xref: /openbsd-src/sys/kern/kern_event.c (revision 3a3fbb3f2e2521ab7c4a56b7ff7462ebd9095ec5)
1 /*	$OpenBSD: kern_event.c,v 1.10 2001/10/26 12:03:27 art 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 	FILE_SET_MATURE(fp);
298 	return (0);
299 }
300 
301 int
302 sys_kevent(struct proc *p, void *v, register_t *retval)
303 {
304 	struct filedesc* fdp = p->p_fd;
305 	struct sys_kevent_args /* {
306 		syscallarg(int)	fd;
307 		syscallarg(const struct kevent *) changelist;
308 		syscallarg(int)	nchanges;
309 		syscallarg(struct kevent *) eventlist;
310 		syscallarg(int)	nevents;
311 		syscallarg(const struct timespec *) timeout;
312 	} */ *uap = v;
313 	struct kevent *kevp;
314 	struct kqueue *kq;
315 	struct file *fp = NULL;
316 	struct timespec ts;
317 	int i, n, nerrors, error;
318 
319 	if ((fp = fd_getfile(fdp, 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 ((fp = fd_getfile(fdp, kev->ident)) == NULL)
407 			return (EBADF);
408 		fp->f_count++;
409 
410 		if (kev->ident < fdp->fd_knlistsize) {
411 			SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
412 				if (kq == kn->kn_kq &&
413 				    kev->filter == kn->kn_filter)
414 					break;
415 		}
416 	} else {
417 		if (fdp->fd_knhashmask != 0) {
418 			struct klist *list;
419 
420 			list = &fdp->fd_knhash[
421 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
422 			SLIST_FOREACH(kn, list, kn_link)
423 				if (kev->ident == kn->kn_id &&
424 				    kq == kn->kn_kq &&
425 				    kev->filter == kn->kn_filter)
426 					break;
427 		}
428 	}
429 
430 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
431 		error = ENOENT;
432 		goto done;
433 	}
434 
435 	/*
436 	 * kn now contains the matching knote, or NULL if no match
437 	 */
438 	if (kev->flags & EV_ADD) {
439 
440 		if (kn == NULL) {
441 			kn = knote_alloc();
442 			if (kn == NULL) {
443 				error = ENOMEM;
444 				goto done;
445 			}
446 			kn->kn_fp = fp;
447 			kn->kn_kq = kq;
448 			kn->kn_fop = fops;
449 
450 			/*
451 			 * apply reference count to knote structure, and
452 			 * do not release it at the end of this routine.
453 			 */
454 			fp = NULL;
455 
456 			kn->kn_sfflags = kev->fflags;
457 			kn->kn_sdata = kev->data;
458 			kev->fflags = 0;
459 			kev->data = 0;
460 			kn->kn_kevent = *kev;
461 
462 			knote_attach(kn, fdp);
463 			if ((error = fops->f_attach(kn)) != 0) {
464 				knote_drop(kn, p);
465 				goto done;
466 			}
467 		} else {
468 			/*
469 			 * The user may change some filter values after the
470 			 * initial EV_ADD, but doing so will not reset any
471 			 * filter which have already been triggered.
