xref: /dflybsd-src/sys/kern/kern_event.c (revision a1fa5d8d095e6aa52c4a9ac5d2ab52b03fee54b9)
1 /*-
2  * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD: src/sys/kern/kern_event.c,v 1.2.2.10 2004/04/04 07:03:14 cperciva Exp $
27  * $DragonFly: src/sys/kern/kern_event.c,v 1.33 2007/02/03 17:05:57 corecode Exp $
28  */
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/proc.h>
34 #include <sys/malloc.h>
35 #include <sys/unistd.h>
36 #include <sys/file.h>
37 #include <sys/lock.h>
38 #include <sys/fcntl.h>
39 #include <sys/queue.h>
40 #include <sys/event.h>
41 #include <sys/eventvar.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/stat.h>
46 #include <sys/sysctl.h>
47 #include <sys/sysproto.h>
48 #include <sys/thread.h>
49 #include <sys/uio.h>
50 #include <sys/signalvar.h>
51 #include <sys/filio.h>
52 #include <sys/ktr.h>
53 
54 #include <sys/thread2.h>
55 #include <sys/file2.h>
56 #include <sys/mplock2.h>
57 
58 /*
59  * Global token for kqueue subsystem
60  */
61 struct lwkt_token kq_token = LWKT_TOKEN_INITIALIZER(kq_token);
62 SYSCTL_LONG(_lwkt, OID_AUTO, kq_collisions,
63     CTLFLAG_RW, &kq_token.t_collisions, 0,
64     "Collision counter of kq_token");
65 
66 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
67 
68 struct kevent_copyin_args {
69 	struct kevent_args	*ka;
70 	int			pchanges;
71 };
72 
73 static int	kqueue_sleep(struct kqueue *kq, struct timespec *tsp);
74 static int	kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count,
75 		    struct knote *marker);
76 static int 	kqueue_read(struct file *fp, struct uio *uio,
77 		    struct ucred *cred, int flags);
78 static int	kqueue_write(struct file *fp, struct uio *uio,
79 		    struct ucred *cred, int flags);
80 static int	kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
81 		    struct ucred *cred, struct sysmsg *msg);
82 static int 	kqueue_kqfilter(struct file *fp, struct knote *kn);
83 static int 	kqueue_stat(struct file *fp, struct stat *st,
84 		    struct ucred *cred);
85 static int 	kqueue_close(struct file *fp);
86 static void	kqueue_wakeup(struct kqueue *kq);
87 static int	filter_attach(struct knote *kn);
88 static int	filter_event(struct knote *kn, long hint);
89 
90 /*
91  * MPSAFE
92  */
93 static struct fileops kqueueops = {
94 	.fo_read = kqueue_read,
95 	.fo_write = kqueue_write,
96 	.fo_ioctl = kqueue_ioctl,
97 	.fo_kqfilter = kqueue_kqfilter,
98 	.fo_stat = kqueue_stat,
99 	.fo_close = kqueue_close,
100 	.fo_shutdown = nofo_shutdown
101 };
102 
103 static void 	knote_attach(struct knote *kn);
104 static void 	knote_drop(struct knote *kn);
105 static void	knote_detach_and_drop(struct knote *kn);
106 static void 	knote_enqueue(struct knote *kn);
107 static void 	knote_dequeue(struct knote *kn);
108 static struct 	knote *knote_alloc(void);
109 static void 	knote_free(struct knote *kn);
110 
111 static void	filt_kqdetach(struct knote *kn);
112 static int	filt_kqueue(struct knote *kn, long hint);
113 static int	filt_procattach(struct knote *kn);
114 static void	filt_procdetach(struct knote *kn);
115 static int	filt_proc(struct knote *kn, long hint);
116 static int	filt_fileattach(struct knote *kn);
117 static void	filt_timerexpire(void *knx);
118 static int	filt_timerattach(struct knote *kn);
119 static void	filt_timerdetach(struct knote *kn);
120 static int	filt_timer(struct knote *kn, long hint);
121 
122 static struct filterops file_filtops =
123 	{ FILTEROP_ISFD, filt_fileattach, NULL, NULL };
124 static struct filterops kqread_filtops =
125 	{ FILTEROP_ISFD, NULL, filt_kqdetach, filt_kqueue };
126 static struct filterops proc_filtops =
127 	{ 0, filt_procattach, filt_procdetach, filt_proc };
128 static struct filterops timer_filtops =
129 	{ 0, filt_timerattach, filt_timerdetach, filt_timer };
130 
131 static int 		kq_ncallouts = 0;
132 static int 		kq_calloutmax = (4 * 1024);
133 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
134     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
135 static int		kq_checkloop = 1000000;
136 SYSCTL_INT(_kern, OID_AUTO, kq_checkloop, CTLFLAG_RW,
137     &kq_checkloop, 0, "Maximum number of callouts allocated for kqueue");
138 
139 #define KNOTE_ACTIVATE(kn) do { 					\
140 	kn->kn_status |= KN_ACTIVE;					\
141 	if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
142 		knote_enqueue(kn);					\
143 } while(0)
144 
145 #define	KN_HASHSIZE		64		/* XXX should be tunable */
146 #define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))
147 
148 extern struct filterops aio_filtops;
149 extern struct filterops sig_filtops;
150 
151 /*
152  * Table for for all system-defined filters.
