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/8.3/sys/kern/kern_event.c 225585 2011-09-15 12:27:26Z attilio $");
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_data = p->p_xstat;
415 kn->kn_ptr.p_proc = NULL;
416 return (1);
417 }
418
419 return (kn->kn_fflags != 0);
420 }
421
422 /*
423 * Called when the process forked. It mostly does the same as the
424 * knote(), activating all knotes registered to be activated when the
425 * process forked. Additionally, for each knote attached to the
426 * parent, check whether user wants to track the new process. If so
427 * attach a new knote to it, and immediately report an event with the
428 * child's pid.
429 */
430 void
431 knote_fork(struct knlist *list, int pid)
432 {
433 struct kqueue *kq;
434 struct knote *kn;
435 struct kevent kev;
436 int error;
437
438 if (list == NULL)
439 return;
440 list->kl_lock(list->kl_lockarg);
441
442 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
443 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
444 continue;
445 kq = kn->kn_kq;
446 KQ_LOCK(kq);
447 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
448 KQ_UNLOCK(kq);
449 continue;
450 }
451
452 /*
453 * The same as knote(), activate the event.
454 */
455 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
456 kn->kn_status |= KN_HASKQLOCK;
457 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid))
458 KNOTE_ACTIVATE(kn, 1);
459 kn->kn_status &= ~KN_HASKQLOCK;
460 KQ_UNLOCK(kq);
461 continue;
462 }
463
464 /*
465 * The NOTE_TRACK case. In addition to the activation
466 * of the event, we need to register new event to
467 * track the child. Drop the locks in preparation for
468 * the call to kqueue_register().
469 */
470 kn->kn_status |= KN_INFLUX;
471 KQ_UNLOCK(kq);
472 list->kl_unlock(list->kl_lockarg);
473
474 /*
475 * Activate existing knote and register a knote with
476 * new process.
477 */
478 kev.ident = pid;
479 kev.filter = kn->kn_filter;
480 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
481 kev.fflags = kn->kn_sfflags;
482 kev.data = kn->kn_id; /* parent */
483 kev.udata = kn->kn_kevent.udata;/* preserve udata */
484 error = kqueue_register(kq, &kev, NULL, 0);
485 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid))
486 KNOTE_ACTIVATE(kn, 0);
487 if (error)
488 kn->kn_fflags |= NOTE_TRACKERR;
489 KQ_LOCK(kq);
490 kn->kn_status &= ~KN_INFLUX;
491 KQ_UNLOCK_FLUX(kq);
492 list->kl_lock(list->kl_lockarg);
493 }
494 list->kl_unlock(list->kl_lockarg);
495 }
496
497 static int
498 timertoticks(intptr_t data)
499 {
500 struct timeval tv;
501 int tticks;
502
503 tv.tv_sec = data / 1000;
504 tv.tv_usec = (data % 1000) * 1000;
505 tticks = tvtohz(&tv);
506
507 return tticks;
508 }
509
510 /* XXX - move to kern_timeout.c? */
511 static void
512 filt_timerexpire(void *knx)
513 {
514 struct knote *kn = knx;
515 struct callout *calloutp;
516
517 kn->kn_data++;
518 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
519
520 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
521 calloutp = (struct callout *)kn->kn_hook;
522 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata),
523 filt_timerexpire, kn);
524 }
525 }
526
527 /*
528 * data contains amount of time to sleep, in milliseconds
529 */
530 /* XXX - move to kern_timeout.c? */
531 static int
532 filt_timerattach(struct knote *kn)
533 {
534 struct callout *calloutp;
535
536 atomic_add_int(&kq_ncallouts, 1);
537
538 if (kq_ncallouts >= kq_calloutmax) {
539 atomic_add_int(&kq_ncallouts, -1);
540 return (ENOMEM);
541 }
542
543 kn->kn_flags |= EV_CLEAR; /* automatically set */
544 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */
545 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK);
546 callout_init(calloutp, CALLOUT_MPSAFE);
547 kn->kn_hook = calloutp;
548 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata),
549 filt_timerexpire, kn);
550
551 return (0);
552 }
553
554 /* XXX - move to kern_timeout.c? */
555 static void
556 filt_timerdetach(struct knote *kn)
557 {
558 struct callout *calloutp;
559
560 calloutp = (struct callout *)kn->kn_hook;
561 callout_drain(calloutp);
562 free(calloutp, M_KQUEUE);
563 atomic_add_int(&kq_ncallouts, -1);
564 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */
565 }
566
567 /* XXX - move to kern_timeout.c? */
568 static int
569 filt_timer(struct knote *kn, long hint)
570 {
571
572 return (kn->kn_data != 0);
573 }
574
575 static int
576 filt_userattach(struct knote *kn)
577 {
578
579 /*
580 * EVFILT_USER knotes are not attached to anything in the kernel.
581 */
582 kn->kn_hook = NULL;
583 if (kn->kn_fflags & NOTE_TRIGGER)
584 kn->kn_hookid = 1;
585 else
586 kn->kn_hookid = 0;
587 return (0);
588 }
589
590 static void
591 filt_userdetach(__unused struct knote *kn)
592 {
593
594 /*
595 * EVFILT_USER knotes are not attached to anything in the kernel.
