xref: /netbsd-src/sys/kern/kern_event.c (revision da5f4674a3fc214be3572d358b66af40ab9401e7)
1 /*	$NetBSD: kern_event.c,v 1.17 2003/07/18 17:34:07 fvdl 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 1.17 2003/07/18 17:34:07 fvdl 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 int	kqueue_scan(struct file *fp, size_t maxevents,
58 		    struct kevent *ulistp, const struct timespec *timeout,
59 		    struct proc *p, register_t *retval);
60 static void	kqueue_wakeup(struct kqueue *kq);
61 
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 proc *p);
68 static int	kqueue_fcntl(struct file *fp, u_int com, void *data,
69 		    struct proc *p);
70 static int	kqueue_poll(struct file *fp, int events, struct proc *p);
71 static int	kqueue_kqfilter(struct file *fp, struct knote *kn);
72 static int	kqueue_stat(struct file *fp, struct stat *sp, struct proc *p);
73 static int	kqueue_close(struct file *fp, struct proc *p);
74 
75 static 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 proc *p,
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 struct pool	kqueue_pool;
107 struct pool	knote_pool;
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  * kqueue_init:
156  *
157  *	Initialize the kqueue/knote facility.
158  */
159 void
160 kqueue_init(void)
161 {
162 
163 	pool_init(&kqueue_pool, sizeof(struct kqueue), 0, 0, 0, "kqueuepl",
164 	    NULL);
165 	pool_init(&knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl",
166 	    NULL);
167 }
168 
169 /*
170  * Find kfilter entry by name, or NULL if not found.
171  */
172 static const struct kfilter *
173 kfilter_byname_sys(const char *name)
174 {
175 	int i;
176 
177 	for (i = 0; sys_kfilters[i].name != NULL; i++) {
178 		if (strcmp(name, sys_kfilters[i].name) == 0)
179 			return (&sys_kfilters[i]);
180 	}
181 	return (NULL);
182 }
183 
184 static struct kfilter *
185 kfilter_byname_user(const char *name)
186 {
187 	int i;
188 
189 	/* user_kfilters[] could be NULL if no filters were registered */
190 	if (!user_kfilters)
191 		return (NULL);
192 
193 	for (i = 0; user_kfilters[i].name != NULL; i++) {
194 		if (user_kfilters[i].name != '\0' &&
195 		    strcmp(name, user_kfilters[i].name) == 0)
196 			return (&user_kfilters[i]);
197 	}
198 	return (NULL);
199 }
200 
201 static const struct kfilter *
202 kfilter_byname(const char *name)
203 {
204 	const struct kfilter *kfilter;
205 
206 	if ((kfilter = kfilter_byname_sys(name)) != NULL)
207 		return (kfilter);
208 
209 	return (kfilter_byname_user(name));
210 }
211 
212 /*
213  * Find kfilter entry by filter id, or NULL if not found.
214  * Assumes entries are indexed in filter id order, for speed.
215  */
216 static const struct kfilter *
217 kfilter_byfilter(uint32_t filter)
218 {
219 	const struct kfilter *kfilter;
220 
221 	if (filter < EVFILT_SYSCOUNT)	/* it's a system filter */
222 		kfilter = &sys_kfilters[filter];
223 	else if (user_kfilters != NULL &&
224 	    filter < EVFILT_SYSCOUNT + user_kfilterc)
225 					/* it's a user filter */
226 		kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT];
227 	else
228 		return (NULL);		/* out of range */
229 	KASSERT(kfilter->filter == filter);	/* sanity check! */
230 	return (kfilter);
231 }
232 
233 /*
234  * Register a new kfilter. Stores the entry in user_kfilters.
235  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
236  * If retfilter != NULL, the new filterid is returned in it.
237  */
238 int
239 kfilter_register(const char *name, const struct filterops *filtops,
240     int *retfilter)
241 {
242 	struct kfilter *kfilter;
243 	void *space;
244 	int len;
245 
246 	if (name == NULL || name[0] == '\0' || filtops == NULL)
247 		return (EINVAL);	/* invalid args */
248 	if (kfilter_byname(name) != NULL)
249 		return (EEXIST);	/* already exists */
250 	if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT)
251 		return (EINVAL);	/* too many */
252 
253 	/* check if need to grow user_kfilters */
254 	if (user_kfilterc + 1 > user_kfiltermaxc) {
255 		/*
256 		 * Grow in KFILTER_EXTENT chunks. Use malloc(9), because we
257 		 * want to traverse user_kfilters as an array.
258 		 */
259 		user_kfiltermaxc += KFILTER_EXTENT;
260 		kfilter = malloc(user_kfiltermaxc * sizeof(struct filter *),
261 		    M_KEVENT, M_WAITOK);
262 
263 		/* copy existing user_kfilters */
264 		if (user_kfilters != NULL)
265 			memcpy((caddr_t)kfilter, (caddr_t)user_kfilters,
266 			    user_kfilterc * sizeof(struct kfilter *));
267 					/* zero new sections */
268 		memset((caddr_t)kfilter +
269 		    user_kfilterc * sizeof(struct kfilter *), 0,
270 		    (user_kfiltermaxc - user_kfilterc) *
271 		    sizeof(struct kfilter *));
272 					/* switch to new kfilter */
273 		if (user_kfilters != NULL)
274 			free(user_kfilters, M_KEVENT);
275 		user_kfilters = kfilter;
276 	}
277 	len = strlen(name) + 1;		/* copy name */
278 	space = malloc(len, M_KEVENT, M_WAITOK);
279 	memcpy(space, name, len);
280 	user_kfilters[user_kfilterc].name = space;
281 
282 	user_kfilters[user_kfilterc].filter = user_kfilterc + EVFILT_SYSCOUNT;
283 
284 	len = sizeof(struct filterops);	/* copy filtops */
285 	space = malloc(len, M_KEVENT, M_WAITOK);
286 	memcpy(space, filtops, len);
287 	user_kfilters[user_kfilterc].filtops = space;
288 
289 	if (retfilter != NULL)
290 		*retfilter = user_kfilters[user_kfilterc].filter;
291 	user_kfilterc++;		/* finally, increment count */
292 	return (0);
293 }
294 
295 /*
296  * Unregister a kfilter previously registered with kfilter_register.
