2 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
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 $
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
34 #include <sys/malloc.h>
35 #include <sys/unistd.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>
46 #include <sys/sysctl.h>
47 #include <sys/sysproto.h>
48 #include <sys/thread.h>
50 #include <sys/signalvar.h>
51 #include <sys/filio.h>
54 #include <sys/thread2.h>
55 #include <sys/file2.h>
56 #include <sys/mplock2.h>
59 * Global token for kqueue subsystem
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");
66 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
68 struct kevent_copyin_args {
69 struct kevent_args *ka;
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,
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);
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
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);
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);
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 };
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");
139 #define KNOTE_ACTIVATE(kn) do { \
140 kn->kn_status |= KN_ACTIVE; \
141 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
145 #define KN_HASHSIZE 64 /* XXX should be tunable */
146 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
148 extern struct filterops aio_filtops;
149 extern struct filterops sig_filtops;
152 * Table for for all system-defined filters.
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 */
166 filt_fileattach(struct knote *kn)
168 return (fo_kqfilter(kn->kn_fp, kn));
175 kqueue_kqfilter(struct file *fp, struct knote *kn)
177 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
179 if (kn->kn_filter != EVFILT_READ)
182 kn->kn_fop = &kqread_filtops;
183 knote_insert(&kq->kq_kqinfo.ki_note, kn);
188 filt_kqdetach(struct knote *kn)
190 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
192 knote_remove(&kq->kq_kqinfo.ki_note, kn);
197 filt_kqueue(struct knote *kn, long hint)
199 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
201 kn->kn_data = kq->kq_count;
202 return (kn->kn_data > 0);
206 filt_procattach(struct knote *kn)
212 p = pfind(kn->kn_id);
213 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
214 p = zpfind(kn->kn_id);
220 if (!PRISON_CHECK(curthread->td_ucred, p->p_ucred)) {
226 lwkt_gettoken(&p->p_token);
227 kn->kn_ptr.p_proc = p;
228 kn->kn_flags |= EV_CLEAR; /* automatically set */
231 * internal flag indicating registration done by kernel
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;
239 knote_insert(&p->p_klist, kn);
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.
246 if (immediate && filt_proc(kn, NOTE_EXIT))
248 lwkt_reltoken(&p->p_token);
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.
263 filt_procdetach(struct knote *kn)
267 if (kn->kn_status & KN_DETACHED)
269 /* XXX locking? take proc_token here? */
270 p = kn->kn_ptr.p_proc;
271 knote_remove(&p->p_klist, kn);
275 filt_proc(struct knote *kn, long hint)
280 * mask off extra data
282 event = (u_int)hint & NOTE_PCTRLMASK;
285 * if the user is interested in this event, record it.
287 if (kn->kn_sfflags & event)
288 kn->kn_fflags |= event;
291 * Process is gone, so flag the event as finished. Detach the
292 * knote from the process now because the process will be poof,
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;
303 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
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.
312 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
317 * register knote with new process.
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);
327 kn->kn_fflags |= NOTE_TRACKERR;
330 return (kn->kn_fflags != 0);
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
339 filt_timerexpire(void *knx)
341 struct knote *kn = knx;
342 struct callout *calloutp;
346 lwkt_gettoken(&kq_token);
347 if ((kn->kn_status & KN_DELETING) == 0) {
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);
359 lwkt_reltoken(&kq_token);
363 * data contains amount of time to sleep, in milliseconds
366 filt_timerattach(struct knote *kn)
368 struct callout *calloutp;
372 if (kq_ncallouts >= kq_calloutmax) {
378 tv.tv_sec = kn->kn_sdata / 1000;
379 tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
380 tticks = tvtohz_high(&tv);
382 kn->kn_flags |= EV_CLEAR; /* automatically set */
383 MALLOC(calloutp, struct callout *, sizeof(*calloutp),
385 callout_init(calloutp);
386 kn->kn_hook = (caddr_t)calloutp;
387 callout_reset(calloutp, tticks, filt_timerexpire, kn);
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
399 filt_timerdetach(struct knote *kn)
401 struct callout *calloutp;
403 calloutp = (struct callout *)kn->kn_hook;
404 callout_terminate(calloutp);
405 FREE(calloutp, M_KQUEUE);
410 filt_timer(struct knote *kn, long hint)
413 return (kn->kn_data != 0);
417 * Acquire a knote, return non-zero on success, 0 on failure.