596 */
597 }
598
599 static int
600 filt_user(struct knote *kn, __unused long hint)
601 {
602
603 return (kn->kn_hookid);
604 }
605
606 static void
607 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
608 {
609 u_int ffctrl;
610
611 switch (type) {
612 case EVENT_REGISTER:
613 if (kev->fflags & NOTE_TRIGGER)
614 kn->kn_hookid = 1;
615
616 ffctrl = kev->fflags & NOTE_FFCTRLMASK;
617 kev->fflags &= NOTE_FFLAGSMASK;
618 switch (ffctrl) {
619 case NOTE_FFNOP:
620 break;
621
622 case NOTE_FFAND:
623 kn->kn_sfflags &= kev->fflags;
624 break;
625
626 case NOTE_FFOR:
627 kn->kn_sfflags |= kev->fflags;
628 break;
629
630 case NOTE_FFCOPY:
631 kn->kn_sfflags = kev->fflags;
632 break;
633
634 default:
635 /* XXX Return error? */
636 break;
637 }
638 kn->kn_sdata = kev->data;
639 if (kev->flags & EV_CLEAR) {
640 kn->kn_hookid = 0;
641 kn->kn_data = 0;
642 kn->kn_fflags = 0;
643 }
644 break;
645
646 case EVENT_PROCESS:
647 *kev = kn->kn_kevent;
648 kev->fflags = kn->kn_sfflags;
649 kev->data = kn->kn_sdata;
650 if (kn->kn_flags & EV_CLEAR) {
651 kn->kn_hookid = 0;
652 kn->kn_data = 0;
653 kn->kn_fflags = 0;
654 }
655 break;
656
657 default:
658 panic("filt_usertouch() - invalid type (%ld)", type);
659 break;
660 }
661 }
662
663 int
664 kqueue(struct thread *td, struct kqueue_args *uap)
665 {
666 struct filedesc *fdp;
667 struct kqueue *kq;
668 struct file *fp;
669 int fd, error;
670
671 fdp = td->td_proc->p_fd;
672 error = falloc(td, &fp, &fd);
673 if (error)
674 goto done2;
675
676 /* An extra reference on `nfp' has been held for us by falloc(). */
677 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
678 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
679 TAILQ_INIT(&kq->kq_head);
680 kq->kq_fdp = fdp;
681 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
682 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
683
684 FILEDESC_XLOCK(fdp);
685 SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
686 FILEDESC_XUNLOCK(fdp);
687
688 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
689 fdrop(fp, td);
690
691 td->td_retval[0] = fd;
692 done2:
693 return (error);
694 }
695
696 #ifndef _SYS_SYSPROTO_H_
697 struct kevent_args {
698 int fd;
699 const struct kevent *changelist;
700 int nchanges;
701 struct kevent *eventlist;
702 int nevents;
703 const struct timespec *timeout;
704 };
705 #endif
706 int
707 kevent(struct thread *td, struct kevent_args *uap)
708 {
709 struct timespec ts, *tsp;
710 struct kevent_copyops k_ops = { uap,
711 kevent_copyout,
712 kevent_copyin};
713 int error;
714 #ifdef KTRACE
715 struct uio ktruio;
716 struct iovec ktriov;
717 struct uio *ktruioin = NULL;
718 struct uio *ktruioout = NULL;
719 #endif
720
721 if (uap->timeout != NULL) {
722 error = copyin(uap->timeout, &ts, sizeof(ts));
723 if (error)
724 return (error);
725 tsp = &ts;
726 } else
727 tsp = NULL;
728
729 #ifdef KTRACE
730 if (KTRPOINT(td, KTR_GENIO)) {
731 ktriov.iov_base = uap->changelist;
732 ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
733 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
734 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
735 .uio_td = td };
736 ktruioin = cloneuio(&ktruio);
737 ktriov.iov_base = uap->eventlist;
738 ktriov.iov_len = uap->nevents * sizeof(struct kevent);
739 ktruioout = cloneuio(&ktruio);
740 }
741 #endif
742
743 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
744 &k_ops, tsp);
745
746 #ifdef KTRACE
747 if (ktruioin != NULL) {
748 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
749 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
750 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
751 ktrgenio(uap->fd, UIO_READ, ktruioout, error);
752 }
753 #endif
754
755 return (error);
756 }
757
758 /*
759 * Copy 'count' items into the destination list pointed to by uap->eventlist.
760 */
761 static int
762 kevent_copyout(void *arg, struct kevent *kevp, int count)
763 {
764 struct kevent_args *uap;
765 int error;
766
767 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
768 uap = (struct kevent_args *)arg;
769
770 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
771 if (error == 0)
772 uap->eventlist += count;
773 return (error);
774 }
775
776 /*
777 * Copy 'count' items from the list pointed to by uap->changelist.
778 */
779 static int
780 kevent_copyin(void *arg, struct kevent *kevp, int count)
781 {
782 struct kevent_args *uap;
783 int error;
784
785 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
786 uap = (struct kevent_args *)arg;
787
788 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
789 if (error == 0)
790 uap->changelist += count;
791 return (error);
792 }
793
794 int
795 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
796 struct kevent_copyops *k_ops, const struct timespec *timeout)
797 {
798 struct kevent keva[KQ_NEVENTS];
799 struct kevent *kevp, *changes;
800 struct kqueue *kq;
801 struct file *fp;
802 int i, n, nerrors, error;
803
804 if ((error = fget(td, fd, &fp)) != 0)
805 return (error);
806 if ((error = kqueue_acquire(fp, &kq)) != 0)
807 goto done_norel;
808
809 nerrors = 0;
810
811 while (nchanges > 0) {
812 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
813 error = k_ops->k_copyin(k_ops->arg, keva, n);
814 if (error)
815 goto done;
816 changes = keva;
817 for (i = 0; i < n; i++) {
818 kevp = &changes[i];
819 if (!kevp->filter)
820 continue;
821 kevp->flags &= ~EV_SYSFLAGS;
822 error = kqueue_register(kq, kevp, td, 1);
823 if (error || (kevp->flags & EV_RECEIPT)) {
824 if (nevents != 0) {
825 kevp->flags = EV_ERROR;
826 kevp->data = error;
827 (void) k_ops->k_copyout(k_ops->arg,
828 kevp, 1);
829 nevents--;
830 nerrors++;
831 } else {
832 goto done;
833 }
834 }
835 }
836 nchanges -= n;
837 }
838 if (nerrors) {
839 td->td_retval[0] = nerrors;
840 error = 0;
841 goto done;
842 }
843
844 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
845 done:
846 kqueue_release(kq, 0);
847 done_norel:
848 fdrop(fp, td);
849 return (error);
850 }
851
852 int
853 kqueue_add_filteropts(int filt, struct filterops *filtops)
854 {
855 int error;
856
857 error = 0;
858 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
859 printf(
860 "trying to add a filterop that is out of range: %d is beyond %d\n",
861 ~filt, EVFILT_SYSCOUNT);
862 return EINVAL;
863 }
864 mtx_lock(&filterops_lock);
865 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
866 sysfilt_ops[~filt].for_fop != NULL)
867 error = EEXIST;
868 else {
869 sysfilt_ops[~filt].for_fop = filtops;
870 sysfilt_ops[~filt].for_refcnt = 0;
871 }
872 mtx_unlock(&filterops_lock);
873
874 return (error);
875 }
876
877 int
878 kqueue_del_filteropts(int filt)
879 {
880 int error;
881
882 error = 0;
883 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
884 return EINVAL;
885
886 mtx_lock(&filterops_lock);
887 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
888 sysfilt_ops[~filt].for_fop == NULL)
889 error = EINVAL;
890 else if (sysfilt_ops[~filt].for_refcnt != 0)
891 error = EBUSY;
892 else {
893 sysfilt_ops[~filt].for_fop = &null_filtops;
894 sysfilt_ops[~filt].for_refcnt = 0;
895 }
896 mtx_unlock(&filterops_lock);
897
898 return error;
899 }
900
901 static struct filterops *
902 kqueue_fo_find(int filt)
903 {
904
905 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
906 return NULL;
907
908 mtx_lock(&filterops_lock);
909 sysfilt_ops[~filt].for_refcnt++;
910 if (sysfilt_ops[~filt].for_fop == NULL)
911 sysfilt_ops[~filt].for_fop = &null_filtops;
912 mtx_unlock(&filterops_lock);
913
914 return sysfilt_ops[~filt].for_fop;
915 }
916
917 static void
918 kqueue_fo_release(int filt)
919 {
920
921 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
922 return;
923
924 mtx_lock(&filterops_lock);
925 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
926 ("filter object refcount not valid on release"));
927 sysfilt_ops[~filt].for_refcnt--;
928 mtx_unlock(&filterops_lock);
929 }
930
931 /*
932 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will
933 * influence if memory allocation should wait. Make sure it is 0 if you
934 * hold any mutexes.
