xref: /netbsd-src/sys/kern/kern_event.c (revision 404fbe5fb94ca1e054339640cabb2801ce52dd30)
1 /*	$NetBSD: kern_event.c,v 1.61 2009/01/11 02:45:52 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*-
30  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
31  * All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  *
42  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
52  * SUCH DAMAGE.
53  *
54  * FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp
55  */
56 
57 #include <sys/cdefs.h>
58 __KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.61 2009/01/11 02:45:52 christos Exp $");
59 
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/proc.h>
64 #include <sys/file.h>
65 #include <sys/select.h>
66 #include <sys/queue.h>
67 #include <sys/event.h>
68 #include <sys/eventvar.h>
69 #include <sys/poll.h>
70 #include <sys/kmem.h>
71 #include <sys/stat.h>
72 #include <sys/filedesc.h>
73 #include <sys/syscallargs.h>
74 #include <sys/kauth.h>
75 #include <sys/conf.h>
76 #include <sys/atomic.h>
77 
78 static int	kqueue_scan(file_t *, size_t, struct kevent *,
79 			    const struct timespec *, register_t *,
80 			    const struct kevent_ops *, struct kevent *,
81 			    size_t);
82 static int	kqueue_ioctl(file_t *, u_long, void *);
83 static int	kqueue_fcntl(file_t *, u_int, void *);
84 static int	kqueue_poll(file_t *, int);
85 static int	kqueue_kqfilter(file_t *, struct knote *);
86 static int	kqueue_stat(file_t *, struct stat *);
87 static int	kqueue_close(file_t *);
88 static int	kqueue_register(struct kqueue *, struct kevent *);
89 static void	kqueue_doclose(struct kqueue *, struct klist *, int);
90 
91 static void	knote_detach(struct knote *, filedesc_t *fdp, bool);
92 static void	knote_enqueue(struct knote *);
93 static void	knote_activate(struct knote *);
94 
95 static void	filt_kqdetach(struct knote *);
96 static int	filt_kqueue(struct knote *, long hint);
97 static int	filt_procattach(struct knote *);
98 static void	filt_procdetach(struct knote *);
99 static int	filt_proc(struct knote *, long hint);
100 static int	filt_fileattach(struct knote *);
101 static void	filt_timerexpire(void *x);
102 static int	filt_timerattach(struct knote *);
103 static void	filt_timerdetach(struct knote *);
104 static int	filt_timer(struct knote *, long hint);
105 
106 static const struct fileops kqueueops = {
107 	(void *)enxio, (void *)enxio, kqueue_ioctl, kqueue_fcntl, kqueue_poll,
108 	kqueue_stat, kqueue_close, kqueue_kqfilter
109 };
110 
111 static const struct filterops kqread_filtops =
112 	{ 1, NULL, filt_kqdetach, filt_kqueue };
113 static const struct filterops proc_filtops =
114 	{ 0, filt_procattach, filt_procdetach, filt_proc };
115 static const struct filterops file_filtops =
116 	{ 1, filt_fileattach, NULL, NULL };
117 static const struct filterops timer_filtops =
118 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
119 
120 static u_int	kq_ncallouts = 0;
121 static int	kq_calloutmax = (4 * 1024);
122 
123 #define	KN_HASHSIZE		64		/* XXX should be tunable */
124 #define	KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
125 
126 extern const struct filterops sig_filtops;
127 
128 /*
129  * Table for for all system-defined filters.
130  * These should be listed in the numeric order of the EVFILT_* defines.
131  * If filtops is NULL, the filter isn't implemented in NetBSD.
132  * End of list is when name is NULL.
133  *
134  * Note that 'refcnt' is meaningless for built-in filters.
135  */
136 struct kfilter {
137 	const char	*name;		/* name of filter */
138 	uint32_t	filter;		/* id of filter */
139 	unsigned	refcnt;		/* reference count */
140 	const struct filterops *filtops;/* operations for filter */
141 	size_t		namelen;	/* length of name string */
142 };
143 
144 /* System defined filters */
145 static struct kfilter sys_kfilters[] = {
146 	{ "EVFILT_READ",	EVFILT_READ,	0, &file_filtops, 0 },
147 	{ "EVFILT_WRITE",	EVFILT_WRITE,	0, &file_filtops, 0, },
148 	{ "EVFILT_AIO",		EVFILT_AIO,	0, NULL, 0 },
149 	{ "EVFILT_VNODE",	EVFILT_VNODE,	0, &file_filtops, 0 },
150 	{ "EVFILT_PROC",	EVFILT_PROC,	0, &proc_filtops, 0 },
151 	{ "EVFILT_SIGNAL",	EVFILT_SIGNAL,	0, &sig_filtops, 0 },
152 	{ "EVFILT_TIMER",	EVFILT_TIMER,	0, &timer_filtops, 0 },
153 	{ NULL,			0,		0, NULL, 0 },
154 };
155 
156 /* User defined kfilters */
157 static struct kfilter	*user_kfilters;		/* array */
158 static int		user_kfilterc;		/* current offset */
159 static int		user_kfiltermaxc;	/* max size so far */
160 static size_t		user_kfiltersz;		/* size of allocated memory */
161 
162 /* Locks */
163 static krwlock_t	kqueue_filter_lock;	/* lock on filter lists */
164 static kmutex_t		kqueue_misc_lock;	/* miscellaneous */
165 
166 /*
167  * Initialize the kqueue subsystem.
168  */
169 void
170 kqueue_init(void)
171 {
172 
173 	rw_init(&kqueue_filter_lock);
174 	mutex_init(&kqueue_misc_lock, MUTEX_DEFAULT, IPL_NONE);
175 }
176 
177 /*
178  * Find kfilter entry by name, or NULL if not found.
179  */
180 static struct kfilter *
181 kfilter_byname_sys(const char *name)
182 {
183 	int i;
184 
185 	KASSERT(rw_lock_held(&kqueue_filter_lock));
186 
187 	for (i = 0; sys_kfilters[i].name != NULL; i++) {
188 		if (strcmp(name, sys_kfilters[i].name) == 0)
189 			return &sys_kfilters[i];
190 	}
191 	return NULL;
192 }
193 
194 static struct kfilter *
195 kfilter_byname_user(const char *name)
196 {
197 	int i;
198 
199 	KASSERT(rw_lock_held(&kqueue_filter_lock));
200 
201 	/* user filter slots have a NULL name if previously deregistered */
202 	for (i = 0; i < user_kfilterc ; i++) {
203 		if (user_kfilters[i].name != NULL &&
204 		    strcmp(name, user_kfilters[i].name) == 0)
205 			return &user_kfilters[i];
206 	}
207 	return NULL;
208 }
209 
210 static struct kfilter *
211 kfilter_byname(const char *name)
212 {
213 	struct kfilter *kfilter;
214 
215 	KASSERT(rw_lock_held(&kqueue_filter_lock));
216 
217 	if ((kfilter = kfilter_byname_sys(name)) != NULL)
218 		return kfilter;
219 
220 	return kfilter_byname_user(name);
221 }
222 
223 /*
224  * Find kfilter entry by filter id, or NULL if not found.