297  * This retains the filter id, but clears the name and frees filtops (filter
298  * operations), so that the number isn't reused during a boot.
299  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
300  */
301 int
302 kfilter_unregister(const char *name)
303 {
304 	struct kfilter *kfilter;
305 
306 	if (name == NULL || name[0] == '\0')
307 		return (EINVAL);	/* invalid name */
308 
309 	if (kfilter_byname_sys(name) != NULL)
310 		return (EINVAL);	/* can't detach system filters */
311 
312 	kfilter = kfilter_byname_user(name);
313 	if (kfilter == NULL)		/* not found */
314 		return (ENOENT);
315 
316 	if (kfilter->name[0] != '\0') {
317 		/* XXX Cast away const (but we know it's safe. */
318 		free((void *) kfilter->name, M_KEVENT);
319 		kfilter->name = "";	/* mark as `not implemented' */
320 	}
321 	if (kfilter->filtops != NULL) {
322 		/* XXX Cast away const (but we know it's safe. */
323 		free((void *) kfilter->filtops, M_KEVENT);
324 		kfilter->filtops = NULL; /* mark as `not implemented' */
325 	}
326 	return (0);
327 }
328 
329 
330 /*
331  * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file
332  * descriptors. Calls struct fileops kqfilter method for given file descriptor.
333  */
334 static int
335 filt_fileattach(struct knote *kn)
336 {
337 	struct file *fp;
338 
339 	fp = kn->kn_fp;
340 	return ((*fp->f_ops->fo_kqfilter)(fp, kn));
341 }
342 
343 /*
344  * Filter detach method for EVFILT_READ on kqueue descriptor.
345  */
346 static void
347 filt_kqdetach(struct knote *kn)
348 {
349 	struct kqueue *kq;
350 
351 	kq = (struct kqueue *)kn->kn_fp->f_data;
352 	SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext);
353 }
354 
355 /*
356  * Filter event method for EVFILT_READ on kqueue descriptor.
357  */
358 /*ARGSUSED*/
359 static int
360 filt_kqueue(struct knote *kn, long hint)
361 {
362 	struct kqueue *kq;
363 
364 	kq = (struct kqueue *)kn->kn_fp->f_data;
365 	kn->kn_data = kq->kq_count;
366 	return (kn->kn_data > 0);
367 }
368 
369 /*
370  * Filter attach method for EVFILT_PROC.
371  */
372 static int
373 filt_procattach(struct knote *kn)
374 {
375 	struct proc *p;
376 
377 	p = pfind(kn->kn_id);
378 	if (p == NULL)
379 		return (ESRCH);
380 
381 	/*
382 	 * Fail if it's not owned by you, or the last exec gave us
383 	 * setuid/setgid privs (unless you're root).
384 	 */
385 	if ((p->p_cred->p_ruid != curproc->p_cred->p_ruid ||
386 		(p->p_flag & P_SUGID))
387 	    && suser(curproc->p_ucred, &curproc->p_acflag) != 0)
388 		return (EACCES);
389 
390 	kn->kn_ptr.p_proc = p;
391 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
392 
393 	/*
394 	 * internal flag indicating registration done by kernel
395 	 */
396 	if (kn->kn_flags & EV_FLAG1) {
397 		kn->kn_data = kn->kn_sdata;	/* ppid */
398 		kn->kn_fflags = NOTE_CHILD;
399 		kn->kn_flags &= ~EV_FLAG1;
400 	}
401 
402 	/* XXXSMP lock the process? */
403 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
404 
405 	return (0);
406 }
407 
408 /*
409  * Filter detach method for EVFILT_PROC.
410  *
411  * The knote may be attached to a different process, which may exit,
412  * leaving nothing for the knote to be attached to.  So when the process
413  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
414  * it will be deleted when read out.  However, as part of the knote deletion,
415  * this routine is called, so a check is needed to avoid actually performing
416  * a detach, because the original process might not exist any more.
417  */
418 static void
419 filt_procdetach(struct knote *kn)
420 {
421 	struct proc *p;
422 
423 	if (kn->kn_status & KN_DETACHED)
424 		return;
425 
426 	p = kn->kn_ptr.p_proc;
427 	KASSERT(p->p_stat == SDEAD || pfind(kn->kn_id) == p);
428 
429 	/* XXXSMP lock the process? */
430 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
431 }
432 
433 /*
434  * Filter event method for EVFILT_PROC.
435  */
436 static int
437 filt_proc(struct knote *kn, long hint)
438 {
439 	u_int event;
440 
441 	/*
442 	 * mask off extra data
443 	 */
444 	event = (u_int)hint & NOTE_PCTRLMASK;
445 
446 	/*
447 	 * if the user is interested in this event, record it.
