1 /* $NetBSD: kern_event.c,v 1.30 2006/07/23 22:06:11 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp $ 29 */ 30 31 #include <sys/cdefs.h> 32 __KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.30 2006/07/23 22:06:11 ad Exp $"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kernel.h> 37 #include <sys/proc.h> 38 #include <sys/malloc.h> 39 #include <sys/unistd.h> 40 #include <sys/file.h> 41 #include <sys/fcntl.h> 42 #include <sys/select.h> 43 #include <sys/queue.h> 44 #include <sys/event.h> 45 #include <sys/eventvar.h> 46 #include <sys/poll.h> 47 #include <sys/pool.h> 48 #include <sys/protosw.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/stat.h> 52 #include <sys/uio.h> 53 #include <sys/mount.h> 54 #include <sys/filedesc.h> 55 #include <sys/sa.h> 56 #include <sys/syscallargs.h> 57 #include <sys/kauth.h> 58 59 static void kqueue_wakeup(struct kqueue *kq); 60 61 static int kqueue_scan(struct file *, size_t, struct kevent *, 62 const struct timespec *, struct lwp *, register_t *, 63 const struct kevent_ops *); 64 static int kqueue_read(struct file *fp, off_t *offset, struct uio *uio, 65 kauth_cred_t cred, int flags); 66 static int kqueue_write(struct file *fp, off_t *offset, struct uio *uio, 67 kauth_cred_t cred, int flags); 68 static int kqueue_ioctl(struct file *fp, u_long com, void *data, 69 struct lwp *l); 70 static int kqueue_fcntl(struct file *fp, u_int com, void *data, 71 struct lwp *l); 72 static int kqueue_poll(struct file *fp, int events, struct lwp *l); 73 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 74 static int kqueue_stat(struct file *fp, struct stat *sp, struct lwp *l); 75 static int kqueue_close(struct file *fp, struct lwp *l); 76 77 static const struct fileops kqueueops = { 78 kqueue_read, kqueue_write, kqueue_ioctl, kqueue_fcntl, kqueue_poll, 79 kqueue_stat, kqueue_close, kqueue_kqfilter 80 }; 81 82 static void knote_attach(struct knote *kn, struct filedesc *fdp); 83 static void knote_drop(struct knote *kn, struct lwp *l, 84 struct filedesc *fdp); 85 static void knote_enqueue(struct knote *kn); 86 static void knote_dequeue(struct knote *kn); 87 88 static void filt_kqdetach(struct knote *kn); 89 static int filt_kqueue(struct knote *kn, long hint); 90 static int filt_procattach(struct knote *kn); 91 static void filt_procdetach(struct knote *kn); 92 static int filt_proc(struct knote *kn, long hint); 93 static int filt_fileattach(struct knote *kn); 94 static void filt_timerexpire(void *knx); 95 static int filt_timerattach(struct knote *kn); 96 static void filt_timerdetach(struct knote *kn); 97 static int filt_timer(struct knote *kn, long hint); 98 99 static const struct filterops kqread_filtops = 100 { 1, NULL, filt_kqdetach, filt_kqueue }; 101 static const struct filterops proc_filtops = 102 { 0, filt_procattach, filt_procdetach, filt_proc }; 103 static const struct filterops file_filtops = 104 { 1, filt_fileattach, NULL, NULL }; 105 static const struct filterops timer_filtops = 106 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 107 108 static POOL_INIT(kqueue_pool, sizeof(struct kqueue), 0, 0, 0, "kqueuepl", NULL); 109 static POOL_INIT(knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl", NULL); 110 static int kq_ncallouts = 0; 111 static int kq_calloutmax = (4 * 1024); 112 113 MALLOC_DEFINE(M_KEVENT, "kevent", "kevents/knotes"); 114 115 #define KNOTE_ACTIVATE(kn) \ 116 do { \ 117 kn->kn_status |= KN_ACTIVE; \ 118 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 119 knote_enqueue(kn); \ 120 } while(0) 121 122 #define KN_HASHSIZE 64 /* XXX should be tunable */ 123 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 124 125 extern const struct filterops sig_filtops; 126 127 /* 128 * Table for for all system-defined filters. 129 * These should be listed in the numeric order of the EVFILT_* defines. 130 * If filtops is NULL, the filter isn't implemented in NetBSD. 131 * End of list is when name is NULL. 132 */ 133 struct kfilter { 134 const char *name; /* name of filter */ 135 uint32_t filter; /* id of filter */ 136 const struct filterops *filtops;/* operations for filter */ 137 }; 138 139 /* System defined filters */ 140 static const struct kfilter sys_kfilters[] = { 141 { "EVFILT_READ", EVFILT_READ, &file_filtops }, 142 { "EVFILT_WRITE", EVFILT_WRITE, &file_filtops }, 143 { "EVFILT_AIO", EVFILT_AIO, NULL }, 144 { "EVFILT_VNODE", EVFILT_VNODE, &file_filtops }, 145 { "EVFILT_PROC", EVFILT_PROC, &proc_filtops }, 146 { "EVFILT_SIGNAL", EVFILT_SIGNAL, &sig_filtops }, 147 { "EVFILT_TIMER", EVFILT_TIMER, &timer_filtops }, 148 { NULL, 0, NULL }, /* end of list */ 149 }; 150 151 /* User defined kfilters */ 152 static struct kfilter *user_kfilters; /* array */ 153 static int user_kfilterc; /* current offset */ 154 static int user_kfiltermaxc; /* max size so far */ 155 156 /* 157 * Find kfilter entry by name, or NULL if not found. 