xref: /dflybsd-src/sys/kern/kern_event.c (revision 6693db176654a0f25095ec64d0a74d58dcf0e47e)
1 /*-
2  * Copyright (c) 1999,2000,2001 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: src/sys/kern/kern_event.c,v 1.2.2.10 2004/04/04 07:03:14 cperciva Exp $
27  * $DragonFly: src/sys/kern/kern_event.c,v 1.33 2007/02/03 17:05:57 corecode Exp $
28  */
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/proc.h>
34 #include <sys/malloc.h>
35 #include <sys/unistd.h>
36 #include <sys/file.h>
37 #include <sys/lock.h>
38 #include <sys/fcntl.h>
39 #include <sys/select.h>
40 #include <sys/queue.h>
41 #include <sys/event.h>
42 #include <sys/eventvar.h>
43 #include <sys/poll.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/stat.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysproto.h>
50 #include <sys/uio.h>
51 #include <sys/signalvar.h>
52 #include <sys/filio.h>
53 
54 #include <sys/thread2.h>
55 #include <sys/file2.h>
56 #include <sys/mplock2.h>
57 
58 #include <vm/vm_zone.h>
59 
60 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
61 
62 static int	kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count,
63 		    struct timespec *tsp, int *errorp);
64 static int 	kqueue_read(struct file *fp, struct uio *uio,
65 		    struct ucred *cred, int flags);
66 static int	kqueue_write(struct file *fp, struct uio *uio,
67 		    struct ucred *cred, int flags);
68 static int	kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
69 		    struct ucred *cred, struct sysmsg *msg);
70 static int 	kqueue_poll(struct file *fp, int events, struct ucred *cred);
71 static int 	kqueue_kqfilter(struct file *fp, struct knote *kn);
72 static int 	kqueue_stat(struct file *fp, struct stat *st,
73 		    struct ucred *cred);
74 static int 	kqueue_close(struct file *fp);
75 static void 	kqueue_wakeup(struct kqueue *kq);
76 
77 /*
78  * MPSAFE
79  */
80 static struct fileops kqueueops = {
81 	.fo_read = kqueue_read,
82 	.fo_write = kqueue_write,
83 	.fo_ioctl = kqueue_ioctl,
84 	.fo_poll = kqueue_poll,
85 	.fo_kqfilter = kqueue_kqfilter,
86 	.fo_stat = kqueue_stat,
87 	.fo_close = kqueue_close,
88 	.fo_shutdown = nofo_shutdown
89 };
90 
91 static void 	knote_attach(struct knote *kn);
92 static void 	knote_drop(struct knote *kn);
93 static void 	knote_enqueue(struct knote *kn);
94 static void 	knote_dequeue(struct knote *kn);
95 static void 	knote_init(void);
96 static struct 	knote *knote_alloc(void);
97 static void 	knote_free(struct knote *kn);
98 
99 static void	filt_kqdetach(struct knote *kn);
100 static int	filt_kqueue(struct knote *kn, long hint);
101 static int	filt_procattach(struct knote *kn);
102 static void	filt_procdetach(struct knote *kn);
103 static int	filt_proc(struct knote *kn, long hint);
104 static int	filt_fileattach(struct knote *kn);
105 static void	filt_timerexpire(void *knx);
106 static int	filt_timerattach(struct knote *kn);
107 static void	filt_timerdetach(struct knote *kn);
108 static int	filt_timer(struct knote *kn, long hint);
109 
110 static struct filterops file_filtops =
111 	{ 1, filt_fileattach, NULL, NULL };
112 static struct filterops kqread_filtops =
113 	{ 1, NULL, filt_kqdetach, filt_kqueue };
114 static struct filterops proc_filtops =
115 	{ 0, filt_procattach, filt_procdetach, filt_proc };
116 static struct filterops timer_filtops =
117 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
118 
119 static vm_zone_t	knote_zone;
120 static int 		kq_ncallouts = 0;
121 static int 		kq_calloutmax = (4 * 1024);
122 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
123     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
124 
125 #define KNOTE_ACTIVATE(kn) do { 					\
126 	kn->kn_status |= KN_ACTIVE;					\
127 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
128 		knote_enqueue(kn);					\
129 } while(0)
130 
131 #define	KN_HASHSIZE		64		/* XXX should be tunable */
132 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
133 
134 extern struct filterops aio_filtops;
135 extern struct filterops sig_filtops;
136 
137 /*
138  * Table for for all system-defined filters.
