xref: /dflybsd-src/sys/kern/kern_event.c (revision e7302aa08274de307cd2c3345fc64c56dbe56e21)
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  * Copy 'count' items into the destination list pointed to by uap->eventlist.
463  */
464 static int
465 kevent_copyout(void *arg, struct kevent *kevp, int count)
466 {
467 	struct kevent_args *uap;
468 	int error;
469 
470 	uap = (struct kevent_args *)arg;
471 
472 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
473 	if (error == 0)
474 		uap->eventlist += count;
475 	return (error);
476 }
477 
478 /*
479  * Copy 'count' items from the list pointed to by uap->changelist.
480  */
481 static int
482 kevent_copyin(void *arg, struct kevent *kevp, int count)
483 {
484 	struct kevent_args *uap;
485 	int error;
486 
487 	uap = (struct kevent_args *)arg;
488 
489 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
490 	if (error == 0)
491 		uap->changelist += count;
492 	return (error);
493 }
494 
495 /*
496  * MPALMOSTSAFE
497  */
498 int
499 kern_kevent(int fd, int nchanges, int nevents, struct kevent_args *uap,
500     k_copyin_fn kevent_copyinfn, k_copyout_fn kevent_copyoutfn,
501     struct timespec *tsp_in)
502 {
503 	struct thread *td = curthread;
504 	struct proc *p = td->td_proc;
505 	struct kevent *kevp;
506 	struct kqueue *kq;
507 	struct file *fp = NULL;
508 	struct timespec ts;
509 	struct timespec *tsp;
510 	int i, n, total, nerrors, error;
511 	struct kevent kev[KQ_NEVENTS];
512 
513 	tsp = tsp_in;
514 
515 	fp = holdfp(p->p_fd, fd, -1);
516 	if (fp == NULL)
517 		return (EBADF);
518 	if (fp->f_type != DTYPE_KQUEUE) {
519 		fdrop(fp);
520 		return (EBADF);
521 	}
522 
523 	kq = (struct kqueue *)fp->f_data;
524 	nerrors = 0;
525 
526 	get_mplock();
527 	while (nchanges > 0) {
528 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
529 		error = kevent_copyinfn(uap, kev, n);
530 		if (error)
531 			goto done;
532 		for (i = 0; i < n; i++) {
533 			kevp = &kev[i];
534 			kevp->flags &= ~EV_SYSFLAGS;
535 			error = kqueue_register(kq, kevp);
536 			if (error) {
537 				if (nevents != 0) {
538 					kevp->flags = EV_ERROR;
539 					kevp->data = error;
540 					kevent_copyoutfn(uap, kevp, 1);
541 					nevents--;
542 					nerrors++;
543 				} else {
544 					goto done;
545 				}
546 			}
547 		}
548 		nchanges -= n;
549 	}
550 	if (nerrors) {
551         	uap->sysmsg_result = nerrors;
552 		error = 0;
553 		goto done;
554 	}
555 
556 	/*
557 	 * Acquire/wait for events - setup timeout
558 	 */
559 	if (tsp != NULL) {
560 		struct timespec ats;
561 
562 		if (tsp->tv_sec || tsp->tv_nsec) {
563 			nanouptime(&ats);
564 			timespecadd(tsp, &ats);		/* tsp = target time */
565 		}
566 	}
567 
568 	/*
569 	 * Loop as required.
570 	 *
571 	 * Collect as many events as we can.  The timeout on successive
572 	 * loops is disabled (kqueue_scan() becomes non-blocking).
