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: stable/9/sys/kern/kern_event.c 303220 2016-07-23 11:26:24Z kib $");
31
32 #include "opt_ktrace.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/capability.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/malloc.h>
42 #include <sys/unistd.h>
43 #include <sys/file.h>
44 #include <sys/filedesc.h>
45 #include <sys/filio.h>
46 #include <sys/fcntl.h>
47 #include <sys/kthread.h>
48 #include <sys/limits.h>
49 #include <sys/selinfo.h>
50 #include <sys/queue.h>
51 #include <sys/event.h>
52 #include <sys/eventvar.h>
53 #include <sys/poll.h>
54 #include <sys/protosw.h>
55 #include <sys/sigio.h>
56 #include <sys/signalvar.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/stat.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysproto.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/taskqueue.h>
64 #include <sys/uio.h>
65 #ifdef KTRACE
66 #include <sys/ktrace.h>
67 #endif
68
69 #include <vm/uma.h>
70
71 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
72
73 /*
74 * This lock is used if multiple kq locks are required. This possibly
75 * should be made into a per proc lock.
76 */
77 static struct mtx kq_global;
78 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
79 #define KQ_GLOBAL_LOCK(lck, haslck) do { \
80 if (!haslck) \
81 mtx_lock(lck); \
82 haslck = 1; \
83 } while (0)
84 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \
85 if (haslck) \
86 mtx_unlock(lck); \
87 haslck = 0; \
88 } while (0)
89
90 TASKQUEUE_DEFINE_THREAD(kqueue);
91
92 static int kevent_copyout(void *arg, struct kevent *kevp, int count);
93 static int kevent_copyin(void *arg, struct kevent *kevp, int count);
94 static int kqueue_register(struct kqueue *kq, struct kevent *kev,
95 struct thread *td, int waitok);
96 static int kqueue_acquire(struct file *fp, struct kqueue **kqp);
97 static void kqueue_release(struct kqueue *kq, int locked);
98 static int kqueue_expand(struct kqueue *kq, struct filterops *fops,
99 uintptr_t ident, int waitok);
100 static void kqueue_task(void *arg, int pending);
101 static int kqueue_scan(struct kqueue *kq, int maxevents,
102 struct kevent_copyops *k_ops,
103 const struct timespec *timeout,
104 struct kevent *keva, struct thread *td);
105 static void kqueue_wakeup(struct kqueue *kq);
106 static struct filterops *kqueue_fo_find(int filt);
107 static void kqueue_fo_release(int filt);
108
109 static fo_rdwr_t kqueue_read;
110 static fo_rdwr_t kqueue_write;
111 static fo_truncate_t kqueue_truncate;
112 static fo_ioctl_t kqueue_ioctl;
113 static fo_poll_t kqueue_poll;
114 static fo_kqfilter_t kqueue_kqfilter;
115 static fo_stat_t kqueue_stat;
116 static fo_close_t kqueue_close;
117
118 static struct fileops kqueueops = {
119 .fo_read = kqueue_read,
120 .fo_write = kqueue_write,
121 .fo_truncate = kqueue_truncate,
122 .fo_ioctl = kqueue_ioctl,
123 .fo_poll = kqueue_poll,
124 .fo_kqfilter = kqueue_kqfilter,
125 .fo_stat = kqueue_stat,
126 .fo_close = kqueue_close,
127 .fo_chmod = invfo_chmod,
128 .fo_chown = invfo_chown,
129 };
130
131 static int knote_attach(struct knote *kn, struct kqueue *kq);
132 static void knote_drop(struct knote *kn, struct thread *td);
133 static void knote_enqueue(struct knote *kn);
134 static void knote_dequeue(struct knote *kn);
135 static void knote_init(void);
136 static struct knote *knote_alloc(int waitok);
137 static void knote_free(struct knote *kn);
138
139 static void filt_kqdetach(struct knote *kn);
140 static int filt_kqueue(struct knote *kn, long hint);
141 static int filt_procattach(struct knote *kn);
142 static void filt_procdetach(struct knote *kn);
143 static int filt_proc(struct knote *kn, long hint);
144 static int filt_fileattach(struct knote *kn);
145 static void filt_timerexpire(void *knx);
146 static int filt_timerattach(struct knote *kn);
147 static void filt_timerdetach(struct knote *kn);
148 static int filt_timer(struct knote *kn, long hint);
149 static int filt_userattach(struct knote *kn);
150 static void filt_userdetach(struct knote *kn);
151 static int filt_user(struct knote *kn, long hint);
152 static void filt_usertouch(struct knote *kn, struct kevent *kev,
153 u_long type);
154
155 static struct filterops file_filtops = {
156 .f_isfd = 1,
157 .f_attach = filt_fileattach,
158 };
159 static struct filterops kqread_filtops = {
160 .f_isfd = 1,
161 .f_detach = filt_kqdetach,
162 .f_event = filt_kqueue,
163 };
164 /* XXX - move to kern_proc.c? */
165 static struct filterops proc_filtops = {
166 .f_isfd = 0,
167 .f_attach = filt_procattach,
168 .f_detach = filt_procdetach,
169 .f_event = filt_proc,
170 };
171 static struct filterops timer_filtops = {
172 .f_isfd = 0,
173 .f_attach = filt_timerattach,
174 .f_detach = filt_timerdetach,
175 .f_event = filt_timer,
176 };
177 static struct filterops user_filtops = {
178 .f_attach = filt_userattach,
179 .f_detach = filt_userdetach,
180 .f_event = filt_user,
181 .f_touch = filt_usertouch,
182 };
183
184 static uma_zone_t knote_zone;
185 static int kq_ncallouts = 0;
186 static int kq_calloutmax = (4 * 1024);
187 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
188 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
189
190 /* XXX - ensure not KN_INFLUX?? */
191 #define KNOTE_ACTIVATE(kn, islock) do { \
192 if ((islock)) \
193 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \
194 else \
195 KQ_LOCK((kn)->kn_kq); \
196 (kn)->kn_status |= KN_ACTIVE; \
197 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
198 knote_enqueue((kn)); \
199 if (!(islock)) \
200 KQ_UNLOCK((kn)->kn_kq); \
201 } while(0)
202 #define KQ_LOCK(kq) do { \
203 mtx_lock(&(kq)->kq_lock); \
204 } while (0)
205 #define KQ_FLUX_WAKEUP(kq) do { \
206 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
207 (kq)->kq_state &= ~KQ_FLUXWAIT; \
208 wakeup((kq)); \
209 } \
210 } while (0)
211 #define KQ_UNLOCK_FLUX(kq) do { \
212 KQ_FLUX_WAKEUP(kq); \
213 mtx_unlock(&(kq)->kq_lock); \
214 } while (0)
215 #define KQ_UNLOCK(kq) do { \
216 mtx_unlock(&(kq)->kq_lock); \
217 } while (0)
218 #define KQ_OWNED(kq) do { \
219 mtx_assert(&(kq)->kq_lock, MA_OWNED); \
220 } while (0)
221 #define KQ_NOTOWNED(kq) do { \
222 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \
223 } while (0)
224 #define KN_LIST_LOCK(kn) do { \
225 if (kn->kn_knlist != NULL) \
226 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \
227 } while (0)
228 #define KN_LIST_UNLOCK(kn) do { \
229 if (kn->kn_knlist != NULL) \
230 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \
231 } while (0)
232 #define KNL_ASSERT_LOCK(knl, islocked) do { \
233 if (islocked) \
234 KNL_ASSERT_LOCKED(knl); \
235 else \
236 KNL_ASSERT_UNLOCKED(knl); \
237 } while (0)
238 #ifdef INVARIANTS
239 #define KNL_ASSERT_LOCKED(knl) do { \
240 knl->kl_assert_locked((knl)->kl_lockarg); \
241 } while (0)
242 #define KNL_ASSERT_UNLOCKED(knl) do { \
243 knl->kl_assert_unlocked((knl)->kl_lockarg); \
244 } while (0)
245 #else /* !INVARIANTS */
246 #define KNL_ASSERT_LOCKED(knl) do {} while(0)
247 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0)
248 #endif /* INVARIANTS */
249
250 #define KN_HASHSIZE 64 /* XXX should be tunable */
251 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
252
253 static int
filt_nullattach(struct knote * kn)254 filt_nullattach(struct knote *kn)
255 {
256
257 return (ENXIO);
258 };
259
260 struct filterops null_filtops = {
261 .f_isfd = 0,
262 .f_attach = filt_nullattach,
263 };
264
265 /* XXX - make SYSINIT to add these, and move into respective modules. */
266 extern struct filterops sig_filtops;
267 extern struct filterops fs_filtops;
268
269 /*
270 * Table for for all system-defined filters.
