1 /*- 2 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD: src/sys/kern/kern_event.c,v 1.2.2.10 2004/04/04 07:03:14 cperciva Exp $ 27 * $DragonFly: src/sys/kern/kern_event.c,v 1.33 2007/02/03 17:05:57 corecode Exp $ 28 */ 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/proc.h> 34 #include <sys/malloc.h> 35 #include <sys/unistd.h> 36 #include <sys/file.h> 37 #include <sys/lock.h> 38 #include <sys/fcntl.h> 39 #include <sys/queue.h> 40 #include <sys/event.h> 41 #include <sys/eventvar.h> 42 #include <sys/protosw.h> 43 #include <sys/socket.h> 44 #include <sys/socketvar.h> 45 #include <sys/stat.h> 46 #include <sys/sysctl.h> 47 #include <sys/sysproto.h> 48 #include <sys/thread.h> 49 #include <sys/uio.h> 50 #include <sys/signalvar.h> 51 #include <sys/filio.h> 52 #include <sys/ktr.h> 53 54 #include <sys/thread2.h> 55 #include <sys/file2.h> 56 #include <sys/mplock2.h> 57 58 /* 59 * Global token for kqueue subsystem 60 */ 61 struct lwkt_token kq_token = LWKT_TOKEN_INITIALIZER(kq_token); 62 SYSCTL_LONG(_lwkt, OID_AUTO, kq_collisions, 63 CTLFLAG_RW, &kq_token.t_collisions, 0, 64 "Collision counter of kq_token"); 65 66 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 67 68 struct kevent_copyin_args { 69 struct kevent_args *ka; 70 int pchanges; 71 }; 72 73 static int kqueue_sleep(struct kqueue *kq, struct timespec *tsp); 74 static int kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 75 struct knote *marker); 76 static int kqueue_read(struct file *fp, struct uio *uio, 77 struct ucred *cred, int flags); 78 static int kqueue_write(struct file *fp, struct uio *uio, 79 struct ucred *cred, int flags); 80 static int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 81 struct ucred *cred, struct sysmsg *msg); 82 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 83 static int kqueue_stat(struct file *fp, struct stat *st, 84 struct ucred *cred); 85 static int kqueue_close(struct file *fp); 86 static void kqueue_wakeup(struct kqueue *kq); 87 static int filter_attach(struct knote *kn); 88 static int filter_event(struct knote *kn, long hint); 89 90 /* 91 * MPSAFE 92 */ 93 static struct fileops kqueueops = { 94 .fo_read = kqueue_read, 95 .fo_write = kqueue_write, 96 .fo_ioctl = kqueue_ioctl, 97 .fo_kqfilter = kqueue_kqfilter, 98 .fo_stat = kqueue_stat, 99 .fo_close = kqueue_close, 100 .fo_shutdown = nofo_shutdown 101 }; 102 103 static void knote_attach(struct knote *kn); 104 static void knote_drop(struct knote *kn); 105 static void knote_detach_and_drop(struct knote *kn); 106 static void knote_enqueue(struct knote *kn); 107 static void knote_dequeue(struct knote *kn); 108 static struct knote *knote_alloc(void); 109 static void knote_free(struct knote *kn); 110 111 static void filt_kqdetach(struct knote *kn); 112 static int filt_kqueue(struct knote *kn, long hint); 113 static int filt_procattach(struct knote *kn); 114 static void filt_procdetach(struct knote *kn); 115 static int filt_proc(struct knote *kn, long hint); 116 static int filt_fileattach(struct knote *kn); 117 static void filt_timerexpire(void *knx); 118 static int filt_timerattach(struct knote *kn); 119 static void filt_timerdetach(struct knote *kn); 120 static int filt_timer(struct knote *kn, long hint); 121 122 static struct filterops file_filtops = 123 { FILTEROP_ISFD, filt_fileattach, NULL, NULL }; 124 static struct filterops kqread_filtops = 125 { FILTEROP_ISFD, NULL, filt_kqdetach, filt_kqueue }; 126 static struct filterops proc_filtops = 127 { 0, filt_procattach, filt_procdetach, filt_proc }; 128 static struct filterops timer_filtops = 129 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 130 131 static int kq_ncallouts = 0; 132 static int kq_calloutmax = (4 * 1024); 133 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 134 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 135 static int kq_checkloop = 1000000; 136 SYSCTL_INT(_kern, OID_AUTO, kq_checkloop, CTLFLAG_RW, 137 &kq_checkloop, 0, "Maximum number of callouts allocated for kqueue"); 138 139 #define KNOTE_ACTIVATE(kn) do { \ 140 kn->kn_status |= KN_ACTIVE; \ 141 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 142 knote_enqueue(kn); \ 143 } while(0) 144 145 #define KN_HASHSIZE 64 /* XXX should be tunable */ 146 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 147 148 extern struct filterops aio_filtops; 149 extern struct filterops sig_filtops; 150 151 /* 152 * Table for for all system-defined filters. 