139  */
140 static struct filterops *sysfilt_ops[] = {
141 	&file_filtops,			/* EVFILT_READ */
142 	&file_filtops,			/* EVFILT_WRITE */
143 	&aio_filtops,			/* EVFILT_AIO */
144 	&file_filtops,			/* EVFILT_VNODE */
145 	&proc_filtops,			/* EVFILT_PROC */
146 	&sig_filtops,			/* EVFILT_SIGNAL */
147 	&timer_filtops,			/* EVFILT_TIMER */
148 };
149 
150 static int
151 filt_fileattach(struct knote *kn)
152 {
153 	return (fo_kqfilter(kn->kn_fp, kn));
154 }
155 
156 /*
157  * MPALMOSTSAFE - acquires mplock
158  */
159 static int
160 kqueue_kqfilter(struct file *fp, struct knote *kn)
161 {
162 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
163 
164 	get_mplock();
165 	if (kn->kn_filter != EVFILT_READ) {
166 		rel_mplock();
167 		return (1);
168 	}
169 
170 	kn->kn_fop = &kqread_filtops;
171 	SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext);
172 	rel_mplock();
173 	return (0);
174 }
175 
176 static void
177 filt_kqdetach(struct knote *kn)
178 {
179 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
180 
181 	SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext);
182 }
183 
184 /*ARGSUSED*/
185 static int
186 filt_kqueue(struct knote *kn, long hint)
187 {
188 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
189 
190 	kn->kn_data = kq->kq_count;
191 	return (kn->kn_data > 0);
192 }
193 
194 static int
195 filt_procattach(struct knote *kn)
196 {
197 	struct proc *p;
198 	int immediate;
199 
200 	immediate = 0;
201 	p = pfind(kn->kn_id);
202 	if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
203 		p = zpfind(kn->kn_id);
204 		immediate = 1;
205 	}
206 	if (p == NULL)
207 		return (ESRCH);
208 	if (!PRISON_CHECK(curthread->td_ucred, p->p_ucred))
209 		return (EACCES);
210 
211 	kn->kn_ptr.p_proc = p;
212 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
213 
214 	/*
215 	 * internal flag indicating registration done by kernel
216 	 */
217 	if (kn->kn_flags & EV_FLAG1) {
218 		kn->kn_data = kn->kn_sdata;		/* ppid */
219 		kn->kn_fflags = NOTE_CHILD;
220 		kn->kn_flags &= ~EV_FLAG1;
221 	}
222 
223 	/* XXX lock the proc here while adding to the list? */
224 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
225 
226 	/*
227 	 * Immediately activate any exit notes if the target process is a
228 	 * zombie.  This is necessary to handle the case where the target
229 	 * process, e.g. a child, dies before the kevent is registered.
230 	 */
231 	if (immediate && filt_proc(kn, NOTE_EXIT))
232 		KNOTE_ACTIVATE(kn);
233 
234 	return (0);
235 }
236 
237 /*
238  * The knote may be attached to a different process, which may exit,
239  * leaving nothing for the knote to be attached to.  So when the process
240  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
241  * it will be deleted when read out.  However, as part of the knote deletion,
242  * this routine is called, so a check is needed to avoid actually performing
243  * a detach, because the original process does not exist any more.
244  */
245 static void
246 filt_procdetach(struct knote *kn)
247 {
248 	struct proc *p;
249 
250 	if (kn->kn_status & KN_DETACHED)
251 		return;
252 	/* XXX locking?  this might modify another process. */
253 	p = kn->kn_ptr.p_proc;
254 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
255 }
256 
257 static int
258 filt_proc(struct knote *kn, long hint)
259 {
260 	u_int event;
261 
262 	/*
263 	 * mask off extra data
264 	 */
265 	event = (u_int)hint & NOTE_PCTRLMASK;
266 
267 	/*
268 	 * if the user is interested in this event, record it.
