xref: /freebsd/sys/kern/kern_event.c (revision d1d015864103b253b3fcb2f72a0da5b0cfeb31b6)
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 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/capability.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/rwlock.h>
41 #include <sys/proc.h>
42 #include <sys/malloc.h>
43 #include <sys/unistd.h>
44 #include <sys/file.h>
45 #include <sys/filedesc.h>
46 #include <sys/filio.h>
47 #include <sys/fcntl.h>
48 #include <sys/kthread.h>
49 #include <sys/selinfo.h>
50 #include <sys/stdatomic.h>
51 #include <sys/queue.h>
52 #include <sys/event.h>
53 #include <sys/eventvar.h>
54 #include <sys/poll.h>
55 #include <sys/protosw.h>
56 #include <sys/sigio.h>
57 #include <sys/signalvar.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysproto.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/taskqueue.h>
65 #include <sys/uio.h>
66 #ifdef KTRACE
67 #include <sys/ktrace.h>
68 #endif
69 
70 #include <vm/uma.h>
71 
72 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
73 
74 /*
75  * This lock is used if multiple kq locks are required.  This possibly
76  * should be made into a per proc lock.
77  */
78 static struct mtx	kq_global;
79 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
80 #define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
81 	if (!haslck)				\
82 		mtx_lock(lck);			\
83 	haslck = 1;				\
84 } while (0)
85 #define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
86 	if (haslck)				\
87 		mtx_unlock(lck);			\
88 	haslck = 0;				\
89 } while (0)
90 
91 TASKQUEUE_DEFINE_THREAD(kqueue);
92 
93 static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
94 static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
95 static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
96 		    struct thread *td, int waitok);
97 static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
98 static void	kqueue_release(struct kqueue *kq, int locked);
99 static int	kqueue_expand(struct kqueue *kq, struct filterops *fops,
100 		    uintptr_t ident, int waitok);
101 static void	kqueue_task(void *arg, int pending);
102 static int	kqueue_scan(struct kqueue *kq, int maxevents,
103 		    struct kevent_copyops *k_ops,
104 		    const struct timespec *timeout,
105 		    struct kevent *keva, struct thread *td);
106 static void 	kqueue_wakeup(struct kqueue *kq);
107 static struct filterops *kqueue_fo_find(int filt);
108 static void	kqueue_fo_release(int filt);
109 
110 static fo_rdwr_t	kqueue_read;
111 static fo_rdwr_t	kqueue_write;
112 static fo_truncate_t	kqueue_truncate;
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 
119 static struct fileops kqueueops = {
120 	.fo_read = kqueue_read,
121 	.fo_write = kqueue_write,
122 	.fo_truncate = kqueue_truncate,
123 	.fo_ioctl = kqueue_ioctl,
124 	.fo_poll = kqueue_poll,
125 	.fo_kqfilter = kqueue_kqfilter,
126 	.fo_stat = kqueue_stat,
127 	.fo_close = kqueue_close,
128 	.fo_chmod = invfo_chmod,
129 	.fo_chown = invfo_chown,
130 	.fo_sendfile = invfo_sendfile,
131 };
132 
133 static int 	knote_attach(struct knote *kn, struct kqueue *kq);
134 static void 	knote_drop(struct knote *kn, struct thread *td);
135 static void 	knote_enqueue(struct knote *kn);
136 static void 	knote_dequeue(struct knote *kn);
137 static void 	knote_init(void);
138 static struct 	knote *knote_alloc(int waitok);
139 static void 	knote_free(struct knote *kn);
140 
141 static void	filt_kqdetach(struct knote *kn);
142 static int	filt_kqueue(struct knote *kn, long hint);
143 static int	filt_procattach(struct knote *kn);
144 static void	filt_procdetach(struct knote *kn);
145 static int	filt_proc(struct knote *kn, long hint);
146 static int	filt_fileattach(struct knote *kn);
147 static void	filt_timerexpire(void *knx);
148 static int	filt_timerattach(struct knote *kn);
149 static void	filt_timerdetach(struct knote *kn);
150 static int	filt_timer(struct knote *kn, long hint);
151 static int	filt_userattach(struct knote *kn);
152 static void	filt_userdetach(struct knote *kn);
153 static int	filt_user(struct knote *kn, long hint);
154 static void	filt_usertouch(struct knote *kn, struct kevent *kev,
155 		    u_long type);
156 
157 static struct filterops file_filtops = {
158 	.f_isfd = 1,
159 	.f_attach = filt_fileattach,
160 };
161 static struct filterops kqread_filtops = {
162 	.f_isfd = 1,
163 	.f_detach = filt_kqdetach,
164 	.f_event = filt_kqueue,
165 };
166 /* XXX - move to kern_proc.c?  */
167 static struct filterops proc_filtops = {
168 	.f_isfd = 0,
169 	.f_attach = filt_procattach,
170 	.f_detach = filt_procdetach,
171 	.f_event = filt_proc,
172 };
173 static struct filterops timer_filtops = {
174 	.f_isfd = 0,
175 	.f_attach = filt_timerattach,
176 	.f_detach = filt_timerdetach,
177 	.f_event = filt_timer,
178 };
179 static struct filterops user_filtops = {
180 	.f_attach = filt_userattach,
181 	.f_detach = filt_userdetach,
182 	.f_event = filt_user,
183 	.f_touch = filt_usertouch,
184 };
185 
186 static uma_zone_t	knote_zone;
187 static atomic_uint	kq_ncallouts = ATOMIC_VAR_INIT(0);
188 static unsigned int 	kq_calloutmax = 4 * 1024;
189 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
190     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
191 
192 /* XXX - ensure not KN_INFLUX?? */
193 #define KNOTE_ACTIVATE(kn, islock) do { 				\
194 	if ((islock))							\
195 		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
196 	else								\
197 		KQ_LOCK((kn)->kn_kq);					\
198 	(kn)->kn_status |= KN_ACTIVE;					\
199 	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
200 		knote_enqueue((kn));					\
201 	if (!(islock))							\
202 		KQ_UNLOCK((kn)->kn_kq);					\
203 } while(0)
204 #define KQ_LOCK(kq) do {						\
205 	mtx_lock(&(kq)->kq_lock);					\
206 } while (0)
207 #define KQ_FLUX_WAKEUP(kq) do {						\
208 	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
209 		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
210 		wakeup((kq));						\
211 	}								\
212 } while (0)
213 #define KQ_UNLOCK_FLUX(kq) do {						\
214 	KQ_FLUX_WAKEUP(kq);						\
215 	mtx_unlock(&(kq)->kq_lock);					\
216 } while (0)
217 #define KQ_UNLOCK(kq) do {						\
218 	mtx_unlock(&(kq)->kq_lock);					\
219 } while (0)
220 #define KQ_OWNED(kq) do {						\
221 	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
222 } while (0)
223 #define KQ_NOTOWNED(kq) do {						\
224 	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
225 } while (0)
226 #define KN_LIST_LOCK(kn) do {						\
227 	if (kn->kn_knlist != NULL)					\
228 		kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg);	\
229 } while (0)
230 #define KN_LIST_UNLOCK(kn) do {						\
231 	if (kn->kn_knlist != NULL) 					\
232 		kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg);	\
233 } while (0)
234 #define	KNL_ASSERT_LOCK(knl, islocked) do {				\
235 	if (islocked)							\
236 		KNL_ASSERT_LOCKED(knl);				\
237 	else								\
238 		KNL_ASSERT_UNLOCKED(knl);				\
239 } while (0)
240 #ifdef INVARIANTS
241 #define	KNL_ASSERT_LOCKED(knl) do {					\
242 	knl->kl_assert_locked((knl)->kl_lockarg);			\
243 } while (0)
244 #define	KNL_ASSERT_UNLOCKED(knl) do {					\
245 	knl->kl_assert_unlocked((knl)->kl_lockarg);			\
246 } while (0)
247 #else /* !INVARIANTS */
248 #define	KNL_ASSERT_LOCKED(knl) do {} while(0)
249 #define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
250 #endif /* INVARIANTS */
251 
252 #define	KN_HASHSIZE		64		/* XXX should be tunable */
253 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
254 
255 static int
256 filt_nullattach(struct knote *kn)
257 {
258 
259 	return (ENXIO);
260 };
261 
262 struct filterops null_filtops = {
263 	.f_isfd = 0,
264 	.f_attach = filt_nullattach,
265 };
266 
267 /* XXX - make SYSINIT to add these, and move into respective modules. */
268 extern struct filterops sig_filtops;
269 extern struct filterops fs_filtops;
270 
271 /*
272  * Table for for all system-defined filters.
