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: releng/9.2/sys/kern/kern_event.c 239915 2012-08-30 18:29:49Z jhb $");
31
32 #include "opt_ktrace.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/capability.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/malloc.h>
42 #include <sys/unistd.h>
43 #include <sys/file.h>
44 #include <sys/filedesc.h>
45 #include <sys/filio.h>
46 #include <sys/fcntl.h>
47 #include <sys/kthread.h>
48 #include <sys/selinfo.h>
49 #include <sys/queue.h>
50 #include <sys/event.h>
51 #include <sys/eventvar.h>
52 #include <sys/poll.h>
53 #include <sys/protosw.h>
54 #include <sys/sigio.h>
55 #include <sys/signalvar.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/stat.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysproto.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/taskqueue.h>
63 #include <sys/uio.h>
64 #ifdef KTRACE
65 #include <sys/ktrace.h>
66 #endif
67
68 #include <vm/uma.h>
69
70 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
71
72 /*
73 * This lock is used if multiple kq locks are required. This possibly
74 * should be made into a per proc lock.
75 */
76 static struct mtx kq_global;
77 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
78 #define KQ_GLOBAL_LOCK(lck, haslck) do { \
79 if (!haslck) \
80 mtx_lock(lck); \
81 haslck = 1; \
82 } while (0)
83 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \
84 if (haslck) \
85 mtx_unlock(lck); \
86 haslck = 0; \
87 } while (0)
88
89 TASKQUEUE_DEFINE_THREAD(kqueue);
90
91 static int kevent_copyout(void *arg, struct kevent *kevp, int count);
92 static int kevent_copyin(void *arg, struct kevent *kevp, int count);
93 static int kqueue_register(struct kqueue *kq, struct kevent *kev,
94 struct thread *td, int waitok);
95 static int kqueue_acquire(struct file *fp, struct kqueue **kqp);
96 static void kqueue_release(struct kqueue *kq, int locked);
97 static int kqueue_expand(struct kqueue *kq, struct filterops *fops,
98 uintptr_t ident, int waitok);
99 static void kqueue_task(void *arg, int pending);
100 static int kqueue_scan(struct kqueue *kq, int maxevents,
101 struct kevent_copyops *k_ops,
102 const struct timespec *timeout,
103 struct kevent *keva, struct thread *td);
104 static void kqueue_wakeup(struct kqueue *kq);
105 static struct filterops *kqueue_fo_find(int filt);
106 static void kqueue_fo_release(int filt);
107
108 static fo_rdwr_t kqueue_read;
109 static fo_rdwr_t kqueue_write;
110 static fo_truncate_t kqueue_truncate;
111 static fo_ioctl_t kqueue_ioctl;
112 static fo_poll_t kqueue_poll;
113 static fo_kqfilter_t kqueue_kqfilter;
114 static fo_stat_t kqueue_stat;
115 static fo_close_t kqueue_close;
116
117 static struct fileops kqueueops = {
118 .fo_read = kqueue_read,
119 .fo_write = kqueue_write,
120 .fo_truncate = kqueue_truncate,
121 .fo_ioctl = kqueue_ioctl,
122 .fo_poll = kqueue_poll,
123 .fo_kqfilter = kqueue_kqfilter,
124 .fo_stat = kqueue_stat,
125 .fo_close = kqueue_close,
126 .fo_chmod = invfo_chmod,
127 .fo_chown = invfo_chown,
128 };
129
130 static int knote_attach(struct knote *kn, struct kqueue *kq);
131 static void knote_drop(struct knote *kn, struct thread *td);
132 static void knote_enqueue(struct knote *kn);
133 static void knote_dequeue(struct knote *kn);
134 static void knote_init(void);
135 static struct knote *knote_alloc(int waitok);
136 static void knote_free(struct knote *kn);
137
138 static void filt_kqdetach(struct knote *kn);
139 static int filt_kqueue(struct knote *kn, long hint);
140 static int filt_procattach(struct knote *kn);
141 static void filt_procdetach(struct knote *kn);
142 static int filt_proc(struct knote *kn, long hint);
143 static int filt_fileattach(struct knote *kn);
144 static void filt_timerexpire(void *knx);
145 static int filt_timerattach(struct knote *kn);
146 static void filt_timerdetach(struct knote *kn);
147 static int filt_timer(struct knote *kn, long hint);
148 static int filt_userattach(struct knote *kn);
149 static void filt_userdetach(struct knote *kn);
150 static int filt_user(struct knote *kn, long hint);
151 static void filt_usertouch(struct knote *kn, struct kevent *kev,
152 u_long type);
153
154 static struct filterops file_filtops = {
155 .f_isfd = 1,
156 .f_attach = filt_fileattach,
157 };
158 static struct filterops kqread_filtops = {
159 .f_isfd = 1,
160 .f_detach = filt_kqdetach,
161 .f_event = filt_kqueue,
162 };
163 /* XXX - move to kern_proc.c? */
164 static struct filterops proc_filtops = {
165 .f_isfd = 0,
166 .f_attach = filt_procattach,
167 .f_detach = filt_procdetach,
168 .f_event = filt_proc,
169 };
170 static struct filterops timer_filtops = {
171 .f_isfd = 0,
172 .f_attach = filt_timerattach,
173 .f_detach = filt_timerdetach,
174 .f_event = filt_timer,
175 };
176 static struct filterops user_filtops = {
177 .f_attach = filt_userattach,
178 .f_detach = filt_userdetach,
179 .f_event = filt_user,
180 .f_touch = filt_usertouch,
181 };
182
183 static uma_zone_t knote_zone;
184 static int kq_ncallouts = 0;
185 static int kq_calloutmax = (4 * 1024);
186 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
187 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
188
189 /* XXX - ensure not KN_INFLUX?? */
190 #define KNOTE_ACTIVATE(kn, islock) do { \
191 if ((islock)) \
192 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \
193 else \
194 KQ_LOCK((kn)->kn_kq); \
195 (kn)->kn_status |= KN_ACTIVE; \
196 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
197 knote_enqueue((kn)); \
198 if (!(islock)) \
199 KQ_UNLOCK((kn)->kn_kq); \
200 } while(0)
201 #define KQ_LOCK(kq) do { \
202 mtx_lock(&(kq)->kq_lock); \
203 } while (0)
204 #define KQ_FLUX_WAKEUP(kq) do { \
205 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
206 (kq)->kq_state &= ~KQ_FLUXWAIT; \
207 wakeup((kq)); \
208 } \
209 } while (0)
210 #define KQ_UNLOCK_FLUX(kq) do { \
211 KQ_FLUX_WAKEUP(kq); \
212 mtx_unlock(&(kq)->kq_lock); \
213 } while (0)
214 #define KQ_UNLOCK(kq) do { \
215 mtx_unlock(&(kq)->kq_lock); \
216 } while (0)
217 #define KQ_OWNED(kq) do { \
218 mtx_assert(&(kq)->kq_lock, MA_OWNED); \
219 } while (0)
220 #define KQ_NOTOWNED(kq) do { \
221 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \
222 } while (0)
223 #define KN_LIST_LOCK(kn) do { \
224 if (kn->kn_knlist != NULL) \
225 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \
226 } while (0)
227 #define KN_LIST_UNLOCK(kn) do { \
228 if (kn->kn_knlist != NULL) \
229 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \
230 } while (0)
231 #define KNL_ASSERT_LOCK(knl, islocked) do { \
232 if (islocked) \
233 KNL_ASSERT_LOCKED(knl); \
234 else \
235 KNL_ASSERT_UNLOCKED(knl); \
236 } while (0)
237 #ifdef INVARIANTS
238 #define KNL_ASSERT_LOCKED(knl) do { \
239 knl->kl_assert_locked((knl)->kl_lockarg); \
240 } while (0)
241 #define KNL_ASSERT_UNLOCKED(knl) do { \
242 knl->kl_assert_unlocked((knl)->kl_lockarg); \
243 } while (0)
244 #else /* !INVARIANTS */
245 #define KNL_ASSERT_LOCKED(knl) do {} while(0)
246 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0)
247 #endif /* INVARIANTS */
248
249 #define KN_HASHSIZE 64 /* XXX should be tunable */
250 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
251
252 static int
253 filt_nullattach(struct knote *kn)
254 {
255
256 return (ENXIO);
257 };
258
259 struct filterops null_filtops = {
260 .f_isfd = 0,
261 .f_attach = filt_nullattach,
262 };
263
264 /* XXX - make SYSINIT to add these, and move into respective modules. */
265 extern struct filterops sig_filtops;
266 extern struct filterops fs_filtops;
267
268 /*
269 * Table for for all system-defined filters.
