FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_event.c
1 /*-
2 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
3 * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org>
4 * Copyright (c) 2009 Apple, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31
32 #include "opt_ktrace.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/proc.h>
40 #include <sys/malloc.h>
41 #include <sys/unistd.h>
42 #include <sys/file.h>
43 #include <sys/filedesc.h>
44 #include <sys/filio.h>
45 #include <sys/fcntl.h>
46 #include <sys/kthread.h>
47 #include <sys/selinfo.h>
48 #include <sys/queue.h>
49 #include <sys/event.h>
50 #include <sys/eventvar.h>
51 #include <sys/poll.h>
52 #include <sys/protosw.h>
53 #include <sys/sigio.h>
54 #include <sys/signalvar.h>
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/stat.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysproto.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/taskqueue.h>
62 #include <sys/uio.h>
63 #ifdef KTRACE
64 #include <sys/ktrace.h>
65 #endif
66
67 #include <vm/uma.h>
68
69 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
70
71 /*
72 * This lock is used if multiple kq locks are required. This possibly
73 * should be made into a per proc lock.
74 */
75 static struct mtx kq_global;
76 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
77 #define KQ_GLOBAL_LOCK(lck, haslck) do { \
78 if (!haslck) \
79 mtx_lock(lck); \
80 haslck = 1; \
81 } while (0)
82 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \
83 if (haslck) \
84 mtx_unlock(lck); \
85 haslck = 0; \
86 } while (0)
87
88 TASKQUEUE_DEFINE_THREAD(kqueue);
89
90 static int kevent_copyout(void *arg, struct kevent *kevp, int count);
91 static int kevent_copyin(void *arg, struct kevent *kevp, int count);
92 static int kqueue_register(struct kqueue *kq, struct kevent *kev,
93 struct thread *td, int waitok);
94 static int kqueue_acquire(struct file *fp, struct kqueue **kqp);
95 static void kqueue_release(struct kqueue *kq, int locked);
96 static int kqueue_expand(struct kqueue *kq, struct filterops *fops,
97 uintptr_t ident, int waitok);
98 static void kqueue_task(void *arg, int pending);
99 static int kqueue_scan(struct kqueue *kq, int maxevents,
100 struct kevent_copyops *k_ops,
101 const struct timespec *timeout,
102 struct kevent *keva, struct thread *td);
103 static void kqueue_wakeup(struct kqueue *kq);
104 static struct filterops *kqueue_fo_find(int filt);
105 static void kqueue_fo_release(int filt);
106
107 static fo_rdwr_t kqueue_read;
108 static fo_rdwr_t kqueue_write;
109 static fo_truncate_t kqueue_truncate;
110 static fo_ioctl_t kqueue_ioctl;
111 static fo_poll_t kqueue_poll;
112 static fo_kqfilter_t kqueue_kqfilter;
113 static fo_stat_t kqueue_stat;
114 static fo_close_t kqueue_close;
115
116 static struct fileops kqueueops = {
117 .fo_read = kqueue_read,
118 .fo_write = kqueue_write,
119 .fo_truncate = kqueue_truncate,
120 .fo_ioctl = kqueue_ioctl,
121 .fo_poll = kqueue_poll,
122 .fo_kqfilter = kqueue_kqfilter,
123 .fo_stat = kqueue_stat,
124 .fo_close = kqueue_close,
125 };
126
127 static int knote_attach(struct knote *kn, struct kqueue *kq);
128 static void knote_drop(struct knote *kn, struct thread *td);
129 static void knote_enqueue(struct knote *kn);
130 static void knote_dequeue(struct knote *kn);
131 static void knote_init(void);
132 static struct knote *knote_alloc(int waitok);
133 static void knote_free(struct knote *kn);
134
135 static void filt_kqdetach(struct knote *kn);
136 static int filt_kqueue(struct knote *kn, long hint);
137 static int filt_procattach(struct knote *kn);
138 static void filt_procdetach(struct knote *kn);
139 static int filt_proc(struct knote *kn, long hint);
140 static int filt_fileattach(struct knote *kn);
141 static void filt_timerexpire(void *knx);
142 static int filt_timerattach(struct knote *kn);
143 static void filt_timerdetach(struct knote *kn);
144 static int filt_timer(struct knote *kn, long hint);
145 static int filt_userattach(struct knote *kn);
146 static void filt_userdetach(struct knote *kn);
147 static int filt_user(struct knote *kn, long hint);
148 static void filt_usertouch(struct knote *kn, struct kevent *kev,
149 u_long type);
150
151 static struct filterops file_filtops =
152 { 1, filt_fileattach, NULL, NULL };
153 static struct filterops kqread_filtops =
154 { 1, NULL, filt_kqdetach, filt_kqueue };
155 /* XXX - move to kern_proc.c? */
156 static struct filterops proc_filtops =
157 { 0, filt_procattach, filt_procdetach, filt_proc };
158 static struct filterops timer_filtops =
159 { 0, filt_timerattach, filt_timerdetach, filt_timer };
160 static struct filterops user_filtops = {
161 .f_attach = filt_userattach,
162 .f_detach = filt_userdetach,
163 .f_event = filt_user,
164 .f_touch = filt_usertouch,
165 };
166
167 static uma_zone_t knote_zone;
168 static int kq_ncallouts = 0;
169 static int kq_calloutmax = (4 * 1024);
170 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
171 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
172
173 /* XXX - ensure not KN_INFLUX?? */
174 #define KNOTE_ACTIVATE(kn, islock) do { \
175 if ((islock)) \
176 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \
177 else \
178 KQ_LOCK((kn)->kn_kq); \
179 (kn)->kn_status |= KN_ACTIVE; \
180 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
181 knote_enqueue((kn)); \
182 if (!(islock)) \
183 KQ_UNLOCK((kn)->kn_kq); \
184 } while(0)
185 #define KQ_LOCK(kq) do { \
186 mtx_lock(&(kq)->kq_lock); \
187 } while (0)
188 #define KQ_FLUX_WAKEUP(kq) do { \
189 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
190 (kq)->kq_state &= ~KQ_FLUXWAIT; \
191 wakeup((kq)); \
192 } \
193 } while (0)
194 #define KQ_UNLOCK_FLUX(kq) do { \
195 KQ_FLUX_WAKEUP(kq); \
196 mtx_unlock(&(kq)->kq_lock); \
197 } while (0)
198 #define KQ_UNLOCK(kq) do { \
199 mtx_unlock(&(kq)->kq_lock); \
200 } while (0)
201 #define KQ_OWNED(kq) do { \
202 mtx_assert(&(kq)->kq_lock, MA_OWNED); \
203 } while (0)
204 #define KQ_NOTOWNED(kq) do { \
205 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \
206 } while (0)
207 #define KN_LIST_LOCK(kn) do { \
208 if (kn->kn_knlist != NULL) \
209 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \
210 } while (0)
211 #define KN_LIST_UNLOCK(kn) do { \
212 if (kn->kn_knlist != NULL) \
213 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \
214 } while (0)
215 #define KNL_ASSERT_LOCK(knl, islocked) do { \
216 if (islocked) \
217 KNL_ASSERT_LOCKED(knl); \
218 else \
219 KNL_ASSERT_UNLOCKED(knl); \
220 } while (0)
221 #ifdef INVARIANTS
222 #define KNL_ASSERT_LOCKED(knl) do { \
223 knl->kl_assert_locked((knl)->kl_lockarg); \
224 } while (0)
225 #define KNL_ASSERT_UNLOCKED(knl) do { \
226 knl->kl_assert_unlocked((knl)->kl_lockarg); \
227 } while (0)
228 #else /* !INVARIANTS */
229 #define KNL_ASSERT_LOCKED(knl) do {} while(0)
230 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0)
231 #endif /* INVARIANTS */
232
233 #define KN_HASHSIZE 64 /* XXX should be tunable */
234 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
235
236 static int
237 filt_nullattach(struct knote *kn)
238 {
239
240 return (ENXIO);
241 };
242
243 struct filterops null_filtops =
244 { 0, filt_nullattach, NULL, NULL };
245
246 /* XXX - make SYSINIT to add these, and move into respective modules. */
247 extern struct filterops sig_filtops;
248 extern struct filterops fs_filtops;
249
250 /*
251 * Table for for all system-defined filters.
