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_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_stop() (or should), so the
335 * knote should still be a valid structure. However the timeout
336 * can race a deletion so if KN_DELETING is set we just don't touch
340 filt_timerexpire(void *knx
)
342 struct knote
*kn
= knx
;
343 struct callout
*calloutp
;
347 lwkt_gettoken(&kq_token
);
348 if ((kn
->kn_status
& KN_DELETING
) == 0) {
352 if ((kn
->kn_flags
& EV_ONESHOT
) == 0) {
353 tv
.tv_sec
= kn
->kn_sdata
/ 1000;
354 tv
.tv_usec
= (kn
->kn_sdata
% 1000) * 1000;
355 tticks
= tvtohz_high(&tv
);
356 calloutp
= (struct callout
*)kn
->kn_hook
;
357 callout_reset(calloutp
, tticks
, filt_timerexpire
, kn
);
360 lwkt_reltoken(&kq_token
);
364 * data contains amount of time to sleep, in milliseconds
367 filt_timerattach(struct knote
*kn
)
369 struct callout
*calloutp
;
373 if (kq_ncallouts
>= kq_calloutmax
) {
379 tv
.tv_sec
= kn
->kn_sdata
/ 1000;
380 tv
.tv_usec
= (kn
->kn_sdata
% 1000) * 1000;
381 tticks
= tvtohz_high(&tv
);
383 kn
->kn_flags
|= EV_CLEAR
; /* automatically set */
384 MALLOC(calloutp
, struct callout
*, sizeof(*calloutp
),
386 callout_init(calloutp
);
387 kn
->kn_hook
= (caddr_t
)calloutp
;
388 callout_reset(calloutp
, tticks
, filt_timerexpire
, kn
);
394 filt_timerdetach(struct knote
*kn
)
396 struct callout
*calloutp
;
398 calloutp
= (struct callout
*)kn
->kn_hook
;
399 callout_stop(calloutp
);
400 FREE(calloutp
, M_KQUEUE
);
405 filt_timer(struct knote
*kn
, long hint
)
408 return (kn
->kn_data
!= 0);
412 * Acquire a knote, return non-zero on success, 0 on failure.
414 * If we cannot acquire the knote we sleep and return 0. The knote
415 * may be stale on return in this case and the caller must restart
416 * whatever loop they are in.
420 knote_acquire(struct knote
*kn
)
422 if (kn
->kn_status
& KN_PROCESSING
) {
423 kn
->kn_status
|= KN_WAITING
| KN_REPROCESS
;
424 tsleep(kn
, 0, "kqepts", hz
);
425 /* knote may be stale now */
428 kn
->kn_status
|= KN_PROCESSING
;
433 * Release an acquired knote, clearing KN_PROCESSING and handling any
434 * KN_REPROCESS events.
436 * Non-zero is returned if the knote is destroyed.
440 knote_release(struct knote
*kn
)
442 while (kn
->kn_status
& KN_REPROCESS
) {
443 kn
->kn_status
&= ~KN_REPROCESS
;
444 if (kn
->kn_status
& KN_WAITING
) {
445 kn
->kn_status
&= ~KN_WAITING
;
448 if (kn
->kn_status
& KN_DELETING
) {
449 knote_detach_and_drop(kn
);
453 if (filter_event(kn
, 0))
456 kn
->kn_status
&= ~KN_PROCESSING
;
461 * Initialize a kqueue.
463 * NOTE: The lwp/proc code initializes a kqueue for select/poll ops.
468 kqueue_init(struct kqueue
*kq
, struct filedesc
*fdp
)
470 TAILQ_INIT(&kq
->kq_knpend
);
471 TAILQ_INIT(&kq
->kq_knlist
);
474 SLIST_INIT(&kq
->kq_kqinfo
.ki_note
);
478 * Terminate a kqueue. Freeing the actual kq itself is left up to the
479 * caller (it might be embedded in a lwp so we don't do it here).
