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/11.1/sys/kern/kern_event.c 320290 2017-06-23 18:57:56Z kib $");
31
32 #include "opt_ktrace.h"
33 #include "opt_kqueue.h"
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/capsicum.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/rwlock.h>
42 #include <sys/proc.h>
43 #include <sys/malloc.h>
44 #include <sys/unistd.h>
45 #include <sys/file.h>
46 #include <sys/filedesc.h>
47 #include <sys/filio.h>
48 #include <sys/fcntl.h>
49 #include <sys/kthread.h>
50 #include <sys/selinfo.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/resourcevar.h>
57 #include <sys/sigio.h>
58 #include <sys/signalvar.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysproto.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/taskqueue.h>
66 #include <sys/uio.h>
67 #include <sys/user.h>
68 #ifdef KTRACE
69 #include <sys/ktrace.h>
70 #endif
71 #include <machine/atomic.h>
72
73 #include <vm/uma.h>
74
75 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
76
77 /*
78 * This lock is used if multiple kq locks are required. This possibly
79 * should be made into a per proc lock.
80 */
81 static struct mtx kq_global;
82 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
83 #define KQ_GLOBAL_LOCK(lck, haslck) do { \
84 if (!haslck) \
85 mtx_lock(lck); \
86 haslck = 1; \
87 } while (0)
88 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \
89 if (haslck) \
90 mtx_unlock(lck); \
91 haslck = 0; \
92 } while (0)
93
94 TASKQUEUE_DEFINE_THREAD(kqueue_ctx);
95
96 static int kevent_copyout(void *arg, struct kevent *kevp, int count);
97 static int kevent_copyin(void *arg, struct kevent *kevp, int count);
98 static int kqueue_register(struct kqueue *kq, struct kevent *kev,
99 struct thread *td, int waitok);
100 static int kqueue_acquire(struct file *fp, struct kqueue **kqp);
101 static void kqueue_release(struct kqueue *kq, int locked);
102 static void kqueue_destroy(struct kqueue *kq);
103 static void kqueue_drain(struct kqueue *kq, struct thread *td);
104 static int kqueue_expand(struct kqueue *kq, struct filterops *fops,
105 uintptr_t ident, int waitok);
106 static void kqueue_task(void *arg, int pending);
107 static int kqueue_scan(struct kqueue *kq, int maxevents,
108 struct kevent_copyops *k_ops,
109 const struct timespec *timeout,
110 struct kevent *keva, struct thread *td);
111 static void kqueue_wakeup(struct kqueue *kq);
112 static struct filterops *kqueue_fo_find(int filt);
113 static void kqueue_fo_release(int filt);
114
115 static fo_ioctl_t kqueue_ioctl;
116 static fo_poll_t kqueue_poll;
117 static fo_kqfilter_t kqueue_kqfilter;
118 static fo_stat_t kqueue_stat;
119 static fo_close_t kqueue_close;
120 static fo_fill_kinfo_t kqueue_fill_kinfo;
121
122 static struct fileops kqueueops = {
123 .fo_read = invfo_rdwr,
124 .fo_write = invfo_rdwr,
125 .fo_truncate = invfo_truncate,
126 .fo_ioctl = kqueue_ioctl,
127 .fo_poll = kqueue_poll,
128 .fo_kqfilter = kqueue_kqfilter,
129 .fo_stat = kqueue_stat,
130 .fo_close = kqueue_close,
131 .fo_chmod = invfo_chmod,
132 .fo_chown = invfo_chown,
133 .fo_sendfile = invfo_sendfile,
134 .fo_fill_kinfo = kqueue_fill_kinfo,
135 };
136
137 static int knote_attach(struct knote *kn, struct kqueue *kq);
138 static void knote_drop(struct knote *kn, struct thread *td);
139 static void knote_enqueue(struct knote *kn);
140 static void knote_dequeue(struct knote *kn);
141 static void knote_init(void);
142 static struct knote *knote_alloc(int waitok);
143 static void knote_free(struct knote *kn);
144
145 static void filt_kqdetach(struct knote *kn);
146 static int filt_kqueue(struct knote *kn, long hint);
147 static int filt_procattach(struct knote *kn);
148 static void filt_procdetach(struct knote *kn);
149 static int filt_proc(struct knote *kn, long hint);
150 static int filt_fileattach(struct knote *kn);
151 static void filt_timerexpire(void *knx);
152 static int filt_timerattach(struct knote *kn);
153 static void filt_timerdetach(struct knote *kn);
154 static int filt_timer(struct knote *kn, long hint);
155 static int filt_userattach(struct knote *kn);
156 static void filt_userdetach(struct knote *kn);
157 static int filt_user(struct knote *kn, long hint);
158 static void filt_usertouch(struct knote *kn, struct kevent *kev,
159 u_long type);
160
161 static struct filterops file_filtops = {
162 .f_isfd = 1,
163 .f_attach = filt_fileattach,
164 };
165 static struct filterops kqread_filtops = {
166 .f_isfd = 1,
167 .f_detach = filt_kqdetach,
168 .f_event = filt_kqueue,
169 };
170 /* XXX - move to kern_proc.c? */
171 static struct filterops proc_filtops = {
172 .f_isfd = 0,
173 .f_attach = filt_procattach,
174 .f_detach = filt_procdetach,
175 .f_event = filt_proc,
176 };
177 static struct filterops timer_filtops = {
178 .f_isfd = 0,
179 .f_attach = filt_timerattach,
180 .f_detach = filt_timerdetach,
181 .f_event = filt_timer,
182 };
183 static struct filterops user_filtops = {
184 .f_attach = filt_userattach,
185 .f_detach = filt_userdetach,
186 .f_event = filt_user,
187 .f_touch = filt_usertouch,
188 };
189
190 static uma_zone_t knote_zone;
191 static unsigned int kq_ncallouts = 0;
192 static unsigned int kq_calloutmax = 4 * 1024;
193 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
194 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
195
196 /* XXX - ensure not KN_INFLUX?? */
197 #define KNOTE_ACTIVATE(kn, islock) do { \
198 if ((islock)) \
199 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \
200 else \
201 KQ_LOCK((kn)->kn_kq); \
202 (kn)->kn_status |= KN_ACTIVE; \
203 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
204 knote_enqueue((kn)); \
205 if (!(islock)) \
206 KQ_UNLOCK((kn)->kn_kq); \
207 } while(0)
208 #define KQ_LOCK(kq) do { \
209 mtx_lock(&(kq)->kq_lock); \
210 } while (0)
211 #define KQ_FLUX_WAKEUP(kq) do { \
212 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
213 (kq)->kq_state &= ~KQ_FLUXWAIT; \
214 wakeup((kq)); \
215 } \
216 } while (0)
217 #define KQ_UNLOCK_FLUX(kq) do { \
218 KQ_FLUX_WAKEUP(kq); \
219 mtx_unlock(&(kq)->kq_lock); \
220 } while (0)
221 #define KQ_UNLOCK(kq) do { \
222 mtx_unlock(&(kq)->kq_lock); \
223 } while (0)
224 #define KQ_OWNED(kq) do { \
225 mtx_assert(&(kq)->kq_lock, MA_OWNED); \
226 } while (0)
227 #define KQ_NOTOWNED(kq) do { \
228 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \
229 } while (0)
230
231 static struct knlist *
232 kn_list_lock(struct knote *kn)
233 {
234 struct knlist *knl;
235
236 knl = kn->kn_knlist;
237 if (knl != NULL)
238 knl->kl_lock(knl->kl_lockarg);
239 return (knl);
240 }
241
242 static void
243 kn_list_unlock(struct knlist *knl)
244 {
245 bool do_free;
246
247 if (knl == NULL)
248 return;
249 do_free = knl->kl_autodestroy && knlist_empty(knl);
250 knl->kl_unlock(knl->kl_lockarg);
251 if (do_free) {
252 knlist_destroy(knl);
253 free(knl, M_KQUEUE);
254 }
255 }
256
257 #define KNL_ASSERT_LOCK(knl, islocked) do { \
258 if (islocked) \
259 KNL_ASSERT_LOCKED(knl); \
260 else \
261 KNL_ASSERT_UNLOCKED(knl); \
262 } while (0)
263 #ifdef INVARIANTS
264 #define KNL_ASSERT_LOCKED(knl) do { \
265 knl->kl_assert_locked((knl)->kl_lockarg); \
266 } while (0)
267 #define KNL_ASSERT_UNLOCKED(knl) do { \
268 knl->kl_assert_unlocked((knl)->kl_lockarg); \
269 } while (0)
270 #else /* !INVARIANTS */
271 #define KNL_ASSERT_LOCKED(knl) do {} while(0)
272 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0)
273 #endif /* INVARIANTS */
274
275 #ifndef KN_HASHSIZE
276 #define KN_HASHSIZE 64 /* XXX should be tunable */
277 #endif
278
279 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
280
281 static int
282 filt_nullattach(struct knote *kn)
283 {
284
285 return (ENXIO);
286 };
287
288 struct filterops null_filtops = {
289 .f_isfd = 0,
290 .f_attach = filt_nullattach,
291 };
292
293 /* XXX - make SYSINIT to add these, and move into respective modules. */
294 extern struct filterops sig_filtops;
295 extern struct filterops fs_filtops;
296
297 /*
298 * Table for for all system-defined filters.
