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 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 3 * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include "opt_ktrace.h" 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/kernel.h> 36 #include <sys/lock.h> 37 #include <sys/mutex.h> 38 #include <sys/proc.h> 39 #include <sys/malloc.h> 40 #include <sys/unistd.h> 41 #include <sys/file.h> 42 #include <sys/filedesc.h> 43 #include <sys/filio.h> 44 #include <sys/fcntl.h> 45 #include <sys/kthread.h> 46 #include <sys/selinfo.h> 47 #include <sys/queue.h> 48 #include <sys/event.h> 49 #include <sys/eventvar.h> 50 #include <sys/poll.h> 51 #include <sys/protosw.h> 52 #include <sys/sigio.h> 53 #include <sys/signalvar.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/stat.h> 57 #include <sys/sysctl.h> 58 #include <sys/sysproto.h> 59 #include <sys/syscallsubr.h> 60 #include <sys/taskqueue.h> 61 #include <sys/uio.h> 62 #ifdef KTRACE 63 #include <sys/ktrace.h> 64 #endif 65 66 #include <vm/uma.h> 67 68 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 69 70 /* 71 * This lock is used if multiple kq locks are required. This possibly 72 * should be made into a per proc lock. 73 */ 74 static struct mtx kq_global; 75 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 76 #define KQ_GLOBAL_LOCK(lck, haslck) do { \ 77 if (!haslck) \ 78 mtx_lock(lck); \ 79 haslck = 1; \ 80 } while (0) 81 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 82 if (haslck) \ 83 mtx_unlock(lck); \ 84 haslck = 0; \ 85 } while (0) 86 87 TASKQUEUE_DEFINE_THREAD(kqueue); 88 89 static int kevent_copyout(void *arg, struct kevent *kevp, int count); 90 static int kevent_copyin(void *arg, struct kevent *kevp, int count); 91 static int kqueue_register(struct kqueue *kq, struct kevent *kev, 92 struct thread *td, int waitok); 93 static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 94 static void kqueue_release(struct kqueue *kq, int locked); 95 static int kqueue_expand(struct kqueue *kq, struct filterops *fops, 96 uintptr_t ident, int waitok); 97 static void kqueue_task(void *arg, int pending); 98 static int kqueue_scan(struct kqueue *kq, int maxevents, 99 struct kevent_copyops *k_ops, 100 const struct timespec *timeout, 101 struct kevent *keva, struct thread *td); 102 static void kqueue_wakeup(struct kqueue *kq); 103 static struct filterops *kqueue_fo_find(int filt); 104 static void kqueue_fo_release(int filt); 105 106 static fo_rdwr_t kqueue_read; 107 static fo_rdwr_t kqueue_write; 108 static fo_ioctl_t kqueue_ioctl; 109 static fo_poll_t kqueue_poll; 110 static fo_kqfilter_t kqueue_kqfilter; 111 static fo_stat_t kqueue_stat; 112 static fo_close_t kqueue_close; 113 114 static struct fileops kqueueops = { 115 .fo_read = kqueue_read, 116 .fo_write = kqueue_write, 117 .fo_ioctl = kqueue_ioctl, 118 .fo_poll = kqueue_poll, 119 .fo_kqfilter = kqueue_kqfilter, 120 .fo_stat = kqueue_stat, 121 .fo_close = kqueue_close, 122 }; 123 124 static int knote_attach(struct knote *kn, struct kqueue *kq); 125 static void knote_drop(struct knote *kn, struct thread *td); 126 static void knote_enqueue(struct knote *kn); 127 static void knote_dequeue(struct knote *kn); 128 static void knote_init(void); 129 static struct knote *knote_alloc(int waitok); 130 static void knote_free(struct knote *kn); 131 132 static void filt_kqdetach(struct knote *kn); 133 static int filt_kqueue(struct knote *kn, long hint); 134 static int filt_procattach(struct knote *kn); 135 static void filt_procdetach(struct knote *kn); 136 static int filt_proc(struct knote *kn, long hint); 137 static int filt_fileattach(struct knote *kn); 138 static void filt_timerexpire(void *knx); 139 static int filt_timerattach(struct knote *kn); 140 static void filt_timerdetach(struct knote *kn); 141 static int filt_timer(struct knote *kn, long hint); 142 143 static struct filterops file_filtops = 144 { 1, filt_fileattach, NULL, NULL }; 145 static struct filterops kqread_filtops = 146 { 1, NULL, filt_kqdetach, filt_kqueue }; 147 /* XXX - move to kern_proc.c? */ 148 static struct filterops proc_filtops = 149 { 0, filt_procattach, filt_procdetach, filt_proc }; 150 static struct filterops timer_filtops = 151 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 152 153 static uma_zone_t knote_zone; 154 static int kq_ncallouts = 0; 155 static int kq_calloutmax = (4 * 1024); 156 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 157 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 158 159 /* XXX - ensure not KN_INFLUX?? */ 160 #define KNOTE_ACTIVATE(kn, islock) do { \ 161 if ((islock)) \ 162 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 163 else \ 164 KQ_LOCK((kn)->kn_kq); \ 165 (kn)->kn_status |= KN_ACTIVE; \ 166 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 167 knote_enqueue((kn)); \ 168 if (!(islock)) \ 169 KQ_UNLOCK((kn)->kn_kq); \ 170 } while(0) 171 #define KQ_LOCK(kq) do { \ 172 mtx_lock(&(kq)->kq_lock); \ 173 } while (0) 174 #define KQ_FLUX_WAKEUP(kq) do { \ 175 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 176 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 177 wakeup((kq)); \ 178 } \ 179 } while (0) 180 #define KQ_UNLOCK_FLUX(kq) do { \ 181 KQ_FLUX_WAKEUP(kq); \ 182 mtx_unlock(&(kq)->kq_lock); \ 183 } while (0) 184 #define KQ_UNLOCK(kq) do { \ 185 mtx_unlock(&(kq)->kq_lock); \ 186 } while (0) 187 #define KQ_OWNED(kq) do { \ 188 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 189 } while (0) 190 #define KQ_NOTOWNED(kq) do { \ 191 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 192 } while (0) 193 #define KN_LIST_LOCK(kn) do { \ 194 if (kn->kn_knlist != NULL) \ 195 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \ 196 } while (0) 197 #define KN_LIST_UNLOCK(kn) do { \ 198 if (kn->kn_knlist != NULL) \ 199 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \ 200 } while (0) 201 #define KNL_ASSERT_LOCK(knl, islocked) do { \ 202 if (islocked) \ 203 KNL_ASSERT_LOCKED(knl); \ 204 else \ 205 KNL_ASSERT_UNLOCKED(knl); \ 206 } while (0) 207 #ifdef INVARIANTS 208 #define KNL_ASSERT_LOCKED(knl) do { \ 209 if (!knl->kl_locked((knl)->kl_lockarg)) \ 210 panic("knlist not locked, but should be"); \ 211 } while (0) 212 #define KNL_ASSERT_UNLOCKED(knl) do { \ 213 if (knl->kl_locked((knl)->kl_lockarg)) \ 214 panic("knlist locked, but should not be"); \ 215 } while (0) 216 #else /* !INVARIANTS */ 217 #define KNL_ASSERT_LOCKED(knl) do {} while(0) 218 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 219 #endif /* INVARIANTS */ 220 221 #define KN_HASHSIZE 64 /* XXX should be tunable */ 222 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 223 224 static int 225 filt_nullattach(struct knote *kn) 226 { 227 228 return (ENXIO); 229 }; 230 231 struct filterops null_filtops = 232 { 0, filt_nullattach, NULL, NULL }; 233 234 /* XXX - make SYSINIT to add these, and move into respective modules. */ 235 extern struct filterops sig_filtops; 236 extern struct filterops fs_filtops; 237 238 /* 239 * Table for for all system-defined filters. 