269 	 */
270 	if (kn->kn_sfflags & event)
271 		kn->kn_fflags |= event;
272 
273 	/*
274 	 * Process is gone, so flag the event as finished.  Detach the
275 	 * knote from the process now because the process will be poof,
276 	 * gone later on.
277 	 */
278 	if (event == NOTE_EXIT) {
279 		struct proc *p = kn->kn_ptr.p_proc;
280 		if ((kn->kn_status & KN_DETACHED) == 0) {
281 			SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
282 			kn->kn_status |= KN_DETACHED;
283 			kn->kn_data = p->p_xstat;
284 			kn->kn_ptr.p_proc = NULL;
285 		}
286 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
287 		return (1);
288 	}
289 
290 	/*
291 	 * process forked, and user wants to track the new process,
292 	 * so attach a new knote to it, and immediately report an
293 	 * event with the parent's pid.
294 	 */
295 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
296 		struct kevent kev;
297 		int error;
298 
299 		/*
300 		 * register knote with new process.
301 		 */
302 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
303 		kev.filter = kn->kn_filter;
304 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
305 		kev.fflags = kn->kn_sfflags;
306 		kev.data = kn->kn_id;			/* parent */
307 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
308 		error = kqueue_register(kn->kn_kq, &kev);
309 		if (error)
310 			kn->kn_fflags |= NOTE_TRACKERR;
311 	}
312 
313 	return (kn->kn_fflags != 0);
314 }
315 
316 static void
317 filt_timerexpire(void *knx)
318 {
319 	struct knote *kn = knx;
320 	struct callout *calloutp;
321 	struct timeval tv;
322 	int tticks;
323 
324 	kn->kn_data++;
325 	KNOTE_ACTIVATE(kn);
326 
327 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
328 		tv.tv_sec = kn->kn_sdata / 1000;
329 		tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
330 		tticks = tvtohz_high(&tv);
331 		calloutp = (struct callout *)kn->kn_hook;
332 		callout_reset(calloutp, tticks, filt_timerexpire, kn);
333 	}
334 }
335 
336 /*
337  * data contains amount of time to sleep, in milliseconds
338  */
339 static int
340 filt_timerattach(struct knote *kn)
341 {
342 	struct callout *calloutp;
343 	struct timeval tv;
344 	int tticks;
345 
346 	if (kq_ncallouts >= kq_calloutmax)
347 		return (ENOMEM);
348 	kq_ncallouts++;
349 
350 	tv.tv_sec = kn->kn_sdata / 1000;
351 	tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
352 	tticks = tvtohz_high(&tv);
353 
354 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
355 	MALLOC(calloutp, struct callout *, sizeof(*calloutp),
356 	    M_KQUEUE, M_WAITOK);
357 	callout_init(calloutp);
358 	kn->kn_hook = (caddr_t)calloutp;
359 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
360 
361 	return (0);
362 }
363 
364 static void
365 filt_timerdetach(struct knote *kn)
366 {
367 	struct callout *calloutp;
368 
369 	calloutp = (struct callout *)kn->kn_hook;
370 	callout_stop(calloutp);
371 	FREE(calloutp, M_KQUEUE);
372 	kq_ncallouts--;
373 }
374 
375 static int
376 filt_timer(struct knote *kn, long hint)
377 {
378 
379 	return (kn->kn_data != 0);
380 }
381 
382 /*
383  * Initialize a kqueue.
384  *
385  * NOTE: The lwp/proc code initializes a kqueue for select/poll ops.
386  *
387  * MPSAFE
388  */
389 void
390 kqueue_init(struct kqueue *kq, struct filedesc *fdp)
391 {
392 	TAILQ_INIT(&kq->kq_knpend);
393 	TAILQ_INIT(&kq->kq_knlist);
394 	kq->kq_fdp = fdp;
395 }
396 
397 /*
398  * Terminate a kqueue.  Freeing the actual kq itself is left up to the
399  * caller (it might be embedded in a lwp so we don't do it here).
