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 kn->kn_hook = NULL; 375 return (ENOMEM); 376 } 377 kq_ncallouts++; 378 379 tv.tv_sec = kn->kn_sdata / 1000; 380 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 381 tticks = tvtohz_high(&tv); 382 383 kn->kn_flags |= EV_CLEAR; /* automatically set */ 384 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 385 M_KQUEUE, M_WAITOK); 386 callout_init(calloutp); 387 kn->kn_hook = (caddr_t)calloutp; 388 callout_reset(calloutp, tticks, filt_timerexpire, kn); 389 390 return (0); 391 } 392 393 static void 394 filt_timerdetach(struct knote *kn) 395 { 396 struct callout *calloutp; 397 398 calloutp = (struct callout *)kn->kn_hook; 399 callout_stop(calloutp); 400 FREE(calloutp, M_KQUEUE); 401 kq_ncallouts--; 402 } 403 404 static int 405 filt_timer(struct knote *kn, long hint) 406 { 407 408 return (kn->kn_data != 0); 409 } 410 411 /* 412 * Acquire a knote, return non-zero on success, 0 on failure. 413 * 414 * If we cannot acquire the knote we sleep and return 0. The knote 415 * may be stale on return in this case and the caller must restart 416 * whatever loop they are in. 417 */ 418 static __inline 419 int 420 knote_acquire(struct knote *kn) 421 { 422 if (kn->kn_status & KN_PROCESSING) { 423 kn->kn_status |= KN_WAITING | KN_REPROCESS; 424 tsleep(kn, 0, "kqepts", hz); 425 /* knote may be stale now */ 426 return(0); 427 } 428 kn->kn_status |= KN_PROCESSING; 429 return(1); 430 } 431 432 /* 433 * Release an acquired knote, clearing KN_PROCESSING and handling any 434 * KN_REPROCESS events. 435 * 436 * Non-zero is returned if the knote is destroyed. 437 */ 438 static __inline 439 int 440 knote_release(struct knote *kn) 441 { 442 while (kn->kn_status & KN_REPROCESS) { 443 kn->kn_status &= ~KN_REPROCESS; 444 if (kn->kn_status & KN_WAITING) { 445 kn->kn_status &= ~KN_WAITING; 446 wakeup(kn); 447 } 448 if (kn->kn_status & KN_DELETING) { 449 knote_detach_and_drop(kn); 450 return(1); 451 /* NOT REACHED */ 452 } 453 if (filter_event(kn, 0)) 454 KNOTE_ACTIVATE(kn); 455 } 456 kn->kn_status &= ~KN_PROCESSING; 457 return(0); 458 } 459 460 /* 461 * Initialize a kqueue. 462 * 463 * NOTE: The lwp/proc code initializes a kqueue for select/poll ops. 464 * 465 * MPSAFE 466 */ 467 void 468 kqueue_init(struct kqueue *kq, struct filedesc *fdp) 469 { 470 TAILQ_INIT(&kq->kq_knpend); 471 TAILQ_INIT(&kq->kq_knlist); 472 kq->kq_count = 0; 473 kq->kq_fdp = fdp; 474 SLIST_INIT(&kq->kq_kqinfo.ki_note); 475 } 476 477 /* 478 * Terminate a kqueue. Freeing the actual kq itself is left up to the 479 * caller (it might be embedded in a lwp so we don't do it here). 480 * 481 * The kq's knlist must be completely eradicated so block on any 482 * processing races. 483 */ 484 void 485 kqueue_terminate(struct kqueue *kq) 486 { 487 struct knote *kn; 488 489 lwkt_gettoken(&kq_token); 490 while ((kn = TAILQ_FIRST(&kq->kq_knlist)) != NULL) { 491 if (knote_acquire(kn)) 492 knote_detach_and_drop(kn); 493 } 494 if (kq->kq_knhash) { 495 kfree(kq->kq_knhash, M_KQUEUE); 496 kq->kq_knhash = NULL; 497 kq->kq_knhashmask = 0; 498 } 499 lwkt_reltoken(&kq_token); 500 } 501 502 /* 503 * MPSAFE 504 */ 505 int 506 sys_kqueue(struct kqueue_args *uap) 507 { 508 struct thread *td = curthread; 509 struct kqueue *kq; 510 struct file *fp; 511 int fd, error; 512 513 error = falloc(td->td_lwp, &fp, &fd); 514 if (error) 515 return (error); 516 fp->f_flag = FREAD | FWRITE; 517 fp->f_type = DTYPE_KQUEUE; 518 fp->f_ops = &kqueueops; 519 520 kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 521 kqueue_init(kq, td->td_proc->p_fd); 522 fp->f_data = kq; 523 524 fsetfd(kq->kq_fdp, fp, fd); 525 uap->sysmsg_result = fd; 526 fdrop(fp); 527 return (error); 528 } 529 530 /* 531 * Copy 'count' items into the destination list pointed to by uap->eventlist. 