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