400  */
401 void
402 kqueue_terminate(struct kqueue *kq)
403 {
404 	struct knote *kn;
405 	struct klist *list;
406 	int hv;
407 
408 	while ((kn = TAILQ_FIRST(&kq->kq_knlist)) != NULL) {
409 		kn->kn_fop->f_detach(kn);
410 		if (kn->kn_fop->f_isfd) {
411 			list = &kn->kn_fp->f_klist;
412 			SLIST_REMOVE(list, kn, knote, kn_link);
413 			fdrop(kn->kn_fp);
414 			kn->kn_fp = NULL;
415 		} else {
416 			hv = KN_HASH(kn->kn_id, kq->kq_knhashmask);
417 			list = &kq->kq_knhash[hv];
418 			SLIST_REMOVE(list, kn, knote, kn_link);
419 		}
420 		TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink);
421 		if (kn->kn_status & KN_QUEUED)
422 			knote_dequeue(kn);
423 		knote_free(kn);
424 	}
425 
426 	if (kq->kq_knhash) {
427 		kfree(kq->kq_knhash, M_KQUEUE);
428 		kq->kq_knhash = NULL;
429 		kq->kq_knhashmask = 0;
430 	}
431 }
432 
433 /*
434  * MPSAFE
435  */
436 int
437 sys_kqueue(struct kqueue_args *uap)
438 {
439 	struct thread *td = curthread;
440 	struct kqueue *kq;
441 	struct file *fp;
442 	int fd, error;
443 
444 	error = falloc(td->td_lwp, &fp, &fd);
445 	if (error)
446 		return (error);
447 	fp->f_flag = FREAD | FWRITE;
448 	fp->f_type = DTYPE_KQUEUE;
449 	fp->f_ops = &kqueueops;
450 
451 	kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO);
452 	kqueue_init(kq, td->td_proc->p_fd);
453 	fp->f_data = kq;
454 
455 	fsetfd(kq->kq_fdp, fp, fd);
456 	uap->sysmsg_result = fd;
457 	fdrop(fp);
458 	return (error);
459 }
460 
461 /*
462  * MPALMOSTSAFE
463  */
464 int
465 sys_kevent(struct kevent_args *uap)
466 {
467 	struct thread *td = curthread;
468 	struct proc *p = td->td_proc;
469 	struct kevent *kevp;
470 	struct kqueue *kq;
471 	struct file *fp = NULL;
472 	struct timespec ts;
473 	struct timespec *tsp;
474 	int i, n, total, nerrors, error;
475 	struct kevent kev[KQ_NEVENTS];
476 
477 	fp = holdfp(p->p_fd, uap->fd, -1);
478 	if (fp == NULL)
479 		return (EBADF);
480 	if (fp->f_type != DTYPE_KQUEUE) {
481 		fdrop(fp);
482 		return (EBADF);
483 	}
484 
485 	if (uap->timeout) {
486 		error = copyin(uap->timeout, &ts, sizeof(ts));
487 		if (error)
488 			goto done;
489 		tsp = &ts;
490 	} else {
491 		tsp = NULL;
492 	}
493 
494 	kq = (struct kqueue *)fp->f_data;
495 	nerrors = 0;
496 
497 	get_mplock();
498 	while (uap->nchanges > 0) {
499 		n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges;
500 		error = copyin(uap->changelist, kev, n * sizeof(struct kevent));
501 		if (error)
502 			goto done;
503 		for (i = 0; i < n; i++) {
504 			kevp = &kev[i];
505 			kevp->flags &= ~EV_SYSFLAGS;
506 			error = kqueue_register(kq, kevp);
507 			if (error) {
508 				if (uap->nevents != 0) {
509 					kevp->flags = EV_ERROR;
510 					kevp->data = error;
511 					copyout(kevp, uap->eventlist,
512 						sizeof(*kevp));
513 					uap->eventlist++;
514 					uap->nevents--;
515 					nerrors++;
516 				} else {
517 					goto done;
518 				}
519 			}
520 		}
521 		uap->nchanges -= n;
522 		uap->changelist += n;
523 	}
524 	if (nerrors) {
525         	uap->sysmsg_result = nerrors;
526 		error = 0;
527 		goto done;
528 	}
529 
530 	/*
531 	 * Acquire/wait for events - setup timeout
532 	 */
533 	if (tsp != NULL) {
534 		struct timespec ats;
535 
536 		if (tsp->tv_sec || tsp->tv_nsec) {
537 			nanouptime(&ats);
538 			timespecadd(tsp, &ats);		/* tsp = target time */
539 		}
540 	}
541 
542 	/*
543 	 * Loop as required.