573 	 */
574 	total = 0;
575 	error = 0;
576 	while ((n = nevents - total) > 0) {
577 		if (n > KQ_NEVENTS)
578 			n = KQ_NEVENTS;
579 		i = kqueue_scan(kq, kev, n, tsp, &error);
580 		if (i == 0)
581 			break;
582 		error = kevent_copyoutfn(uap, kev, i);
583 		total += i;
584 		if (error || i != n)
585 			break;
586 		tsp = &ts;		/* successive loops non-blocking */
587 		tsp->tv_sec = 0;
588 		tsp->tv_nsec = 0;
589 	}
590 	uap->sysmsg_result = total;
591 done:
592 	rel_mplock();
593 	if (fp != NULL)
594 		fdrop(fp);
595 	return (error);
596 }
597 
598 /*
599  * MPALMOSTSAFE
600  */
601 int
602 sys_kevent(struct kevent_args *uap)
603 {
604 	struct timespec ts, *tsp;
605 	int error;
606 
607 	if (uap->timeout) {
608 		error = copyin(uap->timeout, &ts, sizeof(ts));
609 		if (error)
610 			return (error);
611 		tsp = &ts;
612 	} else {
613 		tsp = NULL;
614 	}
615 
616 	error = kern_kevent(uap->fd, uap->nchanges, uap->nevents,
617 	    uap, kevent_copyin, kevent_copyout, tsp);
618 
619 	return (error);
620 }
621 
622 int
623 kqueue_register(struct kqueue *kq, struct kevent *kev)
624 {
625 	struct filedesc *fdp = kq->kq_fdp;
626 	struct filterops *fops;
627 	struct file *fp = NULL;
628 	struct knote *kn = NULL;
629 	int error = 0;
630 
631 	if (kev->filter < 0) {
632 		if (kev->filter + EVFILT_SYSCOUNT < 0)
633 			return (EINVAL);
634 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
635 	} else {
636 		/*
637 		 * XXX
638 		 * filter attach routine is responsible for insuring that
639 		 * the identifier can be attached to it.
640 		 */
641 		kprintf("unknown filter: %d\n", kev->filter);
642 		return (EINVAL);
643 	}
644 
645 	if (fops->f_isfd) {
646 		/* validate descriptor */
647 		fp = holdfp(fdp, kev->ident, -1);
648 		if (fp == NULL)
649 			return (EBADF);
650 
651 		SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
652 			if (kn->kn_kq == kq &&
653 			    kn->kn_filter == kev->filter &&
654 			    kn->kn_id == kev->ident) {
655 				break;
656 			}
657 		}
658 	} else {
659 		if (kq->kq_knhashmask) {
660 			struct klist *list;
661 
662 			list = &kq->kq_knhash[
663 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
664 			SLIST_FOREACH(kn, list, kn_link) {
665 				if (kn->kn_id == kev->ident &&
666 				    kn->kn_filter == kev->filter)
667 					break;
668 			}
669 		}
670 	}
671 
672 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
673 		error = ENOENT;
674 		goto done;
675 	}
676 
677 	/*
678 	 * kn now contains the matching knote, or NULL if no match
679 	 */
680 	if (kev->flags & EV_ADD) {
681 		if (kn == NULL) {
682 			kn = knote_alloc();
683 			if (kn == NULL) {
684 				error = ENOMEM;
685 				goto done;
686 			}
687 			kn->kn_fp = fp;
688 			kn->kn_kq = kq;
689 			kn->kn_fop = fops;
690 
691 			/*
692 			 * apply reference count to knote structure, and
693 			 * do not release it at the end of this routine.
694 			 */
695 			fp = NULL;
696 
697 			kn->kn_sfflags = kev->fflags;
698 			kn->kn_sdata = kev->data;
699 			kev->fflags = 0;
700 			kev->data = 0;
701 			kn->kn_kevent = *kev;
702 
703 			knote_attach(kn);
704 			if ((error = fops->f_attach(kn)) != 0) {
705 				knote_drop(kn);
706 				goto done;
707 			}
708 		} else {
709 			/*
710 			 * The user may change some filter values after the
711 			 * initial EV_ADD, but doing so will not reset any
712 			 * filter which have already been triggered.
