xref: /openbsd-src/sys/kern/kern_event.c (revision db3296cf5c1dd9058ceecc3a29fe4aaa0bd26000)
1 /*	$OpenBSD: kern_event.c,v 1.19 2003/06/27 16:20:58 nate Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD: src/sys/kern/kern_event.c,v 1.22 2001/02/23 20:32:42 jlemon Exp $
29  */
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/proc.h>
35 #include <sys/malloc.h>
36 #include <sys/unistd.h>
37 #include <sys/file.h>
38 #include <sys/filedesc.h>
39 #include <sys/fcntl.h>
40 #include <sys/select.h>
41 #include <sys/queue.h>
42 #include <sys/event.h>
43 #include <sys/eventvar.h>
44 #include <sys/pool.h>
45 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/stat.h>
49 #include <sys/uio.h>
50 #include <sys/mount.h>
51 #include <sys/syscallargs.h>
52 
53 int	kqueue_scan(struct file *fp, int maxevents,
54 		    struct kevent *ulistp, const struct timespec *timeout,
55 		    struct proc *p, int *retval);
56 
57 int	kqueue_read(struct file *fp, off_t *poff, struct uio *uio,
58 		    struct ucred *cred);
59 int	kqueue_write(struct file *fp, off_t *poff, struct uio *uio,
60 		    struct ucred *cred);
61 int	kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
62 		    struct proc *p);
63 int	kqueue_select(struct file *fp, int which, struct proc *p);
64 int 	kqueue_kqfilter(struct file *fp, struct knote *kn);
65 int	kqueue_stat(struct file *fp, struct stat *st, struct proc *p);
66 int	kqueue_close(struct file *fp, struct proc *p);
67 void	kqueue_wakeup(struct kqueue *kq);
68 
69 struct fileops kqueueops = {
70 	kqueue_read,
71 	kqueue_write,
72 	kqueue_ioctl,
73 	kqueue_select,
74 	kqueue_kqfilter,
75 	kqueue_stat,
76 	kqueue_close
77 };
78 
79 void	knote_attach(struct knote *kn, struct filedesc *fdp);
80 void	knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp);
81 void	knote_enqueue(struct knote *kn);
82 void	knote_dequeue(struct knote *kn);
83 #define knote_alloc() ((struct knote *)pool_get(&knote_pool, PR_WAITOK))
84 #define knote_free(kn) (pool_put(&knote_pool, kn))
85 
86 void	filt_kqdetach(struct knote *kn);
87 int	filt_kqueue(struct knote *kn, long hint);
88 int	filt_procattach(struct knote *kn);
89 void	filt_procdetach(struct knote *kn);
90 int	filt_proc(struct knote *kn, long hint);
91 int	filt_fileattach(struct knote *kn);
92 
93 struct filterops kqread_filtops =
94 	{ 1, NULL, filt_kqdetach, filt_kqueue };
95 struct filterops proc_filtops =
96 	{ 0, filt_procattach, filt_procdetach, filt_proc };
97 struct filterops file_filtops =
98 	{ 1, filt_fileattach, NULL, NULL };
99 
100 struct	pool knote_pool;
101 struct	pool kqueue_pool;
102 
103 #define KNOTE_ACTIVATE(kn) do {						\
104 	kn->kn_status |= KN_ACTIVE;					\
105 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
106 		knote_enqueue(kn);					\
107 } while(0)
108 
109 #define	KN_HASHSIZE		64		/* XXX should be tunable */
110 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
111 
112 extern struct filterops sig_filtops;
113 #ifdef notyet
114 extern struct filterops aio_filtops;
115 #endif
116 
117 /*
118  * Table for for all system-defined filters.
