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/10.3/sys/kern/kern_event.c 293545 2016-01-09 16:39:15Z dchagin $");
31
32 #include "opt_ktrace.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/capsicum.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/rwlock.h>
41 #include <sys/proc.h>
42 #include <sys/malloc.h>
43 #include <sys/unistd.h>
44 #include <sys/file.h>
45 #include <sys/filedesc.h>
46 #include <sys/filio.h>
47 #include <sys/fcntl.h>
48 #include <sys/kthread.h>
49 #include <sys/selinfo.h>
50 #include <sys/stdatomic.h>
51 #include <sys/queue.h>
52 #include <sys/event.h>
53 #include <sys/eventvar.h>
54 #include <sys/poll.h>
55 #include <sys/protosw.h>
56 #include <sys/sigio.h>
57 #include <sys/signalvar.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysproto.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/taskqueue.h>
65 #include <sys/uio.h>
66 #ifdef KTRACE
67 #include <sys/ktrace.h>
68 #endif
69
70 #include <vm/uma.h>
71
72 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
73
74 /*
75 * This lock is used if multiple kq locks are required. This possibly
76 * should be made into a per proc lock.
77 */
78 static struct mtx kq_global;
79 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
80 #define KQ_GLOBAL_LOCK(lck, haslck) do { \
81 if (!haslck) \
82 mtx_lock(lck); \
83 haslck = 1; \
84 } while (0)
85 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \
86 if (haslck) \
87 mtx_unlock(lck); \
88 haslck = 0; \
89 } while (0)
90
91 TASKQUEUE_DEFINE_THREAD(kqueue);
92
93 static int kevent_copyout(void *arg, struct kevent *kevp, int count);
94 static int kevent_copyin(void *arg, struct kevent *kevp, int count);
95 static int kqueue_register(struct kqueue *kq, struct kevent *kev,
96 struct thread *td, int waitok);
97 static int kqueue_acquire(struct file *fp, struct kqueue **kqp);
98 static void kqueue_release(struct kqueue *kq, int locked);
99 static int kqueue_expand(struct kqueue *kq, struct filterops *fops,
100 uintptr_t ident, int waitok);
101 static void kqueue_task(void *arg, int pending);
102 static int kqueue_scan(struct kqueue *kq, int maxevents,
103 struct kevent_copyops *k_ops,
104 const struct timespec *timeout,
105 struct kevent *keva, struct thread *td);
106 static void kqueue_wakeup(struct kqueue *kq);
107 static struct filterops *kqueue_fo_find(int filt);
108 static void kqueue_fo_release(int filt);
109
110 static fo_rdwr_t kqueue_read;
111 static fo_rdwr_t kqueue_write;
112 static fo_truncate_t kqueue_truncate;
113 static fo_ioctl_t kqueue_ioctl;
114 static fo_poll_t kqueue_poll;
115 static fo_kqfilter_t kqueue_kqfilter;
116 static fo_stat_t kqueue_stat;
117 static fo_close_t kqueue_close;
118
119 static struct fileops kqueueops = {
120 .fo_read = kqueue_read,
121 .fo_write = kqueue_write,
122 .fo_truncate = kqueue_truncate,
123 .fo_ioctl = kqueue_ioctl,
124 .fo_poll = kqueue_poll,
125 .fo_kqfilter = kqueue_kqfilter,
126 .fo_stat = kqueue_stat,
127 .fo_close = kqueue_close,
128 .fo_chmod = invfo_chmod,
129 .fo_chown = invfo_chown,
130 .fo_sendfile = invfo_sendfile,
131 };
132
133 static int knote_attach(struct knote *kn, struct kqueue *kq);
134 static void knote_drop(struct knote *kn, struct thread *td);
135 static void knote_enqueue(struct knote *kn);
136 static void knote_dequeue(struct knote *kn);
137 static void knote_init(void);
138 static struct knote *knote_alloc(int waitok);
139 static void knote_free(struct knote *kn);
140
141 static void filt_kqdetach(struct knote *kn);
142 static int filt_kqueue(struct knote *kn, long hint);
143 static int filt_procattach(struct knote *kn);
144 static void filt_procdetach(struct knote *kn);
145 static int filt_proc(struct knote *kn, long hint);
146 static int filt_fileattach(struct knote *kn);
147 static void filt_timerexpire(void *knx);
148 static int filt_timerattach(struct knote *kn);
149 static void filt_timerdetach(struct knote *kn);
150 static int filt_timer(struct knote *kn, long hint);
151 static int filt_userattach(struct knote *kn);
152 static void filt_userdetach(struct knote *kn);
153 static int filt_user(struct knote *kn, long hint);
154 static void filt_usertouch(struct knote *kn, struct kevent *kev,
155 u_long type);
156
157 static struct filterops file_filtops = {
158 .f_isfd = 1,
159 .f_attach = filt_fileattach,
160 };
161 static struct filterops kqread_filtops = {
162 .f_isfd = 1,
163 .f_detach = filt_kqdetach,
164 .f_event = filt_kqueue,
165 };
166 /* XXX - move to kern_proc.c? */
167 static struct filterops proc_filtops = {
168 .f_isfd = 0,
169 .f_attach = filt_procattach,
170 .f_detach = filt_procdetach,
171 .f_event = filt_proc,
172 };
173 static struct filterops timer_filtops = {
174 .f_isfd = 0,
175 .f_attach = filt_timerattach,
176 .f_detach = filt_timerdetach,
177 .f_event = filt_timer,
178 };
179 static struct filterops user_filtops = {
180 .f_attach = filt_userattach,
181 .f_detach = filt_userdetach,
182 .f_event = filt_user,
183 .f_touch = filt_usertouch,
184 };
185
186 static uma_zone_t knote_zone;
187 static atomic_uint kq_ncallouts = ATOMIC_VAR_INIT(0);
188 static unsigned int kq_calloutmax = 4 * 1024;
189 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
190 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
191
192 /* XXX - ensure not KN_INFLUX?? */
193 #define KNOTE_ACTIVATE(kn, islock) do { \
194 if ((islock)) \
195 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \
196 else \
197 KQ_LOCK((kn)->kn_kq); \
198 (kn)->kn_status |= KN_ACTIVE; \
199 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
200 knote_enqueue((kn)); \
201 if (!(islock)) \
202 KQ_UNLOCK((kn)->kn_kq); \
203 } while(0)
204 #define KQ_LOCK(kq) do { \
205 mtx_lock(&(kq)->kq_lock); \
206 } while (0)
207 #define KQ_FLUX_WAKEUP(kq) do { \
208 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
209 (kq)->kq_state &= ~KQ_FLUXWAIT; \
210 wakeup((kq)); \
211 } \
212 } while (0)
213 #define KQ_UNLOCK_FLUX(kq) do { \
214 KQ_FLUX_WAKEUP(kq); \
215 mtx_unlock(&(kq)->kq_lock); \
216 } while (0)
217 #define KQ_UNLOCK(kq) do { \
218 mtx_unlock(&(kq)->kq_lock); \
219 } while (0)
220 #define KQ_OWNED(kq) do { \
221 mtx_assert(&(kq)->kq_lock, MA_OWNED); \
222 } while (0)
223 #define KQ_NOTOWNED(kq) do { \
224 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \
225 } while (0)
226 #define KN_LIST_LOCK(kn) do { \
227 if (kn->kn_knlist != NULL) \
228 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \
229 } while (0)
230 #define KN_LIST_UNLOCK(kn) do { \
231 if (kn->kn_knlist != NULL) \
232 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \
233 } while (0)
234 #define KNL_ASSERT_LOCK(knl, islocked) do { \
235 if (islocked) \
236 KNL_ASSERT_LOCKED(knl); \
237 else \
238 KNL_ASSERT_UNLOCKED(knl); \
239 } while (0)
240 #ifdef INVARIANTS
241 #define KNL_ASSERT_LOCKED(knl) do { \
242 knl->kl_assert_locked((knl)->kl_lockarg); \
243 } while (0)
244 #define KNL_ASSERT_UNLOCKED(knl) do { \
245 knl->kl_assert_unlocked((knl)->kl_lockarg); \
246 } while (0)
247 #else /* !INVARIANTS */
248 #define KNL_ASSERT_LOCKED(knl) do {} while(0)
249 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0)
250 #endif /* INVARIANTS */
251
252 #define KN_HASHSIZE 64 /* XXX should be tunable */
253 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
254
255 static int
256 filt_nullattach(struct knote *kn)
257 {
258
259 return (ENXIO);
260 };
261
262 struct filterops null_filtops = {
263 .f_isfd = 0,
264 .f_attach = filt_nullattach,
265 };
266
267 /* XXX - make SYSINIT to add these, and move into respective modules. */
268 extern struct filterops sig_filtops;
269 extern struct filterops fs_filtops;
270
271 /*
272 * Table for for all system-defined filters.
