xref: /netbsd-src/sys/kern/kern_event.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: kern_event.c,v 1.25 2005/12/11 12:24:29 christos 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/cdefs.h>
31 __KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.25 2005/12/11 12:24:29 christos Exp $");
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/proc.h>
37 #include <sys/malloc.h>
38 #include <sys/unistd.h>
39 #include <sys/file.h>
40 #include <sys/fcntl.h>
41 #include <sys/select.h>
42 #include <sys/queue.h>
43 #include <sys/event.h>
44 #include <sys/eventvar.h>
45 #include <sys/poll.h>
46 #include <sys/pool.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/stat.h>
51 #include <sys/uio.h>
52 #include <sys/mount.h>
53 #include <sys/filedesc.h>
54 #include <sys/sa.h>
55 #include <sys/syscallargs.h>
56 
57 static void	kqueue_wakeup(struct kqueue *kq);
58 
59 static int	kqueue_scan(struct file *, size_t, struct kevent *,
60     const struct timespec *, struct lwp *, register_t *,
61     const struct kevent_ops *);
62 static int	kqueue_read(struct file *fp, off_t *offset, struct uio *uio,
63 		    struct ucred *cred, int flags);
64 static int	kqueue_write(struct file *fp, off_t *offset, struct uio *uio,
65 		    struct ucred *cred, int flags);
66 static int	kqueue_ioctl(struct file *fp, u_long com, void *data,
67 		    struct lwp *l);
68 static int	kqueue_fcntl(struct file *fp, u_int com, void *data,
69 		    struct lwp *l);
70 static int	kqueue_poll(struct file *fp, int events, struct lwp *l);
71 static int	kqueue_kqfilter(struct file *fp, struct knote *kn);
72 static int	kqueue_stat(struct file *fp, struct stat *sp, struct lwp *l);
73 static int	kqueue_close(struct file *fp, struct lwp *l);
74 
75 static const struct fileops kqueueops = {
76 	kqueue_read, kqueue_write, kqueue_ioctl, kqueue_fcntl, kqueue_poll,
77 	kqueue_stat, kqueue_close, kqueue_kqfilter
78 };
79 
80 static void	knote_attach(struct knote *kn, struct filedesc *fdp);
81 static void	knote_drop(struct knote *kn, struct lwp *l,
82 		    struct filedesc *fdp);
83 static void	knote_enqueue(struct knote *kn);
84 static void	knote_dequeue(struct knote *kn);
85 
86 static void	filt_kqdetach(struct knote *kn);
87 static int	filt_kqueue(struct knote *kn, long hint);
88 static int	filt_procattach(struct knote *kn);
89 static void	filt_procdetach(struct knote *kn);
90 static int	filt_proc(struct knote *kn, long hint);
91 static int	filt_fileattach(struct knote *kn);
92 static void	filt_timerexpire(void *knx);
93 static int	filt_timerattach(struct knote *kn);
94 static void	filt_timerdetach(struct knote *kn);
95 static int	filt_timer(struct knote *kn, long hint);
96 
97 static const struct filterops kqread_filtops =
98 	{ 1, NULL, filt_kqdetach, filt_kqueue };
99 static const struct filterops proc_filtops =
100 	{ 0, filt_procattach, filt_procdetach, filt_proc };
101 static const struct filterops file_filtops =
102 	{ 1, filt_fileattach, NULL, NULL };
103 static struct filterops timer_filtops =
104 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
105 
106 POOL_INIT(kqueue_pool, sizeof(struct kqueue), 0, 0, 0, "kqueuepl", NULL);
107 POOL_INIT(knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl", NULL);
108 static int	kq_ncallouts = 0;
109 static int	kq_calloutmax = (4 * 1024);
110 
111 MALLOC_DEFINE(M_KEVENT, "kevent", "kevents/knotes");
112 
113 #define	KNOTE_ACTIVATE(kn)						\
114 do {									\
115 	kn->kn_status |= KN_ACTIVE;					\
116 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
117 		knote_enqueue(kn);					\
118 } while(0)
119 
120 #define	KN_HASHSIZE		64		/* XXX should be tunable */
121 #define	KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
122 
123 extern const struct filterops sig_filtops;
124 
125 /*
126  * Table for for all system-defined filters.
127  * These should be listed in the numeric order of the EVFILT_* defines.
128  * If filtops is NULL, the filter isn't implemented in NetBSD.
129  * End of list is when name is NULL.
130  */
131 struct kfilter {
132 	const char	 *name;		/* name of filter */
133 	uint32_t	  filter;	/* id of filter */
134 	const struct filterops *filtops;/* operations for filter */
135 };
136 
137 		/* System defined filters */
138 static const struct kfilter sys_kfilters[] = {
139 	{ "EVFILT_READ",	EVFILT_READ,	&file_filtops },
140 	{ "EVFILT_WRITE",	EVFILT_WRITE,	&file_filtops },
141 	{ "EVFILT_AIO",		EVFILT_AIO,	NULL },
142 	{ "EVFILT_VNODE",	EVFILT_VNODE,	&file_filtops },
143 	{ "EVFILT_PROC",	EVFILT_PROC,	&proc_filtops },
144 	{ "EVFILT_SIGNAL",	EVFILT_SIGNAL,	&sig_filtops },
145 	{ "EVFILT_TIMER",	EVFILT_TIMER,	&timer_filtops },
146 	{ NULL,			0,		NULL },	/* end of list */
147 };
148 
149 		/* User defined kfilters */
150 static struct kfilter	*user_kfilters;		/* array */
151 static int		user_kfilterc;		/* current offset */
152 static int		user_kfiltermaxc;	/* max size so far */
153 
154 /*
155  * Find kfilter entry by name, or NULL if not found.