240 */ 241 static struct mtx filterops_lock; 242 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", 243 MTX_DEF); 244 static struct { 245 struct filterops *for_fop; 246 int for_refcnt; 247 } sysfilt_ops[EVFILT_SYSCOUNT] = { 248 { &file_filtops }, /* EVFILT_READ */ 249 { &file_filtops }, /* EVFILT_WRITE */ 250 { &null_filtops }, /* EVFILT_AIO */ 251 { &file_filtops }, /* EVFILT_VNODE */ 252 { &proc_filtops }, /* EVFILT_PROC */ 253 { &sig_filtops }, /* EVFILT_SIGNAL */ 254 { &timer_filtops }, /* EVFILT_TIMER */ 255 { &file_filtops }, /* EVFILT_NETDEV */ 256 { &fs_filtops }, /* EVFILT_FS */ 257 { &null_filtops }, /* EVFILT_LIO */ 258 }; 259 260 /* 261 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 262 * method. 263 */ 264 static int 265 filt_fileattach(struct knote *kn) 266 { 267 268 return (fo_kqfilter(kn->kn_fp, kn)); 269 } 270 271 /*ARGSUSED*/ 272 static int 273 kqueue_kqfilter(struct file *fp, struct knote *kn) 274 { 275 struct kqueue *kq = kn->kn_fp->f_data; 276 277 if (kn->kn_filter != EVFILT_READ) 278 return (EINVAL); 279 280 kn->kn_status |= KN_KQUEUE; 281 kn->kn_fop = &kqread_filtops; 282 knlist_add(&kq->kq_sel.si_note, kn, 0); 283 284 return (0); 285 } 286 287 static void 288 filt_kqdetach(struct knote *kn) 289 { 290 struct kqueue *kq = kn->kn_fp->f_data; 291 292 knlist_remove(&kq->kq_sel.si_note, kn, 0); 293 } 294 295 /*ARGSUSED*/ 296 static int 297 filt_kqueue(struct knote *kn, long hint) 298 { 299 struct kqueue *kq = kn->kn_fp->f_data; 300 301 kn->kn_data = kq->kq_count; 302 return (kn->kn_data > 0); 303 } 304 305 /* XXX - move to kern_proc.c? */ 306 static int 307 filt_procattach(struct knote *kn) 308 { 309 struct proc *p; 310 int immediate; 311 int error; 312 313 immediate = 0; 314 p = pfind(kn->kn_id); 315 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 316 p = zpfind(kn->kn_id); 317 immediate = 1; 318 } else if (p != NULL && (p->p_flag & P_WEXIT)) { 319 immediate = 1; 320 } 321 322 if (p == NULL) 323 return (ESRCH); 324 if ((error = p_cansee(curthread, p))) 325 return (error); 326 327 kn->kn_ptr.p_proc = p; 328 kn->kn_flags |= EV_CLEAR; /* automatically set */ 329 330 /* 331 * internal flag indicating registration done by kernel 332 */ 333 if (kn->kn_flags & EV_FLAG1) { 334 kn->kn_data = kn->kn_sdata; /* ppid */ 335 kn->kn_fflags = NOTE_CHILD; 336 kn->kn_flags &= ~EV_FLAG1; 337 } 338 339 if (immediate == 0) 340 knlist_add(&p->p_klist, kn, 1); 341 342 /* 343 * Immediately activate any exit notes if the target process is a 344 * zombie. This is necessary to handle the case where the target 345 * process, e.g. a child, dies before the kevent is registered. 346 */ 347 if (immediate && filt_proc(kn, NOTE_EXIT)) 348 KNOTE_ACTIVATE(kn, 0); 349 350 PROC_UNLOCK(p); 351 352 return (0); 353 } 354 355 /* 356 * The knote may be attached to a different process, which may exit, 357 * leaving nothing for the knote to be attached to. So when the process 358 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 359 * it will be deleted when read out. However, as part of the knote deletion, 360 * this routine is called, so a check is needed to avoid actually performing 361 * a detach, because the original process does not exist any more. 362 */ 363 /* XXX - move to kern_proc.c? */ 364 static void 365 filt_procdetach(struct knote *kn) 366 { 367 struct proc *p; 368 369 p = kn->kn_ptr.p_proc; 370 knlist_remove(&p->p_klist, kn, 0); 371 kn->kn_ptr.p_proc = NULL; 372 } 373 374 /* XXX - move to kern_proc.c? */ 375 static int 376 filt_proc(struct knote *kn, long hint) 377 { 378 struct proc *p = kn->kn_ptr.p_proc; 379 u_int event; 380 381 /* 382 * mask off extra data 383 */ 384 event = (u_int)hint & NOTE_PCTRLMASK; 385 386 /* 387 * if the user is interested in this event, record it. 388 */ 389 if (kn->kn_sfflags & event) 390 kn->kn_fflags |= event; 391 392 /* 393 * process is gone, so flag the event as finished. 394 */ 395 if (event == NOTE_EXIT) { 396 if (!(kn->kn_status & KN_DETACHED)) 397 knlist_remove_inevent(&p->p_klist, kn); 398 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 399 kn->kn_data = p->p_xstat; 400 kn->kn_ptr.p_proc = NULL; 401 return (1); 402 } 403 404 return (kn->kn_fflags != 0); 405 } 406 407 /* 408 * Called when the process forked. It mostly does the same as the 409 * knote(), activating all knotes registered to be activated when the 410 * process forked. Additionally, for each knote attached to the 411 * parent, check whether user wants to track the new process. If so 412 * attach a new knote to it, and immediately report an event with the 413 * child's pid. 414 */ 415 void 416 knote_fork(struct knlist *list, int pid) 417 { 418 struct kqueue *kq; 419 struct knote *kn; 420 struct kevent kev; 421 int error; 422 423 if (list == NULL) 424 return; 425 list->kl_lock(list->kl_lockarg); 426 427 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 428 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) 429 continue; 430 kq = kn->kn_kq; 431 KQ_LOCK(kq); 432 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 433 KQ_UNLOCK(kq); 434 continue; 435 } 436 437 /* 438 * The same as knote(), activate the event. 439 */ 440 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 441 kn->kn_status |= KN_HASKQLOCK; 442 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid)) 443 KNOTE_ACTIVATE(kn, 1); 444 kn->kn_status &= ~KN_HASKQLOCK; 445 KQ_UNLOCK(kq); 446 continue; 447 } 448 449 /* 450 * The NOTE_TRACK case. In addition to the activation 451 * of the event, we need to register new event to 452 * track the child. Drop the locks in preparation for 453 * the call to kqueue_register(). 454 */ 455 kn->kn_status |= KN_INFLUX; 456 KQ_UNLOCK(kq); 457 list->kl_unlock(list->kl_lockarg); 458 459 /* 460 * Activate existing knote and register a knote with 461 * new process. 