532 */ 533 static int 534 kevent_copyout(void *arg, struct kevent *kevp, int count, int *res) 535 { 536 struct kevent_copyin_args *kap; 537 int error; 538 539 kap = (struct kevent_copyin_args *)arg; 540 541 error = copyout(kevp, kap->ka->eventlist, count * sizeof(*kevp)); 542 if (error == 0) { 543 kap->ka->eventlist += count; 544 *res += count; 545 } else { 546 *res = -1; 547 } 548 549 return (error); 550 } 551 552 /* 553 * Copy at most 'max' items from the list pointed to by kap->changelist, 554 * return number of items in 'events'. 555 */ 556 static int 557 kevent_copyin(void *arg, struct kevent *kevp, int max, int *events) 558 { 559 struct kevent_copyin_args *kap; 560 int error, count; 561 562 kap = (struct kevent_copyin_args *)arg; 563 564 count = min(kap->ka->nchanges - kap->pchanges, max); 565 error = copyin(kap->ka->changelist, kevp, count * sizeof *kevp); 566 if (error == 0) { 567 kap->ka->changelist += count; 568 kap->pchanges += count; 569 *events = count; 570 } 571 572 return (error); 573 } 574 575 /* 576 * MPSAFE 577 */ 578 int 579 kern_kevent(struct kqueue *kq, int nevents, int *res, void *uap, 580 k_copyin_fn kevent_copyinfn, k_copyout_fn kevent_copyoutfn, 581 struct timespec *tsp_in) 582 { 583 struct kevent *kevp; 584 struct timespec *tsp; 585 int i, n, total, error, nerrors = 0; 586 int lres; 587 int limit = kq_checkloop; 588 struct kevent kev[KQ_NEVENTS]; 589 struct knote marker; 590 591 tsp = tsp_in; 592 *res = 0; 593 594 lwkt_gettoken(&kq_token); 595 for ( ;; ) { 596 n = 0; 597 error = kevent_copyinfn(uap, kev, KQ_NEVENTS, &n); 598 if (error) 599 goto done; 600 if (n == 0) 601 break; 602 for (i = 0; i < n; i++) { 603 kevp = &kev[i]; 604 kevp->flags &= ~EV_SYSFLAGS; 605 error = kqueue_register(kq, kevp); 606 607 /* 608 * If a registration returns an error we 609 * immediately post the error. The kevent() 610 * call itself will fail with the error if 611 * no space is available for posting. 612 * 613 * Such errors normally bypass the timeout/blocking 614 * code. However, if the copyoutfn function refuses 615 * to post the error (see sys_poll()), then we 616 * ignore it too. 617 */ 618 if (error) { 619 kevp->flags = EV_ERROR; 620 kevp->data = error; 621 lres = *res; 622 kevent_copyoutfn(uap, kevp, 1, res); 623 if (lres != *res) { 624 nevents--; 625 nerrors++; 626 } 627 } 628 } 629 } 630 if (nerrors) { 631 error = 0; 632 goto done; 633 } 634 635 /* 636 * Acquire/wait for events - setup timeout 637 */ 638 if (tsp != NULL) { 639 struct timespec ats; 640 641 if (tsp->tv_sec || tsp->tv_nsec) { 642 nanouptime(&ats); 643 timespecadd(tsp, &ats); /* tsp = target time */ 644 } 645 } 646 647 /* 648 * Loop as required. 649 * 650 * Collect as many events as we can. Sleeping on successive 651 * loops is disabled if copyoutfn has incremented (*res). 652 * 653 * The loop stops if an error occurs, all events have been 654 * scanned (the marker has been reached), or fewer than the 655 * maximum number of events is found. 656 * 657 * The copyoutfn function does not have to increment (*res) in 658 * order for the loop to continue. 659 * 660 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents. 661 */ 662 total = 0; 663 error = 0; 664 marker.kn_filter = EVFILT_MARKER; 665 marker.kn_status = KN_PROCESSING; 666 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 667 while ((n = nevents - total) > 0) { 668 if (n > KQ_NEVENTS) 669 n = KQ_NEVENTS; 670 671 /* 672 * If no events are pending sleep until timeout (if any) 673 * or an event occurs. 