156  */
157 static const struct kfilter *
158 kfilter_byname_sys(const char *name)
159 {
160 	int i;
161 
162 	for (i = 0; sys_kfilters[i].name != NULL; i++) {
163 		if (strcmp(name, sys_kfilters[i].name) == 0)
164 			return (&sys_kfilters[i]);
165 	}
166 	return (NULL);
167 }
168 
169 static struct kfilter *
170 kfilter_byname_user(const char *name)
171 {
172 	int i;
173 
174 	/* user_kfilters[] could be NULL if no filters were registered */
175 	if (!user_kfilters)
176 		return (NULL);
177 
178 	for (i = 0; user_kfilters[i].name != NULL; i++) {
179 		if (user_kfilters[i].name != '\0' &&
180 		    strcmp(name, user_kfilters[i].name) == 0)
181 			return (&user_kfilters[i]);
182 	}
183 	return (NULL);
184 }
185 
186 static const struct kfilter *
187 kfilter_byname(const char *name)
188 {
189 	const struct kfilter *kfilter;
190 
191 	if ((kfilter = kfilter_byname_sys(name)) != NULL)
192 		return (kfilter);
193 
194 	return (kfilter_byname_user(name));
195 }
196 
197 /*
198  * Find kfilter entry by filter id, or NULL if not found.
199  * Assumes entries are indexed in filter id order, for speed.
200  */
201 static const struct kfilter *
202 kfilter_byfilter(uint32_t filter)
203 {
204 	const struct kfilter *kfilter;
205 
206 	if (filter < EVFILT_SYSCOUNT)	/* it's a system filter */
207 		kfilter = &sys_kfilters[filter];
208 	else if (user_kfilters != NULL &&
209 	    filter < EVFILT_SYSCOUNT + user_kfilterc)
210 					/* it's a user filter */
211 		kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT];
212 	else
213 		return (NULL);		/* out of range */
214 	KASSERT(kfilter->filter == filter);	/* sanity check! */
215 	return (kfilter);
216 }
217 
218 /*
219  * Register a new kfilter. Stores the entry in user_kfilters.
220  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
221  * If retfilter != NULL, the new filterid is returned in it.
222  */
223 int
224 kfilter_register(const char *name, const struct filterops *filtops,
225     int *retfilter)
226 {
227 	struct kfilter *kfilter;
228 	void *space;
229 	int len;
230 
231 	if (name == NULL || name[0] == '\0' || filtops == NULL)
232 		return (EINVAL);	/* invalid args */
233 	if (kfilter_byname(name) != NULL)
234 		return (EEXIST);	/* already exists */
235 	if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT)
236 		return (EINVAL);	/* too many */
237 
238 	/* check if need to grow user_kfilters */
239 	if (user_kfilterc + 1 > user_kfiltermaxc) {
240 		/*
241 		 * Grow in KFILTER_EXTENT chunks. Use malloc(9), because we
242 		 * want to traverse user_kfilters as an array.
243 		 */
244 		user_kfiltermaxc += KFILTER_EXTENT;
245 		kfilter = malloc(user_kfiltermaxc * sizeof(struct filter *),
246 		    M_KEVENT, M_WAITOK);
247 
248 		/* copy existing user_kfilters */
249 		if (user_kfilters != NULL)
250 			memcpy((caddr_t)kfilter, (caddr_t)user_kfilters,
251 			    user_kfilterc * sizeof(struct kfilter *));
252 					/* zero new sections */
253 		memset((caddr_t)kfilter +
254 		    user_kfilterc * sizeof(struct kfilter *), 0,
255 		    (user_kfiltermaxc - user_kfilterc) *
256 		    sizeof(struct kfilter *));
257 					/* switch to new kfilter */
258 		if (user_kfilters != NULL)
259 			free(user_kfilters, M_KEVENT);
260 		user_kfilters = kfilter;
261 	}
262 	len = strlen(name) + 1;		/* copy name */
263 	space = malloc(len, M_KEVENT, M_WAITOK);
264 	memcpy(space, name, len);
265 	user_kfilters[user_kfilterc].name = space;
266 
267 	user_kfilters[user_kfilterc].filter = user_kfilterc + EVFILT_SYSCOUNT;
268 
269 	len = sizeof(struct filterops);	/* copy filtops */
270 	space = malloc(len, M_KEVENT, M_WAITOK);
271 	memcpy(space, filtops, len);
272 	user_kfilters[user_kfilterc].filtops = space;
273 
274 	if (retfilter != NULL)
275 		*retfilter = user_kfilters[user_kfilterc].filter;
276 	user_kfilterc++;		/* finally, increment count */
277 	return (0);
278 }
279 
280 /*
281  * Unregister a kfilter previously registered with kfilter_register.
282  * This retains the filter id, but clears the name and frees filtops (filter
283  * operations), so that the number isn't reused during a boot.
284  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
285  */
286 int
287 kfilter_unregister(const char *name)
288 {
289 	struct kfilter *kfilter;
290 
291 	if (name == NULL || name[0] == '\0')
292 		return (EINVAL);	/* invalid name */
293 
294 	if (kfilter_byname_sys(name) != NULL)
295 		return (EINVAL);	/* can't detach system filters */
296 
297 	kfilter = kfilter_byname_user(name);
298 	if (kfilter == NULL)		/* not found */
299 		return (ENOENT);
300 
301 	if (kfilter->name[0] != '\0') {
302 		/* XXXUNCONST Cast away const (but we know it's safe. */
303 		free(__UNCONST(kfilter->name), M_KEVENT);
304 		kfilter->name = "";	/* mark as `not implemented' */
305 	}
306 	if (kfilter->filtops != NULL) {
307 		/* XXXUNCONST Cast away const (but we know it's safe. */
308 		free(__UNCONST(kfilter->filtops), M_KEVENT);
309 		kfilter->filtops = NULL; /* mark as `not implemented' */
310 	}
311 	return (0);
312 }
313 
314 
315 /*
316  * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file
317  * descriptors. Calls struct fileops kqfilter method for given file descriptor.
318  */
319 static int
320 filt_fileattach(struct knote *kn)
321 {
322 	struct file *fp;
323 
324 	fp = kn->kn_fp;
325 	return ((*fp->f_ops->fo_kqfilter)(fp, kn));
326 }
327 
328 /*
329  * Filter detach method for EVFILT_READ on kqueue descriptor.
330  */
331 static void
332 filt_kqdetach(struct knote *kn)
333 {
334 	struct kqueue *kq;
335 
336 	kq = (struct kqueue *)kn->kn_fp->f_data;
337 	SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext);
338 }
339 
340 /*
341  * Filter event method for EVFILT_READ on kqueue descriptor.