153 */ 154 static struct filterops *sysfilt_ops[] = { 155 &file_filtops, /* EVFILT_READ */ 156 &file_filtops, /* EVFILT_WRITE */ 157 &aio_filtops, /* EVFILT_AIO */ 158 &file_filtops, /* EVFILT_VNODE */ 159 &proc_filtops, /* EVFILT_PROC */ 160 &sig_filtops, /* EVFILT_SIGNAL */ 161 &timer_filtops, /* EVFILT_TIMER */ 162 &file_filtops, /* EVFILT_EXCEPT */ 163 }; 164 165 static int 166 filt_fileattach(struct knote *kn) 167 { 168 return (fo_kqfilter(kn->kn_fp, kn)); 169 } 170 171 /* 172 * MPSAFE 173 */ 174 static int 175 kqueue_kqfilter(struct file *fp, struct knote *kn) 176 { 177 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 178 179 if (kn->kn_filter != EVFILT_READ) 180 return (EOPNOTSUPP); 181 182 kn->kn_fop = &kqread_filtops; 183 knote_insert(&kq->kq_kqinfo.ki_note, kn); 184 return (0); 185 } 186 187 static void 188 filt_kqdetach(struct knote *kn) 189 { 190 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 191 192 knote_remove(&kq->kq_kqinfo.ki_note, kn); 193 } 194 195 /*ARGSUSED*/ 196 static int 197 filt_kqueue(struct knote *kn, long hint) 198 { 199 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 200 201 kn->kn_data = kq->kq_count; 202 return (kn->kn_data > 0); 203 } 204 205 static int 206 filt_procattach(struct knote *kn) 207 { 208 struct proc *p; 209 int immediate; 210 211 immediate = 0; 212 p = pfind(kn->kn_id); 213 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 214 p = zpfind(kn->kn_id); 215 immediate = 1; 216 } 217 if (p == NULL) { 218 return (ESRCH); 219 } 220 if (!PRISON_CHECK(curthread->td_ucred, p->p_ucred)) { 221 if (p) 222 PRELE(p); 223 return (EACCES); 224 } 225 226 lwkt_gettoken(&p->p_token); 227 kn->kn_ptr.p_proc = p; 228 kn->kn_flags |= EV_CLEAR; /* automatically set */ 229 230 /* 231 * internal flag indicating registration done by kernel 232 */ 233 if (kn->kn_flags & EV_FLAG1) { 234 kn->kn_data = kn->kn_sdata; /* ppid */ 235 kn->kn_fflags = NOTE_CHILD; 236 kn->kn_flags &= ~EV_FLAG1; 237 } 238 239 knote_insert(&p->p_klist, kn); 240 241 /* 242 * Immediately activate any exit notes if the target process is a 243 * zombie. This is necessary to handle the case where the target 244 * process, e.g. a child, dies before the kevent is negistered. 245 */ 246 if (immediate && filt_proc(kn, NOTE_EXIT)) 247 KNOTE_ACTIVATE(kn); 248 lwkt_reltoken(&p->p_token); 249 PRELE(p); 250 251 return (0); 252 } 253 254 /* 255 * The knote may be attached to a different process, which may exit, 256 * leaving nothing for the knote to be attached to. So when the process 257 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 258 * it will be deleted when read out. However, as part of the knote deletion, 259 * this routine is called, so a check is needed to avoid actually performing 260 * a detach, because the original process does not exist any more. 261 */ 262 static void 263 filt_procdetach(struct knote *kn) 264 { 265 struct proc *p; 266 267 if (kn->kn_status & KN_DETACHED) 268 return; 269 /* XXX locking? take proc_token here? */ 270 p = kn->kn_ptr.p_proc; 271 knote_remove(&p->p_klist, kn); 272 } 273 274 static int 275 filt_proc(struct knote *kn, long hint) 276 { 277 u_int event; 278 279 /* 280 * mask off extra data 281 */ 282 event = (u_int)hint & NOTE_PCTRLMASK; 283 284 /* 285 * if the user is interested in this event, record it. 286 */ 287 if (kn->kn_sfflags & event) 288 kn->kn_fflags |= event; 289 290 /* 291 * Process is gone, so flag the event as finished. Detach the 292 * knote from the process now because the process will be poof, 293 * gone later on. 294 */ 295 if (event == NOTE_EXIT) { 296 struct proc *p = kn->kn_ptr.p_proc; 297 if ((kn->kn_status & KN_DETACHED) == 0) { 298 knote_remove(&p->p_klist, kn); 299 kn->kn_status |= KN_DETACHED; 300 kn->kn_data = p->p_xstat; 301 kn->kn_ptr.p_proc = NULL; 302 } 303 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 304 return (1); 305 } 306 307 /* 308 * process forked, and user wants to track the new process, 309 * so attach a new knote to it, and immediately report an 310 * event with the parent's pid. 311 */ 312 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 313 struct kevent kev; 314 int error; 315 316 /* 317 * register knote with new process. 318 */ 319 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 320 kev.filter = kn->kn_filter; 321 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 322 kev.fflags = kn->kn_sfflags; 323 kev.data = kn->kn_id; /* parent */ 324 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 325 error = kqueue_register(kn->kn_kq, &kev); 326 if (error) 327 kn->kn_fflags |= NOTE_TRACKERR; 328 } 329 330 return (kn->kn_fflags != 0); 331 } 332 333 /* 334 * The callout interlocks with callout_stop() (or should), so the 335 * knote should still be a valid structure. However the timeout 336 * can race a deletion so if KN_DELETING is set we just don't touch 337 * the knote. 