935 */
936 static int
937 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
938 {
939 struct filterops *fops;
940 struct file *fp;
941 struct knote *kn, *tkn;
942 int error, filt, event;
943 int haskqglobal;
944
945 fp = NULL;
946 kn = NULL;
947 error = 0;
948 haskqglobal = 0;
949
950 filt = kev->filter;
951 fops = kqueue_fo_find(filt);
952 if (fops == NULL)
953 return EINVAL;
954
955 tkn = knote_alloc(waitok); /* prevent waiting with locks */
956
957 findkn:
958 if (fops->f_isfd) {
959 KASSERT(td != NULL, ("td is NULL"));
960 error = fget(td, kev->ident, &fp);
961 if (error)
962 goto done;
963
964 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
965 kev->ident, 0) != 0) {
966 /* try again */
967 fdrop(fp, td);
968 fp = NULL;
969 error = kqueue_expand(kq, fops, kev->ident, waitok);
970 if (error)
971 goto done;
972 goto findkn;
973 }
974
975 if (fp->f_type == DTYPE_KQUEUE) {
976 /*
977 * if we add some inteligence about what we are doing,
978 * we should be able to support events on ourselves.
979 * We need to know when we are doing this to prevent
980 * getting both the knlist lock and the kq lock since
981 * they are the same thing.
982 */
983 if (fp->f_data == kq) {
984 error = EINVAL;
985 goto done;
986 }
987
988 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
989 }
990
991 KQ_LOCK(kq);
992 if (kev->ident < kq->kq_knlistsize) {
993 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
994 if (kev->filter == kn->kn_filter)
995 break;
996 }
997 } else {
998 if ((kev->flags & EV_ADD) == EV_ADD)
999 kqueue_expand(kq, fops, kev->ident, waitok);
1000
1001 KQ_LOCK(kq);
1002 if (kq->kq_knhashmask != 0) {
1003 struct klist *list;
1004
1005 list = &kq->kq_knhash[
1006 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1007 SLIST_FOREACH(kn, list, kn_link)
1008 if (kev->ident == kn->kn_id &&
1009 kev->filter == kn->kn_filter)
1010 break;
1011 }
1012 }
1013
1014 /* knote is in the process of changing, wait for it to stablize. */
1015 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1016 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1017 kq->kq_state |= KQ_FLUXWAIT;
1018 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1019 if (fp != NULL) {
1020 fdrop(fp, td);
1021 fp = NULL;
1022 }
1023 goto findkn;
1024 }
1025
1026 /*
1027 * kn now contains the matching knote, or NULL if no match
1028 */
1029 if (kn == NULL) {
1030 if (kev->flags & EV_ADD) {
1031 kn = tkn;
1032 tkn = NULL;
1033 if (kn == NULL) {
1034 KQ_UNLOCK(kq);
1035 error = ENOMEM;
1036 goto done;
1037 }
1038 kn->kn_fp = fp;
1039 kn->kn_kq = kq;
1040 kn->kn_fop = fops;
1041 /*
1042 * apply reference counts to knote structure, and
1043 * do not release it at the end of this routine.
1044 */
1045 fops = NULL;
1046 fp = NULL;
1047
1048 kn->kn_sfflags = kev->fflags;
1049 kn->kn_sdata = kev->data;
1050 kev->fflags = 0;
1051 kev->data = 0;
1052 kn->kn_kevent = *kev;
1053 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1054 EV_ENABLE | EV_DISABLE);
1055 kn->kn_status = KN_INFLUX|KN_DETACHED;
1056
1057 error = knote_attach(kn, kq);
1058 KQ_UNLOCK(kq);
1059 if (error != 0) {
1060 tkn = kn;
1061 goto done;
1062 }
1063
1064 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1065 knote_drop(kn, td);
1066 goto done;
1067 }
1068 KN_LIST_LOCK(kn);
1069 goto done_ev_add;
1070 } else {
1071 /* No matching knote and the EV_ADD flag is not set. */
1072 KQ_UNLOCK(kq);
1073 error = ENOENT;
1074 goto done;
1075 }
1076 }
1077
1078 if (kev->flags & EV_DELETE) {
1079 kn->kn_status |= KN_INFLUX;
1080 KQ_UNLOCK(kq);
1081 if (!(kn->kn_status & KN_DETACHED))
1082 kn->kn_fop->f_detach(kn);
1083 knote_drop(kn, td);
1084 goto done;
1085 }
1086
1087 /*
1088 * The user may change some filter values after the initial EV_ADD,
1089 * but doing so will not reset any filter which has already been
1090 * triggered.
1091 */
1092 kn->kn_status |= KN_INFLUX;
1093 KQ_UNLOCK(kq);
1094 KN_LIST_LOCK(kn);
1095 kn->kn_kevent.udata = kev->udata;
1096 if (!fops->f_isfd && fops->f_touch != NULL) {
1097 fops->f_touch(kn, kev, EVENT_REGISTER);
1098 } else {
1099 kn->kn_sfflags = kev->fflags;
1100 kn->kn_sdata = kev->data;
1101 }
1102
1103 /*
1104 * We can get here with kn->kn_knlist == NULL. This can happen when
1105 * the initial attach event decides that the event is "completed"
1106 * already. i.e. filt_procattach is called on a zombie process. It
1107 * will call filt_proc which will remove it from the list, and NULL
1108 * kn_knlist.