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.
425 knote_acquire(struct knote *kn)
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 */
433 kn->kn_status |= KN_PROCESSING;
438 * Release an acquired knote, clearing KN_PROCESSING and handling any
439 * KN_REPROCESS events.
441 * Non-zero is returned if the knote is destroyed.
445 knote_release(struct knote *kn)
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;
453 if (kn->kn_status & KN_DELETING) {
454 knote_detach_and_drop(kn);
458 if (filter_event(kn, 0))
461 kn->kn_status &= ~KN_PROCESSING;
466 * Initialize a kqueue.
468 * NOTE: The lwp/proc code initializes a kqueue for select/poll ops.
473 kqueue_init(struct kqueue *kq, struct filedesc *fdp)
475 TAILQ_INIT(&kq->kq_knpend);
476 TAILQ_INIT(&kq->kq_knlist);
479 SLIST_INIT(&kq->kq_kqinfo.ki_note);
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).
486 * The kq's knlist must be completely eradicated so block on any
490 kqueue_terminate(struct kqueue *kq)
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);
500 kfree(kq->kq_knhash, M_KQUEUE);
501 kq->kq_knhash = NULL;
502 kq->kq_knhashmask = 0;
504 lwkt_reltoken(&kq_token);
511 sys_kqueue(struct kqueue_args *uap)
513 struct thread *td = curthread;
518 error = falloc(td->td_lwp, &fp, &fd);
521 fp->f_flag = FREAD | FWRITE;
522 fp->f_type = DTYPE_KQUEUE;
523 fp->f_ops = &kqueueops;
525 kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO);
526 kqueue_init(kq, td->td_proc->p_fd);
529 fsetfd(kq->kq_fdp, fp, fd);
530 uap->sysmsg_result = fd;
536 * Copy 'count' items into the destination list pointed to by uap->eventlist.
539 kevent_copyout(void *arg, struct kevent *kevp, int count, int *res)
541 struct kevent_copyin_args *kap;
544 kap = (struct kevent_copyin_args *)arg;
546 error = copyout(kevp, kap->ka->eventlist, count * sizeof(*kevp));
548 kap->ka->eventlist += count;
558 * Copy at most 'max' items from the list pointed to by kap->changelist,
559 * return number of items in 'events'.
562 kevent_copyin(void *arg, struct kevent *kevp, int max, int *events)
564 struct kevent_copyin_args *kap;
567 kap = (struct kevent_copyin_args *)arg;
569 count = min(kap->ka->nchanges - kap->pchanges, max);
570 error = copyin(kap->ka->changelist, kevp, count * sizeof *kevp);
572 kap->ka->changelist += count;
573 kap->pchanges += count;
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)
589 struct timespec *tsp;
590 int i, n, total, error, nerrors = 0;
592 int limit = kq_checkloop;
593 struct kevent kev[KQ_NEVENTS];
599 lwkt_gettoken(&kq_token);
602 error = kevent_copyinfn(uap, kev, KQ_NEVENTS, &n);
607 for (i = 0; i < n; i++) {
609 kevp->flags &= ~EV_SYSFLAGS;
610 error = kqueue_register(kq, kevp);
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.
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
624 kevp->flags = EV_ERROR;
627 kevent_copyoutfn(uap, kevp, 1, res);
630 } else if (lres != *res) {
643 * Acquire/wait for events - setup timeout
648 if (tsp->tv_sec || tsp->tv_nsec) {
650 timespecadd(tsp, &ats); /* tsp = target time */
657 * Collect as many events as we can. Sleeping on successive
658 * loops is disabled if copyoutfn has incremented (*res).
660 * The loop stops if an error occurs, all events have been
661 * scanned (the marker has been reached), or fewer than the
662 * maximum number of events is found.
664 * The copyoutfn function does not have to increment (*res) in
665 * order for the loop to continue.
667 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents.
671 marker.kn_filter = EVFILT_MARKER;
672 marker.kn_status = KN_PROCESSING;
673 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
674 while ((n = nevents - total) > 0) {
679 * If no events are pending sleep until timeout (if any)
680 * or an event occurs.