158 */ 159 static const struct kfilter * 160 kfilter_byname_sys(const char *name) 161 { 162 int i; 163 164 for (i = 0; sys_kfilters[i].name != NULL; i++) { 165 if (strcmp(name, sys_kfilters[i].name) == 0) 166 return (&sys_kfilters[i]); 167 } 168 return (NULL); 169 } 170 171 static struct kfilter * 172 kfilter_byname_user(const char *name) 173 { 174 int i; 175 176 /* user_kfilters[] could be NULL if no filters were registered */ 177 if (!user_kfilters) 178 return (NULL); 179 180 for (i = 0; user_kfilters[i].name != NULL; i++) { 181 if (user_kfilters[i].name != '\0' && 182 strcmp(name, user_kfilters[i].name) == 0) 183 return (&user_kfilters[i]); 184 } 185 return (NULL); 186 } 187 188 static const struct kfilter * 189 kfilter_byname(const char *name) 190 { 191 const struct kfilter *kfilter; 192 193 if ((kfilter = kfilter_byname_sys(name)) != NULL) 194 return (kfilter); 195 196 return (kfilter_byname_user(name)); 197 } 198 199 /* 200 * Find kfilter entry by filter id, or NULL if not found. 201 * Assumes entries are indexed in filter id order, for speed. 202 */ 203 static const struct kfilter * 204 kfilter_byfilter(uint32_t filter) 205 { 206 const struct kfilter *kfilter; 207 208 if (filter < EVFILT_SYSCOUNT) /* it's a system filter */ 209 kfilter = &sys_kfilters[filter]; 210 else if (user_kfilters != NULL && 211 filter < EVFILT_SYSCOUNT + user_kfilterc) 212 /* it's a user filter */ 213 kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT]; 214 else 215 return (NULL); /* out of range */ 216 KASSERT(kfilter->filter == filter); /* sanity check! */ 217 return (kfilter); 218 } 219 220 /* 221 * Register a new kfilter. Stores the entry in user_kfilters. 222 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 223 * If retfilter != NULL, the new filterid is returned in it. 224 */ 225 int 226 kfilter_register(const char *name, const struct filterops *filtops, 227 int *retfilter) 228 { 229 struct kfilter *kfilter; 230 void *space; 231 int len; 232 233 if (name == NULL || name[0] == '\0' || filtops == NULL) 234 return (EINVAL); /* invalid args */ 235 if (kfilter_byname(name) != NULL) 236 return (EEXIST); /* already exists */ 237 if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT) 238 return (EINVAL); /* too many */ 239 240 /* check if need to grow user_kfilters */ 241 if (user_kfilterc + 1 > user_kfiltermaxc) { 242 /* 243 * Grow in KFILTER_EXTENT chunks. Use malloc(9), because we 244 * want to traverse user_kfilters as an array. 245 */ 246 user_kfiltermaxc += KFILTER_EXTENT; 247 kfilter = malloc(user_kfiltermaxc * sizeof(struct filter *), 248 M_KEVENT, M_WAITOK); 249 250 /* copy existing user_kfilters */ 251 if (user_kfilters != NULL) 252 memcpy((caddr_t)kfilter, (caddr_t)user_kfilters, 253 user_kfilterc * sizeof(struct kfilter *)); 254 /* zero new sections */ 255 memset((caddr_t)kfilter + 256 user_kfilterc * sizeof(struct kfilter *), 0, 257 (user_kfiltermaxc - user_kfilterc) * 258 sizeof(struct kfilter *)); 259 /* switch to new kfilter */ 260 if (user_kfilters != NULL) 261 free(user_kfilters, M_KEVENT); 262 user_kfilters = kfilter; 263 } 264 len = strlen(name) + 1; /* copy name */ 265 space = malloc(len, M_KEVENT, M_WAITOK); 266 memcpy(space, name, len); 267 user_kfilters[user_kfilterc].name = space; 268 269 user_kfilters[user_kfilterc].filter = user_kfilterc + EVFILT_SYSCOUNT; 270 271 len = sizeof(struct filterops); /* copy filtops */ 272 space = malloc(len, M_KEVENT, M_WAITOK); 273 memcpy(space, filtops, len); 274 user_kfilters[user_kfilterc].filtops = space; 275 276 if (retfilter != NULL) 277 *retfilter = user_kfilters[user_kfilterc].filter; 278 user_kfilterc++; /* finally, increment count */ 279 return (0); 280 } 281 282 /* 283 * Unregister a kfilter previously registered with kfilter_register. 284 * This retains the filter id, but clears the name and frees filtops (filter 285 * operations), so that the number isn't reused during a boot. 286 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 287 */ 288 int 289 kfilter_unregister(const char *name) 290 { 291 struct kfilter *kfilter; 292 293 if (name == NULL || name[0] == '\0') 294 return (EINVAL); /* invalid name */ 295 296 if (kfilter_byname_sys(name) != NULL) 297 return (EINVAL); /* can't detach system filters */ 298 299 kfilter = kfilter_byname_user(name); 300 if (kfilter == NULL) /* not found */ 301 return (ENOENT); 302 303 if (kfilter->name[0] != '\0') { 304 /* XXXUNCONST Cast away const (but we know it's safe. */ 305 free(__UNCONST(kfilter->name), M_KEVENT); 306 kfilter->name = ""; /* mark as `not implemented' */ 307 } 308 if (kfilter->filtops != NULL) { 309 /* XXXUNCONST Cast away const (but we know it's safe. */ 310 free(__UNCONST(kfilter->filtops), M_KEVENT); 311 kfilter->filtops = NULL; /* mark as `not implemented' */ 312 } 313 return (0); 314 } 315 316 317 /* 318 * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file 319 * descriptors. Calls struct fileops kqfilter method for given file descriptor. 