338 */ 339 static void 340 filt_timerexpire(void *knx) 341 { 342 struct knote *kn = knx; 343 struct callout *calloutp; 344 struct timeval tv; 345 int tticks; 346 347 lwkt_gettoken(&kq_token); 348 if ((kn->kn_status & KN_DELETING) == 0) { 349 kn->kn_data++; 350 KNOTE_ACTIVATE(kn); 351 352 if ((kn->kn_flags & EV_ONESHOT) == 0) { 353 tv.tv_sec = kn->kn_sdata / 1000; 354 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 355 tticks = tvtohz_high(&tv); 356 calloutp = (struct callout *)kn->kn_hook; 357 callout_reset(calloutp, tticks, filt_timerexpire, kn); 358 } 359 } 360 lwkt_reltoken(&kq_token); 361 } 362 363 /* 364 * data contains amount of time to sleep, in milliseconds 365 */ 366 static int 367 filt_timerattach(struct knote *kn) 368 { 369 struct callout *calloutp; 370 struct timeval tv; 371 int tticks; 372 373 if (kq_ncallouts >= kq_calloutmax) 374 return (ENOMEM); 375 kq_ncallouts++; 376 377 tv.tv_sec = kn->kn_sdata / 1000; 378 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 379 tticks = tvtohz_high(&tv); 380 381 kn->kn_flags |= EV_CLEAR; /* automatically set */ 382 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 383 M_KQUEUE, M_WAITOK); 384 callout_init(calloutp); 385 kn->kn_hook = (caddr_t)calloutp; 386 callout_reset(calloutp, tticks, filt_timerexpire, kn); 387 388 return (0); 389 } 390 391 static void 392 filt_timerdetach(struct knote *kn) 393 { 394 struct callout *calloutp; 395 396 calloutp = (struct callout *)kn->kn_hook; 397 callout_stop(calloutp); 398 FREE(calloutp, M_KQUEUE); 399 kq_ncallouts--; 400 } 401 402 static int 403 filt_timer(struct knote *kn, long hint) 404 { 405 406 return (kn->kn_data != 0); 407 } 408 409 /* 410 * Acquire a knote, return non-zero on success, 0 on failure. 411 * 412 * If we cannot acquire the knote we sleep and return 0. The knote 413 * may be stale on return in this case and the caller must restart 414 * whatever loop they are in. 415 */ 416 static __inline 417 int 418 knote_acquire(struct knote *kn) 419 { 420 if (kn->kn_status & KN_PROCESSING) { 421 kn->kn_status |= KN_WAITING | KN_REPROCESS; 422 tsleep(kn, 0, "kqepts", hz); 423 /* knote may be stale now */ 424 return(0); 425 } 426 kn->kn_status |= KN_PROCESSING; 427 return(1); 428 } 429 430 /* 431 * Release an acquired knote, clearing KN_PROCESSING and handling any 432 * KN_REPROCESS events. 433 * 434 * Non-zero is returned if the knote is destroyed. 435 */ 436 static __inline 437 int 438 knote_release(struct knote *kn) 439 { 440 while (kn->kn_status & KN_REPROCESS) { 441 kn->kn_status &= ~KN_REPROCESS; 442 if (kn->kn_status & KN_WAITING) { 443 kn->kn_status &= ~KN_WAITING; 444 wakeup(kn); 445 } 446 if (kn->kn_status & KN_DELETING) { 447 knote_detach_and_drop(kn); 448 return(1); 449 /* NOT REACHED */ 450 } 451 if (filter_event(kn, 0)) 452 KNOTE_ACTIVATE(kn); 453 } 454 kn->kn_status &= ~KN_PROCESSING; 455 return(0); 456 } 457 458 /* 459 * Initialize a kqueue. 460 * 461 * NOTE: The lwp/proc code initializes a kqueue for select/poll ops. 462 * 463 * MPSAFE 464 */ 465 void 466 kqueue_init(struct kqueue *kq, struct filedesc *fdp) 467 { 468 TAILQ_INIT(&kq->kq_knpend); 469 TAILQ_INIT(&kq->kq_knlist); 470 kq->kq_count = 0; 471 kq->kq_fdp = fdp; 472 SLIST_INIT(&kq->kq_kqinfo.ki_note); 473 } 474 475 /* 476 * Terminate a kqueue. Freeing the actual kq itself is left up to the 477 * caller (it might be embedded in a lwp so we don't do it here). 478 * 479 * The kq's knlist must be completely eradicated so block on any 480 * processing races. 481 */ 482 void 483 kqueue_terminate(struct kqueue *kq) 484 { 485 struct knote *kn; 486 487 lwkt_gettoken(&kq_token); 488 while ((kn = TAILQ_FIRST(&kq->kq_knlist)) != NULL) { 489 if (knote_acquire(kn)) 490 knote_detach_and_drop(kn); 491 } 492 if (kq->kq_knhash) { 493 kfree(kq->kq_knhash, M_KQUEUE); 494 kq->kq_knhash = NULL; 495 kq->kq_knhashmask = 0; 496 } 497 lwkt_reltoken(&kq_token); 498 } 499 500 /* 501 * MPSAFE 502 */ 503 int 504 sys_kqueue(struct kqueue_args *uap) 505 { 506 struct thread *td = curthread; 507 struct kqueue *kq; 508 struct file *fp; 509 int fd, error; 510 511 error = falloc(td->td_lwp, &fp, &fd); 512 if (error) 513 return (error); 514 fp->f_flag = FREAD | FWRITE; 515 fp->f_type = DTYPE_KQUEUE; 516 fp->f_ops = &kqueueops; 517 518 kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 519 kqueue_init(kq, td->td_proc->p_fd); 520 fp->f_data = kq; 521 522 fsetfd(kq->kq_fdp, fp, fd); 523 uap->sysmsg_result = fd; 524 fdrop(fp); 525 return (error); 526 } 527 528 /* 529 * Copy 'count' items into the destination list pointed to by uap->eventlist. 530 */ 531 static int 532 kevent_copyout(void *arg, struct kevent *kevp, int count, int *res) 533 { 534 struct kevent_copyin_args *kap; 535 int error; 536 537 kap = (struct kevent_copyin_args *)arg; 538 539 error = copyout(kevp, kap->ka->eventlist, count * sizeof(*kevp)); 540 if (error == 0) { 541 kap->ka->eventlist += count; 542 *res += count; 543 } else { 544 *res = -1; 545 } 546 547 return (error); 548 } 549 550 /* 551 * Copy at most 'max' items from the list pointed to by kap->changelist, 552 * return number of items in 'events'. 