xref: /netbsd-src/sys/kern/kern_event.c (revision 5bbd2a12505d72a8177929a37b5cee489d0a1cfd)
1 /*	$NetBSD: kern_event.c,v 1.76 2012/06/02 15:54:02 martin Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
34  * All rights reserved.
35  *
36  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  *
57  * FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp
58  */
59 
60 #include <sys/cdefs.h>
61 __KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.76 2012/06/02 15:54:02 martin Exp $");
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/kernel.h>
66 #include <sys/proc.h>
67 #include <sys/file.h>
68 #include <sys/select.h>
69 #include <sys/queue.h>
70 #include <sys/event.h>
71 #include <sys/eventvar.h>
72 #include <sys/poll.h>
73 #include <sys/kmem.h>
74 #include <sys/stat.h>
75 #include <sys/filedesc.h>
76 #include <sys/syscallargs.h>
77 #include <sys/kauth.h>
78 #include <sys/conf.h>
79 #include <sys/atomic.h>
80 
81 static int	kqueue_scan(file_t *, size_t, struct kevent *,
82 			    const struct timespec *, register_t *,
83 			    const struct kevent_ops *, struct kevent *,
84 			    size_t);
85 static int	kqueue_ioctl(file_t *, u_long, void *);
86 static int	kqueue_fcntl(file_t *, u_int, void *);
87 static int	kqueue_poll(file_t *, int);
88 static int	kqueue_kqfilter(file_t *, struct knote *);
89 static int	kqueue_stat(file_t *, struct stat *);
90 static int	kqueue_close(file_t *);
91 static int	kqueue_register(struct kqueue *, struct kevent *);
92 static void	kqueue_doclose(struct kqueue *, struct klist *, int);
93 
94 static void	knote_detach(struct knote *, filedesc_t *fdp, bool);
95 static void	knote_enqueue(struct knote *);
96 static void	knote_activate(struct knote *);
97 
98 static void	filt_kqdetach(struct knote *);
99 static int	filt_kqueue(struct knote *, long hint);
100 static int	filt_procattach(struct knote *);
101 static void	filt_procdetach(struct knote *);
102 static int	filt_proc(struct knote *, long hint);
103 static int	filt_fileattach(struct knote *);
104 static void	filt_timerexpire(void *x);
105 static int	filt_timerattach(struct knote *);
106 static void	filt_timerdetach(struct knote *);
107 static int	filt_timer(struct knote *, long hint);
108 
109 static const struct fileops kqueueops = {
110 	.fo_read = (void *)enxio,
111 	.fo_write = (void *)enxio,
112 	.fo_ioctl = kqueue_ioctl,
113 	.fo_fcntl = kqueue_fcntl,
114 	.fo_poll = kqueue_poll,
115 	.fo_stat = kqueue_stat,
116 	.fo_close = kqueue_close,
117 	.fo_kqfilter = kqueue_kqfilter,
118 	.fo_restart = fnullop_restart,
119 };
120 
121 static const struct filterops kqread_filtops =
122 	{ 1, NULL, filt_kqdetach, filt_kqueue };
123 static const struct filterops proc_filtops =
124 	{ 0, filt_procattach, filt_procdetach, filt_proc };
125 static const struct filterops file_filtops =
126 	{ 1, filt_fileattach, NULL, NULL };
127 static const struct filterops timer_filtops =
128 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
129 
130 static u_int	kq_ncallouts = 0;
131 static int	kq_calloutmax = (4 * 1024);
132 
133 #define	KN_HASHSIZE		64		/* XXX should be tunable */
134 #define	KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
135 
136 extern const struct filterops sig_filtops;
137 
138 /*
139  * Table for for all system-defined filters.
140  * These should be listed in the numeric order of the EVFILT_* defines.
141  * If filtops is NULL, the filter isn't implemented in NetBSD.
142  * End of list is when name is NULL.
143  *
144  * Note that 'refcnt' is meaningless for built-in filters.
145  */
146 struct kfilter {
147 	const char	*name;		/* name of filter */
148 	uint32_t	filter;		/* id of filter */
149 	unsigned	refcnt;		/* reference count */
150 	const struct filterops *filtops;/* operations for filter */
151 	size_t		namelen;	/* length of name string */
152 };
153 
154 /* System defined filters */
155 static struct kfilter sys_kfilters[] = {
156 	{ "EVFILT_READ",	EVFILT_READ,	0, &file_filtops, 0 },
157 	{ "EVFILT_WRITE",	EVFILT_WRITE,	0, &file_filtops, 0, },
158 	{ "EVFILT_AIO",		EVFILT_AIO,	0, NULL, 0 },
159 	{ "EVFILT_VNODE",	EVFILT_VNODE,	0, &file_filtops, 0 },
160 	{ "EVFILT_PROC",	EVFILT_PROC,	0, &proc_filtops, 0 },
161 	{ "EVFILT_SIGNAL",	EVFILT_SIGNAL,	0, &sig_filtops, 0 },
162 	{ "EVFILT_TIMER",	EVFILT_TIMER,	0, &timer_filtops, 0 },
163 	{ NULL,			0,		0, NULL, 0 },
164 };
165 
166 /* User defined kfilters */
167 static struct kfilter	*user_kfilters;		/* array */
168 static int		user_kfilterc;		/* current offset */
169 static int		user_kfiltermaxc;	/* max size so far */
170 static size_t		user_kfiltersz;		/* size of allocated memory */
171 
172 /* Locks */
173 static krwlock_t	kqueue_filter_lock;	/* lock on filter lists */
174 static kmutex_t		kqueue_misc_lock;	/* miscellaneous */
175 
176 static kauth_listener_t	kqueue_listener;
177 
178 static int
179 kqueue_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
180     void *arg0, void *arg1, void *arg2, void *arg3)
181 {
182 	struct proc *p;
183 	int result;
184 
185 	result = KAUTH_RESULT_DEFER;
186 	p = arg0;
187 
188 	if (action != KAUTH_PROCESS_KEVENT_FILTER)
189 		return result;
190 
191 	if ((kauth_cred_getuid(p->p_cred) != kauth_cred_getuid(cred) ||
192 	    ISSET(p->p_flag, PK_SUGID)))
193 		return result;
194 
195 	result = KAUTH_RESULT_ALLOW;
196 
197 	return result;
198 }
199 
200 /*
201  * Initialize the kqueue subsystem.
202  */
203 void
204 kqueue_init(void)
205 {
206 
207 	rw_init(&kqueue_filter_lock);
208 	mutex_init(&kqueue_misc_lock, MUTEX_DEFAULT, IPL_NONE);
209 
210 	kqueue_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS,
211 	    kqueue_listener_cb, NULL);
212 }
213 
214 /*
215  * Find kfilter entry by name, or NULL if not found.