270 */
271 static struct mtx filterops_lock;
272 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
273 MTX_DEF);
274 static struct {
275 struct filterops *for_fop;
276 int for_refcnt;
277 } sysfilt_ops[EVFILT_SYSCOUNT] = {
278 { &file_filtops }, /* EVFILT_READ */
279 { &file_filtops }, /* EVFILT_WRITE */
280 { &null_filtops }, /* EVFILT_AIO */
281 { &file_filtops }, /* EVFILT_VNODE */
282 { &proc_filtops }, /* EVFILT_PROC */
283 { &sig_filtops }, /* EVFILT_SIGNAL */
284 { &timer_filtops }, /* EVFILT_TIMER */
285 { &null_filtops }, /* former EVFILT_NETDEV */
286 { &fs_filtops }, /* EVFILT_FS */
287 { &null_filtops }, /* EVFILT_LIO */
288 { &user_filtops }, /* EVFILT_USER */
289 };
290
291 /*
292 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
293 * method.
294 */
295 static int
296 filt_fileattach(struct knote *kn)
297 {
298
299 return (fo_kqfilter(kn->kn_fp, kn));
300 }
301
302 /*ARGSUSED*/
303 static int
304 kqueue_kqfilter(struct file *fp, struct knote *kn)
305 {
306 struct kqueue *kq = kn->kn_fp->f_data;
307
308 if (kn->kn_filter != EVFILT_READ)
309 return (EINVAL);
310
311 kn->kn_status |= KN_KQUEUE;
312 kn->kn_fop = &kqread_filtops;
313 knlist_add(&kq->kq_sel.si_note, kn, 0);
314
315 return (0);
316 }
317
318 static void
319 filt_kqdetach(struct knote *kn)
320 {
321 struct kqueue *kq = kn->kn_fp->f_data;
322
323 knlist_remove(&kq->kq_sel.si_note, kn, 0);
324 }
325
326 /*ARGSUSED*/
327 static int
328 filt_kqueue(struct knote *kn, long hint)
329 {
330 struct kqueue *kq = kn->kn_fp->f_data;
331
332 kn->kn_data = kq->kq_count;
333 return (kn->kn_data > 0);
334 }
335
336 /* XXX - move to kern_proc.c? */
337 static int
338 filt_procattach(struct knote *kn)
339 {
340 struct proc *p;
341 int immediate;
342 int error;
343
344 immediate = 0;
345 p = pfind(kn->kn_id);
346 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
347 p = zpfind(kn->kn_id);
348 immediate = 1;
349 } else if (p != NULL && (p->p_flag & P_WEXIT)) {
350 immediate = 1;
351 }
352
353 if (p == NULL)
354 return (ESRCH);
355 if ((error = p_cansee(curthread, p))) {
356 PROC_UNLOCK(p);
357 return (error);
358 }
359
360 kn->kn_ptr.p_proc = p;
361 kn->kn_flags |= EV_CLEAR; /* automatically set */
362
363 /*
364 * internal flag indicating registration done by kernel
365 */
366 if (kn->kn_flags & EV_FLAG1) {
367 kn->kn_data = kn->kn_sdata; /* ppid */
368 kn->kn_fflags = NOTE_CHILD;
369 kn->kn_flags &= ~EV_FLAG1;
370 }
371
372 if (immediate == 0)
373 knlist_add(&p->p_klist, kn, 1);
374
375 /*
376 * Immediately activate any exit notes if the target process is a
377 * zombie. This is necessary to handle the case where the target
378 * process, e.g. a child, dies before the kevent is registered.
379 */
380 if (immediate && filt_proc(kn, NOTE_EXIT))
381 KNOTE_ACTIVATE(kn, 0);
382
383 PROC_UNLOCK(p);
384
385 return (0);
386 }
387
388 /*
389 * The knote may be attached to a different process, which may exit,
390 * leaving nothing for the knote to be attached to. So when the process
391 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
392 * it will be deleted when read out. However, as part of the knote deletion,
393 * this routine is called, so a check is needed to avoid actually performing
394 * a detach, because the original process does not exist any more.
395 */
396 /* XXX - move to kern_proc.c? */
397 static void
398 filt_procdetach(struct knote *kn)
399 {
400 struct proc *p;
401
402 p = kn->kn_ptr.p_proc;
403 knlist_remove(&p->p_klist, kn, 0);
404 kn->kn_ptr.p_proc = NULL;
405 }
406
407 /* XXX - move to kern_proc.c? */
408 static int
409 filt_proc(struct knote *kn, long hint)
410 {
411 struct proc *p = kn->kn_ptr.p_proc;
412 u_int event;
413
414 /*
415 * mask off extra data
416 */
417 event = (u_int)hint & NOTE_PCTRLMASK;
418
419 /*
420 * if the user is interested in this event, record it.
421 */
422 if (kn->kn_sfflags & event)
423 kn->kn_fflags |= event;
424
425 /*
426 * process is gone, so flag the event as finished.
427 */
428 if (event == NOTE_EXIT) {
429 if (!(kn->kn_status & KN_DETACHED))
430 knlist_remove_inevent(&p->p_klist, kn);
431 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
432 kn->kn_data = p->p_xstat;
433 kn->kn_ptr.p_proc = NULL;
434 return (1);
435 }
436
437 return (kn->kn_fflags != 0);
438 }
439
440 /*
441 * Called when the process forked. It mostly does the same as the
442 * knote(), activating all knotes registered to be activated when the
443 * process forked. Additionally, for each knote attached to the
444 * parent, check whether user wants to track the new process. If so
445 * attach a new knote to it, and immediately report an event with the
446 * child's pid.
447 */
448 void
449 knote_fork(struct knlist *list, int pid)
450 {
451 struct kqueue *kq;
452 struct knote *kn;
453 struct kevent kev;
454 int error;
455
456 if (list == NULL)
457 return;
458 list->kl_lock(list->kl_lockarg);
459
460 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
461 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
462 continue;
463 kq = kn->kn_kq;
464 KQ_LOCK(kq);
465 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
466 KQ_UNLOCK(kq);
467 continue;
468 }
469
470 /*
471 * The same as knote(), activate the event.
472 */
473 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
474 kn->kn_status |= KN_HASKQLOCK;
475 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid))
476 KNOTE_ACTIVATE(kn, 1);
477 kn->kn_status &= ~KN_HASKQLOCK;
478 KQ_UNLOCK(kq);
479 continue;
480 }
481
482 /*
483 * The NOTE_TRACK case. In addition to the activation
484 * of the event, we need to register new event to
485 * track the child. Drop the locks in preparation for
486 * the call to kqueue_register().
487 */
488 kn->kn_status |= KN_INFLUX;
489 KQ_UNLOCK(kq);
490 list->kl_unlock(list->kl_lockarg);
491
492 /*
493 * Activate existing knote and register a knote with
494 * new process.
495 */
496 kev.ident = pid;
497 kev.filter = kn->kn_filter;
498 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
499 kev.fflags = kn->kn_sfflags;
500 kev.data = kn->kn_id; /* parent */
501 kev.udata = kn->kn_kevent.udata;/* preserve udata */
502 error = kqueue_register(kq, &kev, NULL, 0);
503 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid))
504 KNOTE_ACTIVATE(kn, 0);
505 if (error)
506 kn->kn_fflags |= NOTE_TRACKERR;
507 KQ_LOCK(kq);
508 kn->kn_status &= ~KN_INFLUX;
509 KQ_UNLOCK_FLUX(kq);
510 list->kl_lock(list->kl_lockarg);
511 }
512 list->kl_unlock(list->kl_lockarg);
513 }
514
515 /*
516 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
517 * interval timer support code.