544 	 *
545 	 * Collect as many events as we can.  The timeout on successive
546 	 * loops is disabled (kqueue_scan() becomes non-blocking).
547 	 */
548 	total = 0;
549 	error = 0;
550 	while ((n = uap->nevents - total) > 0) {
551 		if (n > KQ_NEVENTS)
552 			n = KQ_NEVENTS;
553 		i = kqueue_scan(kq, kev, n, tsp, &error);
554 		if (i == 0)
555 			break;
556 		error = copyout(kev, uap->eventlist + total,
557 				(size_t)i * sizeof(struct kevent));
558 		total += i;
559 		if (error || i != n)
560 			break;
561 		tsp = &ts;		/* successive loops non-blocking */
562 		tsp->tv_sec = 0;
563 		tsp->tv_nsec = 0;
564 	}
565 	uap->sysmsg_result = total;
566 done:
567 	rel_mplock();
568 	if (fp != NULL)
569 		fdrop(fp);
570 	return (error);
571 }
572 
573 int
574 kqueue_register(struct kqueue *kq, struct kevent *kev)
575 {
576 	struct filedesc *fdp = kq->kq_fdp;
577 	struct filterops *fops;
578 	struct file *fp = NULL;
579 	struct knote *kn = NULL;
580 	int error = 0;
581 
582 	if (kev->filter < 0) {
583 		if (kev->filter + EVFILT_SYSCOUNT < 0)
584 			return (EINVAL);
585 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
586 	} else {
587 		/*
588 		 * XXX
589 		 * filter attach routine is responsible for insuring that
590 		 * the identifier can be attached to it.
591 		 */
592 		kprintf("unknown filter: %d\n", kev->filter);
593 		return (EINVAL);
594 	}
595 
596 	if (fops->f_isfd) {
597 		/* validate descriptor */
598 		fp = holdfp(fdp, kev->ident, -1);
599 		if (fp == NULL)
600 			return (EBADF);
601 
602 		SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
603 			if (kn->kn_kq == kq &&
604 			    kn->kn_filter == kev->filter &&
605 			    kn->kn_id == kev->ident) {
606 				break;
607 			}
608 		}
609 	} else {
610 		if (kq->kq_knhashmask) {
611 			struct klist *list;
612 
613 			list = &kq->kq_knhash[
614 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
615 			SLIST_FOREACH(kn, list, kn_link) {
616 				if (kn->kn_id == kev->ident &&
617 				    kn->kn_filter == kev->filter)
618 					break;
619 			}
620 		}
621 	}
622 
623 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
624 		error = ENOENT;
625 		goto done;
626 	}
627 
628 	/*
629 	 * kn now contains the matching knote, or NULL if no match
630 	 */
631 	if (kev->flags & EV_ADD) {
632 		if (kn == NULL) {
633 			kn = knote_alloc();
634 			if (kn == NULL) {
635 				error = ENOMEM;
636 				goto done;
637 			}
638 			kn->kn_fp = fp;
639 			kn->kn_kq = kq;
640 			kn->kn_fop = fops;
641 
642 			/*
643 			 * apply reference count to knote structure, and
644 			 * do not release it at the end of this routine.