299 */
300 static struct mtx filterops_lock;
301 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
302 MTX_DEF);
303 static struct {
304 struct filterops *for_fop;
305 int for_nolock;
306 int for_refcnt;
307 } sysfilt_ops[EVFILT_SYSCOUNT] = {
308 { &file_filtops, 1 }, /* EVFILT_READ */
309 { &file_filtops, 1 }, /* EVFILT_WRITE */
310 { &null_filtops }, /* EVFILT_AIO */
311 { &file_filtops, 1 }, /* EVFILT_VNODE */
312 { &proc_filtops, 1 }, /* EVFILT_PROC */
313 { &sig_filtops, 1 }, /* EVFILT_SIGNAL */
314 { &timer_filtops, 1 }, /* EVFILT_TIMER */
315 { &file_filtops, 1 }, /* EVFILT_PROCDESC */
316 { &fs_filtops, 1 }, /* EVFILT_FS */
317 { &null_filtops }, /* EVFILT_LIO */
318 { &user_filtops, 1 }, /* EVFILT_USER */
319 { &null_filtops }, /* EVFILT_SENDFILE */
320 };
321
322 /*
323 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
324 * method.
325 */
326 static int
327 filt_fileattach(struct knote *kn)
328 {
329
330 return (fo_kqfilter(kn->kn_fp, kn));
331 }
332
333 /*ARGSUSED*/
334 static int
335 kqueue_kqfilter(struct file *fp, struct knote *kn)
336 {
337 struct kqueue *kq = kn->kn_fp->f_data;
338
339 if (kn->kn_filter != EVFILT_READ)
340 return (EINVAL);
341
342 kn->kn_status |= KN_KQUEUE;
343 kn->kn_fop = &kqread_filtops;
344 knlist_add(&kq->kq_sel.si_note, kn, 0);
345
346 return (0);
347 }
348
349 static void
350 filt_kqdetach(struct knote *kn)
351 {
352 struct kqueue *kq = kn->kn_fp->f_data;
353
354 knlist_remove(&kq->kq_sel.si_note, kn, 0);
355 }
356
357 /*ARGSUSED*/
358 static int
359 filt_kqueue(struct knote *kn, long hint)
360 {
361 struct kqueue *kq = kn->kn_fp->f_data;
362
363 kn->kn_data = kq->kq_count;
364 return (kn->kn_data > 0);
365 }
366
367 /* XXX - move to kern_proc.c? */
368 static int
369 filt_procattach(struct knote *kn)
370 {
371 struct proc *p;
372 int error;
373 bool exiting, immediate;
374
375 exiting = immediate = false;
376 p = pfind(kn->kn_id);
377 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
378 p = zpfind(kn->kn_id);
379 exiting = true;
380 } else if (p != NULL && (p->p_flag & P_WEXIT)) {
381 exiting = true;
382 }
383
384 if (p == NULL)
385 return (ESRCH);
386 if ((error = p_cansee(curthread, p))) {
387 PROC_UNLOCK(p);
388 return (error);
389 }
390
391 kn->kn_ptr.p_proc = p;
392 kn->kn_flags |= EV_CLEAR; /* automatically set */
393
394 /*
395 * Internal flag indicating registration done by kernel for the
396 * purposes of getting a NOTE_CHILD notification.
397 */
398 if (kn->kn_flags & EV_FLAG2) {
399 kn->kn_flags &= ~EV_FLAG2;
400 kn->kn_data = kn->kn_sdata; /* ppid */
401 kn->kn_fflags = NOTE_CHILD;
402 kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
403 immediate = true; /* Force immediate activation of child note. */
404 }
405 /*
406 * Internal flag indicating registration done by kernel (for other than
407 * NOTE_CHILD).
408 */
409 if (kn->kn_flags & EV_FLAG1) {
410 kn->kn_flags &= ~EV_FLAG1;
411 }
412
413 knlist_add(p->p_klist, kn, 1);
414
415 /*
416 * Immediately activate any child notes or, in the case of a zombie
417 * target process, exit notes. The latter is necessary to handle the
418 * case where the target process, e.g. a child, dies before the kevent
419 * is registered.
420 */
421 if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
422 KNOTE_ACTIVATE(kn, 0);
423
424 PROC_UNLOCK(p);
425
426 return (0);
427 }
428
429 /*
430 * The knote may be attached to a different process, which may exit,
431 * leaving nothing for the knote to be attached to. So when the process
432 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
433 * it will be deleted when read out. However, as part of the knote deletion,
434 * this routine is called, so a check is needed to avoid actually performing
435 * a detach, because the original process does not exist any more.
436 */
437 /* XXX - move to kern_proc.c? */
438 static void
439 filt_procdetach(struct knote *kn)
440 {
441
442 knlist_remove(kn->kn_knlist, kn, 0);
443 kn->kn_ptr.p_proc = NULL;
444 }
445
446 /* XXX - move to kern_proc.c? */
447 static int
448 filt_proc(struct knote *kn, long hint)
449 {
450 struct proc *p;
451 u_int event;
452
453 p = kn->kn_ptr.p_proc;
454 if (p == NULL) /* already activated, from attach filter */
455 return (0);
456
457 /* Mask off extra data. */
458 event = (u_int)hint & NOTE_PCTRLMASK;
459
460 /* If the user is interested in this event, record it. */
461 if (kn->kn_sfflags & event)
462 kn->kn_fflags |= event;
463
464 /* Process is gone, so flag the event as finished. */
465 if (event == NOTE_EXIT) {
466 kn->kn_flags |= EV_EOF | EV_ONESHOT;
467 kn->kn_ptr.p_proc = NULL;
468 if (kn->kn_fflags & NOTE_EXIT)
469 kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
470 if (kn->kn_fflags == 0)
471 kn->kn_flags |= EV_DROP;
472 return (1);
473 }
474
475 return (kn->kn_fflags != 0);
476 }
477
478 /*
479 * Called when the process forked. It mostly does the same as the
480 * knote(), activating all knotes registered to be activated when the
481 * process forked. Additionally, for each knote attached to the
482 * parent, check whether user wants to track the new process. If so
483 * attach a new knote to it, and immediately report an event with the
484 * child's pid.
485 */
486 void
487 knote_fork(struct knlist *list, int pid)
488 {
489 struct kqueue *kq;
490 struct knote *kn;
491 struct kevent kev;
492 int error;
493
494 if (list == NULL)
495 return;
496 list->kl_lock(list->kl_lockarg);
497
498 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
499 kq = kn->kn_kq;
500 KQ_LOCK(kq);
501 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
502 KQ_UNLOCK(kq);
503 continue;
504 }
505
506 /*
507 * The same as knote(), activate the event.
508 */
509 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
510 kn->kn_status |= KN_HASKQLOCK;
511 if (kn->kn_fop->f_event(kn, NOTE_FORK))
512 KNOTE_ACTIVATE(kn, 1);
513 kn->kn_status &= ~KN_HASKQLOCK;
514 KQ_UNLOCK(kq);
515 continue;
516 }
517
518 /*
519 * The NOTE_TRACK case. In addition to the activation
520 * of the event, we need to register new events to
521 * track the child. Drop the locks in preparation for
522 * the call to kqueue_register().
523 */
524 kn->kn_status |= KN_INFLUX;
525 KQ_UNLOCK(kq);
526 list->kl_unlock(list->kl_lockarg);
527
528 /*
529 * Activate existing knote and register tracking knotes with
530 * new process.
531 *
532 * First register a knote to get just the child notice. This
533 * must be a separate note from a potential NOTE_EXIT
534 * notification since both NOTE_CHILD and NOTE_EXIT are defined
535 * to use the data field (in conflicting ways).
536 */
537 kev.ident = pid;
538 kev.filter = kn->kn_filter;
539 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
540 EV_FLAG2;
541 kev.fflags = kn->kn_sfflags;
542 kev.data = kn->kn_id; /* parent */
543 kev.udata = kn->kn_kevent.udata;/* preserve udata */
544 error = kqueue_register(kq, &kev, NULL, 0);
545 if (error)
546 kn->kn_fflags |= NOTE_TRACKERR;
547
548 /*
549 * Then register another knote to track other potential events
550 * from the new process.
551 */
552 kev.ident = pid;
553 kev.filter = kn->kn_filter;
554 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
555 kev.fflags = kn->kn_sfflags;
556 kev.data = kn->kn_id; /* parent */
557 kev.udata = kn->kn_kevent.udata;/* preserve udata */
558 error = kqueue_register(kq, &kev, NULL, 0);
559 if (error)
560 kn->kn_fflags |= NOTE_TRACKERR;
561 if (kn->kn_fop->f_event(kn, NOTE_FORK))
562 KNOTE_ACTIVATE(kn, 0);
563 KQ_LOCK(kq);
564 kn->kn_status &= ~KN_INFLUX;
565 KQ_UNLOCK_FLUX(kq);
566 list->kl_lock(list->kl_lockarg);
567 }
568 list->kl_unlock(list->kl_lockarg);
569 }
570
571 /*
572 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
573 * interval timer support code.
574 */
575
576 #define NOTE_TIMER_PRECMASK \
577 (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)
578
579 static sbintime_t
580 timer2sbintime(intptr_t data, int flags)
581 {
582 int64_t secs;
583
584 /*
585 * Macros for converting to the fractional second portion of an
586 * sbintime_t using 64bit multiplication to improve precision.