342  */
343 /*ARGSUSED*/
344 static int
345 filt_kqueue(struct knote *kn, long hint)
346 {
347 	struct kqueue *kq;
348 
349 	kq = (struct kqueue *)kn->kn_fp->f_data;
350 	kn->kn_data = kq->kq_count;
351 	return (kn->kn_data > 0);
352 }
353 
354 /*
355  * Filter attach method for EVFILT_PROC.
356  */
357 static int
358 filt_procattach(struct knote *kn)
359 {
360 	struct proc *p;
361 
362 	p = pfind(kn->kn_id);
363 	if (p == NULL)
364 		return (ESRCH);
365 
366 	/*
367 	 * Fail if it's not owned by you, or the last exec gave us
368 	 * setuid/setgid privs (unless you're root).
369 	 */
370 	if ((p->p_cred->p_ruid != curproc->p_cred->p_ruid ||
371 		(p->p_flag & P_SUGID))
372 	    && suser(curproc->p_ucred, &curproc->p_acflag) != 0)
373 		return (EACCES);
374 
375 	kn->kn_ptr.p_proc = p;
376 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
377 
378 	/*
379 	 * internal flag indicating registration done by kernel
380 	 */
381 	if (kn->kn_flags & EV_FLAG1) {
382 		kn->kn_data = kn->kn_sdata;	/* ppid */
383 		kn->kn_fflags = NOTE_CHILD;
384 		kn->kn_flags &= ~EV_FLAG1;
385 	}
386 
387 	/* XXXSMP lock the process? */
388 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
389 
390 	return (0);
391 }
392 
393 /*
394  * Filter detach method for EVFILT_PROC.
395  *
396  * The knote may be attached to a different process, which may exit,
397  * leaving nothing for the knote to be attached to.  So when the process
398  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
399  * it will be deleted when read out.  However, as part of the knote deletion,
400  * this routine is called, so a check is needed to avoid actually performing
401  * a detach, because the original process might not exist any more.
402  */
403 static void
404 filt_procdetach(struct knote *kn)
405 {
406 	struct proc *p;
407 
408 	if (kn->kn_status & KN_DETACHED)
409 		return;
410 
411 	p = kn->kn_ptr.p_proc;
412 	KASSERT(p->p_stat == SZOMB || pfind(kn->kn_id) == p);
413 
414 	/* XXXSMP lock the process? */
415 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
416 }
417 
418 /*
419  * Filter event method for EVFILT_PROC.
420  */
421 static int
422 filt_proc(struct knote *kn, long hint)
423 {
424 	u_int event;
425 
426 	/*
427 	 * mask off extra data
428 	 */
429 	event = (u_int)hint & NOTE_PCTRLMASK;
430 
431 	/*
432 	 * if the user is interested in this event, record it.
433 	 */
434 	if (kn->kn_sfflags & event)
435 		kn->kn_fflags |= event;
436 
437 	/*
438 	 * process is gone, so flag the event as finished.
439 	 */
440 	if (event == NOTE_EXIT) {
441 		/*
442 		 * Detach the knote from watched process and mark
443 		 * it as such. We can't leave this to kqueue_scan(),
444 		 * since the process might not exist by then. And we
445 		 * have to do this now, since psignal KNOTE() is called
446 		 * also for zombies and we might end up reading freed
447 		 * memory if the kevent would already be picked up
448 		 * and knote g/c'ed.
449 		 */
450 		kn->kn_fop->f_detach(kn);
451 		kn->kn_status |= KN_DETACHED;
452 
453 		/* Mark as ONESHOT, so that the knote it g/c'ed when read */
454 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
455 		return (1);
456 	}
457 
458 	/*
459 	 * process forked, and user wants to track the new process,
460 	 * so attach a new knote to it, and immediately report an
461 	 * event with the parent's pid.
462 	 */
463 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
464 		struct kevent kev;
465 		int error;
466 
467 		/*
468 		 * register knote with new process.
469 		 */
470 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
471 		kev.filter = kn->kn_filter;
472 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
473 		kev.fflags = kn->kn_sfflags;
474 		kev.data = kn->kn_id;			/* parent */
475 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
476 		error = kqueue_register(kn->kn_kq, &kev, NULL);
477 		if (error)
478 			kn->kn_fflags |= NOTE_TRACKERR;
479 	}
480 
481 	return (kn->kn_fflags != 0);
482 }
483 
484 static void
485 filt_timerexpire(void *knx)
486 {
487 	struct knote *kn = knx;
488 	int tticks;
489 
490 	kn->kn_data++;
491 	KNOTE_ACTIVATE(kn);
492 
493 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
494 		tticks = mstohz(kn->kn_sdata);
495 		callout_schedule((struct callout *)kn->kn_hook, tticks);
496 	}
497 }
498 
499 /*
500  * data contains amount of time to sleep, in milliseconds
501  */
502 static int
503 filt_timerattach(struct knote *kn)
504 {
505 	struct callout *calloutp;
506 	int tticks;
507 
508 	if (kq_ncallouts >= kq_calloutmax)
509 		return (ENOMEM);
510 	kq_ncallouts++;
511 
512 	tticks = mstohz(kn->kn_sdata);
513 
514 	/* if the supplied value is under our resolution, use 1 tick */
515 	if (tticks == 0) {
516 		if (kn->kn_sdata == 0)
517 			return (EINVAL);
518 		tticks = 1;
519 	}
520 
521 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
522 	MALLOC(calloutp, struct callout *, sizeof(*calloutp),
523 	    M_KEVENT, 0);
524 	callout_init(calloutp);
525 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
526 	kn->kn_hook = calloutp;
527 
528 	return (0);
529 }
530 
531 static void
532 filt_timerdetach(struct knote *kn)
533 {
534 	struct callout *calloutp;
535 
536 	calloutp = (struct callout *)kn->kn_hook;
537 	callout_stop(calloutp);
538 	FREE(calloutp, M_KEVENT);
539 	kq_ncallouts--;
540 }
541 
542 static int
543 filt_timer(struct knote *kn, long hint)
544 {
545 	return (kn->kn_data != 0);
546 }
547 
548 /*
549  * filt_seltrue:
550  *
551  *	This filter "event" routine simulates seltrue().
