Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_event.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_ktrace.h" 35 #include "opt_kqueue.h" 36 37 #ifdef COMPAT_FREEBSD11 38 #define _WANT_FREEBSD11_KEVENT 39 #endif 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/capsicum.h> 44 #include <sys/kernel.h> 45 #include <sys/limits.h> 46 #include <sys/lock.h> 47 #include <sys/mutex.h> 48 #include <sys/rwlock.h> 49 #include <sys/proc.h> 50 #include <sys/malloc.h> 51 #include <sys/unistd.h> 52 #include <sys/file.h> 53 #include <sys/filedesc.h> 54 #include <sys/filio.h> 55 #include <sys/fcntl.h> 56 #include <sys/kthread.h> 57 #include <sys/selinfo.h> 58 #include <sys/queue.h> 59 #include <sys/event.h> 60 #include <sys/eventvar.h> 61 #include <sys/poll.h> 62 #include <sys/protosw.h> 63 #include <sys/resourcevar.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/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 80 #include <vm/uma.h> 81 82 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 83 84 /* 85 * This lock is used if multiple kq locks are required. This possibly 86 * should be made into a per proc lock. 87 */ 88 static struct mtx kq_global; 89 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 90 #define KQ_GLOBAL_LOCK(lck, haslck) do { \ 91 if (!haslck) \ 92 mtx_lock(lck); \ 93 haslck = 1; \ 94 } while (0) 95 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 96 if (haslck) \ 97 mtx_unlock(lck); \ 98 haslck = 0; \ 99 } while (0) 100 101 TASKQUEUE_DEFINE_THREAD(kqueue_ctx); 102 103 static int kevent_copyout(void *arg, struct kevent *kevp, int count); 104 static int kevent_copyin(void *arg, struct kevent *kevp, int count); 105 static int kqueue_register(struct kqueue *kq, struct kevent *kev, 106 struct thread *td, int mflag); 107 static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 108 static void kqueue_release(struct kqueue *kq, int locked); 109 static void kqueue_destroy(struct kqueue *kq); 110 static void kqueue_drain(struct kqueue *kq, struct thread *td); 111 static int kqueue_expand(struct kqueue *kq, struct filterops *fops, 112 uintptr_t ident, int mflag); 113 static void kqueue_task(void *arg, int pending); 114 static int kqueue_scan(struct kqueue *kq, int maxevents, 115 struct kevent_copyops *k_ops, 116 const struct timespec *timeout, 117 struct kevent *keva, struct thread *td); 118 static void kqueue_wakeup(struct kqueue *kq); 119 static struct filterops *kqueue_fo_find(int filt); 120 static void kqueue_fo_release(int filt); 121 struct g_kevent_args; 122 static int kern_kevent_generic(struct thread *td, 123 struct g_kevent_args *uap, 124 struct kevent_copyops *k_ops, const char *struct_name); 125 126 static fo_ioctl_t kqueue_ioctl; 127 static fo_poll_t kqueue_poll; 128 static fo_kqfilter_t kqueue_kqfilter; 129 static fo_stat_t kqueue_stat; 130 static fo_close_t kqueue_close; 131 static fo_fill_kinfo_t kqueue_fill_kinfo; 132 133 static struct fileops kqueueops = { 134 .fo_read = invfo_rdwr, 135 .fo_write = invfo_rdwr, 136 .fo_truncate = invfo_truncate, 137 .fo_ioctl = kqueue_ioctl, 138 .fo_poll = kqueue_poll, 139 .fo_kqfilter = kqueue_kqfilter, 140 .fo_stat = kqueue_stat, 141 .fo_close = kqueue_close, 142 .fo_chmod = invfo_chmod, 143 .fo_chown = invfo_chown, 144 .fo_sendfile = invfo_sendfile, 145 .fo_fill_kinfo = kqueue_fill_kinfo, 146 }; 147 148 static int knote_attach(struct knote *kn, struct kqueue *kq); 149 static void knote_drop(struct knote *kn, struct thread *td); 150 static void knote_drop_detached(struct knote *kn, struct thread *td); 151 static void knote_enqueue(struct knote *kn); 152 static void knote_dequeue(struct knote *kn); 153 static void knote_init(void); 154 static struct knote *knote_alloc(int mflag); 155 static void knote_free(struct knote *kn); 156 157 static void filt_kqdetach(struct knote *kn); 158 static int filt_kqueue(struct knote *kn, long hint); 159 static int filt_procattach(struct knote *kn); 160 static void filt_procdetach(struct knote *kn); 161 static int filt_proc(struct knote *kn, long hint); 162 static int filt_fileattach(struct knote *kn); 163 static void filt_timerexpire(void *knx); 164 static int filt_timerattach(struct knote *kn); 165 static void filt_timerdetach(struct knote *kn); 166 static void filt_timerstart(struct knote *kn, sbintime_t to); 167 static void filt_timertouch(struct knote *kn, struct kevent *kev, 168 u_long type); 169 static int filt_timervalidate(struct knote *kn, sbintime_t *to); 170 static int filt_timer(struct knote *kn, long hint); 171 static int filt_userattach(struct knote *kn); 172 static void filt_userdetach(struct knote *kn); 173 static int filt_user(struct knote *kn, long hint); 174 static void filt_usertouch(struct knote *kn, struct kevent *kev, 175 u_long type); 176 177 static struct filterops file_filtops = { 178 .f_isfd = 1, 179 .f_attach = filt_fileattach, 180 }; 181 static struct filterops kqread_filtops = { 182 .f_isfd = 1, 183 .f_detach = filt_kqdetach, 184 .f_event = filt_kqueue, 185 }; 186 /* XXX - move to kern_proc.c? */ 187 static struct filterops proc_filtops = { 188 .f_isfd = 0, 189 .f_attach = filt_procattach, 190 .f_detach = filt_procdetach, 191 .f_event = filt_proc, 192 }; 193 static struct filterops timer_filtops = { 194 .f_isfd = 0, 195 .f_attach = filt_timerattach, 196 .f_detach = filt_timerdetach, 197 .f_event = filt_timer, 198 .f_touch = filt_timertouch, 199 }; 200 static struct filterops user_filtops = { 201 .f_attach = filt_userattach, 202 .f_detach = filt_userdetach, 203 .f_event = filt_user, 204 .f_touch = filt_usertouch, 205 }; 206 207 static uma_zone_t knote_zone; 208 static unsigned int kq_ncallouts = 0; 209 static unsigned int kq_calloutmax = 4 * 1024; 210 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 211 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 212 213 /* XXX - ensure not influx ? */ 214 #define KNOTE_ACTIVATE(kn, islock) do { \ 215 if ((islock)) \ 216 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 217 else \ 218 KQ_LOCK((kn)->kn_kq); \ 219 (kn)->kn_status |= KN_ACTIVE; \ 220 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 221 knote_enqueue((kn)); \ 222 if (!(islock)) \ 223 KQ_UNLOCK((kn)->kn_kq); \ 224 } while(0) 225 #define KQ_LOCK(kq) do { \ 226 mtx_lock(&(kq)->kq_lock); \ 227 } while (0) 228 #define KQ_FLUX_WAKEUP(kq) do { \ 229 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 230 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 231 wakeup((kq)); \ 232 } \ 233 } while (0) 234 #define KQ_UNLOCK_FLUX(kq) do { \ 235 KQ_FLUX_WAKEUP(kq); \ 236 mtx_unlock(&(kq)->kq_lock); \ 237 } while (0) 238 #define KQ_UNLOCK(kq) do { \ 239 mtx_unlock(&(kq)->kq_lock); \ 240 } while (0) 241 #define KQ_OWNED(kq) do { \ 242 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 243 } while (0) 244 #define KQ_NOTOWNED(kq) do { \ 245 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 246 } while (0) 247 248 static struct knlist * 249 kn_list_lock(struct knote *kn) 250 { 251 struct knlist *knl; 252 253 knl = kn->kn_knlist; 254 if (knl != NULL) 255 knl->kl_lock(knl->kl_lockarg); 256 return (knl); 257 } 258 259 static void 260 kn_list_unlock(struct knlist *knl) 261 { 262 bool do_free; 263 264 if (knl == NULL) 265 return; 266 do_free = knl->kl_autodestroy && knlist_empty(knl); 267 knl->kl_unlock(knl->kl_lockarg); 268 if (do_free) { 269 knlist_destroy(knl); 270 free(knl, M_KQUEUE); 271 } 272 } 273 274 static bool 275 kn_in_flux(struct knote *kn) 276 { 277 278 return (kn->kn_influx > 0); 279 } 280 281 static void 282 kn_enter_flux(struct knote *kn) 283 { 284 285 KQ_OWNED(kn->kn_kq); 286 MPASS(kn->kn_influx < INT_MAX); 287 kn->kn_influx++; 288 } 289 290 static bool 291 kn_leave_flux(struct knote *kn) 292 { 293 294 KQ_OWNED(kn->kn_kq); 295 MPASS(kn->kn_influx > 0); 296 kn->kn_influx--; 297 return (kn->kn_influx == 0); 298 } 299 300 #define KNL_ASSERT_LOCK(knl, islocked) do { \ 301 if (islocked) \ 302 KNL_ASSERT_LOCKED(knl); \ 303 else \ 304 KNL_ASSERT_UNLOCKED(knl); \ 305 } while (0) 306 #ifdef INVARIANTS 307 #define KNL_ASSERT_LOCKED(knl) do { \ 308 knl->kl_assert_locked((knl)->kl_lockarg); \ 309 } while (0) 310 #define KNL_ASSERT_UNLOCKED(knl) do { \ 311 knl->kl_assert_unlocked((knl)->kl_lockarg); \ 312 } while (0) 313 #else /* !INVARIANTS */ 314 #define KNL_ASSERT_LOCKED(knl) do {} while(0) 315 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 316 #endif /* INVARIANTS */ 317 318 #ifndef KN_HASHSIZE 319 #define KN_HASHSIZE 64 /* XXX should be tunable */ 320 #endif 321 322 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 323 324 static int 325 filt_nullattach(struct knote *kn) 326 { 327 328 return (ENXIO); 329 }; 330 331 struct filterops null_filtops = { 332 .f_isfd = 0, 333 .f_attach = filt_nullattach, 334 }; 335 336 /* XXX - make SYSINIT to add these, and move into respective modules. */ 337 extern struct filterops sig_filtops; 338 extern struct filterops fs_filtops; 339 340 /* 341 * Table for all system-defined filters. 