481 * The kq's knlist must be completely eradicated so block on any
485 kqueue_terminate(struct kqueue
*kq
)
489 lwkt_gettoken(&kq_token
);
490 while ((kn
= TAILQ_FIRST(&kq
->kq_knlist
)) != NULL
) {
491 if (knote_acquire(kn
))
492 knote_detach_and_drop(kn
);
495 kfree(kq
->kq_knhash
, M_KQUEUE
);
496 kq
->kq_knhash
= NULL
;
497 kq
->kq_knhashmask
= 0;
499 lwkt_reltoken(&kq_token
);
506 sys_kqueue(struct kqueue_args
*uap
)
508 struct thread
*td
= curthread
;
513 error
= falloc(td
->td_lwp
, &fp
, &fd
);
516 fp
->f_flag
= FREAD
| FWRITE
;
517 fp
->f_type
= DTYPE_KQUEUE
;
518 fp
->f_ops
= &kqueueops
;
520 kq
= kmalloc(sizeof(struct kqueue
), M_KQUEUE
, M_WAITOK
| M_ZERO
);
521 kqueue_init(kq
, td
->td_proc
->p_fd
);
524 fsetfd(kq
->kq_fdp
, fp
, fd
);
525 uap
->sysmsg_result
= fd
;
531 * Copy 'count' items into the destination list pointed to by uap->eventlist.
534 kevent_copyout(void *arg
, struct kevent
*kevp
, int count
, int *res
)
536 struct kevent_copyin_args
*kap
;
539 kap
= (struct kevent_copyin_args
*)arg
;
541 error
= copyout(kevp
, kap
->ka
->eventlist
, count
* sizeof(*kevp
));
543 kap
->ka
->eventlist
+= count
;
553 * Copy at most 'max' items from the list pointed to by kap->changelist,
554 * return number of items in 'events'.
557 kevent_copyin(void *arg
, struct kevent
*kevp
, int max
, int *events
)
559 struct kevent_copyin_args
*kap
;
562 kap
= (struct kevent_copyin_args
*)arg
;
564 count
= min(kap
->ka
->nchanges
- kap
->pchanges
, max
);
565 error
= copyin(kap
->ka
->changelist
, kevp
, count
* sizeof *kevp
);
567 kap
->ka
->changelist
+= count
;
568 kap
->pchanges
+= count
;
579 kern_kevent(struct kqueue
*kq
, int nevents
, int *res
, void *uap
,
580 k_copyin_fn kevent_copyinfn
, k_copyout_fn kevent_copyoutfn
,
581 struct timespec
*tsp_in
)
584 struct timespec
*tsp
;
585 int i
, n
, total
, error
, nerrors
= 0;
587 int limit
= kq_checkloop
;
588 struct kevent kev
[KQ_NEVENTS
];
594 lwkt_gettoken(&kq_token
);
597 error
= kevent_copyinfn(uap
, kev
, KQ_NEVENTS
, &n
);
602 for (i
= 0; i
< n
; i
++) {
604 kevp
->flags
&= ~EV_SYSFLAGS
;
605 error
= kqueue_register(kq
, kevp
);
608 * If a registration returns an error we
609 * immediately post the error. The kevent()
610 * call itself will fail with the error if
611 * no space is available for posting.
613 * Such errors normally bypass the timeout/blocking
614 * code. However, if the copyoutfn function refuses
615 * to post the error (see sys_poll()), then we
619 kevp
->flags
= EV_ERROR
;
622 kevent_copyoutfn(uap
, kevp
, 1, res
);
636 * Acquire/wait for events - setup timeout
641 if (tsp
->tv_sec
|| tsp
->tv_nsec
) {
643 timespecadd(tsp
, &ats
); /* tsp = target time */
650 * Collect as many events as we can. Sleeping on successive
651 * loops is disabled if copyoutfn has incremented (*res).