153  */
154 static struct filterops *sysfilt_ops[] = {
155 	&file_filtops,			/* EVFILT_READ */
156 	&file_filtops,			/* EVFILT_WRITE */
157 	&aio_filtops,			/* EVFILT_AIO */
158 	&file_filtops,			/* EVFILT_VNODE */
159 	&proc_filtops,			/* EVFILT_PROC */
160 	&sig_filtops,			/* EVFILT_SIGNAL */
161 	&timer_filtops,			/* EVFILT_TIMER */
162 	&file_filtops,			/* EVFILT_EXCEPT */
163 };
164 
165 static int
166 filt_fileattach(struct knote *kn)
167 {
168 	return (fo_kqfilter(kn->kn_fp, kn));
169 }
170 
171 /*
172  * MPSAFE
173  */
174 static int
175 kqueue_kqfilter(struct file *fp, struct knote *kn)
176 {
177 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
178 
179 	if (kn->kn_filter != EVFILT_READ)
180 		return (EOPNOTSUPP);
181 
182 	kn->kn_fop = &kqread_filtops;
183 	knote_insert(&kq->kq_kqinfo.ki_note, kn);
184 	return (0);
185 }
186 
187 static void
188 filt_kqdetach(struct knote *kn)
189 {
190 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
191 
192 	knote_remove(&kq->kq_kqinfo.ki_note, kn);
193 }
194 
195 /*ARGSUSED*/
196 static int
197 filt_kqueue(struct knote *kn, long hint)
198 {
199 	struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
200 
201 	kn->kn_data = kq->kq_count;
202 	return (kn->kn_data > 0);
203 }
204 
205 static int
206 filt_procattach(struct knote *kn)
207 {
208 	struct proc *p;
209 	int immediate;
210 
211 	immediate = 0;
212 	p = pfind(kn->kn_id);
213 	if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
214 		p = zpfind(kn->kn_id);
215 		immediate = 1;
216 	}
217 	if (p == NULL) {
218 		return (ESRCH);
219 	}
220 	if (!PRISON_CHECK(curthread->td_ucred, p->p_ucred)) {
221 		if (p)
222 			PRELE(p);
223 		return (EACCES);
224 	}
225 
226 	lwkt_gettoken(&p->p_token);
227 	kn->kn_ptr.p_proc = p;
228 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
229 
230 	/*
231 	 * internal flag indicating registration done by kernel
232 	 */
233 	if (kn->kn_flags & EV_FLAG1) {
234 		kn->kn_data = kn->kn_sdata;		/* ppid */
235 		kn->kn_fflags = NOTE_CHILD;
236 		kn->kn_flags &= ~EV_FLAG1;
237 	}
238 
239 	knote_insert(&p->p_klist, kn);
240 
241 	/*
242 	 * Immediately activate any exit notes if the target process is a
243 	 * zombie.  This is necessary to handle the case where the target
244 	 * process, e.g. a child, dies before the kevent is negistered.
245 	 */
246 	if (immediate && filt_proc(kn, NOTE_EXIT))
247 		KNOTE_ACTIVATE(kn);
248 	lwkt_reltoken(&p->p_token);
249 	PRELE(p);
250 
251 	return (0);
252 }
253 
254 /*
255  * The knote may be attached to a different process, which may exit,
256  * leaving nothing for the knote to be attached to.  So when the process
257  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
258  * it will be deleted when read out.  However, as part of the knote deletion,
259  * this routine is called, so a check is needed to avoid actually performing
260  * a detach, because the original process does not exist any more.
261  */
262 static void
263 filt_procdetach(struct knote *kn)
264 {
265 	struct proc *p;
266 
267 	if (kn->kn_status & KN_DETACHED)
268 		return;
269 	/* XXX locking? take proc_token here? */
270 	p = kn->kn_ptr.p_proc;
271 	knote_remove(&p->p_klist, kn);
272 }
273 
274 static int
275 filt_proc(struct knote *kn, long hint)
276 {
277 	u_int event;
278 
279 	/*
280 	 * mask off extra data
281 	 */
282 	event = (u_int)hint & NOTE_PCTRLMASK;
283 
284 	/*
285 	 * if the user is interested in this event, record it.
286 	 */
287 	if (kn->kn_sfflags & event)
288 		kn->kn_fflags |= event;
289 
290 	/*
291 	 * Process is gone, so flag the event as finished.  Detach the
292 	 * knote from the process now because the process will be poof,
293 	 * gone later on.
294 	 */
295 	if (event == NOTE_EXIT) {
296 		struct proc *p = kn->kn_ptr.p_proc;
297 		if ((kn->kn_status & KN_DETACHED) == 0) {
298 			knote_remove(&p->p_klist, kn);
299 			kn->kn_status |= KN_DETACHED;
300 			kn->kn_data = p->p_xstat;
301 			kn->kn_ptr.p_proc = NULL;
302 		}
303 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
304 		return (1);
305 	}
306 
307 	/*
308 	 * process forked, and user wants to track the new process,
309 	 * so attach a new knote to it, and immediately report an
310 	 * event with the parent's pid.
311 	 */
312 	if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
313 		struct kevent kev;
314 		int error;
315 
316 		/*
317 		 * register knote with new process.
318 		 */
319 		kev.ident = hint & NOTE_PDATAMASK;	/* pid */
320 		kev.filter = kn->kn_filter;
321 		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
322 		kev.fflags = kn->kn_sfflags;
323 		kev.data = kn->kn_id;			/* parent */
324 		kev.udata = kn->kn_kevent.udata;	/* preserve udata */
325 		error = kqueue_register(kn->kn_kq, &kev);
326 		if (error)
327 			kn->kn_fflags |= NOTE_TRACKERR;
328 	}
329 
330 	return (kn->kn_fflags != 0);
331 }
332 
333 /*
334  * The callout interlocks with callout_terminate() but can still
335  * race a deletion so if KN_DELETING is set we just don't touch
336  * the knote.
337  */
338 static void
339 filt_timerexpire(void *knx)
340 {
341 	struct knote *kn = knx;
342 	struct callout *calloutp;
343 	struct timeval tv;
344 	int tticks;
345 
346 	lwkt_gettoken(&kq_token);
347 	if ((kn->kn_status & KN_DELETING) == 0) {
348 		kn->kn_data++;
349 		KNOTE_ACTIVATE(kn);
350 
351 		if ((kn->kn_flags & EV_ONESHOT) == 0) {
352 			tv.tv_sec = kn->kn_sdata / 1000;
353 			tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
354 			tticks = tvtohz_high(&tv);
355 			calloutp = (struct callout *)kn->kn_hook;
356 			callout_reset(calloutp, tticks, filt_timerexpire, kn);
357 		}
358 	}
359 	lwkt_reltoken(&kq_token);
360 }
361 
362 /*
363  * data contains amount of time to sleep, in milliseconds
364  */
365 static int
366 filt_timerattach(struct knote *kn)
367 {
368 	struct callout *calloutp;
369 	struct timeval tv;
370 	int tticks;
371 
372 	if (kq_ncallouts >= kq_calloutmax) {
373 		kn->kn_hook = NULL;
374 		return (ENOMEM);
375 	}
376 	kq_ncallouts++;
377 
378 	tv.tv_sec = kn->kn_sdata / 1000;
379 	tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
380 	tticks = tvtohz_high(&tv);
381 
382 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
383 	MALLOC(calloutp, struct callout *, sizeof(*calloutp),
384 	    M_KQUEUE, M_WAITOK);
385 	callout_init(calloutp);
386 	kn->kn_hook = (caddr_t)calloutp;
387 	callout_reset(calloutp, tticks, filt_timerexpire, kn);
388 
389 	return (0);
390 }
391 
392 /*
393  * This function is called with the knote flagged locked but it is
394  * still possible to race a callout event due to the callback blocking.