273 */
274 static struct mtx filterops_lock;
275 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
276 MTX_DEF);
277 static struct {
278 struct filterops *for_fop;
279 int for_refcnt;
280 } sysfilt_ops[EVFILT_SYSCOUNT] = {
281 { &file_filtops }, /* EVFILT_READ */
282 { &file_filtops }, /* EVFILT_WRITE */
283 { &null_filtops }, /* EVFILT_AIO */
284 { &file_filtops }, /* EVFILT_VNODE */
285 { &proc_filtops }, /* EVFILT_PROC */
286 { &sig_filtops }, /* EVFILT_SIGNAL */
287 { &timer_filtops }, /* EVFILT_TIMER */
288 { &null_filtops }, /* former EVFILT_NETDEV */
289 { &fs_filtops }, /* EVFILT_FS */
290 { &null_filtops }, /* EVFILT_LIO */
291 { &user_filtops }, /* EVFILT_USER */
292 };
293
294 /*
295 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
296 * method.
297 */
298 static int
299 filt_fileattach(struct knote *kn)
300 {
301
302 return (fo_kqfilter(kn->kn_fp, kn));
303 }
304
305 /*ARGSUSED*/
306 static int
307 kqueue_kqfilter(struct file *fp, struct knote *kn)
308 {
309 struct kqueue *kq = kn->kn_fp->f_data;
310
311 if (kn->kn_filter != EVFILT_READ)
312 return (EINVAL);
313
314 kn->kn_status |= KN_KQUEUE;
315 kn->kn_fop = &kqread_filtops;
316 knlist_add(&kq->kq_sel.si_note, kn, 0);
317
318 return (0);
319 }
320
321 static void
322 filt_kqdetach(struct knote *kn)
323 {
324 struct kqueue *kq = kn->kn_fp->f_data;
325
326 knlist_remove(&kq->kq_sel.si_note, kn, 0);
327 }
328
329 /*ARGSUSED*/
330 static int
331 filt_kqueue(struct knote *kn, long hint)
332 {
333 struct kqueue *kq = kn->kn_fp->f_data;
334
335 kn->kn_data = kq->kq_count;
336 return (kn->kn_data > 0);
337 }
338
339 /* XXX - move to kern_proc.c? */
340 static int
341 filt_procattach(struct knote *kn)
342 {
343 struct proc *p;
344 int immediate;
345 int error;
346
347 immediate = 0;
348 p = pfind(kn->kn_id);
349 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
350 p = zpfind(kn->kn_id);
351 immediate = 1;
352 } else if (p != NULL && (p->p_flag & P_WEXIT)) {
353 immediate = 1;
354 }
355
356 if (p == NULL)
357 return (ESRCH);
358 if ((error = p_cansee(curthread, p))) {
359 PROC_UNLOCK(p);
360 return (error);
361 }
362
363 kn->kn_ptr.p_proc = p;
364 kn->kn_flags |= EV_CLEAR; /* automatically set */
365
366 /*
367 * internal flag indicating registration done by kernel
368 */
369 if (kn->kn_flags & EV_FLAG1) {
370 kn->kn_data = kn->kn_sdata; /* ppid */
371 kn->kn_fflags = NOTE_CHILD;
372 kn->kn_flags &= ~EV_FLAG1;
373 }
374
375 if (immediate == 0)
376 knlist_add(&p->p_klist, kn, 1);
377
378 /*
379 * Immediately activate any exit notes if the target process is a
380 * zombie. This is necessary to handle the case where the target
381 * process, e.g. a child, dies before the kevent is registered.
382 */
383 if (immediate && filt_proc(kn, NOTE_EXIT))
384 KNOTE_ACTIVATE(kn, 0);
385
386 PROC_UNLOCK(p);
387
388 return (0);
389 }
390
391 /*
392 * The knote may be attached to a different process, which may exit,
393 * leaving nothing for the knote to be attached to. So when the process
394 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
395 * it will be deleted when read out. However, as part of the knote deletion,
396 * this routine is called, so a check is needed to avoid actually performing
397 * a detach, because the original process does not exist any more.
398 */
399 /* XXX - move to kern_proc.c? */
400 static void
401 filt_procdetach(struct knote *kn)
402 {
403 struct proc *p;
404
405 p = kn->kn_ptr.p_proc;
406 knlist_remove(&p->p_klist, kn, 0);
407 kn->kn_ptr.p_proc = NULL;
408 }
409
410 /* XXX - move to kern_proc.c? */
411 static int
412 filt_proc(struct knote *kn, long hint)
413 {
414 struct proc *p = kn->kn_ptr.p_proc;
415 u_int event;
416
417 /*
418 * mask off extra data
419 */
420 event = (u_int)hint & NOTE_PCTRLMASK;
421
422 /*
423 * if the user is interested in this event, record it.
424 */
425 if (kn->kn_sfflags & event)
426 kn->kn_fflags |= event;
427
428 /*
429 * process is gone, so flag the event as finished.
430 */
431 if (event == NOTE_EXIT) {
432 if (!(kn->kn_status & KN_DETACHED))
433 knlist_remove_inevent(&p->p_klist, kn);
434 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
435 kn->kn_ptr.p_proc = NULL;
436 if (kn->kn_fflags & NOTE_EXIT)
437 kn->kn_data = p->p_xstat;
438 if (kn->kn_fflags == 0)
439 kn->kn_flags |= EV_DROP;
440 return (1);
441 }
442
443 return (kn->kn_fflags != 0);
444 }
445
446 /*
447 * Called when the process forked. It mostly does the same as the
448 * knote(), activating all knotes registered to be activated when the
449 * process forked. Additionally, for each knote attached to the
450 * parent, check whether user wants to track the new process. If so
451 * attach a new knote to it, and immediately report an event with the
452 * child's pid.
453 */
454 void
455 knote_fork(struct knlist *list, int pid)
456 {
457 struct kqueue *kq;
458 struct knote *kn;
459 struct kevent kev;
460 int error;
461
462 if (list == NULL)
463 return;
464 list->kl_lock(list->kl_lockarg);
465
466 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
467 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
468 continue;
469 kq = kn->kn_kq;
470 KQ_LOCK(kq);
471 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
472 KQ_UNLOCK(kq);
473 continue;
474 }
475
476 /*
477 * The same as knote(), activate the event.
478 */
479 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
480 kn->kn_status |= KN_HASKQLOCK;
481 if (kn->kn_fop->f_event(kn, NOTE_FORK))
482 KNOTE_ACTIVATE(kn, 1);
483 kn->kn_status &= ~KN_HASKQLOCK;
484 KQ_UNLOCK(kq);
485 continue;
486 }
487
488 /*
489 * The NOTE_TRACK case. In addition to the activation
490 * of the event, we need to register new event to
491 * track the child. Drop the locks in preparation for
492 * the call to kqueue_register().
493 */
494 kn->kn_status |= KN_INFLUX;
495 KQ_UNLOCK(kq);
496 list->kl_unlock(list->kl_lockarg);
497
498 /*
499 * Activate existing knote and register a knote with
500 * new process.
501 */
502 kev.ident = pid;
503 kev.filter = kn->kn_filter;
504 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
505 kev.fflags = kn->kn_sfflags;
506 kev.data = kn->kn_id; /* parent */
507 kev.udata = kn->kn_kevent.udata;/* preserve udata */
508 error = kqueue_register(kq, &kev, NULL, 0);
509 if (error)
510 kn->kn_fflags |= NOTE_TRACKERR;
511 if (kn->kn_fop->f_event(kn, NOTE_FORK))
512 KNOTE_ACTIVATE(kn, 0);
513 KQ_LOCK(kq);
514 kn->kn_status &= ~KN_INFLUX;
515 KQ_UNLOCK_FLUX(kq);
516 list->kl_lock(list->kl_lockarg);
517 }
518 list->kl_unlock(list->kl_lockarg);
519 }
520
521 /*
522 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
523 * interval timer support code.