518 */
519 static int
520 timertoticks(intptr_t data)
521 {
522 struct timeval tv;
523 int tticks;
524
525 tv.tv_sec = data / 1000;
526 tv.tv_usec = (data % 1000) * 1000;
527 tticks = tvtohz(&tv);
528
529 return tticks;
530 }
531
532 static void
533 filt_timerexpire(void *knx)
534 {
535 struct knote *kn = knx;
536 struct callout *calloutp;
537
538 kn->kn_data++;
539 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
540
541 /*
542 * timertoticks() uses tvtohz() which always adds 1 to allow
543 * for the time until the next clock interrupt being strictly
544 * less than 1 clock tick. We don't want that here since we
545 * want to appear to be in sync with the clock interrupt even
546 * when we're delayed.
547 */
548 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
549 calloutp = (struct callout *)kn->kn_hook;
550 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata) - 1,
551 filt_timerexpire, kn);
552 }
553 }
554
555 /*
556 * data contains amount of time to sleep, in milliseconds
557 */
558 static int
559 filt_timerattach(struct knote *kn)
560 {
561 struct callout *calloutp;
562
563 atomic_add_int(&kq_ncallouts, 1);
564
565 if (kq_ncallouts >= kq_calloutmax) {
566 atomic_add_int(&kq_ncallouts, -1);
567 return (ENOMEM);
568 }
569
570 kn->kn_flags |= EV_CLEAR; /* automatically set */
571 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */
572 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK);
573 callout_init(calloutp, CALLOUT_MPSAFE);
574 kn->kn_hook = calloutp;
575 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata),
576 filt_timerexpire, kn);
577
578 return (0);
579 }
580
581 static void
582 filt_timerdetach(struct knote *kn)
583 {
584 struct callout *calloutp;
585
586 calloutp = (struct callout *)kn->kn_hook;
587 callout_drain(calloutp);
588 free(calloutp, M_KQUEUE);
589 atomic_add_int(&kq_ncallouts, -1);
590 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */
591 }
592
593 static int
594 filt_timer(struct knote *kn, long hint)
595 {
596
597 return (kn->kn_data != 0);
598 }
599
600 static int
601 filt_userattach(struct knote *kn)
602 {
603
604 /*
605 * EVFILT_USER knotes are not attached to anything in the kernel.
606 */
607 kn->kn_hook = NULL;
608 if (kn->kn_fflags & NOTE_TRIGGER)
609 kn->kn_hookid = 1;
610 else
611 kn->kn_hookid = 0;
612 return (0);
613 }
614
615 static void
616 filt_userdetach(__unused struct knote *kn)
617 {
618
619 /*
620 * EVFILT_USER knotes are not attached to anything in the kernel.
621 */
622 }
623
624 static int
625 filt_user(struct knote *kn, __unused long hint)
626 {
627
628 return (kn->kn_hookid);
629 }
630
631 static void
632 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
633 {
634 u_int ffctrl;
635
636 switch (type) {
637 case EVENT_REGISTER:
638 if (kev->fflags & NOTE_TRIGGER)
639 kn->kn_hookid = 1;
640
641 ffctrl = kev->fflags & NOTE_FFCTRLMASK;
642 kev->fflags &= NOTE_FFLAGSMASK;
643 switch (ffctrl) {
644 case NOTE_FFNOP:
645 break;
646
647 case NOTE_FFAND:
648 kn->kn_sfflags &= kev->fflags;
649 break;
650
651 case NOTE_FFOR:
652 kn->kn_sfflags |= kev->fflags;
653 break;
654
655 case NOTE_FFCOPY:
656 kn->kn_sfflags = kev->fflags;
657 break;
658
659 default:
660 /* XXX Return error? */
661 break;
662 }
663 kn->kn_sdata = kev->data;
664 if (kev->flags & EV_CLEAR) {
665 kn->kn_hookid = 0;
666 kn->kn_data = 0;
667 kn->kn_fflags = 0;
668 }
669 break;
670
671 case EVENT_PROCESS:
672 *kev = kn->kn_kevent;
673 kev->fflags = kn->kn_sfflags;
674 kev->data = kn->kn_sdata;
675 if (kn->kn_flags & EV_CLEAR) {
676 kn->kn_hookid = 0;
677 kn->kn_data = 0;
678 kn->kn_fflags = 0;
679 }
680 break;
681
682 default:
683 panic("filt_usertouch() - invalid type (%ld)", type);
684 break;
685 }
686 }
687
688 int
689 sys_kqueue(struct thread *td, struct kqueue_args *uap)
690 {
691 struct filedesc *fdp;
692 struct kqueue *kq;
693 struct file *fp;
694 int fd, error;
695
696 fdp = td->td_proc->p_fd;
697 error = falloc(td, &fp, &fd, 0);
698 if (error)
699 goto done2;
700
701 /* An extra reference on `nfp' has been held for us by falloc(). */
702 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
703 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
704 TAILQ_INIT(&kq->kq_head);
705 kq->kq_fdp = fdp;
706 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
707 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
708
709 FILEDESC_XLOCK(fdp);
710 SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
711 FILEDESC_XUNLOCK(fdp);
712
713 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
714 fdrop(fp, td);
715
716 td->td_retval[0] = fd;
717 done2:
718 return (error);
719 }
720
721 #ifndef _SYS_SYSPROTO_H_
722 struct kevent_args {
723 int fd;
724 const struct kevent *changelist;
725 int nchanges;
726 struct kevent *eventlist;
727 int nevents;
728 const struct timespec *timeout;
729 };
730 #endif
731 int
732 sys_kevent(struct thread *td, struct kevent_args *uap)
733 {
734 struct timespec ts, *tsp;
735 struct kevent_copyops k_ops = { uap,
736 kevent_copyout,
737 kevent_copyin};
738 int error;
739 #ifdef KTRACE
740 struct uio ktruio;
741 struct iovec ktriov;
742 struct uio *ktruioin = NULL;
743 struct uio *ktruioout = NULL;
744 #endif
745
746 if (uap->timeout != NULL) {
747 error = copyin(uap->timeout, &ts, sizeof(ts));
748 if (error)
749 return (error);
750 tsp = &ts;
751 } else
752 tsp = NULL;
753
754 #ifdef KTRACE
755 if (KTRPOINT(td, KTR_GENIO)) {
756 ktriov.iov_base = uap->changelist;
757 ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
758 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
759 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
760 .uio_td = td };
761 ktruioin = cloneuio(&ktruio);
762 ktriov.iov_base = uap->eventlist;
763 ktriov.iov_len = uap->nevents * sizeof(struct kevent);
764 ktruioout = cloneuio(&ktruio);
765 }
766 #endif
767
768 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
769 &k_ops, tsp);
770
771 #ifdef KTRACE
772 if (ktruioin != NULL) {
773 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
774 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
775 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
776 ktrgenio(uap->fd, UIO_READ, ktruioout, error);
777 }
778 #endif
779
780 return (error);
781 }
782
783 /*
784 * Copy 'count' items into the destination list pointed to by uap->eventlist.
785 */
786 static int
787 kevent_copyout(void *arg, struct kevent *kevp, int count)
788 {
789 struct kevent_args *uap;
790 int error;
791
792 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
793 uap = (struct kevent_args *)arg;
794
795 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
796 if (error == 0)
797 uap->eventlist += count;
798 return (error);
799 }
800
801 /*
802 * Copy 'count' items from the list pointed to by uap->changelist.
