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 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD: releng/6.4/sys/kern/kern_event.c 197715 2009-10-02 18:09:56Z simon $");
30
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/proc.h>
37 #include <sys/malloc.h>
38 #include <sys/unistd.h>
39 #include <sys/file.h>
40 #include <sys/filedesc.h>
41 #include <sys/filio.h>
42 #include <sys/fcntl.h>
43 #include <sys/kthread.h>
44 #include <sys/selinfo.h>
45 #include <sys/queue.h>
46 #include <sys/event.h>
47 #include <sys/eventvar.h>
48 #include <sys/poll.h>
49 #include <sys/protosw.h>
50 #include <sys/sigio.h>
51 #include <sys/signalvar.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/stat.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysproto.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/taskqueue.h>
59 #include <sys/uio.h>
60
61 #include <vm/uma.h>
62
63 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
64
65 /*
66 * This lock is used if multiple kq locks are required. This possibly
67 * should be made into a per proc lock.
68 */
69 static struct mtx kq_global;
70 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
71 #define KQ_GLOBAL_LOCK(lck, haslck) do { \
72 if (!haslck) \
73 mtx_lock(lck); \
74 haslck = 1; \
75 } while (0)
76 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \
77 if (haslck) \
78 mtx_unlock(lck); \
79 haslck = 0; \
80 } while (0)
81
82 TASKQUEUE_DEFINE_THREAD(kqueue);
83
84 static int kevent_copyout(void *arg, struct kevent *kevp, int count);
85 static int kevent_copyin(void *arg, struct kevent *kevp, int count);
86 static int kqueue_aquire(struct file *fp, struct kqueue **kqp);
87 static void kqueue_release(struct kqueue *kq, int locked);
88 static int kqueue_expand(struct kqueue *kq, struct filterops *fops,
89 uintptr_t ident, int waitok);
90 static void kqueue_task(void *arg, int pending);
91 static int kqueue_scan(struct kqueue *kq, int maxevents,
92 struct kevent_copyops *k_ops,
93 const struct timespec *timeout,
94 struct kevent *keva, struct thread *td);
95 static void kqueue_wakeup(struct kqueue *kq);
96 static struct filterops *kqueue_fo_find(int filt);
97 static void kqueue_fo_release(int filt);
98
99 static fo_rdwr_t kqueue_read;
100 static fo_rdwr_t kqueue_write;
101 static fo_ioctl_t kqueue_ioctl;
102 static fo_poll_t kqueue_poll;
103 static fo_kqfilter_t kqueue_kqfilter;
104 static fo_stat_t kqueue_stat;
105 static fo_close_t kqueue_close;
106
107 static struct fileops kqueueops = {
108 .fo_read = kqueue_read,
109 .fo_write = kqueue_write,
110 .fo_ioctl = kqueue_ioctl,
111 .fo_poll = kqueue_poll,
112 .fo_kqfilter = kqueue_kqfilter,
113 .fo_stat = kqueue_stat,
114 .fo_close = kqueue_close,
115 };
116
117 static int knote_attach(struct knote *kn, struct kqueue *kq);
118 static void knote_drop(struct knote *kn, struct thread *td);
119 static void knote_enqueue(struct knote *kn);
120 static void knote_dequeue(struct knote *kn);
121 static void knote_init(void);
122 static struct knote *knote_alloc(int waitok);
123 static void knote_free(struct knote *kn);
124
125 static void filt_kqdetach(struct knote *kn);
126 static int filt_kqueue(struct knote *kn, long hint);
127 static int filt_procattach(struct knote *kn);
128 static void filt_procdetach(struct knote *kn);
129 static int filt_proc(struct knote *kn, long hint);
130 static int filt_fileattach(struct knote *kn);
131 static void filt_timerexpire(void *knx);
132 static int filt_timerattach(struct knote *kn);
133 static void filt_timerdetach(struct knote *kn);
134 static int filt_timer(struct knote *kn, long hint);
135
136 static struct filterops file_filtops =
137 { 1, filt_fileattach, NULL, NULL };
138 static struct filterops kqread_filtops =
139 { 1, NULL, filt_kqdetach, filt_kqueue };
140 /* XXX - move to kern_proc.c? */
141 static struct filterops proc_filtops =
142 { 0, filt_procattach, filt_procdetach, filt_proc };
143 static struct filterops timer_filtops =
144 { 0, filt_timerattach, filt_timerdetach, filt_timer };
145
146 static uma_zone_t knote_zone;
147 static int kq_ncallouts = 0;
148 static int kq_calloutmax = (4 * 1024);
149 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
150 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
151
152 /* XXX - ensure not KN_INFLUX?? */
153 #define KNOTE_ACTIVATE(kn, islock) do { \
154 if ((islock)) \
155 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \
156 else \
157 KQ_LOCK((kn)->kn_kq); \
158 (kn)->kn_status |= KN_ACTIVE; \
159 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
160 knote_enqueue((kn)); \
161 if (!(islock)) \
162 KQ_UNLOCK((kn)->kn_kq); \
163 } while(0)
164 #define KQ_LOCK(kq) do { \
165 mtx_lock(&(kq)->kq_lock); \
166 } while (0)
167 #define KQ_FLUX_WAKEUP(kq) do { \
168 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
169 (kq)->kq_state &= ~KQ_FLUXWAIT; \
170 wakeup((kq)); \
171 } \
172 } while (0)
173 #define KQ_UNLOCK_FLUX(kq) do { \
174 KQ_FLUX_WAKEUP(kq); \
175 mtx_unlock(&(kq)->kq_lock); \
176 } while (0)
177 #define KQ_UNLOCK(kq) do { \
178 mtx_unlock(&(kq)->kq_lock); \
179 } while (0)
180 #define KQ_OWNED(kq) do { \
181 mtx_assert(&(kq)->kq_lock, MA_OWNED); \
182 } while (0)
183 #define KQ_NOTOWNED(kq) do { \
184 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \
185 } while (0)
186 #define KN_LIST_LOCK(kn) do { \
187 if (kn->kn_knlist != NULL) \
188 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \
189 } while (0)
190 #define KN_LIST_UNLOCK(kn) do { \
191 if (kn->kn_knlist != NULL) \
192 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \
193 } while (0)
194 #define KNL_ASSERT_LOCK(knl, islocked) do { \
195 if (islocked) \
196 KNL_ASSERT_LOCKED(knl); \
197 else \
198 KNL_ASSERT_UNLOCKED(knl); \
199 } while (0)
200 #ifdef INVARIANTS
201 #define KNL_ASSERT_LOCKED(knl) do { \
202 if (!knl->kl_locked((knl)->kl_lockarg)) \
203 panic("knlist not locked, but should be"); \
204 } while (0)
205 #define KNL_ASSERT_UNLOCKED(knl) do { \
206 if (knl->kl_locked((knl)->kl_lockarg)) \
207 panic("knlist locked, but should not be"); \
208 } while (0)
209 #else /* !INVARIANTS */
210 #define KNL_ASSERT_LOCKED(knl) do {} while(0)
211 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0)
212 #endif /* INVARIANTS */
213
214 #define KN_HASHSIZE 64 /* XXX should be tunable */
215 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
216
217 static int
218 filt_nullattach(struct knote *kn)
219 {
220
221 return (ENXIO);
222 };
223
224 struct filterops null_filtops =
225 { 0, filt_nullattach, NULL, NULL };
226
227 /* XXX - make SYSINIT to add these, and move into respective modules. */
228 extern struct filterops sig_filtops;
229 extern struct filterops fs_filtops;
230
231 /*
232 * Table for for all system-defined filters.