524 */
525
526 #define NOTE_TIMER_PRECMASK (NOTE_SECONDS|NOTE_MSECONDS|NOTE_USECONDS| \
527 NOTE_NSECONDS)
528
529 static __inline sbintime_t
530 timer2sbintime(intptr_t data, int flags)
531 {
532 sbintime_t modifier;
533
534 switch (flags & NOTE_TIMER_PRECMASK) {
535 case NOTE_SECONDS:
536 modifier = SBT_1S;
537 break;
538 case NOTE_MSECONDS: /* FALLTHROUGH */
539 case 0:
540 modifier = SBT_1MS;
541 break;
542 case NOTE_USECONDS:
543 modifier = SBT_1US;
544 break;
545 case NOTE_NSECONDS:
546 modifier = SBT_1NS;
547 break;
548 default:
549 return (-1);
550 }
551
552 #ifdef __LP64__
553 if (data > SBT_MAX / modifier)
554 return (SBT_MAX);
555 #endif
556 return (modifier * data);
557 }
558
559 static void
560 filt_timerexpire(void *knx)
561 {
562 struct callout *calloutp;
563 struct knote *kn;
564
565 kn = knx;
566 kn->kn_data++;
567 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
568
569 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
570 calloutp = (struct callout *)kn->kn_hook;
571 *kn->kn_ptr.p_nexttime += timer2sbintime(kn->kn_sdata,
572 kn->kn_sfflags);
573 callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0,
574 filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE);
575 }
576 }
577
578 /*
579 * data contains amount of time to sleep
580 */
581 static int
582 filt_timerattach(struct knote *kn)
583 {
584 struct callout *calloutp;
585 sbintime_t to;
586 unsigned int ncallouts;
587
588 if ((intptr_t)kn->kn_sdata < 0)
589 return (EINVAL);
590 if ((intptr_t)kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
591 kn->kn_sdata = 1;
592 /* Only precision unit are supported in flags so far */
593 if (kn->kn_sfflags & ~NOTE_TIMER_PRECMASK)
594 return (EINVAL);
595
596 to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
597 if (to < 0)
598 return (EINVAL);
599
600 ncallouts = atomic_load_explicit(&kq_ncallouts, memory_order_relaxed);
601 do {
602 if (ncallouts >= kq_calloutmax)
603 return (ENOMEM);
604 } while (!atomic_compare_exchange_weak_explicit(&kq_ncallouts,
605 &ncallouts, ncallouts + 1, memory_order_relaxed,
606 memory_order_relaxed));
607
608 kn->kn_flags |= EV_CLEAR; /* automatically set */
609 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */
610 kn->kn_ptr.p_nexttime = malloc(sizeof(sbintime_t), M_KQUEUE, M_WAITOK);
611 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK);
612 callout_init(calloutp, CALLOUT_MPSAFE);
613 kn->kn_hook = calloutp;
614 *kn->kn_ptr.p_nexttime = to + sbinuptime();
615 callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0,
616 filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE);
617
618 return (0);
619 }
620
621 static void
622 filt_timerdetach(struct knote *kn)
623 {
624 struct callout *calloutp;
625 unsigned int old;
626
627 calloutp = (struct callout *)kn->kn_hook;
628 callout_drain(calloutp);
629 free(calloutp, M_KQUEUE);
630 free(kn->kn_ptr.p_nexttime, M_KQUEUE);
631 old = atomic_fetch_sub_explicit(&kq_ncallouts, 1, memory_order_relaxed);
632 KASSERT(old > 0, ("Number of callouts cannot become negative"));
633 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */
634 }
635
636 static int
637 filt_timer(struct knote *kn, long hint)
638 {
639
640 return (kn->kn_data != 0);
641 }
642
643 static int
644 filt_userattach(struct knote *kn)
645 {
646
647 /*
648 * EVFILT_USER knotes are not attached to anything in the kernel.
649 */
650 kn->kn_hook = NULL;
651 if (kn->kn_fflags & NOTE_TRIGGER)
652 kn->kn_hookid = 1;
653 else
654 kn->kn_hookid = 0;
655 return (0);
656 }
657
658 static void
659 filt_userdetach(__unused struct knote *kn)
660 {
661
662 /*
663 * EVFILT_USER knotes are not attached to anything in the kernel.
664 */
665 }
666
667 static int
668 filt_user(struct knote *kn, __unused long hint)
669 {
670
671 return (kn->kn_hookid);
672 }
673
674 static void
675 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
676 {
677 u_int ffctrl;
678
679 switch (type) {
680 case EVENT_REGISTER:
681 if (kev->fflags & NOTE_TRIGGER)
682 kn->kn_hookid = 1;
683
684 ffctrl = kev->fflags & NOTE_FFCTRLMASK;
685 kev->fflags &= NOTE_FFLAGSMASK;
686 switch (ffctrl) {
687 case NOTE_FFNOP:
688 break;
689
690 case NOTE_FFAND:
691 kn->kn_sfflags &= kev->fflags;
692 break;
693
694 case NOTE_FFOR:
695 kn->kn_sfflags |= kev->fflags;
696 break;
697
698 case NOTE_FFCOPY:
699 kn->kn_sfflags = kev->fflags;
700 break;
701
702 default:
703 /* XXX Return error? */
704 break;
705 }
706 kn->kn_sdata = kev->data;
707 if (kev->flags & EV_CLEAR) {
708 kn->kn_hookid = 0;
709 kn->kn_data = 0;
710 kn->kn_fflags = 0;
711 }
712 break;
713
714 case EVENT_PROCESS:
715 *kev = kn->kn_kevent;
716 kev->fflags = kn->kn_sfflags;
717 kev->data = kn->kn_sdata;
718 if (kn->kn_flags & EV_CLEAR) {
719 kn->kn_hookid = 0;
720 kn->kn_data = 0;
721 kn->kn_fflags = 0;
722 }
723 break;
724
725 default:
726 panic("filt_usertouch() - invalid type (%ld)", type);
727 break;
728 }
729 }
730
731 int
732 sys_kqueue(struct thread *td, struct kqueue_args *uap)
733 {
734
735 return (kern_kqueue(td, 0));
736 }
737
738 int
739 kern_kqueue(struct thread *td, int flags)
740 {
741 struct filedesc *fdp;
742 struct kqueue *kq;
743 struct file *fp;
744 int fd, error;
745
746 fdp = td->td_proc->p_fd;
747 error = falloc(td, &fp, &fd, flags);
748 if (error)
749 goto done2;
750
751 /* An extra reference on `fp' has been held for us by falloc(). */
752 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
753 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
754 TAILQ_INIT(&kq->kq_head);
755 kq->kq_fdp = fdp;
756 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
757 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
758
759 FILEDESC_XLOCK(fdp);
760 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
761 FILEDESC_XUNLOCK(fdp);
762
763 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
764 fdrop(fp, td);
765
766 td->td_retval[0] = fd;
767 done2:
768 return (error);
769 }
770
771 #ifndef _SYS_SYSPROTO_H_
772 struct kevent_args {
773 int fd;
774 const struct kevent *changelist;
775 int nchanges;
776 struct kevent *eventlist;
777 int nevents;
778 const struct timespec *timeout;
779 };
780 #endif
781 int
782 sys_kevent(struct thread *td, struct kevent_args *uap)
783 {
784 struct timespec ts, *tsp;
785 struct kevent_copyops k_ops = { uap,
786 kevent_copyout,
787 kevent_copyin};
788 int error;
789 #ifdef KTRACE
790 struct uio ktruio;
791 struct iovec ktriov;
792 struct uio *ktruioin = NULL;
793 struct uio *ktruioout = NULL;
794 #endif
795
796 if (uap->timeout != NULL) {
797 error = copyin(uap->timeout, &ts, sizeof(ts));
798 if (error)
799 return (error);
800 tsp = &ts;
801 } else
802 tsp = NULL;
803
804 #ifdef KTRACE
805 if (KTRPOINT(td, KTR_GENIO)) {
806 ktriov.iov_base = uap->changelist;
807 ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
808 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
809 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
810 .uio_td = td };
811 ktruioin = cloneuio(&ktruio);
812 ktriov.iov_base = uap->eventlist;
813 ktriov.iov_len = uap->nevents * sizeof(struct kevent);
814 ktruioout = cloneuio(&ktruio);
815 }
816 #endif
817
818 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
819 &k_ops, tsp);
820
821 #ifdef KTRACE
822 if (ktruioin != NULL) {
823 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
824 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
825 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
826 ktrgenio(uap->fd, UIO_READ, ktruioout, error);
827 }
828 #endif
829
830 return (error);
831 }
832
833 /*
834 * Copy 'count' items into the destination list pointed to by uap->eventlist.
835 */
836 static int
837 kevent_copyout(void *arg, struct kevent *kevp, int count)
838 {
839 struct kevent_args *uap;
840 int error;
841
842 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
843 uap = (struct kevent_args *)arg;
844
845 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
846 if (error == 0)
847 uap->eventlist += count;
848 return (error);
849 }
850
851 /*
852 * Copy 'count' items from the list pointed to by uap->changelist.