803 */
804 static int
805 kevent_copyin(void *arg, struct kevent *kevp, int count)
806 {
807 struct kevent_args *uap;
808 int error;
809
810 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
811 uap = (struct kevent_args *)arg;
812
813 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
814 if (error == 0)
815 uap->changelist += count;
816 return (error);
817 }
818
819 int
820 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
821 struct kevent_copyops *k_ops, const struct timespec *timeout)
822 {
823 struct kevent keva[KQ_NEVENTS];
824 struct kevent *kevp, *changes;
825 struct kqueue *kq;
826 struct file *fp;
827 int i, n, nerrors, error;
828
829 if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0)
830 return (error);
831 if ((error = kqueue_acquire(fp, &kq)) != 0)
832 goto done_norel;
833
834 nerrors = 0;
835
836 while (nchanges > 0) {
837 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
838 error = k_ops->k_copyin(k_ops->arg, keva, n);
839 if (error)
840 goto done;
841 changes = keva;
842 for (i = 0; i < n; i++) {
843 kevp = &changes[i];
844 if (!kevp->filter)
845 continue;
846 kevp->flags &= ~EV_SYSFLAGS;
847 error = kqueue_register(kq, kevp, td, 1);
848 if (error || (kevp->flags & EV_RECEIPT)) {
849 if (nevents != 0) {
850 kevp->flags = EV_ERROR;
851 kevp->data = error;
852 (void) k_ops->k_copyout(k_ops->arg,
853 kevp, 1);
854 nevents--;
855 nerrors++;
856 } else {
857 goto done;
858 }
859 }
860 }
861 nchanges -= n;
862 }
863 if (nerrors) {
864 td->td_retval[0] = nerrors;
865 error = 0;
866 goto done;
867 }
868
869 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
870 done:
871 kqueue_release(kq, 0);
872 done_norel:
873 fdrop(fp, td);
874 return (error);
875 }
876
877 int
878 kqueue_add_filteropts(int filt, struct filterops *filtops)
879 {
880 int error;
881
882 error = 0;
883 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
884 printf(
885 "trying to add a filterop that is out of range: %d is beyond %d\n",
886 ~filt, EVFILT_SYSCOUNT);
887 return EINVAL;
888 }
889 mtx_lock(&filterops_lock);
890 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
891 sysfilt_ops[~filt].for_fop != NULL)
892 error = EEXIST;
893 else {
894 sysfilt_ops[~filt].for_fop = filtops;
895 sysfilt_ops[~filt].for_refcnt = 0;
896 }
897 mtx_unlock(&filterops_lock);
898
899 return (error);
900 }
901
902 int
903 kqueue_del_filteropts(int filt)
904 {
905 int error;
906
907 error = 0;
908 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
909 return EINVAL;
910
911 mtx_lock(&filterops_lock);
912 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
913 sysfilt_ops[~filt].for_fop == NULL)
914 error = EINVAL;
915 else if (sysfilt_ops[~filt].for_refcnt != 0)
916 error = EBUSY;
917 else {
918 sysfilt_ops[~filt].for_fop = &null_filtops;
919 sysfilt_ops[~filt].for_refcnt = 0;
920 }
921 mtx_unlock(&filterops_lock);
922
923 return error;
924 }
925
926 static struct filterops *
927 kqueue_fo_find(int filt)
928 {
929
930 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
931 return NULL;
932
933 mtx_lock(&filterops_lock);
934 sysfilt_ops[~filt].for_refcnt++;
935 if (sysfilt_ops[~filt].for_fop == NULL)
936 sysfilt_ops[~filt].for_fop = &null_filtops;
937 mtx_unlock(&filterops_lock);
938
939 return sysfilt_ops[~filt].for_fop;
940 }
941
942 static void
943 kqueue_fo_release(int filt)
944 {
945
946 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
947 return;
948
949 mtx_lock(&filterops_lock);
950 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
951 ("filter object refcount not valid on release"));
952 sysfilt_ops[~filt].for_refcnt--;
953 mtx_unlock(&filterops_lock);
954 }
955
956 /*
957 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will
958 * influence if memory allocation should wait. Make sure it is 0 if you
959 * hold any mutexes.
960 */
961 static int
962 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
963 {
964 struct filterops *fops;
965 struct file *fp;
966 struct knote *kn, *tkn;
967 int error, filt, event;
968 int haskqglobal;
969
970 fp = NULL;
971 kn = NULL;
972 error = 0;
973 haskqglobal = 0;
974
975 filt = kev->filter;
976 fops = kqueue_fo_find(filt);
977 if (fops == NULL)
978 return EINVAL;
979
980 tkn = knote_alloc(waitok); /* prevent waiting with locks */
981
982 findkn:
983 if (fops->f_isfd) {
984 KASSERT(td != NULL, ("td is NULL"));
985 error = fget(td, kev->ident, CAP_POLL_EVENT, &fp);
986 if (error)
987 goto done;
988
989 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
990 kev->ident, 0) != 0) {
991 /* try again */
992 fdrop(fp, td);
993 fp = NULL;
994 error = kqueue_expand(kq, fops, kev->ident, waitok);
995 if (error)
996 goto done;
997 goto findkn;
998 }
999
1000 if (fp->f_type == DTYPE_KQUEUE) {
1001 /*
1002 * if we add some inteligence about what we are doing,
1003 * we should be able to support events on ourselves.
1004 * We need to know when we are doing this to prevent
1005 * getting both the knlist lock and the kq lock since
1006 * they are the same thing.
1007 */
1008 if (fp->f_data == kq) {
1009 error = EINVAL;
1010 goto done;
1011 }
1012
1013 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1014 }
1015
1016 KQ_LOCK(kq);
1017 if (kev->ident < kq->kq_knlistsize) {
1018 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1019 if (kev->filter == kn->kn_filter)
1020 break;
1021 }
1022 } else {
1023 if ((kev->flags & EV_ADD) == EV_ADD)
1024 kqueue_expand(kq, fops, kev->ident, waitok);
1025
1026 KQ_LOCK(kq);
1027 if (kq->kq_knhashmask != 0) {
1028 struct klist *list;
1029
1030 list = &kq->kq_knhash[
1031 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1032 SLIST_FOREACH(kn, list, kn_link)
1033 if (kev->ident == kn->kn_id &&
1034 kev->filter == kn->kn_filter)
1035 break;
1036 }
1037 }
1038
1039 /* knote is in the process of changing, wait for it to stablize. */
1040 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1041 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1042 kq->kq_state |= KQ_FLUXWAIT;
1043 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1044 if (fp != NULL) {
1045 fdrop(fp, td);
1046 fp = NULL;
1047 }
1048 goto findkn;
1049 }
1050
1051 /*
1052 * kn now contains the matching knote, or NULL if no match
1053 */
1054 if (kn == NULL) {
1055 if (kev->flags & EV_ADD) {
1056 kn = tkn;
1057 tkn = NULL;
1058 if (kn == NULL) {
1059 KQ_UNLOCK(kq);
1060 error = ENOMEM;
1061 goto done;
1062 }
1063 kn->kn_fp = fp;
1064 kn->kn_kq = kq;
1065 kn->kn_fop = fops;
1066 /*
1067 * apply reference counts to knote structure, and
1068 * do not release it at the end of this routine.
1069 */
1070 fops = NULL;
1071 fp = NULL;
1072
1073 kn->kn_sfflags = kev->fflags;
1074 kn->kn_sdata = kev->data;
1075 kev->fflags = 0;
1076 kev->data = 0;
1077 kn->kn_kevent = *kev;
1078 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1079 EV_ENABLE | EV_DISABLE);
1080 kn->kn_status = KN_INFLUX|KN_DETACHED;
1081
1082 error = knote_attach(kn, kq);
1083 KQ_UNLOCK(kq);
1084 if (error != 0) {
1085 tkn = kn;
1086 goto done;
1087 }
1088
1089 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1090 knote_drop(kn, td);
1091 goto done;
1092 }
1093 KN_LIST_LOCK(kn);
1094 goto done_ev_add;
1095 } else {
1096 /* No matching knote and the EV_ADD flag is not set. */
1097 KQ_UNLOCK(kq);
1098 error = ENOENT;
1099 goto done;
1100 }
1101 }
1102
1103 if (kev->flags & EV_DELETE) {
1104 kn->kn_status |= KN_INFLUX;
1105 KQ_UNLOCK(kq);
1106 if (!(kn->kn_status & KN_DETACHED))
1107 kn->kn_fop->f_detach(kn);
1108 knote_drop(kn, td);
1109 goto done;
1110 }
1111
1112 /*
1113 * The user may change some filter values after the initial EV_ADD,
1114 * but doing so will not reset any filter which has already been
1115 * triggered.
1116 */
1117 kn->kn_status |= KN_INFLUX;
1118 KQ_UNLOCK(kq);
1119 KN_LIST_LOCK(kn);
1120 kn->kn_kevent.udata = kev->udata;
1121 if (!fops->f_isfd && fops->f_touch != NULL) {
1122 fops->f_touch(kn, kev, EVENT_REGISTER);
1123 } else {
1124 kn->kn_sfflags = kev->fflags;
1125 kn->kn_sdata = kev->data;
1126 }
1127
1128 /*
1129 * We can get here with kn->kn_knlist == NULL. This can happen when
1130 * the initial attach event decides that the event is "completed"
1131 * already. i.e. filt_procattach is called on a zombie process. It
1132 * will call filt_proc which will remove it from the list, and NULL
1133 * kn_knlist.