674 * 675 * After the sleep completes the marker is moved to the 676 * end of the list, making any received events available 677 * to our scan. 678 */ 679 if (kq->kq_count == 0 && *res == 0) { 680 error = kqueue_sleep(kq, tsp); 681 if (error) 682 break; 683 684 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 685 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 686 } 687 688 /* 689 * Process all received events 690 * Account for all non-spurious events in our total 691 */ 692 i = kqueue_scan(kq, kev, n, &marker); 693 if (i) { 694 lres = *res; 695 error = kevent_copyoutfn(uap, kev, i, res); 696 total += *res - lres; 697 if (error) 698 break; 699 } 700 if (limit && --limit == 0) 701 panic("kqueue: checkloop failed i=%d", i); 702 703 /* 704 * Normally when fewer events are returned than requested 705 * we can stop. However, if only spurious events were 706 * collected the copyout will not bump (*res) and we have 707 * to continue. 708 */ 709 if (i < n && *res) 710 break; 711 712 /* 713 * Deal with an edge case where spurious events can cause 714 * a loop to occur without moving the marker. This can 715 * prevent kqueue_scan() from picking up new events which 716 * race us. We must be sure to move the marker for this 717 * case. 718 * 719 * NOTE: We do not want to move the marker if events 720 * were scanned because normal kqueue operations 721 * may reactivate events. Moving the marker in 722 * that case could result in duplicates for the 723 * same event. 724 */ 725 if (i == 0) { 726 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 727 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 728 } 729 } 730 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 731 732 /* Timeouts do not return EWOULDBLOCK. */ 733 if (error == EWOULDBLOCK) 734 error = 0; 735 736 done: 737 lwkt_reltoken(&kq_token); 738 return (error); 739 } 740 741 /* 742 * MPALMOSTSAFE 743 */ 744 int 745 sys_kevent(struct kevent_args *uap) 746 { 747 struct thread *td = curthread; 748 struct proc *p = td->td_proc; 749 struct timespec ts, *tsp; 750 struct kqueue *kq; 751 struct file *fp = NULL; 752 struct kevent_copyin_args *kap, ka; 753 int error; 754 755 if (uap->timeout) { 756 error = copyin(uap->timeout, &ts, sizeof(ts)); 757 if (error) 758 return (error); 759 tsp = &ts; 760 } else { 761 tsp = NULL; 762 } 763 764 fp = holdfp(p->p_fd, uap->fd, -1); 765 if (fp == NULL) 766 return (EBADF); 767 if (fp->f_type != DTYPE_KQUEUE) { 768 fdrop(fp); 769 return (EBADF); 770 } 771 772 kq = (struct kqueue *)fp->f_data; 773 774 kap = &ka; 775 kap->ka = uap; 776 kap->pchanges = 0; 777 778 error = kern_kevent(kq, uap->nevents, &uap->sysmsg_result, kap, 779 kevent_copyin, kevent_copyout, tsp); 780 781 fdrop(fp); 782 783 return (error); 784 } 785 786 int 787 kqueue_register(struct kqueue *kq, struct kevent *kev) 788 { 789 struct filedesc *fdp = kq->kq_fdp; 790 struct filterops *fops; 791 struct file *fp = NULL; 792 struct knote *kn = NULL; 793 int error = 0; 794 795 if (kev->filter < 0) { 796 if (kev->filter + EVFILT_SYSCOUNT < 0) 797 return (EINVAL); 798 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 799 } else { 800 /* 801 * XXX 802 * filter attach routine is responsible for insuring that 803 * the identifier can be attached to it. 804 */ 805 kprintf("unknown filter: %d\n", kev->filter); 806 return (EINVAL); 807 } 808 809 lwkt_gettoken(&kq_token); 810 if (fops->f_flags & FILTEROP_ISFD) { 811 /* validate descriptor */ 812 fp = holdfp(fdp, kev->ident, -1); 813 if (fp == NULL) { 814 lwkt_reltoken(&kq_token); 815 return (EBADF); 816 } 817 818 again1: 819 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 