1109 */
1110 done_ev_add:
1111 event = kn->kn_fop->f_event(kn, 0);
1112 KQ_LOCK(kq);
1113 if (event)
1114 KNOTE_ACTIVATE(kn, 1);
1115 kn->kn_status &= ~KN_INFLUX;
1116 KN_LIST_UNLOCK(kn);
1117
1118 if ((kev->flags & EV_DISABLE) &&
1119 ((kn->kn_status & KN_DISABLED) == 0)) {
1120 kn->kn_status |= KN_DISABLED;
1121 }
1122
1123 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1124 kn->kn_status &= ~KN_DISABLED;
1125 if ((kn->kn_status & KN_ACTIVE) &&
1126 ((kn->kn_status & KN_QUEUED) == 0))
1127 knote_enqueue(kn);
1128 }
1129 KQ_UNLOCK_FLUX(kq);
1130
1131 done:
1132 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1133 if (fp != NULL)
1134 fdrop(fp, td);
1135 if (tkn != NULL)
1136 knote_free(tkn);
1137 if (fops != NULL)
1138 kqueue_fo_release(filt);
1139 return (error);
1140 }
1141
1142 static int
1143 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1144 {
1145 int error;
1146 struct kqueue *kq;
1147
1148 error = 0;
1149
1150 kq = fp->f_data;
1151 if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1152 return (EBADF);
1153 *kqp = kq;
1154 KQ_LOCK(kq);
1155 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1156 KQ_UNLOCK(kq);
1157 return (EBADF);
1158 }
1159 kq->kq_refcnt++;
1160 KQ_UNLOCK(kq);
1161
1162 return error;
1163 }
1164
1165 static void
1166 kqueue_release(struct kqueue *kq, int locked)
1167 {
1168 if (locked)
1169 KQ_OWNED(kq);
1170 else
1171 KQ_LOCK(kq);
1172 kq->kq_refcnt--;
1173 if (kq->kq_refcnt == 1)
1174 wakeup(&kq->kq_refcnt);
1175 if (!locked)
1176 KQ_UNLOCK(kq);
1177 }
1178
1179 static void
1180 kqueue_schedtask(struct kqueue *kq)
1181 {
1182
1183 KQ_OWNED(kq);
1184 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1185 ("scheduling kqueue task while draining"));
1186
1187 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1188 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1189 kq->kq_state |= KQ_TASKSCHED;
1190 }
1191 }
1192
1193 /*
1194 * Expand the kq to make sure we have storage for fops/ident pair.
1195 *
1196 * Return 0 on success (or no work necessary), return errno on failure.
1197 *
1198 * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1199 * If kqueue_register is called from a non-fd context, there usually/should
1200 * be no locks held.
1201 */
1202 static int
1203 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1204 int waitok)
1205 {
1206 struct klist *list, *tmp_knhash, *to_free;
1207 u_long tmp_knhashmask;
1208 int size;
1209 int fd;
1210 int mflag = waitok ? M_WAITOK : M_NOWAIT;
1211
1212 KQ_NOTOWNED(kq);
1213
1214 to_free = NULL;
1215 if (fops->f_isfd) {
1216 fd = ident;
1217 if (kq->kq_knlistsize <= fd) {
1218 size = kq->kq_knlistsize;
1219 while (size <= fd)
1220 size += KQEXTENT;
1221 list = malloc(size * sizeof list, M_KQUEUE, mflag);
1222 if (list == NULL)
1223 return ENOMEM;
1224 KQ_LOCK(kq);
1225 if (kq->kq_knlistsize > fd) {
1226 to_free = list;
1227 list = NULL;
1228 } else {
1229 if (kq->kq_knlist != NULL) {
1230 bcopy(kq->kq_knlist, list,
1231 kq->kq_knlistsize * sizeof list);
1232 to_free = kq->kq_knlist;
1233 kq->kq_knlist = NULL;
1234 }
1235 bzero((caddr_t)list +
1236 kq->kq_knlistsize * sizeof list,
1237 (size - kq->kq_knlistsize) * sizeof list);
1238 kq->kq_knlistsize = size;
1239 kq->kq_knlist = list;
1240 }
1241 KQ_UNLOCK(kq);
1242 }
1243 } else {
1244 if (kq->kq_knhashmask == 0) {
1245 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1246 &tmp_knhashmask);
1247 if (tmp_knhash == NULL)
1248 return ENOMEM;
1249 KQ_LOCK(kq);
1250 if (kq->kq_knhashmask == 0) {
1251 kq->kq_knhash = tmp_knhash;
1252 kq->kq_knhashmask = tmp_knhashmask;
1253 } else {
1254 to_free = tmp_knhash;
1255 }
1256 KQ_UNLOCK(kq);
1257 }
1258 }
1259 free(to_free, M_KQUEUE);
1260
1261 KQ_NOTOWNED(kq);
1262 return 0;
1263 }
1264
1265 static void
1266 kqueue_task(void *arg, int pending)
1267 {
1268 struct kqueue *kq;
1269 int haskqglobal;
1270
1271 haskqglobal = 0;
1272 kq = arg;
1273
1274 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1275 KQ_LOCK(kq);
1276
1277 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1278
1279 kq->kq_state &= ~KQ_TASKSCHED;
1280 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1281 wakeup(&kq->kq_state);
1282 }
1283 KQ_UNLOCK(kq);
1284 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1285 }
1286
1287 /*
1288 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1289 * We treat KN_MARKER knotes as if they are INFLUX.
1290 */
1291 static int
1292 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1293 const struct timespec *tsp, struct kevent *keva, struct thread *td)
1294 {
1295 struct kevent *kevp;
1296 struct timeval atv, rtv, ttv;
1297 struct knote *kn, *marker;
1298 int count, timeout, nkev, error, influx;
1299 int haskqglobal, touch;
1300
1301 count = maxevents;
1302 nkev = 0;
1303 error = 0;
1304 haskqglobal = 0;
1305
1306 if (maxevents == 0)
1307 goto done_nl;
1308
1309 if (tsp != NULL) {
1310 TIMESPEC_TO_TIMEVAL(&atv, tsp);
1311 if (itimerfix(&atv)) {
1312 error = EINVAL;
1313 goto done_nl;
1314 }
1315 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1316 timeout = -1;
1317 else
1318 timeout = atv.tv_sec > 24 * 60 * 60 ?
