1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
5 * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org>
6 * Copyright (c) 2009 Apple, Inc.
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 #include "opt_ktrace.h"
32 #include "opt_kqueue.h"
33
34 #ifdef COMPAT_FREEBSD11
35 #define _WANT_FREEBSD11_KEVENT
36 #endif
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/capsicum.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/malloc.h>
47 #include <sys/unistd.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/fcntl.h>
52 #include <sys/imgact.h>
53 #include <sys/jail.h>
54 #include <sys/jaildesc.h>
55 #include <sys/kthread.h>
56 #include <sys/selinfo.h>
57 #include <sys/queue.h>
58 #include <sys/event.h>
59 #include <sys/eventvar.h>
60 #include <sys/poll.h>
61 #include <sys/protosw.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sbuf.h>
64 #include <sys/sigio.h>
65 #include <sys/signalvar.h>
66 #include <sys/socket.h>
67 #include <sys/socketvar.h>
68 #include <sys/stat.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/sysproto.h>
72 #include <sys/syscallsubr.h>
73 #include <sys/taskqueue.h>
74 #include <sys/uio.h>
75 #include <sys/user.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79 #include <machine/atomic.h>
80 #ifdef COMPAT_FREEBSD32
81 #include <compat/freebsd32/freebsd32.h>
82 #include <compat/freebsd32/freebsd32_util.h>
83 #endif
84
85 #include <vm/uma.h>
86
87 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
88
89 /*
90 * This lock is used if multiple kq locks are required. This possibly
91 * should be made into a per proc lock.
92 */
93 static struct mtx kq_global;
94 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
95 #define KQ_GLOBAL_LOCK(lck, haslck) do { \
96 if (!haslck) \
97 mtx_lock(lck); \
98 haslck = 1; \
99 } while (0)
100 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \
101 if (haslck) \
102 mtx_unlock(lck); \
103 haslck = 0; \
104 } while (0)
105
106 TASKQUEUE_DEFINE_THREAD(kqueue_ctx);
107
108 static int kevent_copyout(void *arg, struct kevent *kevp, int count);
109 static int kevent_copyin(void *arg, struct kevent *kevp, int count);
110 static int kqueue_register(struct kqueue *kq, struct kevent *kev,
111 struct thread *td, int mflag);
112 static int kqueue_acquire(struct file *fp, struct kqueue **kqp);
113 static void kqueue_release(struct kqueue *kq, int locked);
114 static void kqueue_destroy(struct kqueue *kq);
115 static void kqueue_drain(struct kqueue *kq, struct thread *td);
116 static int kqueue_expand(struct kqueue *kq, const struct filterops *fops,
117 uintptr_t ident, int mflag);
118 static void kqueue_task(void *arg, int pending);
119 static int kqueue_scan(struct kqueue *kq, int maxevents,
120 struct kevent_copyops *k_ops,
121 const struct timespec *timeout,
122 struct kevent *keva, struct thread *td);
123 static void kqueue_wakeup(struct kqueue *kq);
124 static const struct filterops *kqueue_fo_find(int filt);
125 static void kqueue_fo_release(int filt);
126 struct g_kevent_args;
127 static int kern_kevent_generic(struct thread *td,
128 struct g_kevent_args *uap,
129 struct kevent_copyops *k_ops, const char *struct_name);
130
131 static fo_ioctl_t kqueue_ioctl;
132 static fo_poll_t kqueue_poll;
133 static fo_kqfilter_t kqueue_kqfilter;
134 static fo_stat_t kqueue_stat;
135 static fo_close_t kqueue_close;
136 static fo_fill_kinfo_t kqueue_fill_kinfo;
137 static fo_fork_t kqueue_fork;
138
139 static const struct fileops kqueueops = {
140 .fo_read = invfo_rdwr,
141 .fo_write = invfo_rdwr,
142 .fo_truncate = invfo_truncate,
143 .fo_ioctl = kqueue_ioctl,
144 .fo_poll = kqueue_poll,
145 .fo_kqfilter = kqueue_kqfilter,
146 .fo_stat = kqueue_stat,
147 .fo_close = kqueue_close,
148 .fo_chmod = invfo_chmod,
149 .fo_chown = invfo_chown,
150 .fo_sendfile = invfo_sendfile,
151 .fo_cmp = file_kcmp_generic,
152 .fo_fork = kqueue_fork,
153 .fo_fill_kinfo = kqueue_fill_kinfo,
154 .fo_flags = DFLAG_FORK,
155 };
156
157 static int knote_attach(struct knote *kn, struct kqueue *kq);
158 static void knote_drop(struct knote *kn, struct thread *td);
159 static void knote_drop_detached(struct knote *kn, struct thread *td);
160 static void knote_enqueue(struct knote *kn);
161 static void knote_dequeue(struct knote *kn);
162 static void knote_init(void *);
163 static struct knote *knote_alloc(int mflag);
164 static void knote_free(struct knote *kn);
165
166 static void filt_kqdetach(struct knote *kn);
167 static int filt_kqueue(struct knote *kn, long hint);
168 static int filt_procattach(struct knote *kn);
169 static void filt_procdetach(struct knote *kn);
170 static int filt_proc(struct knote *kn, long hint);
171 static int filt_jailattach(struct knote *kn);
172 static void filt_jaildetach(struct knote *kn);
173 static int filt_jail(struct knote *kn, long hint);
174 static int filt_fileattach(struct knote *kn);
175 static void filt_timerexpire(void *knx);
176 static void filt_timerexpire_l(struct knote *kn, bool proc_locked);
177 static int filt_timerattach(struct knote *kn);
178 static void filt_timerdetach(struct knote *kn);
179 static void filt_timerstart(struct knote *kn, sbintime_t to);
180 static void filt_timertouch(struct knote *kn, struct kevent *kev,
181 u_long type);
182 static int filt_timercopy(struct knote *kn, struct proc *p1);
183 static int filt_timervalidate(struct knote *kn, sbintime_t *to);
184 static int filt_timer(struct knote *kn, long hint);
185 static int filt_userattach(struct knote *kn);
186 static void filt_userdetach(struct knote *kn);
187 static int filt_user(struct knote *kn, long hint);
188 static void filt_usertouch(struct knote *kn, struct kevent *kev,
189 u_long type);
190
191 static const struct filterops file_filtops = {
192 .f_isfd = 1,
193 .f_attach = filt_fileattach,
194 .f_copy = knote_triv_copy,
195 };
196 static const struct filterops kqread_filtops = {
197 .f_isfd = 1,
198 .f_detach = filt_kqdetach,
199 .f_event = filt_kqueue,
200 .f_copy = knote_triv_copy,
201 };
202 /* XXX - move to kern_proc.c? */
203 static const struct filterops proc_filtops = {
204 .f_isfd = 0,
205 .f_attach = filt_procattach,
206 .f_detach = filt_procdetach,
207 .f_event = filt_proc,
208 .f_copy = knote_triv_copy,
209 };
210 static const struct filterops jail_filtops = {
211 .f_isfd = 0,
212 .f_attach = filt_jailattach,
213 .f_detach = filt_jaildetach,
214 .f_event = filt_jail,
215 .f_copy = knote_triv_copy,
216 };
217 static const struct filterops timer_filtops = {
218 .f_isfd = 0,
219 .f_attach = filt_timerattach,
220 .f_detach = filt_timerdetach,
221 .f_event = filt_timer,
222 .f_touch = filt_timertouch,
223 .f_copy = filt_timercopy,
224 };
225 static const struct filterops user_filtops = {
226 .f_attach = filt_userattach,
227 .f_detach = filt_userdetach,
228 .f_event = filt_user,
229 .f_touch = filt_usertouch,
230 .f_copy = knote_triv_copy,
231 };
232
233 static uma_zone_t knote_zone;
234 static unsigned int __exclusive_cache_line kq_ncallouts;
235 static unsigned int kq_calloutmax = 4 * 1024;
236 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
237 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
238
239 /* XXX - ensure not influx ? */
240 #define KNOTE_ACTIVATE(kn, islock) do { \
241 if ((islock)) \
242 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \
243 else \
244 KQ_LOCK((kn)->kn_kq); \
245 (kn)->kn_status |= KN_ACTIVE; \
246 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \
247 knote_enqueue((kn)); \
248 if (!(islock)) \
249 KQ_UNLOCK((kn)->kn_kq); \
250 } while (0)
251 #define KQ_LOCK(kq) do { \
252 mtx_lock(&(kq)->kq_lock); \
253 } while (0)
254 #define KQ_FLUX_WAKEUP(kq) do { \
255 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \
256 (kq)->kq_state &= ~KQ_FLUXWAIT; \
257 wakeup((kq)); \
258 } \
259 } while (0)
260 #define KQ_UNLOCK_FLUX(kq) do { \
261 KQ_FLUX_WAKEUP(kq); \
262 mtx_unlock(&(kq)->kq_lock); \
263 } while (0)
264 #define KQ_UNLOCK(kq) do { \
265 mtx_unlock(&(kq)->kq_lock); \
266 } while (0)
267 #define KQ_OWNED(kq) do { \
268 mtx_assert(&(kq)->kq_lock, MA_OWNED); \
269 } while (0)
270 #define KQ_NOTOWNED(kq) do { \
271 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \
272 } while (0)
273
274 static struct knlist *
kn_list_lock(struct knote * kn)275 kn_list_lock(struct knote *kn)
276 {
277 struct knlist *knl;
278
279 knl = kn->kn_knlist;
280 if (knl != NULL)
281 knl->kl_lock(knl->kl_lockarg);
282 return (knl);
283 }
284
285 static void
kn_list_unlock(struct knlist * knl)286 kn_list_unlock(struct knlist *knl)
287 {
288 bool do_free;
289
290 if (knl == NULL)
291 return;
292 do_free = knl->kl_autodestroy && knlist_empty(knl);
293 knl->kl_unlock(knl->kl_lockarg);
294 if (do_free) {
295 knlist_destroy(knl);
296 free(knl, M_KQUEUE);
297 }
298 }
299
300 static bool
kn_in_flux(struct knote * kn)301 kn_in_flux(struct knote *kn)
302 {
303
304 return (kn->kn_influx > 0);
305 }
306
307 static void
kn_enter_flux(struct knote * kn)308 kn_enter_flux(struct knote *kn)
309 {
310
311 KQ_OWNED(kn->kn_kq);
312 MPASS(kn->kn_influx < INT_MAX);
313 kn->kn_influx++;
314 }
315
316 static bool
kn_leave_flux(struct knote * kn)317 kn_leave_flux(struct knote *kn)
318 {
319
320 KQ_OWNED(kn->kn_kq);
321 MPASS(kn->kn_influx > 0);
322 kn->kn_influx--;
323 return (kn->kn_influx == 0);
324 }
325
326 #define KNL_ASSERT_LOCK(knl, islocked) do { \
327 if (islocked) \
328 KNL_ASSERT_LOCKED(knl); \
329 else \
330 KNL_ASSERT_UNLOCKED(knl); \
331 } while (0)
332 #ifdef INVARIANTS
333 #define KNL_ASSERT_LOCKED(knl) do { \
334 knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED); \
335 } while (0)
336 #define KNL_ASSERT_UNLOCKED(knl) do { \
337 knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED); \
338 } while (0)
339 #else /* !INVARIANTS */
340 #define KNL_ASSERT_LOCKED(knl) do {} while (0)
341 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0)
342 #endif /* INVARIANTS */
343
344 #ifndef KN_HASHSIZE
345 #define KN_HASHSIZE 64 /* XXX should be tunable */
346 #endif
347
348 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
349
350 static int
filt_nullattach(struct knote * kn)351 filt_nullattach(struct knote *kn)
352 {
353
354 return (ENXIO);
355 };
356
357 static const struct filterops null_filtops = {
358 .f_isfd = 0,
359 .f_attach = filt_nullattach,
360 .f_copy = knote_triv_copy,
361 };
362
363 /* XXX - make SYSINIT to add these, and move into respective modules. */
364 extern const struct filterops sig_filtops;
365 extern const struct filterops fs_filtops;
366
367 /*
368 * Table for all system-defined filters.
369 */
370 static struct mtx filterops_lock;
371 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF);
372 static struct {
373 const struct filterops *for_fop;
374 int for_nolock;
375 int for_refcnt;
376 } sysfilt_ops[EVFILT_SYSCOUNT] = {
377 [~EVFILT_READ] = { &file_filtops, 1 },
378 [~EVFILT_WRITE] = { &file_filtops, 1 },
379 [~EVFILT_AIO] = { &null_filtops },
380 [~EVFILT_VNODE] = { &file_filtops, 1 },
381 [~EVFILT_PROC] = { &proc_filtops, 1 },
382 [~EVFILT_SIGNAL] = { &sig_filtops, 1 },
383 [~EVFILT_TIMER] = { &timer_filtops, 1 },
384 [~EVFILT_PROCDESC] = { &file_filtops, 1 },
385 [~EVFILT_FS] = { &fs_filtops, 1 },
386 [~EVFILT_LIO] = { &null_filtops },
387 [~EVFILT_USER] = { &user_filtops, 1 },
388 [~EVFILT_SENDFILE] = { &null_filtops },
389 [~EVFILT_EMPTY] = { &file_filtops, 1 },
390 [~EVFILT_JAIL] = { &jail_filtops, 1 },
391 [~EVFILT_JAILDESC] = { &file_filtops, 1 },
392 };
393
394 /*
395 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
396 * method.
397 */
398 static int
filt_fileattach(struct knote * kn)399 filt_fileattach(struct knote *kn)
400 {
401
402 return (fo_kqfilter(kn->kn_fp, kn));
403 }
404
405 /*ARGSUSED*/
406 static int
kqueue_kqfilter(struct file * fp,struct knote * kn)407 kqueue_kqfilter(struct file *fp, struct knote *kn)
408 {
409 struct kqueue *kq = kn->kn_fp->f_data;
410
411 if (kn->kn_filter != EVFILT_READ)
412 return (EINVAL);
413
414 kn->kn_status |= KN_KQUEUE;
415 kn->kn_fop = &kqread_filtops;
416 knlist_add(&kq->kq_sel.si_note, kn, 0);
417
418 return (0);
419 }
420
421 static void
filt_kqdetach(struct knote * kn)422 filt_kqdetach(struct knote *kn)
423 {
424 struct kqueue *kq = kn->kn_fp->f_data;
425
426 knlist_remove(&kq->kq_sel.si_note, kn, 0);
427 }
428
429 /*ARGSUSED*/
430 static int
filt_kqueue(struct knote * kn,long hint)431 filt_kqueue(struct knote *kn, long hint)
432 {
433 struct kqueue *kq = kn->kn_fp->f_data;
434
435 kn->kn_data = kq->kq_count;
436 return (kn->kn_data > 0);
437 }
438
439 /* XXX - move to kern_proc.c? */
440 static int
filt_procattach(struct knote * kn)441 filt_procattach(struct knote *kn)
442 {
443 struct proc *p;
444 int error;
445 bool exiting, immediate;
446
447 exiting = immediate = false;
448 if (kn->kn_sfflags & NOTE_EXIT)
449 p = pfind_any(kn->kn_id);
450 else
451 p = pfind(kn->kn_id);
452 if (p == NULL)
453 return (ESRCH);
454 if (p->p_flag & P_WEXIT)
455 exiting = true;
456
457 if ((error = p_cansee(curthread, p))) {
458 PROC_UNLOCK(p);
459 return (error);
460 }
461
462 kn->kn_ptr.p_proc = p;
463 kn->kn_flags |= EV_CLEAR; /* automatically set */
464
465 /*
466 * Internal flag indicating registration done by kernel for the
467 * purposes of getting a NOTE_CHILD notification.