395  * We must call callout_terminate() instead of callout_stop() to deal
396  * with the race.
397  */
398 static void
399 filt_timerdetach(struct knote *kn)
400 {
401 	struct callout *calloutp;
402 
403 	calloutp = (struct callout *)kn->kn_hook;
404 	callout_terminate(calloutp);
405 	FREE(calloutp, M_KQUEUE);
406 	kq_ncallouts--;
407 }
408 
409 static int
410 filt_timer(struct knote *kn, long hint)
411 {
412 
413 	return (kn->kn_data != 0);
414 }
415 
416 /*
417  * Acquire a knote, return non-zero on success, 0 on failure.
418  *
419  * If we cannot acquire the knote we sleep and return 0.  The knote
420  * may be stale on return in this case and the caller must restart
421  * whatever loop they are in.
422  */
423 static __inline
424 int
425 knote_acquire(struct knote *kn)
426 {
427 	if (kn->kn_status & KN_PROCESSING) {
428 		kn->kn_status |= KN_WAITING | KN_REPROCESS;
429 		tsleep(kn, 0, "kqepts", hz);
430 		/* knote may be stale now */
431 		return(0);
432 	}
433 	kn->kn_status |= KN_PROCESSING;
434 	return(1);
435 }
436 
437 /*
438  * Release an acquired knote, clearing KN_PROCESSING and handling any
439  * KN_REPROCESS events.
440  *
441  * Non-zero is returned if the knote is destroyed.
442  */
443 static __inline
444 int
445 knote_release(struct knote *kn)
446 {
447 	while (kn->kn_status & KN_REPROCESS) {
448 		kn->kn_status &= ~KN_REPROCESS;
449 		if (kn->kn_status & KN_WAITING) {
450 			kn->kn_status &= ~KN_WAITING;
451 			wakeup(kn);
452 		}
453 		if (kn->kn_status & KN_DELETING) {
454 			knote_detach_and_drop(kn);
455 			return(1);
456 			/* NOT REACHED */
457 		}
458 		if (filter_event(kn, 0))
459 			KNOTE_ACTIVATE(kn);
460 	}
461 	kn->kn_status &= ~KN_PROCESSING;
462 	return(0);
463 }
464 
465 /*
466  * Initialize a kqueue.
467  *
468  * NOTE: The lwp/proc code initializes a kqueue for select/poll ops.
469  *
470  * MPSAFE
471  */
472 void
473 kqueue_init(struct kqueue *kq, struct filedesc *fdp)
474 {
475 	TAILQ_INIT(&kq->kq_knpend);
476 	TAILQ_INIT(&kq->kq_knlist);
477 	kq->kq_count = 0;
478 	kq->kq_fdp = fdp;
479 	SLIST_INIT(&kq->kq_kqinfo.ki_note);
480 }
481 
482 /*
483  * Terminate a kqueue.  Freeing the actual kq itself is left up to the
484  * caller (it might be embedded in a lwp so we don't do it here).
485  *
486  * The kq's knlist must be completely eradicated so block on any
487  * processing races.
488  */
489 void
490 kqueue_terminate(struct kqueue *kq)
491 {
492 	struct knote *kn;
493 
494 	lwkt_gettoken(&kq_token);
495 	while ((kn = TAILQ_FIRST(&kq->kq_knlist)) != NULL) {
496 		if (knote_acquire(kn))
497 			knote_detach_and_drop(kn);
498 	}
499 	if (kq->kq_knhash) {
500 		kfree(kq->kq_knhash, M_KQUEUE);
501 		kq->kq_knhash = NULL;
502 		kq->kq_knhashmask = 0;
503 	}
504 	lwkt_reltoken(&kq_token);
505 }
506 
507 /*
508  * MPSAFE
509  */
510 int
511 sys_kqueue(struct kqueue_args *uap)
512 {
513 	struct thread *td = curthread;
514 	struct kqueue *kq;
515 	struct file *fp;
516 	int fd, error;
517 
518 	error = falloc(td->td_lwp, &fp, &fd);
519 	if (error)
520 		return (error);
521 	fp->f_flag = FREAD | FWRITE;
522 	fp->f_type = DTYPE_KQUEUE;
523 	fp->f_ops = &kqueueops;
524 
525 	kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO);
526 	kqueue_init(kq, td->td_proc->p_fd);
527 	fp->f_data = kq;
528 
529 	fsetfd(kq->kq_fdp, fp, fd);
530 	uap->sysmsg_result = fd;
531 	fdrop(fp);
532 	return (error);
533 }
534 
535 /*
536  * Copy 'count' items into the destination list pointed to by uap->eventlist.
537  */
538 static int
539 kevent_copyout(void *arg, struct kevent *kevp, int count, int *res)
540 {
541 	struct kevent_copyin_args *kap;
542 	int error;
543 
544 	kap = (struct kevent_copyin_args *)arg;
545 
546 	error = copyout(kevp, kap->ka->eventlist, count * sizeof(*kevp));
547 	if (error == 0) {
548 		kap->ka->eventlist += count;
549 		*res += count;
550 	} else {
551 		*res = -1;
552 	}
553 
554 	return (error);
555 }
556 
557 /*
558  * Copy at most 'max' items from the list pointed to by kap->changelist,
559  * return number of items in 'events'.
