xref: /freebsd/sys/kern/kern_event.c (revision 3ef90571c1feac738dd60ea1516df2f974f18b56)
1 /*-
2  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
3  * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org>
4  * Copyright (c) 2009 Apple, 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 AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_ktrace.h"
33 #include "opt_kqueue.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/capsicum.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/rwlock.h>
42 #include <sys/proc.h>
43 #include <sys/malloc.h>
44 #include <sys/unistd.h>
45 #include <sys/file.h>
46 #include <sys/filedesc.h>
47 #include <sys/filio.h>
48 #include <sys/fcntl.h>
49 #include <sys/kthread.h>
50 #include <sys/selinfo.h>
51 #include <sys/stdatomic.h>
52 #include <sys/queue.h>
53 #include <sys/event.h>
54 #include <sys/eventvar.h>
55 #include <sys/poll.h>
56 #include <sys/protosw.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sigio.h>
59 #include <sys/signalvar.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysproto.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/taskqueue.h>
67 #include <sys/uio.h>
68 #include <sys/user.h>
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72 
73 #include <vm/uma.h>
74 
75 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
76 
77 /*
78  * This lock is used if multiple kq locks are required.  This possibly
79  * should be made into a per proc lock.
80  */
81 static struct mtx	kq_global;
82 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
83 #define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
84 	if (!haslck)				\
85 		mtx_lock(lck);			\
86 	haslck = 1;				\
87 } while (0)
88 #define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
89 	if (haslck)				\
90 		mtx_unlock(lck);			\
91 	haslck = 0;				\
92 } while (0)
93 
94 TASKQUEUE_DEFINE_THREAD(kqueue);
95 
96 static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
97 static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
98 static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
99 		    struct thread *td, int waitok);
100 static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
101 static void	kqueue_release(struct kqueue *kq, int locked);
102 static int	kqueue_expand(struct kqueue *kq, struct filterops *fops,
103 		    uintptr_t ident, int waitok);
104 static void	kqueue_task(void *arg, int pending);
105 static int	kqueue_scan(struct kqueue *kq, int maxevents,
106 		    struct kevent_copyops *k_ops,
107 		    const struct timespec *timeout,
108 		    struct kevent *keva, struct thread *td);
109 static void 	kqueue_wakeup(struct kqueue *kq);
110 static struct filterops *kqueue_fo_find(int filt);
111 static void	kqueue_fo_release(int filt);
112 
113 static fo_ioctl_t	kqueue_ioctl;
114 static fo_poll_t	kqueue_poll;
115 static fo_kqfilter_t	kqueue_kqfilter;
116 static fo_stat_t	kqueue_stat;
117 static fo_close_t	kqueue_close;
118 static fo_fill_kinfo_t	kqueue_fill_kinfo;
119 
120 static struct fileops kqueueops = {
121 	.fo_read = invfo_rdwr,
122 	.fo_write = invfo_rdwr,
123 	.fo_truncate = invfo_truncate,
124 	.fo_ioctl = kqueue_ioctl,
125 	.fo_poll = kqueue_poll,
126 	.fo_kqfilter = kqueue_kqfilter,
127 	.fo_stat = kqueue_stat,
128 	.fo_close = kqueue_close,
129 	.fo_chmod = invfo_chmod,
130 	.fo_chown = invfo_chown,
131 	.fo_sendfile = invfo_sendfile,
132 	.fo_fill_kinfo = kqueue_fill_kinfo,
133 };
134 
135 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
136 static void 	knote_drop(struct knote *kn, struct thread *td);
137 static void 	knote_enqueue(struct knote *kn);
138 static void 	knote_dequeue(struct knote *kn);
139 static void 	knote_init(void);
140 static struct 	knote *knote_alloc(int waitok);
141 static void 	knote_free(struct knote *kn);
142 
143 static void	filt_kqdetach(struct knote *kn);
144 static int	filt_kqueue(struct knote *kn, long hint);
145 static int	filt_procattach(struct knote *kn);
146 static void	filt_procdetach(struct knote *kn);
147 static int	filt_proc(struct knote *kn, long hint);
148 static int	filt_fileattach(struct knote *kn);
149 static void	filt_timerexpire(void *knx);
150 static int	filt_timerattach(struct knote *kn);
151 static void	filt_timerdetach(struct knote *kn);
152 static int	filt_timer(struct knote *kn, long hint);
153 static int	filt_userattach(struct knote *kn);
154 static void	filt_userdetach(struct knote *kn);
155 static int	filt_user(struct knote *kn, long hint);
156 static void	filt_usertouch(struct knote *kn, struct kevent *kev,
157 		    u_long type);
158 
159 static struct filterops file_filtops = {
160 	.f_isfd = 1,
161 	.f_attach = filt_fileattach,
162 };
163 static struct filterops kqread_filtops = {
164 	.f_isfd = 1,
165 	.f_detach = filt_kqdetach,
166 	.f_event = filt_kqueue,
167 };
168 /* XXX - move to kern_proc.c?  */
169 static struct filterops proc_filtops = {
170 	.f_isfd = 0,
171 	.f_attach = filt_procattach,
172 	.f_detach = filt_procdetach,
173 	.f_event = filt_proc,
174 };
175 static struct filterops timer_filtops = {
176 	.f_isfd = 0,
177 	.f_attach = filt_timerattach,
178 	.f_detach = filt_timerdetach,
179 	.f_event = filt_timer,
180 };
181 static struct filterops user_filtops = {
182 	.f_attach = filt_userattach,
183 	.f_detach = filt_userdetach,
184 	.f_event = filt_user,
185 	.f_touch = filt_usertouch,
186 };
187 
188 static uma_zone_t	knote_zone;
189 static atomic_uint	kq_ncallouts = ATOMIC_VAR_INIT(0);
190 static unsigned int 	kq_calloutmax = 4 * 1024;
191 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
192     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
193 
194 /* XXX - ensure not KN_INFLUX?? */
195 #define KNOTE_ACTIVATE(kn, islock) do { 				\
196 	if ((islock))							\
197 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
198 	else								\
199 		KQ_LOCK((kn)->kn_kq);					\
200 	(kn)->kn_status |= KN_ACTIVE;					\
201 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
202 		knote_enqueue((kn));					\
203 	if (!(islock))							\
204 		KQ_UNLOCK((kn)->kn_kq);					\
205 } while(0)
206 #define KQ_LOCK(kq) do {						\
207 	mtx_lock(&(kq)->kq_lock);					\
208 } while (0)
209 #define KQ_FLUX_WAKEUP(kq) do {						\
210 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
211 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
212 		wakeup((kq));						\
213 	}								\
214 } while (0)
215 #define KQ_UNLOCK_FLUX(kq) do {						\
216 	KQ_FLUX_WAKEUP(kq);						\
217 	mtx_unlock(&(kq)->kq_lock);					\
218 } while (0)
219 #define KQ_UNLOCK(kq) do {						\
220 	mtx_unlock(&(kq)->kq_lock);					\
221 } while (0)
222 #define KQ_OWNED(kq) do {						\
223 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
224 } while (0)
225 #define KQ_NOTOWNED(kq) do {						\
226 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
227 } while (0)
228 #define KN_LIST_LOCK(kn) do {						\
229 	if (kn->kn_knlist != NULL)					\
230 		kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg);	\
231 } while (0)
232 #define KN_LIST_UNLOCK(kn) do {						\
233 	if (kn->kn_knlist != NULL) 					\
234 		kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg);	\
235 } while (0)
236 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
237 	if (islocked)							\
238 		KNL_ASSERT_LOCKED(knl);				\
239 	else								\
240 		KNL_ASSERT_UNLOCKED(knl);				\
241 } while (0)
242 #ifdef INVARIANTS
243 #define	KNL_ASSERT_LOCKED(knl) do {					\
244 	knl->kl_assert_locked((knl)->kl_lockarg);			\
245 } while (0)
246 #define	KNL_ASSERT_UNLOCKED(knl) do {					\
247 	knl->kl_assert_unlocked((knl)->kl_lockarg);			\
248 } while (0)
249 #else /* !INVARIANTS */
250 #define	KNL_ASSERT_LOCKED(knl) do {} while(0)
251 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
252 #endif /* INVARIANTS */
253 
254 #ifndef	KN_HASHSIZE
255 #define	KN_HASHSIZE		64		/* XXX should be tunable */
256 #endif
257 
258 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
259 
260 static int
261 filt_nullattach(struct knote *kn)
262 {
263 
264 	return (ENXIO);
265 };
266 
267 struct filterops null_filtops = {
268 	.f_isfd = 0,
269 	.f_attach = filt_nullattach,
270 };
271 
272 /* XXX - make SYSINIT to add these, and move into respective modules. */
273 extern struct filterops sig_filtops;
274 extern struct filterops fs_filtops;
275 
276 /*
277  * Table for for all system-defined filters.
278  */
279 static struct mtx	filterops_lock;
280 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
281 	MTX_DEF);
282 static struct {
283 	struct filterops *for_fop;
284 	int for_nolock;
285 	int for_refcnt;
286 } sysfilt_ops[EVFILT_SYSCOUNT] = {
287 	{ &file_filtops, 1 },			/* EVFILT_READ */
288 	{ &file_filtops, 1 },			/* EVFILT_WRITE */
289 	{ &null_filtops },			/* EVFILT_AIO */
290 	{ &file_filtops, 1 },			/* EVFILT_VNODE */
291 	{ &proc_filtops, 1 },			/* EVFILT_PROC */
292 	{ &sig_filtops, 1 },			/* EVFILT_SIGNAL */
293 	{ &timer_filtops, 1 },			/* EVFILT_TIMER */
294 	{ &file_filtops, 1 },			/* EVFILT_PROCDESC */
295 	{ &fs_filtops, 1 },			/* EVFILT_FS */
296 	{ &null_filtops },			/* EVFILT_LIO */
297 	{ &user_filtops, 1 },			/* EVFILT_USER */
298 	{ &null_filtops },			/* EVFILT_SENDFILE */
299 };
300 
301 /*
302  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
303  * method.