587 */
588 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
589 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
590 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
591 switch (flags & NOTE_TIMER_PRECMASK) {
592 case NOTE_SECONDS:
593 #ifdef __LP64__
594 if (data > (SBT_MAX / SBT_1S))
595 return (SBT_MAX);
596 #endif
597 return ((sbintime_t)data << 32);
598 case NOTE_MSECONDS: /* FALLTHROUGH */
599 case 0:
600 if (data >= 1000) {
601 secs = data / 1000;
602 #ifdef __LP64__
603 if (secs > (SBT_MAX / SBT_1S))
604 return (SBT_MAX);
605 #endif
606 return (secs << 32 | MS_TO_SBT(data % 1000));
607 }
608 return (MS_TO_SBT(data));
609 case NOTE_USECONDS:
610 if (data >= 1000000) {
611 secs = data / 1000000;
612 #ifdef __LP64__
613 if (secs > (SBT_MAX / SBT_1S))
614 return (SBT_MAX);
615 #endif
616 return (secs << 32 | US_TO_SBT(data % 1000000));
617 }
618 return (US_TO_SBT(data));
619 case NOTE_NSECONDS:
620 if (data >= 1000000000) {
621 secs = data / 1000000000;
622 #ifdef __LP64__
623 if (secs > (SBT_MAX / SBT_1S))
624 return (SBT_MAX);
625 #endif
626 return (secs << 32 | US_TO_SBT(data % 1000000000));
627 }
628 return (NS_TO_SBT(data));
629 default:
630 break;
631 }
632 return (-1);
633 }
634
635 struct kq_timer_cb_data {
636 struct callout c;
637 sbintime_t next; /* next timer event fires at */
638 sbintime_t to; /* precalculated timer period */
639 };
640
641 static void
642 filt_timerexpire(void *knx)
643 {
644 struct knote *kn;
645 struct kq_timer_cb_data *kc;
646
647 kn = knx;
648 kn->kn_data++;
649 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
650
651 if ((kn->kn_flags & EV_ONESHOT) != 0)
652 return;
653
654 kc = kn->kn_ptr.p_v;
655 kc->next += kc->to;
656 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn,
657 PCPU_GET(cpuid), C_ABSOLUTE);
658 }
659
660 /*
661 * data contains amount of time to sleep
662 */
663 static int
664 filt_timerattach(struct knote *kn)
665 {
666 struct kq_timer_cb_data *kc;
667 sbintime_t to;
668 unsigned int ncallouts;
669
670 if (kn->kn_sdata < 0)
671 return (EINVAL);
672 if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
673 kn->kn_sdata = 1;
674 /* Only precision unit are supported in flags so far */
675 if ((kn->kn_sfflags & ~NOTE_TIMER_PRECMASK) != 0)
676 return (EINVAL);
677
678 to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
679 if (to < 0)
680 return (EINVAL);
681
682 do {
683 ncallouts = kq_ncallouts;
684 if (ncallouts >= kq_calloutmax)
685 return (ENOMEM);
686 } while (!atomic_cmpset_int(&kq_ncallouts, ncallouts, ncallouts + 1));
687
688 kn->kn_flags |= EV_CLEAR; /* automatically set */
689 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */
690 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
691 callout_init(&kc->c, 1);
692 kc->next = to + sbinuptime();
693 kc->to = to;
694 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn,
695 PCPU_GET(cpuid), C_ABSOLUTE);
696
697 return (0);
698 }
699
700 static void
701 filt_timerdetach(struct knote *kn)
702 {
703 struct kq_timer_cb_data *kc;
704 unsigned int old;
705
706 kc = kn->kn_ptr.p_v;
707 callout_drain(&kc->c);
708 free(kc, M_KQUEUE);
709 old = atomic_fetchadd_int(&kq_ncallouts, -1);
710 KASSERT(old > 0, ("Number of callouts cannot become negative"));
711 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */
712 }
713
714 static int
715 filt_timer(struct knote *kn, long hint)
716 {
717
718 return (kn->kn_data != 0);
719 }
720
721 static int
722 filt_userattach(struct knote *kn)
723 {
724
725 /*
726 * EVFILT_USER knotes are not attached to anything in the kernel.
727 */
728 kn->kn_hook = NULL;
729 if (kn->kn_fflags & NOTE_TRIGGER)
730 kn->kn_hookid = 1;
731 else
732 kn->kn_hookid = 0;
733 return (0);
734 }
735
736 static void
737 filt_userdetach(__unused struct knote *kn)
738 {
739
740 /*
741 * EVFILT_USER knotes are not attached to anything in the kernel.
742 */
743 }
744
745 static int
746 filt_user(struct knote *kn, __unused long hint)
747 {
748
749 return (kn->kn_hookid);
750 }
751
752 static void
753 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
754 {
755 u_int ffctrl;
756
757 switch (type) {
758 case EVENT_REGISTER:
759 if (kev->fflags & NOTE_TRIGGER)
760 kn->kn_hookid = 1;
761
762 ffctrl = kev->fflags & NOTE_FFCTRLMASK;
763 kev->fflags &= NOTE_FFLAGSMASK;
764 switch (ffctrl) {
765 case NOTE_FFNOP:
766 break;
767
768 case NOTE_FFAND:
769 kn->kn_sfflags &= kev->fflags;
770 break;
771
772 case NOTE_FFOR:
773 kn->kn_sfflags |= kev->fflags;
774 break;
775
776 case NOTE_FFCOPY:
777 kn->kn_sfflags = kev->fflags;
778 break;
779
780 default:
781 /* XXX Return error? */
782 break;
783 }
784 kn->kn_sdata = kev->data;
785 if (kev->flags & EV_CLEAR) {
786 kn->kn_hookid = 0;
787 kn->kn_data = 0;
788 kn->kn_fflags = 0;
789 }
790 break;
791
792 case EVENT_PROCESS:
793 *kev = kn->kn_kevent;
794 kev->fflags = kn->kn_sfflags;
795 kev->data = kn->kn_sdata;
796 if (kn->kn_flags & EV_CLEAR) {
797 kn->kn_hookid = 0;
798 kn->kn_data = 0;
799 kn->kn_fflags = 0;
800 }
801 break;
802
803 default:
804 panic("filt_usertouch() - invalid type (%ld)", type);
805 break;
806 }
807 }
808
809 int
810 sys_kqueue(struct thread *td, struct kqueue_args *uap)
811 {
812
813 return (kern_kqueue(td, 0, NULL));
814 }
815
816 static void
817 kqueue_init(struct kqueue *kq)
818 {
819
820 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
821 TAILQ_INIT(&kq->kq_head);
822 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
823 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
824 }
825
826 int
827 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps)
828 {
829 struct filedesc *fdp;
830 struct kqueue *kq;
831 struct file *fp;
832 struct ucred *cred;
833 int fd, error;
834
835 fdp = td->td_proc->p_fd;
836 cred = td->td_ucred;
837 if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
838 return (ENOMEM);
839
840 error = falloc_caps(td, &fp, &fd, flags, fcaps);
841 if (error != 0) {
842 chgkqcnt(cred->cr_ruidinfo, -1, 0);
843 return (error);
844 }
845
846 /* An extra reference on `fp' has been held for us by falloc(). */
847 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
848 kqueue_init(kq);
849 kq->kq_fdp = fdp;
850 kq->kq_cred = crhold(cred);
851
852 FILEDESC_XLOCK(fdp);
853 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
854 FILEDESC_XUNLOCK(fdp);
855
856 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
857 fdrop(fp, td);
858
859 td->td_retval[0] = fd;
860 return (0);
861 }
862
863 #ifdef KTRACE
864 static size_t
865 kev_iovlen(int n, u_int kgio)
866 {
867
868 if (n < 0 || n >= kgio / sizeof(struct kevent))
869 return (kgio);
870 return (n * sizeof(struct kevent));
871 }
872 #endif
873
874 #ifndef _SYS_SYSPROTO_H_
875 struct kevent_args {
876 int fd;
877 const struct kevent *changelist;
878 int nchanges;
879 struct kevent *eventlist;
880 int nevents;
881 const struct timespec *timeout;
882 };
883 #endif
884 int
885 sys_kevent(struct thread *td, struct kevent_args *uap)
886 {
887 struct timespec ts, *tsp;
888 struct kevent_copyops k_ops = {
889 .arg = uap,
890 .k_copyout = kevent_copyout,
891 .k_copyin = kevent_copyin,
892 };
893 int error;
894 #ifdef KTRACE
895 struct uio ktruio;
896 struct iovec ktriov;
897 struct uio *ktruioin = NULL;
898 struct uio *ktruioout = NULL;
899 u_int kgio;
900 #endif
901
902 if (uap->timeout != NULL) {
903 error = copyin(uap->timeout, &ts, sizeof(ts));
904 if (error)
905 return (error);
906 tsp = &ts;
907 } else
908 tsp = NULL;
909
910 #ifdef KTRACE
911 if (KTRPOINT(td, KTR_GENIO)) {
912 kgio = ktr_geniosize;
913 ktriov.iov_base = uap->changelist;
914 ktriov.iov_len = kev_iovlen(uap->nchanges, kgio);
915 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
916 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
917 .uio_td = td };
918 ktruioin = cloneuio(&ktruio);
919 ktriov.iov_base = uap->eventlist;
920 ktriov.iov_len = kev_iovlen(uap->nevents, kgio);
921 ktriov.iov_len = uap->nevents * sizeof(struct kevent);
922 ktruioout = cloneuio(&ktruio);
923 }
924 #endif
925
926 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
927 &k_ops, tsp);
928
929 #ifdef KTRACE
930 if (ktruioin != NULL) {
931 ktruioin->uio_resid = kev_iovlen(uap->nchanges, kgio);
932 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
933 ktruioout->uio_resid = kev_iovlen(td->td_retval[0], kgio);
934 ktrgenio(uap->fd, UIO_READ, ktruioout, error);
935 }
936 #endif
937
938 return (error);
939 }
940
941 /*
942 * Copy 'count' items into the destination list pointed to by uap->eventlist.