225  * Assumes entries are indexed in filter id order, for speed.
226  */
227 static struct kfilter *
228 kfilter_byfilter(uint32_t filter)
229 {
230 	struct kfilter *kfilter;
231 
232 	KASSERT(rw_lock_held(&kqueue_filter_lock));
233 
234 	if (filter < EVFILT_SYSCOUNT)	/* it's a system filter */
235 		kfilter = &sys_kfilters[filter];
236 	else if (user_kfilters != NULL &&
237 	    filter < EVFILT_SYSCOUNT + user_kfilterc)
238 					/* it's a user filter */
239 		kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT];
240 	else
241 		return (NULL);		/* out of range */
242 	KASSERT(kfilter->filter == filter);	/* sanity check! */
243 	return (kfilter);
244 }
245 
246 /*
247  * Register a new kfilter. Stores the entry in user_kfilters.
248  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
249  * If retfilter != NULL, the new filterid is returned in it.
250  */
251 int
252 kfilter_register(const char *name, const struct filterops *filtops,
253 		 int *retfilter)
254 {
255 	struct kfilter *kfilter;
256 	size_t len;
257 	int i;
258 
259 	if (name == NULL || name[0] == '\0' || filtops == NULL)
260 		return (EINVAL);	/* invalid args */
261 
262 	rw_enter(&kqueue_filter_lock, RW_WRITER);
263 	if (kfilter_byname(name) != NULL) {
264 		rw_exit(&kqueue_filter_lock);
265 		return (EEXIST);	/* already exists */
266 	}
267 	if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT) {
268 		rw_exit(&kqueue_filter_lock);
269 		return (EINVAL);	/* too many */
270 	}
271 
272 	for (i = 0; i < user_kfilterc; i++) {
273 		kfilter = &user_kfilters[i];
274 		if (kfilter->name == NULL) {
275 			/* Previously deregistered slot.  Reuse. */
276 			goto reuse;
277 		}
278 	}
279 
280 	/* check if need to grow user_kfilters */
281 	if (user_kfilterc + 1 > user_kfiltermaxc) {
282 		/* Grow in KFILTER_EXTENT chunks. */
283 		user_kfiltermaxc += KFILTER_EXTENT;
284 		len = user_kfiltermaxc * sizeof(struct filter *);
285 		kfilter = kmem_alloc(len, KM_SLEEP);
286 		memset((char *)kfilter + user_kfiltersz, 0, len - user_kfiltersz);
287 		if (user_kfilters != NULL) {
288 			memcpy(kfilter, user_kfilters, user_kfiltersz);
289 			kmem_free(user_kfilters, user_kfiltersz);
290 		}
291 		user_kfiltersz = len;
292 		user_kfilters = kfilter;
293 	}
294 	/* Adding new slot */
295 	kfilter = &user_kfilters[user_kfilterc++];
296 reuse:
297 	kfilter->namelen = strlen(name) + 1;
298 	kfilter->name = kmem_alloc(kfilter->namelen, KM_SLEEP);
299 	memcpy(__UNCONST(kfilter->name), name, kfilter->namelen);
300 
301 	kfilter->filter = (kfilter - user_kfilters) + EVFILT_SYSCOUNT;
302 
303 	kfilter->filtops = kmem_alloc(sizeof(*filtops), KM_SLEEP);
304 	memcpy(__UNCONST(kfilter->filtops), filtops, sizeof(*filtops));
305 
306 	if (retfilter != NULL)
307 		*retfilter = kfilter->filter;
308 	rw_exit(&kqueue_filter_lock);
309 
310 	return (0);
311 }
312 
313 /*
314  * Unregister a kfilter previously registered with kfilter_register.
315  * This retains the filter id, but clears the name and frees filtops (filter
316  * operations), so that the number isn't reused during a boot.
317  * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise.
318  */
319 int
320 kfilter_unregister(const char *name)
321 {
322 	struct kfilter *kfilter;
323 
324 	if (name == NULL || name[0] == '\0')
325 		return (EINVAL);	/* invalid name */
326 
327 	rw_enter(&kqueue_filter_lock, RW_WRITER);
328 	if (kfilter_byname_sys(name) != NULL) {
329 		rw_exit(&kqueue_filter_lock);
330 		return (EINVAL);	/* can't detach system filters */
331 	}
332 
333 	kfilter = kfilter_byname_user(name);
334 	if (kfilter == NULL) {
335 		rw_exit(&kqueue_filter_lock);
336 		return (ENOENT);
337 	}
338 	if (kfilter->refcnt != 0) {
339 		rw_exit(&kqueue_filter_lock);
340 		return (EBUSY);
341 	}
342 
343 	/* Cast away const (but we know it's safe. */
344 	kmem_free(__UNCONST(kfilter->name), kfilter->namelen);
345 	kfilter->name = NULL;	/* mark as `not implemented' */
346 
347 	if (kfilter->filtops != NULL) {
348 		/* Cast away const (but we know it's safe. */
349 		kmem_free(__UNCONST(kfilter->filtops),
350 		    sizeof(*kfilter->filtops));
351 		kfilter->filtops = NULL; /* mark as `not implemented' */
352 	}
353 	rw_exit(&kqueue_filter_lock);
354 
355 	return (0);
356 }
357 
358 
359 /*
360  * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file
361  * descriptors. Calls fileops kqfilter method for given file descriptor.
362  */
363 static int
364 filt_fileattach(struct knote *kn)
365 {
366 	file_t *fp;
367 
368 	fp = kn->kn_obj;
369 
370 	return (*fp->f_ops->fo_kqfilter)(fp, kn);
371 }
372 
373 /*
374  * Filter detach method for EVFILT_READ on kqueue descriptor.
375  */
376 static void
377 filt_kqdetach(struct knote *kn)
378 {
379 	struct kqueue *kq;
380 
381 	kq = ((file_t *)kn->kn_obj)->f_data;
382 
383 	mutex_spin_enter(&kq->kq_lock);
384 	SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext);
385 	mutex_spin_exit(&kq->kq_lock);
386 }
387 
388 /*
389  * Filter event method for EVFILT_READ on kqueue descriptor.