119  */
120 struct filterops *sysfilt_ops[] = {
121 	&file_filtops,			/* EVFILT_READ */
122 	&file_filtops,			/* EVFILT_WRITE */
123 	NULL, /*&aio_filtops,*/		/* EVFILT_AIO */
124 	&file_filtops,			/* EVFILT_VNODE */
125 	&proc_filtops,			/* EVFILT_PROC */
126 	&sig_filtops,			/* EVFILT_SIGNAL */
127 };
128 
129 /* XXX - call this on startup instead. */
130 void kqueue_init(void);
131 
132 int kqueue_initialized;
133 
134 void
135 kqueue_init(void)
136 {
137 	pool_init(&kqueue_pool, sizeof(struct kqueue), 0, 0, 0, "kqeuepl",
138 	    &pool_allocator_nointr);
139 	pool_init(&knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl",
140 	    &pool_allocator_nointr);
141 
142 	kqueue_initialized = 1;
143 }
144 
145 int
146 filt_fileattach(struct knote *kn)
147 {
148 	struct file *fp = kn->kn_fp;
149 
150 	return ((*fp->f_ops->fo_kqfilter)(fp, kn));
151 }
152 
153 int
154 kqueue_kqfilter(struct file *fp, struct knote *kn)
155 {
156 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
157 
158 	if (kn->kn_filter != EVFILT_READ)
159 		return (1);
160 
161 	kn->kn_fop = &kqread_filtops;
162 	SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext);
163 	return (0);
164 }
165 
166 void
167 filt_kqdetach(struct knote *kn)
168 {
169 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
170 
171 	SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext);
172 }
173 
174 /*ARGSUSED*/
175 int
176 filt_kqueue(struct knote *kn, long hint)
177 {
178 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
179 
180 	kn->kn_data = kq->kq_count;
181 	return (kn->kn_data > 0);
182 }
183 
184 int
185 filt_procattach(struct knote *kn)
186 {
187 	struct proc *p;
188 
189 	p = pfind(kn->kn_id);
190 	if (p == NULL)
191 		return (ESRCH);
192 
193 	/*
194 	 * Fail if it's not owned by you, or the last exec gave us
195 	 * setuid/setgid privs (unless you're root).
196 	 */
197 	if ((p->p_cred->p_ruid != curproc->p_cred->p_ruid ||
198 	        (p->p_flag & P_SUGID)) &&
199 	    suser(curproc->p_ucred, &curproc->p_acflag) != 0)
200 		return (EACCES);
201 
202 	kn->kn_ptr.p_proc = p;
203 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
204 
205 	/*
206 	 * internal flag indicating registration done by kernel
207 	 */
208 	if (kn->kn_flags & EV_FLAG1) {
209 		kn->kn_data = kn->kn_sdata;		/* ppid */
210 		kn->kn_fflags = NOTE_CHILD;
211 		kn->kn_flags &= ~EV_FLAG1;
212 	}
213 
214 	/* XXX lock the proc here while adding to the list? */
215 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
216 
217 	return (0);
218 }
219 
220 /*
221  * The knote may be attached to a different process, which may exit,
222  * leaving nothing for the knote to be attached to.  So when the process
223  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
224  * it will be deleted when read out.  However, as part of the knote deletion,
225  * this routine is called, so a check is needed to avoid actually performing
226  * a detach, because the original process does not exist any more.
227  */
228 void
229 filt_procdetach(struct knote *kn)
230 {
231 	struct proc *p = kn->kn_ptr.p_proc;
232 
233 	if (kn->kn_status & KN_DETACHED)
234 		return;
235 
236 	/* XXX locking?  this might modify another process. */
237 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
238 }
239 
240 int
241 filt_proc(struct knote *kn, long hint)
242 {
243 	u_int event;
244 
245 	/*
246 	 * mask off extra data
247 	 */
248 	event = (u_int)hint & NOTE_PCTRLMASK;
249 
250 	/*
251 	 * if the user is interested in this event, record it.
252 	 */
253 	if (kn->kn_sfflags & event)
254 		kn->kn_fflags |= event;
255 
256 	/*
257 	 * process is gone, so flag the event as finished.
258 	 */
259 	if (event == NOTE_EXIT) {
260 		kn->kn_status |= KN_DETACHED;
261 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
262 		return (1);
263 	}
264 
265 	/*
266 	 * process forked, and user wants to track the new process,
267 	 * so attach a new knote to it, and immediately report an
268 	 * event with the parent's pid.