320 */ 321 static int 322 filt_fileattach(struct knote *kn) 323 { 324 struct file *fp; 325 326 fp = kn->kn_fp; 327 return ((*fp->f_ops->fo_kqfilter)(fp, kn)); 328 } 329 330 /* 331 * Filter detach method for EVFILT_READ on kqueue descriptor. 332 */ 333 static void 334 filt_kqdetach(struct knote *kn) 335 { 336 struct kqueue *kq; 337 338 kq = (struct kqueue *)kn->kn_fp->f_data; 339 SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext); 340 } 341 342 /* 343 * Filter event method for EVFILT_READ on kqueue descriptor. 344 */ 345 /*ARGSUSED*/ 346 static int 347 filt_kqueue(struct knote *kn, long hint) 348 { 349 struct kqueue *kq; 350 351 kq = (struct kqueue *)kn->kn_fp->f_data; 352 kn->kn_data = kq->kq_count; 353 return (kn->kn_data > 0); 354 } 355 356 /* 357 * Filter attach method for EVFILT_PROC. 358 */ 359 static int 360 filt_procattach(struct knote *kn) 361 { 362 struct proc *p, *curp; 363 struct lwp *curl; 364 365 curl = curlwp; 366 curp = curl->l_proc; 367 368 p = pfind(kn->kn_id); 369 if (p == NULL) 370 return (ESRCH); 371 372 /* 373 * Fail if it's not owned by you, or the last exec gave us 374 * setuid/setgid privs (unless you're root). 375 */ 376 if ((kauth_cred_getuid(p->p_cred) != kauth_cred_getuid(curl->l_cred) || 377 (p->p_flag & P_SUGID)) && kauth_authorize_generic(curl->l_cred, 378 KAUTH_GENERIC_ISSUSER, &curl->l_acflag) != 0) 379 return (EACCES); 380 381 kn->kn_ptr.p_proc = p; 382 kn->kn_flags |= EV_CLEAR; /* automatically set */ 383 384 /* 385 * internal flag indicating registration done by kernel 386 */ 387 if (kn->kn_flags & EV_FLAG1) { 388 kn->kn_data = kn->kn_sdata; /* ppid */ 389 kn->kn_fflags = NOTE_CHILD; 390 kn->kn_flags &= ~EV_FLAG1; 391 } 392 393 /* XXXSMP lock the process? */ 394 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 395 396 return (0); 397 } 398 399 /* 400 * Filter detach method for EVFILT_PROC. 401 * 402 * The knote may be attached to a different process, which may exit, 403 * leaving nothing for the knote to be attached to. So when the process 404 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 405 * it will be deleted when read out. However, as part of the knote deletion, 406 * this routine is called, so a check is needed to avoid actually performing 407 * a detach, because the original process might not exist any more. 408 */ 409 static void 410 filt_procdetach(struct knote *kn) 411 { 412 struct proc *p; 413 414 if (kn->kn_status & KN_DETACHED) 415 return; 416 417 p = kn->kn_ptr.p_proc; 418 KASSERT(p->p_stat == SZOMB || pfind(kn->kn_id) == p); 419 420 /* XXXSMP lock the process? */ 421 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 422 } 423 424 /* 425 * Filter event method for EVFILT_PROC. 426 */ 427 static int 428 filt_proc(struct knote *kn, long hint) 429 { 430 u_int event; 431 432 /* 433 * mask off extra data 434 */ 435 event = (u_int)hint & NOTE_PCTRLMASK; 436 437 /* 438 * if the user is interested in this event, record it. 439 */ 440 if (kn->kn_sfflags & event) 441 kn->kn_fflags |= event; 442 443 /* 444 * process is gone, so flag the event as finished. 445 */ 446 if (event == NOTE_EXIT) { 447 /* 448 * Detach the knote from watched process and mark 449 * it as such. We can't leave this to kqueue_scan(), 450 * since the process might not exist by then. And we 451 * have to do this now, since psignal KNOTE() is called 452 * also for zombies and we might end up reading freed 453 * memory if the kevent would already be picked up 454 * and knote g/c'ed. 455 */ 456 kn->kn_fop->f_detach(kn); 457 kn->kn_status |= KN_DETACHED; 458 459 /* Mark as ONESHOT, so that the knote it g/c'ed when read */ 460 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 461 return (1); 462 } 463 464 /* 465 * process forked, and user wants to track the new process, 466 * so attach a new knote to it, and immediately report an 467 * event with the parent's pid. 468 */ 469 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 470 struct kevent kev; 471 int error; 472 473 /* 474 * register knote with new process. 