252 */
253 static struct mtx filterops_lock;
254 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
255 MTX_DEF);
256 static struct {
257 struct filterops *for_fop;
258 int for_refcnt;
259 } sysfilt_ops[EVFILT_SYSCOUNT] = {
260 { &file_filtops }, /* EVFILT_READ */
261 { &file_filtops }, /* EVFILT_WRITE */
262 { &null_filtops }, /* EVFILT_AIO */
263 { &file_filtops }, /* EVFILT_VNODE */
264 { &proc_filtops }, /* EVFILT_PROC */
265 { &sig_filtops }, /* EVFILT_SIGNAL */
266 { &timer_filtops }, /* EVFILT_TIMER */
267 { &null_filtops }, /* former EVFILT_NETDEV */
268 { &fs_filtops }, /* EVFILT_FS */
269 { &null_filtops }, /* EVFILT_LIO */
270 { &user_filtops }, /* EVFILT_USER */
271 };
272
273 /*
274 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
275 * method.
276 */
277 static int
278 filt_fileattach(struct knote *kn)
279 {
280
281 return (fo_kqfilter(kn->kn_fp, kn));
282 }
283
284 /*ARGSUSED*/
285 static int
286 kqueue_kqfilter(struct file *fp, struct knote *kn)
287 {
288 struct kqueue *kq = kn->kn_fp->f_data;
289
290 if (kn->kn_filter != EVFILT_READ)
291 return (EINVAL);
292
293 kn->kn_status |= KN_KQUEUE;
294 kn->kn_fop = &kqread_filtops;
295 knlist_add(&kq->kq_sel.si_note, kn, 0);
296
297 return (0);
298 }
299
300 static void
301 filt_kqdetach(struct knote *kn)
302 {
303 struct kqueue *kq = kn->kn_fp->f_data;
304
305 knlist_remove(&kq->kq_sel.si_note, kn, 0);
306 }
307
308 /*ARGSUSED*/
309 static int
310 filt_kqueue(struct knote *kn, long hint)
311 {
312 struct kqueue *kq = kn->kn_fp->f_data;
313
314 kn->kn_data = kq->kq_count;
315 return (kn->kn_data > 0);
316 }
317
318 /* XXX - move to kern_proc.c? */
319 static int
320 filt_procattach(struct knote *kn)
321 {
322 struct proc *p;
323 int immediate;
324 int error;
325
326 immediate = 0;
327 p = pfind(kn->kn_id);
328 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
329 p = zpfind(kn->kn_id);
330 immediate = 1;
331 } else if (p != NULL && (p->p_flag & P_WEXIT)) {
332 immediate = 1;
333 }
334
335 if (p == NULL)
336 return (ESRCH);
337 if ((error = p_cansee(curthread, p))) {
338 PROC_UNLOCK(p);
339 return (error);
340 }
341
342 kn->kn_ptr.p_proc = p;
343 kn->kn_flags |= EV_CLEAR; /* automatically set */
344
345 /*
346 * internal flag indicating registration done by kernel
347 */
348 if (kn->kn_flags & EV_FLAG1) {
349 kn->kn_data = kn->kn_sdata; /* ppid */
350 kn->kn_fflags = NOTE_CHILD;
351 kn->kn_flags &= ~EV_FLAG1;
352 }
353
354 if (immediate == 0)
355 knlist_add(&p->p_klist, kn, 1);
356
357 /*
358 * Immediately activate any exit notes if the target process is a
359 * zombie. This is necessary to handle the case where the target
360 * process, e.g. a child, dies before the kevent is registered.
361 */
362 if (immediate && filt_proc(kn, NOTE_EXIT))
363 KNOTE_ACTIVATE(kn, 0);
364
365 PROC_UNLOCK(p);
366
367 return (0);
368 }
369
370 /*
371 * The knote may be attached to a different process, which may exit,
372 * leaving nothing for the knote to be attached to. So when the process
373 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
374 * it will be deleted when read out. However, as part of the knote deletion,
375 * this routine is called, so a check is needed to avoid actually performing
376 * a detach, because the original process does not exist any more.
377 */
378 /* XXX - move to kern_proc.c? */
379 static void
380 filt_procdetach(struct knote *kn)
381 {
382 struct proc *p;
383
384 p = kn->kn_ptr.p_proc;
385 knlist_remove(&p->p_klist, kn, 0);
386 kn->kn_ptr.p_proc = NULL;
387 }
388
389 /* XXX - move to kern_proc.c? */
390 static int
391 filt_proc(struct knote *kn, long hint)
392 {
393 struct proc *p = kn->kn_ptr.p_proc;
394 u_int event;
395
396 /*
397 * mask off extra data
398 */
399 event = (u_int)hint & NOTE_PCTRLMASK;
400
401 /*
402 * if the user is interested in this event, record it.
403 */
404 if (kn->kn_sfflags & event)
405 kn->kn_fflags |= event;
406
407 /*
408 * process is gone, so flag the event as finished.
409 */
410 if (event == NOTE_EXIT) {
411 if (!(kn->kn_status & KN_DETACHED))
412 knlist_remove_inevent(&p->p_klist, kn);
413 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
414 kn->kn_ptr.p_proc = NULL;
415 if (kn->kn_fflags & NOTE_EXIT)
416 kn->kn_data = p->p_xstat;
417 if (kn->kn_fflags == 0)
418 kn->kn_flags |= EV_DROP;
419 return (1);
420 }
421
422 return (kn->kn_fflags != 0);
423 }
424
425 /*
426 * Called when the process forked. It mostly does the same as the
427 * knote(), activating all knotes registered to be activated when the
428 * process forked. Additionally, for each knote attached to the
429 * parent, check whether user wants to track the new process. If so
430 * attach a new knote to it, and immediately report an event with the
431 * child's pid.
432 */
433 void
434 knote_fork(struct knlist *list, int pid)
435 {
436 struct kqueue *kq;
437 struct knote *kn;
438 struct kevent kev;
439 int error;
440
441 if (list == NULL)
442 return;
443 list->kl_lock(list->kl_lockarg);
444
445 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
446 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
447 continue;
448 kq = kn->kn_kq;
449 KQ_LOCK(kq);
450 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
451 KQ_UNLOCK(kq);
452 continue;
453 }
454
455 /*
456 * The same as knote(), activate the event.
457 */
458 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
459 kn->kn_status |= KN_HASKQLOCK;
460 if (kn->kn_fop->f_event(kn, NOTE_FORK))
461 KNOTE_ACTIVATE(kn, 1);
462 kn->kn_status &= ~KN_HASKQLOCK;
463 KQ_UNLOCK(kq);
464 continue;
465 }
466
467 /*
468 * The NOTE_TRACK case. In addition to the activation
469 * of the event, we need to register new event to
470 * track the child. Drop the locks in preparation for
471 * the call to kqueue_register().
472 */
473 kn->kn_status |= KN_INFLUX;
474 KQ_UNLOCK(kq);
475 list->kl_unlock(list->kl_lockarg);
476
477 /*
478 * Activate existing knote and register a knote with
479 * new process.
480 */
481 kev.ident = pid;
482 kev.filter = kn->kn_filter;
483 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
484 kev.fflags = kn->kn_sfflags;
485 kev.data = kn->kn_id; /* parent */
486 kev.udata = kn->kn_kevent.udata;/* preserve udata */
487 error = kqueue_register(kq, &kev, NULL, 0);
488 if (error)
489 kn->kn_fflags |= NOTE_TRACKERR;
490 if (kn->kn_fop->f_event(kn, NOTE_FORK))
491 KNOTE_ACTIVATE(kn, 0);
492 KQ_LOCK(kq);
493 kn->kn_status &= ~KN_INFLUX;
494 KQ_UNLOCK_FLUX(kq);
495 list->kl_lock(list->kl_lockarg);
496 }
497 list->kl_unlock(list->kl_lockarg);
498 }
499
500 /*
501 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
502 * interval timer support code.
503 */
504 static int
505 timertoticks(intptr_t data)
506 {
507 struct timeval tv;
508 int tticks;
509
510 tv.tv_sec = data / 1000;
511 tv.tv_usec = (data % 1000) * 1000;
512 tticks = tvtohz(&tv);
513
514 return tticks;
515 }
516
517 static void
518 filt_timerexpire(void *knx)
519 {
520 struct knote *kn = knx;
521 struct callout *calloutp;
522
523 kn->kn_data++;
524 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
525
526 /*
527 * timertoticks() uses tvtohz() which always adds 1 to allow
528 * for the time until the next clock interrupt being strictly
529 * less than 1 clock tick. We don't want that here since we
530 * want to appear to be in sync with the clock interrupt even
531 * when we're delayed.