233 */
234 static struct mtx filterops_lock;
235 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops",
236 MTX_DEF);
237 static struct {
238 struct filterops *for_fop;
239 int for_refcnt;
240 } sysfilt_ops[EVFILT_SYSCOUNT] = {
241 { &file_filtops }, /* EVFILT_READ */
242 { &file_filtops }, /* EVFILT_WRITE */
243 { &null_filtops }, /* EVFILT_AIO */
244 { &file_filtops }, /* EVFILT_VNODE */
245 { &proc_filtops }, /* EVFILT_PROC */
246 { &sig_filtops }, /* EVFILT_SIGNAL */
247 { &timer_filtops }, /* EVFILT_TIMER */
248 { &file_filtops }, /* EVFILT_NETDEV */
249 { &fs_filtops }, /* EVFILT_FS */
250 };
251
252 /*
253 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
254 * method.
255 */
256 static int
257 filt_fileattach(struct knote *kn)
258 {
259
260 return (fo_kqfilter(kn->kn_fp, kn));
261 }
262
263 /*ARGSUSED*/
264 static int
265 kqueue_kqfilter(struct file *fp, struct knote *kn)
266 {
267 struct kqueue *kq = kn->kn_fp->f_data;
268
269 if (kn->kn_filter != EVFILT_READ)
270 return (EINVAL);
271
272 kn->kn_status |= KN_KQUEUE;
273 kn->kn_fop = &kqread_filtops;
274 knlist_add(&kq->kq_sel.si_note, kn, 0);
275
276 return (0);
277 }
278
279 static void
280 filt_kqdetach(struct knote *kn)
281 {
282 struct kqueue *kq = kn->kn_fp->f_data;
283
284 knlist_remove(&kq->kq_sel.si_note, kn, 0);
285 }
286
287 /*ARGSUSED*/
288 static int
289 filt_kqueue(struct knote *kn, long hint)
290 {
291 struct kqueue *kq = kn->kn_fp->f_data;
292
293 kn->kn_data = kq->kq_count;
294 return (kn->kn_data > 0);
295 }
296
297 /* XXX - move to kern_proc.c? */
298 static int
299 filt_procattach(struct knote *kn)
300 {
301 struct proc *p;
302 int immediate;
303 int error;
304
305 immediate = 0;
306 p = pfind(kn->kn_id);
307 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
308 p = zpfind(kn->kn_id);
309 immediate = 1;
310 } else if (p != NULL && (p->p_flag & P_WEXIT)) {
311 immediate = 1;
312 }
313
314 if (p == NULL)
315 return (ESRCH);
316 if ((error = p_cansee(curthread, p)))
317 return (error);
318
319 kn->kn_ptr.p_proc = p;
320 kn->kn_flags |= EV_CLEAR; /* automatically set */
321
322 /*
323 * internal flag indicating registration done by kernel
324 */
325 if (kn->kn_flags & EV_FLAG1) {
326 kn->kn_data = kn->kn_sdata; /* ppid */
327 kn->kn_fflags = NOTE_CHILD;
328 kn->kn_flags &= ~EV_FLAG1;
329 }
330
331 if (immediate == 0)
332 knlist_add(&p->p_klist, kn, 1);
333
334 /*
335 * Immediately activate any exit notes if the target process is a
336 * zombie. This is necessary to handle the case where the target
337 * process, e.g. a child, dies before the kevent is registered.
338 */
339 if (immediate && filt_proc(kn, NOTE_EXIT))
340 KNOTE_ACTIVATE(kn, 0);
341
342 PROC_UNLOCK(p);
343
344 return (0);
345 }
346
347 /*
348 * The knote may be attached to a different process, which may exit,
349 * leaving nothing for the knote to be attached to. So when the process
350 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
351 * it will be deleted when read out. However, as part of the knote deletion,
352 * this routine is called, so a check is needed to avoid actually performing
353 * a detach, because the original process does not exist any more.
354 */
355 /* XXX - move to kern_proc.c? */
356 static void
357 filt_procdetach(struct knote *kn)
358 {
359 struct proc *p;
360
361 p = kn->kn_ptr.p_proc;
362 knlist_remove(&p->p_klist, kn, 0);
363 kn->kn_ptr.p_proc = NULL;
364 }
365
366 /* XXX - move to kern_proc.c? */
367 static int
368 filt_proc(struct knote *kn, long hint)
369 {
370 struct proc *p = kn->kn_ptr.p_proc;
371 u_int event;
372
373 /*
374 * mask off extra data
375 */
376 event = (u_int)hint & NOTE_PCTRLMASK;
377
378 /*
379 * if the user is interested in this event, record it.
380 */
381 if (kn->kn_sfflags & event)
382 kn->kn_fflags |= event;
383
384 /*
385 * process is gone, so flag the event as finished.
386 */
387 if (event == NOTE_EXIT) {
388 if (!(kn->kn_status & KN_DETACHED))
389 knlist_remove_inevent(&p->p_klist, kn);
390 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
391 kn->kn_ptr.p_proc = NULL;
392 return (1);
393 }
394
395 return (kn->kn_fflags != 0);
396 }
397
398 /*
399 * Called when the process forked. It mostly does the same as the
400 * knote(), activating all knotes registered to be activated when the
401 * process forked. Additionally, for each knote attached to the
402 * parent, check whether user wants to track the new process. If so
403 * attach a new knote to it, and immediately report an event with the
404 * child's pid.
405 */
406 void
407 knote_fork(struct knlist *list, int pid)
408 {
409 struct kqueue *kq;
410 struct knote *kn;
411 struct kevent kev;
412 int error;
413
414 if (list == NULL)
415 return;
416 list->kl_lock(list->kl_lockarg);
417
418 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
419 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX)
420 continue;
421 kq = kn->kn_kq;
422 KQ_LOCK(kq);
423 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
424 KQ_UNLOCK(kq);
425 continue;
426 }
427
428 /*
429 * The same as knote(), activate the event.
430 */
431 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
432 kn->kn_status |= KN_HASKQLOCK;
433 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid))
434 KNOTE_ACTIVATE(kn, 1);
435 kn->kn_status &= ~KN_HASKQLOCK;
436 KQ_UNLOCK(kq);
437 continue;
438 }
439
440 /*
441 * The NOTE_TRACK case. In addition to the activation
442 * of the event, we need to register new event to
443 * track the child. Drop the locks in preparation for
444 * the call to kqueue_register().
445 */
446 kn->kn_status |= KN_INFLUX;
447 KQ_UNLOCK(kq);
448 list->kl_unlock(list->kl_lockarg);
449
450 /*
451 * Activate existing knote and register a knote with
452 * new process.
