xref: /openbsd-src/sys/kern/kern_event.c (revision b2ea75c1b17e1a9a339660e7ed45cd24946b230e)
1 /*	$OpenBSD: kern_event.c,v 1.9 2001/07/17 01:51:37 provos 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);
81 void	knote_enqueue(struct knote *kn);
82 void	knote_dequeue(struct knote *kn);
83 void	knote_init(void);
84 struct	knote *knote_alloc(void);
85 void	knote_free(struct knote *kn);
86 
87 void	filt_kqdetach(struct knote *kn);
88 int	filt_kqueue(struct knote *kn, long hint);
89 int	filt_procattach(struct knote *kn);
90 void	filt_procdetach(struct knote *kn);
91 int	filt_proc(struct knote *kn, long hint);
92 int	filt_fileattach(struct knote *kn);
93 
94 struct filterops kqread_filtops =
95 	{ 1, NULL, filt_kqdetach, filt_kqueue };
96 struct filterops proc_filtops =
97 	{ 0, filt_procattach, filt_procdetach, filt_proc };
98 struct filterops file_filtops =
99 	{ 1, filt_fileattach, NULL, NULL };
100 
101 struct	pool knote_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 int
130 filt_fileattach(struct knote *kn)
131 {
132 	struct file *fp = kn->kn_fp;
133 
134 	return ((*fp->f_ops->fo_kqfilter)(fp, kn));
135 }
136 
137 int
138 kqueue_kqfilter(struct file *fp, struct knote *kn)
139 {
140 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
141 
142 	if (kn->kn_filter != EVFILT_READ)
143 		return (1);
144 
145 	kn->kn_fop = &kqread_filtops;
146 	SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext);
147 	return (0);
148 }
149 
150 void
151 filt_kqdetach(struct knote *kn)
152 {
153 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
154 
155 	SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext);
156 }
157 
158 /*ARGSUSED*/
159 int
160 filt_kqueue(struct knote *kn, long hint)
161 {
162 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
163 
164 	kn->kn_data = kq->kq_count;
165 	return (kn->kn_data > 0);
166 }
167 
168 int
169 filt_procattach(struct knote *kn)
170 {
171 	struct proc *p;
172 
173 	p = pfind(kn->kn_id);
174 	if (p == NULL)
175 		return (ESRCH);
176 
177 	/*
178 	 * Fail if it's not owned by you, or the last exec gave us
179 	 * setuid/setgid privs (unless you're root).
180 	 */
181 	if ((p->p_cred->p_ruid != curproc->p_cred->p_ruid ||
182 	        (p->p_flag & P_SUGID)) &&
183 	    suser(curproc->p_ucred, &curproc->p_acflag) != 0)
184 		return (EACCES);
185 
186 	kn->kn_ptr.p_proc = p;
187 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
188 
189 	/*
190 	 * internal flag indicating registration done by kernel
191 	 */
192 	if (kn->kn_flags & EV_FLAG1) {
193 		kn->kn_data = kn->kn_sdata;		/* ppid */
194 		kn->kn_fflags = NOTE_CHILD;
195 		kn->kn_flags &= ~EV_FLAG1;
196 	}
197 
198 	/* XXX lock the proc here while adding to the list? */
199 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
200 
201 	return (0);
202 }
203 
204 /*
205  * The knote may be attached to a different process, which may exit,
206  * leaving nothing for the knote to be attached to.  So when the process
207  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
208  * it will be deleted when read out.  However, as part of the knote deletion,
209  * this routine is called, so a check is needed to avoid actually performing
210  * a detach, because the original process does not exist any more.
211  */
212 void
213 filt_procdetach(struct knote *kn)
214 {
215 	struct proc *p = kn->kn_ptr.p_proc;
216 
217 	if (kn->kn_status & KN_DETACHED)
218 		return;
219 
220 	/* XXX locking?  this might modify another process. */
221 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
222 }
223 
224 int
225 filt_proc(struct knote *kn, long hint)
226 {
227 	u_int event;
228 
229 	/*
230 	 * mask off extra data
231 	 */
232 	event = (u_int)hint & NOTE_PCTRLMASK;
233 
234 	/*
235 	 * if the user is interested in this event, record it.