269 	 */
270 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
271 		struct kevent kev;
272 		int error;
273 
274 		/*
275 		 * register knote with new process.
276 		 */
277 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
278 		kev.filter = kn->kn_filter;
279 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
280 		kev.fflags = kn->kn_sfflags;
281 		kev.data = kn->kn_id;			/* parent */
282 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
283 		error = kqueue_register(kn->kn_kq, &kev, NULL);
284 		if (error)
285 			kn->kn_fflags |= NOTE_TRACKERR;
286 	}
287 
288 	return (kn->kn_fflags != 0);
289 }
290 
291 /*
292  * filt_seltrue:
293  *
294  *	This filter "event" routine simulates seltrue().
295  */
296 int
297 filt_seltrue(struct knote *kn, long hint)
298 {
299 
300 	/*
301 	 * We don't know how much data can be read/written,
302 	 * but we know that it *can* be.  This is about as
303 	 * good as select/poll does as well.
304 	 */
305 	kn->kn_data = 0;
306 	return (1);
307 }
308 
309 int
310 sys_kqueue(struct proc *p, void *v, register_t *retval)
311 {
312 	struct filedesc *fdp = p->p_fd;
313 	struct kqueue *kq;
314 	struct file *fp;
315 	int fd, error;
316 
317 	if (!kqueue_initialized)
318 		kqueue_init();
319 
320 	error = falloc(p, &fp, &fd);
321 	if (error)
322 		return (error);
323 	fp->f_flag = FREAD | FWRITE;
324 	fp->f_type = DTYPE_KQUEUE;
325 	fp->f_ops = &kqueueops;
326 	kq = pool_get(&kqueue_pool, PR_WAITOK);
327 	bzero(kq, sizeof(*kq));
328 	TAILQ_INIT(&kq->kq_head);
329 	fp->f_data = (caddr_t)kq;
330 	*retval = fd;
331 	if (fdp->fd_knlistsize < 0)
332 		fdp->fd_knlistsize = 0;		/* this process has a kq */
333 	kq->kq_fdp = fdp;
334 	FILE_SET_MATURE(fp);
335 	return (0);
336 }
337 
338 int
339 sys_kevent(struct proc *p, void *v, register_t *retval)
340 {
341 	struct filedesc* fdp = p->p_fd;
342 	struct sys_kevent_args /* {
343 		syscallarg(int)	fd;
344 		syscallarg(const struct kevent *) changelist;
345 		syscallarg(int)	nchanges;
346 		syscallarg(struct kevent *) eventlist;
347 		syscallarg(int)	nevents;
348 		syscallarg(const struct timespec *) timeout;
349 	} */ *uap = v;
350 	struct kevent *kevp;
351 	struct kqueue *kq;
352 	struct file *fp;
353 	struct timespec ts;
354 	int i, n, nerrors, error;
355 
356 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL ||
357 	    (fp->f_type != DTYPE_KQUEUE))
358 		return (EBADF);
359 
360 	FREF(fp);
361 
362 	if (SCARG(uap, timeout) != NULL) {
363 		error = copyin(SCARG(uap, timeout), &ts, sizeof(ts));
364 		if (error)
365 			goto done;
366 		SCARG(uap, timeout) = &ts;
367 	}
368 
369 	kq = (struct kqueue *)fp->f_data;
370 	nerrors = 0;
371 
372 	while (SCARG(uap, nchanges) > 0) {
373 		n = SCARG(uap, nchanges) > KQ_NEVENTS
374 			? KQ_NEVENTS : SCARG(uap, nchanges);
375 		error = copyin(SCARG(uap, changelist), kq->kq_kev,
376 		    n * sizeof(struct kevent));
377 		if (error)
378 			goto done;
379 		for (i = 0; i < n; i++) {
380 			kevp = &kq->kq_kev[i];
381 			kevp->flags &= ~EV_SYSFLAGS;
382 			error = kqueue_register(kq, kevp, p);
383 			if (error) {
384 				if (SCARG(uap, nevents) != 0) {