560  */
561 static int
562 kevent_copyin(void *arg, struct kevent *kevp, int max, int *events)
563 {
564 	struct kevent_copyin_args *kap;
565 	int error, count;
566 
567 	kap = (struct kevent_copyin_args *)arg;
568 
569 	count = min(kap->ka->nchanges - kap->pchanges, max);
570 	error = copyin(kap->ka->changelist, kevp, count * sizeof *kevp);
571 	if (error == 0) {
572 		kap->ka->changelist += count;
573 		kap->pchanges += count;
574 		*events = count;
575 	}
576 
577 	return (error);
578 }
579 
580 /*
581  * MPSAFE
582  */
583 int
584 kern_kevent(struct kqueue *kq, int nevents, int *res, void *uap,
585 	    k_copyin_fn kevent_copyinfn, k_copyout_fn kevent_copyoutfn,
586 	    struct timespec *tsp_in)
587 {
588 	struct kevent *kevp;
589 	struct timespec *tsp;
590 	int i, n, total, error, nerrors = 0;
591 	int lres;
592 	int limit = kq_checkloop;
593 	struct kevent kev[KQ_NEVENTS];
594 	struct knote marker;
595 
596 	tsp = tsp_in;
597 	*res = 0;
598 
599 	lwkt_gettoken(&kq_token);
600 	for ( ;; ) {
601 		n = 0;
602 		error = kevent_copyinfn(uap, kev, KQ_NEVENTS, &n);
603 		if (error)
604 			goto done;
605 		if (n == 0)
606 			break;
607 		for (i = 0; i < n; i++) {
608 			kevp = &kev[i];
609 			kevp->flags &= ~EV_SYSFLAGS;
610 			error = kqueue_register(kq, kevp);
611 
612 			/*
613 			 * If a registration returns an error we
614 			 * immediately post the error.  The kevent()
615 			 * call itself will fail with the error if
616 			 * no space is available for posting.
617 			 *
618 			 * Such errors normally bypass the timeout/blocking
619 			 * code.  However, if the copyoutfn function refuses
620 			 * to post the error (see sys_poll()), then we
621 			 * ignore it too.
622 			 */
623 			if (error) {
624 				kevp->flags = EV_ERROR;
625 				kevp->data = error;
626 				lres = *res;
627 				kevent_copyoutfn(uap, kevp, 1, res);
628 				if (lres != *res) {
629 					nevents--;
630 					nerrors++;
631 				}
632 			}
633 		}
634 	}
635 	if (nerrors) {
636 		error = 0;
637 		goto done;
638 	}
639 
640 	/*
641 	 * Acquire/wait for events - setup timeout
642 	 */
643 	if (tsp != NULL) {
644 		struct timespec ats;
645 
646 		if (tsp->tv_sec || tsp->tv_nsec) {
647 			nanouptime(&ats);
648 			timespecadd(tsp, &ats);		/* tsp = target time */
649 		}
650 	}
651 
652 	/*
653 	 * Loop as required.
654 	 *
655 	 * Collect as many events as we can. Sleeping on successive
656 	 * loops is disabled if copyoutfn has incremented (*res).
657 	 *
658 	 * The loop stops if an error occurs, all events have been
659 	 * scanned (the marker has been reached), or fewer than the
660 	 * maximum number of events is found.
661 	 *
662 	 * The copyoutfn function does not have to increment (*res) in
663 	 * order for the loop to continue.
664 	 *
665 	 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents.
666 	 */
667 	total = 0;
668 	error = 0;
669 	marker.kn_filter = EVFILT_MARKER;
670 	marker.kn_status = KN_PROCESSING;
671 	TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
672 	while ((n = nevents - total) > 0) {
673 		if (n > KQ_NEVENTS)
674 			n = KQ_NEVENTS;
675 
676 		/*
677 		 * If no events are pending sleep until timeout (if any)
678 		 * or an event occurs.
679 		 *
680 		 * After the sleep completes the marker is moved to the
681 		 * end of the list, making any received events available
682 		 * to our scan.
683 		 */
684 		if (kq->kq_count == 0 && *res == 0) {
685 			error = kqueue_sleep(kq, tsp);
686 			if (error)
687 				break;
688 
689 			TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
690 			TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
691 		}
692 
693 		/*
694 		 * Process all received events
695 		 * Account for all non-spurious events in our total
696 		 */
697 		i = kqueue_scan(kq, kev, n, &marker);
698 		if (i) {
699 			lres = *res;
700 			error = kevent_copyoutfn(uap, kev, i, res);
701 			total += *res - lres;
702 			if (error)
703 				break;
704 		}
705 		if (limit && --limit == 0)
706 			panic("kqueue: checkloop failed i=%d", i);
707 
708 		/*
709 		 * Normally when fewer events are returned than requested
710 		 * we can stop.  However, if only spurious events were
711 		 * collected the copyout will not bump (*res) and we have
712 		 * to continue.
713 		 */
714 		if (i < n && *res)
715 			break;
716 
717 		/*
718 		 * Deal with an edge case where spurious events can cause
719 		 * a loop to occur without moving the marker.  This can
720 		 * prevent kqueue_scan() from picking up new events which
721 		 * race us.  We must be sure to move the marker for this
722 		 * case.
723 		 *
724 		 * NOTE: We do not want to move the marker if events
725 		 *	 were scanned because normal kqueue operations
726 		 *	 may reactivate events.  Moving the marker in
727 		 *	 that case could result in duplicates for the
728 		 *	 same event.
