xref: /netbsd-src/sys/kern/kern_event.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: kern_event.c,v 1.84 2015/12/08 14:52:06 christos 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 1.84 2015/12/08 14:52:06 christos 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_kqueue;
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_kqueue;
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;
454 	struct lwp *curl;
455 
456 	curl = curlwp;
457 
458 	mutex_enter(proc_lock);
459 	if (kn->kn_flags & EV_FLAG1) {
460 		/*
461 		 * NOTE_TRACK attaches to the child process too early
462 		 * for proc_find, so do a raw look up and check the state
463 		 * explicitly.
464 		 */
465 		p = proc_find_raw(kn->kn_id);
466 		if (p != NULL && p->p_stat != SIDL)
467 			p = NULL;
468 	} else {
469 		p = proc_find(kn->kn_id);
470 	}
471 
472 	if (p == NULL) {
473 		mutex_exit(proc_lock);
474 		return ESRCH;
475 	}
476 
477 	/*
478 	 * Fail if it's not owned by you, or the last exec gave us
479 	 * setuid/setgid privs (unless you're root).
480 	 */
481 	mutex_enter(p->p_lock);
482 	mutex_exit(proc_lock);
483 	if (kauth_authorize_process(curl->l_cred, KAUTH_PROCESS_KEVENT_FILTER,
484 	    p, NULL, NULL, NULL) != 0) {
485 	    	mutex_exit(p->p_lock);
486 		return EACCES;
487 	}
488 
489 	kn->kn_obj = p;
490 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
491 
492 	/*
493 	 * internal flag indicating registration done by kernel
494 	 */
495 	if (kn->kn_flags & EV_FLAG1) {
496 		kn->kn_data = kn->kn_sdata;	/* ppid */
497 		kn->kn_fflags = NOTE_CHILD;
498 		kn->kn_flags &= ~EV_FLAG1;
499 	}
500 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
501     	mutex_exit(p->p_lock);
502 
503 	return 0;
504 }
505 
506 /*
507  * Filter detach method for EVFILT_PROC.
508  *
509  * The knote may be attached to a different process, which may exit,
510  * leaving nothing for the knote to be attached to.  So when the process
511  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
512  * it will be deleted when read out.  However, as part of the knote deletion,
513  * this routine is called, so a check is needed to avoid actually performing
514  * a detach, because the original process might not exist any more.
515  */
516 static void
517 filt_procdetach(struct knote *kn)
518 {
519 	struct proc *p;
520 
521 	if (kn->kn_status & KN_DETACHED)
522 		return;
523 
524 	p = kn->kn_obj;
525 
526 	mutex_enter(p->p_lock);
527 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
528 	mutex_exit(p->p_lock);
529 }
530 
531 /*
532  * Filter event method for EVFILT_PROC.
533  */
534 static int
535 filt_proc(struct knote *kn, long hint)
536 {
537 	u_int event, fflag;
538 	struct kevent kev;
539 	struct kqueue *kq;
540 	int error;
541 
542 	event = (u_int)hint & NOTE_PCTRLMASK;
543 	kq = kn->kn_kq;
544 	fflag = 0;
545 
546 	/* If the user is interested in this event, record it. */
547 	if (kn->kn_sfflags & event)
548 		fflag |= event;
549 
550 	if (event == NOTE_EXIT) {
551 		struct proc *p = kn->kn_obj;
552 
553 		if (p != NULL)
554 			kn->kn_data = p->p_xstat;
555 		/*
556 		 * Process is gone, so flag the event as finished.
557 		 *
558 		 * Detach the knote from watched process and mark
559 		 * it as such. We can't leave this to kqueue_scan(),
560 		 * since the process might not exist by then. And we
561 		 * have to do this now, since psignal KNOTE() is called
562 		 * also for zombies and we might end up reading freed
563 		 * memory if the kevent would already be picked up
564 		 * and knote g/c'ed.
565 		 */
566 		filt_procdetach(kn);
567 
568 		mutex_spin_enter(&kq->kq_lock);
569 		kn->kn_status |= KN_DETACHED;
570 		/* Mark as ONESHOT, so that the knote it g/c'ed when read */
571 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
572 		kn->kn_fflags |= fflag;
573 		mutex_spin_exit(&kq->kq_lock);
574 
575 		return 1;
576 	}
577 
578 	mutex_spin_enter(&kq->kq_lock);
579 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
580 		/*
581 		 * Process forked, and user wants to track the new process,
582 		 * so attach a new knote to it, and immediately report an
583 		 * event with the parent's pid.  Register knote with new
584 		 * process.