271 */
272 static struct mtx filterops_lock;
273 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
274 MTX_DEF);
275 static struct {
276 struct filterops *for_fop;
277 int for_refcnt;
278 } sysfilt_ops[EVFILT_SYSCOUNT] = {
279 { &file_filtops }, /* EVFILT_READ */
280 { &file_filtops }, /* EVFILT_WRITE */
281 { &null_filtops }, /* EVFILT_AIO */
282 { &file_filtops }, /* EVFILT_VNODE */
283 { &proc_filtops }, /* EVFILT_PROC */
284 { &sig_filtops }, /* EVFILT_SIGNAL */
285 { &timer_filtops }, /* EVFILT_TIMER */
286 { &null_filtops }, /* former EVFILT_NETDEV */
287 { &fs_filtops }, /* EVFILT_FS */
288 { &null_filtops }, /* EVFILT_LIO */
289 { &user_filtops }, /* EVFILT_USER */
290 };
291
292 /*
293 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
294 * method.
295 */
296 static int
filt_fileattach(struct knote * kn)297 filt_fileattach(struct knote *kn)
298 {
299
300 return (fo_kqfilter(kn->kn_fp, kn));
301 }
302
303 /*ARGSUSED*/
304 static int
kqueue_kqfilter(struct file * fp,struct knote * kn)305 kqueue_kqfilter(struct file *fp, struct knote *kn)
306 {
307 struct kqueue *kq = kn->kn_fp->f_data;
308
309 if (kn->kn_filter != EVFILT_READ)
310 return (EINVAL);
311
312 kn->kn_status |= KN_KQUEUE;
313 kn->kn_fop = &kqread_filtops;
314 knlist_add(&kq->kq_sel.si_note, kn, 0);
315
316 return (0);
317 }
318
319 static void
filt_kqdetach(struct knote * kn)320 filt_kqdetach(struct knote *kn)
321 {
322 struct kqueue *kq = kn->kn_fp->f_data;
323
324 knlist_remove(&kq->kq_sel.si_note, kn, 0);
325 }
326
327 /*ARGSUSED*/
328 static int
filt_kqueue(struct knote * kn,long hint)329 filt_kqueue(struct knote *kn, long hint)
330 {
331 struct kqueue *kq = kn->kn_fp->f_data;
332
333 kn->kn_data = kq->kq_count;
334 return (kn->kn_data > 0);
335 }
336
337 /* XXX - move to kern_proc.c? */
338 static int
filt_procattach(struct knote * kn)339 filt_procattach(struct knote *kn)
340 {
341 struct proc *p;
342 int immediate;
343 int error;
344
345 immediate = 0;
346 p = pfind(kn->kn_id);
347 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
348 p = zpfind(kn->kn_id);
349 immediate = 1;
350 } else if (p != NULL && (p->p_flag & P_WEXIT)) {
351 immediate = 1;
352 }
353
354 if (p == NULL)
355 return (ESRCH);
356 if ((error = p_cansee(curthread, p))) {
357 PROC_UNLOCK(p);
358 return (error);
359 }
360
361 kn->kn_ptr.p_proc = p;
362 kn->kn_flags |= EV_CLEAR; /* automatically set */
363
364 /*
365 * internal flag indicating registration done by kernel
366 */
367 if (kn->kn_flags & EV_FLAG1) {
368 kn->kn_data = kn->kn_sdata; /* ppid */
369 kn->kn_fflags = NOTE_CHILD;
370 kn->kn_flags &= ~EV_FLAG1;
371 }
372
373 if (immediate == 0)
374 knlist_add(&p->p_klist, kn, 1);
375
376 /*
377 * Immediately activate any exit notes if the target process is a
378 * zombie. This is necessary to handle the case where the target
379 * process, e.g. a child, dies before the kevent is registered.
380 */
381 if (immediate && filt_proc(kn, NOTE_EXIT))
382 KNOTE_ACTIVATE(kn, 0);
383
384 PROC_UNLOCK(p);
385
386 return (0);
387 }
388
389 /*
390 * The knote may be attached to a different process, which may exit,
391 * leaving nothing for the knote to be attached to. So when the process
392 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
393 * it will be deleted when read out. However, as part of the knote deletion,
394 * this routine is called, so a check is needed to avoid actually performing
395 * a detach, because the original process does not exist any more.
396 */
397 /* XXX - move to kern_proc.c? */
398 static void
filt_procdetach(struct knote * kn)399 filt_procdetach(struct knote *kn)
400 {
401 struct proc *p;
402
403 p = kn->kn_ptr.p_proc;
404 knlist_remove(&p->p_klist, kn, 0);
405 kn->kn_ptr.p_proc = NULL;
406 }
407
408 /* XXX - move to kern_proc.c? */
409 static int
filt_proc(struct knote * kn,long hint)410 filt_proc(struct knote *kn, long hint)
411 {
412 struct proc *p = kn->kn_ptr.p_proc;
413 u_int event;
414
415 /*
416 * mask off extra data
417 */
418 event = (u_int)hint & NOTE_PCTRLMASK;
419
420 /*
421 * if the user is interested in this event, record it.
422 */
423 if (kn->kn_sfflags & event)
424 kn->kn_fflags |= event;
425
426 /*
427 * process is gone, so flag the event as finished.
428 */
429 if (event == NOTE_EXIT) {
430 if (!(kn->kn_status & KN_DETACHED))
431 knlist_remove_inevent(&p->p_klist, kn);
432 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
433 kn->kn_ptr.p_proc = NULL;
434 if (kn->kn_fflags & NOTE_EXIT)
435 kn->kn_data = p->p_xstat;
436 if (kn->kn_fflags == 0)
437 kn->kn_flags |= EV_DROP;
438 return (1);
439 }
440
441 return (kn->kn_fflags != 0);
442 }
443
444 /*
445 * Called when the process forked. It mostly does the same as the
446 * knote(), activating all knotes registered to be activated when the
447 * process forked. Additionally, for each knote attached to the
448 * parent, check whether user wants to track the new process. If so
449 * attach a new knote to it, and immediately report an event with the
450 * child's pid.
451 */
452 void
knote_fork(struct knlist * list,int pid)453 knote_fork(struct knlist *list, int pid)
454 {
455 struct kqueue *kq;
456 struct knote *kn;
457 struct kevent kev;
458 int error;
459
460 if (list == NULL)
461 return;
462 list->kl_lock(list->kl_lockarg);
463
464 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
465 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
466 continue;
467 kq = kn->kn_kq;
468 KQ_LOCK(kq);
469 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
470 KQ_UNLOCK(kq);
471 continue;
472 }
473
474 /*
475 * The same as knote(), activate the event.
476 */
477 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
478 kn->kn_status |= KN_HASKQLOCK;
479 if (kn->kn_fop->f_event(kn, NOTE_FORK))
480 KNOTE_ACTIVATE(kn, 1);
481 kn->kn_status &= ~KN_HASKQLOCK;
482 KQ_UNLOCK(kq);
483 continue;
484 }
485
486 /*
487 * The NOTE_TRACK case. In addition to the activation
488 * of the event, we need to register new event to
489 * track the child. Drop the locks in preparation for
490 * the call to kqueue_register().
491 */
492 kn->kn_status |= KN_INFLUX;
493 KQ_UNLOCK(kq);
494 list->kl_unlock(list->kl_lockarg);
495
496 /*
497 * Activate existing knote and register a knote with
498 * new process.
499 */
500 kev.ident = pid;
501 kev.filter = kn->kn_filter;
502 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
503 kev.fflags = kn->kn_sfflags;
504 kev.data = kn->kn_id; /* parent */
505 kev.udata = kn->kn_kevent.udata;/* preserve udata */
506 error = kqueue_register(kq, &kev, NULL, 0);
507 if (error)
508 kn->kn_fflags |= NOTE_TRACKERR;
509 if (kn->kn_fop->f_event(kn, NOTE_FORK))
510 KNOTE_ACTIVATE(kn, 0);
511 KQ_LOCK(kq);
512 kn->kn_status &= ~KN_INFLUX;
513 KQ_UNLOCK_FLUX(kq);
514 list->kl_lock(list->kl_lockarg);
515 }
516 list->kl_unlock(list->kl_lockarg);
517 }
518
519 /*
520 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
521 * interval timer support code.
522 */
523 static int
timertoticks(intptr_t data)524 timertoticks(intptr_t data)
525 {
526 struct timeval tv;
527 int tticks;
528
529 tv.tv_sec = data / 1000;
530 tv.tv_usec = (data % 1000) * 1000;
531 tticks = tvtohz(&tv);
532
533 return tticks;
534 }
535
536 static void
filt_timerexpire(void * knx)537 filt_timerexpire(void *knx)
538 {
539 struct knote *kn = knx;
540 struct callout *calloutp;
541
542 kn->kn_data++;
543 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
544
545 /*
546 * timertoticks() uses tvtohz() which always adds 1 to allow
547 * for the time until the next clock interrupt being strictly
548 * less than 1 clock tick. We don't want that here since we
549 * want to appear to be in sync with the clock interrupt even
550 * when we're delayed.