385 					kevp->flags = EV_ERROR;
386 					kevp->data = error;
387 					(void) copyout((caddr_t)kevp,
388 					    (caddr_t)SCARG(uap, eventlist),
389 					    sizeof(*kevp));
390 					SCARG(uap, eventlist)++;
391 					SCARG(uap, nevents)--;
392 					nerrors++;
393 				} else {
394 					goto done;
395 				}
396 			}
397 		}
398 		SCARG(uap, nchanges) -= n;
399 		SCARG(uap, changelist) += n;
400 	}
401 	if (nerrors) {
402 		*retval = nerrors;
403 		error = 0;
404 		goto done;
405 	}
406 
407 	error = kqueue_scan(fp, SCARG(uap, nevents), SCARG(uap, eventlist),
408 			    SCARG(uap, timeout), p, &n);
409 	*retval = n;
410  done:
411 	FRELE(fp);
412 	return (error);
413 }
414 
415 int
416 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p)
417 {
418 	struct filedesc *fdp = kq->kq_fdp;
419 	struct filterops *fops = NULL;
420 	struct file *fp = NULL;
421 	struct knote *kn = NULL;
422 	int s, error = 0;
423 
424 	if (kev->filter < 0) {
425 		if (kev->filter + EVFILT_SYSCOUNT < 0)
426 			return (EINVAL);
427 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
428 	}
429 
430 	if (fops == NULL) {
431 		/*
432 		 * XXX
433 		 * filter attach routine is responsible for insuring that
434 		 * the identifier can be attached to it.
435 		 */
436 		return (EINVAL);
437 	}
438 
439 	if (fops->f_isfd) {
440 		/* validate descriptor */
441 		if ((fp = fd_getfile(fdp, kev->ident)) == NULL)
442 			return (EBADF);
443 		FREF(fp);
444 		fp->f_count++;
445 
446 		if (kev->ident < fdp->fd_knlistsize) {
447 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
448 				if (kq == kn->kn_kq &&
449 				    kev->filter == kn->kn_filter)
450 					break;
451 		}
452 	} else {
453 		if (fdp->fd_knhashmask != 0) {
454 			struct klist *list;
455 
456 			list = &fdp->fd_knhash[
457 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
458 			SLIST_FOREACH(kn, list, kn_link)
459 				if (kev->ident == kn->kn_id &&
460 				    kq == kn->kn_kq &&
461 				    kev->filter == kn->kn_filter)
462 					break;
463 		}
464 	}
465 
466 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
467 		error = ENOENT;
468 		goto done;
469 	}
470 
471 	/*
472 	 * kn now contains the matching knote, or NULL if no match
473 	 */
474 	if (kev->flags & EV_ADD) {
475 
476 		if (kn == NULL) {
477 			kn = knote_alloc();
478 			if (kn == NULL) {
479 				error = ENOMEM;
480 				goto done;
481 			}
482 			kn->kn_fp = fp;
483 			kn->kn_kq = kq;
484 			kn->kn_fop = fops;
485 
486 			/*
487 			 * apply reference count to knote structure, and
488 			 * do not release it at the end of this routine.
489 			 */
490 			if (fp != NULL)
491 				FRELE(fp);
492 			fp = NULL;
493 
494 			kn->kn_sfflags = kev->fflags;
495 			kn->kn_sdata = kev->data;
496 			kev->fflags = 0;
497 			kev->data = 0;
498 			kn->kn_kevent = *kev;
499 
500 			knote_attach(kn, fdp);
501 			if ((error = fops->f_attach(kn)) != 0) {
502 				knote_drop(kn, p, fdp);
503 				goto done;
504 			}
505 		} else {
506 			/*
507 			 * The user may change some filter values after the
508 			 * initial EV_ADD, but doing so will not reset any
509 			 * filter which have already been triggered.