273  */
274 static struct mtx	filterops_lock;
275 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
276 	MTX_DEF);
277 static struct {
278 	struct filterops *for_fop;
279 	int for_refcnt;
280 } sysfilt_ops[EVFILT_SYSCOUNT] = {
281 	{ &file_filtops },			/* EVFILT_READ */
282 	{ &file_filtops },			/* EVFILT_WRITE */
283 	{ &null_filtops },			/* EVFILT_AIO */
284 	{ &file_filtops },			/* EVFILT_VNODE */
285 	{ &proc_filtops },			/* EVFILT_PROC */
286 	{ &sig_filtops },			/* EVFILT_SIGNAL */
287 	{ &timer_filtops },			/* EVFILT_TIMER */
288 	{ &null_filtops },			/* former EVFILT_NETDEV */
289 	{ &fs_filtops },			/* EVFILT_FS */
290 	{ &null_filtops },			/* EVFILT_LIO */
291 	{ &user_filtops },			/* EVFILT_USER */
292 };
293 
294 /*
295  * Simple redirection for all cdevsw style objects to call their fo_kqfilter
296  * method.
297  */
298 static int
299 filt_fileattach(struct knote *kn)
300 {
301 
302 	return (fo_kqfilter(kn->kn_fp, kn));
303 }
304 
305 /*ARGSUSED*/
306 static int
307 kqueue_kqfilter(struct file *fp, struct knote *kn)
308 {
309 	struct kqueue *kq = kn->kn_fp->f_data;
310 
311 	if (kn->kn_filter != EVFILT_READ)
312 		return (EINVAL);
313 
314 	kn->kn_status |= KN_KQUEUE;
315 	kn->kn_fop = &kqread_filtops;
316 	knlist_add(&kq->kq_sel.si_note, kn, 0);
317 
318 	return (0);
319 }
320 
321 static void
322 filt_kqdetach(struct knote *kn)
323 {
324 	struct kqueue *kq = kn->kn_fp->f_data;
325 
326 	knlist_remove(&kq->kq_sel.si_note, kn, 0);
327 }
328 
329 /*ARGSUSED*/
330 static int
331 filt_kqueue(struct knote *kn, long hint)
332 {
333 	struct kqueue *kq = kn->kn_fp->f_data;
334 
335 	kn->kn_data = kq->kq_count;
336 	return (kn->kn_data > 0);
337 }
338 
339 /* XXX - move to kern_proc.c?  */
340 static int
341 filt_procattach(struct knote *kn)
342 {
343 	struct proc *p;
344 	int immediate;
345 	int error;
346 
347 	immediate = 0;
348 	p = pfind(kn->kn_id);
349 	if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
350 		p = zpfind(kn->kn_id);
351 		immediate = 1;
352 	} else if (p != NULL && (p->p_flag & P_WEXIT)) {
353 		immediate = 1;
354 	}
355 
356 	if (p == NULL)
357 		return (ESRCH);
358 	if ((error = p_cansee(curthread, p))) {
359 		PROC_UNLOCK(p);
360 		return (error);
361 	}
362 
363 	kn->kn_ptr.p_proc = p;
364 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
365 
366 	/*
367 	 * internal flag indicating registration done by kernel
368 	 */
369 	if (kn->kn_flags & EV_FLAG1) {
370 		kn->kn_data = kn->kn_sdata;		/* ppid */
371 		kn->kn_fflags = NOTE_CHILD;
372 		kn->kn_flags &= ~EV_FLAG1;
373 	}
374 
375 	if (immediate == 0)
376 		knlist_add(&p->p_klist, kn, 1);
377 
378 	/*
379 	 * Immediately activate any exit notes if the target process is a
380 	 * zombie.  This is necessary to handle the case where the target
381 	 * process, e.g. a child, dies before the kevent is registered.
382 	 */
383 	if (immediate && filt_proc(kn, NOTE_EXIT))
384 		KNOTE_ACTIVATE(kn, 0);
385 
386 	PROC_UNLOCK(p);
387 
388 	return (0);
389 }
390 
391 /*
392  * The knote may be attached to a different process, which may exit,
393  * leaving nothing for the knote to be attached to.  So when the process
394  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
395  * it will be deleted when read out.  However, as part of the knote deletion,
396  * this routine is called, so a check is needed to avoid actually performing
397  * a detach, because the original process does not exist any more.
398  */
399 /* XXX - move to kern_proc.c?  */
400 static void
401 filt_procdetach(struct knote *kn)
402 {
403 	struct proc *p;
404 
405 	p = kn->kn_ptr.p_proc;
406 	knlist_remove(&p->p_klist, kn, 0);
407 	kn->kn_ptr.p_proc = NULL;
408 }
409 
410 /* XXX - move to kern_proc.c?  */
411 static int
412 filt_proc(struct knote *kn, long hint)
413 {
414 	struct proc *p = kn->kn_ptr.p_proc;
415 	u_int event;
416 
417 	/*
418 	 * mask off extra data
419 	 */
420 	event = (u_int)hint & NOTE_PCTRLMASK;
421 
422 	/*
423 	 * if the user is interested in this event, record it.
424 	 */
425 	if (kn->kn_sfflags & event)
426 		kn->kn_fflags |= event;
427 
428 	/*
429 	 * process is gone, so flag the event as finished.
430 	 */
431 	if (event == NOTE_EXIT) {
432 		if (!(kn->kn_status & KN_DETACHED))
433 			knlist_remove_inevent(&p->p_klist, kn);
434 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
435 		kn->kn_ptr.p_proc = NULL;
436 		if (kn->kn_fflags & NOTE_EXIT)
437 			kn->kn_data = p->p_xstat;
438 		if (kn->kn_fflags == 0)
439 			kn->kn_flags |= EV_DROP;
440 		return (1);
441 	}
442 
443 	return (kn->kn_fflags != 0);
444 }
445 
446 /*
447  * Called when the process forked. It mostly does the same as the
448  * knote(), activating all knotes registered to be activated when the
449  * process forked. Additionally, for each knote attached to the
450  * parent, check whether user wants to track the new process. If so
451  * attach a new knote to it, and immediately report an event with the
452  * child's pid.
453  */
454 void
455 knote_fork(struct knlist *list, int pid)
456 {
457 	struct kqueue *kq;
458 	struct knote *kn;
459 	struct kevent kev;
460 	int error;
461 
462 	if (list == NULL)
463 		return;
464 	list->kl_lock(list->kl_lockarg);
465 
466 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
467 		if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
468 			continue;
469 		kq = kn->kn_kq;
470 		KQ_LOCK(kq);
471 		if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
472 			KQ_UNLOCK(kq);
473 			continue;
474 		}
475 
476 		/*
477 		 * The same as knote(), activate the event.
478 		 */
479 		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
480 			kn->kn_status |= KN_HASKQLOCK;
481 			if (kn->kn_fop->f_event(kn, NOTE_FORK))
482 				KNOTE_ACTIVATE(kn, 1);
483 			kn->kn_status &= ~KN_HASKQLOCK;
484 			KQ_UNLOCK(kq);
485 			continue;
486 		}
487 
488 		/*
489 		 * The NOTE_TRACK case. In addition to the activation
490 		 * of the event, we need to register new event to
491 		 * track the child. Drop the locks in preparation for
492 		 * the call to kqueue_register().
493 		 */
494 		kn->kn_status |= KN_INFLUX;
495 		KQ_UNLOCK(kq);
496 		list->kl_unlock(list->kl_lockarg);
497 
498 		/*
499 		 * Activate existing knote and register a knote with
500 		 * new process.
501 		 */
502 		kev.ident = pid;
503 		kev.filter = kn->kn_filter;
504 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
505 		kev.fflags = kn->kn_sfflags;
506 		kev.data = kn->kn_id;		/* parent */
507 		kev.udata = kn->kn_kevent.udata;/* preserve udata */
508 		error = kqueue_register(kq, &kev, NULL, 0);
509 		if (error)
510 			kn->kn_fflags |= NOTE_TRACKERR;
511 		if (kn->kn_fop->f_event(kn, NOTE_FORK))
512 			KNOTE_ACTIVATE(kn, 0);
513 		KQ_LOCK(kq);
514 		kn->kn_status &= ~KN_INFLUX;
515 		KQ_UNLOCK_FLUX(kq);
516 		list->kl_lock(list->kl_lockarg);
517 	}
518 	list->kl_unlock(list->kl_lockarg);
519 }
520 
521 /*
522  * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
523  * interval timer support code.