729 		 */
730 		if (i == 0) {
731 			TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
732 			TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
733 		}
734 	}
735 	TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
736 
737 	/* Timeouts do not return EWOULDBLOCK. */
738 	if (error == EWOULDBLOCK)
739 		error = 0;
740 
741 done:
742 	lwkt_reltoken(&kq_token);
743 	return (error);
744 }
745 
746 /*
747  * MPALMOSTSAFE
748  */
749 int
750 sys_kevent(struct kevent_args *uap)
751 {
752 	struct thread *td = curthread;
753 	struct proc *p = td->td_proc;
754 	struct timespec ts, *tsp;
755 	struct kqueue *kq;
756 	struct file *fp = NULL;
757 	struct kevent_copyin_args *kap, ka;
758 	int error;
759 
760 	if (uap->timeout) {
761 		error = copyin(uap->timeout, &ts, sizeof(ts));
762 		if (error)
763 			return (error);
764 		tsp = &ts;
765 	} else {
766 		tsp = NULL;
767 	}
768 
769 	fp = holdfp(p->p_fd, uap->fd, -1);
770 	if (fp == NULL)
771 		return (EBADF);
772 	if (fp->f_type != DTYPE_KQUEUE) {
773 		fdrop(fp);
774 		return (EBADF);
775 	}
776 
777 	kq = (struct kqueue *)fp->f_data;
778 
779 	kap = &ka;
780 	kap->ka = uap;
781 	kap->pchanges = 0;
782 
783 	error = kern_kevent(kq, uap->nevents, &uap->sysmsg_result, kap,
784 			    kevent_copyin, kevent_copyout, tsp);
785 
786 	fdrop(fp);
787 
788 	return (error);
789 }
790 
791 int
792 kqueue_register(struct kqueue *kq, struct kevent *kev)
793 {
794 	struct filedesc *fdp = kq->kq_fdp;
795 	struct filterops *fops;
796 	struct file *fp = NULL;
797 	struct knote *kn = NULL;
798 	int error = 0;
799 
800 	if (kev->filter < 0) {
801 		if (kev->filter + EVFILT_SYSCOUNT < 0)
802 			return (EINVAL);
803 		fops = sysfilt_ops[~kev->filter];	/* to 0-base index */
804 	} else {
805 		/*
806 		 * XXX
807 		 * filter attach routine is responsible for insuring that
808 		 * the identifier can be attached to it.
809 		 */
810 		kprintf("unknown filter: %d\n", kev->filter);
811 		return (EINVAL);
812 	}
813 
814 	lwkt_gettoken(&kq_token);
815 	if (fops->f_flags & FILTEROP_ISFD) {
816 		/* validate descriptor */
817 		fp = holdfp(fdp, kev->ident, -1);
818 		if (fp == NULL) {
819 			lwkt_reltoken(&kq_token);
820 			return (EBADF);
821 		}
822 
823 again1:
824 		SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
825 			if (kn->kn_kq == kq &&
826 			    kn->kn_filter == kev->filter &&
827 			    kn->kn_id == kev->ident) {
828 				if (knote_acquire(kn) == 0)
829 					goto again1;
830 				break;
831 			}
832 		}
833 	} else {
834 		if (kq->kq_knhashmask) {
835 			struct klist *list;
836 
837 			list = &kq->kq_knhash[
838 			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
839 again2:
840 			SLIST_FOREACH(kn, list, kn_link) {
841 				if (kn->kn_id == kev->ident &&
842 				    kn->kn_filter == kev->filter) {
843 					if (knote_acquire(kn) == 0)
844 						goto again2;
845 					break;
846 				}
847 			}
848 		}
849 	}
850 
851 	/*
852 	 * NOTE: At this point if kn is non-NULL we will have acquired
853 	 *	 it and set KN_PROCESSING.
854 	 */
855 	if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
856 		error = ENOENT;
857 		goto done;
858 	}
859 
860 	/*
861 	 * kn now contains the matching knote, or NULL if no match
862 	 */
863 	if (kev->flags & EV_ADD) {
864 		if (kn == NULL) {
865 			kn = knote_alloc();
866 			if (kn == NULL) {
867 				error = ENOMEM;
868 				goto done;
869 			}
870 			kn->kn_fp = fp;
871 			kn->kn_kq = kq;
872 			kn->kn_fop = fops;
873 
874 			/*
875 			 * apply reference count to knote structure, and
876 			 * do not release it at the end of this routine.
877 			 */
878 			fp = NULL;
879 
880 			kn->kn_sfflags = kev->fflags;
881 			kn->kn_sdata = kev->data;
882 			kev->fflags = 0;
883 			kev->data = 0;
884 			kn->kn_kevent = *kev;
885 
886 			/*
887 			 * KN_PROCESSING prevents the knote from getting
888 			 * ripped out from under us while we are trying
889 			 * to attach it, in case the attach blocks.
890 			 */
891 			kn->kn_status = KN_PROCESSING;
892 			knote_attach(kn);
893 			if ((error = filter_attach(kn)) != 0) {
894 				kn->kn_status |= KN_DELETING | KN_REPROCESS;
895 				knote_drop(kn);
896 				goto done;
897 			}
898 
899 			/*
900 			 * Interlock against close races which either tried
901 			 * to remove our knote while we were blocked or missed
902 			 * it entirely prior to our attachment.  We do not
903 			 * want to end up with a knote on a closed descriptor.
904 			 */
905 			if ((fops->f_flags & FILTEROP_ISFD) &&
906 			    checkfdclosed(fdp, kev->ident, kn->kn_fp)) {
907 				kn->kn_status |= KN_DELETING | KN_REPROCESS;
908 			}
909 		} else {
910 			/*
911 			 * The user may change some filter values after the
912 			 * initial EV_ADD, but doing so will not reset any
913 			 * filter which have already been triggered.
914 			 */
915 			KKASSERT(kn->kn_status & KN_PROCESSING);
916 			kn->kn_sfflags = kev->fflags;
917 			kn->kn_sdata = kev->data;
918 			kn->kn_kevent.udata = kev->udata;
919 		}
920 
921 		/*
922 		 * Execute the filter event to immediately activate the
923 		 * knote if necessary.  If reprocessing events are pending
924 		 * due to blocking above we do not run the filter here
925 		 * but instead let knote_release() do it.  Otherwise we
926 		 * might run the filter on a deleted event.
927 		 */
928 		if ((kn->kn_status & KN_REPROCESS) == 0) {
929 			if (filter_event(kn, 0))
930 				KNOTE_ACTIVATE(kn);
931 		}
932 	} else if (kev->flags & EV_DELETE) {
933 		/*
934 		 * Delete the existing knote
935 		 */
936 		knote_detach_and_drop(kn);
937 		goto done;
938 	}
939 
940 	/*
941 	 * Disablement does not deactivate a knote here.
942 	 */
943 	if ((kev->flags & EV_DISABLE) &&
944 	    ((kn->kn_status & KN_DISABLED) == 0)) {
945 		kn->kn_status |= KN_DISABLED;
946 	}
947 
948 	/*
949 	 * Re-enablement may have to immediately enqueue an active knote.