551 */
552 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
553 calloutp = (struct callout *)kn->kn_hook;
554 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata) - 1,
555 filt_timerexpire, kn);
556 }
557 }
558
559 /*
560 * data contains amount of time to sleep, in milliseconds
561 */
562 static int
filt_timerattach(struct knote * kn)563 filt_timerattach(struct knote *kn)
564 {
565 struct callout *calloutp;
566
567 atomic_add_int(&kq_ncallouts, 1);
568
569 if (kq_ncallouts >= kq_calloutmax) {
570 atomic_add_int(&kq_ncallouts, -1);
571 return (ENOMEM);
572 }
573
574 kn->kn_flags |= EV_CLEAR; /* automatically set */
575 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */
576 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK);
577 callout_init(calloutp, CALLOUT_MPSAFE);
578 kn->kn_hook = calloutp;
579 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata),
580 filt_timerexpire, kn);
581
582 return (0);
583 }
584
585 static void
filt_timerdetach(struct knote * kn)586 filt_timerdetach(struct knote *kn)
587 {
588 struct callout *calloutp;
589
590 calloutp = (struct callout *)kn->kn_hook;
591 callout_drain(calloutp);
592 free(calloutp, M_KQUEUE);
593 atomic_add_int(&kq_ncallouts, -1);
594 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */
595 }
596
597 static int
filt_timer(struct knote * kn,long hint)598 filt_timer(struct knote *kn, long hint)
599 {
600
601 return (kn->kn_data != 0);
602 }
603
604 static int
filt_userattach(struct knote * kn)605 filt_userattach(struct knote *kn)
606 {
607
608 /*
609 * EVFILT_USER knotes are not attached to anything in the kernel.
610 */
611 kn->kn_hook = NULL;
612 if (kn->kn_fflags & NOTE_TRIGGER)
613 kn->kn_hookid = 1;
614 else
615 kn->kn_hookid = 0;
616 return (0);
617 }
618
619 static void
filt_userdetach(__unused struct knote * kn)620 filt_userdetach(__unused struct knote *kn)
621 {
622
623 /*
624 * EVFILT_USER knotes are not attached to anything in the kernel.
625 */
626 }
627
628 static int
filt_user(struct knote * kn,__unused long hint)629 filt_user(struct knote *kn, __unused long hint)
630 {
631
632 return (kn->kn_hookid);
633 }
634
635 static void
filt_usertouch(struct knote * kn,struct kevent * kev,u_long type)636 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
637 {
638 u_int ffctrl;
639
640 switch (type) {
641 case EVENT_REGISTER:
642 if (kev->fflags & NOTE_TRIGGER)
643 kn->kn_hookid = 1;
644
645 ffctrl = kev->fflags & NOTE_FFCTRLMASK;
646 kev->fflags &= NOTE_FFLAGSMASK;
647 switch (ffctrl) {
648 case NOTE_FFNOP:
649 break;
650
651 case NOTE_FFAND:
652 kn->kn_sfflags &= kev->fflags;
653 break;
654
655 case NOTE_FFOR:
656 kn->kn_sfflags |= kev->fflags;
657 break;
658
659 case NOTE_FFCOPY:
660 kn->kn_sfflags = kev->fflags;
661 break;
662
663 default:
664 /* XXX Return error? */
665 break;
666 }
667 kn->kn_sdata = kev->data;
668 if (kev->flags & EV_CLEAR) {
669 kn->kn_hookid = 0;
670 kn->kn_data = 0;
671 kn->kn_fflags = 0;
672 }
673 break;
674
675 case EVENT_PROCESS:
676 *kev = kn->kn_kevent;
677 kev->fflags = kn->kn_sfflags;
678 kev->data = kn->kn_sdata;
679 if (kn->kn_flags & EV_CLEAR) {
680 kn->kn_hookid = 0;
681 kn->kn_data = 0;
682 kn->kn_fflags = 0;
683 }
684 break;
685
686 default:
687 panic("filt_usertouch() - invalid type (%ld)", type);
688 break;
689 }
690 }
691
692 int
sys_kqueue(struct thread * td,struct kqueue_args * uap)693 sys_kqueue(struct thread *td, struct kqueue_args *uap)
694 {
695 struct filedesc *fdp;
696 struct kqueue *kq;
697 struct file *fp;
698 int fd, error;
699
700 fdp = td->td_proc->p_fd;
701 error = falloc(td, &fp, &fd, 0);
702 if (error)
703 goto done2;
704
705 /* An extra reference on `nfp' has been held for us by falloc(). */
706 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
707 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
708 TAILQ_INIT(&kq->kq_head);
709 kq->kq_fdp = fdp;
710 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
711 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
712
713 FILEDESC_XLOCK(fdp);
714 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
715 FILEDESC_XUNLOCK(fdp);
716
717 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
718 fdrop(fp, td);
719
720 td->td_retval[0] = fd;
721 done2:
722 return (error);
723 }
724
725 #ifndef _SYS_SYSPROTO_H_
726 struct kevent_args {
727 int fd;
728 const struct kevent *changelist;
729 int nchanges;
730 struct kevent *eventlist;
731 int nevents;
732 const struct timespec *timeout;
733 };
734 #endif
735 int
sys_kevent(struct thread * td,struct kevent_args * uap)736 sys_kevent(struct thread *td, struct kevent_args *uap)
737 {
738 struct timespec ts, *tsp;
739 struct kevent_copyops k_ops = { uap,
740 kevent_copyout,
741 kevent_copyin};
742 int error;
743 #ifdef KTRACE
744 struct uio ktruio;
745 struct iovec ktriov;
746 struct uio *ktruioin = NULL;
747 struct uio *ktruioout = NULL;
748 #endif
749
750 if (uap->timeout != NULL) {
751 error = copyin(uap->timeout, &ts, sizeof(ts));
752 if (error)
753 return (error);
754 tsp = &ts;
755 } else
756 tsp = NULL;
757
758 #ifdef KTRACE
759 if (KTRPOINT(td, KTR_GENIO)) {
760 ktriov.iov_base = uap->changelist;
761 ktriov.iov_len = uap->nchanges * sizeof(struct kevent);
762 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1,
763 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ,
764 .uio_td = td };
765 ktruioin = cloneuio(&ktruio);
766 ktriov.iov_base = uap->eventlist;
767 ktriov.iov_len = uap->nevents * sizeof(struct kevent);
768 ktruioout = cloneuio(&ktruio);
769 }
770 #endif
771
772 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
773 &k_ops, tsp);
774
775 #ifdef KTRACE
776 if (ktruioin != NULL) {
777 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent);
778 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0);
779 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent);
780 ktrgenio(uap->fd, UIO_READ, ktruioout, error);
781 }
782 #endif
783
784 return (error);
785 }
786
787 /*
788 * Copy 'count' items into the destination list pointed to by uap->eventlist.
789 */
790 static int
kevent_copyout(void * arg,struct kevent * kevp,int count)791 kevent_copyout(void *arg, struct kevent *kevp, int count)
792 {
793 struct kevent_args *uap;
794 int error;
795
796 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
797 uap = (struct kevent_args *)arg;
798
799 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
800 if (error == 0)
801 uap->eventlist += count;
802 return (error);
803 }
804
805 /*
806 * Copy 'count' items from the list pointed to by uap->changelist.