553 */ 554 static int 555 kevent_copyin(void *arg, struct kevent *kevp, int max, int *events) 556 { 557 struct kevent_copyin_args *kap; 558 int error, count; 559 560 kap = (struct kevent_copyin_args *)arg; 561 562 count = min(kap->ka->nchanges - kap->pchanges, max); 563 error = copyin(kap->ka->changelist, kevp, count * sizeof *kevp); 564 if (error == 0) { 565 kap->ka->changelist += count; 566 kap->pchanges += count; 567 *events = count; 568 } 569 570 return (error); 571 } 572 573 /* 574 * MPSAFE 575 */ 576 int 577 kern_kevent(struct kqueue *kq, int nevents, int *res, void *uap, 578 k_copyin_fn kevent_copyinfn, k_copyout_fn kevent_copyoutfn, 579 struct timespec *tsp_in) 580 { 581 struct kevent *kevp; 582 struct timespec *tsp; 583 int i, n, total, error, nerrors = 0; 584 int lres; 585 int limit = kq_checkloop; 586 struct kevent kev[KQ_NEVENTS]; 587 struct knote marker; 588 589 tsp = tsp_in; 590 *res = 0; 591 592 lwkt_gettoken(&kq_token); 593 for ( ;; ) { 594 n = 0; 595 error = kevent_copyinfn(uap, kev, KQ_NEVENTS, &n); 596 if (error) 597 goto done; 598 if (n == 0) 599 break; 600 for (i = 0; i < n; i++) { 601 kevp = &kev[i]; 602 kevp->flags &= ~EV_SYSFLAGS; 603 error = kqueue_register(kq, kevp); 604 605 /* 606 * If a registration returns an error we 607 * immediately post the error. The kevent() 608 * call itself will fail with the error if 609 * no space is available for posting. 610 * 611 * Such errors normally bypass the timeout/blocking 612 * code. However, if the copyoutfn function refuses 613 * to post the error (see sys_poll()), then we 614 * ignore it too. 615 */ 616 if (error) { 617 kevp->flags = EV_ERROR; 618 kevp->data = error; 619 lres = *res; 620 kevent_copyoutfn(uap, kevp, 1, res); 621 if (lres != *res) { 622 nevents--; 623 nerrors++; 624 } 625 } 626 } 627 } 628 if (nerrors) { 629 error = 0; 630 goto done; 631 } 632 633 /* 634 * Acquire/wait for events - setup timeout 635 */ 636 if (tsp != NULL) { 637 struct timespec ats; 638 639 if (tsp->tv_sec || tsp->tv_nsec) { 640 nanouptime(&ats); 641 timespecadd(tsp, &ats); /* tsp = target time */ 642 } 643 } 644 645 /* 646 * Loop as required. 647 * 648 * Collect as many events as we can. Sleeping on successive 649 * loops is disabled if copyoutfn has incremented (*res). 650 * 651 * The loop stops if an error occurs, all events have been 652 * scanned (the marker has been reached), or fewer than the 653 * maximum number of events is found. 654 * 655 * The copyoutfn function does not have to increment (*res) in 656 * order for the loop to continue. 657 * 658 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents. 659 */ 660 total = 0; 661 error = 0; 662 marker.kn_filter = EVFILT_MARKER; 663 marker.kn_status = KN_PROCESSING; 664 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 665 while ((n = nevents - total) > 0) { 666 if (n > KQ_NEVENTS) 667 n = KQ_NEVENTS; 668 669 /* 670 * If no events are pending sleep until timeout (if any) 671 * or an event occurs. 672 * 673 * After the sleep completes the marker is moved to the 674 * end of the list, making any received events available 675 * to our scan. 676 */ 677 if (kq->kq_count == 0 && *res == 0) { 678 error = kqueue_sleep(kq, tsp); 679 if (error) 680 break; 681 682 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 683 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 684 } 685 686 /* 687 * Process all received events 688 * Account for all non-spurious events in our total 689 */ 690 i = kqueue_scan(kq, kev, n, &marker); 691 if (i) { 692 lres = *res; 693 error = kevent_copyoutfn(uap, kev, i, res); 694 total += *res - lres; 695 if (error) 696 break; 697 } 698 if (limit && --limit == 0) 699 panic("kqueue: checkloop failed i=%d", i); 700 701 /* 702 * Normally when fewer events are returned than requested 703 * we can stop. However, if only spurious events were 704 * collected the copyout will not bump (*res) and we have 705 * to continue. 706 */ 707 if (i < n && *res) 708 break; 709 710 /* 711 * Deal with an edge case where spurious events can cause 712 * a loop to occur without moving the marker. This can 713 * prevent kqueue_scan() from picking up new events which 714 * race us. We must be sure to move the marker for this 715 * case. 716 * 717 * NOTE: We do not want to move the marker if events 718 * were scanned because normal kqueue operations 719 * may reactivate events. Moving the marker in 720 * that case could result in duplicates for the 721 * same event. 