943 */
944 static int
945 kevent_copyout(void *arg, struct kevent *kevp, int count)
946 {
947 struct kevent_args *uap;
948 int error;
949
950 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
951 uap = (struct kevent_args *)arg;
952
953 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
954 if (error == 0)
955 uap->eventlist += count;
956 return (error);
957 }
958
959 /*
960 * Copy 'count' items from the list pointed to by uap->changelist.
961 */
962 static int
963 kevent_copyin(void *arg, struct kevent *kevp, int count)
964 {
965 struct kevent_args *uap;
966 int error;
967
968 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
969 uap = (struct kevent_args *)arg;
970
971 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
972 if (error == 0)
973 uap->changelist += count;
974 return (error);
975 }
976
977 int
978 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
979 struct kevent_copyops *k_ops, const struct timespec *timeout)
980 {
981 cap_rights_t rights;
982 struct file *fp;
983 int error;
984
985 cap_rights_init(&rights);
986 if (nchanges > 0)
987 cap_rights_set(&rights, CAP_KQUEUE_CHANGE);
988 if (nevents > 0)
989 cap_rights_set(&rights, CAP_KQUEUE_EVENT);
990 error = fget(td, fd, &rights, &fp);
991 if (error != 0)
992 return (error);
993
994 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
995 fdrop(fp, td);
996
997 return (error);
998 }
999
1000 static int
1001 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
1002 struct kevent_copyops *k_ops, const struct timespec *timeout)
1003 {
1004 struct kevent keva[KQ_NEVENTS];
1005 struct kevent *kevp, *changes;
1006 int i, n, nerrors, error;
1007
1008 nerrors = 0;
1009 while (nchanges > 0) {
1010 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
1011 error = k_ops->k_copyin(k_ops->arg, keva, n);
1012 if (error)
1013 return (error);
1014 changes = keva;
1015 for (i = 0; i < n; i++) {
1016 kevp = &changes[i];
1017 if (!kevp->filter)
1018 continue;
1019 kevp->flags &= ~EV_SYSFLAGS;
1020 error = kqueue_register(kq, kevp, td, 1);
1021 if (error || (kevp->flags & EV_RECEIPT)) {
1022 if (nevents == 0)
1023 return (error);
1024 kevp->flags = EV_ERROR;
1025 kevp->data = error;
1026 (void)k_ops->k_copyout(k_ops->arg, kevp, 1);
1027 nevents--;
1028 nerrors++;
1029 }
1030 }
1031 nchanges -= n;
1032 }
1033 if (nerrors) {
1034 td->td_retval[0] = nerrors;
1035 return (0);
1036 }
1037
1038 return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
1039 }
1040
1041 int
1042 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
1043 struct kevent_copyops *k_ops, const struct timespec *timeout)
1044 {
1045 struct kqueue *kq;
1046 int error;
1047
1048 error = kqueue_acquire(fp, &kq);
1049 if (error != 0)
1050 return (error);
1051 error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
1052 kqueue_release(kq, 0);
1053 return (error);
1054 }
1055
1056 /*
1057 * Performs a kevent() call on a temporarily created kqueue. This can be
1058 * used to perform one-shot polling, similar to poll() and select().
1059 */
1060 int
1061 kern_kevent_anonymous(struct thread *td, int nevents,
1062 struct kevent_copyops *k_ops)
1063 {
1064 struct kqueue kq = {};
1065 int error;
1066
1067 kqueue_init(&kq);
1068 kq.kq_refcnt = 1;
1069 error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
1070 kqueue_drain(&kq, td);
1071 kqueue_destroy(&kq);
1072 return (error);
1073 }
1074
1075 int
1076 kqueue_add_filteropts(int filt, struct filterops *filtops)
1077 {
1078 int error;
1079
1080 error = 0;
1081 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
1082 printf(
1083 "trying to add a filterop that is out of range: %d is beyond %d\n",
1084 ~filt, EVFILT_SYSCOUNT);
1085 return EINVAL;
1086 }
1087 mtx_lock(&filterops_lock);
1088 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
1089 sysfilt_ops[~filt].for_fop != NULL)
1090 error = EEXIST;
1091 else {
1092 sysfilt_ops[~filt].for_fop = filtops;
1093 sysfilt_ops[~filt].for_refcnt = 0;
1094 }
1095 mtx_unlock(&filterops_lock);
1096
1097 return (error);
1098 }
1099
1100 int
1101 kqueue_del_filteropts(int filt)
1102 {
1103 int error;
1104
1105 error = 0;
1106 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1107 return EINVAL;
1108
1109 mtx_lock(&filterops_lock);
1110 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1111 sysfilt_ops[~filt].for_fop == NULL)
1112 error = EINVAL;
1113 else if (sysfilt_ops[~filt].for_refcnt != 0)
1114 error = EBUSY;
1115 else {
1116 sysfilt_ops[~filt].for_fop = &null_filtops;
1117 sysfilt_ops[~filt].for_refcnt = 0;
1118 }
1119 mtx_unlock(&filterops_lock);
1120
1121 return error;
1122 }
1123
1124 static struct filterops *
1125 kqueue_fo_find(int filt)
1126 {
1127
1128 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1129 return NULL;
1130
1131 if (sysfilt_ops[~filt].for_nolock)
1132 return sysfilt_ops[~filt].for_fop;
1133
1134 mtx_lock(&filterops_lock);
1135 sysfilt_ops[~filt].for_refcnt++;
1136 if (sysfilt_ops[~filt].for_fop == NULL)
1137 sysfilt_ops[~filt].for_fop = &null_filtops;
1138 mtx_unlock(&filterops_lock);
1139
1140 return sysfilt_ops[~filt].for_fop;
1141 }
1142
1143 static void
1144 kqueue_fo_release(int filt)
1145 {
1146
1147 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1148 return;
1149
1150 if (sysfilt_ops[~filt].for_nolock)
1151 return;
1152
1153 mtx_lock(&filterops_lock);
1154 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1155 ("filter object refcount not valid on release"));
1156 sysfilt_ops[~filt].for_refcnt--;
1157 mtx_unlock(&filterops_lock);
1158 }
1159
1160 /*
1161 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will
1162 * influence if memory allocation should wait. Make sure it is 0 if you
1163 * hold any mutexes.
1164 */
1165 static int
1166 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
1167 {
1168 struct filterops *fops;
1169 struct file *fp;
1170 struct knote *kn, *tkn;
1171 struct knlist *knl;
1172 cap_rights_t rights;
1173 int error, filt, event;
1174 int haskqglobal, filedesc_unlock;
1175
1176 if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
1177 return (EINVAL);
1178
1179 fp = NULL;
1180 kn = NULL;
1181 knl = NULL;
1182 error = 0;
1183 haskqglobal = 0;
1184 filedesc_unlock = 0;
1185
1186 filt = kev->filter;
1187 fops = kqueue_fo_find(filt);
1188 if (fops == NULL)
1189 return EINVAL;
1190
1191 if (kev->flags & EV_ADD) {
1192 /*
1193 * Prevent waiting with locks. Non-sleepable
1194 * allocation failures are handled in the loop, only
1195 * if the spare knote appears to be actually required.
1196 */
1197 tkn = knote_alloc(waitok);
1198 } else {
1199 tkn = NULL;
1200 }
1201
1202 findkn:
1203 if (fops->f_isfd) {
1204 KASSERT(td != NULL, ("td is NULL"));
1205 if (kev->ident > INT_MAX)
1206 error = EBADF;
1207 else
1208 error = fget(td, kev->ident,
1209 cap_rights_init(&rights, CAP_EVENT), &fp);
1210 if (error)
1211 goto done;
1212
1213 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1214 kev->ident, 0) != 0) {
1215 /* try again */
1216 fdrop(fp, td);
1217 fp = NULL;
1218 error = kqueue_expand(kq, fops, kev->ident, waitok);
1219 if (error)
1220 goto done;
1221 goto findkn;
1222 }
1223
1224 if (fp->f_type == DTYPE_KQUEUE) {
1225 /*
1226 * If we add some intelligence about what we are doing,
1227 * we should be able to support events on ourselves.
1228 * We need to know when we are doing this to prevent
1229 * getting both the knlist lock and the kq lock since
1230 * they are the same thing.
1231 */
1232 if (fp->f_data == kq) {
1233 error = EINVAL;
1234 goto done;
1235 }
1236
1237 /*
1238 * Pre-lock the filedesc before the global
1239 * lock mutex, see the comment in
1240 * kqueue_close().
1241 */
1242 FILEDESC_XLOCK(td->td_proc->p_fd);
1243 filedesc_unlock = 1;
1244 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1245 }
1246
1247 KQ_LOCK(kq);
1248 if (kev->ident < kq->kq_knlistsize) {
1249 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1250 if (kev->filter == kn->kn_filter)
1251 break;
1252 }
1253 } else {
1254 if ((kev->flags & EV_ADD) == EV_ADD)
1255 kqueue_expand(kq, fops, kev->ident, waitok);
1256
1257 KQ_LOCK(kq);
1258
1259 /*
1260 * If possible, find an existing knote to use for this kevent.
1261 */
1262 if (kev->filter == EVFILT_PROC &&
1263 (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
1264 /* This is an internal creation of a process tracking
1265 * note. Don't attempt to coalesce this with an
1266 * existing note.