462 */ 463 kev.ident = pid; 464 kev.filter = kn->kn_filter; 465 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 466 kev.fflags = kn->kn_sfflags; 467 kev.data = kn->kn_id; /* parent */ 468 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 469 error = kqueue_register(kq, &kev, NULL, 0); 470 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid)) 471 KNOTE_ACTIVATE(kn, 0); 472 if (error) 473 kn->kn_fflags |= NOTE_TRACKERR; 474 KQ_LOCK(kq); 475 kn->kn_status &= ~KN_INFLUX; 476 KQ_UNLOCK_FLUX(kq); 477 list->kl_lock(list->kl_lockarg); 478 } 479 list->kl_unlock(list->kl_lockarg); 480 } 481 482 static int 483 timertoticks(intptr_t data) 484 { 485 struct timeval tv; 486 int tticks; 487 488 tv.tv_sec = data / 1000; 489 tv.tv_usec = (data % 1000) * 1000; 490 tticks = tvtohz(&tv); 491 492 return tticks; 493 } 494 495 /* XXX - move to kern_timeout.c? */ 496 static void 497 filt_timerexpire(void *knx) 498 { 499 struct knote *kn = knx; 500 struct callout *calloutp; 501 502 kn->kn_data++; 503 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 504 505 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) { 506 calloutp = (struct callout *)kn->kn_hook; 507 callout_reset(calloutp, timertoticks(kn->kn_sdata), 508 filt_timerexpire, kn); 509 } 510 } 511 512 /* 513 * data contains amount of time to sleep, in milliseconds 514 */ 515 /* XXX - move to kern_timeout.c? */ 516 static int 517 filt_timerattach(struct knote *kn) 518 { 519 struct callout *calloutp; 520 521 atomic_add_int(&kq_ncallouts, 1); 522 523 if (kq_ncallouts >= kq_calloutmax) { 524 atomic_add_int(&kq_ncallouts, -1); 525 return (ENOMEM); 526 } 527 528 kn->kn_flags |= EV_CLEAR; /* automatically set */ 529 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */ 530 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 531 M_KQUEUE, M_WAITOK); 532 callout_init(calloutp, CALLOUT_MPSAFE); 533 kn->kn_hook = calloutp; 534 callout_reset(calloutp, timertoticks(kn->kn_sdata), filt_timerexpire, 535 kn); 536 537 return (0); 538 } 539 540 /* XXX - move to kern_timeout.c? */ 541 static void 542 filt_timerdetach(struct knote *kn) 543 { 544 struct callout *calloutp; 545 546 calloutp = (struct callout *)kn->kn_hook; 547 callout_drain(calloutp); 548 FREE(calloutp, M_KQUEUE); 549 atomic_add_int(&kq_ncallouts, -1); 550 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */ 551 } 552 553 /* XXX - move to kern_timeout.c? */ 554 static int 555 filt_timer(struct knote *kn, long hint) 556 { 557 558 return (kn->kn_data != 0); 559 } 560 561 int 562 kqueue(struct thread *td, struct kqueue_args *uap) 563 { 564 struct filedesc *fdp; 565 struct kqueue *kq; 566 struct file *fp; 567 int fd, error; 568 569 fdp = td->td_proc->p_fd; 570 error = falloc(td, &fp, &fd); 571 if (error) 572 goto done2; 573 574 /* An extra reference on `nfp' has been held for us by falloc(). */ 575 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 576 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK); 577 TAILQ_INIT(&kq->kq_head); 578 kq->kq_fdp = fdp; 579 knlist_init(&kq->kq_sel.si_note, &kq->kq_lock, NULL, NULL, NULL); 580 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 581 582 FILEDESC_XLOCK(fdp); 583 SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 584 FILEDESC_XUNLOCK(fdp); 585 586 FILE_LOCK(fp); 587 fp->f_flag = FREAD | FWRITE; 588 fp->f_type = DTYPE_KQUEUE; 589 fp->f_data = kq; 590 fp->f_ops = &kqueueops; 591 FILE_UNLOCK(fp); 592 fdrop(fp, td); 593 594 td->td_retval[0] = fd; 595 done2: 596 return (error); 597 } 598 599 #ifndef _SYS_SYSPROTO_H_ 600 struct kevent_args { 601 int fd; 602 const struct kevent *changelist; 603 int nchanges; 604 struct kevent *eventlist; 605 int nevents; 606 const struct timespec *timeout; 607 }; 608 #endif 609 int 610 kevent(struct thread *td, struct kevent_args *uap) 611 { 612 struct timespec ts, *tsp; 613 struct kevent_copyops k_ops = { uap, 614 kevent_copyout, 615 kevent_copyin}; 616 int error; 617 #ifdef KTRACE 618 struct uio ktruio; 619 struct iovec ktriov; 620 struct uio *ktruioin = NULL; 621 struct uio *ktruioout = NULL; 622 #endif 623 624 if (uap->timeout != NULL) { 625 error = copyin(uap->timeout, &ts, sizeof(ts)); 626 if (error) 627 return (error); 628 tsp = &ts; 629 } else 630 tsp = NULL; 631 632 #ifdef KTRACE 633 if (KTRPOINT(td, KTR_GENIO)) { 634 ktriov.iov_base = uap->changelist; 635 ktriov.iov_len = uap->nchanges * sizeof(struct kevent); 636 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1, 637 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ, 638 .uio_td = td }; 639 ktruioin = cloneuio(&ktruio); 640 ktriov.iov_base = uap->eventlist; 641 ktriov.iov_len = uap->nevents * sizeof(struct kevent); 642 ktruioout = cloneuio(&ktruio); 643 } 644 #endif 645 646 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 647 &k_ops, tsp); 648 649 #ifdef KTRACE 650 if (ktruioin != NULL) { 651 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent); 652 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0); 653 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent); 654 ktrgenio(uap->fd, UIO_READ, ktruioout, error); 655 } 656 #endif 657 658 return (error); 659 } 660 661 /* 662 * Copy 'count' items into the destination list pointed to by uap->eventlist. 663 */ 664 static int 665 kevent_copyout(void *arg, struct kevent *kevp, int count) 666 { 667 struct kevent_args *uap; 668 int error; 669 670 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 671 uap = (struct kevent_args *)arg; 672 673 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 674 if (error == 0) 675 uap->eventlist += count; 676 return (error); 677 } 678 679 /* 680 * Copy 'count' items from the list pointed to by uap->changelist. 681 */ 682 static int 683 kevent_copyin(void *arg, struct kevent *kevp, int count) 684 { 685 struct kevent_args *uap; 686 int error; 687 688 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 689 uap = (struct kevent_args *)arg; 690 691 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 692 if (error == 0) 693 uap->changelist += count; 694 return (error); 695 } 696 697 int 698 kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 699 struct kevent_copyops *k_ops, const struct timespec *timeout) 700 { 701 struct kevent keva[KQ_NEVENTS]; 702 struct kevent *kevp, *changes; 703 struct kqueue *kq; 704 struct file *fp; 705 int i, n, nerrors, error; 706 707 if ((error = fget(td, fd, &fp)) != 0) 708 return (error); 709 if ((error = kqueue_acquire(fp, &kq)) != 0) 710 goto done_norel; 711 712 nerrors = 0; 713 714 while (nchanges > 0) { 715 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 716 error = k_ops->k_copyin(k_ops->arg, keva, n); 717 if (error) 718 goto done; 719 changes = keva; 720 for (i = 0; i < n; i++) { 721 kevp = &changes[i]; 722 if (!kevp->filter) 723 continue; 724 kevp->flags &= ~EV_SYSFLAGS; 725 error = kqueue_register(kq, kevp, td, 1); 726 if (error) { 727 if (nevents != 0) { 728 kevp->flags = EV_ERROR; 729 kevp->data = error; 730 (void) k_ops->k_copyout(k_ops->arg, 731 kevp, 1); 732 nevents--; 