475 */ 476 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 477 kev.filter = kn->kn_filter; 478 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 479 kev.fflags = kn->kn_sfflags; 480 kev.data = kn->kn_id; /* parent */ 481 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 482 error = kqueue_register(kn->kn_kq, &kev, NULL); 483 if (error) 484 kn->kn_fflags |= NOTE_TRACKERR; 485 } 486 487 return (kn->kn_fflags != 0); 488 } 489 490 static void 491 filt_timerexpire(void *knx) 492 { 493 struct knote *kn = knx; 494 int tticks; 495 496 kn->kn_data++; 497 KNOTE_ACTIVATE(kn); 498 499 if ((kn->kn_flags & EV_ONESHOT) == 0) { 500 tticks = mstohz(kn->kn_sdata); 501 callout_schedule((struct callout *)kn->kn_hook, tticks); 502 } 503 } 504 505 /* 506 * data contains amount of time to sleep, in milliseconds 507 */ 508 static int 509 filt_timerattach(struct knote *kn) 510 { 511 struct callout *calloutp; 512 int tticks; 513 514 if (kq_ncallouts >= kq_calloutmax) 515 return (ENOMEM); 516 kq_ncallouts++; 517 518 tticks = mstohz(kn->kn_sdata); 519 520 /* if the supplied value is under our resolution, use 1 tick */ 521 if (tticks == 0) { 522 if (kn->kn_sdata == 0) 523 return (EINVAL); 524 tticks = 1; 525 } 526 527 kn->kn_flags |= EV_CLEAR; /* automatically set */ 528 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 529 M_KEVENT, 0); 530 callout_init(calloutp); 531 callout_reset(calloutp, tticks, filt_timerexpire, kn); 532 kn->kn_hook = calloutp; 533 534 return (0); 535 } 536 537 static void 538 filt_timerdetach(struct knote *kn) 539 { 540 struct callout *calloutp; 541 542 calloutp = (struct callout *)kn->kn_hook; 543 callout_stop(calloutp); 544 FREE(calloutp, M_KEVENT); 545 kq_ncallouts--; 546 } 547 548 static int 549 filt_timer(struct knote *kn, long hint) 550 { 551 return (kn->kn_data != 0); 552 } 553 554 /* 555 * filt_seltrue: 556 * 557 * This filter "event" routine simulates seltrue(). 558 */ 559 int 560 filt_seltrue(struct knote *kn, long hint) 561 { 562 563 /* 564 * We don't know how much data can be read/written, 565 * but we know that it *can* be. This is about as 566 * good as select/poll does as well. 567 */ 568 kn->kn_data = 0; 569 return (1); 570 } 571 572 /* 573 * This provides full kqfilter entry for device switch tables, which 574 * has same effect as filter using filt_seltrue() as filter method. 575 */ 576 static void 577 filt_seltruedetach(struct knote *kn) 578 { 579 /* Nothing to do */ 580 } 581 582 static const struct filterops seltrue_filtops = 583 { 1, NULL, filt_seltruedetach, filt_seltrue }; 584 585 int 586 seltrue_kqfilter(dev_t dev, struct knote *kn) 587 { 588 switch (kn->kn_filter) { 589 case EVFILT_READ: 590 case EVFILT_WRITE: 591 kn->kn_fop = &seltrue_filtops; 592 break; 593 default: 594 return (1); 595 } 596 597 /* Nothing more to do */ 598 return (0); 599 } 600 601 /* 602 * kqueue(2) system call. 603 */ 604 int 605 sys_kqueue(struct lwp *l, void *v, register_t *retval) 606 { 607 struct filedesc *fdp; 608 struct kqueue *kq; 609 struct file *fp; 610 int fd, error; 611 612 fdp = l->l_proc->p_fd; 613 error = falloc(l, &fp, &fd); /* setup a new file descriptor */ 614 if (error) 615 return (error); 616 fp->f_flag = FREAD | FWRITE; 617 fp->f_type = DTYPE_KQUEUE; 618 fp->f_ops = &kqueueops; 619 kq = pool_get(&kqueue_pool, PR_WAITOK); 620 memset((char *)kq, 0, sizeof(struct kqueue)); 621 simple_lock_init(&kq->kq_lock); 622 TAILQ_INIT(&kq->kq_head); 623 fp->f_data = (caddr_t)kq; /* store the kqueue with the fp */ 624 *retval = fd; 625 if (fdp->fd_knlistsize < 0) 626 fdp->fd_knlistsize = 0; /* this process has a kq */ 627 kq->kq_fdp = fdp; 628 FILE_SET_MATURE(fp); 629 FILE_UNUSE(fp, l); /* falloc() does FILE_USE() */ 630 return (error); 631 } 632 633 /* 634 * kevent(2) system call. 635 */ 636 static int 637 kevent_fetch_changes(void *private, const struct kevent *changelist, 638 struct kevent *changes, size_t index, int n) 639 { 640 return copyin(changelist + index, changes, n * sizeof(*changes)); 641 } 642 643 static int 644 kevent_put_events(void *private, struct kevent *events, 645 struct kevent *eventlist, size_t index, int n) 646 { 647 return copyout(events, eventlist + index, n * sizeof(*events)); 648 } 649 650 static const struct kevent_ops kevent_native_ops = { 651 keo_private: NULL, 652 keo_fetch_timeout: copyin, 653 keo_fetch_changes: kevent_fetch_changes, 654 keo_put_events: kevent_put_events, 655 }; 656 657 int 658 sys_kevent(struct lwp *l, void *v, register_t *retval) 659 { 660 struct sys_kevent_args /* { 661 syscallarg(int) fd; 662 syscallarg(const struct kevent *) changelist; 663 syscallarg(size_t) nchanges; 664 syscallarg(struct kevent *) eventlist; 665 syscallarg(size_t) nevents; 666 syscallarg(const struct timespec *) timeout; 667 } */ *uap = v; 668 669 return kevent1(l, retval, SCARG(uap, fd), SCARG(uap, changelist), 670 SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents), 671 SCARG(uap, timeout), &kevent_native_ops); 672 } 673 674 int 675 kevent1(struct lwp *l, register_t *retval, int fd, 676 const struct kevent *changelist, size_t nchanges, struct kevent *eventlist, 677 size_t nevents, const struct timespec *timeout, 678 const struct kevent_ops *keops) 679 { 680 struct kevent *kevp; 681 struct