xref: /freebsd/sys/kern/kern_event.c (revision d940bfec8c329dd82d8d54efebd81c8aa420503b)
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 	int i, n, nerrors, error;
828 
829 	if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0)
830 		return (error);
831 	if ((error = kqueue_acquire(fp, &kq)) != 0)
832 		goto done_norel;
833 
834 	nerrors = 0;
835 
836 	while (nchanges > 0) {
837 		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
838 		error = k_ops->k_copyin(k_ops->arg, keva, n);
839 		if (error)
840 			goto done;
841 		changes = keva;
842 		for (i = 0; i < n; i++) {
843 			kevp = &changes[i];
844 			if (!kevp->filter)
845 				continue;
846 			kevp->flags &= ~EV_SYSFLAGS;
847 			error = kqueue_register(kq, kevp, td, 1);
848 			if (error || (kevp->flags & EV_RECEIPT)) {
849 				if (nevents != 0) {
850 					kevp->flags = EV_ERROR;
851 					kevp->data = error;
852 					(void) k_ops->k_copyout(k_ops->arg,
853 					    kevp, 1);
854 					nevents--;
855 					nerrors++;
856 				} else {
857 					goto done;
858 				}
859 			}
860 		}
861 		nchanges -= n;
862 	}
863 	if (nerrors) {
864 		td->td_retval[0] = nerrors;
865 		error = 0;
866 		goto done;
867 	}
868 
869 	error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
870 done:
871 	kqueue_release(kq, 0);
872 done_norel:
873 	fdrop(fp, td);
874 	return (error);
875 }
876 
877 int
878 kqueue_add_filteropts(int filt, struct filterops *filtops)
879 {
880 	int error;
881 
882 	error = 0;
883 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
884 		printf(
885 "trying to add a filterop that is out of range: %d is beyond %d\n",
886 		    ~filt, EVFILT_SYSCOUNT);
887 		return EINVAL;
888 	}
889 	mtx_lock(&filterops_lock);
890 	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
891 	    sysfilt_ops[~filt].for_fop != NULL)
892 		error = EEXIST;
893 	else {
894 		sysfilt_ops[~filt].for_fop = filtops;
895 		sysfilt_ops[~filt].for_refcnt = 0;
896 	}
897 	mtx_unlock(&filterops_lock);
898 
899 	return (error);
900 }
901 
902 int
903 kqueue_del_filteropts(int filt)
904 {
905 	int error;
906 
907 	error = 0;
908 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
909 		return EINVAL;
910 
911 	mtx_lock(&filterops_lock);
912 	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
913 	    sysfilt_ops[~filt].for_fop == NULL)
914 		error = EINVAL;
915 	else if (sysfilt_ops[~filt].for_refcnt != 0)
916 		error = EBUSY;
917 	else {
918 		sysfilt_ops[~filt].for_fop = &null_filtops;
919 		sysfilt_ops[~filt].for_refcnt = 0;
920 	}
921 	mtx_unlock(&filterops_lock);
922 
923 	return error;
924 }
925 
926 static struct filterops *
927 kqueue_fo_find(int filt)
928 {
929 
930 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
931 		return NULL;
932 
933 	mtx_lock(&filterops_lock);
934 	sysfilt_ops[~filt].for_refcnt++;
935 	if (sysfilt_ops[~filt].for_fop == NULL)
936 		sysfilt_ops[~filt].for_fop = &null_filtops;
937 	mtx_unlock(&filterops_lock);
938 
939 	return sysfilt_ops[~filt].for_fop;
940 }
941 
942 static void
943 kqueue_fo_release(int filt)
944 {
945 
946 	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
947 		return;
948 
949 	mtx_lock(&filterops_lock);
950 	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
951 	    ("filter object refcount not valid on release"));
952 	sysfilt_ops[~filt].for_refcnt--;
953 	mtx_unlock(&filterops_lock);
954 }
955 
956 /*
957  * A ref to kq (obtained via kqueue_acquire) must be held.  waitok will
958  * influence if memory allocation should wait.  Make sure it is 0 if you
959  * hold any mutexes.
960  */
961 static int
962 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
963 {
964 	struct filterops *fops;
965 	struct file *fp;
966 	struct knote *kn, *tkn;
967 	int error, filt, event;
968 	int haskqglobal;
969 
970 	fp = NULL;
971 	kn = NULL;
972 	error = 0;
973 	haskqglobal = 0;
974 
975 	filt = kev->filter;
976 	fops = kqueue_fo_find(filt);
977 	if (fops == NULL)
978 		return EINVAL;
979 
980 	tkn = knote_alloc(waitok);		/* prevent waiting with locks */
981 
982 findkn:
983 	if (fops->f_isfd) {
984 		KASSERT(td != NULL, ("td is NULL"));
985 		error = fget(td, kev->ident, CAP_POLL_EVENT, &fp);
986 		if (error)
987 			goto done;
988 
989 		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
990 		    kev->ident, 0) != 0) {
991 			/* try again */
992 			fdrop(fp, td);
993 			fp = NULL;
994 			error = kqueue_expand(kq, fops, kev->ident, waitok);
995 			if (error)
996 				goto done;
997 			goto findkn;
998 		}
999 
1000 		if (fp->f_type == DTYPE_KQUEUE) {
1001 			/*
1002 			 * if we add some inteligence about what we are doing,
1003 			 * we should be able to support events on ourselves.
1004 			 * We need to know when we are doing this to prevent
1005 			 * getting both the knlist lock and the kq lock since
1006 			 * they are the same thing.
