xref: /freebsd/sys/kern/kern_event.c (revision b601c69bdbe8755d26570261d7fd4c02ee4eff74)
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 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
287 		error = kqueue_register(kn->kn_kq, &kev, NULL);
288 		if (error)
289 			kn->kn_fflags |= NOTE_TRACKERR;
290 	}
291 
292 	return (kn->kn_fflags != 0);
293 }
294 
295 int
296 kqueue(struct proc *p, struct kqueue_args *uap)
297 {
298 	struct filedesc *fdp = p->p_fd;
299 	struct kqueue *kq;
300 	struct file *fp;
301 	int fd, error;
302 
303 	error = falloc(p, &fp, &fd);
304 	if (error)
305 		return (error);
306 	fp->f_flag = FREAD | FWRITE;
307 	fp->f_type = DTYPE_KQUEUE;
308 	fp->f_ops = &kqueueops;
309 	kq = malloc(sizeof(struct kqueue), M_TEMP, M_WAITOK);
310 	bzero(kq, sizeof(*kq));
311 	TAILQ_INIT(&kq->kq_head);
312 	fp->f_data = (caddr_t)kq;
313 	p->p_retval[0] = fd;
314 	if (fdp->fd_knlistsize < 0)
315 		fdp->fd_knlistsize = 0;		/* this process has a kq */
316 	kq->kq_fdp = fdp;
317 	return (error);
318 }
319 
320 #ifndef _SYS_SYSPROTO_H_
321 struct kevent_args {
322 	int	fd;
323 	struct	kevent *changelist;
324 	int	nchanges;
325 	struct	kevent *eventlist;
326 	int	nevents;
327 	struct	timespec *timeout;
328 };
329 #endif
330 int
331 kevent(struct proc *p, struct kevent_args *uap)
332 {
333 	struct filedesc* fdp = p->p_fd;
334 	struct kevent *kevp;
335 	struct kqueue *kq;
336 	struct file *fp;
337 	struct timespec ts;
338 	int i, n, nerrors, error;
339 
340         if (((u_int)uap->fd) >= fdp->fd_nfiles ||
341             (fp = fdp->fd_ofiles[uap->fd]) == NULL ||
342 	    (fp->f_type != DTYPE_KQUEUE))
343 		return (EBADF);
344 
345 	if (uap->timeout != NULL) {
346 		error = copyin(uap->timeout, &ts, sizeof(ts));
347 		if (error)
348 			return error;
349 		uap->timeout = &ts;
350 	}
351 
352 	kq = (struct kqueue *)fp->f_data;
353 	nerrors = 0;
354 
355 	while (uap->nchanges > 0) {
356 		n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges;
357 		error = copyin(uap->changelist, kq->kq_kev,
358 		    n * sizeof(struct kevent));
359 		if (error)
360 			return (error);
361 		for (i = 0; i < n; i++) {
362 			kevp = &kq->kq_kev[i];
363 			kevp->flags &= ~EV_SYSFLAGS;
364 			error = kqueue_register(kq, kevp, p);
365 			if (error) {
366 				if (uap->nevents != 0) {
367 					kevp->flags = EV_ERROR;
368 					kevp->data = error;
369 					(void) copyout((caddr_t)kevp,
370 					    (caddr_t)uap->eventlist,
371 					    sizeof(*kevp));
372 					uap->eventlist++;
373 					uap->nevents--;
374 					nerrors++;
375 				} else {
376 					return (error);
377 				}
378 			}
379 		}
380 		uap->nchanges -= n;
381 		uap->changelist += n;
382 	}
383 	if (nerrors) {
384         	p->p_retval[0] = nerrors;
385 		return (0);
386 	}
387 
388 	error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, p);
389 	return (error);
390 }
391 
392 int
393 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p)
394 {
395 	struct filedesc *fdp = kq->kq_fdp;
396 	struct filterops *fops;
397 	struct file *fp = NULL;
398 	struct knote *kn = NULL;
399 	int s, error = 0;
400 
401 	if (kev->filter < 0) {
402 		if (kev->filter + EVFILT_SYSCOUNT < 0)
403 			return (EINVAL);
404 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
405 	} else {
406 		/*
407 		 * XXX
408 		 * filter attach routine is responsible for insuring that
409 		 * the identifier can be attached to it.