653 * The loop stops if an error occurs, all events have been
654 * scanned (the marker has been reached), or fewer than the
655 * maximum number of events is found.
657 * The copyoutfn function does not have to increment (*res) in
658 * order for the loop to continue.
660 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents.
664 marker
.kn_filter
= EVFILT_MARKER
;
665 marker
.kn_status
= KN_PROCESSING
;
666 TAILQ_INSERT_TAIL(&kq
->kq_knpend
, &marker
, kn_tqe
);
667 while ((n
= nevents
- total
) > 0) {
672 * If no events are pending sleep until timeout (if any)
673 * or an event occurs.
675 * After the sleep completes the marker is moved to the
676 * end of the list, making any received events available
679 if (kq
->kq_count
== 0 && *res
== 0) {
680 error
= kqueue_sleep(kq
, tsp
);
684 TAILQ_REMOVE(&kq
->kq_knpend
, &marker
, kn_tqe
);
685 TAILQ_INSERT_TAIL(&kq
->kq_knpend
, &marker
, kn_tqe
);
689 * Process all received events
690 * Account for all non-spurious events in our total
692 i
= kqueue_scan(kq
, kev
, n
, &marker
);
695 error
= kevent_copyoutfn(uap
, kev
, i
, res
);
696 total
+= *res
- lres
;
700 if (limit
&& --limit
== 0)
701 panic("kqueue: checkloop failed i=%d", i
);
704 * Normally when fewer events are returned than requested
705 * we can stop. However, if only spurious events were
706 * collected the copyout will not bump (*res) and we have
713 * Deal with an edge case where spurious events can cause
714 * a loop to occur without moving the marker. This can
715 * prevent kqueue_scan() from picking up new events which
716 * race us. We must be sure to move the marker for this
719 * NOTE: We do not want to move the marker if events
720 * were scanned because normal kqueue operations
721 * may reactivate events. Moving the marker in
722 * that case could result in duplicates for the
726 TAILQ_REMOVE(&kq
->kq_knpend
, &marker
, kn_tqe
);
727 TAILQ_INSERT_TAIL(&kq
->kq_knpend
, &marker
, kn_tqe
);
730 TAILQ_REMOVE(&kq
->kq_knpend
, &marker
, kn_tqe
);
732 /* Timeouts do not return EWOULDBLOCK. */
733 if (error
== EWOULDBLOCK
)
737 lwkt_reltoken(&kq_token
);
745 sys_kevent(struct kevent_args
*uap
)
747 struct thread
*td
= curthread
;
748 struct proc
*p
= td
->td_proc
;
749 struct timespec ts
, *tsp
;
751 struct file
*fp
= NULL
;
752 struct kevent_copyin_args
*kap
, ka
;
756 error
= copyin(uap
->timeout
, &ts
, sizeof(ts
));
764 fp
= holdfp(p
->p_fd
, uap
->fd
, -1);
767 if (fp
->f_type
!= DTYPE_KQUEUE
) {
772 kq
= (struct kqueue
*)fp
->f_data
;
778 error
= kern_kevent(kq
, uap
->nevents
, &uap
->sysmsg_result
, kap
,
779 kevent_copyin
, kevent_copyout
, tsp
);
787 kqueue_register(struct kqueue
*kq
, struct kevent
*kev
)
789 struct filedesc
*fdp
= kq
->kq_fdp
;
790 struct filterops
*fops
;
791 struct file
*fp
= NULL
;
792 struct knote
*kn
= NULL
;
795 if (kev
->filter
< 0) {
796 if (kev
->filter
+ EVFILT_SYSCOUNT
< 0)
798 fops
= sysfilt_ops
[~kev
->filter
]; /* to 0-base index */
802 * filter attach routine is responsible for insuring that
803 * the identifier can be attached to it.
