xref: /freebsd/sys/kern/kern_event.c (revision a79b71281cd63ad7a6cc43a6d5673a2510b51630)
1 /*-
2  * Copyright (c) 1999,2000 Jonathan Lemon <jlemon@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  *	$FreeBSD$
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/kernel.h>
32 #include <sys/proc.h>
33 #include <sys/malloc.h>
34 #include <sys/unistd.h>
35 #include <sys/file.h>
36 #include <sys/fcntl.h>
37 #include <sys/select.h>
38 #include <sys/queue.h>
39 #include <sys/event.h>
40 #include <sys/eventvar.h>
41 #include <sys/poll.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/stat.h>
46 #include <sys/sysproto.h>
47 #include <sys/uio.h>
48 
49 #include <vm/vm_zone.h>
50 
51 static int 	filt_nullattach(struct knote *kn);
52 static int 	filt_rwtypattach(struct knote *kn);
53 static int	filt_kqattach(struct knote *kn);
54 static void	filt_kqdetach(struct knote *kn);
55 static int	filt_kqueue(struct knote *kn, long hint);
56 static int	filt_procattach(struct knote *kn);
57 static void	filt_procdetach(struct knote *kn);
58 static int	filt_proc(struct knote *kn, long hint);
59 
60 static int	kqueue_scan(struct file *fp, int maxevents,
61 		    struct kevent *ulistp, struct timespec *timeout,
62 		    struct proc *p);
63 static int 	kqueue_read(struct file *fp, struct uio *uio,
64 		    struct ucred *cred, int flags, struct proc *p);
65 static int	kqueue_write(struct file *fp, struct uio *uio,
66 		    struct ucred *cred, int flags, struct proc *p);
67 static int	kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
68 		    struct proc *p);
69 static int 	kqueue_poll(struct file *fp, int events, struct ucred *cred,
70 		    struct proc *p);
71 static int 	kqueue_stat(struct file *fp, struct stat *st, struct proc *p);
72 static int 	kqueue_close(struct file *fp, struct proc *p);
73 static void 	kqueue_wakeup(struct kqueue *kq);
74 
75 static void 	knote_attach(struct knote *kn, struct filedesc *fdp);
76 static void 	knote_drop(struct knote *kn, struct proc *p);
77 static void 	knote_enqueue(struct knote *kn);
78 static void 	knote_dequeue(struct knote *kn);
79 static void 	knote_init(void);
80 static struct 	knote *knote_alloc(void);
81 static void 	knote_free(struct knote *kn);
82 
83 static vm_zone_t	knote_zone;
84 
85 #define KNOTE_ACTIVATE(kn) do { 					\
86 	kn->kn_status |= KN_ACTIVE;					\
87 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
88 		knote_enqueue(kn);					\
89 } while(0)
90 
91 #define	KN_HASHSIZE		64		/* XXX should be tunable */
92 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
93 
94 static struct fileops kqueueops = {
95 	kqueue_read,
96 	kqueue_write,
97 	kqueue_ioctl,
98 	kqueue_poll,
99 	kqueue_stat,
100 	kqueue_close
101 };
102 
103 extern struct filterops so_rwfiltops[];
104 extern struct filterops fifo_rwfiltops[];
105 extern struct filterops pipe_rwfiltops[];
106 extern struct filterops vn_rwfiltops[];
107 
108 static struct filterops kq_rwfiltops[] = {
109     { 1, filt_kqattach, filt_kqdetach, filt_kqueue },
110     { 1, filt_nullattach, NULL, NULL },
111 };
112 
113 extern struct filterops aio_filtops;
114 extern struct filterops sig_filtops;
115 extern struct filterops vn_filtops;
116 
117 static struct filterops rwtype_filtops =
118 	{ 1, filt_rwtypattach, NULL, NULL };
119 static struct filterops proc_filtops =
120 	{ 0, filt_procattach, filt_procdetach, filt_proc };
121 
122 /*
123  * XXX
124  * These must match the order of defines in <sys/file.h>
125  */
126 static struct filterops *rwtypfilt_sw[] = {
127 	NULL,				/* 0 */
128 	vn_rwfiltops,			/* DTYPE_VNODE */
129 	so_rwfiltops,			/* DTYPE_SOCKET */
130 	pipe_rwfiltops,			/* DTYPE_PIPE */
131 	fifo_rwfiltops,			/* DTYPE_FIFO */
132 	kq_rwfiltops,			/* DTYPE_KQUEUE */
133 };
134 
135 /*
136  * table for for all system-defined filters.