585 		 */
586 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
587 		kev.filter = kn->kn_filter;
588 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
589 		kev.fflags = kn->kn_sfflags;
590 		kev.data = kn->kn_id;			/* parent */
591 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
592 		mutex_spin_exit(&kq->kq_lock);
593 		error = kqueue_register(kq, &kev);
594 		mutex_spin_enter(&kq->kq_lock);
595 		if (error != 0)
596 			kn->kn_fflags |= NOTE_TRACKERR;
597 	}
598 	kn->kn_fflags |= fflag;
599 	fflag = kn->kn_fflags;
600 	mutex_spin_exit(&kq->kq_lock);
601 
602 	return fflag != 0;
603 }
604 
605 static void
606 filt_timerexpire(void *knx)
607 {
608 	struct knote *kn = knx;
609 	int tticks;
610 
611 	mutex_enter(&kqueue_misc_lock);
612 	kn->kn_data++;
613 	knote_activate(kn);
614 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
615 		tticks = mstohz(kn->kn_sdata);
616 		if (tticks <= 0)
617 			tticks = 1;
618 		callout_schedule((callout_t *)kn->kn_hook, tticks);
619 	}
620 	mutex_exit(&kqueue_misc_lock);
621 }
622 
623 /*
624  * data contains amount of time to sleep, in milliseconds
625  */
626 static int
627 filt_timerattach(struct knote *kn)
628 {
629 	callout_t *calloutp;
630 	struct kqueue *kq;
631 	int tticks;
632 
633 	tticks = mstohz(kn->kn_sdata);
634 
635 	/* if the supplied value is under our resolution, use 1 tick */
636 	if (tticks == 0) {
637 		if (kn->kn_sdata == 0)
638 			return EINVAL;
639 		tticks = 1;
640 	}
641 
642 	if (atomic_inc_uint_nv(&kq_ncallouts) >= kq_calloutmax ||
643 	    (calloutp = kmem_alloc(sizeof(*calloutp), KM_NOSLEEP)) == NULL) {
644 		atomic_dec_uint(&kq_ncallouts);
645 		return ENOMEM;
646 	}
647 	callout_init(calloutp, CALLOUT_MPSAFE);
648 
649 	kq = kn->kn_kq;
650 	mutex_spin_enter(&kq->kq_lock);
651 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
652 	kn->kn_hook = calloutp;
653 	mutex_spin_exit(&kq->kq_lock);
654 
655 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
656 
657 	return (0);
658 }
659 
660 static void
661 filt_timerdetach(struct knote *kn)
662 {
663 	callout_t *calloutp;
664 
665 	calloutp = (callout_t *)kn->kn_hook;
666 	callout_halt(calloutp, NULL);
667 	callout_destroy(calloutp);
668 	kmem_free(calloutp, sizeof(*calloutp));
669 	atomic_dec_uint(&kq_ncallouts);
670 }
671 
672 static int
673 filt_timer(struct knote *kn, long hint)
674 {
675 	int rv;
676 
677 	mutex_enter(&kqueue_misc_lock);
678 	rv = (kn->kn_data != 0);
679 	mutex_exit(&kqueue_misc_lock);
680 
681 	return rv;
682 }
683 
684 /*
685  * filt_seltrue:
686  *
687  *	This filter "event" routine simulates seltrue().
688  */
689 int
690 filt_seltrue(struct knote *kn, long hint)
691 {
692 
693 	/*
694 	 * We don't know how much data can be read/written,
695 	 * but we know that it *can* be.  This is about as
696 	 * good as select/poll does as well.
697 	 */
698 	kn->kn_data = 0;
699 	return (1);
700 }
701 
702 /*
703  * This provides full kqfilter entry for device switch tables, which
704  * has same effect as filter using filt_seltrue() as filter method.
705  */
706 static void
707 filt_seltruedetach(struct knote *kn)
708 {
709 	/* Nothing to do */
710 }
711 
712 const struct filterops seltrue_filtops =
713 	{ 1, NULL, filt_seltruedetach, filt_seltrue };
714 
715 int
716 seltrue_kqfilter(dev_t dev, struct knote *kn)
717 {
718 	switch (kn->kn_filter) {
719 	case EVFILT_READ:
720 	case EVFILT_WRITE:
721 		kn->kn_fop = &seltrue_filtops;
722 		break;
723 	default:
724 		return (EINVAL);
725 	}
726 
727 	/* Nothing more to do */
728 	return (0);
729 }
730 
731 /*
732  * kqueue(2) system call.
733  */
734 static int
735 kqueue1(struct lwp *l, int flags, register_t *retval)
736 {
737 	struct kqueue *kq;
738 	file_t *fp;
739 	int fd, error;
740 
741 	if ((error = fd_allocfile(&fp, &fd)) != 0)
742 		return error;
743 	fp->f_flag = FREAD | FWRITE | (flags & (FNONBLOCK|FNOSIGPIPE));
744 	fp->f_type = DTYPE_KQUEUE;
745 	fp->f_ops = &kqueueops;
746 	kq = kmem_zalloc(sizeof(*kq), KM_SLEEP);
747 	mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED);
748 	cv_init(&kq->kq_cv, "kqueue");
749 	selinit(&kq->kq_sel);
750 	TAILQ_INIT(&kq->kq_head);
751 	fp->f_kqueue = kq;
752 	*retval = fd;
753 	kq->kq_fdp = curlwp->l_fd;
754 	fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0);
755 	fd_affix(curproc, fp, fd);
756 	return error;
757 }
758 
759 /*
760  * kqueue(2) system call.
761  */
762 int
763 sys_kqueue(struct lwp *l, const void *v, register_t *retval)
764 {
765 	return kqueue1(l, 0, retval);
766 }
767 
768 int
769 sys_kqueue1(struct lwp *l, const struct sys_kqueue1_args *uap,
770     register_t *retval)
771 {
772 	/* {
773 		syscallarg(int) flags;
774 	} */
775 	return kqueue1(l, SCARG(uap, flags), retval);
776 }
777 
778 /*
779  * kevent(2) system call.