304  */
305 static int
306 filt_fileattach(struct knote *kn)
307 {
308 
309 	return (fo_kqfilter(kn->kn_fp, kn));
310 }
311 
312 /*ARGSUSED*/
313 static int
314 kqueue_kqfilter(struct file *fp, struct knote *kn)
315 {
316 	struct kqueue *kq = kn->kn_fp->f_data;
317 
318 	if (kn->kn_filter != EVFILT_READ)
319 		return (EINVAL);
320 
321 	kn->kn_status |= KN_KQUEUE;
322 	kn->kn_fop = &kqread_filtops;
323 	knlist_add(&kq->kq_sel.si_note, kn, 0);
324 
325 	return (0);
326 }
327 
328 static void
329 filt_kqdetach(struct knote *kn)
330 {
331 	struct kqueue *kq = kn->kn_fp->f_data;
332 
333 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
334 }
335 
336 /*ARGSUSED*/
337 static int
338 filt_kqueue(struct knote *kn, long hint)
339 {
340 	struct kqueue *kq = kn->kn_fp->f_data;
341 
342 	kn->kn_data = kq->kq_count;
343 	return (kn->kn_data > 0);
344 }
345 
346 /* XXX - move to kern_proc.c?  */
347 static int
348 filt_procattach(struct knote *kn)
349 {
350 	struct proc *p;
351 	int immediate;
352 	int error;
353 
354 	immediate = 0;
355 	p = pfind(kn->kn_id);
356 	if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
357 		p = zpfind(kn->kn_id);
358 		immediate = 1;
359 	} else if (p != NULL && (p->p_flag & P_WEXIT)) {
360 		immediate = 1;
361 	}
362 
363 	if (p == NULL)
364 		return (ESRCH);
365 	if ((error = p_cansee(curthread, p))) {
366 		PROC_UNLOCK(p);
367 		return (error);
368 	}
369 
370 	kn->kn_ptr.p_proc = p;
371 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
372 
373 	/*
374 	 * internal flag indicating registration done by kernel
375 	 */
376 	if (kn->kn_flags & EV_FLAG1) {
377 		kn->kn_data = kn->kn_sdata;		/* ppid */
378 		kn->kn_fflags = NOTE_CHILD;
379 		kn->kn_flags &= ~EV_FLAG1;
380 	}
381 
382 	if (immediate == 0)
383 		knlist_add(&p->p_klist, kn, 1);
384 
385 	/*
386 	 * Immediately activate any exit notes if the target process is a
387 	 * zombie.  This is necessary to handle the case where the target
388 	 * process, e.g. a child, dies before the kevent is registered.
389 	 */
390 	if (immediate && filt_proc(kn, NOTE_EXIT))
391 		KNOTE_ACTIVATE(kn, 0);
392 
393 	PROC_UNLOCK(p);
394 
395 	return (0);
396 }
397 
398 /*
399  * The knote may be attached to a different process, which may exit,
400  * leaving nothing for the knote to be attached to.  So when the process
401  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
402  * it will be deleted when read out.  However, as part of the knote deletion,
403  * this routine is called, so a check is needed to avoid actually performing
404  * a detach, because the original process does not exist any more.
405  */
406 /* XXX - move to kern_proc.c?  */
407 static void
408 filt_procdetach(struct knote *kn)
409 {
410 	struct proc *p;
411 
412 	p = kn->kn_ptr.p_proc;
413 	knlist_remove(&p->p_klist, kn, 0);
414 	kn->kn_ptr.p_proc = NULL;
415 }
416 
417 /* XXX - move to kern_proc.c?  */
418 static int
419 filt_proc(struct knote *kn, long hint)
420 {
421 	struct proc *p;
422 	u_int event;
423 
424 	p = kn->kn_ptr.p_proc;
425 	/* Mask off extra data. */
426 	event = (u_int)hint & NOTE_PCTRLMASK;
427 
428 	/* If the user is interested in this event, record it. */
429 	if (kn->kn_sfflags & event)
430 		kn->kn_fflags |= event;
431 
432 	/* Process is gone, so flag the event as finished. */
433 	if (event == NOTE_EXIT) {
434 		if (!(kn->kn_status & KN_DETACHED))
435 			knlist_remove_inevent(&p->p_klist, kn);
436 		kn->kn_flags |= EV_EOF | EV_ONESHOT;
437 		kn->kn_ptr.p_proc = NULL;
438 		if (kn->kn_fflags & NOTE_EXIT)
439 			kn->kn_data = p->p_xstat;
440 		if (kn->kn_fflags == 0)
441 			kn->kn_flags |= EV_DROP;
442 		return (1);
443 	}
444 
445 	return (kn->kn_fflags != 0);
446 }
447 
448 /*
449  * Called when the process forked. It mostly does the same as the
450  * knote(), activating all knotes registered to be activated when the
451  * process forked. Additionally, for each knote attached to the
452  * parent, check whether user wants to track the new process. If so
453  * attach a new knote to it, and immediately report an event with the
454  * child's pid.
455  */
456 void
457 knote_fork(struct knlist *list, int pid)
458 {
459 	struct kqueue *kq;
460 	struct knote *kn;
461 	struct kevent kev;
462 	int error;
463 
464 	if (list == NULL)
465 		return;
466 	list->kl_lock(list->kl_lockarg);
467 
468 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
469 		/*
470 		 * XXX - Why do we skip the kn if it is _INFLUX?  Does this
471 		 * mean we will not properly wake up some notes?
472 		 */
473 		if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
474 			continue;
475 		kq = kn->kn_kq;
476 		KQ_LOCK(kq);
477 		if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
478 			KQ_UNLOCK(kq);
479 			continue;
480 		}
481 
482 		/*
483 		 * The same as knote(), activate the event.
484 		 */
485 		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
486 			kn->kn_status |= KN_HASKQLOCK;
487 			if (kn->kn_fop->f_event(kn, NOTE_FORK))
488 				KNOTE_ACTIVATE(kn, 1);
489 			kn->kn_status &= ~KN_HASKQLOCK;
490 			KQ_UNLOCK(kq);
491 			continue;
492 		}
493 
494 		/*
495 		 * The NOTE_TRACK case. In addition to the activation
496 		 * of the event, we need to register new event to
497 		 * track the child. Drop the locks in preparation for
498 		 * the call to kqueue_register().
499 		 */
500 		kn->kn_status |= KN_INFLUX;
501 		KQ_UNLOCK(kq);
502 		list->kl_unlock(list->kl_lockarg);
503 
504 		/*
505 		 * Activate existing knote and register a knote with
506 		 * new process.
507 		 */
508 		kev.ident = pid;
509 		kev.filter = kn->kn_filter;
510 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
511 		kev.fflags = kn->kn_sfflags;
512 		kev.data = kn->kn_id;		/* parent */
513 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
514 		error = kqueue_register(kq, &kev, NULL, 0);
515 		if (error)
516 			kn->kn_fflags |= NOTE_TRACKERR;
517 		if (kn->kn_fop->f_event(kn, NOTE_FORK))
518 			KNOTE_ACTIVATE(kn, 0);
519 		KQ_LOCK(kq);
520 		kn->kn_status &= ~KN_INFLUX;
521 		KQ_UNLOCK_FLUX(kq);
522 		list->kl_lock(list->kl_lockarg);
523 	}
524 	list->kl_unlock(list->kl_lockarg);
525 }
526 
527 /*
528  * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
529  * interval timer support code.