448 	 */
449 	if (kn->kn_sfflags & event)
450 		kn->kn_fflags |= event;
451 
452 	/*
453 	 * process is gone, so flag the event as finished.
454 	 */
455 	if (event == NOTE_EXIT) {
456 		/*
457 		 * Detach the knote from watched process and mark
458 		 * it as such. We can't leave this to kqueue_scan(),
459 		 * since the process might not exist by then. And we
460 		 * have to do this now, since psignal KNOTE() is called
461 		 * also for zombies and we might end up reading freed
462 		 * memory if the kevent would already be picked up
463 		 * and knote g/c'ed.
464 		 */
465 		kn->kn_fop->f_detach(kn);
466 		kn->kn_status |= KN_DETACHED;
467 
468 		/* Mark as ONESHOT, so that the knote it g/c'ed when read */
469 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
470 		return (1);
471 	}
472 
473 	/*
474 	 * process forked, and user wants to track the new process,
475 	 * so attach a new knote to it, and immediately report an
476 	 * event with the parent's pid.
477 	 */
478 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
479 		struct kevent kev;
480 		int error;
481 
482 		/*
483 		 * register knote with new process.
484 		 */
485 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
486 		kev.filter = kn->kn_filter;
487 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
488 		kev.fflags = kn->kn_sfflags;
489 		kev.data = kn->kn_id;			/* parent */
490 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
491 		error = kqueue_register(kn->kn_kq, &kev, NULL);
492 		if (error)
493 			kn->kn_fflags |= NOTE_TRACKERR;
494 	}
495 
496 	return (kn->kn_fflags != 0);
497 }
498 
499 static void
500 filt_timerexpire(void *knx)
501 {
502 	struct knote *kn = knx;
503 	int tticks;
504 
505 	kn->kn_data++;
506 	KNOTE_ACTIVATE(kn);
507 
508 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
509 		tticks = mstohz(kn->kn_sdata);
510 		callout_schedule((struct callout *)kn->kn_hook, tticks);
511 	}
512 }
513 
514 /*
515  * data contains amount of time to sleep, in milliseconds
516  */
517 static int
518 filt_timerattach(struct knote *kn)
519 {
520 	struct callout *calloutp;
521 	int tticks;
522 
523 	if (kq_ncallouts >= kq_calloutmax)
524 		return (ENOMEM);
525 	kq_ncallouts++;
526 
527 	tticks = mstohz(kn->kn_sdata);
528 
529 	/* if the supplied value is under our resolution, use 1 tick */
530 	if (tticks == 0) {
531 		if (kn->kn_sdata == 0)
532 			return (EINVAL);
533 		tticks = 1;
534 	}
535 
536 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
537 	MALLOC(calloutp, struct callout *, sizeof(*calloutp),
538 	    M_KEVENT, 0);
539 	callout_init(calloutp);
540 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
541 	kn->kn_hook = calloutp;
542 
543 	return (0);
544 }
545 
546 static void
547 filt_timerdetach(struct knote *kn)
548 {
549 	struct callout *calloutp;
550 
551 	calloutp = (struct callout *)kn->kn_hook;
552 	callout_stop(calloutp);
553 	FREE(calloutp, M_KEVENT);
554 	kq_ncallouts--;
555 }
556 
557 static int
558 filt_timer(struct knote *kn, long hint)
559 {
560 	return (kn->kn_data != 0);
561 }
562 
563 /*
564  * filt_seltrue:
565  *
566  *	This filter "event" routine simulates seltrue().
567  */
568 int
569 filt_seltrue(struct knote *kn, long hint)
570 {
571 
572 	/*
573 	 * We don't know how much data can be read/written,
574 	 * but we know that it *can* be.  This is about as
575 	 * good as select/poll does as well.
576 	 */
577 	kn->kn_data = 0;
578 	return (1);
579 }
580 
581 /*
582  * This provides full kqfilter entry for device switch tables, which
583  * has same effect as filter using filt_seltrue() as filter method.
584  */
585 static void
586 filt_seltruedetach(struct knote *kn)
587 {
588 	/* Nothing to do */
589 }
590 
591 static const struct filterops seltrue_filtops =
592 	{ 1, NULL, filt_seltruedetach, filt_seltrue };
593 
594 int
595 seltrue_kqfilter(dev_t dev, struct knote *kn)
596 {
597 	switch (kn->kn_filter) {
598 	case EVFILT_READ:
599 	case EVFILT_WRITE:
600 		kn->kn_fop = &seltrue_filtops;
601 		break;
602 	default:
603 		return (1);
604 	}
605 
606 	/* Nothing more to do */
607 	return (0);
608 }
609 
610 /*
611  * kqueue(2) system call.