853 */
854 static int
855 kevent_copyin(void *arg, struct kevent *kevp, int count)
856 {
857 struct kevent_args *uap;
858 int error;
859
860 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
861 uap = (struct kevent_args *)arg;
862
863 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
864 if (error == 0)
865 uap->changelist += count;
866 return (error);
867 }
868
869 int
870 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
871 struct kevent_copyops *k_ops, const struct timespec *timeout)
872 {
873 cap_rights_t rights;
874 struct file *fp;
875 int error;
876
877 cap_rights_init(&rights);
878 if (nchanges > 0)
879 cap_rights_set(&rights, CAP_KQUEUE_CHANGE);
880 if (nevents > 0)
881 cap_rights_set(&rights, CAP_KQUEUE_EVENT);
882 error = fget(td, fd, &rights, &fp);
883 if (error != 0)
884 return (error);
885
886 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
887 fdrop(fp, td);
888
889 return (error);
890 }
891
892 int
893 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
894 struct kevent_copyops *k_ops, const struct timespec *timeout)
895 {
896 struct kevent keva[KQ_NEVENTS];
897 struct kevent *kevp, *changes;
898 struct kqueue *kq;
899 int i, n, nerrors, error;
900
901 error = kqueue_acquire(fp, &kq);
902 if (error != 0)
903 return (error);
904
905 nerrors = 0;
906
907 while (nchanges > 0) {
908 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
909 error = k_ops->k_copyin(k_ops->arg, keva, n);
910 if (error)
911 goto done;
912 changes = keva;
913 for (i = 0; i < n; i++) {
914 kevp = &changes[i];
915 if (!kevp->filter)
916 continue;
917 kevp->flags &= ~EV_SYSFLAGS;
918 error = kqueue_register(kq, kevp, td, 1);
919 if (error || (kevp->flags & EV_RECEIPT)) {
920 if (nevents != 0) {
921 kevp->flags = EV_ERROR;
922 kevp->data = error;
923 (void) k_ops->k_copyout(k_ops->arg,
924 kevp, 1);
925 nevents--;
926 nerrors++;
927 } else {
928 goto done;
929 }
930 }
931 }
932 nchanges -= n;
933 }
934 if (nerrors) {
935 td->td_retval[0] = nerrors;
936 error = 0;
937 goto done;
938 }
939
940 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
941 done:
942 kqueue_release(kq, 0);
943 return (error);
944 }
945
946 int
947 kqueue_add_filteropts(int filt, struct filterops *filtops)
948 {
949 int error;
950
951 error = 0;
952 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
953 printf(
954 "trying to add a filterop that is out of range: %d is beyond %d\n",
955 ~filt, EVFILT_SYSCOUNT);
956 return EINVAL;
957 }
958 mtx_lock(&filterops_lock);
959 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
960 sysfilt_ops[~filt].for_fop != NULL)
961 error = EEXIST;
962 else {
963 sysfilt_ops[~filt].for_fop = filtops;
964 sysfilt_ops[~filt].for_refcnt = 0;
965 }
966 mtx_unlock(&filterops_lock);
967
968 return (error);
969 }
970
971 int
972 kqueue_del_filteropts(int filt)
973 {
974 int error;
975
976 error = 0;
977 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
978 return EINVAL;
979
980 mtx_lock(&filterops_lock);
981 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
982 sysfilt_ops[~filt].for_fop == NULL)
983 error = EINVAL;
984 else if (sysfilt_ops[~filt].for_refcnt != 0)
985 error = EBUSY;
986 else {
987 sysfilt_ops[~filt].for_fop = &null_filtops;
988 sysfilt_ops[~filt].for_refcnt = 0;
989 }
990 mtx_unlock(&filterops_lock);
991
992 return error;
993 }
994
995 static struct filterops *
996 kqueue_fo_find(int filt)
997 {
998
999 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1000 return NULL;
1001
1002 mtx_lock(&filterops_lock);
1003 sysfilt_ops[~filt].for_refcnt++;
1004 if (sysfilt_ops[~filt].for_fop == NULL)
1005 sysfilt_ops[~filt].for_fop = &null_filtops;
1006 mtx_unlock(&filterops_lock);
1007
1008 return sysfilt_ops[~filt].for_fop;
1009 }
1010
1011 static void
1012 kqueue_fo_release(int filt)
1013 {
1014
1015 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1016 return;
1017
1018 mtx_lock(&filterops_lock);
1019 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1020 ("filter object refcount not valid on release"));
1021 sysfilt_ops[~filt].for_refcnt--;
1022 mtx_unlock(&filterops_lock);
1023 }
1024
1025 /*
1026 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will
1027 * influence if memory allocation should wait. Make sure it is 0 if you
1028 * hold any mutexes.
1029 */
1030 static int
1031 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
1032 {
1033 struct filterops *fops;
1034 struct file *fp;
1035 struct knote *kn, *tkn;
1036 cap_rights_t rights;
1037 int error, filt, event;
1038 int haskqglobal, filedesc_unlock;
1039
1040 fp = NULL;
1041 kn = NULL;
1042 error = 0;
1043 haskqglobal = 0;
1044 filedesc_unlock = 0;
1045
1046 filt = kev->filter;
1047 fops = kqueue_fo_find(filt);
1048 if (fops == NULL)
1049 return EINVAL;
1050
1051 tkn = knote_alloc(waitok); /* prevent waiting with locks */
1052
1053 findkn:
1054 if (fops->f_isfd) {
1055 KASSERT(td != NULL, ("td is NULL"));
1056 error = fget(td, kev->ident,
1057 cap_rights_init(&rights, CAP_EVENT), &fp);
1058 if (error)
1059 goto done;
1060
1061 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1062 kev->ident, 0) != 0) {
1063 /* try again */
1064 fdrop(fp, td);
1065 fp = NULL;
1066 error = kqueue_expand(kq, fops, kev->ident, waitok);
1067 if (error)
1068 goto done;
1069 goto findkn;
1070 }
1071
1072 if (fp->f_type == DTYPE_KQUEUE) {
1073 /*
1074 * if we add some inteligence about what we are doing,
1075 * we should be able to support events on ourselves.
1076 * We need to know when we are doing this to prevent
1077 * getting both the knlist lock and the kq lock since
1078 * they are the same thing.
1079 */
1080 if (fp->f_data == kq) {
1081 error = EINVAL;
1082 goto done;
1083 }
1084
1085 /*
1086 * Pre-lock the filedesc before the global
1087 * lock mutex, see the comment in
1088 * kqueue_close().
1089 */
1090 FILEDESC_XLOCK(td->td_proc->p_fd);
1091 filedesc_unlock = 1;
1092 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1093 }
1094
1095 KQ_LOCK(kq);
1096 if (kev->ident < kq->kq_knlistsize) {
1097 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1098 if (kev->filter == kn->kn_filter)
1099 break;
1100 }
1101 } else {
1102 if ((kev->flags & EV_ADD) == EV_ADD)
1103 kqueue_expand(kq, fops, kev->ident, waitok);
1104
1105 KQ_LOCK(kq);
1106 if (kq->kq_knhashmask != 0) {
1107 struct klist *list;
1108
1109 list = &kq->kq_knhash[
1110 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1111 SLIST_FOREACH(kn, list, kn_link)
1112 if (kev->ident == kn->kn_id &&
1113 kev->filter == kn->kn_filter)
1114 break;
1115 }
1116 }
1117
1118 /* knote is in the process of changing, wait for it to stablize. */
1119 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1120 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1121 if (filedesc_unlock) {
1122 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1123 filedesc_unlock = 0;
1124 }
1125 kq->kq_state |= KQ_FLUXWAIT;
1126 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1127 if (fp != NULL) {
1128 fdrop(fp, td);
1129 fp = NULL;
1130 }
1131 goto findkn;
1132 }
1133
1134 /*
1135 * kn now contains the matching knote, or NULL if no match
1136 */
1137 if (kn == NULL) {
1138 if (kev->flags & EV_ADD) {
1139 kn = tkn;
1140 tkn = NULL;
1141 if (kn == NULL) {
1142 KQ_UNLOCK(kq);
1143 error = ENOMEM;
1144 goto done;
1145 }
1146 kn->kn_fp = fp;
1147 kn->kn_kq = kq;
1148 kn->kn_fop = fops;
1149 /*
1150 * apply reference counts to knote structure, and
1151 * do not release it at the end of this routine.
1152 */
1153 fops = NULL;
1154 fp = NULL;
1155
1156 kn->kn_sfflags = kev->fflags;
1157 kn->kn_sdata = kev->data;
1158 kev->fflags = 0;
1159 kev->data = 0;
1160 kn->kn_kevent = *kev;
1161 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1162 EV_ENABLE | EV_DISABLE);
1163 kn->kn_status = KN_INFLUX|KN_DETACHED;
1164
1165 error = knote_attach(kn, kq);
1166 KQ_UNLOCK(kq);
1167 if (error != 0) {
1168 tkn = kn;
1169 goto done;
1170 }
1171
1172 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1173 knote_drop(kn, td);
1174 goto done;
1175 }
1176 KN_LIST_LOCK(kn);
1177 goto done_ev_add;
1178 } else {
1179 /* No matching knote and the EV_ADD flag is not set. */
1180 KQ_UNLOCK(kq);
1181 error = ENOENT;
1182 goto done;
1183 }
1184 }
1185
1186 if (kev->flags & EV_DELETE) {
1187 kn->kn_status |= KN_INFLUX;
1188 KQ_UNLOCK(kq);
1189 if (!(kn->kn_status & KN_DETACHED))
1190 kn->kn_fop->f_detach(kn);
1191 knote_drop(kn, td);
1192 goto done;
1193 }
1194
1195 /*
1196 * The user may change some filter values after the initial EV_ADD,
1197 * but doing so will not reset any filter which has already been
1198 * triggered.