807 */
808 static int
kevent_copyin(void * arg,struct kevent * kevp,int count)809 kevent_copyin(void *arg, struct kevent *kevp, int count)
810 {
811 struct kevent_args *uap;
812 int error;
813
814 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
815 uap = (struct kevent_args *)arg;
816
817 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
818 if (error == 0)
819 uap->changelist += count;
820 return (error);
821 }
822
823 int
kern_kevent(struct thread * td,int fd,int nchanges,int nevents,struct kevent_copyops * k_ops,const struct timespec * timeout)824 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
825 struct kevent_copyops *k_ops, const struct timespec *timeout)
826 {
827 struct kevent keva[KQ_NEVENTS];
828 struct kevent *kevp, *changes;
829 struct kqueue *kq;
830 struct file *fp;
831 int i, n, nerrors, error;
832
833 if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0)
834 return (error);
835 if ((error = kqueue_acquire(fp, &kq)) != 0)
836 goto done_norel;
837
838 nerrors = 0;
839
840 while (nchanges > 0) {
841 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
842 error = k_ops->k_copyin(k_ops->arg, keva, n);
843 if (error)
844 goto done;
845 changes = keva;
846 for (i = 0; i < n; i++) {
847 kevp = &changes[i];
848 if (!kevp->filter)
849 continue;
850 kevp->flags &= ~EV_SYSFLAGS;
851 error = kqueue_register(kq, kevp, td, 1);
852 if (error || (kevp->flags & EV_RECEIPT)) {
853 if (nevents != 0) {
854 kevp->flags = EV_ERROR;
855 kevp->data = error;
856 (void) k_ops->k_copyout(k_ops->arg,
857 kevp, 1);
858 nevents--;
859 nerrors++;
860 } else {
861 goto done;
862 }
863 }
864 }
865 nchanges -= n;
866 }
867 if (nerrors) {
868 td->td_retval[0] = nerrors;
869 error = 0;
870 goto done;
871 }
872
873 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
874 done:
875 kqueue_release(kq, 0);
876 done_norel:
877 fdrop(fp, td);
878 return (error);
879 }
880
881 int
kqueue_add_filteropts(int filt,struct filterops * filtops)882 kqueue_add_filteropts(int filt, struct filterops *filtops)
883 {
884 int error;
885
886 error = 0;
887 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
888 printf(
889 "trying to add a filterop that is out of range: %d is beyond %d\n",
890 ~filt, EVFILT_SYSCOUNT);
891 return EINVAL;
892 }
893 mtx_lock(&filterops_lock);
894 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
895 sysfilt_ops[~filt].for_fop != NULL)
896 error = EEXIST;
897 else {
898 sysfilt_ops[~filt].for_fop = filtops;
899 sysfilt_ops[~filt].for_refcnt = 0;
900 }
901 mtx_unlock(&filterops_lock);
902
903 return (error);
904 }
905
906 int
kqueue_del_filteropts(int filt)907 kqueue_del_filteropts(int filt)
908 {
909 int error;
910
911 error = 0;
912 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
913 return EINVAL;
914
915 mtx_lock(&filterops_lock);
916 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
917 sysfilt_ops[~filt].for_fop == NULL)
918 error = EINVAL;
919 else if (sysfilt_ops[~filt].for_refcnt != 0)
920 error = EBUSY;
921 else {
922 sysfilt_ops[~filt].for_fop = &null_filtops;
923 sysfilt_ops[~filt].for_refcnt = 0;
924 }
925 mtx_unlock(&filterops_lock);
926
927 return error;
928 }
929
930 static struct filterops *
kqueue_fo_find(int filt)931 kqueue_fo_find(int filt)
932 {
933
934 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
935 return NULL;
936
937 mtx_lock(&filterops_lock);
938 sysfilt_ops[~filt].for_refcnt++;
939 if (sysfilt_ops[~filt].for_fop == NULL)
940 sysfilt_ops[~filt].for_fop = &null_filtops;
941 mtx_unlock(&filterops_lock);
942
943 return sysfilt_ops[~filt].for_fop;
944 }
945
946 static void
kqueue_fo_release(int filt)947 kqueue_fo_release(int filt)
948 {
949
950 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
951 return;
952
953 mtx_lock(&filterops_lock);
954 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
955 ("filter object refcount not valid on release"));
956 sysfilt_ops[~filt].for_refcnt--;
957 mtx_unlock(&filterops_lock);
958 }
959
960 /*
961 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will
962 * influence if memory allocation should wait. Make sure it is 0 if you
963 * hold any mutexes.
964 */
965 static int
kqueue_register(struct kqueue * kq,struct kevent * kev,struct thread * td,int waitok)966 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
967 {
968 struct filterops *fops;
969 struct file *fp;
970 struct knote *kn, *tkn;
971 int error, filt, event;
972 int haskqglobal, filedesc_unlock;
973
974 fp = NULL;
975 kn = NULL;
976 error = 0;
977 haskqglobal = 0;
978 filedesc_unlock = 0;
979
980 filt = kev->filter;
981 fops = kqueue_fo_find(filt);
982 if (fops == NULL)
983 return EINVAL;
984
985 tkn = knote_alloc(waitok); /* prevent waiting with locks */
986
987 findkn:
988 if (fops->f_isfd) {
989 KASSERT(td != NULL, ("td is NULL"));
990 if (kev->ident > INT_MAX)
991 error = EBADF;
992 else
993 error = fget(td, kev->ident, CAP_POLL_EVENT, &fp);
994 if (error)
995 goto done;
996
997 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
998 kev->ident, 0) != 0) {
999 /* try again */
1000 fdrop(fp, td);
1001 fp = NULL;
1002 error = kqueue_expand(kq, fops, kev->ident, waitok);
1003 if (error)
1004 goto done;
1005 goto findkn;
1006 }
1007
1008 if (fp->f_type == DTYPE_KQUEUE) {
1009 /*
1010 * if we add some inteligence about what we are doing,
1011 * we should be able to support events on ourselves.
1012 * We need to know when we are doing this to prevent
1013 * getting both the knlist lock and the kq lock since
1014 * they are the same thing.
1015 */
1016 if (fp->f_data == kq) {
1017 error = EINVAL;
1018 goto done;
1019 }
1020
1021 /*
1022 * Pre-lock the filedesc before the global
1023 * lock mutex, see the comment in
1024 * kqueue_close().
1025 */
1026 FILEDESC_XLOCK(td->td_proc->p_fd);
1027 filedesc_unlock = 1;
1028 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1029 }
1030
1031 KQ_LOCK(kq);
1032 if (kev->ident < kq->kq_knlistsize) {
1033 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
1034 if (kev->filter == kn->kn_filter)
1035 break;
1036 }
1037 } else {
1038 if ((kev->flags & EV_ADD) == EV_ADD)
1039 kqueue_expand(kq, fops, kev->ident, waitok);
1040
1041 KQ_LOCK(kq);
1042 if (kq->kq_knhashmask != 0) {
1043 struct klist *list;
1044
1045 list = &kq->kq_knhash[
1046 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1047 SLIST_FOREACH(kn, list, kn_link)
1048 if (kev->ident == kn->kn_id &&
1049 kev->filter == kn->kn_filter)
1050 break;
1051 }
1052 }
1053
1054 /* knote is in the process of changing, wait for it to stablize. */
1055 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1056 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1057 if (filedesc_unlock) {
1058 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1059 filedesc_unlock = 0;
1060 }
1061 kq->kq_state |= KQ_FLUXWAIT;
1062 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1063 if (fp != NULL) {
1064 fdrop(fp, td);
1065 fp = NULL;
1066 }
1067 goto findkn;
1068 }
1069
1070 /*
1071 * kn now contains the matching knote, or NULL if no match
1072 */
1073 if (kn == NULL) {
1074 if (kev->flags & EV_ADD) {
1075 kn = tkn;
1076 tkn = NULL;
1077 if (kn == NULL) {
1078 KQ_UNLOCK(kq);
1079 error = ENOMEM;
1080 goto done;
1081 }
1082 kn->kn_fp = fp;
1083 kn->kn_kq = kq;
1084 kn->kn_fop = fops;
1085 /*
1086 * apply reference counts to knote structure, and
1087 * do not release it at the end of this routine.
1088 */
1089 fops = NULL;
1090 fp = NULL;
1091
1092 kn->kn_sfflags = kev->fflags;
1093 kn->kn_sdata = kev->data;
1094 kev->fflags = 0;
1095 kev->data = 0;
1096 kn->kn_kevent = *kev;
1097 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1098 EV_ENABLE | EV_DISABLE);
1099 kn->kn_status = KN_INFLUX|KN_DETACHED;
1100
1101 error = knote_attach(kn, kq);
1102 KQ_UNLOCK(kq);
1103 if (error != 0) {
1104 tkn = kn;
1105 goto done;
1106 }
1107
1108 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1109 knote_drop(kn, td);
1110 goto done;
1111 }
1112 KN_LIST_LOCK(kn);
1113 goto done_ev_add;
1114 } else {
1115 /* No matching knote and the EV_ADD flag is not set. */
1116 KQ_UNLOCK(kq);
1117 error = ENOENT;
1118 goto done;
1119 }
1120 }
1121
1122 if (kev->flags & EV_DELETE) {
1123 kn->kn_status |= KN_INFLUX;
1124 KQ_UNLOCK(kq);
1125 if (!(kn->kn_status & KN_DETACHED))
1126 kn->kn_fop->f_detach(kn);
1127 knote_drop(kn, td);
1128 goto done;
1129 }
1130
1131 /*
1132 * The user may change some filter values after the initial EV_ADD,
1133 * but doing so will not reset any filter which has already been
1134 * triggered.
1135 */
1136 kn->kn_status |= KN_INFLUX | KN_SCAN;
1137 KQ_UNLOCK(kq);
1138 KN_LIST_LOCK(kn);
1139 kn->kn_kevent.udata = kev->udata;
1140 if (!fops->f_isfd && fops->f_touch != NULL) {
1141 fops->f_touch(kn, kev, EVENT_REGISTER);
1142 } else {
1143 kn->kn_sfflags = kev->fflags;
1144 kn->kn_sdata = kev->data;
1145 }
1146
1147 /*
1148 * We can get here with kn->kn_knlist == NULL. This can happen when
1149 * the initial attach event decides that the event is "completed"
1150 * already. i.e. filt_procattach is called on a zombie process. It
1151 * will call filt_proc which will remove it from the list, and NULL
1152 * kn_knlist.