137  */
138 static struct filterops *sysfilt_ops[] = {
139 	&rwtype_filtops,		/* EVFILT_READ */
140 	&rwtype_filtops,		/* EVFILT_WRITE */
141 	&aio_filtops,			/* EVFILT_AIO */
142 	&vn_filtops,			/* EVFILT_VNODE */
143 	&proc_filtops,			/* EVFILT_PROC */
144 	&sig_filtops,			/* EVFILT_SIGNAL */
145 };
146 
147 static int
148 filt_nullattach(struct knote *kn)
149 {
150 	return (ENXIO);
151 }
152 
153 /*
154  * file-type specific attach routine for read/write filters
155  */
156 static int
157 filt_rwtypattach(struct knote *kn)
158 {
159 	struct filterops *fops;
160 
161 	fops = rwtypfilt_sw[kn->kn_fp->f_type];
162 	if (fops == NULL)
163 		return (EINVAL);
164 	kn->kn_fop = &fops[~kn->kn_filter];	/* convert to 0-base index */
165 	return (kn->kn_fop->f_attach(kn));
166 }
167 
168 static int
169 filt_kqattach(struct knote *kn)
170 {
171 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
172 
173 	SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext);
174 	return (0);
175 }
176 
177 static void
178 filt_kqdetach(struct knote *kn)
179 {
180 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
181 
182 	SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext);
183 }
184 
185 /*ARGSUSED*/
186 static int
187 filt_kqueue(struct knote *kn, long hint)
188 {
189 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
190 
191 	kn->kn_data = kq->kq_count;
192 	return (kn->kn_data > 0);
193 }
194 
195 static int
196 filt_procattach(struct knote *kn)
197 {
198 	struct proc *p;
199 
200 	p = pfind(kn->kn_id);
201 	if (p == NULL)
202 		return (ESRCH);
203 	if (! PRISON_CHECK(curproc, p))
204 		return (EACCES);
205 
206 	kn->kn_ptr.p_proc = p;
207 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
208 
209 	/*
210 	 * internal flag indicating registration done by kernel
211 	 */
212 	if (kn->kn_flags & EV_FLAG1) {
213 		kn->kn_data = kn->kn_sdata;		/* ppid */
214 		kn->kn_fflags = NOTE_CHILD;
215 		kn->kn_flags &= ~EV_FLAG1;
216 	}
217 
218 	/* XXX lock the proc here while adding to the list? */
219 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
220 
221 	return (0);
222 }
223 
224 /*
225  * The knote may be attached to a different process, which may exit,
226  * leaving nothing for the knote to be attached to.  So when the process
227  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
228  * it will be deleted when read out.  However, as part of the knote deletion,
229  * this routine is called, so a check is needed to avoid actually performing
230  * a detach, because the original process does not exist any more.
231  */
232 static void
233 filt_procdetach(struct knote *kn)
234 {
235 	struct proc *p = kn->kn_ptr.p_proc;
236 
237 	if (kn->kn_status & KN_DETACHED)
238 		return;
239 
240 	/* XXX locking?  this might modify another process. */
241 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
242 }
243 
244 static int
245 filt_proc(struct knote *kn, long hint)
246 {
247 	u_int event;
248 
249 	/*
250 	 * mask off extra data
251 	 */
252 	event = (u_int)hint & NOTE_PCTRLMASK;
253 
254 	/*
255 	 * if the user is interested in this event, record it.