950 	 */
951 	if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
952 		kn->kn_status &= ~KN_DISABLED;
953 		if ((kn->kn_status & KN_ACTIVE) &&
954 		    ((kn->kn_status & KN_QUEUED) == 0)) {
955 			knote_enqueue(kn);
956 		}
957 	}
958 
959 	/*
960 	 * Handle any required reprocessing
961 	 */
962 	knote_release(kn);
963 	/* kn may be invalid now */
964 
965 done:
966 	lwkt_reltoken(&kq_token);
967 	if (fp != NULL)
968 		fdrop(fp);
969 	return (error);
970 }
971 
972 /*
973  * Block as necessary until the target time is reached.
974  * If tsp is NULL we block indefinitely.  If tsp->ts_secs/nsecs are both
975  * 0 we do not block at all.
976  */
977 static int
978 kqueue_sleep(struct kqueue *kq, struct timespec *tsp)
979 {
980 	int error = 0;
981 
982 	if (tsp == NULL) {
983 		kq->kq_state |= KQ_SLEEP;
984 		error = tsleep(kq, PCATCH, "kqread", 0);
985 	} else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
986 		error = EWOULDBLOCK;
987 	} else {
988 		struct timespec ats;
989 		struct timespec atx = *tsp;
990 		int timeout;
991 
992 		nanouptime(&ats);
993 		timespecsub(&atx, &ats);
994 		if (ats.tv_sec < 0) {
995 			error = EWOULDBLOCK;
996 		} else {
997 			timeout = atx.tv_sec > 24 * 60 * 60 ?
998 				24 * 60 * 60 * hz : tstohz_high(&atx);
999 			kq->kq_state |= KQ_SLEEP;
1000 			error = tsleep(kq, PCATCH, "kqread", timeout);
1001 		}
1002 	}
1003 
1004 	/* don't restart after signals... */
1005 	if (error == ERESTART)
1006 		return (EINTR);
1007 
1008 	return (error);
1009 }
1010 
1011 /*
1012  * Scan the kqueue, return the number of active events placed in kevp up
1013  * to count.
1014  *
1015  * Continuous mode events may get recycled, do not continue scanning past
1016  * marker unless no events have been collected.
1017  */
1018 static int
1019 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count,
1020             struct knote *marker)
1021 {
1022         struct knote *kn, local_marker;
1023         int total;
1024 
1025         total = 0;
1026 	local_marker.kn_filter = EVFILT_MARKER;
1027 	local_marker.kn_status = KN_PROCESSING;
1028 
1029 	/*
1030 	 * Collect events.
1031 	 */
1032 	TAILQ_INSERT_HEAD(&kq->kq_knpend, &local_marker, kn_tqe);
1033 	while (count) {
1034 		kn = TAILQ_NEXT(&local_marker, kn_tqe);
1035 		if (kn->kn_filter == EVFILT_MARKER) {
1036 			/* Marker reached, we are done */
1037 			if (kn == marker)
1038 				break;
1039 
1040 			/* Move local marker past some other threads marker */
1041 			kn = TAILQ_NEXT(kn, kn_tqe);
1042 			TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe);
1043 			TAILQ_INSERT_BEFORE(kn, &local_marker, kn_tqe);
1044 			continue;
1045 		}
1046 
1047 		/*
1048 		 * We can't skip a knote undergoing processing, otherwise
1049 		 * we risk not returning it when the user process expects
1050 		 * it should be returned.  Sleep and retry.
1051 		 */
1052 		if (knote_acquire(kn) == 0)
1053 			continue;
1054 
1055 		/*
1056 		 * Remove the event for processing.
1057 		 *
1058 		 * WARNING!  We must leave KN_QUEUED set to prevent the
1059 		 *	     event from being KNOTE_ACTIVATE()d while
1060 		 *	     the queue state is in limbo, in case we
1061 		 *	     block.
1062 		 *
1063 		 * WARNING!  We must set KN_PROCESSING to avoid races
1064 		 *	     against deletion or another thread's
1065 		 *	     processing.
1066 		 */
1067 		TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
1068 		kq->kq_count--;
1069 
1070 		/*
1071 		 * We have to deal with an extremely important race against
1072 		 * file descriptor close()s here.  The file descriptor can
1073 		 * disappear MPSAFE, and there is a small window of
1074 		 * opportunity between that and the call to knote_fdclose().
1075 		 *
1076 		 * If we hit that window here while doselect or dopoll is
1077 		 * trying to delete a spurious event they will not be able
1078 		 * to match up the event against a knote and will go haywire.
1079 		 */
1080 		if ((kn->kn_fop->f_flags & FILTEROP_ISFD) &&
1081 		    checkfdclosed(kq->kq_fdp, kn->kn_kevent.ident, kn->kn_fp)) {
1082 			kn->kn_status |= KN_DELETING | KN_REPROCESS;
1083 		}
1084 
1085 		if (kn->kn_status & KN_DISABLED) {
1086 			/*
1087 			 * If disabled we ensure the event is not queued
1088 			 * but leave its active bit set.  On re-enablement
1089 			 * the event may be immediately triggered.
1090 			 */
1091 			kn->kn_status &= ~KN_QUEUED;
1092 		} else if ((kn->kn_flags & EV_ONESHOT) == 0 &&
1093 			   (kn->kn_status & KN_DELETING) == 0 &&
1094 			   filter_event(kn, 0) == 0) {
1095 			/*
1096 			 * If not running in one-shot mode and the event
1097 			 * is no longer present we ensure it is removed
1098 			 * from the queue and ignore it.
1099 			 */
1100 			kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1101 		} else {
1102 			/*
1103 			 * Post the event
1104 			 */
1105 			*kevp++ = kn->kn_kevent;
1106 			++total;
1107 			--count;
1108 
1109 			if (kn->kn_flags & EV_ONESHOT) {
1110 				kn->kn_status &= ~KN_QUEUED;
1111 				kn->kn_status |= KN_DELETING | KN_REPROCESS;
1112 			} else if (kn->kn_flags & EV_CLEAR) {
1113 				kn->kn_data = 0;
1114 				kn->kn_fflags = 0;
1115 				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1116 			} else {
1117 				TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
1118 				kq->kq_count++;
1119 			}
1120 		}
1121 
1122 		/*
1123 		 * Handle any post-processing states
1124 		 */
1125 		knote_release(kn);
1126 	}
1127 	TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe);
1128 
1129 	return (total);
1130 }
1131 
1132 /*
1133  * XXX
1134  * This could be expanded to call kqueue_scan, if desired.