1134 */
1135 done_ev_add:
1136 event = kn->kn_fop->f_event(kn, 0);
1137 KQ_LOCK(kq);
1138 if (event)
1139 KNOTE_ACTIVATE(kn, 1);
1140 kn->kn_status &= ~KN_INFLUX;
1141 KN_LIST_UNLOCK(kn);
1142
1143 if ((kev->flags & EV_DISABLE) &&
1144 ((kn->kn_status & KN_DISABLED) == 0)) {
1145 kn->kn_status |= KN_DISABLED;
1146 }
1147
1148 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1149 kn->kn_status &= ~KN_DISABLED;
1150 if ((kn->kn_status & KN_ACTIVE) &&
1151 ((kn->kn_status & KN_QUEUED) == 0))
1152 knote_enqueue(kn);
1153 }
1154 KQ_UNLOCK_FLUX(kq);
1155
1156 done:
1157 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1158 if (fp != NULL)
1159 fdrop(fp, td);
1160 if (tkn != NULL)
1161 knote_free(tkn);
1162 if (fops != NULL)
1163 kqueue_fo_release(filt);
1164 return (error);
1165 }
1166
1167 static int
1168 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1169 {
1170 int error;
1171 struct kqueue *kq;
1172
1173 error = 0;
1174
1175 kq = fp->f_data;
1176 if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1177 return (EBADF);
1178 *kqp = kq;
1179 KQ_LOCK(kq);
1180 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1181 KQ_UNLOCK(kq);
1182 return (EBADF);
1183 }
1184 kq->kq_refcnt++;
1185 KQ_UNLOCK(kq);
1186
1187 return error;
1188 }
1189
1190 static void
1191 kqueue_release(struct kqueue *kq, int locked)
1192 {
1193 if (locked)
1194 KQ_OWNED(kq);
1195 else
1196 KQ_LOCK(kq);
1197 kq->kq_refcnt--;
1198 if (kq->kq_refcnt == 1)
1199 wakeup(&kq->kq_refcnt);
1200 if (!locked)
1201 KQ_UNLOCK(kq);
1202 }
1203
1204 static void
1205 kqueue_schedtask(struct kqueue *kq)
1206 {
1207
1208 KQ_OWNED(kq);
1209 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1210 ("scheduling kqueue task while draining"));
1211
1212 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1213 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1214 kq->kq_state |= KQ_TASKSCHED;
1215 }
1216 }
1217
1218 /*
1219 * Expand the kq to make sure we have storage for fops/ident pair.
1220 *
1221 * Return 0 on success (or no work necessary), return errno on failure.
1222 *
1223 * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1224 * If kqueue_register is called from a non-fd context, there usually/should
1225 * be no locks held.
1226 */
1227 static int
1228 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1229 int waitok)
1230 {
1231 struct klist *list, *tmp_knhash, *to_free;
1232 u_long tmp_knhashmask;
1233 int size;
1234 int fd;
1235 int mflag = waitok ? M_WAITOK : M_NOWAIT;
1236
1237 KQ_NOTOWNED(kq);
1238
1239 to_free = NULL;
1240 if (fops->f_isfd) {
1241 fd = ident;
1242 if (kq->kq_knlistsize <= fd) {
1243 size = kq->kq_knlistsize;
1244 while (size <= fd)
1245 size += KQEXTENT;
1246 list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1247 if (list == NULL)
1248 return ENOMEM;
1249 KQ_LOCK(kq);
1250 if (kq->kq_knlistsize > fd) {
1251 to_free = list;
1252 list = NULL;
1253 } else {
1254 if (kq->kq_knlist != NULL) {
1255 bcopy(kq->kq_knlist, list,
1256 kq->kq_knlistsize * sizeof(*list));
1257 to_free = kq->kq_knlist;
1258 kq->kq_knlist = NULL;
1259 }
1260 bzero((caddr_t)list +
1261 kq->kq_knlistsize * sizeof(*list),
1262 (size - kq->kq_knlistsize) * sizeof(*list));
1263 kq->kq_knlistsize = size;
1264 kq->kq_knlist = list;
1265 }
1266 KQ_UNLOCK(kq);
1267 }
1268 } else {
1269 if (kq->kq_knhashmask == 0) {
1270 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1271 &tmp_knhashmask);
1272 if (tmp_knhash == NULL)
1273 return ENOMEM;
1274 KQ_LOCK(kq);
1275 if (kq->kq_knhashmask == 0) {
1276 kq->kq_knhash = tmp_knhash;
1277 kq->kq_knhashmask = tmp_knhashmask;
1278 } else {
1279 to_free = tmp_knhash;
1280 }
1281 KQ_UNLOCK(kq);
1282 }
1283 }
1284 free(to_free, M_KQUEUE);
1285
1286 KQ_NOTOWNED(kq);
1287 return 0;
1288 }
1289
1290 static void
1291 kqueue_task(void *arg, int pending)
1292 {
1293 struct kqueue *kq;
1294 int haskqglobal;
1295
1296 haskqglobal = 0;
1297 kq = arg;
1298
1299 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1300 KQ_LOCK(kq);
1301
1302 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1303
1304 kq->kq_state &= ~KQ_TASKSCHED;
1305 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1306 wakeup(&kq->kq_state);
1307 }
1308 KQ_UNLOCK(kq);
1309 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1310 }
1311
1312 /*
1313 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1314 * We treat KN_MARKER knotes as if they are INFLUX.
1315 */
1316 static int
1317 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1318 const struct timespec *tsp, struct kevent *keva, struct thread *td)
1319 {
1320 struct kevent *kevp;
1321 struct timeval atv, rtv, ttv;
1322 struct knote *kn, *marker;
1323 int count, timeout, nkev, error, influx;
1324 int haskqglobal, touch;
1325
1326 count = maxevents;
1327 nkev = 0;
1328 error = 0;
1329 haskqglobal = 0;
1330
1331 if (maxevents == 0)
1332 goto done_nl;
1333
1334 if (tsp != NULL) {
1335 TIMESPEC_TO_TIMEVAL(&atv, tsp);
1336 if (itimerfix(&atv)) {
1337 error = EINVAL;
1338 goto done_nl;
1339 }
1340 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1341 timeout = -1;
1342 else
1343 timeout = atv.tv_sec > 24 * 60 * 60 ?
1344 24 * 60 * 60 * hz : tvtohz(&atv);
1345 getmicrouptime(&rtv);
1346 timevaladd(&atv, &rtv);
1347 } else {
1348 atv.tv_sec = 0;
1349 atv.tv_usec = 0;
1350 timeout = 0;
1351 }
1352 marker = knote_alloc(1);
1353 if (marker == NULL) {
1354 error = ENOMEM;
1355 goto done_nl;
1356 }
1357 marker->kn_status = KN_MARKER;
1358 KQ_LOCK(kq);
1359 goto start;
1360
1361 retry:
1362 if (atv.tv_sec || atv.tv_usec) {
1363 getmicrouptime(&rtv);
1364 if (timevalcmp(&rtv, &atv, >=))
1365 goto done;
1366 ttv = atv;
1367 timevalsub(&ttv, &rtv);
1368 timeout = ttv.tv_sec > 24 * 60 * 60 ?