1267 */
1268 ;
1269 } else if (kq->kq_knhashmask != 0) {
1270 struct klist *list;
1271
1272 list = &kq->kq_knhash[
1273 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1274 SLIST_FOREACH(kn, list, kn_link)
1275 if (kev->ident == kn->kn_id &&
1276 kev->filter == kn->kn_filter)
1277 break;
1278 }
1279 }
1280
1281 /* knote is in the process of changing, wait for it to stabilize. */
1282 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1283 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1284 if (filedesc_unlock) {
1285 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1286 filedesc_unlock = 0;
1287 }
1288 kq->kq_state |= KQ_FLUXWAIT;
1289 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1290 if (fp != NULL) {
1291 fdrop(fp, td);
1292 fp = NULL;
1293 }
1294 goto findkn;
1295 }
1296
1297 /*
1298 * kn now contains the matching knote, or NULL if no match
1299 */
1300 if (kn == NULL) {
1301 if (kev->flags & EV_ADD) {
1302 kn = tkn;
1303 tkn = NULL;
1304 if (kn == NULL) {
1305 KQ_UNLOCK(kq);
1306 error = ENOMEM;
1307 goto done;
1308 }
1309 kn->kn_fp = fp;
1310 kn->kn_kq = kq;
1311 kn->kn_fop = fops;
1312 /*
1313 * apply reference counts to knote structure, and
1314 * do not release it at the end of this routine.
1315 */
1316 fops = NULL;
1317 fp = NULL;
1318
1319 kn->kn_sfflags = kev->fflags;
1320 kn->kn_sdata = kev->data;
1321 kev->fflags = 0;
1322 kev->data = 0;
1323 kn->kn_kevent = *kev;
1324 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1325 EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1326 kn->kn_status = KN_INFLUX|KN_DETACHED;
1327
1328 error = knote_attach(kn, kq);
1329 KQ_UNLOCK(kq);
1330 if (error != 0) {
1331 tkn = kn;
1332 goto done;
1333 }
1334
1335 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1336 knote_drop(kn, td);
1337 goto done;
1338 }
1339 knl = kn_list_lock(kn);
1340 goto done_ev_add;
1341 } else {
1342 /* No matching knote and the EV_ADD flag is not set. */
1343 KQ_UNLOCK(kq);
1344 error = ENOENT;
1345 goto done;
1346 }
1347 }
1348
1349 if (kev->flags & EV_DELETE) {
1350 kn->kn_status |= KN_INFLUX;
1351 KQ_UNLOCK(kq);
1352 if (!(kn->kn_status & KN_DETACHED))
1353 kn->kn_fop->f_detach(kn);
1354 knote_drop(kn, td);
1355 goto done;
1356 }
1357
1358 if (kev->flags & EV_FORCEONESHOT) {
1359 kn->kn_flags |= EV_ONESHOT;
1360 KNOTE_ACTIVATE(kn, 1);
1361 }
1362
1363 /*
1364 * The user may change some filter values after the initial EV_ADD,
1365 * but doing so will not reset any filter which has already been
1366 * triggered.
1367 */
1368 kn->kn_status |= KN_INFLUX | KN_SCAN;
1369 KQ_UNLOCK(kq);
1370 knl = kn_list_lock(kn);
1371 kn->kn_kevent.udata = kev->udata;
1372 if (!fops->f_isfd && fops->f_touch != NULL) {
1373 fops->f_touch(kn, kev, EVENT_REGISTER);
1374 } else {
1375 kn->kn_sfflags = kev->fflags;
1376 kn->kn_sdata = kev->data;
1377 }
1378
1379 /*
1380 * We can get here with kn->kn_knlist == NULL. This can happen when
1381 * the initial attach event decides that the event is "completed"
1382 * already. i.e. filt_procattach is called on a zombie process. It
1383 * will call filt_proc which will remove it from the list, and NULL
1384 * kn_knlist.
1385 */
1386 done_ev_add:
1387 if ((kev->flags & EV_ENABLE) != 0)
1388 kn->kn_status &= ~KN_DISABLED;
1389 else if ((kev->flags & EV_DISABLE) != 0)
1390 kn->kn_status |= KN_DISABLED;
1391
1392 if ((kn->kn_status & KN_DISABLED) == 0)
1393 event = kn->kn_fop->f_event(kn, 0);
1394 else
1395 event = 0;
1396
1397 KQ_LOCK(kq);
1398 if (event)
1399 kn->kn_status |= KN_ACTIVE;
1400 if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
1401 KN_ACTIVE)
1402 knote_enqueue(kn);
1403 kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1404 kn_list_unlock(knl);
1405 KQ_UNLOCK_FLUX(kq);
1406
1407 done:
1408 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1409 if (filedesc_unlock)
1410 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1411 if (fp != NULL)
1412 fdrop(fp, td);
1413 knote_free(tkn);
1414 if (fops != NULL)
1415 kqueue_fo_release(filt);
1416 return (error);
1417 }
1418
1419 static int
1420 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1421 {
1422 int error;
1423 struct kqueue *kq;
1424
1425 error = 0;
1426
1427 kq = fp->f_data;
1428 if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1429 return (EBADF);
1430 *kqp = kq;
1431 KQ_LOCK(kq);
1432 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1433 KQ_UNLOCK(kq);
1434 return (EBADF);
1435 }
1436 kq->kq_refcnt++;
1437 KQ_UNLOCK(kq);
1438
1439 return error;
1440 }
1441
1442 static void
1443 kqueue_release(struct kqueue *kq, int locked)
1444 {
1445 if (locked)
1446 KQ_OWNED(kq);
1447 else
1448 KQ_LOCK(kq);
1449 kq->kq_refcnt--;
1450 if (kq->kq_refcnt == 1)
1451 wakeup(&kq->kq_refcnt);
1452 if (!locked)
1453 KQ_UNLOCK(kq);
1454 }
1455
1456 static void
1457 kqueue_schedtask(struct kqueue *kq)
1458 {
1459
1460 KQ_OWNED(kq);
1461 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1462 ("scheduling kqueue task while draining"));
1463
1464 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1465 taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
1466 kq->kq_state |= KQ_TASKSCHED;
1467 }
1468 }
1469
1470 /*
1471 * Expand the kq to make sure we have storage for fops/ident pair.
1472 *
1473 * Return 0 on success (or no work necessary), return errno on failure.
1474 *
1475 * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1476 * If kqueue_register is called from a non-fd context, there usually/should
1477 * be no locks held.
1478 */
1479 static int
1480 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1481 int waitok)
1482 {
1483 struct klist *list, *tmp_knhash, *to_free;
1484 u_long tmp_knhashmask;
1485 int size;
1486 int fd;
1487 int mflag = waitok ? M_WAITOK : M_NOWAIT;
1488
1489 KQ_NOTOWNED(kq);
1490
1491 to_free = NULL;
1492 if (fops->f_isfd) {
1493 fd = ident;
1494 if (kq->kq_knlistsize <= fd) {
1495 size = kq->kq_knlistsize;
1496 while (size <= fd)
1497 size += KQEXTENT;
1498 list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1499 if (list == NULL)
1500 return ENOMEM;
1501 KQ_LOCK(kq);
1502 if (kq->kq_knlistsize > fd) {
1503 to_free = list;
1504 list = NULL;
1505 } else {
1506 if (kq->kq_knlist != NULL) {
1507 bcopy(kq->kq_knlist, list,
1508 kq->kq_knlistsize * sizeof(*list));
1509 to_free = kq->kq_knlist;
1510 kq->kq_knlist = NULL;
1511 }
1512 bzero((caddr_t)list +
1513 kq->kq_knlistsize * sizeof(*list),
1514 (size - kq->kq_knlistsize) * sizeof(*list));
1515 kq->kq_knlistsize = size;
1516 kq->kq_knlist = list;
1517 }
1518 KQ_UNLOCK(kq);
1519 }
1520 } else {
1521 if (kq->kq_knhashmask == 0) {
1522 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1523 &tmp_knhashmask);
1524 if (tmp_knhash == NULL)
1525 return ENOMEM;
1526 KQ_LOCK(kq);
1527 if (kq->kq_knhashmask == 0) {
1528 kq->kq_knhash = tmp_knhash;
1529 kq->kq_knhashmask = tmp_knhashmask;
1530 } else {
1531 to_free = tmp_knhash;
1532 }
1533 KQ_UNLOCK(kq);
1534 }
1535 }
1536 free(to_free, M_KQUEUE);
1537
1538 KQ_NOTOWNED(kq);
1539 return 0;
1540 }
1541
1542 static void
1543 kqueue_task(void *arg, int pending)
1544 {
1545 struct kqueue *kq;
1546 int haskqglobal;
1547
1548 haskqglobal = 0;
1549 kq = arg;
1550
1551 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1552 KQ_LOCK(kq);
1553
1554 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1555
1556 kq->kq_state &= ~KQ_TASKSCHED;
1557 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1558 wakeup(&kq->kq_state);
1559 }
1560 KQ_UNLOCK(kq);
1561 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1562 }
1563
1564 /*
1565 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1566 * We treat KN_MARKER knotes as if they are INFLUX.