524  */
525 static __inline sbintime_t
526 timer2sbintime(intptr_t data)
527 {
528 
529 	return (SBT_1MS * data);
530 }
531 
532 static void
533 filt_timerexpire(void *knx)
534 {
535 	struct callout *calloutp;
536 	struct knote *kn;
537 
538 	kn = knx;
539 	kn->kn_data++;
540 	KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
541 
542 	if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
543 		calloutp = (struct callout *)kn->kn_hook;
544 		callout_reset_sbt_on(calloutp,
545 		    timer2sbintime(kn->kn_sdata), 0 /* 1ms? */,
546 		    filt_timerexpire, kn, PCPU_GET(cpuid), 0);
547 	}
548 }
549 
550 /*
551  * data contains amount of time to sleep, in milliseconds
552  */
553 static int
554 filt_timerattach(struct knote *kn)
555 {
556 	struct callout *calloutp;
557 	unsigned int ncallouts;
558 
559 	ncallouts = atomic_load_explicit(&kq_ncallouts, memory_order_relaxed);
560 	do {
561 		if (ncallouts >= kq_calloutmax)
562 			return (ENOMEM);
563 	} while (!atomic_compare_exchange_weak_explicit(&kq_ncallouts,
564 	    &ncallouts, ncallouts + 1, memory_order_relaxed,
565 	    memory_order_relaxed));
566 
567 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
568 	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
569 	calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK);
570 	callout_init(calloutp, CALLOUT_MPSAFE);
571 	kn->kn_hook = calloutp;
572 	callout_reset_sbt_on(calloutp,
573 	    timer2sbintime(kn->kn_sdata), 0 /* 1ms? */,
574 	    filt_timerexpire, kn, PCPU_GET(cpuid), 0);
575 
576 	return (0);
577 }
578 
579 static void
580 filt_timerdetach(struct knote *kn)
581 {
582 	struct callout *calloutp;
583 	unsigned int old;
584 
585 	calloutp = (struct callout *)kn->kn_hook;
586 	callout_drain(calloutp);
587 	free(calloutp, M_KQUEUE);
588 	old = atomic_fetch_sub_explicit(&kq_ncallouts, 1, memory_order_relaxed);
589 	KASSERT(old > 0, ("Number of callouts cannot become negative"));
590 	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
591 }
592 
593 static int
594 filt_timer(struct knote *kn, long hint)
595 {
596 
597 	return (kn->kn_data != 0);
598 }
599 
600 static int
601 filt_userattach(struct knote *kn)
602 {
603 
604 	/*
605 	 * EVFILT_USER knotes are not attached to anything in the kernel.
606 	 */
607 	kn->kn_hook = NULL;
608 	if (kn->kn_fflags & NOTE_TRIGGER)
609 		kn->kn_hookid = 1;
610 	else
611 		kn->kn_hookid = 0;
612 	return (0);
613 }
614 
615 static void
616 filt_userdetach(__unused struct knote *kn)
617 {
618 
619 	/*
620 	 * EVFILT_USER knotes are not attached to anything in the kernel.
621 	 */
622 }
623 
624 static int
625 filt_user(struct knote *kn, __unused long hint)
626 {
627 
628 	return (kn->kn_hookid);
629 }
630 
631 static void
632 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
633 {
634 	u_int ffctrl;
635 
636 	switch (type) {
637 	case EVENT_REGISTER:
638 		if (kev->fflags & NOTE_TRIGGER)
639 			kn->kn_hookid = 1;
640 
641 		ffctrl = kev->fflags & NOTE_FFCTRLMASK;
642 		kev->fflags &= NOTE_FFLAGSMASK;
643 		switch (ffctrl) {
644 		case NOTE_FFNOP:
645 			break;
646 
647 		case NOTE_FFAND:
648 			kn->kn_sfflags &= kev->fflags;
649 			break;
650 
651 		case NOTE_FFOR:
652 			kn->kn_sfflags |= kev->fflags;
653 			break;
654 
655 		case NOTE_FFCOPY:
656 			kn->kn_sfflags = kev->fflags;
657 			break;
658 
659 		default:
660 			/* XXX Return error? */
661 			break;
662 		}
663 		kn->kn_sdata = kev->data;
664 		if (kev->flags & EV_CLEAR) {
665 			kn->kn_hookid = 0;
666 			kn->kn_data = 0;
667 			kn->kn_fflags = 0;
668 		}
669 		break;
670 
671         case EVENT_PROCESS:
672 		*kev = kn->kn_kevent;
673 		kev->fflags = kn->kn_sfflags;
674 		kev->data = kn->kn_sdata;
675 		if (kn->kn_flags & EV_CLEAR) {
676 			kn->kn_hookid = 0;
677 			kn->kn_data = 0;
678 			kn->kn_fflags = 0;
679 		}
680 		break;
681 
682 	default:
683 		panic("filt_usertouch() - invalid type (%ld)", type);
684 		break;
685 	}
686 }
687 
688 int
689 sys_kqueue(struct thread *td, struct kqueue_args *uap)
690 {
691 	struct filedesc *fdp;
692 	struct kqueue *kq;
693 	struct file *fp;
694 	int fd, error;
695 
696 	fdp = td->td_proc->p_fd;
697 	error = falloc(td, &fp, &fd, 0);
698 	if (error)
699 		goto done2;
700 
701 	/* An extra reference on `fp' has been held for us by falloc(). */
702 	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
703 	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
704 	TAILQ_INIT(&kq->kq_head);
705 	kq->kq_fdp = fdp;
706 	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
707 	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
708 
709 	FILEDESC_XLOCK(fdp);
710 	SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
711 	FILEDESC_XUNLOCK(fdp);
712 
713 	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
714 	fdrop(fp, td);
715 
716 	td->td_retval[0] = fd;
717 done2:
718 	return (error);
719 }
720 
721 #ifndef _SYS_SYSPROTO_H_
722 struct kevent_args {
723 	int	fd;
724 	const struct kevent *changelist;
725 	int	nchanges;
726 	struct	kevent *eventlist;
727 	int	nevents;
728 	const struct timespec *timeout;
729 };
730 #endif
731 int
732 sys_kevent(struct thread *td, struct kevent_args *uap)
733 {
734 	struct timespec ts, *tsp;
735 	struct kevent_copyops k_ops = { uap,
736 					kevent_copyout,
737 					kevent_copyin};
738 	int error;
739 #ifdef KTRACE
740 	struct uio ktruio;
741 	struct iovec ktriov;
742 	struct uio *ktruioin = NULL;
743 	struct uio *ktruioout = NULL;
744 #endif
745 
746 	if (uap->timeout != NULL) {
747 		error = copyin(uap->timeout, &ts, sizeof(ts));
748 		if (error)
749 			return (error);
750 		tsp = &ts;
751 	} else
752 		tsp = NULL;
753 
754 #ifdef KTRACE
755 	if (KTRPOINT(td, KTR_GENIO)) {
756 		ktriov.iov_base = uap->changelist;
757 		ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
758 		ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
759 		    .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
760 		    .uio_td = td };
761 		ktruioin = cloneuio(&ktruio);
762 		ktriov.iov_base = uap->eventlist;
763 		ktriov.iov_len = uap->nevents * sizeof(struct kevent);
764 		ktruioout = cloneuio(&ktruio);
765 	}
766 #endif
767 
768 	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
769 	    &k_ops, tsp);
770 
771 #ifdef KTRACE
772 	if (ktruioin != NULL) {
773 		ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
774 		ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
775 		ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
776 		ktrgenio(uap->fd, UIO_READ, ktruioout, error);
777 	}
778 #endif
779 
780 	return (error);
781 }
782 
783 /*
784  * Copy 'count' items into the destination list pointed to by uap->eventlist.
785  */
786 static int
787 kevent_copyout(void *arg, struct kevent *kevp, int count)
788 {
789 	struct kevent_args *uap;
790 	int error;
791 
792 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
793 	uap = (struct kevent_args *)arg;
794 
795 	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
796 	if (error == 0)
797 		uap->eventlist += count;
798 	return (error);
799 }
800 
801 /*
802  * Copy 'count' items from the list pointed to by uap->changelist.