1369 24 * 60 * 60 * hz : tvtohz(&ttv);
1370 }
1371
1372 start:
1373 kevp = keva;
1374 if (kq->kq_count == 0) {
1375 if (timeout < 0) {
1376 error = EWOULDBLOCK;
1377 } else {
1378 kq->kq_state |= KQ_SLEEP;
1379 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH,
1380 "kqread", timeout);
1381 }
1382 if (error == 0)
1383 goto retry;
1384 /* don't restart after signals... */
1385 if (error == ERESTART)
1386 error = EINTR;
1387 else if (error == EWOULDBLOCK)
1388 error = 0;
1389 goto done;
1390 }
1391
1392 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1393 influx = 0;
1394 while (count) {
1395 KQ_OWNED(kq);
1396 kn = TAILQ_FIRST(&kq->kq_head);
1397
1398 if ((kn->kn_status == KN_MARKER && kn != marker) ||
1399 (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1400 if (influx) {
1401 influx = 0;
1402 KQ_FLUX_WAKEUP(kq);
1403 }
1404 kq->kq_state |= KQ_FLUXWAIT;
1405 error = msleep(kq, &kq->kq_lock, PSOCK,
1406 "kqflxwt", 0);
1407 continue;
1408 }
1409
1410 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1411 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1412 kn->kn_status &= ~KN_QUEUED;
1413 kq->kq_count--;
1414 continue;
1415 }
1416 if (kn == marker) {
1417 KQ_FLUX_WAKEUP(kq);
1418 if (count == maxevents)
1419 goto retry;
1420 goto done;
1421 }
1422 KASSERT((kn->kn_status & KN_INFLUX) == 0,
1423 ("KN_INFLUX set when not suppose to be"));
1424
1425 if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1426 kn->kn_status &= ~KN_QUEUED;
1427 kn->kn_status |= KN_INFLUX;
1428 kq->kq_count--;
1429 KQ_UNLOCK(kq);
1430 /*
1431 * We don't need to lock the list since we've marked
1432 * it _INFLUX.
1433 */
1434 *kevp = kn->kn_kevent;
1435 if (!(kn->kn_status & KN_DETACHED))
1436 kn->kn_fop->f_detach(kn);
1437 knote_drop(kn, td);
1438 KQ_LOCK(kq);
1439 kn = NULL;
1440 } else {
1441 kn->kn_status |= KN_INFLUX;
1442 KQ_UNLOCK(kq);
1443 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1444 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1445 KN_LIST_LOCK(kn);
1446 if (kn->kn_fop->f_event(kn, 0) == 0) {
1447 KQ_LOCK(kq);
1448 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1449 kn->kn_status &=
1450 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX);
1451 kq->kq_count--;
1452 KN_LIST_UNLOCK(kn);
1453 influx = 1;
1454 continue;
1455 }
1456 touch = (!kn->kn_fop->f_isfd &&
1457 kn->kn_fop->f_touch != NULL);
1458 if (touch)
1459 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1460 else
1461 *kevp = kn->kn_kevent;
1462 KQ_LOCK(kq);
1463 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1464 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1465 /*
1466 * Manually clear knotes who weren't
1467 * 'touch'ed.
1468 */
1469 if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1470 kn->kn_data = 0;
1471 kn->kn_fflags = 0;
1472 }
1473 if (kn->kn_flags & EV_DISPATCH)
1474 kn->kn_status |= KN_DISABLED;
1475 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1476 kq->kq_count--;
1477 } else
1478 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1479
1480 kn->kn_status &= ~(KN_INFLUX);
1481 KN_LIST_UNLOCK(kn);
1482 influx = 1;
1483 }
1484
1485 /* we are returning a copy to the user */
1486 kevp++;
1487 nkev++;
1488 count--;
1489
1490 if (nkev == KQ_NEVENTS) {
1491 influx = 0;
1492 KQ_UNLOCK_FLUX(kq);
1493 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1494 nkev = 0;
1495 kevp = keva;
1496 KQ_LOCK(kq);
1497 if (error)
1498 break;
1499 }
1500 }
1501 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1502 done:
1503 KQ_OWNED(kq);
1504 KQ_UNLOCK_FLUX(kq);
1505 knote_free(marker);
1506 done_nl:
1507 KQ_NOTOWNED(kq);
1508 if (nkev != 0)
1509 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1510 td->td_retval[0] = maxevents - count;
1511 return (error);
1512 }
1513
1514 /*
1515 * XXX
1516 * This could be expanded to call kqueue_scan, if desired.
1517 */
1518 /*ARGSUSED*/
1519 static int
1520 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1521 int flags, struct thread *td)
1522 {
1523 return (ENXIO);
1524 }
1525
1526 /*ARGSUSED*/
1527 static int
1528 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1529 int flags, struct thread *td)
1530 {
1531 return (ENXIO);
1532 }
1533
1534 /*ARGSUSED*/
1535 static int
1536 kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1537 struct thread *td)
1538 {
1539
1540 return (EINVAL);
1541 }
1542
1543 /*ARGSUSED*/
1544 static int
1545 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1546 struct ucred *active_cred, struct thread *td)
1547 {
1548 /*
1549 * Enabling sigio causes two major problems:
1550 * 1) infinite recursion:
1551 * Synopsys: kevent is being used to track signals and have FIOASYNC
1552 * set. On receipt of a signal this will cause a kqueue to recurse
1553 * into itself over and over. Sending the sigio causes the kqueue
1554 * to become ready, which in turn posts sigio again, forever.
1555 * Solution: this can be solved by setting a flag in the kqueue that
1556 * we have a SIGIO in progress.
1557 * 2) locking problems:
1558 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1559 * us above the proc and pgrp locks.
1560 * Solution: Post a signal using an async mechanism, being sure to
1561 * record a generation count in the delivery so that we do not deliver
1562 * a signal to the wrong process.
1563 *
1564 * Note, these two mechanisms are somewhat mutually exclusive!
1565 */
1566 #if 0
1567 struct kqueue *kq;
1568
1569 kq = fp->f_data;
1570 switch (cmd) {
1571 case FIOASYNC:
1572 if (*(int *)data) {
1573 kq->kq_state |= KQ_ASYNC;
1574 } else {
1575 kq->kq_state &= ~KQ_ASYNC;
1576 }
1577 return (0);
1578
1579 case FIOSETOWN:
1580 return (fsetown(*(int *)data, &kq->kq_sigio));
1581
1582 case FIOGETOWN:
1583 *(int *)data = fgetown(&kq->kq_sigio);
1584 return (0);
1585 }
1586 #endif
1587
1588 return (ENOTTY);
1589 }
1590
1591 /*ARGSUSED*/
1592 static int
1593 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1594 struct thread *td)
1595 {
1596 struct kqueue *kq;
1597 int revents = 0;
1598 int error;
1599
1600 if ((error = kqueue_acquire(fp, &kq)))
1601 return POLLERR;
1602
1603 KQ_LOCK(kq);
1604 if (events & (POLLIN | POLLRDNORM)) {
1605 if (kq->kq_count) {
1606 revents |= events & (POLLIN | POLLRDNORM);
1607 } else {
1608 selrecord(td, &kq->kq_sel);
1609 if (SEL_WAITING(&kq->kq_sel))
1610 kq->kq_state |= KQ_SEL;
1611 }
1612 }
1613 kqueue_release(kq, 1);
1614 KQ_UNLOCK(kq);
1615 return (revents);
1616 }
1617
1618 /*ARGSUSED*/
1619 static int
1620 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1621 struct thread *td)
1622 {
1623
1624 bzero((void *)st, sizeof *st);
1625 /*
1626 * We no longer return kq_count because the unlocked value is useless.
1627 * If you spent all this time getting the count, why not spend your
1628 * syscall better by calling kevent?
1629 *
1630 * XXX - This is needed for libc_r.