820 if (kn->kn_kq == kq && 821 kn->kn_filter == kev->filter && 822 kn->kn_id == kev->ident) { 823 if (knote_acquire(kn) == 0) 824 goto again1; 825 break; 826 } 827 } 828 } else { 829 if (kq->kq_knhashmask) { 830 struct klist *list; 831 832 list = &kq->kq_knhash[ 833 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 834 again2: 835 SLIST_FOREACH(kn, list, kn_link) { 836 if (kn->kn_id == kev->ident && 837 kn->kn_filter == kev->filter) { 838 if (knote_acquire(kn) == 0) 839 goto again2; 840 break; 841 } 842 } 843 } 844 } 845 846 /* 847 * NOTE: At this point if kn is non-NULL we will have acquired 848 * it and set KN_PROCESSING. 849 */ 850 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 851 error = ENOENT; 852 goto done; 853 } 854 855 /* 856 * kn now contains the matching knote, or NULL if no match 857 */ 858 if (kev->flags & EV_ADD) { 859 if (kn == NULL) { 860 kn = knote_alloc(); 861 if (kn == NULL) { 862 error = ENOMEM; 863 goto done; 864 } 865 kn->kn_fp = fp; 866 kn->kn_kq = kq; 867 kn->kn_fop = fops; 868 869 /* 870 * apply reference count to knote structure, and 871 * do not release it at the end of this routine. 872 */ 873 fp = NULL; 874 875 kn->kn_sfflags = kev->fflags; 876 kn->kn_sdata = kev->data; 877 kev->fflags = 0; 878 kev->data = 0; 879 kn->kn_kevent = *kev; 880 881 /* 882 * KN_PROCESSING prevents the knote from getting 883 * ripped out from under us while we are trying 884 * to attach it, in case the attach blocks. 885 */ 886 kn->kn_status = KN_PROCESSING; 887 knote_attach(kn); 888 if ((error = filter_attach(kn)) != 0) { 889 kn->kn_status |= KN_DELETING | KN_REPROCESS; 890 knote_drop(kn); 891 goto done; 892 } 893 894 /* 895 * Interlock against close races which either tried 896 * to remove our knote while we were blocked or missed 897 * it entirely prior to our attachment. We do not 898 * want to end up with a knote on a closed descriptor. 899 */ 900 if ((fops->f_flags & FILTEROP_ISFD) && 901 checkfdclosed(fdp, kev->ident, kn->kn_fp)) { 902 kn->kn_status |= KN_DELETING | KN_REPROCESS; 903 } 904 } else { 905 /* 906 * The user may change some filter values after the 907 * initial EV_ADD, but doing so will not reset any 908 * filter which have already been triggered. 909 */ 910 KKASSERT(kn->kn_status & KN_PROCESSING); 911 kn->kn_sfflags = kev->fflags; 912 kn->kn_sdata = kev->data; 913 kn->kn_kevent.udata = kev->udata; 914 } 915 916 /* 917 * Execute the filter event to immediately activate the 918 * knote if necessary. If reprocessing events are pending 919 * due to blocking above we do not run the filter here 920 * but instead let knote_release() do it. Otherwise we 921 * might run the filter on a deleted event. 922 */ 923 if ((kn->kn_status & KN_REPROCESS) == 0) { 924 if (filter_event(kn, 0)) 925 KNOTE_ACTIVATE(kn); 926 } 927 } else if (kev->flags & EV_DELETE) { 928 /* 929 * Delete the existing knote 930 */ 931 knote_detach_and_drop(kn); 932 goto done; 933 } 934 935 /* 936 * Disablement does not deactivate a knote here. 937 */ 938 if ((kev->flags & EV_DISABLE) && 939 ((kn->kn_status & KN_DISABLED) == 0)) { 940 kn->kn_status |= KN_DISABLED; 941 } 942 943 /* 944 * Re-enablement may have to immediately enqueue an active knote. 945 */ 946 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 947 kn->kn_status &= ~KN_DISABLED; 948 if ((kn->kn_status & KN_ACTIVE) && 949 ((kn->kn_status & KN_QUEUED) == 0)) { 950 knote_enqueue(kn); 951 } 952 } 953 954 /* 955 * Handle any required reprocessing 956 */ 957 knote_release(kn); 958 /* kn may be invalid now */ 959 960 done: 961 lwkt_reltoken(&kq_token); 962 if (fp != NULL) 963 fdrop(fp); 964 return (error); 965 } 966 967 /* 968 * Block as necessary until the target time is reached. 