733 nerrors++; 734 } else { 735 goto done; 736 } 737 } 738 } 739 nchanges -= n; 740 } 741 if (nerrors) { 742 td->td_retval[0] = nerrors; 743 error = 0; 744 goto done; 745 } 746 747 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td); 748 done: 749 kqueue_release(kq, 0); 750 done_norel: 751 fdrop(fp, td); 752 return (error); 753 } 754 755 int 756 kqueue_add_filteropts(int filt, struct filterops *filtops) 757 { 758 int error; 759 760 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 761 printf( 762 "trying to add a filterop that is out of range: %d is beyond %d\n", 763 ~filt, EVFILT_SYSCOUNT); 764 return EINVAL; 765 } 766 mtx_lock(&filterops_lock); 767 if (sysfilt_ops[~filt].for_fop != &null_filtops && 768 sysfilt_ops[~filt].for_fop != NULL) 769 error = EEXIST; 770 else { 771 sysfilt_ops[~filt].for_fop = filtops; 772 sysfilt_ops[~filt].for_refcnt = 0; 773 } 774 mtx_unlock(&filterops_lock); 775 776 return (0); 777 } 778 779 int 780 kqueue_del_filteropts(int filt) 781 { 782 int error; 783 784 error = 0; 785 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 786 return EINVAL; 787 788 mtx_lock(&filterops_lock); 789 if (sysfilt_ops[~filt].for_fop == &null_filtops || 790 sysfilt_ops[~filt].for_fop == NULL) 791 error = EINVAL; 792 else if (sysfilt_ops[~filt].for_refcnt != 0) 793 error = EBUSY; 794 else { 795 sysfilt_ops[~filt].for_fop = &null_filtops; 796 sysfilt_ops[~filt].for_refcnt = 0; 797 } 798 mtx_unlock(&filterops_lock); 799 800 return error; 801 } 802 803 static struct filterops * 804 kqueue_fo_find(int filt) 805 { 806 807 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 808 return NULL; 809 810 mtx_lock(&filterops_lock); 811 sysfilt_ops[~filt].for_refcnt++; 812 if (sysfilt_ops[~filt].for_fop == NULL) 813 sysfilt_ops[~filt].for_fop = &null_filtops; 814 mtx_unlock(&filterops_lock); 815 816 return sysfilt_ops[~filt].for_fop; 817 } 818 819 static void 820 kqueue_fo_release(int filt) 821 { 822 823 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 824 return; 825 826 mtx_lock(&filterops_lock); 827 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 828 ("filter object refcount not valid on release")); 829 sysfilt_ops[~filt].for_refcnt--; 830 mtx_unlock(&filterops_lock); 831 } 832 833 /* 834 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will 835 * influence if memory allocation should wait. Make sure it is 0 if you 836 * hold any mutexes. 837 */ 838 static int 839 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok) 840 { 841 struct filterops *fops; 842 struct file *fp; 843 struct knote *kn, *tkn; 844 int error, filt, event; 845 int haskqglobal; 846 847 fp = NULL; 848 kn = NULL; 849 error = 0; 850 haskqglobal = 0; 851 852 filt = kev->filter; 853 fops = kqueue_fo_find(filt); 854 if (fops == NULL) 855 return EINVAL; 856 857 tkn = knote_alloc(waitok); /* prevent waiting with locks */ 858 859 findkn: 860 if (fops->f_isfd) { 861 KASSERT(td != NULL, ("td is NULL")); 862 error = fget(td, kev->ident, &fp); 863 if (error) 864 goto done; 865 866 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 867 kev->ident, 0) != 0) { 868 /* try again */ 869 fdrop(fp, td); 870 fp = NULL; 871 error = kqueue_expand(kq, fops, kev->ident, waitok); 872 if (error) 873 goto done; 874 goto findkn; 875 } 876 877 if (fp->f_type == DTYPE_KQUEUE) { 878 /* 879 * if we add some inteligence about what we are doing, 880 * we should be able to support events on ourselves. 881 * We need to know when we are doing this to prevent 882 * getting both the knlist lock and the kq lock since 883 * they are the same thing. 884 */ 885 if (fp->f_data == kq) { 886 error = EINVAL; 887 goto done; 888 } 889 890 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 891 } 892 893 KQ_LOCK(kq); 894 if (kev->ident < kq->kq_knlistsize) { 895 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 896 if (kev->filter == kn->kn_filter) 897 break; 898 } 899 } else { 900 if ((kev->flags & EV_ADD) == EV_ADD) 901 kqueue_expand(kq, fops, kev->ident, waitok); 902 903 KQ_LOCK(kq); 904 if (kq->kq_knhashmask != 0) { 905 struct klist *list; 906 907 list = &kq->kq_knhash[ 908 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 909 SLIST_FOREACH(kn, list, kn_link) 910 if (kev->ident == kn->kn_id && 911 kev->filter == kn->kn_filter) 912 break; 913 } 914 } 915 916 /* knote is in the process of changing, wait for it to stablize. */ 917 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 918 if (fp != NULL) { 919 fdrop(fp, td); 920 fp = NULL; 921 } 922 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 923 kq->kq_state |= KQ_FLUXWAIT; 924 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 925 goto findkn; 926 } 927 928 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 929 KQ_UNLOCK(kq); 930 error = ENOENT; 931 goto done; 932 } 933 934 /* 935 * kn now contains the matching knote, or NULL if no match 936 */ 937 if (kev->flags & EV_ADD) { 938 if (kn == NULL) { 939 kn = tkn; 940 tkn = NULL; 941 if (kn == NULL) { 942 KQ_UNLOCK(kq); 943 error = ENOMEM; 944 goto done; 945 } 946 kn->kn_fp = fp; 947 kn->kn_kq = kq; 948 kn->kn_fop = fops; 949 /* 950 * apply reference counts to knote structure, and 951 * do not release it at the end of this routine. 952 */ 953 fops = NULL; 954 fp = NULL; 955 956 kn->kn_sfflags = kev->fflags; 957 kn->kn_sdata = kev->data; 958 kev->fflags = 0; 959 kev->data = 0; 960 kn->kn_kevent = *kev; 961 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 962 EV_ENABLE | EV_DISABLE); 963 kn->kn_status = KN_INFLUX|KN_DETACHED; 964 965 error = knote_attach(kn, kq); 966 KQ_UNLOCK(kq); 967 if (error != 0) { 968 tkn = kn; 969 goto done; 970 } 971 972 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 973 knote_drop(kn, td); 974 goto done; 975 } 976 KN_LIST_LOCK(kn); 977 } else { 978 /* 979 * The user may change some filter values after the 980 * initial EV_ADD, but doing so will not reset any 981 * filter which has already been triggered. 982 */ 983 kn->kn_status |= KN_INFLUX; 984 KQ_UNLOCK(kq); 985 KN_LIST_LOCK(kn); 986 kn->kn_sfflags = kev->fflags; 987 kn->kn_sdata = kev->data; 988 kn->kn_kevent.udata = kev->udata; 989 } 990 991 /* 992 * We can get here with kn->kn_knlist == NULL. 993 * This can happen when the initial attach event decides that 994 * the event is "completed" already. i.e. filt_procattach 995 * is called on a zombie process. It will call filt_proc 996 * which will remove it from the list, and NULL kn_knlist. 997 */ 998 event = kn->kn_fop->f_event(kn, 0); 999 KQ_LOCK(kq); 1000 if (event) 1001 KNOTE_ACTIVATE(kn, 1); 1002 kn->kn_status &= ~KN_INFLUX; 1003 KN_LIST_UNLOCK(kn); 1004 } else if (kev->flags & EV_DELETE) { 1005 kn->kn_status |= KN_INFLUX; 1006 KQ_UNLOCK(kq); 1007 if (!