780  */
781 int
782 kevent_fetch_changes(void *ctx, const struct kevent *changelist,
783     struct kevent *changes, size_t index, int n)
784 {
785 
786 	return copyin(changelist + index, changes, n * sizeof(*changes));
787 }
788 
789 int
790 kevent_put_events(void *ctx, struct kevent *events,
791     struct kevent *eventlist, size_t index, int n)
792 {
793 
794 	return copyout(events, eventlist + index, n * sizeof(*events));
795 }
796 
797 static const struct kevent_ops kevent_native_ops = {
798 	.keo_private = NULL,
799 	.keo_fetch_timeout = copyin,
800 	.keo_fetch_changes = kevent_fetch_changes,
801 	.keo_put_events = kevent_put_events,
802 };
803 
804 int
805 sys___kevent50(struct lwp *l, const struct sys___kevent50_args *uap,
806     register_t *retval)
807 {
808 	/* {
809 		syscallarg(int) fd;
810 		syscallarg(const struct kevent *) changelist;
811 		syscallarg(size_t) nchanges;
812 		syscallarg(struct kevent *) eventlist;
813 		syscallarg(size_t) nevents;
814 		syscallarg(const struct timespec *) timeout;
815 	} */
816 
817 	return kevent1(retval, SCARG(uap, fd), SCARG(uap, changelist),
818 	    SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents),
819 	    SCARG(uap, timeout), &kevent_native_ops);
820 }
821 
822 int
823 kevent1(register_t *retval, int fd,
824 	const struct kevent *changelist, size_t nchanges,
825 	struct kevent *eventlist, size_t nevents,
826 	const struct timespec *timeout,
827 	const struct kevent_ops *keops)
828 {
829 	struct kevent *kevp;
830 	struct kqueue *kq;
831 	struct timespec	ts;
832 	size_t i, n, ichange;
833 	int nerrors, error;
834 	struct kevent kevbuf[KQ_NEVENTS];	/* approx 300 bytes on 64-bit */
835 	file_t *fp;
836 
837 	/* check that we're dealing with a kq */
838 	fp = fd_getfile(fd);
839 	if (fp == NULL)
840 		return (EBADF);
841 
842 	if (fp->f_type != DTYPE_KQUEUE) {
843 		fd_putfile(fd);
844 		return (EBADF);
845 	}
846 
847 	if (timeout != NULL) {
848 		error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts));
849 		if (error)
850 			goto done;
851 		timeout = &ts;
852 	}
853 
854 	kq = fp->f_kqueue;
855 	nerrors = 0;
856 	ichange = 0;
857 
858 	/* traverse list of events to register */
859 	while (nchanges > 0) {
860 		n = MIN(nchanges, __arraycount(kevbuf));
861 		error = (*keops->keo_fetch_changes)(keops->keo_private,
862 		    changelist, kevbuf, ichange, n);
863 		if (error)
864 			goto done;
865 		for (i = 0; i < n; i++) {
866 			kevp = &kevbuf[i];
867 			kevp->flags &= ~EV_SYSFLAGS;
868 			/* register each knote */
869 			error = kqueue_register(kq, kevp);
870 			if (error || (kevp->flags & EV_RECEIPT)) {
871 				if (nevents != 0) {
872 					kevp->flags = EV_ERROR;
873 					kevp->data = error;
874 					error = (*keops->keo_put_events)
875 					    (keops->keo_private, kevp,
876 					    eventlist, nerrors, 1);
877 					if (error)
878 						goto done;
879 					nevents--;
880 					nerrors++;
881 				} else {
882 					goto done;
883 				}
884 			}
885 		}
886 		nchanges -= n;	/* update the results */
887 		ichange += n;
888 	}
889 	if (nerrors) {
890 		*retval = nerrors;
891 		error = 0;
892 		goto done;
893 	}
894 
895 	/* actually scan through the events */
896 	error = kqueue_scan(fp, nevents, eventlist, timeout, retval, keops,
897 	    kevbuf, __arraycount(kevbuf));
898  done:
899 	fd_putfile(fd);
900 	return (error);
901 }
902 
903 /*
904  * Register a given kevent kev onto the kqueue
905  */
906 static int
907 kqueue_register(struct kqueue *kq, struct kevent *kev)
908 {
909 	struct kfilter *kfilter;
910 	filedesc_t *fdp;
911 	file_t *fp;
912 	fdfile_t *ff;
913 	struct knote *kn, *newkn;
914 	struct klist *list;
915 	int error, fd, rv;
916 
917 	fdp = kq->kq_fdp;
918 	fp = NULL;
919 	kn = NULL;
920 	error = 0;
921 	fd = 0;
922 
923 	newkn = kmem_zalloc(sizeof(*newkn), KM_SLEEP);