552  */
553 int
554 filt_seltrue(struct knote *kn, long hint)
555 {
556 
557 	/*
558 	 * We don't know how much data can be read/written,
559 	 * but we know that it *can* be.  This is about as
560 	 * good as select/poll does as well.
561 	 */
562 	kn->kn_data = 0;
563 	return (1);
564 }
565 
566 /*
567  * This provides full kqfilter entry for device switch tables, which
568  * has same effect as filter using filt_seltrue() as filter method.
569  */
570 static void
571 filt_seltruedetach(struct knote *kn)
572 {
573 	/* Nothing to do */
574 }
575 
576 static const struct filterops seltrue_filtops =
577 	{ 1, NULL, filt_seltruedetach, filt_seltrue };
578 
579 int
580 seltrue_kqfilter(dev_t dev, struct knote *kn)
581 {
582 	switch (kn->kn_filter) {
583 	case EVFILT_READ:
584 	case EVFILT_WRITE:
585 		kn->kn_fop = &seltrue_filtops;
586 		break;
587 	default:
588 		return (1);
589 	}
590 
591 	/* Nothing more to do */
592 	return (0);
593 }
594 
595 /*
596  * kqueue(2) system call.
597  */
598 int
599 sys_kqueue(struct lwp *l, void *v, register_t *retval)
600 {
601 	struct filedesc	*fdp;
602 	struct kqueue	*kq;
603 	struct file	*fp;
604 	struct proc	*p;
605 	int		fd, error;
606 
607 	p = l->l_proc;
608 	fdp = p->p_fd;
609 	error = falloc(p, &fp, &fd);	/* setup a new file descriptor */
610 	if (error)
611 		return (error);
612 	fp->f_flag = FREAD | FWRITE;
613 	fp->f_type = DTYPE_KQUEUE;
614 	fp->f_ops = &kqueueops;
615 	kq = pool_get(&kqueue_pool, PR_WAITOK);
616 	memset((char *)kq, 0, sizeof(struct kqueue));
617 	simple_lock_init(&kq->kq_lock);
618 	TAILQ_INIT(&kq->kq_head);
619 	fp->f_data = (caddr_t)kq;	/* store the kqueue with the fp */
620 	*retval = fd;
621 	if (fdp->fd_knlistsize < 0)
622 		fdp->fd_knlistsize = 0;	/* this process has a kq */
623 	kq->kq_fdp = fdp;
624 	FILE_SET_MATURE(fp);
625 	FILE_UNUSE(fp, l);		/* falloc() does FILE_USE() */
626 	return (error);
627 }
628 
629 /*
630  * kevent(2) system call.
631  */
632 static int
633 kevent_fetch_changes(void *private, const struct kevent *changelist,
634     struct kevent *changes, size_t index, int n)
635 {
636 	return copyin(changelist + index, changes, n * sizeof(*changes));
637 }
638 
639 static int
640 kevent_put_events(void *private, struct kevent *events,
641     struct kevent *eventlist, size_t index, int n)
642 {
643 	return copyout(events, eventlist + index, n * sizeof(*events));
644 }
645 
646 static const struct kevent_ops kevent_native_ops = {
647 	keo_private: NULL,
648 	keo_fetch_timeout: copyin,
649 	keo_fetch_changes: kevent_fetch_changes,
650 	keo_put_events: kevent_put_events,
651 };
652 
653 int
654 sys_kevent(struct lwp *l, void *v, register_t *retval)
655 {
656 	struct sys_kevent_args /* {
657 		syscallarg(int) fd;
658 		syscallarg(const struct kevent *) changelist;
659 		syscallarg(size_t) nchanges;
660 		syscallarg(struct kevent *) eventlist;
661 		syscallarg(size_t) nevents;
662 		syscallarg(const struct timespec *) timeout;
663 	} */ *uap = v;
664 
665 	return kevent1(l, retval, SCARG(uap, fd), SCARG(uap, changelist),
666 	    SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents),
667 	    SCARG(uap, timeout), &kevent_native_ops);
668 }
669 
670 int
671 kevent1(struct lwp *l, register_t *retval, int fd,
672     const struct kevent *changelist, size_t nchanges, struct kevent *eventlist,
673     size_t nevents, const struct timespec *timeout,
674     const struct kevent_ops *keops)
675 {
676 	struct kevent	*kevp;
677 	struct kqueue	*kq;
678 	struct file	*fp;
679 	struct timespec	ts;
680 	struct proc	*p;
681 	size_t		i, n, ichange;
682 	int		nerrors, error;
683 
684 	p = l->l_proc;
685 	/* check that we're dealing with a kq */
686 	fp = fd_getfile(p->p_fd, fd);
687 	if (fp == NULL)
688 		return (EBADF);
689 
690 	if (fp->f_type != DTYPE_KQUEUE) {
691 		simple_unlock(&fp->f_slock);
692 		return (EBADF);
693 	}
694 
695 	FILE_USE(fp);
696 
697 	if (timeout != NULL) {
698 		error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts));
699 		if (error)
700 			goto done;
701 		timeout = &ts;
702 	}
703 
704 	kq = (struct kqueue *)fp->f_data;
705 	nerrors = 0;
706 	ichange = 0;
707 
708 	/* traverse list of events to register */
709 	while (nchanges > 0) {
710 		/* copyin a maximum of KQ_EVENTS at each pass */
711 		n = MIN(nchanges, KQ_NEVENTS);
712 		error = (*keops->keo_fetch_changes)(keops->keo_private,
713 		    changelist, kq->kq_kev, ichange, n);
714 		if (error)
715 			goto done;
716 		for (i = 0; i < n; i++) {
717 			kevp = &kq->kq_kev[i];
718 			kevp->flags &= ~EV_SYSFLAGS;
719 			/* register each knote */
720 			error = kqueue_register(kq, kevp, l);