713 			 */
714 			kn->kn_sfflags = kev->fflags;
715 			kn->kn_sdata = kev->data;
716 			kn->kn_kevent.udata = kev->udata;
717 		}
718 
719 		crit_enter();
720 		if (kn->kn_fop->f_event(kn, 0))
721 			KNOTE_ACTIVATE(kn);
722 		crit_exit();
723 	} else if (kev->flags & EV_DELETE) {
724 		kn->kn_fop->f_detach(kn);
725 		knote_drop(kn);
726 		goto done;
727 	}
728 
729 	if ((kev->flags & EV_DISABLE) &&
730 	    ((kn->kn_status & KN_DISABLED) == 0)) {
731 		crit_enter();
732 		kn->kn_status |= KN_DISABLED;
733 		crit_exit();
734 	}
735 
736 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
737 		crit_enter();
738 		kn->kn_status &= ~KN_DISABLED;
739 		if ((kn->kn_status & KN_ACTIVE) &&
740 		    ((kn->kn_status & KN_QUEUED) == 0))
741 			knote_enqueue(kn);
742 		crit_exit();
743 	}
744 
745 done:
746 	if (fp != NULL)
747 		fdrop(fp);
748 	return (error);
749 }
750 
751 /*
752  * Scan the kqueue, blocking if necessary until the target time is reached.
753  * If tsp is NULL we block indefinitely.  If tsp->ts_secs/nsecs are both
754  * 0 we do not block at all.
755  */
756 static int
757 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count,
758 	    struct timespec *tsp, int *errorp)
759 {
760 	struct knote *kn, marker;
761 	int total;
762 
763 	total = 0;
764 again:
765 	crit_enter();
766 	if (kq->kq_count == 0) {
767 		if (tsp == NULL) {
768 			kq->kq_state |= KQ_SLEEP;
769 			*errorp = tsleep(kq, PCATCH, "kqread", 0);
770 		} else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
771 			*errorp = EWOULDBLOCK;
772 		} else {
773 			struct timespec ats;
774 			struct timespec atx = *tsp;
775 			int timeout;
776 
777 			nanouptime(&ats);
778 			timespecsub(&atx, &ats);
779 			if (ats.tv_sec < 0) {
780 				*errorp = EWOULDBLOCK;
781 			} else {
782 				timeout = atx.tv_sec > 24 * 60 * 60 ?
783 					24 * 60 * 60 * hz : tstohz_high(&atx);
784 				kq->kq_state |= KQ_SLEEP;
785 				*errorp = tsleep(kq, PCATCH, "kqread", timeout);
786 			}
787 		}
788 		crit_exit();
789 		if (*errorp == 0)
790 			goto again;
791 		/* don't restart after signals... */
792 		if (*errorp == ERESTART)
793 			*errorp = EINTR;
794 		else if (*errorp == EWOULDBLOCK)
795 			*errorp = 0;
796 		goto done;
797 	}
798 
799 	/*
800 	 * Collect events.  Continuous mode events may get recycled
801 	 * past the marker so we stop when we hit it unless no events
802 	 * have been collected.
803 	 */
804 	TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
805 	while (count) {
806 		kn = TAILQ_FIRST(&kq->kq_knpend);
807 		if (kn == &marker)
808 			break;
809 		TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
810 		if (kn->kn_status & KN_DISABLED) {
811 			kn->kn_status &= ~KN_QUEUED;
812 			kq->kq_count--;
813 			continue;
814 		}
815 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
816 		    kn->kn_fop->f_event(kn, 0) == 0) {
817 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
818 			kq->kq_count--;
819 			continue;
820 		}
821 		*kevp++ = kn->kn_kevent;
822 		++total;
823 		--count;
824 
825 		/*
826 		 * Post-event action on the note
827 		 */
828 		if (kn->kn_flags & EV_ONESHOT) {
829 			kn->kn_status &= ~KN_QUEUED;
830 			kq->kq_count--;
831 			crit_exit();
832 			kn->kn_fop->f_detach(kn);
833 			knote_drop(kn);
834 			crit_enter();
835 		} else if (kn->kn_flags & EV_CLEAR) {
836 			kn->kn_data = 0;
837 			kn->kn_fflags = 0;
838 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
839 			kq->kq_count--;
840 		} else {
841 			TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
842 		}
843 	}
844 	TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
845 	crit_exit();
846 	if (total == 0)
847 		goto again;
848 done:
849 	return (total);
850 }
851 
852 /*
853  * XXX
854  * This could be expanded to call kqueue_scan, if desired.