1007 			 */
1008 			if (fp->f_data == kq) {
1009 				error = EINVAL;
1010 				goto done;
1011 			}
1012 
1013 			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1014 		}
1015 
1016 		KQ_LOCK(kq);
1017 		if (kev->ident < kq->kq_knlistsize) {
1018 			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1019 				if (kev->filter == kn->kn_filter)
1020 					break;
1021 		}
1022 	} else {
1023 		if ((kev->flags & EV_ADD) == EV_ADD)
1024 			kqueue_expand(kq, fops, kev->ident, waitok);
1025 
1026 		KQ_LOCK(kq);
1027 		if (kq->kq_knhashmask != 0) {
1028 			struct klist *list;
1029 
1030 			list = &kq->kq_knhash[
1031 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1032 			SLIST_FOREACH(kn, list, kn_link)
1033 				if (kev->ident == kn->kn_id &&
1034 				    kev->filter == kn->kn_filter)
1035 					break;
1036 		}
1037 	}
1038 
1039 	/* knote is in the process of changing, wait for it to stablize. */
1040 	if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1041 		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1042 		kq->kq_state |= KQ_FLUXWAIT;
1043 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1044 		if (fp != NULL) {
1045 			fdrop(fp, td);
1046 			fp = NULL;
1047 		}
1048 		goto findkn;
1049 	}
1050 
1051 	/*
1052 	 * kn now contains the matching knote, or NULL if no match
1053 	 */
1054 	if (kn == NULL) {
1055 		if (kev->flags & EV_ADD) {
1056 			kn = tkn;
1057 			tkn = NULL;
1058 			if (kn == NULL) {
1059 				KQ_UNLOCK(kq);
1060 				error = ENOMEM;
1061 				goto done;
1062 			}
1063 			kn->kn_fp = fp;
1064 			kn->kn_kq = kq;
1065 			kn->kn_fop = fops;
1066 			/*
1067 			 * apply reference counts to knote structure, and
1068 			 * do not release it at the end of this routine.
1069 			 */
1070 			fops = NULL;
1071 			fp = NULL;
1072 
1073 			kn->kn_sfflags = kev->fflags;
1074 			kn->kn_sdata = kev->data;
1075 			kev->fflags = 0;
1076 			kev->data = 0;
1077 			kn->kn_kevent = *kev;
1078 			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1079 			    EV_ENABLE | EV_DISABLE);
1080 			kn->kn_status = KN_INFLUX|KN_DETACHED;
1081 
1082 			error = knote_attach(kn, kq);
1083 			KQ_UNLOCK(kq);
1084 			if (error != 0) {
1085 				tkn = kn;
1086 				goto done;
1087 			}
1088 
1089 			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1090 				knote_drop(kn, td);
1091 				goto done;
1092 			}
1093 			KN_LIST_LOCK(kn);
1094 			goto done_ev_add;
1095 		} else {
1096 			/* No matching knote and the EV_ADD flag is not set. */
1097 			KQ_UNLOCK(kq);
1098 			error = ENOENT;
1099 			goto done;
1100 		}
1101 	}
1102 
1103 	if (kev->flags & EV_DELETE) {
1104 		kn->kn_status |= KN_INFLUX;
1105 		KQ_UNLOCK(kq);
1106 		if (!(kn->kn_status & KN_DETACHED))
1107 			kn->kn_fop->f_detach(kn);
1108 		knote_drop(kn, td);
1109 		goto done;
1110 	}
1111 
1112 	/*
1113 	 * The user may change some filter values after the initial EV_ADD,
1114 	 * but doing so will not reset any filter which has already been
1115 	 * triggered.
1116 	 */
1117 	kn->kn_status |= KN_INFLUX;
1118 	KQ_UNLOCK(kq);
1119 	KN_LIST_LOCK(kn);
1120 	kn->kn_kevent.udata = kev->udata;
1121 	if (!fops->f_isfd && fops->f_touch != NULL) {
1122 		fops->f_touch(kn, kev, EVENT_REGISTER);
1123 	} else {
1124 		kn->kn_sfflags = kev->fflags;
1125 		kn->kn_sdata = kev->data;
1126 	}
1127 
1128 	/*
1129 	 * We can get here with kn->kn_knlist == NULL.  This can happen when
1130 	 * the initial attach event decides that the event is "completed"
1131 	 * already.  i.e. filt_procattach is called on a zombie process.  It
1132 	 * will call filt_proc which will remove it from the list, and NULL
1133 	 * kn_knlist.
1134 	 */
1135 done_ev_add:
1136 	event = kn->kn_fop->f_event(kn, 0);
1137 	KQ_LOCK(kq);
1138 	if (event)
1139 		KNOTE_ACTIVATE(kn, 1);
1140 	kn->kn_status &= ~KN_INFLUX;
1141 	KN_LIST_UNLOCK(kn);
1142 
1143 	if ((kev->flags & EV_DISABLE) &&
1144 	    ((kn->kn_status & KN_DISABLED) == 0)) {
1145 		kn->kn_status |= KN_DISABLED;
1146 	}
1147 
1148 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1149 		kn->kn_status &= ~KN_DISABLED;
1150 		if ((kn->kn_status & KN_ACTIVE) &&
1151 		    ((kn->kn_status & KN_QUEUED) == 0))
1152 			knote_enqueue(kn);
1153 	}
1154 	KQ_UNLOCK_FLUX(kq);
1155 
1156 done:
1157 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1158 	if (fp != NULL)
1159 		fdrop(fp, td);
1160 	if (tkn != NULL)
1161 		knote_free(tkn);
1162 	if (fops != NULL)
1163 		kqueue_fo_release(filt);
1164 	return (error);
1165 }
1166 
1167 static int
1168 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1169 {
1170 	int error;
1171 	struct kqueue *kq;
1172 
1173 	error = 0;
1174 
1175 	kq = fp->f_data;
1176 	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1177 		return (EBADF);
1178 	*kqp = kq;
1179 	KQ_LOCK(kq);
1180 	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1181 		KQ_UNLOCK(kq);
1182 		return (EBADF);
1183 	}
1184 	kq->kq_refcnt++;
1185 	KQ_UNLOCK(kq);
1186 
1187 	return error;
1188 }
1189 
1190 static void
1191 kqueue_release(struct kqueue *kq, int locked)
1192 {
1193 	if (locked)
1194 		KQ_OWNED(kq);
1195 	else
1196 		KQ_LOCK(kq);
1197 	kq->kq_refcnt--;
1198 	if (kq->kq_refcnt == 1)
1199 		wakeup(&kq->kq_refcnt);
1200 	if (!locked)
1201 		KQ_UNLOCK(kq);
1202 }
1203 
1204 static void
1205 kqueue_schedtask(struct kqueue *kq)
1206 {
1207 
1208 	KQ_OWNED(kq);
1209 	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1210 	    ("scheduling kqueue task while draining"));
1211 
1212 	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1213 		taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1214 		kq->kq_state |= KQ_TASKSCHED;
1215 	}
1216 }
1217 
1218 /*
1219  * Expand the kq to make sure we have storage for fops/ident pair.