721 			if (error) {
722 				if (nevents != 0) {
723 					kevp->flags = EV_ERROR;
724 					kevp->data = error;
725 					error = (*keops->keo_put_events)
726 					    (keops->keo_private, kevp,
727 					    eventlist, nerrors, 1);
728 					if (error)
729 						goto done;
730 					nevents--;
731 					nerrors++;
732 				} else {
733 					goto done;
734 				}
735 			}
736 		}
737 		nchanges -= n;	/* update the results */
738 		ichange += n;
739 	}
740 	if (nerrors) {
741 		*retval = nerrors;
742 		error = 0;
743 		goto done;
744 	}
745 
746 	/* actually scan through the events */
747 	error = kqueue_scan(fp, nevents, eventlist, timeout, l, retval, keops);
748  done:
749 	FILE_UNUSE(fp, l);
750 	return (error);
751 }
752 
753 /*
754  * Register a given kevent kev onto the kqueue
755  */
756 int
757 kqueue_register(struct kqueue *kq, struct kevent *kev, struct lwp *l)
758 {
759 	const struct kfilter *kfilter;
760 	struct filedesc	*fdp;
761 	struct file	*fp;
762 	struct knote	*kn;
763 	int		s, error;
764 
765 	fdp = kq->kq_fdp;
766 	fp = NULL;
767 	kn = NULL;
768 	error = 0;
769 	kfilter = kfilter_byfilter(kev->filter);
770 	if (kfilter == NULL || kfilter->filtops == NULL) {
771 		/* filter not found nor implemented */
772 		return (EINVAL);
773 	}
774 
775 	/* search if knote already exists */
776 	if (kfilter->filtops->f_isfd) {
777 		/* monitoring a file descriptor */
778 		if ((fp = fd_getfile(fdp, kev->ident)) == NULL)
779 			return (EBADF);	/* validate descriptor */
780 		FILE_USE(fp);
781 
782 		if (kev->ident < fdp->fd_knlistsize) {
783 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
784 				if (kq == kn->kn_kq &&
785 				    kev->filter == kn->kn_filter)
786 					break;
787 		}
788 	} else {
789 		/*
790 		 * not monitoring a file descriptor, so
791 		 * lookup knotes in internal hash table
792 		 */
793 		if (fdp->fd_knhashmask != 0) {
794 			struct klist *list;
795 
796 			list = &fdp->fd_knhash[
797 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
798 			SLIST_FOREACH(kn, list, kn_link)
799 				if (kev->ident == kn->kn_id &&
800 				    kq == kn->kn_kq &&
801 				    kev->filter == kn->kn_filter)
802 					break;
803 		}
804 	}
805 
806 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
807 		error = ENOENT;		/* filter not found */
808 		goto done;
809 	}
810 
811 	/*
812 	 * kn now contains the matching knote, or NULL if no match
813 	 */
814 	if (kev->flags & EV_ADD) {
815 		/* add knote */
816 
817 		if (kn == NULL) {
818 			/* create new knote */
819 			kn = pool_get(&knote_pool, PR_WAITOK);
820 			if (kn == NULL) {
821 				error = ENOMEM;
822 				goto done;
823 			}
824 			kn->kn_fp = fp;
825 			kn->kn_kq = kq;
826 			kn->kn_fop = kfilter->filtops;
827 
828 			/*
829 			 * apply reference count to knote structure, and
830 			 * do not release it at the end of this routine.
831 			 */
832 			fp = NULL;
833 
834 			kn->kn_sfflags = kev->fflags;
835 			kn->kn_sdata = kev->data;
836 			kev->fflags = 0;
837 			kev->data = 0;
838 			kn->kn_kevent = *kev;
839 
840 			knote_attach(kn, fdp);
841 			if ((error = kfilter->filtops->f_attach(kn)) != 0) {
842 				knote_drop(kn, l, fdp);
843 				goto done;
844 			}
845 		} else {
846 			/* modify existing knote */
847 
848 			/*
849 			 * The user may change some filter values after the
850 			 * initial EV_ADD, but doing so will not reset any
851 			 * filter which have already been triggered.
852 			 */
853 			kn->kn_sfflags = kev->fflags;
854 			kn->kn_sdata = kev->data;
855 			kn->kn_kevent.udata = kev->udata;
856 		}
857 
858 		s = splsched();
859 		if (kn->kn_fop->f_event(kn, 0))
860 			KNOTE_ACTIVATE(kn);
861 		splx(s);
862 
863 	} else if (kev->flags & EV_DELETE) {	/* delete knote */
864 		kn->kn_fop->f_detach(kn);
865 		knote_drop(kn, l, fdp);
866 		goto done;
867 	}
868 
869 	/* disable knote */
870 	if ((kev->flags & EV_DISABLE) &&
871 	    ((kn->kn_status & KN_DISABLED) == 0)) {
872 		s = splsched();
873 		kn->kn_status |= KN_DISABLED;
874 		splx(s);
875 	}
876 
877 	/* enable knote */
878 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
879 		s = splsched();
880 		kn->kn_status &= ~KN_DISABLED;
881 		if ((kn->kn_status & KN_ACTIVE) &&
882 		    ((kn->kn_status & KN_QUEUED) == 0))
883 			knote_enqueue(kn);
884 		splx(s);
885 	}
886 
887  done:
888 	if (fp != NULL)
889 		FILE_UNUSE(fp, l);
890 	return (error);
891 }
892 
893 /*
894  * Scan through the list of events on fp (for a maximum of maxevents),
895  * returning the results in to ulistp. Timeout is determined by tsp; if
896  * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait
897  * as appropriate.