530  */
531 
532 #define NOTE_TIMER_PRECMASK	(NOTE_SECONDS|NOTE_MSECONDS|NOTE_USECONDS| \
533 				NOTE_NSECONDS)
534 
535 static __inline sbintime_t
536 timer2sbintime(intptr_t data, int flags)
537 {
538 	sbintime_t modifier;
539 
540 	switch (flags & NOTE_TIMER_PRECMASK) {
541 	case NOTE_SECONDS:
542 		modifier = SBT_1S;
543 		break;
544 	case NOTE_MSECONDS: /* FALLTHROUGH */
545 	case 0:
546 		modifier = SBT_1MS;
547 		break;
548 	case NOTE_USECONDS:
549 		modifier = SBT_1US;
550 		break;
551 	case NOTE_NSECONDS:
552 		modifier = SBT_1NS;
553 		break;
554 	default:
555 		return (-1);
556 	}
557 
558 #ifdef __LP64__
559 	if (data > SBT_MAX / modifier)
560 		return (SBT_MAX);
561 #endif
562 	return (modifier * data);
563 }
564 
565 static void
566 filt_timerexpire(void *knx)
567 {
568 	struct callout *calloutp;
569 	struct knote *kn;
570 
571 	kn = knx;
572 	kn->kn_data++;
573 	KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
574 
575 	if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
576 		calloutp = (struct callout *)kn->kn_hook;
577 		*kn->kn_ptr.p_nexttime += timer2sbintime(kn->kn_sdata,
578 		    kn->kn_sfflags);
579 		callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0,
580 		    filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE);
581 	}
582 }
583 
584 /*
585  * data contains amount of time to sleep
586  */
587 static int
588 filt_timerattach(struct knote *kn)
589 {
590 	struct callout *calloutp;
591 	sbintime_t to;
592 	unsigned int ncallouts;
593 
594 	if ((intptr_t)kn->kn_sdata < 0)
595 		return (EINVAL);
596 	if ((intptr_t)kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
597 		kn->kn_sdata = 1;
598 	/* Only precision unit are supported in flags so far */
599 	if (kn->kn_sfflags & ~NOTE_TIMER_PRECMASK)
600 		return (EINVAL);
601 
602 	to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
603 	if (to < 0)
604 		return (EINVAL);
605 
606 	ncallouts = atomic_load_explicit(&kq_ncallouts, memory_order_relaxed);
607 	do {
608 		if (ncallouts >= kq_calloutmax)
609 			return (ENOMEM);
610 	} while (!atomic_compare_exchange_weak_explicit(&kq_ncallouts,
611 	    &ncallouts, ncallouts + 1, memory_order_relaxed,
612 	    memory_order_relaxed));
613 
614 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
615 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
616 	kn->kn_ptr.p_nexttime = malloc(sizeof(sbintime_t), M_KQUEUE, M_WAITOK);
617 	calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK);
618 	callout_init(calloutp, 1);
619 	kn->kn_hook = calloutp;
620 	*kn->kn_ptr.p_nexttime = to + sbinuptime();
621 	callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0,
622 	    filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE);
623 
624 	return (0);
625 }
626 
627 static void
628 filt_timerdetach(struct knote *kn)
629 {
630 	struct callout *calloutp;
631 	unsigned int old;
632 
633 	calloutp = (struct callout *)kn->kn_hook;
634 	callout_drain(calloutp);
635 	free(calloutp, M_KQUEUE);
636 	free(kn->kn_ptr.p_nexttime, M_KQUEUE);
637 	old = atomic_fetch_sub_explicit(&kq_ncallouts, 1, memory_order_relaxed);
638 	KASSERT(old > 0, ("Number of callouts cannot become negative"));
639 	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
640 }
641 
642 static int
643 filt_timer(struct knote *kn, long hint)
644 {
645 
646 	return (kn->kn_data != 0);
647 }
648 
649 static int
650 filt_userattach(struct knote *kn)
651 {
652 
653 	/*
654 	 * EVFILT_USER knotes are not attached to anything in the kernel.
655 	 */
656 	kn->kn_hook = NULL;
657 	if (kn->kn_fflags & NOTE_TRIGGER)
658 		kn->kn_hookid = 1;
659 	else
660 		kn->kn_hookid = 0;
661 	return (0);
662 }
663 
664 static void
665 filt_userdetach(__unused struct knote *kn)
666 {
667 
668 	/*
669 	 * EVFILT_USER knotes are not attached to anything in the kernel.
670 	 */
671 }
672 
673 static int
674 filt_user(struct knote *kn, __unused long hint)
675 {
676 
677 	return (kn->kn_hookid);
678 }
679 
680 static void
681 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
682 {
683 	u_int ffctrl;
684 
685 	switch (type) {
686 	case EVENT_REGISTER:
687 		if (kev->fflags & NOTE_TRIGGER)
688 			kn->kn_hookid = 1;
689 
690 		ffctrl = kev->fflags & NOTE_FFCTRLMASK;
691 		kev->fflags &= NOTE_FFLAGSMASK;
692 		switch (ffctrl) {
693 		case NOTE_FFNOP:
694 			break;
695 
696 		case NOTE_FFAND:
697 			kn->kn_sfflags &= kev->fflags;
698 			break;
699 
700 		case NOTE_FFOR:
701 			kn->kn_sfflags |= kev->fflags;
702 			break;
703 
704 		case NOTE_FFCOPY:
705 			kn->kn_sfflags = kev->fflags;
706 			break;
707 
708 		default:
709 			/* XXX Return error? */
710 			break;
711 		}
712 		kn->kn_sdata = kev->data;
713 		if (kev->flags & EV_CLEAR) {
714 			kn->kn_hookid = 0;
715 			kn->kn_data = 0;
716 			kn->kn_fflags = 0;
717 		}
718 		break;
719 
720         case EVENT_PROCESS:
721 		*kev = kn->kn_kevent;
722 		kev->fflags = kn->kn_sfflags;
723 		kev->data = kn->kn_sdata;
724 		if (kn->kn_flags & EV_CLEAR) {
725 			kn->kn_hookid = 0;
726 			kn->kn_data = 0;
727 			kn->kn_fflags = 0;
728 		}
729 		break;
730 
731 	default:
732 		panic("filt_usertouch() - invalid type (%ld)", type);
733 		break;
734 	}
735 }
736 
737 int
738 sys_kqueue(struct thread *td, struct kqueue_args *uap)
739 {
740 
741 	return (kern_kqueue(td, 0));
742 }
743 
744 int
745 kern_kqueue(struct thread *td, int flags)
746 {
747 	struct filedesc *fdp;
748 	struct kqueue *kq;
749 	struct file *fp;
750 	struct proc *p;
751 	struct ucred *cred;
752 	int fd, error;
753 
754 	p = td->td_proc;
755 	cred = td->td_ucred;
756 	crhold(cred);
757 	PROC_LOCK(p);
758 	if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td->td_proc,
759 	    RLIMIT_KQUEUES))) {
760 		PROC_UNLOCK(p);
761 		crfree(cred);
762 		return (ENOMEM);
763 	}
764 	PROC_UNLOCK(p);
765 
766 	fdp = p->p_fd;
767 	error = falloc(td, &fp, &fd, flags);
768 	if (error)
769 		goto done2;
770 
771 	/* An extra reference on `fp' has been held for us by falloc(). */
772 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
773 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
774 	TAILQ_INIT(&kq->kq_head);
775 	kq->kq_fdp = fdp;
776 	kq->kq_cred = cred;
777 	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
778 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
779 
780 	FILEDESC_XLOCK(fdp);
781 	TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
782 	FILEDESC_XUNLOCK(fdp);
783 
784 	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
785 	fdrop(fp, td);
786 
787 	td->td_retval[0] = fd;
788 done2:
789 	if (error != 0) {
790 		chgkqcnt(cred->cr_ruidinfo, -1, 0);
791 		crfree(cred);
792 	}
793 	return (error);
794 }
795 
796 #ifndef _SYS_SYSPROTO_H_
797 struct kevent_args {
798 	int	fd;
799 	const struct kevent *changelist;
800 	int	nchanges;
801 	struct	kevent *eventlist;
802 	int	nevents;
803 	const struct timespec *timeout;
804 };
805 #endif
806 int
807 sys_kevent(struct thread *td, struct kevent_args *uap)
808 {
809 	struct timespec ts, *tsp;
810 	struct kevent_copyops k_ops = { uap,
811 					kevent_copyout,
812 					kevent_copyin};
813 	int error;
814 #ifdef KTRACE
815 	struct uio ktruio;
816 	struct iovec ktriov;
817 	struct uio *ktruioin = NULL;
818 	struct uio *ktruioout = NULL;
819 #endif
820 
821 	if (uap->timeout != NULL) {
822 		error = copyin(uap->timeout, &ts, sizeof(ts));
823 		if (error)
824 			return (error);
825 		tsp = &ts;
826 	} else
827 		tsp = NULL;
828 
829 #ifdef KTRACE
830 	if (KTRPOINT(td, KTR_GENIO)) {
831 		ktriov.iov_base = uap->changelist;
832 		ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
833 		ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
834 		    .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
835 		    .uio_td = td };
836 		ktruioin = cloneuio(&ktruio);
837 		ktriov.iov_base = uap->eventlist;
838 		ktriov.iov_len = uap->nevents * sizeof(struct kevent);
839 		ktruioout = cloneuio(&ktruio);
840 	}
841 #endif
842 
843 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
844 	    &k_ops, tsp);
845 
846 #ifdef KTRACE
847 	if (ktruioin != NULL) {
848 		ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
849 		ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
850 		ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
851 		ktrgenio(uap->fd, UIO_READ, ktruioout, error);
852 	}
853 #endif
854 
855 	return (error);
856 }
857 
858 /*
859  * Copy 'count' items into the destination list pointed to by uap->eventlist.
860  */
861 static int
862 kevent_copyout(void *arg, struct kevent *kevp, int count)
863 {
864 	struct kevent_args *uap;
865 	int error;
866 
867 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
868 	uap = (struct kevent_args *)arg;
869 
870 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
871 	if (error == 0)
872 		uap->eventlist += count;
873 	return (error);
874 }
875 
876 /*
877  * Copy 'count' items from the list pointed to by uap->changelist.