256 	 */
257 	if (kn->kn_sfflags & event)
258 		kn->kn_fflags |= event;
259 
260 	/*
261 	 * process is gone, so flag the event as finished.
262 	 */
263 	if (event == NOTE_EXIT) {
264 		kn->kn_status |= KN_DETACHED;
265 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
266 		return (1);
267 	}
268 
269 	/*
270 	 * process forked, and user wants to track the new process,
271 	 * so attach a new knote to it, and immediately report an
272 	 * event with the parent's pid.
273 	 */
274 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
275 		struct kevent kev;
276 		int error;
277 
278 		/*
279 		 * register knote with new process.
280 		 */
281 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
282 		kev.filter = kn->kn_filter;
283 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
284 		kev.fflags = kn->kn_sfflags;
285 		kev.data = kn->kn_id;			/* parent */
286 		error = kqueue_register(kn->kn_kq, &kev, NULL);
287 		if (error)
288 			kn->kn_fflags |= NOTE_TRACKERR;
289 	}
290 
291 	return (kn->kn_fflags != 0);
292 }
293 
294 int
295 kqueue(struct proc *p, struct kqueue_args *uap)
296 {
297 	struct filedesc *fdp = p->p_fd;
298 	struct kqueue *kq;
299 	struct file *fp;
300 	int fd, error;
301 
302 	error = falloc(p, &fp, &fd);
303 	if (error)
304 		return (error);
305 	fp->f_flag = FREAD | FWRITE;
306 	fp->f_type = DTYPE_KQUEUE;
307 	fp->f_ops = &kqueueops;
308 	kq = malloc(sizeof(struct kqueue), M_TEMP, M_WAITOK);
309 	bzero(kq, sizeof(*kq));
310 	TAILQ_INIT(&kq->kq_head);
311 	fp->f_data = (caddr_t)kq;
312 	p->p_retval[0] = fd;
313 	fdp->fd_knlistsize = 0;		/* mark this fdesc as having a kq */
314 	kq->kq_fdp = fdp;
315 	return (error);
316 }
317 
318 #ifndef _SYS_SYSPROTO_H_
319 struct kevent_args {
320 	int	fd;
321 	int	nchanges;
322 	struct	kevent **changelist;
323 	int	nevents;
324 	struct	kevent *eventlist;
325 	struct	timespec *timeout;
326 };
327 #endif
328 int
329 kevent(struct proc *p, struct kevent_args *uap)
330 {
331         struct filedesc* fdp = p->p_fd;
332 	struct kevent kev;
333 	struct kqueue *kq;
334         struct file *fp;
335 	struct timespec ts;
336 	int i, n, nerrors, error;
337 
338         if (((u_int)uap->fd) >= fdp->fd_nfiles ||
339             (fp = fdp->fd_ofiles[uap->fd]) == NULL ||
340 	    (fp->f_type != DTYPE_KQUEUE))
341 		return (EBADF);
342 
343 	if (uap->timeout != NULL) {
344 		error = copyin((caddr_t)uap->timeout, (caddr_t)&ts,
345 		    sizeof(ts));
346 		if (error)
347 			return error;
348 		uap->timeout = &ts;
349 	}
350 
351 	kq = (struct kqueue *)fp->f_data;
352 	nerrors = 0;
353 
354 	while (uap->nchanges > 0) {
355 		n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges;
356 		error = copyin((caddr_t)uap->changelist, (caddr_t)kq->kq_kevp,
357 		    n * sizeof(struct kevent *));
358 		if (error)
359 			return (error);
360 		for (i = 0; i < n; i++) {
361 			error = copyin((caddr_t)kq->kq_kevp[i],
362 			    (caddr_t)&kev, sizeof(kev));
363 			if (error)
364 				return (error);
365 			kev.flags &= ~EV_SYSFLAGS;
366 			error = kqueue_register(kq, &kev, p);
367 			if (error) {
368 				if (uap->nevents != 0) {
369 					kev.flags = EV_ERROR;
370 					kev.data = error;
371 					(void) copyout((caddr_t)&kev,
372 					    (caddr_t)uap->eventlist,
373 					    sizeof(kev));
374 					uap->eventlist++;
375 					uap->nevents--;
376 					nerrors++;
377 				} else {
378 					return (error);
379 				}
380 			}
381 		}
382 		uap->nchanges -= n;
383 		uap->changelist += n;
384 	}
385 	if (nerrors) {
386         	p->p_retval[0] = nerrors;
387 		return (0);
388 	}
389 
390 	error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, p);
391 	return (error);
392 }
393 
394 int
395 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p)
396 {
397 	struct filedesc *fdp = kq->kq_fdp;
398 	struct filterops *fops;
399 	struct file *fp = NULL;
400 	struct knote *kn = NULL;
401 	int s, error = 0;
402 
403 	if (kev->filter < 0) {
404 		if (kev->filter + EVFILT_SYSCOUNT < 0)
405 			return (EINVAL);
406 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
407 	} else {
408 		/*
409 		 * XXX
410 		 * filter attach routine is responsible for insuring that
411 		 * the identifier can be attached to it.