410 		 */
411 		printf("unknown filter: %d\n", kev->filter);
412 		return (EINVAL);
413 	}
414 
415 	if (fops->f_isfd) {
416 		/* validate descriptor; ignore invalid descriptors */
417 		if ((u_int)kev->ident >= fdp->fd_nfiles ||
418 		    (fp = fdp->fd_ofiles[kev->ident]) == NULL)
419 			return (0);
420 
421 		if (kev->ident < fdp->fd_knlistsize) {
422 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
423 				if (kq == kn->kn_kq &&
424 				    kev->filter == kn->kn_filter)
425 					break;
426 		}
427 	} else {
428 		if (fdp->fd_knhashmask != 0) {
429 			struct klist *list;
430 
431 			list = &fdp->fd_knhash[
432 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
433 			SLIST_FOREACH(kn, list, kn_link)
434 				if (kev->ident == kn->kn_id &&
435 				    kq == kn->kn_kq &&
436 				    kev->filter == kn->kn_filter)
437 					break;
438 		}
439 	}
440 
441 	if (kn == NULL && ((kev->flags & EV_ADD) == 0))
442 		goto done;
443 
444 	/*
445 	 * kn now contains the matching knote, or NULL if no match
446 	 */
447 	if (kev->flags & EV_ADD) {
448 
449 		if (kn == NULL) {
450 			kn = knote_alloc();
451 			if (kn == NULL)
452 				return (ENOMEM);
453 			if (fp != NULL)
454 				fhold(fp);
455 			kn->kn_fp = fp;
456 			kn->kn_kq = kq;
457 			kn->kn_fop = fops;
458 
459 			kn->kn_sfflags = kev->fflags;
460 			kn->kn_sdata = kev->data;
461 			kev->fflags = 0;
462 			kev->data = 0;
463 			kn->kn_kevent = *kev;
464 
465 			knote_attach(kn, fdp);
466 			if ((error = fops->f_attach(kn)) != 0) {
467 				knote_drop(kn, p);
468 				goto done;
469 			}
470 		} else {
471 			/*
472 			 * The user may change some filter values after the
473 			 * initial EV_ADD, but doing so will not reset any
474 			 * filter which have already been triggered.
475 			 */
476 			kn->kn_sfflags = kev->fflags;
477 			kn->kn_sdata = kev->data;
478 			kn->kn_kevent.udata = kev->udata;
479 		}
480 
481 		s = splhigh();
482 		if (kn->kn_fop->f_event(kn, 0))
483 			KNOTE_ACTIVATE(kn);
484 		splx(s);
485 
486 	} else if (kev->flags & EV_DELETE) {
487 		kn->kn_fop->f_detach(kn);
488 		knote_drop(kn, p);
489 		goto done;
490 	}
491 
492 	if ((kev->flags & EV_DISABLE) &&
493 	    ((kn->kn_status & KN_DISABLED) == 0)) {
494 		s = splhigh();
495 		kn->kn_status |= KN_DISABLED;
496 		splx(s);
497 	}
498 
499 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
500 		s = splhigh();
501 		kn->kn_status &= ~KN_DISABLED;
502 		if ((kn->kn_status & KN_ACTIVE) &&
503 		    ((kn->kn_status & KN_QUEUED) == 0))
504 			knote_enqueue(kn);
505 		splx(s);
506 	}
507 
508 done:
509 	return (error);
510 }
511 
512 static int
513 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
514 	struct timespec *tsp, struct proc *p)
515 {
516 	struct kqueue *kq = (struct kqueue *)fp->f_data;
517 	struct kevent *kevp;
518 	struct timeval atv, rtv, ttv;
519 	struct knote *kn, marker;
520 	int s, count, timeout, nkev = 0, error = 0;
521 
522 	count = maxevents;
523 	if (count == 0)
524 		goto done;
525 
526         if (tsp != NULL) {
527 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
528                 if (itimerfix(&atv)) {
529 			error = EINVAL;
530 			goto done;
531 		}
532 		timeout = atv.tv_sec > 24 * 60 * 60 ?