468 */
469 if (kn->kn_flags & EV_FLAG2) {
470 kn->kn_flags &= ~EV_FLAG2;
471 kn->kn_data = kn->kn_sdata; /* ppid */
472 kn->kn_fflags = NOTE_CHILD;
473 kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
474 immediate = true; /* Force immediate activation of child note. */
475 }
476 /*
477 * Internal flag indicating registration done by kernel (for other than
478 * NOTE_CHILD).
479 */
480 if (kn->kn_flags & EV_FLAG1) {
481 kn->kn_flags &= ~EV_FLAG1;
482 }
483
484 knlist_add(p->p_klist, kn, 1);
485
486 /*
487 * Immediately activate any child notes or, in the case of a zombie
488 * target process, exit notes. The latter is necessary to handle the
489 * case where the target process, e.g. a child, dies before the kevent
490 * is registered.
491 */
492 if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
493 KNOTE_ACTIVATE(kn, 0);
494
495 PROC_UNLOCK(p);
496
497 return (0);
498 }
499
500 /*
501 * The knote may be attached to a different process, which may exit,
502 * leaving nothing for the knote to be attached to. So when the process
503 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
504 * it will be deleted when read out. However, as part of the knote deletion,
505 * this routine is called, so a check is needed to avoid actually performing
506 * a detach, because the original process does not exist any more.
507 */
508 /* XXX - move to kern_proc.c? */
509 static void
filt_procdetach(struct knote * kn)510 filt_procdetach(struct knote *kn)
511 {
512
513 knlist_remove(kn->kn_knlist, kn, 0);
514 kn->kn_ptr.p_proc = NULL;
515 }
516
517 /* XXX - move to kern_proc.c? */
518 static int
filt_proc(struct knote * kn,long hint)519 filt_proc(struct knote *kn, long hint)
520 {
521 struct proc *p;
522 u_int event;
523
524 p = kn->kn_ptr.p_proc;
525 if (p == NULL) /* already activated, from attach filter */
526 return (0);
527
528 /* Mask off extra data. */
529 event = (u_int)hint & NOTE_PCTRLMASK;
530
531 /* If the user is interested in this event, record it. */
532 if (kn->kn_sfflags & event)
533 kn->kn_fflags |= event;
534
535 /* Process is gone, so flag the event as finished. */
536 if (event == NOTE_EXIT) {
537 kn->kn_flags |= EV_EOF | EV_ONESHOT;
538 kn->kn_ptr.p_proc = NULL;
539 if (kn->kn_fflags & NOTE_EXIT)
540 kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
541 if (kn->kn_fflags == 0)
542 kn->kn_flags |= EV_DROP;
543 return (1);
544 }
545
546 return (kn->kn_fflags != 0);
547 }
548
549 /*
550 * Called when the process forked. It mostly does the same as the
551 * knote(), activating all knotes registered to be activated when the
552 * process forked. Additionally, for each knote attached to the
553 * parent, check whether user wants to track the new process. If so
554 * attach a new knote to it, and immediately report an event with the
555 * child's pid.
556 */
557 void
knote_fork(struct knlist * list,int pid)558 knote_fork(struct knlist *list, int pid)
559 {
560 struct kqueue *kq;
561 struct knote *kn;
562 struct kevent kev;
563 int error;
564
565 MPASS(list != NULL);
566 KNL_ASSERT_LOCKED(list);
567 if (SLIST_EMPTY(&list->kl_list))
568 return;
569
570 memset(&kev, 0, sizeof(kev));
571 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
572 kq = kn->kn_kq;
573 KQ_LOCK(kq);
574 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
575 KQ_UNLOCK(kq);
576 continue;
577 }
578
579 /*
580 * The same as knote(), activate the event.
581 */
582 if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
583 if (kn->kn_fop->f_event(kn, NOTE_FORK))
584 KNOTE_ACTIVATE(kn, 1);
585 KQ_UNLOCK(kq);
586 continue;
587 }
588
589 /*
590 * The NOTE_TRACK case. In addition to the activation
591 * of the event, we need to register new events to
592 * track the child. Drop the locks in preparation for
593 * the call to kqueue_register().
594 */
595 kn_enter_flux(kn);
596 KQ_UNLOCK(kq);
597 list->kl_unlock(list->kl_lockarg);
598
599 /*
600 * Activate existing knote and register tracking knotes with
601 * new process.
602 *
603 * First register a knote to get just the child notice. This
604 * must be a separate note from a potential NOTE_EXIT
605 * notification since both NOTE_CHILD and NOTE_EXIT are defined
606 * to use the data field (in conflicting ways).
607 */
608 kev.ident = pid;
609 kev.filter = kn->kn_filter;
610 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
611 EV_FLAG2;
612 kev.fflags = kn->kn_sfflags;
613 kev.data = kn->kn_id; /* parent */
614 kev.udata = kn->kn_kevent.udata;/* preserve udata */
615 error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
616 if (error)
617 kn->kn_fflags |= NOTE_TRACKERR;
618
619 /*
620 * Then register another knote to track other potential events
621 * from the new process.
622 */
623 kev.ident = pid;
624 kev.filter = kn->kn_filter;
625 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
626 kev.fflags = kn->kn_sfflags;
627 kev.data = kn->kn_id; /* parent */
628 kev.udata = kn->kn_kevent.udata;/* preserve udata */
629 error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
630
631 /*
632 * Serialize updates to the kn_kevent fields with threads
633 * scanning the queue.
634 */
635 list->kl_lock(list->kl_lockarg);
636 if (error)
637 kn->kn_fflags |= NOTE_TRACKERR;
638 if (kn->kn_fop->f_event(kn, NOTE_FORK)) {
639 KQ_LOCK(kq);
640 KNOTE_ACTIVATE(kn, 1);
641 } else {
642 KQ_LOCK(kq);
643 }
644 kn_leave_flux(kn);
645 KQ_UNLOCK_FLUX(kq);
646 }
647 }
648
649 int
filt_jailattach(struct knote * kn)650 filt_jailattach(struct knote *kn)
651 {
652 struct prison *pr;
653
654 if (kn->kn_id == 0) {
655 /* Let jid=0 watch the current prison (including prison0). */
656 pr = curthread->td_ucred->cr_prison;
657 mtx_lock(&pr->pr_mtx);
658 } else {
659 sx_slock(&allprison_lock);
660 pr = prison_find_child(curthread->td_ucred->cr_prison,
661 kn->kn_id);
662 sx_sunlock(&allprison_lock);
663 if (pr == NULL)
664 return (ENOENT);
665 if (!prison_isalive(pr)) {
666 mtx_unlock(&pr->pr_mtx);
667 return (ENOENT);
668 }
669 }
670 kn->kn_ptr.p_prison = pr;
671 kn->kn_flags |= EV_CLEAR;
672 knlist_add(pr->pr_klist, kn, 1);
673 mtx_unlock(&pr->pr_mtx);
674 return (0);
675 }
676
677 void
filt_jaildetach(struct knote * kn)678 filt_jaildetach(struct knote *kn)
679 {
680 if (kn->kn_ptr.p_prison != NULL) {
681 knlist_remove(kn->kn_knlist, kn, 0);
682 kn->kn_ptr.p_prison = NULL;
683 } else
684 kn->kn_status |= KN_DETACHED;
685 }
686
687 int
filt_jail(struct knote * kn,long hint)688 filt_jail(struct knote *kn, long hint)
689 {
690 struct prison *pr;
691 u_int event;
692
693 pr = kn->kn_ptr.p_prison;
694 if (pr == NULL) /* already activated, from attach filter */
695 return (0);
696
697 /*
698 * Mask off extra data. In the NOTE_JAIL_CHILD case, that's
699 * everything except the NOTE_JAIL_CHILD bit itself, since a
700 * JID is any positive integer.
701 */
702 event = ((u_int)hint & NOTE_JAIL_CHILD) ? NOTE_JAIL_CHILD :
703 (u_int)hint & NOTE_JAIL_CTRLMASK;
704
705 /* If the user is interested in this event, record it. */
706 if (kn->kn_sfflags & event) {
707 kn->kn_fflags |= event;
708 /* Report the created jail id or attached process id. */
709 if (event == NOTE_JAIL_CHILD || event == NOTE_JAIL_ATTACH) {
710 if (kn->kn_data != 0)
711 kn->kn_fflags |= NOTE_JAIL_MULTI;
712 kn->kn_data = (kn->kn_fflags & NOTE_JAIL_MULTI) ? 0U :
713 (u_int)hint & ~event;
714 }
715 }
716
717 /* Prison is gone, so flag the event as finished. */
718 if (event == NOTE_JAIL_REMOVE) {
719 kn->kn_flags |= EV_EOF | EV_ONESHOT;
720 kn->kn_ptr.p_prison = NULL;
721 if (kn->kn_fflags == 0)
722 kn->kn_flags |= EV_DROP;
723 return (1);
724 }
725
726 return (kn->kn_fflags != 0);
727 }
728
729 /*
730 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
731 * interval timer support code.
732 */
733
734 #define NOTE_TIMER_PRECMASK \
735 (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)
736
737 static sbintime_t
timer2sbintime(int64_t data,unsigned int flags)738 timer2sbintime(int64_t data, unsigned int flags)
739 {
740 int64_t secs;
741
742 /*
743 * Macros for converting to the fractional second portion of an
744 * sbintime_t using 64bit multiplication to improve precision.
745 */
746 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
747 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
748 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
749 switch (flags & NOTE_TIMER_PRECMASK) {
750 case NOTE_SECONDS:
751 #ifdef __LP64__
752 if (data > (SBT_MAX / SBT_1S))
753 return (SBT_MAX);
754 #endif
755 return ((sbintime_t)data << 32);
756 case NOTE_MSECONDS: /* FALLTHROUGH */
757 case 0:
758 if (data >= 1000) {
759 secs = data / 1000;
760 #ifdef __LP64__
761 if (secs > (SBT_MAX / SBT_1S))
762 return (SBT_MAX);
763 #endif
764 return (secs << 32 | MS_TO_SBT(data % 1000));
765 }
766 return (MS_TO_SBT(data));
767 case NOTE_USECONDS:
768 if (data >= 1000000) {
769 secs = data / 1000000;
770 #ifdef __LP64__
771 if (secs > (SBT_MAX / SBT_1S))
772 return (SBT_MAX);
773 #endif
774 return (secs << 32 | US_TO_SBT(data % 1000000));
775 }
776 return (US_TO_SBT(data));
777 case NOTE_NSECONDS:
778 if (data >= 1000000000) {
779 secs = data / 1000000000;
780 #ifdef __LP64__
781 if (secs > (SBT_MAX / SBT_1S))
782 return (SBT_MAX);
783 #endif
784 return (secs << 32 | NS_TO_SBT(data % 1000000000));
785 }
786 return (NS_TO_SBT(data));
787 default:
788 break;
789 }
790 return (-1);
791 }
792
793 struct kq_timer_cb_data {
794 struct callout c;
795 struct proc *p;
796 struct knote *kn;
797 int cpuid;
798 int flags;
799 TAILQ_ENTRY(kq_timer_cb_data) link;
800 sbintime_t next; /* next timer event fires at */
801 sbintime_t to; /* precalculated timer period, 0 for abs */
802 };
803
804 #define KQ_TIMER_CB_ENQUEUED 0x01
805
806 static void
kqtimer_sched_callout(struct kq_timer_cb_data * kc)807 kqtimer_sched_callout(struct kq_timer_cb_data *kc)
808 {
809 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn,
810 kc->cpuid, C_ABSOLUTE);
811 }
812
813 void
kqtimer_proc_continue(struct proc * p)814 kqtimer_proc_continue(struct proc *p)
815 {
816 struct kq_timer_cb_data *kc, *kc1;
817 sbintime_t now;
818
819 PROC_LOCK_ASSERT(p, MA_OWNED);
820
821 now = sbinuptime();
822 TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) {
823 TAILQ_REMOVE(&p->p_kqtim_stop, kc, link);
824 kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
825 if (kc->next <= now)
826 filt_timerexpire_l(kc->kn, true);
827 else
828 kqtimer_sched_callout(kc);
829 }
830 }
831
832 static void
filt_timerexpire_l(struct knote * kn,bool proc_locked)833 filt_timerexpire_l(struct knote *kn, bool proc_locked)
834 {
835 struct kq_timer_cb_data *kc;
836 struct proc *p;
837 uint64_t delta;
838 sbintime_t now;
839
840 kc = kn->kn_ptr.p_v;
841
842 if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) {
843 kn->kn_data++;
844 KNOTE_ACTIVATE(kn, 0);
845 return;
846 }
847
848 now = sbinuptime();
849 if (now >= kc->next) {
850 delta = (now - kc->next) / kc->to;
851 if (delta == 0)
852 delta = 1;
853 kn->kn_data += delta;
854 kc->next += delta * kc->to;
855 if (now >= kc->next) /* overflow */
856 kc->next = now + kc->to;
857 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */
858 }
859
860 /*
861 * Initial check for stopped kc->p is racy. It is fine to
862 * miss the set of the stop flags, at worst we would schedule
863 * one more callout. On the other hand, it is not fine to not
864 * schedule when we we missed clearing of the flags, we
865 * recheck them under the lock and observe consistent state.
866 */
867 p = kc->p;
868 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
869 if (!proc_locked)
870 PROC_LOCK(p);
871 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
872 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) {
873 /*
874 * Insert into head so that
875 * kqtimer_proc_continue() does not
876 * iterate into us again.