855  *
856  * MPSAFE
857  */
858 static int
859 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
860 {
861 	return (ENXIO);
862 }
863 
864 /*
865  * MPSAFE
866  */
867 static int
868 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
869 {
870 	return (ENXIO);
871 }
872 
873 /*
874  * MPALMOSTSAFE
875  */
876 static int
877 kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
878 	     struct ucred *cred, struct sysmsg *msg)
879 {
880 	struct kqueue *kq;
881 	int error;
882 
883 	get_mplock();
884 	kq = (struct kqueue *)fp->f_data;
885 
886 	switch(com) {
887 	case FIOASYNC:
888 		if (*(int *)data)
889 			kq->kq_state |= KQ_ASYNC;
890 		else
891 			kq->kq_state &= ~KQ_ASYNC;
892 		error = 0;
893 		break;
894 	case FIOSETOWN:
895 		error = fsetown(*(int *)data, &kq->kq_sigio);
896 		break;
897 	default:
898 		error = ENOTTY;
899 		break;
900 	}
901 	rel_mplock();
902 	return (error);
903 }
904 
905 /*
906  * MPALMOSTSAFE - acquires mplock
907  */
908 static int
909 kqueue_poll(struct file *fp, int events, struct ucred *cred)
910 {
911 	struct kqueue *kq = (struct kqueue *)fp->f_data;
912 	int revents = 0;
913 
914 	get_mplock();
915 	crit_enter();
916         if (events & (POLLIN | POLLRDNORM)) {
917                 if (kq->kq_count) {
918                         revents |= events & (POLLIN | POLLRDNORM);
919 		} else {
920                         selrecord(curthread, &kq->kq_sel);
921 			kq->kq_state |= KQ_SEL;
922 		}
923 	}
924 	crit_exit();
925 	rel_mplock();
926 	return (revents);
927 }
928 
929 /*
930  * MPSAFE
931  */
932 static int
933 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
934 {
935 	struct kqueue *kq = (struct kqueue *)fp->f_data;
936 
937 	bzero((void *)st, sizeof(*st));
938 	st->st_size = kq->kq_count;
939 	st->st_blksize = sizeof(struct kevent);
940 	st->st_mode = S_IFIFO;
941 	return (0);
942 }
943 
944 /*
945  * MPALMOSTSAFE - acquires mplock
946  */
947 static int
948 kqueue_close(struct file *fp)
949 {
950 	struct kqueue *kq = (struct kqueue *)fp->f_data;
951 
952 	get_mplock();
953 
954 	kqueue_terminate(kq);
955 
956 	fp->f_data = NULL;
957 	funsetown(kq->kq_sigio);
958 	rel_mplock();
959 
960 	kfree(kq, M_KQUEUE);
961 	return (0);
962 }
963 
964 static void
965 kqueue_wakeup(struct kqueue *kq)
966 {
967 	if (kq->kq_state & KQ_SLEEP) {
968 		kq->kq_state &= ~KQ_SLEEP;
969 		wakeup(kq);
970 	}
971 	if (kq->kq_state & KQ_SEL) {
972 		kq->kq_state &= ~KQ_SEL;
973 		selwakeup(&kq->kq_sel);
974 	}
975 	KNOTE(&kq->kq_sel.si_note, 0);
976 }
977 
978 /*
979  * walk down a list of knotes, activating them if their event has triggered.