924 
925 	rw_enter(&kqueue_filter_lock, RW_READER);
926 	kfilter = kfilter_byfilter(kev->filter);
927 	if (kfilter == NULL || kfilter->filtops == NULL) {
928 		/* filter not found nor implemented */
929 		rw_exit(&kqueue_filter_lock);
930 		kmem_free(newkn, sizeof(*newkn));
931 		return (EINVAL);
932 	}
933 
934 	/* search if knote already exists */
935 	if (kfilter->filtops->f_isfd) {
936 		/* monitoring a file descriptor */
937 		fd = kev->ident;
938 		if ((fp = fd_getfile(fd)) == NULL) {
939 			rw_exit(&kqueue_filter_lock);
940 			kmem_free(newkn, sizeof(*newkn));
941 			return EBADF;
942 		}
943 		mutex_enter(&fdp->fd_lock);
944 		ff = fdp->fd_dt->dt_ff[fd];
945 		if (fd <= fdp->fd_lastkqfile) {
946 			SLIST_FOREACH(kn, &ff->ff_knlist, kn_link) {
947 				if (kq == kn->kn_kq &&
948 				    kev->filter == kn->kn_filter)
949 					break;
950 			}
951 		}
952 	} else {
953 		/*
954 		 * not monitoring a file descriptor, so
955 		 * lookup knotes in internal hash table
956 		 */
957 		mutex_enter(&fdp->fd_lock);
958 		if (fdp->fd_knhashmask != 0) {
959 			list = &fdp->fd_knhash[
960 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
961 			SLIST_FOREACH(kn, list, kn_link) {
962 				if (kev->ident == kn->kn_id &&
963 				    kq == kn->kn_kq &&
964 				    kev->filter == kn->kn_filter)
965 					break;
966 			}
967 		}
968 	}
969 
970 	/*
971 	 * kn now contains the matching knote, or NULL if no match
972 	 */
973 	if (kev->flags & EV_ADD) {
974 		if (kn == NULL) {
975 			/* create new knote */
976 			kn = newkn;
977 			newkn = NULL;
978 			kn->kn_obj = fp;
979 			kn->kn_id = kev->ident;
980 			kn->kn_kq = kq;
981 			kn->kn_fop = kfilter->filtops;
982 			kn->kn_kfilter = kfilter;
983 			kn->kn_sfflags = kev->fflags;
984 			kn->kn_sdata = kev->data;
985 			kev->fflags = 0;
986 			kev->data = 0;
987 			kn->kn_kevent = *kev;
988 
989 			/*
990 			 * apply reference count to knote structure, and
991 			 * do not release it at the end of this routine.
992 			 */
993 			fp = NULL;
994 
995 			if (!kn->kn_fop->f_isfd) {
996 				/*
997 				 * If knote is not on an fd, store on
998 				 * internal hash table.
999 				 */
1000 				if (fdp->fd_knhashmask == 0) {
1001 					/* XXXAD can block with fd_lock held */
1002 					fdp->fd_knhash = hashinit(KN_HASHSIZE,
1003 					    HASH_LIST, true,
1004 					    &fdp->fd_knhashmask);
1005 				}
1006 				list = &fdp->fd_knhash[KN_HASH(kn->kn_id,
1007 				    fdp->fd_knhashmask)];
1008 			} else {
1009 				/* Otherwise, knote is on an fd. */
1010 				list = (struct klist *)
1011 				    &fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist;
1012 				if ((int)kn->kn_id > fdp->fd_lastkqfile)
1013 					fdp->fd_lastkqfile = kn->kn_id;
1014 			}
1015 			SLIST_INSERT_HEAD(list, kn, kn_link);
1016 
1017 			KERNEL_LOCK(1, NULL);		/* XXXSMP */
1018 			error = (*kfilter->filtops->f_attach)(kn);
1019 			KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1020 			if (error != 0) {
1021 #ifdef DIAGNOSTIC
1022 				printf("%s: event not supported for file type"
1023 				    " %d\n", __func__, fp ? fp->f_type : -1);
1024 #endif
1025 				/* knote_detach() drops fdp->fd_lock */
1026 				knote_detach(kn, fdp, false);
1027 				goto done;
1028 			}
1029 			atomic_inc_uint(&kfilter->refcnt);
1030 		} else {
1031 			/*
1032 			 * The user may change some filter values after the
1033 			 * initial EV_ADD, but doing so will not reset any
1034 			 * filter which have already been triggered.
1035 			 */
1036 			kn->kn_sfflags = kev->fflags;
1037 			kn->kn_sdata = kev->data;
1038 			kn->kn_kevent.udata = kev->udata;
1039 		}
1040 		/*
1041 		 * We can get here if we are trying to attach
1042 		 * an event to a file descriptor that does not
1043 		 * support events, and the attach routine is
1044 		 * broken and does not return an error.