(kn->kn_status & KN_DETACHED)) 1008 kn->kn_fop->f_detach(kn); 1009 knote_drop(kn, td); 1010 goto done; 1011 } 1012 1013 if ((kev->flags & EV_DISABLE) && 1014 ((kn->kn_status & KN_DISABLED) == 0)) { 1015 kn->kn_status |= KN_DISABLED; 1016 } 1017 1018 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 1019 kn->kn_status &= ~KN_DISABLED; 1020 if ((kn->kn_status & KN_ACTIVE) && 1021 ((kn->kn_status & KN_QUEUED) == 0)) 1022 knote_enqueue(kn); 1023 } 1024 KQ_UNLOCK_FLUX(kq); 1025 1026 done: 1027 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1028 if (fp != NULL) 1029 fdrop(fp, td); 1030 if (tkn != NULL) 1031 knote_free(tkn); 1032 if (fops != NULL) 1033 kqueue_fo_release(filt); 1034 return (error); 1035 } 1036 1037 static int 1038 kqueue_acquire(struct file *fp, struct kqueue **kqp) 1039 { 1040 int error; 1041 struct kqueue *kq; 1042 1043 error = 0; 1044 1045 FILE_LOCK(fp); 1046 do { 1047 kq = fp->f_data; 1048 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) { 1049 error = EBADF; 1050 break; 1051 } 1052 *kqp = kq; 1053 KQ_LOCK(kq); 1054 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1055 KQ_UNLOCK(kq); 1056 error = EBADF; 1057 break; 1058 } 1059 kq->kq_refcnt++; 1060 KQ_UNLOCK(kq); 1061 } while (0); 1062 FILE_UNLOCK(fp); 1063 1064 return error; 1065 } 1066 1067 static void 1068 kqueue_release(struct kqueue *kq, int locked) 1069 { 1070 if (locked) 1071 KQ_OWNED(kq); 1072 else 1073 KQ_LOCK(kq); 1074 kq->kq_refcnt--; 1075 if (kq->kq_refcnt == 1) 1076 wakeup(&kq->kq_refcnt); 1077 if (!locked) 1078 KQ_UNLOCK(kq); 1079 } 1080 1081 static void 1082 kqueue_schedtask(struct kqueue *kq) 1083 { 1084 1085 KQ_OWNED(kq); 1086 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1087 ("scheduling kqueue task while draining")); 1088 1089 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1090 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task); 1091 kq->kq_state |= KQ_TASKSCHED; 1092 } 1093 } 1094 1095 /* 1096 * Expand the kq to make sure we have storage for fops/ident pair. 1097 * 1098 * Return 0 on success (or no work necessary), return errno on failure. 1099 * 1100 * Not calling hashinit w/ waitok (proper malloc flag) should be safe. 1101 * If kqueue_register is called from a non-fd context, there usually/should 1102 * be no locks held. 1103 */ 1104 static int 1105 kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, 1106 int waitok) 1107 { 1108 struct klist *list, *tmp_knhash; 1109 u_long tmp_knhashmask; 1110 int size; 1111 int fd; 1112 int mflag = waitok ? M_WAITOK : M_NOWAIT; 1113 1114 KQ_NOTOWNED(kq); 1115 1116 if (fops->f_isfd) { 1117 fd = ident; 1118 if (kq->kq_knlistsize <= fd) { 1119 size = kq->kq_knlistsize; 1120 while (size <= fd) 1121 size += KQEXTENT; 1122 MALLOC(list, struct klist *, 1123 size * sizeof list, M_KQUEUE, mflag); 1124 if (list == NULL) 1125 return ENOMEM; 1126 KQ_LOCK(kq); 1127 if (kq->kq_knlistsize > fd) { 1128 FREE(list, M_KQUEUE); 1129 list = NULL; 1130 } else { 1131 if (kq->kq_knlist != NULL) { 1132 bcopy(kq->kq_knlist, list, 1133 kq->kq_knlistsize * sizeof list); 1134 FREE(kq->kq_knlist, M_KQUEUE); 1135 kq->kq_knlist = NULL; 1136 } 1137 bzero((caddr_t)list + 1138 kq->kq_knlistsize * sizeof list, 1139 (size - kq->kq_knlistsize) * sizeof list); 1140 kq->kq_knlistsize = size; 1141 kq->kq_knlist = list; 1142 } 1143 KQ_UNLOCK(kq); 1144 } 1145 } else { 1146 if (kq->kq_knhashmask == 0) { 1147 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1148 &tmp_knhashmask); 1149 if (tmp_knhash == NULL) 1150 return ENOMEM; 1151 KQ_LOCK(kq); 1152 if (kq->kq_knhashmask == 0) { 1153 kq->kq_knhash = tmp_knhash; 1154 kq->kq_knhashmask = tmp_knhashmask; 1155 } else { 1156 free(tmp_knhash, M_KQUEUE); 1157 } 1158 KQ_UNLOCK(kq); 1159 } 1160 } 1161 1162 KQ_NOTOWNED(kq); 1163 return 0; 1164 } 1165 1166 static void 1167 kqueue_task(void *arg, int pending) 1168 { 1169 struct kqueue *kq; 1170 int haskqglobal; 1171 1172 haskqglobal = 0; 1173 kq = arg; 1174 1175 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1176 KQ_LOCK(kq); 1177 1178 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 1179 1180 kq->kq_state &= ~KQ_TASKSCHED; 1181 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 1182 wakeup(&kq->kq_state); 1183 } 1184 KQ_UNLOCK(kq); 1185 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1186 } 1187 1188 /* 1189 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 1190 * We treat KN_MARKER knotes as if they are INFLUX. 1191 */ 1192 static int 1193 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 1194 const struct timespec *tsp, struct kevent *keva, struct thread *td) 1195 { 1196 struct kevent *kevp; 1197 struct timeval atv, rtv, ttv; 1198 struct knote *kn, *marker; 1199 int count, timeout, nkev, error, influx; 1200 int haskqglobal; 1201 1202 count = maxevents; 1203 nkev = 0; 1204 error = 0; 1205 haskqglobal = 0; 1206 1207 if (maxevents == 0) 1208 goto done_nl; 1209 1210 if (tsp != NULL) { 1211 TIMESPEC_TO_TIMEVAL(&atv, tsp); 1212 if (itimerfix(&atv)) { 1213 error = EINVAL; 1214 goto done_nl; 1215 } 1216 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 1217 timeout = -1; 1218 else 1219 timeout = atv.tv_sec > 24 * 60 * 60 ? 1220 24 * 60 * 60 * hz : tvtohz(&atv); 1221 getmicrouptime(&rtv); 1222 timevaladd(&atv, &rtv); 1223 } else { 1224 atv.tv_sec = 0; 1225 atv.tv_usec = 0; 1226 timeout = 0; 1227 } 1228 marker = knote_alloc(1); 1229 if (marker == NULL) { 1230 error = ENOMEM; 1231 goto done_nl; 1232 } 1233 marker->kn_status = KN_MARKER; 1234 KQ_LOCK(kq); 1235 goto start; 1236 1237 retry: 1238 if (atv.tv_sec || atv.tv_usec) { 1239 getmicrouptime(&rtv); 1240 if (timevalcmp(&rtv, &atv, >=)) 1241 goto done; 1242 ttv = atv; 1243 timevalsub(&ttv, &rtv); 1244 timeout = ttv.tv_sec > 24 * 60 * 60 ? 