453 */
454 kev.ident = pid;
455 kev.filter = kn->kn_filter;
456 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
457 kev.fflags = kn->kn_sfflags;
458 kev.data = kn->kn_id; /* parent */
459 kev.udata = kn->kn_kevent.udata;/* preserve udata */
460 error = kqueue_register(kq, &kev, NULL, 0);
461 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid))
462 KNOTE_ACTIVATE(kn, 0);
463 if (error)
464 kn->kn_fflags |= NOTE_TRACKERR;
465 KQ_LOCK(kq);
466 kn->kn_status &= ~KN_INFLUX;
467 KQ_UNLOCK_FLUX(kq);
468 list->kl_lock(list->kl_lockarg);
469 }
470 list->kl_unlock(list->kl_lockarg);
471 }
472
473 static int
474 timertoticks(intptr_t data)
475 {
476 struct timeval tv;
477 int tticks;
478
479 tv.tv_sec = data / 1000;
480 tv.tv_usec = (data % 1000) * 1000;
481 tticks = tvtohz(&tv);
482
483 return tticks;
484 }
485
486 /* XXX - move to kern_timeout.c? */
487 static void
488 filt_timerexpire(void *knx)
489 {
490 struct knote *kn = knx;
491 struct callout *calloutp;
492
493 kn->kn_data++;
494 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
495
496 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) {
497 calloutp = (struct callout *)kn->kn_hook;
498 callout_reset(calloutp, timertoticks(kn->kn_sdata),
499 filt_timerexpire, kn);
500 }
501 }
502
503 /*
504 * data contains amount of time to sleep, in milliseconds
505 */
506 /* XXX - move to kern_timeout.c? */
507 static int
508 filt_timerattach(struct knote *kn)
509 {
510 struct callout *calloutp;
511
512 atomic_add_int(&kq_ncallouts, 1);
513
514 if (kq_ncallouts >= kq_calloutmax) {
515 atomic_add_int(&kq_ncallouts, -1);
516 return (ENOMEM);
517 }
518
519 kn->kn_flags |= EV_CLEAR; /* automatically set */
520 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */
521 MALLOC(calloutp, struct callout *, sizeof(*calloutp),
522 M_KQUEUE, M_WAITOK);
523 callout_init(calloutp, CALLOUT_MPSAFE);
524 kn->kn_hook = calloutp;
525 callout_reset(calloutp, timertoticks(kn->kn_sdata), filt_timerexpire,
526 kn);
527
528 return (0);
529 }
530
531 /* XXX - move to kern_timeout.c? */
532 static void
533 filt_timerdetach(struct knote *kn)
534 {
535 struct callout *calloutp;
536
537 calloutp = (struct callout *)kn->kn_hook;
538 callout_drain(calloutp);
539 FREE(calloutp, M_KQUEUE);
540 atomic_add_int(&kq_ncallouts, -1);
541 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */
542 }
543
544 /* XXX - move to kern_timeout.c? */
545 static int
546 filt_timer(struct knote *kn, long hint)
547 {
548
549 return (kn->kn_data != 0);
550 }
551
552 /*
553 * MPSAFE
554 */
555 int
556 kqueue(struct thread *td, struct kqueue_args *uap)
557 {
558 struct filedesc *fdp;
559 struct kqueue *kq;
560 struct file *fp;
561 int fd, error;
562
563 fdp = td->td_proc->p_fd;
564 error = falloc(td, &fp, &fd);
565 if (error)
566 goto done2;
567
568 /* An extra reference on `nfp' has been held for us by falloc(). */
569 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
570 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK);
571 TAILQ_INIT(&kq->kq_head);
572 kq->kq_fdp = fdp;
573 knlist_init(&kq->kq_sel.si_note, &kq->kq_lock, NULL, NULL, NULL);
574 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
575
576 FILEDESC_LOCK_FAST(fdp);
577 SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
578 FILEDESC_UNLOCK_FAST(fdp);
579
580 FILE_LOCK(fp);
581 fp->f_flag = FREAD | FWRITE;
582 fp->f_type = DTYPE_KQUEUE;
583 fp->f_ops = &kqueueops;
584 fp->f_data = kq;
585 FILE_UNLOCK(fp);
586 fdrop(fp, td);
587
588 td->td_retval[0] = fd;
589 done2:
590 return (error);
591 }
592
593 #ifndef _SYS_SYSPROTO_H_
594 struct kevent_args {
595 int fd;
596 const struct kevent *changelist;
597 int nchanges;
598 struct kevent *eventlist;
599 int nevents;
600 const struct timespec *timeout;
601 };
602 #endif
603 /*
604 * MPSAFE
605 */
606 int
607 kevent(struct thread *td, struct kevent_args *uap)
608 {
609 struct timespec ts, *tsp;
610 struct kevent_copyops k_ops = { uap,
611 kevent_copyout,
612 kevent_copyin};
613 int error;
614
615 if (uap->timeout != NULL) {
616 error = copyin(uap->timeout, &ts, sizeof(ts));
617 if (error)
618 return (error);
619 tsp = &ts;
620 } else
621 tsp = NULL;
622
623 return (kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
624 &k_ops, tsp));
625 }
626
627 /*
628 * Copy 'count' items into the destination list pointed to by uap->eventlist.
629 */
630 static int
631 kevent_copyout(void *arg, struct kevent *kevp, int count)
632 {
633 struct kevent_args *uap;
634 int error;
635
636 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
637 uap = (struct kevent_args *)arg;
638
639 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
640 if (error == 0)
641 uap->eventlist += count;
642 return (error);
643 }
644
645 /*
646 * Copy 'count' items from the list pointed to by uap->changelist.
647 */
648 static int
649 kevent_copyin(void *arg, struct kevent *kevp, int count)
650 {
651 struct kevent_args *uap;
652 int error;
653
654 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
655 uap = (struct kevent_args *)arg;
656
657 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
658 if (error == 0)
659 uap->changelist += count;
660 return (error);
661 }
662
663 int
664 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
665 struct kevent_copyops *k_ops, const struct timespec *timeout)
666 {
667 struct kevent keva[KQ_NEVENTS];
668 struct kevent *kevp, *changes;
669 struct kqueue *kq;
670 struct file *fp;
671 int i, n, nerrors, error;
672
673 if ((error = fget(td, fd, &fp)) != 0)
674 return (error);
675 if ((error = kqueue_aquire(fp, &kq)) != 0)
676 goto done_norel;
677
678 nerrors = 0;
679
680 while (nchanges > 0) {
681 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
682 error = k_ops->k_copyin(k_ops->arg, keva, n);
683 if (error)
684 goto done;
685 changes = keva;
686 for (i = 0; i < n; i++) {
687 kevp = &changes[i];
688 kevp->flags &= ~EV_SYSFLAGS;
689 error = kqueue_register(kq, kevp, td, 1);
690 if (error) {
691 if (nevents != 0) {
692 kevp->flags = EV_ERROR;
693 kevp->data = error;
694 (void) k_ops->k_copyout(k_ops->arg,
695 kevp, 1);
696 nevents--;
697 nerrors++;
698 } else {
699 goto done;
700 }
701 }
702 }
703 nchanges -= n;
704 }
705 if (nerrors) {
706 td->td_retval[0] = nerrors;
707 error = 0;
708 goto done;
709 }
710
711 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td);
712 done:
713 kqueue_release(kq, 0);
714 done_norel:
715 if (fp != NULL)
716 fdrop(fp, td);
717 return (error);
718 }
719
720 int
721 kqueue_add_filteropts(int filt, struct filterops *filtops)
722 {
723 int error;
724
725 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
726 printf(
727 "trying to add a filterop that is out of range: %d is beyond %d\n",
728 ~filt, EVFILT_SYSCOUNT);
729 return EINVAL;
730 }
731 mtx_lock(&filterops_lock);
732 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
733 sysfilt_ops[~filt].for_fop != NULL)
734 error = EEXIST;
735 else {
736 sysfilt_ops[~filt].for_fop = filtops;
737 sysfilt_ops[~filt].for_refcnt = 0;
738 }
739 mtx_unlock(&filterops_lock);
740
741 return (0);
742 }
743
744 int
745 kqueue_del_filteropts(int filt)
746 {
747 int error;
748
749 error = 0;
750 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
751 return EINVAL;
752
753 mtx_lock(&filterops_lock);
754 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
755 sysfilt_ops[~filt].for_fop == NULL)
756 error = EINVAL;
757 else if (sysfilt_ops[~filt].for_refcnt != 0)
758 error = EBUSY;
759 else {
760 sysfilt_ops[~filt].for_fop = &null_filtops;
761 sysfilt_ops[~filt].for_refcnt = 0;
762 }
763 mtx_unlock(&filterops_lock);
764
765 return error;
766 }
767
768 static struct filterops *
769 kqueue_fo_find(int filt)
770 {
771
772 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
773 return NULL;
774
775 mtx_lock(&filterops_lock);
776 sysfilt_ops[~filt].for_refcnt++;
777 if (sysfilt_ops[~filt].for_fop == NULL)
778 sysfilt_ops[~filt].for_fop = &null_filtops;
779 mtx_unlock(&filterops_lock);
780
781 return sysfilt_ops[~filt].for_fop;
782 }
783
784 static void
785 kqueue_fo_release(int filt)
786 {
787
788 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
789 return;
790
791 mtx_lock(&filterops_lock);
792 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
793 ("filter object refcount not valid on release"));
794 sysfilt_ops[~filt].for_refcnt--;
795 mtx_unlock(&filterops_lock);
796 }
797
798 /*
799 * A ref to kq (obtained via kqueue_aquire) should be held. waitok will
800 * influence if memory allocation should wait. Make sure it is 0 if you
801 * hold any mutexes.