612  */
613 int
614 sys_kqueue(struct lwp *l, void *v, register_t *retval)
615 {
616 	struct filedesc	*fdp;
617 	struct kqueue	*kq;
618 	struct file	*fp;
619 	struct proc	*p;
620 	int		fd, error;
621 
622 	p = l->l_proc;
623 	fdp = p->p_fd;
624 	error = falloc(p, &fp, &fd);	/* setup a new file descriptor */
625 	if (error)
626 		return (error);
627 	fp->f_flag = FREAD | FWRITE;
628 	fp->f_type = DTYPE_KQUEUE;
629 	fp->f_ops = &kqueueops;
630 	kq = pool_get(&kqueue_pool, PR_WAITOK);
631 	memset((char *)kq, 0, sizeof(struct kqueue));
632 	simple_lock_init(&kq->kq_lock);
633 	TAILQ_INIT(&kq->kq_head);
634 	fp->f_data = (caddr_t)kq;	/* store the kqueue with the fp */
635 	*retval = fd;
636 	if (fdp->fd_knlistsize < 0)
637 		fdp->fd_knlistsize = 0;	/* this process has a kq */
638 	kq->kq_fdp = fdp;
639 	FILE_SET_MATURE(fp);
640 	FILE_UNUSE(fp, p);		/* falloc() does FILE_USE() */
641 	return (error);
642 }
643 
644 /*
645  * kevent(2) system call.
646  */
647 int
648 sys_kevent(struct lwp *l, void *v, register_t *retval)
649 {
650 	struct sys_kevent_args /* {
651 		syscallarg(int) fd;
652 		syscallarg(const struct kevent *) changelist;
653 		syscallarg(size_t) nchanges;
654 		syscallarg(struct kevent *) eventlist;
655 		syscallarg(size_t) nevents;
656 		syscallarg(const struct timespec *) timeout;
657 	} */ *uap = v;
658 	struct kevent	*kevp;
659 	struct kqueue	*kq;
660 	struct file	*fp;
661 	struct timespec	ts;
662 	struct proc	*p;
663 	size_t		i, n;
664 	int		nerrors, error;
665 
666 	p = l->l_proc;
667 	/* check that we're dealing with a kq */
668 	fp = fd_getfile(p->p_fd, SCARG(uap, fd));
669 	if (fp == NULL)
670 		return (EBADF);
671 
672 	if (fp->f_type != DTYPE_KQUEUE) {
673 		simple_unlock(&fp->f_slock);
674 		return (EBADF);
675 	}
676 
677 	FILE_USE(fp);
678 
679 	if (SCARG(uap, timeout) != NULL) {
680 		error = copyin(SCARG(uap, timeout), &ts, sizeof(ts));
681 		if (error)
682 			goto done;
683 		SCARG(uap, timeout) = &ts;
684 	}
685 
686 	kq = (struct kqueue *)fp->f_data;
687 	nerrors = 0;
688 
689 	/* traverse list of events to register */
690 	while (SCARG(uap, nchanges) > 0) {
691 		/* copyin a maximum of KQ_EVENTS at each pass */
692 		n = MIN(SCARG(uap, nchanges), KQ_NEVENTS);
693 		error = copyin(SCARG(uap, changelist), kq->kq_kev,
694 		    n * sizeof(struct kevent));
695 		if (error)
696 			goto done;
697 		for (i = 0; i < n; i++) {
698 			kevp = &kq->kq_kev[i];
699 			kevp->flags &= ~EV_SYSFLAGS;
700 			/* register each knote */
701 			error = kqueue_register(kq, kevp, p);
702 			if (error) {
703 				if (SCARG(uap, nevents) != 0) {
704 					kevp->flags = EV_ERROR;
705 					kevp->data = error;
706 					error = copyout((caddr_t)kevp,
707 					    (caddr_t)SCARG(uap, eventlist),
708 					    sizeof(*kevp));
709 					if (error)
710 						goto done;
711 					SCARG(uap, eventlist)++;
712 					SCARG(uap, nevents)--;
713 					nerrors++;
714 				} else {
715 					goto done;
716 				}
717 			}
718 		}
719 		SCARG(uap, nchanges) -= n;	/* update the results */
720 		SCARG(uap, changelist) += n;
721 	}
722 	if (nerrors) {
723 		*retval = nerrors;
724 		error = 0;
725 		goto done;
726 	}
727 
728 	/* actually scan through the events */
729 	error = kqueue_scan(fp, SCARG(uap, nevents), SCARG(uap, eventlist),
730 	    SCARG(uap, timeout), p, retval);
731  done:
732 	FILE_UNUSE(fp, p);
733 	return (error);
734 }
735 
736 /*
737  * Register a given kevent kev onto the kqueue
738  */
739 int
740 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p)
741 {
742 	const struct kfilter *kfilter;
743 	struct filedesc	*fdp;
744 	struct file	*fp;
745 	struct knote	*kn;
746 	int		s, error;
747 
748 	fdp = kq->kq_fdp;
749 	fp = NULL;
750 	kn = NULL;
751 	error = 0;
752 	kfilter = kfilter_byfilter(kev->filter);
753 	if (kfilter == NULL || kfilter->filtops == NULL) {
754 		/* filter not found nor implemented */
755 		return (EINVAL);
756 	}
757 
758 	/* search if knote already exists */
759 	if (kfilter->filtops->f_isfd) {
760 		/* monitoring a file descriptor */
761 		if ((fp = fd_getfile(fdp, kev->ident)) == NULL)
762 			return (EBADF);	/* validate descriptor */
763 		FILE_USE(fp);
764 
765 		if (kev->ident < fdp->fd_knlistsize) {
766 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
767 				if (kq == kn->kn_kq &&
768 				    kev->filter == kn->kn_filter)
769 					break;
770 		}
771 	} else {
772 		/*
773 		 * not monitoring a file descriptor, so
774 		 * lookup knotes in internal hash table
775 		 */
776 		if (fdp->fd_knhashmask != 0) {
777 			struct klist *list;
778 
779 			list = &fdp->fd_knhash[
780 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
781 			SLIST_FOREACH(kn, list, kn_link)
782 				if (kev->ident == kn->kn_id &&
783 				    kq == kn->kn_kq &&
784 				    kev->filter == kn->kn_filter)
785 					break;
786 		}
787 	}
788 
789 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
790 		error = ENOENT;		/* filter not found */
791 		goto done;
792 	}
793 
794 	/*
795 	 * kn now contains the matching knote, or NULL if no match
796 	 */
797 	if (kev->flags & EV_ADD) {
798 		/* add knote */
799 
800 		if (kn == NULL) {
801 			/* create new knote */
802 			kn = pool_get(&knote_pool, PR_WAITOK);
803 			if (kn == NULL) {
804 				error = ENOMEM;
805 				goto done;
806 			}
807 			kn->kn_fp = fp;
808 			kn->kn_kq = kq;
809 			kn->kn_fop = kfilter->filtops;
810 
811 			/*
812 			 * apply reference count to knote structure, and
813 			 * do not release it at the end of this routine.