1631 */
1632 st->st_mode = S_IFIFO;
1633 return (0);
1634 }
1635
1636 /*ARGSUSED*/
1637 static int
1638 kqueue_close(struct file *fp, struct thread *td)
1639 {
1640 struct kqueue *kq = fp->f_data;
1641 struct filedesc *fdp;
1642 struct knote *kn;
1643 int i;
1644 int error;
1645
1646 if ((error = kqueue_acquire(fp, &kq)))
1647 return error;
1648
1649 KQ_LOCK(kq);
1650
1651 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1652 ("kqueue already closing"));
1653 kq->kq_state |= KQ_CLOSING;
1654 if (kq->kq_refcnt > 1)
1655 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1656
1657 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1658 fdp = kq->kq_fdp;
1659
1660 KASSERT(knlist_empty(&kq->kq_sel.si_note),
1661 ("kqueue's knlist not empty"));
1662
1663 for (i = 0; i < kq->kq_knlistsize; i++) {
1664 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1665 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1666 kq->kq_state |= KQ_FLUXWAIT;
1667 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1668 continue;
1669 }
1670 kn->kn_status |= KN_INFLUX;
1671 KQ_UNLOCK(kq);
1672 if (!(kn->kn_status & KN_DETACHED))
1673 kn->kn_fop->f_detach(kn);
1674 knote_drop(kn, td);
1675 KQ_LOCK(kq);
1676 }
1677 }
1678 if (kq->kq_knhashmask != 0) {
1679 for (i = 0; i <= kq->kq_knhashmask; i++) {
1680 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1681 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1682 kq->kq_state |= KQ_FLUXWAIT;
1683 msleep(kq, &kq->kq_lock, PSOCK,
1684 "kqclo2", 0);
1685 continue;
1686 }
1687 kn->kn_status |= KN_INFLUX;
1688 KQ_UNLOCK(kq);
1689 if (!(kn->kn_status & KN_DETACHED))
1690 kn->kn_fop->f_detach(kn);
1691 knote_drop(kn, td);
1692 KQ_LOCK(kq);
1693 }
1694 }
1695 }
1696
1697 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1698 kq->kq_state |= KQ_TASKDRAIN;
1699 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1700 }
1701
1702 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1703 selwakeuppri(&kq->kq_sel, PSOCK);
1704 if (!SEL_WAITING(&kq->kq_sel))
1705 kq->kq_state &= ~KQ_SEL;
1706 }
1707
1708 KQ_UNLOCK(kq);
1709
1710 FILEDESC_XLOCK(fdp);
1711 SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list);
1712 FILEDESC_XUNLOCK(fdp);
1713
1714 seldrain(&kq->kq_sel);
1715 knlist_destroy(&kq->kq_sel.si_note);
1716 mtx_destroy(&kq->kq_lock);
1717 kq->kq_fdp = NULL;
1718
1719 if (kq->kq_knhash != NULL)
1720 free(kq->kq_knhash, M_KQUEUE);
1721 if (kq->kq_knlist != NULL)
1722 free(kq->kq_knlist, M_KQUEUE);
1723
1724 funsetown(&kq->kq_sigio);
1725 free(kq, M_KQUEUE);
1726 fp->f_data = NULL;
1727
1728 return (0);
1729 }
1730
1731 static void
1732 kqueue_wakeup(struct kqueue *kq)
1733 {
1734 KQ_OWNED(kq);
1735
1736 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1737 kq->kq_state &= ~KQ_SLEEP;
1738 wakeup(kq);
1739 }
1740 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1741 selwakeuppri(&kq->kq_sel, PSOCK);
1742 if (!SEL_WAITING(&kq->kq_sel))
1743 kq->kq_state &= ~KQ_SEL;
1744 }
1745 if (!knlist_empty(&kq->kq_sel.si_note))
1746 kqueue_schedtask(kq);
1747 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1748 pgsigio(&kq->kq_sigio, SIGIO, 0);
1749 }
1750 }
1751
1752 /*
1753 * Walk down a list of knotes, activating them if their event has triggered.
1754 *
1755 * There is a possibility to optimize in the case of one kq watching another.
1756 * Instead of scheduling a task to wake it up, you could pass enough state
1757 * down the chain to make up the parent kqueue. Make this code functional
1758 * first.
1759 */
1760 void
1761 knote(struct knlist *list, long hint, int lockflags)
1762 {
1763 struct kqueue *kq;
1764 struct knote *kn;
1765 int error;
1766
1767 if (list == NULL)
1768 return;
1769
1770 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1771
1772 if ((lockflags & KNF_LISTLOCKED) == 0)
1773 list->kl_lock(list->kl_lockarg);
1774
1775 /*
1776 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1777 * the kqueue scheduling, but this will introduce four
1778 * lock/unlock's for each knote to test. If we do, continue to use
1779 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1780 * only safe if you want to remove the current item, which we are
1781 * not doing.
1782 */
1783 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1784 kq = kn->kn_kq;
1785 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1786 KQ_LOCK(kq);
1787 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1788 KQ_UNLOCK(kq);
1789 } else if ((lockflags & KNF_NOKQLOCK) != 0) {
1790 kn->kn_status |= KN_INFLUX;
1791 KQ_UNLOCK(kq);
1792 error = kn->kn_fop->f_event(kn, hint);
1793 KQ_LOCK(kq);
1794 kn->kn_status &= ~KN_INFLUX;
1795 if (error)
1796 KNOTE_ACTIVATE(kn, 1);
1797 KQ_UNLOCK_FLUX(kq);
1798 } else {
1799 kn->kn_status |= KN_HASKQLOCK;
1800 if (kn->kn_fop->f_event(kn, hint))
1801 KNOTE_ACTIVATE(kn, 1);
1802 kn->kn_status &= ~KN_HASKQLOCK;
1803 KQ_UNLOCK(kq);
1804 }
1805 }
1806 kq = NULL;
1807 }
1808 if ((lockflags & KNF_LISTLOCKED) == 0)
1809 list->kl_unlock(list->kl_lockarg);
1810 }
1811
1812 /*
1813 * add a knote to a knlist
1814 */
1815 void
1816 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1817 {
1818 KNL_ASSERT_LOCK(knl, islocked);
1819 KQ_NOTOWNED(kn->kn_kq);
1820 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1821 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1822 if (!islocked)
1823 knl->kl_lock(knl->kl_lockarg);
1824 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1825 if (!islocked)
1826 knl->kl_unlock(knl->kl_lockarg);
1827 KQ_LOCK(kn->kn_kq);
1828 kn->kn_knlist = knl;
1829 kn->kn_status &= ~KN_DETACHED;
1830 KQ_UNLOCK(kn->kn_kq);
1831 }
1832
1833 static void
1834 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1835 {
1836 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1837 KNL_ASSERT_LOCK(knl, knlislocked);
1838 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1839 if (!kqislocked)
1840 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1841 ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1842 if (!knlislocked)
1843 knl->kl_lock(knl->kl_lockarg);
1844 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1845 kn->kn_knlist = NULL;
1846 if (!knlislocked)
1847 knl->kl_unlock(knl->kl_lockarg);
1848 if (!kqislocked)
1849 KQ_LOCK(kn->kn_kq);
1850 kn->kn_status |= KN_DETACHED;
1851 if (!kqislocked)
1852 KQ_UNLOCK(kn->kn_kq);
1853 }
1854
1855 /*
1856 * remove all knotes from a specified klist
1857 */
1858 void
1859 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1860 {
1861
1862 knlist_remove_kq(knl, kn, islocked, 0);
1863 }
1864
1865 /*
1866 * remove knote from a specified klist while in f_event handler.
1867 */
1868 void
1869 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1870 {
1871
1872 knlist_remove_kq(knl, kn, 1,
1873 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1874 }
1875
1876 int
1877 knlist_empty(struct knlist *knl)
1878 {
1879 KNL_ASSERT_LOCKED(knl);
1880 return SLIST_EMPTY(&knl->kl_list);
1881 }
1882
1883 static struct mtx knlist_lock;
1884 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1885 MTX_DEF);
1886 static void knlist_mtx_lock(void *arg);
1887 static void knlist_mtx_unlock(void *arg);
1888
1889 static void
1890 knlist_mtx_lock(void *arg)
1891 {
1892 mtx_lock((struct mtx *)arg);
1893 }
1894
1895 static void
1896 knlist_mtx_unlock(void *arg)
1897 {
1898 mtx_unlock((struct mtx *)arg);
1899 }
1900
1901 static void
1902 knlist_mtx_assert_locked(void *arg)
1903 {
1904 mtx_assert((struct mtx *)arg, MA_OWNED);
1905 }
1906
1907 static void
1908 knlist_mtx_assert_unlocked(void *arg)
1909 {
1910 mtx_assert((struct mtx *)arg, MA_NOTOWNED);
1911 }
1912
1913 void
1914 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1915 void (*kl_unlock)(void *),
1916 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
1917 {
1918
1919 if (lock == NULL)
1920 knl->kl_lockarg = &knlist_lock;
1921 else
1922 knl->kl_lockarg = lock;
1923
1924 if (kl_lock == NULL)
1925 knl->kl_lock = knlist_mtx_lock;
1926 else
1927 knl->kl_lock = kl_lock;
1928 if (kl_unlock == NULL)
1929 knl->kl_unlock = knlist_mtx_unlock;
1930 else
1931 knl->kl_unlock = kl_unlock;
1932 if (kl_assert_locked == NULL)
1933 knl->kl_assert_locked = knlist_mtx_assert_locked;
1934 else
1935 knl->kl_assert_locked = kl_assert_locked;
1936 if (kl_assert_unlocked == NULL)
1937 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
1938 else
1939 knl->kl_assert_unlocked = kl_assert_unlocked;
1940
1941 SLIST_INIT(&knl->kl_list);
1942 }
1943
1944 void
1945 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
1946 {
1947
1948 knlist_init(knl, lock, NULL, NULL, NULL, NULL);
1949 }
1950
1951 void
1952 knlist_destroy(struct knlist *knl)
1953 {
1954
1955 #ifdef INVARIANTS
1956 /*
1957 * if we run across this error, we need to find the offending
1958 * driver and have it call knlist_clear.