1220  *
1221  * Return 0 on success (or no work necessary), return errno on failure.
1222  *
1223  * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1224  * If kqueue_register is called from a non-fd context, there usually/should
1225  * be no locks held.
1226  */
1227 static int
1228 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1229 	int waitok)
1230 {
1231 	struct klist *list, *tmp_knhash, *to_free;
1232 	u_long tmp_knhashmask;
1233 	int size;
1234 	int fd;
1235 	int mflag = waitok ? M_WAITOK : M_NOWAIT;
1236 
1237 	KQ_NOTOWNED(kq);
1238 
1239 	to_free = NULL;
1240 	if (fops->f_isfd) {
1241 		fd = ident;
1242 		if (kq->kq_knlistsize <= fd) {
1243 			size = kq->kq_knlistsize;
1244 			while (size <= fd)
1245 				size += KQEXTENT;
1246 			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1247 			if (list == NULL)
1248 				return ENOMEM;
1249 			KQ_LOCK(kq);
1250 			if (kq->kq_knlistsize > fd) {
1251 				to_free = list;
1252 				list = NULL;
1253 			} else {
1254 				if (kq->kq_knlist != NULL) {
1255 					bcopy(kq->kq_knlist, list,
1256 					    kq->kq_knlistsize * sizeof(*list));
1257 					to_free = kq->kq_knlist;
1258 					kq->kq_knlist = NULL;
1259 				}
1260 				bzero((caddr_t)list +
1261 				    kq->kq_knlistsize * sizeof(*list),
1262 				    (size - kq->kq_knlistsize) * sizeof(*list));
1263 				kq->kq_knlistsize = size;
1264 				kq->kq_knlist = list;
1265 			}
1266 			KQ_UNLOCK(kq);
1267 		}
1268 	} else {
1269 		if (kq->kq_knhashmask == 0) {
1270 			tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1271 			    &tmp_knhashmask);
1272 			if (tmp_knhash == NULL)
1273 				return ENOMEM;
1274 			KQ_LOCK(kq);
1275 			if (kq->kq_knhashmask == 0) {
1276 				kq->kq_knhash = tmp_knhash;
1277 				kq->kq_knhashmask = tmp_knhashmask;
1278 			} else {
1279 				to_free = tmp_knhash;
1280 			}
1281 			KQ_UNLOCK(kq);
1282 		}
1283 	}
1284 	free(to_free, M_KQUEUE);
1285 
1286 	KQ_NOTOWNED(kq);
1287 	return 0;
1288 }
1289 
1290 static void
1291 kqueue_task(void *arg, int pending)
1292 {
1293 	struct kqueue *kq;
1294 	int haskqglobal;
1295 
1296 	haskqglobal = 0;
1297 	kq = arg;
1298 
1299 	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1300 	KQ_LOCK(kq);
1301 
1302 	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1303 
1304 	kq->kq_state &= ~KQ_TASKSCHED;
1305 	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1306 		wakeup(&kq->kq_state);
1307 	}
1308 	KQ_UNLOCK(kq);
1309 	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1310 }
1311 
1312 /*
1313  * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1314  * We treat KN_MARKER knotes as if they are INFLUX.