814 			 */
815 			fp = NULL;
816 
817 			kn->kn_sfflags = kev->fflags;
818 			kn->kn_sdata = kev->data;
819 			kev->fflags = 0;
820 			kev->data = 0;
821 			kn->kn_kevent = *kev;
822 
823 			knote_attach(kn, fdp);
824 			if ((error = kfilter->filtops->f_attach(kn)) != 0) {
825 				knote_drop(kn, p, fdp);
826 				goto done;
827 			}
828 		} else {
829 			/* modify existing knote */
830 
831 			/*
832 			 * The user may change some filter values after the
833 			 * initial EV_ADD, but doing so will not reset any
834 			 * filter which have already been triggered.
835 			 */
836 			kn->kn_sfflags = kev->fflags;
837 			kn->kn_sdata = kev->data;
838 			kn->kn_kevent.udata = kev->udata;
839 		}
840 
841 		s = splsched();
842 		if (kn->kn_fop->f_event(kn, 0))
843 			KNOTE_ACTIVATE(kn);
844 		splx(s);
845 
846 	} else if (kev->flags & EV_DELETE) {	/* delete knote */
847 		kn->kn_fop->f_detach(kn);
848 		knote_drop(kn, p, fdp);
849 		goto done;
850 	}
851 
852 	/* disable knote */
853 	if ((kev->flags & EV_DISABLE) &&
854 	    ((kn->kn_status & KN_DISABLED) == 0)) {
855 		s = splsched();
856 		kn->kn_status |= KN_DISABLED;
857 		splx(s);
858 	}
859 
860 	/* enable knote */
861 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
862 		s = splsched();
863 		kn->kn_status &= ~KN_DISABLED;
864 		if ((kn->kn_status & KN_ACTIVE) &&
865 		    ((kn->kn_status & KN_QUEUED) == 0))
866 			knote_enqueue(kn);
867 		splx(s);
868 	}
869 
870  done:
871 	if (fp != NULL)
872 		FILE_UNUSE(fp, p);
873 	return (error);
874 }
875 
876 /*
877  * Scan through the list of events on fp (for a maximum of maxevents),
878  * returning the results in to ulistp. Timeout is determined by tsp; if
879  * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait
880  * as appropriate.
881  */
882 static int
883 kqueue_scan(struct file *fp, size_t maxevents, struct kevent *ulistp,
884 	const struct timespec *tsp, struct proc *p, register_t *retval)
885 {
886 	struct kqueue	*kq;
887 	struct kevent	*kevp;
888 	struct timeval	atv;
889 	struct knote	*kn, marker;
890 	size_t		count, nkev;
891 	int		s, timeout, error;
892 
893 	kq = (struct kqueue *)fp->f_data;
894 	count = maxevents;
895 	nkev = error = 0;
896 	if (count == 0)
897 		goto done;
898 
899 	if (tsp) {				/* timeout supplied */
900 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
901 		if (itimerfix(&atv)) {
902 			error = EINVAL;
903 			goto done;
904 		}
905 		s = splclock();
906 		timeradd(&atv, &time, &atv);	/* calc. time to wait until */
907 		splx(s);
908 		timeout = hzto(&atv);
909 		if (timeout <= 0)
910 			timeout = -1;		/* do poll */
911 	} else {
912 		/* no timeout, wait forever */
913 		timeout = 0;
914 	}
915 	goto start;
916 
917  retry:
918 	if (tsp) {
919 		/*
920 		 * We have to recalculate the timeout on every retry.