1135  *
1136  * MPSAFE
1137  */
1138 static int
1139 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
1140 {
1141 	return (ENXIO);
1142 }
1143 
1144 /*
1145  * MPSAFE
1146  */
1147 static int
1148 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
1149 {
1150 	return (ENXIO);
1151 }
1152 
1153 /*
1154  * MPALMOSTSAFE
1155  */
1156 static int
1157 kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
1158 	     struct ucred *cred, struct sysmsg *msg)
1159 {
1160 	struct kqueue *kq;
1161 	int error;
1162 
1163 	lwkt_gettoken(&kq_token);
1164 	kq = (struct kqueue *)fp->f_data;
1165 
1166 	switch(com) {
1167 	case FIOASYNC:
1168 		if (*(int *)data)
1169 			kq->kq_state |= KQ_ASYNC;
1170 		else
1171 			kq->kq_state &= ~KQ_ASYNC;
1172 		error = 0;
1173 		break;
1174 	case FIOSETOWN:
1175 		error = fsetown(*(int *)data, &kq->kq_sigio);
1176 		break;
1177 	default:
1178 		error = ENOTTY;
1179 		break;
1180 	}
1181 	lwkt_reltoken(&kq_token);
1182 	return (error);
1183 }
1184 
1185 /*
1186  * MPSAFE
1187  */
1188 static int
1189 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
1190 {
1191 	struct kqueue *kq = (struct kqueue *)fp->f_data;
1192 
1193 	bzero((void *)st, sizeof(*st));
1194 	st->st_size = kq->kq_count;
1195 	st->st_blksize = sizeof(struct kevent);
1196 	st->st_mode = S_IFIFO;
1197 	return (0);
1198 }
1199 
1200 /*
1201  * MPSAFE
1202  */
1203 static int
1204 kqueue_close(struct file *fp)
1205 {
1206 	struct kqueue *kq = (struct kqueue *)fp->f_data;
1207 
1208 	kqueue_terminate(kq);
1209 
1210 	fp->f_data = NULL;
1211 	funsetown(&kq->kq_sigio);
1212 
1213 	kfree(kq, M_KQUEUE);
1214 	return (0);
1215 }
1216 
1217 static void
1218 kqueue_wakeup(struct kqueue *kq)
1219 {
1220 	if (kq->kq_state & KQ_SLEEP) {
1221 		kq->kq_state &= ~KQ_SLEEP;
1222 		wakeup(kq);
1223 	}
1224 	KNOTE(&kq->kq_kqinfo.ki_note, 0);
1225 }
1226 
1227 /*
1228  * Calls filterops f_attach function, acquiring mplock if filter is not
1229  * marked as FILTEROP_MPSAFE.
1230  */
1231 static int
1232 filter_attach(struct knote *kn)
1233 {
1234 	int ret;
1235 
1236 	if (!(kn->kn_fop->f_flags & FILTEROP_MPSAFE)) {
1237 		get_mplock();
1238 		ret = kn->kn_fop->f_attach(kn);
1239 		rel_mplock();
1240 	} else {
1241 		ret = kn->kn_fop->f_attach(kn);
1242 	}
1243 
1244 	return (ret);
1245 }
1246 
1247 /*
1248  * Detach the knote and drop it, destroying the knote.
1249  *
1250  * Calls filterops f_detach function, acquiring mplock if filter is not
1251  * marked as FILTEROP_MPSAFE.
1252  */
1253 static void
1254 knote_detach_and_drop(struct knote *kn)
1255 {
1256 	kn->kn_status |= KN_DELETING | KN_REPROCESS;
1257 	if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
1258 		kn->kn_fop->f_detach(kn);
1259 	} else {
1260 		get_mplock();
1261 		kn->kn_fop->f_detach(kn);
1262 		rel_mplock();
1263 	}
1264 	knote_drop(kn);
1265 }
1266 
1267 /*
1268  * Calls filterops f_event function, acquiring mplock if filter is not
1269  * marked as FILTEROP_MPSAFE.
1270  *
1271  * If the knote is in the middle of being created or deleted we cannot
1272  * safely call the filter op.
1273  */
1274 static int
1275 filter_event(struct knote *kn, long hint)
1276 {
1277 	int ret;
1278 
1279 	if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
1280 		ret = kn->kn_fop->f_event(kn, hint);
1281 	} else {
1282 		get_mplock();
1283 		ret = kn->kn_fop->f_event(kn, hint);
1284 		rel_mplock();
1285 	}
1286 	return (ret);
1287 }
1288 
1289 /*
1290  * Walk down a list of knotes, activating them if their event has triggered.
1291  *
1292  * If we encounter any knotes which are undergoing processing we just mark
1293  * them for reprocessing and do not try to [re]activate the knote.  However,
1294  * if a hint is being passed we have to wait and that makes things a bit
1295  * sticky.
1296  */
1297 void
1298 knote(struct klist *list, long hint)
1299 {
1300 	struct knote *kn;
1301 
1302 	lwkt_gettoken(&kq_token);
1303 restart:
1304 	SLIST_FOREACH(kn, list, kn_next) {
1305 		if (kn->kn_status & KN_PROCESSING) {
1306 			/*
1307 			 * Someone else is processing the knote, ask the
1308 			 * other thread to reprocess it and don't mess
1309 			 * with it otherwise.
1310 			 */
1311 			if (hint == 0) {
1312 				kn->kn_status |= KN_REPROCESS;
1313 				continue;
1314 			}
1315 
1316 			/*
1317 			 * If the hint is non-zero we have to wait or risk
1318 			 * losing the state the caller is trying to update.
1319 			 *
1320 			 * XXX This is a real problem, certain process
1321 			 *     and signal filters will bump kn_data for
1322 			 *     already-processed notes more than once if
1323 			 *     we restart the list scan.  FIXME.
1324 			 */
1325 			kn->kn_status |= KN_WAITING | KN_REPROCESS;
1326 			tsleep(kn, 0, "knotec", hz);
1327 			goto restart;
1328 		}
1329 
1330 		/*
1331 		 * Become the reprocessing master ourselves.
1332 		 *
1333 		 * If hint is non-zer running the event is mandatory
1334 		 * when not deleting so do it whether reprocessing is
1335 		 * set or not.