216  */
217 static struct kfilter *
218 kfilter_byname_sys(const char *name)
219 {
220 	int i;
221 
222 	KASSERT(rw_lock_held(&kqueue_filter_lock));
223 
224 	for (i = 0; sys_kfilters[i].name != NULL; i++) {
225 		if (strcmp(name, sys_kfilters[i].name) == 0)
226 			return &sys_kfilters[i];
227 	}
228 	return NULL;
229 }
230 
231 static struct kfilter *
232 kfilter_byname_user(const char *name)
233 {
234 	int i;
235 
236 	KASSERT(rw_lock_held(&kqueue_filter_lock));
237 
238 	/* user filter slots have a NULL name if previously deregistered */
239 	for (i = 0; i < user_kfilterc ; i++) {
240 		if (user_kfilters[i].name != NULL &&
241 		    strcmp(name, user_kfilters[i].name) == 0)
242 			return &user_kfilters[i];
243 	}
244 	return NULL;
245 }
246 
247 static struct kfilter *
248 kfilter_byname(const char *name)
249 {
250 	struct kfilter *kfilter;
251 
252 	KASSERT(rw_lock_held(&kqueue_filter_lock));
253 
254 	if ((kfilter = kfilter_byname_sys(name)) != NULL)
255 		return kfilter;
256 
257 	return kfilter_byname_user(name);
258 }
259 
260 /*
261  * Find kfilter entry by filter id, or NULL if not found.
262  * Assumes entries are indexed in filter id order, for speed.
263  */
264 static struct kfilter *
265 kfilter_byfilter(uint32_t filter)
266 {
267 	struct kfilter *kfilter;
268 
269 	KASSERT(rw_lock_held(&kqueue_filter_lock));
270 
271 	if (filter < EVFILT_SYSCOUNT)	/* it's a system filter */
272 		kfilter = &sys_kfilters[filter];
273 	else if (user_kfilters != NULL &&
274 	    filter < EVFILT_SYSCOUNT + user_kfilterc)
275 					/* it's a user filter */
276 		kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT];
277 	else
278 		return (NULL);		/* out of range */
279 	KASSERT(kfilter->filter == filter);	/* sanity check! */
280 	return (kfilter);
281 }
282 
283 /*
284  * Register a new kfilter. Stores the entry in user_kfilters.
285  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
286  * If retfilter != NULL, the new filterid is returned in it.
287  */
288 int
289 kfilter_register(const char *name, const struct filterops *filtops,
290 		 int *retfilter)
291 {
292 	struct kfilter *kfilter;
293 	size_t len;
294 	int i;
295 
296 	if (name == NULL || name[0] == '\0' || filtops == NULL)
297 		return (EINVAL);	/* invalid args */
298 
299 	rw_enter(&kqueue_filter_lock, RW_WRITER);
300 	if (kfilter_byname(name) != NULL) {
301 		rw_exit(&kqueue_filter_lock);
302 		return (EEXIST);	/* already exists */
303 	}
304 	if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT) {
305 		rw_exit(&kqueue_filter_lock);
306 		return (EINVAL);	/* too many */
307 	}
308 
309 	for (i = 0; i < user_kfilterc; i++) {
310 		kfilter = &user_kfilters[i];
311 		if (kfilter->name == NULL) {
312 			/* Previously deregistered slot.  Reuse. */
313 			goto reuse;
314 		}
315 	}
316 
317 	/* check if need to grow user_kfilters */
318 	if (user_kfilterc + 1 > user_kfiltermaxc) {
319 		/* Grow in KFILTER_EXTENT chunks. */
320 		user_kfiltermaxc += KFILTER_EXTENT;
321 		len = user_kfiltermaxc * sizeof(*kfilter);
322 		kfilter = kmem_alloc(len, KM_SLEEP);
323 		memset((char *)kfilter + user_kfiltersz, 0, len - user_kfiltersz);
324 		if (user_kfilters != NULL) {
325 			memcpy(kfilter, user_kfilters, user_kfiltersz);
326 			kmem_free(user_kfilters, user_kfiltersz);
327 		}
328 		user_kfiltersz = len;
329 		user_kfilters = kfilter;
330 	}
331 	/* Adding new slot */
332 	kfilter = &user_kfilters[user_kfilterc++];
333 reuse:
334 	kfilter->namelen = strlen(name) + 1;
335 	kfilter->name = kmem_alloc(kfilter->namelen, KM_SLEEP);
336 	memcpy(__UNCONST(kfilter->name), name, kfilter->namelen);
337 
338 	kfilter->filter = (kfilter - user_kfilters) + EVFILT_SYSCOUNT;
339 
340 	kfilter->filtops = kmem_alloc(sizeof(*filtops), KM_SLEEP);
341 	memcpy(__UNCONST(kfilter->filtops), filtops, sizeof(*filtops));
342 
343 	if (retfilter != NULL)
344 		*retfilter = kfilter->filter;
345 	rw_exit(&kqueue_filter_lock);
346 
347 	return (0);
348 }
349 
350 /*
351  * Unregister a kfilter previously registered with kfilter_register.
352  * This retains the filter id, but clears the name and frees filtops (filter
353  * operations), so that the number isn't reused during a boot.
354  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
355  */
356 int
357 kfilter_unregister(const char *name)
358 {
359 	struct kfilter *kfilter;
360 
361 	if (name == NULL || name[0] == '\0')
362 		return (EINVAL);	/* invalid name */
363 
364 	rw_enter(&kqueue_filter_lock, RW_WRITER);
365 	if (kfilter_byname_sys(name) != NULL) {
366 		rw_exit(&kqueue_filter_lock);
367 		return (EINVAL);	/* can't detach system filters */
368 	}
369 
370 	kfilter = kfilter_byname_user(name);
371 	if (kfilter == NULL) {
372 		rw_exit(&kqueue_filter_lock);
373 		return (ENOENT);
374 	}
375 	if (kfilter->refcnt != 0) {
376 		rw_exit(&kqueue_filter_lock);
377 		return (EBUSY);
378 	}
379 
380 	/* Cast away const (but we know it's safe. */
381 	kmem_free(__UNCONST(kfilter->name), kfilter->namelen);
382 	kfilter->name = NULL;	/* mark as `not implemented' */
383 
384 	if (kfilter->filtops != NULL) {
385 		/* Cast away const (but we know it's safe. */
386 		kmem_free(__UNCONST(kfilter->filtops),
387 		    sizeof(*kfilter->filtops));
388 		kfilter->filtops = NULL; /* mark as `not implemented' */
389 	}
390 	rw_exit(&kqueue_filter_lock);
391 
392 	return (0);
393 }
394 
395 
396 /*
397  * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file
398  * descriptors. Calls fileops kqfilter method for given file descriptor.
399  */
400 static int
401 filt_fileattach(struct knote *kn)
402 {
403 	file_t *fp;
404 
405 	fp = kn->kn_obj;
406 
407 	return (*fp->f_ops->fo_kqfilter)(fp, kn);
408 }
409 
410 /*
411  * Filter detach method for EVFILT_READ on kqueue descriptor.