412 		 */
413 		printf("unknown filter: %d\n", kev->filter);
414 		return (EINVAL);
415 	}
416 
417 	if (fops->f_isfd) {
418 		/* validate descriptor; ignore invalid descriptors */
419 		if ((u_int)kev->ident >= fdp->fd_nfiles ||
420 		    (fp = fdp->fd_ofiles[kev->ident]) == NULL)
421 			return (0);
422 
423 		if (kev->ident < fdp->fd_knlistsize) {
424 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
425 				if (kq == kn->kn_kq &&
426 				    kev->filter == kn->kn_filter)
427 					break;
428 		}
429 	} else {
430 		if (fdp->fd_knhashmask != 0) {
431 			struct klist *list;
432 
433 			list = &fdp->fd_knhash[
434 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
435 			SLIST_FOREACH(kn, list, kn_link)
436 				if (kev->ident == kn->kn_id &&
437 				    kq == kn->kn_kq &&
438 				    kev->filter == kn->kn_filter)
439 					break;
440 		}
441 	}
442 
443 	if (kn == NULL && ((kev->flags & EV_ADD) == 0))
444 		goto done;
445 
446 	/*
447 	 * kn now contains the matching knote, or NULL if no match
448 	 */
449 	if (kev->flags & EV_ADD) {
450 		int attach = 0;
451 
452 		if (kn == NULL) {
453 			kn = knote_alloc();
454 			if (kn == NULL)
455 				return (ENOMEM);
456 			if (fp != NULL)
457 				fhold(fp);
458 			kn->kn_fp = fp;
459 			kn->kn_kq = kq;
460 			kn->kn_fop = fops;
461 			attach = 1;
462 		}
463 		kn->kn_sfflags = kev->fflags;
464 		kn->kn_sdata = kev->data;
465 		kev->fflags = 0;
466 		kev->data = 0;
467 		kn->kn_kevent = *kev;
468 
469 		if (attach) {
470 			knote_attach(kn, fdp);
471 			if ((error = fops->f_attach(kn)) != 0) {
472 				knote_drop(kn, p);
473 				goto done;
474 			}
475 		}
476 		s = splhigh();
477 		if (kn->kn_fop->f_event(kn, 0))
478 			KNOTE_ACTIVATE(kn);
479 		splx(s);
480 	} else if (kev->flags & EV_DELETE) {
481 		kn->kn_fop->f_detach(kn);
482 		knote_drop(kn, p);
483 		goto done;
484 	}
485 
486 	if ((kev->flags & EV_DISABLE) &&
487 	    ((kn->kn_status & KN_DISABLED) == 0)) {
488 		s = splhigh();
489 		kn->kn_status |= KN_DISABLED;
490 		splx(s);
491 	}
492 
493 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
494 		s = splhigh();
495 		kn->kn_status &= ~KN_DISABLED;
496 		if ((kn->kn_status & KN_ACTIVE) &&
497 		    ((kn->kn_status & KN_QUEUED) == 0))
498 			knote_enqueue(kn);
499 		splx(s);
500 	}
501 
502 done:
503 	return (error);
504 }
505 
506 static int
507 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
508 	struct timespec *tsp, struct proc *p)
509 {
510 	struct kqueue *kq = (struct kqueue *)fp->f_data;
511 	struct kevent *kevp;
512 	struct timeval atv, rtv, ttv;
513 	struct knote *kn, marker;
514 	int s, count, timeout, nkev = 0, error = 0;
515 
516 	count = maxevents;
517 	if (count == 0)
518 		goto done;
519 
520         if (tsp != NULL) {
521 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
522                 if (itimerfix(&atv)) {
523 			error = EINVAL;
524 			goto done;
525 		}
526 		timeout = atv.tv_sec > 24 * 60 * 60 ?