390  */
391 /*ARGSUSED*/
392 static int
393 filt_kqueue(struct knote *kn, long hint)
394 {
395 	struct kqueue *kq;
396 	int rv;
397 
398 	kq = ((file_t *)kn->kn_obj)->f_data;
399 
400 	if (hint != NOTE_SUBMIT)
401 		mutex_spin_enter(&kq->kq_lock);
402 	kn->kn_data = kq->kq_count;
403 	rv = (kn->kn_data > 0);
404 	if (hint != NOTE_SUBMIT)
405 		mutex_spin_exit(&kq->kq_lock);
406 
407 	return rv;
408 }
409 
410 /*
411  * Filter attach method for EVFILT_PROC.
412  */
413 static int
414 filt_procattach(struct knote *kn)
415 {
416 	struct proc *p, *curp;
417 	struct lwp *curl;
418 
419 	curl = curlwp;
420 	curp = curl->l_proc;
421 
422 	mutex_enter(proc_lock);
423 	p = p_find(kn->kn_id, PFIND_LOCKED);
424 	if (p == NULL) {
425 		mutex_exit(proc_lock);
426 		return ESRCH;
427 	}
428 
429 	/*
430 	 * Fail if it's not owned by you, or the last exec gave us
431 	 * setuid/setgid privs (unless you're root).
432 	 */
433 	mutex_enter(p->p_lock);
434 	mutex_exit(proc_lock);
435 	if (kauth_authorize_process(curl->l_cred, KAUTH_PROCESS_KEVENT_FILTER,
436 	    p, NULL, NULL, NULL) != 0) {
437 	    	mutex_exit(p->p_lock);
438 		return EACCES;
439 	}
440 
441 	kn->kn_obj = p;
442 	kn->kn_flags |= EV_CLEAR;	/* automatically set */
443 
444 	/*
445 	 * internal flag indicating registration done by kernel
446 	 */
447 	if (kn->kn_flags & EV_FLAG1) {
448 		kn->kn_data = kn->kn_sdata;	/* ppid */
449 		kn->kn_fflags = NOTE_CHILD;
450 		kn->kn_flags &= ~EV_FLAG1;
451 	}
452 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
453     	mutex_exit(p->p_lock);
454 
455 	return 0;
456 }
457 
458 /*
459  * Filter detach method for EVFILT_PROC.
460  *
461  * The knote may be attached to a different process, which may exit,
462  * leaving nothing for the knote to be attached to.  So when the process
463  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
464  * it will be deleted when read out.  However, as part of the knote deletion,
465  * this routine is called, so a check is needed to avoid actually performing
466  * a detach, because the original process might not exist any more.
467  */
468 static void
469 filt_procdetach(struct knote *kn)
470 {
471 	struct proc *p;
472 
473 	if (kn->kn_status & KN_DETACHED)
474 		return;
475 
476 	p = kn->kn_obj;
477 
478 	mutex_enter(p->p_lock);
479 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
480 	mutex_exit(p->p_lock);
481 }
482 
483 /*
484  * Filter event method for EVFILT_PROC.
485  */
486 static int
487 filt_proc(struct knote *kn, long hint)
488 {
489 	u_int event, fflag;
490 	struct kevent kev;
491 	struct kqueue *kq;
492 	int error;
493 
494 	event = (u_int)hint & NOTE_PCTRLMASK;
495 	kq = kn->kn_kq;
496 	fflag = 0;
497 
498 	/* If the user is interested in this event, record it. */
499 	if (kn->kn_sfflags & event)
500 		fflag |= event;
501 
502 	if (event == NOTE_EXIT) {
503 		/*
504 		 * Process is gone, so flag the event as finished.
505 		 *
506 		 * Detach the knote from watched process and mark
507 		 * it as such. We can't leave this to kqueue_scan(),
508 		 * since the process might not exist by then. And we
509 		 * have to do this now, since psignal KNOTE() is called
510 		 * also for zombies and we might end up reading freed
511 		 * memory if the kevent would already be picked up
512 		 * and knote g/c'ed.
513 		 */
514 		filt_procdetach(kn);
515 
516 		mutex_spin_enter(&kq->kq_lock);
517 		kn->kn_status |= KN_DETACHED;
518 		/* Mark as ONESHOT, so that the knote it g/c'ed when read */
519 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
520 		kn->kn_fflags |= fflag;
521 		mutex_spin_exit(&kq->kq_lock);
522 
523 		return 1;
524 	}
525 
526 	mutex_spin_enter(&kq->kq_lock);
527 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
528 		/*
529 		 * Process forked, and user wants to track the new process,
530 		 * so attach a new knote to it, and immediately report an
531 		 * event with the parent's pid.  Register knote with new
532 		 * process.
533 		 */
534 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
535 		kev.filter = kn->kn_filter;
536 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
537 		kev.fflags = kn->kn_sfflags;
538 		kev.data = kn->kn_id;			/* parent */
539 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
540 		mutex_spin_exit(&kq->kq_lock);
541 		error = kqueue_register(kq, &kev);
542 		mutex_spin_enter(&kq->kq_lock);
543 		if (error != 0)
544 			kn->kn_fflags |= NOTE_TRACKERR;
545 	}
546 	kn->kn_fflags |= fflag;
547 	fflag = kn->kn_fflags;
548 	mutex_spin_exit(&kq->kq_lock);
549 
550 	return fflag != 0;
551 }
552 
553 static void
554 filt_timerexpire(void *knx)
555 {
556 	struct knote *kn = knx;
557 	int tticks;
558 
559 	mutex_enter(&kqueue_misc_lock);
560 	kn->kn_data++;
561 	knote_activate(kn);
562 	if ((kn->kn_flags & EV_ONESHOT) == 0) {
563 		tticks = mstohz(kn->kn_sdata);
564 		callout_schedule((callout_t *)kn->kn_hook, tticks);
565 	}
566 	mutex_exit(&kqueue_misc_lock);
567 }
568 
569 /*
570  * data contains amount of time to sleep, in milliseconds
571  */
572 static int
573 filt_timerattach(struct knote *kn)
574 {
575 	callout_t *calloutp;
576 	struct kqueue *kq;
577 	int tticks;
578 
579 	tticks = mstohz(kn->kn_sdata);
580 
581 	/* if the supplied value is under our resolution, use 1 tick */
582 	if (tticks == 0) {
583 		if (kn->kn_sdata == 0)
584 			return EINVAL;
585 		tticks = 1;
586 	}
587 
588 	if (atomic_inc_uint_nv(&kq_ncallouts) >= kq_calloutmax ||
589 	    (calloutp = kmem_alloc(sizeof(*calloutp), KM_NOSLEEP)) == NULL) {
590 		atomic_dec_uint(&kq_ncallouts);
591 		return ENOMEM;
592 	}
593 	callout_init(calloutp, CALLOUT_MPSAFE);
594 
595 	kq = kn->kn_kq;
596 	mutex_spin_enter(&kq->kq_lock);
597 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
598 	kn->kn_hook = calloutp;
599 	mutex_spin_exit(&kq->kq_lock);
600 
601 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
602 
603 	return (0);
604 }
605 
606 static void
607 filt_timerdetach(struct knote *kn)
608 {
609 	callout_t *calloutp;
610 
611 	calloutp = (callout_t *)kn->kn_hook;
612 	callout_halt(calloutp, NULL);
613 	callout_destroy(calloutp);
614 	kmem_free(calloutp, sizeof(*calloutp));
615 	atomic_dec_uint(&kq_ncallouts);
616 }
617 
618 static int
619 filt_timer(struct knote *kn, long hint)
620 {
621 	int rv;
622 
623 	mutex_enter(&kqueue_misc_lock);
624 	rv = (kn->kn_data != 0);
625 	mutex_exit(&kqueue_misc_lock);
626 
627 	return rv;
628 }
629 
630 /*
631  * filt_seltrue:
632  *
633  *	This filter "event" routine simulates seltrue().