1315  */
1316 static int
1317 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1318     const struct timespec *tsp, struct kevent *keva, struct thread *td)
1319 {
1320 	struct kevent *kevp;
1321 	struct knote *kn, *marker;
1322 	sbintime_t asbt, rsbt;
1323 	int count, error, haskqglobal, influx, nkev, touch;
1324 
1325 	count = maxevents;
1326 	nkev = 0;
1327 	error = 0;
1328 	haskqglobal = 0;
1329 
1330 	if (maxevents == 0)
1331 		goto done_nl;
1332 
1333 	rsbt = 0;
1334 	if (tsp != NULL) {
1335 		if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 ||
1336 		    tsp->tv_nsec >= 1000000000) {
1337 			error = EINVAL;
1338 			goto done_nl;
1339 		}
1340 		if (timespecisset(tsp)) {
1341 			if (tsp->tv_sec <= INT32_MAX) {
1342 				rsbt = tstosbt(*tsp);
1343 				if (TIMESEL(&asbt, rsbt))
1344 					asbt += tc_tick_sbt;
1345 				if (asbt <= INT64_MAX - rsbt)
1346 					asbt += rsbt;
1347 				else
1348 					asbt = 0;
1349 				rsbt >>= tc_precexp;
1350 			} else
1351 				asbt = 0;
1352 		} else
1353 			asbt = -1;
1354 	} else
1355 		asbt = 0;
1356 	marker = knote_alloc(1);
1357 	if (marker == NULL) {
1358 		error = ENOMEM;
1359 		goto done_nl;
1360 	}
1361 	marker->kn_status = KN_MARKER;
1362 	KQ_LOCK(kq);
1363 
1364 retry:
1365 	kevp = keva;
1366 	if (kq->kq_count == 0) {
1367 		if (asbt == -1) {
1368 			error = EWOULDBLOCK;
1369 		} else {
1370 			kq->kq_state |= KQ_SLEEP;
1371 			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1372 			    "kqread", asbt, rsbt, C_ABSOLUTE);
1373 		}
1374 		if (error == 0)
1375 			goto retry;
1376 		/* don't restart after signals... */
1377 		if (error == ERESTART)
1378 			error = EINTR;
1379 		else if (error == EWOULDBLOCK)
1380 			error = 0;
1381 		goto done;
1382 	}
1383 
1384 	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1385 	influx = 0;
1386 	while (count) {
1387 		KQ_OWNED(kq);
1388 		kn = TAILQ_FIRST(&kq->kq_head);
1389 
1390 		if ((kn->kn_status == KN_MARKER && kn != marker) ||
1391 		    (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1392 			if (influx) {
1393 				influx = 0;
1394 				KQ_FLUX_WAKEUP(kq);
1395 			}
1396 			kq->kq_state |= KQ_FLUXWAIT;
1397 			error = msleep(kq, &kq->kq_lock, PSOCK,
1398 			    "kqflxwt", 0);
1399 			continue;
1400 		}
1401 
1402 		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1403 		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1404 			kn->kn_status &= ~KN_QUEUED;
1405 			kq->kq_count--;
1406 			continue;
1407 		}
1408 		if (kn == marker) {
1409 			KQ_FLUX_WAKEUP(kq);
1410 			if (count == maxevents)
1411 				goto retry;
1412 			goto done;
1413 		}
1414 		KASSERT((kn->kn_status & KN_INFLUX) == 0,
1415 		    ("KN_INFLUX set when not suppose to be"));
1416 
1417 		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1418 			kn->kn_status &= ~KN_QUEUED;
1419 			kn->kn_status |= KN_INFLUX;
1420 			kq->kq_count--;
1421 			KQ_UNLOCK(kq);
1422 			/*
1423 			 * We don't need to lock the list since we've marked
1424 			 * it _INFLUX.
1425 			 */
1426 			if (!(kn->kn_status & KN_DETACHED))
1427 				kn->kn_fop->f_detach(kn);
1428 			knote_drop(kn, td);
1429 			KQ_LOCK(kq);
1430 			continue;
1431 		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1432 			kn->kn_status &= ~KN_QUEUED;
1433 			kn->kn_status |= KN_INFLUX;
1434 			kq->kq_count--;
1435 			KQ_UNLOCK(kq);
1436 			/*
1437 			 * We don't need to lock the list since we've marked
1438 			 * it _INFLUX.
1439 			 */
1440 			*kevp = kn->kn_kevent;
1441 			if (!(kn->kn_status & KN_DETACHED))
1442 				kn->kn_fop->f_detach(kn);
1443 			knote_drop(kn, td);
1444 			KQ_LOCK(kq);
1445 			kn = NULL;
1446 		} else {
1447 			kn->kn_status |= KN_INFLUX;
1448 			KQ_UNLOCK(kq);
1449 			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1450 				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1451 			KN_LIST_LOCK(kn);
1452 			if (kn->kn_fop->f_event(kn, 0) == 0) {
1453 				KQ_LOCK(kq);
1454 				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1455 				kn->kn_status &=
1456 				    ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX);
1457 				kq->kq_count--;
1458 				KN_LIST_UNLOCK(kn);
1459 				influx = 1;
1460 				continue;
1461 			}
1462 			touch = (!kn->kn_fop->f_isfd &&
1463 			    kn->kn_fop->f_touch != NULL);
1464 			if (touch)
1465 				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1466 			else
1467 				*kevp = kn->kn_kevent;
1468 			KQ_LOCK(kq);
1469 			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1470 			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1471 				/*
1472 				 * Manually clear knotes who weren't
1473 				 * 'touch'ed.
1474 				 */
1475 				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1476 					kn->kn_data = 0;
1477 					kn->kn_fflags = 0;
1478 				}
1479 				if (kn->kn_flags & EV_DISPATCH)
1480 					kn->kn_status |= KN_DISABLED;
1481 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1482 				kq->kq_count--;
1483 			} else
1484 				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1485 
1486 			kn->kn_status &= ~(KN_INFLUX);
1487 			KN_LIST_UNLOCK(kn);
1488 			influx = 1;
1489 		}
1490 
1491 		/* we are returning a copy to the user */
1492 		kevp++;
1493 		nkev++;
1494 		count--;
1495 
1496 		if (nkev == KQ_NEVENTS) {
1497 			influx = 0;
1498 			KQ_UNLOCK_FLUX(kq);
1499 			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1500 			nkev = 0;
1501 			kevp = keva;
1502 			KQ_LOCK(kq);
1503 			if (error)
1504 				break;
1505 		}
1506 	}
1507 	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1508 done:
1509 	KQ_OWNED(kq);
1510 	KQ_UNLOCK_FLUX(kq);
1511 	knote_free(marker);
1512 done_nl:
1513 	KQ_NOTOWNED(kq);
1514 	if (nkev != 0)
1515 		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1516 	td->td_retval[0] = maxevents - count;
1517 	return (error);
1518 }
1519 
1520 /*
1521  * XXX
1522  * This could be expanded to call kqueue_scan, if desired.