803  */
804 static int
805 kevent_copyin(void *arg, struct kevent *kevp, int count)
806 {
807 	struct kevent_args *uap;
808 	int error;
809 
810 	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
811 	uap = (struct kevent_args *)arg;
812 
813 	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
814 	if (error == 0)
815 		uap->changelist += count;
816 	return (error);
817 }
818 
819 int
820 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
821     struct kevent_copyops *k_ops, const struct timespec *timeout)
822 {
823 	struct kevent keva[KQ_NEVENTS];
824 	struct kevent *kevp, *changes;
825 	struct kqueue *kq;
826 	struct file *fp;
827 	cap_rights_t rights;
828 	int i, n, nerrors, error;
829 
830 	error = fget(td, fd, cap_rights_init(&rights, CAP_POST_EVENT), &fp);
831 	if (error != 0)
832 		return (error);
833 	if ((error = kqueue_acquire(fp, &kq)) != 0)
834 		goto done_norel;
835 
836 	nerrors = 0;
837 
838 	while (nchanges > 0) {
839 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
840 		error = k_ops->k_copyin(k_ops->arg, keva, n);
841 		if (error)
842 			goto done;
843 		changes = keva;
844 		for (i = 0; i < n; i++) {
845 			kevp = &changes[i];
846 			if (!kevp->filter)
847 				continue;
848 			kevp->flags &= ~EV_SYSFLAGS;
849 			error = kqueue_register(kq, kevp, td, 1);
850 			if (error || (kevp->flags & EV_RECEIPT)) {
851 				if (nevents != 0) {
852 					kevp->flags = EV_ERROR;
853 					kevp->data = error;
854 					(void) k_ops->k_copyout(k_ops->arg,
855 					    kevp, 1);
856 					nevents--;
857 					nerrors++;
858 				} else {
859 					goto done;
860 				}
861 			}
862 		}
863 		nchanges -= n;
864 	}
865 	if (nerrors) {
866 		td->td_retval[0] = nerrors;
867 		error = 0;
868 		goto done;
869 	}
870 
871 	error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
872 done:
873 	kqueue_release(kq, 0);
874 done_norel:
875 	fdrop(fp, td);
876 	return (error);
877 }
878 
879 int
880 kqueue_add_filteropts(int filt, struct filterops *filtops)
881 {
882 	int error;
883 
884 	error = 0;
885 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
886 		printf(
887 "trying to add a filterop that is out of range: %d is beyond %d\n",
888 		    ~filt, EVFILT_SYSCOUNT);
889 		return EINVAL;
890 	}
891 	mtx_lock(&filterops_lock);
892 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
893 	    sysfilt_ops[~filt].for_fop != NULL)
894 		error = EEXIST;
895 	else {
896 		sysfilt_ops[~filt].for_fop = filtops;
897 		sysfilt_ops[~filt].for_refcnt = 0;
898 	}
899 	mtx_unlock(&filterops_lock);
900 
901 	return (error);
902 }
903 
904 int
905 kqueue_del_filteropts(int filt)
906 {
907 	int error;
908 
909 	error = 0;
910 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
911 		return EINVAL;
912 
913 	mtx_lock(&filterops_lock);
914 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
915 	    sysfilt_ops[~filt].for_fop == NULL)
916 		error = EINVAL;
917 	else if (sysfilt_ops[~filt].for_refcnt != 0)
918 		error = EBUSY;
919 	else {
920 		sysfilt_ops[~filt].for_fop = &null_filtops;
921 		sysfilt_ops[~filt].for_refcnt = 0;
922 	}
923 	mtx_unlock(&filterops_lock);
924 
925 	return error;
926 }
927 
928 static struct filterops *
929 kqueue_fo_find(int filt)
930 {
931 
932 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
933 		return NULL;
934 
935 	mtx_lock(&filterops_lock);
936 	sysfilt_ops[~filt].for_refcnt++;
937 	if (sysfilt_ops[~filt].for_fop == NULL)
938 		sysfilt_ops[~filt].for_fop = &null_filtops;
939 	mtx_unlock(&filterops_lock);
940 
941 	return sysfilt_ops[~filt].for_fop;
942 }
943 
944 static void
945 kqueue_fo_release(int filt)
946 {
947 
948 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
949 		return;
950 
951 	mtx_lock(&filterops_lock);
952 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
953 	    ("filter object refcount not valid on release"));
954 	sysfilt_ops[~filt].for_refcnt--;
955 	mtx_unlock(&filterops_lock);
956 }
957 
958 /*
959  * A ref to kq (obtained via kqueue_acquire) must be held.  waitok will
960  * influence if memory allocation should wait.  Make sure it is 0 if you
961  * hold any mutexes.
962  */
963 static int
964 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
965 {
966 	struct filterops *fops;
967 	struct file *fp;
968 	struct knote *kn, *tkn;
969 	cap_rights_t rights;
970 	int error, filt, event;
971 	int haskqglobal;
972 
973 	fp = NULL;
974 	kn = NULL;
975 	error = 0;
976 	haskqglobal = 0;
977 
978 	filt = kev->filter;
979 	fops = kqueue_fo_find(filt);
980 	if (fops == NULL)
981 		return EINVAL;
982 
983 	tkn = knote_alloc(waitok);		/* prevent waiting with locks */
984 
985 findkn:
986 	if (fops->f_isfd) {
987 		KASSERT(td != NULL, ("td is NULL"));
988 		error = fget(td, kev->ident,
989 		    cap_rights_init(&rights, CAP_POLL_EVENT), &fp);
990 		if (error)
991 			goto done;
992 
993 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
994 		    kev->ident, 0) != 0) {
995 			/* try again */
996 			fdrop(fp, td);
997 			fp = NULL;
998 			error = kqueue_expand(kq, fops, kev->ident, waitok);
999 			if (error)
1000 				goto done;
1001 			goto findkn;
1002 		}
1003 
1004 		if (fp->f_type == DTYPE_KQUEUE) {
1005 			/*
1006 			 * if we add some inteligence about what we are doing,
1007 			 * we should be able to support events on ourselves.
1008 			 * We need to know when we are doing this to prevent
1009 			 * getting both the knlist lock and the kq lock since
1010 			 * they are the same thing.
1011 			 */
1012 			if (fp->f_data == kq) {
1013 				error = EINVAL;
1014 				goto done;
1015 			}
1016 
1017 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1018 		}
1019 
1020 		KQ_LOCK(kq);
1021 		if (kev->ident < kq->kq_knlistsize) {
1022 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1023 				if (kev->filter == kn->kn_filter)
1024 					break;
1025 		}
1026 	} else {
1027 		if ((kev->flags & EV_ADD) == EV_ADD)
1028 			kqueue_expand(kq, fops, kev->ident, waitok);
1029 
1030 		KQ_LOCK(kq);
1031 		if (kq->kq_knhashmask != 0) {
1032 			struct klist *list;
1033 
1034 			list = &kq->kq_knhash[
1035 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1036 			SLIST_FOREACH(kn, list, kn_link)
1037 				if (kev->ident == kn->kn_id &&
1038 				    kev->filter == kn->kn_filter)
1039 					break;
1040 		}
1041 	}
1042 
1043 	/* knote is in the process of changing, wait for it to stablize. */
1044 	if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1045 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1046 		kq->kq_state |= KQ_FLUXWAIT;
1047 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1048 		if (fp != NULL) {
1049 			fdrop(fp, td);
1050 			fp = NULL;
1051 		}
1052 		goto findkn;
1053 	}
1054 
1055 	/*
1056 	 * kn now contains the matching knote, or NULL if no match
1057 	 */
1058 	if (kn == NULL) {
1059 		if (kev->flags & EV_ADD) {
1060 			kn = tkn;
1061 			tkn = NULL;
1062 			if (kn == NULL) {
1063 				KQ_UNLOCK(kq);
1064 				error = ENOMEM;
1065 				goto done;
1066 			}
1067 			kn->kn_fp = fp;
1068 			kn->kn_kq = kq;
1069 			kn->kn_fop = fops;
1070 			/*
1071 			 * apply reference counts to knote structure, and
1072 			 * do not release it at the end of this routine.
1073 			 */
1074 			fops = NULL;
1075 			fp = NULL;
1076 
1077 			kn->kn_sfflags = kev->fflags;
1078 			kn->kn_sdata = kev->data;
1079 			kev->fflags = 0;
1080 			kev->data = 0;
1081 			kn->kn_kevent = *kev;
1082 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1083 			    EV_ENABLE | EV_DISABLE);
1084 			kn->kn_status = KN_INFLUX|KN_DETACHED;
1085 
1086 			error = knote_attach(kn, kq);
1087 			KQ_UNLOCK(kq);
1088 			if (error != 0) {
1089 				tkn = kn;
1090 				goto done;
1091 			}
1092 
1093 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1094 				knote_drop(kn, td);
1095 				goto done;
1096 			}
1097 			KN_LIST_LOCK(kn);
1098 			goto done_ev_add;
1099 		} else {
1100 			/* No matching knote and the EV_ADD flag is not set. */
1101 			KQ_UNLOCK(kq);
1102 			error = ENOENT;
1103 			goto done;
1104 		}
1105 	}
1106 
1107 	if (kev->flags & EV_DELETE) {
1108 		kn->kn_status |= KN_INFLUX;
1109 		KQ_UNLOCK(kq);
1110 		if (!(kn->kn_status & KN_DETACHED))
1111 			kn->kn_fop->f_detach(kn);
1112 		knote_drop(kn, td);
1113 		goto done;
1114 	}
1115 
1116 	/*
1117 	 * The user may change some filter values after the initial EV_ADD,
1118 	 * but doing so will not reset any filter which has already been
1119 	 * triggered.
1120 	 */
1121 	kn->kn_status |= KN_INFLUX;
1122 	KQ_UNLOCK(kq);
1123 	KN_LIST_LOCK(kn);
1124 	kn->kn_kevent.udata = kev->udata;
1125 	if (!fops->f_isfd && fops->f_touch != NULL) {
1126 		fops->f_touch(kn, kev, EVENT_REGISTER);
1127 	} else {
1128 		kn->kn_sfflags = kev->fflags;
1129 		kn->kn_sdata = kev->data;
1130 	}
1131 
1132 	/*
1133 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1134 	 * the initial attach event decides that the event is "completed"
1135 	 * already.  i.e. filt_procattach is called on a zombie process.  It
1136 	 * will call filt_proc which will remove it from the list, and NULL
1137 	 * kn_knlist.