1567 */
1568 static int
1569 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1570 const struct timespec *tsp, struct kevent *keva, struct thread *td)
1571 {
1572 struct kevent *kevp;
1573 struct knote *kn, *marker;
1574 struct knlist *knl;
1575 sbintime_t asbt, rsbt;
1576 int count, error, haskqglobal, influx, nkev, touch;
1577
1578 count = maxevents;
1579 nkev = 0;
1580 error = 0;
1581 haskqglobal = 0;
1582
1583 if (maxevents == 0)
1584 goto done_nl;
1585
1586 rsbt = 0;
1587 if (tsp != NULL) {
1588 if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 ||
1589 tsp->tv_nsec >= 1000000000) {
1590 error = EINVAL;
1591 goto done_nl;
1592 }
1593 if (timespecisset(tsp)) {
1594 if (tsp->tv_sec <= INT32_MAX) {
1595 rsbt = tstosbt(*tsp);
1596 if (TIMESEL(&asbt, rsbt))
1597 asbt += tc_tick_sbt;
1598 if (asbt <= SBT_MAX - rsbt)
1599 asbt += rsbt;
1600 else
1601 asbt = 0;
1602 rsbt >>= tc_precexp;
1603 } else
1604 asbt = 0;
1605 } else
1606 asbt = -1;
1607 } else
1608 asbt = 0;
1609 marker = knote_alloc(1);
1610 marker->kn_status = KN_MARKER;
1611 KQ_LOCK(kq);
1612
1613 retry:
1614 kevp = keva;
1615 if (kq->kq_count == 0) {
1616 if (asbt == -1) {
1617 error = EWOULDBLOCK;
1618 } else {
1619 kq->kq_state |= KQ_SLEEP;
1620 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
1621 "kqread", asbt, rsbt, C_ABSOLUTE);
1622 }
1623 if (error == 0)
1624 goto retry;
1625 /* don't restart after signals... */
1626 if (error == ERESTART)
1627 error = EINTR;
1628 else if (error == EWOULDBLOCK)
1629 error = 0;
1630 goto done;
1631 }
1632
1633 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1634 influx = 0;
1635 while (count) {
1636 KQ_OWNED(kq);
1637 kn = TAILQ_FIRST(&kq->kq_head);
1638
1639 if ((kn->kn_status == KN_MARKER && kn != marker) ||
1640 (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1641 if (influx) {
1642 influx = 0;
1643 KQ_FLUX_WAKEUP(kq);
1644 }
1645 kq->kq_state |= KQ_FLUXWAIT;
1646 error = msleep(kq, &kq->kq_lock, PSOCK,
1647 "kqflxwt", 0);
1648 continue;
1649 }
1650
1651 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1652 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1653 kn->kn_status &= ~KN_QUEUED;
1654 kq->kq_count--;
1655 continue;
1656 }
1657 if (kn == marker) {
1658 KQ_FLUX_WAKEUP(kq);
1659 if (count == maxevents)
1660 goto retry;
1661 goto done;
1662 }
1663 KASSERT((kn->kn_status & KN_INFLUX) == 0,
1664 ("KN_INFLUX set when not suppose to be"));
1665
1666 if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1667 kn->kn_status &= ~KN_QUEUED;
1668 kn->kn_status |= KN_INFLUX;
1669 kq->kq_count--;
1670 KQ_UNLOCK(kq);
1671 /*
1672 * We don't need to lock the list since we've marked
1673 * it _INFLUX.
1674 */
1675 if (!(kn->kn_status & KN_DETACHED))
1676 kn->kn_fop->f_detach(kn);
1677 knote_drop(kn, td);
1678 KQ_LOCK(kq);
1679 continue;
1680 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1681 kn->kn_status &= ~KN_QUEUED;
1682 kn->kn_status |= KN_INFLUX;
1683 kq->kq_count--;
1684 KQ_UNLOCK(kq);
1685 /*
1686 * We don't need to lock the list since we've marked
1687 * it _INFLUX.
1688 */
1689 *kevp = kn->kn_kevent;
1690 if (!(kn->kn_status & KN_DETACHED))
1691 kn->kn_fop->f_detach(kn);
1692 knote_drop(kn, td);
1693 KQ_LOCK(kq);
1694 kn = NULL;
1695 } else {
1696 kn->kn_status |= KN_INFLUX | KN_SCAN;
1697 KQ_UNLOCK(kq);
1698 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1699 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1700 knl = kn_list_lock(kn);
1701 if (kn->kn_fop->f_event(kn, 0) == 0) {
1702 KQ_LOCK(kq);
1703 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1704 kn->kn_status &=
1705 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX |
1706 KN_SCAN);
1707 kq->kq_count--;
1708 kn_list_unlock(knl);
1709 influx = 1;
1710 continue;
1711 }
1712 touch = (!kn->kn_fop->f_isfd &&
1713 kn->kn_fop->f_touch != NULL);
1714 if (touch)
1715 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1716 else
1717 *kevp = kn->kn_kevent;
1718 KQ_LOCK(kq);
1719 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1720 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1721 /*
1722 * Manually clear knotes who weren't
1723 * 'touch'ed.
1724 */
1725 if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1726 kn->kn_data = 0;
1727 kn->kn_fflags = 0;
1728 }
1729 if (kn->kn_flags & EV_DISPATCH)
1730 kn->kn_status |= KN_DISABLED;
1731 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1732 kq->kq_count--;
1733 } else
1734 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1735
1736 kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1737 kn_list_unlock(knl);
1738 influx = 1;
1739 }
1740
1741 /* we are returning a copy to the user */
1742 kevp++;
1743 nkev++;
1744 count--;
1745
1746 if (nkev == KQ_NEVENTS) {
1747 influx = 0;
1748 KQ_UNLOCK_FLUX(kq);
1749 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1750 nkev = 0;
1751 kevp = keva;
1752 KQ_LOCK(kq);
1753 if (error)
1754 break;
1755 }
1756 }
1757 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1758 done:
1759 KQ_OWNED(kq);
1760 KQ_UNLOCK_FLUX(kq);
1761 knote_free(marker);
1762 done_nl:
1763 KQ_NOTOWNED(kq);
1764 if (nkev != 0)
1765 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1766 td->td_retval[0] = maxevents - count;
1767 return (error);
1768 }
1769
1770 /*ARGSUSED*/
1771 static int
1772 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1773 struct ucred *active_cred, struct thread *td)
1774 {
1775 /*
1776 * Enabling sigio causes two major problems:
1777 * 1) infinite recursion:
1778 * Synopsys: kevent is being used to track signals and have FIOASYNC
1779 * set. On receipt of a signal this will cause a kqueue to recurse
1780 * into itself over and over. Sending the sigio causes the kqueue
1781 * to become ready, which in turn posts sigio again, forever.
1782 * Solution: this can be solved by setting a flag in the kqueue that
1783 * we have a SIGIO in progress.
1784 * 2) locking problems:
1785 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1786 * us above the proc and pgrp locks.
1787 * Solution: Post a signal using an async mechanism, being sure to
1788 * record a generation count in the delivery so that we do not deliver
1789 * a signal to the wrong process.
1790 *
1791 * Note, these two mechanisms are somewhat mutually exclusive!
1792 */
1793 #if 0
1794 struct kqueue *kq;
1795
1796 kq = fp->f_data;
1797 switch (cmd) {
1798 case FIOASYNC:
1799 if (*(int *)data) {
1800 kq->kq_state |= KQ_ASYNC;
1801 } else {
1802 kq->kq_state &= ~KQ_ASYNC;
1803 }
1804 return (0);
1805
1806 case FIOSETOWN:
1807 return (fsetown(*(int *)data, &kq->kq_sigio));
1808
1809 case FIOGETOWN:
1810 *(int *)data = fgetown(&kq->kq_sigio);
1811 return (0);
1812 }
1813 #endif
1814
1815 return (ENOTTY);
1816 }
1817
1818 /*ARGSUSED*/
1819 static int
1820 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1821 struct thread *td)
1822 {
1823 struct kqueue *kq;
1824 int revents = 0;
1825 int error;
1826
1827 if ((error = kqueue_acquire(fp, &kq)))
1828 return POLLERR;
1829
1830 KQ_LOCK(kq);
1831 if (events & (POLLIN | POLLRDNORM)) {
1832 if (kq->kq_count) {
1833 revents |= events & (POLLIN | POLLRDNORM);
1834 } else {
1835 selrecord(td, &kq->kq_sel);
1836 if (SEL_WAITING(&kq->kq_sel))
1837 kq->kq_state |= KQ_SEL;
1838 }
1839 }
1840 kqueue_release(kq, 1);
1841 KQ_UNLOCK(kq);
1842 return (revents);
1843 }
1844
1845 /*ARGSUSED*/
1846 static int
1847 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1848 struct thread *td)
1849 {
1850
1851 bzero((void *)st, sizeof *st);
1852 /*
1853 * We no longer return kq_count because the unlocked value is useless.
1854 * If you spent all this time getting the count, why not spend your
1855 * syscall better by calling kevent?
1856 *
1857 * XXX - This is needed for libc_r.