342 */ 343 static struct mtx filterops_lock; 344 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", 345 MTX_DEF); 346 static struct { 347 struct filterops *for_fop; 348 int for_nolock; 349 int for_refcnt; 350 } sysfilt_ops[EVFILT_SYSCOUNT] = { 351 { &file_filtops, 1 }, /* EVFILT_READ */ 352 { &file_filtops, 1 }, /* EVFILT_WRITE */ 353 { &null_filtops }, /* EVFILT_AIO */ 354 { &file_filtops, 1 }, /* EVFILT_VNODE */ 355 { &proc_filtops, 1 }, /* EVFILT_PROC */ 356 { &sig_filtops, 1 }, /* EVFILT_SIGNAL */ 357 { &timer_filtops, 1 }, /* EVFILT_TIMER */ 358 { &file_filtops, 1 }, /* EVFILT_PROCDESC */ 359 { &fs_filtops, 1 }, /* EVFILT_FS */ 360 { &null_filtops }, /* EVFILT_LIO */ 361 { &user_filtops, 1 }, /* EVFILT_USER */ 362 { &null_filtops }, /* EVFILT_SENDFILE */ 363 { &file_filtops, 1 }, /* EVFILT_EMPTY */ 364 }; 365 366 /* 367 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 368 * method. 369 */ 370 static int 371 filt_fileattach(struct knote *kn) 372 { 373 374 return (fo_kqfilter(kn->kn_fp, kn)); 375 } 376 377 /*ARGSUSED*/ 378 static int 379 kqueue_kqfilter(struct file *fp, struct knote *kn) 380 { 381 struct kqueue *kq = kn->kn_fp->f_data; 382 383 if (kn->kn_filter != EVFILT_READ) 384 return (EINVAL); 385 386 kn->kn_status |= KN_KQUEUE; 387 kn->kn_fop = &kqread_filtops; 388 knlist_add(&kq->kq_sel.si_note, kn, 0); 389 390 return (0); 391 } 392 393 static void 394 filt_kqdetach(struct knote *kn) 395 { 396 struct kqueue *kq = kn->kn_fp->f_data; 397 398 knlist_remove(&kq->kq_sel.si_note, kn, 0); 399 } 400 401 /*ARGSUSED*/ 402 static int 403 filt_kqueue(struct knote *kn, long hint) 404 { 405 struct kqueue *kq = kn->kn_fp->f_data; 406 407 kn->kn_data = kq->kq_count; 408 return (kn->kn_data > 0); 409 } 410 411 /* XXX - move to kern_proc.c? */ 412 static int 413 filt_procattach(struct knote *kn) 414 { 415 struct proc *p; 416 int error; 417 bool exiting, immediate; 418 419 exiting = immediate = false; 420 if (kn->kn_sfflags & NOTE_EXIT) 421 p = pfind_any(kn->kn_id); 422 else 423 p = pfind(kn->kn_id); 424 if (p == NULL) 425 return (ESRCH); 426 if (p->p_flag & P_WEXIT) 427 exiting = true; 428 429 if ((error = p_cansee(curthread, p))) { 430 PROC_UNLOCK(p); 431 return (error); 432 } 433 434 kn->kn_ptr.p_proc = p; 435 kn->kn_flags |= EV_CLEAR; /* automatically set */ 436 437 /* 438 * Internal flag indicating registration done by kernel for the 439 * purposes of getting a NOTE_CHILD notification. 440 */ 441 if (kn->kn_flags & EV_FLAG2) { 442 kn->kn_flags &= ~EV_FLAG2; 443 kn->kn_data = kn->kn_sdata; /* ppid */ 444 kn->kn_fflags = NOTE_CHILD; 445 kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK); 446 immediate = true; /* Force immediate activation of child note. */ 447 } 448 /* 449 * Internal flag indicating registration done by kernel (for other than 450 * NOTE_CHILD). 451 */ 452 if (kn->kn_flags & EV_FLAG1) { 453 kn->kn_flags &= ~EV_FLAG1; 454 } 455 456 knlist_add(p->p_klist, kn, 1); 457 458 /* 459 * Immediately activate any child notes or, in the case of a zombie 460 * target process, exit notes. The latter is necessary to handle the 461 * case where the target process, e.g. a child, dies before the kevent 462 * is registered. 463 */ 464 if (immediate || (exiting && filt_proc(kn, NOTE_EXIT))) 465 KNOTE_ACTIVATE(kn, 0); 466 467 PROC_UNLOCK(p); 468 469 return (0); 470 } 471 472 /* 473 * The knote may be attached to a different process, which may exit, 474 * leaving nothing for the knote to be attached to. So when the process 475 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 476 * it will be deleted when read out. However, as part of the knote deletion, 477 * this routine is called, so a check is needed to avoid actually performing 478 * a detach, because the original process does not exist any more. 479 */ 480 /* XXX - move to kern_proc.c? */ 481 static void 482 filt_procdetach(struct knote *kn) 483 { 484 485 knlist_remove(kn->kn_knlist, kn, 0); 486 kn->kn_ptr.p_proc = NULL; 487 } 488 489 /* XXX - move to kern_proc.c? */ 490 static int 491 filt_proc(struct knote *kn, long hint) 492 { 493 struct proc *p; 494 u_int event; 495 496 p = kn->kn_ptr.p_proc; 497 if (p == NULL) /* already activated, from attach filter */ 498 return (0); 499 500 /* Mask off extra data. */ 501 event = (u_int)hint & NOTE_PCTRLMASK; 502 503 /* If the user is interested in this event, record it. */ 504 if (kn->kn_sfflags & event) 505 kn->kn_fflags |= event; 506 507 /* Process is gone, so flag the event as finished. */ 508 if (event == NOTE_EXIT) { 509 kn->kn_flags |= EV_EOF | EV_ONESHOT; 510 kn->kn_ptr.p_proc = NULL; 511 if (kn->kn_fflags & NOTE_EXIT) 512 kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig); 513 if (kn->kn_fflags == 0) 514 kn->kn_flags |= EV_DROP; 515 return (1); 516 } 517 518 return (kn->kn_fflags != 0); 519 } 520 521 /* 522 * Called when the process forked. It mostly does the same as the 523 * knote(), activating all knotes registered to be activated when the 524 * process forked. Additionally, for each knote attached to the 525 * parent, check whether user wants to track the new process. If so 526 * attach a new knote to it, and immediately report an event with the 527 * child's pid. 528 */ 529 void 530 knote_fork(struct knlist *list, int pid) 531 { 532 struct kqueue *kq; 533 struct knote *kn; 534 struct kevent kev; 535 int error; 536 537 if (list == NULL) 538 return; 539 540 memset(&kev, 0, sizeof(kev)); 541 list->kl_lock(list->kl_lockarg); 542 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 543 kq = kn->kn_kq; 544 KQ_LOCK(kq); 545 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 546 KQ_UNLOCK(kq); 547 continue; 548 } 549 550 /* 551 * The same as knote(), activate the event. 552 */ 553 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 554 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 555 KNOTE_ACTIVATE(kn, 1); 556 KQ_UNLOCK(kq); 557 continue; 558 } 559 560 /* 561 * The NOTE_TRACK case. In addition to the activation 562 * of the event, we need to register new events to 563 * track the child. Drop the locks in preparation for 564 * the call to kqueue_register(). 565 */ 566 kn_enter_flux(kn); 567 KQ_UNLOCK(kq); 568 list->kl_unlock(list->kl_lockarg); 569 570 /* 571 * Activate existing knote and register tracking knotes with 572 * new process. 573 * 574 * First register a knote to get just the child notice. This 575 * must be a separate note from a potential NOTE_EXIT 576 * notification since both NOTE_CHILD and NOTE_EXIT are defined 577 * to use the data field (in conflicting ways). 578 */ 579 kev.ident = pid; 580 kev.filter = kn->kn_filter; 581 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT | 582 EV_FLAG2; 583 kev.fflags = kn->kn_sfflags; 584 kev.data = kn->kn_id; /* parent */ 585 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 586 error = kqueue_register(kq, &kev, NULL, M_NOWAIT); 587 if (error) 588 kn->kn_fflags |= NOTE_TRACKERR; 589 590 /* 591 * Then register another knote to track other potential events 592 * from the new process. 593 */ 594 kev.ident = pid; 595 kev.filter = kn->kn_filter; 596 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 597 kev.fflags = kn->kn_sfflags; 598 kev.data = kn->kn_id; /* parent */ 599 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 600 error = kqueue_register(kq, &kev, NULL, M_NOWAIT); 601 if (error) 602 kn->kn_fflags |= NOTE_TRACKERR; 603 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 604 KNOTE_ACTIVATE(kn, 0); 605 list->kl_lock(list->kl_lockarg); 606 KQ_LOCK(kq); 607 kn_leave_flux(kn); 608 KQ_UNLOCK_FLUX(kq); 609 } 610 list->kl_unlock(list->kl_lockarg); 611 } 612 613 /* 614 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the 615 * interval timer support code. 616 */ 617 618 #define NOTE_TIMER_PRECMASK \ 619 (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS) 620 621 static sbintime_t 622 timer2sbintime(int64_t data, int flags) 623 { 624 int64_t secs; 625 626 /* 627 * Macros for converting to the fractional second portion of an 628 * sbintime_t using 64bit multiplication to improve precision. 