1245 24 * 60 * 60 * hz : tvtohz(&ttv); 1246 } 1247 1248 start: 1249 kevp = keva; 1250 if (kq->kq_count == 0) { 1251 if (timeout < 0) { 1252 error = EWOULDBLOCK; 1253 } else { 1254 kq->kq_state |= KQ_SLEEP; 1255 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH, 1256 "kqread", timeout); 1257 } 1258 if (error == 0) 1259 goto retry; 1260 /* don't restart after signals... */ 1261 if (error == ERESTART) 1262 error = EINTR; 1263 else if (error == EWOULDBLOCK) 1264 error = 0; 1265 goto done; 1266 } 1267 1268 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1269 influx = 0; 1270 while (count) { 1271 KQ_OWNED(kq); 1272 kn = TAILQ_FIRST(&kq->kq_head); 1273 1274 if ((kn->kn_status == KN_MARKER && kn != marker) || 1275 (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1276 if (influx) { 1277 influx = 0; 1278 KQ_FLUX_WAKEUP(kq); 1279 } 1280 kq->kq_state |= KQ_FLUXWAIT; 1281 error = msleep(kq, &kq->kq_lock, PSOCK, 1282 "kqflxwt", 0); 1283 continue; 1284 } 1285 1286 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1287 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 1288 kn->kn_status &= ~KN_QUEUED; 1289 kq->kq_count--; 1290 continue; 1291 } 1292 if (kn == marker) { 1293 KQ_FLUX_WAKEUP(kq); 1294 if (count == maxevents) 1295 goto retry; 1296 goto done; 1297 } 1298 KASSERT((kn->kn_status & KN_INFLUX) == 0, 1299 ("KN_INFLUX set when not suppose to be")); 1300 1301 if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 1302 kn->kn_status &= ~KN_QUEUED; 1303 kn->kn_status |= KN_INFLUX; 1304 kq->kq_count--; 1305 KQ_UNLOCK(kq); 1306 /* 1307 * We don't need to lock the list since we've marked 1308 * it _INFLUX. 1309 */ 1310 *kevp = kn->kn_kevent; 1311 if (!(kn->kn_status & KN_DETACHED)) 1312 kn->kn_fop->f_detach(kn); 1313 knote_drop(kn, td); 1314 KQ_LOCK(kq); 1315 kn = NULL; 1316 } else { 1317 kn->kn_status |= KN_INFLUX; 1318 KQ_UNLOCK(kq); 1319 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 1320 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1321 KN_LIST_LOCK(kn); 1322 if (kn->kn_fop->f_event(kn, 0) == 0) { 1323 KQ_LOCK(kq); 1324 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1325 kn->kn_status &= 1326 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX); 1327 kq->kq_count--; 1328 KN_LIST_UNLOCK(kn); 1329 influx = 1; 1330 continue; 1331 } 1332 *kevp = kn->kn_kevent; 1333 KQ_LOCK(kq); 1334 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1335 if (kn->kn_flags & EV_CLEAR) { 1336 kn->kn_data = 0; 1337 kn->kn_fflags = 0; 1338 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1339 kq->kq_count--; 1340 } else 1341 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1342 1343 kn->kn_status &= ~(KN_INFLUX); 1344 KN_LIST_UNLOCK(kn); 1345 influx = 1; 1346 } 1347 1348 /* we are returning a copy to the user */ 1349 kevp++; 1350 nkev++; 1351 count--; 1352 1353 if (nkev == KQ_NEVENTS) { 1354 influx = 0; 1355 KQ_UNLOCK_FLUX(kq); 1356 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1357 nkev = 0; 1358 kevp = keva; 1359 KQ_LOCK(kq); 1360 if (error) 1361 break; 1362 } 1363 } 1364 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1365 done: 1366 KQ_OWNED(kq); 1367 KQ_UNLOCK_FLUX(kq); 1368 knote_free(marker); 1369 done_nl: 1370 KQ_NOTOWNED(kq); 1371 if (nkev != 0) 1372 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1373 td->td_retval[0] = maxevents - count; 1374 return (error); 1375 } 1376 1377 /* 1378 * XXX 1379 * This could be expanded to call kqueue_scan, if desired. 1380 */ 1381 /*ARGSUSED*/ 1382 static int 1383 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 1384 int flags, struct thread *td) 1385 { 1386 return (ENXIO); 1387 } 1388 1389 /*ARGSUSED*/ 1390 static int 1391 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 1392 int flags, struct thread *td) 1393 { 1394 return (ENXIO); 1395 } 1396 1397 /*ARGSUSED*/ 1398 static int 1399 kqueue_ioctl(struct file *fp, u_long cmd, void *data, 1400 struct ucred *active_cred, struct thread *td) 1401 { 1402 /* 1403 * Enabling sigio causes two major problems: 1404 * 1) infinite recursion: 1405 * Synopsys: kevent is being used to track signals and have FIOASYNC 1406 * set. On receipt of a signal this will cause a kqueue to recurse 1407 * into itself over and over. Sending the sigio causes the kqueue 1408 * to become ready, which in turn posts sigio again, forever. 1409 * Solution: this can be solved by setting a flag in the kqueue that 1410 * we have a SIGIO in progress. 1411 * 2) locking problems: 1412 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 1413 * us above the proc and pgrp locks. 1414 * Solution: Post a signal using an async mechanism, being sure to 1415 * record a generation count in the delivery so that we do not deliver 1416 * a signal to the wrong process. 1417 * 1418 * Note, these two mechanisms are somewhat mutually exclusive! 1419 */ 1420 #if 0 1421 struct kqueue *kq; 1422 1423 kq = fp->f_data; 1424 switch (cmd) { 1425 case FIOASYNC: 1426 if (*(int *)data) { 1427 kq->kq_state |= KQ_ASYNC; 1428 } else { 1429 kq->kq_state &= ~KQ_ASYNC; 1430 } 1431 return (0); 1432 1433 case FIOSETOWN: 1434 return (fsetown(*(int *)data, &kq->kq_sigio)); 1435 1436 case FIOGETOWN: 1437 *(int *)data = fgetown(&kq->kq_sigio); 1438 return (0); 1439 } 1440 #endif 1441 1442 return (ENOTTY); 1443 } 1444 1445 /*ARGSUSED*/ 1446 static int 1447 kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 1448 struct thread *td) 1449 { 1450 struct kqueue *kq; 1451 int revents = 0; 1452 int error; 1453 1454 if ((error = kqueue_acquire(fp, &kq))) 1455 return POLLERR; 1456 1457 KQ_LOCK(kq); 1458 if (events & (POLLIN | POLLRDNORM)) { 1459 if (kq->kq_count) { 1460 revents |= events & (POLLIN | POLLRDNORM); 1461 } else { 1462 selrecord(td, &kq->kq_sel); 1463 kq->kq_state |= KQ_SEL; 1464 } 1465 } 1466 kqueue_release(kq, 1); 1467 KQ_UNLOCK(kq); 1468 return (revents); 1469 } 1470 1471 /*ARGSUSED*/ 1472 static int 1473 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 1474 struct thread *td) 1475 { 1476 1477 bzero((void *)st, sizeof *st); 1478 /* 1479 * We no longer return kq_count because the unlocked value is useless. 1480 * If you spent all this time getting the count, why not spend your 1481 * syscall better by calling kevent? 1482 * 1483 * XXX - This is needed for libc_r. 1484 */ 1485 st->st_mode = S_IFIFO; 1486 return (0); 1487 } 1488 1489 /*ARGSUSED*/ 1490 static int 1491 kqueue_close(struct file *fp, struct thread *td) 1492 { 1493 struct kqueue *kq = fp->f_data; 1494 struct filedesc *fdp; 1495 struct knote *kn; 1496 int i; 1497 int error; 1498 1499 if ((error = kqueue_acquire(fp, &kq))) 1500 return error; 1501 1502 KQ_LOCK(kq); 1503 1504 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 1505 ("kqueue already closing")); 1506 kq->kq_state |= KQ_CLOSING; 1507 if (kq->kq_refcnt > 1) 1508 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 1509 1510 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 1511 fdp = kq->kq_fdp; 1512 1513 KASSERT(knlist_empty(&kq->kq_sel.si_note), 1514 ("kqueue's knlist not empty")); 1515 1516 for (i = 0; i < kq->kq_knlistsize; i++) { 1517 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 1518 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1519 kq->kq_state |= KQ_FLUXWAIT; 1520 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 1521 continue; 1522 } 1523 kn->kn_status |= KN_INFLUX; 1524 KQ_UNLOCK(kq); 1525 if (!