533 			24 * 60 * 60 * hz : tvtohz(&atv);
534                 getmicrouptime(&rtv);
535                 timevaladd(&atv, &rtv);
536         } else {
537                 atv.tv_sec = 0;
538 		timeout = 0;
539 	}
540 	goto start;
541 
542 retry:
543 	if (atv.tv_sec) {
544 		getmicrouptime(&rtv);
545 		if (timevalcmp(&rtv, &atv, >=))
546 			goto done;
547 		ttv = atv;
548 		timevalsub(&ttv, &rtv);
549 		timeout = ttv.tv_sec > 24 * 60 * 60 ?
550 			24 * 60 * 60 * hz : tvtohz(&ttv);
551 	}
552 
553 start:
554 	kevp = kq->kq_kev;
555 	s = splhigh();
556 	if (kq->kq_count == 0) {
557 		kq->kq_state |= KQ_SLEEP;
558 		error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout);
559 		splx(s);
560 		if (error == 0)
561 			goto retry;
562 		/* don't restart after signals... */
563 		if (error == ERESTART)
564 			error = EINTR;
565 		else if (error == EWOULDBLOCK)
566 			error = 0;
567 		goto done;
568 	}
569 
570 	TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
571 	while (count) {
572 		kn = TAILQ_FIRST(&kq->kq_head);
573 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
574 		if (kn == &marker) {
575 			splx(s);
576 			if (count == maxevents)
577 				goto retry;
578 			goto done;
579 		}
580 		if (kn->kn_status & KN_DISABLED) {
581 			kn->kn_status &= ~KN_QUEUED;
582 			kq->kq_count--;
583 			continue;
584 		}
585 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
586 		    kn->kn_fop->f_event(kn, 0) == 0) {
587 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
588 			kq->kq_count--;
589 			continue;
590 		}
591 		*kevp = kn->kn_kevent;
592 		kevp++;
593 		nkev++;
594 		if (kn->kn_flags & EV_ONESHOT) {
595 			kn->kn_status &= ~KN_QUEUED;
596 			kq->kq_count--;
597 			splx(s);
598 			kn->kn_fop->f_detach(kn);
599 			knote_drop(kn, p);
600 			s = splhigh();
601 		} else if (kn->kn_flags & EV_CLEAR) {
602 			kn->kn_data = 0;
603 			kn->kn_fflags = 0;
604 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
605 			kq->kq_count--;
606 		} else {
607 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
608 		}
609 		count--;
610 		if (nkev == KQ_NEVENTS) {
611 			splx(s);
612 			error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
613 			    sizeof(struct kevent) * nkev);
614 			ulistp += nkev;
615 			nkev = 0;
616 			kevp = kq->kq_kev;
617 			s = splhigh();
618 			if (error)
619 				break;
620 		}
621 	}
622 	TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
623 	splx(s);
624 done:
625 	if (nkev != 0)
626 		error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
627 		    sizeof(struct kevent) * nkev);
628         p->p_retval[0] = maxevents - count;
629 	return (error);
630 }
631 
632 /*
633  * XXX
634  * This could be expanded to call kqueue_scan, if desired.