1199 */
1200 kn->kn_status |= KN_INFLUX | KN_SCAN;
1201 KQ_UNLOCK(kq);
1202 KN_LIST_LOCK(kn);
1203 kn->kn_kevent.udata = kev->udata;
1204 if (!fops->f_isfd && fops->f_touch != NULL) {
1205 fops->f_touch(kn, kev, EVENT_REGISTER);
1206 } else {
1207 kn->kn_sfflags = kev->fflags;
1208 kn->kn_sdata = kev->data;
1209 }
1210
1211 /*
1212 * We can get here with kn->kn_knlist == NULL. This can happen when
1213 * the initial attach event decides that the event is "completed"
1214 * already. i.e. filt_procattach is called on a zombie process. It
1215 * will call filt_proc which will remove it from the list, and NULL
1216 * kn_knlist.
1217 */
1218 done_ev_add:
1219 event = kn->kn_fop->f_event(kn, 0);
1220 KQ_LOCK(kq);
1221 if (event)
1222 KNOTE_ACTIVATE(kn, 1);
1223 kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1224 KN_LIST_UNLOCK(kn);
1225
1226 if ((kev->flags & EV_DISABLE) &&
1227 ((kn->kn_status & KN_DISABLED) == 0)) {
1228 kn->kn_status |= KN_DISABLED;
1229 }
1230
1231 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1232 kn->kn_status &= ~KN_DISABLED;
1233 if ((kn->kn_status & KN_ACTIVE) &&
1234 ((kn->kn_status & KN_QUEUED) == 0))
1235 knote_enqueue(kn);
1236 }
1237 KQ_UNLOCK_FLUX(kq);
1238
1239 done:
1240 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1241 if (filedesc_unlock)
1242 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1243 if (fp != NULL)
1244 fdrop(fp, td);
1245 if (tkn != NULL)
1246 knote_free(tkn);
1247 if (fops != NULL)
1248 kqueue_fo_release(filt);
1249 return (error);
1250 }
1251
1252 static int
1253 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1254 {
1255 int error;
1256 struct kqueue *kq;
1257
1258 error = 0;
1259
1260 kq = fp->f_data;
1261 if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1262 return (EBADF);
1263 *kqp = kq;
1264 KQ_LOCK(kq);
1265 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1266 KQ_UNLOCK(kq);
1267 return (EBADF);
1268 }
1269 kq->kq_refcnt++;
1270 KQ_UNLOCK(kq);
1271
1272 return error;
1273 }
1274
1275 static void
1276 kqueue_release(struct kqueue *kq, int locked)
1277 {
1278 if (locked)
1279 KQ_OWNED(kq);
1280 else
1281 KQ_LOCK(kq);
1282 kq->kq_refcnt--;
1283 if (kq->kq_refcnt == 1)
1284 wakeup(&kq->kq_refcnt);
1285 if (!locked)
1286 KQ_UNLOCK(kq);
1287 }
1288
1289 static void
1290 kqueue_schedtask(struct kqueue *kq)
1291 {
1292
1293 KQ_OWNED(kq);
1294 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1295 ("scheduling kqueue task while draining"));
1296
1297 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1298 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1299 kq->kq_state |= KQ_TASKSCHED;
1300 }
1301 }
1302
1303 /*
1304 * Expand the kq to make sure we have storage for fops/ident pair.
1305 *
1306 * Return 0 on success (or no work necessary), return errno on failure.
1307 *
1308 * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1309 * If kqueue_register is called from a non-fd context, there usually/should
1310 * be no locks held.
1311 */
1312 static int
1313 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1314 int waitok)
1315 {
1316 struct klist *list, *tmp_knhash, *to_free;
1317 u_long tmp_knhashmask;
1318 int size;
1319 int fd;
1320 int mflag = waitok ? M_WAITOK : M_NOWAIT;
1321
1322 KQ_NOTOWNED(kq);
1323
1324 to_free = NULL;
1325 if (fops->f_isfd) {
1326 fd = ident;
1327 if (kq->kq_knlistsize <= fd) {
1328 size = kq->kq_knlistsize;
1329 while (size <= fd)
1330 size += KQEXTENT;
1331 list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1332 if (list == NULL)
1333 return ENOMEM;
1334 KQ_LOCK(kq);
1335 if (kq->kq_knlistsize > fd) {
1336 to_free = list;
1337 list = NULL;
1338 } else {
1339 if (kq->kq_knlist != NULL) {
1340 bcopy(kq->kq_knlist, list,
1341 kq->kq_knlistsize * sizeof(*list));
1342 to_free = kq->kq_knlist;
1343 kq->kq_knlist = NULL;
1344 }
1345 bzero((caddr_t)list +
1346 kq->kq_knlistsize * sizeof(*list),
1347 (size - kq->kq_knlistsize) * sizeof(*list));
1348 kq->kq_knlistsize = size;
1349 kq->kq_knlist = list;
1350 }
1351 KQ_UNLOCK(kq);
1352 }
1353 } else {
1354 if (kq->kq_knhashmask == 0) {
1355 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1356 &tmp_knhashmask);
1357 if (tmp_knhash == NULL)
1358 return ENOMEM;
1359 KQ_LOCK(kq);
1360 if (kq->kq_knhashmask == 0) {
1361 kq->kq_knhash = tmp_knhash;
1362 kq->kq_knhashmask = tmp_knhashmask;
1363 } else {
1364 to_free = tmp_knhash;
1365 }
1366 KQ_UNLOCK(kq);
1367 }
1368 }
1369 free(to_free, M_KQUEUE);
1370
1371 KQ_NOTOWNED(kq);
1372 return 0;
1373 }
1374
1375 static void
1376 kqueue_task(void *arg, int pending)
1377 {
1378 struct kqueue *kq;
1379 int haskqglobal;
1380
1381 haskqglobal = 0;
1382 kq = arg;
1383
1384 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1385 KQ_LOCK(kq);
1386
1387 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1388
1389 kq->kq_state &= ~KQ_TASKSCHED;
1390 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1391 wakeup(&kq->kq_state);
1392 }
1393 KQ_UNLOCK(kq);
1394 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1395 }
1396
1397 /*
1398 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1399 * We treat KN_MARKER knotes as if they are INFLUX.
1400 */
1401 static int
1402 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1403 const struct timespec *tsp, struct kevent *keva, struct thread *td)
1404 {
1405 struct kevent *kevp;
1406 struct knote *kn, *marker;
1407 sbintime_t asbt, rsbt;
1408 int count, error, haskqglobal, influx, nkev, touch;
1409
1410 count = maxevents;
1411 nkev = 0;
1412 error = 0;
1413 haskqglobal = 0;
1414
1415 if (maxevents == 0)
1416 goto done_nl;
1417
1418 rsbt = 0;
1419 if (tsp != NULL) {
1420 if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 ||
1421 tsp->tv_nsec >= 1000000000) {
1422 error = EINVAL;
1423 goto done_nl;
1424 }
1425 if (timespecisset(tsp)) {
1426 if (tsp->tv_sec <= INT32_MAX) {
1427 rsbt = tstosbt(*tsp);
1428 if (TIMESEL(&asbt, rsbt))
1429 asbt += tc_tick_sbt;
1430 if (asbt <= INT64_MAX - rsbt)
1431 asbt += rsbt;
1432 else
1433 asbt = 0;
1434 rsbt >>= tc_precexp;
1435 } else
1436 asbt = 0;
1437 } else
1438 asbt = -1;
1439 } else
1440 asbt = 0;
1441 marker = knote_alloc(1);
1442 if (marker == NULL) {
1443 error = ENOMEM;
1444 goto done_nl;
1445 }
1446 marker->kn_status = KN_MARKER;
1447 KQ_LOCK(kq);
1448
1449 retry:
1450 kevp = keva;
1451 if (kq->kq_count == 0) {
1452 if (asbt == -1) {
1453 error = EWOULDBLOCK;
1454 } else {
1455 kq->kq_state |= KQ_SLEEP;
1456 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1457 "kqread", asbt, rsbt, C_ABSOLUTE);
1458 }
1459 if (error == 0)
1460 goto retry;
1461 /* don't restart after signals... */
1462 if (error == ERESTART)
1463 error = EINTR;
1464 else if (error == EWOULDBLOCK)
1465 error = 0;
1466 goto done;
1467 }
1468
1469 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1470 influx = 0;
1471 while (count) {
1472 KQ_OWNED(kq);
1473 kn = TAILQ_FIRST(&kq->kq_head);
1474
1475 if ((kn->kn_status == KN_MARKER && kn != marker) ||
1476 (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1477 if (influx) {
1478 influx = 0;
1479 KQ_FLUX_WAKEUP(kq);
1480 }
1481 kq->kq_state |= KQ_FLUXWAIT;
1482 error = msleep(kq, &kq->kq_lock, PSOCK,
1483 "kqflxwt", 0);
1484 continue;
1485 }
1486
1487 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1488 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1489 kn->kn_status &= ~KN_QUEUED;
1490 kq->kq_count--;
1491 continue;
1492 }
1493 if (kn == marker) {
1494 KQ_FLUX_WAKEUP(kq);
1495 if (count == maxevents)
1496 goto retry;
1497 goto done;
1498 }
1499 KASSERT((kn->kn_status & KN_INFLUX) == 0,
1500 ("KN_INFLUX set when not suppose to be"));
1501
1502 if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1503 kn->kn_status &= ~KN_QUEUED;
1504 kn->kn_status |= KN_INFLUX;
1505 kq->kq_count--;
1506 KQ_UNLOCK(kq);
1507 /*
1508 * We don't need to lock the list since we've marked
1509 * it _INFLUX.