878  */
879 static int
880 kevent_copyin(void *arg, struct kevent *kevp, int count)
881 {
882 	struct kevent_args *uap;
883 	int error;
884 
885 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
886 	uap = (struct kevent_args *)arg;
887 
888 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
889 	if (error == 0)
890 		uap->changelist += count;
891 	return (error);
892 }
893 
894 int
895 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
896     struct kevent_copyops *k_ops, const struct timespec *timeout)
897 {
898 	cap_rights_t rights;
899 	struct file *fp;
900 	int error;
901 
902 	cap_rights_init(&rights);
903 	if (nchanges > 0)
904 		cap_rights_set(&rights, CAP_KQUEUE_CHANGE);
905 	if (nevents > 0)
906 		cap_rights_set(&rights, CAP_KQUEUE_EVENT);
907 	error = fget(td, fd, &rights, &fp);
908 	if (error != 0)
909 		return (error);
910 
911 	error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
912 	fdrop(fp, td);
913 
914 	return (error);
915 }
916 
917 int
918 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
919     struct kevent_copyops *k_ops, const struct timespec *timeout)
920 {
921 	struct kevent keva[KQ_NEVENTS];
922 	struct kevent *kevp, *changes;
923 	struct kqueue *kq;
924 	int i, n, nerrors, error;
925 
926 	error = kqueue_acquire(fp, &kq);
927 	if (error != 0)
928 		return (error);
929 
930 	nerrors = 0;
931 
932 	while (nchanges > 0) {
933 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
934 		error = k_ops->k_copyin(k_ops->arg, keva, n);
935 		if (error)
936 			goto done;
937 		changes = keva;
938 		for (i = 0; i < n; i++) {
939 			kevp = &changes[i];
940 			if (!kevp->filter)
941 				continue;
942 			kevp->flags &= ~EV_SYSFLAGS;
943 			error = kqueue_register(kq, kevp, td, 1);
944 			if (error || (kevp->flags & EV_RECEIPT)) {
945 				if (nevents != 0) {
946 					kevp->flags = EV_ERROR;
947 					kevp->data = error;
948 					(void) k_ops->k_copyout(k_ops->arg,
949 					    kevp, 1);
950 					nevents--;
951 					nerrors++;
952 				} else {
953 					goto done;
954 				}
955 			}
956 		}
957 		nchanges -= n;
958 	}
959 	if (nerrors) {
960 		td->td_retval[0] = nerrors;
961 		error = 0;
962 		goto done;
963 	}
964 
965 	error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
966 done:
967 	kqueue_release(kq, 0);
968 	return (error);
969 }
970 
971 int
972 kqueue_add_filteropts(int filt, struct filterops *filtops)
973 {
974 	int error;
975 
976 	error = 0;
977 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
978 		printf(
979 "trying to add a filterop that is out of range: %d is beyond %d\n",
980 		    ~filt, EVFILT_SYSCOUNT);
981 		return EINVAL;
982 	}
983 	mtx_lock(&filterops_lock);
984 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
985 	    sysfilt_ops[~filt].for_fop != NULL)
986 		error = EEXIST;
987 	else {
988 		sysfilt_ops[~filt].for_fop = filtops;
989 		sysfilt_ops[~filt].for_refcnt = 0;
990 	}
991 	mtx_unlock(&filterops_lock);
992 
993 	return (error);
994 }
995 
996 int
997 kqueue_del_filteropts(int filt)
998 {
999 	int error;
1000 
1001 	error = 0;
1002 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1003 		return EINVAL;
1004 
1005 	mtx_lock(&filterops_lock);
1006 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1007 	    sysfilt_ops[~filt].for_fop == NULL)
1008 		error = EINVAL;
1009 	else if (sysfilt_ops[~filt].for_refcnt != 0)
1010 		error = EBUSY;
1011 	else {
1012 		sysfilt_ops[~filt].for_fop = &null_filtops;
1013 		sysfilt_ops[~filt].for_refcnt = 0;
1014 	}
1015 	mtx_unlock(&filterops_lock);
1016 
1017 	return error;
1018 }
1019 
1020 static struct filterops *
1021 kqueue_fo_find(int filt)
1022 {
1023 
1024 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1025 		return NULL;
1026 
1027 	if (sysfilt_ops[~filt].for_nolock)
1028 		return sysfilt_ops[~filt].for_fop;
1029 
1030 	mtx_lock(&filterops_lock);
1031 	sysfilt_ops[~filt].for_refcnt++;
1032 	if (sysfilt_ops[~filt].for_fop == NULL)
1033 		sysfilt_ops[~filt].for_fop = &null_filtops;
1034 	mtx_unlock(&filterops_lock);
1035 
1036 	return sysfilt_ops[~filt].for_fop;
1037 }
1038 
1039 static void
1040 kqueue_fo_release(int filt)
1041 {
1042 
1043 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1044 		return;
1045 
1046 	if (sysfilt_ops[~filt].for_nolock)
1047 		return;
1048 
1049 	mtx_lock(&filterops_lock);
1050 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1051 	    ("filter object refcount not valid on release"));
1052 	sysfilt_ops[~filt].for_refcnt--;
1053 	mtx_unlock(&filterops_lock);
1054 }
1055 
1056 /*
1057  * A ref to kq (obtained via kqueue_acquire) must be held.  waitok will
1058  * influence if memory allocation should wait.  Make sure it is 0 if you
1059  * hold any mutexes.
1060  */
1061 static int
1062 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
1063 {
1064 	struct filterops *fops;
1065 	struct file *fp;
1066 	struct knote *kn, *tkn;
1067 	cap_rights_t rights;
1068 	int error, filt, event;
1069 	int haskqglobal, filedesc_unlock;
1070 
1071 	fp = NULL;
1072 	kn = NULL;
1073 	error = 0;
1074 	haskqglobal = 0;
1075 	filedesc_unlock = 0;
1076 
1077 	filt = kev->filter;
1078 	fops = kqueue_fo_find(filt);
1079 	if (fops == NULL)
1080 		return EINVAL;
1081 
1082 	if (kev->flags & EV_ADD)
1083 		tkn = knote_alloc(waitok);	/* prevent waiting with locks */
1084 	else
1085 		tkn = NULL;
1086 
1087 findkn:
1088 	if (fops->f_isfd) {
1089 		KASSERT(td != NULL, ("td is NULL"));
1090 		error = fget(td, kev->ident,
1091 		    cap_rights_init(&rights, CAP_EVENT), &fp);
1092 		if (error)
1093 			goto done;
1094 
1095 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1096 		    kev->ident, 0) != 0) {
1097 			/* try again */
1098 			fdrop(fp, td);
1099 			fp = NULL;
1100 			error = kqueue_expand(kq, fops, kev->ident, waitok);
1101 			if (error)
1102 				goto done;
1103 			goto findkn;
1104 		}
1105 
1106 		if (fp->f_type == DTYPE_KQUEUE) {
1107 			/*
1108 			 * if we add some inteligence about what we are doing,
1109 			 * we should be able to support events on ourselves.
1110 			 * We need to know when we are doing this to prevent
1111 			 * getting both the knlist lock and the kq lock since
1112 			 * they are the same thing.
1113 			 */
1114 			if (fp->f_data == kq) {
1115 				error = EINVAL;
1116 				goto done;
1117 			}
1118 
1119 			/*
1120 			 * Pre-lock the filedesc before the global
1121 			 * lock mutex, see the comment in
1122 			 * kqueue_close().
1123 			 */
1124 			FILEDESC_XLOCK(td->td_proc->p_fd);
1125 			filedesc_unlock = 1;
1126 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1127 		}
1128 
1129 		KQ_LOCK(kq);
1130 		if (kev->ident < kq->kq_knlistsize) {
1131 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1132 				if (kev->filter == kn->kn_filter)
1133 					break;
1134 		}
1135 	} else {
1136 		if ((kev->flags & EV_ADD) == EV_ADD)
1137 			kqueue_expand(kq, fops, kev->ident, waitok);
1138 
1139 		KQ_LOCK(kq);
1140 		if (kq->kq_knhashmask != 0) {
1141 			struct klist *list;
1142 
1143 			list = &kq->kq_knhash[
1144 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1145 			SLIST_FOREACH(kn, list, kn_link)
1146 				if (kev->ident == kn->kn_id &&
1147 				    kev->filter == kn->kn_filter)
1148 					break;
1149 		}
1150 	}
1151 
1152 	/* knote is in the process of changing, wait for it to stablize. */
1153 	if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1154 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1155 		if (filedesc_unlock) {
1156 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1157 			filedesc_unlock = 0;
1158 		}
1159 		kq->kq_state |= KQ_FLUXWAIT;
1160 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1161 		if (fp != NULL) {
1162 			fdrop(fp, td);
1163 			fp = NULL;
1164 		}
1165 		goto findkn;
1166 	}
1167 
1168 	/*
1169 	 * kn now contains the matching knote, or NULL if no match
1170 	 */
1171 	if (kn == NULL) {
1172 		if (kev->flags & EV_ADD) {
1173 			kn = tkn;
1174 			tkn = NULL;
1175 			if (kn == NULL) {
1176 				KQ_UNLOCK(kq);
1177 				error = ENOMEM;
1178 				goto done;
1179 			}
1180 			kn->kn_fp = fp;
1181 			kn->kn_kq = kq;
1182 			kn->kn_fop = fops;
1183 			/*
1184 			 * apply reference counts to knote structure, and
1185 			 * do not release it at the end of this routine.