877 */
878 kc->flags |= KQ_TIMER_CB_ENQUEUED;
879 TAILQ_INSERT_HEAD(&p->p_kqtim_stop, kc, link);
880 }
881 if (!proc_locked)
882 PROC_UNLOCK(p);
883 return;
884 }
885 if (!proc_locked)
886 PROC_UNLOCK(p);
887 }
888 kqtimer_sched_callout(kc);
889 }
890
891 static void
filt_timerexpire(void * knx)892 filt_timerexpire(void *knx)
893 {
894 filt_timerexpire_l(knx, false);
895 }
896
897 /*
898 * data contains amount of time to sleep
899 */
900 static int
filt_timervalidate(struct knote * kn,sbintime_t * to)901 filt_timervalidate(struct knote *kn, sbintime_t *to)
902 {
903 struct bintime bt;
904 sbintime_t sbt;
905
906 if (kn->kn_sdata < 0)
907 return (EINVAL);
908 if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
909 kn->kn_sdata = 1;
910 /*
911 * The only fflags values supported are the timer unit
912 * (precision) and the absolute time indicator.
913 */
914 if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0)
915 return (EINVAL);
916
917 *to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
918 if (*to < 0)
919 return (EINVAL);
920 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
921 getboottimebin(&bt);
922 sbt = bttosbt(bt);
923 *to = MAX(0, *to - sbt);
924 }
925 return (0);
926 }
927
928 static int
filt_timerattach(struct knote * kn)929 filt_timerattach(struct knote *kn)
930 {
931 struct kq_timer_cb_data *kc;
932 sbintime_t to;
933 int error;
934
935 to = -1;
936 error = filt_timervalidate(kn, &to);
937 if (error != 0)
938 return (error);
939 KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 ||
940 (kn->kn_sfflags & NOTE_ABSTIME) != 0,
941 ("%s: periodic timer has a calculated zero timeout", __func__));
942 KASSERT(to >= 0,
943 ("%s: timer has a calculated negative timeout", __func__));
944
945 if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) {
946 atomic_subtract_int(&kq_ncallouts, 1);
947 return (ENOMEM);
948 }
949
950 if ((kn->kn_sfflags & NOTE_ABSTIME) == 0)
951 kn->kn_flags |= EV_CLEAR; /* automatically set */
952 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */
953 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
954 kc->kn = kn;
955 kc->p = curproc;
956 kc->cpuid = PCPU_GET(cpuid);
957 kc->flags = 0;
958 callout_init(&kc->c, 1);
959 filt_timerstart(kn, to);
960
961 return (0);
962 }
963
964 static int
filt_timercopy(struct knote * kn,struct proc * p)965 filt_timercopy(struct knote *kn, struct proc *p)
966 {
967 struct kq_timer_cb_data *kc_src, *kc;
968
969 if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) {
970 atomic_subtract_int(&kq_ncallouts, 1);
971 return (ENOMEM);
972 }
973
974 kn->kn_status &= ~KN_DETACHED;
975 kc_src = kn->kn_ptr.p_v;
976 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
977 kc->kn = kn;
978 kc->p = p;
979 kc->flags = kc_src->flags & ~KQ_TIMER_CB_ENQUEUED;
980 kc->next = kc_src->next;
981 kc->to = kc_src->to;
982 kc->cpuid = PCPU_GET(cpuid);
983 callout_init(&kc->c, 1);
984 kqtimer_sched_callout(kc);
985 return (0);
986 }
987
988 static void
filt_timerstart(struct knote * kn,sbintime_t to)989 filt_timerstart(struct knote *kn, sbintime_t to)
990 {
991 struct kq_timer_cb_data *kc;
992
993 kc = kn->kn_ptr.p_v;
994 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
995 kc->next = to;
996 kc->to = 0;
997 } else {
998 kc->next = to + sbinuptime();
999 kc->to = to;
1000 }
1001 kqtimer_sched_callout(kc);
1002 }
1003
1004 static void
filt_timerdetach(struct knote * kn)1005 filt_timerdetach(struct knote *kn)
1006 {
1007 struct kq_timer_cb_data *kc;
1008 unsigned int old __unused;
1009 bool pending;
1010
1011 kc = kn->kn_ptr.p_v;
1012 do {
1013 callout_drain(&kc->c);
1014
1015 /*
1016 * kqtimer_proc_continue() might have rescheduled this callout.
1017 * Double-check, using the process mutex as an interlock.
1018 */
1019 PROC_LOCK(kc->p);
1020 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) {
1021 kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
1022 TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link);
1023 }
1024 pending = callout_pending(&kc->c);
1025 PROC_UNLOCK(kc->p);
1026 } while (pending);
1027 free(kc, M_KQUEUE);
1028 old = atomic_fetchadd_int(&kq_ncallouts, -1);
1029 KASSERT(old > 0, ("Number of callouts cannot become negative"));
1030 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */
1031 }
1032
1033 static void
filt_timertouch(struct knote * kn,struct kevent * kev,u_long type)1034 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type)
1035 {
1036 struct kq_timer_cb_data *kc;
1037 struct kqueue *kq;
1038 sbintime_t to;
1039 int error;
1040
1041 switch (type) {
1042 case EVENT_REGISTER:
1043 /* Handle re-added timers that update data/fflags */
1044 if (kev->flags & EV_ADD) {
1045 kc = kn->kn_ptr.p_v;
1046
1047 /* Drain any existing callout. */
1048 callout_drain(&kc->c);
1049
1050 /* Throw away any existing undelivered record
1051 * of the timer expiration. This is done under
1052 * the presumption that if a process is
1053 * re-adding this timer with new parameters,
1054 * it is no longer interested in what may have
1055 * happened under the old parameters. If it is
1056 * interested, it can wait for the expiration,
1057 * delete the old timer definition, and then
1058 * add the new one.
1059 *
1060 * This has to be done while the kq is locked:
1061 * - if enqueued, dequeue
1062 * - make it no longer active
1063 * - clear the count of expiration events
1064 */
1065 kq = kn->kn_kq;
1066 KQ_LOCK(kq);
1067 if (kn->kn_status & KN_QUEUED)
1068 knote_dequeue(kn);
1069
1070 kn->kn_status &= ~KN_ACTIVE;
1071 kn->kn_data = 0;
1072 KQ_UNLOCK(kq);
1073
1074 /* Reschedule timer based on new data/fflags */
1075 kn->kn_sfflags = kev->fflags;
1076 kn->kn_sdata = kev->data;
1077 error = filt_timervalidate(kn, &to);
1078 if (error != 0) {
1079 kn->kn_flags |= EV_ERROR;
1080 kn->kn_data = error;
1081 } else
1082 filt_timerstart(kn, to);
1083 }
1084 break;
1085
1086 case EVENT_PROCESS:
1087 *kev = kn->kn_kevent;
1088 if (kn->kn_flags & EV_CLEAR) {
1089 kn->kn_data = 0;
1090 kn->kn_fflags = 0;
1091 }
1092 break;
1093
1094 default:
1095 panic("filt_timertouch() - invalid type (%ld)", type);
1096 break;
1097 }
1098 }
1099
1100 static int
filt_timer(struct knote * kn,long hint)1101 filt_timer(struct knote *kn, long hint)
1102 {
1103
1104 return (kn->kn_data != 0);
1105 }
1106
1107 static int
filt_userattach(struct knote * kn)1108 filt_userattach(struct knote *kn)
1109 {
1110
1111 /*
1112 * EVFILT_USER knotes are not attached to anything in the kernel.
1113 */
1114 kn->kn_hook = NULL;
1115 if (kn->kn_fflags & NOTE_TRIGGER)
1116 kn->kn_hookid = 1;
1117 else
1118 kn->kn_hookid = 0;
1119 return (0);
1120 }
1121
1122 static void
filt_userdetach(__unused struct knote * kn)1123 filt_userdetach(__unused struct knote *kn)
1124 {
1125
1126 /*
1127 * EVFILT_USER knotes are not attached to anything in the kernel.
1128 */
1129 }
1130
1131 static int
filt_user(struct knote * kn,__unused long hint)1132 filt_user(struct knote *kn, __unused long hint)
1133 {
1134
1135 return (kn->kn_hookid);
1136 }
1137
1138 static void
filt_usertouch(struct knote * kn,struct kevent * kev,u_long type)1139 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type)
1140 {
1141 u_int ffctrl;
1142
1143 switch (type) {
1144 case EVENT_REGISTER:
1145 if (kev->fflags & NOTE_TRIGGER)
1146 kn->kn_hookid = 1;
1147
1148 ffctrl = kev->fflags & NOTE_FFCTRLMASK;
1149 kev->fflags &= NOTE_FFLAGSMASK;
1150 switch (ffctrl) {
1151 case NOTE_FFNOP:
1152 break;
1153
1154 case NOTE_FFAND:
1155 kn->kn_sfflags &= kev->fflags;
1156 break;
1157
1158 case NOTE_FFOR:
1159 kn->kn_sfflags |= kev->fflags;
1160 break;
1161
1162 case NOTE_FFCOPY:
1163 kn->kn_sfflags = kev->fflags;
1164 break;
1165
1166 default:
1167 /* XXX Return error? */
1168 break;
1169 }
1170 kn->kn_sdata = kev->data;
1171 if (kev->flags & EV_CLEAR) {
1172 kn->kn_hookid = 0;
1173 kn->kn_data = 0;
1174 kn->kn_fflags = 0;
1175 }
1176 break;
1177
1178 case EVENT_PROCESS:
1179 *kev = kn->kn_kevent;
1180 kev->fflags = kn->kn_sfflags;
1181 kev->data = kn->kn_sdata;
1182 if (kn->kn_flags & EV_CLEAR) {
1183 kn->kn_hookid = 0;
1184 kn->kn_data = 0;
1185 kn->kn_fflags = 0;
1186 }
1187 break;
1188
1189 default:
1190 panic("filt_usertouch() - invalid type (%ld)", type);
1191 break;
1192 }
1193 }
1194
1195 int
sys_kqueue(struct thread * td,struct kqueue_args * uap)1196 sys_kqueue(struct thread *td, struct kqueue_args *uap)
1197 {
1198
1199 return (kern_kqueue(td, 0, false, NULL));
1200 }
1201
1202 int
sys_kqueuex(struct thread * td,struct kqueuex_args * uap)1203 sys_kqueuex(struct thread *td, struct kqueuex_args *uap)
1204 {
1205 int flags;
1206
1207 if ((uap->flags & ~(KQUEUE_CLOEXEC | KQUEUE_CPONFORK)) != 0)
1208 return (EINVAL);
1209 flags = 0;
1210 if ((uap->flags & KQUEUE_CLOEXEC) != 0)
1211 flags |= O_CLOEXEC;
1212 return (kern_kqueue(td, flags, (uap->flags & KQUEUE_CPONFORK) != 0,
1213 NULL));
1214 }
1215
1216 static void
kqueue_init(struct kqueue * kq,bool cponfork)1217 kqueue_init(struct kqueue *kq, bool cponfork)
1218 {
1219
1220 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
1221 TAILQ_INIT(&kq->kq_head);
1222 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
1223 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
1224 if (cponfork)
1225 kq->kq_state |= KQ_CPONFORK;
1226 }
1227
1228 static int
kern_kqueue_alloc(struct thread * td,struct filedesc * fdp,int * fdip,struct file ** fpp,int flags,struct filecaps * fcaps,bool cponfork,struct kqueue ** kqp)1229 kern_kqueue_alloc(struct thread *td, struct filedesc *fdp, int *fdip,
1230 struct file **fpp, int flags, struct filecaps *fcaps, bool cponfork,
1231 struct kqueue **kqp)
1232 {
1233 struct ucred *cred;
1234 struct kqueue *kq;
1235 int error;
1236
1237 cred = td->td_ucred;
1238 if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
1239 return (ENOMEM);
1240
1241 error = fdip != NULL ? falloc_caps(td, fpp, fdip, flags, fcaps) :
1242 _falloc_noinstall(td, fpp, 1);
1243 if (error != 0) {
1244 chgkqcnt(cred->cr_ruidinfo, -1, 0);
1245 return (error);
1246 }
1247
1248 /* An extra reference on `fp' has been held for us by falloc(). */
1249 kq = malloc(sizeof(*kq), M_KQUEUE, M_WAITOK | M_ZERO);
1250 kqueue_init(kq, cponfork);
1251 kq->kq_fdp = fdp;
1252 kq->kq_cred = crhold(cred);
1253
1254 if (fdip != NULL)
1255 FILEDESC_XLOCK(fdp);
1256 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
1257 if (fdip != NULL)
1258 FILEDESC_XUNLOCK(fdp);
1259
1260 finit(*fpp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
1261 *kqp = kq;
1262 return (0);
1263 }
1264
1265 int
kern_kqueue(struct thread * td,int flags,bool cponfork,struct filecaps * fcaps)1266 kern_kqueue(struct thread *td, int flags, bool cponfork, struct filecaps *fcaps)
1267 {
1268 struct kqueue *kq;
1269 struct file *fp;
1270 int fd, error;
1271
1272 error = kern_kqueue_alloc(td, td->td_proc->p_fd, &fd, &fp, flags,
1273 fcaps, cponfork, &kq);
1274 if (error != 0)
1275 return (error);
1276
1277 fdrop(fp, td);
1278
1279 td->td_retval[0] = fd;
1280 return (0);
1281 }
1282
1283 struct g_kevent_args {
1284 int fd;
1285 const void *changelist;
1286 int nchanges;
1287 void *eventlist;
1288 int nevents;
1289 const struct timespec *timeout;
1290 };
1291
1292 int
sys_kevent(struct thread * td,struct kevent_args * uap)1293 sys_kevent(struct thread *td, struct kevent_args *uap)
1294 {
1295 struct kevent_copyops k_ops = {
1296 .arg = uap,
1297 .k_copyout = kevent_copyout,
1298 .k_copyin = kevent_copyin,
1299 .kevent_size = sizeof(struct kevent),
1300 };
1301 struct g_kevent_args gk_args = {
1302 .fd = uap->fd,
1303 .changelist = uap->changelist,
1304 .nchanges = uap->nchanges,
1305 .eventlist = uap->eventlist,
1306 .nevents = uap->nevents,
1307 .timeout = uap->timeout,
1308 };
1309
1310 return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent"));
1311 }
1312
1313 static int
kern_kevent_generic(struct thread * td,struct g_kevent_args * uap,struct kevent_copyops * k_ops,const char * struct_name)1314 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap,
1315 struct kevent_copyops *k_ops, const char *struct_name)
1316 {
1317 struct timespec ts, *tsp;
1318 #ifdef KTRACE
1319 struct kevent *eventlist = uap->eventlist;
1320 #endif
1321 int error;
1322
1323 if (uap->timeout != NULL) {
1324 error = copyin(uap->timeout, &ts, sizeof(ts));
1325 if (error)
1326 return (error);
1327 tsp = &ts;
1328 } else
1329 tsp = NULL;
1330
1331 #ifdef KTRACE
1332 if (KTRPOINT(td, KTR_STRUCT_ARRAY))
1333 ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist,
1334 uap->nchanges, k_ops->kevent_size);
1335 #endif
1336
1337 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
1338 k_ops, tsp);
1339
1340 #ifdef KTRACE
1341 if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
1342 ktrstructarray(struct_name, UIO_USERSPACE, eventlist,
1343 td->td_retval[0], k_ops->kevent_size);
1344 #endif
1345
1346 return (error);
1347 }
1348
1349 /*
1350 * Copy 'count' items into the destination list pointed to by uap->eventlist.