921 		 */
922 		timeout = hzto(&atv);
923 		if (timeout <= 0)
924 			goto done;
925 	}
926 
927  start:
928 	kevp = kq->kq_kev;
929 	s = splsched();
930 	simple_lock(&kq->kq_lock);
931 	if (kq->kq_count == 0) {
932 		if (timeout < 0) {
933 			error = EWOULDBLOCK;
934 			simple_unlock(&kq->kq_lock);
935 		} else {
936 			kq->kq_state |= KQ_SLEEP;
937 			error = ltsleep(kq, PSOCK | PCATCH | PNORELOCK,
938 					"kqread", timeout, &kq->kq_lock);
939 		}
940 		splx(s);
941 		if (error == 0)
942 			goto retry;
943 		/* don't restart after signals... */
944 		if (error == ERESTART)
945 			error = EINTR;
946 		else if (error == EWOULDBLOCK)
947 			error = 0;
948 		goto done;
949 	}
950 
951 	/* mark end of knote list */
952 	TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
953 	simple_unlock(&kq->kq_lock);
954 
955 	while (count) {				/* while user wants data ... */
956 		simple_lock(&kq->kq_lock);
957 		kn = TAILQ_FIRST(&kq->kq_head);	/* get next knote */
958 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
959 		if (kn == &marker) {		/* if it's our marker, stop */
960 			/* What if it's some else's marker? */
961 			simple_unlock(&kq->kq_lock);
962 			splx(s);
963 			if (count == maxevents)
964 				goto retry;
965 			goto done;
966 		}
967 		kq->kq_count--;
968 		simple_unlock(&kq->kq_lock);
969 
970 		if (kn->kn_status & KN_DISABLED) {
971 			/* don't want disabled events */
972 			kn->kn_status &= ~KN_QUEUED;
973 			continue;
974 		}
975 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
976 		    kn->kn_fop->f_event(kn, 0) == 0) {
977 			/*
978 			 * non-ONESHOT event that hasn't
979 			 * triggered again, so de-queue.
980 			 */
981 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
982 			continue;
983 		}
984 		*kevp = kn->kn_kevent;
985 		kevp++;
986 		nkev++;
987 		if (kn->kn_flags & EV_ONESHOT) {
988 			/* delete ONESHOT events after retrieval */
989 			kn->kn_status &= ~KN_QUEUED;
990 			splx(s);
991 			kn->kn_fop->f_detach(kn);
992 			knote_drop(kn, p, p->p_fd);
993 			s = splsched();
994 		} else if (kn->kn_flags & EV_CLEAR) {
995 			/* clear state after retrieval */
996 			kn->kn_data = 0;
997 			kn->kn_fflags = 0;
998 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
999 		} else {
1000 			/* add event back on list */
1001 			simple_lock(&kq->kq_lock);
1002 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1003 			kq->kq_count++;
1004 			simple_unlock(&kq->kq_lock);
1005 		}
1006 		count--;
1007 		if (nkev == KQ_NEVENTS) {
1008 			/* do copyouts in KQ_NEVENTS chunks */
1009 			splx(s);
1010 			error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
1011 			    sizeof(struct kevent) * nkev);
1012 			ulistp += nkev;
1013 			nkev = 0;
1014 			kevp = kq->kq_kev;
1015 			s = splsched();
1016 			if (error)
1017 				break;
1018 		}
1019 	}
1020 
1021 	/* remove marker */
1022 	simple_lock(&kq->kq_lock);
1023 	TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
1024 	simple_unlock(&kq->kq_lock);
1025 	splx(s);
1026  done:
1027 	if (nkev != 0) {
1028 		/* copyout remaining events */
1029 		error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
1030 		    sizeof(struct kevent) * nkev);
1031 	}
1032 	*retval = maxevents - count;
1033 
1034 	return (error);
1035 }
1036 
1037 /*
1038  * struct fileops read method for a kqueue descriptor.
1039  * Not implemented.
1040  * XXX: This could be expanded to call kqueue_scan, if desired.
1041  */
1042 /*ARGSUSED*/
1043 static int
1044 kqueue_read(struct file *fp, off_t *offset, struct uio *uio,
1045 	struct ucred *cred, int flags)
1046 {
1047 
1048 	return (ENXIO);
1049 }
1050 
1051 /*
1052  * struct fileops write method for a kqueue descriptor.
1053  * Not implemented.
1054  */
1055 /*ARGSUSED*/
1056 static int
1057 kqueue_write(struct file *fp, off_t *offset, struct uio *uio,
1058 	struct ucred *cred, int flags)
1059 {
1060 
1061 	return (ENXIO);
1062 }
1063 
1064 /*
1065  * struct fileops ioctl method for a kqueue descriptor.
1066  *
1067  * Two ioctls are currently supported. They both use struct kfilter_mapping:
1068  *	KFILTER_BYNAME		find name for filter, and return result in
1069  *				name, which is of size len.
1070  *	KFILTER_BYFILTER	find filter for name. len is ignored.
1071  */
1072 /*ARGSUSED*/
1073 static int
1074 kqueue_ioctl(struct file *fp, u_long com, void *data, struct proc *p)
1075 {
1076 	struct kfilter_mapping	*km;
1077 	const struct kfilter	*kfilter;
1078 	char			*name;
1079 	int			error;
1080 
1081 	km = (struct kfilter_mapping *)data;
1082 	error = 0;
1083 
1084 	switch (com) {
1085 	case KFILTER_BYFILTER:	/* convert filter -> name */
1086 		kfilter = kfilter_byfilter(km->filter);
1087 		if (kfilter != NULL)
1088 			error = copyoutstr(kfilter->name, km->name, km->len,
1089 			    NULL);
1090 		else
1091 			error = ENOENT;
1092 		break;
1093 
1094 	case KFILTER_BYNAME:	/* convert name -> filter */
1095 		MALLOC(name, char *, KFILTER_MAXNAME, M_KEVENT, M_WAITOK);
1096 		error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL);
1097 		if (error) {
1098 			FREE(name, M_KEVENT);
1099 			break;
1100 		}
1101 		kfilter = kfilter_byname(name);
1102 		if (kfilter != NULL)
1103 			km->filter = kfilter->filter;
1104 		else
1105 			error = ENOENT;
1106 		FREE(name, M_KEVENT);
1107 		break;
1108 
1109 	default:
1110 		error = ENOTTY;
1111 
1112 	}
1113 	return (error);
1114 }
1115 
1116 /*
1117  * struct fileops fcntl method for a kqueue descriptor.