1510 */
1511 if (!(kn->kn_status & KN_DETACHED))
1512 kn->kn_fop->f_detach(kn);
1513 knote_drop(kn, td);
1514 KQ_LOCK(kq);
1515 continue;
1516 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1517 kn->kn_status &= ~KN_QUEUED;
1518 kn->kn_status |= KN_INFLUX;
1519 kq->kq_count--;
1520 KQ_UNLOCK(kq);
1521 /*
1522 * We don't need to lock the list since we've marked
1523 * it _INFLUX.
1524 */
1525 *kevp = kn->kn_kevent;
1526 if (!(kn->kn_status & KN_DETACHED))
1527 kn->kn_fop->f_detach(kn);
1528 knote_drop(kn, td);
1529 KQ_LOCK(kq);
1530 kn = NULL;
1531 } else {
1532 kn->kn_status |= KN_INFLUX | KN_SCAN;
1533 KQ_UNLOCK(kq);
1534 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1535 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1536 KN_LIST_LOCK(kn);
1537 if (kn->kn_fop->f_event(kn, 0) == 0) {
1538 KQ_LOCK(kq);
1539 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1540 kn->kn_status &=
1541 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX |
1542 KN_SCAN);
1543 kq->kq_count--;
1544 KN_LIST_UNLOCK(kn);
1545 influx = 1;
1546 continue;
1547 }
1548 touch = (!kn->kn_fop->f_isfd &&
1549 kn->kn_fop->f_touch != NULL);
1550 if (touch)
1551 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1552 else
1553 *kevp = kn->kn_kevent;
1554 KQ_LOCK(kq);
1555 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1556 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1557 /*
1558 * Manually clear knotes who weren't
1559 * 'touch'ed.
1560 */
1561 if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1562 kn->kn_data = 0;
1563 kn->kn_fflags = 0;
1564 }
1565 if (kn->kn_flags & EV_DISPATCH)
1566 kn->kn_status |= KN_DISABLED;
1567 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1568 kq->kq_count--;
1569 } else
1570 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1571
1572 kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1573 KN_LIST_UNLOCK(kn);
1574 influx = 1;
1575 }
1576
1577 /* we are returning a copy to the user */
1578 kevp++;
1579 nkev++;
1580 count--;
1581
1582 if (nkev == KQ_NEVENTS) {
1583 influx = 0;
1584 KQ_UNLOCK_FLUX(kq);
1585 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1586 nkev = 0;
1587 kevp = keva;
1588 KQ_LOCK(kq);
1589 if (error)
1590 break;
1591 }
1592 }
1593 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1594 done:
1595 KQ_OWNED(kq);
1596 KQ_UNLOCK_FLUX(kq);
1597 knote_free(marker);
1598 done_nl:
1599 KQ_NOTOWNED(kq);
1600 if (nkev != 0)
1601 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1602 td->td_retval[0] = maxevents - count;
1603 return (error);
1604 }
1605
1606 /*
1607 * XXX
1608 * This could be expanded to call kqueue_scan, if desired.
1609 */
1610 /*ARGSUSED*/
1611 static int
1612 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1613 int flags, struct thread *td)
1614 {
1615 return (ENXIO);
1616 }
1617
1618 /*ARGSUSED*/
1619 static int
1620 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1621 int flags, struct thread *td)
1622 {
1623 return (ENXIO);
1624 }
1625
1626 /*ARGSUSED*/
1627 static int
1628 kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1629 struct thread *td)
1630 {
1631
1632 return (EINVAL);
1633 }
1634
1635 /*ARGSUSED*/
1636 static int
1637 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1638 struct ucred *active_cred, struct thread *td)
1639 {
1640 /*
1641 * Enabling sigio causes two major problems:
1642 * 1) infinite recursion:
1643 * Synopsys: kevent is being used to track signals and have FIOASYNC
1644 * set. On receipt of a signal this will cause a kqueue to recurse
1645 * into itself over and over. Sending the sigio causes the kqueue
1646 * to become ready, which in turn posts sigio again, forever.
1647 * Solution: this can be solved by setting a flag in the kqueue that
1648 * we have a SIGIO in progress.
1649 * 2) locking problems:
1650 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1651 * us above the proc and pgrp locks.
1652 * Solution: Post a signal using an async mechanism, being sure to
1653 * record a generation count in the delivery so that we do not deliver
1654 * a signal to the wrong process.
1655 *
1656 * Note, these two mechanisms are somewhat mutually exclusive!
1657 */
1658 #if 0
1659 struct kqueue *kq;
1660
1661 kq = fp->f_data;
1662 switch (cmd) {
1663 case FIOASYNC:
1664 if (*(int *)data) {
1665 kq->kq_state |= KQ_ASYNC;
1666 } else {
1667 kq->kq_state &= ~KQ_ASYNC;
1668 }
1669 return (0);
1670
1671 case FIOSETOWN:
1672 return (fsetown(*(int *)data, &kq->kq_sigio));
1673
1674 case FIOGETOWN:
1675 *(int *)data = fgetown(&kq->kq_sigio);
1676 return (0);
1677 }
1678 #endif
1679
1680 return (ENOTTY);
1681 }
1682
1683 /*ARGSUSED*/
1684 static int
1685 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1686 struct thread *td)
1687 {
1688 struct kqueue *kq;
1689 int revents = 0;
1690 int error;
1691
1692 if ((error = kqueue_acquire(fp, &kq)))
1693 return POLLERR;
1694
1695 KQ_LOCK(kq);
1696 if (events & (POLLIN | POLLRDNORM)) {
1697 if (kq->kq_count) {
1698 revents |= events & (POLLIN | POLLRDNORM);
1699 } else {
1700 selrecord(td, &kq->kq_sel);
1701 if (SEL_WAITING(&kq->kq_sel))
1702 kq->kq_state |= KQ_SEL;
1703 }
1704 }
1705 kqueue_release(kq, 1);
1706 KQ_UNLOCK(kq);
1707 return (revents);
1708 }
1709
1710 /*ARGSUSED*/
1711 static int
1712 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1713 struct thread *td)
1714 {
1715
1716 bzero((void *)st, sizeof *st);
1717 /*
1718 * We no longer return kq_count because the unlocked value is useless.
1719 * If you spent all this time getting the count, why not spend your
1720 * syscall better by calling kevent?
1721 *
1722 * XXX - This is needed for libc_r.
1723 */
1724 st->st_mode = S_IFIFO;
1725 return (0);
1726 }
1727
1728 /*ARGSUSED*/
1729 static int
1730 kqueue_close(struct file *fp, struct thread *td)
1731 {
1732 struct kqueue *kq = fp->f_data;
1733 struct filedesc *fdp;
1734 struct knote *kn;
1735 int i;
1736 int error;
1737 int filedesc_unlock;
1738
1739 if ((error = kqueue_acquire(fp, &kq)))
1740 return error;
1741
1742 filedesc_unlock = 0;
1743 KQ_LOCK(kq);
1744
1745 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1746 ("kqueue already closing"));
1747 kq->kq_state |= KQ_CLOSING;
1748 if (kq->kq_refcnt > 1)
1749 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1750
1751 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1752 fdp = kq->kq_fdp;
1753
1754 KASSERT(knlist_empty(&kq->kq_sel.si_note),
1755 ("kqueue's knlist not empty"));
1756
1757 for (i = 0; i < kq->kq_knlistsize; i++) {
1758 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1759 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1760 kq->kq_state |= KQ_FLUXWAIT;
1761 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1762 continue;
1763 }
1764 kn->kn_status |= KN_INFLUX;
1765 KQ_UNLOCK(kq);
1766 if (!(kn->kn_status & KN_DETACHED))
1767 kn->kn_fop->f_detach(kn);
1768 knote_drop(kn, td);
1769 KQ_LOCK(kq);
1770 }
1771 }
1772 if (kq->kq_knhashmask != 0) {
1773 for (i = 0; i <= kq->kq_knhashmask; i++) {
1774 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1775 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1776 kq->kq_state |= KQ_FLUXWAIT;
1777 msleep(kq, &kq->kq_lock, PSOCK,
1778 "kqclo2", 0);
1779 continue;
1780 }
1781 kn->kn_status |= KN_INFLUX;
1782 KQ_UNLOCK(kq);
1783 if (!(kn->kn_status & KN_DETACHED))
1784 kn->kn_fop->f_detach(kn);
1785 knote_drop(kn, td);
1786 KQ_LOCK(kq);
1787 }
1788 }
1789 }
1790
1791 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1792 kq->kq_state |= KQ_TASKDRAIN;
1793 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1794 }
1795
1796 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1797 selwakeuppri(&kq->kq_sel, PSOCK);
1798 if (!SEL_WAITING(&kq->kq_sel))
1799 kq->kq_state &= ~KQ_SEL;
1800 }
1801
1802 KQ_UNLOCK(kq);
1803
1804 /*
1805 * We could be called due to the knote_drop() doing fdrop(),
1806 * called from kqueue_register(). In this case the global
1807 * lock is owned, and filedesc sx is locked before, to not
1808 * take the sleepable lock after non-sleepable.