532 */
533 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
534 calloutp = (struct callout *)kn->kn_hook;
535 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata) - 1,
536 filt_timerexpire, kn);
537 }
538 }
539
540 /*
541 * data contains amount of time to sleep, in milliseconds
542 */
543 static int
544 filt_timerattach(struct knote *kn)
545 {
546 struct callout *calloutp;
547
548 atomic_add_int(&kq_ncallouts, 1);
549
550 if (kq_ncallouts >= kq_calloutmax) {
551 atomic_add_int(&kq_ncallouts, -1);
552 return (ENOMEM);
553 }
554
555 kn->kn_flags |= EV_CLEAR; /* automatically set */
556 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */
557 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK);
558 callout_init(calloutp, CALLOUT_MPSAFE);
559 kn->kn_hook = calloutp;
560 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata),
561 filt_timerexpire, kn);
562
563 return (0);
564 }
565
566 static void
567 filt_timerdetach(struct knote *kn)
568 {
569 struct callout *calloutp;
570
571 calloutp = (struct callout *)kn->kn_hook;
572 callout_drain(calloutp);
573 free(calloutp, M_KQUEUE);
574 atomic_add_int(&kq_ncallouts, -1);
575 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */
576 }
577
578 static int
579 filt_timer(struct knote *kn, long hint)
580 {
581
582 return (kn->kn_data != 0);
583 }
584
585 static int
586 filt_userattach(struct knote *kn)
587 {
588
589 /*
590 * EVFILT_USER knotes are not attached to anything in the kernel.
591 */
592 kn->kn_hook = NULL;
593 if (kn->kn_fflags & NOTE_TRIGGER)
594 kn->kn_hookid = 1;
595 else
596 kn->kn_hookid = 0;
597 return (0);
598 }
599
600 static void
601 filt_userdetach(__unused struct knote *kn)
602 {
603
604 /*
605 * EVFILT_USER knotes are not attached to anything in the kernel.
606 */
607 }
608
609 static int
610 filt_user(struct knote *kn, __unused long hint)
611 {
612
613 return (kn->kn_hookid);
614 }
615
616 static void
617 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
618 {
619 u_int ffctrl;
620
621 switch (type) {
622 case EVENT_REGISTER:
623 if (kev->fflags & NOTE_TRIGGER)
624 kn->kn_hookid = 1;
625
626 ffctrl = kev->fflags & NOTE_FFCTRLMASK;
627 kev->fflags &= NOTE_FFLAGSMASK;
628 switch (ffctrl) {
629 case NOTE_FFNOP:
630 break;
631
632 case NOTE_FFAND:
633 kn->kn_sfflags &= kev->fflags;
634 break;
635
636 case NOTE_FFOR:
637 kn->kn_sfflags |= kev->fflags;
638 break;
639
640 case NOTE_FFCOPY:
641 kn->kn_sfflags = kev->fflags;
642 break;
643
644 default:
645 /* XXX Return error? */
646 break;
647 }
648 kn->kn_sdata = kev->data;
649 if (kev->flags & EV_CLEAR) {
650 kn->kn_hookid = 0;
651 kn->kn_data = 0;
652 kn->kn_fflags = 0;
653 }
654 break;
655
656 case EVENT_PROCESS:
657 *kev = kn->kn_kevent;
658 kev->fflags = kn->kn_sfflags;
659 kev->data = kn->kn_sdata;
660 if (kn->kn_flags & EV_CLEAR) {
661 kn->kn_hookid = 0;
662 kn->kn_data = 0;
663 kn->kn_fflags = 0;
664 }
665 break;
666
667 default:
668 panic("filt_usertouch() - invalid type (%ld)", type);
669 break;
670 }
671 }
672
673 int
674 kqueue(struct thread *td, struct kqueue_args *uap)
675 {
676 struct filedesc *fdp;
677 struct kqueue *kq;
678 struct file *fp;
679 int fd, error;
680
681 fdp = td->td_proc->p_fd;
682 error = falloc(td, &fp, &fd);
683 if (error)
684 goto done2;
685
686 /* An extra reference on `nfp' has been held for us by falloc(). */
687 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
688 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
689 TAILQ_INIT(&kq->kq_head);
690 kq->kq_fdp = fdp;
691 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
692 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
693
694 FILEDESC_XLOCK(fdp);
695 SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
696 FILEDESC_XUNLOCK(fdp);
697
698 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
699 fdrop(fp, td);
700
701 td->td_retval[0] = fd;
702 done2:
703 return (error);
704 }
705
706 #ifndef _SYS_SYSPROTO_H_
707 struct kevent_args {
708 int fd;
709 const struct kevent *changelist;
710 int nchanges;
711 struct kevent *eventlist;
712 int nevents;
713 const struct timespec *timeout;
714 };
715 #endif
716 int
717 kevent(struct thread *td, struct kevent_args *uap)
718 {
719 struct timespec ts, *tsp;
720 struct kevent_copyops k_ops = { uap,
721 kevent_copyout,
722 kevent_copyin};
723 int error;
724 #ifdef KTRACE
725 struct uio ktruio;
726 struct iovec ktriov;
727 struct uio *ktruioin = NULL;
728 struct uio *ktruioout = NULL;
729 #endif
730
731 if (uap->timeout != NULL) {
732 error = copyin(uap->timeout, &ts, sizeof(ts));
733 if (error)
734 return (error);
735 tsp = &ts;
736 } else
737 tsp = NULL;
738
739 #ifdef KTRACE
740 if (KTRPOINT(td, KTR_GENIO)) {
741 ktriov.iov_base = uap->changelist;
742 ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
743 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
744 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
745 .uio_td = td };
746 ktruioin = cloneuio(&ktruio);
747 ktriov.iov_base = uap->eventlist;
748 ktriov.iov_len = uap->nevents * sizeof(struct kevent);
749 ktruioout = cloneuio(&ktruio);
750 }
751 #endif
752
753 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
754 &k_ops, tsp);
755
756 #ifdef KTRACE
757 if (ktruioin != NULL) {
758 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
759 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
760 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
761 ktrgenio(uap->fd, UIO_READ, ktruioout, error);
762 }
763 #endif
764
765 return (error);
766 }
767
768 /*
769 * Copy 'count' items into the destination list pointed to by uap->eventlist.
770 */
771 static int
772 kevent_copyout(void *arg, struct kevent *kevp, int count)
773 {
774 struct kevent_args *uap;
775 int error;
776
777 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
778 uap = (struct kevent_args *)arg;
779
780 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
781 if (error == 0)
782 uap->eventlist += count;
783 return (error);
784 }
785
786 /*
787 * Copy 'count' items from the list pointed to by uap->changelist.