412  */
413 static void
414 filt_kqdetach(struct knote *kn)
415 {
416 	struct kqueue *kq;
417 
418 	kq = ((file_t *)kn->kn_obj)->f_data;
419 
420 	mutex_spin_enter(&kq->kq_lock);
421 	SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext);
422 	mutex_spin_exit(&kq->kq_lock);
423 }
424 
425 /*
426  * Filter event method for EVFILT_READ on kqueue descriptor.
427  */
428 /*ARGSUSED*/
429 static int
430 filt_kqueue(struct knote *kn, long hint)
431 {
432 	struct kqueue *kq;
433 	int rv;
434 
435 	kq = ((file_t *)kn->kn_obj)->f_data;
436 
437 	if (hint != NOTE_SUBMIT)
438 		mutex_spin_enter(&kq->kq_lock);
439 	kn->kn_data = kq->kq_count;
440 	rv = (kn->kn_data > 0);
441 	if (hint != NOTE_SUBMIT)
442 		mutex_spin_exit(&kq->kq_lock);
443 
444 	return rv;
445 }
446 
447 /*
448  * Filter attach method for EVFILT_PROC.
449  */
450 static int
451 filt_procattach(struct knote *kn)
452 {
453 	struct proc *p, *curp;
454 	struct lwp *curl;
455 
456 	curl = curlwp;
457 	curp = curl->l_proc;
458 
459 	mutex_enter(proc_lock);
460 	p = proc_find(kn->kn_id);
461 	if (p == NULL) {
462 		mutex_exit(proc_lock);
463 		return ESRCH;
464 	}
465 
466 	/*
467 	 * Fail if it's not owned by you, or the last exec gave us
468 	 * setuid/setgid privs (unless you're root).
469 	 */
470 	mutex_enter(p->p_lock);
471 	mutex_exit(proc_lock);
472 	if (kauth_authorize_process(curl->l_cred, KAUTH_PROCESS_KEVENT_FILTER,
473 	    p, NULL, NULL, NULL) != 0) {
474 	    	mutex_exit(p->p_lock);
475 		return EACCES;
476 	}
477 
478 	kn->kn_obj = p;
479 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
480 
481 	/*
482 	 * internal flag indicating registration done by kernel
483 	 */
484 	if (kn->kn_flags & EV_FLAG1) {
485 		kn->kn_data = kn->kn_sdata;	/* ppid */
486 		kn->kn_fflags = NOTE_CHILD;
487 		kn->kn_flags &= ~EV_FLAG1;
488 	}
489 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
490     	mutex_exit(p->p_lock);
491 
492 	return 0;
493 }
494 
495 /*
496  * Filter detach method for EVFILT_PROC.
497  *
498  * The knote may be attached to a different process, which may exit,
499  * leaving nothing for the knote to be attached to.  So when the process
500  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
501  * it will be deleted when read out.  However, as part of the knote deletion,
502  * this routine is called, so a check is needed to avoid actually performing
503  * a detach, because the original process might not exist any more.
504  */
505 static void
506 filt_procdetach(struct knote *kn)
507 {
508 	struct proc *p;
509 
510 	if (kn->kn_status & KN_DETACHED)
511 		return;
512 
513 	p = kn->kn_obj;
514 
515 	mutex_enter(p->p_lock);
516 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
517 	mutex_exit(p->p_lock);
518 }
519 
520 /*
521  * Filter event method for EVFILT_PROC.
522  */
523 static int
524 filt_proc(struct knote *kn, long hint)
525 {
526 	u_int event, fflag;
527 	struct kevent kev;
528 	struct kqueue *kq;
529 	int error;
530 
531 	event = (u_int)hint & NOTE_PCTRLMASK;
532 	kq = kn->kn_kq;
533 	fflag = 0;
534 
535 	/* If the user is interested in this event, record it. */
536 	if (kn->kn_sfflags & event)
537 		fflag |= event;
538 
539 	if (event == NOTE_EXIT) {
540 		/*
541 		 * Process is gone, so flag the event as finished.
542 		 *
543 		 * Detach the knote from watched process and mark
544 		 * it as such. We can't leave this to kqueue_scan(),
545 		 * since the process might not exist by then. And we
546 		 * have to do this now, since psignal KNOTE() is called
547 		 * also for zombies and we might end up reading freed
548 		 * memory if the kevent would already be picked up
549 		 * and knote g/c'ed.
550 		 */
551 		filt_procdetach(kn);
552 
553 		mutex_spin_enter(&kq->kq_lock);
554 		kn->kn_status |= KN_DETACHED;
555 		/* Mark as ONESHOT, so that the knote it g/c'ed when read */
556 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
557 		kn->kn_fflags |= fflag;
558 		mutex_spin_exit(&kq->kq_lock);
559 
560 		return 1;
561 	}
562 
563 	mutex_spin_enter(&kq->kq_lock);
564 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
565 		/*
566 		 * Process forked, and user wants to track the new process,
567 		 * so attach a new knote to it, and immediately report an
568 		 * event with the parent's pid.  Register knote with new
569 		 * process.
570 		 */
571 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
572 		kev.filter = kn->kn_filter;
573 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
574 		kev.fflags = kn->kn_sfflags;
575 		kev.data = kn->kn_id;			/* parent */
576 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
577 		mutex_spin_exit(&kq->kq_lock);
578 		error = kqueue_register(kq, &kev);
579 		mutex_spin_enter(&kq->kq_lock);
580 		if (error != 0)
581 			kn->kn_fflags |= NOTE_TRACKERR;
582 	}
583 	kn->kn_fflags |= fflag;
584 	fflag = kn->kn_fflags;
585 	mutex_spin_exit(&kq->kq_lock);
586 
587 	return fflag != 0;
588 }
589 
590 static void
591 filt_timerexpire(void *knx)
592 {
593 	struct knote *kn = knx;
594 	int tticks;
595 
596 	mutex_enter(&kqueue_misc_lock);
597 	kn->kn_data++;
598 	knote_activate(kn);
599 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
600 		tticks = mstohz(kn->kn_sdata);
601 		if (tticks <= 0)
602 			tticks = 1;
603 		callout_schedule((callout_t *)kn->kn_hook, tticks);
604 	}
605 	mutex_exit(&kqueue_misc_lock);
606 }
607 
608 /*
609  * data contains amount of time to sleep, in milliseconds
610  */
611 static int
612 filt_timerattach(struct knote *kn)
613 {
614 	callout_t *calloutp;
615 	struct kqueue *kq;
616 	int tticks;
617 
618 	tticks = mstohz(kn->kn_sdata);
619 
620 	/* if the supplied value is under our resolution, use 1 tick */
621 	if (tticks == 0) {
622 		if (kn->kn_sdata == 0)
623 			return EINVAL;
624 		tticks = 1;
625 	}
626 
627 	if (atomic_inc_uint_nv(&kq_ncallouts) >= kq_calloutmax ||
628 	    (calloutp = kmem_alloc(sizeof(*calloutp), KM_NOSLEEP)) == NULL) {
629 		atomic_dec_uint(&kq_ncallouts);
630 		return ENOMEM;
631 	}
632 	callout_init(calloutp, CALLOUT_MPSAFE);
633 
634 	kq = kn->kn_kq;
635 	mutex_spin_enter(&kq->kq_lock);
636 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
637 	kn->kn_hook = calloutp;
638 	mutex_spin_exit(&kq->kq_lock);
639 
640 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
641 
642 	return (0);
643 }
644 
645 static void
646 filt_timerdetach(struct knote *kn)
647 {
648 	callout_t *calloutp;
649 
650 	calloutp = (callout_t *)kn->kn_hook;
651 	callout_halt(calloutp, NULL);
652 	callout_destroy(calloutp);
653 	kmem_free(calloutp, sizeof(*calloutp));
654 	atomic_dec_uint(&kq_ncallouts);
655 }
656 
657 static int
658 filt_timer(struct knote *kn, long hint)
659 {
660 	int rv;
661 
662 	mutex_enter(&kqueue_misc_lock);
663 	rv = (kn->kn_data != 0);
664 	mutex_exit(&kqueue_misc_lock);
665 
666 	return rv;
667 }
668 
669 /*
670  * filt_seltrue:
671  *
672  *	This filter "event" routine simulates seltrue().