634  */
635 int
636 filt_seltrue(struct knote *kn, long hint)
637 {
638 
639 	/*
640 	 * We don't know how much data can be read/written,
641 	 * but we know that it *can* be.  This is about as
642 	 * good as select/poll does as well.
643 	 */
644 	kn->kn_data = 0;
645 	return (1);
646 }
647 
648 /*
649  * This provides full kqfilter entry for device switch tables, which
650  * has same effect as filter using filt_seltrue() as filter method.
651  */
652 static void
653 filt_seltruedetach(struct knote *kn)
654 {
655 	/* Nothing to do */
656 }
657 
658 const struct filterops seltrue_filtops =
659 	{ 1, NULL, filt_seltruedetach, filt_seltrue };
660 
661 int
662 seltrue_kqfilter(dev_t dev, struct knote *kn)
663 {
664 	switch (kn->kn_filter) {
665 	case EVFILT_READ:
666 	case EVFILT_WRITE:
667 		kn->kn_fop = &seltrue_filtops;
668 		break;
669 	default:
670 		return (EINVAL);
671 	}
672 
673 	/* Nothing more to do */
674 	return (0);
675 }
676 
677 /*
678  * kqueue(2) system call.
679  */
680 int
681 sys_kqueue(struct lwp *l, const void *v, register_t *retval)
682 {
683 	struct kqueue *kq;
684 	file_t *fp;
685 	int fd, error;
686 
687 	if ((error = fd_allocfile(&fp, &fd)) != 0)
688 		return error;
689 	fp->f_flag = FREAD | FWRITE;
690 	fp->f_type = DTYPE_KQUEUE;
691 	fp->f_ops = &kqueueops;
692 	kq = kmem_zalloc(sizeof(*kq), KM_SLEEP);
693 	mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED);
694 	cv_init(&kq->kq_cv, "kqueue");
695 	selinit(&kq->kq_sel);
696 	TAILQ_INIT(&kq->kq_head);
697 	fp->f_data = kq;
698 	*retval = fd;
699 	kq->kq_fdp = curlwp->l_fd;
700 	fd_affix(curproc, fp, fd);
701 	return error;
702 }
703 
704 /*
705  * kevent(2) system call.
706  */
707 int
708 kevent_fetch_changes(void *private, const struct kevent *changelist,
709     struct kevent *changes, size_t index, int n)
710 {
711 
712 	return copyin(changelist + index, changes, n * sizeof(*changes));
713 }
714 
715 int
716 kevent_put_events(void *private, struct kevent *events,
717     struct kevent *eventlist, size_t index, int n)
718 {
719 
720 	return copyout(events, eventlist + index, n * sizeof(*events));
721 }
722 
723 static const struct kevent_ops kevent_native_ops = {
724 	.keo_private = NULL,
725 	.keo_fetch_timeout = copyin,
726 	.keo_fetch_changes = kevent_fetch_changes,
727 	.keo_put_events = kevent_put_events,
728 };
729 
730 int
731 sys___kevent50(struct lwp *l, const struct sys___kevent50_args *uap,
732     register_t *retval)
733 {
734 	/* {
735 		syscallarg(int) fd;
736 		syscallarg(const struct kevent *) changelist;
737 		syscallarg(size_t) nchanges;
738 		syscallarg(struct kevent *) eventlist;
739 		syscallarg(size_t) nevents;
740 		syscallarg(const struct timespec *) timeout;
741 	} */
742 
743 	return kevent1(retval, SCARG(uap, fd), SCARG(uap, changelist),
744 	    SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents),
745 	    SCARG(uap, timeout), &kevent_native_ops);
746 }
747 
748 int
749 kevent1(register_t *retval, int fd,
750 	const struct kevent *changelist, size_t nchanges,
751 	struct kevent *eventlist, size_t nevents,
752 	const struct timespec *timeout,
753 	const struct kevent_ops *keops)
754 {
755 	struct kevent *kevp;
756 	struct kqueue *kq;
757 	struct timespec	ts;
758 	size_t i, n, ichange;
759 	int nerrors, error;
760 	struct kevent kevbuf[8];	/* approx 300 bytes on 64-bit */
761 	file_t *fp;
762 
763 	/* check that we're dealing with a kq */
764 	fp = fd_getfile(fd);
765 	if (fp == NULL)
766 		return (EBADF);
767 
768 	if (fp->f_type != DTYPE_KQUEUE) {
769 		fd_putfile(fd);
770 		return (EBADF);
771 	}
772 
773 	if (timeout != NULL) {
774 		error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts));
775 		if (error)
776 			goto done;
777 		timeout = &ts;
778 	}
779 
780 	kq = (struct kqueue *)fp->f_data;
781 	nerrors = 0;
782 	ichange = 0;
783 
784 	/* traverse list of events to register */
785 	while (nchanges > 0) {
786 		n = MIN(nchanges, __arraycount(kevbuf));
787 		error = (*keops->keo_fetch_changes)(keops->keo_private,
788 		    changelist, kevbuf, ichange, n);
789 		if (error)
790 			goto done;
791 		for (i = 0; i < n; i++) {
792 			kevp = &kevbuf[i];
793 			kevp->flags &= ~EV_SYSFLAGS;
794 			/* register each knote */
795 			error = kqueue_register(kq, kevp);
796 			if (error) {
797 				if (nevents != 0) {
798 					kevp->flags = EV_ERROR;
799 					kevp->data = error;