788 */
789 static int
790 kevent_copyin(void *arg, struct kevent *kevp, int count)
791 {
792 struct kevent_args *uap;
793 int error;
794
795 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
796 uap = (struct kevent_args *)arg;
797
798 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
799 if (error == 0)
800 uap->changelist += count;
801 return (error);
802 }
803
804 int
805 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
806 struct kevent_copyops *k_ops, const struct timespec *timeout)
807 {
808 struct kevent keva[KQ_NEVENTS];
809 struct kevent *kevp, *changes;
810 struct kqueue *kq;
811 struct file *fp;
812 int i, n, nerrors, error;
813
814 if ((error = fget(td, fd, &fp)) != 0)
815 return (error);
816 if ((error = kqueue_acquire(fp, &kq)) != 0)
817 goto done_norel;
818
819 nerrors = 0;
820
821 while (nchanges > 0) {
822 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
823 error = k_ops->k_copyin(k_ops->arg, keva, n);
824 if (error)
825 goto done;
826 changes = keva;
827 for (i = 0; i < n; i++) {
828 kevp = &changes[i];
829 if (!kevp->filter)
830 continue;
831 kevp->flags &= ~EV_SYSFLAGS;
832 error = kqueue_register(kq, kevp, td, 1);
833 if (error || (kevp->flags & EV_RECEIPT)) {
834 if (nevents != 0) {
835 kevp->flags = EV_ERROR;
836 kevp->data = error;
837 (void) k_ops->k_copyout(k_ops->arg,
838 kevp, 1);
839 nevents--;
840 nerrors++;
841 } else {
842 goto done;
843 }
844 }
845 }
846 nchanges -= n;
847 }
848 if (nerrors) {
849 td->td_retval[0] = nerrors;
850 error = 0;
851 goto done;
852 }
853
854 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
855 done:
856 kqueue_release(kq, 0);
857 done_norel:
858 fdrop(fp, td);
859 return (error);
860 }
861
862 int
863 kqueue_add_filteropts(int filt, struct filterops *filtops)
864 {
865 int error;
866
867 error = 0;
868 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
869 printf(
870 "trying to add a filterop that is out of range: %d is beyond %d\n",
871 ~filt, EVFILT_SYSCOUNT);
872 return EINVAL;
873 }
874 mtx_lock(&filterops_lock);
875 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
876 sysfilt_ops[~filt].for_fop != NULL)
877 error = EEXIST;
878 else {
879 sysfilt_ops[~filt].for_fop = filtops;
880 sysfilt_ops[~filt].for_refcnt = 0;
881 }
882 mtx_unlock(&filterops_lock);
883
884 return (error);
885 }
886
887 int
888 kqueue_del_filteropts(int filt)
889 {
890 int error;
891
892 error = 0;
893 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
894 return EINVAL;
895
896 mtx_lock(&filterops_lock);
897 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
898 sysfilt_ops[~filt].for_fop == NULL)
899 error = EINVAL;
900 else if (sysfilt_ops[~filt].for_refcnt != 0)
901 error = EBUSY;
902 else {
903 sysfilt_ops[~filt].for_fop = &null_filtops;
904 sysfilt_ops[~filt].for_refcnt = 0;
905 }
906 mtx_unlock(&filterops_lock);
907
908 return error;
909 }
910
911 static struct filterops *
912 kqueue_fo_find(int filt)
913 {
914
915 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
916 return NULL;
917
918 mtx_lock(&filterops_lock);
919 sysfilt_ops[~filt].for_refcnt++;
920 if (sysfilt_ops[~filt].for_fop == NULL)
921 sysfilt_ops[~filt].for_fop = &null_filtops;
922 mtx_unlock(&filterops_lock);
923
924 return sysfilt_ops[~filt].for_fop;
925 }
926
927 static void
928 kqueue_fo_release(int filt)
929 {
930
931 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
932 return;
933
934 mtx_lock(&filterops_lock);
935 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
936 ("filter object refcount not valid on release"));
937 sysfilt_ops[~filt].for_refcnt--;
938 mtx_unlock(&filterops_lock);
939 }
940
941 /*
942 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will
943 * influence if memory allocation should wait. Make sure it is 0 if you
944 * hold any mutexes.
945 */
946 static int
947 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
948 {
949 struct filterops *fops;
950 struct file *fp;
951 struct knote *kn, *tkn;
952 int error, filt, event;
953 int haskqglobal;
954
955 fp = NULL;
956 kn = NULL;
957 error = 0;
958 haskqglobal = 0;
959
960 filt = kev->filter;
961 fops = kqueue_fo_find(filt);
962 if (fops == NULL)
963 return EINVAL;
964
965 tkn = knote_alloc(waitok); /* prevent waiting with locks */
966
967 findkn:
968 if (fops->f_isfd) {
969 KASSERT(td != NULL, ("td is NULL"));
970 error = fget(td, kev->ident, &fp);
971 if (error)
972 goto done;
973
974 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
975 kev->ident, 0) != 0) {
976 /* try again */
977 fdrop(fp, td);
978 fp = NULL;
979 error = kqueue_expand(kq, fops, kev->ident, waitok);
980 if (error)
981 goto done;
982 goto findkn;
983 }
984
985 if (fp->f_type == DTYPE_KQUEUE) {
986 /*
987 * if we add some inteligence about what we are doing,
988 * we should be able to support events on ourselves.
989 * We need to know when we are doing this to prevent
990 * getting both the knlist lock and the kq lock since
991 * they are the same thing.
992 */
993 if (fp->f_data == kq) {
994 error = EINVAL;
995 goto done;
996 }
997
998 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
999 }
1000
1001 KQ_LOCK(kq);
1002 if (kev->ident < kq->kq_knlistsize) {
1003 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1004 if (kev->filter == kn->kn_filter)
1005 break;
1006 }
1007 } else {
1008 if ((kev->flags & EV_ADD) == EV_ADD)
1009 kqueue_expand(kq, fops, kev->ident, waitok);
1010
1011 KQ_LOCK(kq);
1012 if (kq->kq_knhashmask != 0) {
1013 struct klist *list;
1014
1015 list = &kq->kq_knhash[
1016 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1017 SLIST_FOREACH(kn, list, kn_link)
1018 if (kev->ident == kn->kn_id &&
1019 kev->filter == kn->kn_filter)
1020 break;
1021 }
1022 }
1023
1024 /* knote is in the process of changing, wait for it to stablize. */
1025 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1026 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1027 kq->kq_state |= KQ_FLUXWAIT;
1028 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1029 if (fp != NULL) {
1030 fdrop(fp, td);
1031 fp = NULL;
1032 }
1033 goto findkn;
1034 }
1035
1036 /*
1037 * kn now contains the matching knote, or NULL if no match
1038 */
1039 if (kn == NULL) {
1040 if (kev->flags & EV_ADD) {
1041 kn = tkn;
1042 tkn = NULL;
1043 if (kn == NULL) {
1044 KQ_UNLOCK(kq);
1045 error = ENOMEM;
1046 goto done;
1047 }
1048 kn->kn_fp = fp;
1049 kn->kn_kq = kq;
1050 kn->kn_fop = fops;
1051 /*
1052 * apply reference counts to knote structure, and
1053 * do not release it at the end of this routine.
1054 */
1055 fops = NULL;
1056 fp = NULL;
1057
1058 kn->kn_sfflags = kev->fflags;
1059 kn->kn_sdata = kev->data;
1060 kev->fflags = 0;
1061 kev->data = 0;
1062 kn->kn_kevent = *kev;
1063 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1064 EV_ENABLE | EV_DISABLE);
1065 kn->kn_status = KN_INFLUX|KN_DETACHED;
1066
1067 error = knote_attach(kn, kq);
1068 KQ_UNLOCK(kq);
1069 if (error != 0) {
1070 tkn = kn;
1071 goto done;
1072 }
1073
1074 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1075 knote_drop(kn, td);
1076 goto done;
1077 }
1078 KN_LIST_LOCK(kn);
1079 goto done_ev_add;
1080 } else {
1081 /* No matching knote and the EV_ADD flag is not set. */
1082 KQ_UNLOCK(kq);
1083 error = ENOENT;
1084 goto done;
1085 }
1086 }
1087
1088 if (kev->flags & EV_DELETE) {
1089 kn->kn_status |= KN_INFLUX;
1090 KQ_UNLOCK(kq);
1091 if (!(kn->kn_status & KN_DETACHED))
1092 kn->kn_fop->f_detach(kn);
1093 knote_drop(kn, td);
1094 goto done;
1095 }
1096
1097 /*
1098 * The user may change some filter values after the initial EV_ADD,
1099 * but doing so will not reset any filter which has already been
1100 * triggered.
1101 */
1102 kn->kn_status |= KN_INFLUX;
1103 KQ_UNLOCK(kq);
1104 KN_LIST_LOCK(kn);
1105 kn->kn_kevent.udata = kev->udata;
1106 if (!fops->f_isfd && fops->f_touch != NULL) {
1107 fops->f_touch(kn, kev, EVENT_REGISTER);
1108 } else {
1109 kn->kn_sfflags = kev->fflags;
1110 kn->kn_sdata = kev->data;
1111 }
1112
1113 /*
1114 * We can get here with kn->kn_knlist == NULL. This can happen when
1115 * the initial attach event decides that the event is "completed"
1116 * already. i.e. filt_procattach is called on a zombie process. It
1117 * will call filt_proc which will remove it from the list, and NULL
1118 * kn_knlist.