1809 */
1810 if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
1811 FILEDESC_XLOCK(fdp);
1812 filedesc_unlock = 1;
1813 } else
1814 filedesc_unlock = 0;
1815 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
1816 if (filedesc_unlock)
1817 FILEDESC_XUNLOCK(fdp);
1818
1819 seldrain(&kq->kq_sel);
1820 knlist_destroy(&kq->kq_sel.si_note);
1821 mtx_destroy(&kq->kq_lock);
1822 kq->kq_fdp = NULL;
1823
1824 if (kq->kq_knhash != NULL)
1825 free(kq->kq_knhash, M_KQUEUE);
1826 if (kq->kq_knlist != NULL)
1827 free(kq->kq_knlist, M_KQUEUE);
1828
1829 funsetown(&kq->kq_sigio);
1830 free(kq, M_KQUEUE);
1831 fp->f_data = NULL;
1832
1833 return (0);
1834 }
1835
1836 static void
1837 kqueue_wakeup(struct kqueue *kq)
1838 {
1839 KQ_OWNED(kq);
1840
1841 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1842 kq->kq_state &= ~KQ_SLEEP;
1843 wakeup(kq);
1844 }
1845 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1846 selwakeuppri(&kq->kq_sel, PSOCK);
1847 if (!SEL_WAITING(&kq->kq_sel))
1848 kq->kq_state &= ~KQ_SEL;
1849 }
1850 if (!knlist_empty(&kq->kq_sel.si_note))
1851 kqueue_schedtask(kq);
1852 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1853 pgsigio(&kq->kq_sigio, SIGIO, 0);
1854 }
1855 }
1856
1857 /*
1858 * Walk down a list of knotes, activating them if their event has triggered.
1859 *
1860 * There is a possibility to optimize in the case of one kq watching another.
1861 * Instead of scheduling a task to wake it up, you could pass enough state
1862 * down the chain to make up the parent kqueue. Make this code functional
1863 * first.
1864 */
1865 void
1866 knote(struct knlist *list, long hint, int lockflags)
1867 {
1868 struct kqueue *kq;
1869 struct knote *kn, *tkn;
1870 int error;
1871
1872 if (list == NULL)
1873 return;
1874
1875 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1876
1877 if ((lockflags & KNF_LISTLOCKED) == 0)
1878 list->kl_lock(list->kl_lockarg);
1879
1880 /*
1881 * If we unlock the list lock (and set KN_INFLUX), we can
1882 * eliminate the kqueue scheduling, but this will introduce
1883 * four lock/unlock's for each knote to test. Also, marker
1884 * would be needed to keep iteration position, since filters
1885 * or other threads could remove events.
1886 */
1887 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
1888 kq = kn->kn_kq;
1889 KQ_LOCK(kq);
1890 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
1891 /*
1892 * Do not process the influx notes, except for
1893 * the influx coming from the kq unlock in the
1894 * kqueue_scan(). In the later case, we do
1895 * not interfere with the scan, since the code
1896 * fragment in kqueue_scan() locks the knlist,
1897 * and cannot proceed until we finished.
1898 */
1899 KQ_UNLOCK(kq);
1900 } else if ((lockflags & KNF_NOKQLOCK) != 0) {
1901 kn->kn_status |= KN_INFLUX;
1902 KQ_UNLOCK(kq);
1903 error = kn->kn_fop->f_event(kn, hint);
1904 KQ_LOCK(kq);
1905 kn->kn_status &= ~KN_INFLUX;
1906 if (error)
1907 KNOTE_ACTIVATE(kn, 1);
1908 KQ_UNLOCK_FLUX(kq);
1909 } else {
1910 kn->kn_status |= KN_HASKQLOCK;
1911 if (kn->kn_fop->f_event(kn, hint))
1912 KNOTE_ACTIVATE(kn, 1);
1913 kn->kn_status &= ~KN_HASKQLOCK;
1914 KQ_UNLOCK(kq);
1915 }
1916 }
1917 if ((lockflags & KNF_LISTLOCKED) == 0)
1918 list->kl_unlock(list->kl_lockarg);
1919 }
1920
1921 /*
1922 * add a knote to a knlist
1923 */
1924 void
1925 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1926 {
1927 KNL_ASSERT_LOCK(knl, islocked);
1928 KQ_NOTOWNED(kn->kn_kq);
1929 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1930 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1931 if (!islocked)
1932 knl->kl_lock(knl->kl_lockarg);
1933 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1934 if (!islocked)
1935 knl->kl_unlock(knl->kl_lockarg);
1936 KQ_LOCK(kn->kn_kq);
1937 kn->kn_knlist = knl;
1938 kn->kn_status &= ~KN_DETACHED;
1939 KQ_UNLOCK(kn->kn_kq);
1940 }
1941
1942 static void
1943 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1944 {
1945 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1946 KNL_ASSERT_LOCK(knl, knlislocked);
1947 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1948 if (!kqislocked)
1949 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1950 ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1951 if (!knlislocked)
1952 knl->kl_lock(knl->kl_lockarg);
1953 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1954 kn->kn_knlist = NULL;
1955 if (!knlislocked)
1956 knl->kl_unlock(knl->kl_lockarg);
1957 if (!kqislocked)
1958 KQ_LOCK(kn->kn_kq);
1959 kn->kn_status |= KN_DETACHED;
1960 if (!kqislocked)
1961 KQ_UNLOCK(kn->kn_kq);
1962 }
1963
1964 /*
1965 * remove knote from the specified knlist
1966 */
1967 void
1968 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1969 {
1970
1971 knlist_remove_kq(knl, kn, islocked, 0);
1972 }
1973
1974 /*
1975 * remove knote from the specified knlist while in f_event handler.
1976 */
1977 void
1978 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1979 {
1980
1981 knlist_remove_kq(knl, kn, 1,
1982 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1983 }
1984
1985 int
1986 knlist_empty(struct knlist *knl)
1987 {
1988
1989 KNL_ASSERT_LOCKED(knl);
1990 return SLIST_EMPTY(&knl->kl_list);
1991 }
1992
1993 static struct mtx knlist_lock;
1994 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1995 MTX_DEF);
1996 static void knlist_mtx_lock(void *arg);
1997 static void knlist_mtx_unlock(void *arg);
1998
1999 static void
2000 knlist_mtx_lock(void *arg)
2001 {
2002
2003 mtx_lock((struct mtx *)arg);
2004 }
2005
2006 static void
2007 knlist_mtx_unlock(void *arg)
2008 {
2009
2010 mtx_unlock((struct mtx *)arg);
2011 }
2012
2013 static void
2014 knlist_mtx_assert_locked(void *arg)
2015 {
2016
2017 mtx_assert((struct mtx *)arg, MA_OWNED);
2018 }
2019
2020 static void
2021 knlist_mtx_assert_unlocked(void *arg)
2022 {
2023
2024 mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2025 }
2026
2027 static void
2028 knlist_rw_rlock(void *arg)
2029 {
2030
2031 rw_rlock((struct rwlock *)arg);
2032 }
2033
2034 static void
2035 knlist_rw_runlock(void *arg)
2036 {
2037
2038 rw_runlock((struct rwlock *)arg);
2039 }
2040
2041 static void
2042 knlist_rw_assert_locked(void *arg)
2043 {
2044
2045 rw_assert((struct rwlock *)arg, RA_LOCKED);
2046 }
2047
2048 static void
2049 knlist_rw_assert_unlocked(void *arg)
2050 {
2051
2052 rw_assert((struct rwlock *)arg, RA_UNLOCKED);
2053 }
2054
2055 void
2056 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2057 void (*kl_unlock)(void *),
2058 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
2059 {
2060
2061 if (lock == NULL)
2062 knl->kl_lockarg = &knlist_lock;
2063 else
2064 knl->kl_lockarg = lock;
2065
2066 if (kl_lock == NULL)
2067 knl->kl_lock = knlist_mtx_lock;
2068 else
2069 knl->kl_lock = kl_lock;
2070 if (kl_unlock == NULL)
2071 knl->kl_unlock = knlist_mtx_unlock;
2072 else
2073 knl->kl_unlock = kl_unlock;
2074 if (kl_assert_locked == NULL)
2075 knl->kl_assert_locked = knlist_mtx_assert_locked;
2076 else
2077 knl->kl_assert_locked = kl_assert_locked;
2078 if (kl_assert_unlocked == NULL)
2079 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
2080 else
2081 knl->kl_assert_unlocked = kl_assert_unlocked;
2082
2083 SLIST_INIT(&knl->kl_list);
2084 }
2085
2086 void
2087 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2088 {
2089
2090 knlist_init(knl, lock, NULL, NULL, NULL, NULL);
2091 }
2092
2093 void
2094 knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock)
2095 {
2096
2097 knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock,
2098 knlist_rw_assert_locked, knlist_rw_assert_unlocked);
2099 }
2100
2101 void
2102 knlist_destroy(struct knlist *knl)
2103 {
2104
2105 #ifdef INVARIANTS
2106 /*
2107 * if we run across this error, we need to find the offending
2108 * driver and have it call knlist_clear or knlist_delete.