722 */ 723 if (i == 0) { 724 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 725 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 726 } 727 } 728 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 729 730 /* Timeouts do not return EWOULDBLOCK. */ 731 if (error == EWOULDBLOCK) 732 error = 0; 733 734 done: 735 lwkt_reltoken(&kq_token); 736 return (error); 737 } 738 739 /* 740 * MPALMOSTSAFE 741 */ 742 int 743 sys_kevent(struct kevent_args *uap) 744 { 745 struct thread *td = curthread; 746 struct proc *p = td->td_proc; 747 struct timespec ts, *tsp; 748 struct kqueue *kq; 749 struct file *fp = NULL; 750 struct kevent_copyin_args *kap, ka; 751 int error; 752 753 if (uap->timeout) { 754 error = copyin(uap->timeout, &ts, sizeof(ts)); 755 if (error) 756 return (error); 757 tsp = &ts; 758 } else { 759 tsp = NULL; 760 } 761 762 fp = holdfp(p->p_fd, uap->fd, -1); 763 if (fp == NULL) 764 return (EBADF); 765 if (fp->f_type != DTYPE_KQUEUE) { 766 fdrop(fp); 767 return (EBADF); 768 } 769 770 kq = (struct kqueue *)fp->f_data; 771 772 kap = &ka; 773 kap->ka = uap; 774 kap->pchanges = 0; 775 776 error = kern_kevent(kq, uap->nevents, &uap->sysmsg_result, kap, 777 kevent_copyin, kevent_copyout, tsp); 778 779 fdrop(fp); 780 781 return (error); 782 } 783 784 int 785 kqueue_register(struct kqueue *kq, struct kevent *kev) 786 { 787 struct filedesc *fdp = kq->kq_fdp; 788 struct filterops *fops; 789 struct file *fp = NULL; 790 struct knote *kn = NULL; 791 int error = 0; 792 793 if (kev->filter < 0) { 794 if (kev->filter + EVFILT_SYSCOUNT < 0) 795 return (EINVAL); 796 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 797 } else { 798 /* 799 * XXX 800 * filter attach routine is responsible for insuring that 801 * the identifier can be attached to it. 802 */ 803 kprintf("unknown filter: %d\n", kev->filter); 804 return (EINVAL); 805 } 806 807 lwkt_gettoken(&kq_token); 808 if (fops->f_flags & FILTEROP_ISFD) { 809 /* validate descriptor */ 810 fp = holdfp(fdp, kev->ident, -1); 811 if (fp == NULL) { 812 lwkt_reltoken(&kq_token); 813 return (EBADF); 814 } 815 816 again1: 817 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 818 if (kn->kn_kq == kq && 819 kn->kn_filter == kev->filter && 820 kn->kn_id == kev->ident) { 821 if (knote_acquire(kn) == 0) 822 goto again1; 823 break; 824 } 825 } 826 } else { 827 if (kq->kq_knhashmask) { 828 struct klist *list; 829 830 list = &kq->kq_knhash[ 831 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 832 again2: 833 SLIST_FOREACH(kn, list, kn_link) { 834 if (kn->kn_id == kev->ident && 835 kn->kn_filter == kev->filter) { 836 if (knote_acquire(kn) == 0) 837 goto again2; 838 break; 839 } 840 } 841 } 842 } 843 844 /* 845 * NOTE: At this point if kn is non-NULL we will have acquired 846 * it and set KN_PROCESSING. 847 */ 848 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 849 error = ENOENT; 850 goto done; 851 } 852 853 /* 854 * kn now contains the matching knote, or NULL if no match 855 */ 856 if (kev->flags & EV_ADD) { 857 if (kn == NULL) { 858 kn = knote_alloc(); 859 if (kn == NULL) { 860 error = ENOMEM; 861 goto done; 862 } 863 kn->kn_fp = fp; 864 kn->kn_kq = kq; 865 kn->kn_fop = fops; 866 867 /* 868 * apply reference count to knote structure, and 869 * do not release it at the end of this routine. 870 */ 871 fp = NULL; 872 873 kn->kn_sfflags = kev->fflags; 874 kn->kn_sdata = kev->data; 875 kev->fflags = 0; 876 kev->data = 0; 877 kn->kn_kevent = *kev; 878 879 /* 880 * KN_PROCESSING prevents the knote from getting 881 * ripped out from under us while we are trying 882 * to attach it, in case the attach blocks. 883 */ 884 kn->kn_status = KN_PROCESSING; 885 knote_attach(kn); 886 if ((error = filter_attach(kn)) != 0) { 887 kn->kn_status |= KN_DELETING | KN_REPROCESS; 888 knote_drop(kn); 889 goto done; 890 } 891 892 /* 893 * Interlock against close races which either tried 894 * to remove our knote while we were blocked or missed 895 * it entirely prior to our attachment. We do not 896 * want to end up with a knote on a closed descriptor. 897 */ 898 if ((fops->f_flags & FILTEROP_ISFD) && 899 checkfdclosed(fdp, kev->ident, kn->kn_fp)) { 900 kn->kn_status |= KN_DELETING | KN_REPROCESS; 901 } 902 } else { 903 /* 904 * The user may change some filter values after the 905 * initial EV_ADD, but doing so will not reset any 906 * filter which have already been triggered. 907 */ 908 KKASSERT(kn->kn_status & KN_PROCESSING); 909 kn->kn_sfflags = kev->fflags; 910 kn->kn_sdata = kev->data; 911 kn->kn_kevent.udata = kev->udata; 912 } 913 914 /* 915 * Execute the filter event to immediately activate the 916 * knote if necessary. If reprocessing events are pending 917 * due to blocking above we do not run the filter here 918 * but instead let knote_release() do it. Otherwise we 919 * might run the filter on a deleted event. 920 */ 921 if ((kn->kn_status & KN_REPROCESS) == 0) { 922 if (filter_event(kn, 0)) 923 KNOTE_ACTIVATE(kn); 924 } 925 } else if (kev->flags & EV_DELETE) { 926 /* 927 * Delete the existing knote 928 */ 929 knote_detach_and_drop(kn); 930 goto done; 931 } 932 933 /* 934 * Disablement does not deactivate a knote here. 935 */ 936 if ((kev->flags & EV_DISABLE) && 937 ((kn->kn_status & KN_DISABLED) == 0)) { 938 kn->kn_status |= KN_DISABLED; 939 } 940 941 /* 942 * Re-enablement may have to immediately enqueue an active knote. 943 */ 944 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 945 kn->kn_status &= ~KN_DISABLED; 946 if ((kn->kn_status & KN_ACTIVE) && 947 ((kn->kn_status & KN_QUEUED) == 0)) { 948 knote_enqueue(kn); 949 } 950 } 951 952 /* 953 * Handle any required reprocessing 954 */ 955 knote_release(kn); 956 /* kn may be invalid now */ 957 958 done: 959 lwkt_reltoken(&kq_token); 960 if (fp != NULL) 961 fdrop(fp); 962 return (error); 963 } 964 965 /* 966 * Block as necessary until the target time is reached. 