673  */
674 int
675 filt_seltrue(struct knote *kn, long hint)
676 {
677 
678 	/*
679 	 * We don't know how much data can be read/written,
680 	 * but we know that it *can* be.  This is about as
681 	 * good as select/poll does as well.
682 	 */
683 	kn->kn_data = 0;
684 	return (1);
685 }
686 
687 /*
688  * This provides full kqfilter entry for device switch tables, which
689  * has same effect as filter using filt_seltrue() as filter method.
690  */
691 static void
692 filt_seltruedetach(struct knote *kn)
693 {
694 	/* Nothing to do */
695 }
696 
697 const struct filterops seltrue_filtops =
698 	{ 1, NULL, filt_seltruedetach, filt_seltrue };
699 
700 int
701 seltrue_kqfilter(dev_t dev, struct knote *kn)
702 {
703 	switch (kn->kn_filter) {
704 	case EVFILT_READ:
705 	case EVFILT_WRITE:
706 		kn->kn_fop = &seltrue_filtops;
707 		break;
708 	default:
709 		return (EINVAL);
710 	}
711 
712 	/* Nothing more to do */
713 	return (0);
714 }
715 
716 /*
717  * kqueue(2) system call.
718  */
719 static int
720 kqueue1(struct lwp *l, int flags, register_t *retval)
721 {
722 	struct kqueue *kq;
723 	file_t *fp;
724 	int fd, error;
725 
726 	if ((error = fd_allocfile(&fp, &fd)) != 0)
727 		return error;
728 	fp->f_flag = FREAD | FWRITE | (flags & (FNONBLOCK|FNOSIGPIPE));
729 	fp->f_type = DTYPE_KQUEUE;
730 	fp->f_ops = &kqueueops;
731 	kq = kmem_zalloc(sizeof(*kq), KM_SLEEP);
732 	mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED);
733 	cv_init(&kq->kq_cv, "kqueue");
734 	selinit(&kq->kq_sel);
735 	TAILQ_INIT(&kq->kq_head);
736 	fp->f_data = kq;
737 	*retval = fd;
738 	kq->kq_fdp = curlwp->l_fd;
739 	fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0);
740 	fd_affix(curproc, fp, fd);
741 	return error;
742 }
743 
744 /*
745  * kqueue(2) system call.
746  */
747 int
748 sys_kqueue(struct lwp *l, const void *v, register_t *retval)
749 {
750 	return kqueue1(l, 0, retval);
751 }
752 
753 int
754 sys_kqueue1(struct lwp *l, const struct sys_kqueue1_args *uap,
755     register_t *retval)
756 {
757 	/* {
758 		syscallarg(int) flags;
759 	} */
760 	return kqueue1(l, SCARG(uap, flags), retval);
761 }
762 
763 /*
764  * kevent(2) system call.
765  */
766 int
767 kevent_fetch_changes(void *private, const struct kevent *changelist,
768     struct kevent *changes, size_t index, int n)
769 {
770 
771 	return copyin(changelist + index, changes, n * sizeof(*changes));
772 }
773 
774 int
775 kevent_put_events(void *private, struct kevent *events,
776     struct kevent *eventlist, size_t index, int n)
777 {
778 
779 	return copyout(events, eventlist + index, n * sizeof(*events));
780 }
781 
782 static const struct kevent_ops kevent_native_ops = {
783 	.keo_private = NULL,
784 	.keo_fetch_timeout = copyin,
785 	.keo_fetch_changes = kevent_fetch_changes,
786 	.keo_put_events = kevent_put_events,
787 };
788 
789 int
790 sys___kevent50(struct lwp *l, const struct sys___kevent50_args *uap,
791     register_t *retval)
792 {
793 	/* {
794 		syscallarg(int) fd;
795 		syscallarg(const struct kevent *) changelist;
796 		syscallarg(size_t) nchanges;
797 		syscallarg(struct kevent *) eventlist;
798 		syscallarg(size_t) nevents;
799 		syscallarg(const struct timespec *) timeout;
800 	} */
801 
802 	return kevent1(retval, SCARG(uap, fd), SCARG(uap, changelist),
803 	    SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents),
804 	    SCARG(uap, timeout), &kevent_native_ops);
805 }
806 
807 int
808 kevent1(register_t *retval, int fd,
809 	const struct kevent *changelist, size_t nchanges,
810 	struct kevent *eventlist, size_t nevents,
811 	const struct timespec *timeout,
812 	const struct kevent_ops *keops)
813 {
814 	struct kevent *kevp;
815 	struct kqueue *kq;
816 	struct timespec	ts;
817 	size_t i, n, ichange;
818 	int nerrors, error;
819 	struct kevent kevbuf[8];	/* approx 300 bytes on 64-bit */
820 	file_t *fp;
821 
822 	/* check that we're dealing with a kq */
823 	fp = fd_getfile(fd);
824 	if (fp == NULL)
825 		return (EBADF);
826 
827 	if (fp->f_type != DTYPE_KQUEUE) {
828 		fd_putfile(fd);
829 		return (EBADF);
830 	}
831 
832 	if (timeout != NULL) {
833 		error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts));
834 		if (error)
835 			goto done;
836 		timeout = &ts;
837 	}
838 
839 	kq = (struct kqueue *)fp->f_data;
840 	nerrors = 0;
841 	ichange = 0;
842 
843 	/* traverse list of events to register */
844 	while (nchanges > 0) {
845 		n = MIN(nchanges, __arraycount(kevbuf));
846 		error = (*keops->keo_fetch_changes)(keops->keo_private,
847 		    changelist, kevbuf, ichange, n);
848 		if (error)
849 			goto done;
850 		for (i = 0; i < n; i++) {