800 					error = (*keops->keo_put_events)
801 					    (keops->keo_private, kevp,
802 					    eventlist, nerrors, 1);
803 					if (error)
804 						goto done;
805 					nevents--;
806 					nerrors++;
807 				} else {
808 					goto done;
809 				}
810 			}
811 		}
812 		nchanges -= n;	/* update the results */
813 		ichange += n;
814 	}
815 	if (nerrors) {
816 		*retval = nerrors;
817 		error = 0;
818 		goto done;
819 	}
820 
821 	/* actually scan through the events */
822 	error = kqueue_scan(fp, nevents, eventlist, timeout, retval, keops,
823 	    kevbuf, __arraycount(kevbuf));
824  done:
825 	fd_putfile(fd);
826 	return (error);
827 }
828 
829 /*
830  * Register a given kevent kev onto the kqueue
831  */
832 static int
833 kqueue_register(struct kqueue *kq, struct kevent *kev)
834 {
835 	struct kfilter *kfilter;
836 	filedesc_t *fdp;
837 	file_t *fp;
838 	fdfile_t *ff;
839 	struct knote *kn, *newkn;
840 	struct klist *list;
841 	int error, fd, rv;
842 
843 	fdp = kq->kq_fdp;
844 	fp = NULL;
845 	kn = NULL;
846 	error = 0;
847 	fd = 0;
848 
849 	newkn = kmem_zalloc(sizeof(*newkn), KM_SLEEP);
850 
851 	rw_enter(&kqueue_filter_lock, RW_READER);
852 	kfilter = kfilter_byfilter(kev->filter);
853 	if (kfilter == NULL || kfilter->filtops == NULL) {
854 		/* filter not found nor implemented */
855 		rw_exit(&kqueue_filter_lock);
856 		kmem_free(newkn, sizeof(*newkn));
857 		return (EINVAL);
858 	}
859 
860  	mutex_enter(&fdp->fd_lock);
861 
862 	/* search if knote already exists */
863 	if (kfilter->filtops->f_isfd) {
864 		/* monitoring a file descriptor */
865 		fd = kev->ident;
866 		if ((fp = fd_getfile(fd)) == NULL) {
867 		 	mutex_exit(&fdp->fd_lock);
868 			rw_exit(&kqueue_filter_lock);
869 			kmem_free(newkn, sizeof(*newkn));
870 			return EBADF;
871 		}
872 		ff = fdp->fd_ofiles[fd];
873 		if (fd <= fdp->fd_lastkqfile) {
874 			SLIST_FOREACH(kn, &ff->ff_knlist, kn_link) {
875 				if (kq == kn->kn_kq &&
876 				    kev->filter == kn->kn_filter)
877 					break;
878 			}
879 		}
880 	} else {
881 		/*
882 		 * not monitoring a file descriptor, so
883 		 * lookup knotes in internal hash table
884 		 */
885 		if (fdp->fd_knhashmask != 0) {
886 			list = &fdp->fd_knhash[
887 			    KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
888 			SLIST_FOREACH(kn, list, kn_link) {
889 				if (kev->ident == kn->kn_id &&
890 				    kq == kn->kn_kq &&
891 				    kev->filter == kn->kn_filter)
892 					break;
893 			}
894 		}
895 	}
896 
897 	/*
898 	 * kn now contains the matching knote, or NULL if no match
899 	 */
900 	if (kev->flags & EV_ADD) {
901 		if (kn == NULL) {
902 			/* create new knote */
903 			kn = newkn;
904 			newkn = NULL;
905 			kn->kn_obj = fp;
906 			kn->kn_kq = kq;
907 			kn->kn_fop = kfilter->filtops;
908 			kn->kn_kfilter = kfilter;
909 			kn->kn_sfflags = kev->fflags;
910 			kn->kn_sdata = kev->data;
911 			kev->fflags = 0;
912 			kev->data = 0;
913 			kn->kn_kevent = *kev;
914 
915 			/*
916 			 * apply reference count to knote structure, and
917 			 * do not release it at the end of this routine.
918 			 */
919 			fp = NULL;
920 
921 			if (!kn->kn_fop->f_isfd) {
922 				/*
923 				 * If knote is not on an fd, store on
924 				 * internal hash table.
925 				 */
926 				if (fdp->fd_knhashmask == 0) {
927 					/* XXXAD can block with fd_lock held */
928 					fdp->fd_knhash = hashinit(KN_HASHSIZE,
929 					    HASH_LIST, true,
930 					    &fdp->fd_knhashmask);
931 				}
932 				list = &fdp->fd_knhash[KN_HASH(kn->kn_id,
933 				    fdp->fd_knhashmask)];
934 			} else {
935 				/* Otherwise, knote is on an fd. */
936 				list = (struct klist *)
937 				    &fdp->fd_ofiles[kn->kn_id]->ff_knlist;
938 				if ((int)kn->kn_id > fdp->fd_lastkqfile)
939 					fdp->fd_lastkqfile = kn->kn_id;
940 			}
941 			SLIST_INSERT_HEAD(list, kn, kn_link);
942 
943 			KERNEL_LOCK(1, NULL);		/* XXXSMP */
944 			error = (*kfilter->filtops->f_attach)(kn);
945 			KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
946 			if (error != 0) {
947 				/* knote_detach() drops fdp->fd_lock */
948 				knote_detach(kn, fdp, false);
949 				goto done;
950 			}
951 			atomic_inc_uint(&kfilter->refcnt);
952 		} else {
953 			/*
954 			 * The user may change some filter values after the
955 			 * initial EV_ADD, but doing so will not reset any
956 			 * filter which have already been triggered.