629 */ 630 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32) 631 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32) 632 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32) 633 switch (flags & NOTE_TIMER_PRECMASK) { 634 case NOTE_SECONDS: 635 #ifdef __LP64__ 636 if (data > (SBT_MAX / SBT_1S)) 637 return (SBT_MAX); 638 #endif 639 return ((sbintime_t)data << 32); 640 case NOTE_MSECONDS: /* FALLTHROUGH */ 641 case 0: 642 if (data >= 1000) { 643 secs = data / 1000; 644 #ifdef __LP64__ 645 if (secs > (SBT_MAX / SBT_1S)) 646 return (SBT_MAX); 647 #endif 648 return (secs << 32 | MS_TO_SBT(data % 1000)); 649 } 650 return (MS_TO_SBT(data)); 651 case NOTE_USECONDS: 652 if (data >= 1000000) { 653 secs = data / 1000000; 654 #ifdef __LP64__ 655 if (secs > (SBT_MAX / SBT_1S)) 656 return (SBT_MAX); 657 #endif 658 return (secs << 32 | US_TO_SBT(data % 1000000)); 659 } 660 return (US_TO_SBT(data)); 661 case NOTE_NSECONDS: 662 if (data >= 1000000000) { 663 secs = data / 1000000000; 664 #ifdef __LP64__ 665 if (secs > (SBT_MAX / SBT_1S)) 666 return (SBT_MAX); 667 #endif 668 return (secs << 32 | NS_TO_SBT(data % 1000000000)); 669 } 670 return (NS_TO_SBT(data)); 671 default: 672 break; 673 } 674 return (-1); 675 } 676 677 struct kq_timer_cb_data { 678 struct callout c; 679 sbintime_t next; /* next timer event fires at */ 680 sbintime_t to; /* precalculated timer period, 0 for abs */ 681 }; 682 683 static void 684 filt_timerexpire(void *knx) 685 { 686 struct knote *kn; 687 struct kq_timer_cb_data *kc; 688 689 kn = knx; 690 kn->kn_data++; 691 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 692 693 if ((kn->kn_flags & EV_ONESHOT) != 0) 694 return; 695 kc = kn->kn_ptr.p_v; 696 if (kc->to == 0) 697 return; 698 kc->next += kc->to; 699 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, 700 PCPU_GET(cpuid), C_ABSOLUTE); 701 } 702 703 /* 704 * data contains amount of time to sleep 705 */ 706 static int 707 filt_timervalidate(struct knote *kn, sbintime_t *to) 708 { 709 struct bintime bt; 710 sbintime_t sbt; 711 712 if (kn->kn_sdata < 0) 713 return (EINVAL); 714 if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) 715 kn->kn_sdata = 1; 716 /* 717 * The only fflags values supported are the timer unit 718 * (precision) and the absolute time indicator. 719 */ 720 if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0) 721 return (EINVAL); 722 723 *to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); 724 if (*to < 0) 725 return (EINVAL); 726 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 727 getboottimebin(&bt); 728 sbt = bttosbt(bt); 729 *to = MAX(0, *to - sbt); 730 } 731 return (0); 732 } 733 734 static int 735 filt_timerattach(struct knote *kn) 736 { 737 struct kq_timer_cb_data *kc; 738 sbintime_t to; 739 unsigned int ncallouts; 740 int error; 741 742 to = -1; 743 error = filt_timervalidate(kn, &to); 744 if (error != 0) 745 return (error); 746 KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 || 747 (kn->kn_sfflags & NOTE_ABSTIME) != 0, 748 ("%s: periodic timer has a calculated zero timeout", __func__)); 749 KASSERT(to >= 0, 750 ("%s: timer has a calculated negative timeout", __func__)); 751 752 do { 753 ncallouts = kq_ncallouts; 754 if (ncallouts >= kq_calloutmax) 755 return (ENOMEM); 756 } while (!atomic_cmpset_int(&kq_ncallouts, ncallouts, ncallouts + 1)); 757 758 if ((kn->kn_sfflags & NOTE_ABSTIME) == 0) 759 kn->kn_flags |= EV_CLEAR; /* automatically set */ 760 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ 761 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK); 762 callout_init(&kc->c, 1); 763 filt_timerstart(kn, to); 764 765 return (0); 766 } 767 768 static void 769 filt_timerstart(struct knote *kn, sbintime_t to) 770 { 771 struct kq_timer_cb_data *kc; 772 773 kc = kn->kn_ptr.p_v; 774 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 775 kc->next = to; 776 kc->to = 0; 777 } else { 778 kc->next = to + sbinuptime(); 779 kc->to = to; 780 } 781 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, 782 PCPU_GET(cpuid), C_ABSOLUTE); 783 } 784 785 static void 786 filt_timerdetach(struct knote *kn) 787 { 788 struct kq_timer_cb_data *kc; 789 unsigned int old __unused; 790 791 kc = kn->kn_ptr.p_v; 792 callout_drain(&kc->c); 793 free(kc, M_KQUEUE); 794 old = atomic_fetchadd_int(&kq_ncallouts, -1); 795 KASSERT(old > 0, ("Number of callouts cannot become negative")); 796 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ 797 } 798 799 static void 800 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type) 801 { 802 struct kq_timer_cb_data *kc; 803 struct kqueue *kq; 804 sbintime_t to; 805 int error; 806 807 switch (type) { 808 case EVENT_REGISTER: 809 /* Handle re-added timers that update data/fflags */ 810 if (kev->flags & EV_ADD) { 811 kc = kn->kn_ptr.p_v; 812 813 /* Drain any existing callout. */ 814 callout_drain(&kc->c); 815 816 /* Throw away any existing undelivered record 817 * of the timer expiration. This is done under 818 * the presumption that if a process is 819 * re-adding this timer with new parameters, 820 * it is no longer interested in what may have 821 * happened under the old parameters. If it is 822 * interested, it can wait for the expiration, 823 * delete the old timer definition, and then 824 * add the new one. 825 * 826 * This has to be done while the kq is locked: 827 * - if enqueued, dequeue 828 * - make it no longer active 829 * - clear the count of expiration events 830 */ 831 kq = kn->kn_kq; 832 KQ_LOCK(kq); 833 if (kn->kn_status & KN_QUEUED) 834 knote_dequeue(kn); 835 836 kn->kn_status &= ~KN_ACTIVE; 837 kn->kn_data = 0; 838 KQ_UNLOCK(kq); 839 840 /* Reschedule timer based on new data/fflags */ 841 kn->kn_sfflags = kev->fflags; 842 kn->kn_sdata = kev->data; 843 error = filt_timervalidate(kn, &to); 844 if (error != 0) { 845 kn->kn_flags |= EV_ERROR; 846 kn->kn_data = error; 847 } else 848 filt_timerstart(kn, to); 849 } 850 break; 851 852 case EVENT_PROCESS: 853 *kev = kn->kn_kevent; 854 if (kn->kn_flags & EV_CLEAR) { 855 kn->kn_data = 0; 856 kn->kn_fflags = 0; 857 } 858 break; 859 860 default: 861 panic("filt_timertouch() - invalid type (%ld)", type); 862 break; 863 } 864 } 865 866 static int 867 filt_timer(struct knote *kn, long hint) 868 { 869 870 return (kn->kn_data != 0); 871 } 872 873 static int 874 filt_userattach(struct knote *kn) 875 { 876 877 /* 878 * EVFILT_USER knotes are not attached to anything in the kernel. 879 */ 880 kn->kn_hook = NULL; 881 if (kn->kn_fflags & NOTE_TRIGGER) 882 kn->kn_hookid = 1; 883 else 884 kn->kn_hookid = 0; 885 return (0); 886 } 887 888 static void 889 filt_userdetach(__unused struct knote *kn) 890 { 891 892 /* 893 * EVFILT_USER knotes are not attached to anything in the kernel. 894 */ 895 } 896 897 static int 898 filt_user(struct knote *kn, __unused long hint) 899 { 900 901 return (kn->kn_hookid); 902 } 903 904 static void 905 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 906 { 907 u_int ffctrl; 908 909 switch (type) { 910 case EVENT_REGISTER: 911 if (kev->fflags & NOTE_TRIGGER) 912 kn->kn_hookid = 1; 913 914 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 915 kev->fflags &= NOTE_FFLAGSMASK; 916 switch (ffctrl) { 917 case NOTE_FFNOP: 918 break; 919 920 case NOTE_FFAND: 921 kn->kn_sfflags &= kev->fflags; 922 break; 923 924 case NOTE_FFOR: 925 kn->kn_sfflags |= kev->fflags; 926 break; 927 928 case NOTE_FFCOPY: 929 kn->kn_sfflags = kev->fflags; 930 break; 931 932 default: 933 /* XXX Return error? */ 934 break; 935 } 936 kn->kn_sdata = kev->data; 937 if (kev->flags & EV_CLEAR) { 938 kn->kn_hookid = 0; 939 kn->kn_data = 0; 940 kn->kn_fflags = 0; 941 } 942 break; 943 944 case EVENT_PROCESS: 945 *kev = kn->kn_kevent; 946 kev->fflags = kn->kn_sfflags; 947 kev->data = kn->kn_sdata; 948 if (kn->kn_flags & EV_CLEAR) { 949 kn->kn_hookid = 0; 950 kn->kn_data = 0; 951 kn->kn_fflags = 0; 952 } 953 break; 954 955 default: 956 panic("filt_usertouch() - invalid type (%ld)", type); 957 break; 958 } 959 } 960 961 int 962 sys_kqueue(struct thread *td, struct kqueue_args *uap) 963 { 964 965 return (kern_kqueue(td, 0, NULL)); 966 } 967 968 static void 969 kqueue_init(struct kqueue *kq) 970 { 971 972 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK); 973 TAILQ_INIT(&kq->kq_head); 974 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 975 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 976 } 977 978 int 979 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps) 980 { 981 struct filedesc *fdp; 982 struct kqueue *kq; 983 struct file *fp; 984 struct ucred *cred; 985 int fd, error; 986 987 fdp = td->td_proc->p_fd; 988 cred = td->td_ucred; 989 if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) 990 return (ENOMEM); 991 992 error = falloc_caps(td, &fp, &fd, flags, fcaps); 993 if (error != 0) { 994 chgkqcnt(cred->cr_ruidinfo, -1, 0); 995 return (error); 996 } 997 998 /* An extra reference on `fp' has been held for us by falloc(). */ 999 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 1000 kqueue_init(kq); 1001 kq->kq_fdp = fdp; 1002 kq->kq_cred = crhold(cred); 1003 1004 FILEDESC_XLOCK(fdp); 1005 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 1006 FILEDESC_XUNLOCK(fdp); 1007 1008 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 1009 fdrop(fp, td); 1010 1011 td->td_retval[0] = fd; 1012 return (0); 1013 } 1014 1015 struct g_kevent_args { 1016 int fd; 1017 void *changelist; 1018 int nchanges; 1019 void *eventlist; 1020 int nevents; 1021 const struct timespec *timeout; 1022 }; 1023 1024 int 1025 sys_kevent(struct thread *td, struct kevent_args *uap) 1026 { 1027 struct kevent_copyops k_ops = { 1028 .arg = uap, 1029 .k_copyout = kevent_copyout, 1030 .k_copyin = kevent_copyin, 1031 .kevent_size = sizeof(struct kevent), 1032 }; 1033 struct g_kevent_args gk_args = { 1034 .fd = uap->fd, 1035 .changelist = uap->changelist, 1036 .nchanges = uap->nchanges, 1037 .eventlist = uap->eventlist, 1038 .nevents = uap->nevents, 1039 .timeout = uap->timeout, 1040 }; 1041 1042 return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent")); 1043 } 1044 1045 static int 1046 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, 1047 struct kevent_copyops *k_ops, const char *struct_name) 1048 { 1049 struct timespec ts, *tsp; 1050 #ifdef KTRACE 1051 struct kevent *eventlist = uap->eventlist; 1052 #endif 1053 int error; 1054 1055 if (uap->timeout != NULL) { 1056 error = copyin(uap->timeout, &ts, sizeof(ts)); 1057 if (error) 1058 return (error); 1059 tsp = &ts; 1060 } else 1061 tsp = NULL; 1062 1063 #ifdef KTRACE 1064 if (KTRPOINT(td, KTR_STRUCT_ARRAY)) 1065 ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist, 1066 uap->nchanges, k_ops->kevent_size); 1067 #endif 1068 1069 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 1070 k_ops, tsp); 1071 1072 #ifdef KTRACE 1073 if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) 1074 ktrstructarray(struct_name, UIO_USERSPACE, eventlist, 1075 td->td_retval[0], k_ops->kevent_size); 1076 #endif 1077 1078 return (error); 1079 } 1080 1081 /* 1082 * Copy 'count' items into the destination list pointed to by uap->eventlist. 1083 */ 1084 static int 1085 kevent_copyout(void *arg, struct kevent *kevp, int count) 1086 { 1087 struct kevent_args *uap; 1088 int error; 1089 1090 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1091 uap = (struct kevent_args *)arg; 1092 1093 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 1094 if (error == 0) 1095 uap->eventlist += count; 1096 return (error); 1097 } 1098 1099 /* 1100 * Copy 'count' items from the list pointed to by uap->changelist. 1101 */ 1102 static int 1103 kevent_copyin(void *arg, struct kevent *kevp, int count) 1104 { 1105 struct kevent_args *uap; 1106 int error; 1107 1108 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1109 uap = (struct kevent_args *)arg; 1110 1111 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 1112 if (error == 0) 1113 uap->changelist += count; 1114 return (error); 1115 } 1116 1117 #ifdef COMPAT_FREEBSD11 1118 static int 1119 kevent11_copyout(void *arg, struct kevent *kevp, int count) 1120 { 1121 struct freebsd11_kevent_args *uap; 1122 struct kevent_freebsd11 kev11; 1123 int error, i; 1124 1125 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1126 uap = (struct freebsd11_kevent_args *)arg; 1127 1128 for (i = 0; i < count; i++) { 1129 kev11.ident = kevp->ident; 1130 kev11.filter = kevp->filter; 1131 kev11.flags = kevp->flags; 1132 kev11.fflags = kevp->fflags; 1133 kev11.data = kevp->data; 1134 kev11.udata = kevp->udata; 1135 error = copyout(&kev11, uap->eventlist, sizeof(kev11)); 1136 if (error != 0) 1137 break; 1138 uap->eventlist++; 1139 kevp++; 1140 } 1141 return (error); 1142 } 1143 1144 /* 1145 * Copy 'count' items from the list pointed to by uap->changelist. 1146 */ 1147 static int 1148 kevent11_copyin(void *arg, struct kevent *kevp, int count) 1149 { 1150 struct freebsd11_kevent_args *uap; 1151 struct kevent_freebsd11 kev11; 1152 int error, i; 1153 1154 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1155 uap = (struct freebsd11_kevent_args *)arg; 1156 1157 for (i = 0; i < count; i++) { 1158 error = copyin(uap->changelist, &kev11, sizeof(kev11)); 1159 if (error != 0) 1160 break; 1161 kevp->ident = kev11.ident; 1162 kevp->filter = kev11.filter; 1163 kevp->flags = kev11.flags; 1164 kevp->fflags = kev11.fflags; 1165 kevp->data = (uintptr_t)kev11.data; 1166 kevp->udata = kev11.udata; 1167 bzero(&kevp->ext, sizeof(kevp->ext)); 1168 uap->changelist++; 1169 kevp++; 1170 } 1171 return (error); 1172 } 1173 1174 int 1175 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap) 1176 { 1177 struct kevent_copyops k_ops = { 1178 .arg = uap, 1179 .k_copyout = kevent11_copyout, 1180 .k_copyin = kevent11_copyin, 1181 .kevent_size = sizeof(struct kevent_freebsd11), 1182 }; 1183 struct g_kevent_args gk_args = { 1184 .fd = uap->fd, 1185 .changelist = uap->changelist, 1186 .nchanges = uap->nchanges, 1187 .eventlist = uap->eventlist, 1188 .nevents = uap->nevents, 1189 .timeout = uap->timeout, 1190 }; 1191 1192 return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent_freebsd11")); 1193 } 1194 #endif 1195 1196 int 1197 kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 1198 struct kevent_copyops *k_ops, const struct timespec *timeout) 1199 { 1200 cap_rights_t rights; 1201 struct file *fp; 1202 int error; 1203 1204 cap_rights_init(&rights); 1205 if (nchanges > 0) 1206 cap_rights_set(&rights, CAP_KQUEUE_CHANGE); 1207 if (nevents > 0) 1208 cap_rights_set(&rights, CAP_KQUEUE_EVENT); 1209 error = fget(td, fd, &rights, &fp); 1210 if (error != 0) 1211 return (error); 1212 1213 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); 1214 fdrop(fp, td); 1215 1216 return (error); 1217 } 1218 1219 static int 1220 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents, 1221 struct kevent_copyops *k_ops, const struct timespec *timeout) 1222 { 1223 struct kevent keva[KQ_NEVENTS]; 1224 struct kevent *kevp, *changes; 1225 int i, n, nerrors, error; 1226 1227 nerrors = 0; 1228 while (nchanges > 0) { 1229 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 1230 error = k_ops->k_copyin(k_ops->arg, keva, n); 1231 if (error) 1232 return (error); 1233 changes = keva; 1234 for (i = 0; i < n; i++) { 1235 kevp = &changes[i]; 1236 if (!kevp->filter) 1237 continue; 1238 kevp->flags &= ~EV_SYSFLAGS; 1239 error = kqueue_register(kq, kevp, td, M_WAITOK); 1240 if (error || (kevp->flags & EV_RECEIPT)) { 1241 if (nevents == 0) 1242 return (error); 1243 kevp->flags = EV_ERROR; 1244 kevp->data = error; 1245 (void)k_ops->k_copyout(k_ops->arg, kevp, 1); 1246 nevents--; 1247 nerrors++; 1248 } 1249 } 1250 nchanges -= n; 1251 } 1252 if (nerrors) { 1253 td->td_retval[0] = nerrors; 1254 return (0); 1255 } 1256 1257 return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td)); 1258 } 1259 1260 int 1261 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, 1262 struct kevent_copyops *k_ops, const struct timespec *timeout) 1263 { 1264 struct kqueue *kq; 1265 int error; 1266 1267 error = kqueue_acquire(fp, &kq); 1268 if (error != 0) 1269 return (error); 1270 error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout); 1271 kqueue_release(kq, 0); 1272 return (error); 1273 } 1274 1275 /* 1276 * Performs a kevent() call on a temporarily created kqueue. This can be 1277 * used to perform one-shot polling, similar to poll() and select(). 1278 */ 1279 int 1280 kern_kevent_anonymous(struct thread *td, int nevents, 1281 struct kevent_copyops *k_ops) 1282 { 1283 struct kqueue kq = {}; 1284 int error; 1285 1286 kqueue_init(&kq); 1287 kq.kq_refcnt = 1; 1288 error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL); 1289 kqueue_drain(&kq, td); 1290 kqueue_destroy(&kq); 1291 return (error); 1292 } 1293 1294 int 1295 kqueue_add_filteropts(int filt, struct filterops *filtops) 1296 { 1297 int error; 1298 1299 error = 0; 1300 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 1301 printf( 1302 "trying to add a filterop that is out of range: %d is beyond %d\n", 1303 ~filt, EVFILT_SYSCOUNT); 1304 return EINVAL; 1305 } 1306 mtx_lock(&filterops_lock); 1307 if (sysfilt_ops[~filt].for_fop != &null_filtops && 1308 sysfilt_ops[~filt].for_fop != NULL) 1309 error = EEXIST; 1310 else { 1311 sysfilt_ops[~filt].for_fop = filtops; 1312 sysfilt_ops[~filt].for_refcnt = 0; 1313 } 1314 mtx_unlock(&filterops_lock); 1315 1316 return (error); 1317 } 1318 1319 int 1320 kqueue_del_filteropts(int filt) 1321 { 1322 int error; 1323 1324 error = 0; 1325 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1326 return EINVAL; 1327 1328 mtx_lock(&filterops_lock); 1329 if (sysfilt_ops[~filt].for_fop == &null_filtops || 1330 sysfilt_ops[~filt].for_fop == NULL) 1331 error = EINVAL; 1332 else if (sysfilt_ops[~filt].for_refcnt != 0) 1333 error = EBUSY; 1334 else { 1335 sysfilt_ops[~filt].for_fop = &null_filtops; 1336 sysfilt_ops[~filt].for_refcnt = 0; 1337 } 1338 mtx_unlock(&filterops_lock); 1339 1340 return error; 1341 } 1342 1343 static struct filterops * 1344 kqueue_fo_find(int filt) 1345 { 1346 1347 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1348 return NULL; 1349 1350 if (sysfilt_ops[~filt].for_nolock) 1351 return sysfilt_ops[~filt].for_fop; 1352 1353 mtx_lock(&filterops_lock); 1354 sysfilt_ops[~filt].for_refcnt++; 1355 if (sysfilt_ops[~filt].for_fop == NULL) 1356 sysfilt_ops[~filt].for_fop = &null_filtops; 1357 mtx_unlock(&filterops_lock); 1358 1359 return sysfilt_ops[~filt].for_fop; 1360 } 1361 1362 static void 1363 kqueue_fo_release(int filt) 1364 { 1365 1366 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1367 return; 1368 1369 if (sysfilt_ops[~filt].for_nolock) 1370 return; 1371 1372 mtx_lock(&filterops_lock); 1373 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 1374 ("filter object refcount not valid on release")); 1375 sysfilt_ops[~filt].for_refcnt--; 1376 mtx_unlock(&filterops_lock); 1377 } 1378 1379 /* 1380 * A ref to kq (obtained via kqueue_acquire) must be held. 1381 */ 1382 static int 1383 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, 1384 int mflag) 1385 { 1386 struct filterops *fops; 1387 struct file *fp; 1388 struct knote *kn, *tkn; 1389 struct knlist *knl; 1390 int error, filt, event; 1391 int haskqglobal, filedesc_unlock; 1392 1393 if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE)) 1394 return (EINVAL); 1395 1396 fp = NULL; 1397 kn = NULL; 1398 knl = NULL; 1399 error = 0; 1400 haskqglobal = 0; 1401 filedesc_unlock = 0; 1402 1403 filt = kev->filter; 1404 fops = kqueue_fo_find(filt); 1405 if (fops == NULL) 1406 return EINVAL; 1407 1408 if (kev->flags & EV_ADD) { 1409 /* 1410 * Prevent waiting with locks. Non-sleepable 1411 * allocation failures are handled in the loop, only 1412 * if the spare knote appears to be actually required. 