1045 		 */
1046 		KASSERT(kn->kn_fop->f_event != NULL);
1047 		KERNEL_LOCK(1, NULL);			/* XXXSMP */
1048 		rv = (*kn->kn_fop->f_event)(kn, 0);
1049 		KERNEL_UNLOCK_ONE(NULL);		/* XXXSMP */
1050 		if (rv)
1051 			knote_activate(kn);
1052 	} else {
1053 		if (kn == NULL) {
1054 			error = ENOENT;
1055 		 	mutex_exit(&fdp->fd_lock);
1056 			goto done;
1057 		}
1058 		if (kev->flags & EV_DELETE) {
1059 			/* knote_detach() drops fdp->fd_lock */
1060 			knote_detach(kn, fdp, true);
1061 			goto done;
1062 		}
1063 	}
1064 
1065 	/* disable knote */
1066 	if ((kev->flags & EV_DISABLE)) {
1067 		mutex_spin_enter(&kq->kq_lock);
1068 		if ((kn->kn_status & KN_DISABLED) == 0)
1069 			kn->kn_status |= KN_DISABLED;
1070 		mutex_spin_exit(&kq->kq_lock);
1071 	}
1072 
1073 	/* enable knote */
1074 	if ((kev->flags & EV_ENABLE)) {
1075 		knote_enqueue(kn);
1076 	}
1077 	mutex_exit(&fdp->fd_lock);
1078  done:
1079 	rw_exit(&kqueue_filter_lock);
1080 	if (newkn != NULL)
1081 		kmem_free(newkn, sizeof(*newkn));
1082 	if (fp != NULL)
1083 		fd_putfile(fd);
1084 	return (error);
1085 }
1086 
1087 #if defined(DEBUG)
1088 static void
1089 kq_check(struct kqueue *kq)
1090 {
1091 	const struct knote *kn;
1092 	int count;
1093 	int nmarker;
1094 
1095 	KASSERT(mutex_owned(&kq->kq_lock));
1096 	KASSERT(kq->kq_count >= 0);
1097 
1098 	count = 0;
1099 	nmarker = 0;
1100 	TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) {
1101 		if ((kn->kn_status & (KN_MARKER | KN_QUEUED)) == 0) {
1102 			panic("%s: kq=%p kn=%p inconsist 1", __func__, kq, kn);
1103 		}
1104 		if ((kn->kn_status & KN_MARKER) == 0) {
1105 			if (kn->kn_kq != kq) {
1106 				panic("%s: kq=%p kn=%p inconsist 2",
1107 				    __func__, kq, kn);
1108 			}
1109 			if ((kn->kn_status & KN_ACTIVE) == 0) {
1110 				panic("%s: kq=%p kn=%p: not active",
1111 				    __func__, kq, kn);
1112 			}
1113 			count++;
1114 			if (count > kq->kq_count) {
1115 				goto bad;
1116 			}
1117 		} else {
1118 			nmarker++;
1119 #if 0
1120 			if (nmarker > 10000) {
1121 				panic("%s: kq=%p too many markers: %d != %d, "
1122 				    "nmarker=%d",
1123 				    __func__, kq, kq->kq_count, count, nmarker);
1124 			}
1125 #endif
1126 		}
1127 	}
1128 	if (kq->kq_count != count) {
1129 bad:
1130 		panic("%s: kq=%p inconsist 3: %d != %d, nmarker=%d",
1131 		    __func__, kq, kq->kq_count, count, nmarker);
1132 	}
1133 }
1134 #else /* defined(DEBUG) */
1135 #define	kq_check(a)	/* nothing */
1136 #endif /* defined(DEBUG) */
1137 
1138 /*
1139  * Scan through the list of events on fp (for a maximum of maxevents),
1140  * returning the results in to ulistp. Timeout is determined by tsp; if
1141  * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait
1142  * as appropriate.
1143  */
1144 static int
1145 kqueue_scan(file_t *fp, size_t maxevents, struct kevent *ulistp,
1146 	    const struct timespec *tsp, register_t *retval,
1147 	    const struct kevent_ops *keops, struct kevent *kevbuf,
1148 	    size_t kevcnt)
1149 {
1150 	struct kqueue	*kq;
1151 	struct kevent	*kevp;
1152 	struct timespec	ats, sleepts;
1153 	struct knote	*kn, *marker;
1154 	size_t		count, nkev, nevents;
1155 	int		timeout, error, rv;
1156 	filedesc_t	*fdp;
1157 
1158 	fdp = curlwp->l_fd;
1159 	kq = fp->f_kqueue;
1160 	count = maxevents;
1161 	nkev = nevents = error = 0;
1162 	if (count == 0) {
1163 		*retval = 0;
1164 		return 0;
1165 	}
1166 
1167 	if (tsp) {				/* timeout supplied */
1168 		ats = *tsp;
1169 		if (inittimeleft(&ats, &sleepts) == -1) {
1170 			*retval = maxevents;
1171 			return EINVAL;
1172 		}
1173 		timeout = tstohz(&ats);
1174 		if (timeout <= 0)
1175 			timeout = -1;           /* do poll */
1176 	} else {
1177 		/* no timeout, wait forever */
1178 		timeout = 0;
1179 	}
1180 
1181 	marker = kmem_zalloc(sizeof(*marker), KM_SLEEP);
1182 	marker->kn_status = KN_MARKER;
1183 	mutex_spin_enter(&kq->kq_lock);
1184  retry:
1185 	kevp = kevbuf;
1186 	if (kq->kq_count == 0) {
1187 		if (timeout >= 0) {
1188 			error = cv_timedwait_sig(&kq->kq_cv,
1189 			    &kq->kq_lock, timeout);
1190 			if (error == 0) {