957 			 */
958 			kn->kn_sfflags = kev->fflags;
959 			kn->kn_sdata = kev->data;
960 			kn->kn_kevent.udata = kev->udata;
961 		}
962 		KERNEL_LOCK(1, NULL);			/* XXXSMP */
963 		rv = (*kn->kn_fop->f_event)(kn, 0);
964 		KERNEL_UNLOCK_ONE(NULL);		/* XXXSMP */
965 		if (rv)
966 			knote_activate(kn);
967 	} else {
968 		if (kn == NULL) {
969 			error = ENOENT;
970 		 	mutex_exit(&fdp->fd_lock);
971 			goto done;
972 		}
973 		if (kev->flags & EV_DELETE) {
974 			/* knote_detach() drops fdp->fd_lock */
975 			knote_detach(kn, fdp, true);
976 			goto done;
977 		}
978 	}
979 
980 	/* disable knote */
981 	if ((kev->flags & EV_DISABLE)) {
982 		mutex_spin_enter(&kq->kq_lock);
983 		if ((kn->kn_status & KN_DISABLED) == 0)
984 			kn->kn_status |= KN_DISABLED;
985 		mutex_spin_exit(&kq->kq_lock);
986 	}
987 
988 	/* enable knote */
989 	if ((kev->flags & EV_ENABLE)) {
990 		knote_enqueue(kn);
991 	}
992 	mutex_exit(&fdp->fd_lock);
993  done:
994 	rw_exit(&kqueue_filter_lock);
995 	if (newkn != NULL)
996 		kmem_free(newkn, sizeof(*newkn));
997 	if (fp != NULL)
998 		fd_putfile(fd);
999 	return (error);
1000 }
1001 
1002 #if defined(DEBUG)
1003 static void
1004 kq_check(struct kqueue *kq)
1005 {
1006 	const struct knote *kn;
1007 	int count;
1008 	int nmarker;
1009 
1010 	KASSERT(mutex_owned(&kq->kq_lock));
1011 	KASSERT(kq->kq_count >= 0);
1012 
1013 	count = 0;
1014 	nmarker = 0;
1015 	TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) {
1016 		if ((kn->kn_status & (KN_MARKER | KN_QUEUED)) == 0) {
1017 			panic("%s: kq=%p kn=%p inconsist 1", __func__, kq, kn);
1018 		}
1019 		if ((kn->kn_status & KN_MARKER) == 0) {
1020 			if (kn->kn_kq != kq) {
1021 				panic("%s: kq=%p kn=%p inconsist 2",
1022 				    __func__, kq, kn);
1023 			}
1024 			if ((kn->kn_status & KN_ACTIVE) == 0) {
1025 				panic("%s: kq=%p kn=%p: not active",
1026 				    __func__, kq, kn);
1027 			}
1028 			count++;
1029 			if (count > kq->kq_count) {
1030 				goto bad;
1031 			}
1032 		} else {
1033 			nmarker++;
1034 #if 0
1035 			if (nmarker > 10000) {
1036 				panic("%s: kq=%p too many markers: %d != %d, "
1037 				    "nmarker=%d",
1038 				    __func__, kq, kq->kq_count, count, nmarker);
1039 			}
1040 #endif
1041 		}
1042 	}
1043 	if (kq->kq_count != count) {
1044 bad:
1045 		panic("%s: kq=%p inconsist 3: %d != %d, nmarker=%d",
1046 		    __func__, kq, kq->kq_count, count, nmarker);
1047 	}
1048 }
1049 #else /* defined(DEBUG) */
1050 #define	kq_check(a)	/* nothing */
1051 #endif /* defined(DEBUG) */
1052 
1053 /*
1054  * Scan through the list of events on fp (for a maximum of maxevents),
1055  * returning the results in to ulistp. Timeout is determined by tsp; if
1056  * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait
1057  * as appropriate.
1058  */
1059 static int
1060 kqueue_scan(file_t *fp, size_t maxevents, struct kevent *ulistp,
1061 	    const struct timespec *tsp, register_t *retval,
1062 	    const struct kevent_ops *keops, struct kevent *kevbuf,
1063 	    size_t kevcnt)
1064 {
1065 	struct kqueue	*kq;
1066 	struct kevent	*kevp;
1067 	struct timeval	atv, sleeptv;
1068 	struct knote	*kn, *marker;
1069 	size_t		count, nkev, nevents;
1070 	int		timeout, error, rv;
1071 	filedesc_t	*fdp;
1072 
1073 	fdp = curlwp->l_fd;
1074 	kq = fp->f_data;
1075 	count = maxevents;
1076 	nkev = nevents = error = 0;
1077 	if (count == 0) {
1078 		*retval = 0;
1079 		return 0;
1080 	}
1081 
1082 	if (tsp) {				/* timeout supplied */
1083 		TIMESPEC_TO_TIMEVAL(&atv, tsp);
1084 		if (inittimeleft(&atv, &sleeptv) == -1) {
1085 			*retval = maxevents;
1086 			return EINVAL;
1087 		}
1088 		timeout = tvtohz(&atv);
1089 		if (timeout <= 0)
1090 			timeout = -1;           /* do poll */
1091 	} else {
1092 		/* no timeout, wait forever */
1093 		timeout = 0;
1094 	}
1095 
1096 	marker = kmem_zalloc(sizeof(*marker), KM_SLEEP);
1097 	marker->kn_status = KN_MARKER;
1098 	mutex_spin_enter(&kq->kq_lock);
1099  retry:
1100 	kevp = kevbuf;
1101 	if (kq->kq_count == 0) {
1102 		if (timeout >= 0) {
1103 			error = cv_timedwait_sig(&kq->kq_cv,
1104 			    &kq->kq_lock, timeout);
1105 			if (error == 0) {
1106 				 if (tsp == NULL || (timeout =
1107 				     gettimeleft(&atv, &sleeptv)) > 0)
1108 					goto retry;
1109 			} else {
1110 				/* don't restart after signals... */
1111 				if (error == ERESTART)
1112 					error = EINTR;
1113 				if (error == EWOULDBLOCK)
1114 					error = 0;
1115 			}
1116 		}
1117 	} else {
1118 		/* mark end of knote list */
1119 		TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1120 
1121 		while (count != 0) {
1122 			kn = TAILQ_FIRST(&kq->kq_head);	/* get next knote */
1123 			while ((kn->kn_status & KN_MARKER) != 0) {
1124 				if (kn == marker) {
1125 					/* it's our marker, stop */
1126 					TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1127 					if (count < maxevents || (tsp != NULL &&
1128 					    (timeout = gettimeleft(&atv,
1129 					    &sleeptv)) <= 0))
1130 						goto done;
1131 					goto retry;
1132 				}
1133 				/* someone else's marker. */
1134 				kn = TAILQ_NEXT(kn, kn_tqe);
1135 			}
1136 			kq_check(kq);
1137 			TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1138 			kq->kq_count--;
1139 			kn->kn_status &= ~KN_QUEUED;
1140 			kq_check(kq);
1141 			if (kn->kn_status & KN_DISABLED) {
1142 				/* don't want disabled events */
1143 				continue;
1144 			}
1145 			if ((kn->kn_flags & EV_ONESHOT) == 0) {
1146 				mutex_spin_exit(&kq->kq_lock);
1147 				KERNEL_LOCK(1, NULL);		/* XXXSMP */
1148 				rv = (*kn->kn_fop->f_event)(kn, 0);
1149 				KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1150 				mutex_spin_enter(&kq->kq_lock);
1151 				/* Re-poll if note was re-enqueued. */
1152 				if ((kn->kn_status & KN_QUEUED) != 0)
1153 					continue;
1154 				if (rv == 0) {
1155 					/*
1156 					 * non-ONESHOT event that hasn't
1157 					 * triggered again, so de-queue.