236 	 */
237 	if (kn->kn_sfflags & event)
238 		kn->kn_fflags |= event;
239 
240 	/*
241 	 * process is gone, so flag the event as finished.
242 	 */
243 	if (event == NOTE_EXIT) {
244 		kn->kn_status |= KN_DETACHED;
245 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
246 		return (1);
247 	}
248 
249 	/*
250 	 * process forked, and user wants to track the new process,
251 	 * so attach a new knote to it, and immediately report an
252 	 * event with the parent's pid.
253 	 */
254 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
255 		struct kevent kev;
256 		int error;
257 
258 		/*
259 		 * register knote with new process.
260 		 */
261 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
262 		kev.filter = kn->kn_filter;
263 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
264 		kev.fflags = kn->kn_sfflags;
265 		kev.data = kn->kn_id;			/* parent */
266 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
267 		error = kqueue_register(kn->kn_kq, &kev, NULL);
268 		if (error)
269 			kn->kn_fflags |= NOTE_TRACKERR;
270 	}
271 
272 	return (kn->kn_fflags != 0);
273 }
274 
275 int
276 sys_kqueue(struct proc *p, void *v, register_t *retval)
277 {
278 	struct filedesc *fdp = p->p_fd;
279 	struct kqueue *kq;
280 	struct file *fp;
281 	int fd, error;
282 
283 	error = falloc(p, &fp, &fd);
284 	if (error)
285 		return (error);
286 	fp->f_flag = FREAD | FWRITE;
287 	fp->f_type = DTYPE_KQUEUE;
288 	fp->f_ops = &kqueueops;
289 	kq = malloc(sizeof(struct kqueue), M_TEMP, M_WAITOK);
290 	bzero(kq, sizeof(*kq));
291 	TAILQ_INIT(&kq->kq_head);
292 	fp->f_data = (caddr_t)kq;
293 	*retval = fd;
294 	if (fdp->fd_knlistsize < 0)
295 		fdp->fd_knlistsize = 0;		/* this process has a kq */
296 	kq->kq_fdp = fdp;
297 	return (error);
298 }
299 
300 int
301 sys_kevent(struct proc *p, void *v, register_t *retval)
302 {
303 	struct filedesc* fdp = p->p_fd;
304 	struct sys_kevent_args /* {
305 		syscallarg(int)	fd;
306 		syscallarg(const struct kevent *) changelist;
307 		syscallarg(int)	nchanges;
308 		syscallarg(struct kevent *) eventlist;
309 		syscallarg(int)	nevents;
310 		syscallarg(const struct timespec *) timeout;
311 	} */ *uap = v;
312 	struct kevent *kevp;
313 	struct kqueue *kq;
314 	struct file *fp = NULL;
315 	struct timespec ts;
316 	int i, n, nerrors, error;
317 
318 	if (((u_int)SCARG(uap, fd)) >= fdp->fd_nfiles ||
319 	    (fp = fdp->fd_ofiles[SCARG(uap, fd)]) == NULL ||
320 	    (fp->f_type != DTYPE_KQUEUE))
321 		return (EBADF);
322 
323 	fp->f_count++;
324 
325 	if (SCARG(uap, timeout) != NULL) {
326 		error = copyin(SCARG(uap, timeout), &ts, sizeof(ts));
327 		if (error)
328 			goto done;
329 		SCARG(uap, timeout) = &ts;
330 	}
331 
332 	kq = (struct kqueue *)fp->f_data;
333 	nerrors = 0;
334 
335 	while (SCARG(uap, nchanges) > 0) {
336 		n = SCARG(uap, nchanges) > KQ_NEVENTS
337 			? KQ_NEVENTS : SCARG(uap, nchanges);
338 		error = copyin(SCARG(uap, changelist), kq->kq_kev,
339 		    n * sizeof(struct kevent));
340 		if (error)
341 			goto done;
342 		for (i = 0; i < n; i++) {
343 			kevp = &kq->kq_kev[i];
344 			kevp->flags &= ~EV_SYSFLAGS;
345 			error = kqueue_register(kq, kevp, p);
346 			if (error) {
347 				if (SCARG(uap, nevents) != 0) {