527 			24 * 60 * 60 * hz : tvtohz(&atv);
528                 getmicrouptime(&rtv);
529                 timevaladd(&atv, &rtv);
530         } else {
531                 atv.tv_sec = 0;
532 		timeout = 0;
533 	}
534 	goto start;
535 
536 retry:
537 	if (atv.tv_sec) {
538 		getmicrouptime(&rtv);
539 		if (timevalcmp(&rtv, &atv, >=))
540 			goto done;
541 		ttv = atv;
542 		timevalsub(&ttv, &rtv);
543 		timeout = ttv.tv_sec > 24 * 60 * 60 ?
544 			24 * 60 * 60 * hz : tvtohz(&ttv);
545 	}
546 
547 start:
548 	kevp = kq->kq_kev;
549 	s = splhigh();
550 	if (kq->kq_count == 0) {
551 		kq->kq_state |= KQ_SLEEP;
552 		error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout);
553 		splx(s);
554 		if (error == 0)
555 			goto retry;
556 		/* don't restart after signals... */
557 		if (error == ERESTART)
558 			error = EINTR;
559 		else if (error == EWOULDBLOCK)
560 			error = 0;
561 		goto done;
562 	}
563 
564 	TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
565 	while (count) {
566 		kn = TAILQ_FIRST(&kq->kq_head);
567 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
568 		if (kn == &marker) {
569 			splx(s);
570 			if (count == maxevents)
571 				goto retry;
572 			goto done;
573 		}
574 		if (kn->kn_status & KN_DISABLED) {
575 			kn->kn_status &= ~KN_QUEUED;
576 			kq->kq_count--;
577 			continue;
578 		}
579 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
580 		    kn->kn_fop->f_event(kn, 0) == 0) {
581 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
582 			kq->kq_count--;
583 			continue;
584 		}
585 		*kevp = kn->kn_kevent;
586 		kevp++;
587 		nkev++;
588 		if (kn->kn_flags & EV_ONESHOT) {
589 			kn->kn_status &= ~KN_QUEUED;
590 			kq->kq_count--;
591 			splx(s);
592 			kn->kn_fop->f_detach(kn);
593 			knote_drop(kn, p);
594 			s = splhigh();
595 		} else if (kn->kn_flags & EV_CLEAR) {
596 			kn->kn_data = 0;
597 			kn->kn_fflags = 0;
598 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
599 			kq->kq_count--;
600 		} else {
601 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
602 		}
603 		count--;
604 		if (nkev == KQ_NEVENTS) {
605 			splx(s);
606 			error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
607 			    sizeof(struct kevent) * nkev);
608 			ulistp += nkev;
609 			nkev = 0;
610 			kevp = kq->kq_kev;
611 			s = splhigh();
612 			if (error)
613 				break;
614 		}
615 	}
616 	TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
617 	splx(s);
618 done:
619 	if (nkev != 0)
620 		error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
621 		    sizeof(struct kevent) * nkev);
622         p->p_retval[0] = maxevents - count;
623 	return (error);
624 }
625 
626 /*
627  * XXX
628  * This could be expanded to call kqueue_scan, if desired.