kqueue *kq; 682 struct file *fp; 683 struct timespec ts; 684 struct proc *p; 685 size_t i, n, ichange; 686 int nerrors, error; 687 688 p = l->l_proc; 689 /* check that we're dealing with a kq */ 690 fp = fd_getfile(p->p_fd, fd); 691 if (fp == NULL) 692 return (EBADF); 693 694 if (fp->f_type != DTYPE_KQUEUE) { 695 simple_unlock(&fp->f_slock); 696 return (EBADF); 697 } 698 699 FILE_USE(fp); 700 701 if (timeout != NULL) { 702 error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts)); 703 if (error) 704 goto done; 705 timeout = &ts; 706 } 707 708 kq = (struct kqueue *)fp->f_data; 709 nerrors = 0; 710 ichange = 0; 711 712 /* traverse list of events to register */ 713 while (nchanges > 0) { 714 /* copyin a maximum of KQ_EVENTS at each pass */ 715 n = MIN(nchanges, KQ_NEVENTS); 716 error = (*keops->keo_fetch_changes)(keops->keo_private, 717 changelist, kq->kq_kev, ichange, n); 718 if (error) 719 goto done; 720 for (i = 0; i < n; i++) { 721 kevp = &kq->kq_kev[i]; 722 kevp->flags &= ~EV_SYSFLAGS; 723 /* register each knote */ 724 error = kqueue_register(kq, kevp, l); 725 if (error) { 726 if (nevents != 0) { 727 kevp->flags = EV_ERROR; 728 kevp->data = error; 729 error = (*keops->keo_put_events) 730 (keops->keo_private, kevp, 731 eventlist, nerrors, 1); 732 if (error) 733 goto done; 734 nevents--; 735 nerrors++; 736 } else { 737 goto done; 738 } 739 } 740 } 741 nchanges -= n; /* update the results */ 742 ichange += n; 743 } 744 if (nerrors) { 745 *retval = nerrors; 746 error = 0; 747 goto done; 748 } 749 750 /* actually scan through the events */ 751 error = kqueue_scan(fp, nevents, eventlist, timeout, l, retval, keops); 752 done: 753 FILE_UNUSE(fp, l); 754 return (error); 755 } 756 757 /* 758 * Register a given kevent kev onto the kqueue 759 */ 760 int 761 kqueue_register(struct kqueue *kq, struct kevent *kev, struct lwp *l) 762 { 763 const struct kfilter *kfilter; 764 struct filedesc *fdp; 765 struct file *fp; 766 struct knote *kn; 767 int s, error; 768 769 fdp = kq->kq_fdp; 770 fp = NULL; 771 kn = NULL; 772 error = 0; 773 kfilter = kfilter_byfilter(kev->filter); 774 if (kfilter == NULL || kfilter->filtops == NULL) { 775 /* filter not found nor implemented */ 776 return (EINVAL); 777 } 778 779 /* search if knote already exists */ 780 if (kfilter->filtops->f_isfd) { 781 /* monitoring a file descriptor */ 782 if ((fp = fd_getfile(fdp, kev->ident)) == NULL) 783 return (EBADF); /* validate descriptor */ 784 FILE_USE(fp); 785 786 if (kev->ident < fdp->fd_knlistsize) { 787 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 788 if (kq == kn->kn_kq && 789 kev->filter == kn->kn_filter) 790 break; 791 } 792 } else { 793 /* 794 * not monitoring a file descriptor, so 795 * lookup knotes in internal hash table 796 */ 797 if (fdp->fd_knhashmask != 0) { 798 struct klist *list; 799 800 list = &fdp->fd_knhash[ 801 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 802 SLIST_FOREACH(kn, list, kn_link) 803 if (kev->ident == kn->kn_id && 804 kq == kn->kn_kq && 805 kev->filter == kn->kn_filter) 806 break; 807 } 808 } 809 810 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 811 error = ENOENT; /* filter not found */ 812 goto done; 813 } 814 815 /* 816 * kn now contains the matching knote, or NULL if no match 817 */ 818 if (kev->flags & EV_ADD) { 819 /* add knote */ 820 821 if (kn == NULL) { 822 /* create new knote */ 823 kn = pool_get(&knote_pool, PR_WAITOK); 824 if (kn == NULL) { 825 error = ENOMEM; 826 goto done; 827 } 828 kn->kn_fp = fp; 829 kn->kn_kq = kq; 830 kn->kn_fop = kfilter->filtops; 831 832 /* 833 * apply reference count to knote structure, and 834 * do not release it at the end of this routine. 835 */ 836 fp = NULL; 837 838 kn->kn_sfflags = kev->fflags; 839 kn->kn_sdata = kev->data; 840 kev->fflags = 0; 841 kev->data = 0; 842 kn->kn_kevent = *kev; 843 844 knote_attach(kn, fdp); 845 if ((error = kfilter->filtops->f_attach(kn)) != 0) { 846 knote_drop(kn, l, fdp); 847 goto done; 848 } 849 } else { 850 /* modify existing knote */ 851 852 /* 853 * The user may change some filter values after the 854 * initial EV_ADD, but doing so will not reset any 855 * filter which have already been triggered. 856 */ 857 kn->kn_sfflags = kev->fflags; 858 kn->kn_sdata = kev->data; 859 kn->kn_kevent.udata = kev->udata; 860 } 861 862 s = splsched(); 863 if (kn->kn_fop->f_event(kn, 0)) 864 KNOTE_ACTIVATE(kn); 865 splx(s); 866 867 } else if (kev->flags & EV_DELETE) { /* delete knote */ 868 kn->kn_fop->f_detach(kn); 869 knote_drop(kn, l, fdp); 870 goto done; 871 } 872 873 /* disable knote */ 874 if ((kev->flags & EV_DISABLE) && 875 ((kn->kn_status & KN_DISABLED) == 0)) { 876 s = splsched(); 877 kn->kn_status |= KN_DISABLED; 878 splx(s); 879 } 880 881 /* enable knote */ 882 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 883 s = splsched(); 884 kn->kn_status &= ~KN_DISABLED; 885 if ((kn->kn_status & KN_ACTIVE) && 886 ((kn->kn_status & KN_QUEUED) == 0)) 887 knote_enqueue(kn); 888 splx(s); 889 } 890 891 done: 892 if (fp != NULL) 893 FILE_UNUSE(fp, l); 894 return (error); 895 } 896 897 /* 898 * Scan through the list of events on fp (for a maximum of maxevents), 899 * returning the results in to ulistp. Timeout is determined by tsp; if 900 * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait 901 * as appropriate. 