645 			 */
646 			fp = NULL;
647 
648 			kn->kn_sfflags = kev->fflags;
649 			kn->kn_sdata = kev->data;
650 			kev->fflags = 0;
651 			kev->data = 0;
652 			kn->kn_kevent = *kev;
653 
654 			knote_attach(kn);
655 			if ((error = fops->f_attach(kn)) != 0) {
656 				knote_drop(kn);
657 				goto done;
658 			}
659 		} else {
660 			/*
661 			 * The user may change some filter values after the
662 			 * initial EV_ADD, but doing so will not reset any
663 			 * filter which have already been triggered.
664 			 */
665 			kn->kn_sfflags = kev->fflags;
666 			kn->kn_sdata = kev->data;
667 			kn->kn_kevent.udata = kev->udata;
668 		}
669 
670 		crit_enter();
671 		if (kn->kn_fop->f_event(kn, 0))
672 			KNOTE_ACTIVATE(kn);
673 		crit_exit();
674 	} else if (kev->flags & EV_DELETE) {
675 		kn->kn_fop->f_detach(kn);
676 		knote_drop(kn);
677 		goto done;
678 	}
679 
680 	if ((kev->flags & EV_DISABLE) &&
681 	    ((kn->kn_status & KN_DISABLED) == 0)) {
682 		crit_enter();
683 		kn->kn_status |= KN_DISABLED;
684 		crit_exit();
685 	}
686 
687 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
688 		crit_enter();
689 		kn->kn_status &= ~KN_DISABLED;
690 		if ((kn->kn_status & KN_ACTIVE) &&
691 		    ((kn->kn_status & KN_QUEUED) == 0))
692 			knote_enqueue(kn);
693 		crit_exit();
694 	}
695 
696 done:
697 	if (fp != NULL)
698 		fdrop(fp);
699 	return (error);
700 }
701 
702 /*
703  * Scan the kqueue, blocking if necessary until the target time is reached.
704  * If tsp is NULL we block indefinitely.  If tsp->ts_secs/nsecs are both
705  * 0 we do not block at all.
706  */
707 static int
708 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count,
709 	    struct timespec *tsp, int *errorp)
710 {
711 	struct knote *kn, marker;
712 	int total;
713 
714 	total = 0;
715 again:
716 	crit_enter();
717 	if (kq->kq_count == 0) {
718 		if (tsp == NULL) {
719 			kq->kq_state |= KQ_SLEEP;
720 			*errorp = tsleep(kq, PCATCH, "kqread", 0);
721 		} else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
722 			*errorp = EWOULDBLOCK;
723 		} else {
724 			struct timespec ats;
725 			struct timespec atx = *tsp;
726 			int timeout;
727 
728 			nanouptime(&ats);
729 			timespecsub(&atx, &ats);
730 			if (ats.tv_sec < 0) {
731 				*errorp = EWOULDBLOCK;
732 			} else {
733 				timeout = atx.tv_sec > 24 * 60 * 60 ?
734 					24 * 60 * 60 * hz : tstohz_high(&atx);
735 				kq->kq_state |= KQ_SLEEP;
736 				*errorp = tsleep(kq, PCATCH, "kqread", timeout);
737 			}
738 		}
739 		crit_exit();
740 		if (*errorp == 0)
741 			goto again;
742 		/* don't restart after signals... */
743 		if (*errorp == ERESTART)
744 			*errorp = EINTR;
745 		else if (*errorp == EWOULDBLOCK)
746 			*errorp = 0;
747 		goto done;
748 	}
749 
750 	/*
751 	 * Collect events.  Continuous mode events may get recycled
752 	 * past the marker so we stop when we hit it unless no events
753 	 * have been collected.