969 * If tsp is NULL we block indefinitely. If tsp->ts_secs/nsecs are both 970 * 0 we do not block at all. 971 */ 972 static int 973 kqueue_sleep(struct kqueue *kq, struct timespec *tsp) 974 { 975 int error = 0; 976 977 if (tsp == NULL) { 978 kq->kq_state |= KQ_SLEEP; 979 error = tsleep(kq, PCATCH, "kqread", 0); 980 } else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 981 error = EWOULDBLOCK; 982 } else { 983 struct timespec ats; 984 struct timespec atx = *tsp; 985 int timeout; 986 987 nanouptime(&ats); 988 timespecsub(&atx, &ats); 989 if (ats.tv_sec < 0) { 990 error = EWOULDBLOCK; 991 } else { 992 timeout = atx.tv_sec > 24 * 60 * 60 ? 993 24 * 60 * 60 * hz : tstohz_high(&atx); 994 kq->kq_state |= KQ_SLEEP; 995 error = tsleep(kq, PCATCH, "kqread", timeout); 996 } 997 } 998 999 /* don't restart after signals... */ 1000 if (error == ERESTART) 1001 return (EINTR); 1002 1003 return (error); 1004 } 1005 1006 /* 1007 * Scan the kqueue, return the number of active events placed in kevp up 1008 * to count. 1009 * 1010 * Continuous mode events may get recycled, do not continue scanning past 1011 * marker unless no events have been collected. 1012 */ 1013 static int 1014 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 1015 struct knote *marker) 1016 { 1017 struct knote *kn, local_marker; 1018 int total; 1019 1020 total = 0; 1021 local_marker.kn_filter = EVFILT_MARKER; 1022 local_marker.kn_status = KN_PROCESSING; 1023 1024 /* 1025 * Collect events. 1026 */ 1027 TAILQ_INSERT_HEAD(&kq->kq_knpend, &local_marker, kn_tqe); 1028 while (count) { 1029 kn = TAILQ_NEXT(&local_marker, kn_tqe); 1030 if (kn->kn_filter == EVFILT_MARKER) { 1031 /* Marker reached, we are done */ 1032 if (kn == marker) 1033 break; 1034 1035 /* Move local marker past some other threads marker */ 1036 kn = TAILQ_NEXT(kn, kn_tqe); 1037 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1038 TAILQ_INSERT_BEFORE(kn, &local_marker, kn_tqe); 1039 continue; 1040 } 1041 1042 /* 1043 * We can't skip a knote undergoing processing, otherwise 1044 * we risk not returning it when the user process expects 1045 * it should be returned. Sleep and retry. 1046 */ 1047 if (knote_acquire(kn) == 0) 1048 continue; 1049 1050 /* 1051 * Remove the event for processing. 1052 * 1053 * WARNING! We must leave KN_QUEUED set to prevent the 1054 * event from being KNOTE_ACTIVATE()d while 1055 * the queue state is in limbo, in case we 1056 * block. 1057 * 1058 * WARNING! We must set KN_PROCESSING to avoid races 1059 * against deletion or another thread's 1060 * processing. 1061 */ 1062 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1063 kq->kq_count--; 1064 1065 /* 1066 * We have to deal with an extremely important race against 1067 * file descriptor close()s here. The file descriptor can 1068 * disappear MPSAFE, and there is a small window of 1069 * opportunity between that and the call to knote_fdclose(). 1070 * 1071 * If we hit that window here while doselect or dopoll is 1072 * trying to delete a spurious event they will not be able 1073 * to match up the event against a knote and will go haywire. 1074 */ 1075 if ((kn->kn_fop->f_flags & FILTEROP_ISFD) && 1076 checkfdclosed(kq->kq_fdp, kn->kn_kevent.ident, kn->kn_fp)) { 1077 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1078 } 1079 1080 if (kn->kn_status & KN_DISABLED) { 1081 /* 1082 * If disabled we ensure the event is not queued 1083 * but leave its active bit set. On re-enablement 1084 * the event may be immediately triggered. 