805 kprintf("unknown filter: %d\n", kev
->filter
);
809 lwkt_gettoken(&kq_token
);
810 if (fops
->f_flags
& FILTEROP_ISFD
) {
811 /* validate descriptor */
812 fp
= holdfp(fdp
, kev
->ident
, -1);
814 lwkt_reltoken(&kq_token
);
819 SLIST_FOREACH(kn
, &fp
->f_klist
, kn_link
) {
820 if (kn
->kn_kq
== kq
&&
821 kn
->kn_filter
== kev
->filter
&&
822 kn
->kn_id
== kev
->ident
) {
823 if (knote_acquire(kn
) == 0)
829 if (kq
->kq_knhashmask
) {
832 list
= &kq
->kq_knhash
[
833 KN_HASH((u_long
)kev
->ident
, kq
->kq_knhashmask
)];
835 SLIST_FOREACH(kn
, list
, kn_link
) {
836 if (kn
->kn_id
== kev
->ident
&&
837 kn
->kn_filter
== kev
->filter
) {
838 if (knote_acquire(kn
) == 0)
847 * NOTE: At this point if kn is non-NULL we will have acquired
848 * it and set KN_PROCESSING.
850 if (kn
== NULL
&& ((kev
->flags
& EV_ADD
) == 0)) {
856 * kn now contains the matching knote, or NULL if no match
858 if (kev
->flags
& EV_ADD
) {
870 * apply reference count to knote structure, and
871 * do not release it at the end of this routine.
875 kn
->kn_sfflags
= kev
->fflags
;
876 kn
->kn_sdata
= kev
->data
;
879 kn
->kn_kevent
= *kev
;
882 * KN_PROCESSING prevents the knote from getting
883 * ripped out from under us while we are trying
884 * to attach it, in case the attach blocks.
886 kn
->kn_status
= KN_PROCESSING
;
888 if ((error
= filter_attach(kn
)) != 0) {
889 kn
->kn_status
|= KN_DELETING
| KN_REPROCESS
;
895 * Interlock against close races which either tried
896 * to remove our knote while we were blocked or missed
897 * it entirely prior to our attachment. We do not
898 * want to end up with a knote on a closed descriptor.
900 if ((fops
->f_flags
& FILTEROP_ISFD
) &&
901 checkfdclosed(fdp
, kev
->ident
, kn
->kn_fp
)) {
902 kn
->kn_status
|= KN_DELETING
| KN_REPROCESS
;
906 * The user may change some filter values after the
907 * initial EV_ADD, but doing so will not reset any
908 * filter which have already been triggered.
910 KKASSERT(kn
->kn_status
& KN_PROCESSING
);
911 kn
->kn_sfflags
= kev
->fflags
;
912 kn
->kn_sdata
= kev
->data
;
913 kn
->kn_kevent
.udata
= kev
->udata
;
917 * Execute the filter event to immediately activate the
918 * knote if necessary. If reprocessing events are pending
919 * due to blocking above we do not run the filter here
920 * but instead let knote_release() do it. Otherwise we
921 * might run the filter on a deleted event.
923 if ((kn
->kn_status
& KN_REPROCESS
) == 0) {
924 if (filter_event(kn
, 0))
927 } else if (kev
->flags
& EV_DELETE
) {
929 * Delete the existing knote
931 knote_detach_and_drop(kn
);
936 * Disablement does not deactivate a knote here.
938 if ((kev
->flags
& EV_DISABLE
) &&
939 ((kn
->kn_status
& KN_DISABLED
) == 0)) {
940 kn
->kn_status
|= KN_DISABLED
;
944 * Re-enablement may have to immediately enqueue an active knote.
946 if ((kev
->flags
& EV_ENABLE
) && (kn
->kn_status
& KN_DISABLED
)) {
947 kn
->kn_status
&= ~KN_DISABLED
;
948 if ((kn
->kn_status
& KN_ACTIVE
) &&
949 ((kn
->kn_status
& KN_QUEUED
) == 0)) {
955 * Handle any required reprocessing
958 /* kn may be invalid now */
961 lwkt_reltoken(&kq_token
);
968 * Block as necessary until the target time is reached.