510 			 */
511 			kn->kn_sfflags = kev->fflags;
512 			kn->kn_sdata = kev->data;
513 			kn->kn_kevent.udata = kev->udata;
514 		}
515 
516 		s = splhigh();
517 		if (kn->kn_fop->f_event(kn, 0))
518 			KNOTE_ACTIVATE(kn);
519 		splx(s);
520 
521 	} else if (kev->flags & EV_DELETE) {
522 		kn->kn_fop->f_detach(kn);
523 		knote_drop(kn, p, p->p_fd);
524 		goto done;
525 	}
526 
527 	if ((kev->flags & EV_DISABLE) &&
528 	    ((kn->kn_status & KN_DISABLED) == 0)) {
529 		s = splhigh();
530 		kn->kn_status |= KN_DISABLED;
531 		splx(s);
532 	}
533 
534 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
535 		s = splhigh();
536 		kn->kn_status &= ~KN_DISABLED;
537 		if ((kn->kn_status & KN_ACTIVE) &&
538 		    ((kn->kn_status & KN_QUEUED) == 0))
539 			knote_enqueue(kn);
540 		splx(s);
541 	}
542 
543 done:
544 	if (fp != NULL)
545 		closef(fp, p);
546 	return (error);
547 }
548 
549 int
550 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
551 	const struct timespec *tsp, struct proc *p, int *retval)
552 {
553 	struct kqueue *kq = (struct kqueue *)fp->f_data;
554 	struct kevent *kevp;
555 	struct timeval atv;
556 	struct knote *kn, marker;
557 	int s, count, timeout, nkev = 0, error = 0;
558 
559 	count = maxevents;
560 	if (count == 0)
561 		goto done;
562 
563 	if (tsp != NULL) {
564 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
565 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
566 			/* No timeout, just poll */
567 			timeout = -1;
568 			goto start;
569 		}
570 		if (itimerfix(&atv)) {
571 			error = EINVAL;
572 			goto done;
573 		}
574 
575 		s = splclock();
576 		timeradd(&atv, &time, &atv);
577 		timeout = hzto(&atv);
578 		splx(s);
579 	} else {
580 		atv.tv_sec = 0;
581 		atv.tv_usec = 0;
582 		timeout = 0;
583 	}
584 	goto start;
585 
586 retry:
587 	if (atv.tv_sec || atv.tv_usec) {
588 		timeout = hzto(&atv);
589 		if (timeout <= 0)
590 			goto done;
591 	}
592 
593 start:
594 	kevp = kq->kq_kev;
595 	s = splhigh();
596 	if (kq->kq_count == 0) {
597 		if (timeout < 0) {
598 			error = EWOULDBLOCK;
599 		} else {
600 			kq->kq_state |= KQ_SLEEP;
601 			error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout);
602 		}
603 		splx(s);
604 		if (error == 0)
605 			goto retry;
606 		/* don't restart after signals... */
607 		if (error == ERESTART)
608 			error = EINTR;
609 		else if (error == EWOULDBLOCK)
610 			error = 0;
611 		goto done;
612 	}
613 
614 	TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
615 	while (count) {
616 		kn = TAILQ_FIRST(&kq->kq_head);
617 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
618 		if (kn == &marker) {
619 			splx(s);
620 			if (count == maxevents)
621 				goto retry;
622 			goto done;
623 		}
624 		if (kn->kn_status & KN_DISABLED) {
625 			kn->kn_status &= ~KN_QUEUED;
626 			kq->kq_count--;
627 			continue;
628 		}
629 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
630 		    kn->kn_fop->f_event(kn, 0) == 0) {
631 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
632 			kq->kq_count--;
633 			continue;
634 		}
635 		*kevp = kn->kn_kevent;
636 		kevp++;
637 		nkev++;
638 		if (kn->kn_flags & EV_ONESHOT) {
639 			kn->kn_status &= ~KN_QUEUED;
640 			kq->kq_count--;
641 			splx(s);
642 			kn->kn_fop->f_detach(kn);
643 			knote_drop(kn, p, p->p_fd);
644 			s = splhigh();
645 		} else if (kn->kn_flags & EV_CLEAR) {
646 			kn->kn_data = 0;
647 			kn->kn_fflags = 0;
648 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
649 			kq->kq_count--;
650 		} else {
651 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
652 		}
653 		count--;
654 		if (nkev == KQ_NEVENTS) {
655 			splx(s);
656 			error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
657 			    sizeof(struct kevent) * nkev);
658 			ulistp += nkev;
659 			nkev = 0;
660 			kevp = kq->kq_kev;
661 			s = splhigh();
662 			if (error)
663 				break;
664 		}
665 	}
666 	TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
667 	splx(s);
668 done:
669 	if (nkev != 0)
670 		error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
671 		    sizeof(struct kevent) * nkev);
672 	*retval = maxevents - count;
673 	return (error);
674 }
675 
676 /*
677  * XXX
678  * This could be expanded to call kqueue_scan, if desired.