629  */
630 /*ARGSUSED*/
631 static int
632 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred,
633 	int flags, struct proc *p)
634 {
635 	return (ENXIO);
636 }
637 
638 /*ARGSUSED*/
639 static int
640 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred,
641 	 int flags, struct proc *p)
642 {
643 	return (ENXIO);
644 }
645 
646 /*ARGSUSED*/
647 static int
648 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
649 {
650 	return (ENOTTY);
651 }
652 
653 /*ARGSUSED*/
654 static int
655 kqueue_poll(struct file *fp, int events, struct ucred *cred, struct proc *p)
656 {
657 	struct kqueue *kq = (struct kqueue *)fp->f_data;
658 	int revents = 0;
659 	int s = splnet();
660 
661         if (events & (POLLIN | POLLRDNORM)) {
662                 if (kq->kq_count) {
663                         revents |= events & (POLLIN | POLLRDNORM);
664 		} else {
665                         selrecord(p, &kq->kq_sel);
666 			kq->kq_state |= KQ_SEL;
667 		}
668 	}
669 	splx(s);
670 	return (revents);
671 }
672 
673 /*ARGSUSED*/
674 static int
675 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
676 {
677 	struct kqueue *kq = (struct kqueue *)fp->f_data;
678 
679 	bzero((void *)st, sizeof(*st));
680 	st->st_size = kq->kq_count;
681 	st->st_blksize = sizeof(struct kevent);
682 	return (0);
683 }
684 
685 /*ARGSUSED*/
686 static int
687 kqueue_close(struct file *fp, struct proc *p)
688 {
689 	struct kqueue *kq = (struct kqueue *)fp->f_data;
690 	struct filedesc *fdp = p->p_fd;
691 	struct knote **knp, *kn, *kn0;
692 	int i;
693 
694 	for (i = 0; i < fdp->fd_knlistsize; i++) {
695 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
696 		kn = *knp;
697 		while (kn != NULL) {
698 			kn0 = SLIST_NEXT(kn, kn_link);
699 			if (kq == kn->kn_kq) {
700 				kn->kn_fop->f_detach(kn);
701 				fdrop(kn->kn_fp, p);
702 				knote_free(kn);
703 				*knp = kn0;
704 			} else {
705 				knp = &SLIST_NEXT(kn, kn_link);
706 			}
707 			kn = kn0;
708 		}
709 	}
710 	if (fdp->fd_knhashmask != 0) {
711 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
712 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
713 			kn = *knp;
714 			while (kn != NULL) {
715 				kn0 = SLIST_NEXT(kn, kn_link);
716 				if (kq == kn->kn_kq) {
717 					kn->kn_fop->f_detach(kn);
718 		/* XXX non-fd release of kn->kn_ptr */
719 					knote_free(kn);
720 					*knp = kn0;
721 				} else {
722 					knp = &SLIST_NEXT(kn, kn_link);
723 				}
724 				kn = kn0;
725 			}
726 		}
727 	}
728 	free(kq, M_TEMP);
729 	fp->f_data = NULL;
730 
731 	return (0);
732 }
733 
734 static void
735 kqueue_wakeup(struct kqueue *kq)
736 {
737 
738 	if (kq->kq_state & KQ_SLEEP) {
739 		kq->kq_state &= ~KQ_SLEEP;
740 		wakeup(kq);
741 	}
742 	if (kq->kq_state & KQ_SEL) {
743 		kq->kq_state &= ~KQ_SEL;
744 		selwakeup(&kq->kq_sel);
745 	}
746 	KNOTE(&kq->kq_sel.si_note, 0);
747 }
748 
749 /*
750  * walk down a list of knotes, activating them if their event has triggered.