1118  * Not implemented.
1119  */
1120 /*ARGSUSED*/
1121 static int
1122 kqueue_fcntl(struct file *fp, u_int com, void *data, struct proc *p)
1123 {
1124 
1125 	return (ENOTTY);
1126 }
1127 
1128 /*
1129  * struct fileops poll method for a kqueue descriptor.
1130  * Determine if kqueue has events pending.
1131  */
1132 static int
1133 kqueue_poll(struct file *fp, int events, struct proc *p)
1134 {
1135 	struct kqueue	*kq;
1136 	int		revents;
1137 
1138 	kq = (struct kqueue *)fp->f_data;
1139 	revents = 0;
1140 	if (events & (POLLIN | POLLRDNORM)) {
1141 		if (kq->kq_count) {
1142 			revents |= events & (POLLIN | POLLRDNORM);
1143 		} else {
1144 			selrecord(p, &kq->kq_sel);
1145 		}
1146 	}
1147 	return (revents);
1148 }
1149 
1150 /*
1151  * struct fileops stat method for a kqueue descriptor.
1152  * Returns dummy info, with st_size being number of events pending.
1153  */
1154 static int
1155 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
1156 {
1157 	struct kqueue	*kq;
1158 
1159 	kq = (struct kqueue *)fp->f_data;
1160 	memset((void *)st, 0, sizeof(*st));
1161 	st->st_size = kq->kq_count;
1162 	st->st_blksize = sizeof(struct kevent);
1163 	st->st_mode = S_IFIFO;
1164 	return (0);
1165 }
1166 
1167 /*
1168  * struct fileops close method for a kqueue descriptor.
1169  * Cleans up kqueue.
1170  */
1171 static int
1172 kqueue_close(struct file *fp, struct proc *p)
1173 {
1174 	struct kqueue	*kq;
1175 	struct filedesc	*fdp;
1176 	struct knote	**knp, *kn, *kn0;
1177 	int		i;
1178 
1179 	kq = (struct kqueue *)fp->f_data;
1180 	fdp = p->p_fd;
1181 	for (i = 0; i < fdp->fd_knlistsize; i++) {
1182 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
1183 		kn = *knp;
1184 		while (kn != NULL) {
1185 			kn0 = SLIST_NEXT(kn, kn_link);
1186 			if (kq == kn->kn_kq) {
1187 				kn->kn_fop->f_detach(kn);
1188 				FILE_UNUSE(kn->kn_fp, p);
1189 				pool_put(&knote_pool, kn);
1190 				*knp = kn0;
1191 			} else {
1192 				knp = &SLIST_NEXT(kn, kn_link);
1193 			}
1194 			kn = kn0;
1195 		}
1196 	}
1197 	if (fdp->fd_knhashmask != 0) {
1198 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
1199 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
1200 			kn = *knp;
1201 			while (kn != NULL) {
1202 				kn0 = SLIST_NEXT(kn, kn_link);
1203 				if (kq == kn->kn_kq) {
1204 					kn->kn_fop->f_detach(kn);
1205 					/* XXX non-fd release of kn->kn_ptr */
1206 					pool_put(&knote_pool, kn);
1207 					*knp = kn0;
1208 				} else {
1209 					knp = &SLIST_NEXT(kn, kn_link);
1210 				}
1211 				kn = kn0;
1212 			}
1213 		}
1214 	}
1215 	pool_put(&kqueue_pool, kq);
1216 	fp->f_data = NULL;
1217 
1218 	return (0);
1219 }
1220 
1221 /*
1222  * wakeup a kqueue
1223  */
1224 static void
1225 kqueue_wakeup(struct kqueue *kq)
1226 {
1227 	int s;
1228 
1229 	s = splsched();
1230 	simple_lock(&kq->kq_lock);
1231 	if (kq->kq_state & KQ_SLEEP) {		/* if currently sleeping ...  */
1232 		kq->kq_state &= ~KQ_SLEEP;
1233 		wakeup(kq);			/* ... wakeup */
1234 	}
1235 
1236 	/* Notify select/poll and kevent. */
1237 	selnotify(&kq->kq_sel, 0);
1238 	simple_unlock(&kq->kq_lock);
1239 	splx(s);
1240 }
1241 
1242 /*
1243  * struct fileops kqfilter method for a kqueue descriptor.
1244  * Event triggered when monitored kqueue changes.
1245  */
1246 /*ARGSUSED*/
1247 static int
1248 kqueue_kqfilter(struct file *fp, struct knote *kn)
1249 {
1250 	struct kqueue *kq;
1251 
1252 	KASSERT(fp == kn->kn_fp);
1253 	kq = (struct kqueue *)kn->kn_fp->f_data;
1254 	if (kn->kn_filter != EVFILT_READ)
1255 		return (1);
1256 	kn->kn_fop = &kqread_filtops;
1257 	SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext);
1258 	return (0);
1259 }
1260 
1261 
1262 /*
1263  * Walk down a list of knotes, activating them if their event has triggered.