851 			kevp = &kevbuf[i];
852 			kevp->flags &= ~EV_SYSFLAGS;
853 			/* register each knote */
854 			error = kqueue_register(kq, kevp);
855 			if (error) {
856 				if (nevents != 0) {
857 					kevp->flags = EV_ERROR;
858 					kevp->data = error;
859 					error = (*keops->keo_put_events)
860 					    (keops->keo_private, kevp,
861 					    eventlist, nerrors, 1);
862 					if (error)
863 						goto done;
864 					nevents--;
865 					nerrors++;
866 				} else {
867 					goto done;
868 				}
869 			}
870 		}
871 		nchanges -= n;	/* update the results */
872 		ichange += n;
873 	}
874 	if (nerrors) {
875 		*retval = nerrors;
876 		error = 0;
877 		goto done;
878 	}
879 
880 	/* actually scan through the events */
881 	error = kqueue_scan(fp, nevents, eventlist, timeout, retval, keops,
882 	    kevbuf, __arraycount(kevbuf));
883  done:
884 	fd_putfile(fd);
885 	return (error);
886 }
887 
888 /*
889  * Register a given kevent kev onto the kqueue
890  */
891 static int
892 kqueue_register(struct kqueue *kq, struct kevent *kev)
893 {
894 	struct kfilter *kfilter;
895 	filedesc_t *fdp;
896 	file_t *fp;
897 	fdfile_t *ff;
898 	struct knote *kn, *newkn;
899 	struct klist *list;
900 	int error, fd, rv;
901 
902 	fdp = kq->kq_fdp;
903 	fp = NULL;
904 	kn = NULL;
905 	error = 0;
906 	fd = 0;
907 
908 	newkn = kmem_zalloc(sizeof(*newkn), KM_SLEEP);
909 
910 	rw_enter(&kqueue_filter_lock, RW_READER);
911 	kfilter = kfilter_byfilter(kev->filter);
912 	if (kfilter == NULL || kfilter->filtops == NULL) {
913 		/* filter not found nor implemented */
914 		rw_exit(&kqueue_filter_lock);
915 		kmem_free(newkn, sizeof(*newkn));
916 		return (EINVAL);
917 	}
918 
919 	/* search if knote already exists */
920 	if (kfilter->filtops->f_isfd) {
921 		/* monitoring a file descriptor */
922 		fd = kev->ident;
923 		if ((fp = fd_getfile(fd)) == NULL) {
924 			rw_exit(&kqueue_filter_lock);
925 			kmem_free(newkn, sizeof(*newkn));
926 			return EBADF;
927 		}
928 		mutex_enter(&fdp->fd_lock);
929 		ff = fdp->fd_dt->dt_ff[fd];
930 		if (fd <= fdp->fd_lastkqfile) {
931 			SLIST_FOREACH(kn, &ff->ff_knlist, kn_link) {
932 				if (kq == kn->kn_kq &&
933 				    kev->filter == kn->kn_filter)
934 					break;
935 			}
936 		}
937 	} else {
938 		/*
939 		 * not monitoring a file descriptor, so
940 		 * lookup knotes in internal hash table
941 		 */
942 		mutex_enter(&fdp->fd_lock);
943 		if (fdp->fd_knhashmask != 0) {
944 			list = &fdp->fd_knhash[
945 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
946 			SLIST_FOREACH(kn, list, kn_link) {
947 				if (kev->ident == kn->kn_id &&
948 				    kq == kn->kn_kq &&
949 				    kev->filter == kn->kn_filter)
950 					break;
951 			}
952 		}
953 	}
954 
955 	/*
956 	 * kn now contains the matching knote, or NULL if no match
957 	 */
958 	if (kev->flags & EV_ADD) {
959 		if (kn == NULL) {
960 			/* create new knote */
961 			kn = newkn;
962 			newkn = NULL;
963 			kn->kn_obj = fp;
964 			kn->kn_kq = kq;
965 			kn->kn_fop = kfilter->filtops;
966 			kn->kn_kfilter = kfilter;
967 			kn->kn_sfflags = kev->fflags;
968 			kn->kn_sdata = kev->data;
969 			kev->fflags = 0;
970 			kev->data = 0;
971 			kn->kn_kevent = *kev;
972 
973 			/*
974 			 * apply reference count to knote structure, and
975 			 * do not release it at the end of this routine.
976 			 */
977 			fp = NULL;
978 
979 			if (!kn->kn_fop->f_isfd) {
980 				/*
981 				 * If knote is not on an fd, store on
982 				 * internal hash table.
983 				 */
984 				if (fdp->fd_knhashmask == 0) {
985 					/* XXXAD can block with fd_lock held */
986 					fdp->fd_knhash = hashinit(KN_HASHSIZE,
987 					    HASH_LIST, true,
988 					    &fdp->fd_knhashmask);
989 				}
990 				list = &fdp->fd_knhash[KN_HASH(kn->kn_id,
991 				    fdp->fd_knhashmask)];
992 			} else {
993 				/* Otherwise, knote is on an fd. */
994 				list = (struct klist *)
995 				    &fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist;
996 				if ((int)kn->kn_id > fdp->fd_lastkqfile)
997 					fdp->fd_lastkqfile = kn->kn_id;
998 			}
999 			SLIST_INSERT_HEAD(list, kn, kn_link);
1000 
1001 			KERNEL_LOCK(1, NULL);		/* XXXSMP */
1002 			error = (*kfilter->filtops->f_attach)(kn);
1003 			KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1004 			if (error != 0) {
1005 				/* knote_detach() drops fdp->fd_lock */
1006 				knote_detach(kn, fdp, false);
1007 				goto done;
1008 			}
1009 			atomic_inc_uint(&kfilter->refcnt);
1010 		} else {
1011 			/*
1012 			 * The user may change some filter values after the
1013 			 * initial EV_ADD, but doing so will not reset any
1014 			 * filter which have already been triggered.