967 * If tsp is NULL we block indefinitely. If tsp->ts_secs/nsecs are both 968 * 0 we do not block at all. 969 */ 970 static int 971 kqueue_sleep(struct kqueue *kq, struct timespec *tsp) 972 { 973 int error = 0; 974 975 if (tsp == NULL) { 976 kq->kq_state |= KQ_SLEEP; 977 error = tsleep(kq, PCATCH, "kqread", 0); 978 } else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 979 error = EWOULDBLOCK; 980 } else { 981 struct timespec ats; 982 struct timespec atx = *tsp; 983 int timeout; 984 985 nanouptime(&ats); 986 timespecsub(&atx, &ats); 987 if (ats.tv_sec < 0) { 988 error = EWOULDBLOCK; 989 } else { 990 timeout = atx.tv_sec > 24 * 60 * 60 ? 991 24 * 60 * 60 * hz : tstohz_high(&atx); 992 kq->kq_state |= KQ_SLEEP; 993 error = tsleep(kq, PCATCH, "kqread", timeout); 994 } 995 } 996 997 /* don't restart after signals... */ 998 if (error == ERESTART) 999 return (EINTR); 1000 1001 return (error); 1002 } 1003 1004 /* 1005 * Scan the kqueue, return the number of active events placed in kevp up 1006 * to count. 1007 * 1008 * Continuous mode events may get recycled, do not continue scanning past 1009 * marker unless no events have been collected. 1010 */ 1011 static int 1012 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 1013 struct knote *marker) 1014 { 1015 struct knote *kn, local_marker; 1016 int total; 1017 1018 total = 0; 1019 local_marker.kn_filter = EVFILT_MARKER; 1020 local_marker.kn_status = KN_PROCESSING; 1021 1022 /* 1023 * Collect events. 1024 */ 1025 TAILQ_INSERT_HEAD(&kq->kq_knpend, &local_marker, kn_tqe); 1026 while (count) { 1027 kn = TAILQ_NEXT(&local_marker, kn_tqe); 1028 if (kn->kn_filter == EVFILT_MARKER) { 1029 /* Marker reached, we are done */ 1030 if (kn == marker) 1031 break; 1032 1033 /* Move local marker past some other threads marker */ 1034 kn = TAILQ_NEXT(kn, kn_tqe); 1035 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1036 TAILQ_INSERT_BEFORE(kn, &local_marker, kn_tqe); 1037 continue; 1038 } 1039 1040 /* 1041 * We can't skip a knote undergoing processing, otherwise 1042 * we risk not returning it when the user process expects 1043 * it should be returned. Sleep and retry. 1044 */ 1045 if (knote_acquire(kn) == 0) 1046 continue; 1047 1048 /* 1049 * Remove the event for processing. 1050 * 1051 * WARNING! We must leave KN_QUEUED set to prevent the 1052 * event from being KNOTE_ACTIVATE()d while 1053 * the queue state is in limbo, in case we 1054 * block. 1055 * 1056 * WARNING! We must set KN_PROCESSING to avoid races 1057 * against deletion or another thread's 1058 * processing. 1059 */ 1060 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1061 kq->kq_count--; 1062 1063 /* 1064 * We have to deal with an extremely important race against 1065 * file descriptor close()s here. The file descriptor can 1066 * disappear MPSAFE, and there is a small window of 1067 * opportunity between that and the call to knote_fdclose(). 1068 * 1069 * If we hit that window here while doselect or dopoll is 1070 * trying to delete a spurious event they will not be able 1071 * to match up the event against a knote and will go haywire. 1072 */ 1073 if ((kn->kn_fop->f_flags & FILTEROP_ISFD) && 1074 checkfdclosed(kq->kq_fdp, kn->kn_kevent.ident, kn->kn_fp)) { 1075 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1076 } 1077 1078 if (kn->kn_status & KN_DISABLED) { 1079 /* 1080 * If disabled we ensure the event is not queued 1081 * but leave its active bit set. On re-enablement 1082 * the event may be immediately triggered. 1083 */ 1084 kn->kn_status &= ~KN_QUEUED; 1085 } else if ((kn->kn_flags & EV_ONESHOT) == 0 && 1086 (kn->kn_status & KN_DELETING) == 0 && 1087 filter_event(kn, 0) == 0) { 1088 /* 1089 * If not running in one-shot mode and the event 1090 * is no longer present we ensure it is removed 1091 * from the queue and ignore it. 1092 */ 1093 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1094 } else { 1095 /* 1096 * Post the event 1097 */ 1098 *kevp++ = kn->kn_kevent; 1099 ++total; 1100 --count; 1101 1102 if (kn->kn_flags & EV_ONESHOT) { 1103 kn->kn_status &= ~KN_QUEUED; 1104 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1105 } else if (kn->kn_flags & EV_CLEAR) { 1106 kn->kn_data = 0; 1107 kn->kn_fflags = 0; 1108 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1109 } else { 1110 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1111 kq->kq_count++; 1112 } 1113 } 1114 1115 /* 1116 * Handle any post-processing states 1117 */ 1118 knote_release(kn); 1119 } 1120 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1121 1122 return (total); 1123 } 1124 1125 /* 1126 * XXX 1127 * This could be expanded to call kqueue_scan, if desired. 