1186 			 */
1187 			fops = NULL;
1188 			fp = NULL;
1189 
1190 			kn->kn_sfflags = kev->fflags;
1191 			kn->kn_sdata = kev->data;
1192 			kev->fflags = 0;
1193 			kev->data = 0;
1194 			kn->kn_kevent = *kev;
1195 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1196 			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1197 			kn->kn_status = KN_INFLUX|KN_DETACHED;
1198 
1199 			error = knote_attach(kn, kq);
1200 			KQ_UNLOCK(kq);
1201 			if (error != 0) {
1202 				tkn = kn;
1203 				goto done;
1204 			}
1205 
1206 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1207 				knote_drop(kn, td);
1208 				goto done;
1209 			}
1210 			KN_LIST_LOCK(kn);
1211 			goto done_ev_add;
1212 		} else {
1213 			/* No matching knote and the EV_ADD flag is not set. */
1214 			KQ_UNLOCK(kq);
1215 			error = ENOENT;
1216 			goto done;
1217 		}
1218 	}
1219 
1220 	if (kev->flags & EV_DELETE) {
1221 		kn->kn_status |= KN_INFLUX;
1222 		KQ_UNLOCK(kq);
1223 		if (!(kn->kn_status & KN_DETACHED))
1224 			kn->kn_fop->f_detach(kn);
1225 		knote_drop(kn, td);
1226 		goto done;
1227 	}
1228 
1229 	if (kev->flags & EV_FORCEONESHOT) {
1230 		kn->kn_flags |= EV_ONESHOT;
1231 		KNOTE_ACTIVATE(kn, 1);
1232 	}
1233 
1234 	/*
1235 	 * The user may change some filter values after the initial EV_ADD,
1236 	 * but doing so will not reset any filter which has already been
1237 	 * triggered.
1238 	 */
1239 	kn->kn_status |= KN_INFLUX | KN_SCAN;
1240 	KQ_UNLOCK(kq);
1241 	KN_LIST_LOCK(kn);
1242 	kn->kn_kevent.udata = kev->udata;
1243 	if (!fops->f_isfd && fops->f_touch != NULL) {
1244 		fops->f_touch(kn, kev, EVENT_REGISTER);
1245 	} else {
1246 		kn->kn_sfflags = kev->fflags;
1247 		kn->kn_sdata = kev->data;
1248 	}
1249 
1250 	/*
1251 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1252 	 * the initial attach event decides that the event is "completed"
1253 	 * already.  i.e. filt_procattach is called on a zombie process.  It
1254 	 * will call filt_proc which will remove it from the list, and NULL
1255 	 * kn_knlist.
1256 	 */
1257 done_ev_add:
1258 	if ((kev->flags & EV_DISABLE) &&
1259 	    ((kn->kn_status & KN_DISABLED) == 0)) {
1260 		kn->kn_status |= KN_DISABLED;
1261 	}
1262 
1263 	if ((kn->kn_status & KN_DISABLED) == 0)
1264 		event = kn->kn_fop->f_event(kn, 0);
1265 	else
1266 		event = 0;
1267 	KQ_LOCK(kq);
1268 	if (event)
1269 		KNOTE_ACTIVATE(kn, 1);
1270 	kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1271 	KN_LIST_UNLOCK(kn);
1272 
1273 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1274 		kn->kn_status &= ~KN_DISABLED;
1275 		if ((kn->kn_status & KN_ACTIVE) &&
1276 		    ((kn->kn_status & KN_QUEUED) == 0))
1277 			knote_enqueue(kn);
1278 	}
1279 	KQ_UNLOCK_FLUX(kq);
1280 
1281 done:
1282 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1283 	if (filedesc_unlock)
1284 		FILEDESC_XUNLOCK(td->td_proc->p_fd);
1285 	if (fp != NULL)
1286 		fdrop(fp, td);
1287 	if (tkn != NULL)
1288 		knote_free(tkn);
1289 	if (fops != NULL)
1290 		kqueue_fo_release(filt);
1291 	return (error);
1292 }
1293 
1294 static int
1295 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1296 {
1297 	int error;
1298 	struct kqueue *kq;
1299 
1300 	error = 0;
1301 
1302 	kq = fp->f_data;
1303 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1304 		return (EBADF);
1305 	*kqp = kq;
1306 	KQ_LOCK(kq);
1307 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1308 		KQ_UNLOCK(kq);
1309 		return (EBADF);
1310 	}
1311 	kq->kq_refcnt++;
1312 	KQ_UNLOCK(kq);
1313 
1314 	return error;
1315 }
1316 
1317 static void
1318 kqueue_release(struct kqueue *kq, int locked)
1319 {
1320 	if (locked)
1321 		KQ_OWNED(kq);
1322 	else
1323 		KQ_LOCK(kq);
1324 	kq->kq_refcnt--;
1325 	if (kq->kq_refcnt == 1)
1326 		wakeup(&kq->kq_refcnt);
1327 	if (!locked)
1328 		KQ_UNLOCK(kq);
1329 }
1330 
1331 static void
1332 kqueue_schedtask(struct kqueue *kq)
1333 {
1334 
1335 	KQ_OWNED(kq);
1336 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1337 	    ("scheduling kqueue task while draining"));
1338 
1339 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1340 		taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1341 		kq->kq_state |= KQ_TASKSCHED;
1342 	}
1343 }
1344 
1345 /*
1346  * Expand the kq to make sure we have storage for fops/ident pair.
1347  *
1348  * Return 0 on success (or no work necessary), return errno on failure.
1349  *
1350  * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1351  * If kqueue_register is called from a non-fd context, there usually/should
1352  * be no locks held.
1353  */
1354 static int
1355 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1356 	int waitok)
1357 {
1358 	struct klist *list, *tmp_knhash, *to_free;
1359 	u_long tmp_knhashmask;
1360 	int size;
1361 	int fd;
1362 	int mflag = waitok ? M_WAITOK : M_NOWAIT;
1363 
1364 	KQ_NOTOWNED(kq);
1365 
1366 	to_free = NULL;
1367 	if (fops->f_isfd) {
1368 		fd = ident;
1369 		if (kq->kq_knlistsize <= fd) {
1370 			size = kq->kq_knlistsize;
1371 			while (size <= fd)
1372 				size += KQEXTENT;
1373 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1374 			if (list == NULL)
1375 				return ENOMEM;
1376 			KQ_LOCK(kq);
1377 			if (kq->kq_knlistsize > fd) {
1378 				to_free = list;
1379 				list = NULL;
1380 			} else {
1381 				if (kq->kq_knlist != NULL) {
1382 					bcopy(kq->kq_knlist, list,
1383 					    kq->kq_knlistsize * sizeof(*list));
1384 					to_free = kq->kq_knlist;
1385 					kq->kq_knlist = NULL;
1386 				}
1387 				bzero((caddr_t)list +
1388 				    kq->kq_knlistsize * sizeof(*list),
1389 				    (size - kq->kq_knlistsize) * sizeof(*list));
1390 				kq->kq_knlistsize = size;
1391 				kq->kq_knlist = list;
1392 			}
1393 			KQ_UNLOCK(kq);
1394 		}
1395 	} else {
1396 		if (kq->kq_knhashmask == 0) {
1397 			tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1398 			    &tmp_knhashmask);
1399 			if (tmp_knhash == NULL)
1400 				return ENOMEM;
1401 			KQ_LOCK(kq);
1402 			if (kq->kq_knhashmask == 0) {
1403 				kq->kq_knhash = tmp_knhash;
1404 				kq->kq_knhashmask = tmp_knhashmask;
1405 			} else {
1406 				to_free = tmp_knhash;
1407 			}
1408 			KQ_UNLOCK(kq);
1409 		}
1410 	}
1411 	free(to_free, M_KQUEUE);
1412 
1413 	KQ_NOTOWNED(kq);
1414 	return 0;
1415 }
1416 
1417 static void
1418 kqueue_task(void *arg, int pending)
1419 {
1420 	struct kqueue *kq;
1421 	int haskqglobal;
1422 
1423 	haskqglobal = 0;
1424 	kq = arg;
1425 
1426 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1427 	KQ_LOCK(kq);
1428 
1429 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1430 
1431 	kq->kq_state &= ~KQ_TASKSCHED;
1432 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1433 		wakeup(&kq->kq_state);
1434 	}
1435 	KQ_UNLOCK(kq);
1436 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1437 }
1438 
1439 /*
1440  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1441  * We treat KN_MARKER knotes as if they are INFLUX.
1442  */
1443 static int
1444 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1445     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1446 {
1447 	struct kevent *kevp;
1448 	struct knote *kn, *marker;
1449 	sbintime_t asbt, rsbt;
1450 	int count, error, haskqglobal, influx, nkev, touch;
1451 
1452 	count = maxevents;
1453 	nkev = 0;
1454 	error = 0;
1455 	haskqglobal = 0;
1456 
1457 	if (maxevents == 0)
1458 		goto done_nl;
1459 
1460 	rsbt = 0;
1461 	if (tsp != NULL) {
1462 		if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 ||
1463 		    tsp->tv_nsec >= 1000000000) {
1464 			error = EINVAL;
1465 			goto done_nl;
1466 		}
1467 		if (timespecisset(tsp)) {
1468 			if (tsp->tv_sec <= INT32_MAX) {
1469 				rsbt = tstosbt(*tsp);
1470 				if (TIMESEL(&asbt, rsbt))
1471 					asbt += tc_tick_sbt;
1472 				if (asbt <= SBT_MAX - rsbt)
1473 					asbt += rsbt;
1474 				else
1475 					asbt = 0;
1476 				rsbt >>= tc_precexp;
1477 			} else
1478 				asbt = 0;
1479 		} else
1480 			asbt = -1;
1481 	} else
1482 		asbt = 0;
1483 	marker = knote_alloc(1);
1484 	if (marker == NULL) {
1485 		error = ENOMEM;
1486 		goto done_nl;
1487 	}
1488 	marker->kn_status = KN_MARKER;
1489 	KQ_LOCK(kq);
1490 
1491 retry:
1492 	kevp = keva;
1493 	if (kq->kq_count == 0) {
1494 		if (asbt == -1) {
1495 			error = EWOULDBLOCK;
1496 		} else {
1497 			kq->kq_state |= KQ_SLEEP;
1498 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1499 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1500 		}
1501 		if (error == 0)
1502 			goto retry;
1503 		/* don't restart after signals... */
1504 		if (error == ERESTART)
1505 			error = EINTR;
1506 		else if (error == EWOULDBLOCK)
1507 			error = 0;
1508 		goto done;
1509 	}
1510 
1511 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1512 	influx = 0;
1513 	while (count) {
1514 		KQ_OWNED(kq);
1515 		kn = TAILQ_FIRST(&kq->kq_head);
1516 
1517 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1518 		    (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1519 			if (influx) {
1520 				influx = 0;
1521 				KQ_FLUX_WAKEUP(kq);
1522 			}
1523 			kq->kq_state |= KQ_FLUXWAIT;
1524 			error = msleep(kq, &kq->kq_lock, PSOCK,
1525 			    "kqflxwt", 0);
1526 			continue;
1527 		}
1528 
1529 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1530 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1531 			kn->kn_status &= ~KN_QUEUED;
1532 			kq->kq_count--;
1533 			continue;
1534 		}
1535 		if (kn == marker) {
1536 			KQ_FLUX_WAKEUP(kq);
1537 			if (count == maxevents)
1538 				goto retry;
1539 			goto done;
1540 		}
1541 		KASSERT((kn->kn_status & KN_INFLUX) == 0,
1542 		    ("KN_INFLUX set when not suppose to be"));
1543 
1544 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1545 			kn->kn_status &= ~KN_QUEUED;
1546 			kn->kn_status |= KN_INFLUX;
1547 			kq->kq_count--;
1548 			KQ_UNLOCK(kq);
1549 			/*
1550 			 * We don't need to lock the list since we've marked
1551 			 * it _INFLUX.