472 			 */
473 			kn->kn_sfflags = kev->fflags;
474 			kn->kn_sdata = kev->data;
475 			kn->kn_kevent.udata = kev->udata;
476 		}
477 
478 		s = splhigh();
479 		if (kn->kn_fop->f_event(kn, 0))
480 			KNOTE_ACTIVATE(kn);
481 		splx(s);
482 
483 	} else if (kev->flags & EV_DELETE) {
484 		kn->kn_fop->f_detach(kn);
485 		knote_drop(kn, p);
486 		goto done;
487 	}
488 
489 	if ((kev->flags & EV_DISABLE) &&
490 	    ((kn->kn_status & KN_DISABLED) == 0)) {
491 		s = splhigh();
492 		kn->kn_status |= KN_DISABLED;
493 		splx(s);
494 	}
495 
496 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
497 		s = splhigh();
498 		kn->kn_status &= ~KN_DISABLED;
499 		if ((kn->kn_status & KN_ACTIVE) &&
500 		    ((kn->kn_status & KN_QUEUED) == 0))
501 			knote_enqueue(kn);
502 		splx(s);
503 	}
504 
505 done:
506 	if (fp != NULL)
507 		closef(fp, p);
508 	return (error);
509 }
510 
511 int
512 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
513 	const struct timespec *tsp, struct proc *p, int *retval)
514 {
515 	struct kqueue *kq = (struct kqueue *)fp->f_data;
516 	struct kevent *kevp;
517 	struct timeval atv;
518 	struct knote *kn, marker;
519 	int s, count, timeout, nkev = 0, error = 0;
520 
521 	count = maxevents;
522 	if (count == 0)
523 		goto done;
524 
525 	if (tsp != NULL) {
526 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
527 		if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
528 			/* No timeout, just poll */
529 			timeout = -1;
530 			goto start;
531 		}
532 		if (itimerfix(&atv)) {
533 			error = EINVAL;
534 			goto done;
535 		}
536 
537 		s = splclock();
538 		timeradd(&atv, &time, &atv);
539 		timeout = hzto(&atv);
540 		splx(s);
541 	} else {
542 		atv.tv_sec = 0;
543 		atv.tv_usec = 0;
544 		timeout = 0;
545 	}
546 	goto start;
547 
548 retry:
549 	if (atv.tv_sec || atv.tv_usec) {
550 		timeout = hzto(&atv);
551 		if (timeout <= 0)
552 			goto done;
553 	}
554 
555 start:
556 	kevp = kq->kq_kev;
557 	s = splhigh();
558 	if (kq->kq_count == 0) {
559 		if (timeout < 0) {
560 			error = EWOULDBLOCK;
561 		} else {
562 			kq->kq_state |= KQ_SLEEP;
563 			error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout);
564 		}
565 		splx(s);
566 		if (error == 0)
567 			goto retry;
568 		/* don't restart after signals... */
569 		if (error == ERESTART)
570 			error = EINTR;
571 		else if (error == EWOULDBLOCK)
572 			error = 0;
573 		goto done;
574 	}
575 
576 	TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
577 	while (count) {
578 		kn = TAILQ_FIRST(&kq->kq_head);
579 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
580 		if (kn == &marker) {
581 			splx(s);
582 			if (count == maxevents)
583 				goto retry;
584 			goto done;
585 		}
586 		if (kn->kn_status & KN_DISABLED) {
587 			kn->kn_status &= ~KN_QUEUED;
588 			kq->kq_count--;
589 			continue;
590 		}
591 		if ((kn->kn_flags & EV_ONESHOT) == 0 &&
592 		    kn->kn_fop->f_event(kn, 0) == 0) {
593 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
594 			kq->kq_count--;
595 			continue;
596 		}
597 		*kevp = kn->kn_kevent;
598 		kevp++;
599 		nkev++;
600 		if (kn->kn_flags & EV_ONESHOT) {
601 			kn->kn_status &= ~KN_QUEUED;
602 			kq->kq_count--;
603 			splx(s);
604 			kn->kn_fop->f_detach(kn);
605 			knote_drop(kn, p);
606 			s = splhigh();
607 		} else if (kn->kn_flags & EV_CLEAR) {
608 			kn->kn_data = 0;
609 			kn->kn_fflags = 0;
610 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
611 			kq->kq_count--;
612 		} else {
613 			TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
614 		}
615 		count--;
616 		if (nkev == KQ_NEVENTS) {
617 			splx(s);
618 			error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
619 			    sizeof(struct kevent) * nkev);
620 			ulistp += nkev;
621 			nkev = 0;
622 			kevp = kq->kq_kev;
623 			s = splhigh();
624 			if (error)
625 				break;
626 		}
627 	}
628 	TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
629 	splx(s);
630 done:
631 	if (nkev != 0)
632 		error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
633 		    sizeof(struct kevent) * nkev);
634 	*retval = maxevents - count;
635 	return (error);
636 }
637 
638 /*
639  * XXX
640  * This could be expanded to call kqueue_scan, if desired.