1858 */
1859 st->st_mode = S_IFIFO;
1860 return (0);
1861 }
1862
1863 static void
1864 kqueue_drain(struct kqueue *kq, struct thread *td)
1865 {
1866 struct knote *kn;
1867 int i;
1868
1869 KQ_LOCK(kq);
1870
1871 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1872 ("kqueue already closing"));
1873 kq->kq_state |= KQ_CLOSING;
1874 if (kq->kq_refcnt > 1)
1875 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1876
1877 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1878
1879 KASSERT(knlist_empty(&kq->kq_sel.si_note),
1880 ("kqueue's knlist not empty"));
1881
1882 for (i = 0; i < kq->kq_knlistsize; i++) {
1883 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1884 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1885 kq->kq_state |= KQ_FLUXWAIT;
1886 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1887 continue;
1888 }
1889 kn->kn_status |= KN_INFLUX;
1890 KQ_UNLOCK(kq);
1891 if (!(kn->kn_status & KN_DETACHED))
1892 kn->kn_fop->f_detach(kn);
1893 knote_drop(kn, td);
1894 KQ_LOCK(kq);
1895 }
1896 }
1897 if (kq->kq_knhashmask != 0) {
1898 for (i = 0; i <= kq->kq_knhashmask; i++) {
1899 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1900 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1901 kq->kq_state |= KQ_FLUXWAIT;
1902 msleep(kq, &kq->kq_lock, PSOCK,
1903 "kqclo2", 0);
1904 continue;
1905 }
1906 kn->kn_status |= KN_INFLUX;
1907 KQ_UNLOCK(kq);
1908 if (!(kn->kn_status & KN_DETACHED))
1909 kn->kn_fop->f_detach(kn);
1910 knote_drop(kn, td);
1911 KQ_LOCK(kq);
1912 }
1913 }
1914 }
1915
1916 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1917 kq->kq_state |= KQ_TASKDRAIN;
1918 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1919 }
1920
1921 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1922 selwakeuppri(&kq->kq_sel, PSOCK);
1923 if (!SEL_WAITING(&kq->kq_sel))
1924 kq->kq_state &= ~KQ_SEL;
1925 }
1926
1927 KQ_UNLOCK(kq);
1928 }
1929
1930 static void
1931 kqueue_destroy(struct kqueue *kq)
1932 {
1933
1934 KASSERT(kq->kq_fdp == NULL,
1935 ("kqueue still attached to a file descriptor"));
1936 seldrain(&kq->kq_sel);
1937 knlist_destroy(&kq->kq_sel.si_note);
1938 mtx_destroy(&kq->kq_lock);
1939
1940 if (kq->kq_knhash != NULL)
1941 free(kq->kq_knhash, M_KQUEUE);
1942 if (kq->kq_knlist != NULL)
1943 free(kq->kq_knlist, M_KQUEUE);
1944
1945 funsetown(&kq->kq_sigio);
1946 }
1947
1948 /*ARGSUSED*/
1949 static int
1950 kqueue_close(struct file *fp, struct thread *td)
1951 {
1952 struct kqueue *kq = fp->f_data;
1953 struct filedesc *fdp;
1954 int error;
1955 int filedesc_unlock;
1956
1957 if ((error = kqueue_acquire(fp, &kq)))
1958 return error;
1959 kqueue_drain(kq, td);
1960
1961 /*
1962 * We could be called due to the knote_drop() doing fdrop(),
1963 * called from kqueue_register(). In this case the global
1964 * lock is owned, and filedesc sx is locked before, to not
1965 * take the sleepable lock after non-sleepable.
1966 */
1967 fdp = kq->kq_fdp;
1968 kq->kq_fdp = NULL;
1969 if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
1970 FILEDESC_XLOCK(fdp);
1971 filedesc_unlock = 1;
1972 } else
1973 filedesc_unlock = 0;
1974 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
1975 if (filedesc_unlock)
1976 FILEDESC_XUNLOCK(fdp);
1977
1978 kqueue_destroy(kq);
1979 chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
1980 crfree(kq->kq_cred);
1981 free(kq, M_KQUEUE);
1982 fp->f_data = NULL;
1983
1984 return (0);
1985 }
1986
1987 static int
1988 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
1989 {
1990
1991 kif->kf_type = KF_TYPE_KQUEUE;
1992 return (0);
1993 }
1994
1995 static void
1996 kqueue_wakeup(struct kqueue *kq)
1997 {
1998 KQ_OWNED(kq);
1999
2000 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
2001 kq->kq_state &= ~KQ_SLEEP;
2002 wakeup(kq);
2003 }
2004 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2005 selwakeuppri(&kq->kq_sel, PSOCK);
2006 if (!SEL_WAITING(&kq->kq_sel))
2007 kq->kq_state &= ~KQ_SEL;
2008 }
2009 if (!knlist_empty(&kq->kq_sel.si_note))
2010 kqueue_schedtask(kq);
2011 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
2012 pgsigio(&kq->kq_sigio, SIGIO, 0);
2013 }
2014 }
2015
2016 /*
2017 * Walk down a list of knotes, activating them if their event has triggered.
2018 *
2019 * There is a possibility to optimize in the case of one kq watching another.
2020 * Instead of scheduling a task to wake it up, you could pass enough state
2021 * down the chain to make up the parent kqueue. Make this code functional
2022 * first.
2023 */
2024 void
2025 knote(struct knlist *list, long hint, int lockflags)
2026 {
2027 struct kqueue *kq;
2028 struct knote *kn, *tkn;
2029 int error;
2030 bool own_influx;
2031
2032 if (list == NULL)
2033 return;
2034
2035 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
2036
2037 if ((lockflags & KNF_LISTLOCKED) == 0)
2038 list->kl_lock(list->kl_lockarg);
2039
2040 /*
2041 * If we unlock the list lock (and set KN_INFLUX), we can
2042 * eliminate the kqueue scheduling, but this will introduce
2043 * four lock/unlock's for each knote to test. Also, marker
2044 * would be needed to keep iteration position, since filters
2045 * or other threads could remove events.
2046 */
2047 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
2048 kq = kn->kn_kq;
2049 KQ_LOCK(kq);
2050 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
2051 /*
2052 * Do not process the influx notes, except for
2053 * the influx coming from the kq unlock in the
2054 * kqueue_scan(). In the later case, we do
2055 * not interfere with the scan, since the code
2056 * fragment in kqueue_scan() locks the knlist,
2057 * and cannot proceed until we finished.
2058 */
2059 KQ_UNLOCK(kq);
2060 } else if ((lockflags & KNF_NOKQLOCK) != 0) {
2061 own_influx = (kn->kn_status & KN_INFLUX) == 0;
2062 if (own_influx)
2063 kn->kn_status |= KN_INFLUX;
2064 KQ_UNLOCK(kq);
2065 error = kn->kn_fop->f_event(kn, hint);
2066 KQ_LOCK(kq);
2067 if (own_influx)
2068 kn->kn_status &= ~KN_INFLUX;
2069 if (error)
2070 KNOTE_ACTIVATE(kn, 1);
2071 KQ_UNLOCK_FLUX(kq);
2072 } else {
2073 kn->kn_status |= KN_HASKQLOCK;
2074 if (kn->kn_fop->f_event(kn, hint))
2075 KNOTE_ACTIVATE(kn, 1);
2076 kn->kn_status &= ~KN_HASKQLOCK;
2077 KQ_UNLOCK(kq);
2078 }
2079 }
2080 if ((lockflags & KNF_LISTLOCKED) == 0)
2081 list->kl_unlock(list->kl_lockarg);
2082 }
2083
2084 /*
2085 * add a knote to a knlist
2086 */
2087 void
2088 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
2089 {
2090 KNL_ASSERT_LOCK(knl, islocked);
2091 KQ_NOTOWNED(kn->kn_kq);
2092 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
2093 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
2094 if (!islocked)
2095 knl->kl_lock(knl->kl_lockarg);
2096 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
2097 if (!islocked)
2098 knl->kl_unlock(knl->kl_lockarg);
2099 KQ_LOCK(kn->kn_kq);
2100 kn->kn_knlist = knl;
2101 kn->kn_status &= ~KN_DETACHED;
2102 KQ_UNLOCK(kn->kn_kq);
2103 }
2104
2105 static void
2106 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
2107 int kqislocked)
2108 {
2109 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
2110 KNL_ASSERT_LOCK(knl, knlislocked);
2111 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
2112 if (!kqislocked)
2113 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
2114 ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
2115 if (!knlislocked)
2116 knl->kl_lock(knl->kl_lockarg);
2117 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
2118 kn->kn_knlist = NULL;
2119 if (!knlislocked)
2120 kn_list_unlock(knl);
2121 if (!kqislocked)
2122 KQ_LOCK(kn->kn_kq);
2123 kn->kn_status |= KN_DETACHED;
2124 if (!kqislocked)
2125 KQ_UNLOCK(kn->kn_kq);
2126 }
2127
2128 /*
2129 * remove knote from the specified knlist
2130 */
2131 void
2132 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
2133 {
2134
2135 knlist_remove_kq(knl, kn, islocked, 0);
2136 }
2137
2138 int
2139 knlist_empty(struct knlist *knl)
2140 {
2141
2142 KNL_ASSERT_LOCKED(knl);
2143 return (SLIST_EMPTY(&knl->kl_list));
2144 }
2145
2146 static struct mtx knlist_lock;
2147 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2148 MTX_DEF);
2149 static void knlist_mtx_lock(void *arg);
2150 static void knlist_mtx_unlock(void *arg);
2151
2152 static void
2153 knlist_mtx_lock(void *arg)
2154 {
2155
2156 mtx_lock((struct mtx *)arg);
2157 }
2158
2159 static void
2160 knlist_mtx_unlock(void *arg)
2161 {
2162
2163 mtx_unlock((struct mtx *)arg);
2164 }
2165
2166 static void
2167 knlist_mtx_assert_locked(void *arg)
2168 {
2169
2170 mtx_assert((struct mtx *)arg, MA_OWNED);
2171 }
2172
2173 static void
2174 knlist_mtx_assert_unlocked(void *arg)
2175 {
2176
2177 mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2178 }
2179
2180 static void
2181 knlist_rw_rlock(void *arg)
2182 {
2183
2184 rw_rlock((struct rwlock *)arg);
2185 }
2186
2187 static void
2188 knlist_rw_runlock(void *arg)
2189 {
2190
2191 rw_runlock((struct rwlock *)arg);
2192 }
2193
2194 static void
2195 knlist_rw_assert_locked(void *arg)
2196 {
2197
2198 rw_assert((struct rwlock *)arg, RA_LOCKED);
2199 }
2200
2201 static void
2202 knlist_rw_assert_unlocked(void *arg)
2203 {
2204
2205 rw_assert((struct rwlock *)arg, RA_UNLOCKED);
2206 }
2207
2208 void
2209 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2210 void (*kl_unlock)(void *),
2211 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
2212 {
2213
2214 if (lock == NULL)
2215 knl->kl_lockarg = &knlist_lock;
2216 else
2217 knl->kl_lockarg = lock;
2218
2219 if (kl_lock == NULL)
2220 knl->kl_lock = knlist_mtx_lock;
2221 else
2222 knl->kl_lock = kl_lock;
2223 if (kl_unlock == NULL)
2224 knl->kl_unlock = knlist_mtx_unlock;
2225 else
2226 knl->kl_unlock = kl_unlock;
2227 if (kl_assert_locked == NULL)
2228 knl->kl_assert_locked = knlist_mtx_assert_locked;
2229 else
2230 knl->kl_assert_locked = kl_assert_locked;
2231 if (kl_assert_unlocked == NULL)
2232 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
2233 else
2234 knl->kl_assert_unlocked = kl_assert_unlocked;
2235
2236 knl->kl_autodestroy = 0;
2237 SLIST_INIT(&knl->kl_list);
2238 }
2239
2240 void
2241 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2242 {
2243
2244 knlist_init(knl, lock, NULL, NULL, NULL, NULL);
2245 }
2246
2247 struct knlist *
2248 knlist_alloc(struct mtx *lock)
2249 {
2250 struct knlist *knl;
2251
2252 knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
2253 knlist_init_mtx(knl, lock);
2254 return (knl);
2255 }
2256
2257 void
2258 knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock)
2259 {
2260
2261 knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock,
2262 knlist_rw_assert_locked, knlist_rw_assert_unlocked);
2263 }
2264
2265 void
2266 knlist_destroy(struct knlist *knl)
2267 {
2268
2269 KASSERT(KNLIST_EMPTY(knl),
2270 ("destroying knlist %p with knotes on it", knl));
2271 }
2272
2273 void
2274 knlist_detach(struct knlist *knl)
2275 {
2276
2277 KNL_ASSERT_LOCKED(knl);
2278 knl->kl_autodestroy = 1;
2279 if (knlist_empty(knl)) {
2280 knlist_destroy(knl);
2281 free(knl, M_KQUEUE);
2282 }
2283 }
2284
2285 /*
2286 * Even if we are locked, we may need to drop the lock to allow any influx
2287 * knotes time to "settle".