980  */
981 void
982 knote(struct klist *list, long hint)
983 {
984 	struct knote *kn;
985 
986 	SLIST_FOREACH(kn, list, kn_selnext)
987 		if (kn->kn_fop->f_event(kn, hint))
988 			KNOTE_ACTIVATE(kn);
989 }
990 
991 /*
992  * remove all knotes from a specified klist
993  */
994 void
995 knote_remove(struct klist *list)
996 {
997 	struct knote *kn;
998 
999 	while ((kn = SLIST_FIRST(list)) != NULL) {
1000 		kn->kn_fop->f_detach(kn);
1001 		knote_drop(kn);
1002 	}
1003 }
1004 
1005 /*
1006  * remove all knotes referencing a specified fd
1007  */
1008 void
1009 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd)
1010 {
1011 	struct knote *kn;
1012 
1013 restart:
1014 	SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
1015 		if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) {
1016 			kn->kn_fop->f_detach(kn);
1017 			knote_drop(kn);
1018 			goto restart;
1019 		}
1020 	}
1021 }
1022 
1023 static void
1024 knote_attach(struct knote *kn)
1025 {
1026 	struct klist *list;
1027 	struct kqueue *kq = kn->kn_kq;
1028 
1029 	if (kn->kn_fop->f_isfd) {
1030 		KKASSERT(kn->kn_fp);
1031 		list = &kn->kn_fp->f_klist;
1032 	} else {
1033 		if (kq->kq_knhashmask == 0)
1034 			kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1035 						 &kq->kq_knhashmask);
1036 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1037 	}
1038 	SLIST_INSERT_HEAD(list, kn, kn_link);
1039 	TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink);
1040 	kn->kn_status = 0;
1041 }
1042 
1043 /*
1044  * should be called outside of a critical section, since we don't want to
1045  * hold a critical section while calling fdrop and free.
1046  */
1047 static void
1048 knote_drop(struct knote *kn)
1049 {
1050 	struct kqueue *kq;
1051 	struct klist *list;
1052 
1053 	kq = kn->kn_kq;
1054 
1055 	if (kn->kn_fop->f_isfd)
1056 		list = &kn->kn_fp->f_klist;
1057 	else
1058 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1059 
1060 	SLIST_REMOVE(list, kn, knote, kn_link);
1061 	TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink);
1062 	if (kn->kn_status & KN_QUEUED)
1063 		knote_dequeue(kn);
1064 	if (kn->kn_fop->f_isfd)
1065 		fdrop(kn->kn_fp);
1066 	knote_free(kn);
1067 }
1068 
1069 
1070 static void
1071 knote_enqueue(struct knote *kn)
1072 {
1073 	struct kqueue *kq = kn->kn_kq;
1074 
1075 	crit_enter();
1076 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1077 
1078 	TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
1079 	kn->kn_status |= KN_QUEUED;
1080 	++kq->kq_count;
1081 
1082 	/*
1083 	 * Send SIGIO on request (typically set up as a mailbox signal)
1084 	 */
1085 	if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1086 		pgsigio(kq->kq_sigio, SIGIO, 0);
1087 	crit_exit();
1088 	kqueue_wakeup(kq);
1089 }
1090 
1091 static void
1092 knote_dequeue(struct knote *kn)
1093 {
1094 	struct kqueue *kq = kn->kn_kq;
1095 
1096 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1097 	crit_enter();
1098 
1099 	TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
1100 	kn->kn_status &= ~KN_QUEUED;
1101 	kq->kq_count--;
1102 	crit_exit();
1103 }
1104 
1105 static void
1106 knote_init(void)
1107 {
1108 	knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
1109 }
1110 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1111 
1112 static struct knote *
1113 knote_alloc(void)
1114 {
1115 	return ((struct knote *)zalloc(knote_zone));
1116 }
1117 
1118 static void
1119 knote_free(struct knote *kn)
1120 {
1121 	zfree(knote_zone, kn);
1122 }
1123