 	simple_lock(&kq->kq_lock);
1019 	TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
1020 	simple_unlock(&kq->kq_lock);
1021 	splx(s);
1022  done:
1023 	if (nkev != 0) {
1024 		/* copyout remaining events */
1025 		error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
1026 		    sizeof(struct kevent) * nkev);
1027 	}
1028 	*retval = maxevents - count;
1029 
1030 	return (error);
1031 }
1032 
1033 /*
1034  * struct fileops read method for a kqueue descriptor.
1035  * Not implemented.
1036  * XXX: This could be expanded to call kqueue_scan, if desired.
1037  */
1038 /*ARGSUSED*/
1039 static int
1040 kqueue_read(struct file *fp, off_t *offset, struct uio *uio,
1041 	struct ucred *cred, int flags)
1042 {
1043 
1044 	return (ENXIO);
1045 }
1046 
1047 /*
1048  * struct fileops write method for a kqueue descriptor.
1049  * Not implemented.
1050  */
1051 /*ARGSUSED*/
1052 static int
1053 kqueue_write(struct file *fp, off_t *offset, struct uio *uio,
1054 	struct ucred *cred, int flags)
1055 {
1056 
1057 	return (ENXIO);
1058 }
1059 
1060 /*
1061  * struct fileops ioctl method for a kqueue descriptor.
1062  *
1063  * Two ioctls are currently supported. They both use struct kfilter_mapping:
1064  *	KFILTER_BYNAME		find name for filter, and return result in
1065  *				name, which is of size len.
1066  *	KFILTER_BYFILTER	find filter for name. len is ignored.
1067  */
1068 /*ARGSUSED*/
1069 static int
1070 kqueue_ioctl(struct file *fp, u_long com, void *data, struct proc *p)
1071 {
1072 	struct kfilter_mapping	*km;
1073 	const struct kfilter	*kfilter;
1074 	char			*name;
1075 	int			error;
1076 
1077 	km = (struct kfilter_mapping *)data;
1078 	error = 0;
1079 
1080 	switch (com) {
1081 	case KFILTER_BYFILTER:	/* convert filter -> name */
1082 		kfilter = kfilter_byfilter(km->filter);
1083 		if (kfilter != NULL)
1084 			error = copyoutstr(kfilter->name, km->name, km->len,
1085 			    NULL);
1086 		else
1087 			error = ENOENT;
1088 		break;
1089 
1090 	case KFILTER_BYNAME:	/* convert name -> filter */
1091 		MALLOC(name, char *, KFILTER_MAXNAME, M_KEVENT, M_WAITOK);
1092 		error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL);
1093 		if (error) {
1094 			FREE(name, M_KEVENT);
1095 			break;
1096 		}
1097 		kfilter = kfilter_byname(name);
1098 		if (kfilter != NULL)
1099 			km->filter = kfilter->filter;
1100 		else
1101 			error = ENOENT;
1102 		FREE(name, M_KEVENT);
1103 		break;
1104 
1105 	default:
1106 		error = ENOTTY;
1107 
1108 	}
1109 	return (error);
1110 }
1111 
1112 /*
1113  * struct fileops fcntl method for a kqueue descriptor.
1114  * Not implemented.
1115  */
1116 /*ARGSUSED*/
1117 static int
1118 kqueue_fcntl(struct file *fp, u_int com, void *data, struct proc *p)
1119 {
1120 
1121 	return (ENOTTY);
1122 }
1123 
1124 /*
1125  * struct fileops poll method for a kqueue descriptor.
1126  * Determine if kqueue has events pending.
1127  */
1128 static int
1129 kqueue_poll(struct file *fp, int events, struct proc *p)
1130 {
1131 	struct kqueue	*kq;
1132 	int		revents;
1133 
1134 	kq = (struct kqueue *)fp->f_data;
1135 	revents = 0;
1136 	if (events & (POLLIN | POLLRDNORM)) {
1137 		if (kq->kq_count) {
1138 			revents |= events & (POLLIN | POLLRDNORM);
1139 		} else {
1140 			selrecord(p, &kq->kq_sel);
1141 		}
1142 	}
1143 	return (revents);
1144 }
1145 
1146 /*
1147  * struct fileops stat method for a kqueue descriptor.
1148  * Returns dummy info, with st_size being number of events pending.
1149  */
1150 static int
1151 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
1152 {
1153 	struct kqueue	*kq;
1154 
1155 	kq = (struct kqueue *)fp->f_data;
1156 	memset((void *)st, 0, sizeof(*st));
1157 	st->st_size = kq->kq_count;
1158 	st->st_blksize = sizeof(struct kevent);
1159 	st->st_mode = S_IFIFO;
1160 	return (0);
1161 }
1162 
1163 /*
1164  * struct fileops close method for a kqueue descriptor.