1319 24 * 60 * 60 * hz : tvtohz(&atv);
1320 getmicrouptime(&rtv);
1321 timevaladd(&atv, &rtv);
1322 } else {
1323 atv.tv_sec = 0;
1324 atv.tv_usec = 0;
1325 timeout = 0;
1326 }
1327 marker = knote_alloc(1);
1328 if (marker == NULL) {
1329 error = ENOMEM;
1330 goto done_nl;
1331 }
1332 marker->kn_status = KN_MARKER;
1333 KQ_LOCK(kq);
1334 goto start;
1335
1336 retry:
1337 if (atv.tv_sec || atv.tv_usec) {
1338 getmicrouptime(&rtv);
1339 if (timevalcmp(&rtv, &atv, >=))
1340 goto done;
1341 ttv = atv;
1342 timevalsub(&ttv, &rtv);
1343 timeout = ttv.tv_sec > 24 * 60 * 60 ?
1344 24 * 60 * 60 * hz : tvtohz(&ttv);
1345 }
1346
1347 start:
1348 kevp = keva;
1349 if (kq->kq_count == 0) {
1350 if (timeout < 0) {
1351 error = EWOULDBLOCK;
1352 } else {
1353 kq->kq_state |= KQ_SLEEP;
1354 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH,
1355 "kqread", timeout);
1356 }
1357 if (error == 0)
1358 goto retry;
1359 /* don't restart after signals... */
1360 if (error == ERESTART)
1361 error = EINTR;
1362 else if (error == EWOULDBLOCK)
1363 error = 0;
1364 goto done;
1365 }
1366
1367 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1368 influx = 0;
1369 while (count) {
1370 KQ_OWNED(kq);
1371 kn = TAILQ_FIRST(&kq->kq_head);
1372
1373 if ((kn->kn_status == KN_MARKER && kn != marker) ||
1374 (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1375 if (influx) {
1376 influx = 0;
1377 KQ_FLUX_WAKEUP(kq);
1378 }
1379 kq->kq_state |= KQ_FLUXWAIT;
1380 error = msleep(kq, &kq->kq_lock, PSOCK,
1381 "kqflxwt", 0);
1382 continue;
1383 }
1384
1385 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1386 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1387 kn->kn_status &= ~KN_QUEUED;
1388 kq->kq_count--;
1389 continue;
1390 }
1391 if (kn == marker) {
1392 KQ_FLUX_WAKEUP(kq);
1393 if (count == maxevents)
1394 goto retry;
1395 goto done;
1396 }
1397 KASSERT((kn->kn_status & KN_INFLUX) == 0,
1398 ("KN_INFLUX set when not suppose to be"));
1399
1400 if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1401 kn->kn_status &= ~KN_QUEUED;
1402 kn->kn_status |= KN_INFLUX;
1403 kq->kq_count--;
1404 KQ_UNLOCK(kq);
1405 /*
1406 * We don't need to lock the list since we've marked
1407 * it _INFLUX.
1408 */
1409 *kevp = kn->kn_kevent;
1410 if (!(kn->kn_status & KN_DETACHED))
1411 kn->kn_fop->f_detach(kn);
1412 knote_drop(kn, td);
1413 KQ_LOCK(kq);
1414 kn = NULL;
1415 } else {
1416 kn->kn_status |= KN_INFLUX;
1417 KQ_UNLOCK(kq);
1418 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1419 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1420 KN_LIST_LOCK(kn);
1421 if (kn->kn_fop->f_event(kn, 0) == 0) {
1422 KQ_LOCK(kq);
1423 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1424 kn->kn_status &=
1425 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX);
1426 kq->kq_count--;
1427 KN_LIST_UNLOCK(kn);
1428 influx = 1;
1429 continue;
1430 }
1431 touch = (!kn->kn_fop->f_isfd &&
1432 kn->kn_fop->f_touch != NULL);
1433 if (touch)
1434 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1435 else
1436 *kevp = kn->kn_kevent;
1437 KQ_LOCK(kq);
1438 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1439 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1440 /*
1441 * Manually clear knotes who weren't
1442 * 'touch'ed.
1443 */
1444 if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1445 kn->kn_data = 0;
1446 kn->kn_fflags = 0;
1447 }
1448 if (kn->kn_flags & EV_DISPATCH)
1449 kn->kn_status |= KN_DISABLED;
1450 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1451 kq->kq_count--;
1452 } else
1453 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1454
1455 kn->kn_status &= ~(KN_INFLUX);
1456 KN_LIST_UNLOCK(kn);
1457 influx = 1;
1458 }
1459
1460 /* we are returning a copy to the user */
1461 kevp++;
1462 nkev++;
1463 count--;
1464
1465 if (nkev == KQ_NEVENTS) {
1466 influx = 0;
1467 KQ_UNLOCK_FLUX(kq);
1468 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1469 nkev = 0;
1470 kevp = keva;
1471 KQ_LOCK(kq);
1472 if (error)
1473 break;
1474 }
1475 }
1476 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1477 done:
1478 KQ_OWNED(kq);
1479 KQ_UNLOCK_FLUX(kq);
1480 knote_free(marker);
1481 done_nl:
1482 KQ_NOTOWNED(kq);
1483 if (nkev != 0)
1484 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1485 td->td_retval[0] = maxevents - count;
1486 return (error);
1487 }
1488
1489 /*
1490 * XXX
1491 * This could be expanded to call kqueue_scan, if desired.
1492 */
1493 /*ARGSUSED*/
1494 static int
1495 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1496 int flags, struct thread *td)
1497 {
1498 return (ENXIO);
1499 }
1500
1501 /*ARGSUSED*/
1502 static int
1503 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1504 int flags, struct thread *td)
1505 {
1506 return (ENXIO);
1507 }
1508
1509 /*ARGSUSED*/
1510 static int
1511 kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1512 struct thread *td)
1513 {
1514
1515 return (EINVAL);
1516 }
1517
1518 /*ARGSUSED*/
1519 static int
1520 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1521 struct ucred *active_cred, struct thread *td)
1522 {
1523 /*
1524 * Enabling sigio causes two major problems:
1525 * 1) infinite recursion:
1526 * Synopsys: kevent is being used to track signals and have FIOASYNC
1527 * set. On receipt of a signal this will cause a kqueue to recurse
1528 * into itself over and over. Sending the sigio causes the kqueue
1529 * to become ready, which in turn posts sigio again, forever.
1530 * Solution: this can be solved by setting a flag in the kqueue that
1531 * we have a SIGIO in progress.
1532 * 2) locking problems:
1533 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1534 * us above the proc and pgrp locks.
1535 * Solution: Post a signal using an async mechanism, being sure to
1536 * record a generation count in the delivery so that we do not deliver
1537 * a signal to the wrong process.
1538 *
1539 * Note, these two mechanisms are somewhat mutually exclusive!