902 */ 903 static int 904 kqueue_scan(struct file *fp, size_t maxevents, struct kevent *ulistp, 905 const struct timespec *tsp, struct lwp *l, register_t *retval, 906 const struct kevent_ops *keops) 907 { 908 struct proc *p = l->l_proc; 909 struct kqueue *kq; 910 struct kevent *kevp; 911 struct timeval atv, sleeptv; 912 struct knote *kn, *marker=NULL; 913 size_t count, nkev, nevents; 914 int s, timeout, error; 915 916 kq = (struct kqueue *)fp->f_data; 917 count = maxevents; 918 nkev = nevents = error = 0; 919 if (count == 0) 920 goto done; 921 922 if (tsp) { /* timeout supplied */ 923 TIMESPEC_TO_TIMEVAL(&atv, tsp); 924 if (inittimeleft(&atv, &sleeptv) == -1) { 925 error = EINVAL; 926 goto done; 927 } 928 timeout = tvtohz(&atv); 929 if (timeout <= 0) 930 timeout = -1; /* do poll */ 931 } else { 932 /* no timeout, wait forever */ 933 timeout = 0; 934 } 935 936 MALLOC(marker, struct knote *, sizeof(*marker), M_KEVENT, M_WAITOK); 937 memset(marker, 0, sizeof(*marker)); 938 939 goto start; 940 941 retry: 942 if (tsp && (timeout = gettimeleft(&atv, &sleeptv)) <= 0) { 943 goto done; 944 } 945 946 start: 947 kevp = kq->kq_kev; 948 s = splsched(); 949 simple_lock(&kq->kq_lock); 950 if (kq->kq_count == 0) { 951 if (timeout < 0) { 952 error = EWOULDBLOCK; 953 simple_unlock(&kq->kq_lock); 954 } else { 955 kq->kq_state |= KQ_SLEEP; 956 error = ltsleep(kq, PSOCK | PCATCH | PNORELOCK, 957 "kqread", timeout, &kq->kq_lock); 958 } 959 splx(s); 960 if (error == 0) 961 goto retry; 962 /* don't restart after signals... */ 963 if (error == ERESTART) 964 error = EINTR; 965 else if (error == EWOULDBLOCK) 966 error = 0; 967 goto done; 968 } 969 970 /* mark end of knote list */ 971 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 972 simple_unlock(&kq->kq_lock); 973 974 while (count) { /* while user wants data ... */ 975 simple_lock(&kq->kq_lock); 976 kn = TAILQ_FIRST(&kq->kq_head); /* get next knote */ 977 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 978 if (kn == marker) { /* if it's our marker, stop */ 979 /* What if it's some else's marker? */ 980 simple_unlock(&kq->kq_lock); 981 splx(s); 982 if (count == maxevents) 983 goto retry; 984 goto done; 985 } 986 kq->kq_count--; 987 simple_unlock(&kq->kq_lock); 988 989 if (kn->kn_status & KN_DISABLED) { 990 /* don't want disabled events */ 991 kn->kn_status &= ~KN_QUEUED; 992 continue; 993 } 994 if ((kn->kn_flags & EV_ONESHOT) == 0 && 995 kn->kn_fop->f_event(kn, 0) == 0) { 996 /* 997 * non-ONESHOT event that hasn't 998 * triggered again, so de-queue. 999 */ 1000 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1001 continue; 1002 } 1003 *kevp = kn->kn_kevent; 1004 kevp++; 1005 nkev++; 1006 if (kn->kn_flags & EV_ONESHOT) { 1007 /* delete ONESHOT events after retrieval */ 1008 kn->kn_status &= ~KN_QUEUED; 1009 splx(s); 1010 kn->kn_fop->f_detach(kn); 1011 knote_drop(kn, l, p->p_fd); 1012 s = splsched(); 1013 } else if (kn->kn_flags & EV_CLEAR) { 1014 /* clear state after retrieval */ 1015 kn->kn_data = 0; 1016 kn->kn_fflags = 0; 1017 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1018 } else { 1019 /* add event back on list */ 1020 simple_lock(&kq->kq_lock); 1021 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1022 kq->kq_count++; 1023 simple_unlock(&kq->kq_lock); 1024 } 1025 count--; 1026 if (nkev == KQ_NEVENTS) { 1027 /* do copyouts in KQ_NEVENTS chunks */ 1028 splx(s); 1029 error = (*keops->keo_put_events)(keops->keo_private, 1030 &kq->kq_kev[0], ulistp, nevents, nkev); 1031 nevents += nkev; 1032 nkev = 0; 1033 kevp = kq->kq_kev; 1034 s = splsched(); 1035 if (error) 1036 break; 1037 } 1038 } 1039 1040 /* remove marker */ 1041 simple_lock(&kq->kq_lock); 1042 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1043 simple_unlock(&kq->kq_lock); 1044 splx(s); 1045 done: 1046 if (marker) 1047 FREE(marker, M_KEVENT); 1048 1049 if (nkev != 0) 1050 /* copyout remaining events */ 1051 error = (*keops->keo_put_events)(keops->keo_private, 1052 &kq->kq_kev[0], ulistp, nevents, nkev); 1053 *retval = maxevents - count; 1054 1055 return (error); 1056 } 1057 1058 /* 1059 * struct fileops read method for a kqueue descriptor. 1060 * Not implemented. 1061 * XXX: This could be expanded to call kqueue_scan, if desired. 