1138 	 */
1139 done_ev_add:
1140 	event = kn->kn_fop->f_event(kn, 0);
1141 	KQ_LOCK(kq);
1142 	if (event)
1143 		KNOTE_ACTIVATE(kn, 1);
1144 	kn->kn_status &= ~KN_INFLUX;
1145 	KN_LIST_UNLOCK(kn);
1146 
1147 	if ((kev->flags & EV_DISABLE) &&
1148 	    ((kn->kn_status & KN_DISABLED) == 0)) {
1149 		kn->kn_status |= KN_DISABLED;
1150 	}
1151 
1152 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1153 		kn->kn_status &= ~KN_DISABLED;
1154 		if ((kn->kn_status & KN_ACTIVE) &&
1155 		    ((kn->kn_status & KN_QUEUED) == 0))
1156 			knote_enqueue(kn);
1157 	}
1158 	KQ_UNLOCK_FLUX(kq);
1159 
1160 done:
1161 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1162 	if (fp != NULL)
1163 		fdrop(fp, td);
1164 	if (tkn != NULL)
1165 		knote_free(tkn);
1166 	if (fops != NULL)
1167 		kqueue_fo_release(filt);
1168 	return (error);
1169 }
1170 
1171 static int
1172 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1173 {
1174 	int error;
1175 	struct kqueue *kq;
1176 
1177 	error = 0;
1178 
1179 	kq = fp->f_data;
1180 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1181 		return (EBADF);
1182 	*kqp = kq;
1183 	KQ_LOCK(kq);
1184 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1185 		KQ_UNLOCK(kq);
1186 		return (EBADF);
1187 	}
1188 	kq->kq_refcnt++;
1189 	KQ_UNLOCK(kq);
1190 
1191 	return error;
1192 }
1193 
1194 static void
1195 kqueue_release(struct kqueue *kq, int locked)
1196 {
1197 	if (locked)
1198 		KQ_OWNED(kq);
1199 	else
1200 		KQ_LOCK(kq);
1201 	kq->kq_refcnt--;
1202 	if (kq->kq_refcnt == 1)
1203 		wakeup(&kq->kq_refcnt);
1204 	if (!locked)
1205 		KQ_UNLOCK(kq);
1206 }
1207 
1208 static void
1209 kqueue_schedtask(struct kqueue *kq)
1210 {
1211 
1212 	KQ_OWNED(kq);
1213 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1214 	    ("scheduling kqueue task while draining"));
1215 
1216 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1217 		taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1218 		kq->kq_state |= KQ_TASKSCHED;
1219 	}
1220 }
1221 
1222 /*
1223  * Expand the kq to make sure we have storage for fops/ident pair.
1224  *
1225  * Return 0 on success (or no work necessary), return errno on failure.
1226  *
1227  * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1228  * If kqueue_register is called from a non-fd context, there usually/should
1229  * be no locks held.
1230  */
1231 static int
1232 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1233 	int waitok)
1234 {
1235 	struct klist *list, *tmp_knhash, *to_free;
1236 	u_long tmp_knhashmask;
1237 	int size;
1238 	int fd;
1239 	int mflag = waitok ? M_WAITOK : M_NOWAIT;
1240 
1241 	KQ_NOTOWNED(kq);
1242 
1243 	to_free = NULL;
1244 	if (fops->f_isfd) {
1245 		fd = ident;
1246 		if (kq->kq_knlistsize <= fd) {
1247 			size = kq->kq_knlistsize;
1248 			while (size <= fd)
1249 				size += KQEXTENT;
1250 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1251 			if (list == NULL)
1252 				return ENOMEM;
1253 			KQ_LOCK(kq);
1254 			if (kq->kq_knlistsize > fd) {
1255 				to_free = list;
1256 				list = NULL;
1257 			} else {
1258 				if (kq->kq_knlist != NULL) {
1259 					bcopy(kq->kq_knlist, list,
1260 					    kq->kq_knlistsize * sizeof(*list));
1261 					to_free = kq->kq_knlist;
1262 					kq->kq_knlist = NULL;
1263 				}
1264 				bzero((caddr_t)list +
1265 				    kq->kq_knlistsize * sizeof(*list),
1266 				    (size - kq->kq_knlistsize) * sizeof(*list));
1267 				kq->kq_knlistsize = size;
1268 				kq->kq_knlist = list;
1269 			}
1270 			KQ_UNLOCK(kq);
1271 		}
1272 	} else {
1273 		if (kq->kq_knhashmask == 0) {
1274 			tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1275 			    &tmp_knhashmask);
1276 			if (tmp_knhash == NULL)
1277 				return ENOMEM;
1278 			KQ_LOCK(kq);
1279 			if (kq->kq_knhashmask == 0) {
1280 				kq->kq_knhash = tmp_knhash;
1281 				kq->kq_knhashmask = tmp_knhashmask;
1282 			} else {
1283 				to_free = tmp_knhash;
1284 			}
1285 			KQ_UNLOCK(kq);
1286 		}
1287 	}
1288 	free(to_free, M_KQUEUE);
1289 
1290 	KQ_NOTOWNED(kq);
1291 	return 0;
1292 }
1293 
1294 static void
1295 kqueue_task(void *arg, int pending)
1296 {
1297 	struct kqueue *kq;
1298 	int haskqglobal;
1299 
1300 	haskqglobal = 0;
1301 	kq = arg;
1302 
1303 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1304 	KQ_LOCK(kq);
1305 
1306 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1307 
1308 	kq->kq_state &= ~KQ_TASKSCHED;
1309 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1310 		wakeup(&kq->kq_state);
1311 	}
1312 	KQ_UNLOCK(kq);
1313 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1314 }
1315 
1316 /*
1317  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1318  * We treat KN_MARKER knotes as if they are INFLUX.
1319  */
1320 static int
1321 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1322     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1323 {
1324 	struct kevent *kevp;
1325 	struct knote *kn, *marker;
1326 	sbintime_t asbt, rsbt;
1327 	int count, error, haskqglobal, influx, nkev, touch;
1328 
1329 	count = maxevents;
1330 	nkev = 0;
1331 	error = 0;
1332 	haskqglobal = 0;
1333 
1334 	if (maxevents == 0)
1335 		goto done_nl;
1336 
1337 	rsbt = 0;
1338 	if (tsp != NULL) {
1339 		if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 ||
1340 		    tsp->tv_nsec >= 1000000000) {
1341 			error = EINVAL;
1342 			goto done_nl;
1343 		}
1344 		if (timespecisset(tsp)) {
1345 			if (tsp->tv_sec <= INT32_MAX) {
1346 				rsbt = tstosbt(*tsp);
1347 				if (TIMESEL(&asbt, rsbt))
1348 					asbt += tc_tick_sbt;
1349 				if (asbt <= INT64_MAX - rsbt)
1350 					asbt += rsbt;
1351 				else
1352 					asbt = 0;
1353 				rsbt >>= tc_precexp;
1354 			} else
1355 				asbt = 0;
1356 		} else
1357 			asbt = -1;
1358 	} else
1359 		asbt = 0;
1360 	marker = knote_alloc(1);
1361 	if (marker == NULL) {
1362 		error = ENOMEM;
1363 		goto done_nl;
1364 	}
1365 	marker->kn_status = KN_MARKER;
1366 	KQ_LOCK(kq);
1367 
1368 retry:
1369 	kevp = keva;
1370 	if (kq->kq_count == 0) {
1371 		if (asbt == -1) {
1372 			error = EWOULDBLOCK;
1373 		} else {
1374 			kq->kq_state |= KQ_SLEEP;
1375 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1376 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1377 		}
1378 		if (error == 0)
1379 			goto retry;
1380 		/* don't restart after signals... */
1381 		if (error == ERESTART)
1382 			error = EINTR;
1383 		else if (error == EWOULDBLOCK)
1384 			error = 0;
1385 		goto done;
1386 	}
1387 
1388 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1389 	influx = 0;
1390 	while (count) {
1391 		KQ_OWNED(kq);
1392 		kn = TAILQ_FIRST(&kq->kq_head);
1393 
1394 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1395 		    (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1396 			if (influx) {
1397 				influx = 0;
1398 				KQ_FLUX_WAKEUP(kq);
1399 			}
1400 			kq->kq_state |= KQ_FLUXWAIT;
1401 			error = msleep(kq, &kq->kq_lock, PSOCK,
1402 			    "kqflxwt", 0);
1403 			continue;
1404 		}
1405 
1406 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1407 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1408 			kn->kn_status &= ~KN_QUEUED;
1409 			kq->kq_count--;
1410 			continue;
1411 		}
1412 		if (kn == marker) {
1413 			KQ_FLUX_WAKEUP(kq);
1414 			if (count == maxevents)
1415 				goto retry;
1416 			goto done;
1417 		}
1418 		KASSERT((kn->kn_status & KN_INFLUX) == 0,
1419 		    ("KN_INFLUX set when not suppose to be"));
1420 
1421 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1422 			kn->kn_status &= ~KN_QUEUED;
1423 			kn->kn_status |= KN_INFLUX;
1424 			kq->kq_count--;
1425 			KQ_UNLOCK(kq);
1426 			/*
1427 			 * We don't need to lock the list since we've marked
1428 			 * it _INFLUX.