1165  * Cleans up kqueue.
1166  */
1167 static int
1168 kqueue_close(struct file *fp, struct proc *p)
1169 {
1170 	struct kqueue	*kq;
1171 	struct filedesc	*fdp;
1172 	struct knote	**knp, *kn, *kn0;
1173 	int		i;
1174 
1175 	kq = (struct kqueue *)fp->f_data;
1176 	fdp = p->p_fd;
1177 	for (i = 0; i < fdp->fd_knlistsize; i++) {
1178 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
1179 		kn = *knp;
1180 		while (kn != NULL) {
1181 			kn0 = SLIST_NEXT(kn, kn_link);
1182 			if (kq == kn->kn_kq) {
1183 				kn->kn_fop->f_detach(kn);
1184 				FILE_UNUSE(kn->kn_fp, p);
1185 				pool_put(&knote_pool, kn);
1186 				*knp = kn0;
1187 			} else {
1188 				knp = &SLIST_NEXT(kn, kn_link);
1189 			}
1190 			kn = kn0;
1191 		}
1192 	}
1193 	if (fdp->fd_knhashmask != 0) {
1194 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
1195 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
1196 			kn = *knp;
1197 			while (kn != NULL) {
1198 				kn0 = SLIST_NEXT(kn, kn_link);
1199 				if (kq == kn->kn_kq) {
1200 					kn->kn_fop->f_detach(kn);
1201 					/* XXX non-fd release of kn->kn_ptr */
1202 					pool_put(&knote_pool, kn);
1203 					*knp = kn0;
1204 				} else {
1205 					knp = &SLIST_NEXT(kn, kn_link);
1206 				}
1207 				kn = kn0;
1208 			}
1209 		}
1210 	}
1211 	pool_put(&kqueue_pool, kq);
1212 	fp->f_data = NULL;
1213 
1214 	return (0);
1215 }
1216 
1217 /*
1218  * wakeup a kqueue
1219  */
1220 static void
1221 kqueue_wakeup(struct kqueue *kq)
1222 {
1223 	int s;
1224 
1225 	s = splsched();
1226 	simple_lock(&kq->kq_lock);
1227 	if (kq->kq_state & KQ_SLEEP) {		/* if currently sleeping ...  */
1228 		kq->kq_state &= ~KQ_SLEEP;
1229 		wakeup(kq);			/* ... wakeup */
1230 	}
1231 
1232 	/* Notify select/poll and kevent. */
1233 	selnotify(&kq->kq_sel, 0);
1234 	simple_unlock(&kq->kq_lock);
1235 	splx(s);
1236 }
1237 
1238 /*
1239  * struct fileops kqfilter method for a kqueue descriptor.
1240  * Event triggered when monitored kqueue changes.
1241  */
1242 /*ARGSUSED*/
1243 static int
1244 kqueue_kqfilter(struct file *fp, struct knote *kn)
1245 {
1246 	struct kqueue *kq;
1247 
1248 	KASSERT(fp == kn->kn_fp);
1249 	kq = (struct kqueue *)kn->kn_fp->f_data;
1250 	if (kn->kn_filter != EVFILT_READ)
1251 		return (1);
1252 	kn->kn_fop = &kqread_filtops;
1253 	SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext);
1254 	return (0);
1255 }
1256 
1257 
1258 /*
1259  * Walk down a list of knotes, activating them if their event has triggered.
1260  */
1261 void
1262 knote(struct klist *list, long hint)
1263 {
1264 	struct knote *kn;
1265 
1266 	SLIST_FOREACH(kn, list, kn_selnext)
1267 		if (kn->kn_fop->f_event(kn, hint))
1268 			KNOTE_ACTIVATE(kn);
1269 }
1270 
1271 /*
1272  * Remove all knotes from a specified klist
1273  */
1274 void
1275 knote_remove(struct proc *p, struct klist *list)
1276 {
1277 	struct knote *kn;
1278 
1279 	while ((kn = SLIST_FIRST(list)) != NULL) {
1280 		kn->kn_fop->f_detach(kn);
1281 		knote_drop(kn, p, p->p_fd);
1282 	}
1283 }
1284 
1285 /*
1286  * Remove all knotes referencing a specified fd
1287  */
1288 void
1289 knote_fdclose(struct proc *p, int fd)
1290 {
1291 	struct filedesc	*fdp;
1292 	struct klist	*list;
1293 
1294 	fdp = p->p_fd;
1295 	list = &fdp->fd_knlist[fd];
1296 	knote_remove(p, list);
1297 }
1298 
1299 /*
1300  * Attach a new knote to a file descriptor
1301  */
1302 static void
1303 knote_attach(struct knote *kn, struct filedesc *fdp)
1304 {
1305 	struct klist	*list;
1306 	int		size;
1307 
1308 	if (! kn->kn_fop->f_isfd) {
1309 		/* if knote is not on an fd, store on internal hash table */
1310 		if (fdp->fd_knhashmask == 0)
1311 			fdp->fd_knhash = hashinit(KN_HASHSIZE, HASH_LIST,
1312 			    M_KEVENT, M_WAITOK, &fdp->fd_knhashmask);
1313 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1314 		goto done;
1315 	}
1316 
1317 	/*
1318 	 * otherwise, knote is on an fd.