969 * If tsp is NULL we block indefinitely. If tsp->ts_secs/nsecs are both
970 * 0 we do not block at all.
973 kqueue_sleep(struct kqueue
*kq
, struct timespec
*tsp
)
978 kq
->kq_state
|= KQ_SLEEP
;
979 error
= tsleep(kq
, PCATCH
, "kqread", 0);
980 } else if (tsp
->tv_sec
== 0 && tsp
->tv_nsec
== 0) {
984 struct timespec atx
= *tsp
;
988 timespecsub(&atx
, &ats
);
989 if (ats
.tv_sec
< 0) {
992 timeout
= atx
.tv_sec
> 24 * 60 * 60 ?
993 24 * 60 * 60 * hz
: tstohz_high(&atx
);
994 kq
->kq_state
|= KQ_SLEEP
;
995 error
= tsleep(kq
, PCATCH
, "kqread", timeout
);
999 /* don't restart after signals... */
1000 if (error
== ERESTART
)
1007 * Scan the kqueue, return the number of active events placed in kevp up
1010 * Continuous mode events may get recycled, do not continue scanning past
1011 * marker unless no events have been collected.
1014 kqueue_scan(struct kqueue
*kq
, struct kevent
*kevp
, int count
,
1015 struct knote
*marker
)
1017 struct knote
*kn
, local_marker
;
1021 local_marker
.kn_filter
= EVFILT_MARKER
;
1022 local_marker
.kn_status
= KN_PROCESSING
;
1027 TAILQ_INSERT_HEAD(&kq
->kq_knpend
, &local_marker
, kn_tqe
);
1029 kn
= TAILQ_NEXT(&local_marker
, kn_tqe
);
1030 if (kn
->kn_filter
== EVFILT_MARKER
) {
1031 /* Marker reached, we are done */
1035 /* Move local marker past some other threads marker */
1036 kn
= TAILQ_NEXT(kn
, kn_tqe
);
1037 TAILQ_REMOVE(&kq
->kq_knpend
, &local_marker
, kn_tqe
);
1038 TAILQ_INSERT_BEFORE(kn
, &local_marker
, kn_tqe
);
1043 * We can't skip a knote undergoing processing, otherwise
1044 * we risk not returning it when the user process expects
1045 * it should be returned. Sleep and retry.
1047 if (knote_acquire(kn
) == 0)
1051 * Remove the event for processing.
1053 * WARNING! We must leave KN_QUEUED set to prevent the
1054 * event from being KNOTE_ACTIVATE()d while
1055 * the queue state is in limbo, in case we
1058 * WARNING! We must set KN_PROCESSING to avoid races
1059 * against deletion or another thread's
1062 TAILQ_REMOVE(&kq
->kq_knpend
, kn
, kn_tqe
);
1066 * We have to deal with an extremely important race against
1067 * file descriptor close()s here. The file descriptor can
1068 * disappear MPSAFE, and there is a small window of
1069 * opportunity between that and the call to knote_fdclose().
1071 * If we hit that window here while doselect or dopoll is
1072 * trying to delete a spurious event they will not be able
1073 * to match up the event against a knote and will go haywire.
1075 if ((kn
->kn_fop
->f_flags
& FILTEROP_ISFD
) &&
1076 checkfdclosed(kq
->kq_fdp
, kn
->kn_kevent
.ident
, kn
->kn_fp
)) {
1077 kn
->kn_status
|= KN_DELETING
| KN_REPROCESS
;
1080 if (kn
->kn_status
& KN_DISABLED
) {
1082 * If disabled we ensure the event is not queued
1083 * but leave its active bit set. On re-enablement
1084 * the event may be immediately triggered.