(kn->kn_status & KN_DETACHED)) 1526 kn->kn_fop->f_detach(kn); 1527 knote_drop(kn, td); 1528 KQ_LOCK(kq); 1529 } 1530 } 1531 if (kq->kq_knhashmask != 0) { 1532 for (i = 0; i <= kq->kq_knhashmask; i++) { 1533 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 1534 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1535 kq->kq_state |= KQ_FLUXWAIT; 1536 msleep(kq, &kq->kq_lock, PSOCK, 1537 "kqclo2", 0); 1538 continue; 1539 } 1540 kn->kn_status |= KN_INFLUX; 1541 KQ_UNLOCK(kq); 1542 if (!(kn->kn_status & KN_DETACHED)) 1543 kn->kn_fop->f_detach(kn); 1544 knote_drop(kn, td); 1545 KQ_LOCK(kq); 1546 } 1547 } 1548 } 1549 1550 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 1551 kq->kq_state |= KQ_TASKDRAIN; 1552 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 1553 } 1554 1555 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1556 kq->kq_state &= ~KQ_SEL; 1557 selwakeuppri(&kq->kq_sel, PSOCK); 1558 } 1559 1560 KQ_UNLOCK(kq); 1561 1562 FILEDESC_XLOCK(fdp); 1563 SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list); 1564 FILEDESC_XUNLOCK(fdp); 1565 1566 knlist_destroy(&kq->kq_sel.si_note); 1567 mtx_destroy(&kq->kq_lock); 1568 kq->kq_fdp = NULL; 1569 1570 if (kq->kq_knhash != NULL) 1571 free(kq->kq_knhash, M_KQUEUE); 1572 if (kq->kq_knlist != NULL) 1573 free(kq->kq_knlist, M_KQUEUE); 1574 1575 funsetown(&kq->kq_sigio); 1576 free(kq, M_KQUEUE); 1577 fp->f_data = NULL; 1578 1579 return (0); 1580 } 1581 1582 static void 1583 kqueue_wakeup(struct kqueue *kq) 1584 { 1585 KQ_OWNED(kq); 1586 1587 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 1588 kq->kq_state &= ~KQ_SLEEP; 1589 wakeup(kq); 1590 } 1591 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1592 kq->kq_state &= ~KQ_SEL; 1593 selwakeuppri(&kq->kq_sel, PSOCK); 1594 } 1595 if (!knlist_empty(&kq->kq_sel.si_note)) 1596 kqueue_schedtask(kq); 1597 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 1598 pgsigio(&kq->kq_sigio, SIGIO, 0); 1599 } 1600 } 1601 1602 /* 1603 * Walk down a list of knotes, activating them if their event has triggered. 1604 * 1605 * There is a possibility to optimize in the case of one kq watching another. 1606 * Instead of scheduling a task to wake it up, you could pass enough state 1607 * down the chain to make up the parent kqueue. Make this code functional 1608 * first. 1609 */ 1610 void 1611 knote(struct knlist *list, long hint, int islocked) 1612 { 1613 struct kqueue *kq; 1614 struct knote *kn; 1615 1616 if (list == NULL) 1617 return; 1618 1619 KNL_ASSERT_LOCK(list, islocked); 1620 1621 if (!islocked) 1622 list->kl_lock(list->kl_lockarg); 1623 1624 /* 1625 * If we unlock the list lock (and set KN_INFLUX), we can eliminate 1626 * the kqueue scheduling, but this will introduce four 1627 * lock/unlock's for each knote to test. If we do, continue to use 1628 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is 1629 * only safe if you want to remove the current item, which we are 1630 * not doing. 1631 */ 1632 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 1633 kq = kn->kn_kq; 1634 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) { 1635 KQ_LOCK(kq); 1636 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) { 1637 kn->kn_status |= KN_HASKQLOCK; 1638 if (kn->kn_fop->f_event(kn, hint)) 1639 KNOTE_ACTIVATE(kn, 1); 1640 kn->kn_status &= ~KN_HASKQLOCK; 1641 } 1642 KQ_UNLOCK(kq); 1643 } 1644 kq = NULL; 1645 } 1646 if (!islocked) 1647 list->kl_unlock(list->kl_lockarg); 1648 } 1649 1650 /* 1651 * add a knote to a knlist 1652 */ 1653 void 1654 knlist_add(struct knlist *knl, struct knote *kn, int islocked) 1655 { 1656 KNL_ASSERT_LOCK(knl, islocked); 1657 KQ_NOTOWNED(kn->kn_kq); 1658 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == 1659 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED")); 1660 if (!islocked) 1661 knl->kl_lock(knl->kl_lockarg); 1662 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 1663 if (!islocked) 1664 knl->kl_unlock(knl->kl_lockarg); 1665 KQ_LOCK(kn->kn_kq); 1666 kn->kn_knlist = knl; 1667 kn->kn_status &= ~KN_DETACHED; 1668 KQ_UNLOCK(kn->kn_kq); 1669 } 1670 1671 static void 1672 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked) 1673 { 1674 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked")); 1675 KNL_ASSERT_LOCK(knl, knlislocked); 1676 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 1677 if (!kqislocked) 1678 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX, 1679 ("knlist_remove called w/o knote being KN_INFLUX or already removed")); 1680 if (!knlislocked) 1681 knl->kl_lock(knl->kl_lockarg); 1682 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 1683 kn->kn_knlist = NULL; 1684 if (!knlislocked) 1685 knl->kl_unlock(knl->kl_lockarg); 1686 if (!kqislocked) 1687 KQ_LOCK(kn->kn_kq); 1688 kn->kn_status |= KN_DETACHED; 1689 if (!kqislocked) 1690 KQ_UNLOCK(kn->kn_kq); 1691 } 1692 1693 /* 1694 * remove all knotes from a specified klist 1695 */ 1696 void 1697 knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 1698 { 1699 1700 knlist_remove_kq(knl, kn, islocked, 0); 1701 } 1702 1703 /* 1704 * remove knote from a specified klist while in f_event handler. 1705 */ 1706 void 1707 knlist_remove_inevent(struct knlist *knl, struct knote *kn) 1708 { 1709 1710 knlist_remove_kq(knl, kn, 1, 1711 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK); 1712 } 1713 1714 int 1715 knlist_empty(struct knlist *knl) 1716 { 1717 KNL_ASSERT_LOCKED(knl); 1718 return SLIST_EMPTY(&knl->kl_list); 1719 } 1720 1721 static struct mtx knlist_lock; 1722 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 1723 MTX_DEF); 1724 static void knlist_mtx_lock(void *arg); 1725 static void knlist_mtx_unlock(void *arg); 1726 static int knlist_mtx_locked(void *arg); 1727 1728 static void 1729 knlist_mtx_lock(void *arg) 1730 { 1731 mtx_lock((struct mtx *)arg); 1732 } 1733 1734 static void 1735 knlist_mtx_unlock(void *arg) 1736 { 1737 mtx_unlock((struct mtx *)arg); 1738 } 1739 1740 static int 1741 knlist_mtx_locked(void *arg) 1742 { 1743 return (mtx_owned((struct mtx *)arg)); 1744 } 1745 1746 void 1747 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 1748 void (*kl_unlock)(void *), int (*kl_locked)(void *)) 1749 { 1750 1751 if (lock == NULL) 1752 knl->kl_lockarg = &knlist_lock; 1753 else 1754 knl->kl_lockarg = lock; 1755 1756 if (kl_lock == NULL) 1757 knl->kl_lock = knlist_mtx_lock; 1758 else 1759 knl->kl_lock = kl_lock; 1760 if (kl_unlock == NULL) 1761 knl->kl_unlock = knlist_mtx_unlock; 1762 else 1763 knl->kl_unlock = kl_unlock; 1764 if (kl_locked == NULL) 1765 knl->kl_locked = knlist_mtx_locked; 1766 else 1767 knl->kl_locked = kl_locked; 1768 1769 SLIST_INIT(&knl->kl_list); 1770 } 1771 1772 void 1773 knlist_destroy(struct knlist *knl) 1774 { 1775 1776 #ifdef INVARIANTS 1777 /* 1778 * if we run across this error, we need to find the offending 1779 * driver and have it call knlist_clear. 