682 * After the sleep completes the marker is moved to the
683 * end of the list, making any received events available
686 if (kq->kq_count == 0 && *res == 0) {
687 error = kqueue_sleep(kq, tsp);
691 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
692 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
696 * Process all received events
697 * Account for all non-spurious events in our total
699 i = kqueue_scan(kq, kev, n, &marker);
702 error = kevent_copyoutfn(uap, kev, i, res);
703 total += *res - lres;
707 if (limit && --limit == 0)
708 panic("kqueue: checkloop failed i=%d", i);
711 * Normally when fewer events are returned than requested
712 * we can stop. However, if only spurious events were
713 * collected the copyout will not bump (*res) and we have
720 * Deal with an edge case where spurious events can cause
721 * a loop to occur without moving the marker. This can
722 * prevent kqueue_scan() from picking up new events which
723 * race us. We must be sure to move the marker for this
726 * NOTE: We do not want to move the marker if events
727 * were scanned because normal kqueue operations
728 * may reactivate events. Moving the marker in
729 * that case could result in duplicates for the
733 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
734 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe);
737 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe);
739 /* Timeouts do not return EWOULDBLOCK. */
740 if (error == EWOULDBLOCK)
744 lwkt_reltoken(&kq_token);
752 sys_kevent(struct kevent_args *uap)
754 struct thread *td = curthread;
755 struct proc *p = td->td_proc;
756 struct timespec ts, *tsp;
758 struct file *fp = NULL;
759 struct kevent_copyin_args *kap, ka;
763 error = copyin(uap->timeout, &ts, sizeof(ts));
771 fp = holdfp(p->p_fd, uap->fd, -1);
774 if (fp->f_type != DTYPE_KQUEUE) {
779 kq = (struct kqueue *)fp->f_data;
785 error = kern_kevent(kq, uap->nevents, &uap->sysmsg_result, kap,
786 kevent_copyin, kevent_copyout, tsp);
794 kqueue_register(struct kqueue *kq, struct kevent *kev)
796 struct filedesc *fdp = kq->kq_fdp;
797 struct filterops *fops;
798 struct file *fp = NULL;
799 struct knote *kn = NULL;
802 if (kev->filter < 0) {
803 if (kev->filter + EVFILT_SYSCOUNT < 0)
805 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */
809 * filter attach routine is responsible for insuring that
810 * the identifier can be attached to it.
812 kprintf("unknown filter: %d\n", kev->filter);
816 lwkt_gettoken(&kq_token);
817 if (fops->f_flags & FILTEROP_ISFD) {
818 /* validate descriptor */
819 fp = holdfp(fdp, kev->ident, -1);
821 lwkt_reltoken(&kq_token);
826 SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
827 if (kn->kn_kq == kq &&
828 kn->kn_filter == kev->filter &&
829 kn->kn_id == kev->ident) {
830 if (knote_acquire(kn) == 0)
836 if (kq->kq_knhashmask) {
839 list = &kq->kq_knhash[
840 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
842 SLIST_FOREACH(kn, list, kn_link) {
843 if (kn->kn_id == kev->ident &&
844 kn->kn_filter == kev->filter) {
845 if (knote_acquire(kn) == 0)
854 * NOTE: At this point if kn is non-NULL we will have acquired
855 * it and set KN_PROCESSING.
857 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
863 * kn now contains the matching knote, or NULL if no match
865 if (kev->flags & EV_ADD) {
877 * apply reference count to knote structure, and
878 * do not release it at the end of this routine.
882 kn->kn_sfflags = kev->fflags;
883 kn->kn_sdata = kev->data;
886 kn->kn_kevent = *kev;
889 * KN_PROCESSING prevents the knote from getting
890 * ripped out from under us while we are trying
891 * to attach it, in case the attach blocks.
893 kn->kn_status = KN_PROCESSING;
895 if ((error = filter_attach(kn)) != 0) {
896 kn->kn_status |= KN_DELETING | KN_REPROCESS;
902 * Interlock against close races which either tried
903 * to remove our knote while we were blocked or missed
904 * it entirely prior to our attachment. We do not
905 * want to end up with a knote on a closed descriptor.
907 if ((fops->f_flags & FILTEROP_ISFD) &&
908 checkfdclosed(fdp, kev->ident, kn->kn_fp)) {
909 kn->kn_status |= KN_DELETING | KN_REPROCESS;
913 * The user may change some filter values after the
914 * initial EV_ADD, but doing so will not reset any
915 * filter which have already been triggered.