1085 */ 1086 kn->kn_status &= ~KN_QUEUED; 1087 } else if ((kn->kn_flags & EV_ONESHOT) == 0 && 1088 (kn->kn_status & KN_DELETING) == 0 && 1089 filter_event(kn, 0) == 0) { 1090 /* 1091 * If not running in one-shot mode and the event 1092 * is no longer present we ensure it is removed 1093 * from the queue and ignore it. 1094 */ 1095 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1096 } else { 1097 /* 1098 * Post the event 1099 */ 1100 *kevp++ = kn->kn_kevent; 1101 ++total; 1102 --count; 1103 1104 if (kn->kn_flags & EV_ONESHOT) { 1105 kn->kn_status &= ~KN_QUEUED; 1106 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1107 } else if (kn->kn_flags & EV_CLEAR) { 1108 kn->kn_data = 0; 1109 kn->kn_fflags = 0; 1110 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1111 } else { 1112 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1113 kq->kq_count++; 1114 } 1115 } 1116 1117 /* 1118 * Handle any post-processing states 1119 */ 1120 knote_release(kn); 1121 } 1122 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1123 1124 return (total); 1125 } 1126 1127 /* 1128 * XXX 1129 * This could be expanded to call kqueue_scan, if desired. 1130 * 1131 * MPSAFE 1132 */ 1133 static int 1134 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1135 { 1136 return (ENXIO); 1137 } 1138 1139 /* 1140 * MPSAFE 1141 */ 1142 static int 1143 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1144 { 1145 return (ENXIO); 1146 } 1147 1148 /* 1149 * MPALMOSTSAFE 1150 */ 1151 static int 1152 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 1153 struct ucred *cred, struct sysmsg *msg) 1154 { 1155 struct kqueue *kq; 1156 int error; 1157 1158 lwkt_gettoken(&kq_token); 1159 kq = (struct kqueue *)fp->f_data; 1160 1161 switch(com) { 1162 case FIOASYNC: 1163 if (*(int *)data) 1164 kq->kq_state |= KQ_ASYNC; 1165 else 1166 kq->kq_state &= ~KQ_ASYNC; 1167 error = 0; 1168 break; 1169 case FIOSETOWN: 1170 error = fsetown(*(int *)data, &kq->kq_sigio); 1171 break; 1172 default: 1173 error = ENOTTY; 1174 break; 1175 } 1176 lwkt_reltoken(&kq_token); 1177 return (error); 1178 } 1179 1180 /* 1181 * MPSAFE 1182 */ 1183 static int 1184 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred) 1185 { 1186 struct kqueue *kq = (struct kqueue *)fp->f_data; 1187 1188 bzero((void *)st, sizeof(*st)); 1189 st->st_size = kq->kq_count; 1190 st->st_blksize = sizeof(struct kevent); 1191 st->st_mode = S_IFIFO; 1192 return (0); 1193 } 1194 1195 /* 1196 * MPSAFE 1197 */ 1198 static int 1199 kqueue_close(struct file *fp) 1200 { 1201 struct kqueue *kq = (struct kqueue *)fp->f_data; 1202 1203 kqueue_terminate(kq); 1204 1205 fp->f_data = NULL; 1206 funsetown(&kq->kq_sigio); 1207 1208 kfree(kq, M_KQUEUE); 1209 return (0); 1210 } 1211 1212 static void 1213 kqueue_wakeup(struct kqueue *kq) 1214 { 1215 if (kq->kq_state & KQ_SLEEP) { 1216 kq->kq_state &= ~KQ_SLEEP; 1217 wakeup(kq); 1218 } 1219 KNOTE(&kq->kq_kqinfo.ki_note, 0); 1220 } 1221 1222 /* 1223 * Calls filterops f_attach function, acquiring mplock if filter is not 1224 * marked as FILTEROP_MPSAFE. 1225 */ 1226 static int 1227 filter_attach(struct knote *kn) 1228 { 1229 int ret; 1230 1231 if (!(kn->kn_fop->f_flags & FILTEROP_MPSAFE)) { 1232 get_mplock(); 1233 ret = kn->kn_fop->f_attach(kn); 1234 rel_mplock(); 1235 } else { 1236 ret = kn->kn_fop->f_attach(kn); 1237 } 1238 1239 return (ret); 1240 } 1241 1242 /* 1243 * Detach the knote and drop it, destroying the knote. 1244 * 1245 * Calls filterops f_detach function, acquiring mplock if filter is not 1246 * marked as FILTEROP_MPSAFE. 