802 */
803 int
804 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok)
805 {
806 struct filedesc *fdp;
807 struct filterops *fops;
808 struct file *fp;
809 struct knote *kn, *tkn;
810 int error, filt, event;
811 int haskqglobal;
812 int fd;
813
814 fdp = NULL;
815 fp = NULL;
816 kn = NULL;
817 error = 0;
818 haskqglobal = 0;
819
820 filt = kev->filter;
821 fops = kqueue_fo_find(filt);
822 if (fops == NULL)
823 return EINVAL;
824
825 tkn = knote_alloc(waitok); /* prevent waiting with locks */
826
827 findkn:
828 if (fops->f_isfd) {
829 KASSERT(td != NULL, ("td is NULL"));
830 fdp = td->td_proc->p_fd;
831 FILEDESC_LOCK(fdp);
832 /* validate descriptor */
833 fd = kev->ident;
834 if (fd < 0 || fd >= fdp->fd_nfiles ||
835 (fp = fdp->fd_ofiles[fd]) == NULL) {
836 FILEDESC_UNLOCK(fdp);
837 error = EBADF;
838 goto done;
839 }
840 fhold(fp);
841
842 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
843 kev->ident, 0) != 0) {
844 /* unlock and try again */
845 FILEDESC_UNLOCK(fdp);
846 fdrop(fp, td);
847 fp = NULL;
848 error = kqueue_expand(kq, fops, kev->ident, waitok);
849 if (error)
850 goto done;
851 goto findkn;
852 }
853
854 if (fp->f_type == DTYPE_KQUEUE) {
855 /*
856 * if we add some inteligence about what we are doing,
857 * we should be able to support events on ourselves.
858 * We need to know when we are doing this to prevent
859 * getting both the knlist lock and the kq lock since
860 * they are the same thing.
861 */
862 if (fp->f_data == kq) {
863 FILEDESC_UNLOCK(fdp);
864 error = EINVAL;
865 goto done;
866 }
867
868 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
869 }
870
871 FILEDESC_UNLOCK(fdp);
872 KQ_LOCK(kq);
873 if (kev->ident < kq->kq_knlistsize) {
874 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
875 if (kev->filter == kn->kn_filter)
876 break;
877 }
878 } else {
879 if ((kev->flags & EV_ADD) == EV_ADD)
880 kqueue_expand(kq, fops, kev->ident, waitok);
881
882 KQ_LOCK(kq);
883 if (kq->kq_knhashmask != 0) {
884 struct klist *list;
885
886 list = &kq->kq_knhash[
887 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
888 SLIST_FOREACH(kn, list, kn_link)
889 if (kev->ident == kn->kn_id &&
890 kev->filter == kn->kn_filter)
891 break;
892 }
893 }
894
895 /* knote is in the process of changing, wait for it to stablize. */
896 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
897 if (fp != NULL) {
898 fdrop(fp, td);
899 fp = NULL;
900 }
901 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
902 kq->kq_state |= KQ_FLUXWAIT;
903 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
904 goto findkn;
905 }
906
907 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
908 KQ_UNLOCK(kq);
909 error = ENOENT;
910 goto done;
911 }
912
913 /*
914 * kn now contains the matching knote, or NULL if no match
915 */
916 if (kev->flags & EV_ADD) {
917 if (kn == NULL) {
918 kn = tkn;
919 tkn = NULL;
920 if (kn == NULL) {
921 KQ_UNLOCK(kq);
922 error = ENOMEM;
923 goto done;
924 }
925 kn->kn_fp = fp;
926 kn->kn_kq = kq;
927 kn->kn_fop = fops;
928 /*
929 * apply reference counts to knote structure, and
930 * do not release it at the end of this routine.
931 */
932 fops = NULL;
933 fp = NULL;
934
935 kn->kn_sfflags = kev->fflags;
936 kn->kn_sdata = kev->data;
937 kev->fflags = 0;
938 kev->data = 0;
939 kn->kn_kevent = *kev;
940 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
941 EV_ENABLE | EV_DISABLE);
942 kn->kn_status = KN_INFLUX|KN_DETACHED;
943
944 error = knote_attach(kn, kq);
945 KQ_UNLOCK(kq);
946 if (error != 0) {
947 tkn = kn;
948 goto done;
949 }
950
951 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
952 knote_drop(kn, td);
953 goto done;
954 }
955 KN_LIST_LOCK(kn);
956 } else {
957 /*
958 * The user may change some filter values after the
959 * initial EV_ADD, but doing so will not reset any
960 * filter which has already been triggered.
961 */
962 kn->kn_status |= KN_INFLUX;
963 KQ_UNLOCK(kq);
964 KN_LIST_LOCK(kn);
965 kn->kn_sfflags = kev->fflags;
966 kn->kn_sdata = kev->data;
967 kn->kn_kevent.udata = kev->udata;
968 }
969
970 /*
971 * We can get here with kn->kn_knlist == NULL.
972 * This can happen when the initial attach event decides that
973 * the event is "completed" already. i.e. filt_procattach
974 * is called on a zombie process. It will call filt_proc
975 * which will remove it from the list, and NULL kn_knlist.