1413 */ 1414 tkn = knote_alloc(mflag); 1415 } else { 1416 tkn = NULL; 1417 } 1418 1419 findkn: 1420 if (fops->f_isfd) { 1421 KASSERT(td != NULL, ("td is NULL")); 1422 if (kev->ident > INT_MAX) 1423 error = EBADF; 1424 else 1425 error = fget(td, kev->ident, &cap_event_rights, &fp); 1426 if (error) 1427 goto done; 1428 1429 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 1430 kev->ident, M_NOWAIT) != 0) { 1431 /* try again */ 1432 fdrop(fp, td); 1433 fp = NULL; 1434 error = kqueue_expand(kq, fops, kev->ident, mflag); 1435 if (error) 1436 goto done; 1437 goto findkn; 1438 } 1439 1440 if (fp->f_type == DTYPE_KQUEUE) { 1441 /* 1442 * If we add some intelligence about what we are doing, 1443 * we should be able to support events on ourselves. 1444 * We need to know when we are doing this to prevent 1445 * getting both the knlist lock and the kq lock since 1446 * they are the same thing. 1447 */ 1448 if (fp->f_data == kq) { 1449 error = EINVAL; 1450 goto done; 1451 } 1452 1453 /* 1454 * Pre-lock the filedesc before the global 1455 * lock mutex, see the comment in 1456 * kqueue_close(). 1457 */ 1458 FILEDESC_XLOCK(td->td_proc->p_fd); 1459 filedesc_unlock = 1; 1460 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1461 } 1462 1463 KQ_LOCK(kq); 1464 if (kev->ident < kq->kq_knlistsize) { 1465 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1466 if (kev->filter == kn->kn_filter) 1467 break; 1468 } 1469 } else { 1470 if ((kev->flags & EV_ADD) == EV_ADD) { 1471 error = kqueue_expand(kq, fops, kev->ident, mflag); 1472 if (error != 0) 1473 goto done; 1474 } 1475 1476 KQ_LOCK(kq); 1477 1478 /* 1479 * If possible, find an existing knote to use for this kevent. 1480 */ 1481 if (kev->filter == EVFILT_PROC && 1482 (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) { 1483 /* This is an internal creation of a process tracking 1484 * note. Don't attempt to coalesce this with an 1485 * existing note. 1486 */ 1487 ; 1488 } else if (kq->kq_knhashmask != 0) { 1489 struct klist *list; 1490 1491 list = &kq->kq_knhash[ 1492 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1493 SLIST_FOREACH(kn, list, kn_link) 1494 if (kev->ident == kn->kn_id && 1495 kev->filter == kn->kn_filter) 1496 break; 1497 } 1498 } 1499 1500 /* knote is in the process of changing, wait for it to stabilize. */ 1501 if (kn != NULL && kn_in_flux(kn)) { 1502 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1503 if (filedesc_unlock) { 1504 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1505 filedesc_unlock = 0; 1506 } 1507 kq->kq_state |= KQ_FLUXWAIT; 1508 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1509 if (fp != NULL) { 1510 fdrop(fp, td); 1511 fp = NULL; 1512 } 1513 goto findkn; 1514 } 1515 1516 /* 1517 * kn now contains the matching knote, or NULL if no match 1518 */ 1519 if (kn == NULL) { 1520 if (kev->flags & EV_ADD) { 1521 kn = tkn; 1522 tkn = NULL; 1523 if (kn == NULL) { 1524 KQ_UNLOCK(kq); 1525 error = ENOMEM; 1526 goto done; 1527 } 1528 kn->kn_fp = fp; 1529 kn->kn_kq = kq; 1530 kn->kn_fop = fops; 1531 /* 1532 * apply reference counts to knote structure, and 1533 * do not release it at the end of this routine. 1534 */ 1535 fops = NULL; 1536 fp = NULL; 1537 1538 kn->kn_sfflags = kev->fflags; 1539 kn->kn_sdata = kev->data; 1540 kev->fflags = 0; 1541 kev->data = 0; 1542 kn->kn_kevent = *kev; 1543 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1544 EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); 1545 kn->kn_status = KN_DETACHED; 1546 if ((kev->flags & EV_DISABLE) != 0) 1547 kn->kn_status |= KN_DISABLED; 1548 kn_enter_flux(kn); 1549 1550 error = knote_attach(kn, kq); 1551 KQ_UNLOCK(kq); 1552 if (error != 0) { 1553 tkn = kn; 1554 goto done; 1555 } 1556 1557 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1558 knote_drop_detached(kn, td); 1559 goto done; 1560 } 1561 knl = kn_list_lock(kn); 1562 goto done_ev_add; 1563 } else { 1564 /* No matching knote and the EV_ADD flag is not set. */ 1565 KQ_UNLOCK(kq); 1566 error = ENOENT; 1567 goto done; 1568 } 1569 } 1570 1571 if (kev->flags & EV_DELETE) { 1572 kn_enter_flux(kn); 1573 KQ_UNLOCK(kq); 1574 knote_drop(kn, td); 1575 goto done; 1576 } 1577 1578 if (kev->flags & EV_FORCEONESHOT) { 1579 kn->kn_flags |= EV_ONESHOT; 1580 KNOTE_ACTIVATE(kn, 1); 1581 } 1582 1583 if ((kev->flags & EV_ENABLE) != 0) 1584 kn->kn_status &= ~KN_DISABLED; 1585 else if ((kev->flags & EV_DISABLE) != 0) 1586 kn->kn_status |= KN_DISABLED; 1587 1588 /* 1589 * The user may change some filter values after the initial EV_ADD, 1590 * but doing so will not reset any filter which has already been 1591 * triggered. 1592 */ 1593 kn->kn_status |= KN_SCAN; 1594 kn_enter_flux(kn); 1595 KQ_UNLOCK(kq); 1596 knl = kn_list_lock(kn); 1597 kn->kn_kevent.udata = kev->udata; 1598 if (!fops->f_isfd && fops->f_touch != NULL) { 1599 fops->f_touch(kn, kev, EVENT_REGISTER); 1600 } else { 1601 kn->kn_sfflags = kev->fflags; 1602 kn->kn_sdata = kev->data; 1603 } 1604 1605 done_ev_add: 1606 /* 1607 * We can get here with kn->kn_knlist == NULL. This can happen when 1608 * the initial attach event decides that the event is "completed" 1609 * already, e.g., filt_procattach() is called on a zombie process. It 1610 * will call filt_proc() which will remove it from the list, and NULL 1611 * kn_knlist. 1612 * 1613 * KN_DISABLED will be stable while the knote is in flux, so the 1614 * unlocked read will not race with an update. 1615 */ 1616 if ((kn->kn_status & KN_DISABLED) == 0) 1617 event = kn->kn_fop->f_event(kn, 0); 1618 else 1619 event = 0; 1620 1621 KQ_LOCK(kq); 1622 if (event) 1623 kn->kn_status |= KN_ACTIVE; 1624 if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) == 1625 KN_ACTIVE) 1626 knote_enqueue(kn); 1627 kn->kn_status &= ~KN_SCAN; 1628 kn_leave_flux(kn); 1629 kn_list_unlock(knl); 1630 KQ_UNLOCK_FLUX(kq); 1631 1632 done: 1633 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1634 if (filedesc_unlock) 1635 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1636 if (fp != NULL) 1637 fdrop(fp, td); 1638 knote_free(tkn); 1639 if (fops != NULL) 1640 kqueue_fo_release(filt); 1641 return (error); 1642 } 1643 1644 static int 1645 kqueue_acquire(struct file *fp, struct kqueue **kqp) 1646 { 1647 int error; 1648 struct kqueue *kq; 1649 1650 error = 0; 1651 1652 kq = fp->f_data; 1653 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1654 return (EBADF); 1655 *kqp = kq; 1656 KQ_LOCK(kq); 1657 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1658 KQ_UNLOCK(kq); 1659 return (EBADF); 1660 } 1661 kq->kq_refcnt++; 1662 KQ_UNLOCK(kq); 1663 1664 return error; 1665 } 1666 1667 static void 1668 kqueue_release(struct kqueue *kq, int locked) 1669 { 1670 if (locked) 1671 KQ_OWNED(kq); 1672 else 1673 KQ_LOCK(kq); 1674 kq->kq_refcnt--; 1675 if (kq->kq_refcnt == 1) 1676 wakeup(&kq->kq_refcnt); 1677 if (!locked) 1678 KQ_UNLOCK(kq); 1679 } 1680 1681 static void 1682 kqueue_schedtask(struct kqueue *kq) 1683 { 1684 1685 KQ_OWNED(kq); 1686 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1687 ("scheduling kqueue task while draining")); 1688 1689 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1690 taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task); 1691 kq->kq_state |= KQ_TASKSCHED; 1692 } 1693 } 1694 1695 /* 1696 * Expand the kq to make sure we have storage for fops/ident pair. 1697 * 1698 * Return 0 on success (or no work necessary), return errno on failure. 