1128 * 1129 * MPSAFE 1130 */ 1131 static int 1132 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1133 { 1134 return (ENXIO); 1135 } 1136 1137 /* 1138 * MPSAFE 1139 */ 1140 static int 1141 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1142 { 1143 return (ENXIO); 1144 } 1145 1146 /* 1147 * MPALMOSTSAFE 1148 */ 1149 static int 1150 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 1151 struct ucred *cred, struct sysmsg *msg) 1152 { 1153 struct kqueue *kq; 1154 int error; 1155 1156 lwkt_gettoken(&kq_token); 1157 kq = (struct kqueue *)fp->f_data; 1158 1159 switch(com) { 1160 case FIOASYNC: 1161 if (*(int *)data) 1162 kq->kq_state |= KQ_ASYNC; 1163 else 1164 kq->kq_state &= ~KQ_ASYNC; 1165 error = 0; 1166 break; 1167 case FIOSETOWN: 1168 error = fsetown(*(int *)data, &kq->kq_sigio); 1169 break; 1170 default: 1171 error = ENOTTY; 1172 break; 1173 } 1174 lwkt_reltoken(&kq_token); 1175 return (error); 1176 } 1177 1178 /* 1179 * MPSAFE 1180 */ 1181 static int 1182 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred) 1183 { 1184 struct kqueue *kq = (struct kqueue *)fp->f_data; 1185 1186 bzero((void *)st, sizeof(*st)); 1187 st->st_size = kq->kq_count; 1188 st->st_blksize = sizeof(struct kevent); 1189 st->st_mode = S_IFIFO; 1190 return (0); 1191 } 1192 1193 /* 1194 * MPSAFE 1195 */ 1196 static int 1197 kqueue_close(struct file *fp) 1198 { 1199 struct kqueue *kq = (struct kqueue *)fp->f_data; 1200 1201 kqueue_terminate(kq); 1202 1203 fp->f_data = NULL; 1204 funsetown(&kq->kq_sigio); 1205 1206 kfree(kq, M_KQUEUE); 1207 return (0); 1208 } 1209 1210 static void 1211 kqueue_wakeup(struct kqueue *kq) 1212 { 1213 if (kq->kq_state & KQ_SLEEP) { 1214 kq->kq_state &= ~KQ_SLEEP; 1215 wakeup(kq); 1216 } 1217 KNOTE(&kq->kq_kqinfo.ki_note, 0); 1218 } 1219 1220 /* 1221 * Calls filterops f_attach function, acquiring mplock if filter is not 1222 * marked as FILTEROP_MPSAFE. 1223 */ 1224 static int 1225 filter_attach(struct knote *kn) 1226 { 1227 int ret; 1228 1229 if (!(kn->kn_fop->f_flags & FILTEROP_MPSAFE)) { 1230 get_mplock(); 1231 ret = kn->kn_fop->f_attach(kn); 1232 rel_mplock(); 1233 } else { 1234 ret = kn->kn_fop->f_attach(kn); 1235 } 1236 1237 return (ret); 1238 } 1239 1240 /* 1241 * Detach the knote and drop it, destroying the knote. 1242 * 1243 * Calls filterops f_detach function, acquiring mplock if filter is not 1244 * marked as FILTEROP_MPSAFE. 1245 */ 1246 static void 1247 knote_detach_and_drop(struct knote *kn) 1248 { 1249 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1250 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1251 kn->kn_fop->f_detach(kn); 1252 } else { 1253 get_mplock(); 1254 kn->kn_fop->f_detach(kn); 1255 rel_mplock(); 1256 } 1257 knote_drop(kn); 1258 } 1259 1260 /* 1261 * Calls filterops f_event function, acquiring mplock if filter is not 1262 * marked as FILTEROP_MPSAFE. 1263 * 1264 * If the knote is in the middle of being created or deleted we cannot 1265 * safely call the filter op. 1266 */ 1267 static int 1268 filter_event(struct knote *kn, long hint) 1269 { 1270 int ret; 1271 1272 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1273 ret = kn->kn_fop->f_event(kn, hint); 1274 } else { 1275 get_mplock(); 1276 ret = kn->kn_fop->f_event(kn, hint); 1277 rel_mplock(); 1278 } 1279 return (ret); 1280 } 1281 1282 /* 1283 * Walk down a list of knotes, activating them if their event has triggered. 1284 * 1285 * If we encounter any knotes which are undergoing processing we just mark 1286 * them for reprocessing and do not try to [re]activate the knote. However, 1287 * if a hint is being passed we have to wait and that makes things a bit 1288 * sticky. 1289 */ 1290 void 1291 knote(struct klist *list, long hint) 1292 { 1293 struct knote *kn; 1294 1295 lwkt_gettoken(&kq_token); 1296 restart: 1297 SLIST_FOREACH(kn, list, kn_next) { 1298 if (kn->kn_status & KN_PROCESSING) { 1299 /* 1300 * Someone else is processing the knote, ask the 1301 * other thread to reprocess it and don't mess 1302 * with it otherwise. 1303 */ 1304 if (hint == 0) { 1305 kn->kn_status |= KN_REPROCESS; 1306 continue; 1307 } 1308 1309 /* 1310 * If the hint is non-zero we have to wait or risk 1311 * losing the state the caller is trying to update. 1312 * 1313 * XXX This is a real problem, certain process 1314 * and signal filters will bump kn_data for 1315 * already-processed notes more than once if 1316 * we restart the list scan. FIXME. 1317 */ 1318 kn->kn_status |= KN_WAITING | KN_REPROCESS; 1319 tsleep(kn, 0, "knotec", hz); 1320 goto restart; 1321 } 1322 1323 /* 1324 * Become the reprocessing master ourselves. 1325 * 1326 * If hint is non-zer running the event is mandatory 1327 * when not deleting so do it whether reprocessing is 1328 * set or not. 1329 */ 1330 kn->kn_status |= KN_PROCESSING; 1331 if ((kn->kn_status & KN_DELETING) == 0) { 1332 if (filter_event(kn, hint)) 1333 KNOTE_ACTIVATE(kn); 1334 } 1335 if (knote_release(kn)) 1336 goto restart; 1337 } 1338 lwkt_reltoken(&kq_token); 1339 } 1340 1341 /* 1342 * Insert knote at head of klist. 