1015 			 */
1016 			kn->kn_sfflags = kev->fflags;
1017 			kn->kn_sdata = kev->data;
1018 			kn->kn_kevent.udata = kev->udata;
1019 		}
1020 		KERNEL_LOCK(1, NULL);			/* XXXSMP */
1021 		rv = (*kn->kn_fop->f_event)(kn, 0);
1022 		KERNEL_UNLOCK_ONE(NULL);		/* XXXSMP */
1023 		if (rv)
1024 			knote_activate(kn);
1025 	} else {
1026 		if (kn == NULL) {
1027 			error = ENOENT;
1028 		 	mutex_exit(&fdp->fd_lock);
1029 			goto done;
1030 		}
1031 		if (kev->flags & EV_DELETE) {
1032 			/* knote_detach() drops fdp->fd_lock */
1033 			knote_detach(kn, fdp, true);
1034 			goto done;
1035 		}
1036 	}
1037 
1038 	/* disable knote */
1039 	if ((kev->flags & EV_DISABLE)) {
1040 		mutex_spin_enter(&kq->kq_lock);
1041 		if ((kn->kn_status & KN_DISABLED) == 0)
1042 			kn->kn_status |= KN_DISABLED;
1043 		mutex_spin_exit(&kq->kq_lock);
1044 	}
1045 
1046 	/* enable knote */
1047 	if ((kev->flags & EV_ENABLE)) {
1048 		knote_enqueue(kn);
1049 	}
1050 	mutex_exit(&fdp->fd_lock);
1051  done:
1052 	rw_exit(&kqueue_filter_lock);
1053 	if (newkn != NULL)
1054 		kmem_free(newkn, sizeof(*newkn));
1055 	if (fp != NULL)
1056 		fd_putfile(fd);
1057 	return (error);
1058 }
1059 
1060 #if defined(DEBUG)
1061 static void
1062 kq_check(struct kqueue *kq)
1063 {
1064 	const struct knote *kn;
1065 	int count;
1066 	int nmarker;
1067 
1068 	KASSERT(mutex_owned(&kq->kq_lock));
1069 	KASSERT(kq->kq_count >= 0);
1070 
1071 	count = 0;
1072 	nmarker = 0;
1073 	TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) {
1074 		if ((kn->kn_status & (KN_MARKER | KN_QUEUED)) == 0) {
1075 			panic("%s: kq=%p kn=%p inconsist 1", __func__, kq, kn);
1076 		}
1077 		if ((kn->kn_status & KN_MARKER) == 0) {
1078 			if (kn->kn_kq != kq) {
1079 				panic("%s: kq=%p kn=%p inconsist 2",
1080 				    __func__, kq, kn);
1081 			}
1082 			if ((kn->kn_status & KN_ACTIVE) == 0) {
1083 				panic("%s: kq=%p kn=%p: not active",
1084 				    __func__, kq, kn);
1085 			}
1086 			count++;
1087 			if (count > kq->kq_count) {
1088 				goto bad;
1089 			}
1090 		} else {
1091 			nmarker++;
1092 #if 0
1093 			if (nmarker > 10000) {
1094 				panic("%s: kq=%p too many markers: %d != %d, "
1095 				    "nmarker=%d",
1096 				    __func__, kq, kq->kq_count, count, nmarker);
1097 			}
1098 #endif
1099 		}
1100 	}
1101 	if (kq->kq_count != count) {
1102 bad:
1103 		panic("%s: kq=%p inconsist 3: %d != %d, nmarker=%d",
1104 		    __func__, kq, kq->kq_count, count, nmarker);
1105 	}
1106 }
1107 #else /* defined(DEBUG) */
1108 #define	kq_check(a)	/* nothing */
1109 #endif /* defined(DEBUG) */
1110 
1111 /*
1112  * Scan through the list of events on fp (for a maximum of maxevents),
1113  * returning the results in to ulistp. Timeout is determined by tsp; if
1114  * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait
1115  * as appropriate.
1116  */
1117 static int
1118 kqueue_scan(file_t *fp, size_t maxevents, struct kevent *ulistp,
1119 	    const struct timespec *tsp, register_t *retval,
1120 	    const struct kevent_ops *keops, struct kevent *kevbuf,
1121 	    size_t kevcnt)
1122 {
1123 	struct kqueue	*kq;
1124 	struct kevent	*kevp;
1125 	struct timespec	ats, sleepts;
1126 	struct knote	*kn, *marker;
1127 	size_t		count, nkev, nevents;
1128 	int		timeout, error, rv;
1129 	filedesc_t	*fdp;
1130 
1131 	fdp = curlwp->l_fd;
1132 	kq = fp->f_data;
1133 	count = maxevents;
1134 	nkev = nevents = error = 0;
1135 	if (count == 0) {
1136 		*retval = 0;
1137 		return 0;
1138 	}
1139 
1140 	if (tsp) {				/* timeout supplied */
1141 		ats = *tsp;
1142 		if (inittimeleft(&ats, &sleepts) == -1) {
1143 			*retval = maxevents;
1144 			return EINVAL;
1145 		}
1146 		timeout = tstohz(&ats);
1147 		if (timeout <= 0)
1148 			timeout = -1;           /* do poll */
1149 	} else {
1150 		/* no timeout, wait forever */
1151 		timeout = 0;
1152 	}
1153 
1154 	marker = kmem_zalloc(sizeof(*marker), KM_SLEEP);
1155 	marker->kn_status = KN_MARKER;
1156 	mutex_spin_enter(&kq->kq_lock);
1157  retry:
1158 	kevp = kevbuf;
1159 	if (kq->kq_count == 0) {
1160 		if (timeout >= 0) {
1161 			error = cv_timedwait_sig(&kq->kq_cv,
1162 			    &kq->kq_lock, timeout);
1163 			if (error == 0) {
1164 				 if (tsp == NULL || (timeout =
1165 				     gettimeleft(&ats, &sleepts)) > 0)
1166 					goto retry;
1167 			} else {
1168 				/* don't restart after signals... */
1169 				if (error == ERESTART)
1170 					error = EINTR;
1171 				if (error == EWOULDBLOCK)
1172 					error = 0;
1173 			}
1174 		}
1175 	} else {
1176 		/* mark end of knote list */
1177 		TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1178 
1179 		while (count != 0) {
1180 			kn = TAILQ_FIRST(&kq->kq_head);	/* get next knote */
1181 			while ((kn->kn_status & KN_MARKER) != 0) {
1182 				if (kn == marker) {
1183 					/* it's our marker, stop */
1184 					TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1185 					if (count < maxevents || (tsp != NULL &&
1186 					    (timeout = gettimeleft(&ats,
1187 					    &sleepts)) <= 0))
1188 						goto done;
1189 					goto retry;
1190 				}
1191 				/* someone else's marker. */
1192 				kn = TAILQ_NEXT(kn, kn_tqe);
1193 			}
1194 			kq_check(kq);
1195 			TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1196 			kq->kq_count--;
1197 			kn->kn_status &= ~KN_QUEUED;
1198 			kq_check(kq);
1199 			if (kn->kn_status & KN_DISABLED) {
1200 				/* don't want disabled events */
1201 				continue;
1202 			}
1203 			if ((kn->kn_flags & EV_ONESHOT) == 0) {
1204 				mutex_spin_exit(&kq->kq_lock);
1205 				KERNEL_LOCK(1, NULL);		/* XXXSMP */
1206 				rv = (*kn->kn_fop->f_event)(kn, 0);
1207 				KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1208 				mutex_spin_enter(&kq->kq_lock);
1209 				/* Re-poll if note was re-enqueued. */
1210 				if ((kn->kn_status & KN_QUEUED) != 0)
1211 					continue;
1212 				if (rv == 0) {
1213 					/*
1214 					 * non-ONESHOT event that hasn't
1215 					 * triggered again, so de-queue.