1062 */ 1063 /*ARGSUSED*/ 1064 static int 1065 kqueue_read(struct file *fp, off_t *offset, struct uio *uio, 1066 kauth_cred_t cred, int flags) 1067 { 1068 1069 return (ENXIO); 1070 } 1071 1072 /* 1073 * struct fileops write method for a kqueue descriptor. 1074 * Not implemented. 1075 */ 1076 /*ARGSUSED*/ 1077 static int 1078 kqueue_write(struct file *fp, off_t *offset, struct uio *uio, 1079 kauth_cred_t cred, int flags) 1080 { 1081 1082 return (ENXIO); 1083 } 1084 1085 /* 1086 * struct fileops ioctl method for a kqueue descriptor. 1087 * 1088 * Two ioctls are currently supported. They both use struct kfilter_mapping: 1089 * KFILTER_BYNAME find name for filter, and return result in 1090 * name, which is of size len. 1091 * KFILTER_BYFILTER find filter for name. len is ignored. 1092 */ 1093 /*ARGSUSED*/ 1094 static int 1095 kqueue_ioctl(struct file *fp, u_long com, void *data, struct lwp *l) 1096 { 1097 struct kfilter_mapping *km; 1098 const struct kfilter *kfilter; 1099 char *name; 1100 int error; 1101 1102 km = (struct kfilter_mapping *)data; 1103 error = 0; 1104 1105 switch (com) { 1106 case KFILTER_BYFILTER: /* convert filter -> name */ 1107 kfilter = kfilter_byfilter(km->filter); 1108 if (kfilter != NULL) 1109 error = copyoutstr(kfilter->name, km->name, km->len, 1110 NULL); 1111 else 1112 error = ENOENT; 1113 break; 1114 1115 case KFILTER_BYNAME: /* convert name -> filter */ 1116 MALLOC(name, char *, KFILTER_MAXNAME, M_KEVENT, M_WAITOK); 1117 error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL); 1118 if (error) { 1119 FREE(name, M_KEVENT); 1120 break; 1121 } 1122 kfilter = kfilter_byname(name); 1123 if (kfilter != NULL) 1124 km->filter = kfilter->filter; 1125 else 1126 error = ENOENT; 1127 FREE(name, M_KEVENT); 1128 break; 1129 1130 default: 1131 error = ENOTTY; 1132 1133 } 1134 return (error); 1135 } 1136 1137 /* 1138 * struct fileops fcntl method for a kqueue descriptor. 1139 * Not implemented. 1140 */ 1141 /*ARGSUSED*/ 1142 static int 1143 kqueue_fcntl(struct file *fp, u_int com, void *data, struct lwp *l) 1144 { 1145 1146 return (ENOTTY); 1147 } 1148 1149 /* 1150 * struct fileops poll method for a kqueue descriptor. 1151 * Determine if kqueue has events pending. 1152 */ 1153 static int 1154 kqueue_poll(struct file *fp, int events, struct lwp *l) 1155 { 1156 struct kqueue *kq; 1157 int revents; 1158 1159 kq = (struct kqueue *)fp->f_data; 1160 revents = 0; 1161 if (events & (POLLIN | POLLRDNORM)) { 1162 if (kq->kq_count) { 1163 revents |= events & (POLLIN | POLLRDNORM); 1164 } else { 1165 selrecord(l, &kq->kq_sel); 1166 } 1167 } 1168 return (revents); 1169 } 1170 1171 /* 1172 * struct fileops stat method for a kqueue descriptor. 1173 * Returns dummy info, with st_size being number of events pending. 1174 */ 1175 static int 1176 kqueue_stat(struct file *fp, struct stat *st, struct lwp *l) 1177 { 1178 struct kqueue *kq; 1179 1180 kq = (struct kqueue *)fp->f_data; 1181 memset((void *)st, 0, sizeof(*st)); 1182 st->st_size = kq->kq_count; 1183 st->st_blksize = sizeof(struct kevent); 1184 st->st_mode = S_IFIFO; 1185 return (0); 1186 } 1187 1188 /* 1189 * struct fileops close method for a kqueue descriptor. 1190 * Cleans up kqueue. 1191 */ 1192 static int 1193 kqueue_close(struct file *fp, struct lwp *l) 1194 { 1195 struct proc *p = l->l_proc; 1196 struct kqueue *kq; 1197 struct filedesc *fdp; 1198 struct knote **knp, *kn, *kn0; 1199 int i; 1200 1201 kq = (struct kqueue *)fp->f_data; 1202 fdp = p->p_fd; 1203 for (i = 0; i < fdp->fd_knlistsize; i++) { 1204 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 1205 kn = *knp; 1206 while (kn != NULL) { 1207 kn0 = SLIST_NEXT(kn, kn_link); 1208 if (kq == kn->kn_kq) { 1209 kn->kn_fop->f_detach(kn); 1210 FILE_UNUSE(kn->kn_fp, l); 1211 pool_put(&knote_pool, kn); 1212 *knp = kn0; 1213 } else { 1214 knp = &SLIST_NEXT(kn, kn_link); 1215 } 1216 kn = kn0; 1217 } 1218 } 1219 if (fdp->fd_knhashmask != 0) { 1220 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 1221 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 1222 kn = *knp; 1223 while (kn != NULL) { 1224 kn0 = SLIST_NEXT(kn, kn_link); 1225 if (kq == kn->kn_kq) { 1226 kn->kn_fop->f_detach(kn); 1227 /* XXX non-fd release of kn->kn_ptr */ 1228 pool_put(&knote_pool, kn); 1229 *knp = kn0; 1230 } else { 1231 knp = &SLIST_NEXT(kn, kn_link); 1232 } 1233 kn = kn0; 1234 } 1235 } 1236 } 1237 pool_put(&kqueue_pool, kq); 1238 fp->f_data = NULL; 1239 1240 return (0); 1241 } 1242 1243 /* 1244 * wakeup a kqueue 1245 */ 1246 static void 1247 kqueue_wakeup(struct kqueue *kq) 1248 { 1249 int s; 1250 1251 s = splsched(); 1252 simple_lock(&kq->kq_lock); 1253 if (kq->kq_state & KQ_SLEEP) { /* if currently sleeping ... */ 1254 kq->kq_state &= ~KQ_SLEEP; 1255 wakeup(kq); /* ... wakeup */ 1256 } 1257 1258 /* Notify select/poll and kevent. */ 1259 selnotify(&kq->kq_sel, 0); 1260 simple_unlock(&kq->kq_lock); 1261 splx(s); 1262 } 1263 1264 /* 1265 * struct fileops kqfilter method for a kqueue descriptor. 