1158 					 */
1159 					kn->kn_status &= ~KN_ACTIVE;
1160 					continue;
1161 				}
1162 			}
1163 			/* XXXAD should be got from f_event if !oneshot. */
1164 			*kevp++ = kn->kn_kevent;
1165 			nkev++;
1166 			if (kn->kn_flags & EV_ONESHOT) {
1167 				/* delete ONESHOT events after retrieval */
1168 				mutex_spin_exit(&kq->kq_lock);
1169 				mutex_enter(&fdp->fd_lock);
1170 				knote_detach(kn, fdp, true);
1171 				mutex_spin_enter(&kq->kq_lock);
1172 			} else if (kn->kn_flags & EV_CLEAR) {
1173 				/* clear state after retrieval */
1174 				kn->kn_data = 0;
1175 				kn->kn_fflags = 0;
1176 				kn->kn_status &= ~KN_ACTIVE;
1177 			} else {
1178 				/* add event back on list */
1179 				kq_check(kq);
1180 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1181 				kq->kq_count++;
1182 				kn->kn_status |= KN_QUEUED;
1183 				kq_check(kq);
1184 			}
1185 			if (nkev == kevcnt) {
1186 				/* do copyouts in kevcnt chunks */
1187 				mutex_spin_exit(&kq->kq_lock);
1188 				error = (*keops->keo_put_events)
1189 				    (keops->keo_private,
1190 				    kevbuf, ulistp, nevents, nkev);
1191 				mutex_spin_enter(&kq->kq_lock);
1192 				nevents += nkev;
1193 				nkev = 0;
1194 				kevp = kevbuf;
1195 			}
1196 			count--;
1197 			if (error != 0 || count == 0) {
1198 				/* remove marker */
1199 				TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1200 				break;
1201 			}
1202 		}
1203 	}
1204  done:
1205  	mutex_spin_exit(&kq->kq_lock);
1206 	if (marker != NULL)
1207 		kmem_free(marker, sizeof(*marker));
1208 	if (nkev != 0) {
1209 		/* copyout remaining events */
1210 		error = (*keops->keo_put_events)(keops->keo_private,
1211 		    kevbuf, ulistp, nevents, nkev);
1212 	}
1213 	*retval = maxevents - count;
1214 
1215 	return error;
1216 }
1217 
1218 /*
1219  * fileops ioctl method for a kqueue descriptor.
1220  *
1221  * Two ioctls are currently supported. They both use struct kfilter_mapping:
1222  *	KFILTER_BYNAME		find name for filter, and return result in
1223  *				name, which is of size len.
1224  *	KFILTER_BYFILTER	find filter for name. len is ignored.
1225  */
1226 /*ARGSUSED*/
1227 static int
1228 kqueue_ioctl(file_t *fp, u_long com, void *data)
1229 {
1230 	struct kfilter_mapping	*km;
1231 	const struct kfilter	*kfilter;
1232 	char			*name;
1233 	int			error;
1234 
1235 	km = data;
1236 	error = 0;
1237 	name = kmem_alloc(KFILTER_MAXNAME, KM_SLEEP);
1238 
1239 	switch (com) {
1240 	case KFILTER_BYFILTER:	/* convert filter -> name */
1241 		rw_enter(&kqueue_filter_lock, RW_READER);
1242 		kfilter = kfilter_byfilter(km->filter);
1243 		if (kfilter != NULL) {
1244 			strlcpy(name, kfilter->name, KFILTER_MAXNAME);
1245 			rw_exit(&kqueue_filter_lock);
1246 			error = copyoutstr(name, km->name, km->len, NULL);
1247 		} else {
1248 			rw_exit(&kqueue_filter_lock);
1249 			error = ENOENT;
1250 		}
1251 		break;
1252 
1253 	case KFILTER_BYNAME:	/* convert name -> filter */
1254 		error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL);
1255 		if (error) {
1256 			break;
1257 		}
1258 		rw_enter(&kqueue_filter_lock, RW_READER);
1259 		kfilter = kfilter_byname(name);
1260 		if (kfilter != NULL)
1261 			km->filter = kfilter->filter;
1262 		else
1263 			error = ENOENT;
1264 		rw_exit(&kqueue_filter_lock);
1265 		break;
1266 
1267 	default:
1268 		error = ENOTTY;
1269 		break;
1270 
1271 	}
1272 	kmem_free(name, KFILTER_MAXNAME);
1273 	return (error);
1274 }
1275 
1276 /*
1277  * fileops fcntl method for a kqueue descriptor.
1278  */
1279 static int
1280 kqueue_fcntl(file_t *fp, u_int com, void *data)
1281 {
1282 
1283 	return (ENOTTY);
1284 }
1285 
1286 /*
1287  * fileops poll method for a kqueue descriptor.
1288  * Determine if kqueue has events pending.
1289  */
1290 static int
1291 kqueue_poll(file_t *fp, int events)
1292 {
1293 	struct kqueue	*kq;
1294 	int		revents;
1295 
1296 	kq = fp->f_data;
1297 
1298 	revents = 0;
1299 	if (events & (POLLIN | POLLRDNORM)) {
1300 		mutex_spin_enter(&kq->kq_lock);
1301 		if (kq->kq_count != 0) {
1302 			revents |= events & (POLLIN | POLLRDNORM);
1303 		} else {
1304 			selrecord(curlwp, &kq->kq_sel);
1305 		}
1306 		kq_check(kq);
1307 		mutex_spin_exit(&kq->kq_lock);
1308 	}
1309 
1310 	return revents;
1311 }
1312 
1313 /*
1314  * fileops stat method for a kqueue descriptor.
1315  * Returns dummy info, with st_size being number of events pending.
1316  */
1317 static int
1318 kqueue_stat(file_t *fp, struct stat *st)
1319 {
1320 	struct kqueue *kq;
1321 
1322 	kq = fp->f_data;
1323 
1324 	memset(st, 0, sizeof(*st));
1325 	st->st_size = kq->kq_count;
1326 	st->st_blksize = sizeof(struct kevent);
1327 	st->st_mode = S_IFIFO;
1328 
1329 	return 0;
1330 }
1331 
1332 static void
1333 kqueue_doclose(struct kqueue *kq, struct klist *list, int fd)
1334 {
1335 	struct knote *kn;
1336 	filedesc_t *fdp;
1337 
1338 	fdp = kq->kq_fdp;
1339 
1340 	KASSERT(mutex_owned(&fdp->fd_lock));
1341 
1342 	for (kn = SLIST_FIRST(list); kn != NULL;) {
1343 		if (kq != kn->kn_kq) {
1344 			kn = SLIST_NEXT(kn, kn_link);
1345 			continue;
1346 		}
1347 		knote_detach(kn, fdp, true);
1348 		mutex_enter(&fdp->fd_lock);
1349 		kn = SLIST_FIRST(list);
1350 	}
1351 }
1352 
1353 
1354 /*
1355  * fileops close method for a kqueue descriptor.