1780 */ 1781 if (!SLIST_EMPTY(&knl->kl_list)) 1782 printf("WARNING: destroying knlist w/ knotes on it!\n"); 1783 #endif 1784 1785 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL; 1786 SLIST_INIT(&knl->kl_list); 1787 } 1788 1789 /* 1790 * Even if we are locked, we may need to drop the lock to allow any influx 1791 * knotes time to "settle". 1792 */ 1793 void 1794 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 1795 { 1796 struct knote *kn, *kn2; 1797 struct kqueue *kq; 1798 1799 if (islocked) 1800 KNL_ASSERT_LOCKED(knl); 1801 else { 1802 KNL_ASSERT_UNLOCKED(knl); 1803 again: /* need to reacquire lock since we have dropped it */ 1804 knl->kl_lock(knl->kl_lockarg); 1805 } 1806 1807 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 1808 kq = kn->kn_kq; 1809 KQ_LOCK(kq); 1810 if ((kn->kn_status & KN_INFLUX)) { 1811 KQ_UNLOCK(kq); 1812 continue; 1813 } 1814 knlist_remove_kq(knl, kn, 1, 1); 1815 if (killkn) { 1816 kn->kn_status |= KN_INFLUX | KN_DETACHED; 1817 KQ_UNLOCK(kq); 1818 knote_drop(kn, td); 1819 } else { 1820 /* Make sure cleared knotes disappear soon */ 1821 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 1822 KQ_UNLOCK(kq); 1823 } 1824 kq = NULL; 1825 } 1826 1827 if (!SLIST_EMPTY(&knl->kl_list)) { 1828 /* there are still KN_INFLUX remaining */ 1829 kn = SLIST_FIRST(&knl->kl_list); 1830 kq = kn->kn_kq; 1831 KQ_LOCK(kq); 1832 KASSERT(kn->kn_status & KN_INFLUX, 1833 ("knote removed w/o list lock")); 1834 knl->kl_unlock(knl->kl_lockarg); 1835 kq->kq_state |= KQ_FLUXWAIT; 1836 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 1837 kq = NULL; 1838 goto again; 1839 } 1840 1841 if (islocked) 1842 KNL_ASSERT_LOCKED(knl); 1843 else { 1844 knl->kl_unlock(knl->kl_lockarg); 1845 KNL_ASSERT_UNLOCKED(knl); 1846 } 1847 } 1848 1849 /* 1850 * Remove all knotes referencing a specified fd must be called with FILEDESC 1851 * lock. This prevents a race where a new fd comes along and occupies the 1852 * entry and we attach a knote to the fd. 1853 */ 1854 void 1855 knote_fdclose(struct thread *td, int fd) 1856 { 1857 struct filedesc *fdp = td->td_proc->p_fd; 1858 struct kqueue *kq; 1859 struct knote *kn; 1860 int influx; 1861 1862 FILEDESC_XLOCK_ASSERT(fdp); 1863 1864 /* 1865 * We shouldn't have to worry about new kevents appearing on fd 1866 * since filedesc is locked. 1867 */ 1868 SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 1869 KQ_LOCK(kq); 1870 1871 again: 1872 influx = 0; 1873 while (kq->kq_knlistsize > fd && 1874 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 1875 if (kn->kn_status & KN_INFLUX) { 1876 /* someone else might be waiting on our knote */ 1877 if (influx) 1878 wakeup(kq); 1879 kq->kq_state |= KQ_FLUXWAIT; 1880 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 1881 goto again; 1882 } 1883 kn->kn_status |= KN_INFLUX; 1884 KQ_UNLOCK(kq); 1885 if (!(kn->kn_status & KN_DETACHED)) 1886 kn->kn_fop->f_detach(kn); 1887 knote_drop(kn, td); 1888 influx = 1; 1889 KQ_LOCK(kq); 1890 } 1891 KQ_UNLOCK_FLUX(kq); 1892 } 1893 } 1894 1895 static int 1896 knote_attach(struct knote *kn, struct kqueue *kq) 1897 { 1898 struct klist *list; 1899 1900 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX")); 1901 KQ_OWNED(kq); 1902 1903 if (kn->kn_fop->f_isfd) { 1904 if (kn->kn_id >= kq->kq_knlistsize) 1905 return ENOMEM; 1906 list = &kq->kq_knlist[kn->kn_id]; 1907 } else { 1908 if (kq->kq_knhash == NULL) 1909 return ENOMEM; 1910 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1911 } 1912 1913 SLIST_INSERT_HEAD(list, kn, kn_link); 1914 1915 return 0; 1916 } 1917 1918 /* 1919 * knote must already have been detached using the f_detach method. 1920 * no lock need to be held, it is assumed that the KN_INFLUX flag is set 1921 * to prevent other removal. 1922 */ 1923 static void 1924 knote_drop(struct knote *kn, struct thread *td) 1925 { 1926 struct kqueue *kq; 1927 struct klist *list; 1928 1929 kq = kn->kn_kq; 1930 1931 KQ_NOTOWNED(kq); 1932 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX, 1933 ("knote_drop called without KN_INFLUX set in kn_status")); 1934 1935 KQ_LOCK(kq); 1936 if (kn->kn_fop->f_isfd) 1937 list = &kq->kq_knlist[kn->kn_id]; 1938 else 1939 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1940 1941 if (!SLIST_EMPTY(list)) 1942 SLIST_REMOVE(list, kn, knote, kn_link); 1943 if (kn->kn_status & KN_QUEUED) 1944 knote_dequeue(kn); 1945 KQ_UNLOCK_FLUX(kq); 1946 1947 if (kn->kn_fop->f_isfd) { 1948 fdrop(kn->kn_fp, td); 1949 kn->kn_fp = NULL; 1950 } 1951 kqueue_fo_release(kn->kn_kevent.filter); 1952 kn->kn_fop = NULL; 1953 knote_free(kn); 1954 } 1955 1956 static void 1957 knote_enqueue(struct knote *kn) 1958 { 1959 struct kqueue *kq = kn->kn_kq; 1960 1961 KQ_OWNED(kn->kn_kq); 1962 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1963 1964 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1965 kn->kn_status |= KN_QUEUED; 1966 kq->kq_count++; 1967 kqueue_wakeup(kq); 1968 } 1969 1970 static void 1971 knote_dequeue(struct knote *kn) 1972 { 1973 struct kqueue *kq = kn->kn_kq; 1974 1975 KQ_OWNED(kn->kn_kq); 1976 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1977 1978 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1979 kn->kn_status &= ~KN_QUEUED; 1980 kq->kq_count--; 1981 } 1982 1983 static void 1984 knote_init(void) 1985 { 1986 1987 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 1988 NULL, NULL, UMA_ALIGN_PTR, 0); 1989 } 1990 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 1991 1992 static struct knote * 1993 knote_alloc(int waitok) 1994 { 1995 return ((struct knote *)uma_zalloc(knote_zone, 1996 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO)); 1997 } 1998 1999 static void 2000 knote_free(struct knote *kn) 2001 { 2002 if (kn != NULL) 2003 uma_zfree(knote_zone, kn); 2004 } 2005 2006 /* 2007 * Register the kev w/ the kq specified by fd. 2008 */ 2009 int 2010 kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok) 2011 { 2012 struct kqueue *kq; 2013 struct file *fp; 2014 int error; 2015 2016 if ((error = fget(td, fd, &fp)) != 0) 2017 return (error); 2018 if ((error = kqueue_acquire(fp, &kq)) != 0) 2019 goto noacquire; 2020 2021 error = kqueue_register(kq, kev, td, waitok); 2022 2023 kqueue_release(kq, 0); 2024 2025 noacquire: 2026 fdrop(fp, td); 2027 2028 return error; 2029 }
Cache object: eac1429f19f190cced6692248d9102d7
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]