1351 */
1352 static int
kevent_copyout(void * arg,struct kevent * kevp,int count)1353 kevent_copyout(void *arg, struct kevent *kevp, int count)
1354 {
1355 struct kevent_args *uap;
1356 int error;
1357
1358 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1359 uap = (struct kevent_args *)arg;
1360
1361 error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
1362 if (error == 0)
1363 uap->eventlist += count;
1364 return (error);
1365 }
1366
1367 /*
1368 * Copy 'count' items from the list pointed to by uap->changelist.
1369 */
1370 static int
kevent_copyin(void * arg,struct kevent * kevp,int count)1371 kevent_copyin(void *arg, struct kevent *kevp, int count)
1372 {
1373 struct kevent_args *uap;
1374 int error;
1375
1376 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1377 uap = (struct kevent_args *)arg;
1378
1379 error = copyin(uap->changelist, kevp, count * sizeof *kevp);
1380 if (error == 0)
1381 uap->changelist += count;
1382 return (error);
1383 }
1384
1385 #ifdef COMPAT_FREEBSD11
1386 static int
kevent11_copyout(void * arg,struct kevent * kevp,int count)1387 kevent11_copyout(void *arg, struct kevent *kevp, int count)
1388 {
1389 struct freebsd11_kevent_args *uap;
1390 struct freebsd11_kevent kev11;
1391 int error, i;
1392
1393 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1394 uap = (struct freebsd11_kevent_args *)arg;
1395
1396 for (i = 0; i < count; i++) {
1397 kev11.ident = kevp->ident;
1398 kev11.filter = kevp->filter;
1399 kev11.flags = kevp->flags;
1400 kev11.fflags = kevp->fflags;
1401 kev11.data = kevp->data;
1402 kev11.udata = kevp->udata;
1403 error = copyout(&kev11, uap->eventlist, sizeof(kev11));
1404 if (error != 0)
1405 break;
1406 uap->eventlist++;
1407 kevp++;
1408 }
1409 return (error);
1410 }
1411
1412 /*
1413 * Copy 'count' items from the list pointed to by uap->changelist.
1414 */
1415 static int
kevent11_copyin(void * arg,struct kevent * kevp,int count)1416 kevent11_copyin(void *arg, struct kevent *kevp, int count)
1417 {
1418 struct freebsd11_kevent_args *uap;
1419 struct freebsd11_kevent kev11;
1420 int error, i;
1421
1422 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
1423 uap = (struct freebsd11_kevent_args *)arg;
1424
1425 for (i = 0; i < count; i++) {
1426 error = copyin(uap->changelist, &kev11, sizeof(kev11));
1427 if (error != 0)
1428 break;
1429 kevp->ident = kev11.ident;
1430 kevp->filter = kev11.filter;
1431 kevp->flags = kev11.flags;
1432 kevp->fflags = kev11.fflags;
1433 kevp->data = (uintptr_t)kev11.data;
1434 kevp->udata = kev11.udata;
1435 bzero(&kevp->ext, sizeof(kevp->ext));
1436 uap->changelist++;
1437 kevp++;
1438 }
1439 return (error);
1440 }
1441
1442 int
freebsd11_kevent(struct thread * td,struct freebsd11_kevent_args * uap)1443 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap)
1444 {
1445 struct kevent_copyops k_ops = {
1446 .arg = uap,
1447 .k_copyout = kevent11_copyout,
1448 .k_copyin = kevent11_copyin,
1449 .kevent_size = sizeof(struct freebsd11_kevent),
1450 };
1451 struct g_kevent_args gk_args = {
1452 .fd = uap->fd,
1453 .changelist = uap->changelist,
1454 .nchanges = uap->nchanges,
1455 .eventlist = uap->eventlist,
1456 .nevents = uap->nevents,
1457 .timeout = uap->timeout,
1458 };
1459
1460 return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent"));
1461 }
1462 #endif
1463
1464 int
kern_kevent(struct thread * td,int fd,int nchanges,int nevents,struct kevent_copyops * k_ops,const struct timespec * timeout)1465 kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
1466 struct kevent_copyops *k_ops, const struct timespec *timeout)
1467 {
1468 cap_rights_t rights;
1469 struct file *fp;
1470 int error;
1471
1472 cap_rights_init_zero(&rights);
1473 if (nchanges > 0)
1474 cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE);
1475 if (nevents > 0)
1476 cap_rights_set_one(&rights, CAP_KQUEUE_EVENT);
1477 error = fget(td, fd, &rights, &fp);
1478 if (error != 0)
1479 return (error);
1480
1481 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
1482 fdrop(fp, td);
1483
1484 return (error);
1485 }
1486
1487 static int
kqueue_kevent(struct kqueue * kq,struct thread * td,int nchanges,int nevents,struct kevent_copyops * k_ops,const struct timespec * timeout)1488 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
1489 struct kevent_copyops *k_ops, const struct timespec *timeout)
1490 {
1491 struct kevent keva[KQ_NEVENTS];
1492 struct kevent *kevp, *changes;
1493 int i, n, nerrors, error;
1494
1495 if (nchanges < 0)
1496 return (EINVAL);
1497
1498 nerrors = 0;
1499 while (nchanges > 0) {
1500 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
1501 error = k_ops->k_copyin(k_ops->arg, keva, n);
1502 if (error)
1503 return (error);
1504 changes = keva;
1505 for (i = 0; i < n; i++) {
1506 kevp = &changes[i];
1507 if (!kevp->filter)
1508 continue;
1509 kevp->flags &= ~EV_SYSFLAGS;
1510 error = kqueue_register(kq, kevp, td, M_WAITOK);
1511 if (error || (kevp->flags & EV_RECEIPT)) {
1512 if (nevents == 0)
1513 return (error);
1514 kevp->flags = EV_ERROR;
1515 kevp->data = error;
1516 (void)k_ops->k_copyout(k_ops->arg, kevp, 1);
1517 nevents--;
1518 nerrors++;
1519 }
1520 }
1521 nchanges -= n;
1522 }
1523 if (nerrors) {
1524 td->td_retval[0] = nerrors;
1525 return (0);
1526 }
1527
1528 return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
1529 }
1530
1531 int
kern_kevent_fp(struct thread * td,struct file * fp,int nchanges,int nevents,struct kevent_copyops * k_ops,const struct timespec * timeout)1532 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
1533 struct kevent_copyops *k_ops, const struct timespec *timeout)
1534 {
1535 struct kqueue *kq;
1536 int error;
1537
1538 error = kqueue_acquire(fp, &kq);
1539 if (error != 0)
1540 return (error);
1541 error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
1542 kqueue_release(kq, 0);
1543 return (error);
1544 }
1545
1546 /*
1547 * Performs a kevent() call on a temporarily created kqueue. This can be
1548 * used to perform one-shot polling, similar to poll() and select().
1549 */
1550 int
kern_kevent_anonymous(struct thread * td,int nevents,struct kevent_copyops * k_ops)1551 kern_kevent_anonymous(struct thread *td, int nevents,
1552 struct kevent_copyops *k_ops)
1553 {
1554 struct kqueue kq = {};
1555 int error;
1556
1557 kqueue_init(&kq, false);
1558 kq.kq_refcnt = 1;
1559 error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
1560 kqueue_drain(&kq, td);
1561 kqueue_destroy(&kq);
1562 return (error);
1563 }
1564
1565 int
kqueue_add_filteropts(int filt,const struct filterops * filtops)1566 kqueue_add_filteropts(int filt, const struct filterops *filtops)
1567 {
1568 int error;
1569
1570 error = 0;
1571 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
1572 printf(
1573 "trying to add a filterop that is out of range: %d is beyond %d\n",
1574 ~filt, EVFILT_SYSCOUNT);
1575 return EINVAL;
1576 }
1577 mtx_lock(&filterops_lock);
1578 if (sysfilt_ops[~filt].for_fop != &null_filtops &&
1579 sysfilt_ops[~filt].for_fop != NULL)
1580 error = EEXIST;
1581 else {
1582 sysfilt_ops[~filt].for_fop = filtops;
1583 sysfilt_ops[~filt].for_refcnt = 0;
1584 }
1585 mtx_unlock(&filterops_lock);
1586
1587 return (error);
1588 }
1589
1590 int
kqueue_del_filteropts(int filt)1591 kqueue_del_filteropts(int filt)
1592 {
1593 int error;
1594
1595 error = 0;
1596 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1597 return EINVAL;
1598
1599 mtx_lock(&filterops_lock);
1600 if (sysfilt_ops[~filt].for_fop == &null_filtops ||
1601 sysfilt_ops[~filt].for_fop == NULL)
1602 error = EINVAL;
1603 else if (sysfilt_ops[~filt].for_refcnt != 0)
1604 error = EBUSY;
1605 else {
1606 sysfilt_ops[~filt].for_fop = &null_filtops;
1607 sysfilt_ops[~filt].for_refcnt = 0;
1608 }
1609 mtx_unlock(&filterops_lock);
1610
1611 return error;
1612 }
1613
1614 static const struct filterops *
kqueue_fo_find(int filt)1615 kqueue_fo_find(int filt)
1616 {
1617
1618 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1619 return NULL;
1620
1621 if (sysfilt_ops[~filt].for_nolock)
1622 return sysfilt_ops[~filt].for_fop;
1623
1624 mtx_lock(&filterops_lock);
1625 sysfilt_ops[~filt].for_refcnt++;
1626 if (sysfilt_ops[~filt].for_fop == NULL)
1627 sysfilt_ops[~filt].for_fop = &null_filtops;
1628 mtx_unlock(&filterops_lock);
1629
1630 return sysfilt_ops[~filt].for_fop;
1631 }
1632
1633 static void
kqueue_fo_release(int filt)1634 kqueue_fo_release(int filt)
1635 {
1636
1637 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
1638 return;
1639
1640 if (sysfilt_ops[~filt].for_nolock)
1641 return;
1642
1643 mtx_lock(&filterops_lock);
1644 KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
1645 ("filter object %d refcount not valid on release", filt));
1646 sysfilt_ops[~filt].for_refcnt--;
1647 mtx_unlock(&filterops_lock);
1648 }
1649
1650 /*
1651 * A ref to kq (obtained via kqueue_acquire) must be held.
1652 */
1653 static int
kqueue_register(struct kqueue * kq,struct kevent * kev,struct thread * td,int mflag)1654 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td,
1655 int mflag)
1656 {
1657 const struct filterops *fops;
1658 struct file *fp;
1659 struct knote *kn, *tkn;
1660 struct knlist *knl;
1661 int error, filt, event;
1662 int haskqglobal, filedesc_unlock;
1663
1664 if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
1665 return (EINVAL);
1666
1667 fp = NULL;
1668 kn = NULL;
1669 knl = NULL;
1670 error = 0;
1671 haskqglobal = 0;
1672 filedesc_unlock = 0;
1673
1674 filt = kev->filter;
1675 fops = kqueue_fo_find(filt);
1676 if (fops == NULL)
1677 return EINVAL;
1678
1679 if (kev->flags & EV_ADD) {
1680 /* Reject an invalid flag pair early */
1681 if (kev->flags & EV_KEEPUDATA) {
1682 tkn = NULL;
1683 error = EINVAL;
1684 goto done;
1685 }
1686
1687 /*
1688 * Prevent waiting with locks. Non-sleepable
1689 * allocation failures are handled in the loop, only
1690 * if the spare knote appears to be actually required.
1691 */
1692 tkn = knote_alloc(mflag);
1693 } else {
1694 tkn = NULL;
1695 }
1696
1697 findkn:
1698 if (fops->f_isfd) {
1699 KASSERT(td != NULL, ("td is NULL"));
1700 if (kev->ident > INT_MAX)
1701 error = EBADF;
1702 else
1703 error = fget(td, kev->ident, &cap_event_rights, &fp);
1704 if (error)
1705 goto done;
1706
1707 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
1708 kev->ident, M_NOWAIT) != 0) {
1709 /* try again */
1710 fdrop(fp, td);
1711 fp = NULL;
1712 error = kqueue_expand(kq, fops, kev->ident, mflag);
1713 if (error)
1714 goto done;
1715 goto findkn;
1716 }
1717
1718 if (fp->f_type == DTYPE_KQUEUE) {
1719 /*
1720 * If we add some intelligence about what we are doing,
1721 * we should be able to support events on ourselves.
1722 * We need to know when we are doing this to prevent
1723 * getting both the knlist lock and the kq lock since
1724 * they are the same thing.
1725 */
1726 if (fp->f_data == kq) {
1727 error = EINVAL;
1728 goto done;
1729 }
1730
1731 /*
1732 * Pre-lock the filedesc before the global
1733 * lock mutex, see the comment in
1734 * kqueue_close().
1735 */
1736 FILEDESC_XLOCK(td->td_proc->p_fd);
1737 filedesc_unlock = 1;
1738 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
1739 }
1740
1741 KQ_LOCK(kq);
1742 if (kev->ident < kq->kq_knlistsize) {
1743 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) {
1744 MPASS(kn->kn_kq == kq);
1745 if (kev->filter == kn->kn_filter)
1746 break;
1747 }
1748 }
1749 } else {
1750 if ((kev->flags & EV_ADD) == EV_ADD) {
1751 error = kqueue_expand(kq, fops, kev->ident, mflag);
1752 if (error != 0)
1753 goto done;
1754 }
1755
1756 KQ_LOCK(kq);
1757
1758 /*
1759 * If possible, find an existing knote to use for this kevent.
1760 */
1761 if (kev->filter == EVFILT_PROC &&
1762 (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
1763 /* This is an internal creation of a process tracking
1764 * note. Don't attempt to coalesce this with an
1765 * existing note.