917 KKASSERT(kn->kn_status & KN_PROCESSING);
918 kn->kn_sfflags = kev->fflags;
919 kn->kn_sdata = kev->data;
920 kn->kn_kevent.udata = kev->udata;
924 * Execute the filter event to immediately activate the
925 * knote if necessary. If reprocessing events are pending
926 * due to blocking above we do not run the filter here
927 * but instead let knote_release() do it. Otherwise we
928 * might run the filter on a deleted event.
930 if ((kn->kn_status & KN_REPROCESS) == 0) {
931 if (filter_event(kn, 0))
934 } else if (kev->flags & EV_DELETE) {
936 * Delete the existing knote
938 knote_detach_and_drop(kn);
943 * Disablement does not deactivate a knote here.
945 if ((kev->flags & EV_DISABLE) &&
946 ((kn->kn_status & KN_DISABLED) == 0)) {
947 kn->kn_status |= KN_DISABLED;
951 * Re-enablement may have to immediately enqueue an active knote.
953 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
954 kn->kn_status &= ~KN_DISABLED;
955 if ((kn->kn_status & KN_ACTIVE) &&
956 ((kn->kn_status & KN_QUEUED) == 0)) {
962 * Handle any required reprocessing
965 /* kn may be invalid now */
968 lwkt_reltoken(&kq_token);
975 * Block as necessary until the target time is reached.
976 * If tsp is NULL we block indefinitely. If tsp->ts_secs/nsecs are both
977 * 0 we do not block at all.
980 kqueue_sleep(struct kqueue *kq, struct timespec *tsp)
985 kq->kq_state |= KQ_SLEEP;
986 error = tsleep(kq, PCATCH, "kqread", 0);
987 } else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) {
991 struct timespec atx = *tsp;
995 timespecsub(&atx, &ats);
996 if (ats.tv_sec < 0) {
999 timeout = atx.tv_sec > 24 * 60 * 60 ?
1000 24 * 60 * 60 * hz : tstohz_high(&atx);
1001 kq->kq_state |= KQ_SLEEP;
1002 error = tsleep(kq, PCATCH, "kqread", timeout);
1006 /* don't restart after signals... */
1007 if (error == ERESTART)
1014 * Scan the kqueue, return the number of active events placed in kevp up
1017 * Continuous mode events may get recycled, do not continue scanning past
1018 * marker unless no events have been collected.
1021 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count,
1022 struct knote *marker)
1024 struct knote *kn, local_marker;
1028 local_marker.kn_filter = EVFILT_MARKER;
1029 local_marker.kn_status = KN_PROCESSING;
1034 TAILQ_INSERT_HEAD(&kq->kq_knpend, &local_marker, kn_tqe);
1036 kn = TAILQ_NEXT(&local_marker, kn_tqe);
1037 if (kn->kn_filter == EVFILT_MARKER) {
1038 /* Marker reached, we are done */
1042 /* Move local marker past some other threads marker */
1043 kn = TAILQ_NEXT(kn, kn_tqe);
1044 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe);
1045 TAILQ_INSERT_BEFORE(kn, &local_marker, kn_tqe);
1050 * We can't skip a knote undergoing processing, otherwise
1051 * we risk not returning it when the user process expects
1052 * it should be returned. Sleep and retry.
1054 if (knote_acquire(kn) == 0)
1058 * Remove the event for processing.
1060 * WARNING! We must leave KN_QUEUED set to prevent the
1061 * event from being KNOTE_ACTIVATE()d while
1062 * the queue state is in limbo, in case we
1065 * WARNING! We must set KN_PROCESSING to avoid races
1066 * against deletion or another thread's
1069 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
1073 * We have to deal with an extremely important race against
1074 * file descriptor close()s here. The file descriptor can
1075 * disappear MPSAFE, and there is a small window of
1076 * opportunity between that and the call to knote_fdclose().
1078 * If we hit that window here while doselect or dopoll is
1079 * trying to delete a spurious event they will not be able
1080 * to match up the event against a knote and will go haywire.
1082 if ((kn->kn_fop->f_flags & FILTEROP_ISFD) &&
1083 checkfdclosed(kq->kq_fdp, kn->kn_kevent.ident, kn->kn_fp)) {
1084 kn->kn_status |= KN_DELETING | KN_REPROCESS;
1087 if (kn->kn_status & KN_DISABLED) {
1089 * If disabled we ensure the event is not queued
1090 * but leave its active bit set. On re-enablement
1091 * the event may be immediately triggered.