1216 					 */
1217 					kn->kn_status &= ~KN_ACTIVE;
1218 					continue;
1219 				}
1220 			}
1221 			/* XXXAD should be got from f_event if !oneshot. */
1222 			*kevp++ = kn->kn_kevent;
1223 			nkev++;
1224 			if (kn->kn_flags & EV_ONESHOT) {
1225 				/* delete ONESHOT events after retrieval */
1226 				mutex_spin_exit(&kq->kq_lock);
1227 				mutex_enter(&fdp->fd_lock);
1228 				knote_detach(kn, fdp, true);
1229 				mutex_spin_enter(&kq->kq_lock);
1230 			} else if (kn->kn_flags & EV_CLEAR) {
1231 				/* clear state after retrieval */
1232 				kn->kn_data = 0;
1233 				kn->kn_fflags = 0;
1234 				kn->kn_status &= ~KN_ACTIVE;
1235 			} else {
1236 				/* add event back on list */
1237 				kq_check(kq);
1238 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1239 				kq->kq_count++;
1240 				kn->kn_status |= KN_QUEUED;
1241 				kq_check(kq);
1242 			}
1243 			if (nkev == kevcnt) {
1244 				/* do copyouts in kevcnt chunks */
1245 				mutex_spin_exit(&kq->kq_lock);
1246 				error = (*keops->keo_put_events)
1247 				    (keops->keo_private,
1248 				    kevbuf, ulistp, nevents, nkev);
1249 				mutex_spin_enter(&kq->kq_lock);
1250 				nevents += nkev;
1251 				nkev = 0;
1252 				kevp = kevbuf;
1253 			}
1254 			count--;
1255 			if (error != 0 || count == 0) {
1256 				/* remove marker */
1257 				TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1258 				break;
1259 			}
1260 		}
1261 	}
1262  done:
1263  	mutex_spin_exit(&kq->kq_lock);
1264 	if (marker != NULL)
1265 		kmem_free(marker, sizeof(*marker));
1266 	if (nkev != 0) {
1267 		/* copyout remaining events */
1268 		error = (*keops->keo_put_events)(keops->keo_private,
1269 		    kevbuf, ulistp, nevents, nkev);
1270 	}
1271 	*retval = maxevents - count;
1272 
1273 	return error;
1274 }
1275 
1276 /*
1277  * fileops ioctl method for a kqueue descriptor.
1278  *
1279  * Two ioctls are currently supported. They both use struct kfilter_mapping:
1280  *	KFILTER_BYNAME		find name for filter, and return result in
1281  *				name, which is of size len.
1282  *	KFILTER_BYFILTER	find filter for name. len is ignored.
1283  */
1284 /*ARGSUSED*/
1285 static int
1286 kqueue_ioctl(file_t *fp, u_long com, void *data)
1287 {
1288 	struct kfilter_mapping	*km;
1289 	const struct kfilter	*kfilter;
1290 	char			*name;
1291 	int			error;
1292 
1293 	km = data;
1294 	error = 0;
1295 	name = kmem_alloc(KFILTER_MAXNAME, KM_SLEEP);
1296 
1297 	switch (com) {
1298 	case KFILTER_BYFILTER:	/* convert filter -> name */
1299 		rw_enter(&kqueue_filter_lock, RW_READER);
1300 		kfilter = kfilter_byfilter(km->filter);
1301 		if (kfilter != NULL) {
1302 			strlcpy(name, kfilter->name, KFILTER_MAXNAME);
1303 			rw_exit(&kqueue_filter_lock);
1304 			error = copyoutstr(name, km->name, km->len, NULL);
1305 		} else {
1306 			rw_exit(&kqueue_filter_lock);
1307 			error = ENOENT;
1308 		}
1309 		break;
1310 
1311 	case KFILTER_BYNAME:	/* convert name -> filter */
1312 		error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL);
1313 		if (error) {
1314 			break;
1315 		}
1316 		rw_enter(&kqueue_filter_lock, RW_READER);
1317 		kfilter = kfilter_byname(name);
1318 		if (kfilter != NULL)
1319 			km->filter = kfilter->filter;
1320 		else
1321 			error = ENOENT;
1322 		rw_exit(&kqueue_filter_lock);
1323 		break;
1324 
1325 	default:
1326 		error = ENOTTY;
1327 		break;
1328 
1329 	}
1330 	kmem_free(name, KFILTER_MAXNAME);
1331 	return (error);
1332 }
1333 
1334 /*
1335  * fileops fcntl method for a kqueue descriptor.
1336  */
1337 static int
1338 kqueue_fcntl(file_t *fp, u_int com, void *data)
1339 {
1340 
1341 	return (ENOTTY);
1342 }
1343 
1344 /*
1345  * fileops poll method for a kqueue descriptor.
1346  * Determine if kqueue has events pending.
1347  */
1348 static int
1349 kqueue_poll(file_t *fp, int events)
1350 {
1351 	struct kqueue	*kq;
1352 	int		revents;
1353 
1354 	kq = fp->f_data;
1355 
1356 	revents = 0;
1357 	if (events & (POLLIN | POLLRDNORM)) {
1358 		mutex_spin_enter(&kq->kq_lock);
1359 		if (kq->kq_count != 0) {
1360 			revents |= events & (POLLIN | POLLRDNORM);
1361 		} else {
1362 			selrecord(curlwp, &kq->kq_sel);
1363 		}
1364 		kq_check(kq);
1365 		mutex_spin_exit(&kq->kq_lock);
1366 	}
1367 
1368 	return revents;
1369 }
1370 
1371 /*
1372  * fileops stat method for a kqueue descriptor.
1373  * Returns dummy info, with st_size being number of events pending.
1374  */
1375 static int
1376 kqueue_stat(file_t *fp, struct stat *st)
1377 {
1378 	struct kqueue *kq;
1379 
1380 	kq = fp->f_data;
1381 
1382 	memset(st, 0, sizeof(*st));
1383 	st->st_size = kq->kq_count;
1384 	st->st_blksize = sizeof(struct kevent);
1385 	st->st_mode = S_IFIFO;
1386 
1387 	return 0;
1388 }
1389 
1390 static void
1391 kqueue_doclose(struct kqueue *kq, struct klist *list, int fd)
1392 {
1393 	struct knote *kn;
1394 	filedesc_t *fdp;
1395 
1396 	fdp = kq->kq_fdp;
1397 
1398 	KASSERT(mutex_owned(&fdp->fd_lock));
1399 
1400 	for (kn = SLIST_FIRST(list); kn != NULL;) {
1401 		if (kq != kn->kn_kq) {
1402 			kn = SLIST_NEXT(kn, kn_link);
1403 			continue;
1404 		}
1405 		knote_detach(kn, fdp, true);
1406 		mutex_enter(&fdp->fd_lock);
1407 		kn = SLIST_FIRST(list);
1408 	}
1409 }
1410 
1411 
1412 /*
1413  * fileops close method for a kqueue descriptor.