898  */
899 static int
900 kqueue_scan(struct file *fp, size_t maxevents, struct kevent *ulistp,
901     const struct timespec *tsp, struct lwp *l, register_t *retval,
902     const struct kevent_ops *keops)
903 {
904 	struct proc	*p = l->l_proc;
905 	struct kqueue	*kq;
906 	struct kevent	*kevp;
907 	struct timeval	atv;
908 	struct knote	*kn, *marker=NULL;
909 	size_t		count, nkev, nevents;
910 	int		s, timeout, error;
911 
912 	kq = (struct kqueue *)fp->f_data;
913 	count = maxevents;
914 	nkev = nevents = error = 0;
915 	if (count == 0)
916 		goto done;
917 
918 	if (tsp) {				/* timeout supplied */
919 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
920 		if (itimerfix(&atv)) {
921 			error = EINVAL;
922 			goto done;
923 		}
924 		s = splclock();
925 		timeradd(&atv, &time, &atv);	/* calc. time to wait until */
926 		splx(s);
927 		timeout = hzto(&atv);
928 		if (timeout <= 0)
929 			timeout = -1;		/* do poll */
930 	} else {
931 		/* no timeout, wait forever */
932 		timeout = 0;
933 	}
934 
935 	MALLOC(marker, struct knote *, sizeof(*marker), M_KEVENT, M_WAITOK);
936 	memset(marker, 0, sizeof(*marker));
937 
938 	goto start;
939 
940  retry:
941 	if (tsp) {
942 		/*
943 		 * We have to recalculate the timeout on every retry.
944 		 */
945 		timeout = hzto(&atv);
946 		if (timeout <= 0)
947 			goto done;
948 	}
949 
950  start:
951 	kevp = kq->kq_kev;
952 	s = splsched();
953 	simple_lock(&kq->kq_lock);
954 	if (kq->kq_count == 0) {
955 		if (timeout < 0) {
956 			error = EWOULDBLOCK;
957 			simple_unlock(&kq->kq_lock);
958 		} else {
959 			kq->kq_state |= KQ_SLEEP;
960 			error = ltsleep(kq, PSOCK | PCATCH | PNORELOCK,
961 					"kqread", timeout, &kq->kq_lock);
962 		}
963 		splx(s);
964 		if (error == 0)
965 			goto retry;
966 		/* don't restart after signals... */
967 		if (error == ERESTART)
968 			error = EINTR;
969 		else if (error == EWOULDBLOCK)
970 			error = 0;
971 		goto done;
972 	}
973 
974 	/* mark end of knote list */
975 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
976 	simple_unlock(&kq->kq_lock);
977 
978 	while (count) {				/* while user wants data ... */
979 		simple_lock(&kq->kq_lock);
980 		kn = TAILQ_FIRST(&kq->kq_head);	/* get next knote */
981 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
982 		if (kn == marker) {		/* if it's our marker, stop */
983 			/* What if it's some else's marker? */
984 			simple_unlock(&kq->kq_lock);
985 			splx(s);
986 			if (count == maxevents)
987 				goto retry;
988 			goto done;
989 		}
990 		kq->kq_count--;
991 		simple_unlock(&kq->kq_lock);
992 
993 		if (kn->kn_status & KN_DISABLED) {
994 			/* don't want disabled events */
995 			kn->kn_status &= ~KN_QUEUED;
996 			continue;
997 		}
998 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
999 		    kn->kn_fop->f_event(kn, 0) == 0) {
1000 			/*
1001 			 * non-ONESHOT event that hasn't
1002 			 * triggered again, so de-queue.
1003 			 */
1004 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1005 			continue;
1006 		}
1007 		*kevp = kn->kn_kevent;
1008 		kevp++;
1009 		nkev++;
1010 		if (kn->kn_flags & EV_ONESHOT) {
1011 			/* delete ONESHOT events after retrieval */
1012 			kn->kn_status &= ~KN_QUEUED;
1013 			splx(s);
1014 			kn->kn_fop->f_detach(kn);
1015 			knote_drop(kn, l, p->p_fd);
1016 			s = splsched();
1017 		} else if (kn->kn_flags & EV_CLEAR) {
1018 			/* clear state after retrieval */
1019 			kn->kn_data = 0;
1020 			kn->kn_fflags = 0;
1021 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1022 		} else {
1023 			/* add event back on list */
1024 			simple_lock(&kq->kq_lock);
1025 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1026 			kq->kq_count++;
1027 			simple_unlock(&kq->kq_lock);
1028 		}
1029 		count--;
1030 		if (nkev == KQ_NEVENTS) {
1031 			/* do copyouts in KQ_NEVENTS chunks */
1032 			splx(s);
1033 			error = (*keops->keo_put_events)(keops->keo_private,
1034 			    &kq->kq_kev[0], ulistp, nevents, nkev);
1035 			nevents += nkev;
1036 			nkev = 0;
1037 			kevp = kq->kq_kev;
1038 			s = splsched();
1039 			if (error)
1040 				break;
1041 		}
1042 	}
1043 
1044 	/* remove marker */
1045 	simple_lock(&kq->kq_lock);
1046 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1047 	simple_unlock(&kq->kq_lock);
1048 	splx(s);
1049  done:
1050 	if (marker)
1051 		FREE(marker, M_KEVENT);
1052 
1053 	if (nkev != 0)
1054 		/* copyout remaining events */
1055 		error = (*keops->keo_put_events)(keops->keo_private,
1056 		    &kq->kq_kev[0], ulistp, nevents, nkev);
1057 	*retval = maxevents - count;
1058 
1059 	return (error);
1060 }
1061 
1062 /*
1063  * struct fileops read method for a kqueue descriptor.
1064  * Not implemented.
1065  * XXX: This could be expanded to call kqueue_scan, if desired.
1066  */
1067 /*ARGSUSED*/
1068 static int
1069 kqueue_read(struct file *fp, off_t *offset, struct uio *uio,
1070 	struct ucred *cred, int flags)
1071 {
1072 
1073 	return (ENXIO);
1074 }
1075 
1076 /*
1077  * struct fileops write method for a kqueue descriptor.
1078  * Not implemented.
1079  */
1080 /*ARGSUSED*/
1081 static int
1082 kqueue_write(struct file *fp, off_t *offset, struct uio *uio,
1083 	struct ucred *cred, int flags)
1084 {
1085 
1086 	return (ENXIO);
1087 }
1088 
1089 /*
1090  * struct fileops ioctl method for a kqueue descriptor.
1091  *
1092  * Two ioctls are currently supported. They both use struct kfilter_mapping:
1093  *	KFILTER_BYNAME		find name for filter, and return result in
1094  *				name, which is of size len.
1095  *	KFILTER_BYFILTER	find filter for name. len is ignored.