1766 */
1767 ;
1768 } else if (kq->kq_knhashmask != 0) {
1769 struct klist *list;
1770
1771 list = &kq->kq_knhash[
1772 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
1773 SLIST_FOREACH(kn, list, kn_link) {
1774 MPASS(kn->kn_kq == kq);
1775 if (kev->ident == kn->kn_id &&
1776 kev->filter == kn->kn_filter)
1777 break;
1778 }
1779 }
1780 }
1781
1782 /* knote is in the process of changing, wait for it to stabilize. */
1783 if (kn != NULL && kn_in_flux(kn)) {
1784 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1785 if (filedesc_unlock) {
1786 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1787 filedesc_unlock = 0;
1788 }
1789 kq->kq_state |= KQ_FLUXWAIT;
1790 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
1791 if (fp != NULL) {
1792 fdrop(fp, td);
1793 fp = NULL;
1794 }
1795 goto findkn;
1796 }
1797
1798 /*
1799 * kn now contains the matching knote, or NULL if no match
1800 */
1801 if (kn == NULL) {
1802 if (kev->flags & EV_ADD) {
1803 kn = tkn;
1804 tkn = NULL;
1805 if (kn == NULL) {
1806 KQ_UNLOCK(kq);
1807 error = ENOMEM;
1808 goto done;
1809 }
1810
1811 /*
1812 * Now that the kqueue is locked, make sure the fd
1813 * didn't change out from under us.
1814 */
1815 if (fops->f_isfd &&
1816 fget_noref_unlocked(td->td_proc->p_fd,
1817 kev->ident) != fp) {
1818 KQ_UNLOCK(kq);
1819 tkn = kn;
1820 error = EBADF;
1821 goto done;
1822 }
1823 kn->kn_fp = fp;
1824 kn->kn_kq = kq;
1825 kn->kn_fop = fops;
1826
1827 kn->kn_sfflags = kev->fflags;
1828 kn->kn_sdata = kev->data;
1829 kev->fflags = 0;
1830 kev->data = 0;
1831 kn->kn_kevent = *kev;
1832 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
1833 EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
1834 kn->kn_status = KN_DETACHED;
1835 if ((kev->flags & EV_DISABLE) != 0)
1836 kn->kn_status |= KN_DISABLED;
1837 kn_enter_flux(kn);
1838
1839 error = knote_attach(kn, kq);
1840 KQ_UNLOCK(kq);
1841 if (error != 0) {
1842 tkn = kn;
1843 goto done;
1844 }
1845
1846 /*
1847 * We transfer ownership of fops/fp to the knote
1848 * structure and avoid releasing them at the end of
1849 * this routine, now that all of the remaining exit
1850 * paths will knote_drop() to release the reference
1851 * counts we held on them above.
1852 */
1853 fops = NULL;
1854 fp = NULL;
1855
1856 if ((error = kn->kn_fop->f_attach(kn)) != 0) {
1857 knote_drop_detached(kn, td);
1858 goto done;
1859 }
1860 knl = kn_list_lock(kn);
1861 goto done_ev_add;
1862 } else {
1863 /* No matching knote and the EV_ADD flag is not set. */
1864 KQ_UNLOCK(kq);
1865 error = ENOENT;
1866 goto done;
1867 }
1868 }
1869
1870 if (kev->flags & EV_DELETE) {
1871 kn_enter_flux(kn);
1872 KQ_UNLOCK(kq);
1873 knote_drop(kn, td);
1874 goto done;
1875 }
1876
1877 if (kev->flags & EV_FORCEONESHOT) {
1878 kn->kn_flags |= EV_ONESHOT;
1879 KNOTE_ACTIVATE(kn, 1);
1880 }
1881
1882 if ((kev->flags & EV_ENABLE) != 0)
1883 kn->kn_status &= ~KN_DISABLED;
1884 else if ((kev->flags & EV_DISABLE) != 0)
1885 kn->kn_status |= KN_DISABLED;
1886
1887 /*
1888 * The user may change some filter values after the initial EV_ADD,
1889 * but doing so will not reset any filter which has already been
1890 * triggered.
1891 */
1892 kn->kn_status |= KN_SCAN;
1893 kn_enter_flux(kn);
1894 KQ_UNLOCK(kq);
1895 knl = kn_list_lock(kn);
1896 if ((kev->flags & EV_KEEPUDATA) == 0)
1897 kn->kn_kevent.udata = kev->udata;
1898 if (!fops->f_isfd && fops->f_touch != NULL) {
1899 fops->f_touch(kn, kev, EVENT_REGISTER);
1900 } else {
1901 kn->kn_sfflags = kev->fflags;
1902 kn->kn_sdata = kev->data;
1903 }
1904
1905 done_ev_add:
1906 /*
1907 * We can get here with kn->kn_knlist == NULL. This can happen when
1908 * the initial attach event decides that the event is "completed"
1909 * already, e.g., filt_procattach() is called on a zombie process. It
1910 * will call filt_proc() which will remove it from the list, and NULL
1911 * kn_knlist.
1912 *
1913 * KN_DISABLED will be stable while the knote is in flux, so the
1914 * unlocked read will not race with an update.
1915 */
1916 if ((kn->kn_status & KN_DISABLED) == 0)
1917 event = kn->kn_fop->f_event(kn, 0);
1918 else
1919 event = 0;
1920
1921 KQ_LOCK(kq);
1922 if (event)
1923 kn->kn_status |= KN_ACTIVE;
1924 if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
1925 KN_ACTIVE)
1926 knote_enqueue(kn);
1927 kn->kn_status &= ~KN_SCAN;
1928 kn_leave_flux(kn);
1929 kn_list_unlock(knl);
1930 KQ_UNLOCK_FLUX(kq);
1931
1932 done:
1933 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
1934 if (filedesc_unlock)
1935 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1936 if (fp != NULL)
1937 fdrop(fp, td);
1938 knote_free(tkn);
1939 if (fops != NULL)
1940 kqueue_fo_release(filt);
1941 return (error);
1942 }
1943
1944 static int
kqueue_acquire_ref(struct kqueue * kq)1945 kqueue_acquire_ref(struct kqueue *kq)
1946 {
1947 KQ_LOCK(kq);
1948 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
1949 KQ_UNLOCK(kq);
1950 return (EBADF);
1951 }
1952 kq->kq_refcnt++;
1953 KQ_UNLOCK(kq);
1954 return (0);
1955 }
1956
1957 static int
kqueue_acquire(struct file * fp,struct kqueue ** kqp)1958 kqueue_acquire(struct file *fp, struct kqueue **kqp)
1959 {
1960 struct kqueue *kq;
1961 int error;
1962
1963 kq = fp->f_data;
1964 if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
1965 return (EINVAL);
1966 error = kqueue_acquire_ref(kq);
1967 if (error == 0)
1968 *kqp = kq;
1969 return (error);
1970 }
1971
1972 static void
kqueue_release(struct kqueue * kq,int locked)1973 kqueue_release(struct kqueue *kq, int locked)
1974 {
1975 if (locked)
1976 KQ_OWNED(kq);
1977 else
1978 KQ_LOCK(kq);
1979 kq->kq_refcnt--;
1980 if (kq->kq_refcnt == 1)
1981 wakeup(&kq->kq_refcnt);
1982 if (!locked)
1983 KQ_UNLOCK(kq);
1984 }
1985
1986 static void
ast_kqueue(struct thread * td,int tda __unused)1987 ast_kqueue(struct thread *td, int tda __unused)
1988 {
1989 taskqueue_quiesce(taskqueue_kqueue_ctx);
1990 }
1991
1992 static void
kqueue_schedtask(struct kqueue * kq)1993 kqueue_schedtask(struct kqueue *kq)
1994 {
1995 KQ_OWNED(kq);
1996 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
1997 ("scheduling kqueue task while draining"));
1998
1999 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
2000 taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
2001 kq->kq_state |= KQ_TASKSCHED;
2002 ast_sched(curthread, TDA_KQUEUE);
2003 }
2004 }
2005
2006 /*
2007 * Expand the kq to make sure we have storage for fops/ident pair.
2008 *
2009 * Return 0 on success (or no work necessary), return errno on failure.
2010 */
2011 static int
kqueue_expand(struct kqueue * kq,const struct filterops * fops,uintptr_t ident,int mflag)2012 kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident,
2013 int mflag)
2014 {
2015 struct klist *list, *tmp_knhash, *to_free;
2016 u_long tmp_knhashmask;
2017 int error, fd, size;
2018
2019 KQ_NOTOWNED(kq);
2020
2021 error = 0;
2022 to_free = NULL;
2023 if (fops->f_isfd) {
2024 fd = ident;
2025 size = atomic_load_int(&kq->kq_knlistsize);
2026 if (size <= fd) {
2027 do {
2028 size += KQEXTENT;
2029 } while (size <= fd);
2030 list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
2031 if (list == NULL)
2032 return ENOMEM;
2033 KQ_LOCK(kq);
2034 if ((kq->kq_state & KQ_CLOSING) != 0) {
2035 to_free = list;
2036 error = EBADF;
2037 } else if (kq->kq_knlistsize >= size) {
2038 to_free = list;
2039 } else {
2040 if (kq->kq_knlist != NULL) {
2041 bcopy(kq->kq_knlist, list,
2042 kq->kq_knlistsize * sizeof(*list));
2043 to_free = kq->kq_knlist;
2044 kq->kq_knlist = NULL;
2045 }
2046 bzero((caddr_t)list +
2047 kq->kq_knlistsize * sizeof(*list),
2048 (size - kq->kq_knlistsize) * sizeof(*list));
2049 kq->kq_knlistsize = size;
2050 kq->kq_knlist = list;
2051 }
2052 MPASS(error != 0 || kq->kq_knlistsize > fd);
2053 KQ_UNLOCK(kq);
2054 }
2055 } else {
2056 if (kq->kq_knhashmask == 0) {
2057 tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE,
2058 &tmp_knhashmask, (mflag & M_WAITOK) != 0 ?
2059 HASH_WAITOK : HASH_NOWAIT);
2060 if (tmp_knhash == NULL)
2061 return (ENOMEM);
2062 KQ_LOCK(kq);
2063 if ((kq->kq_state & KQ_CLOSING) != 0) {
2064 to_free = tmp_knhash;
2065 error = EBADF;
2066 } else if (kq->kq_knhashmask == 0) {
2067 kq->kq_knhash = tmp_knhash;
2068 kq->kq_knhashmask = tmp_knhashmask;
2069 } else {
2070 to_free = tmp_knhash;
2071 }
2072 KQ_UNLOCK(kq);
2073 }
2074 }
2075 free(to_free, M_KQUEUE);
2076
2077 KQ_NOTOWNED(kq);
2078 return (error);
2079 }
2080
2081 static void
kqueue_task(void * arg,int pending)2082 kqueue_task(void *arg, int pending)
2083 {
2084 struct kqueue *kq;
2085 int haskqglobal;
2086
2087 haskqglobal = 0;
2088 kq = arg;
2089
2090 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
2091 KQ_LOCK(kq);
2092
2093 KNOTE_LOCKED(&kq->kq_sel.si_note, 0);
2094
2095 kq->kq_state &= ~KQ_TASKSCHED;
2096 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
2097 wakeup(&kq->kq_state);
2098 }
2099 KQ_UNLOCK(kq);
2100 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2101 }
2102
2103 /*
2104 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
2105 * We treat KN_MARKER knotes as if they are in flux.
2106 */
2107 static int
kqueue_scan(struct kqueue * kq,int maxevents,struct kevent_copyops * k_ops,const struct timespec * tsp,struct kevent * keva,struct thread * td)2108 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
2109 const struct timespec *tsp, struct kevent *keva, struct thread *td)
2110 {
2111 struct kevent *kevp;
2112 struct knote *kn, *marker;
2113 struct knlist *knl;
2114 sbintime_t asbt, rsbt;
2115 int count, error, haskqglobal, influx, nkev, touch;
2116
2117 count = maxevents;
2118 nkev = 0;
2119 error = 0;
2120 haskqglobal = 0;
2121
2122 if (maxevents == 0)
2123 goto done_nl;
2124 if (maxevents < 0) {
2125 error = EINVAL;
2126 goto done_nl;
2127 }
2128
2129 rsbt = 0;
2130 if (tsp != NULL) {
2131 if (!timespecvalid_interval(tsp)) {
2132 error = EINVAL;
2133 goto done_nl;
2134 }
2135 if (timespecisset(tsp)) {
2136 if (tsp->tv_sec <= INT32_MAX) {
2137 rsbt = tstosbt(*tsp);
2138 if (TIMESEL(&asbt, rsbt))
2139 asbt += tc_tick_sbt;
2140 if (asbt <= SBT_MAX - rsbt)
2141 asbt += rsbt;
2142 else
2143 asbt = 0;
2144 rsbt >>= tc_precexp;
2145 } else
2146 asbt = 0;
2147 } else
2148 asbt = -1;
2149 } else
2150 asbt = 0;
2151 marker = knote_alloc(M_WAITOK);
2152 marker->kn_status = KN_MARKER;
2153 KQ_LOCK(kq);
2154
2155 retry:
2156 kevp = keva;
2157 if (kq->kq_count == 0) {
2158 if (asbt == -1) {
2159 error = EWOULDBLOCK;
2160 } else {
2161 kq->kq_state |= KQ_SLEEP;
2162 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
2163 "kqread", asbt, rsbt, C_ABSOLUTE);
2164 }
2165 if (error == 0)
2166 goto retry;
2167 /* don't restart after signals... */
2168 if (error == ERESTART)
2169 error = EINTR;
2170 else if (error == EWOULDBLOCK)
2171 error = 0;
2172 goto done;
2173 }
2174
2175 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
2176 influx = 0;
2177 while (count) {
2178 KQ_OWNED(kq);
2179 kn = TAILQ_FIRST(&kq->kq_head);
2180
2181 if ((kn->kn_status == KN_MARKER && kn != marker) ||
2182 kn_in_flux(kn)) {
2183 if (influx) {
2184 influx = 0;
2185 KQ_FLUX_WAKEUP(kq);
2186 }
2187 kq->kq_state |= KQ_FLUXWAIT;
2188 error = msleep(kq, &kq->kq_lock, PSOCK,
2189 "kqflxwt", 0);
2190 continue;
2191 }
2192
2193 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2194 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
2195 kn->kn_status &= ~KN_QUEUED;
2196 kq->kq_count--;
2197 continue;
2198 }
2199 if (kn == marker) {
2200 KQ_FLUX_WAKEUP(kq);
2201 if (count == maxevents)
2202 goto retry;
2203 goto done;
2204 }
2205 KASSERT(!kn_in_flux(kn),
2206 ("knote %p is unexpectedly in flux", kn));
2207
2208 if ((kn->kn_flags & EV_DROP) == EV_DROP) {
2209 kn->kn_status &= ~KN_QUEUED;
2210 kn_enter_flux(kn);
2211 kq->kq_count--;
2212 KQ_UNLOCK(kq);
2213 /*
2214 * We don't need to lock the list since we've
2215 * marked it as in flux.