1343 * 1344 * This function may only be called via a filter function and thus 1345 * kq_token should already be held and marked for processing. 1346 */ 1347 void 1348 knote_insert(struct klist *klist, struct knote *kn) 1349 { 1350 KKASSERT(kn->kn_status & KN_PROCESSING); 1351 ASSERT_LWKT_TOKEN_HELD(&kq_token); 1352 SLIST_INSERT_HEAD(klist, kn, kn_next); 1353 } 1354 1355 /* 1356 * Remove knote from a klist 1357 * 1358 * This function may only be called via a filter function and thus 1359 * kq_token should already be held and marked for processing. 1360 */ 1361 void 1362 knote_remove(struct klist *klist, struct knote *kn) 1363 { 1364 KKASSERT(kn->kn_status & KN_PROCESSING); 1365 ASSERT_LWKT_TOKEN_HELD(&kq_token); 1366 SLIST_REMOVE(klist, kn, knote, kn_next); 1367 } 1368 1369 /* 1370 * Remove all knotes from a specified klist 1371 * 1372 * Only called from aio. 1373 */ 1374 void 1375 knote_empty(struct klist *list) 1376 { 1377 struct knote *kn; 1378 1379 lwkt_gettoken(&kq_token); 1380 while ((kn = SLIST_FIRST(list)) != NULL) { 1381 if (knote_acquire(kn)) 1382 knote_detach_and_drop(kn); 1383 } 1384 lwkt_reltoken(&kq_token); 1385 } 1386 1387 void 1388 knote_assume_knotes(struct kqinfo *src, struct kqinfo *dst, 1389 struct filterops *ops, void *hook) 1390 { 1391 struct knote *kn; 1392 1393 lwkt_gettoken(&kq_token); 1394 while ((kn = SLIST_FIRST(&src->ki_note)) != NULL) { 1395 if (knote_acquire(kn)) { 1396 knote_remove(&src->ki_note, kn); 1397 kn->kn_fop = ops; 1398 kn->kn_hook = hook; 1399 knote_insert(&dst->ki_note, kn); 1400 knote_release(kn); 1401 /* kn may be invalid now */ 1402 } 1403 } 1404 lwkt_reltoken(&kq_token); 1405 } 1406 1407 /* 1408 * Remove all knotes referencing a specified fd 1409 */ 1410 void 1411 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd) 1412 { 1413 struct knote *kn; 1414 1415 lwkt_gettoken(&kq_token); 1416 restart: 1417 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 1418 if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) { 1419 if (knote_acquire(kn)) 1420 knote_detach_and_drop(kn); 1421 goto restart; 1422 } 1423 } 1424 lwkt_reltoken(&kq_token); 1425 } 1426 1427 /* 1428 * Low level attach function. 1429 * 1430 * The knote should already be marked for processing. 1431 */ 1432 static void 1433 knote_attach(struct knote *kn) 1434 { 1435 struct klist *list; 1436 struct kqueue *kq = kn->kn_kq; 1437 1438 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1439 KKASSERT(kn->kn_fp); 1440 list = &kn->kn_fp->f_klist; 1441 } else { 1442 if (kq->kq_knhashmask == 0) 1443 kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1444 &kq->kq_knhashmask); 1445 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1446 } 1447 SLIST_INSERT_HEAD(list, kn, kn_link); 1448 TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink); 1449 } 1450 1451 /* 1452 * Low level drop function. 1453 * 1454 * The knote should already be marked for processing. 1455 */ 1456 static void 1457 knote_drop(struct knote *kn) 1458 { 1459 struct kqueue *kq; 1460 struct klist *list; 1461 1462 kq = kn->kn_kq; 1463 1464 if (kn->kn_fop->f_flags & FILTEROP_ISFD) 1465 list = &kn->kn_fp->f_klist; 1466 else 1467 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1468 1469 SLIST_REMOVE(list, kn, knote, kn_link); 1470 TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink); 1471 if (kn->kn_status & KN_QUEUED) 1472 knote_dequeue(kn); 1473 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1474 fdrop(kn->kn_fp); 1475 kn->kn_fp = NULL; 1476 } 1477 knote_free(kn); 1478 } 1479 1480 /* 1481 * Low level enqueue function. 1482 * 1483 * The knote should already be marked for processing. 1484 */ 1485 static void 1486 knote_enqueue(struct knote *kn) 1487 { 1488 struct kqueue *kq = kn->kn_kq; 1489 1490 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1491 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1492 kn->kn_status |= KN_QUEUED; 1493 ++kq->kq_count; 1494 1495 /* 1496 * Send SIGIO on request (typically set up as a mailbox signal) 1497 */ 1498 if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1) 1499 pgsigio(kq->kq_sigio, SIGIO, 0); 1500 1501 kqueue_wakeup(kq); 1502 } 1503 1504 /* 1505 * Low level dequeue function. 1506 * 1507 * The knote should already be marked for processing. 1508 */ 1509 static void 1510 knote_dequeue(struct knote *kn) 1511 { 1512 struct kqueue *kq = kn->kn_kq; 1513 1514 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1515 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1516 kn->kn_status &= ~KN_QUEUED; 1517 kq->kq_count--; 1518 } 1519 1520 static struct knote * 1521 knote_alloc(void) 1522 { 1523 return kmalloc(sizeof(struct knote), M_KQUEUE, M_WAITOK); 1524 } 1525 1526 static void 1527 knote_free(struct knote *kn) 1528 { 1529 kfree(kn, M_KQUEUE); 1530 } 1531