635  */
636 /*ARGSUSED*/
637 static int
638 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred,
639 	int flags, struct proc *p)
640 {
641 	return (ENXIO);
642 }
643 
644 /*ARGSUSED*/
645 static int
646 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred,
647 	 int flags, struct proc *p)
648 {
649 	return (ENXIO);
650 }
651 
652 /*ARGSUSED*/
653 static int
654 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
655 {
656 	return (ENOTTY);
657 }
658 
659 /*ARGSUSED*/
660 static int
661 kqueue_poll(struct file *fp, int events, struct ucred *cred, struct proc *p)
662 {
663 	struct kqueue *kq = (struct kqueue *)fp->f_data;
664 	int revents = 0;
665 	int s = splnet();
666 
667         if (events & (POLLIN | POLLRDNORM)) {
668                 if (kq->kq_count) {
669                         revents |= events & (POLLIN | POLLRDNORM);
670 		} else {
671                         selrecord(p, &kq->kq_sel);
672 			kq->kq_state |= KQ_SEL;
673 		}
674 	}
675 	splx(s);
676 	return (revents);
677 }
678 
679 /*ARGSUSED*/
680 static int
681 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
682 {
683 	struct kqueue *kq = (struct kqueue *)fp->f_data;
684 
685 	bzero((void *)st, sizeof(*st));
686 	st->st_size = kq->kq_count;
687 	st->st_blksize = sizeof(struct kevent);
688 	st->st_mode = S_IFIFO;
689 	return (0);
690 }
691 
692 /*ARGSUSED*/
693 static int
694 kqueue_close(struct file *fp, struct proc *p)
695 {
696 	struct kqueue *kq = (struct kqueue *)fp->f_data;
697 	struct filedesc *fdp = p->p_fd;
698 	struct knote **knp, *kn, *kn0;
699 	int i;
700 
701 	for (i = 0; i < fdp->fd_knlistsize; i++) {
702 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
703 		kn = *knp;
704 		while (kn != NULL) {
705 			kn0 = SLIST_NEXT(kn, kn_link);
706 			if (kq == kn->kn_kq) {
707 				kn->kn_fop->f_detach(kn);
708 				fdrop(kn->kn_fp, p);
709 				knote_free(kn);
710 				*knp = kn0;
711 			} else {
712 				knp = &SLIST_NEXT(kn, kn_link);
713 			}
714 			kn = kn0;
715 		}
716 	}
717 	if (fdp->fd_knhashmask != 0) {
718 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
719 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
720 			kn = *knp;
721 			while (kn != NULL) {
722 				kn0 = SLIST_NEXT(kn, kn_link);
723 				if (kq == kn->kn_kq) {
724 					kn->kn_fop->f_detach(kn);
725 		/* XXX non-fd release of kn->kn_ptr */
726 					knote_free(kn);
727 					*knp = kn0;
728 				} else {
729 					knp = &SLIST_NEXT(kn, kn_link);
730 				}
731 				kn = kn0;
732 			}
733 		}
734 	}
735 	free(kq, M_TEMP);
736 	fp->f_data = NULL;
737 
738 	return (0);
739 }
740 
741 static void
742 kqueue_wakeup(struct kqueue *kq)
743 {
744 
745 	if (kq->kq_state & KQ_SLEEP) {
746 		kq->kq_state &= ~KQ_SLEEP;
747 		wakeup(kq);
748 	}
749 	if (kq->kq_state & KQ_SEL) {
750 		kq->kq_state &= ~KQ_SEL;
751 		selwakeup(&kq->kq_sel);
752 	}
753 	KNOTE(&kq->kq_sel.si_note, 0);
754 }
755 
756 /*
757  * walk down a list of knotes, activating them if their event has triggered.