1429 			 */
1430 			if (!(kn->kn_status & KN_DETACHED))
1431 				kn->kn_fop->f_detach(kn);
1432 			knote_drop(kn, td);
1433 			KQ_LOCK(kq);
1434 			continue;
1435 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1436 			kn->kn_status &= ~KN_QUEUED;
1437 			kn->kn_status |= KN_INFLUX;
1438 			kq->kq_count--;
1439 			KQ_UNLOCK(kq);
1440 			/*
1441 			 * We don't need to lock the list since we've marked
1442 			 * it _INFLUX.
1443 			 */
1444 			*kevp = kn->kn_kevent;
1445 			if (!(kn->kn_status & KN_DETACHED))
1446 				kn->kn_fop->f_detach(kn);
1447 			knote_drop(kn, td);
1448 			KQ_LOCK(kq);
1449 			kn = NULL;
1450 		} else {
1451 			kn->kn_status |= KN_INFLUX;
1452 			KQ_UNLOCK(kq);
1453 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1454 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1455 			KN_LIST_LOCK(kn);
1456 			if (kn->kn_fop->f_event(kn, 0) == 0) {
1457 				KQ_LOCK(kq);
1458 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1459 				kn->kn_status &=
1460 				    ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX);
1461 				kq->kq_count--;
1462 				KN_LIST_UNLOCK(kn);
1463 				influx = 1;
1464 				continue;
1465 			}
1466 			touch = (!kn->kn_fop->f_isfd &&
1467 			    kn->kn_fop->f_touch != NULL);
1468 			if (touch)
1469 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1470 			else
1471 				*kevp = kn->kn_kevent;
1472 			KQ_LOCK(kq);
1473 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1474 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1475 				/*
1476 				 * Manually clear knotes who weren't
1477 				 * 'touch'ed.
1478 				 */
1479 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1480 					kn->kn_data = 0;
1481 					kn->kn_fflags = 0;
1482 				}
1483 				if (kn->kn_flags & EV_DISPATCH)
1484 					kn->kn_status |= KN_DISABLED;
1485 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1486 				kq->kq_count--;
1487 			} else
1488 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1489 
1490 			kn->kn_status &= ~(KN_INFLUX);
1491 			KN_LIST_UNLOCK(kn);
1492 			influx = 1;
1493 		}
1494 
1495 		/* we are returning a copy to the user */
1496 		kevp++;
1497 		nkev++;
1498 		count--;
1499 
1500 		if (nkev == KQ_NEVENTS) {
1501 			influx = 0;
1502 			KQ_UNLOCK_FLUX(kq);
1503 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1504 			nkev = 0;
1505 			kevp = keva;
1506 			KQ_LOCK(kq);
1507 			if (error)
1508 				break;
1509 		}
1510 	}
1511 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1512 done:
1513 	KQ_OWNED(kq);
1514 	KQ_UNLOCK_FLUX(kq);
1515 	knote_free(marker);
1516 done_nl:
1517 	KQ_NOTOWNED(kq);
1518 	if (nkev != 0)
1519 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1520 	td->td_retval[0] = maxevents - count;
1521 	return (error);
1522 }
1523 
1524 /*
1525  * XXX
1526  * This could be expanded to call kqueue_scan, if desired.
1527  */
1528 /*ARGSUSED*/
1529 static int
1530 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1531 	int flags, struct thread *td)
1532 {
1533 	return (ENXIO);
1534 }
1535 
1536 /*ARGSUSED*/
1537 static int
1538 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1539 	 int flags, struct thread *td)
1540 {
1541 	return (ENXIO);
1542 }
1543 
1544 /*ARGSUSED*/
1545 static int
1546 kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1547 	struct thread *td)
1548 {
1549 
1550 	return (EINVAL);
1551 }
1552 
1553 /*ARGSUSED*/
1554 static int
1555 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1556 	struct ucred *active_cred, struct thread *td)
1557 {
1558 	/*
1559 	 * Enabling sigio causes two major problems:
1560 	 * 1) infinite recursion:
1561 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
1562 	 * set.  On receipt of a signal this will cause a kqueue to recurse
1563 	 * into itself over and over.  Sending the sigio causes the kqueue
1564 	 * to become ready, which in turn posts sigio again, forever.
1565 	 * Solution: this can be solved by setting a flag in the kqueue that
1566 	 * we have a SIGIO in progress.
1567 	 * 2) locking problems:
1568 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1569 	 * us above the proc and pgrp locks.
1570 	 * Solution: Post a signal using an async mechanism, being sure to
1571 	 * record a generation count in the delivery so that we do not deliver
1572 	 * a signal to the wrong process.
1573 	 *
1574 	 * Note, these two mechanisms are somewhat mutually exclusive!
1575 	 */
1576 #if 0
1577 	struct kqueue *kq;
1578 
1579 	kq = fp->f_data;
1580 	switch (cmd) {
1581 	case FIOASYNC:
1582 		if (*(int *)data) {
1583 			kq->kq_state |= KQ_ASYNC;
1584 		} else {
1585 			kq->kq_state &= ~KQ_ASYNC;
1586 		}
1587 		return (0);
1588 
1589 	case FIOSETOWN:
1590 		return (fsetown(*(int *)data, &kq->kq_sigio));
1591 
1592 	case FIOGETOWN:
1593 		*(int *)data = fgetown(&kq->kq_sigio);
1594 		return (0);
1595 	}
1596 #endif
1597 
1598 	return (ENOTTY);
1599 }
1600 
1601 /*ARGSUSED*/
1602 static int
1603 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1604 	struct thread *td)
1605 {
1606 	struct kqueue *kq;
1607 	int revents = 0;
1608 	int error;
1609 
1610 	if ((error = kqueue_acquire(fp, &kq)))
1611 		return POLLERR;
1612 
1613 	KQ_LOCK(kq);
1614 	if (events & (POLLIN | POLLRDNORM)) {
1615 		if (kq->kq_count) {
1616 			revents |= events & (POLLIN | POLLRDNORM);
1617 		} else {
1618 			selrecord(td, &kq->kq_sel);
1619 			if (SEL_WAITING(&kq->kq_sel))
1620 				kq->kq_state |= KQ_SEL;
1621 		}
1622 	}
1623 	kqueue_release(kq, 1);
1624 	KQ_UNLOCK(kq);
1625 	return (revents);
1626 }
1627 
1628 /*ARGSUSED*/
1629 static int
1630 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1631 	struct thread *td)
1632 {
1633 
1634 	bzero((void *)st, sizeof *st);
1635 	/*
1636 	 * We no longer return kq_count because the unlocked value is useless.
1637 	 * If you spent all this time getting the count, why not spend your
1638 	 * syscall better by calling kevent?
1639 	 *
1640 	 * XXX - This is needed for libc_r.