1086 kn
->kn_status
&= ~KN_QUEUED
;
1087 } else if ((kn
->kn_flags
& EV_ONESHOT
) == 0 &&
1088 (kn
->kn_status
& KN_DELETING
) == 0 &&
1089 filter_event(kn
, 0) == 0) {
1091 * If not running in one-shot mode and the event
1092 * is no longer present we ensure it is removed
1093 * from the queue and ignore it.
1095 kn
->kn_status
&= ~(KN_QUEUED
| KN_ACTIVE
);
1100 *kevp
++ = kn
->kn_kevent
;
1104 if (kn
->kn_flags
& EV_ONESHOT
) {
1105 kn
->kn_status
&= ~KN_QUEUED
;
1106 kn
->kn_status
|= KN_DELETING
| KN_REPROCESS
;
1107 } else if (kn
->kn_flags
& EV_CLEAR
) {
1110 kn
->kn_status
&= ~(KN_QUEUED
| KN_ACTIVE
);
1112 TAILQ_INSERT_TAIL(&kq
->kq_knpend
, kn
, kn_tqe
);
1118 * Handle any post-processing states
1122 TAILQ_REMOVE(&kq
->kq_knpend
, &local_marker
, kn_tqe
);
1129 * This could be expanded to call kqueue_scan, if desired.
1134 kqueue_read(struct file
*fp
, struct uio
*uio
, struct ucred
*cred
, int flags
)
1143 kqueue_write(struct file
*fp
, struct uio
*uio
, struct ucred
*cred
, int flags
)
1152 kqueue_ioctl(struct file
*fp
, u_long com
, caddr_t data
,
1153 struct ucred
*cred
, struct sysmsg
*msg
)
1158 lwkt_gettoken(&kq_token
);
1159 kq
= (struct kqueue
*)fp
->f_data
;
1164 kq
->kq_state
|= KQ_ASYNC
;
1166 kq
->kq_state
&= ~KQ_ASYNC
;
1170 error
= fsetown(*(int *)data
, &kq
->kq_sigio
);
1176 lwkt_reltoken(&kq_token
);
1184 kqueue_stat(struct file
*fp
, struct stat
*st
, struct ucred
*cred
)
1186 struct kqueue
*kq
= (struct kqueue
*)fp
->f_data
;
1188 bzero((void *)st
, sizeof(*st
));
1189 st
->st_size
= kq
->kq_count
;
1190 st
->st_blksize
= sizeof(struct kevent
);
1191 st
->st_mode
= S_IFIFO
;
1199 kqueue_close(struct file
*fp
)
1201 struct kqueue
*kq
= (struct kqueue
*)fp
->f_data
;
1203 kqueue_terminate(kq
);
1206 funsetown(&kq
->kq_sigio
);
1208 kfree(kq
, M_KQUEUE
);
1213 kqueue_wakeup(struct kqueue
*kq
)
1215 if (kq
->kq_state
& KQ_SLEEP
) {
1216 kq
->kq_state
&= ~KQ_SLEEP
;
1219 KNOTE(&kq
->kq_kqinfo
.ki_note
, 0);
1223 * Calls filterops f_attach function, acquiring mplock if filter is not
1224 * marked as FILTEROP_MPSAFE.
1227 filter_attach(struct knote
*kn
)
1231 if (!(kn
->kn_fop
->f_flags
& FILTEROP_MPSAFE
)) {
1233 ret
= kn
->kn_fop
->f_attach(kn
);
1236 ret
= kn
->kn_fop
->f_attach(kn
);
1243 * Detach the knote and drop it, destroying the knote.
1245 * Calls filterops f_detach function, acquiring mplock if filter is not
1246 * marked as FILTEROP_MPSAFE.
1249 knote_detach_and_drop(struct knote
*kn
)
1251 kn
->kn_status
|= KN_DELETING
| KN_REPROCESS
;
1252 if (kn
->kn_fop
->f_flags
& FILTEROP_MPSAFE
) {
1253 kn
->kn_fop
->f_detach(kn
);
1256 kn
->kn_fop
->f_detach(kn
);
1263 * Calls filterops f_event function, acquiring mplock if filter is not
1264 * marked as FILTEROP_MPSAFE.