1247 */ 1248 static void 1249 knote_detach_and_drop(struct knote *kn) 1250 { 1251 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1252 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1253 kn->kn_fop->f_detach(kn); 1254 } else { 1255 get_mplock(); 1256 kn->kn_fop->f_detach(kn); 1257 rel_mplock(); 1258 } 1259 knote_drop(kn); 1260 } 1261 1262 /* 1263 * Calls filterops f_event function, acquiring mplock if filter is not 1264 * marked as FILTEROP_MPSAFE. 1265 * 1266 * If the knote is in the middle of being created or deleted we cannot 1267 * safely call the filter op. 1268 */ 1269 static int 1270 filter_event(struct knote *kn, long hint) 1271 { 1272 int ret; 1273 1274 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1275 ret = kn->kn_fop->f_event(kn, hint); 1276 } else { 1277 get_mplock(); 1278 ret = kn->kn_fop->f_event(kn, hint); 1279 rel_mplock(); 1280 } 1281 return (ret); 1282 } 1283 1284 /* 1285 * Walk down a list of knotes, activating them if their event has triggered. 1286 * 1287 * If we encounter any knotes which are undergoing processing we just mark 1288 * them for reprocessing and do not try to [re]activate the knote. However, 1289 * if a hint is being passed we have to wait and that makes things a bit 1290 * sticky. 1291 */ 1292 void 1293 knote(struct klist *list, long hint) 1294 { 1295 struct knote *kn; 1296 1297 lwkt_gettoken(&kq_token); 1298 restart: 1299 SLIST_FOREACH(kn, list, kn_next) { 1300 if (kn->kn_status & KN_PROCESSING) { 1301 /* 1302 * Someone else is processing the knote, ask the 1303 * other thread to reprocess it and don't mess 1304 * with it otherwise. 1305 */ 1306 if (hint == 0) { 1307 kn->kn_status |= KN_REPROCESS; 1308 continue; 1309 } 1310 1311 /* 1312 * If the hint is non-zero we have to wait or risk 1313 * losing the state the caller is trying to update. 1314 * 1315 * XXX This is a real problem, certain process 1316 * and signal filters will bump kn_data for 1317 * already-processed notes more than once if 1318 * we restart the list scan. FIXME. 1319 */ 1320 kn->kn_status |= KN_WAITING | KN_REPROCESS; 1321 tsleep(kn, 0, "knotec", hz); 1322 goto restart; 1323 } 1324 1325 /* 1326 * Become the reprocessing master ourselves. 1327 * 1328 * If hint is non-zer running the event is mandatory 1329 * when not deleting so do it whether reprocessing is 1330 * set or not. 1331 */ 1332 kn->kn_status |= KN_PROCESSING; 1333 if ((kn->kn_status & KN_DELETING) == 0) { 1334 if (filter_event(kn, hint)) 1335 KNOTE_ACTIVATE(kn); 1336 } 1337 if (knote_release(kn)) 1338 goto restart; 1339 } 1340 lwkt_reltoken(&kq_token); 1341 } 1342 1343 /* 1344 * Insert knote at head of klist. 1345 * 1346 * This function may only be called via a filter function and thus 1347 * kq_token should already be held and marked for processing. 1348 */ 1349 void 1350 knote_insert(struct klist *klist, struct knote *kn) 1351 { 1352 KKASSERT(kn->kn_status & KN_PROCESSING); 1353 ASSERT_LWKT_TOKEN_HELD(&kq_token); 1354 SLIST_INSERT_HEAD(klist, kn, kn_next); 1355 } 1356 1357 /* 1358 * Remove knote from a klist 1359 * 1360 * This function may only be called via a filter function and thus 1361 * kq_token should already be held and marked for processing. 1362 */ 1363 void 1364 knote_remove(struct klist *klist, struct knote *kn) 1365 { 1366 KKASSERT(kn->kn_status & KN_PROCESSING); 1367 ASSERT_LWKT_TOKEN_HELD(&kq_token); 1368 SLIST_REMOVE(klist, kn, knote, kn_next); 1369 } 1370 1371 /* 1372 * Remove all knotes from a specified klist 1373 * 1374 * Only called from aio. 1375 */ 1376 void 1377 knote_empty(struct klist *list) 1378 { 1379 struct knote *kn; 1380 1381 lwkt_gettoken(&kq_token); 1382 while ((kn = SLIST_FIRST(list)) != NULL) { 1383 if (knote_acquire(kn)) 1384 knote_detach_and_drop(kn); 1385 } 1386 lwkt_reltoken(&kq_token); 