1523  */
1524 /*ARGSUSED*/
1525 static int
1526 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1527 	int flags, struct thread *td)
1528 {
1529 	return (ENXIO);
1530 }
1531 
1532 /*ARGSUSED*/
1533 static int
1534 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1535 	 int flags, struct thread *td)
1536 {
1537 	return (ENXIO);
1538 }
1539 
1540 /*ARGSUSED*/
1541 static int
1542 kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1543 	struct thread *td)
1544 {
1545 
1546 	return (EINVAL);
1547 }
1548 
1549 /*ARGSUSED*/
1550 static int
1551 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1552 	struct ucred *active_cred, struct thread *td)
1553 {
1554 	/*
1555 	 * Enabling sigio causes two major problems:
1556 	 * 1) infinite recursion:
1557 	 * Synopsys: kevent is being used to track signals and have FIOASYNC
1558 	 * set.  On receipt of a signal this will cause a kqueue to recurse
1559 	 * into itself over and over.  Sending the sigio causes the kqueue
1560 	 * to become ready, which in turn posts sigio again, forever.
1561 	 * Solution: this can be solved by setting a flag in the kqueue that
1562 	 * we have a SIGIO in progress.
1563 	 * 2) locking problems:
1564 	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1565 	 * us above the proc and pgrp locks.
1566 	 * Solution: Post a signal using an async mechanism, being sure to
1567 	 * record a generation count in the delivery so that we do not deliver
1568 	 * a signal to the wrong process.
1569 	 *
1570 	 * Note, these two mechanisms are somewhat mutually exclusive!
1571 	 */
1572 #if 0
1573 	struct kqueue *kq;
1574 
1575 	kq = fp->f_data;
1576 	switch (cmd) {
1577 	case FIOASYNC:
1578 		if (*(int *)data) {
1579 			kq->kq_state |= KQ_ASYNC;
1580 		} else {
1581 			kq->kq_state &= ~KQ_ASYNC;
1582 		}
1583 		return (0);
1584 
1585 	case FIOSETOWN:
1586 		return (fsetown(*(int *)data, &kq->kq_sigio));
1587 
1588 	case FIOGETOWN:
1589 		*(int *)data = fgetown(&kq->kq_sigio);
1590 		return (0);
1591 	}
1592 #endif
1593 
1594 	return (ENOTTY);
1595 }
1596 
1597 /*ARGSUSED*/
1598 static int
1599 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1600 	struct thread *td)
1601 {
1602 	struct kqueue *kq;
1603 	int revents = 0;
1604 	int error;
1605 
1606 	if ((error = kqueue_acquire(fp, &kq)))
1607 		return POLLERR;
1608 
1609 	KQ_LOCK(kq);
1610 	if (events & (POLLIN | POLLRDNORM)) {
1611 		if (kq->kq_count) {
1612 			revents |= events & (POLLIN | POLLRDNORM);
1613 		} else {
1614 			selrecord(td, &kq->kq_sel);
1615 			if (SEL_WAITING(&kq->kq_sel))
1616 				kq->kq_state |= KQ_SEL;
1617 		}
1618 	}
1619 	kqueue_release(kq, 1);
1620 	KQ_UNLOCK(kq);
1621 	return (revents);
1622 }
1623 
1624 /*ARGSUSED*/
1625 static int
1626 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1627 	struct thread *td)
1628 {
1629 
1630 	bzero((void *)st, sizeof *st);
1631 	/*
1632 	 * We no longer return kq_count because the unlocked value is useless.
1633 	 * If you spent all this time getting the count, why not spend your
1634 	 * syscall better by calling kevent?
1635 	 *
1636 	 * XXX - This is needed for libc_r.
1637 	 */
1638 	st->st_mode = S_IFIFO;
1639 	return (0);
1640 }
1641 
1642 /*ARGSUSED*/
1643 static int
1644 kqueue_close(struct file *fp, struct thread *td)
1645 {
1646 	struct kqueue *kq = fp->f_data;
1647 	struct filedesc *fdp;
1648 	struct knote *kn;
1649 	int i;
1650 	int error;
1651 
1652 	if ((error = kqueue_acquire(fp, &kq)))
1653 		return error;
1654 
1655 	KQ_LOCK(kq);
1656 
1657 	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1658 	    ("kqueue already closing"));
1659 	kq->kq_state |= KQ_CLOSING;
1660 	if (kq->kq_refcnt > 1)
1661 		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1662 
1663 	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1664 	fdp = kq->kq_fdp;
1665 
1666 	KASSERT(knlist_empty(&kq->kq_sel.si_note),
1667 	    ("kqueue's knlist not empty"));
1668 
1669 	for (i = 0; i < kq->kq_knlistsize; i++) {
1670 		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1671 			if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1672 				kq->kq_state |= KQ_FLUXWAIT;
1673 				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1674 				continue;
1675 			}
1676 			kn->kn_status |= KN_INFLUX;
1677 			KQ_UNLOCK(kq);
1678 			if (!(kn->kn_status & KN_DETACHED))
1679 				kn->kn_fop->f_detach(kn);
1680 			knote_drop(kn, td);
1681 			KQ_LOCK(kq);
1682 		}
1683 	}
1684 	if (kq->kq_knhashmask != 0) {
1685 		for (i = 0; i <= kq->kq_knhashmask; i++) {
1686 			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1687 				if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1688 					kq->kq_state |= KQ_FLUXWAIT;
1689 					msleep(kq, &kq->kq_lock, PSOCK,
1690 					       "kqclo2", 0);
1691 					continue;
1692 				}
1693 				kn->kn_status |= KN_INFLUX;
1694 				KQ_UNLOCK(kq);
1695 				if (!(kn->kn_status & KN_DETACHED))
1696 					kn->kn_fop->f_detach(kn);
1697 				knote_drop(kn, td);
1698 				KQ_LOCK(kq);
1699 			}
1700 		}
1701 	}
1702 
1703 	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1704 		kq->kq_state |= KQ_TASKDRAIN;
