1/* $NetBSD: kern_event.c,v 1.75 2012/01/25 00:28:35 christos Exp $ */ 2 3/*- 4 * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Doran. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32/*- 33 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 * 57 * FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp 58 */ 59 60#include <sys/cdefs.h> 61__KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.75 2012/01/25 00:28:35 christos Exp $"); 62 63#include <sys/param.h> 64#include <sys/systm.h> 65#include <sys/kernel.h> 66#include <sys/proc.h> 67#include <sys/file.h> 68#include <sys/select.h> 69#include <sys/queue.h> 70#include <sys/event.h> 71#include <sys/eventvar.h> 72#include <sys/poll.h> 73#include <sys/kmem.h> 74#include <sys/stat.h> 75#include <sys/filedesc.h> 76#include <sys/syscallargs.h> 77#include <sys/kauth.h> 78#include <sys/conf.h> 79#include <sys/atomic.h> 80 81static int kqueue_scan(file_t *, size_t, struct kevent *, 82 const struct timespec *, register_t *, 83 const struct kevent_ops *, struct kevent *, 84 size_t); 85static int kqueue_ioctl(file_t *, u_long, void *); 86static int kqueue_fcntl(file_t *, u_int, void *); 87static int kqueue_poll(file_t *, int); 88static int kqueue_kqfilter(file_t *, struct knote *); 89static int kqueue_stat(file_t *, struct stat *); 90static int kqueue_close(file_t *); 91static int kqueue_register(struct kqueue *, struct kevent *); 92static void kqueue_doclose(struct kqueue *, struct klist *, int); 93 94static void knote_detach(struct knote *, filedesc_t *fdp, bool); 95static void knote_enqueue(struct knote *); 96static void knote_activate(struct knote *); 97 98static void filt_kqdetach(struct knote *); 99static int filt_kqueue(struct knote *, long hint); 100static int filt_procattach(struct knote *); 101static void filt_procdetach(struct knote *); 102static int filt_proc(struct knote *, long hint); 103static int filt_fileattach(struct knote *); 104static void filt_timerexpire(void *x); 105static int filt_timerattach(struct knote *); 106static void filt_timerdetach(struct knote *); 107static int filt_timer(struct knote *, long hint); 108 109static const struct fileops kqueueops = { 110 .fo_read = (void *)enxio, 111 .fo_write = (void *)enxio, 112 .fo_ioctl = kqueue_ioctl, 113 .fo_fcntl = kqueue_fcntl, 114 .fo_poll = kqueue_poll, 115 .fo_stat = kqueue_stat, 116 .fo_close = kqueue_close, 117 .fo_kqfilter = kqueue_kqfilter, 118 .fo_restart = fnullop_restart, 119}; 120 121static const struct filterops kqread_filtops = 122 { 1, NULL, filt_kqdetach, filt_kqueue }; 123static const struct filterops proc_filtops = 124 { 0, filt_procattach, filt_procdetach, filt_proc }; 125static const struct filterops file_filtops = 126 { 1, filt_fileattach, NULL, NULL }; 127static const struct filterops timer_filtops = 128 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 129 130static u_int kq_ncallouts = 0; 131static int kq_calloutmax = (4 * 1024); 132 133#define KN_HASHSIZE 64 /* XXX should be tunable */ 134#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 135 136extern const struct filterops sig_filtops; 137 138/* 139 * Table for for all system-defined filters. 140 * These should be listed in the numeric order of the EVFILT_* defines. 141 * If filtops is NULL, the filter isn't implemented in NetBSD. 142 * End of list is when name is NULL. 143 * 144 * Note that 'refcnt' is meaningless for built-in filters. 