1336 		 */
1337 		kn->kn_status |= KN_PROCESSING;
1338 		if ((kn->kn_status & KN_DELETING) == 0) {
1339 			if (filter_event(kn, hint))
1340 				KNOTE_ACTIVATE(kn);
1341 		}
1342 		if (knote_release(kn))
1343 			goto restart;
1344 	}
1345 	lwkt_reltoken(&kq_token);
1346 }
1347 
1348 /*
1349  * Insert knote at head of klist.
1350  *
1351  * This function may only be called via a filter function and thus
1352  * kq_token should already be held and marked for processing.
1353  */
1354 void
1355 knote_insert(struct klist *klist, struct knote *kn)
1356 {
1357 	KKASSERT(kn->kn_status & KN_PROCESSING);
1358 	ASSERT_LWKT_TOKEN_HELD(&kq_token);
1359 	SLIST_INSERT_HEAD(klist, kn, kn_next);
1360 }
1361 
1362 /*
1363  * Remove knote from a klist
1364  *
1365  * This function may only be called via a filter function and thus
1366  * kq_token should already be held and marked for processing.
1367  */
1368 void
1369 knote_remove(struct klist *klist, struct knote *kn)
1370 {
1371 	KKASSERT(kn->kn_status & KN_PROCESSING);
1372 	ASSERT_LWKT_TOKEN_HELD(&kq_token);
1373 	SLIST_REMOVE(klist, kn, knote, kn_next);
1374 }
1375 
1376 /*
1377  * Remove all knotes from a specified klist
1378  *
1379  * Only called from aio.
1380  */
1381 void
1382 knote_empty(struct klist *list)
1383 {
1384 	struct knote *kn;
1385 
1386 	lwkt_gettoken(&kq_token);
1387 	while ((kn = SLIST_FIRST(list)) != NULL) {
1388 		if (knote_acquire(kn))
1389 			knote_detach_and_drop(kn);
1390 	}
1391 	lwkt_reltoken(&kq_token);
1392 }
1393 
1394 void
1395 knote_assume_knotes(struct kqinfo *src, struct kqinfo *dst,
1396 		    struct filterops *ops, void *hook)
1397 {
1398 	struct knote *kn;
1399 
1400 	lwkt_gettoken(&kq_token);
1401 	while ((kn = SLIST_FIRST(&src->ki_note)) != NULL) {
1402 		if (knote_acquire(kn)) {
1403 			knote_remove(&src->ki_note, kn);
1404 			kn->kn_fop = ops;
1405 			kn->kn_hook = hook;
1406 			knote_insert(&dst->ki_note, kn);
1407 			knote_release(kn);
1408 			/* kn may be invalid now */
1409 		}
1410 	}
1411 	lwkt_reltoken(&kq_token);
1412 }
1413 
1414 /*
1415  * Remove all knotes referencing a specified fd
1416  */
1417 void
1418 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd)
1419 {
1420 	struct knote *kn;
1421 
1422 	lwkt_gettoken(&kq_token);
1423 restart:
1424 	SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
1425 		if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) {
1426 			if (knote_acquire(kn))
1427 				knote_detach_and_drop(kn);
1428 			goto restart;
1429 		}
1430 	}
1431 	lwkt_reltoken(&kq_token);
1432 }
1433 
1434 /*
1435  * Low level attach function.
1436  *
1437  * The knote should already be marked for processing.
1438  */
1439 static void
1440 knote_attach(struct knote *kn)
1441 {
1442 	struct klist *list;
1443 	struct kqueue *kq = kn->kn_kq;
1444 
1445 	if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
1446 		KKASSERT(kn->kn_fp);
1447 		list = &kn->kn_fp->f_klist;
1448 	} else {
1449 		if (kq->kq_knhashmask == 0)
1450 			kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1451 						 &kq->kq_knhashmask);
1452 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1453 	}
1454 	SLIST_INSERT_HEAD(list, kn, kn_link);
1455 	TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink);
1456 }
1457 
1458 /*
1459  * Low level drop function.
1460  *
1461  * The knote should already be marked for processing.
1462  */
1463 static void
1464 knote_drop(struct knote *kn)
1465 {
1466 	struct kqueue *kq;
1467 	struct klist *list;
1468 
1469 	kq = kn->kn_kq;
1470 
1471 	if (kn->kn_fop->f_flags & FILTEROP_ISFD)
1472 		list = &kn->kn_fp->f_klist;
1473 	else
1474 		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1475 
1476 	SLIST_REMOVE(list, kn, knote, kn_link);
1477 	TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink);
1478 	if (kn->kn_status & KN_QUEUED)
1479 		knote_dequeue(kn);
1480 	if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
1481 		fdrop(kn->kn_fp);
1482 		kn->kn_fp = NULL;
1483 	}
1484 	knote_free(kn);
1485 }
1486 
1487 /*
1488  * Low level enqueue function.
1489  *
1490  * The knote should already be marked for processing.
1491  */
1492 static void
1493 knote_enqueue(struct knote *kn)
1494 {
1495 	struct kqueue *kq = kn->kn_kq;
1496 
1497 	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1498 	TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
1499 	kn->kn_status |= KN_QUEUED;
1500 	++kq->kq_count;
1501 
1502 	/*
1503 	 * Send SIGIO on request (typically set up as a mailbox signal)
1504 	 */
1505 	if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1506 		pgsigio(kq->kq_sigio, SIGIO, 0);
1507 
1508 	kqueue_wakeup(kq);
1509 }
1510 
1511 /*
1512  * Low level dequeue function.
1513  *
1514  * The knote should already be marked for processing.
1515  */
1516 static void
1517 knote_dequeue(struct knote *kn)
1518 {
1519 	struct kqueue *kq = kn->kn_kq;
1520 
1521 	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1522 	TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
1523 	kn->kn_status &= ~KN_QUEUED;
1524 	kq->kq_count--;
1525 }
1526 
1527 static struct knote *
1528 knote_alloc(void)
1529 {
1530 	return kmalloc(sizeof(struct knote), M_KQUEUE, M_WAITOK);
1531 }
1532 
1533 static void
1534 knote_free(struct knote *kn)
1535 {
1536 	kfree(kn, M_KQUEUE);
1537 }
1538