1540 */
1541 #if 0
1542 struct kqueue *kq;
1543
1544 kq = fp->f_data;
1545 switch (cmd) {
1546 case FIOASYNC:
1547 if (*(int *)data) {
1548 kq->kq_state |= KQ_ASYNC;
1549 } else {
1550 kq->kq_state &= ~KQ_ASYNC;
1551 }
1552 return (0);
1553
1554 case FIOSETOWN:
1555 return (fsetown(*(int *)data, &kq->kq_sigio));
1556
1557 case FIOGETOWN:
1558 *(int *)data = fgetown(&kq->kq_sigio);
1559 return (0);
1560 }
1561 #endif
1562
1563 return (ENOTTY);
1564 }
1565
1566 /*ARGSUSED*/
1567 static int
1568 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1569 struct thread *td)
1570 {
1571 struct kqueue *kq;
1572 int revents = 0;
1573 int error;
1574
1575 if ((error = kqueue_acquire(fp, &kq)))
1576 return POLLERR;
1577
1578 KQ_LOCK(kq);
1579 if (events & (POLLIN | POLLRDNORM)) {
1580 if (kq->kq_count) {
1581 revents |= events & (POLLIN | POLLRDNORM);
1582 } else {
1583 selrecord(td, &kq->kq_sel);
1584 if (SEL_WAITING(&kq->kq_sel))
1585 kq->kq_state |= KQ_SEL;
1586 }
1587 }
1588 kqueue_release(kq, 1);
1589 KQ_UNLOCK(kq);
1590 return (revents);
1591 }
1592
1593 /*ARGSUSED*/
1594 static int
1595 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1596 struct thread *td)
1597 {
1598
1599 bzero((void *)st, sizeof *st);
1600 /*
1601 * We no longer return kq_count because the unlocked value is useless.
1602 * If you spent all this time getting the count, why not spend your
1603 * syscall better by calling kevent?
1604 *
1605 * XXX - This is needed for libc_r.
1606 */
1607 st->st_mode = S_IFIFO;
1608 return (0);
1609 }
1610
1611 /*ARGSUSED*/
1612 static int
1613 kqueue_close(struct file *fp, struct thread *td)
1614 {
1615 struct kqueue *kq = fp->f_data;
1616 struct filedesc *fdp;
1617 struct knote *kn;
1618 int i;
1619 int error;
1620
1621 if ((error = kqueue_acquire(fp, &kq)))
1622 return error;
1623
1624 KQ_LOCK(kq);
1625
1626 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1627 ("kqueue already closing"));
1628 kq->kq_state |= KQ_CLOSING;
1629 if (kq->kq_refcnt > 1)
1630 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1631
1632 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1633 fdp = kq->kq_fdp;
1634
1635 KASSERT(knlist_empty(&kq->kq_sel.si_note),
1636 ("kqueue's knlist not empty"));
1637
1638 for (i = 0; i < kq->kq_knlistsize; i++) {
1639 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1640 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1641 kq->kq_state |= KQ_FLUXWAIT;
1642 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1643 continue;
1644 }
1645 kn->kn_status |= KN_INFLUX;
1646 KQ_UNLOCK(kq);
1647 if (!(kn->kn_status & KN_DETACHED))
1648 kn->kn_fop->f_detach(kn);
1649 knote_drop(kn, td);
1650 KQ_LOCK(kq);
1651 }
1652 }
1653 if (kq->kq_knhashmask != 0) {
1654 for (i = 0; i <= kq->kq_knhashmask; i++) {
1655 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1656 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1657 kq->kq_state |= KQ_FLUXWAIT;
1658 msleep(kq, &kq->kq_lock, PSOCK,
1659 "kqclo2", 0);
1660 continue;
1661 }
1662 kn->kn_status |= KN_INFLUX;
1663 KQ_UNLOCK(kq);
1664 if (!(kn->kn_status & KN_DETACHED))
1665 kn->kn_fop->f_detach(kn);
1666 knote_drop(kn, td);
1667 KQ_LOCK(kq);
1668 }
1669 }
1670 }
1671
1672 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1673 kq->kq_state |= KQ_TASKDRAIN;
1674 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1675 }
1676
1677 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1678 selwakeuppri(&kq->kq_sel, PSOCK);
1679 if (!SEL_WAITING(&kq->kq_sel))
1680 kq->kq_state &= ~KQ_SEL;
1681 }
1682
1683 KQ_UNLOCK(kq);
1684
1685 FILEDESC_XLOCK(fdp);
1686 SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list);
1687 FILEDESC_XUNLOCK(fdp);
1688
1689 seldrain(&kq->kq_sel);
1690 knlist_destroy(&kq->kq_sel.si_note);
1691 mtx_destroy(&kq->kq_lock);
1692 kq->kq_fdp = NULL;
1693
1694 if (kq->kq_knhash != NULL)
1695 free(kq->kq_knhash, M_KQUEUE);
1696 if (kq->kq_knlist != NULL)
1697 free(kq->kq_knlist, M_KQUEUE);
1698
1699 funsetown(&kq->kq_sigio);
1700 free(kq, M_KQUEUE);
1701 fp->f_data = NULL;
1702
1703 return (0);
1704 }
1705
1706 static void
1707 kqueue_wakeup(struct kqueue *kq)
1708 {
1709 KQ_OWNED(kq);
1710
1711 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1712 kq->kq_state &= ~KQ_SLEEP;
1713 wakeup(kq);
1714 }
1715 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1716 selwakeuppri(&kq->kq_sel, PSOCK);
1717 if (!SEL_WAITING(&kq->kq_sel))
1718 kq->kq_state &= ~KQ_SEL;
1719 }
1720 if (!knlist_empty(&kq->kq_sel.si_note))
1721 kqueue_schedtask(kq);
1722 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1723 pgsigio(&kq->kq_sigio, SIGIO, 0);
1724 }
1725 }
1726
1727 /*
1728 * Walk down a list of knotes, activating them if their event has triggered.
1729 *
1730 * There is a possibility to optimize in the case of one kq watching another.
1731 * Instead of scheduling a task to wake it up, you could pass enough state
1732 * down the chain to make up the parent kqueue. Make this code functional
1733 * first.
1734 */
1735 void
1736 knote(struct knlist *list, long hint, int lockflags)
1737 {
1738 struct kqueue *kq;
1739 struct knote *kn;
1740 int error;
1741
1742 if (list == NULL)
1743 return;
1744
1745 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1746
1747 if ((lockflags & KNF_LISTLOCKED) == 0)
1748 list->kl_lock(list->kl_lockarg);
1749
1750 /*
1751 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1752 * the kqueue scheduling, but this will introduce four
1753 * lock/unlock's for each knote to test. If we do, continue to use
1754 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1755 * only safe if you want to remove the current item, which we are
1756 * not doing.