1959 */
1960 if (!SLIST_EMPTY(&knl->kl_list))
1961 printf("WARNING: destroying knlist w/ knotes on it!\n");
1962 #endif
1963
1964 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
1965 SLIST_INIT(&knl->kl_list);
1966 }
1967
1968 /*
1969 * Even if we are locked, we may need to drop the lock to allow any influx
1970 * knotes time to "settle".
1971 */
1972 void
1973 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
1974 {
1975 struct knote *kn, *kn2;
1976 struct kqueue *kq;
1977
1978 if (islocked)
1979 KNL_ASSERT_LOCKED(knl);
1980 else {
1981 KNL_ASSERT_UNLOCKED(knl);
1982 again: /* need to reacquire lock since we have dropped it */
1983 knl->kl_lock(knl->kl_lockarg);
1984 }
1985
1986 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
1987 kq = kn->kn_kq;
1988 KQ_LOCK(kq);
1989 if ((kn->kn_status & KN_INFLUX)) {
1990 KQ_UNLOCK(kq);
1991 continue;
1992 }
1993 knlist_remove_kq(knl, kn, 1, 1);
1994 if (killkn) {
1995 kn->kn_status |= KN_INFLUX | KN_DETACHED;
1996 KQ_UNLOCK(kq);
1997 knote_drop(kn, td);
1998 } else {
1999 /* Make sure cleared knotes disappear soon */
2000 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2001 KQ_UNLOCK(kq);
2002 }
2003 kq = NULL;
2004 }
2005
2006 if (!SLIST_EMPTY(&knl->kl_list)) {
2007 /* there are still KN_INFLUX remaining */
2008 kn = SLIST_FIRST(&knl->kl_list);
2009 kq = kn->kn_kq;
2010 KQ_LOCK(kq);
2011 KASSERT(kn->kn_status & KN_INFLUX,
2012 ("knote removed w/o list lock"));
2013 knl->kl_unlock(knl->kl_lockarg);
2014 kq->kq_state |= KQ_FLUXWAIT;
2015 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2016 kq = NULL;
2017 goto again;
2018 }
2019
2020 if (islocked)
2021 KNL_ASSERT_LOCKED(knl);
2022 else {
2023 knl->kl_unlock(knl->kl_lockarg);
2024 KNL_ASSERT_UNLOCKED(knl);
2025 }
2026 }
2027
2028 /*
2029 * Remove all knotes referencing a specified fd must be called with FILEDESC
2030 * lock. This prevents a race where a new fd comes along and occupies the
2031 * entry and we attach a knote to the fd.
2032 */
2033 void
2034 knote_fdclose(struct thread *td, int fd)
2035 {
2036 struct filedesc *fdp = td->td_proc->p_fd;
2037 struct kqueue *kq;
2038 struct knote *kn;
2039 int influx;
2040
2041 FILEDESC_XLOCK_ASSERT(fdp);
2042
2043 /*
2044 * We shouldn't have to worry about new kevents appearing on fd
2045 * since filedesc is locked.
2046 */
2047 SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2048 KQ_LOCK(kq);
2049
2050 again:
2051 influx = 0;
2052 while (kq->kq_knlistsize > fd &&
2053 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2054 if (kn->kn_status & KN_INFLUX) {
2055 /* someone else might be waiting on our knote */
2056 if (influx)
2057 wakeup(kq);
2058 kq->kq_state |= KQ_FLUXWAIT;
2059 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2060 goto again;
2061 }
2062 kn->kn_status |= KN_INFLUX;
2063 KQ_UNLOCK(kq);
2064 if (!(kn->kn_status & KN_DETACHED))
2065 kn->kn_fop->f_detach(kn);
2066 knote_drop(kn, td);
2067 influx = 1;
2068 KQ_LOCK(kq);
2069 }
2070 KQ_UNLOCK_FLUX(kq);
2071 }
2072 }
2073
2074 static int
2075 knote_attach(struct knote *kn, struct kqueue *kq)
2076 {
2077 struct klist *list;
2078
2079 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2080 KQ_OWNED(kq);
2081
2082 if (kn->kn_fop->f_isfd) {
2083 if (kn->kn_id >= kq->kq_knlistsize)
2084 return ENOMEM;
2085 list = &kq->kq_knlist[kn->kn_id];
2086 } else {
2087 if (kq->kq_knhash == NULL)
2088 return ENOMEM;
2089 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2090 }
2091
2092 SLIST_INSERT_HEAD(list, kn, kn_link);
2093
2094 return 0;
2095 }
2096
2097 /*
2098 * knote must already have been detached using the f_detach method.
2099 * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2100 * to prevent other removal.
2101 */
2102 static void
2103 knote_drop(struct knote *kn, struct thread *td)
2104 {
2105 struct kqueue *kq;
2106 struct klist *list;
2107
2108 kq = kn->kn_kq;
2109
2110 KQ_NOTOWNED(kq);
2111 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2112 ("knote_drop called without KN_INFLUX set in kn_status"));
2113
2114 KQ_LOCK(kq);
2115 if (kn->kn_fop->f_isfd)
2116 list = &kq->kq_knlist[kn->kn_id];
2117 else
2118 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2119
2120 if (!SLIST_EMPTY(list))
2121 SLIST_REMOVE(list, kn, knote, kn_link);
2122 if (kn->kn_status & KN_QUEUED)
2123 knote_dequeue(kn);
2124 KQ_UNLOCK_FLUX(kq);
2125
2126 if (kn->kn_fop->f_isfd) {
2127 fdrop(kn->kn_fp, td);
2128 kn->kn_fp = NULL;
2129 }
2130 kqueue_fo_release(kn->kn_kevent.filter);
2131 kn->kn_fop = NULL;
2132 knote_free(kn);
2133 }
2134
2135 static void
2136 knote_enqueue(struct knote *kn)
2137 {
2138 struct kqueue *kq = kn->kn_kq;
2139
2140 KQ_OWNED(kn->kn_kq);
2141 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2142
2143 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2144 kn->kn_status |= KN_QUEUED;
2145 kq->kq_count++;
2146 kqueue_wakeup(kq);
2147 }
2148
2149 static void
2150 knote_dequeue(struct knote *kn)
2151 {
2152 struct kqueue *kq = kn->kn_kq;
2153
2154 KQ_OWNED(kn->kn_kq);
2155 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2156
2157 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2158 kn->kn_status &= ~KN_QUEUED;
2159 kq->kq_count--;
2160 }
2161
2162 static void
2163 knote_init(void)
2164 {
2165
2166 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2167 NULL, NULL, UMA_ALIGN_PTR, 0);
2168 }
2169 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2170
2171 static struct knote *
2172 knote_alloc(int waitok)
2173 {
2174 return ((struct knote *)uma_zalloc(knote_zone,
2175 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2176 }
2177
2178 static void
2179 knote_free(struct knote *kn)
2180 {
2181 if (kn != NULL)
2182 uma_zfree(knote_zone, kn);
2183 }
2184
2185 /*
2186 * Register the kev w/ the kq specified by fd.
2187 */
2188 int
2189 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2190 {
2191 struct kqueue *kq;
2192 struct file *fp;
2193 int error;
2194
2195 if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0)
2196 return (error);
2197 if ((error = kqueue_acquire(fp, &kq)) != 0)
2198 goto noacquire;
2199
2200 error = kqueue_register(kq, kev, td, waitok);
2201
2202 kqueue_release(kq, 0);
2203
2204 noacquire:
2205 fdrop(fp, td);
2206
2207 return error;
2208 }
Cache object: d316cc843fbf43cedd572fb6aac77078
|