2216 */
2217 knote_drop(kn, td);
2218 KQ_LOCK(kq);
2219 continue;
2220 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
2221 kn->kn_status &= ~KN_QUEUED;
2222 kn_enter_flux(kn);
2223 kq->kq_count--;
2224 KQ_UNLOCK(kq);
2225 /*
2226 * We don't need to lock the list since we've
2227 * marked the knote as being in flux.
2228 */
2229 *kevp = kn->kn_kevent;
2230 knote_drop(kn, td);
2231 KQ_LOCK(kq);
2232 kn = NULL;
2233 } else {
2234 kn->kn_status |= KN_SCAN;
2235 kn_enter_flux(kn);
2236 KQ_UNLOCK(kq);
2237 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
2238 KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
2239 knl = kn_list_lock(kn);
2240 if (kn->kn_fop->f_event(kn, 0) == 0) {
2241 KQ_LOCK(kq);
2242 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2243 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
2244 KN_SCAN);
2245 kn_leave_flux(kn);
2246 kq->kq_count--;
2247 kn_list_unlock(knl);
2248 influx = 1;
2249 continue;
2250 }
2251 touch = (!kn->kn_fop->f_isfd &&
2252 kn->kn_fop->f_touch != NULL);
2253 if (touch)
2254 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
2255 else
2256 *kevp = kn->kn_kevent;
2257 KQ_LOCK(kq);
2258 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
2259 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
2260 /*
2261 * Manually clear knotes who weren't
2262 * 'touch'ed.
2263 */
2264 if (touch == 0 && kn->kn_flags & EV_CLEAR) {
2265 kn->kn_data = 0;
2266 kn->kn_fflags = 0;
2267 }
2268 if (kn->kn_flags & EV_DISPATCH)
2269 kn->kn_status |= KN_DISABLED;
2270 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
2271 kq->kq_count--;
2272 } else
2273 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2274
2275 kn->kn_status &= ~KN_SCAN;
2276 kn_leave_flux(kn);
2277 kn_list_unlock(knl);
2278 influx = 1;
2279 }
2280
2281 /* we are returning a copy to the user */
2282 kevp++;
2283 nkev++;
2284 count--;
2285
2286 if (nkev == KQ_NEVENTS) {
2287 influx = 0;
2288 KQ_UNLOCK_FLUX(kq);
2289 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2290 nkev = 0;
2291 kevp = keva;
2292 KQ_LOCK(kq);
2293 if (error)
2294 break;
2295 }
2296 }
2297 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
2298 done:
2299 KQ_OWNED(kq);
2300 KQ_UNLOCK_FLUX(kq);
2301 knote_free(marker);
2302 done_nl:
2303 KQ_NOTOWNED(kq);
2304 if (nkev != 0)
2305 error = k_ops->k_copyout(k_ops->arg, keva, nkev);
2306 td->td_retval[0] = maxevents - count;
2307 return (error);
2308 }
2309
2310 /*ARGSUSED*/
2311 static int
kqueue_ioctl(struct file * fp,u_long cmd,void * data,struct ucred * active_cred,struct thread * td)2312 kqueue_ioctl(struct file *fp, u_long cmd, void *data,
2313 struct ucred *active_cred, struct thread *td)
2314 {
2315 /*
2316 * Enabling sigio causes two major problems:
2317 * 1) infinite recursion:
2318 * Synopsys: kevent is being used to track signals and have FIOASYNC
2319 * set. On receipt of a signal this will cause a kqueue to recurse
2320 * into itself over and over. Sending the sigio causes the kqueue
2321 * to become ready, which in turn posts sigio again, forever.
2322 * Solution: this can be solved by setting a flag in the kqueue that
2323 * we have a SIGIO in progress.
2324 * 2) locking problems:
2325 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
2326 * us above the proc and pgrp locks.
2327 * Solution: Post a signal using an async mechanism, being sure to
2328 * record a generation count in the delivery so that we do not deliver
2329 * a signal to the wrong process.
2330 *
2331 * Note, these two mechanisms are somewhat mutually exclusive!
2332 */
2333 #if 0
2334 struct kqueue *kq;
2335
2336 kq = fp->f_data;
2337 switch (cmd) {
2338 case FIOASYNC:
2339 if (*(int *)data) {
2340 kq->kq_state |= KQ_ASYNC;
2341 } else {
2342 kq->kq_state &= ~KQ_ASYNC;
2343 }
2344 return (0);
2345
2346 case FIOSETOWN:
2347 return (fsetown(*(int *)data, &kq->kq_sigio));
2348
2349 case FIOGETOWN:
2350 *(int *)data = fgetown(&kq->kq_sigio);
2351 return (0);
2352 }
2353 #endif
2354
2355 return (ENOTTY);
2356 }
2357
2358 /*ARGSUSED*/
2359 static int
kqueue_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)2360 kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
2361 struct thread *td)
2362 {
2363 struct kqueue *kq;
2364 int revents = 0;
2365 int error;
2366
2367 if ((error = kqueue_acquire(fp, &kq)))
2368 return POLLERR;
2369
2370 KQ_LOCK(kq);
2371 if (events & (POLLIN | POLLRDNORM)) {
2372 if (kq->kq_count) {
2373 revents |= events & (POLLIN | POLLRDNORM);
2374 } else {
2375 selrecord(td, &kq->kq_sel);
2376 if (SEL_WAITING(&kq->kq_sel))
2377 kq->kq_state |= KQ_SEL;
2378 }
2379 }
2380 kqueue_release(kq, 1);
2381 KQ_UNLOCK(kq);
2382 return (revents);
2383 }
2384
2385 /*ARGSUSED*/
2386 static int
kqueue_stat(struct file * fp,struct stat * st,struct ucred * active_cred)2387 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
2388 {
2389
2390 bzero((void *)st, sizeof *st);
2391 /*
2392 * We no longer return kq_count because the unlocked value is useless.
2393 * If you spent all this time getting the count, why not spend your
2394 * syscall better by calling kevent?
2395 *
2396 * XXX - This is needed for libc_r.
2397 */
2398 st->st_mode = S_IFIFO;
2399 return (0);
2400 }
2401
2402 static void
kqueue_drain(struct kqueue * kq,struct thread * td)2403 kqueue_drain(struct kqueue *kq, struct thread *td)
2404 {
2405 struct knote *kn;
2406 int i;
2407
2408 KQ_LOCK(kq);
2409
2410 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
2411 ("kqueue already closing"));
2412 kq->kq_state |= KQ_CLOSING;
2413 if (kq->kq_refcnt > 1)
2414 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);
2415
2416 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));
2417
2418 KASSERT(knlist_empty(&kq->kq_sel.si_note),
2419 ("kqueue's knlist not empty"));
2420
2421 for (i = 0; i < kq->kq_knlistsize; i++) {
2422 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
2423 if (kn_in_flux(kn)) {
2424 kq->kq_state |= KQ_FLUXWAIT;
2425 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
2426 continue;
2427 }
2428 kn_enter_flux(kn);
2429 KQ_UNLOCK(kq);
2430 knote_drop(kn, td);
2431 KQ_LOCK(kq);
2432 }
2433 }
2434 if (kq->kq_knhashmask != 0) {
2435 for (i = 0; i <= kq->kq_knhashmask; i++) {
2436 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
2437 if (kn_in_flux(kn)) {
2438 kq->kq_state |= KQ_FLUXWAIT;
2439 msleep(kq, &kq->kq_lock, PSOCK,
2440 "kqclo2", 0);
2441 continue;
2442 }
2443 kn_enter_flux(kn);
2444 KQ_UNLOCK(kq);
2445 knote_drop(kn, td);
2446 KQ_LOCK(kq);
2447 }
2448 }
2449 }
2450
2451 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
2452 kq->kq_state |= KQ_TASKDRAIN;
2453 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
2454 }
2455
2456 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2457 selwakeuppri(&kq->kq_sel, PSOCK);
2458 if (!SEL_WAITING(&kq->kq_sel))
2459 kq->kq_state &= ~KQ_SEL;
2460 }
2461
2462 KQ_UNLOCK(kq);
2463 }
2464
2465 static void
kqueue_destroy(struct kqueue * kq)2466 kqueue_destroy(struct kqueue *kq)
2467 {
2468
2469 KASSERT(kq->kq_fdp == NULL,
2470 ("kqueue still attached to a file descriptor"));
2471 seldrain(&kq->kq_sel);
2472 knlist_destroy(&kq->kq_sel.si_note);
2473 mtx_destroy(&kq->kq_lock);
2474
2475 if (kq->kq_knhash != NULL)
2476 free(kq->kq_knhash, M_KQUEUE);
2477 if (kq->kq_knlist != NULL)
2478 free(kq->kq_knlist, M_KQUEUE);
2479
2480 funsetown(&kq->kq_sigio);
2481 }
2482
2483 /*ARGSUSED*/
2484 static int
kqueue_close(struct file * fp,struct thread * td)2485 kqueue_close(struct file *fp, struct thread *td)
2486 {
2487 struct kqueue *kq = fp->f_data;
2488 struct filedesc *fdp;
2489 int error;
2490 int filedesc_unlock;
2491
2492 if ((error = kqueue_acquire(fp, &kq)))
2493 return error;
2494 kqueue_drain(kq, td);
2495
2496 /*
2497 * We could be called due to the knote_drop() doing fdrop(),
2498 * called from kqueue_register(). In this case the global
2499 * lock is owned, and filedesc sx is locked before, to not
2500 * take the sleepable lock after non-sleepable.
2501 */
2502 fdp = kq->kq_fdp;
2503 kq->kq_fdp = NULL;
2504 if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
2505 FILEDESC_XLOCK(fdp);
2506 filedesc_unlock = 1;
2507 } else
2508 filedesc_unlock = 0;
2509 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
2510 if (filedesc_unlock)
2511 FILEDESC_XUNLOCK(fdp);
2512
2513 kqueue_destroy(kq);
2514 chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
2515 crfree(kq->kq_cred);
2516 free(kq, M_KQUEUE);
2517 fp->f_data = NULL;
2518
2519 return (0);
2520 }
2521
2522 static int
kqueue_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)2523 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2524 {
2525 struct kqueue *kq = fp->f_data;
2526
2527 kif->kf_type = KF_TYPE_KQUEUE;
2528 kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq;
2529 kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count;
2530 kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state;
2531 return (0);
2532 }
2533
2534 static void
kqueue_wakeup(struct kqueue * kq)2535 kqueue_wakeup(struct kqueue *kq)
2536 {
2537 KQ_OWNED(kq);
2538
2539 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
2540 kq->kq_state &= ~KQ_SLEEP;
2541 wakeup(kq);
2542 }
2543 if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
2544 selwakeuppri(&kq->kq_sel, PSOCK);
2545 if (!SEL_WAITING(&kq->kq_sel))
2546 kq->kq_state &= ~KQ_SEL;
2547 }
2548 if (!knlist_empty(&kq->kq_sel.si_note))
2549 kqueue_schedtask(kq);
2550 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
2551 pgsigio(&kq->kq_sigio, SIGIO, 0);
2552 }
2553 }
2554
2555 /*
2556 * Walk down a list of knotes, activating them if their event has triggered.
2557 *
2558 * There is a possibility to optimize in the case of one kq watching another.
2559 * Instead of scheduling a task to wake it up, you could pass enough state
2560 * down the chain to make up the parent kqueue. Make this code functional
2561 * first.
2562 */
2563 void
knote(struct knlist * list,long hint,int lockflags)2564 knote(struct knlist *list, long hint, int lockflags)
2565 {
2566 struct kqueue *kq;
2567 struct knote *kn, *tkn;
2568 int error;
2569
2570 if (list == NULL)
2571 return;
2572
2573 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);
2574
2575 if ((lockflags & KNF_LISTLOCKED) == 0)
2576 list->kl_lock(list->kl_lockarg);
2577
2578 /*
2579 * If we unlock the list lock (and enter influx), we can
2580 * eliminate the kqueue scheduling, but this will introduce
2581 * four lock/unlock's for each knote to test. Also, marker
2582 * would be needed to keep iteration position, since filters
2583 * or other threads could remove events.
2584 */
2585 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
2586 kq = kn->kn_kq;
2587 KQ_LOCK(kq);
2588 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
2589 /*
2590 * Do not process the influx notes, except for
2591 * the influx coming from the kq unlock in the
2592 * kqueue_scan(). In the later case, we do
2593 * not interfere with the scan, since the code
2594 * fragment in kqueue_scan() locks the knlist,
2595 * and cannot proceed until we finished.