1699 */ 1700 static int 1701 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, 1702 int mflag) 1703 { 1704 struct klist *list, *tmp_knhash, *to_free; 1705 u_long tmp_knhashmask; 1706 int error, fd, size; 1707 1708 KQ_NOTOWNED(kq); 1709 1710 error = 0; 1711 to_free = NULL; 1712 if (fops->f_isfd) { 1713 fd = ident; 1714 if (kq->kq_knlistsize <= fd) { 1715 size = kq->kq_knlistsize; 1716 while (size <= fd) 1717 size += KQEXTENT; 1718 list = malloc(size * sizeof(*list), M_KQUEUE, mflag); 1719 if (list == NULL) 1720 return ENOMEM; 1721 KQ_LOCK(kq); 1722 if ((kq->kq_state & KQ_CLOSING) != 0) { 1723 to_free = list; 1724 error = EBADF; 1725 } else if (kq->kq_knlistsize > fd) { 1726 to_free = list; 1727 } else { 1728 if (kq->kq_knlist != NULL) { 1729 bcopy(kq->kq_knlist, list, 1730 kq->kq_knlistsize * sizeof(*list)); 1731 to_free = kq->kq_knlist; 1732 kq->kq_knlist = NULL; 1733 } 1734 bzero((caddr_t)list + 1735 kq->kq_knlistsize * sizeof(*list), 1736 (size - kq->kq_knlistsize) * sizeof(*list)); 1737 kq->kq_knlistsize = size; 1738 kq->kq_knlist = list; 1739 } 1740 KQ_UNLOCK(kq); 1741 } 1742 } else { 1743 if (kq->kq_knhashmask == 0) { 1744 tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE, 1745 &tmp_knhashmask, (mflag & M_WAITOK) != 0 ? 1746 HASH_WAITOK : HASH_NOWAIT); 1747 if (tmp_knhash == NULL) 1748 return (ENOMEM); 1749 KQ_LOCK(kq); 1750 if ((kq->kq_state & KQ_CLOSING) != 0) { 1751 to_free = tmp_knhash; 1752 error = EBADF; 1753 } else if (kq->kq_knhashmask == 0) { 1754 kq->kq_knhash = tmp_knhash; 1755 kq->kq_knhashmask = tmp_knhashmask; 1756 } else { 1757 to_free = tmp_knhash; 1758 } 1759 KQ_UNLOCK(kq); 1760 } 1761 } 1762 free(to_free, M_KQUEUE); 1763 1764 KQ_NOTOWNED(kq); 1765 return (error); 1766 } 1767 1768 static void 1769 kqueue_task(void *arg, int pending) 1770 { 1771 struct kqueue *kq; 1772 int haskqglobal; 1773 1774 haskqglobal = 0; 1775 kq = arg; 1776 1777 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1778 KQ_LOCK(kq); 1779 1780 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 1781 1782 kq->kq_state &= ~KQ_TASKSCHED; 1783 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 1784 wakeup(&kq->kq_state); 1785 } 1786 KQ_UNLOCK(kq); 1787 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1788 } 1789 1790 /* 1791 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 1792 * We treat KN_MARKER knotes as if they are in flux. 1793 */ 1794 static int 1795 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 1796 const struct timespec *tsp, struct kevent *keva, struct thread *td) 1797 { 1798 struct kevent *kevp; 1799 struct knote *kn, *marker; 1800 struct knlist *knl; 1801 sbintime_t asbt, rsbt; 1802 int count, error, haskqglobal, influx, nkev, touch; 1803 1804 count = maxevents; 1805 nkev = 0; 1806 error = 0; 1807 haskqglobal = 0; 1808 1809 if (maxevents == 0) 1810 goto done_nl; 1811 1812 rsbt = 0; 1813 if (tsp != NULL) { 1814 if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 || 1815 tsp->tv_nsec >= 1000000000) { 1816 error = EINVAL; 1817 goto done_nl; 1818 } 1819 if (timespecisset(tsp)) { 1820 if (tsp->tv_sec <= INT32_MAX) { 1821 rsbt = tstosbt(*tsp); 1822 if (TIMESEL(&asbt, rsbt)) 1823 asbt += tc_tick_sbt; 1824 if (asbt <= SBT_MAX - rsbt) 1825 asbt += rsbt; 1826 else 1827 asbt = 0; 1828 rsbt >>= tc_precexp; 1829 } else 1830 asbt = 0; 1831 } else 1832 asbt = -1; 1833 } else 1834 asbt = 0; 1835 marker = knote_alloc(M_WAITOK); 1836 marker->kn_status = KN_MARKER; 1837 KQ_LOCK(kq); 1838 1839 retry: 1840 kevp = keva; 1841 if (kq->kq_count == 0) { 1842 if (asbt == -1) { 1843 error = EWOULDBLOCK; 1844 } else { 1845 kq->kq_state |= KQ_SLEEP; 1846 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, 1847 "kqread", asbt, rsbt, C_ABSOLUTE); 1848 } 1849 if (error == 0) 1850 goto retry; 1851 /* don't restart after signals... */ 1852 if (error == ERESTART) 1853 error = EINTR; 1854 else if (error == EWOULDBLOCK) 1855 error = 0; 1856 goto done; 1857 } 1858 1859 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1860 influx = 0; 1861 while (count) { 1862 KQ_OWNED(kq); 1863 kn = TAILQ_FIRST(&kq->kq_head); 1864 1865 if ((kn->kn_status == KN_MARKER && kn != marker) || 1866 kn_in_flux(kn)) { 1867 if (influx) { 1868 influx = 0; 1869 KQ_FLUX_WAKEUP(kq); 1870 } 1871 kq->kq_state |= KQ_FLUXWAIT; 1872 error = msleep(kq, &kq->kq_lock, PSOCK, 1873 "kqflxwt", 0); 1874 continue; 1875 } 1876 1877 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1878 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 1879 kn->kn_status &= ~KN_QUEUED; 1880 kq->kq_count--; 1881 continue; 1882 } 1883 if (kn == marker) { 1884 KQ_FLUX_WAKEUP(kq); 1885 if (count == maxevents) 1886 goto retry; 1887 goto done; 1888 } 1889 KASSERT(!kn_in_flux(kn), 1890 ("knote %p is unexpectedly in flux", kn)); 1891 1892 if ((kn->kn_flags & EV_DROP) == EV_DROP) { 1893 kn->kn_status &= ~KN_QUEUED; 1894 kn_enter_flux(kn); 1895 kq->kq_count--; 1896 KQ_UNLOCK(kq); 1897 /* 1898 * We don't need to lock the list since we've 1899 * marked it as in flux. 1900 */ 1901 knote_drop(kn, td); 1902 KQ_LOCK(kq); 1903 continue; 1904 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 1905 kn->kn_status &= ~KN_QUEUED; 1906 kn_enter_flux(kn); 1907 kq->kq_count--; 1908 KQ_UNLOCK(kq); 1909 /* 1910 * We don't need to lock the list since we've 1911 * marked the knote as being in flux. 1912 */ 1913 *kevp = kn->kn_kevent; 1914 knote_drop(kn, td); 1915 KQ_LOCK(kq); 1916 kn = NULL; 1917 } else { 1918 kn->kn_status |= KN_SCAN; 1919 kn_enter_flux(kn); 1920 KQ_UNLOCK(kq); 1921 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 1922 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1923 knl = kn_list_lock(kn); 1924 if (kn->kn_fop->f_event(kn, 0) == 0) { 1925 KQ_LOCK(kq); 1926 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1927 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE | 1928 KN_SCAN); 1929 kn_leave_flux(kn); 1930 kq->kq_count--; 1931 kn_list_unlock(knl); 1932 influx = 1; 1933 continue; 1934 } 1935 touch = (!kn->kn_fop->f_isfd && 1936 kn->kn_fop->f_touch != NULL); 1937 if (touch) 1938 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 1939 else 1940 *kevp = kn->kn_kevent; 1941 KQ_LOCK(kq); 1942 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1943 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 1944 /* 1945 * Manually clear knotes who weren't 1946 * 'touch'ed. 1947 */ 1948 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 1949 kn->kn_data = 0; 1950 kn->kn_fflags = 0; 1951 } 1952 if (kn->kn_flags & EV_DISPATCH) 1953 kn->kn_status |= KN_DISABLED; 1954 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1955 kq->kq_count--; 1956 } else 1957 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1958 1959 kn->kn_status &= ~KN_SCAN; 1960 kn_leave_flux(kn); 1961 kn_list_unlock(knl); 1962 influx = 1; 1963 } 1964 1965 /* we are returning a copy to the user */ 1966 kevp++; 1967 nkev++; 1968 count--; 1969 1970 if (nkev == KQ_NEVENTS) { 1971 influx = 0; 1972 KQ_UNLOCK_FLUX(kq); 1973 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1974 nkev = 0; 1975 kevp = keva; 1976 KQ_LOCK(kq); 1977 if (error) 1978 break; 1979 } 1980 } 1981 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1982 done: 1983 KQ_OWNED(kq); 1984 KQ_UNLOCK_FLUX(kq); 1985 knote_free(marker); 1986 done_nl: 1987 KQ_NOTOWNED(kq); 1988 if (nkev != 0) 1989 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1990 td->td_retval[0] = maxevents - count; 1991 return (error); 1992 } 1993 1994 /*ARGSUSED*/ 1995 static int 1996 kqueue_ioctl(struct file *fp, u_long cmd, void *data, 1997 struct ucred *active_cred, struct thread *td) 1998 { 1999 /* 2000 * Enabling sigio causes two major problems: 2001 * 1) infinite recursion: 2002 * Synopsys: kevent is being used to track signals and have FIOASYNC 2003 * set. On receipt of a signal this will cause a kqueue to recurse 2004 * into itself over and over. Sending the sigio causes the kqueue 2005 * to become ready, which in turn posts sigio again, forever. 2006 * Solution: this can be solved by setting a flag in the kqueue that 2007 * we have a SIGIO in progress. 2008 * 2) locking problems: 2009 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 2010 * us above the proc and pgrp locks. 2011 * Solution: Post a signal using an async mechanism, being sure to 2012 * record a generation count in the delivery so that we do not deliver 2013 * a signal to the wrong process. 2014 * 2015 * Note, these two mechanisms are somewhat mutually exclusive! 