641  */
642 /*ARGSUSED*/
643 int
644 kqueue_read(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred)
645 {
646 	return (ENXIO);
647 }
648 
649 /*ARGSUSED*/
650 int
651 kqueue_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred)
652 
653 {
654 	return (ENXIO);
655 }
656 
657 /*ARGSUSED*/
658 int
659 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p)
660 {
661 	return (ENOTTY);
662 }
663 
664 /*ARGSUSED*/
665 int
666 kqueue_select(struct file *fp, int which, struct proc *p)
667 {
668 	struct kqueue *kq = (struct kqueue *)fp->f_data;
669 	int res = 0;
670 	int s = splnet();
671 
672 	if (which == FREAD) {
673 		if (kq->kq_count) {
674 			res = 1;
675 		} else {
676 			selrecord(p, &kq->kq_sel);
677 			kq->kq_state |= KQ_SEL;
678 		}
679 	}
680 	splx(s);
681 	return (res);
682 }
683 
684 /*ARGSUSED*/
685 int
686 kqueue_stat(struct file *fp, struct stat *st, struct proc *p)
687 {
688 	struct kqueue *kq = (struct kqueue *)fp->f_data;
689 
690 	bzero((void *)st, sizeof(*st));
691 	st->st_size = kq->kq_count;
692 	st->st_blksize = sizeof(struct kevent);
693 	st->st_mode = S_IFIFO;
694 	return (0);
695 }
696 
697 /*ARGSUSED*/
698 int
699 kqueue_close(struct file *fp, struct proc *p)
700 {
701 	struct kqueue *kq = (struct kqueue *)fp->f_data;
702 	struct filedesc *fdp = p->p_fd;
703 	struct knote **knp, *kn, *kn0;
704 	int i;
705 
706 	for (i = 0; i < fdp->fd_knlistsize; i++) {
707 		knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
708 		kn = *knp;
709 		while (kn != NULL) {
710 			kn0 = SLIST_NEXT(kn, kn_link);
711 			if (kq == kn->kn_kq) {
712 				kn->kn_fop->f_detach(kn);
713 				closef(kn->kn_fp, p);
714 				knote_free(kn);
715 				*knp = kn0;
716 			} else {
717 				knp = &SLIST_NEXT(kn, kn_link);
718 			}
719 			kn = kn0;
720 		}
721 	}
722 	if (fdp->fd_knhashmask != 0) {
723 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
724 			knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
725 			kn = *knp;
726 			while (kn != NULL) {
727 				kn0 = SLIST_NEXT(kn, kn_link);
728 				if (kq == kn->kn_kq) {
729 					kn->kn_fop->f_detach(kn);
730 		/* XXX non-fd release of kn->kn_ptr */
731 					knote_free(kn);
732 					*knp = kn0;
733 				} else {
734 					knp = &SLIST_NEXT(kn, kn_link);
735 				}
736 				kn = kn0;
737 			}
738 		}
739 	}
740 	free(kq, M_TEMP);
741 	fp->f_data = NULL;
742 
743 	return (0);
744 }
745 
746 void
747 kqueue_wakeup(struct kqueue *kq)
748 {
749 
750 	if (kq->kq_state & KQ_SLEEP) {
751 		kq->kq_state &= ~KQ_SLEEP;
752 		wakeup(kq);
753 	}
754 	if (kq->kq_state & KQ_SEL) {
755 		kq->kq_state &= ~KQ_SEL;
756 		selwakeup(&kq->kq_sel);
757 	}
758 	KNOTE(&kq->kq_sel.si_note, 0);
759 }
760 
761 /*
762  * walk down a list of knotes, activating them if their event has triggered.