1191 				 if (tsp == NULL || (timeout =
1192 				     gettimeleft(&ats, &sleepts)) > 0)
1193 					goto retry;
1194 			} else {
1195 				/* don't restart after signals... */
1196 				if (error == ERESTART)
1197 					error = EINTR;
1198 				if (error == EWOULDBLOCK)
1199 					error = 0;
1200 			}
1201 		}
1202 	} else {
1203 		/* mark end of knote list */
1204 		TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1205 
1206 		while (count != 0) {
1207 			kn = TAILQ_FIRST(&kq->kq_head);	/* get next knote */
1208 			while ((kn->kn_status & KN_MARKER) != 0) {
1209 				if (kn == marker) {
1210 					/* it's our marker, stop */
1211 					TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1212 					if (count < maxevents || (tsp != NULL &&
1213 					    (timeout = gettimeleft(&ats,
1214 					    &sleepts)) <= 0))
1215 						goto done;
1216 					goto retry;
1217 				}
1218 				/* someone else's marker. */
1219 				kn = TAILQ_NEXT(kn, kn_tqe);
1220 			}
1221 			kq_check(kq);
1222 			TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1223 			kq->kq_count--;
1224 			kn->kn_status &= ~KN_QUEUED;
1225 			kq_check(kq);
1226 			if (kn->kn_status & KN_DISABLED) {
1227 				/* don't want disabled events */
1228 				continue;
1229 			}
1230 			if ((kn->kn_flags & EV_ONESHOT) == 0) {
1231 				mutex_spin_exit(&kq->kq_lock);
1232 				KASSERT(kn->kn_fop->f_event != NULL);
1233 				KERNEL_LOCK(1, NULL);		/* XXXSMP */
1234 				rv = (*kn->kn_fop->f_event)(kn, 0);
1235 				KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1236 				mutex_spin_enter(&kq->kq_lock);
1237 				/* Re-poll if note was re-enqueued. */
1238 				if ((kn->kn_status & KN_QUEUED) != 0)
1239 					continue;
1240 				if (rv == 0) {
1241 					/*
1242 					 * non-ONESHOT event that hasn't
1243 					 * triggered again, so de-queue.
1244 					 */
1245 					kn->kn_status &= ~KN_ACTIVE;
1246 					continue;
1247 				}
1248 			}
1249 			/* XXXAD should be got from f_event if !oneshot. */
1250 			*kevp++ = kn->kn_kevent;
1251 			nkev++;
1252 			if (kn->kn_flags & EV_ONESHOT) {
1253 				/* delete ONESHOT events after retrieval */
1254 				mutex_spin_exit(&kq->kq_lock);
1255 				mutex_enter(&fdp->fd_lock);
1256 				knote_detach(kn, fdp, true);
1257 				mutex_spin_enter(&kq->kq_lock);
1258 			} else if (kn->kn_flags & EV_CLEAR) {
1259 				/* clear state after retrieval */
1260 				kn->kn_data = 0;
1261 				kn->kn_fflags = 0;
1262 				kn->kn_status &= ~(KN_QUEUED|KN_ACTIVE);
1263 			} else if (kn->kn_flags & EV_DISPATCH) {
1264 				kn->kn_status |= KN_DISABLED;
1265 				kn->kn_status &= ~(KN_QUEUED|KN_ACTIVE);
1266 			} else {
1267 				/* add event back on list */
1268 				kq_check(kq);
1269 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1270 				kq->kq_count++;
1271 				kn->kn_status |= KN_QUEUED;
1272 				kq_check(kq);
1273 			}
1274 			if (nkev == kevcnt) {
1275 				/* do copyouts in kevcnt chunks */
1276 				mutex_spin_exit(&kq->kq_lock);
1277 				error = (*keops->keo_put_events)
1278 				    (keops->keo_private,
1279 				    kevbuf, ulistp, nevents, nkev);
1280 				mutex_spin_enter(&kq->kq_lock);
1281 				nevents += nkev;
1282 				nkev = 0;
1283 				kevp = kevbuf;
1284 			}
1285 			count--;
1286 			if (error != 0 || count == 0) {
1287 				/* remove marker */
1288 				TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1289 				break;
1290 			}
1291 		}
1292 	}
1293  done:
1294  	mutex_spin_exit(&kq->kq_lock);
1295 	if (marker != NULL)
1296 		kmem_free(marker, sizeof(*marker));
1297 	if (nkev != 0) {
1298 		/* copyout remaining events */
1299 		error = (*keops->keo_put_events)(keops->keo_private,
1300 		    kevbuf, ulistp, nevents, nkev);
1301 	}
1302 	*retval = maxevents - count;
1303 
1304 	return error;
1305 }
1306 
1307 /*
1308  * fileops ioctl method for a kqueue descriptor.
1309  *
1310  * Two ioctls are currently supported. They both use struct kfilter_mapping:
1311  *	KFILTER_BYNAME		find name for filter, and return result in
1312  *				name, which is of size len.
1313  *	KFILTER_BYFILTER	find filter for name. len is ignored.