1153 */
1154 done_ev_add:
1155 event = kn->kn_fop->f_event(kn, 0);
1156 KQ_LOCK(kq);
1157 if (event)
1158 KNOTE_ACTIVATE(kn, 1);
1159 kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1160 KN_LIST_UNLOCK(kn);
1161
1162 if ((kev->flags & EV_DISABLE) &&
1163 ((kn->kn_status & KN_DISABLED) == 0)) {
1164 kn->kn_status |= KN_DISABLED;
1165 }
1166
1167 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
1168 kn->kn_status &= ~KN_DISABLED;
1169 if ((kn->kn_status & KN_ACTIVE) &&
1170 ((kn->kn_status & KN_QUEUED) == 0))
1171 knote_enqueue(kn);
1172 }
1173 KQ_UNLOCK_FLUX(kq);
1174
1175 done:
1176 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1177 if (filedesc_unlock)
1178 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1179 if (fp != NULL)
1180 fdrop(fp, td);
1181 if (tkn != NULL)
1182 knote_free(tkn);
1183 if (fops != NULL)
1184 kqueue_fo_release(filt);
1185 return (error);
1186 }
1187
1188 static int
kqueue_acquire(struct file * fp,struct kqueue ** kqp)1189 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1190 {
1191 int error;
1192 struct kqueue *kq;
1193
1194 error = 0;
1195
1196 kq = fp->f_data;
1197 if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1198 return (EBADF);
1199 *kqp = kq;
1200 KQ_LOCK(kq);
1201 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1202 KQ_UNLOCK(kq);
1203 return (EBADF);
1204 }
1205 kq->kq_refcnt++;
1206 KQ_UNLOCK(kq);
1207
1208 return error;
1209 }
1210
1211 static void
kqueue_release(struct kqueue * kq,int locked)1212 kqueue_release(struct kqueue *kq, int locked)
1213 {
1214 if (locked)
1215 KQ_OWNED(kq);
1216 else
1217 KQ_LOCK(kq);
1218 kq->kq_refcnt--;
1219 if (kq->kq_refcnt == 1)
1220 wakeup(&kq->kq_refcnt);
1221 if (!locked)
1222 KQ_UNLOCK(kq);
1223 }
1224
1225 static void
kqueue_schedtask(struct kqueue * kq)1226 kqueue_schedtask(struct kqueue *kq)
1227 {
1228
1229 KQ_OWNED(kq);
1230 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1231 ("scheduling kqueue task while draining"));
1232
1233 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1234 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1235 kq->kq_state |= KQ_TASKSCHED;
1236 }
1237 }
1238
1239 /*
1240 * Expand the kq to make sure we have storage for fops/ident pair.
1241 *
1242 * Return 0 on success (or no work necessary), return errno on failure.
1243 *
1244 * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1245 * If kqueue_register is called from a non-fd context, there usually/should
1246 * be no locks held.
1247 */
1248 static int
kqueue_expand(struct kqueue * kq,struct filterops * fops,uintptr_t ident,int waitok)1249 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1250 int waitok)
1251 {
1252 struct klist *list, *tmp_knhash, *to_free;
1253 u_long tmp_knhashmask;
1254 int size;
1255 int fd;
1256 int mflag = waitok ? M_WAITOK : M_NOWAIT;
1257
1258 KQ_NOTOWNED(kq);
1259
1260 to_free = NULL;
1261 if (fops->f_isfd) {
1262 fd = ident;
1263 if (kq->kq_knlistsize <= fd) {
1264 size = kq->kq_knlistsize;
1265 while (size <= fd)
1266 size += KQEXTENT;
1267 list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
1268 if (list == NULL)
1269 return ENOMEM;
1270 KQ_LOCK(kq);
1271 if (kq->kq_knlistsize > fd) {
1272 to_free = list;
1273 list = NULL;
1274 } else {
1275 if (kq->kq_knlist != NULL) {
1276 bcopy(kq->kq_knlist, list,
1277 kq->kq_knlistsize * sizeof(*list));
1278 to_free = kq->kq_knlist;
1279 kq->kq_knlist = NULL;
1280 }
1281 bzero((caddr_t)list +
1282 kq->kq_knlistsize * sizeof(*list),
1283 (size - kq->kq_knlistsize) * sizeof(*list));
1284 kq->kq_knlistsize = size;
1285 kq->kq_knlist = list;
1286 }
1287 KQ_UNLOCK(kq);
1288 }
1289 } else {
1290 if (kq->kq_knhashmask == 0) {
1291 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1292 &tmp_knhashmask);
1293 if (tmp_knhash == NULL)
1294 return ENOMEM;
1295 KQ_LOCK(kq);
1296 if (kq->kq_knhashmask == 0) {
1297 kq->kq_knhash = tmp_knhash;
1298 kq->kq_knhashmask = tmp_knhashmask;
1299 } else {
1300 to_free = tmp_knhash;
1301 }
1302 KQ_UNLOCK(kq);
1303 }
1304 }
1305 free(to_free, M_KQUEUE);
1306
1307 KQ_NOTOWNED(kq);
1308 return 0;
1309 }
1310
1311 static void
kqueue_task(void * arg,int pending)1312 kqueue_task(void *arg, int pending)
1313 {
1314 struct kqueue *kq;
1315 int haskqglobal;
1316
1317 haskqglobal = 0;
1318 kq = arg;
1319
1320 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1321 KQ_LOCK(kq);
1322
1323 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1324
1325 kq->kq_state &= ~KQ_TASKSCHED;
1326 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1327 wakeup(&kq->kq_state);
1328 }
1329 KQ_UNLOCK(kq);
1330 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1331 }
1332
1333 /*
1334 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1335 * We treat KN_MARKER knotes as if they are INFLUX.
1336 */
1337 static int
kqueue_scan(struct kqueue * kq,int maxevents,struct kevent_copyops * k_ops,const struct timespec * tsp,struct kevent * keva,struct thread * td)1338 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1339 const struct timespec *tsp, struct kevent *keva, struct thread *td)
1340 {
1341 struct kevent *kevp;
1342 struct timeval atv, rtv, ttv;
1343 struct knote *kn, *marker;
1344 int count, timeout, nkev, error, influx;
1345 int haskqglobal, touch;
1346
1347 count = maxevents;
1348 nkev = 0;
1349 error = 0;
1350 haskqglobal = 0;
1351
1352 if (maxevents == 0)
1353 goto done_nl;
1354
1355 if (tsp != NULL) {
1356 TIMESPEC_TO_TIMEVAL(&atv, tsp);
1357 if (itimerfix(&atv)) {
1358 error = EINVAL;
1359 goto done_nl;
1360 }
1361 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1362 timeout = -1;
1363 else
1364 timeout = atv.tv_sec > 24 * 60 * 60 ?
1365 24 * 60 * 60 * hz : tvtohz(&atv);
1366 getmicrouptime(&rtv);
1367 timevaladd(&atv, &rtv);
1368 } else {
1369 atv.tv_sec = 0;
1370 atv.tv_usec = 0;
1371 timeout = 0;
1372 }
1373 marker = knote_alloc(1);
1374 if (marker == NULL) {
1375 error = ENOMEM;
1376 goto done_nl;
1377 }
1378 marker->kn_status = KN_MARKER;
1379 KQ_LOCK(kq);
1380 goto start;
1381
1382 retry:
1383 if (atv.tv_sec || atv.tv_usec) {
1384 getmicrouptime(&rtv);
1385 if (timevalcmp(&rtv, &atv, >=))
1386 goto done;
1387 ttv = atv;
1388 timevalsub(&ttv, &rtv);
1389 timeout = ttv.tv_sec > 24 * 60 * 60 ?
1390 24 * 60 * 60 * hz : tvtohz(&ttv);
1391 }
1392
1393 start:
1394 kevp = keva;
1395 if (kq->kq_count == 0) {
1396 if (timeout < 0) {
1397 error = EWOULDBLOCK;
1398 } else {
1399 kq->kq_state |= KQ_SLEEP;
1400 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH,
1401 "kqread", timeout);
1402 }
1403 if (error == 0)
1404 goto retry;
1405 /* don't restart after signals... */
1406 if (error == ERESTART)
1407 error = EINTR;
1408 else if (error == EWOULDBLOCK)
1409 error = 0;
1410 goto done;
1411 }
1412
1413 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1414 influx = 0;
1415 while (count) {
1416 KQ_OWNED(kq);
1417 kn = TAILQ_FIRST(&kq->kq_head);
1418
1419 if ((kn->kn_status == KN_MARKER && kn != marker) ||
1420 (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1421 if (influx) {
1422 influx = 0;
1423 KQ_FLUX_WAKEUP(kq);
1424 }
1425 kq->kq_state |= KQ_FLUXWAIT;
1426 error = msleep(kq, &kq->kq_lock, PSOCK,
1427 "kqflxwt", 0);
1428 continue;
1429 }
1430
1431 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1432 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1433 kn->kn_status &= ~KN_QUEUED;
1434 kq->kq_count--;
1435 continue;
1436 }
1437 if (kn == marker) {
1438 KQ_FLUX_WAKEUP(kq);
1439 if (count == maxevents)
1440 goto retry;
1441 goto done;
1442 }
1443 KASSERT((kn->kn_status & KN_INFLUX) == 0,
1444 ("KN_INFLUX set when not suppose to be"));
1445
1446 if ((kn->kn_flags & EV_DROP) == EV_DROP) {
1447 kn->kn_status &= ~KN_QUEUED;
1448 kn->kn_status |= KN_INFLUX;
1449 kq->kq_count--;
1450 KQ_UNLOCK(kq);
1451 /*
1452 * We don't need to lock the list since we've marked
1453 * it _INFLUX.