348 					kevp->flags = EV_ERROR;
349 					kevp->data = error;
350 					(void) copyout((caddr_t)kevp,
351 					    (caddr_t)SCARG(uap, eventlist),
352 					    sizeof(*kevp));
353 					SCARG(uap, eventlist)++;
354 					SCARG(uap, nevents)--;
355 					nerrors++;
356 				} else {
357 					goto done;
358 				}
359 			}
360 		}
361 		SCARG(uap, nchanges) -= n;
362 		SCARG(uap, changelist) += n;
363 	}
364 	if (nerrors) {
365 		*retval = nerrors;
366 		error = 0;
367 		goto done;
368 	}
369 
370 	error = kqueue_scan(fp, SCARG(uap, nevents), SCARG(uap, eventlist),
371 			    SCARG(uap, timeout), p, &n);
372 	*retval = n;
373  done:
374 	if (fp != NULL)
375 		closef(fp, p);
376 	return (error);
377 }
378 
379 int
380 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p)
381 {
382 	struct filedesc *fdp = kq->kq_fdp;
383 	struct filterops *fops = NULL;
384 	struct file *fp = NULL;
385 	struct knote *kn = NULL;
386 	int s, error = 0;
387 
388 	if (kev->filter < 0) {
389 		if (kev->filter + EVFILT_SYSCOUNT < 0)
390 			return (EINVAL);
391 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
392 	}
393 
394 	if (fops == NULL) {
395 		/*
396 		 * XXX
397 		 * filter attach routine is responsible for insuring that
398 		 * the identifier can be attached to it.
399 		 */
400 		printf("unknown filter: %d\n", kev->filter);
401 		return (EINVAL);
402 	}
403 
404 	if (fops->f_isfd) {
405 		/* validate descriptor */
406 		if ((u_int)kev->ident >= fdp->fd_nfiles ||
407 		    (fp = fdp->fd_ofiles[kev->ident]) == NULL)
408 			return (EBADF);
409 		fp->f_count++;
410 
411 		if (kev->ident < fdp->fd_knlistsize) {
412 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
413 				if (kq == kn->kn_kq &&
414 				    kev->filter == kn->kn_filter)
415 					break;
416 		}
417 	} else {
418 		if (fdp->fd_knhashmask != 0) {
419 			struct klist *list;
420 
421 			list = &fdp->fd_knhash[
422 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
423 			SLIST_FOREACH(kn, list, kn_link)
424 				if (kev->ident == kn->kn_id &&
425 				    kq == kn->kn_kq &&
426 				    kev->filter == kn->kn_filter)
427 					break;
428 		}
429 	}
430 
431 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
432 		error = ENOENT;
433 		goto done;
434 	}
435 
436 	/*
437 	 * kn now contains the matching knote, or NULL if no match
438 	 */
439 	if (kev->flags & EV_ADD) {
440 
441 		if (kn == NULL) {
442 			kn = knote_alloc();
443 			if (kn == NULL) {
444 				error = ENOMEM;
445 				goto done;
446 			}
447 			kn->kn_fp = fp;
448 			kn->kn_kq = kq;
449 			kn->kn_fop = fops;
450 
451 			/*
452 			 * apply reference count to knote structure, and
453 			 * do not release it at the end of this routine.
454 			 */
455 			fp = NULL;
456 
457 			kn->kn_sfflags = kev->fflags;
458 			kn->kn_sdata = kev->data;
459 			kev->fflags = 0;
460 			kev->data = 0;
461 			kn->kn_kevent = *kev;
462 
463 			knote_attach(kn, fdp);
464 			if ((error = fops->f_attach(kn)) != 0) {
465 				knote_drop(kn, p);
466 				goto done;
467 			}
468 		} else {
469 			/*
470 			 * The user may change some filter values after the
471 			 * initial EV_ADD, but doing so will not reset any
472 			 * filter which have already been triggered.