758  */
759 void
760 knote(struct klist *list, long hint)
761 {
762 	struct knote *kn;
763 
764 	SLIST_FOREACH(kn, list, kn_selnext)
765 		if (kn->kn_fop->f_event(kn, hint))
766 			KNOTE_ACTIVATE(kn);
767 }
768 
769 /*
770  * remove all knotes from a specified klist
771  */
772 void
773 knote_remove(struct proc *p, struct klist *list)
774 {
775 	struct knote *kn;
776 
777 	while ((kn = SLIST_FIRST(list)) != NULL) {
778 		kn->kn_fop->f_detach(kn);
779 		knote_drop(kn, p);
780 	}
781 }
782 
783 /*
784  * remove all knotes referencing a specified fd
785  */
786 void
787 knote_fdclose(struct proc *p, int fd)
788 {
789 	struct filedesc *fdp = p->p_fd;
790 	struct klist *list = &fdp->fd_knlist[fd];
791 
792 	knote_remove(p, list);
793 }
794 
795 static void
796 knote_attach(struct knote *kn, struct filedesc *fdp)
797 {
798 	struct klist *list;
799 	int size;
800 
801 	if (! kn->kn_fop->f_isfd) {
802 		if (fdp->fd_knhashmask == 0)
803 			fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP,
804 			    &fdp->fd_knhashmask);
805 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
806 		goto done;
807 	}
808 
809 	if (fdp->fd_knlistsize <= kn->kn_id) {
810 		size = fdp->fd_knlistsize;
811 		while (size <= kn->kn_id)
812 			size += KQEXTENT;
813 		MALLOC(list, struct klist *,
814 		    size * sizeof(struct klist *), M_TEMP, M_WAITOK);
815 		bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
816 		    fdp->fd_knlistsize * sizeof(struct klist *));
817 		bzero((caddr_t)list +
818 		    fdp->fd_knlistsize * sizeof(struct klist *),
819 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
820 		if (fdp->fd_knlist != NULL)
821 			FREE(fdp->fd_knlist, M_TEMP);
822 		fdp->fd_knlistsize = size;
823 		fdp->fd_knlist = list;
824 	}
825 	list = &fdp->fd_knlist[kn->kn_id];
826 done:
827 	SLIST_INSERT_HEAD(list, kn, kn_link);
828 	kn->kn_status = 0;
829 }
830 
831 /*
832  * should be called at spl == 0, since we don't want to hold spl
833  * while calling fdrop and free.
834  */
835 static void
836 knote_drop(struct knote *kn, struct proc *p)
837 {
838         struct filedesc *fdp = p->p_fd;
839 	struct klist *list;
840 
841 	if (kn->kn_fop->f_isfd)
842 		list = &fdp->fd_knlist[kn->kn_id];
843 	else
844 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
845 
846 	SLIST_REMOVE(list, kn, knote, kn_link);
847 	if (kn->kn_status & KN_QUEUED)
848 		knote_dequeue(kn);
849 	if (kn->kn_fop->f_isfd)
850 		fdrop(kn->kn_fp, p);
851 	knote_free(kn);
852 }
853 
854 
855 static void
856 knote_enqueue(struct knote *kn)
857 {
858 	struct kqueue *kq = kn->kn_kq;
859 	int s = splhigh();
860 
861 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
862 
863 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
864 	kn->kn_status |= KN_QUEUED;
865 	kq->kq_count++;
866 	splx(s);
867 	kqueue_wakeup(kq);
868 }
869 
870 static void
871 knote_dequeue(struct knote *kn)
872 {
873 	struct kqueue *kq = kn->kn_kq;
874 	int s = splhigh();
875 
876 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
877 
878 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
879 	kn->kn_status &= ~KN_QUEUED;
880 	kq->kq_count--;
881 	splx(s);
882 }
883 
884 static void
885 knote_init(void)
886 {
887 	knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
888 }
889 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
890 
891 static struct knote *
892 knote_alloc(void)
893 {
894 	return ((struct knote *)zalloc(knote_zone));
895 }
896 
897 static void
898 knote_free(struct knote *kn)
899 {
900 	zfree(knote_zone, kn);
901 }
902