976 */
977 event = kn->kn_fop->f_event(kn, 0);
978 KQ_LOCK(kq);
979 if (event)
980 KNOTE_ACTIVATE(kn, 1);
981 kn->kn_status &= ~KN_INFLUX;
982 KN_LIST_UNLOCK(kn);
983 } else if (kev->flags & EV_DELETE) {
984 kn->kn_status |= KN_INFLUX;
985 KQ_UNLOCK(kq);
986 if (!(kn->kn_status & KN_DETACHED))
987 kn->kn_fop->f_detach(kn);
988 knote_drop(kn, td);
989 goto done;
990 }
991
992 if ((kev->flags & EV_DISABLE) &&
993 ((kn->kn_status & KN_DISABLED) == 0)) {
994 kn->kn_status |= KN_DISABLED;
995 }
996
997 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
998 kn->kn_status &= ~KN_DISABLED;
999 if ((kn->kn_status & KN_ACTIVE) &&
1000 ((kn->kn_status & KN_QUEUED) == 0))
1001 knote_enqueue(kn);
1002 }
1003 KQ_UNLOCK_FLUX(kq);
1004
1005 done:
1006 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1007 if (fp != NULL)
1008 fdrop(fp, td);
1009 if (tkn != NULL)
1010 knote_free(tkn);
1011 if (fops != NULL)
1012 kqueue_fo_release(filt);
1013 return (error);
1014 }
1015
1016 static int
1017 kqueue_aquire(struct file *fp, struct kqueue **kqp)
1018 {
1019 int error;
1020 struct kqueue *kq;
1021
1022 error = 0;
1023
1024 FILE_LOCK(fp);
1025 do {
1026 kq = fp->f_data;
1027 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) {
1028 error = EBADF;
1029 break;
1030 }
1031 *kqp = kq;
1032 KQ_LOCK(kq);
1033 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1034 KQ_UNLOCK(kq);
1035 error = EBADF;
1036 break;
1037 }
1038 kq->kq_refcnt++;
1039 KQ_UNLOCK(kq);
1040 } while (0);
1041 FILE_UNLOCK(fp);
1042
1043 return error;
1044 }
1045
1046 static void
1047 kqueue_release(struct kqueue *kq, int locked)
1048 {
1049 if (locked)
1050 KQ_OWNED(kq);
1051 else
1052 KQ_LOCK(kq);
1053 kq->kq_refcnt--;
1054 if (kq->kq_refcnt == 1)
1055 wakeup(&kq->kq_refcnt);
1056 if (!locked)
1057 KQ_UNLOCK(kq);
1058 }
1059
1060 static void
1061 kqueue_schedtask(struct kqueue *kq)
1062 {
1063
1064 KQ_OWNED(kq);
1065 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1066 ("scheduling kqueue task while draining"));
1067
1068 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
1069 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task);
1070 kq->kq_state |= KQ_TASKSCHED;
1071 }
1072 }
1073
1074 /*
1075 * Expand the kq to make sure we have storage for fops/ident pair.
1076 *
1077 * Return 0 on success (or no work necessary), return errno on failure.
1078 *
1079 * Not calling hashinit w/ waitok (proper malloc flag) should be safe.
1080 * If kqueue_register is called from a non-fd context, there usually/should
1081 * be no locks held.
1082 */
1083 static int
1084 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident,
1085 int waitok)
1086 {
1087 struct klist *list, *tmp_knhash;
1088 u_long tmp_knhashmask;
1089 int size;
1090 int fd;
1091 int mflag = waitok ? M_WAITOK : M_NOWAIT;
1092
1093 KQ_NOTOWNED(kq);
1094
1095 if (fops->f_isfd) {
1096 fd = ident;
1097 if (kq->kq_knlistsize <= fd) {
1098 size = kq->kq_knlistsize;
1099 while (size <= fd)
1100 size += KQEXTENT;
1101 MALLOC(list, struct klist *,
1102 size * sizeof list, M_KQUEUE, mflag);
1103 if (list == NULL)
1104 return ENOMEM;
1105 KQ_LOCK(kq);
1106 if (kq->kq_knlistsize > fd) {
1107 FREE(list, M_KQUEUE);
1108 list = NULL;
1109 } else {
1110 if (kq->kq_knlist != NULL) {
1111 bcopy(kq->kq_knlist, list,
1112 kq->kq_knlistsize * sizeof list);
1113 FREE(kq->kq_knlist, M_KQUEUE);
1114 kq->kq_knlist = NULL;
1115 }
1116 bzero((caddr_t)list +
1117 kq->kq_knlistsize * sizeof list,
1118 (size - kq->kq_knlistsize) * sizeof list);
1119 kq->kq_knlistsize = size;
1120 kq->kq_knlist = list;
1121 }
1122 KQ_UNLOCK(kq);
1123 }
1124 } else {
1125 if (kq->kq_knhashmask == 0) {
1126 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
1127 &tmp_knhashmask);
1128 if (tmp_knhash == NULL)
1129 return ENOMEM;
1130 KQ_LOCK(kq);
1131 if (kq->kq_knhashmask == 0) {
1132 kq->kq_knhash = tmp_knhash;
1133 kq->kq_knhashmask = tmp_knhashmask;
1134 } else {
1135 free(tmp_knhash, M_KQUEUE);
1136 }
1137 KQ_UNLOCK(kq);
1138 }
1139 }
1140
1141 KQ_NOTOWNED(kq);
1142 return 0;
1143 }
1144
1145 static void
1146 kqueue_task(void *arg, int pending)
1147 {
1148 struct kqueue *kq;
1149 int haskqglobal;
1150
1151 haskqglobal = 0;
1152 kq = arg;
1153
1154 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1155 KQ_LOCK(kq);
1156
1157 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
1158
1159 kq->kq_state &= ~KQ_TASKSCHED;
1160 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
1161 wakeup(&kq->kq_state);
1162 }
1163 KQ_UNLOCK(kq);
1164 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1165 }
1166
1167 /*
1168 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
1169 * We treat KN_MARKER knotes as if they are INFLUX.
1170 */
1171 static int
1172 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
1173 const struct timespec *tsp, struct kevent *keva, struct thread *td)
1174 {
1175 struct kevent *kevp;
1176 struct timeval atv, rtv, ttv;
1177 struct knote *kn, *marker;
1178 int count, timeout, nkev, error, influx;
1179 int haskqglobal;
1180
1181 count = maxevents;
1182 nkev = 0;
1183 error = 0;
1184 haskqglobal = 0;
1185
1186 if (maxevents == 0)
1187 goto done_nl;
1188
1189 if (tsp != NULL) {
1190 TIMESPEC_TO_TIMEVAL(&atv, tsp);
1191 if (itimerfix(&atv)) {
1192 error = EINVAL;
1193 goto done_nl;
1194 }
1195 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
1196 timeout = -1;
1197 else
1198 timeout = atv.tv_sec > 24 * 60 * 60 ?
1199 24 * 60 * 60 * hz : tvtohz(&atv);
1200 getmicrouptime(&rtv);
1201 timevaladd(&atv, &rtv);
1202 } else {
1203 atv.tv_sec = 0;
1204 atv.tv_usec = 0;
1205 timeout = 0;
1206 }
1207 marker = knote_alloc(1);
1208 if (marker == NULL) {
1209 error = ENOMEM;
1210 goto done_nl;
1211 }
1212 marker->kn_status = KN_MARKER;
1213 KQ_LOCK(kq);
1214 goto start;
1215
1216 retry:
1217 if (atv.tv_sec || atv.tv_usec) {
1218 getmicrouptime(&rtv);
1219 if (timevalcmp(&rtv, &atv, >=))
1220 goto done;
1221 ttv = atv;
1222 timevalsub(&ttv, &rtv);
1223 timeout = ttv.tv_sec > 24 * 60 * 60 ?