1119 */
1120 done_ev_add:
1121 event = kn->kn_fop->f_event(kn, 0);
1122 KQ_LOCK(kq);
1123 if (event)
1124 KNOTE_ACTIVATE(kn, 1);
1125 kn->kn_status &= ~KN_INFLUX;
1126 KN_LIST_UNLOCK(kn);
1127
1128 if ((kev->flags & EV_DISABLE) &&
1129 ((kn->kn_status & KN_DISABLED) == 0)) {
1130 kn->kn_status |= KN_DISABLED;
1131 }
1132
1133 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1134 kn->kn_status &= ~KN_DISABLED;
1135 if ((kn->kn_status & KN_ACTIVE) &&
1136 ((kn->kn_status & KN_QUEUED) == 0))
1137 knote_enqueue(kn);
1138 }
1139 KQ_UNLOCK_FLUX(kq);
1140
1141 done:
1142 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1143 if (fp != NULL)
1144 fdrop(fp, td);
1145 if (tkn != NULL)
1146 knote_free(tkn);
1147 if (fops != NULL)
1148 kqueue_fo_release(filt);
1149 return (error);
1150 }
1151
1152 static int
1153 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1154 {
1155 int error;
1156 struct kqueue *kq;
1157
1158 error = 0;
1159
1160 kq = fp->f_data;
1161 if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1162 return (EBADF);
1163 *kqp = kq;
1164 KQ_LOCK(kq);
1165 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1166 KQ_UNLOCK(kq);
1167 return (EBADF);
1168 }
1169 kq->kq_refcnt++;
1170 KQ_UNLOCK(kq);
1171
1172 return error;
1173 }
1174
1175 static void
1176 kqueue_release(struct kqueue *kq, int locked)
1177 {
1178 if (locked)
1179 KQ_OWNED(kq);
1180 else
1181 KQ_LOCK(kq);
1182 kq->kq_refcnt--;
1183 if (kq->kq_refcnt == 1)
1184 wakeup(&kq->kq_refcnt);
1185 if (!locked)
1186 KQ_UNLOCK(kq);
1187 }
1188
1189 static void
1190 kqueue_schedtask(struct kqueue *kq)
1191 {
1192
1193 KQ_OWNED(kq);
1194 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1195 ("scheduling kqueue task while draining"));
1196
1197 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1198 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1199 kq->kq_state |= KQ_TASKSCHED;
1200 }
1201 }
1202
1203 /*
1204 * Expand the kq to make sure we have storage for fops/ident pair.
1205 *
1206 * Return 0 on success (or no work necessary), return errno on failure.
1207 *
1208 * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1209 * If kqueue_register is called from a non-fd context, there usually/should
1210 * be no locks held.
1211 */
1212 static int
1213 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1214 int waitok)
1215 {
1216 struct klist *list, *tmp_knhash, *to_free;
1217 u_long tmp_knhashmask;
1218 int size;
1219 int fd;
1220 int mflag = waitok ? M_WAITOK : M_NOWAIT;
1221
1222 KQ_NOTOWNED(kq);
1223
1224 to_free = NULL;
1225 if (fops->f_isfd) {
1226 fd = ident;
1227 if (kq->kq_knlistsize <= fd) {
1228 size = kq->kq_knlistsize;
1229 while (size <= fd)
1230 size += KQEXTENT;
1231 list = malloc(size * sizeof list, M_KQUEUE, mflag);
1232 if (list == NULL)
1233 return ENOMEM;
1234 KQ_LOCK(kq);
1235 if (kq->kq_knlistsize > fd) {
1236 to_free = list;
1237 list = NULL;
1238 } else {
1239 if (kq->kq_knlist != NULL) {
1240 bcopy(kq->kq_knlist, list,
1241 kq->kq_knlistsize * sizeof list);
1242 to_free = kq->kq_knlist;
1243 kq->kq_knlist = NULL;
1244 }
1245 bzero((caddr_t)list +
1246 kq->kq_knlistsize * sizeof list,
1247 (size - kq->kq_knlistsize) * sizeof list);
1248 kq->kq_knlistsize = size;
1249 kq->kq_knlist = list;
1250 }
1251 KQ_UNLOCK(kq);
1252 }
1253 } else {
1254 if (kq->kq_knhashmask == 0) {
1255 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1256 &tmp_knhashmask);
1257 if (tmp_knhash == NULL)
1258 return ENOMEM;
1259 KQ_LOCK(kq);
1260 if (kq->kq_knhashmask == 0) {
1261 kq->kq_knhash = tmp_knhash;
1262 kq->kq_knhashmask = tmp_knhashmask;
1263 } else {
1264 to_free = tmp_knhash;
1265 }
1266 KQ_UNLOCK(kq);
1267 }
1268 }
1269 free(to_free, M_KQUEUE);
1270
1271 KQ_NOTOWNED(kq);
1272 return 0;
1273 }
1274
1275 static void
1276 kqueue_task(void *arg, int pending)
1277 {
1278 struct kqueue *kq;
1279 int haskqglobal;
1280
1281 haskqglobal = 0;
1282 kq = arg;
1283
1284 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1285 KQ_LOCK(kq);
1286
1287 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1288
1289 kq->kq_state &= ~KQ_TASKSCHED;
1290 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1291 wakeup(&kq->kq_state);
1292 }
1293 KQ_UNLOCK(kq);
1294 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1295 }
1296
1297 /*
1298 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1299 * We treat KN_MARKER knotes as if they are INFLUX.
1300 */
1301 static int
1302 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1303 const struct timespec *tsp, struct kevent *keva, struct thread *td)
1304 {
1305 struct kevent *kevp;
1306 struct timeval atv, rtv, ttv;
1307 struct knote *kn, *marker;
1308 int count, timeout, nkev, error, influx;
1309 int haskqglobal, touch;
1310
1311 count = maxevents;
1312 nkev = 0;
1313 error = 0;
1314 haskqglobal = 0;
1315
1316 if (maxevents == 0)
1317 goto done_nl;
1318
1319 if (tsp != NULL) {
1320 TIMESPEC_TO_TIMEVAL(&atv, tsp);
1321 if (itimerfix(&atv)) {
1322 error = EINVAL;
1323 goto done_nl;
1324 }
1325 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1326 timeout = -1;
1327 else
1328 timeout = atv.tv_sec > 24 * 60 * 60 ?
1329 24 * 60 * 60 * hz : tvtohz(&atv);
1330 getmicrouptime(&rtv);
1331 timevaladd(&atv, &rtv);
1332 } else {
1333 atv.tv_sec = 0;
1334 atv.tv_usec = 0;
1335 timeout = 0;
1336 }
1337 marker = knote_alloc(1);
1338 if (marker == NULL) {
1339 error = ENOMEM;
1340 goto done_nl;
1341 }
1342 marker->kn_status = KN_MARKER;
1343 KQ_LOCK(kq);
1344 goto start;
1345
1346 retry:
1347 if (atv.tv_sec || atv.tv_usec) {
1348 getmicrouptime(&rtv);
1349 if (timevalcmp(&rtv, &atv, >=))
1350 goto done;
1351 ttv = atv;
1352 timevalsub(&ttv, &rtv);
1353 timeout = ttv.tv_sec > 24 * 60 * 60 ?
1354 24 * 60 * 60 * hz : tvtohz(&ttv);
1355 }
1356
1357 start:
1358 kevp = keva;
1359 if (kq->kq_count == 0) {
1360 if (timeout < 0) {
1361 error = EWOULDBLOCK;
1362 } else {
1363 kq->kq_state |= KQ_SLEEP;
1364 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH,
1365 "kqread", timeout);
1366 }
1367 if (error == 0)
1368 goto retry;
1369 /* don't restart after signals... */
1370 if (error == ERESTART)
1371 error = EINTR;
1372 else if (error == EWOULDBLOCK)
1373 error = 0;
1374 goto done;
1375 }
1376
1377 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1378 influx = 0;
1379 while (count) {
1380 KQ_OWNED(kq);
1381 kn = TAILQ_FIRST(&kq->kq_head);
1382
1383 if ((kn->kn_status == KN_MARKER && kn != marker) ||
1384 (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1385 if (influx) {
1386 influx = 0;
1387 KQ_FLUX_WAKEUP(kq);
1388 }
1389 kq->kq_state |= KQ_FLUXWAIT;
1390 error = msleep(kq, &kq->kq_lock, PSOCK,
1391 "kqflxwt", 0);
1392 continue;
1393 }
1394
1395 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1396 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1397 kn->kn_status &= ~KN_QUEUED;
1398 kq->kq_count--;
1399 continue;
1400 }
1401 if (kn == marker) {
1402 KQ_FLUX_WAKEUP(kq);
1403 if (count == maxevents)
1404 goto retry;
1405 goto done;
1406 }
1407 KASSERT((kn->kn_status & KN_INFLUX) == 0,
1408 ("KN_INFLUX set when not suppose to be"));
1409
1410 if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1411 kn->kn_status &= ~KN_QUEUED;
1412 kn->kn_status |= KN_INFLUX;
1413 kq->kq_count--;
1414 KQ_UNLOCK(kq);
1415 /*
1416 * We don't need to lock the list since we've marked
1417 * it _INFLUX.