1356  */
1357 static int
1358 kqueue_close(file_t *fp)
1359 {
1360 	struct kqueue *kq;
1361 	filedesc_t *fdp;
1362 	fdfile_t *ff;
1363 	int i;
1364 
1365 	kq = fp->f_data;
1366 	fdp = curlwp->l_fd;
1367 
1368 	mutex_enter(&fdp->fd_lock);
1369 	for (i = 0; i <= fdp->fd_lastkqfile; i++) {
1370 		if ((ff = fdp->fd_ofiles[i]) == NULL)
1371 			continue;
1372 		kqueue_doclose(kq, (struct klist *)&ff->ff_knlist, i);
1373 	}
1374 	if (fdp->fd_knhashmask != 0) {
1375 		for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
1376 			kqueue_doclose(kq, &fdp->fd_knhash[i], -1);
1377 		}
1378 	}
1379 	mutex_exit(&fdp->fd_lock);
1380 
1381 	KASSERT(kq->kq_count == 0);
1382 	mutex_destroy(&kq->kq_lock);
1383 	cv_destroy(&kq->kq_cv);
1384 	seldestroy(&kq->kq_sel);
1385 	kmem_free(kq, sizeof(*kq));
1386 	fp->f_data = NULL;
1387 
1388 	return (0);
1389 }
1390 
1391 /*
1392  * struct fileops kqfilter method for a kqueue descriptor.
1393  * Event triggered when monitored kqueue changes.
1394  */
1395 static int
1396 kqueue_kqfilter(file_t *fp, struct knote *kn)
1397 {
1398 	struct kqueue *kq;
1399 	filedesc_t *fdp;
1400 
1401 	kq = ((file_t *)kn->kn_obj)->f_data;
1402 
1403 	KASSERT(fp == kn->kn_obj);
1404 
1405 	if (kn->kn_filter != EVFILT_READ)
1406 		return 1;
1407 
1408 	kn->kn_fop = &kqread_filtops;
1409 	fdp = curlwp->l_fd;
1410 	mutex_enter(&kq->kq_lock);
1411 	SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext);
1412 	mutex_exit(&kq->kq_lock);
1413 
1414 	return 0;
1415 }
1416 
1417 
1418 /*
1419  * Walk down a list of knotes, activating them if their event has
1420  * triggered.  The caller's object lock (e.g. device driver lock)
1421  * must be held.
1422  */
1423 void
1424 knote(struct klist *list, long hint)
1425 {
1426 	struct knote *kn;
1427 
1428 	SLIST_FOREACH(kn, list, kn_selnext) {
1429 		if ((*kn->kn_fop->f_event)(kn, hint))
1430 			knote_activate(kn);
1431 	}
1432 }
1433 
1434 /*
1435  * Remove all knotes referencing a specified fd
1436  */
1437 void
1438 knote_fdclose(int fd)
1439 {
1440 	struct klist *list;
1441 	struct knote *kn;
1442 	filedesc_t *fdp;
1443 
1444 	fdp = curlwp->l_fd;
1445 	list = (struct klist *)&fdp->fd_ofiles[fd]->ff_knlist;
1446 	mutex_enter(&fdp->fd_lock);
1447 	while ((kn = SLIST_FIRST(list)) != NULL) {
1448 		knote_detach(kn, fdp, true);
1449 		mutex_enter(&fdp->fd_lock);
1450 	}
1451 	mutex_exit(&fdp->fd_lock);
1452 }
1453 
1454 /*
1455  * Drop knote.  Called with fdp->fd_lock held, and will drop before
1456  * returning.
1457  */
1458 static void
1459 knote_detach(struct knote *kn, filedesc_t *fdp, bool dofop)
1460 {
1461 	struct klist *list;
1462 	struct kqueue *kq;
1463 
1464 	kq = kn->kn_kq;
1465 
1466 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1467 	KASSERT(mutex_owned(&fdp->fd_lock));
1468 
1469 	/* Remove from monitored object. */
1470 	if (dofop) {
1471 		KERNEL_LOCK(1, NULL);		/* XXXSMP */
1472 		(*kn->kn_fop->f_detach)(kn);
1473 		KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
1474 	}
1475 
1476 	/* Remove from descriptor table. */
1477 	if (kn->kn_fop->f_isfd)
1478 		list = (struct klist *)&fdp->fd_ofiles[kn->kn_id]->ff_knlist;
1479 	else
1480 		list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1481 
1482 	SLIST_REMOVE(list, kn, knote, kn_link);
1483 
1484 	/* Remove from kqueue. */
1485 	/* XXXAD should verify not in use by kqueue_scan. */
1486 	mutex_spin_enter(&kq->kq_lock);
1487 	if ((kn->kn_status & KN_QUEUED) != 0) {
1488 		kq_check(kq);
1489 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1490 		kn->kn_status &= ~KN_QUEUED;
1491 		kq->kq_count--;
1492 		kq_check(kq);
1493 	}
1494 	mutex_spin_exit(&kq->kq_lock);
1495 
1496 	mutex_exit(&fdp->fd_lock);
1497 	if (kn->kn_fop->f_isfd)
1498 		fd_putfile(kn->kn_id);
1499 	atomic_dec_uint(&kn->kn_kfilter->refcnt);
1500 	kmem_free(kn, sizeof(*kn));
1501 }
1502 
1503 /*
1504  * Queue new event for knote.
1505  */
1506 static void
1507 knote_enqueue(struct knote *kn)
1508 {
1509 	struct kqueue *kq;
1510 
1511 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1512 
1513 	kq = kn->kn_kq;
1514 
1515 	mutex_spin_enter(&kq->kq_lock);
1516 	if ((kn->kn_status & KN_DISABLED) != 0) {
1517 		kn->kn_status &= ~KN_DISABLED;
1518 	}
1519 	if ((kn->kn_status & (KN_ACTIVE | KN_QUEUED)) == KN_ACTIVE) {
1520 		kq_check(kq);
1521 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1522 		kn->kn_status |= KN_QUEUED;
1523 		kq->kq_count++;
1524 		kq_check(kq);
1525 		cv_broadcast(&kq->kq_cv);
1526 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1527 	}
1528 	mutex_spin_exit(&kq->kq_lock);
1529 }
1530 /*
1531  * Queue new event for knote.
1532  */
1533 static void
1534 knote_activate(struct knote *kn)
1535 {
1536 	struct kqueue *kq;
1537 
1538 	KASSERT((kn->kn_status & KN_MARKER) == 0);
1539 
1540 	kq = kn->kn_kq;
1541 
1542 	mutex_spin_enter(&kq->kq_lock);
1543 	kn->kn_status |= KN_ACTIVE;
1544 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) {
1545 		kq_check(kq);
1546 		TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1547 		kn->kn_status |= KN_QUEUED;
1548 		kq->kq_count++;
1549 		kq_check(kq);
1550 		cv_broadcast(&kq->kq_cv);
1551 		selnotify(&kq->kq_sel, 0, NOTE_SUBMIT);
1552 	}
1553 	mutex_spin_exit(&kq->kq_lock);
1554 }
1555