1454 */
1455 if (!(kn->kn_status & KN_DETACHED))
1456 kn->kn_fop->f_detach(kn);
1457 knote_drop(kn, td);
1458 KQ_LOCK(kq);
1459 continue;
1460 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1461 kn->kn_status &= ~KN_QUEUED;
1462 kn->kn_status |= KN_INFLUX;
1463 kq->kq_count--;
1464 KQ_UNLOCK(kq);
1465 /*
1466 * We don't need to lock the list since we've marked
1467 * it _INFLUX.
1468 */
1469 *kevp = kn->kn_kevent;
1470 if (!(kn->kn_status & KN_DETACHED))
1471 kn->kn_fop->f_detach(kn);
1472 knote_drop(kn, td);
1473 KQ_LOCK(kq);
1474 kn = NULL;
1475 } else {
1476 kn->kn_status |= KN_INFLUX | KN_SCAN;
1477 KQ_UNLOCK(kq);
1478 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1479 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1480 KN_LIST_LOCK(kn);
1481 if (kn->kn_fop->f_event(kn, 0) == 0) {
1482 KQ_LOCK(kq);
1483 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1484 kn->kn_status &=
1485 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX |
1486 KN_SCAN);
1487 kq->kq_count--;
1488 KN_LIST_UNLOCK(kn);
1489 influx = 1;
1490 continue;
1491 }
1492 touch = (!kn->kn_fop->f_isfd &&
1493 kn->kn_fop->f_touch != NULL);
1494 if (touch)
1495 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
1496 else
1497 *kevp = kn->kn_kevent;
1498 KQ_LOCK(kq);
1499 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1500 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
1501 /*
1502 * Manually clear knotes who weren't
1503 * 'touch'ed.
1504 */
1505 if (touch == 0 && kn->kn_flags & EV_CLEAR) {
1506 kn->kn_data = 0;
1507 kn->kn_fflags = 0;
1508 }
1509 if (kn->kn_flags & EV_DISPATCH)
1510 kn->kn_status |= KN_DISABLED;
1511 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1512 kq->kq_count--;
1513 } else
1514 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1515
1516 kn->kn_status &= ~(KN_INFLUX | KN_SCAN);
1517 KN_LIST_UNLOCK(kn);
1518 influx = 1;
1519 }
1520
1521 /* we are returning a copy to the user */
1522 kevp++;
1523 nkev++;
1524 count--;
1525
1526 if (nkev == KQ_NEVENTS) {
1527 influx = 0;
1528 KQ_UNLOCK_FLUX(kq);
1529 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1530 nkev = 0;
1531 kevp = keva;
1532 KQ_LOCK(kq);
1533 if (error)
1534 break;
1535 }
1536 }
1537 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1538 done:
1539 KQ_OWNED(kq);
1540 KQ_UNLOCK_FLUX(kq);
1541 knote_free(marker);
1542 done_nl:
1543 KQ_NOTOWNED(kq);
1544 if (nkev != 0)
1545 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1546 td->td_retval[0] = maxevents - count;
1547 return (error);
1548 }
1549
1550 /*
1551 * XXX
1552 * This could be expanded to call kqueue_scan, if desired.
1553 */
1554 /*ARGSUSED*/
1555 static int
kqueue_read(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)1556 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1557 int flags, struct thread *td)
1558 {
1559 return (ENXIO);
1560 }
1561
1562 /*ARGSUSED*/
1563 static int
kqueue_write(struct file * fp,struct uio * uio,struct ucred * active_cred,int flags,struct thread * td)1564 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1565 int flags, struct thread *td)
1566 {
1567 return (ENXIO);
1568 }
1569
1570 /*ARGSUSED*/
1571 static int
kqueue_truncate(struct file * fp,off_t length,struct ucred * active_cred,struct thread * td)1572 kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1573 struct thread *td)
1574 {
1575
1576 return (EINVAL);
1577 }
1578
1579 /*ARGSUSED*/
1580 static int
kqueue_ioctl(struct file * fp,u_long cmd,void * data,struct ucred * active_cred,struct thread * td)1581 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1582 struct ucred *active_cred, struct thread *td)
1583 {
1584 /*
1585 * Enabling sigio causes two major problems:
1586 * 1) infinite recursion:
1587 * Synopsys: kevent is being used to track signals and have FIOASYNC
1588 * set. On receipt of a signal this will cause a kqueue to recurse
1589 * into itself over and over. Sending the sigio causes the kqueue
1590 * to become ready, which in turn posts sigio again, forever.
1591 * Solution: this can be solved by setting a flag in the kqueue that
1592 * we have a SIGIO in progress.
1593 * 2) locking problems:
1594 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1595 * us above the proc and pgrp locks.
1596 * Solution: Post a signal using an async mechanism, being sure to
1597 * record a generation count in the delivery so that we do not deliver
1598 * a signal to the wrong process.
1599 *
1600 * Note, these two mechanisms are somewhat mutually exclusive!
1601 */
1602 #if 0
1603 struct kqueue *kq;
1604
1605 kq = fp->f_data;
1606 switch (cmd) {
1607 case FIOASYNC:
1608 if (*(int *)data) {
1609 kq->kq_state |= KQ_ASYNC;
1610 } else {
1611 kq->kq_state &= ~KQ_ASYNC;
1612 }
1613 return (0);
1614
1615 case FIOSETOWN:
1616 return (fsetown(*(int *)data, &kq->kq_sigio));
1617
1618 case FIOGETOWN:
1619 *(int *)data = fgetown(&kq->kq_sigio);
1620 return (0);
1621 }
1622 #endif
1623
1624 return (ENOTTY);
1625 }
1626
1627 /*ARGSUSED*/
1628 static int
kqueue_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)1629 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1630 struct thread *td)
1631 {
1632 struct kqueue *kq;
1633 int revents = 0;
1634 int error;
1635
1636 if ((error = kqueue_acquire(fp, &kq)))
1637 return POLLERR;
1638
1639 KQ_LOCK(kq);
1640 if (events & (POLLIN | POLLRDNORM)) {
1641 if (kq->kq_count) {
1642 revents |= events & (POLLIN | POLLRDNORM);
1643 } else {
1644 selrecord(td, &kq->kq_sel);
1645 if (SEL_WAITING(&kq->kq_sel))
1646 kq->kq_state |= KQ_SEL;
1647 }
1648 }
1649 kqueue_release(kq, 1);
1650 KQ_UNLOCK(kq);
1651 return (revents);
1652 }
1653
1654 /*ARGSUSED*/
1655 static int
kqueue_stat(struct file * fp,struct stat * st,struct ucred * active_cred,struct thread * td)1656 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1657 struct thread *td)
1658 {
1659
1660 bzero((void *)st, sizeof *st);
1661 /*
1662 * We no longer return kq_count because the unlocked value is useless.
1663 * If you spent all this time getting the count, why not spend your
1664 * syscall better by calling kevent?
1665 *
1666 * XXX - This is needed for libc_r.
1667 */
1668 st->st_mode = S_IFIFO;
1669 return (0);
1670 }
1671
1672 /*ARGSUSED*/
1673 static int
kqueue_close(struct file * fp,struct thread * td)1674 kqueue_close(struct file *fp, struct thread *td)
1675 {
1676 struct kqueue *kq = fp->f_data;
1677 struct filedesc *fdp;
1678 struct knote *kn;
1679 int i;
1680 int error;
1681 int filedesc_unlock;
1682
1683 if ((error = kqueue_acquire(fp, &kq)))
1684 return error;
1685
1686 filedesc_unlock = 0;
1687 KQ_LOCK(kq);
1688
1689 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1690 ("kqueue already closing"));
1691 kq->kq_state |= KQ_CLOSING;
1692 if (kq->kq_refcnt > 1)
1693 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1694
1695 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1696 fdp = kq->kq_fdp;
1697
1698 KASSERT(knlist_empty(&kq->kq_sel.si_note),
1699 ("kqueue's knlist not empty"));
1700
1701 for (i = 0; i < kq->kq_knlistsize; i++) {
1702 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1703 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1704 kq->kq_state |= KQ_FLUXWAIT;
1705 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1706 continue;
1707 }
1708 kn->kn_status |= KN_INFLUX;
1709 KQ_UNLOCK(kq);
1710 if (!(kn->kn_status & KN_DETACHED))
1711 kn->kn_fop->f_detach(kn);
1712 knote_drop(kn, td);
1713 KQ_LOCK(kq);
1714 }
1715 }
1716 if (kq->kq_knhashmask != 0) {
1717 for (i = 0; i <= kq->kq_knhashmask; i++) {
1718 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1719 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1720 kq->kq_state |= KQ_FLUXWAIT;
1721 msleep(kq, &kq->kq_lock, PSOCK,
1722 "kqclo2", 0);
1723 continue;
1724 }
1725 kn->kn_status |= KN_INFLUX;
1726 KQ_UNLOCK(kq);
1727 if (!(kn->kn_status & KN_DETACHED))
1728 kn->kn_fop->f_detach(kn);
1729 knote_drop(kn, td);
1730 KQ_LOCK(kq);
1731 }
1732 }
1733 }
1734
1735 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1736 kq->kq_state |= KQ_TASKDRAIN;
1737 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1738 }
1739
1740 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1741 selwakeuppri(&kq->kq_sel, PSOCK);
1742 if (!SEL_WAITING(&kq->kq_sel))
1743 kq->kq_state &= ~KQ_SEL;
1744 }
1745
1746 KQ_UNLOCK(kq);
1747
1748 /*
1749 * We could be called due to the knote_drop() doing fdrop(),
1750 * called from kqueue_register(). In this case the global
1751 * lock is owned, and filedesc sx is locked before, to not
1752 * take the sleepable lock after non-sleepable.