473 			 */
474 			kn->kn_sfflags = kev->fflags;
475 			kn->kn_sdata = kev->data;
476 			kn->kn_kevent.udata = kev->udata;
477 		}
478 
479 		s = splhigh();
480 		if (kn->kn_fop->f_event(kn, 0))
481 			KNOTE_ACTIVATE(kn);
482 		splx(s);
483 
484 	} else if (kev->flags & EV_DELETE) {
485 		kn->kn_fop->f_detach(kn);
486 		knote_drop(kn, p);
487 		goto done;
488 	}
489 
490 	if ((kev->flags & EV_DISABLE) &&
491 	    ((kn->kn_status & KN_DISABLED) == 0)) {
492 		s = splhigh();
493 		kn->kn_status |= KN_DISABLED;
494 		splx(s);
495 	}
496 
497 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
498 		s = splhigh();
499 		kn->kn_status &= ~KN_DISABLED;
500 		if ((kn->kn_status & KN_ACTIVE) &&
501 		    ((kn->kn_status & KN_QUEUED) == 0))
502 			knote_enqueue(kn);
503 		splx(s);
504 	}
505 
506 done:
507 	if (fp != NULL)
508 		closef(fp, p);
509 	return (error);
510 }
511 
512 int
513 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
514 	const struct timespec *tsp, struct proc *p, int *retval)
515 {
516 	struct kqueue *kq = (struct kqueue *)fp->f_data;
517 	struct kevent *kevp;
518 	struct timeval atv;
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 (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
529 			/* No timeout, just poll */
530 			timeout = -1;
531 			goto start;
532 		}
533 		if (itimerfix(&atv)) {
534 			error = EINVAL;
535 			goto done;
536 		}
537 
538 		s = splclock();
539 		timeradd(&atv, &time, &atv);
540 		timeout = hzto(&atv);
541 		splx(s);
542 	} else {
543 		atv.tv_sec = 0;
544 		atv.tv_usec = 0;
545 		timeout = 0;
546 	}
547 	goto start;
548 
549 retry:
550 	if (atv.tv_sec || atv.tv_usec) {
551 		timeout = hzto(&atv);
552 		if (timeout <= 0)
553 			goto done;
554 	}
555 
556 start:
557 	kevp = kq->kq_kev;
558 	s = splhigh();
559 	if (kq->kq_count == 0) {
560 		if (timeout < 0) {
561 			error = EWOULDBLOCK;
562 		} else {
563 			kq->kq_state |= KQ_SLEEP;
564 			error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout);
565 		}
566 		splx(s);
567 		if (error == 0)
568 			goto retry;
569 		/* don't restart after signals... */
570 		if (error == ERESTART)
571 			error = EINTR;
572 		else if (error == EWOULDBLOCK)
573 			error = 0;
574 		goto done;
575 	}
576 
577 	TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
578 	while (count) {
579 		kn = TAILQ_FIRST(&kq->kq_head);
580 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
581 		if (kn == &marker) {
582 			splx(s);
583 			if (count == maxevents)
584 				goto retry;
585 			goto done;
586 		}
587 		if (kn->kn_status & KN_DISABLED) {
588 			kn->kn_status &= ~KN_QUEUED;
589 			kq->kq_count--;
590 			continue;
591 		}
592 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
593 		    kn->kn_fop->f_event(kn, 0) == 0) {
594 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
595 			kq->kq_count--;
596 			continue;
597 		}
598 		*kevp = kn->kn_kevent;
599 		kevp++;
600 		nkev++;
601 		if (kn->kn_flags & EV_ONESHOT) {
602 			kn->kn_status &= ~KN_QUEUED;
603 			kq->kq_count--;
604 			splx(s);
605 			kn->kn_fop->f_detach(kn);
606 			knote_drop(kn, p);
607 			s = splhigh();
608 		} else if (kn->kn_flags & EV_CLEAR) {
609 			kn->kn_data = 0;
610 			kn->kn_fflags = 0;
611 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
612 			kq->kq_count--;
613 		} else {
614 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
615 		}
616 		count--;
617 		if (nkev == KQ_NEVENTS) {
618 			splx(s);
619 			error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
620 			    sizeof(struct kevent) * nkev);
621 			ulistp += nkev;
622 			nkev = 0;
623 			kevp = kq->kq_kev;
624 			s = splhigh();
625 			if (error)
626 				break;
627 		}
628 	}
629 	TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
630 	splx(s);
631 done:
632 	if (nkev != 0)
633 		error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
634 		    sizeof(struct kevent) * nkev);
635 	*retval = maxevents - count;
636 	return (error);
637 }
638 
639 /*
640  * XXX
641  * This could be expanded to call kqueue_scan, if desired.