1757 */
1758 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1759 kq = kn->kn_kq;
1760 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1761 KQ_LOCK(kq);
1762 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1763 KQ_UNLOCK(kq);
1764 } else if ((lockflags & KNF_NOKQLOCK) != 0) {
1765 kn->kn_status |= KN_INFLUX;
1766 KQ_UNLOCK(kq);
1767 error = kn->kn_fop->f_event(kn, hint);
1768 KQ_LOCK(kq);
1769 kn->kn_status &= ~KN_INFLUX;
1770 if (error)
1771 KNOTE_ACTIVATE(kn, 1);
1772 KQ_UNLOCK_FLUX(kq);
1773 } else {
1774 kn->kn_status |= KN_HASKQLOCK;
1775 if (kn->kn_fop->f_event(kn, hint))
1776 KNOTE_ACTIVATE(kn, 1);
1777 kn->kn_status &= ~KN_HASKQLOCK;
1778 KQ_UNLOCK(kq);
1779 }
1780 }
1781 kq = NULL;
1782 }
1783 if ((lockflags & KNF_LISTLOCKED) == 0)
1784 list->kl_unlock(list->kl_lockarg);
1785 }
1786
1787 /*
1788 * add a knote to a knlist
1789 */
1790 void
1791 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1792 {
1793 KNL_ASSERT_LOCK(knl, islocked);
1794 KQ_NOTOWNED(kn->kn_kq);
1795 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1796 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1797 if (!islocked)
1798 knl->kl_lock(knl->kl_lockarg);
1799 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1800 if (!islocked)
1801 knl->kl_unlock(knl->kl_lockarg);
1802 KQ_LOCK(kn->kn_kq);
1803 kn->kn_knlist = knl;
1804 kn->kn_status &= ~KN_DETACHED;
1805 KQ_UNLOCK(kn->kn_kq);
1806 }
1807
1808 static void
1809 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1810 {
1811 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1812 KNL_ASSERT_LOCK(knl, knlislocked);
1813 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1814 if (!kqislocked)
1815 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1816 ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1817 if (!knlislocked)
1818 knl->kl_lock(knl->kl_lockarg);
1819 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1820 kn->kn_knlist = NULL;
1821 if (!knlislocked)
1822 knl->kl_unlock(knl->kl_lockarg);
1823 if (!kqislocked)
1824 KQ_LOCK(kn->kn_kq);
1825 kn->kn_status |= KN_DETACHED;
1826 if (!kqislocked)
1827 KQ_UNLOCK(kn->kn_kq);
1828 }
1829
1830 /*
1831 * remove all knotes from a specified klist
1832 */
1833 void
1834 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1835 {
1836
1837 knlist_remove_kq(knl, kn, islocked, 0);
1838 }
1839
1840 /*
1841 * remove knote from a specified klist while in f_event handler.
1842 */
1843 void
1844 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1845 {
1846
1847 knlist_remove_kq(knl, kn, 1,
1848 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1849 }
1850
1851 int
1852 knlist_empty(struct knlist *knl)
1853 {
1854 KNL_ASSERT_LOCKED(knl);
1855 return SLIST_EMPTY(&knl->kl_list);
1856 }
1857
1858 static struct mtx knlist_lock;
1859 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1860 MTX_DEF);
1861 static void knlist_mtx_lock(void *arg);
1862 static void knlist_mtx_unlock(void *arg);
1863
1864 static void
1865 knlist_mtx_lock(void *arg)
1866 {
1867 mtx_lock((struct mtx *)arg);
1868 }
1869
1870 static void
1871 knlist_mtx_unlock(void *arg)
1872 {
1873 mtx_unlock((struct mtx *)arg);
1874 }
1875
1876 static void
1877 knlist_mtx_assert_locked(void *arg)
1878 {
1879 mtx_assert((struct mtx *)arg, MA_OWNED);
1880 }
1881
1882 static void
1883 knlist_mtx_assert_unlocked(void *arg)
1884 {
1885 mtx_assert((struct mtx *)arg, MA_NOTOWNED);
1886 }
1887
1888 void
1889 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1890 void (*kl_unlock)(void *),
1891 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
1892 {
1893
1894 if (lock == NULL)
1895 knl->kl_lockarg = &knlist_lock;
1896 else
1897 knl->kl_lockarg = lock;
1898
1899 if (kl_lock == NULL)
1900 knl->kl_lock = knlist_mtx_lock;
1901 else
1902 knl->kl_lock = kl_lock;
1903 if (kl_unlock == NULL)
1904 knl->kl_unlock = knlist_mtx_unlock;
1905 else
1906 knl->kl_unlock = kl_unlock;
1907 if (kl_assert_locked == NULL)
1908 knl->kl_assert_locked = knlist_mtx_assert_locked;
1909 else
1910 knl->kl_assert_locked = kl_assert_locked;
1911 if (kl_assert_unlocked == NULL)
1912 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
1913 else
1914 knl->kl_assert_unlocked = kl_assert_unlocked;
1915
1916 SLIST_INIT(&knl->kl_list);
1917 }
1918
1919 void
1920 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
1921 {
1922
1923 knlist_init(knl, lock, NULL, NULL, NULL, NULL);
1924 }
1925
1926 void
1927 knlist_destroy(struct knlist *knl)
1928 {
1929
1930 #ifdef INVARIANTS
1931 /*
1932 * if we run across this error, we need to find the offending
1933 * driver and have it call knlist_clear.
1934 */
1935 if (!SLIST_EMPTY(&knl->kl_list))
1936 printf("WARNING: destroying knlist w/ knotes on it!\n");
1937 #endif
1938
1939 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
1940 SLIST_INIT(&knl->kl_list);
1941 }
1942
1943 /*
1944 * Even if we are locked, we may need to drop the lock to allow any influx
1945 * knotes time to "settle".