1705 		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1706 	}
1707 
1708 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1709 		selwakeuppri(&kq->kq_sel, PSOCK);
1710 		if (!SEL_WAITING(&kq->kq_sel))
1711 			kq->kq_state &= ~KQ_SEL;
1712 	}
1713 
1714 	KQ_UNLOCK(kq);
1715 
1716 	FILEDESC_XLOCK(fdp);
1717 	SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list);
1718 	FILEDESC_XUNLOCK(fdp);
1719 
1720 	seldrain(&kq->kq_sel);
1721 	knlist_destroy(&kq->kq_sel.si_note);
1722 	mtx_destroy(&kq->kq_lock);
1723 	kq->kq_fdp = NULL;
1724 
1725 	if (kq->kq_knhash != NULL)
1726 		free(kq->kq_knhash, M_KQUEUE);
1727 	if (kq->kq_knlist != NULL)
1728 		free(kq->kq_knlist, M_KQUEUE);
1729 
1730 	funsetown(&kq->kq_sigio);
1731 	free(kq, M_KQUEUE);
1732 	fp->f_data = NULL;
1733 
1734 	return (0);
1735 }
1736 
1737 static void
1738 kqueue_wakeup(struct kqueue *kq)
1739 {
1740 	KQ_OWNED(kq);
1741 
1742 	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1743 		kq->kq_state &= ~KQ_SLEEP;
1744 		wakeup(kq);
1745 	}
1746 	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1747 		selwakeuppri(&kq->kq_sel, PSOCK);
1748 		if (!SEL_WAITING(&kq->kq_sel))
1749 			kq->kq_state &= ~KQ_SEL;
1750 	}
1751 	if (!knlist_empty(&kq->kq_sel.si_note))
1752 		kqueue_schedtask(kq);
1753 	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1754 		pgsigio(&kq->kq_sigio, SIGIO, 0);
1755 	}
1756 }
1757 
1758 /*
1759  * Walk down a list of knotes, activating them if their event has triggered.
1760  *
1761  * There is a possibility to optimize in the case of one kq watching another.
1762  * Instead of scheduling a task to wake it up, you could pass enough state
1763  * down the chain to make up the parent kqueue.  Make this code functional
1764  * first.
1765  */
1766 void
1767 knote(struct knlist *list, long hint, int lockflags)
1768 {
1769 	struct kqueue *kq;
1770 	struct knote *kn;
1771 	int error;
1772 
1773 	if (list == NULL)
1774 		return;
1775 
1776 	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1777 
1778 	if ((lockflags & KNF_LISTLOCKED) == 0)
1779 		list->kl_lock(list->kl_lockarg);
1780 
1781 	/*
1782 	 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1783 	 * the kqueue scheduling, but this will introduce four
1784 	 * lock/unlock's for each knote to test.  If we do, continue to use
1785 	 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1786 	 * only safe if you want to remove the current item, which we are
1787 	 * not doing.
1788 	 */
1789 	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1790 		kq = kn->kn_kq;
1791 		if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1792 			KQ_LOCK(kq);
1793 			if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1794 				KQ_UNLOCK(kq);
1795 			} else if ((lockflags & KNF_NOKQLOCK) != 0) {
1796 				kn->kn_status |= KN_INFLUX;
1797 				KQ_UNLOCK(kq);
1798 				error = kn->kn_fop->f_event(kn, hint);
1799 				KQ_LOCK(kq);
1800 				kn->kn_status &= ~KN_INFLUX;
1801 				if (error)
1802 					KNOTE_ACTIVATE(kn, 1);
1803 				KQ_UNLOCK_FLUX(kq);
1804 			} else {
1805 				kn->kn_status |= KN_HASKQLOCK;
1806 				if (kn->kn_fop->f_event(kn, hint))
1807 					KNOTE_ACTIVATE(kn, 1);
1808 				kn->kn_status &= ~KN_HASKQLOCK;
1809 				KQ_UNLOCK(kq);
1810 			}
1811 		}
1812 		kq = NULL;
1813 	}
1814 	if ((lockflags & KNF_LISTLOCKED) == 0)
1815 		list->kl_unlock(list->kl_lockarg);
1816 }
1817 
1818 /*
1819  * add a knote to a knlist
1820  */
1821 void
1822 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1823 {
1824 	KNL_ASSERT_LOCK(knl, islocked);
1825 	KQ_NOTOWNED(kn->kn_kq);
1826 	KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1827 	    (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1828 	if (!islocked)
1829 		knl->kl_lock(knl->kl_lockarg);
1830 	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1831 	if (!islocked)
1832 		knl->kl_unlock(knl->kl_lockarg);
1833 	KQ_LOCK(kn->kn_kq);
1834 	kn->kn_knlist = knl;
1835 	kn->kn_status &= ~KN_DETACHED;
1836 	KQ_UNLOCK(kn->kn_kq);
1837 }
1838 
1839 static void
1840 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1841 {
1842 	KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1843 	KNL_ASSERT_LOCK(knl, knlislocked);
1844 	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1845 	if (!kqislocked)
1846 		KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1847     ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1848 	if (!knlislocked)
1849 		knl->kl_lock(knl->kl_lockarg);
1850 	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1851 	kn->kn_knlist = NULL;
1852 	if (!knlislocked)
1853 		knl->kl_unlock(knl->kl_lockarg);
1854 	if (!kqislocked)
1855 		KQ_LOCK(kn->kn_kq);
1856 	kn->kn_status |= KN_DETACHED;
1857 	if (!kqislocked)
1858 		KQ_UNLOCK(kn->kn_kq);
1859 }
1860 
1861 /*
1862  * remove knote from the specified knlist
1863  */
1864 void
1865 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1866 {
1867 
1868 	knlist_remove_kq(knl, kn, islocked, 0);
1869 }
1870 
1871 /*
1872  * remove knote from the specified knlist while in f_event handler.