1641 	 */
1642 	st->st_mode = S_IFIFO;
1643 	return (0);
1644 }
1645 
1646 /*ARGSUSED*/
1647 static int
1648 kqueue_close(struct file *fp, struct thread *td)
1649 {
1650 	struct kqueue *kq = fp->f_data;
1651 	struct filedesc *fdp;
1652 	struct knote *kn;
1653 	int i;
1654 	int error;
1655 
1656 	if ((error = kqueue_acquire(fp, &kq)))
1657 		return error;
1658 
1659 	KQ_LOCK(kq);
1660 
1661 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1662 	    ("kqueue already closing"));
1663 	kq->kq_state |= KQ_CLOSING;
1664 	if (kq->kq_refcnt > 1)
1665 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1666 
1667 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1668 	fdp = kq->kq_fdp;
1669 
1670 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
1671 	    ("kqueue's knlist not empty"));
1672 
1673 	for (i = 0; i < kq->kq_knlistsize; i++) {
1674 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1675 			if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1676 				kq->kq_state |= KQ_FLUXWAIT;
1677 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1678 				continue;
1679 			}
1680 			kn->kn_status |= KN_INFLUX;
1681 			KQ_UNLOCK(kq);
1682 			if (!(kn->kn_status & KN_DETACHED))
1683 				kn->kn_fop->f_detach(kn);
1684 			knote_drop(kn, td);
1685 			KQ_LOCK(kq);
1686 		}
1687 	}
1688 	if (kq->kq_knhashmask != 0) {
1689 		for (i = 0; i <= kq->kq_knhashmask; i++) {
1690 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1691 				if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1692 					kq->kq_state |= KQ_FLUXWAIT;
1693 					msleep(kq, &kq->kq_lock, PSOCK,
1694 					       "kqclo2", 0);
1695 					continue;
1696 				}
1697 				kn->kn_status |= KN_INFLUX;
1698 				KQ_UNLOCK(kq);
1699 				if (!(kn->kn_status & KN_DETACHED))
1700 					kn->kn_fop->f_detach(kn);
1701 				knote_drop(kn, td);
1702 				KQ_LOCK(kq);
1703 			}
1704 		}
1705 	}
1706 
1707 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1708 		kq->kq_state |= KQ_TASKDRAIN;
1709 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1710 	}
1711 
1712 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1713 		selwakeuppri(&kq->kq_sel, PSOCK);
1714 		if (!SEL_WAITING(&kq->kq_sel))
1715 			kq->kq_state &= ~KQ_SEL;
1716 	}
1717 
1718 	KQ_UNLOCK(kq);
1719 
1720 	FILEDESC_XLOCK(fdp);
1721 	SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list);
1722 	FILEDESC_XUNLOCK(fdp);
1723 
1724 	seldrain(&kq->kq_sel);
1725 	knlist_destroy(&kq->kq_sel.si_note);
1726 	mtx_destroy(&kq->kq_lock);
1727 	kq->kq_fdp = NULL;
1728 
1729 	if (kq->kq_knhash != NULL)
1730 		free(kq->kq_knhash, M_KQUEUE);
1731 	if (kq->kq_knlist != NULL)
1732 		free(kq->kq_knlist, M_KQUEUE);
1733 
1734 	funsetown(&kq->kq_sigio);
1735 	free(kq, M_KQUEUE);
1736 	fp->f_data = NULL;
1737 
1738 	return (0);
1739 }
1740 
1741 static void
1742 kqueue_wakeup(struct kqueue *kq)
1743 {
1744 	KQ_OWNED(kq);
1745 
1746 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1747 		kq->kq_state &= ~KQ_SLEEP;
1748 		wakeup(kq);
1749 	}
1750 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1751 		selwakeuppri(&kq->kq_sel, PSOCK);
1752 		if (!SEL_WAITING(&kq->kq_sel))
1753 			kq->kq_state &= ~KQ_SEL;
1754 	}
1755 	if (!knlist_empty(&kq->kq_sel.si_note))
1756 		kqueue_schedtask(kq);
1757 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1758 		pgsigio(&kq->kq_sigio, SIGIO, 0);
1759 	}
1760 }
1761 
1762 /*
1763  * Walk down a list of knotes, activating them if their event has triggered.
1764  *
1765  * There is a possibility to optimize in the case of one kq watching another.
1766  * Instead of scheduling a task to wake it up, you could pass enough state
1767  * down the chain to make up the parent kqueue.  Make this code functional
1768  * first.
1769  */
1770 void
1771 knote(struct knlist *list, long hint, int lockflags)
1772 {
1773 	struct kqueue *kq;
1774 	struct knote *kn;
1775 	int error;
1776 
1777 	if (list == NULL)
1778 		return;
1779 
1780 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1781 
1782 	if ((lockflags & KNF_LISTLOCKED) == 0)
1783 		list->kl_lock(list->kl_lockarg);
1784 
1785 	/*
1786 	 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1787 	 * the kqueue scheduling, but this will introduce four
1788 	 * lock/unlock's for each knote to test.  If we do, continue to use
1789 	 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1790 	 * only safe if you want to remove the current item, which we are
1791 	 * not doing.
1792 	 */
1793 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1794 		kq = kn->kn_kq;
1795 		if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1796 			KQ_LOCK(kq);
1797 			if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1798 				KQ_UNLOCK(kq);
1799 			} else if ((lockflags & KNF_NOKQLOCK) != 0) {
1800 				kn->kn_status |= KN_INFLUX;
1801 				KQ_UNLOCK(kq);
1802 				error = kn->kn_fop->f_event(kn, hint);
1803 				KQ_LOCK(kq);
1804 				kn->kn_status &= ~KN_INFLUX;
1805 				if (error)
1806 					KNOTE_ACTIVATE(kn, 1);
1807 				KQ_UNLOCK_FLUX(kq);
1808 			} else {
1809 				kn->kn_status |= KN_HASKQLOCK;
1810 				if (kn->kn_fop->f_event(kn, hint))
1811 					KNOTE_ACTIVATE(kn, 1);
1812 				kn->kn_status &= ~KN_HASKQLOCK;
1813 				KQ_UNLOCK(kq);
1814 			}
1815 		}
1816 		kq = NULL;
1817 	}
1818 	if ((lockflags & KNF_LISTLOCKED) == 0)
1819 		list->kl_unlock(list->kl_lockarg);
1820 }
1821 
1822 /*
1823  * add a knote to a knlist
1824  */
1825 void
1826 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1827 {
1828 	KNL_ASSERT_LOCK(knl, islocked);
1829 	KQ_NOTOWNED(kn->kn_kq);
1830 	KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1831 	    (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1832 	if (!islocked)
1833 		knl->kl_lock(knl->kl_lockarg);
1834 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1835 	if (!islocked)
1836 		knl->kl_unlock(knl->kl_lockarg);
1837 	KQ_LOCK(kn->kn_kq);
1838 	kn->kn_knlist = knl;
1839 	kn->kn_status &= ~KN_DETACHED;
1840 	KQ_UNLOCK(kn->kn_kq);
1841 }
1842 
1843 static void
1844 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1845 {
1846 	KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1847 	KNL_ASSERT_LOCK(knl, knlislocked);
1848 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1849 	if (!kqislocked)
1850 		KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1851     ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1852 	if (!knlislocked)
1853 		knl->kl_lock(knl->kl_lockarg);
1854 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1855 	kn->kn_knlist = NULL;
1856 	if (!knlislocked)
1857 		knl->kl_unlock(knl->kl_lockarg);
1858 	if (!kqislocked)
1859 		KQ_LOCK(kn->kn_kq);
1860 	kn->kn_status |= KN_DETACHED;
1861 	if (!kqislocked)
1862 		KQ_UNLOCK(kn->kn_kq);
1863 }
1864 
1865 /*
1866  * remove knote from the specified knlist
1867  */
1868 void
1869 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1870 {
1871 
1872 	knlist_remove_kq(knl, kn, islocked, 0);
1873 }
1874 
1875 /*
1876  * remove knote from the specified knlist while in f_event handler.
1877  */
1878 void
1879 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1880 {
1881 
1882 	knlist_remove_kq(knl, kn, 1,
1883 	    (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1884 }
1885 
1886 int
1887 knlist_empty(struct knlist *knl)
1888 {
1889 
1890 	KNL_ASSERT_LOCKED(knl);
1891 	return SLIST_EMPTY(&knl->kl_list);
1892 }
1893 
1894 static struct mtx	knlist_lock;
1895 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1896 	MTX_DEF);
1897 static void knlist_mtx_lock(void *arg);
1898 static void knlist_mtx_unlock(void *arg);
1899 
1900 static void
1901 knlist_mtx_lock(void *arg)
1902 {
1903 
1904 	mtx_lock((struct mtx *)arg);
1905 }
1906 
1907 static void
1908 knlist_mtx_unlock(void *arg)
1909 {
1910 
1911 	mtx_unlock((struct mtx *)arg);
1912 }
1913 
1914 static void
1915 knlist_mtx_assert_locked(void *arg)
1916 {
1917 
1918 	mtx_assert((struct mtx *)arg, MA_OWNED);
1919 }
1920 
1921 static void
1922 knlist_mtx_assert_unlocked(void *arg)
1923 {
1924 
1925 	mtx_assert((struct mtx *)arg, MA_NOTOWNED);
1926 }
1927 
1928 static void
1929 knlist_rw_rlock(void *arg)
1930 {
1931 
1932 	rw_rlock((struct rwlock *)arg);
1933 }
1934 
1935 static void
1936 knlist_rw_runlock(void *arg)
1937 {
1938 
1939 	rw_runlock((struct rwlock *)arg);
1940 }
1941 
1942 static void
1943 knlist_rw_assert_locked(void *arg)
1944 {
1945 
1946 	rw_assert((struct rwlock *)arg, RA_LOCKED);
1947 }
1948 
1949 static void
1950 knlist_rw_assert_unlocked(void *arg)
1951 {
1952 
1953 	rw_assert((struct rwlock *)arg, RA_UNLOCKED);
1954 }
1955 
1956 void
1957 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1958     void (*kl_unlock)(void *),
1959     void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
1960 {
1961 
1962 	if (lock == NULL)
1963 		knl->kl_lockarg = &knlist_lock;
1964 	else
1965 		knl->kl_lockarg = lock;
1966 
1967 	if (kl_lock == NULL)
1968 		knl->kl_lock = knlist_mtx_lock;
1969 	else
1970 		knl->kl_lock = kl_lock;
1971 	if (kl_unlock == NULL)
1972 		knl->kl_unlock = knlist_mtx_unlock;
1973 	else
1974 		knl->kl_unlock = kl_unlock;
1975 	if (kl_assert_locked == NULL)
1976 		knl->kl_assert_locked = knlist_mtx_assert_locked;
1977 	else
1978 		knl->kl_assert_locked = kl_assert_locked;
1979 	if (kl_assert_unlocked == NULL)
1980 		knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
1981 	else
1982 		knl->kl_assert_unlocked = kl_assert_unlocked;
1983 
1984 	SLIST_INIT(&knl->kl_list);
1985 }
1986 
1987 void
1988 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
1989 {
1990 
1991 	knlist_init(knl, lock, NULL, NULL, NULL, NULL);
1992 }
1993 
1994 void
1995 knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock)
1996 {
1997 
1998 	knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock,
1999 	    knlist_rw_assert_locked, knlist_rw_assert_unlocked);
2000 }
2001 
2002 void
2003 knlist_destroy(struct knlist *knl)
2004 {
2005 
2006 #ifdef INVARIANTS
2007 	/*
2008 	 * if we run across this error, we need to find the offending
2009 	 * driver and have it call knlist_clear or knlist_delete.