754 	 */
755 	TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
756 	while (count) {
757 		kn = TAILQ_FIRST(&kq->kq_knpend);
758 		if (kn == &marker)
759 			break;
760 		TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
761 		if (kn->kn_status & KN_DISABLED) {
762 			kn->kn_status &= ~KN_QUEUED;
763 			kq->kq_count--;
764 			continue;
765 		}
766 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
767 		    kn->kn_fop->f_event(kn, 0) == 0) {
768 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
769 			kq->kq_count--;
770 			continue;
771 		}
772 		*kevp++ = kn->kn_kevent;
773 		++total;
774 		--count;
775 
776 		/*
777 		 * Post-event action on the note
778 		 */
779 		if (kn->kn_flags & EV_ONESHOT) {
780 			kn->kn_status &= ~KN_QUEUED;
781 			kq->kq_count--;
782 			crit_exit();
783 			kn->kn_fop->f_detach(kn);
784 			knote_drop(kn);
785 			crit_enter();
786 		} else if (kn->kn_flags & EV_CLEAR) {
787 			kn->kn_data = 0;
788 			kn->kn_fflags = 0;
789 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
790 			kq->kq_count--;
791 		} else {
792 			TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
793 		}
794 	}
795 	TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
796 	crit_exit();
797 	if (total == 0)
798 		goto again;
799 done:
800 	return (total);
801 }
802 
803 /*
804  * XXX
805  * This could be expanded to call kqueue_scan, if desired.
806  *
807  * MPSAFE
808  */
809 static int
810 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
811 {
812 	return (ENXIO);
813 }
814 
815 /*
816  * MPSAFE
817  */
818 static int
819 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
820 {
821 	return (ENXIO);
822 }
823 
824 /*
825  * MPALMOSTSAFE
826  */
827 static int
828 kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
829 	     struct ucred *cred, struct sysmsg *msg)
830 {
831 	struct kqueue *kq;
832 	int error;
833 
834 	get_mplock();
835 	kq = (struct kqueue *)fp->f_data;
836 
837 	switch(com) {
838 	case FIOASYNC:
839 		if (*(int *)data)
840 			kq->kq_state |= KQ_ASYNC;
841 		else
842 			kq->kq_state &= ~KQ_ASYNC;
843 		error = 0;
844 		break;
845 	case FIOSETOWN:
846 		error = fsetown(*(int *)data, &kq->kq_sigio);
847 		break;
848 	default:
849 		error = ENOTTY;
850 		break;
851 	}
852 	rel_mplock();
853 	return (error);
854 }
855 
856 /*
857  * MPALMOSTSAFE - acquires mplock
858  */
859 static int
860 kqueue_poll(struct file *fp, int events, struct ucred *cred)
861 {
862 	struct kqueue *kq = (struct kqueue *)fp->f_data;
863 	int revents = 0;
864 
865 	get_mplock();
866 	crit_enter();
867         if (events & (POLLIN | POLLRDNORM)) {
868                 if (kq->kq_count) {
869                         revents |= events & (POLLIN | POLLRDNORM);
870 		} else {
871                         selrecord(curthread, &kq->kq_sel);
872 			kq->kq_state |= KQ_SEL;
873 		}
874 	}
875 	crit_exit();
876 	rel_mplock();
877 	return (revents);
878 }
879 
880 /*
881  * MPSAFE
882  */
883 static int
884 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
885 {
886 	struct kqueue *kq = (struct kqueue *)fp->f_data;
887 
888 	bzero((void *)st, sizeof(*st));
889 	st->st_size = kq->kq_count;
890 	st->st_blksize = sizeof(struct kevent);
891 	st->st_mode = S_IFIFO;
892 	return (0);
893 }
894 
895 /*
896  * MPALMOSTSAFE - acquires mplock
897  */
898 static int
899 kqueue_close(struct file *fp)
900 {
901 	struct kqueue *kq = (struct kqueue *)fp->f_data;
902 
903 	get_mplock();
904 
905 	kqueue_terminate(kq);
906 
907 	fp->f_data = NULL;
908 	funsetown(kq->kq_sigio);
909 	rel_mplock();
910 
911 	kfree(kq, M_KQUEUE);
912 	return (0);
913 }
914 
915 static void
916 kqueue_wakeup(struct kqueue *kq)
917 {
918 	if (kq->kq_state & KQ_SLEEP) {
919 		kq->kq_state &= ~KQ_SLEEP;
920 		wakeup(kq);
921 	}
922 	if (kq->kq_state & KQ_SEL) {
923 		kq->kq_state &= ~KQ_SEL;
924 		selwakeup(&kq->kq_sel);
925 	}
926 	KNOTE(&kq->kq_sel.si_note, 0);
927 }
928 
929 /*
930  * walk down a list of knotes, activating them if their event has triggered.