1096  */
1097 /*ARGSUSED*/
1098 static int
1099 kqueue_ioctl(struct file *fp, u_long com, void *data, struct lwp *l)
1100 {
1101 	struct kfilter_mapping	*km;
1102 	const struct kfilter	*kfilter;
1103 	char			*name;
1104 	int			error;
1105 
1106 	km = (struct kfilter_mapping *)data;
1107 	error = 0;
1108 
1109 	switch (com) {
1110 	case KFILTER_BYFILTER:	/* convert filter -> name */
1111 		kfilter = kfilter_byfilter(km->filter);
1112 		if (kfilter != NULL)
1113 			error = copyoutstr(kfilter->name, km->name, km->len,
1114 			    NULL);
1115 		else
1116 			error = ENOENT;
1117 		break;
1118 
1119 	case KFILTER_BYNAME:	/* convert name -> filter */
1120 		MALLOC(name, char *, KFILTER_MAXNAME, M_KEVENT, M_WAITOK);
1121 		error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL);
1122 		if (error) {
1123 			FREE(name, M_KEVENT);
1124 			break;
1125 		}
1126 		kfilter = kfilter_byname(name);
1127 		if (kfilter != NULL)
1128 			km->filter = kfilter->filter;
1129 		else
1130 			error = ENOENT;
1131 		FREE(name, M_KEVENT);
1132 		break;
1133 
1134 	default:
1135 		error = ENOTTY;
1136 
1137 	}
1138 	return (error);
1139 }
1140 
1141 /*
1142  * struct fileops fcntl method for a kqueue descriptor.
1143  * Not implemented.
1144  */
1145 /*ARGSUSED*/
1146 static int
1147 kqueue_fcntl(struct file *fp, u_int com, void *data, struct lwp *l)
1148 {
1149 
1150 	return (ENOTTY);
1151 }
1152 
1153 /*
1154  * struct fileops poll method for a kqueue descriptor.
1155  * Determine if kqueue has events pending.
1156  */
1157 static int
1158 kqueue_poll(struct file *fp, int events, struct lwp *l)
1159 {
1160 	struct kqueue	*kq;
1161 	int		revents;
1162 
1163 	kq = (struct kqueue *)fp->f_data;
1164 	revents = 0;
1165 	if (events & (POLLIN | POLLRDNORM)) {
1166 		if (kq->kq_count) {
1167 			revents |= events & (POLLIN | POLLRDNORM);
1168 		} else {
1169 			selrecord(l, &kq->kq_sel);
1170 		}
1171 	}
1172 	return (revents);
1173 }
1174 
1175 /*
1176  * struct fileops stat method for a kqueue descriptor.
1177  * Returns dummy info, with st_size being number of events pending.
1178  */
1179 static int
1180 kqueue_stat(struct file *fp, struct stat *st, struct lwp *l)
1181 {
1182 	struct kqueue	*kq;
1183 
1184 	kq = (struct kqueue *)fp->f_data;
1185 	memset((void *)st, 0, sizeof(*st));
1186 	st->st_size = kq->kq_count;
1187 	st->st_blksize = sizeof(struct kevent);
1188 	st->st_mode = S_IFIFO;
1189 	return (0);
1190 }
1191 
1192 /*
1193  * struct fileops close method for a kqueue descriptor.
1194  * Cleans up kqueue.
1195  */
1196 static int
1197 kqueue_close(struct file *fp, struct lwp *l)
1198 {
1199 	struct proc	*p = l->l_proc;
1200 	struct kqueue	*kq;
1201 	struct filedesc	*fdp;
1202 	struct knote	**knp, *kn, *kn0;
1203 	int		i;
1204 
1205 	kq = (struct kqueue *)fp->f_data;
1206 	fdp = p->p_fd;
1207 	for (i = 0; i < fdp->fd_knlistsize; i++) {
1208 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
1209 		kn = *knp;
1210 		while (kn != NULL) {
1211 			kn0 = SLIST_NEXT(kn, kn_link);
1212 			if (kq == kn->kn_kq) {
1213 				kn->kn_fop->f_detach(kn);
1214 				FILE_UNUSE(kn->kn_fp, l);
1215 				pool_put(&knote_pool, kn);
1216 				*knp = kn0;
1217 			} else {
1218 				knp = &SLIST_NEXT(kn, kn_link);
1219 			}
1220 			kn = kn0;
1221 		}
1222 	}
1223 	if (fdp->fd_knhashmask != 0) {
1224 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
1225 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
1226 			kn = *knp;
1227 			while (kn != NULL) {
1228 				kn0 = SLIST_NEXT(kn, kn_link);
1229 				if (kq == kn->kn_kq) {
1230 					kn->kn_fop->f_detach(kn);
1231 					/* XXX non-fd release of kn->kn_ptr */
1232 					pool_put(&knote_pool, kn);
1233 					*knp = kn0;
1234 				} else {
1235 					knp = &SLIST_NEXT(kn, kn_link);
1236 				}
1237 				kn = kn0;
1238 			}
1239 		}
1240 	}
1241 	pool_put(&kqueue_pool, kq);
1242 	fp->f_data = NULL;
1243 
1244 	return (0);
1245 }
1246 
1247 /*
1248  * wakeup a kqueue
1249  */
1250 static void
1251 kqueue_wakeup(struct kqueue *kq)
1252 {
1253 	int s;
1254 
1255 	s = splsched();
1256 	simple_lock(&kq->kq_lock);
1257 	if (kq->kq_state & KQ_SLEEP) {		/* if currently sleeping ...  */
1258 		kq->kq_state &= ~KQ_SLEEP;
1259 		wakeup(kq);			/* ... wakeup */
1260 	}
1261 
1262 	/* Notify select/poll and kevent. */
1263 	selnotify(&kq->kq_sel, 0);
1264 	simple_unlock(&kq->kq_lock);
1265 	splx(s);
1266 }
1267 
1268 /*
1269  * struct fileops kqfilter method for a kqueue descriptor.
1270  * Event triggered when monitored kqueue changes.