1314  */
1315 /*ARGSUSED*/
1316 static int
1317 kqueue_ioctl(file_t *fp, u_long com, void *data)
1318 {
1319 	struct kfilter_mapping	*km;
1320 	const struct kfilter	*kfilter;
1321 	char			*name;
1322 	int			error;
1323 
1324 	km = data;
1325 	error = 0;
1326 	name = kmem_alloc(KFILTER_MAXNAME, KM_SLEEP);
1327 
1328 	switch (com) {
1329 	case KFILTER_BYFILTER:	/* convert filter -> name */
1330 		rw_enter(&kqueue_filter_lock, RW_READER);
1331 		kfilter = kfilter_byfilter(km->filter);
1332 		if (kfilter != NULL) {
1333 			strlcpy(name, kfilter->name, KFILTER_MAXNAME);
1334 			rw_exit(&kqueue_filter_lock);
1335 			error = copyoutstr(name, km->name, km->len, NULL);
1336 		} else {
1337 			rw_exit(&kqueue_filter_lock);
1338 			error = ENOENT;
1339 		}
1340 		break;
1341 
1342 	case KFILTER_BYNAME:	/* convert name -> filter */
1343 		error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL);
1344 		if (error) {
1345 			break;
1346 		}
1347 		rw_enter(&kqueue_filter_lock, RW_READER);
1348 		kfilter = kfilter_byname(name);
1349 		if (kfilter != NULL)
1350 			km->filter = kfilter->filter;
1351 		else
1352 			error = ENOENT;
1353 		rw_exit(&kqueue_filter_lock);
1354 		break;
1355 
1356 	default:
1357 		error = ENOTTY;
1358 		break;
1359 
1360 	}
1361 	kmem_free(name, KFILTER_MAXNAME);
1362 	return (error);
1363 }
1364 
1365 /*
1366  * fileops fcntl method for a kqueue descriptor.
1367  */
1368 static int
1369 kqueue_fcntl(file_t *fp, u_int com, void *data)
1370 {
1371 
1372 	return (ENOTTY);
1373 }
1374 
1375 /*
1376  * fileops poll method for a kqueue descriptor.
1377  * Determine if kqueue has events pending.
1378  */
1379 static int
1380 kqueue_poll(file_t *fp, int events)
1381 {
1382 	struct kqueue	*kq;
1383 	int		revents;
1384 
1385 	kq = fp->f_kqueue;
1386 
1387 	revents = 0;
1388 	if (events & (POLLIN | POLLRDNORM)) {
1389 		mutex_spin_enter(&kq->kq_lock);
1390 		if (kq->kq_count != 0) {
1391 			revents |= events & (POLLIN | POLLRDNORM);
1392 		} else {
1393 			selrecord(curlwp, &kq->kq_sel);
1394 		}
1395 		kq_check(kq);
1396 		mutex_spin_exit(&kq->kq_lock);
1397 	}
1398 
1399 	return revents;
1400 }
1401 
1402 /*
1403  * fileops stat method for a kqueue descriptor.
1404  * Returns dummy info, with st_size being number of events pending.
1405  */
1406 static int
1407 kqueue_stat(file_t *fp, struct stat *st)
1408 {
1409 	struct kqueue *kq;
1410 
1411 	kq = fp->f_kqueue;
1412 
1413 	memset(st, 0, sizeof(*st));
1414 	st->st_size = kq->kq_count;
1415 	st->st_blksize = sizeof(struct kevent);
1416 	st->st_mode = S_IFIFO;
1417 
1418 	return 0;
1419 }
1420 
1421 static void
1422 kqueue_doclose(struct kqueue *kq, struct klist *list, int fd)
1423 {
1424 	struct knote *kn;
1425 	filedesc_t *fdp;
1426 
1427 	fdp = kq->kq_fdp;
1428 
1429 	KASSERT(mutex_owned(&fdp->fd_lock));
1430 
1431 	for (kn = SLIST_FIRST(list); kn != NULL;) {
1432 		if (kq != kn->kn_kq) {
1433 			kn = SLIST_NEXT(kn, kn_link);
1434 			continue;
1435 		}
1436 		knote_detach(kn, fdp, true);
1437 		mutex_enter(&fdp->fd_lock);
1438 		kn = SLIST_FIRST(list);
1439 	}
1440 }
1441 
1442 
1443 /*
1444  * fileops close method for a kqueue descriptor.
1445  */
1446 static int
1447 kqueue_close(file_t *fp)
1448 {
1449 	struct kqueue *kq;
1450 	filedesc_t *fdp;
1451 	fdfile_t *ff;
1452 	int i;
1453 
1454 	kq = fp->f_kqueue;
1455 	fp->f_kqueue = NULL;
1456 	fp->f_type = 0;
1457 	fdp = curlwp->l_fd;
1458 
1459 	mutex_enter(&fdp->fd_lock);
1460 	for (i = 0; i <= fdp->fd_lastkqfile; i++) {
1461 		if ((ff = fdp->fd_dt->dt_ff[i]) == NULL)
1462 			continue;
1463 		kqueue_doclose(kq, (struct klist *)&ff->ff_knlist, i);
1464 	}
1465 	if (fdp->fd_knhashmask != 0) {
1466 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
1467 			kqueue_doclose(kq, &fdp->fd_knhash[i], -1);
1468 		}
1469 	}
1470 	mutex_exit(&fdp->fd_lock);
1471 
1472 	KASSERT(kq->kq_count == 0);
1473 	mutex_destroy(&kq->kq_lock);
1474 	cv_destroy(&kq->kq_cv);
1475 	seldestroy(&kq->kq_sel);
1476 	kmem_free(kq, sizeof(*kq));
1477 
1478 	return (0);
1479 }
1480 
1481 /*
1482  * struct fileops kqfilter method for a kqueue descriptor.
1483  * Event triggered when monitored kqueue changes.