1093 kn->kn_status &= ~KN_QUEUED;
1094 } else if ((kn->kn_flags & EV_ONESHOT) == 0 &&
1095 (kn->kn_status & KN_DELETING) == 0 &&
1096 filter_event(kn, 0) == 0) {
1098 * If not running in one-shot mode and the event
1099 * is no longer present we ensure it is removed
1100 * from the queue and ignore it.
1102 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1107 *kevp++ = kn->kn_kevent;
1111 if (kn->kn_flags & EV_ONESHOT) {
1112 kn->kn_status &= ~KN_QUEUED;
1113 kn->kn_status |= KN_DELETING | KN_REPROCESS;
1114 } else if (kn->kn_flags & EV_CLEAR) {
1117 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1119 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
1125 * Handle any post-processing states
1129 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe);
1136 * This could be expanded to call kqueue_scan, if desired.
1141 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
1150 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
1159 kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
1160 struct ucred *cred, struct sysmsg *msg)
1165 lwkt_gettoken(&kq_token);
1166 kq = (struct kqueue *)fp->f_data;
1171 kq->kq_state |= KQ_ASYNC;
1173 kq->kq_state &= ~KQ_ASYNC;
1177 error = fsetown(*(int *)data, &kq->kq_sigio);
1183 lwkt_reltoken(&kq_token);
1191 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
1193 struct kqueue *kq = (struct kqueue *)fp->f_data;
1195 bzero((void *)st, sizeof(*st));
1196 st->st_size = kq->kq_count;
1197 st->st_blksize = sizeof(struct kevent);
1198 st->st_mode = S_IFIFO;
1206 kqueue_close(struct file *fp)
1208 struct kqueue *kq = (struct kqueue *)fp->f_data;
1210 kqueue_terminate(kq);
1213 funsetown(&kq->kq_sigio);
1215 kfree(kq, M_KQUEUE);
1220 kqueue_wakeup(struct kqueue *kq)
1222 if (kq->kq_state & KQ_SLEEP) {
1223 kq->kq_state &= ~KQ_SLEEP;
1226 KNOTE(&kq->kq_kqinfo.ki_note, 0);
1230 * Calls filterops f_attach function, acquiring mplock if filter is not
1231 * marked as FILTEROP_MPSAFE.
1234 filter_attach(struct knote *kn)
1238 if (!(kn->kn_fop->f_flags & FILTEROP_MPSAFE)) {
1240 ret = kn->kn_fop->f_attach(kn);
1243 ret = kn->kn_fop->f_attach(kn);
1250 * Detach the knote and drop it, destroying the knote.
1252 * Calls filterops f_detach function, acquiring mplock if filter is not
1253 * marked as FILTEROP_MPSAFE.
1256 knote_detach_and_drop(struct knote *kn)
1258 kn->kn_status |= KN_DELETING | KN_REPROCESS;
1259 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
1260 kn->kn_fop->f_detach(kn);
1263 kn->kn_fop->f_detach(kn);
1270 * Calls filterops f_event function, acquiring mplock if filter is not
1271 * marked as FILTEROP_MPSAFE.
1273 * If the knote is in the middle of being created or deleted we cannot
1274 * safely call the filter op.
1277 filter_event(struct knote *kn, long hint)
1281 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) {
1282 ret = kn->kn_fop->f_event(kn, hint);
1285 ret = kn->kn_fop->f_event(kn, hint);
1292 * Walk down a list of knotes, activating them if their event has triggered.
1294 * If we encounter any knotes which are undergoing processing we just mark
1295 * them for reprocessing and do not try to [re]activate the knote. However,
1296 * if a hint is being passed we have to wait and that makes things a bit
1300 knote(struct klist *list, long hint)
1304 lwkt_gettoken(&kq_token);
1306 SLIST_FOREACH(kn, list, kn_next) {
1307 if (kn->kn_status & KN_PROCESSING) {
1309 * Someone else is processing the knote, ask the
1310 * other thread to reprocess it and don't mess
1311 * with it otherwise.
1314 kn->kn_status |= KN_REPROCESS;
1319 * If the hint is non-zero we have to wait or risk
1320 * losing the state the caller is trying to update.
1322 * XXX This is a real problem, certain process
1323 * and signal filters will bump kn_data for
1324 * already-processed notes more than once if
1325 * we restart the list scan. FIXME.
1327 kn->kn_status |= KN_WAITING | KN_REPROCESS;
1328 tsleep(kn, 0, "knotec", hz);
1333 * Become the reprocessing master ourselves.