1873  */
1874 void
1875 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1876 {
1877 
1878 	knlist_remove_kq(knl, kn, 1,
1879 	    (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1880 }
1881 
1882 int
1883 knlist_empty(struct knlist *knl)
1884 {
1885 
1886 	KNL_ASSERT_LOCKED(knl);
1887 	return SLIST_EMPTY(&knl->kl_list);
1888 }
1889 
1890 static struct mtx	knlist_lock;
1891 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1892 	MTX_DEF);
1893 static void knlist_mtx_lock(void *arg);
1894 static void knlist_mtx_unlock(void *arg);
1895 
1896 static void
1897 knlist_mtx_lock(void *arg)
1898 {
1899 
1900 	mtx_lock((struct mtx *)arg);
1901 }
1902 
1903 static void
1904 knlist_mtx_unlock(void *arg)
1905 {
1906 
1907 	mtx_unlock((struct mtx *)arg);
1908 }
1909 
1910 static void
1911 knlist_mtx_assert_locked(void *arg)
1912 {
1913 
1914 	mtx_assert((struct mtx *)arg, MA_OWNED);
1915 }
1916 
1917 static void
1918 knlist_mtx_assert_unlocked(void *arg)
1919 {
1920 
1921 	mtx_assert((struct mtx *)arg, MA_NOTOWNED);
1922 }
1923 
1924 static void
1925 knlist_rw_rlock(void *arg)
1926 {
1927 
1928 	rw_rlock((struct rwlock *)arg);
1929 }
1930 
1931 static void
1932 knlist_rw_runlock(void *arg)
1933 {
1934 
1935 	rw_runlock((struct rwlock *)arg);
1936 }
1937 
1938 static void
1939 knlist_rw_assert_locked(void *arg)
1940 {
1941 
1942 	rw_assert((struct rwlock *)arg, RA_LOCKED);
1943 }
1944 
1945 static void
1946 knlist_rw_assert_unlocked(void *arg)
1947 {
1948 
1949 	rw_assert((struct rwlock *)arg, RA_UNLOCKED);
1950 }
1951 
1952 void
1953 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1954     void (*kl_unlock)(void *),
1955     void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
1956 {
1957 
1958 	if (lock == NULL)
1959 		knl->kl_lockarg = &knlist_lock;
1960 	else
1961 		knl->kl_lockarg = lock;
1962 
1963 	if (kl_lock == NULL)
1964 		knl->kl_lock = knlist_mtx_lock;
1965 	else
1966 		knl->kl_lock = kl_lock;
1967 	if (kl_unlock == NULL)
1968 		knl->kl_unlock = knlist_mtx_unlock;
1969 	else
1970 		knl->kl_unlock = kl_unlock;
1971 	if (kl_assert_locked == NULL)
1972 		knl->kl_assert_locked = knlist_mtx_assert_locked;
1973 	else
1974 		knl->kl_assert_locked = kl_assert_locked;
1975 	if (kl_assert_unlocked == NULL)
1976 		knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
1977 	else
1978 		knl->kl_assert_unlocked = kl_assert_unlocked;
1979 
1980 	SLIST_INIT(&knl->kl_list);
1981 }
1982 
1983 void
1984 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
1985 {
1986 
1987 	knlist_init(knl, lock, NULL, NULL, NULL, NULL);
1988 }
1989 
1990 void
1991 knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock)
1992 {
1993 
1994 	knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock,
1995 	    knlist_rw_assert_locked, knlist_rw_assert_unlocked);
1996 }
1997 
1998 void
1999 knlist_destroy(struct knlist *knl)
2000 {
2001 
2002 #ifdef INVARIANTS
2003 	/*
2004 	 * if we run across this error, we need to find the offending
2005 	 * driver and have it call knlist_clear or knlist_delete.
2006 	 */
2007 	if (!SLIST_EMPTY(&knl->kl_list))
2008 		printf("WARNING: destroying knlist w/ knotes on it!\n");
2009 #endif
2010 
2011 	knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
2012 	SLIST_INIT(&knl->kl_list);
2013 }
2014 
2015 /*
2016  * Even if we are locked, we may need to drop the lock to allow any influx
2017  * knotes time to "settle".
2018  */
2019 void
2020 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2021 {
2022 	struct knote *kn, *kn2;
2023 	struct kqueue *kq;
2024 
2025 	if (islocked)
2026 		KNL_ASSERT_LOCKED(knl);
2027 	else {
2028 		KNL_ASSERT_UNLOCKED(knl);
2029 again:		/* need to reacquire lock since we have dropped it */
2030 		knl->kl_lock(knl->kl_lockarg);
2031 	}
2032 
2033 	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2034 		kq = kn->kn_kq;
2035 		KQ_LOCK(kq);
2036 		if ((kn->kn_status & KN_INFLUX)) {
2037 			KQ_UNLOCK(kq);
2038 			continue;
2039 		}
2040 		knlist_remove_kq(knl, kn, 1, 1);
2041 		if (killkn) {
2042 			kn->kn_status |= KN_INFLUX | KN_DETACHED;
2043 			KQ_UNLOCK(kq);
2044 			knote_drop(kn, td);
2045 		} else {
2046 			/* Make sure cleared knotes disappear soon */
2047 			kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2048 			KQ_UNLOCK(kq);
2049 		}
2050 		kq = NULL;
2051 	}
2052 
2053 	if (!SLIST_EMPTY(&knl->kl_list)) {
2054 		/* there are still KN_INFLUX remaining */
2055 		kn = SLIST_FIRST(&knl->kl_list);
2056 		kq = kn->kn_kq;
2057 		KQ_LOCK(kq);
2058 		KASSERT(kn->kn_status & KN_INFLUX,
2059 		    ("knote removed w/o list lock"));
2060 		knl->kl_unlock(knl->kl_lockarg);
2061 		kq->kq_state |= KQ_FLUXWAIT;
2062 		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2063 		kq = NULL;
2064 		goto again;
2065 	}
2066 
2067 	if (islocked)
2068 		KNL_ASSERT_LOCKED(knl);
2069 	else {
2070 		knl->kl_unlock(knl->kl_lockarg);
2071 		KNL_ASSERT_UNLOCKED(knl);
2072 	}
2073 }
2074 
2075 /*
2076  * Remove all knotes referencing a specified fd must be called with FILEDESC
2077  * lock.  This prevents a race where a new fd comes along and occupies the
2078  * entry and we attach a knote to the fd.