1319 	 * knotes are stored in fd_knlist indexed by kn->kn_id.
1320 	 */
1321 	if (fdp->fd_knlistsize <= kn->kn_id) {
1322 		/* expand list, it's too small */
1323 		size = fdp->fd_knlistsize;
1324 		while (size <= kn->kn_id) {
1325 			/* grow in KQ_EXTENT chunks */
1326 			size += KQ_EXTENT;
1327 		}
1328 		list = malloc(size * sizeof(struct klist *), M_KEVENT,M_WAITOK);
1329 		if (fdp->fd_knlist) {
1330 			/* copy existing knlist */
1331 			memcpy((caddr_t)list, (caddr_t)fdp->fd_knlist,
1332 			    fdp->fd_knlistsize * sizeof(struct klist *));
1333 		}
1334 		/*
1335 		 * Zero new memory. Stylistically, SLIST_INIT() should be
1336 		 * used here, but that does same thing as the memset() anyway.
1337 		 */
1338 		memset(&list[fdp->fd_knlistsize], 0,
1339 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
1340 
1341 		/* switch to new knlist */
1342 		if (fdp->fd_knlist != NULL)
1343 			free(fdp->fd_knlist, M_KEVENT);
1344 		fdp->fd_knlistsize = size;
1345 		fdp->fd_knlist = list;
1346 	}
1347 
1348 	/* get list head for this fd */
1349 	list = &fdp->fd_knlist[kn->kn_id];
1350  done:
1351 	/* add new knote */
1352 	SLIST_INSERT_HEAD(list, kn, kn_link);
1353 	kn->kn_status = 0;
1354 }
1355 
1356 /*
1357  * Drop knote.
1358  * Should be called at spl == 0, since we don't want to hold spl
1359  * while calling FILE_UNUSE and free.
1360  */
1361 static void
1362 knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp)
1363 {
1364 	struct klist	*list;
1365 
1366 	if (kn->kn_fop->f_isfd)
1367 		list = &fdp->fd_knlist[kn->kn_id];
1368 	else
1369 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1370 
1371 	SLIST_REMOVE(list, kn, knote, kn_link);
1372 	if (kn->kn_status & KN_QUEUED)
1373 		knote_dequeue(kn);
1374 	if (kn->kn_fop->f_isfd)
1375 		FILE_UNUSE(kn->kn_fp, p);
1376 	pool_put(&knote_pool, kn);
1377 }
1378 
1379 
1380 /*
1381  * Queue new event for knote.
1382  */
1383 static void
1384 knote_enqueue(struct knote *kn)
1385 {
1386 	struct kqueue	*kq;
1387 	int		s;
1388 
1389 	kq = kn->kn_kq;
1390 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
1391 
1392 	s = splsched();
1393 	simple_lock(&kq->kq_lock);
1394 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1395 	kn->kn_status |= KN_QUEUED;
1396 	kq->kq_count++;
1397 	simple_unlock(&kq->kq_lock);
1398 	splx(s);
1399 	kqueue_wakeup(kq);
1400 }
1401 
1402 /*
1403  * Dequeue event for knote.
1404  */
1405 static void
1406 knote_dequeue(struct knote *kn)
1407 {
1408 	struct kqueue	*kq;
1409 	int		s;
1410 
1411 	KASSERT(kn->kn_status & KN_QUEUED);
1412 	kq = kn->kn_kq;
1413 
1414 	s = splsched();
1415 	simple_lock(&kq->kq_lock);
1416 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1417 	kn->kn_status &= ~KN_QUEUED;
1418 	kq->kq_count--;
1419 	simple_unlock(&kq->kq_lock);
1420 	splx(s);
1421 }
1422