679  */
680 /*ARGSUSED*/
681 int
682 kqueue_read(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred)
683 {
684 	return (ENXIO);
685 }
686 
687 /*ARGSUSED*/
688 int
689 kqueue_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred)
690 
691 {
692 	return (ENXIO);
693 }
694 
695 /*ARGSUSED*/
696 int
697 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
698 {
699 	return (ENOTTY);
700 }
701 
702 /*ARGSUSED*/
703 int
704 kqueue_select(struct file *fp, int which, struct proc *p)
705 {
706 	struct kqueue *kq = (struct kqueue *)fp->f_data;
707 	int res = 0;
708 	int s = splnet();
709 
710 	if (which == FREAD) {
711 		if (kq->kq_count) {
712 			res = 1;
713 		} else {
714 			selrecord(p, &kq->kq_sel);
715 			kq->kq_state |= KQ_SEL;
716 		}
717 	}
718 	splx(s);
719 	return (res);
720 }
721 
722 /*ARGSUSED*/
723 int
724 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
725 {
726 	struct kqueue *kq = (struct kqueue *)fp->f_data;
727 
728 	bzero((void *)st, sizeof(*st));
729 	st->st_size = kq->kq_count;
730 	st->st_blksize = sizeof(struct kevent);
731 	st->st_mode = S_IFIFO;
732 	return (0);
733 }
734 
735 /*ARGSUSED*/
736 int
737 kqueue_close(struct file *fp, struct proc *p)
738 {
739 	struct kqueue *kq = (struct kqueue *)fp->f_data;
740 	struct filedesc *fdp = p->p_fd;
741 	struct knote **knp, *kn, *kn0;
742 	int i;
743 
744 	for (i = 0; i < fdp->fd_knlistsize; i++) {
745 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
746 		kn = *knp;
747 		while (kn != NULL) {
748 			kn0 = SLIST_NEXT(kn, kn_link);
749 			if (kq == kn->kn_kq) {
750 				FREF(kn->kn_fp);
751 				kn->kn_fop->f_detach(kn);
752 				closef(kn->kn_fp, p);
753 				knote_free(kn);
754 				*knp = kn0;
755 			} else {
756 				knp = &SLIST_NEXT(kn, kn_link);
757 			}
758 			kn = kn0;
759 		}
760 	}
761 	if (fdp->fd_knhashmask != 0) {
762 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
763 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
764 			kn = *knp;
765 			while (kn != NULL) {
766 				kn0 = SLIST_NEXT(kn, kn_link);
767 				if (kq == kn->kn_kq) {
768 					kn->kn_fop->f_detach(kn);
769 		/* XXX non-fd release of kn->kn_ptr */
770 					knote_free(kn);
771 					*knp = kn0;
772 				} else {
773 					knp = &SLIST_NEXT(kn, kn_link);
774 				}
775 				kn = kn0;
776 			}
777 		}
778 	}
779 	pool_put(&kqueue_pool, kq);
780 	fp->f_data = NULL;
781 
782 	return (0);
783 }
784 
785 void
786 kqueue_wakeup(struct kqueue *kq)
787 {
788 
789 	if (kq->kq_state & KQ_SLEEP) {
790 		kq->kq_state &= ~KQ_SLEEP;
791 		wakeup(kq);
792 	}
793 	if (kq->kq_state & KQ_SEL) {
794 		kq->kq_state &= ~KQ_SEL;
795 		selwakeup(&kq->kq_sel);
796 	}
797 	KNOTE(&kq->kq_sel.si_note, 0);
798 }
799 
800 /*
801  * walk down a list of knotes, activating them if their event has triggered.