1484  */
1485 static int
1486 kqueue_kqfilter(file_t *fp, struct knote *kn)
1487 {
1488 	struct kqueue *kq;
1489 
1490 	kq = ((file_t *)kn->kn_obj)->f_kqueue;
1491 
1492 	KASSERT(fp == kn->kn_obj);
1493 
1494 	if (kn->kn_filter != EVFILT_READ)
1495 		return 1;
1496 
1497 	kn->kn_fop = &kqread_filtops;
1498 	mutex_enter(&kq->kq_lock);
1499 	SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext);
1500 	mutex_exit(&kq->kq_lock);
1501 
1502 	return 0;
1503 }
1504 
1505 
1506 /*
1507  * Walk down a list of knotes, activating them if their event has
1508  * triggered.  The caller's object lock (e.g. device driver lock)
1509  * must be held.
1510  */
1511 void
1512 knote(struct klist *list, long hint)
1513 {
1514 	struct knote *kn, *tmpkn;
1515 
1516 	SLIST_FOREACH_SAFE(kn, list, kn_selnext, tmpkn) {
1517 		KASSERT(kn->kn_fop->f_event != NULL);
1518 		if ((*kn->kn_fop->f_event)(kn, hint))
1519 			knote_activate(kn);
1520 	}
1521 }
1522 
1523 /*
1524  * Remove all knotes referencing a specified fd
1525  */
1526 void
1527 knote_fdclose(int fd)
1528 {
1529 	struct klist *list;
1530 	struct knote *kn;
1531 	filedesc_t *fdp;
1532 
1533 	fdp = curlwp->l_fd;
1534 	list = (struct klist *)&fdp->fd_dt->dt_ff[fd]->ff_knlist;
1535 	mutex_enter(&fdp->fd_lock);
1536 	while ((kn = SLIST_FIRST(list)) != NULL) {
1537 		knote_detach(kn, fdp, true);
1538 		mutex_enter(&fdp->fd_lock);
1539 	}
1540 	mutex_exit(&fdp->fd_lock);
1541 }
1542 
1543 /*
1544  * Drop knote.  Called with fdp->fd_lock held, and will drop before
1545  * returning.
1546  */
1547 static void
1548 knote_detach(struct knote *kn, filedesc_t *fdp, bool dofop)
1549 {
1550 	struct klist *list;
1551 	struct kqueue *kq;
1552 
1553 	kq = kn->kn_kq;
1554 
1555 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1556 	KASSERT(mutex_owned(&fdp->fd_lock));
1557 
1558 	/* Remove from monitored object. */
1559 	if (dofop) {
1560 		KERNEL_LOCK(1, NULL);		/* XXXSMP */
1561 		(*kn->kn_fop->f_detach)(kn);
1562 		KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1563 	}
1564 
1565 	/* Remove from descriptor table. */
1566 	if (kn->kn_fop->f_isfd)
1567 		list = (struct klist *)&fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist;
1568 	else
1569 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1570 
1571 	SLIST_REMOVE(list, kn, knote, kn_link);
1572 
1573 	/* Remove from kqueue. */
1574 	/* XXXAD should verify not in use by kqueue_scan. */
1575 	mutex_spin_enter(&kq->kq_lock);
1576 	if ((kn->kn_status & KN_QUEUED) != 0) {
1577 		kq_check(kq);
1578 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1579 		kn->kn_status &= ~KN_QUEUED;
1580 		kq->kq_count--;
1581 		kq_check(kq);
1582 	}
1583 	mutex_spin_exit(&kq->kq_lock);
1584 
1585 	mutex_exit(&fdp->fd_lock);
1586 	if (kn->kn_fop->f_isfd)
1587 		fd_putfile(kn->kn_id);
1588 	atomic_dec_uint(&kn->kn_kfilter->refcnt);
1589 	kmem_free(kn, sizeof(*kn));
1590 }
1591 
1592 /*
1593  * Queue new event for knote.
1594  */
1595 static void
1596 knote_enqueue(struct knote *kn)
1597 {
1598 	struct kqueue *kq;
1599 
1600 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1601 
1602 	kq = kn->kn_kq;
1603 
1604 	mutex_spin_enter(&kq->kq_lock);
1605 	if ((kn->kn_status & KN_DISABLED) != 0) {
1606 		kn->kn_status &= ~KN_DISABLED;
1607 	}
1608 	if ((kn->kn_status & (KN_ACTIVE | KN_QUEUED)) == KN_ACTIVE) {
1609 		kq_check(kq);
1610 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1611 		kn->kn_status |= KN_QUEUED;
1612 		kq->kq_count++;
1613 		kq_check(kq);
1614 		cv_broadcast(&kq->kq_cv);
1615 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1616 	}
1617 	mutex_spin_exit(&kq->kq_lock);
1618 }
1619 /*
1620  * Queue new event for knote.
1621  */
1622 static void
1623 knote_activate(struct knote *kn)
1624 {
1625 	struct kqueue *kq;
1626 
1627 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1628 
1629 	kq = kn->kn_kq;
1630 
1631 	mutex_spin_enter(&kq->kq_lock);
1632 	kn->kn_status |= KN_ACTIVE;
1633 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) {
1634 		kq_check(kq);
1635 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1636 		kn->kn_status |= KN_QUEUED;
1637 		kq->kq_count++;
1638 		kq_check(kq);
1639 		cv_broadcast(&kq->kq_cv);
1640 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1641 	}
1642 	mutex_spin_exit(&kq->kq_lock);
1643 }
1644