1266 * Event triggered when monitored kqueue changes. 1267 */ 1268 /*ARGSUSED*/ 1269 static int 1270 kqueue_kqfilter(struct file *fp, struct knote *kn) 1271 { 1272 struct kqueue *kq; 1273 1274 KASSERT(fp == kn->kn_fp); 1275 kq = (struct kqueue *)kn->kn_fp->f_data; 1276 if (kn->kn_filter != EVFILT_READ) 1277 return (1); 1278 kn->kn_fop = &kqread_filtops; 1279 SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext); 1280 return (0); 1281 } 1282 1283 1284 /* 1285 * Walk down a list of knotes, activating them if their event has triggered. 1286 */ 1287 void 1288 knote(struct klist *list, long hint) 1289 { 1290 struct knote *kn; 1291 1292 SLIST_FOREACH(kn, list, kn_selnext) 1293 if (kn->kn_fop->f_event(kn, hint)) 1294 KNOTE_ACTIVATE(kn); 1295 } 1296 1297 /* 1298 * Remove all knotes from a specified klist 1299 */ 1300 void 1301 knote_remove(struct lwp *l, struct klist *list) 1302 { 1303 struct knote *kn; 1304 1305 while ((kn = SLIST_FIRST(list)) != NULL) { 1306 kn->kn_fop->f_detach(kn); 1307 knote_drop(kn, l, l->l_proc->p_fd); 1308 } 1309 } 1310 1311 /* 1312 * Remove all knotes referencing a specified fd 1313 */ 1314 void 1315 knote_fdclose(struct lwp *l, int fd) 1316 { 1317 struct filedesc *fdp; 1318 struct klist *list; 1319 1320 fdp = l->l_proc->p_fd; 1321 list = &fdp->fd_knlist[fd]; 1322 knote_remove(l, list); 1323 } 1324 1325 /* 1326 * Attach a new knote to a file descriptor 1327 */ 1328 static void 1329 knote_attach(struct knote *kn, struct filedesc *fdp) 1330 { 1331 struct klist *list; 1332 int size; 1333 1334 if (! kn->kn_fop->f_isfd) { 1335 /* if knote is not on an fd, store on internal hash table */ 1336 if (fdp->fd_knhashmask == 0) 1337 fdp->fd_knhash = hashinit(KN_HASHSIZE, HASH_LIST, 1338 M_KEVENT, M_WAITOK, &fdp->fd_knhashmask); 1339 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1340 goto done; 1341 } 1342 1343 /* 1344 * otherwise, knote is on an fd. 1345 * knotes are stored in fd_knlist indexed by kn->kn_id. 1346 */ 1347 if (fdp->fd_knlistsize <= kn->kn_id) { 1348 /* expand list, it's too small */ 1349 size = fdp->fd_knlistsize; 1350 while (size <= kn->kn_id) { 1351 /* grow in KQ_EXTENT chunks */ 1352 size += KQ_EXTENT; 1353 } 1354 list = malloc(size * sizeof(struct klist *), M_KEVENT,M_WAITOK); 1355 if (fdp->fd_knlist) { 1356 /* copy existing knlist */ 1357 memcpy((caddr_t)list, (caddr_t)fdp->fd_knlist, 1358 fdp->fd_knlistsize * sizeof(struct klist *)); 1359 } 1360 /* 1361 * Zero new memory. Stylistically, SLIST_INIT() should be 1362 * used here, but that does same thing as the memset() anyway. 1363 */ 1364 memset(&list[fdp->fd_knlistsize], 0, 1365 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 1366 1367 /* switch to new knlist */ 1368 if (fdp->fd_knlist != NULL) 1369 free(fdp->fd_knlist, M_KEVENT); 1370 fdp->fd_knlistsize = size; 1371 fdp->fd_knlist = list; 1372 } 1373 1374 /* get list head for this fd */ 1375 list = &fdp->fd_knlist[kn->kn_id]; 1376 done: 1377 /* add new knote */ 1378 SLIST_INSERT_HEAD(list, kn, kn_link); 1379 kn->kn_status = 0; 1380 } 1381 1382 /* 1383 * Drop knote. 1384 * Should be called at spl == 0, since we don't want to hold spl 1385 * while calling FILE_UNUSE and free. 1386 */ 1387 static void 1388 knote_drop(struct knote *kn, struct lwp *l, struct filedesc *fdp) 1389 { 1390 struct klist *list; 1391 1392 if (kn->kn_fop->f_isfd) 1393 list = &fdp->fd_knlist[kn->kn_id]; 1394 else 1395 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1396 1397 SLIST_REMOVE(list, kn, knote, kn_link); 1398 if (kn->kn_status & KN_QUEUED) 1399 knote_dequeue(kn); 1400 if (kn->kn_fop->f_isfd) 1401 FILE_UNUSE(kn->kn_fp, l); 1402 pool_put(&knote_pool, kn); 1403 } 1404 1405 1406 /* 1407 * Queue new event for knote. 1408 */ 1409 static void 1410 knote_enqueue(struct knote *kn) 1411 { 1412 struct kqueue *kq; 1413 int s; 1414 1415 kq = kn->kn_kq; 1416 KASSERT((kn->kn_status & KN_QUEUED) == 0); 1417 1418 s = splsched(); 1419 simple_lock(&kq->kq_lock); 1420 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1421 kn->kn_status |= KN_QUEUED; 1422 kq->kq_count++; 1423 simple_unlock(&kq->kq_lock); 1424 splx(s); 1425 kqueue_wakeup(kq); 1426 } 1427 1428 /* 1429 * Dequeue event for knote. 1430 */ 1431 static void 1432 knote_dequeue(struct knote *kn) 1433 { 1434 struct kqueue *kq; 1435 int s; 1436 1437 KASSERT(kn->kn_status & KN_QUEUED); 1438 kq = kn->kn_kq; 1439 1440 s = splsched(); 1441 simple_lock(&kq->kq_lock); 1442 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1443 kn->kn_status &= ~KN_QUEUED; 1444 kq->kq_count--; 1445 simple_unlock(&kq->kq_lock); 1446 splx(s); 1447 } 1448