2010 	 */
2011 	if (!SLIST_EMPTY(&knl->kl_list))
2012 		printf("WARNING: destroying knlist w/ knotes on it!\n");
2013 #endif
2014 
2015 	knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
2016 	SLIST_INIT(&knl->kl_list);
2017 }
2018 
2019 /*
2020  * Even if we are locked, we may need to drop the lock to allow any influx
2021  * knotes time to "settle".
2022  */
2023 void
2024 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2025 {
2026 	struct knote *kn, *kn2;
2027 	struct kqueue *kq;
2028 
2029 	if (islocked)
2030 		KNL_ASSERT_LOCKED(knl);
2031 	else {
2032 		KNL_ASSERT_UNLOCKED(knl);
2033 again:		/* need to reacquire lock since we have dropped it */
2034 		knl->kl_lock(knl->kl_lockarg);
2035 	}
2036 
2037 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2038 		kq = kn->kn_kq;
2039 		KQ_LOCK(kq);
2040 		if ((kn->kn_status & KN_INFLUX)) {
2041 			KQ_UNLOCK(kq);
2042 			continue;
2043 		}
2044 		knlist_remove_kq(knl, kn, 1, 1);
2045 		if (killkn) {
2046 			kn->kn_status |= KN_INFLUX | KN_DETACHED;
2047 			KQ_UNLOCK(kq);
2048 			knote_drop(kn, td);
2049 		} else {
2050 			/* Make sure cleared knotes disappear soon */
2051 			kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2052 			KQ_UNLOCK(kq);
2053 		}
2054 		kq = NULL;
2055 	}
2056 
2057 	if (!SLIST_EMPTY(&knl->kl_list)) {
2058 		/* there are still KN_INFLUX remaining */
2059 		kn = SLIST_FIRST(&knl->kl_list);
2060 		kq = kn->kn_kq;
2061 		KQ_LOCK(kq);
2062 		KASSERT(kn->kn_status & KN_INFLUX,
2063 		    ("knote removed w/o list lock"));
2064 		knl->kl_unlock(knl->kl_lockarg);
2065 		kq->kq_state |= KQ_FLUXWAIT;
2066 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2067 		kq = NULL;
2068 		goto again;
2069 	}
2070 
2071 	if (islocked)
2072 		KNL_ASSERT_LOCKED(knl);
2073 	else {
2074 		knl->kl_unlock(knl->kl_lockarg);
2075 		KNL_ASSERT_UNLOCKED(knl);
2076 	}
2077 }
2078 
2079 /*
2080  * Remove all knotes referencing a specified fd must be called with FILEDESC
2081  * lock.  This prevents a race where a new fd comes along and occupies the
2082  * entry and we attach a knote to the fd.
2083  */
2084 void
2085 knote_fdclose(struct thread *td, int fd)
2086 {
2087 	struct filedesc *fdp = td->td_proc->p_fd;
2088 	struct kqueue *kq;
2089 	struct knote *kn;
2090 	int influx;
2091 
2092 	FILEDESC_XLOCK_ASSERT(fdp);
2093 
2094 	/*
2095 	 * We shouldn't have to worry about new kevents appearing on fd
2096 	 * since filedesc is locked.
2097 	 */
2098 	SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2099 		KQ_LOCK(kq);
2100 
2101 again:
2102 		influx = 0;
2103 		while (kq->kq_knlistsize > fd &&
2104 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2105 			if (kn->kn_status & KN_INFLUX) {
2106 				/* someone else might be waiting on our knote */
2107 				if (influx)
2108 					wakeup(kq);
2109 				kq->kq_state |= KQ_FLUXWAIT;
2110 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2111 				goto again;
2112 			}
2113 			kn->kn_status |= KN_INFLUX;
2114 			KQ_UNLOCK(kq);
2115 			if (!(kn->kn_status & KN_DETACHED))
2116 				kn->kn_fop->f_detach(kn);
2117 			knote_drop(kn, td);
2118 			influx = 1;
2119 			KQ_LOCK(kq);
2120 		}
2121 		KQ_UNLOCK_FLUX(kq);
2122 	}
2123 }
2124 
2125 static int
2126 knote_attach(struct knote *kn, struct kqueue *kq)
2127 {
2128 	struct klist *list;
2129 
2130 	KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2131 	KQ_OWNED(kq);
2132 
2133 	if (kn->kn_fop->f_isfd) {
2134 		if (kn->kn_id >= kq->kq_knlistsize)
2135 			return ENOMEM;
2136 		list = &kq->kq_knlist[kn->kn_id];
2137 	} else {
2138 		if (kq->kq_knhash == NULL)
2139 			return ENOMEM;
2140 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2141 	}
2142 
2143 	SLIST_INSERT_HEAD(list, kn, kn_link);
2144 
2145 	return 0;
2146 }
2147 
2148 /*
2149  * knote must already have been detached using the f_detach method.
2150  * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2151  * to prevent other removal.
2152  */
2153 static void
2154 knote_drop(struct knote *kn, struct thread *td)
2155 {
2156 	struct kqueue *kq;
2157 	struct klist *list;
2158 
2159 	kq = kn->kn_kq;
2160 
2161 	KQ_NOTOWNED(kq);
2162 	KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2163 	    ("knote_drop called without KN_INFLUX set in kn_status"));
2164 
2165 	KQ_LOCK(kq);
2166 	if (kn->kn_fop->f_isfd)
2167 		list = &kq->kq_knlist[kn->kn_id];
2168 	else
2169 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2170 
2171 	if (!SLIST_EMPTY(list))
2172 		SLIST_REMOVE(list, kn, knote, kn_link);
2173 	if (kn->kn_status & KN_QUEUED)
2174 		knote_dequeue(kn);
2175 	KQ_UNLOCK_FLUX(kq);
2176 
2177 	if (kn->kn_fop->f_isfd) {
2178 		fdrop(kn->kn_fp, td);
2179 		kn->kn_fp = NULL;
2180 	}
2181 	kqueue_fo_release(kn->kn_kevent.filter);
2182 	kn->kn_fop = NULL;
2183 	knote_free(kn);
2184 }
2185 
2186 static void
2187 knote_enqueue(struct knote *kn)
2188 {
2189 	struct kqueue *kq = kn->kn_kq;
2190 
2191 	KQ_OWNED(kn->kn_kq);
2192 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2193 
2194 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2195 	kn->kn_status |= KN_QUEUED;
2196 	kq->kq_count++;
2197 	kqueue_wakeup(kq);
2198 }
2199 
2200 static void
2201 knote_dequeue(struct knote *kn)
2202 {
2203 	struct kqueue *kq = kn->kn_kq;
2204 
2205 	KQ_OWNED(kn->kn_kq);
2206 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2207 
2208 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2209 	kn->kn_status &= ~KN_QUEUED;
2210 	kq->kq_count--;
2211 }
2212 
2213 static void
2214 knote_init(void)
2215 {
2216 
2217 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2218 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2219 }
2220 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2221 
2222 static struct knote *
2223 knote_alloc(int waitok)
2224 {
2225 	return ((struct knote *)uma_zalloc(knote_zone,
2226 	    (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2227 }
2228 
2229 static void
2230 knote_free(struct knote *kn)
2231 {
2232 	if (kn != NULL)
2233 		uma_zfree(knote_zone, kn);
2234 }
2235 
2236 /*
2237  * Register the kev w/ the kq specified by fd.
2238  */
2239 int
2240 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2241 {
2242 	struct kqueue *kq;
2243 	struct file *fp;
2244 	cap_rights_t rights;
2245 	int error;
2246 
2247 	error = fget(td, fd, cap_rights_init(&rights, CAP_POST_EVENT), &fp);
2248 	if (error != 0)
2249 		return (error);
2250 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2251 		goto noacquire;
2252 
2253 	error = kqueue_register(kq, kev, td, waitok);
2254 
2255 	kqueue_release(kq, 0);
2256 
2257 noacquire:
2258 	fdrop(fp, td);
2259 
2260 	return error;
2261 }
2262