1418 */
1419 if (!(kn->kn_status & KN_DETACHED))
1420 kn->kn_fop->f_detach(kn);
1421 knote_drop(kn, td);
1422 KQ_LOCK(kq);
1423 continue;
1424 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1425 kn->kn_status &= ~KN_QUEUED;
1426 kn->kn_status |= KN_INFLUX;
1427 kq->kq_count--;
1428 KQ_UNLOCK(kq);
1429 /*
1430 * We don't need to lock the list since we've marked
1431 * it _INFLUX.
1432 */
1433 *kevp = kn->kn_kevent;
1434 if (!(kn->kn_status & KN_DETACHED))
1435 kn->kn_fop->f_detach(kn);
1436 knote_drop(kn, td);
1437 KQ_LOCK(kq);
1438 kn = NULL;
1439 } else {
1440 kn->kn_status |= KN_INFLUX;
1441 KQ_UNLOCK(kq);
1442 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1443 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1444 KN_LIST_LOCK(kn);
1445 if (kn->kn_fop->f_event(kn, 0) == 0) {
1446 KQ_LOCK(kq);
1447 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1448 kn->kn_status &=
1449 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX);
1450 kq->kq_count--;
1451 KN_LIST_UNLOCK(kn);
1452 influx = 1;
1453 continue;
1454 }
1455 touch = (!kn->kn_fop->f_isfd &&
1456 kn->kn_fop->f_touch != NULL);
1457 if (touch)
1458 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1459 else
1460 *kevp = kn->kn_kevent;
1461 KQ_LOCK(kq);
1462 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1463 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1464 /*
1465 * Manually clear knotes who weren't
1466 * 'touch'ed.
1467 */
1468 if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1469 kn->kn_data = 0;
1470 kn->kn_fflags = 0;
1471 }
1472 if (kn->kn_flags & EV_DISPATCH)
1473 kn->kn_status |= KN_DISABLED;
1474 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1475 kq->kq_count--;
1476 } else
1477 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1478
1479 kn->kn_status &= ~(KN_INFLUX);
1480 KN_LIST_UNLOCK(kn);
1481 influx = 1;
1482 }
1483
1484 /* we are returning a copy to the user */
1485 kevp++;
1486 nkev++;
1487 count--;
1488
1489 if (nkev == KQ_NEVENTS) {
1490 influx = 0;
1491 KQ_UNLOCK_FLUX(kq);
1492 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1493 nkev = 0;
1494 kevp = keva;
1495 KQ_LOCK(kq);
1496 if (error)
1497 break;
1498 }
1499 }
1500 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1501 done:
1502 KQ_OWNED(kq);
1503 KQ_UNLOCK_FLUX(kq);
1504 knote_free(marker);
1505 done_nl:
1506 KQ_NOTOWNED(kq);
1507 if (nkev != 0)
1508 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1509 td->td_retval[0] = maxevents - count;
1510 return (error);
1511 }
1512
1513 /*
1514 * XXX
1515 * This could be expanded to call kqueue_scan, if desired.
1516 */
1517 /*ARGSUSED*/
1518 static int
1519 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1520 int flags, struct thread *td)
1521 {
1522 return (ENXIO);
1523 }
1524
1525 /*ARGSUSED*/
1526 static int
1527 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1528 int flags, struct thread *td)
1529 {
1530 return (ENXIO);
1531 }
1532
1533 /*ARGSUSED*/
1534 static int
1535 kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1536 struct thread *td)
1537 {
1538
1539 return (EINVAL);
1540 }
1541
1542 /*ARGSUSED*/
1543 static int
1544 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1545 struct ucred *active_cred, struct thread *td)
1546 {
1547 /*
1548 * Enabling sigio causes two major problems:
1549 * 1) infinite recursion:
1550 * Synopsys: kevent is being used to track signals and have FIOASYNC
1551 * set. On receipt of a signal this will cause a kqueue to recurse
1552 * into itself over and over. Sending the sigio causes the kqueue
1553 * to become ready, which in turn posts sigio again, forever.
1554 * Solution: this can be solved by setting a flag in the kqueue that
1555 * we have a SIGIO in progress.
1556 * 2) locking problems:
1557 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1558 * us above the proc and pgrp locks.
1559 * Solution: Post a signal using an async mechanism, being sure to
1560 * record a generation count in the delivery so that we do not deliver
1561 * a signal to the wrong process.
1562 *
1563 * Note, these two mechanisms are somewhat mutually exclusive!
1564 */
1565 #if 0
1566 struct kqueue *kq;
1567
1568 kq = fp->f_data;
1569 switch (cmd) {
1570 case FIOASYNC:
1571 if (*(int *)data) {
1572 kq->kq_state |= KQ_ASYNC;
1573 } else {
1574 kq->kq_state &= ~KQ_ASYNC;
1575 }
1576 return (0);
1577
1578 case FIOSETOWN:
1579 return (fsetown(*(int *)data, &kq->kq_sigio));
1580
1581 case FIOGETOWN:
1582 *(int *)data = fgetown(&kq->kq_sigio);
1583 return (0);
1584 }
1585 #endif
1586
1587 return (ENOTTY);
1588 }
1589
1590 /*ARGSUSED*/
1591 static int
1592 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1593 struct thread *td)
1594 {
1595 struct kqueue *kq;
1596 int revents = 0;
1597 int error;
1598
1599 if ((error = kqueue_acquire(fp, &kq)))
1600 return POLLERR;
1601
1602 KQ_LOCK(kq);
1603 if (events & (POLLIN | POLLRDNORM)) {
1604 if (kq->kq_count) {
1605 revents |= events & (POLLIN | POLLRDNORM);
1606 } else {
1607 selrecord(td, &kq->kq_sel);
1608 if (SEL_WAITING(&kq->kq_sel))
1609 kq->kq_state |= KQ_SEL;
1610 }
1611 }
1612 kqueue_release(kq, 1);
1613 KQ_UNLOCK(kq);
1614 return (revents);
1615 }
1616
1617 /*ARGSUSED*/
1618 static int
1619 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1620 struct thread *td)
1621 {
1622
1623 bzero((void *)st, sizeof *st);
1624 /*
1625 * We no longer return kq_count because the unlocked value is useless.
1626 * If you spent all this time getting the count, why not spend your
1627 * syscall better by calling kevent?
1628 *
1629 * XXX - This is needed for libc_r.