2596 */
2597 KQ_UNLOCK(kq);
2598 } else if ((lockflags & KNF_NOKQLOCK) != 0) {
2599 kn_enter_flux(kn);
2600 KQ_UNLOCK(kq);
2601 error = kn->kn_fop->f_event(kn, hint);
2602 KQ_LOCK(kq);
2603 kn_leave_flux(kn);
2604 if (error)
2605 KNOTE_ACTIVATE(kn, 1);
2606 KQ_UNLOCK_FLUX(kq);
2607 } else {
2608 if (kn->kn_fop->f_event(kn, hint))
2609 KNOTE_ACTIVATE(kn, 1);
2610 KQ_UNLOCK(kq);
2611 }
2612 }
2613 if ((lockflags & KNF_LISTLOCKED) == 0)
2614 list->kl_unlock(list->kl_lockarg);
2615 }
2616
2617 /*
2618 * add a knote to a knlist
2619 */
2620 void
knlist_add(struct knlist * knl,struct knote * kn,int islocked)2621 knlist_add(struct knlist *knl, struct knote *kn, int islocked)
2622 {
2623
2624 KNL_ASSERT_LOCK(knl, islocked);
2625 KQ_NOTOWNED(kn->kn_kq);
2626 KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn));
2627 KASSERT((kn->kn_status & KN_DETACHED) != 0,
2628 ("knote %p was not detached", kn));
2629 KASSERT(kn->kn_knlist == NULL,
2630 ("knote %p was already on knlist %p", kn, kn->kn_knlist));
2631 if (!islocked)
2632 knl->kl_lock(knl->kl_lockarg);
2633 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
2634 if (!islocked)
2635 knl->kl_unlock(knl->kl_lockarg);
2636 KQ_LOCK(kn->kn_kq);
2637 kn->kn_knlist = knl;
2638 kn->kn_status &= ~KN_DETACHED;
2639 KQ_UNLOCK(kn->kn_kq);
2640 }
2641
2642 static void
knlist_remove_kq(struct knlist * knl,struct knote * kn,int knlislocked,int kqislocked)2643 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
2644 int kqislocked)
2645 {
2646
2647 KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked"));
2648 KNL_ASSERT_LOCK(knl, knlislocked);
2649 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
2650 KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn));
2651 KASSERT((kn->kn_status & KN_DETACHED) == 0,
2652 ("knote %p was already detached", kn));
2653 KASSERT(kn->kn_knlist == knl,
2654 ("knote %p was not on knlist %p", kn, knl));
2655 if (!knlislocked)
2656 knl->kl_lock(knl->kl_lockarg);
2657 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
2658 kn->kn_knlist = NULL;
2659 if (!knlislocked)
2660 kn_list_unlock(knl);
2661 if (!kqislocked)
2662 KQ_LOCK(kn->kn_kq);
2663 kn->kn_status |= KN_DETACHED;
2664 if (!kqislocked)
2665 KQ_UNLOCK(kn->kn_kq);
2666 }
2667
2668 /*
2669 * remove knote from the specified knlist
2670 */
2671 void
knlist_remove(struct knlist * knl,struct knote * kn,int islocked)2672 knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
2673 {
2674
2675 knlist_remove_kq(knl, kn, islocked, 0);
2676 }
2677
2678 int
knlist_empty(struct knlist * knl)2679 knlist_empty(struct knlist *knl)
2680 {
2681
2682 KNL_ASSERT_LOCKED(knl);
2683 return (SLIST_EMPTY(&knl->kl_list));
2684 }
2685
2686 static struct mtx knlist_lock;
2687 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
2688 MTX_DEF);
2689 static void knlist_mtx_lock(void *arg);
2690 static void knlist_mtx_unlock(void *arg);
2691
2692 static void
knlist_mtx_lock(void * arg)2693 knlist_mtx_lock(void *arg)
2694 {
2695
2696 mtx_lock((struct mtx *)arg);
2697 }
2698
2699 static void
knlist_mtx_unlock(void * arg)2700 knlist_mtx_unlock(void *arg)
2701 {
2702
2703 mtx_unlock((struct mtx *)arg);
2704 }
2705
2706 static void
knlist_mtx_assert_lock(void * arg,int what)2707 knlist_mtx_assert_lock(void *arg, int what)
2708 {
2709
2710 if (what == LA_LOCKED)
2711 mtx_assert((struct mtx *)arg, MA_OWNED);
2712 else
2713 mtx_assert((struct mtx *)arg, MA_NOTOWNED);
2714 }
2715
2716 void
knlist_init(struct knlist * knl,void * lock,void (* kl_lock)(void *),void (* kl_unlock)(void *),void (* kl_assert_lock)(void *,int))2717 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
2718 void (*kl_unlock)(void *),
2719 void (*kl_assert_lock)(void *, int))
2720 {
2721
2722 if (lock == NULL)
2723 knl->kl_lockarg = &knlist_lock;
2724 else
2725 knl->kl_lockarg = lock;
2726
2727 if (kl_lock == NULL)
2728 knl->kl_lock = knlist_mtx_lock;
2729 else
2730 knl->kl_lock = kl_lock;
2731 if (kl_unlock == NULL)
2732 knl->kl_unlock = knlist_mtx_unlock;
2733 else
2734 knl->kl_unlock = kl_unlock;
2735 if (kl_assert_lock == NULL)
2736 knl->kl_assert_lock = knlist_mtx_assert_lock;
2737 else
2738 knl->kl_assert_lock = kl_assert_lock;
2739
2740 knl->kl_autodestroy = 0;
2741 SLIST_INIT(&knl->kl_list);
2742 }
2743
2744 void
knlist_init_mtx(struct knlist * knl,struct mtx * lock)2745 knlist_init_mtx(struct knlist *knl, struct mtx *lock)
2746 {
2747
2748 knlist_init(knl, lock, NULL, NULL, NULL);
2749 }
2750
2751 struct knlist *
knlist_alloc(struct mtx * lock)2752 knlist_alloc(struct mtx *lock)
2753 {
2754 struct knlist *knl;
2755
2756 knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
2757 knlist_init_mtx(knl, lock);
2758 return (knl);
2759 }
2760
2761 void
knlist_destroy(struct knlist * knl)2762 knlist_destroy(struct knlist *knl)
2763 {
2764
2765 KASSERT(KNLIST_EMPTY(knl),
2766 ("destroying knlist %p with knotes on it", knl));
2767 }
2768
2769 void
knlist_detach(struct knlist * knl)2770 knlist_detach(struct knlist *knl)
2771 {
2772
2773 KNL_ASSERT_LOCKED(knl);
2774 knl->kl_autodestroy = 1;
2775 if (knlist_empty(knl)) {
2776 knlist_destroy(knl);
2777 free(knl, M_KQUEUE);
2778 }
2779 }
2780
2781 /*
2782 * Even if we are locked, we may need to drop the lock to allow any influx
2783 * knotes time to "settle".
2784 */
2785 void
knlist_cleardel(struct knlist * knl,struct thread * td,int islocked,int killkn)2786 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
2787 {
2788 struct knote *kn, *kn2;
2789 struct kqueue *kq;
2790
2791 KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
2792 if (islocked)
2793 KNL_ASSERT_LOCKED(knl);
2794 else {
2795 KNL_ASSERT_UNLOCKED(knl);
2796 knl->kl_lock(knl->kl_lockarg);
2797 }
2798
2799 for (;;) {
2800 /*
2801 * Each pass removes as many knotes as we can before dropping
2802 * into FLUXWAIT. Active knotes are simply detached and either
2803 * freed or converted to one-shot, as the attached subject is
2804 * essentially disappearing.
2805 */
2806 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
2807 kq = kn->kn_kq;
2808 KQ_LOCK(kq);
2809 if (kn_in_flux(kn)) {
2810 KQ_UNLOCK(kq);
2811 continue;
2812 }
2813 knlist_remove_kq(knl, kn, 1, 1);
2814 if (killkn) {
2815 kn_enter_flux(kn);
2816 KQ_UNLOCK(kq);
2817 knote_drop_detached(kn, td);
2818 } else {
2819 /* Make sure cleared knotes disappear soon */
2820 kn->kn_flags |= EV_EOF | EV_ONESHOT;
2821 KQ_UNLOCK(kq);
2822 }
2823 kq = NULL;
2824 }
2825
2826 if (SLIST_EMPTY(&knl->kl_list))
2827 break;
2828
2829 /* there are still in flux knotes remaining */
2830 kn = SLIST_FIRST(&knl->kl_list);
2831 kq = kn->kn_kq;
2832 KQ_LOCK(kq);
2833 KASSERT(kn_in_flux(kn), ("knote removed w/o list lock"));
2834 knl->kl_unlock(knl->kl_lockarg);
2835 kq->kq_state |= KQ_FLUXWAIT;
2836 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
2837 kq = NULL;
2838 knl->kl_lock(knl->kl_lockarg);
2839 }
2840
2841 if (islocked)
2842 KNL_ASSERT_LOCKED(knl);
2843 else {
2844 knl->kl_unlock(knl->kl_lockarg);
2845 KNL_ASSERT_UNLOCKED(knl);
2846 }
2847 }
2848
2849 /*
2850 * Remove all knotes referencing a specified fd must be called with FILEDESC
2851 * lock. This prevents a race where a new fd comes along and occupies the
2852 * entry and we attach a knote to the fd.
2853 */
2854 void
knote_fdclose(struct thread * td,int fd)2855 knote_fdclose(struct thread *td, int fd)
2856 {
2857 struct filedesc *fdp = td->td_proc->p_fd;
2858 struct kqueue *kq;
2859 struct knote *kn;
2860
2861 FILEDESC_XLOCK_ASSERT(fdp);
2862
2863 /*
2864 * We shouldn't have to worry about new kevents appearing on fd
2865 * since filedesc is locked.
2866 */
2867 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
2868 KQ_LOCK(kq);
2869 if (kq->kq_knlistsize <= fd ||
2870 SLIST_EMPTY(&kq->kq_knlist[fd])) {
2871 KQ_UNLOCK(kq);
2872 continue;
2873 }
2874
2875 while ((kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
2876 if (kn_in_flux(kn)) {
2877 /*
2878 * Wait for this knote to stabilize, it could be
2879 * the case that it's in the process of being
2880 * dropped anyways.
2881 */
2882 kq->kq_state |= KQ_FLUXWAIT;
2883 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2884 continue;
2885 }
2886 kn_enter_flux(kn);
2887 KQ_UNLOCK(kq);
2888 knote_drop(kn, td);
2889 KQ_LOCK(kq);
2890 }
2891 KQ_UNLOCK_FLUX(kq);
2892 }
2893 }
2894
2895 static int
knote_attach(struct knote * kn,struct kqueue * kq)2896 knote_attach(struct knote *kn, struct kqueue *kq)
2897 {
2898 struct klist *list;
2899
2900 KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn));
2901 KQ_OWNED(kq);
2902 MPASS(kn->kn_kq == kq);
2903
2904 if ((kq->kq_state & KQ_CLOSING) != 0)
2905 return (EBADF);
2906 if (kn->kn_fop->f_isfd) {
2907 if (kn->kn_id >= kq->kq_knlistsize)
2908 return (ENOMEM);
2909 list = &kq->kq_knlist[kn->kn_id];
2910 } else {
2911 if (kq->kq_knhash == NULL)
2912 return (ENOMEM);
2913 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2914 }
2915 SLIST_INSERT_HEAD(list, kn, kn_link);
2916 return (0);
2917 }
2918
2919 static void
knote_drop(struct knote * kn,struct thread * td)2920 knote_drop(struct knote *kn, struct thread *td)
2921 {
2922
2923 if ((kn->kn_status & KN_DETACHED) == 0)
2924 kn->kn_fop->f_detach(kn);
2925 knote_drop_detached(kn, td);
2926 }
2927
2928 static void
knote_drop_detached(struct knote * kn,struct thread * td)2929 knote_drop_detached(struct knote *kn, struct thread *td)
2930 {
2931 struct kqueue *kq;
2932 struct klist *list;
2933
2934 kq = kn->kn_kq;
2935
2936 KASSERT((kn->kn_status & KN_DETACHED) != 0,
2937 ("knote %p still attached", kn));
2938 KQ_NOTOWNED(kq);
2939
2940 KQ_LOCK(kq);
2941 for (;;) {
2942 KASSERT(kn->kn_influx >= 1,
2943 ("knote_drop called on %p with influx %d",
2944 kn, kn->kn_influx));
2945 if (kn->kn_influx == 1)
2946 break;
2947 kq->kq_state |= KQ_FLUXWAIT;
2948 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
2949 }
2950
2951 MPASS(kn->kn_kq == kq);
2952 if (kn->kn_fop->f_isfd)
2953 list = &kq->kq_knlist[kn->kn_id];
2954 else
2955 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
2956
2957 SLIST_REMOVE(list, kn, knote, kn_link);
2958 if (kn->kn_status & KN_QUEUED)
2959 knote_dequeue(kn);
2960 KQ_UNLOCK_FLUX(kq);
2961
2962 if (kn->kn_fop->f_isfd) {
2963 fdrop(kn->kn_fp, td);
2964 kn->kn_fp = NULL;
2965 }
2966 kqueue_fo_release(kn->kn_kevent.filter);
2967 kn->kn_fop = NULL;
2968 knote_free(kn);
2969 }
2970
2971 static void
knote_enqueue(struct knote * kn)2972 knote_enqueue(struct knote *kn)
2973 {
2974 struct kqueue *kq = kn->kn_kq;
2975
2976 KQ_OWNED(kn->kn_kq);
2977 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
2978
2979 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
2980 kn->kn_status |= KN_QUEUED;
2981 kq->kq_count++;
2982 kqueue_wakeup(kq);
2983 }
2984
2985 static void
knote_dequeue(struct knote * kn)2986 knote_dequeue(struct knote *kn)
2987 {
2988 struct kqueue *kq = kn->kn_kq;
2989
2990 KQ_OWNED(kn->kn_kq);
2991 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
2992
2993 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
2994 kn->kn_status &= ~KN_QUEUED;
2995 kq->kq_count--;
2996 }
2997
2998 static void
knote_init(void * dummy __unused)2999 knote_init(void *dummy __unused)
3000 {
3001
3002 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
3003 NULL, NULL, UMA_ALIGN_PTR, 0);
3004 ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue);
3005 prison0.pr_klist = knlist_alloc(&prison0.pr_mtx);
3006 }
3007 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);
3008
3009 static struct knote *
knote_alloc(int mflag)3010 knote_alloc(int mflag)
3011 {
3012
3013 return (uma_zalloc(knote_zone, mflag | M_ZERO));
3014 }
3015
3016 static void
knote_free(struct knote * kn)3017 knote_free(struct knote *kn)
3018 {
3019
3020 uma_zfree(knote_zone, kn);
3021 }
3022
3023 /*
3024 * Register the kev w/ the kq specified by fd.