1387 } 1388 1389 void 1390 knote_assume_knotes(struct kqinfo *src, struct kqinfo *dst, 1391 struct filterops *ops, void *hook) 1392 { 1393 struct knote *kn; 1394 1395 lwkt_gettoken(&kq_token); 1396 while ((kn = SLIST_FIRST(&src->ki_note)) != NULL) { 1397 if (knote_acquire(kn)) { 1398 knote_remove(&src->ki_note, kn); 1399 kn->kn_fop = ops; 1400 kn->kn_hook = hook; 1401 knote_insert(&dst->ki_note, kn); 1402 knote_release(kn); 1403 /* kn may be invalid now */ 1404 } 1405 } 1406 lwkt_reltoken(&kq_token); 1407 } 1408 1409 /* 1410 * Remove all knotes referencing a specified fd 1411 */ 1412 void 1413 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd) 1414 { 1415 struct knote *kn; 1416 1417 lwkt_gettoken(&kq_token); 1418 restart: 1419 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 1420 if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) { 1421 if (knote_acquire(kn)) 1422 knote_detach_and_drop(kn); 1423 goto restart; 1424 } 1425 } 1426 lwkt_reltoken(&kq_token); 1427 } 1428 1429 /* 1430 * Low level attach function. 1431 * 1432 * The knote should already be marked for processing. 1433 */ 1434 static void 1435 knote_attach(struct knote *kn) 1436 { 1437 struct klist *list; 1438 struct kqueue *kq = kn->kn_kq; 1439 1440 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1441 KKASSERT(kn->kn_fp); 1442 list = &kn->kn_fp->f_klist; 1443 } else { 1444 if (kq->kq_knhashmask == 0) 1445 kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1446 &kq->kq_knhashmask); 1447 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1448 } 1449 SLIST_INSERT_HEAD(list, kn, kn_link); 1450 TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink); 1451 } 1452 1453 /* 1454 * Low level drop function. 1455 * 1456 * The knote should already be marked for processing. 1457 */ 1458 static void 1459 knote_drop(struct knote *kn) 1460 { 1461 struct kqueue *kq; 1462 struct klist *list; 1463 1464 kq = kn->kn_kq; 1465 1466 if (kn->kn_fop->f_flags & FILTEROP_ISFD) 1467 list = &kn->kn_fp->f_klist; 1468 else 1469 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1470 1471 SLIST_REMOVE(list, kn, knote, kn_link); 1472 TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink); 1473 if (kn->kn_status & KN_QUEUED) 1474 knote_dequeue(kn); 1475 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1476 fdrop(kn->kn_fp); 1477 kn->kn_fp = NULL; 1478 } 1479 knote_free(kn); 1480 } 1481 1482 /* 1483 * Low level enqueue function. 1484 * 1485 * The knote should already be marked for processing. 1486 */ 1487 static void 1488 knote_enqueue(struct knote *kn) 1489 { 1490 struct kqueue *kq = kn->kn_kq; 1491 1492 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1493 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1494 kn->kn_status |= KN_QUEUED; 1495 ++kq->kq_count; 1496 1497 /* 1498 * Send SIGIO on request (typically set up as a mailbox signal) 1499 */ 1500 if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1) 1501 pgsigio(kq->kq_sigio, SIGIO, 0); 1502 1503 kqueue_wakeup(kq); 1504 } 1505 1506 /* 1507 * Low level dequeue function. 1508 * 1509 * The knote should already be marked for processing. 1510 */ 1511 static void 1512 knote_dequeue(struct knote *kn) 1513 { 1514 struct kqueue *kq = kn->kn_kq; 1515 1516 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1517 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1518 kn->kn_status &= ~KN_QUEUED; 1519 kq->kq_count--; 1520 } 1521 1522 static struct knote * 1523 knote_alloc(void) 1524 { 1525 return kmalloc(sizeof(struct knote), M_KQUEUE, M_WAITOK); 1526 } 1527 1528 static void 1529 knote_free(struct knote *kn) 1530 { 1531 kfree(kn, M_KQUEUE); 1532 } 1533