2288 */
2289 void
2290 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2291 {
2292 struct knote *kn, *kn2;
2293 struct kqueue *kq;
2294
2295 KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
2296 if (islocked)
2297 KNL_ASSERT_LOCKED(knl);
2298 else {
2299 KNL_ASSERT_UNLOCKED(knl);
2300 again: /* need to reacquire lock since we have dropped it */
2301 knl->kl_lock(knl->kl_lockarg);
2302 }
2303
2304 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2305 kq = kn->kn_kq;
2306 KQ_LOCK(kq);
2307 if ((kn->kn_status & KN_INFLUX)) {
2308 KQ_UNLOCK(kq);
2309 continue;
2310 }
2311 knlist_remove_kq(knl, kn, 1, 1);
2312 if (killkn) {
2313 kn->kn_status |= KN_INFLUX | KN_DETACHED;
2314 KQ_UNLOCK(kq);
2315 knote_drop(kn, td);
2316 } else {
2317 /* Make sure cleared knotes disappear soon */
2318 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2319 KQ_UNLOCK(kq);
2320 }
2321 kq = NULL;
2322 }
2323
2324 if (!SLIST_EMPTY(&knl->kl_list)) {
2325 /* there are still KN_INFLUX remaining */
2326 kn = SLIST_FIRST(&knl->kl_list);
2327 kq = kn->kn_kq;
2328 KQ_LOCK(kq);
2329 KASSERT(kn->kn_status & KN_INFLUX,
2330 ("knote removed w/o list lock"));
2331 knl->kl_unlock(knl->kl_lockarg);
2332 kq->kq_state |= KQ_FLUXWAIT;
2333 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2334 kq = NULL;
2335 goto again;
2336 }
2337
2338 if (islocked)
2339 KNL_ASSERT_LOCKED(knl);
2340 else {
2341 knl->kl_unlock(knl->kl_lockarg);
2342 KNL_ASSERT_UNLOCKED(knl);
2343 }
2344 }
2345
2346 /*
2347 * Remove all knotes referencing a specified fd must be called with FILEDESC
2348 * lock. This prevents a race where a new fd comes along and occupies the
2349 * entry and we attach a knote to the fd.
2350 */
2351 void
2352 knote_fdclose(struct thread *td, int fd)
2353 {
2354 struct filedesc *fdp = td->td_proc->p_fd;
2355 struct kqueue *kq;
2356 struct knote *kn;
2357 int influx;
2358
2359 FILEDESC_XLOCK_ASSERT(fdp);
2360
2361 /*
2362 * We shouldn't have to worry about new kevents appearing on fd
2363 * since filedesc is locked.
2364 */
2365 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2366 KQ_LOCK(kq);
2367
2368 again:
2369 influx = 0;
2370 while (kq->kq_knlistsize > fd &&
2371 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2372 if (kn->kn_status & KN_INFLUX) {
2373 /* someone else might be waiting on our knote */
2374 if (influx)
2375 wakeup(kq);
2376 kq->kq_state |= KQ_FLUXWAIT;
2377 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2378 goto again;
2379 }
2380 kn->kn_status |= KN_INFLUX;
2381 KQ_UNLOCK(kq);
2382 if (!(kn->kn_status & KN_DETACHED))
2383 kn->kn_fop->f_detach(kn);
2384 knote_drop(kn, td);
2385 influx = 1;
2386 KQ_LOCK(kq);
2387 }
2388 KQ_UNLOCK_FLUX(kq);
2389 }
2390 }
2391
2392 static int
2393 knote_attach(struct knote *kn, struct kqueue *kq)
2394 {
2395 struct klist *list;
2396
2397 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2398 KQ_OWNED(kq);
2399
2400 if (kn->kn_fop->f_isfd) {
2401 if (kn->kn_id >= kq->kq_knlistsize)
2402 return (ENOMEM);
2403 list = &kq->kq_knlist[kn->kn_id];
2404 } else {
2405 if (kq->kq_knhash == NULL)
2406 return (ENOMEM);
2407 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2408 }
2409 SLIST_INSERT_HEAD(list, kn, kn_link);
2410 return (0);
2411 }
2412
2413 /*
2414 * knote must already have been detached using the f_detach method.
2415 * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2416 * to prevent other removal.
2417 */
2418 static void
2419 knote_drop(struct knote *kn, struct thread *td)
2420 {
2421 struct kqueue *kq;
2422 struct klist *list;
2423
2424 kq = kn->kn_kq;
2425
2426 KQ_NOTOWNED(kq);
2427 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2428 ("knote_drop called without KN_INFLUX set in kn_status"));
2429
2430 KQ_LOCK(kq);
2431 if (kn->kn_fop->f_isfd)
2432 list = &kq->kq_knlist[kn->kn_id];
2433 else
2434 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2435
2436 if (!SLIST_EMPTY(list))
2437 SLIST_REMOVE(list, kn, knote, kn_link);
2438 if (kn->kn_status & KN_QUEUED)
2439 knote_dequeue(kn);
2440 KQ_UNLOCK_FLUX(kq);
2441
2442 if (kn->kn_fop->f_isfd) {
2443 fdrop(kn->kn_fp, td);
2444 kn->kn_fp = NULL;
2445 }
2446 kqueue_fo_release(kn->kn_kevent.filter);
2447 kn->kn_fop = NULL;
2448 knote_free(kn);
2449 }
2450
2451 static void
2452 knote_enqueue(struct knote *kn)
2453 {
2454 struct kqueue *kq = kn->kn_kq;
2455
2456 KQ_OWNED(kn->kn_kq);
2457 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2458
2459 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2460 kn->kn_status |= KN_QUEUED;
2461 kq->kq_count++;
2462 kqueue_wakeup(kq);
2463 }
2464
2465 static void
2466 knote_dequeue(struct knote *kn)
2467 {
2468 struct kqueue *kq = kn->kn_kq;
2469
2470 KQ_OWNED(kn->kn_kq);
2471 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2472
2473 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2474 kn->kn_status &= ~KN_QUEUED;
2475 kq->kq_count--;
2476 }
2477
2478 static void
2479 knote_init(void)
2480 {
2481
2482 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2483 NULL, NULL, UMA_ALIGN_PTR, 0);
2484 }
2485 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2486
2487 static struct knote *
2488 knote_alloc(int waitok)
2489 {
2490
2491 return (uma_zalloc(knote_zone, (waitok ? M_WAITOK : M_NOWAIT) |
2492 M_ZERO));
2493 }
2494
2495 static void
2496 knote_free(struct knote *kn)
2497 {
2498
2499 uma_zfree(knote_zone, kn);
2500 }
2501
2502 /*
2503 * Register the kev w/ the kq specified by fd.
2504 */
2505 int
2506 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2507 {
2508 struct kqueue *kq;
2509 struct file *fp;
2510 cap_rights_t rights;
2511 int error;
2512
2513 error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp);
2514 if (error != 0)
2515 return (error);
2516 if ((error = kqueue_acquire(fp, &kq)) != 0)
2517 goto noacquire;
2518
2519 error = kqueue_register(kq, kev, td, waitok);
2520 kqueue_release(kq, 0);
2521
2522 noacquire:
2523 fdrop(fp, td);
2524 return (error);
2525 }
Cache object: a044f53490a28a794bd8c1c114850870
|