642  */
643 /*ARGSUSED*/
644 int
645 kqueue_read(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred)
646 {
647 	return (ENXIO);
648 }
649 
650 /*ARGSUSED*/
651 int
652 kqueue_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred)
653 
654 {
655 	return (ENXIO);
656 }
657 
658 /*ARGSUSED*/
659 int
660 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
661 {
662 	return (ENOTTY);
663 }
664 
665 /*ARGSUSED*/
666 int
667 kqueue_select(struct file *fp, int which, struct proc *p)
668 {
669 	struct kqueue *kq = (struct kqueue *)fp->f_data;
670 	int res = 0;
671 	int s = splnet();
672 
673 	if (which == FREAD) {
674 		if (kq->kq_count) {
675 			res = 1;
676 		} else {
677 			selrecord(p, &kq->kq_sel);
678 			kq->kq_state |= KQ_SEL;
679 		}
680 	}
681 	splx(s);
682 	return (res);
683 }
684 
685 /*ARGSUSED*/
686 int
687 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
688 {
689 	struct kqueue *kq = (struct kqueue *)fp->f_data;
690 
691 	bzero((void *)st, sizeof(*st));
692 	st->st_size = kq->kq_count;
693 	st->st_blksize = sizeof(struct kevent);
694 	st->st_mode = S_IFIFO;
695 	return (0);
696 }
697 
698 /*ARGSUSED*/
699 int
700 kqueue_close(struct file *fp, struct proc *p)
701 {
702 	struct kqueue *kq = (struct kqueue *)fp->f_data;
703 	struct filedesc *fdp = p->p_fd;
704 	struct knote **knp, *kn, *kn0;
705 	int i;
706 
707 	for (i = 0; i < fdp->fd_knlistsize; i++) {
708 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
709 		kn = *knp;
710 		while (kn != NULL) {
711 			kn0 = SLIST_NEXT(kn, kn_link);
712 			if (kq == kn->kn_kq) {
713 				kn->kn_fop->f_detach(kn);
714 				closef(kn->kn_fp, p);
715 				knote_free(kn);
716 				*knp = kn0;
717 			} else {
718 				knp = &SLIST_NEXT(kn, kn_link);
719 			}
720 			kn = kn0;
721 		}
722 	}
723 	if (fdp->fd_knhashmask != 0) {
724 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
725 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
726 			kn = *knp;
727 			while (kn != NULL) {
728 				kn0 = SLIST_NEXT(kn, kn_link);
729 				if (kq == kn->kn_kq) {
730 					kn->kn_fop->f_detach(kn);
731 		/* XXX non-fd release of kn->kn_ptr */
732 					knote_free(kn);
733 					*knp = kn0;
734 				} else {
735 					knp = &SLIST_NEXT(kn, kn_link);
736 				}
737 				kn = kn0;
738 			}
739 		}
740 	}
741 	free(kq, M_TEMP);
742 	fp->f_data = NULL;
743 
744 	return (0);
745 }
746 
747 void
748 kqueue_wakeup(struct kqueue *kq)
749 {
750 
751 	if (kq->kq_state & KQ_SLEEP) {
752 		kq->kq_state &= ~KQ_SLEEP;
753 		wakeup(kq);
754 	}
755 	if (kq->kq_state & KQ_SEL) {
756 		kq->kq_state &= ~KQ_SEL;
757 		selwakeup(&kq->kq_sel);
758 	}
759 	KNOTE(&kq->kq_sel.si_note, 0);
760 }
761 
762 /*
763  * walk down a list of knotes, activating them if their event has triggered.