1552 			 */
1553 			if (!(kn->kn_status & KN_DETACHED))
1554 				kn->kn_fop->f_detach(kn);
1555 			knote_drop(kn, td);
1556 			KQ_LOCK(kq);
1557 			continue;
1558 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1559 			kn->kn_status &= ~KN_QUEUED;
1560 			kn->kn_status |= KN_INFLUX;
1561 			kq->kq_count--;
1562 			KQ_UNLOCK(kq);
1563 			/*
1564 			 * We don't need to lock the list since we've marked
1565 			 * it _INFLUX.
1566 			 */
1567 			*kevp = kn->kn_kevent;
1568 			if (!(kn->kn_status & KN_DETACHED))
1569 				kn->kn_fop->f_detach(kn);
1570 			knote_drop(kn, td);
1571 			KQ_LOCK(kq);
1572 			kn = NULL;
1573 		} else {
1574 			kn->kn_status |= KN_INFLUX | KN_SCAN;
1575 			KQ_UNLOCK(kq);
1576 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1577 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1578 			KN_LIST_LOCK(kn);
1579 			if (kn->kn_fop->f_event(kn, 0) == 0) {
1580 				KQ_LOCK(kq);
1581 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1582 				kn->kn_status &=
1583 				    ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX |
1584 				    KN_SCAN);
1585 				kq->kq_count--;
1586 				KN_LIST_UNLOCK(kn);
1587 				influx = 1;
1588 				continue;
1589 			}
1590 			touch = (!kn->kn_fop->f_isfd &&
1591 			    kn->kn_fop->f_touch != NULL);
1592 			if (touch)
1593 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1594 			else
1595 				*kevp = kn->kn_kevent;
1596 			KQ_LOCK(kq);
1597 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1598 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1599 				/*
1600 				 * Manually clear knotes who weren't
1601 				 * 'touch'ed.
1602 				 */
1603 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1604 					kn->kn_data = 0;
1605 					kn->kn_fflags = 0;
1606 				}
1607 				if (kn->kn_flags & EV_DISPATCH)
1608 					kn->kn_status |= KN_DISABLED;
1609 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1610 				kq->kq_count--;
1611 			} else
1612 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1613 
1614 			kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1615 			KN_LIST_UNLOCK(kn);
1616 			influx = 1;
1617 		}
1618 
1619 		/* we are returning a copy to the user */
1620 		kevp++;
1621 		nkev++;
1622 		count--;
1623 
1624 		if (nkev == KQ_NEVENTS) {
1625 			influx = 0;
1626 			KQ_UNLOCK_FLUX(kq);
1627 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1628 			nkev = 0;
1629 			kevp = keva;
1630 			KQ_LOCK(kq);
1631 			if (error)
1632 				break;
1633 		}
1634 	}
1635 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1636 done:
1637 	KQ_OWNED(kq);
1638 	KQ_UNLOCK_FLUX(kq);
1639 	knote_free(marker);
1640 done_nl:
1641 	KQ_NOTOWNED(kq);
1642 	if (nkev != 0)
1643 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1644 	td->td_retval[0] = maxevents - count;
1645 	return (error);
1646 }
1647 
1648 /*ARGSUSED*/
1649 static int
1650 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1651 	struct ucred *active_cred, struct thread *td)
1652 {
1653 	/*
1654 	 * Enabling sigio causes two major problems:
1655 	 * 1) infinite recursion:
1656 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
1657 	 * set.  On receipt of a signal this will cause a kqueue to recurse
1658 	 * into itself over and over.  Sending the sigio causes the kqueue
1659 	 * to become ready, which in turn posts sigio again, forever.
1660 	 * Solution: this can be solved by setting a flag in the kqueue that
1661 	 * we have a SIGIO in progress.
1662 	 * 2) locking problems:
1663 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1664 	 * us above the proc and pgrp locks.
1665 	 * Solution: Post a signal using an async mechanism, being sure to
1666 	 * record a generation count in the delivery so that we do not deliver
1667 	 * a signal to the wrong process.
1668 	 *
1669 	 * Note, these two mechanisms are somewhat mutually exclusive!
1670 	 */
1671 #if 0
1672 	struct kqueue *kq;
1673 
1674 	kq = fp->f_data;
1675 	switch (cmd) {
1676 	case FIOASYNC:
1677 		if (*(int *)data) {
1678 			kq->kq_state |= KQ_ASYNC;
1679 		} else {
1680 			kq->kq_state &= ~KQ_ASYNC;
1681 		}
1682 		return (0);
1683 
1684 	case FIOSETOWN:
1685 		return (fsetown(*(int *)data, &kq->kq_sigio));
1686 
1687 	case FIOGETOWN:
1688 		*(int *)data = fgetown(&kq->kq_sigio);
1689 		return (0);
1690 	}
1691 #endif
1692 
1693 	return (ENOTTY);
1694 }
1695 
1696 /*ARGSUSED*/
1697 static int
1698 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1699 	struct thread *td)
1700 {
1701 	struct kqueue *kq;
1702 	int revents = 0;
1703 	int error;
1704 
1705 	if ((error = kqueue_acquire(fp, &kq)))
1706 		return POLLERR;
1707 
1708 	KQ_LOCK(kq);
1709 	if (events & (POLLIN | POLLRDNORM)) {
1710 		if (kq->kq_count) {
1711 			revents |= events & (POLLIN | POLLRDNORM);
1712 		} else {
1713 			selrecord(td, &kq->kq_sel);
1714 			if (SEL_WAITING(&kq->kq_sel))
1715 				kq->kq_state |= KQ_SEL;
1716 		}
1717 	}
1718 	kqueue_release(kq, 1);
1719 	KQ_UNLOCK(kq);
1720 	return (revents);
1721 }
1722 
1723 /*ARGSUSED*/
1724 static int
1725 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1726 	struct thread *td)
1727 {
1728 
1729 	bzero((void *)st, sizeof *st);
1730 	/*
1731 	 * We no longer return kq_count because the unlocked value is useless.
1732 	 * If you spent all this time getting the count, why not spend your
1733 	 * syscall better by calling kevent?
1734 	 *
1735 	 * XXX - This is needed for libc_r.
1736 	 */
1737 	st->st_mode = S_IFIFO;
1738 	return (0);
1739 }
1740 
1741 /*ARGSUSED*/
1742 static int
1743 kqueue_close(struct file *fp, struct thread *td)
1744 {
1745 	struct kqueue *kq = fp->f_data;
1746 	struct filedesc *fdp;
1747 	struct knote *kn;
1748 	int i;
1749 	int error;
1750 	int filedesc_unlock;
1751 
1752 	if ((error = kqueue_acquire(fp, &kq)))
1753 		return error;
1754 
1755 	filedesc_unlock = 0;
1756 	KQ_LOCK(kq);
1757 
1758 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1759 	    ("kqueue already closing"));
1760 	kq->kq_state |= KQ_CLOSING;
1761 	if (kq->kq_refcnt > 1)
1762 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1763 
1764 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1765 	fdp = kq->kq_fdp;
1766 
1767 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
1768 	    ("kqueue's knlist not empty"));
1769 
1770 	for (i = 0; i < kq->kq_knlistsize; i++) {
1771 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1772 			if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1773 				kq->kq_state |= KQ_FLUXWAIT;
1774 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1775 				continue;
1776 			}
1777 			kn->kn_status |= KN_INFLUX;
1778 			KQ_UNLOCK(kq);
1779 			if (!(kn->kn_status & KN_DETACHED))
1780 				kn->kn_fop->f_detach(kn);
1781 			knote_drop(kn, td);
1782 			KQ_LOCK(kq);
1783 		}
1784 	}
1785 	if (kq->kq_knhashmask != 0) {
1786 		for (i = 0; i <= kq->kq_knhashmask; i++) {
1787 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1788 				if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1789 					kq->kq_state |= KQ_FLUXWAIT;
1790 					msleep(kq, &kq->kq_lock, PSOCK,
1791 					       "kqclo2", 0);
1792 					continue;
1793 				}
1794 				kn->kn_status |= KN_INFLUX;
1795 				KQ_UNLOCK(kq);
1796 				if (!(kn->kn_status & KN_DETACHED))
1797 					kn->kn_fop->f_detach(kn);
1798 				knote_drop(kn, td);
1799 				KQ_LOCK(kq);
1800 			}
1801 		}
1802 	}
1803 
1804 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1805 		kq->kq_state |= KQ_TASKDRAIN;
1806 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1807 	}
1808 
1809 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1810 		selwakeuppri(&kq->kq_sel, PSOCK);
1811 		if (!SEL_WAITING(&kq->kq_sel))
1812 			kq->kq_state &= ~KQ_SEL;
1813 	}
1814 
1815 	KQ_UNLOCK(kq);
1816 
1817 	/*
1818 	 * We could be called due to the knote_drop() doing fdrop(),
1819 	 * called from kqueue_register().  In this case the global
1820 	 * lock is owned, and filedesc sx is locked before, to not
1821 	 * take the sleepable lock after non-sleepable.