1946 */
1947 void
1948 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
1949 {
1950 struct knote *kn, *kn2;
1951 struct kqueue *kq;
1952
1953 if (islocked)
1954 KNL_ASSERT_LOCKED(knl);
1955 else {
1956 KNL_ASSERT_UNLOCKED(knl);
1957 again: /* need to reacquire lock since we have dropped it */
1958 knl->kl_lock(knl->kl_lockarg);
1959 }
1960
1961 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
1962 kq = kn->kn_kq;
1963 KQ_LOCK(kq);
1964 if ((kn->kn_status & KN_INFLUX)) {
1965 KQ_UNLOCK(kq);
1966 continue;
1967 }
1968 knlist_remove_kq(knl, kn, 1, 1);
1969 if (killkn) {
1970 kn->kn_status |= KN_INFLUX | KN_DETACHED;
1971 KQ_UNLOCK(kq);
1972 knote_drop(kn, td);
1973 } else {
1974 /* Make sure cleared knotes disappear soon */
1975 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1976 KQ_UNLOCK(kq);
1977 }
1978 kq = NULL;
1979 }
1980
1981 if (!SLIST_EMPTY(&knl->kl_list)) {
1982 /* there are still KN_INFLUX remaining */
1983 kn = SLIST_FIRST(&knl->kl_list);
1984 kq = kn->kn_kq;
1985 KQ_LOCK(kq);
1986 KASSERT(kn->kn_status & KN_INFLUX,
1987 ("knote removed w/o list lock"));
1988 knl->kl_unlock(knl->kl_lockarg);
1989 kq->kq_state |= KQ_FLUXWAIT;
1990 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
1991 kq = NULL;
1992 goto again;
1993 }
1994
1995 if (islocked)
1996 KNL_ASSERT_LOCKED(knl);
1997 else {
1998 knl->kl_unlock(knl->kl_lockarg);
1999 KNL_ASSERT_UNLOCKED(knl);
2000 }
2001 }
2002
2003 /*
2004 * Remove all knotes referencing a specified fd must be called with FILEDESC
2005 * lock. This prevents a race where a new fd comes along and occupies the
2006 * entry and we attach a knote to the fd.
2007 */
2008 void
2009 knote_fdclose(struct thread *td, int fd)
2010 {
2011 struct filedesc *fdp = td->td_proc->p_fd;
2012 struct kqueue *kq;
2013 struct knote *kn;
2014 int influx;
2015
2016 FILEDESC_XLOCK_ASSERT(fdp);
2017
2018 /*
2019 * We shouldn't have to worry about new kevents appearing on fd
2020 * since filedesc is locked.
2021 */
2022 SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2023 KQ_LOCK(kq);
2024
2025 again:
2026 influx = 0;
2027 while (kq->kq_knlistsize > fd &&
2028 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2029 if (kn->kn_status & KN_INFLUX) {
2030 /* someone else might be waiting on our knote */
2031 if (influx)
2032 wakeup(kq);
2033 kq->kq_state |= KQ_FLUXWAIT;
2034 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2035 goto again;
2036 }
2037 kn->kn_status |= KN_INFLUX;
2038 KQ_UNLOCK(kq);
2039 if (!(kn->kn_status & KN_DETACHED))
2040 kn->kn_fop->f_detach(kn);
2041 knote_drop(kn, td);
2042 influx = 1;
2043 KQ_LOCK(kq);
2044 }
2045 KQ_UNLOCK_FLUX(kq);
2046 }
2047 }
2048
2049 static int
2050 knote_attach(struct knote *kn, struct kqueue *kq)
2051 {
2052 struct klist *list;
2053
2054 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2055 KQ_OWNED(kq);
2056
2057 if (kn->kn_fop->f_isfd) {
2058 if (kn->kn_id >= kq->kq_knlistsize)
2059 return ENOMEM;
2060 list = &kq->kq_knlist[kn->kn_id];
2061 } else {
2062 if (kq->kq_knhash == NULL)
2063 return ENOMEM;
2064 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2065 }
2066
2067 SLIST_INSERT_HEAD(list, kn, kn_link);
2068
2069 return 0;
2070 }
2071
2072 /*
2073 * knote must already have been detached using the f_detach method.
2074 * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2075 * to prevent other removal.
2076 */
2077 static void
2078 knote_drop(struct knote *kn, struct thread *td)
2079 {
2080 struct kqueue *kq;
2081 struct klist *list;
2082
2083 kq = kn->kn_kq;
2084
2085 KQ_NOTOWNED(kq);
2086 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2087 ("knote_drop called without KN_INFLUX set in kn_status"));
2088
2089 KQ_LOCK(kq);
2090 if (kn->kn_fop->f_isfd)
2091 list = &kq->kq_knlist[kn->kn_id];
2092 else
2093 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2094
2095 if (!SLIST_EMPTY(list))
2096 SLIST_REMOVE(list, kn, knote, kn_link);
2097 if (kn->kn_status & KN_QUEUED)
2098 knote_dequeue(kn);
2099 KQ_UNLOCK_FLUX(kq);
2100
2101 if (kn->kn_fop->f_isfd) {
2102 fdrop(kn->kn_fp, td);
2103 kn->kn_fp = NULL;
2104 }
2105 kqueue_fo_release(kn->kn_kevent.filter);
2106 kn->kn_fop = NULL;
2107 knote_free(kn);
2108 }
2109
2110 static void
2111 knote_enqueue(struct knote *kn)
2112 {
2113 struct kqueue *kq = kn->kn_kq;
2114
2115 KQ_OWNED(kn->kn_kq);
2116 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2117
2118 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2119 kn->kn_status |= KN_QUEUED;
2120 kq->kq_count++;
2121 kqueue_wakeup(kq);
2122 }
2123
2124 static void
2125 knote_dequeue(struct knote *kn)
2126 {
2127 struct kqueue *kq = kn->kn_kq;
2128
2129 KQ_OWNED(kn->kn_kq);
2130 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2131
2132 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2133 kn->kn_status &= ~KN_QUEUED;
2134 kq->kq_count--;
2135 }
2136
2137 static void
2138 knote_init(void)
2139 {
2140
2141 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2142 NULL, NULL, UMA_ALIGN_PTR, 0);
2143 }
2144 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2145
2146 static struct knote *
2147 knote_alloc(int waitok)
2148 {
2149 return ((struct knote *)uma_zalloc(knote_zone,
2150 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2151 }
2152
2153 static void
2154 knote_free(struct knote *kn)
2155 {
2156 if (kn != NULL)
2157 uma_zfree(knote_zone, kn);
2158 }
2159
2160 /*
2161 * Register the kev w/ the kq specified by fd.
2162 */
2163 int
2164 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2165 {
2166 struct kqueue *kq;
2167 struct file *fp;
2168 int error;
2169
2170 if ((error = fget(td, fd, &fp)) != 0)
2171 return (error);
2172 if ((error = kqueue_acquire(fp, &kq)) != 0)
2173 goto noacquire;
2174
2175 error = kqueue_register(kq, kev, td, waitok);
2176
2177 kqueue_release(kq, 0);
2178
2179 noacquire:
2180 fdrop(fp, td);
2181
2182 return error;
2183 }
Cache object: d96d94dfa50cb9c0e80f4ae0b3dc363f
|