1266 * If the knote is in the middle of being created or deleted we cannot
1267 * safely call the filter op.
1270 filter_event(struct knote
*kn
, long hint
)
1274 if (kn
->kn_fop
->f_flags
& FILTEROP_MPSAFE
) {
1275 ret
= kn
->kn_fop
->f_event(kn
, hint
);
1278 ret
= kn
->kn_fop
->f_event(kn
, hint
);
1285 * Walk down a list of knotes, activating them if their event has triggered.
1287 * If we encounter any knotes which are undergoing processing we just mark
1288 * them for reprocessing and do not try to [re]activate the knote. However,
1289 * if a hint is being passed we have to wait and that makes things a bit
1293 knote(struct klist
*list
, long hint
)
1297 lwkt_gettoken(&kq_token
);
1299 SLIST_FOREACH(kn
, list
, kn_next
) {
1300 if (kn
->kn_status
& KN_PROCESSING
) {
1302 * Someone else is processing the knote, ask the
1303 * other thread to reprocess it and don't mess
1304 * with it otherwise.
1307 kn
->kn_status
|= KN_REPROCESS
;
1312 * If the hint is non-zero we have to wait or risk
1313 * losing the state the caller is trying to update.
1315 * XXX This is a real problem, certain process
1316 * and signal filters will bump kn_data for
1317 * already-processed notes more than once if
1318 * we restart the list scan. FIXME.
1320 kn
->kn_status
|= KN_WAITING
| KN_REPROCESS
;
1321 tsleep(kn
, 0, "knotec", hz
);
1326 * Become the reprocessing master ourselves.
1328 * If hint is non-zer running the event is mandatory
1329 * when not deleting so do it whether reprocessing is
1332 kn
->kn_status
|= KN_PROCESSING
;
1333 if ((kn
->kn_status
& KN_DELETING
) == 0) {
1334 if (filter_event(kn
, hint
))
1337 if (knote_release(kn
))
1340 lwkt_reltoken(&kq_token
);
1344 * Insert knote at head of klist.
1346 * This function may only be called via a filter function and thus
1347 * kq_token should already be held and marked for processing.
1350 knote_insert(struct klist
*klist
, struct knote
*kn
)
1352 KKASSERT(kn
->kn_status
& KN_PROCESSING
);
1353 ASSERT_LWKT_TOKEN_HELD(&kq_token
);
1354 SLIST_INSERT_HEAD(klist
, kn
, kn_next
);
1358 * Remove knote from a klist
1360 * This function may only be called via a filter function and thus
1361 * kq_token should already be held and marked for processing.
1364 knote_remove(struct klist
*klist
, struct knote
*kn
)
1366 KKASSERT(kn
->kn_status
& KN_PROCESSING
);
1367 ASSERT_LWKT_TOKEN_HELD(&kq_token
);
1368 SLIST_REMOVE(klist
, kn
, knote
, kn_next
);
1372 * Remove all knotes from a specified klist
1374 * Only called from aio.