802  */
803 void
804 knote(struct klist *list, long hint)
805 {
806 	struct knote *kn;
807 
808 	SLIST_FOREACH(kn, list, kn_selnext)
809 		if (kn->kn_fop->f_event(kn, hint))
810 			KNOTE_ACTIVATE(kn);
811 }
812 
813 /*
814  * remove all knotes from a specified klist
815  */
816 void
817 knote_remove(struct proc *p, struct klist *list)
818 {
819 	struct knote *kn;
820 
821 	while ((kn = SLIST_FIRST(list)) != NULL) {
822 		kn->kn_fop->f_detach(kn);
823 		knote_drop(kn, p, p->p_fd);
824 	}
825 }
826 
827 /*
828  * remove all knotes referencing a specified fd
829  */
830 void
831 knote_fdclose(struct proc *p, int fd)
832 {
833 	struct filedesc *fdp = p->p_fd;
834 	struct klist *list = &fdp->fd_knlist[fd];
835 
836 	knote_remove(p, list);
837 }
838 
839 void
840 knote_attach(struct knote *kn, struct filedesc *fdp)
841 {
842 	struct klist *list;
843 	int size;
844 
845 	if (! kn->kn_fop->f_isfd) {
846 		if (fdp->fd_knhashmask == 0)
847 			fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP,
848 			    M_WAITOK, &fdp->fd_knhashmask);
849 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
850 		goto done;
851 	}
852 
853 	if (fdp->fd_knlistsize <= kn->kn_id) {
854 		size = fdp->fd_knlistsize;
855 		while (size <= kn->kn_id)
856 			size += KQEXTENT;
857 		list = malloc(size * sizeof(struct klist *), M_TEMP, M_WAITOK);
858 		bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
859 		    fdp->fd_knlistsize * sizeof(struct klist *));
860 		bzero((caddr_t)list +
861 		    fdp->fd_knlistsize * sizeof(struct klist *),
862 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
863 		if (fdp->fd_knlist != NULL)
864 			free(fdp->fd_knlist, M_TEMP);
865 		fdp->fd_knlistsize = size;
866 		fdp->fd_knlist = list;
867 	}
868 	list = &fdp->fd_knlist[kn->kn_id];
869 done:
870 	SLIST_INSERT_HEAD(list, kn, kn_link);
871 	kn->kn_status = 0;
872 }
873 
874 /*
875  * should be called at spl == 0, since we don't want to hold spl
876  * while calling closef and free.
877  */
878 void
879 knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp)
880 {
881 	struct klist *list;
882 
883 	if (kn->kn_fop->f_isfd)
884 		list = &fdp->fd_knlist[kn->kn_id];
885 	else
886 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
887 
888 	SLIST_REMOVE(list, kn, knote, kn_link);
889 	if (kn->kn_status & KN_QUEUED)
890 		knote_dequeue(kn);
891 	if (kn->kn_fop->f_isfd) {
892 		FREF(kn->kn_fp);
893 		closef(kn->kn_fp, p);
894 	}
895 	knote_free(kn);
896 }
897 
898 
899 void
900 knote_enqueue(struct knote *kn)
901 {
902 	struct kqueue *kq = kn->kn_kq;
903 	int s = splhigh();
904 
905 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
906 
907 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
908 	kn->kn_status |= KN_QUEUED;
909 	kq->kq_count++;
910 	splx(s);
911 	kqueue_wakeup(kq);
912 }
913 
914 void
915 knote_dequeue(struct knote *kn)
916 {
917 	struct kqueue *kq = kn->kn_kq;
918 	int s = splhigh();
919 
920 	KASSERT(kn->kn_status & KN_QUEUED);
921 
922 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
923 	kn->kn_status &= ~KN_QUEUED;
924 	kq->kq_count--;
925 	splx(s);
926 }
927