1630 */
1631 st->st_mode = S_IFIFO;
1632 return (0);
1633 }
1634
1635 /*ARGSUSED*/
1636 static int
1637 kqueue_close(struct file *fp, struct thread *td)
1638 {
1639 struct kqueue *kq = fp->f_data;
1640 struct filedesc *fdp;
1641 struct knote *kn;
1642 int i;
1643 int error;
1644
1645 if ((error = kqueue_acquire(fp, &kq)))
1646 return error;
1647
1648 KQ_LOCK(kq);
1649
1650 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1651 ("kqueue already closing"));
1652 kq->kq_state |= KQ_CLOSING;
1653 if (kq->kq_refcnt > 1)
1654 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1655
1656 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1657 fdp = kq->kq_fdp;
1658
1659 KASSERT(knlist_empty(&kq->kq_sel.si_note),
1660 ("kqueue's knlist not empty"));
1661
1662 for (i = 0; i < kq->kq_knlistsize; i++) {
1663 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1664 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1665 kq->kq_state |= KQ_FLUXWAIT;
1666 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1667 continue;
1668 }
1669 kn->kn_status |= KN_INFLUX;
1670 KQ_UNLOCK(kq);
1671 if (!(kn->kn_status & KN_DETACHED))
1672 kn->kn_fop->f_detach(kn);
1673 knote_drop(kn, td);
1674 KQ_LOCK(kq);
1675 }
1676 }
1677 if (kq->kq_knhashmask != 0) {
1678 for (i = 0; i <= kq->kq_knhashmask; i++) {
1679 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1680 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1681 kq->kq_state |= KQ_FLUXWAIT;
1682 msleep(kq, &kq->kq_lock, PSOCK,
1683 "kqclo2", 0);
1684 continue;
1685 }
1686 kn->kn_status |= KN_INFLUX;
1687 KQ_UNLOCK(kq);
1688 if (!(kn->kn_status & KN_DETACHED))
1689 kn->kn_fop->f_detach(kn);
1690 knote_drop(kn, td);
1691 KQ_LOCK(kq);
1692 }
1693 }
1694 }
1695
1696 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1697 kq->kq_state |= KQ_TASKDRAIN;
1698 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1699 }
1700
1701 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1702 selwakeuppri(&kq->kq_sel, PSOCK);
1703 if (!SEL_WAITING(&kq->kq_sel))
1704 kq->kq_state &= ~KQ_SEL;
1705 }
1706
1707 KQ_UNLOCK(kq);
1708
1709 FILEDESC_XLOCK(fdp);
1710 SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list);
1711 FILEDESC_XUNLOCK(fdp);
1712
1713 seldrain(&kq->kq_sel);
1714 knlist_destroy(&kq->kq_sel.si_note);
1715 mtx_destroy(&kq->kq_lock);
1716 kq->kq_fdp = NULL;
1717
1718 if (kq->kq_knhash != NULL)
1719 free(kq->kq_knhash, M_KQUEUE);
1720 if (kq->kq_knlist != NULL)
1721 free(kq->kq_knlist, M_KQUEUE);
1722
1723 funsetown(&kq->kq_sigio);
1724 free(kq, M_KQUEUE);
1725 fp->f_data = NULL;
1726
1727 return (0);
1728 }
1729
1730 static void
1731 kqueue_wakeup(struct kqueue *kq)
1732 {
1733 KQ_OWNED(kq);
1734
1735 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1736 kq->kq_state &= ~KQ_SLEEP;
1737 wakeup(kq);
1738 }
1739 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1740 selwakeuppri(&kq->kq_sel, PSOCK);
1741 if (!SEL_WAITING(&kq->kq_sel))
1742 kq->kq_state &= ~KQ_SEL;
1743 }
1744 if (!knlist_empty(&kq->kq_sel.si_note))
1745 kqueue_schedtask(kq);
1746 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1747 pgsigio(&kq->kq_sigio, SIGIO, 0);
1748 }
1749 }
1750
1751 /*
1752 * Walk down a list of knotes, activating them if their event has triggered.
1753 *
1754 * There is a possibility to optimize in the case of one kq watching another.
1755 * Instead of scheduling a task to wake it up, you could pass enough state
1756 * down the chain to make up the parent kqueue. Make this code functional
1757 * first.
1758 */
1759 void
1760 knote(struct knlist *list, long hint, int lockflags)
1761 {
1762 struct kqueue *kq;
1763 struct knote *kn;
1764 int error;
1765
1766 if (list == NULL)
1767 return;
1768
1769 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1770
1771 if ((lockflags & KNF_LISTLOCKED) == 0)
1772 list->kl_lock(list->kl_lockarg);
1773
1774 /*
1775 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1776 * the kqueue scheduling, but this will introduce four
1777 * lock/unlock's for each knote to test. If we do, continue to use
1778 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1779 * only safe if you want to remove the current item, which we are
1780 * not doing.
1781 */
1782 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1783 kq = kn->kn_kq;
1784 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1785 KQ_LOCK(kq);
1786 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1787 KQ_UNLOCK(kq);
1788 } else if ((lockflags & KNF_NOKQLOCK) != 0) {
1789 kn->kn_status |= KN_INFLUX;
1790 KQ_UNLOCK(kq);
1791 error = kn->kn_fop->f_event(kn, hint);
1792 KQ_LOCK(kq);
1793 kn->kn_status &= ~KN_INFLUX;
1794 if (error)
1795 KNOTE_ACTIVATE(kn, 1);
1796 KQ_UNLOCK_FLUX(kq);
1797 } else {
1798 kn->kn_status |= KN_HASKQLOCK;
1799 if (kn->kn_fop->f_event(kn, hint))
1800 KNOTE_ACTIVATE(kn, 1);
1801 kn->kn_status &= ~KN_HASKQLOCK;
1802 KQ_UNLOCK(kq);
1803 }
1804 }
1805 kq = NULL;
1806 }
1807 if ((lockflags & KNF_LISTLOCKED) == 0)
1808 list->kl_unlock(list->kl_lockarg);
1809 }
1810
1811 /*
1812 * add a knote to a knlist
1813 */
1814 void
1815 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1816 {
1817 KNL_ASSERT_LOCK(knl, islocked);
1818 KQ_NOTOWNED(kn->kn_kq);
1819 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1820 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1821 if (!islocked)
1822 knl->kl_lock(knl->kl_lockarg);
1823 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1824 if (!islocked)
1825 knl->kl_unlock(knl->kl_lockarg);
1826 KQ_LOCK(kn->kn_kq);
1827 kn->kn_knlist = knl;
1828 kn->kn_status &= ~KN_DETACHED;
1829 KQ_UNLOCK(kn->kn_kq);
1830 }
1831
1832 static void
1833 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1834 {
1835 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1836 KNL_ASSERT_LOCK(knl, knlislocked);
1837 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1838 if (!kqislocked)
1839 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1840 ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1841 if (!knlislocked)
1842 knl->kl_lock(knl->kl_lockarg);
1843 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1844 kn->kn_knlist = NULL;
1845 if (!knlislocked)
1846 knl->kl_unlock(knl->kl_lockarg);
1847 if (!kqislocked)
1848 KQ_LOCK(kn->kn_kq);
1849 kn->kn_status |= KN_DETACHED;
1850 if (!kqislocked)
1851 KQ_UNLOCK(kn->kn_kq);
1852 }
1853
1854 /*
1855 * remove all knotes from a specified klist
1856 */
1857 void
1858 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1859 {
1860
1861 knlist_remove_kq(knl, kn, islocked, 0);
1862 }
1863
1864 /*
1865 * remove knote from a specified klist while in f_event handler.
1866 */
1867 void
1868 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1869 {
1870
1871 knlist_remove_kq(knl, kn, 1,
1872 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1873 }
1874
1875 int
1876 knlist_empty(struct knlist *knl)
1877 {
1878 KNL_ASSERT_LOCKED(knl);
1879 return SLIST_EMPTY(&knl->kl_list);
1880 }
1881
1882 static struct mtx knlist_lock;
1883 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1884 MTX_DEF);
1885 static void knlist_mtx_lock(void *arg);
1886 static void knlist_mtx_unlock(void *arg);
1887
1888 static void
1889 knlist_mtx_lock(void *arg)
1890 {
1891 mtx_lock((struct mtx *)arg);
1892 }
1893
1894 static void
1895 knlist_mtx_unlock(void *arg)
1896 {
1897 mtx_unlock((struct mtx *)arg);
1898 }
1899
1900 static void
1901 knlist_mtx_assert_locked(void *arg)
1902 {
1903 mtx_assert((struct mtx *)arg, MA_OWNED);
1904 }
1905
1906 static void
1907 knlist_mtx_assert_unlocked(void *arg)
1908 {
1909 mtx_assert((struct mtx *)arg, MA_NOTOWNED);
1910 }
1911
1912 void
1913 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1914 void (*kl_unlock)(void *),
1915 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
1916 {
1917
1918 if (lock == NULL)
1919 knl->kl_lockarg = &knlist_lock;
1920 else
1921 knl->kl_lockarg = lock;
1922
1923 if (kl_lock == NULL)
1924 knl->kl_lock = knlist_mtx_lock;
1925 else
1926 knl->kl_lock = kl_lock;
1927 if (kl_unlock == NULL)
1928 knl->kl_unlock = knlist_mtx_unlock;
1929 else
1930 knl->kl_unlock = kl_unlock;
1931 if (kl_assert_locked == NULL)
1932 knl->kl_assert_locked = knlist_mtx_assert_locked;
1933 else
1934 knl->kl_assert_locked = kl_assert_locked;
1935 if (kl_assert_unlocked == NULL)
1936 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
1937 else
1938 knl->kl_assert_unlocked = kl_assert_unlocked;
1939
1940 SLIST_INIT(&knl->kl_list);
1941 }
1942
1943 void
1944 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
1945 {
1946
1947 knlist_init(knl, lock, NULL, NULL, NULL, NULL);
1948 }
1949
1950 void
1951 knlist_destroy(struct knlist *knl)
1952 {
1953
1954 #ifdef INVARIANTS
1955 /*
1956 * if we run across this error, we need to find the offending
1957 * driver and have it call knlist_clear.