1224 24 * 60 * 60 * hz : tvtohz(&ttv);
1225 }
1226
1227 start:
1228 kevp = keva;
1229 if (kq->kq_count == 0) {
1230 if (timeout < 0) {
1231 error = EWOULDBLOCK;
1232 } else {
1233 kq->kq_state |= KQ_SLEEP;
1234 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH,
1235 "kqread", timeout);
1236 }
1237 if (error == 0)
1238 goto retry;
1239 /* don't restart after signals... */
1240 if (error == ERESTART)
1241 error = EINTR;
1242 else if (error == EWOULDBLOCK)
1243 error = 0;
1244 goto done;
1245 }
1246
1247 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
1248 influx = 0;
1249 while (count) {
1250 KQ_OWNED(kq);
1251 kn = TAILQ_FIRST(&kq->kq_head);
1252
1253 if ((kn->kn_status == KN_MARKER && kn != marker) ||
1254 (kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1255 if (influx) {
1256 influx = 0;
1257 KQ_FLUX_WAKEUP(kq);
1258 }
1259 kq->kq_state |= KQ_FLUXWAIT;
1260 error = msleep(kq, &kq->kq_lock, PSOCK,
1261 "kqflxwt", 0);
1262 continue;
1263 }
1264
1265 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1266 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
1267 kn->kn_status &= ~KN_QUEUED;
1268 kq->kq_count--;
1269 continue;
1270 }
1271 if (kn == marker) {
1272 KQ_FLUX_WAKEUP(kq);
1273 if (count == maxevents)
1274 goto retry;
1275 goto done;
1276 }
1277 KASSERT((kn->kn_status & KN_INFLUX) == 0,
1278 ("KN_INFLUX set when not suppose to be"));
1279
1280 if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
1281 kn->kn_status &= ~KN_QUEUED;
1282 kn->kn_status |= KN_INFLUX;
1283 kq->kq_count--;
1284 KQ_UNLOCK(kq);
1285 /*
1286 * We don't need to lock the list since we've marked
1287 * it _INFLUX.
1288 */
1289 *kevp = kn->kn_kevent;
1290 if (!(kn->kn_status & KN_DETACHED))
1291 kn->kn_fop->f_detach(kn);
1292 knote_drop(kn, td);
1293 KQ_LOCK(kq);
1294 kn = NULL;
1295 } else {
1296 kn->kn_status |= KN_INFLUX;
1297 KQ_UNLOCK(kq);
1298 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
1299 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1300 KN_LIST_LOCK(kn);
1301 if (kn->kn_fop->f_event(kn, 0) == 0) {
1302 KQ_LOCK(kq);
1303 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1304 kn->kn_status &=
1305 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX);
1306 kq->kq_count--;
1307 KN_LIST_UNLOCK(kn);
1308 influx = 1;
1309 continue;
1310 }
1311 *kevp = kn->kn_kevent;
1312 KQ_LOCK(kq);
1313 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1314 if (kn->kn_flags & EV_CLEAR) {
1315 kn->kn_data = 0;
1316 kn->kn_fflags = 0;
1317 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
1318 kq->kq_count--;
1319 } else
1320 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1321
1322 kn->kn_status &= ~(KN_INFLUX);
1323 KN_LIST_UNLOCK(kn);
1324 influx = 1;
1325 }
1326
1327 /* we are returning a copy to the user */
1328 kevp++;
1329 nkev++;
1330 count--;
1331
1332 if (nkev == KQ_NEVENTS) {
1333 influx = 0;
1334 KQ_UNLOCK_FLUX(kq);
1335 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1336 nkev = 0;
1337 kevp = keva;
1338 KQ_LOCK(kq);
1339 if (error)
1340 break;
1341 }
1342 }
1343 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
1344 done:
1345 KQ_OWNED(kq);
1346 KQ_UNLOCK_FLUX(kq);
1347 knote_free(marker);
1348 done_nl:
1349 KQ_NOTOWNED(kq);
1350 if (nkev != 0)
1351 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
1352 td->td_retval[0] = maxevents - count;
1353 return (error);
1354 }
1355
1356 /*
1357 * XXX
1358 * This could be expanded to call kqueue_scan, if desired.
1359 */
1360 /*ARGSUSED*/
1361 static int
1362 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
1363 int flags, struct thread *td)
1364 {
1365 return (ENXIO);
1366 }
1367
1368 /*ARGSUSED*/
1369 static int
1370 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
1371 int flags, struct thread *td)
1372 {
1373 return (ENXIO);
1374 }
1375
1376 /*ARGSUSED*/
1377 static int
1378 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
1379 struct ucred *active_cred, struct thread *td)
1380 {
1381 /*
1382 * Enabling sigio causes two major problems:
1383 * 1) infinite recursion:
1384 * Synopsys: kevent is being used to track signals and have FIOASYNC
1385 * set. On receipt of a signal this will cause a kqueue to recurse
1386 * into itself over and over. Sending the sigio causes the kqueue
1387 * to become ready, which in turn posts sigio again, forever.
1388 * Solution: this can be solved by setting a flag in the kqueue that
1389 * we have a SIGIO in progress.
1390 * 2) locking problems:
1391 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
1392 * us above the proc and pgrp locks.
1393 * Solution: Post a signal using an async mechanism, being sure to
1394 * record a generation count in the delivery so that we do not deliver
1395 * a signal to the wrong process.
1396 *
1397 * Note, these two mechanisms are somewhat mutually exclusive!
1398 */
1399 #if 0
1400 struct kqueue *kq;
1401
1402 kq = fp->f_data;
1403 switch (cmd) {
1404 case FIOASYNC:
1405 if (*(int *)data) {
1406 kq->kq_state |= KQ_ASYNC;
1407 } else {
1408 kq->kq_state &= ~KQ_ASYNC;
1409 }
1410 return (0);
1411
1412 case FIOSETOWN:
1413 return (fsetown(*(int *)data, &kq->kq_sigio));
1414
1415 case FIOGETOWN:
1416 *(int *)data = fgetown(&kq->kq_sigio);
1417 return (0);
1418 }
1419 #endif
1420
1421 return (ENOTTY);
1422 }
1423
1424 /*ARGSUSED*/
1425 static int
1426 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
1427 struct thread *td)
1428 {
1429 struct kqueue *kq;
1430 int revents = 0;
1431 int error;
1432
1433 if ((error = kqueue_aquire(fp, &kq)))
1434 return POLLERR;
1435
1436 KQ_LOCK(kq);
1437 if (events & (POLLIN | POLLRDNORM)) {
1438 if (kq->kq_count) {
1439 revents |= events & (POLLIN | POLLRDNORM);
1440 } else {
1441 selrecord(td, &kq->kq_sel);
1442 kq->kq_state |= KQ_SEL;
1443 }
1444 }
1445 kqueue_release(kq, 1);
1446 KQ_UNLOCK(kq);
1447 return (revents);
1448 }
1449
1450 /*ARGSUSED*/
1451 static int
1452 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
1453 struct thread *td)
1454 {
1455
1456 bzero((void *)st, sizeof *st);
1457 /*
1458 * We no longer return kq_count because the unlocked value is useless.
1459 * If you spent all this time getting the count, why not spend your
1460 * syscall better by calling kevent?
1461 *
1462 * XXX - This is needed for libc_r.