751  */
752 void
753 knote(struct klist *list, long hint)
754 {
755 	struct knote *kn;
756 
757 	SLIST_FOREACH(kn, list, kn_selnext)
758 		if (kn->kn_fop->f_event(kn, hint))
759 			KNOTE_ACTIVATE(kn);
760 }
761 
762 /*
763  * remove all knotes from a specified klist
764  */
765 void
766 knote_remove(struct proc *p, struct klist *list)
767 {
768 	struct knote *kn;
769 
770 	while ((kn = SLIST_FIRST(list)) != NULL) {
771 		kn->kn_fop->f_detach(kn);
772 		knote_drop(kn, p);
773 	}
774 }
775 
776 /*
777  * remove all knotes referencing a specified fd
778  */
779 void
780 knote_fdclose(struct proc *p, int fd)
781 {
782 	struct filedesc *fdp = p->p_fd;
783 	struct klist *list = &fdp->fd_knlist[fd];
784 
785 	knote_remove(p, list);
786 }
787 
788 static void
789 knote_attach(struct knote *kn, struct filedesc *fdp)
790 {
791 	struct klist *list;
792 	int size;
793 
794 	if (! kn->kn_fop->f_isfd) {
795 		if (fdp->fd_knhashmask == 0)
796 			fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP,
797 			    &fdp->fd_knhashmask);
798 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
799 		goto done;
800 	}
801 
802 	if (fdp->fd_knlistsize <= kn->kn_id) {
803 		size = fdp->fd_knlistsize;
804 		while (size <= kn->kn_id)
805 			size += KQEXTENT;
806 		MALLOC(list, struct klist *,
807 		    size * sizeof(struct klist *), M_TEMP, M_WAITOK);
808 		bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
809 		    fdp->fd_knlistsize * sizeof(struct klist *));
810 		bzero((caddr_t)list +
811 		    fdp->fd_knlistsize * sizeof(struct klist *),
812 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
813 		if (fdp->fd_knlist != NULL)
814 			FREE(fdp->fd_knlist, M_TEMP);
815 		fdp->fd_knlistsize = size;
816 		fdp->fd_knlist = list;
817 	}
818 	list = &fdp->fd_knlist[kn->kn_id];
819 done:
820 	SLIST_INSERT_HEAD(list, kn, kn_link);
821 	kn->kn_status = 0;
822 }
823 
824 /*
825  * should be called at spl == 0, since we don't want to hold spl
826  * while calling fdrop and free.
827  */
828 static void
829 knote_drop(struct knote *kn, struct proc *p)
830 {
831         struct filedesc *fdp = p->p_fd;
832 	struct klist *list;
833 
834 	if (kn->kn_fop->f_isfd)
835 		list = &fdp->fd_knlist[kn->kn_id];
836 	else
837 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
838 
839 	SLIST_REMOVE(list, kn, knote, kn_link);
840 	if (kn->kn_status & KN_QUEUED)
841 		knote_dequeue(kn);
842 	if (kn->kn_fop->f_isfd)
843 		fdrop(kn->kn_fp, p);
844 	knote_free(kn);
845 }
846 
847 
848 static void
849 knote_enqueue(struct knote *kn)
850 {
851 	struct kqueue *kq = kn->kn_kq;
852 	int s = splhigh();
853 
854 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
855 
856 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
857 	kn->kn_status |= KN_QUEUED;
858 	kq->kq_count++;
859 	splx(s);
860 	kqueue_wakeup(kq);
861 }
862 
863 static void
864 knote_dequeue(struct knote *kn)
865 {
866 	struct kqueue *kq = kn->kn_kq;
867 	int s = splhigh();
868 
869 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
870 
871 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
872 	kn->kn_status &= ~KN_QUEUED;
873 	kq->kq_count--;
874 	splx(s);
875 }
876 
877 static void
878 knote_init(void)
879 {
880 	knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
881 }
882 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
883 
884 static struct knote *
885 knote_alloc(void)
886 {
887 	return ((struct knote *)zalloc(knote_zone));
888 }
889 
890 static void
891 knote_free(struct knote *kn)
892 {
893 	zfree(knote_zone, kn);
894 }
895