764  */
765 void
766 knote(struct klist *list, long hint)
767 {
768 	struct knote *kn;
769 
770 	SLIST_FOREACH(kn, list, kn_selnext)
771 		if (kn->kn_fop->f_event(kn, hint))
772 			KNOTE_ACTIVATE(kn);
773 }
774 
775 /*
776  * remove all knotes from a specified klist
777  */
778 void
779 knote_remove(struct proc *p, struct klist *list)
780 {
781 	struct knote *kn;
782 
783 	while ((kn = SLIST_FIRST(list)) != NULL) {
784 		kn->kn_fop->f_detach(kn);
785 		knote_drop(kn, p);
786 	}
787 }
788 
789 /*
790  * remove all knotes referencing a specified fd
791  */
792 void
793 knote_fdclose(struct proc *p, int fd)
794 {
795 	struct filedesc *fdp = p->p_fd;
796 	struct klist *list = &fdp->fd_knlist[fd];
797 
798 	knote_remove(p, list);
799 }
800 
801 void
802 knote_attach(struct knote *kn, struct filedesc *fdp)
803 {
804 	struct klist *list;
805 	int size;
806 
807 	if (! kn->kn_fop->f_isfd) {
808 		if (fdp->fd_knhashmask == 0)
809 			fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP,
810 			    M_WAITOK, &fdp->fd_knhashmask);
811 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
812 		goto done;
813 	}
814 
815 	if (fdp->fd_knlistsize <= kn->kn_id) {
816 		size = fdp->fd_knlistsize;
817 		while (size <= kn->kn_id)
818 			size += KQEXTENT;
819 		MALLOC(list, struct klist *,
820 		    size * sizeof(struct klist *), M_TEMP, M_WAITOK);
821 		bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
822 		    fdp->fd_knlistsize * sizeof(struct klist *));
823 		bzero((caddr_t)list +
824 		    fdp->fd_knlistsize * sizeof(struct klist *),
825 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
826 		if (fdp->fd_knlist != NULL)
827 			FREE(fdp->fd_knlist, M_TEMP);
828 		fdp->fd_knlistsize = size;
829 		fdp->fd_knlist = list;
830 	}
831 	list = &fdp->fd_knlist[kn->kn_id];
832 done:
833 	SLIST_INSERT_HEAD(list, kn, kn_link);
834 	kn->kn_status = 0;
835 }
836 
837 /*
838  * should be called at spl == 0, since we don't want to hold spl
839  * while calling closef and free.
840  */
841 void
842 knote_drop(struct knote *kn, struct proc *p)
843 {
844 	struct filedesc *fdp = p->p_fd;
845 	struct klist *list;
846 
847 	if (kn->kn_fop->f_isfd)
848 		list = &fdp->fd_knlist[kn->kn_id];
849 	else
850 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
851 
852 	SLIST_REMOVE(list, kn, knote, kn_link);
853 	if (kn->kn_status & KN_QUEUED)
854 		knote_dequeue(kn);
855 	if (kn->kn_fop->f_isfd)
856 		closef(kn->kn_fp, p);
857 	knote_free(kn);
858 }
859 
860 
861 void
862 knote_enqueue(struct knote *kn)
863 {
864 	struct kqueue *kq = kn->kn_kq;
865 	int s = splhigh();
866 
867 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
868 
869 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
870 	kn->kn_status |= KN_QUEUED;
871 	kq->kq_count++;
872 	splx(s);
873 	kqueue_wakeup(kq);
874 }
875 
876 void
877 knote_dequeue(struct knote *kn)
878 {
879 	struct kqueue *kq = kn->kn_kq;
880 	int s = splhigh();
881 
882 	KASSERT(kn->kn_status & KN_QUEUED);
883 
884 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
885 	kn->kn_status &= ~KN_QUEUED;
886 	kq->kq_count--;
887 	splx(s);
888 }
889 
890 void
891 knote_init(void)
892 {
893 	pool_init(&knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl",
894 	    0, pool_page_alloc_nointr, pool_page_free_nointr, M_KNOTE);
895 }
896 
897 struct knote *
898 knote_alloc(void)
899 {
900 	static int knote_pool_initialised;
901 
902 	if (!knote_pool_initialised) {
903 		knote_init();
904 		knote_pool_initialised++;
905 	}
906 
907 	return (pool_get(&knote_pool, PR_WAITOK));
908 }
909 
910 void
911 knote_free(struct knote *kn)
912 {
913 	pool_put(&knote_pool, kn);
914 }
915