1753 */
1754 if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
1755 FILEDESC_XLOCK(fdp);
1756 filedesc_unlock = 1;
1757 } else
1758 filedesc_unlock = 0;
1759 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
1760 if (filedesc_unlock)
1761 FILEDESC_XUNLOCK(fdp);
1762
1763 seldrain(&kq->kq_sel);
1764 knlist_destroy(&kq->kq_sel.si_note);
1765 mtx_destroy(&kq->kq_lock);
1766 kq->kq_fdp = NULL;
1767
1768 if (kq->kq_knhash != NULL)
1769 free(kq->kq_knhash, M_KQUEUE);
1770 if (kq->kq_knlist != NULL)
1771 free(kq->kq_knlist, M_KQUEUE);
1772
1773 funsetown(&kq->kq_sigio);
1774 free(kq, M_KQUEUE);
1775 fp->f_data = NULL;
1776
1777 return (0);
1778 }
1779
1780 static void
kqueue_wakeup(struct kqueue * kq)1781 kqueue_wakeup(struct kqueue *kq)
1782 {
1783 KQ_OWNED(kq);
1784
1785 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1786 kq->kq_state &= ~KQ_SLEEP;
1787 wakeup(kq);
1788 }
1789 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1790 selwakeuppri(&kq->kq_sel, PSOCK);
1791 if (!SEL_WAITING(&kq->kq_sel))
1792 kq->kq_state &= ~KQ_SEL;
1793 }
1794 if (!knlist_empty(&kq->kq_sel.si_note))
1795 kqueue_schedtask(kq);
1796 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1797 pgsigio(&kq->kq_sigio, SIGIO, 0);
1798 }
1799 }
1800
1801 /*
1802 * Walk down a list of knotes, activating them if their event has triggered.
1803 *
1804 * There is a possibility to optimize in the case of one kq watching another.
1805 * Instead of scheduling a task to wake it up, you could pass enough state
1806 * down the chain to make up the parent kqueue. Make this code functional
1807 * first.
1808 */
1809 void
knote(struct knlist * list,long hint,int lockflags)1810 knote(struct knlist *list, long hint, int lockflags)
1811 {
1812 struct kqueue *kq;
1813 struct knote *kn;
1814 int error;
1815
1816 if (list == NULL)
1817 return;
1818
1819 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
1820
1821 if ((lockflags & KNF_LISTLOCKED) == 0)
1822 list->kl_lock(list->kl_lockarg);
1823
1824 /*
1825 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1826 * the kqueue scheduling, but this will introduce four
1827 * lock/unlock's for each knote to test. If we do, continue to use
1828 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1829 * only safe if you want to remove the current item, which we are
1830 * not doing.
1831 */
1832 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1833 kq = kn->kn_kq;
1834 KQ_LOCK(kq);
1835 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) {
1836 /*
1837 * Do not process the influx notes, except for
1838 * the influx coming from the kq unlock in the
1839 * kqueue_scan(). In the later case, we do
1840 * not interfere with the scan, since the code
1841 * fragment in kqueue_scan() locks the knlist,
1842 * and cannot proceed until we finished.
1843 */
1844 KQ_UNLOCK(kq);
1845 } else if ((lockflags & KNF_NOKQLOCK) != 0) {
1846 kn->kn_status |= KN_INFLUX;
1847 KQ_UNLOCK(kq);
1848 error = kn->kn_fop->f_event(kn, hint);
1849 KQ_LOCK(kq);
1850 kn->kn_status &= ~KN_INFLUX;
1851 if (error)
1852 KNOTE_ACTIVATE(kn, 1);
1853 KQ_UNLOCK_FLUX(kq);
1854 } else {
1855 kn->kn_status |= KN_HASKQLOCK;
1856 if (kn->kn_fop->f_event(kn, hint))
1857 KNOTE_ACTIVATE(kn, 1);
1858 kn->kn_status &= ~KN_HASKQLOCK;
1859 KQ_UNLOCK(kq);
1860 }
1861 }
1862 if ((lockflags & KNF_LISTLOCKED) == 0)
1863 list->kl_unlock(list->kl_lockarg);
1864 }
1865
1866 /*
1867 * add a knote to a knlist
1868 */
1869 void
knlist_add(struct knlist * knl,struct knote * kn,int islocked)1870 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1871 {
1872 KNL_ASSERT_LOCK(knl, islocked);
1873 KQ_NOTOWNED(kn->kn_kq);
1874 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1875 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1876 if (!islocked)
1877 knl->kl_lock(knl->kl_lockarg);
1878 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1879 if (!islocked)
1880 knl->kl_unlock(knl->kl_lockarg);
1881 KQ_LOCK(kn->kn_kq);
1882 kn->kn_knlist = knl;
1883 kn->kn_status &= ~KN_DETACHED;
1884 KQ_UNLOCK(kn->kn_kq);
1885 }
1886
1887 static void
knlist_remove_kq(struct knlist * knl,struct knote * kn,int knlislocked,int kqislocked)1888 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1889 {
1890 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1891 KNL_ASSERT_LOCK(knl, knlislocked);
1892 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1893 if (!kqislocked)
1894 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1895 ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1896 if (!knlislocked)
1897 knl->kl_lock(knl->kl_lockarg);
1898 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1899 kn->kn_knlist = NULL;
1900 if (!knlislocked)
1901 knl->kl_unlock(knl->kl_lockarg);
1902 if (!kqislocked)
1903 KQ_LOCK(kn->kn_kq);
1904 kn->kn_status |= KN_DETACHED;
1905 if (!kqislocked)
1906 KQ_UNLOCK(kn->kn_kq);
1907 }
1908
1909 /*
1910 * remove all knotes from a specified klist
1911 */
1912 void
knlist_remove(struct knlist * knl,struct knote * kn,int islocked)1913 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1914 {
1915
1916 knlist_remove_kq(knl, kn, islocked, 0);
1917 }
1918
1919 /*
1920 * remove knote from a specified klist while in f_event handler.
1921 */
1922 void
knlist_remove_inevent(struct knlist * knl,struct knote * kn)1923 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1924 {
1925
1926 knlist_remove_kq(knl, kn, 1,
1927 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1928 }
1929
1930 int
knlist_empty(struct knlist * knl)1931 knlist_empty(struct knlist *knl)
1932 {
1933 KNL_ASSERT_LOCKED(knl);
1934 return SLIST_EMPTY(&knl->kl_list);
1935 }
1936
1937 static struct mtx knlist_lock;
1938 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1939 MTX_DEF);
1940 static void knlist_mtx_lock(void *arg);
1941 static void knlist_mtx_unlock(void *arg);
1942
1943 static void
knlist_mtx_lock(void * arg)1944 knlist_mtx_lock(void *arg)
1945 {
1946 mtx_lock((struct mtx *)arg);
1947 }
1948
1949 static void
knlist_mtx_unlock(void * arg)1950 knlist_mtx_unlock(void *arg)
1951 {
1952 mtx_unlock((struct mtx *)arg);
1953 }
1954
1955 static void
knlist_mtx_assert_locked(void * arg)1956 knlist_mtx_assert_locked(void *arg)
1957 {
1958 mtx_assert((struct mtx *)arg, MA_OWNED);
1959 }
1960
1961 static void
knlist_mtx_assert_unlocked(void * arg)1962 knlist_mtx_assert_unlocked(void *arg)
1963 {
1964 mtx_assert((struct mtx *)arg, MA_NOTOWNED);
1965 }
1966
1967 void
knlist_init(struct knlist * knl,void * lock,void (* kl_lock)(void *),void (* kl_unlock)(void *),void (* kl_assert_locked)(void *),void (* kl_assert_unlocked)(void *))1968 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1969 void (*kl_unlock)(void *),
1970 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *))
1971 {
1972
1973 if (lock == NULL)
1974 knl->kl_lockarg = &knlist_lock;
1975 else
1976 knl->kl_lockarg = lock;
1977
1978 if (kl_lock == NULL)
1979 knl->kl_lock = knlist_mtx_lock;
1980 else
1981 knl->kl_lock = kl_lock;
1982 if (kl_unlock == NULL)
1983 knl->kl_unlock = knlist_mtx_unlock;
1984 else
1985 knl->kl_unlock = kl_unlock;
1986 if (kl_assert_locked == NULL)
1987 knl->kl_assert_locked = knlist_mtx_assert_locked;
1988 else
1989 knl->kl_assert_locked = kl_assert_locked;
1990 if (kl_assert_unlocked == NULL)
1991 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked;
1992 else
1993 knl->kl_assert_unlocked = kl_assert_unlocked;
1994
1995 SLIST_INIT(&knl->kl_list);
1996 }
1997
1998 void
knlist_init_mtx(struct knlist * knl,struct mtx * lock)1999 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2000 {
2001
2002 knlist_init(knl, lock, NULL, NULL, NULL, NULL);
2003 }
2004
2005 void
knlist_destroy(struct knlist * knl)2006 knlist_destroy(struct knlist *knl)
2007 {
2008
2009 #ifdef INVARIANTS
2010 /*
2011 * if we run across this error, we need to find the offending
2012 * driver and have it call knlist_clear.