1463 */
1464 st->st_mode = S_IFIFO;
1465 return (0);
1466 }
1467
1468 /*ARGSUSED*/
1469 static int
1470 kqueue_close(struct file *fp, struct thread *td)
1471 {
1472 struct kqueue *kq = fp->f_data;
1473 struct filedesc *fdp;
1474 struct knote *kn;
1475 int i;
1476 int error;
1477
1478 if ((error = kqueue_aquire(fp, &kq)))
1479 return error;
1480
1481 KQ_LOCK(kq);
1482
1483 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
1484 ("kqueue already closing"));
1485 kq->kq_state |= KQ_CLOSING;
1486 if (kq->kq_refcnt > 1)
1487 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
1488
1489 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
1490 fdp = kq->kq_fdp;
1491
1492 KASSERT(knlist_empty(&kq->kq_sel.si_note),
1493 ("kqueue's knlist not empty"));
1494
1495 for (i = 0; i < kq->kq_knlistsize; i++) {
1496 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
1497 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1498 kq->kq_state |= KQ_FLUXWAIT;
1499 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
1500 continue;
1501 }
1502 kn->kn_status |= KN_INFLUX;
1503 KQ_UNLOCK(kq);
1504 if (!(kn->kn_status & KN_DETACHED))
1505 kn->kn_fop->f_detach(kn);
1506 knote_drop(kn, td);
1507 KQ_LOCK(kq);
1508 }
1509 }
1510 if (kq->kq_knhashmask != 0) {
1511 for (i = 0; i <= kq->kq_knhashmask; i++) {
1512 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
1513 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) {
1514 kq->kq_state |= KQ_FLUXWAIT;
1515 msleep(kq, &kq->kq_lock, PSOCK,
1516 "kqclo2", 0);
1517 continue;
1518 }
1519 kn->kn_status |= KN_INFLUX;
1520 KQ_UNLOCK(kq);
1521 if (!(kn->kn_status & KN_DETACHED))
1522 kn->kn_fop->f_detach(kn);
1523 knote_drop(kn, td);
1524 KQ_LOCK(kq);
1525 }
1526 }
1527 }
1528
1529 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
1530 kq->kq_state |= KQ_TASKDRAIN;
1531 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
1532 }
1533
1534 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1535 kq->kq_state &= ~KQ_SEL;
1536 selwakeuppri(&kq->kq_sel, PSOCK);
1537 }
1538
1539 KQ_UNLOCK(kq);
1540
1541 FILEDESC_LOCK_FAST(fdp);
1542 SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list);
1543 FILEDESC_UNLOCK_FAST(fdp);
1544
1545 knlist_destroy(&kq->kq_sel.si_note);
1546 mtx_destroy(&kq->kq_lock);
1547 kq->kq_fdp = NULL;
1548
1549 if (kq->kq_knhash != NULL)
1550 free(kq->kq_knhash, M_KQUEUE);
1551 if (kq->kq_knlist != NULL)
1552 free(kq->kq_knlist, M_KQUEUE);
1553
1554 funsetown(&kq->kq_sigio);
1555 free(kq, M_KQUEUE);
1556 fp->f_data = NULL;
1557
1558 return (0);
1559 }
1560
1561 static void
1562 kqueue_wakeup(struct kqueue *kq)
1563 {
1564 KQ_OWNED(kq);
1565
1566 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
1567 kq->kq_state &= ~KQ_SLEEP;
1568 wakeup(kq);
1569 }
1570 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
1571 kq->kq_state &= ~KQ_SEL;
1572 selwakeuppri(&kq->kq_sel, PSOCK);
1573 }
1574 if (!knlist_empty(&kq->kq_sel.si_note))
1575 kqueue_schedtask(kq);
1576 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
1577 pgsigio(&kq->kq_sigio, SIGIO, 0);
1578 }
1579 }
1580
1581 /*
1582 * Walk down a list of knotes, activating them if their event has triggered.
1583 *
1584 * There is a possibility to optimize in the case of one kq watching another.
1585 * Instead of scheduling a task to wake it up, you could pass enough state
1586 * down the chain to make up the parent kqueue. Make this code functional
1587 * first.
1588 */
1589 void
1590 knote(struct knlist *list, long hint, int islocked)
1591 {
1592 struct kqueue *kq;
1593 struct knote *kn;
1594
1595 if (list == NULL)
1596 return;
1597
1598 KNL_ASSERT_LOCK(list, islocked);
1599
1600 if (!islocked)
1601 list->kl_lock(list->kl_lockarg);
1602
1603 /*
1604 * If we unlock the list lock (and set KN_INFLUX), we can eliminate
1605 * the kqueue scheduling, but this will introduce four
1606 * lock/unlock's for each knote to test. If we do, continue to use
1607 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is
1608 * only safe if you want to remove the current item, which we are
1609 * not doing.
1610 */
1611 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
1612 kq = kn->kn_kq;
1613 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1614 KQ_LOCK(kq);
1615 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) {
1616 kn->kn_status |= KN_HASKQLOCK;
1617 if (kn->kn_fop->f_event(kn, hint))
1618 KNOTE_ACTIVATE(kn, 1);
1619 kn->kn_status &= ~KN_HASKQLOCK;
1620 }
1621 KQ_UNLOCK(kq);
1622 }
1623 kq = NULL;
1624 }
1625 if (!islocked)
1626 list->kl_unlock(list->kl_lockarg);
1627 }
1628
1629 /*
1630 * add a knote to a knlist
1631 */
1632 void
1633 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
1634 {
1635 KNL_ASSERT_LOCK(knl, islocked);
1636 KQ_NOTOWNED(kn->kn_kq);
1637 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) ==
1638 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED"));
1639 if (!islocked)
1640 knl->kl_lock(knl->kl_lockarg);
1641 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
1642 if (!islocked)
1643 knl->kl_unlock(knl->kl_lockarg);
1644 KQ_LOCK(kn->kn_kq);
1645 kn->kn_knlist = knl;
1646 kn->kn_status &= ~KN_DETACHED;
1647 KQ_UNLOCK(kn->kn_kq);
1648 }
1649
1650 static void
1651 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked)
1652 {
1653 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked"));
1654 KNL_ASSERT_LOCK(knl, knlislocked);
1655 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
1656 if (!kqislocked)
1657 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX,
1658 ("knlist_remove called w/o knote being KN_INFLUX or already removed"));
1659 if (!knlislocked)
1660 knl->kl_lock(knl->kl_lockarg);
1661 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
1662 kn->kn_knlist = NULL;
1663 if (!knlislocked)
1664 knl->kl_unlock(knl->kl_lockarg);
1665 if (!kqislocked)
1666 KQ_LOCK(kn->kn_kq);
1667 kn->kn_status |= KN_DETACHED;
1668 if (!kqislocked)
1669 KQ_UNLOCK(kn->kn_kq);
1670 }
1671
1672 /*
1673 * remove all knotes from a specified klist
1674 */
1675 void
1676 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
1677 {
1678
1679 knlist_remove_kq(knl, kn, islocked, 0);
1680 }
1681
1682 /*
1683 * remove knote from a specified klist while in f_event handler.
1684 */
1685 void
1686 knlist_remove_inevent(struct knlist *knl, struct knote *kn)
1687 {
1688
1689 knlist_remove_kq(knl, kn, 1,
1690 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK);
1691 }
1692
1693 int
1694 knlist_empty(struct knlist *knl)
1695 {
1696 KNL_ASSERT_LOCKED(knl);
1697 return SLIST_EMPTY(&knl->kl_list);
1698 }
1699
1700 static struct mtx knlist_lock;
1701 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
1702 MTX_DEF);
1703 static void knlist_mtx_lock(void *arg);
1704 static void knlist_mtx_unlock(void *arg);
1705 static int knlist_mtx_locked(void *arg);
1706
1707 static void
1708 knlist_mtx_lock(void *arg)
1709 {
1710 mtx_lock((struct mtx *)arg);
1711 }
1712
1713 static void
1714 knlist_mtx_unlock(void *arg)
1715 {
1716 mtx_unlock((struct mtx *)arg);
1717 }
1718
1719 static int
1720 knlist_mtx_locked(void *arg)
1721 {
1722 return (mtx_owned((struct mtx *)arg));
1723 }
1724
1725 void
1726 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
1727 void (*kl_unlock)(void *), int (*kl_locked)(void *))
1728 {
1729
1730 if (lock == NULL)
1731 knl->kl_lockarg = &knlist_lock;
1732 else
1733 knl->kl_lockarg = lock;
1734
1735 if (kl_lock == NULL)
1736 knl->kl_lock = knlist_mtx_lock;
1737 else
1738 knl->kl_lock = kl_lock;
1739 if (kl_unlock == NULL)
1740 knl->kl_unlock = knlist_mtx_unlock;
1741 else
1742 knl->kl_unlock = kl_unlock;
1743 if (kl_locked == NULL)
1744 knl->kl_locked = knlist_mtx_locked;
1745 else
1746 knl->kl_locked = kl_locked;
1747
1748 SLIST_INIT(&knl->kl_list);
1749 }
1750
1751 void
1752 knlist_destroy(struct knlist *knl)
1753 {
1754
1755 #ifdef INVARIANTS
1756 /*
1757 * if we run across this error, we need to find the offending
1758 * driver and have it call knlist_clear.