1377 knote_empty(struct klist
*list
)
1381 lwkt_gettoken(&kq_token
);
1382 while ((kn
= SLIST_FIRST(list
)) != NULL
) {
1383 if (knote_acquire(kn
))
1384 knote_detach_and_drop(kn
);
1386 lwkt_reltoken(&kq_token
);
1390 knote_assume_knotes(struct kqinfo
*src
, struct kqinfo
*dst
,
1391 struct filterops
*ops
, void *hook
)
1395 lwkt_gettoken(&kq_token
);
1396 while ((kn
= SLIST_FIRST(&src
->ki_note
)) != NULL
) {
1397 if (knote_acquire(kn
)) {
1398 knote_remove(&src
->ki_note
, kn
);
1401 knote_insert(&dst
->ki_note
, kn
);
1403 /* kn may be invalid now */
1406 lwkt_reltoken(&kq_token
);
1410 * Remove all knotes referencing a specified fd
1413 knote_fdclose(struct file
*fp
, struct filedesc
*fdp
, int fd
)
1417 lwkt_gettoken(&kq_token
);
1419 SLIST_FOREACH(kn
, &fp
->f_klist
, kn_link
) {
1420 if (kn
->kn_kq
->kq_fdp
== fdp
&& kn
->kn_id
== fd
) {
1421 if (knote_acquire(kn
))
1422 knote_detach_and_drop(kn
);
1426 lwkt_reltoken(&kq_token
);
1430 * Low level attach function.
1432 * The knote should already be marked for processing.
1435 knote_attach(struct knote
*kn
)
1438 struct kqueue
*kq
= kn
->kn_kq
;
1440 if (kn
->kn_fop
->f_flags
& FILTEROP_ISFD
) {
1441 KKASSERT(kn
->kn_fp
);
1442 list
= &kn
->kn_fp
->f_klist
;
1444 if (kq
->kq_knhashmask
== 0)
1445 kq
->kq_knhash
= hashinit(KN_HASHSIZE
, M_KQUEUE
,
1446 &kq
->kq_knhashmask
);
1447 list
= &kq
->kq_knhash
[KN_HASH(kn
->kn_id
, kq
->kq_knhashmask
)];
1449 SLIST_INSERT_HEAD(list
, kn
, kn_link
);
1450 TAILQ_INSERT_HEAD(&kq
->kq_knlist
, kn
, kn_kqlink
);
1454 * Low level drop function.
1456 * The knote should already be marked for processing.
1459 knote_drop(struct knote
*kn
)
1466 if (kn
->kn_fop
->f_flags
& FILTEROP_ISFD
)
1467 list
= &kn
->kn_fp
->f_klist
;
1469 list
= &kq
->kq_knhash
[KN_HASH(kn
->kn_id
, kq
->kq_knhashmask
)];
1471 SLIST_REMOVE(list
, kn
, knote
, kn_link
);
1472 TAILQ_REMOVE(&kq
->kq_knlist
, kn
, kn_kqlink
);
1473 if (kn
->kn_status
& KN_QUEUED
)
1475 if (kn
->kn_fop
->f_flags
& FILTEROP_ISFD
) {
1483 * Low level enqueue function.
1485 * The knote should already be marked for processing.
1488 knote_enqueue(struct knote
*kn
)
1490 struct kqueue
*kq
= kn
->kn_kq
;
1492 KASSERT((kn
->kn_status
& KN_QUEUED
) == 0, ("knote already queued"));
1493 TAILQ_INSERT_TAIL(&kq
->kq_knpend
, kn
, kn_tqe
);
1494 kn
->kn_status
|= KN_QUEUED
;
1498 * Send SIGIO on request (typically set up as a mailbox signal)
1500 if (kq
->kq_sigio
&& (kq
->kq_state
& KQ_ASYNC
) && kq
->kq_count
== 1)
1501 pgsigio(kq
->kq_sigio
, SIGIO
, 0);
1507 * Low level dequeue function.
1509 * The knote should already be marked for processing.
1512 knote_dequeue(struct knote
*kn
)
1514 struct kqueue
*kq
= kn
->kn_kq
;
1516 KASSERT(kn
->kn_status
& KN_QUEUED
, ("knote not queued"));
1517 TAILQ_REMOVE(&kq
->kq_knpend
, kn
, kn_tqe
);
1518 kn
->kn_status
&= ~KN_QUEUED
;
1522 static struct knote
*
1525 return kmalloc(sizeof(struct knote
), M_KQUEUE
, M_WAITOK
);
1529 knote_free(struct knote
*kn
)
1531 kfree(kn
, M_KQUEUE
);