1822 	 */
1823 	if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
1824 		FILEDESC_XLOCK(fdp);
1825 		filedesc_unlock = 1;
1826 	} else
1827 		filedesc_unlock = 0;
1828 	TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
1829 	if (filedesc_unlock)
1830 		FILEDESC_XUNLOCK(fdp);
1831 
1832 	seldrain(&kq->kq_sel);
1833 	knlist_destroy(&kq->kq_sel.si_note);
1834 	mtx_destroy(&kq->kq_lock);
1835 	kq->kq_fdp = NULL;
1836 
1837 	if (kq->kq_knhash != NULL)
1838 		free(kq->kq_knhash, M_KQUEUE);
1839 	if (kq->kq_knlist != NULL)
1840 		free(kq->kq_knlist, M_KQUEUE);
1841 
1842 	funsetown(&kq->kq_sigio);
1843 	chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
1844 	crfree(kq->kq_cred);
1845 	free(kq, M_KQUEUE);
1846 	fp->f_data = NULL;
1847 
1848 	return (0);
1849 }
1850 
1851 static int
1852 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
1853 {
1854 
1855 	kif->kf_type = KF_TYPE_KQUEUE;
1856 	return (0);
1857 }
1858 
1859 static void
1860 kqueue_wakeup(struct kqueue *kq)
1861 {
1862 	KQ_OWNED(kq);
1863 
1864 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1865 		kq->kq_state &= ~KQ_SLEEP;
1866 		wakeup(kq);
1867 	}
1868 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1869 		selwakeuppri(&kq->kq_sel, PSOCK);
1870 		if (!SEL_WAITING(&kq->kq_sel))
1871 			kq->kq_state &= ~KQ_SEL;
1872 	}
1873 	if (!knlist_empty(&kq->kq_sel.si_note))
1874 		kqueue_schedtask(kq);
1875 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1876 		pgsigio(&kq->kq_sigio, SIGIO, 0);
1877 	}
1878 }
1879 
1880 /*
1881  * Walk down a list of knotes, activating them if their event has triggered.
1882  *
1883  * There is a possibility to optimize in the case of one kq watching another.
1884  * Instead of scheduling a task to wake it up, you could pass enough state
1885  * down the chain to make up the parent kqueue.  Make this code functional
1886  * first.
1887  */
1888 void
1889 knote(struct knlist *list, long hint, int lockflags)
1890 {
1891 	struct kqueue *kq;
1892 	struct knote *kn;
1893 	int error;
1894 
1895 	if (list == NULL)
1896 		return;
1897 
1898 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1899 
1900 	if ((lockflags & KNF_LISTLOCKED) == 0)
1901 		list->kl_lock(list->kl_lockarg);
1902 
1903 	/*
1904 	 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1905 	 * the kqueue scheduling, but this will introduce four
1906 	 * lock/unlock's for each knote to test.  If we do, continue to use
1907 	 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1908 	 * only safe if you want to remove the current item, which we are
1909 	 * not doing.
1910 	 */
1911 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1912 		kq = kn->kn_kq;
1913 		KQ_LOCK(kq);
1914 		if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
1915 			/*
1916 			 * Do not process the influx notes, except for
1917 			 * the influx coming from the kq unlock in the
1918 			 * kqueue_scan().  In the later case, we do
1919 			 * not interfere with the scan, since the code
1920 			 * fragment in kqueue_scan() locks the knlist,
1921 			 * and cannot proceed until we finished.
1922 			 */
1923 			KQ_UNLOCK(kq);
1924 		} else if ((lockflags & KNF_NOKQLOCK) != 0) {
1925 			kn->kn_status |= KN_INFLUX;
1926 			KQ_UNLOCK(kq);
1927 			error = kn->kn_fop->f_event(kn, hint);
1928 			KQ_LOCK(kq);
1929 			kn->kn_status &= ~KN_INFLUX;
1930 			if (error)
1931 				KNOTE_ACTIVATE(kn, 1);
1932 			KQ_UNLOCK_FLUX(kq);
1933 		} else {
1934 			kn->kn_status |= KN_HASKQLOCK;
1935 			if (kn->kn_fop->f_event(kn, hint))
1936 				KNOTE_ACTIVATE(kn, 1);
1937 			kn->kn_status &= ~KN_HASKQLOCK;
1938 			KQ_UNLOCK(kq);
1939 		}
1940 	}
1941 	if ((lockflags & KNF_LISTLOCKED) == 0)
1942 		list->kl_unlock(list->kl_lockarg);
1943 }
1944 
1945 /*
1946  * add a knote to a knlist
1947  */
1948 void
1949 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1950 {
1951 	KNL_ASSERT_LOCK(knl, islocked);
1952 	KQ_NOTOWNED(kn->kn_kq);
1953 	KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1954 	    (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1955 	if (!islocked)
1956 		knl->kl_lock(knl->kl_lockarg);
1957 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1958 	if (!islocked)
1959 		knl->kl_unlock(knl->kl_lockarg);
1960 	KQ_LOCK(kn->kn_kq);
1961 	kn->kn_knlist = knl;
1962 	kn->kn_status &= ~KN_DETACHED;
1963 	KQ_UNLOCK(kn->kn_kq);
1964 }
1965 
1966 static void
1967 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1968 {
1969 	KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1970 	KNL_ASSERT_LOCK(knl, knlislocked);
1971 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1972 	if (!kqislocked)
1973 		KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1974     ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1975 	if (!knlislocked)
1976 		knl->kl_lock(knl->kl_lockarg);
1977 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1978 	kn->kn_knlist = NULL;
1979 	if (!knlislocked)
1980 		knl->kl_unlock(knl->kl_lockarg);
1981 	if (!kqislocked)
1982 		KQ_LOCK(kn->kn_kq);
1983 	kn->kn_status |= KN_DETACHED;
1984 	if (!kqislocked)
1985 		KQ_UNLOCK(kn->kn_kq);
1986 }
1987 
1988 /*
1989  * remove knote from the specified knlist
1990  */
1991 void
1992 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1993 {
1994 
1995 	knlist_remove_kq(knl, kn, islocked, 0);
1996 }
1997 
1998 /*
1999  * remove knote from the specified knlist while in f_event handler.
2000  */
2001 void
2002 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
2003 {
2004 
2005 	knlist_remove_kq(knl, kn, 1,
2006 	    (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
2007 }
2008 
2009 int
2010 knlist_empty(struct knlist *knl)
2011 {
2012 
2013 	KNL_ASSERT_LOCKED(knl);
2014 	return SLIST_EMPTY(&knl->kl_list);
2015 }
2016 
2017 static struct mtx	knlist_lock;
2018 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2019 	MTX_DEF);
2020 static void knlist_mtx_lock(void *arg);
2021 static void knlist_mtx_unlock(void *arg);
2022 
2023 static void
2024 knlist_mtx_lock(void *arg)
2025 {
2026 
2027 	mtx_lock((struct mtx *)arg);
2028 }
2029 
2030 static void
2031 knlist_mtx_unlock(void *arg)
2032 {
2033 
2034 	mtx_unlock((struct mtx *)arg);
2035 }
2036 
2037 static void
2038 knlist_mtx_assert_locked(void *arg)
2039 {
2040 
2041 	mtx_assert((struct mtx *)arg, MA_OWNED);
2042 }
2043 
2044 static void
2045 knlist_mtx_assert_unlocked(void *arg)
2046 {
2047 
2048 	mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2049 }
2050 
2051 static void
2052 knlist_rw_rlock(void *arg)
2053 {
2054 
2055 	rw_rlock((struct rwlock *)arg);
2056 }
2057 
2058 static void
2059 knlist_rw_runlock(void *arg)
2060 {
2061 
2062 	rw_runlock((struct rwlock *)arg);
2063 }
2064 
2065 static void
2066 knlist_rw_assert_locked(void *arg)
2067 {
2068 
2069 	rw_assert((struct rwlock *)arg, RA_LOCKED);
2070 }
2071 
2072 static void
2073 knlist_rw_assert_unlocked(void *arg)
2074 {
2075 
2076 	rw_assert((struct rwlock *)arg, RA_UNLOCKED);
2077 }
2078 
2079 void
2080 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2081     void (*kl_unlock)(void *),
2082     void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
2083 {
2084 
2085 	if (lock == NULL)
2086 		knl->kl_lockarg = &knlist_lock;
2087 	else
2088 		knl->kl_lockarg = lock;
2089 
2090 	if (kl_lock == NULL)
2091 		knl->kl_lock = knlist_mtx_lock;
2092 	else
2093 		knl->kl_lock = kl_lock;
2094 	if (kl_unlock == NULL)
2095 		knl->kl_unlock = knlist_mtx_unlock;
2096 	else
2097 		knl->kl_unlock = kl_unlock;
2098 	if (kl_assert_locked == NULL)
2099 		knl->kl_assert_locked = knlist_mtx_assert_locked;
2100 	else
2101 		knl->kl_assert_locked = kl_assert_locked;
2102 	if (kl_assert_unlocked == NULL)
2103 		knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
2104 	else
2105 		knl->kl_assert_unlocked = kl_assert_unlocked;
2106 
2107 	SLIST_INIT(&knl->kl_list);
2108 }
2109 
2110 void
2111 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2112 {
2113 
2114 	knlist_init(knl, lock, NULL, NULL, NULL, NULL);
2115 }
2116 
2117 void
2118 knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock)
2119 {
2120 
2121 	knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock,
2122 	    knlist_rw_assert_locked, knlist_rw_assert_unlocked);
2123 }
2124 
2125 void
2126 knlist_destroy(struct knlist *knl)
2127 {
2128 
2129 #ifdef INVARIANTS
2130 	/*
2131 	 * if we run across this error, we need to find the offending
2132 	 * driver and have it call knlist_clear or knlist_delete.