1414  */
1415 static int
1416 kqueue_close(file_t *fp)
1417 {
1418 	struct kqueue *kq;
1419 	filedesc_t *fdp;
1420 	fdfile_t *ff;
1421 	int i;
1422 
1423 	kq = fp->f_data;
1424 	fdp = curlwp->l_fd;
1425 
1426 	mutex_enter(&fdp->fd_lock);
1427 	for (i = 0; i <= fdp->fd_lastkqfile; i++) {
1428 		if ((ff = fdp->fd_dt->dt_ff[i]) == NULL)
1429 			continue;
1430 		kqueue_doclose(kq, (struct klist *)&ff->ff_knlist, i);
1431 	}
1432 	if (fdp->fd_knhashmask != 0) {
1433 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
1434 			kqueue_doclose(kq, &fdp->fd_knhash[i], -1);
1435 		}
1436 	}
1437 	mutex_exit(&fdp->fd_lock);
1438 
1439 	KASSERT(kq->kq_count == 0);
1440 	mutex_destroy(&kq->kq_lock);
1441 	cv_destroy(&kq->kq_cv);
1442 	seldestroy(&kq->kq_sel);
1443 	kmem_free(kq, sizeof(*kq));
1444 	fp->f_data = NULL;
1445 
1446 	return (0);
1447 }
1448 
1449 /*
1450  * struct fileops kqfilter method for a kqueue descriptor.
1451  * Event triggered when monitored kqueue changes.
1452  */
1453 static int
1454 kqueue_kqfilter(file_t *fp, struct knote *kn)
1455 {
1456 	struct kqueue *kq;
1457 
1458 	kq = ((file_t *)kn->kn_obj)->f_data;
1459 
1460 	KASSERT(fp == kn->kn_obj);
1461 
1462 	if (kn->kn_filter != EVFILT_READ)
1463 		return 1;
1464 
1465 	kn->kn_fop = &kqread_filtops;
1466 	mutex_enter(&kq->kq_lock);
1467 	SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext);
1468 	mutex_exit(&kq->kq_lock);
1469 
1470 	return 0;
1471 }
1472 
1473 
1474 /*
1475  * Walk down a list of knotes, activating them if their event has
1476  * triggered.  The caller's object lock (e.g. device driver lock)
1477  * must be held.
1478  */
1479 void
1480 knote(struct klist *list, long hint)
1481 {
1482 	struct knote *kn, *tmpkn;
1483 
1484 	SLIST_FOREACH_SAFE(kn, list, kn_selnext, tmpkn) {
1485 		if ((*kn->kn_fop->f_event)(kn, hint))
1486 			knote_activate(kn);
1487 	}
1488 }
1489 
1490 /*
1491  * Remove all knotes referencing a specified fd
1492  */
1493 void
1494 knote_fdclose(int fd)
1495 {
1496 	struct klist *list;
1497 	struct knote *kn;
1498 	filedesc_t *fdp;
1499 
1500 	fdp = curlwp->l_fd;
1501 	list = (struct klist *)&fdp->fd_dt->dt_ff[fd]->ff_knlist;
1502 	mutex_enter(&fdp->fd_lock);
1503 	while ((kn = SLIST_FIRST(list)) != NULL) {
1504 		knote_detach(kn, fdp, true);
1505 		mutex_enter(&fdp->fd_lock);
1506 	}
1507 	mutex_exit(&fdp->fd_lock);
1508 }
1509 
1510 /*
1511  * Drop knote.  Called with fdp->fd_lock held, and will drop before
1512  * returning.
1513  */
1514 static void
1515 knote_detach(struct knote *kn, filedesc_t *fdp, bool dofop)
1516 {
1517 	struct klist *list;
1518 	struct kqueue *kq;
1519 
1520 	kq = kn->kn_kq;
1521 
1522 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1523 	KASSERT(mutex_owned(&fdp->fd_lock));
1524 
1525 	/* Remove from monitored object. */
1526 	if (dofop) {
1527 		KERNEL_LOCK(1, NULL);		/* XXXSMP */
1528 		(*kn->kn_fop->f_detach)(kn);
1529 		KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1530 	}
1531 
1532 	/* Remove from descriptor table. */
1533 	if (kn->kn_fop->f_isfd)
1534 		list = (struct klist *)&fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist;
1535 	else
1536 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1537 
1538 	SLIST_REMOVE(list, kn, knote, kn_link);
1539 
1540 	/* Remove from kqueue. */
1541 	/* XXXAD should verify not in use by kqueue_scan. */
1542 	mutex_spin_enter(&kq->kq_lock);
1543 	if ((kn->kn_status & KN_QUEUED) != 0) {
1544 		kq_check(kq);
1545 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1546 		kn->kn_status &= ~KN_QUEUED;
1547 		kq->kq_count--;
1548 		kq_check(kq);
1549 	}
1550 	mutex_spin_exit(&kq->kq_lock);
1551 
1552 	mutex_exit(&fdp->fd_lock);
1553 	if (kn->kn_fop->f_isfd)
1554 		fd_putfile(kn->kn_id);
1555 	atomic_dec_uint(&kn->kn_kfilter->refcnt);
1556 	kmem_free(kn, sizeof(*kn));
1557 }
1558 
1559 /*
1560  * Queue new event for knote.
1561  */
1562 static void
1563 knote_enqueue(struct knote *kn)
1564 {
1565 	struct kqueue *kq;
1566 
1567 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1568 
1569 	kq = kn->kn_kq;
1570 
1571 	mutex_spin_enter(&kq->kq_lock);
1572 	if ((kn->kn_status & KN_DISABLED) != 0) {
1573 		kn->kn_status &= ~KN_DISABLED;
1574 	}
1575 	if ((kn->kn_status & (KN_ACTIVE | KN_QUEUED)) == KN_ACTIVE) {
1576 		kq_check(kq);
1577 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1578 		kn->kn_status |= KN_QUEUED;
1579 		kq->kq_count++;
1580 		kq_check(kq);
1581 		cv_broadcast(&kq->kq_cv);
1582 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1583 	}
1584 	mutex_spin_exit(&kq->kq_lock);
1585 }
1586 /*
1587  * Queue new event for knote.
1588  */
1589 static void
1590 knote_activate(struct knote *kn)
1591 {
1592 	struct kqueue *kq;
1593 
1594 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1595 
1596 	kq = kn->kn_kq;
1597 
1598 	mutex_spin_enter(&kq->kq_lock);
1599 	kn->kn_status |= KN_ACTIVE;
1600 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) {
1601 		kq_check(kq);
1602 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1603 		kn->kn_status |= KN_QUEUED;
1604 		kq->kq_count++;
1605 		kq_check(kq);
1606 		cv_broadcast(&kq->kq_cv);
1607 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1608 	}
1609 	mutex_spin_exit(&kq->kq_lock);
1610 }
1611