2016 */ 2017 #if 0 2018 struct kqueue *kq; 2019 2020 kq = fp->f_data; 2021 switch (cmd) { 2022 case FIOASYNC: 2023 if (*(int *)data) { 2024 kq->kq_state |= KQ_ASYNC; 2025 } else { 2026 kq->kq_state &= ~KQ_ASYNC; 2027 } 2028 return (0); 2029 2030 case FIOSETOWN: 2031 return (fsetown(*(int *)data, &kq->kq_sigio)); 2032 2033 case FIOGETOWN: 2034 *(int *)data = fgetown(&kq->kq_sigio); 2035 return (0); 2036 } 2037 #endif 2038 2039 return (ENOTTY); 2040 } 2041 2042 /*ARGSUSED*/ 2043 static int 2044 kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 2045 struct thread *td) 2046 { 2047 struct kqueue *kq; 2048 int revents = 0; 2049 int error; 2050 2051 if ((error = kqueue_acquire(fp, &kq))) 2052 return POLLERR; 2053 2054 KQ_LOCK(kq); 2055 if (events & (POLLIN | POLLRDNORM)) { 2056 if (kq->kq_count) { 2057 revents |= events & (POLLIN | POLLRDNORM); 2058 } else { 2059 selrecord(td, &kq->kq_sel); 2060 if (SEL_WAITING(&kq->kq_sel)) 2061 kq->kq_state |= KQ_SEL; 2062 } 2063 } 2064 kqueue_release(kq, 1); 2065 KQ_UNLOCK(kq); 2066 return (revents); 2067 } 2068 2069 /*ARGSUSED*/ 2070 static int 2071 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 2072 struct thread *td) 2073 { 2074 2075 bzero((void *)st, sizeof *st); 2076 /* 2077 * We no longer return kq_count because the unlocked value is useless. 2078 * If you spent all this time getting the count, why not spend your 2079 * syscall better by calling kevent? 2080 * 2081 * XXX - This is needed for libc_r. 2082 */ 2083 st->st_mode = S_IFIFO; 2084 return (0); 2085 } 2086 2087 static void 2088 kqueue_drain(struct kqueue *kq, struct thread *td) 2089 { 2090 struct knote *kn; 2091 int i; 2092 2093 KQ_LOCK(kq); 2094 2095 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 2096 ("kqueue already closing")); 2097 kq->kq_state |= KQ_CLOSING; 2098 if (kq->kq_refcnt > 1) 2099 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 2100 2101 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 2102 2103 KASSERT(knlist_empty(&kq->kq_sel.si_note), 2104 ("kqueue's knlist not empty")); 2105 2106 for (i = 0; i < kq->kq_knlistsize; i++) { 2107 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 2108 if (kn_in_flux(kn)) { 2109 kq->kq_state |= KQ_FLUXWAIT; 2110 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 2111 continue; 2112 } 2113 kn_enter_flux(kn); 2114 KQ_UNLOCK(kq); 2115 knote_drop(kn, td); 2116 KQ_LOCK(kq); 2117 } 2118 } 2119 if (kq->kq_knhashmask != 0) { 2120 for (i = 0; i <= kq->kq_knhashmask; i++) { 2121 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 2122 if (kn_in_flux(kn)) { 2123 kq->kq_state |= KQ_FLUXWAIT; 2124 msleep(kq, &kq->kq_lock, PSOCK, 2125 "kqclo2", 0); 2126 continue; 2127 } 2128 kn_enter_flux(kn); 2129 KQ_UNLOCK(kq); 2130 knote_drop(kn, td); 2131 KQ_LOCK(kq); 2132 } 2133 } 2134 } 2135 2136 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 2137 kq->kq_state |= KQ_TASKDRAIN; 2138 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 2139 } 2140 2141 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2142 selwakeuppri(&kq->kq_sel, PSOCK); 2143 if (!SEL_WAITING(&kq->kq_sel)) 2144 kq->kq_state &= ~KQ_SEL; 2145 } 2146 2147 KQ_UNLOCK(kq); 2148 } 2149 2150 static void 2151 kqueue_destroy(struct kqueue *kq) 2152 { 2153 2154 KASSERT(kq->kq_fdp == NULL, 2155 ("kqueue still attached to a file descriptor")); 2156 seldrain(&kq->kq_sel); 2157 knlist_destroy(&kq->kq_sel.si_note); 2158 mtx_destroy(&kq->kq_lock); 2159 2160 if (kq->kq_knhash != NULL) 2161 free(kq->kq_knhash, M_KQUEUE); 2162 if (kq->kq_knlist != NULL) 2163 free(kq->kq_knlist, M_KQUEUE); 2164 2165 funsetown(&kq->kq_sigio); 2166 } 2167 2168 /*ARGSUSED*/ 2169 static int 2170 kqueue_close(struct file *fp, struct thread *td) 2171 { 2172 struct kqueue *kq = fp->f_data; 2173 struct filedesc *fdp; 2174 int error; 2175 int filedesc_unlock; 2176 2177 if ((error = kqueue_acquire(fp, &kq))) 2178 return error; 2179 kqueue_drain(kq, td); 2180 2181 /* 2182 * We could be called due to the knote_drop() doing fdrop(), 2183 * called from kqueue_register(). In this case the global 2184 * lock is owned, and filedesc sx is locked before, to not 2185 * take the sleepable lock after non-sleepable. 2186 */ 2187 fdp = kq->kq_fdp; 2188 kq->kq_fdp = NULL; 2189 if (!sx_xlocked(FILEDESC_LOCK(fdp))) { 2190 FILEDESC_XLOCK(fdp); 2191 filedesc_unlock = 1; 2192 } else 2193 filedesc_unlock = 0; 2194 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); 2195 if (filedesc_unlock) 2196 FILEDESC_XUNLOCK(fdp); 2197 2198 kqueue_destroy(kq); 2199 chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); 2200 crfree(kq->kq_cred); 2201 free(kq, M_KQUEUE); 2202 fp->f_data = NULL; 2203 2204 return (0); 2205 } 2206 2207 static int 2208 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2209 { 2210 2211 kif->kf_type = KF_TYPE_KQUEUE; 2212 return (0); 2213 } 2214 2215 static void 2216 kqueue_wakeup(struct kqueue *kq) 2217 { 2218 KQ_OWNED(kq); 2219 2220 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 2221 kq->kq_state &= ~KQ_SLEEP; 2222 wakeup(kq); 2223 } 2224 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2225 selwakeuppri(&kq->kq_sel, PSOCK); 2226 if (!SEL_WAITING(&kq->kq_sel)) 2227 kq->kq_state &= ~KQ_SEL; 2228 } 2229 if (!knlist_empty(&kq->kq_sel.si_note)) 2230 kqueue_schedtask(kq); 2231 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 2232 pgsigio(&kq->kq_sigio, SIGIO, 0); 2233 } 2234 } 2235 2236 /* 2237 * Walk down a list of knotes, activating them if their event has triggered. 2238 * 2239 * There is a possibility to optimize in the case of one kq watching another. 2240 * Instead of scheduling a task to wake it up, you could pass enough state 2241 * down the chain to make up the parent kqueue. Make this code functional 2242 * first. 2243 */ 2244 void 2245 knote(struct knlist *list, long hint, int lockflags) 2246 { 2247 struct kqueue *kq; 2248 struct knote *kn, *tkn; 2249 int error; 2250 2251 if (list == NULL) 2252 return; 2253 2254 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 2255 2256 if ((lockflags & KNF_LISTLOCKED) == 0) 2257 list->kl_lock(list->kl_lockarg); 2258 2259 /* 2260 * If we unlock the list lock (and enter influx), we can 2261 * eliminate the kqueue scheduling, but this will introduce 2262 * four lock/unlock's for each knote to test. Also, marker 2263 * would be needed to keep iteration position, since filters 2264 * or other threads could remove events. 2265 */ 2266 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) { 2267 kq = kn->kn_kq; 2268 KQ_LOCK(kq); 2269 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 2270 /* 2271 * Do not process the influx notes, except for 2272 * the influx coming from the kq unlock in the 2273 * kqueue_scan(). In the later case, we do 2274 * not interfere with the scan, since the code 2275 * fragment in kqueue_scan() locks the knlist, 2276 * and cannot proceed until we finished. 2277 */ 2278 KQ_UNLOCK(kq); 2279 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 2280 kn_enter_flux(kn); 2281 KQ_UNLOCK(kq); 2282 error = kn->kn_fop->f_event(kn, hint); 2283 KQ_LOCK(kq); 2284 kn_leave_flux(kn); 2285 if (error) 2286 KNOTE_ACTIVATE(kn, 1); 2287 KQ_UNLOCK_FLUX(kq); 2288 } else { 2289 if (kn->kn_fop->f_event(kn, hint)) 2290 KNOTE_ACTIVATE(kn, 1); 2291 KQ_UNLOCK(kq); 2292 } 2293 } 2294 if ((lockflags & KNF_LISTLOCKED) == 0) 2295 list->kl_unlock(list->kl_lockarg); 2296 } 2297 2298 /* 2299 * add a knote to a knlist 2300 */ 2301 void 2302 knlist_add(struct knlist *knl, struct knote *kn, int islocked) 2303 { 2304 2305 KNL_ASSERT_LOCK(knl, islocked); 2306 KQ_NOTOWNED(kn->kn_kq); 2307 KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn)); 2308 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2309 ("knote %p was not detached", kn)); 2310 if (!islocked) 2311 knl->kl_lock(knl->kl_lockarg); 2312 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 2313 if (!islocked) 2314 knl->kl_unlock(knl->kl_lockarg); 2315 KQ_LOCK(kn->kn_kq); 2316 kn->kn_knlist = knl; 2317 kn->kn_status &= ~KN_DETACHED; 2318 KQ_UNLOCK(kn->kn_kq); 2319 } 2320 2321 static void 2322 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, 2323 int kqislocked) 2324 { 2325 2326 KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked")); 2327 KNL_ASSERT_LOCK(knl, knlislocked); 2328 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 2329 KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn)); 2330 KASSERT((kn->kn_status & KN_DETACHED) == 0, 2331 ("knote %p was already detached", kn)); 2332 if (!knlislocked) 2333 knl->kl_lock(knl->kl_lockarg); 2334 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 2335 kn->kn_knlist = NULL; 2336 if (!knlislocked) 2337 kn_list_unlock(knl); 2338 if (!kqislocked) 2339 KQ_LOCK(kn->kn_kq); 2340 kn->kn_status |= KN_DETACHED; 2341 if (!kqislocked) 2342 KQ_UNLOCK(kn->kn_kq); 2343 } 2344 2345 /* 2346 * remove knote from the specified knlist 2347 */ 2348 void 2349 knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 2350 { 2351 2352 knlist_remove_kq(knl, kn, islocked, 0); 2353 } 2354 2355 int 2356 knlist_empty(struct knlist *knl) 2357 { 2358 2359 KNL_ASSERT_LOCKED(knl); 2360 return (SLIST_EMPTY(&knl->kl_list)); 2361 } 2362 2363 static struct mtx knlist_lock; 2364 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 2365 MTX_DEF); 2366 static void knlist_mtx_lock(void *arg); 2367 static void knlist_mtx_unlock(void *arg); 2368 2369 static void 2370 knlist_mtx_lock(void *arg) 2371 { 2372 2373 mtx_lock((struct mtx *)arg); 2374 } 2375 2376 static void 2377 knlist_mtx_unlock(void *arg) 2378 { 2379 2380 mtx_unlock((struct mtx *)arg); 2381 } 2382 2383 static void 2384 knlist_mtx_assert_locked(void *arg) 2385 { 2386 2387 mtx_assert((struct mtx *)arg, MA_OWNED); 2388 } 2389 2390 static void 2391 knlist_mtx_assert_unlocked(void *arg) 2392 { 2393 2394 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 2395 } 2396 2397 static void 2398 knlist_rw_rlock(void *arg) 2399 { 2400 2401 rw_rlock((struct rwlock *)arg); 2402 } 2403 2404 static void 2405 knlist_rw_runlock(void *arg) 2406 { 2407 2408 rw_runlock((struct rwlock *)arg); 2409 } 2410 2411 static void 2412 knlist_rw_assert_locked(void *arg) 2413 { 2414 2415 rw_assert((struct rwlock *)arg, RA_LOCKED); 2416 } 2417 2418 static void 2419 knlist_rw_assert_unlocked(void *arg) 2420 { 2421 2422 rw_assert((struct rwlock *)arg, RA_UNLOCKED); 2423 } 2424 2425 void 2426 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 2427 void (*kl_unlock)(void *), 2428 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *)) 2429 { 2430 2431 if (lock == NULL) 2432 knl->kl_lockarg = &knlist_lock; 2433 else 2434 knl->kl_lockarg = lock; 2435 2436 if (kl_lock == NULL) 2437 knl->kl_lock = knlist_mtx_lock; 2438 else 2439 knl->kl_lock = kl_lock; 2440 if (kl_unlock == NULL) 2441 knl->kl_unlock = knlist_mtx_unlock; 2442 else 2443 knl->kl_unlock = kl_unlock; 2444 if (kl_assert_locked == NULL) 2445 knl->kl_assert_locked = knlist_mtx_assert_locked; 2446 else 2447 knl->kl_assert_locked = kl_assert_locked; 2448 if (kl_assert_unlocked == NULL) 2449 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked; 2450 else 2451 knl->kl_assert_unlocked = kl_assert_unlocked; 2452 2453 knl->kl_autodestroy = 0; 2454 SLIST_INIT(&knl->kl_list); 2455 } 2456 2457 void 2458 knlist_init_mtx(struct knlist *knl, struct mtx *lock) 2459 { 2460 2461 knlist_init(knl, lock, NULL, NULL, NULL, NULL); 2462 } 2463 2464 struct knlist * 2465 knlist_alloc(struct mtx *lock) 2466 { 2467 struct knlist *knl; 2468 2469 knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK); 2470 knlist_init_mtx(knl, lock); 2471 return (knl); 2472 } 2473 2474 void 2475 knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock) 2476 { 2477 2478 knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock, 2479 knlist_rw_assert_locked, knlist_rw_assert_unlocked); 2480 } 2481 2482 void 2483 knlist_destroy(struct knlist *knl) 2484 { 2485 2486 KASSERT(KNLIST_EMPTY(knl), 2487 ("destroying knlist %p with knotes on it", knl)); 2488 } 2489 2490 void 2491 knlist_detach(struct knlist *knl) 2492 { 2493 2494 KNL_ASSERT_LOCKED(knl); 2495 knl->kl_autodestroy = 1; 2496 if (knlist_empty(knl)) { 2497 knlist_destroy(knl); 2498 free(knl, M_KQUEUE); 2499 } 2500 } 2501 2502 /* 2503 * Even if we are locked, we may need to drop the lock to allow any influx 2504 * knotes time to "settle". 2505 */ 2506 void 2507 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 2508 { 2509 struct knote *kn, *kn2; 2510 struct kqueue *kq; 2511 2512 KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl)); 2513 if (islocked) 2514 KNL_ASSERT_LOCKED(knl); 2515 else { 2516 KNL_ASSERT_UNLOCKED(knl); 2517 again: /* need to reacquire lock since we have dropped it */ 2518 knl->kl_lock(knl->kl_lockarg); 2519 } 2520 2521 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 2522 kq = kn->kn_kq; 2523 KQ_LOCK(kq); 2524 if (kn_in_flux(kn)) { 2525 KQ_UNLOCK(kq); 2526 continue; 2527 } 2528 knlist_remove_kq(knl, kn, 1, 1); 2529 if (killkn) { 2530 kn_enter_flux(kn); 2531 KQ_UNLOCK(kq); 2532 knote_drop_detached(kn, td); 2533 } else { 2534 /* Make sure cleared knotes disappear soon */ 2535 kn->kn_flags |= EV_EOF | EV_ONESHOT; 2536 KQ_UNLOCK(kq); 2537 } 2538 kq = NULL; 2539 } 2540 2541 if (!SLIST_EMPTY(&knl->kl_list)) { 2542 /* there are still in flux knotes remaining */ 2543 kn = SLIST_FIRST(&knl->kl_list); 2544 kq = kn->kn_kq; 2545 KQ_LOCK(kq); 2546 KASSERT(kn_in_flux(kn), ("knote removed w/o list lock")); 2547 knl->kl_unlock(knl->kl_lockarg); 2548 kq->kq_state |= KQ_FLUXWAIT; 2549 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 2550 kq = NULL; 2551 goto again; 2552 } 2553 2554 if (islocked) 2555 KNL_ASSERT_LOCKED(knl); 2556 else { 2557 knl->kl_unlock(knl->kl_lockarg); 2558 KNL_ASSERT_UNLOCKED(knl); 2559 } 2560 } 2561 2562 /* 2563 * Remove all knotes referencing a specified fd must be called with FILEDESC 2564 * lock. This prevents a race where a new fd comes along and occupies the 2565 * entry and we attach a knote to the fd. 2566 */ 2567 void 2568 knote_fdclose(struct thread *td, int fd) 2569 { 2570 struct filedesc *fdp = td->td_proc->p_fd; 2571 struct kqueue *kq; 2572 struct knote *kn; 2573 int influx; 2574 2575 FILEDESC_XLOCK_ASSERT(fdp); 2576 2577 /* 2578 * We shouldn't have to worry about new kevents appearing on fd 2579 * since filedesc is locked. 2580 */ 2581 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2582 KQ_LOCK(kq); 2583 2584 again: 2585 influx = 0; 2586 while (kq->kq_knlistsize > fd && 2587 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2588 if (kn_in_flux(kn)) { 2589 /* someone else might be waiting on our knote */ 2590 if (influx) 2591 wakeup(kq); 2592 kq->kq_state |= KQ_FLUXWAIT; 2593 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2594 goto again; 2595 } 2596 kn_enter_flux(kn); 2597 KQ_UNLOCK(kq); 2598 influx = 1; 2599 knote_drop(kn, td); 2600 KQ_LOCK(kq); 2601 } 2602 KQ_UNLOCK_FLUX(kq); 2603 } 2604 } 2605 2606 static int 2607 knote_attach(struct knote *kn, struct kqueue *kq) 2608 { 2609 struct klist *list; 2610 2611 KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn)); 2612 KQ_OWNED(kq); 2613 2614 if ((kq->kq_state & KQ_CLOSING) != 0) 2615 return (EBADF); 2616 if (kn->kn_fop->f_isfd) { 2617 if (kn->kn_id >= kq->kq_knlistsize) 2618 return (ENOMEM); 2619 list = &kq->kq_knlist[kn->kn_id]; 2620 } else { 2621 if (kq->kq_knhash == NULL) 2622 return (ENOMEM); 2623 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2624 } 2625 SLIST_INSERT_HEAD(list, kn, kn_link); 2626 return (0); 2627 } 2628 2629 static void 2630 knote_drop(struct knote *kn, struct thread *td) 2631 { 2632 2633 if ((kn->kn_status & KN_DETACHED) == 0) 2634 kn->kn_fop->f_detach(kn); 2635 knote_drop_detached(kn, td); 2636 } 2637 2638 static void 2639 knote_drop_detached(struct knote *kn, struct thread *td) 2640 { 2641 struct kqueue *kq; 2642 struct klist *list; 2643 2644 kq = kn->kn_kq; 2645 2646 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2647 ("knote %p still attached", kn)); 2648 KQ_NOTOWNED(kq); 2649 2650 KQ_LOCK(kq); 2651 KASSERT(kn->kn_influx == 1, 2652 ("knote_drop called on %p with influx %d", kn, kn->kn_influx)); 2653 2654 if (kn->kn_fop->f_isfd) 2655 list = &kq->kq_knlist[kn->kn_id]; 2656 else 2657 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2658 2659 if (!SLIST_EMPTY(list)) 2660 SLIST_REMOVE(list, kn, knote, kn_link); 2661 if (kn->kn_status & KN_QUEUED) 2662 knote_dequeue(kn); 2663 KQ_UNLOCK_FLUX(kq); 2664 2665 if (kn->kn_fop->f_isfd) { 2666 fdrop(kn->kn_fp, td); 2667 kn->kn_fp = NULL; 2668 } 2669 kqueue_fo_release(kn->kn_kevent.filter); 2670 kn->kn_fop = NULL; 2671 knote_free(kn); 2672 } 2673 2674 static void 2675 knote_enqueue(struct knote *kn) 2676 { 2677 struct kqueue *kq = kn->kn_kq; 2678 2679 KQ_OWNED(kn->kn_kq); 2680 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2681 2682 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2683 kn->kn_status |= KN_QUEUED; 2684 kq->kq_count++; 2685 kqueue_wakeup(kq); 2686 } 2687 2688 static void 2689 knote_dequeue(struct knote *kn) 2690 { 2691 struct kqueue *kq = kn->kn_kq; 2692 2693 KQ_OWNED(kn->kn_kq); 2694 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2695 2696 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2697 kn->kn_status &= ~KN_QUEUED; 2698 kq->kq_count--; 2699 } 2700 2701 static void 2702 knote_init(void) 2703 { 2704 2705 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2706 NULL, NULL, UMA_ALIGN_PTR, 0); 2707 } 2708 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2709 2710 static struct knote * 2711 knote_alloc(int mflag) 2712 { 2713 2714 return (uma_zalloc(knote_zone, mflag | M_ZERO)); 2715 } 2716 2717 static void 2718 knote_free(struct knote *kn) 2719 { 2720 2721 uma_zfree(knote_zone, kn); 2722 } 2723 2724 /* 2725 * Register the kev w/ the kq specified by fd. 2726 */ 2727 int 2728 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag) 2729 { 2730 struct kqueue *kq; 2731 struct file *fp; 2732 cap_rights_t rights; 2733 int error; 2734 2735 error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp); 2736 if (error != 0) 2737 return (error); 2738 if ((error = kqueue_acquire(fp, &kq)) != 0) 2739 goto noacquire; 2740 2741 error = kqueue_register(kq, kev, td, mflag); 2742 kqueue_release(kq, 0); 2743 2744 noacquire: 2745 fdrop(fp, td); 2746 return (error); 2747 }
Cache object: f05712d7a82e33c01e2ee80be49f2697
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]