2133 	 */
2134 	if (!SLIST_EMPTY(&knl->kl_list))
2135 		printf("WARNING: destroying knlist w/ knotes on it!\n");
2136 #endif
2137 
2138 	knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
2139 	SLIST_INIT(&knl->kl_list);
2140 }
2141 
2142 /*
2143  * Even if we are locked, we may need to drop the lock to allow any influx
2144  * knotes time to "settle".
2145  */
2146 void
2147 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2148 {
2149 	struct knote *kn, *kn2;
2150 	struct kqueue *kq;
2151 
2152 	if (islocked)
2153 		KNL_ASSERT_LOCKED(knl);
2154 	else {
2155 		KNL_ASSERT_UNLOCKED(knl);
2156 again:		/* need to reacquire lock since we have dropped it */
2157 		knl->kl_lock(knl->kl_lockarg);
2158 	}
2159 
2160 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2161 		kq = kn->kn_kq;
2162 		KQ_LOCK(kq);
2163 		if ((kn->kn_status & KN_INFLUX)) {
2164 			KQ_UNLOCK(kq);
2165 			continue;
2166 		}
2167 		knlist_remove_kq(knl, kn, 1, 1);
2168 		if (killkn) {
2169 			kn->kn_status |= KN_INFLUX | KN_DETACHED;
2170 			KQ_UNLOCK(kq);
2171 			knote_drop(kn, td);
2172 		} else {
2173 			/* Make sure cleared knotes disappear soon */
2174 			kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2175 			KQ_UNLOCK(kq);
2176 		}
2177 		kq = NULL;
2178 	}
2179 
2180 	if (!SLIST_EMPTY(&knl->kl_list)) {
2181 		/* there are still KN_INFLUX remaining */
2182 		kn = SLIST_FIRST(&knl->kl_list);
2183 		kq = kn->kn_kq;
2184 		KQ_LOCK(kq);
2185 		KASSERT(kn->kn_status & KN_INFLUX,
2186 		    ("knote removed w/o list lock"));
2187 		knl->kl_unlock(knl->kl_lockarg);
2188 		kq->kq_state |= KQ_FLUXWAIT;
2189 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2190 		kq = NULL;
2191 		goto again;
2192 	}
2193 
2194 	if (islocked)
2195 		KNL_ASSERT_LOCKED(knl);
2196 	else {
2197 		knl->kl_unlock(knl->kl_lockarg);
2198 		KNL_ASSERT_UNLOCKED(knl);
2199 	}
2200 }
2201 
2202 /*
2203  * Remove all knotes referencing a specified fd must be called with FILEDESC
2204  * lock.  This prevents a race where a new fd comes along and occupies the
2205  * entry and we attach a knote to the fd.
2206  */
2207 void
2208 knote_fdclose(struct thread *td, int fd)
2209 {
2210 	struct filedesc *fdp = td->td_proc->p_fd;
2211 	struct kqueue *kq;
2212 	struct knote *kn;
2213 	int influx;
2214 
2215 	FILEDESC_XLOCK_ASSERT(fdp);
2216 
2217 	/*
2218 	 * We shouldn't have to worry about new kevents appearing on fd
2219 	 * since filedesc is locked.
2220 	 */
2221 	TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2222 		KQ_LOCK(kq);
2223 
2224 again:
2225 		influx = 0;
2226 		while (kq->kq_knlistsize > fd &&
2227 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2228 			if (kn->kn_status & KN_INFLUX) {
2229 				/* someone else might be waiting on our knote */
2230 				if (influx)
2231 					wakeup(kq);
2232 				kq->kq_state |= KQ_FLUXWAIT;
2233 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2234 				goto again;
2235 			}
2236 			kn->kn_status |= KN_INFLUX;
2237 			KQ_UNLOCK(kq);
2238 			if (!(kn->kn_status & KN_DETACHED))
2239 				kn->kn_fop->f_detach(kn);
2240 			knote_drop(kn, td);
2241 			influx = 1;
2242 			KQ_LOCK(kq);
2243 		}
2244 		KQ_UNLOCK_FLUX(kq);
2245 	}
2246 }
2247 
2248 static int
2249 knote_attach(struct knote *kn, struct kqueue *kq)
2250 {
2251 	struct klist *list;
2252 
2253 	KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2254 	KQ_OWNED(kq);
2255 
2256 	if (kn->kn_fop->f_isfd) {
2257 		if (kn->kn_id >= kq->kq_knlistsize)
2258 			return ENOMEM;
2259 		list = &kq->kq_knlist[kn->kn_id];
2260 	} else {
2261 		if (kq->kq_knhash == NULL)
2262 			return ENOMEM;
2263 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2264 	}
2265 
2266 	SLIST_INSERT_HEAD(list, kn, kn_link);
2267 
2268 	return 0;
2269 }
2270 
2271 /*
2272  * knote must already have been detached using the f_detach method.
2273  * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2274  * to prevent other removal.
2275  */
2276 static void
2277 knote_drop(struct knote *kn, struct thread *td)
2278 {
2279 	struct kqueue *kq;
2280 	struct klist *list;
2281 
2282 	kq = kn->kn_kq;
2283 
2284 	KQ_NOTOWNED(kq);
2285 	KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2286 	    ("knote_drop called without KN_INFLUX set in kn_status"));
2287 
2288 	KQ_LOCK(kq);
2289 	if (kn->kn_fop->f_isfd)
2290 		list = &kq->kq_knlist[kn->kn_id];
2291 	else
2292 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2293 
2294 	if (!SLIST_EMPTY(list))
2295 		SLIST_REMOVE(list, kn, knote, kn_link);
2296 	if (kn->kn_status & KN_QUEUED)
2297 		knote_dequeue(kn);
2298 	KQ_UNLOCK_FLUX(kq);
2299 
2300 	if (kn->kn_fop->f_isfd) {
2301 		fdrop(kn->kn_fp, td);
2302 		kn->kn_fp = NULL;
2303 	}
2304 	kqueue_fo_release(kn->kn_kevent.filter);
2305 	kn->kn_fop = NULL;
2306 	knote_free(kn);
2307 }
2308 
2309 static void
2310 knote_enqueue(struct knote *kn)
2311 {
2312 	struct kqueue *kq = kn->kn_kq;
2313 
2314 	KQ_OWNED(kn->kn_kq);
2315 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2316 
2317 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2318 	kn->kn_status |= KN_QUEUED;
2319 	kq->kq_count++;
2320 	kqueue_wakeup(kq);
2321 }
2322 
2323 static void
2324 knote_dequeue(struct knote *kn)
2325 {
2326 	struct kqueue *kq = kn->kn_kq;
2327 
2328 	KQ_OWNED(kn->kn_kq);
2329 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2330 
2331 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2332 	kn->kn_status &= ~KN_QUEUED;
2333 	kq->kq_count--;
2334 }
2335 
2336 static void
2337 knote_init(void)
2338 {
2339 
2340 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2341 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2342 }
2343 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2344 
2345 static struct knote *
2346 knote_alloc(int waitok)
2347 {
2348 	return ((struct knote *)uma_zalloc(knote_zone,
2349 	    (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2350 }
2351 
2352 static void
2353 knote_free(struct knote *kn)
2354 {
2355 	if (kn != NULL)
2356 		uma_zfree(knote_zone, kn);
2357 }
2358 
2359 /*
2360  * Register the kev w/ the kq specified by fd.
2361  */
2362 int
2363 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2364 {
2365 	struct kqueue *kq;
2366 	struct file *fp;
2367 	cap_rights_t rights;
2368 	int error;
2369 
2370 	error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp);
2371 	if (error != 0)
2372 		return (error);
2373 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2374 		goto noacquire;
2375 
2376 	error = kqueue_register(kq, kev, td, waitok);
2377 
2378 	kqueue_release(kq, 0);
2379 
2380 noacquire:
2381 	fdrop(fp, td);
2382 
2383 	return error;
2384 }
2385