3025 */
3026 int
kqfd_register(int fd,struct kevent * kev,struct thread * td,int mflag)3027 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag)
3028 {
3029 struct kqueue *kq;
3030 struct file *fp;
3031 cap_rights_t rights;
3032 int error;
3033
3034 error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE),
3035 &fp);
3036 if (error != 0)
3037 return (error);
3038 if ((error = kqueue_acquire(fp, &kq)) != 0)
3039 goto noacquire;
3040
3041 error = kqueue_register(kq, kev, td, mflag);
3042 kqueue_release(kq, 0);
3043
3044 noacquire:
3045 fdrop(fp, td);
3046 return (error);
3047 }
3048
3049 static int
kqueue_fork_alloc(struct filedesc * fdp,struct file * fp,struct file ** fp1,struct thread * td)3050 kqueue_fork_alloc(struct filedesc *fdp, struct file *fp, struct file **fp1,
3051 struct thread *td)
3052 {
3053 struct kqueue *kq, *kq1;
3054 int error;
3055
3056 MPASS(fp->f_type == DTYPE_KQUEUE);
3057 kq = fp->f_data;
3058 if ((kq->kq_state & KQ_CPONFORK) == 0)
3059 return (EOPNOTSUPP);
3060 error = kqueue_acquire_ref(kq);
3061 if (error != 0)
3062 return (error);
3063 error = kern_kqueue_alloc(td, fdp, NULL, fp1, 0, NULL, true, &kq1);
3064 if (error == 0) {
3065 kq1->kq_forksrc = kq;
3066 (*fp1)->f_flag = fp->f_flag & (FREAD | FWRITE | FEXEC |
3067 O_CLOEXEC | O_CLOFORK);
3068 } else {
3069 kqueue_release(kq, 0);
3070 }
3071 return (error);
3072 }
3073
3074 static void
kqueue_fork_copy_knote(struct kqueue * kq1,struct knote * kn,struct proc * p1,struct filedesc * fdp)3075 kqueue_fork_copy_knote(struct kqueue *kq1, struct knote *kn, struct proc *p1,
3076 struct filedesc *fdp)
3077 {
3078 struct knote *kn1;
3079 const struct filterops *fop;
3080 int error;
3081
3082 fop = kn->kn_fop;
3083 if (fop->f_copy == NULL || (fop->f_isfd &&
3084 fdp->fd_files->fdt_ofiles[kn->kn_kevent.ident].fde_file == NULL))
3085 return;
3086 error = kqueue_expand(kq1, fop, kn->kn_kevent.ident, M_WAITOK);
3087 if (error != 0)
3088 return;
3089
3090 kn1 = knote_alloc(M_WAITOK);
3091 *kn1 = *kn;
3092 kn1->kn_status |= KN_DETACHED;
3093 kn1->kn_status &= ~KN_QUEUED;
3094 kn1->kn_kq = kq1;
3095 kn1->kn_knlist = NULL;
3096 error = fop->f_copy(kn1, p1);
3097 if (error != 0) {
3098 knote_free(kn1);
3099 return;
3100 }
3101 (void)kqueue_fo_find(kn->kn_kevent.filter);
3102 if (fop->f_isfd && !fhold(kn1->kn_fp)) {
3103 fop->f_detach(kn1);
3104 kqueue_fo_release(kn->kn_kevent.filter);
3105 knote_free(kn1);
3106 return;
3107 }
3108 if (kn->kn_knlist != NULL)
3109 knlist_add(kn->kn_knlist, kn1, 0);
3110 KQ_LOCK(kq1);
3111 knote_attach(kn1, kq1);
3112 kn1->kn_influx = 0;
3113 if ((kn->kn_status & KN_QUEUED) != 0)
3114 knote_enqueue(kn1);
3115 KQ_UNLOCK(kq1);
3116 }
3117
3118 static void
kqueue_fork_copy_list(struct klist * knlist,struct knote * marker,struct kqueue * kq,struct kqueue * kq1,struct proc * p1,struct filedesc * fdp)3119 kqueue_fork_copy_list(struct klist *knlist, struct knote *marker,
3120 struct kqueue *kq, struct kqueue *kq1, struct proc *p1,
3121 struct filedesc *fdp)
3122 {
3123 struct knote *kn;
3124
3125 KQ_OWNED(kq);
3126 kn = SLIST_FIRST(knlist);
3127 while (kn != NULL) {
3128 MPASS(kn->kn_kq == kq);
3129 if ((kn->kn_status & KN_DETACHED) != 0 ||
3130 (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0)) {
3131 kn = SLIST_NEXT(kn, kn_link);
3132 continue;
3133 }
3134 kn_enter_flux(kn);
3135 SLIST_INSERT_AFTER(kn, marker, kn_link);
3136 KQ_UNLOCK(kq);
3137 kqueue_fork_copy_knote(kq1, kn, p1, fdp);
3138 KQ_LOCK(kq);
3139 kn_leave_flux(kn);
3140 kn = SLIST_NEXT(marker, kn_link);
3141 /* XXXKIB switch kn_link to LIST? */
3142 SLIST_REMOVE(knlist, marker, knote, kn_link);
3143 }
3144 }
3145
3146 static int
kqueue_fork_copy(struct filedesc * fdp,struct file * fp,struct file * fp1,struct proc * p1,struct thread * td)3147 kqueue_fork_copy(struct filedesc *fdp, struct file *fp, struct file *fp1,
3148 struct proc *p1, struct thread *td)
3149 {
3150 struct kqueue *kq, *kq1;
3151 struct knote *marker;
3152 int error, i;
3153
3154 error = 0;
3155 MPASS(fp == NULL);
3156 MPASS(fp1->f_type == DTYPE_KQUEUE);
3157
3158 kq1 = fp1->f_data;
3159 kq = kq1->kq_forksrc;
3160 marker = knote_alloc(M_WAITOK);
3161 marker->kn_status = KN_MARKER;
3162 marker->kn_kq = kq;
3163
3164 KQ_LOCK(kq);
3165 for (i = 0; i < kq->kq_knlistsize; i++) {
3166 kqueue_fork_copy_list(&kq->kq_knlist[i], marker, kq, kq1,
3167 p1, fdp);
3168 }
3169 if (kq->kq_knhashmask != 0) {
3170 for (i = 0; i <= kq->kq_knhashmask; i++) {
3171 kqueue_fork_copy_list(&kq->kq_knhash[i], marker, kq,
3172 kq1, p1, fdp);
3173 }
3174 }
3175 kqueue_release(kq, 1);
3176 kq1->kq_forksrc = NULL;
3177 KQ_UNLOCK_FLUX(kq);
3178
3179 knote_free(marker);
3180 return (error);
3181 }
3182
3183 static int
kqueue_fork(struct filedesc * fdp,struct file * fp,struct file ** fp1,struct proc * p1,struct thread * td)3184 kqueue_fork(struct filedesc *fdp, struct file *fp, struct file **fp1,
3185 struct proc *p1, struct thread *td)
3186 {
3187 if (*fp1 == NULL)
3188 return (kqueue_fork_alloc(fdp, fp, fp1, td));
3189 return (kqueue_fork_copy(fdp, fp, *fp1, p1, td));
3190 }
3191
3192 int
knote_triv_copy(struct knote * kn __unused,struct proc * p1 __unused)3193 knote_triv_copy(struct knote *kn __unused, struct proc *p1 __unused)
3194 {
3195 return (0);
3196 }
3197
3198 struct knote_status_export_bit {
3199 int kn_status_bit;
3200 int knt_status_bit;
3201 };
3202
3203 #define ST(name) \
3204 { .kn_status_bit = KN_##name, .knt_status_bit = KNOTE_STATUS_##name }
3205 static const struct knote_status_export_bit knote_status_export_bits[] = {
3206 ST(ACTIVE),
3207 ST(QUEUED),
3208 ST(DISABLED),
3209 ST(DETACHED),
3210 ST(KQUEUE),
3211 };
3212 #undef ST
3213
3214 static int
knote_status_export(int kn_status)3215 knote_status_export(int kn_status)
3216 {
3217 const struct knote_status_export_bit *b;
3218 unsigned i;
3219 int res;
3220
3221 res = 0;
3222 for (i = 0; i < nitems(knote_status_export_bits); i++) {
3223 b = &knote_status_export_bits[i];
3224 if ((kn_status & b->kn_status_bit) != 0)
3225 res |= b->knt_status_bit;
3226 }
3227 return (res);
3228 }
3229
3230 static int
kern_proc_kqueue_report_one(struct sbuf * s,struct proc * p,int kq_fd,struct kqueue * kq,struct knote * kn,bool compat32 __unused)3231 kern_proc_kqueue_report_one(struct sbuf *s, struct proc *p,
3232 int kq_fd, struct kqueue *kq, struct knote *kn, bool compat32 __unused)
3233 {
3234 struct kinfo_knote kin;
3235 #ifdef COMPAT_FREEBSD32
3236 struct kinfo_knote32 kin32;
3237 #endif
3238 int error;
3239
3240 if (kn->kn_status == KN_MARKER)
3241 return (0);
3242
3243 memset(&kin, 0, sizeof(kin));
3244 kin.knt_kq_fd = kq_fd;
3245 memcpy(&kin.knt_event, &kn->kn_kevent, sizeof(struct kevent));
3246 kin.knt_status = knote_status_export(kn->kn_status);
3247 kn_enter_flux(kn);
3248 KQ_UNLOCK_FLUX(kq);
3249 if (kn->kn_fop->f_userdump != NULL)
3250 (void)kn->kn_fop->f_userdump(p, kn, &kin);
3251 #ifdef COMPAT_FREEBSD32
3252 if (compat32) {
3253 freebsd32_kinfo_knote_to_32(&kin, &kin32);
3254 error = sbuf_bcat(s, &kin32, sizeof(kin32));
3255 } else
3256 #endif
3257 error = sbuf_bcat(s, &kin, sizeof(kin));
3258 KQ_LOCK(kq);
3259 kn_leave_flux(kn);
3260 return (error);
3261 }
3262
3263 static int
kern_proc_kqueue_report(struct sbuf * s,struct proc * p,int kq_fd,struct kqueue * kq,bool compat32)3264 kern_proc_kqueue_report(struct sbuf *s, struct proc *p, int kq_fd,
3265 struct kqueue *kq, bool compat32)
3266 {
3267 struct knote *kn;
3268 int error, i;
3269
3270 error = 0;
3271 KQ_LOCK(kq);
3272 for (i = 0; i < kq->kq_knlistsize; i++) {
3273 SLIST_FOREACH(kn, &kq->kq_knlist[i], kn_link) {
3274 MPASS(kn->kn_kq == kq);
3275 error = kern_proc_kqueue_report_one(s, p, kq_fd,
3276 kq, kn, compat32);
3277 if (error != 0)
3278 goto out;
3279 }
3280 }
3281 if (kq->kq_knhashmask == 0)
3282 goto out;
3283 for (i = 0; i <= kq->kq_knhashmask; i++) {
3284 SLIST_FOREACH(kn, &kq->kq_knhash[i], kn_link) {
3285 MPASS(kn->kn_kq == kq);
3286 error = kern_proc_kqueue_report_one(s, p, kq_fd,
3287 kq, kn, compat32);
3288 if (error != 0)
3289 goto out;
3290 }
3291 }
3292 out:
3293 KQ_UNLOCK_FLUX(kq);
3294 return (error);
3295 }
3296
3297 struct kern_proc_kqueues_out1_cb_args {
3298 struct sbuf *s;
3299 bool compat32;
3300 };
3301
3302 static int
kern_proc_kqueues_out1_cb(struct proc * p,int fd,struct file * fp,void * arg)3303 kern_proc_kqueues_out1_cb(struct proc *p, int fd, struct file *fp, void *arg)
3304 {
3305 struct kqueue *kq;
3306 struct kern_proc_kqueues_out1_cb_args *a;
3307
3308 if (fp->f_type != DTYPE_KQUEUE)
3309 return (0);
3310 a = arg;
3311 kq = fp->f_data;
3312 return (kern_proc_kqueue_report(a->s, p, fd, kq, a->compat32));
3313 }
3314
3315 static int
kern_proc_kqueues_out1(struct thread * td,struct proc * p,struct sbuf * s,bool compat32)3316 kern_proc_kqueues_out1(struct thread *td, struct proc *p, struct sbuf *s,
3317 bool compat32)
3318 {
3319 struct kern_proc_kqueues_out1_cb_args a;
3320
3321 a.s = s;
3322 a.compat32 = compat32;
3323 return (fget_remote_foreach(td, p, kern_proc_kqueues_out1_cb, &a));
3324 }
3325
3326 int
kern_proc_kqueues_out(struct proc * p,struct sbuf * sb,size_t maxlen,bool compat32)3327 kern_proc_kqueues_out(struct proc *p, struct sbuf *sb, size_t maxlen,
3328 bool compat32)
3329 {
3330 struct sbuf *s, sm;
3331 size_t sb_len;
3332 int error;
3333
3334 if (maxlen == -1 || maxlen == 0)
3335 sb_len = 128;
3336 else
3337 sb_len = maxlen;
3338 s = sbuf_new(&sm, NULL, sb_len, maxlen == -1 ? SBUF_AUTOEXTEND :
3339 SBUF_FIXEDLEN);
3340 error = kern_proc_kqueues_out1(curthread, p, s, compat32);
3341 sbuf_finish(s);
3342 if (error == 0) {
3343 sbuf_bcat(sb, sbuf_data(s), MIN(sbuf_len(s), maxlen == -1 ?
3344 SIZE_T_MAX : maxlen));
3345 }
3346 sbuf_delete(s);
3347 return (error);
3348 }
3349
3350 static int
sysctl_kern_proc_kqueue_one(struct thread * td,struct sbuf * s,struct proc * p,int kq_fd,bool compat32)3351 sysctl_kern_proc_kqueue_one(struct thread *td, struct sbuf *s, struct proc *p,
3352 int kq_fd, bool compat32)
3353 {
3354 struct file *fp;
3355 struct kqueue *kq;
3356 int error;
3357
3358 error = fget_remote(td, p, kq_fd, &fp);
3359 if (error == 0) {
3360 if (fp->f_type != DTYPE_KQUEUE) {
3361 error = EINVAL;
3362 } else {
3363 kq = fp->f_data;
3364 error = kern_proc_kqueue_report(s, p, kq_fd, kq,
3365 compat32);
3366 }
3367 fdrop(fp, td);
3368 }
3369 return (error);
3370 }
3371
3372 static int
sysctl_kern_proc_kqueue(SYSCTL_HANDLER_ARGS)3373 sysctl_kern_proc_kqueue(SYSCTL_HANDLER_ARGS)
3374 {
3375 struct thread *td;
3376 struct proc *p;
3377 struct sbuf *s, sm;
3378 int error, error1, *name;
3379 bool compat32;
3380
3381 name = (int *)arg1;
3382 if ((u_int)arg2 > 2 || (u_int)arg2 == 0)
3383 return (EINVAL);
3384
3385 td = curthread;
3386 #ifdef COMPAT_FREEBSD32
3387 compat32 = SV_CURPROC_FLAG(SV_ILP32);
3388 #else
3389 compat32 = false;
3390 #endif
3391
3392 error = pget((pid_t)name[0], PGET_NOTWEXIT, &p);
3393 if (error != 0)
3394 return (error);
3395
3396 _PHOLD(p);
3397 execve_block_wait(td, p);
3398 error = p_candebug(td, p);
3399 if (error != 0)
3400 goto out1;
3401 PROC_UNLOCK(p);
3402
3403 s = sbuf_new_for_sysctl(&sm, NULL, 0, req);
3404 if (s == NULL) {
3405 error = ENOMEM;
3406 goto out;
3407 }
3408 sbuf_clear_flags(s, SBUF_INCLUDENUL);
3409
3410 if ((u_int)arg2 == 1) {
3411 error = kern_proc_kqueues_out1(td, p, s, compat32);
3412 } else {
3413 error = sysctl_kern_proc_kqueue_one(td, s, p,
3414 name[1] /* kq_fd */, compat32);
3415 }
3416
3417 error1 = sbuf_finish(s);
3418 if (error == 0)
3419 error = error1;
3420 sbuf_delete(s);
3421
3422 out:
3423 PROC_LOCK(p);
3424 out1:
3425 execve_unblock(td, p);
3426 _PRELE(p);
3427 PROC_UNLOCK(p);
3428 return (error);
3429 }
3430
3431 static SYSCTL_NODE(_kern_proc, KERN_PROC_KQUEUE, kq,
3432 CTLFLAG_RD | CTLFLAG_MPSAFE,
3433 sysctl_kern_proc_kqueue, "KQueue events");
3434