1271  */
1272 /*ARGSUSED*/
1273 static int
1274 kqueue_kqfilter(struct file *fp, struct knote *kn)
1275 {
1276 	struct kqueue *kq;
1277 
1278 	KASSERT(fp == kn->kn_fp);
1279 	kq = (struct kqueue *)kn->kn_fp->f_data;
1280 	if (kn->kn_filter != EVFILT_READ)
1281 		return (1);
1282 	kn->kn_fop = &kqread_filtops;
1283 	SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext);
1284 	return (0);
1285 }
1286 
1287 
1288 /*
1289  * Walk down a list of knotes, activating them if their event has triggered.
1290  */
1291 void
1292 knote(struct klist *list, long hint)
1293 {
1294 	struct knote *kn;
1295 
1296 	SLIST_FOREACH(kn, list, kn_selnext)
1297 		if (kn->kn_fop->f_event(kn, hint))
1298 			KNOTE_ACTIVATE(kn);
1299 }
1300 
1301 /*
1302  * Remove all knotes from a specified klist
1303  */
1304 void
1305 knote_remove(struct lwp *l, struct klist *list)
1306 {
1307 	struct knote *kn;
1308 
1309 	while ((kn = SLIST_FIRST(list)) != NULL) {
1310 		kn->kn_fop->f_detach(kn);
1311 		knote_drop(kn, l, l->l_proc->p_fd);
1312 	}
1313 }
1314 
1315 /*
1316  * Remove all knotes referencing a specified fd
1317  */
1318 void
1319 knote_fdclose(struct lwp *l, int fd)
1320 {
1321 	struct filedesc	*fdp;
1322 	struct klist	*list;
1323 
1324 	fdp = l->l_proc->p_fd;
1325 	list = &fdp->fd_knlist[fd];
1326 	knote_remove(l, list);
1327 }
1328 
1329 /*
1330  * Attach a new knote to a file descriptor
1331  */
1332 static void
1333 knote_attach(struct knote *kn, struct filedesc *fdp)
1334 {
1335 	struct klist	*list;
1336 	int		size;
1337 
1338 	if (! kn->kn_fop->f_isfd) {
1339 		/* if knote is not on an fd, store on internal hash table */
1340 		if (fdp->fd_knhashmask == 0)
1341 			fdp->fd_knhash = hashinit(KN_HASHSIZE, HASH_LIST,
1342 			    M_KEVENT, M_WAITOK, &fdp->fd_knhashmask);
1343 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1344 		goto done;
1345 	}
1346 
1347 	/*
1348 	 * otherwise, knote is on an fd.
1349 	 * knotes are stored in fd_knlist indexed by kn->kn_id.
1350 	 */
1351 	if (fdp->fd_knlistsize <= kn->kn_id) {
1352 		/* expand list, it's too small */
1353 		size = fdp->fd_knlistsize;
1354 		while (size <= kn->kn_id) {
1355 			/* grow in KQ_EXTENT chunks */
1356 			size += KQ_EXTENT;
1357 		}
1358 		list = malloc(size * sizeof(struct klist *), M_KEVENT,M_WAITOK);
1359 		if (fdp->fd_knlist) {
1360 			/* copy existing knlist */
1361 			memcpy((caddr_t)list, (caddr_t)fdp->fd_knlist,
1362 			    fdp->fd_knlistsize * sizeof(struct klist *));
1363 		}
1364 		/*
1365 		 * Zero new memory. Stylistically, SLIST_INIT() should be
1366 		 * used here, but that does same thing as the memset() anyway.
1367 		 */
1368 		memset(&list[fdp->fd_knlistsize], 0,
1369 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
1370 
1371 		/* switch to new knlist */
1372 		if (fdp->fd_knlist != NULL)
1373 			free(fdp->fd_knlist, M_KEVENT);
1374 		fdp->fd_knlistsize = size;
1375 		fdp->fd_knlist = list;
1376 	}
1377 
1378 	/* get list head for this fd */
1379 	list = &fdp->fd_knlist[kn->kn_id];
1380  done:
1381 	/* add new knote */
1382 	SLIST_INSERT_HEAD(list, kn, kn_link);
1383 	kn->kn_status = 0;
1384 }
1385 
1386 /*
1387  * Drop knote.
1388  * Should be called at spl == 0, since we don't want to hold spl
1389  * while calling FILE_UNUSE and free.
1390  */
1391 static void
1392 knote_drop(struct knote *kn, struct lwp *l, struct filedesc *fdp)
1393 {
1394 	struct klist	*list;
1395 
1396 	if (kn->kn_fop->f_isfd)
1397 		list = &fdp->fd_knlist[kn->kn_id];
1398 	else
1399 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1400 
1401 	SLIST_REMOVE(list, kn, knote, kn_link);
1402 	if (kn->kn_status & KN_QUEUED)
1403 		knote_dequeue(kn);
1404 	if (kn->kn_fop->f_isfd)
1405 		FILE_UNUSE(kn->kn_fp, l);
1406 	pool_put(&knote_pool, kn);
1407 }
1408 
1409 
1410 /*
1411  * Queue new event for knote.
1412  */
1413 static void
1414 knote_enqueue(struct knote *kn)
1415 {
1416 	struct kqueue	*kq;
1417 	int		s;
1418 
1419 	kq = kn->kn_kq;
1420 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
1421 
1422 	s = splsched();
1423 	simple_lock(&kq->kq_lock);
1424 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1425 	kn->kn_status |= KN_QUEUED;
1426 	kq->kq_count++;
1427 	simple_unlock(&kq->kq_lock);
1428 	splx(s);
1429 	kqueue_wakeup(kq);
1430 }
1431 
1432 /*
1433  * Dequeue event for knote.
1434  */
1435 static void
1436 knote_dequeue(struct knote *kn)
1437 {
1438 	struct kqueue	*kq;
1439 	int		s;
1440 
1441 	KASSERT(kn->kn_status & KN_QUEUED);
1442 	kq = kn->kn_kq;
1443 
1444 	s = splsched();
1445 	simple_lock(&kq->kq_lock);
1446 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1447 	kn->kn_status &= ~KN_QUEUED;
1448 	kq->kq_count--;
1449 	simple_unlock(&kq->kq_lock);
1450 	splx(s);
1451 }
1452