1264  */
1265 void
1266 knote(struct klist *list, long hint)
1267 {
1268 	struct knote *kn;
1269 
1270 	SLIST_FOREACH(kn, list, kn_selnext)
1271 		if (kn->kn_fop->f_event(kn, hint))
1272 			KNOTE_ACTIVATE(kn);
1273 }
1274 
1275 /*
1276  * Remove all knotes from a specified klist
1277  */
1278 void
1279 knote_remove(struct proc *p, struct klist *list)
1280 {
1281 	struct knote *kn;
1282 
1283 	while ((kn = SLIST_FIRST(list)) != NULL) {
1284 		kn->kn_fop->f_detach(kn);
1285 		knote_drop(kn, p, p->p_fd);
1286 	}
1287 }
1288 
1289 /*
1290  * Remove all knotes referencing a specified fd
1291  */
1292 void
1293 knote_fdclose(struct proc *p, int fd)
1294 {
1295 	struct filedesc	*fdp;
1296 	struct klist	*list;
1297 
1298 	fdp = p->p_fd;
1299 	list = &fdp->fd_knlist[fd];
1300 	knote_remove(p, list);
1301 }
1302 
1303 /*
1304  * Attach a new knote to a file descriptor
1305  */
1306 static void
1307 knote_attach(struct knote *kn, struct filedesc *fdp)
1308 {
1309 	struct klist	*list;
1310 	int		size;
1311 
1312 	if (! kn->kn_fop->f_isfd) {
1313 		/* if knote is not on an fd, store on internal hash table */
1314 		if (fdp->fd_knhashmask == 0)
1315 			fdp->fd_knhash = hashinit(KN_HASHSIZE, HASH_LIST,
1316 			    M_KEVENT, M_WAITOK, &fdp->fd_knhashmask);
1317 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1318 		goto done;
1319 	}
1320 
1321 	/*
1322 	 * otherwise, knote is on an fd.
1323 	 * knotes are stored in fd_knlist indexed by kn->kn_id.
1324 	 */
1325 	if (fdp->fd_knlistsize <= kn->kn_id) {
1326 		/* expand list, it's too small */
1327 		size = fdp->fd_knlistsize;
1328 		while (size <= kn->kn_id) {
1329 			/* grow in KQ_EXTENT chunks */
1330 			size += KQ_EXTENT;
1331 		}
1332 		list = malloc(size * sizeof(struct klist *), M_KEVENT,M_WAITOK);
1333 		if (fdp->fd_knlist) {
1334 			/* copy existing knlist */
1335 			memcpy((caddr_t)list, (caddr_t)fdp->fd_knlist,
1336 			    fdp->fd_knlistsize * sizeof(struct klist *));
1337 		}
1338 		/*
1339 		 * Zero new memory. Stylistically, SLIST_INIT() should be
1340 		 * used here, but that does same thing as the memset() anyway.
1341 		 */
1342 		memset(&list[fdp->fd_knlistsize], 0,
1343 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
1344 
1345 		/* switch to new knlist */
1346 		if (fdp->fd_knlist != NULL)
1347 			free(fdp->fd_knlist, M_KEVENT);
1348 		fdp->fd_knlistsize = size;
1349 		fdp->fd_knlist = list;
1350 	}
1351 
1352 	/* get list head for this fd */
1353 	list = &fdp->fd_knlist[kn->kn_id];
1354  done:
1355 	/* add new knote */
1356 	SLIST_INSERT_HEAD(list, kn, kn_link);
1357 	kn->kn_status = 0;
1358 }
1359 
1360 /*
1361  * Drop knote.
1362  * Should be called at spl == 0, since we don't want to hold spl
1363  * while calling FILE_UNUSE and free.
1364  */
1365 static void
1366 knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp)
1367 {
1368 	struct klist	*list;
1369 
1370 	if (kn->kn_fop->f_isfd)
1371 		list = &fdp->fd_knlist[kn->kn_id];
1372 	else
1373 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1374 
1375 	SLIST_REMOVE(list, kn, knote, kn_link);
1376 	if (kn->kn_status & KN_QUEUED)
1377 		knote_dequeue(kn);
1378 	if (kn->kn_fop->f_isfd)
1379 		FILE_UNUSE(kn->kn_fp, p);
1380 	pool_put(&knote_pool, kn);
1381 }
1382 
1383 
1384 /*
1385  * Queue new event for knote.
1386  */
1387 static void
1388 knote_enqueue(struct knote *kn)
1389 {
1390 	struct kqueue	*kq;
1391 	int		s;
1392 
1393 	kq = kn->kn_kq;
1394 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
1395 
1396 	s = splsched();
1397 	simple_lock(&kq->kq_lock);
1398 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1399 	kn->kn_status |= KN_QUEUED;
1400 	kq->kq_count++;
1401 	simple_unlock(&kq->kq_lock);
1402 	splx(s);
1403 	kqueue_wakeup(kq);
1404 }
1405 
1406 /*
1407  * Dequeue event for knote.
1408  */
1409 static void
1410 knote_dequeue(struct knote *kn)
1411 {
1412 	struct kqueue	*kq;
1413 	int		s;
1414 
1415 	KASSERT(kn->kn_status & KN_QUEUED);
1416 	kq = kn->kn_kq;
1417 
1418 	s = splsched();
1419 	simple_lock(&kq->kq_lock);
1420 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1421 	kn->kn_status &= ~KN_QUEUED;
1422 	kq->kq_count--;
1423 	simple_unlock(&kq->kq_lock);
1424 	splx(s);
1425 }
1426