2079  */
2080 void
2081 knote_fdclose(struct thread *td, int fd)
2082 {
2083 	struct filedesc *fdp = td->td_proc->p_fd;
2084 	struct kqueue *kq;
2085 	struct knote *kn;
2086 	int influx;
2087 
2088 	FILEDESC_XLOCK_ASSERT(fdp);
2089 
2090 	/*
2091 	 * We shouldn't have to worry about new kevents appearing on fd
2092 	 * since filedesc is locked.
2093 	 */
2094 	SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2095 		KQ_LOCK(kq);
2096 
2097 again:
2098 		influx = 0;
2099 		while (kq->kq_knlistsize > fd &&
2100 		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2101 			if (kn->kn_status & KN_INFLUX) {
2102 				/* someone else might be waiting on our knote */
2103 				if (influx)
2104 					wakeup(kq);
2105 				kq->kq_state |= KQ_FLUXWAIT;
2106 				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2107 				goto again;
2108 			}
2109 			kn->kn_status |= KN_INFLUX;
2110 			KQ_UNLOCK(kq);
2111 			if (!(kn->kn_status & KN_DETACHED))
2112 				kn->kn_fop->f_detach(kn);
2113 			knote_drop(kn, td);
2114 			influx = 1;
2115 			KQ_LOCK(kq);
2116 		}
2117 		KQ_UNLOCK_FLUX(kq);
2118 	}
2119 }
2120 
2121 static int
2122 knote_attach(struct knote *kn, struct kqueue *kq)
2123 {
2124 	struct klist *list;
2125 
2126 	KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2127 	KQ_OWNED(kq);
2128 
2129 	if (kn->kn_fop->f_isfd) {
2130 		if (kn->kn_id >= kq->kq_knlistsize)
2131 			return ENOMEM;
2132 		list = &kq->kq_knlist[kn->kn_id];
2133 	} else {
2134 		if (kq->kq_knhash == NULL)
2135 			return ENOMEM;
2136 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2137 	}
2138 
2139 	SLIST_INSERT_HEAD(list, kn, kn_link);
2140 
2141 	return 0;
2142 }
2143 
2144 /*
2145  * knote must already have been detached using the f_detach method.
2146  * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2147  * to prevent other removal.
2148  */
2149 static void
2150 knote_drop(struct knote *kn, struct thread *td)
2151 {
2152 	struct kqueue *kq;
2153 	struct klist *list;
2154 
2155 	kq = kn->kn_kq;
2156 
2157 	KQ_NOTOWNED(kq);
2158 	KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2159 	    ("knote_drop called without KN_INFLUX set in kn_status"));
2160 
2161 	KQ_LOCK(kq);
2162 	if (kn->kn_fop->f_isfd)
2163 		list = &kq->kq_knlist[kn->kn_id];
2164 	else
2165 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2166 
2167 	if (!SLIST_EMPTY(list))
2168 		SLIST_REMOVE(list, kn, knote, kn_link);
2169 	if (kn->kn_status & KN_QUEUED)
2170 		knote_dequeue(kn);
2171 	KQ_UNLOCK_FLUX(kq);
2172 
2173 	if (kn->kn_fop->f_isfd) {
2174 		fdrop(kn->kn_fp, td);
2175 		kn->kn_fp = NULL;
2176 	}
2177 	kqueue_fo_release(kn->kn_kevent.filter);
2178 	kn->kn_fop = NULL;
2179 	knote_free(kn);
2180 }
2181 
2182 static void
2183 knote_enqueue(struct knote *kn)
2184 {
2185 	struct kqueue *kq = kn->kn_kq;
2186 
2187 	KQ_OWNED(kn->kn_kq);
2188 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2189 
2190 	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2191 	kn->kn_status |= KN_QUEUED;
2192 	kq->kq_count++;
2193 	kqueue_wakeup(kq);
2194 }
2195 
2196 static void
2197 knote_dequeue(struct knote *kn)
2198 {
2199 	struct kqueue *kq = kn->kn_kq;
2200 
2201 	KQ_OWNED(kn->kn_kq);
2202 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2203 
2204 	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2205 	kn->kn_status &= ~KN_QUEUED;
2206 	kq->kq_count--;
2207 }
2208 
2209 static void
2210 knote_init(void)
2211 {
2212 
2213 	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2214 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2215 }
2216 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2217 
2218 static struct knote *
2219 knote_alloc(int waitok)
2220 {
2221 	return ((struct knote *)uma_zalloc(knote_zone,
2222 	    (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2223 }
2224 
2225 static void
2226 knote_free(struct knote *kn)
2227 {
2228 	if (kn != NULL)
2229 		uma_zfree(knote_zone, kn);
2230 }
2231 
2232 /*
2233  * Register the kev w/ the kq specified by fd.
2234  */
2235 int
2236 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2237 {
2238 	struct kqueue *kq;
2239 	struct file *fp;
2240 	int error;
2241 
2242 	if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0)
2243 		return (error);
2244 	if ((error = kqueue_acquire(fp, &kq)) != 0)
2245 		goto noacquire;
2246 
2247 	error = kqueue_register(kq, kev, td, waitok);
2248 
2249 	kqueue_release(kq, 0);
2250 
2251 noacquire:
2252 	fdrop(fp, td);
2253 
2254 	return error;
2255 }
2256