2109 */
2110 if (!SLIST_EMPTY(&knl->kl_list))
2111 printf("WARNING: destroying knlist w/ knotes on it!\n");
2112 #endif
2113
2114 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
2115 SLIST_INIT(&knl->kl_list);
2116 }
2117
2118 /*
2119 * Even if we are locked, we may need to drop the lock to allow any influx
2120 * knotes time to "settle".
2121 */
2122 void
2123 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2124 {
2125 struct knote *kn, *kn2;
2126 struct kqueue *kq;
2127
2128 if (islocked)
2129 KNL_ASSERT_LOCKED(knl);
2130 else {
2131 KNL_ASSERT_UNLOCKED(knl);
2132 again: /* need to reacquire lock since we have dropped it */
2133 knl->kl_lock(knl->kl_lockarg);
2134 }
2135
2136 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2137 kq = kn->kn_kq;
2138 KQ_LOCK(kq);
2139 if ((kn->kn_status & KN_INFLUX)) {
2140 KQ_UNLOCK(kq);
2141 continue;
2142 }
2143 knlist_remove_kq(knl, kn, 1, 1);
2144 if (killkn) {
2145 kn->kn_status |= KN_INFLUX | KN_DETACHED;
2146 KQ_UNLOCK(kq);
2147 knote_drop(kn, td);
2148 } else {
2149 /* Make sure cleared knotes disappear soon */
2150 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2151 KQ_UNLOCK(kq);
2152 }
2153 kq = NULL;
2154 }
2155
2156 if (!SLIST_EMPTY(&knl->kl_list)) {
2157 /* there are still KN_INFLUX remaining */
2158 kn = SLIST_FIRST(&knl->kl_list);
2159 kq = kn->kn_kq;
2160 KQ_LOCK(kq);
2161 KASSERT(kn->kn_status & KN_INFLUX,
2162 ("knote removed w/o list lock"));
2163 knl->kl_unlock(knl->kl_lockarg);
2164 kq->kq_state |= KQ_FLUXWAIT;
2165 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2166 kq = NULL;
2167 goto again;
2168 }
2169
2170 if (islocked)
2171 KNL_ASSERT_LOCKED(knl);
2172 else {
2173 knl->kl_unlock(knl->kl_lockarg);
2174 KNL_ASSERT_UNLOCKED(knl);
2175 }
2176 }
2177
2178 /*
2179 * Remove all knotes referencing a specified fd must be called with FILEDESC
2180 * lock. This prevents a race where a new fd comes along and occupies the
2181 * entry and we attach a knote to the fd.
2182 */
2183 void
2184 knote_fdclose(struct thread *td, int fd)
2185 {
2186 struct filedesc *fdp = td->td_proc->p_fd;
2187 struct kqueue *kq;
2188 struct knote *kn;
2189 int influx;
2190
2191 FILEDESC_XLOCK_ASSERT(fdp);
2192
2193 /*
2194 * We shouldn't have to worry about new kevents appearing on fd
2195 * since filedesc is locked.
2196 */
2197 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2198 KQ_LOCK(kq);
2199
2200 again:
2201 influx = 0;
2202 while (kq->kq_knlistsize > fd &&
2203 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2204 if (kn->kn_status & KN_INFLUX) {
2205 /* someone else might be waiting on our knote */
2206 if (influx)
2207 wakeup(kq);
2208 kq->kq_state |= KQ_FLUXWAIT;
2209 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2210 goto again;
2211 }
2212 kn->kn_status |= KN_INFLUX;
2213 KQ_UNLOCK(kq);
2214 if (!(kn->kn_status & KN_DETACHED))
2215 kn->kn_fop->f_detach(kn);
2216 knote_drop(kn, td);
2217 influx = 1;
2218 KQ_LOCK(kq);
2219 }
2220 KQ_UNLOCK_FLUX(kq);
2221 }
2222 }
2223
2224 static int
2225 knote_attach(struct knote *kn, struct kqueue *kq)
2226 {
2227 struct klist *list;
2228
2229 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2230 KQ_OWNED(kq);
2231
2232 if (kn->kn_fop->f_isfd) {
2233 if (kn->kn_id >= kq->kq_knlistsize)
2234 return ENOMEM;
2235 list = &kq->kq_knlist[kn->kn_id];
2236 } else {
2237 if (kq->kq_knhash == NULL)
2238 return ENOMEM;
2239 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2240 }
2241
2242 SLIST_INSERT_HEAD(list, kn, kn_link);
2243
2244 return 0;
2245 }
2246
2247 /*
2248 * knote must already have been detached using the f_detach method.
2249 * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2250 * to prevent other removal.
2251 */
2252 static void
2253 knote_drop(struct knote *kn, struct thread *td)
2254 {
2255 struct kqueue *kq;
2256 struct klist *list;
2257
2258 kq = kn->kn_kq;
2259
2260 KQ_NOTOWNED(kq);
2261 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2262 ("knote_drop called without KN_INFLUX set in kn_status"));
2263
2264 KQ_LOCK(kq);
2265 if (kn->kn_fop->f_isfd)
2266 list = &kq->kq_knlist[kn->kn_id];
2267 else
2268 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2269
2270 if (!SLIST_EMPTY(list))
2271 SLIST_REMOVE(list, kn, knote, kn_link);
2272 if (kn->kn_status & KN_QUEUED)
2273 knote_dequeue(kn);
2274 KQ_UNLOCK_FLUX(kq);
2275
2276 if (kn->kn_fop->f_isfd) {
2277 fdrop(kn->kn_fp, td);
2278 kn->kn_fp = NULL;
2279 }
2280 kqueue_fo_release(kn->kn_kevent.filter);
2281 kn->kn_fop = NULL;
2282 knote_free(kn);
2283 }
2284
2285 static void
2286 knote_enqueue(struct knote *kn)
2287 {
2288 struct kqueue *kq = kn->kn_kq;
2289
2290 KQ_OWNED(kn->kn_kq);
2291 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2292
2293 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2294 kn->kn_status |= KN_QUEUED;
2295 kq->kq_count++;
2296 kqueue_wakeup(kq);
2297 }
2298
2299 static void
2300 knote_dequeue(struct knote *kn)
2301 {
2302 struct kqueue *kq = kn->kn_kq;
2303
2304 KQ_OWNED(kn->kn_kq);
2305 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2306
2307 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2308 kn->kn_status &= ~KN_QUEUED;
2309 kq->kq_count--;
2310 }
2311
2312 static void
2313 knote_init(void)
2314 {
2315
2316 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2317 NULL, NULL, UMA_ALIGN_PTR, 0);
2318 }
2319 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2320
2321 static struct knote *
2322 knote_alloc(int waitok)
2323 {
2324 return ((struct knote *)uma_zalloc(knote_zone,
2325 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2326 }
2327
2328 static void
2329 knote_free(struct knote *kn)
2330 {
2331 if (kn != NULL)
2332 uma_zfree(knote_zone, kn);
2333 }
2334
2335 /*
2336 * Register the kev w/ the kq specified by fd.
2337 */
2338 int
2339 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2340 {
2341 struct kqueue *kq;
2342 struct file *fp;
2343 cap_rights_t rights;
2344 int error;
2345
2346 error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp);
2347 if (error != 0)
2348 return (error);
2349 if ((error = kqueue_acquire(fp, &kq)) != 0)
2350 goto noacquire;
2351
2352 error = kqueue_register(kq, kev, td, waitok);
2353
2354 kqueue_release(kq, 0);
2355
2356 noacquire:
2357 fdrop(fp, td);
2358
2359 return error;
2360 }
Cache object: 1ece525eb1adff80b8246bb6c2d76fc2
|