2013 */
2014 if (!SLIST_EMPTY(&knl->kl_list))
2015 printf("WARNING: destroying knlist w/ knotes on it!\n");
2016 #endif
2017
2018 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
2019 SLIST_INIT(&knl->kl_list);
2020 }
2021
2022 /*
2023 * Even if we are locked, we may need to drop the lock to allow any influx
2024 * knotes time to "settle".
2025 */
2026 void
knlist_cleardel(struct knlist * knl,struct thread * td,int islocked,int killkn)2027 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2028 {
2029 struct knote *kn, *kn2;
2030 struct kqueue *kq;
2031
2032 if (islocked)
2033 KNL_ASSERT_LOCKED(knl);
2034 else {
2035 KNL_ASSERT_UNLOCKED(knl);
2036 again: /* need to reacquire lock since we have dropped it */
2037 knl->kl_lock(knl->kl_lockarg);
2038 }
2039
2040 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2041 kq = kn->kn_kq;
2042 KQ_LOCK(kq);
2043 if ((kn->kn_status & KN_INFLUX)) {
2044 KQ_UNLOCK(kq);
2045 continue;
2046 }
2047 knlist_remove_kq(knl, kn, 1, 1);
2048 if (killkn) {
2049 kn->kn_status |= KN_INFLUX | KN_DETACHED;
2050 KQ_UNLOCK(kq);
2051 knote_drop(kn, td);
2052 } else {
2053 /* Make sure cleared knotes disappear soon */
2054 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2055 KQ_UNLOCK(kq);
2056 }
2057 kq = NULL;
2058 }
2059
2060 if (!SLIST_EMPTY(&knl->kl_list)) {
2061 /* there are still KN_INFLUX remaining */
2062 kn = SLIST_FIRST(&knl->kl_list);
2063 kq = kn->kn_kq;
2064 KQ_LOCK(kq);
2065 KASSERT(kn->kn_status & KN_INFLUX,
2066 ("knote removed w/o list lock"));
2067 knl->kl_unlock(knl->kl_lockarg);
2068 kq->kq_state |= KQ_FLUXWAIT;
2069 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2070 kq = NULL;
2071 goto again;
2072 }
2073
2074 if (islocked)
2075 KNL_ASSERT_LOCKED(knl);
2076 else {
2077 knl->kl_unlock(knl->kl_lockarg);
2078 KNL_ASSERT_UNLOCKED(knl);
2079 }
2080 }
2081
2082 /*
2083 * Remove all knotes referencing a specified fd must be called with FILEDESC
2084 * lock. This prevents a race where a new fd comes along and occupies the
2085 * entry and we attach a knote to the fd.
2086 */
2087 void
knote_fdclose(struct thread * td,int fd)2088 knote_fdclose(struct thread *td, int fd)
2089 {
2090 struct filedesc *fdp = td->td_proc->p_fd;
2091 struct kqueue *kq;
2092 struct knote *kn;
2093 int influx;
2094
2095 FILEDESC_XLOCK_ASSERT(fdp);
2096
2097 /*
2098 * We shouldn't have to worry about new kevents appearing on fd
2099 * since filedesc is locked.
2100 */
2101 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2102 KQ_LOCK(kq);
2103
2104 again:
2105 influx = 0;
2106 while (kq->kq_knlistsize > fd &&
2107 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2108 if (kn->kn_status & KN_INFLUX) {
2109 /* someone else might be waiting on our knote */
2110 if (influx)
2111 wakeup(kq);
2112 kq->kq_state |= KQ_FLUXWAIT;
2113 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2114 goto again;
2115 }
2116 kn->kn_status |= KN_INFLUX;
2117 KQ_UNLOCK(kq);
2118 if (!(kn->kn_status & KN_DETACHED))
2119 kn->kn_fop->f_detach(kn);
2120 knote_drop(kn, td);
2121 influx = 1;
2122 KQ_LOCK(kq);
2123 }
2124 KQ_UNLOCK_FLUX(kq);
2125 }
2126 }
2127
2128 static int
knote_attach(struct knote * kn,struct kqueue * kq)2129 knote_attach(struct knote *kn, struct kqueue *kq)
2130 {
2131 struct klist *list;
2132
2133 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
2134 KQ_OWNED(kq);
2135
2136 if (kn->kn_fop->f_isfd) {
2137 if (kn->kn_id >= kq->kq_knlistsize)
2138 return ENOMEM;
2139 list = &kq->kq_knlist[kn->kn_id];
2140 } else {
2141 if (kq->kq_knhash == NULL)
2142 return ENOMEM;
2143 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2144 }
2145
2146 SLIST_INSERT_HEAD(list, kn, kn_link);
2147
2148 return 0;
2149 }
2150
2151 /*
2152 * knote must already have been detached using the f_detach method.
2153 * no lock need to be held, it is assumed that the KN_INFLUX flag is set
2154 * to prevent other removal.
2155 */
2156 static void
knote_drop(struct knote * kn,struct thread * td)2157 knote_drop(struct knote *kn, struct thread *td)
2158 {
2159 struct kqueue *kq;
2160 struct klist *list;
2161
2162 kq = kn->kn_kq;
2163
2164 KQ_NOTOWNED(kq);
2165 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
2166 ("knote_drop called without KN_INFLUX set in kn_status"));
2167
2168 KQ_LOCK(kq);
2169 if (kn->kn_fop->f_isfd)
2170 list = &kq->kq_knlist[kn->kn_id];
2171 else
2172 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2173
2174 if (!SLIST_EMPTY(list))
2175 SLIST_REMOVE(list, kn, knote, kn_link);
2176 if (kn->kn_status & KN_QUEUED)
2177 knote_dequeue(kn);
2178 KQ_UNLOCK_FLUX(kq);
2179
2180 if (kn->kn_fop->f_isfd) {
2181 fdrop(kn->kn_fp, td);
2182 kn->kn_fp = NULL;
2183 }
2184 kqueue_fo_release(kn->kn_kevent.filter);
2185 kn->kn_fop = NULL;
2186 knote_free(kn);
2187 }
2188
2189 static void
knote_enqueue(struct knote * kn)2190 knote_enqueue(struct knote *kn)
2191 {
2192 struct kqueue *kq = kn->kn_kq;
2193
2194 KQ_OWNED(kn->kn_kq);
2195 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2196
2197 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2198 kn->kn_status |= KN_QUEUED;
2199 kq->kq_count++;
2200 kqueue_wakeup(kq);
2201 }
2202
2203 static void
knote_dequeue(struct knote * kn)2204 knote_dequeue(struct knote *kn)
2205 {
2206 struct kqueue *kq = kn->kn_kq;
2207
2208 KQ_OWNED(kn->kn_kq);
2209 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2210
2211 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2212 kn->kn_status &= ~KN_QUEUED;
2213 kq->kq_count--;
2214 }
2215
2216 static void
knote_init(void)2217 knote_init(void)
2218 {
2219
2220 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
2221 NULL, NULL, UMA_ALIGN_PTR, 0);
2222 }
2223 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
2224
2225 static struct knote *
knote_alloc(int waitok)2226 knote_alloc(int waitok)
2227 {
2228 return ((struct knote *)uma_zalloc(knote_zone,
2229 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
2230 }
2231
2232 static void
knote_free(struct knote * kn)2233 knote_free(struct knote *kn)
2234 {
2235 if (kn != NULL)
2236 uma_zfree(knote_zone, kn);
2237 }
2238
2239 /*
2240 * Register the kev w/ the kq specified by fd.
2241 */
2242 int
kqfd_register(int fd,struct kevent * kev,struct thread * td,int waitok)2243 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok)
2244 {
2245 struct kqueue *kq;
2246 struct file *fp;
2247 int error;
2248
2249 if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0)
2250 return (error);
2251 if ((error = kqueue_acquire(fp, &kq)) != 0)
2252 goto noacquire;
2253
2254 error = kqueue_register(kq, kev, td, waitok);
2255
2256 kqueue_release(kq, 0);
2257
2258 noacquire:
2259 fdrop(fp, td);
2260
2261 return error;
2262 }
2263