931  */
932 void
933 knote(struct klist *list, long hint)
934 {
935 	struct knote *kn;
936 
937 	SLIST_FOREACH(kn, list, kn_selnext)
938 		if (kn->kn_fop->f_event(kn, hint))
939 			KNOTE_ACTIVATE(kn);
940 }
941 
942 /*
943  * remove all knotes from a specified klist
944  */
945 void
946 knote_remove(struct klist *list)
947 {
948 	struct knote *kn;
949 
950 	while ((kn = SLIST_FIRST(list)) != NULL) {
951 		kn->kn_fop->f_detach(kn);
952 		knote_drop(kn);
953 	}
954 }
955 
956 /*
957  * remove all knotes referencing a specified fd
958  */
959 void
960 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd)
961 {
962 	struct knote *kn;
963 
964 restart:
965 	SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
966 		if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) {
967 			kn->kn_fop->f_detach(kn);
968 			knote_drop(kn);
969 			goto restart;
970 		}
971 	}
972 }
973 
974 static void
975 knote_attach(struct knote *kn)
976 {
977 	struct klist *list;
978 	struct kqueue *kq = kn->kn_kq;
979 
980 	if (kn->kn_fop->f_isfd) {
981 		KKASSERT(kn->kn_fp);
982 		list = &kn->kn_fp->f_klist;
983 	} else {
984 		if (kq->kq_knhashmask == 0)
985 			kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
986 						 &kq->kq_knhashmask);
987 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
988 	}
989 	SLIST_INSERT_HEAD(list, kn, kn_link);
990 	TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink);
991 	kn->kn_status = 0;
992 }
993 
994 /*
995  * should be called outside of a critical section, since we don't want to
996  * hold a critical section while calling fdrop and free.
997  */
998 static void
999 knote_drop(struct knote *kn)
1000 {
1001 	struct kqueue *kq;
1002 	struct klist *list;
1003 
1004 	kq = kn->kn_kq;
1005 
1006 	if (kn->kn_fop->f_isfd)
1007 		list = &kn->kn_fp->f_klist;
1008 	else
1009 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1010 
1011 	SLIST_REMOVE(list, kn, knote, kn_link);
1012 	TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink);
1013 	if (kn->kn_status & KN_QUEUED)
1014 		knote_dequeue(kn);
1015 	if (kn->kn_fop->f_isfd)
1016 		fdrop(kn->kn_fp);
1017 	knote_free(kn);
1018 }
1019 
1020 
1021 static void
1022 knote_enqueue(struct knote *kn)
1023 {
1024 	struct kqueue *kq = kn->kn_kq;
1025 
1026 	crit_enter();
1027 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1028 
1029 	TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
1030 	kn->kn_status |= KN_QUEUED;
1031 	++kq->kq_count;
1032 
1033 	/*
1034 	 * Send SIGIO on request (typically set up as a mailbox signal)
1035 	 */
1036 	if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1037 		pgsigio(kq->kq_sigio, SIGIO, 0);
1038 	crit_exit();
1039 	kqueue_wakeup(kq);
1040 }
1041 
1042 static void
1043 knote_dequeue(struct knote *kn)
1044 {
1045 	struct kqueue *kq = kn->kn_kq;
1046 
1047 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1048 	crit_enter();
1049 
1050 	TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
1051 	kn->kn_status &= ~KN_QUEUED;
1052 	kq->kq_count--;
1053 	crit_exit();
1054 }
1055 
1056 static void
1057 knote_init(void)
1058 {
1059 	knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
1060 }
1061 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1062 
1063 static struct knote *
1064 knote_alloc(void)
1065 {
1066 	return ((struct knote *)zalloc(knote_zone));
1067 }
1068 
1069 static void
1070 knote_free(struct knote *kn)
1071 {
1072 	zfree(knote_zone, kn);
1073 }
1074