1335 * If hint is non-zer running the event is mandatory
1336 * when not deleting so do it whether reprocessing is
1339 kn->kn_status |= KN_PROCESSING;
1340 if ((kn->kn_status & KN_DELETING) == 0) {
1341 if (filter_event(kn, hint))
1344 if (knote_release(kn))
1347 lwkt_reltoken(&kq_token);
1351 * Insert knote at head of klist.
1353 * This function may only be called via a filter function and thus
1354 * kq_token should already be held and marked for processing.
1357 knote_insert(struct klist *klist, struct knote *kn)
1359 KKASSERT(kn->kn_status & KN_PROCESSING);
1360 ASSERT_LWKT_TOKEN_HELD(&kq_token);
1361 SLIST_INSERT_HEAD(klist, kn, kn_next);
1365 * Remove knote from a klist
1367 * This function may only be called via a filter function and thus
1368 * kq_token should already be held and marked for processing.
1371 knote_remove(struct klist *klist, struct knote *kn)
1373 KKASSERT(kn->kn_status & KN_PROCESSING);
1374 ASSERT_LWKT_TOKEN_HELD(&kq_token);
1375 SLIST_REMOVE(klist, kn, knote, kn_next);
1379 * Remove all knotes from a specified klist
1381 * Only called from aio.
1384 knote_empty(struct klist *list)
1388 lwkt_gettoken(&kq_token);
1389 while ((kn = SLIST_FIRST(list)) != NULL) {
1390 if (knote_acquire(kn))
1391 knote_detach_and_drop(kn);
1393 lwkt_reltoken(&kq_token);
1397 knote_assume_knotes(struct kqinfo *src, struct kqinfo *dst,
1398 struct filterops *ops, void *hook)
1402 lwkt_gettoken(&kq_token);
1403 while ((kn = SLIST_FIRST(&src->ki_note)) != NULL) {
1404 if (knote_acquire(kn)) {
1405 knote_remove(&src->ki_note, kn);
1408 knote_insert(&dst->ki_note, kn);
1410 /* kn may be invalid now */
1413 lwkt_reltoken(&kq_token);
1417 * Remove all knotes referencing a specified fd
1420 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd)
1424 lwkt_gettoken(&kq_token);
1426 SLIST_FOREACH(kn, &fp->f_klist, kn_link) {
1427 if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) {
1428 if (knote_acquire(kn))
1429 knote_detach_and_drop(kn);
1433 lwkt_reltoken(&kq_token);
1437 * Low level attach function.
1439 * The knote should already be marked for processing.
1442 knote_attach(struct knote *kn)
1445 struct kqueue *kq = kn->kn_kq;
1447 if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
1448 KKASSERT(kn->kn_fp);
1449 list = &kn->kn_fp->f_klist;
1451 if (kq->kq_knhashmask == 0)
1452 kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1453 &kq->kq_knhashmask);
1454 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1456 SLIST_INSERT_HEAD(list, kn, kn_link);
1457 TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink);
1461 * Low level drop function.
1463 * The knote should already be marked for processing.
1466 knote_drop(struct knote *kn)
1473 if (kn->kn_fop->f_flags & FILTEROP_ISFD)
1474 list = &kn->kn_fp->f_klist;
1476 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1478 SLIST_REMOVE(list, kn, knote, kn_link);
1479 TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink);
1480 if (kn->kn_status & KN_QUEUED)
1482 if (kn->kn_fop->f_flags & FILTEROP_ISFD) {
1490 * Low level enqueue function.
1492 * The knote should already be marked for processing.
1495 knote_enqueue(struct knote *kn)
1497 struct kqueue *kq = kn->kn_kq;
1499 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1500 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe);
1501 kn->kn_status |= KN_QUEUED;
1505 * Send SIGIO on request (typically set up as a mailbox signal)
1507 if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1508 pgsigio(kq->kq_sigio, SIGIO, 0);
1514 * Low level dequeue function.
1516 * The knote should already be marked for processing.
1519 knote_dequeue(struct knote *kn)
1521 struct kqueue *kq = kn->kn_kq;
1523 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1524 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe);
1525 kn->kn_status &= ~KN_QUEUED;
1529 static struct knote *
1532 return kmalloc(sizeof(struct knote), M_KQUEUE, M_WAITOK);
1536 knote_free(struct knote *kn)
1538 kfree(kn, M_KQUEUE);