1958 */
1959 if (!SLIST_EMPTY(&knl->kl_list))
1960 printf("WARNING: destroying knlist w/ knotes on it!\n");
1961 #endif
1962
1963 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
1964 SLIST_INIT(&knl->kl_list);
1965 }
1966
1967 /*
1968 * Even if we are locked, we may need to drop the lock to allow any influx
1969 * knotes time to "settle".
1970 */
1971 void
1972 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
1973 {
1974 struct knote *kn, *kn2;
1975 struct kqueue *kq;
1976
1977 if (islocked)
1978 KNL_ASSERT_LOCKED(knl);
1979 else {
1980 KNL_ASSERT_UNLOCKED(knl);
1981 again: /* need to reacquire lock since we have dropped it */
1982 knl->kl_lock(knl->kl_lockarg);
1983 }
1984
1985 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
1986 kq = kn->kn_kq;
1987 KQ_LOCK(kq);
1988 if ((kn->kn_status & KN_INFLUX)) {
1989 KQ_UNLOCK(kq);
1990 continue;
1991 }
1992 knlist_remove_kq(knl, kn, 1, 1);
1993 if (killkn) {
1994 kn->kn_status |= KN_INFLUX | KN_DETACHED;
1995 KQ_UNLOCK(kq);
1996 knote_drop(kn, td);
1997 } else {
1998 /* Make sure cleared knotes disappear soon */
1999 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2000 KQ_UNLOCK(kq);
2001 }
2002 kq = NULL;
2003 }
2004
2005 if (!SLIST_EMPTY(&knl->kl_list)) {
2006 /* there are still KN_INFLUX remaining */
2007 kn = SLIST_FIRST(&knl->kl_list);
2008 kq = kn->kn_kq;
2009 KQ_LOCK(kq);
2010 KASSERT(kn->kn_status & KN_INFLUX,
2011 ("knote removed w/o list lock"));
2012 knl->kl_unlock(knl->kl_lockarg);
2013 kq->kq_state |= KQ_FLUXWAIT;
2014 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2015 kq = NULL;
2016 goto again;
2017 }
2018
2019 if (islocked)
2020 KNL_ASSERT_LOCKED(knl);
2021 else {
2022 knl->kl_unlock(knl->kl_lockarg);
2023 KNL_ASSERT_UNLOCKED(knl);
2024 }
2025 }
2026
2027 /*
2028 * Remove all knotes referencing a specified fd must be called with FILEDESC
2029 * lock. This prevents a race where a new fd comes along and occupies the
2030 * entry and we attach a knote to the fd.
2031 */
2032 void
2033 knote_fdclose(struct thread *td, int fd)
2034 {
2035 struct filedesc *fdp = td->td_proc->p_fd;
2036 struct kqueue *kq;
2037 struct knote *kn;
2038 int influx;
2039
2040 FILEDESC_XLOCK_ASSERT(fdp);
2041
2042 /*
2043 * We shouldn't have to worry about new kevents appearing on fd
2044 * since filedesc is locked.
2045 */
2046 SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2047 KQ_LOCK(kq);
2048
2049 again:
2050 influx = 0;
2051 while (kq->kq_knlistsize > fd &&
2052 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2053 if (kn->kn_status & KN_INFLUX) {
2054 /* someone else might be waiting on our knote */
2055 if (influx)
2056 wakeup(kq);
2057 kq->kq_state |= KQ_FLUXWAIT;
2058 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2059 goto again;
2060 }
2061 kn->kn_status |= KN_INFLUX;
2062 KQ_UNLOCK(kq);
2063 if (!(kn->kn_status & KN_DETACHED))
2064 kn->kn_fop->f_detach(kn);
2065 knote_drop(kn, td);
2066 influx = 1;
2067 KQ_LOCK(kq);
2068 }
2069 KQ_UNLOCK_FLUX(kq);
2070 }
2071 }
2072
2073 static int
2074 knote_attach(struct knote *kn, struct kqueue *kq)
2075 {
2076 struct klist *list;
2077
2078 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2079 KQ_OWNED(kq);
2080
2081 if (kn->kn_fop->f_isfd) {
2082 if (kn->kn_id >= kq->kq_knlistsize)
2083 return ENOMEM;
2084 list = &kq->kq_knlist[kn->kn_id];
2085 } else {
2086 if (kq->kq_knhash == NULL)
2087 return ENOMEM;
2088 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2089 }
2090
2091 SLIST_INSERT_HEAD(list, kn, kn_link);
2092
2093 return 0;
2094 }
2095
2096 /*
2097 * knote must already have been detached using the f_detach method.
2098 * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2099 * to prevent other removal.
2100 */
2101 static void
2102 knote_drop(struct knote *kn, struct thread *td)
2103 {
2104 struct kqueue *kq;
2105 struct klist *list;
2106
2107 kq = kn->kn_kq;
2108
2109 KQ_NOTOWNED(kq);
2110 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2111 ("knote_drop called without KN_INFLUX set in kn_status"));
2112
2113 KQ_LOCK(kq);
2114 if (kn->kn_fop->f_isfd)
2115 list = &kq->kq_knlist[kn->kn_id];
2116 else
2117 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2118
2119 if (!SLIST_EMPTY(list))
2120 SLIST_REMOVE(list, kn, knote, kn_link);
2121 if (kn->kn_status & KN_QUEUED)
2122 knote_dequeue(kn);
2123 KQ_UNLOCK_FLUX(kq);
2124
2125 if (kn->kn_fop->f_isfd) {
2126 fdrop(kn->kn_fp, td);
2127 kn->kn_fp = NULL;
2128 }
2129 kqueue_fo_release(kn->kn_kevent.filter);
2130 kn->kn_fop = NULL;
2131 knote_free(kn);
2132 }
2133
2134 static void
2135 knote_enqueue(struct knote *kn)
2136 {
2137 struct kqueue *kq = kn->kn_kq;
2138
2139 KQ_OWNED(kn->kn_kq);
2140 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2141
2142 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2143 kn->kn_status |= KN_QUEUED;
2144 kq->kq_count++;
2145 kqueue_wakeup(kq);
2146 }
2147
2148 static void
2149 knote_dequeue(struct knote *kn)
2150 {
2151 struct kqueue *kq = kn->kn_kq;
2152
2153 KQ_OWNED(kn->kn_kq);
2154 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2155
2156 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2157 kn->kn_status &= ~KN_QUEUED;
2158 kq->kq_count--;
2159 }
2160
2161 static void
2162 knote_init(void)
2163 {
2164
2165 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2166 NULL, NULL, UMA_ALIGN_PTR, 0);
2167 }
2168 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2169
2170 static struct knote *
2171 knote_alloc(int waitok)
2172 {
2173 return ((struct knote *)uma_zalloc(knote_zone,
2174 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2175 }
2176
2177 static void
2178 knote_free(struct knote *kn)
2179 {
2180 if (kn != NULL)
2181 uma_zfree(knote_zone, kn);
2182 }
2183
2184 /*
2185 * Register the kev w/ the kq specified by fd.
2186 */
2187 int
2188 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2189 {
2190 struct kqueue *kq;
2191 struct file *fp;
2192 int error;
2193
2194 if ((error = fget(td, fd, &fp)) != 0)
2195 return (error);
2196 if ((error = kqueue_acquire(fp, &kq)) != 0)
2197 goto noacquire;
2198
2199 error = kqueue_register(kq, kev, td, waitok);
2200
2201 kqueue_release(kq, 0);
2202
2203 noacquire:
2204 fdrop(fp, td);
2205
2206 return error;
2207 }
Cache object: 50a981471efa143cb6577e45618a7dd4
|