763  */
764 void
765 knote(struct klist *list, long hint)
766 {
767 	struct knote *kn;
768 
769 	SLIST_FOREACH(kn, list, kn_selnext)
770 		if (kn->kn_fop->f_event(kn, hint))
771 			KNOTE_ACTIVATE(kn);
772 }
773 
774 /*
775  * remove all knotes from a specified klist
776  */
777 void
778 knote_remove(struct proc *p, struct klist *list)
779 {
780 	struct knote *kn;
781 
782 	while ((kn = SLIST_FIRST(list)) != NULL) {
783 		kn->kn_fop->f_detach(kn);
784 		knote_drop(kn, p);
785 	}
786 }
787 
788 /*
789  * remove all knotes referencing a specified fd
790  */
791 void
792 knote_fdclose(struct proc *p, int fd)
793 {
794 	struct filedesc *fdp = p->p_fd;
795 	struct klist *list = &fdp->fd_knlist[fd];
796 
797 	knote_remove(p, list);
798 }
799 
800 void
801 knote_attach(struct knote *kn, struct filedesc *fdp)
802 {
803 	struct klist *list;
804 	int size;
805 
806 	if (! kn->kn_fop->f_isfd) {
807 		if (fdp->fd_knhashmask == 0)
808 			fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP,
809 			    M_WAITOK, &fdp->fd_knhashmask);
810 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
811 		goto done;
812 	}
813 
814 	if (fdp->fd_knlistsize <= kn->kn_id) {
815 		size = fdp->fd_knlistsize;
816 		while (size <= kn->kn_id)
817 			size += KQEXTENT;
818 		MALLOC(list, struct klist *,
819 		    size * sizeof(struct klist *), M_TEMP, M_WAITOK);
820 		bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
821 		    fdp->fd_knlistsize * sizeof(struct klist *));
822 		bzero((caddr_t)list +
823 		    fdp->fd_knlistsize * sizeof(struct klist *),
824 		    (size - fdp->fd_knlistsize) * sizeof(struct klist *));
825 		if (fdp->fd_knlist != NULL)
826 			FREE(fdp->fd_knlist, M_TEMP);
827 		fdp->fd_knlistsize = size;
828 		fdp->fd_knlist = list;
829 	}
830 	list = &fdp->fd_knlist[kn->kn_id];
831 done:
832 	SLIST_INSERT_HEAD(list, kn, kn_link);
833 	kn->kn_status = 0;
834 }
835 
836 /*
837  * should be called at spl == 0, since we don't want to hold spl
838  * while calling closef and free.
839  */
840 void
841 knote_drop(struct knote *kn, struct proc *p)
842 {
843 	struct filedesc *fdp = p->p_fd;
844 	struct klist *list;
845 
846 	if (kn->kn_fop->f_isfd)
847 		list = &fdp->fd_knlist[kn->kn_id];
848 	else
849 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
850 
851 	SLIST_REMOVE(list, kn, knote, kn_link);
852 	if (kn->kn_status & KN_QUEUED)
853 		knote_dequeue(kn);
854 	if (kn->kn_fop->f_isfd)
855 		closef(kn->kn_fp, p);
856 	knote_free(kn);
857 }
858 
859 
860 void
861 knote_enqueue(struct knote *kn)
862 {
863 	struct kqueue *kq = kn->kn_kq;
864 	int s = splhigh();
865 
866 	KASSERT((kn->kn_status & KN_QUEUED) == 0);
867 
868 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
869 	kn->kn_status |= KN_QUEUED;
870 	kq->kq_count++;
871 	splx(s);
872 	kqueue_wakeup(kq);
873 }
874 
875 void
876 knote_dequeue(struct knote *kn)
877 {
878 	struct kqueue *kq = kn->kn_kq;
879 	int s = splhigh();
880 
881 	KASSERT(kn->kn_status & KN_QUEUED);
882 
883 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
884 	kn->kn_status &= ~KN_QUEUED;
885 	kq->kq_count--;
886 	splx(s);
887 }
888 
889 void
890 knote_init(void)
891 {
892 	pool_init(&knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl",
893 	    0, pool_page_alloc_nointr, pool_page_free_nointr, M_KNOTE);
894 }
895 
896 struct knote *
897 knote_alloc(void)
898 {
899 	static int knote_pool_initialised;
900 
901 	if (!knote_pool_initialised) {
902 		knote_init();
903 		knote_pool_initialised++;
904 	}
905 
906 	return (pool_get(&knote_pool, PR_WAITOK));
907 }
908 
909 void
910 knote_free(struct knote *kn)
911 {
912 	pool_put(&knote_pool, kn);
913 }
914