1759 */
1760 if (!SLIST_EMPTY(&knl->kl_list))
1761 printf("WARNING: destroying knlist w/ knotes on it!\n");
1762 #endif
1763
1764 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL;
1765 SLIST_INIT(&knl->kl_list);
1766 }
1767
1768 /*
1769 * Even if we are locked, we may need to drop the lock to allow any influx
1770 * knotes time to "settle".
1771 */
1772 void
1773 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
1774 {
1775 struct knote *kn, *kn2;
1776 struct kqueue *kq;
1777
1778 if (islocked)
1779 KNL_ASSERT_LOCKED(knl);
1780 else {
1781 KNL_ASSERT_UNLOCKED(knl);
1782 again: /* need to reaquire lock since we have dropped it */
1783 knl->kl_lock(knl->kl_lockarg);
1784 }
1785
1786 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
1787 kq = kn->kn_kq;
1788 KQ_LOCK(kq);
1789 if ((kn->kn_status & KN_INFLUX)) {
1790 KQ_UNLOCK(kq);
1791 continue;
1792 }
1793 knlist_remove_kq(knl, kn, 1, 1);
1794 if (killkn) {
1795 kn->kn_status |= KN_INFLUX | KN_DETACHED;
1796 KQ_UNLOCK(kq);
1797 knote_drop(kn, td);
1798 } else {
1799 /* Make sure cleared knotes disappear soon */
1800 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1801 KQ_UNLOCK(kq);
1802 }
1803 kq = NULL;
1804 }
1805
1806 if (!SLIST_EMPTY(&knl->kl_list)) {
1807 /* there are still KN_INFLUX remaining */
1808 kn = SLIST_FIRST(&knl->kl_list);
1809 kq = kn->kn_kq;
1810 KQ_LOCK(kq);
1811 KASSERT(kn->kn_status & KN_INFLUX,
1812 ("knote removed w/o list lock"));
1813 knl->kl_unlock(knl->kl_lockarg);
1814 kq->kq_state |= KQ_FLUXWAIT;
1815 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
1816 kq = NULL;
1817 goto again;
1818 }
1819
1820 if (islocked)
1821 KNL_ASSERT_LOCKED(knl);
1822 else {
1823 knl->kl_unlock(knl->kl_lockarg);
1824 KNL_ASSERT_UNLOCKED(knl);
1825 }
1826 }
1827
1828 /*
1829 * remove all knotes referencing a specified fd
1830 * must be called with FILEDESC lock. This prevents a race where a new fd
1831 * comes along and occupies the entry and we attach a knote to the fd.
1832 */
1833 void
1834 knote_fdclose(struct thread *td, int fd)
1835 {
1836 struct filedesc *fdp = td->td_proc->p_fd;
1837 struct kqueue *kq;
1838 struct knote *kn;
1839 int influx;
1840
1841 FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1842
1843 /*
1844 * We shouldn't have to worry about new kevents appearing on fd
1845 * since filedesc is locked.
1846 */
1847 SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
1848 KQ_LOCK(kq);
1849
1850 again:
1851 influx = 0;
1852 while (kq->kq_knlistsize > fd &&
1853 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
1854 if (kn->kn_status & KN_INFLUX) {
1855 /* someone else might be waiting on our knote */
1856 if (influx)
1857 wakeup(kq);
1858 kq->kq_state |= KQ_FLUXWAIT;
1859 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
1860 goto again;
1861 }
1862 kn->kn_status |= KN_INFLUX;
1863 KQ_UNLOCK(kq);
1864 if (!(kn->kn_status & KN_DETACHED))
1865 kn->kn_fop->f_detach(kn);
1866 knote_drop(kn, td);
1867 influx = 1;
1868 KQ_LOCK(kq);
1869 }
1870 KQ_UNLOCK_FLUX(kq);
1871 }
1872 }
1873
1874 static int
1875 knote_attach(struct knote *kn, struct kqueue *kq)
1876 {
1877 struct klist *list;
1878
1879 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX"));
1880 KQ_OWNED(kq);
1881
1882 if (kn->kn_fop->f_isfd) {
1883 if (kn->kn_id >= kq->kq_knlistsize)
1884 return ENOMEM;
1885 list = &kq->kq_knlist[kn->kn_id];
1886 } else {
1887 if (kq->kq_knhash == NULL)
1888 return ENOMEM;
1889 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1890 }
1891
1892 SLIST_INSERT_HEAD(list, kn, kn_link);
1893
1894 return 0;
1895 }
1896
1897 /*
1898 * knote must already have been detatched using the f_detach method.
1899 * no lock need to be held, it is assumed that the KN_INFLUX flag is set
1900 * to prevent other removal.
1901 */
1902 static void
1903 knote_drop(struct knote *kn, struct thread *td)
1904 {
1905 struct kqueue *kq;
1906 struct klist *list;
1907
1908 kq = kn->kn_kq;
1909
1910 KQ_NOTOWNED(kq);
1911 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX,
1912 ("knote_drop called without KN_INFLUX set in kn_status"));
1913
1914 KQ_LOCK(kq);
1915 if (kn->kn_fop->f_isfd)
1916 list = &kq->kq_knlist[kn->kn_id];
1917 else
1918 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
1919
1920 SLIST_REMOVE(list, kn, knote, kn_link);
1921 if (kn->kn_status & KN_QUEUED)
1922 knote_dequeue(kn);
1923 KQ_UNLOCK_FLUX(kq);
1924
1925 if (kn->kn_fop->f_isfd) {
1926 fdrop(kn->kn_fp, td);
1927 kn->kn_fp = NULL;
1928 }
1929 kqueue_fo_release(kn->kn_kevent.filter);
1930 kn->kn_fop = NULL;
1931 knote_free(kn);
1932 }
1933
1934 static void
1935 knote_enqueue(struct knote *kn)
1936 {
1937 struct kqueue *kq = kn->kn_kq;
1938
1939 KQ_OWNED(kn->kn_kq);
1940 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1941
1942 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1943 kn->kn_status |= KN_QUEUED;
1944 kq->kq_count++;
1945 kqueue_wakeup(kq);
1946 }
1947
1948 static void
1949 knote_dequeue(struct knote *kn)
1950 {
1951 struct kqueue *kq = kn->kn_kq;
1952
1953 KQ_OWNED(kn->kn_kq);
1954 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1955
1956 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1957 kn->kn_status &= ~KN_QUEUED;
1958 kq->kq_count--;
1959 }
1960
1961 static void
1962 knote_init(void)
1963 {
1964
1965 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
1966 NULL, NULL, UMA_ALIGN_PTR, 0);
1967 }
1968 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1969
1970 static struct knote *
1971 knote_alloc(int waitok)
1972 {
1973 return ((struct knote *)uma_zalloc(knote_zone,
1974 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO));
1975 }
1976
1977 static void
1978 knote_free(struct knote *kn)
1979 {
1980 if (kn != NULL)
1981 uma_zfree(knote_zone, kn);
1982 }
Cache object: e9f7ad2d934267b573bcedc51f318729
|