145 */ 146struct kfilter { 147 const char *name; /* name of filter */ 148 uint32_t filter; /* id of filter */ 149 unsigned refcnt; /* reference count */ 150 const struct filterops *filtops;/* operations for filter */ 151 size_t namelen; /* length of name string */ 152}; 153 154/* System defined filters */ 155static struct kfilter sys_kfilters[] = { 156 { "EVFILT_READ", EVFILT_READ, 0, &file_filtops, 0 }, 157 { "EVFILT_WRITE", EVFILT_WRITE, 0, &file_filtops, 0, }, 158 { "EVFILT_AIO", EVFILT_AIO, 0, NULL, 0 }, 159 { "EVFILT_VNODE", EVFILT_VNODE, 0, &file_filtops, 0 }, 160 { "EVFILT_PROC", EVFILT_PROC, 0, &proc_filtops, 0 }, 161 { "EVFILT_SIGNAL", EVFILT_SIGNAL, 0, &sig_filtops, 0 }, 162 { "EVFILT_TIMER", EVFILT_TIMER, 0, &timer_filtops, 0 }, 163 { NULL, 0, 0, NULL, 0 }, 164}; 165 166/* User defined kfilters */ 167static struct kfilter *user_kfilters; /* array */ 168static int user_kfilterc; /* current offset */ 169static int user_kfiltermaxc; /* max size so far */ 170static size_t user_kfiltersz; /* size of allocated memory */ 171 172/* Locks */ 173static krwlock_t kqueue_filter_lock; /* lock on filter lists */ 174static kmutex_t kqueue_misc_lock; /* miscellaneous */ 175 176static kauth_listener_t kqueue_listener; 177 178static int 179kqueue_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 180 void *arg0, void *arg1, void *arg2, void *arg3) 181{ 182 struct proc *p; 183 int result; 184 185 result = KAUTH_RESULT_DEFER; 186 p = arg0; 187 188 if (action != KAUTH_PROCESS_KEVENT_FILTER) 189 return result; 190 191 if ((kauth_cred_getuid(p->p_cred) != kauth_cred_getuid(cred) || 192 ISSET(p->p_flag, PK_SUGID))) 193 return result; 194 195 result = KAUTH_RESULT_ALLOW; 196 197 return result; 198} 199 200/* 201 * Initialize the kqueue subsystem. 202 */ 203void 204kqueue_init(void) 205{ 206 207 rw_init(&kqueue_filter_lock); 208 mutex_init(&kqueue_misc_lock, MUTEX_DEFAULT, IPL_NONE); 209 210 kqueue_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS, 211 kqueue_listener_cb, NULL); 212} 213 214/* 215 * Find kfilter entry by name, or NULL if not found. 216 */ 217static struct kfilter * 218kfilter_byname_sys(const char *name) 219{ 220 int i; 221 222 KASSERT(rw_lock_held(&kqueue_filter_lock)); 223 224 for (i = 0; sys_kfilters[i].name != NULL; i++) { 225 if (strcmp(name, sys_kfilters[i].name) == 0) 226 return &sys_kfilters[i]; 227 } 228 return NULL; 229} 230 231static struct kfilter * 232kfilter_byname_user(const char *name) 233{ 234 int i; 235 236 KASSERT(rw_lock_held(&kqueue_filter_lock)); 237 238 /* user filter slots have a NULL name if previously deregistered */ 239 for (i = 0; i < user_kfilterc ; i++) { 240 if (user_kfilters[i].name != NULL && 241 strcmp(name, user_kfilters[i].name) == 0) 242 return &user_kfilters[i]; 243 } 244 return NULL; 245} 246 247static struct kfilter * 248kfilter_byname(const char *name) 249{ 250 struct kfilter *kfilter; 251 252 KASSERT(rw_lock_held(&kqueue_filter_lock)); 253 254 if ((kfilter = kfilter_byname_sys(name)) != NULL) 255 return kfilter; 256 257 return kfilter_byname_user(name); 258} 259 260/* 261 * Find kfilter entry by filter id, or NULL if not found. 262 * Assumes entries are indexed in filter id order, for speed. 263 */ 264static struct kfilter * 265kfilter_byfilter(uint32_t filter) 266{ 267 struct kfilter *kfilter; 268 269 KASSERT(rw_lock_held(&kqueue_filter_lock)); 270 271 if (filter < EVFILT_SYSCOUNT) /* it's a system filter */ 272 kfilter = &sys_kfilters[filter]; 273 else if (user_kfilters != NULL && 274 filter < EVFILT_SYSCOUNT + user_kfilterc) 275 /* it's a user filter */ 276 kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT]; 277 else 278 return (NULL); /* out of range */ 279 KASSERT(kfilter->filter == filter); /* sanity check! */ 280 return (kfilter); 281} 282 283/* 284 * Register a new kfilter. Stores the entry in user_kfilters. 285 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 286 * If retfilter != NULL, the new filterid is returned in it. 287 */ 288int 289kfilter_register(const char *name, const struct filterops *filtops, 290 int *retfilter) 291{ 292 struct kfilter *kfilter; 293 size_t len; 294 int i; 295 296 if (name == NULL || name[0] == '\0' || filtops == NULL) 297 return (EINVAL); /* invalid args */ 298 299 rw_enter(&kqueue_filter_lock, RW_WRITER); 300 if (kfilter_byname(name) != NULL) { 301 rw_exit(&kqueue_filter_lock); 302 return (EEXIST); /* already exists */ 303 } 304 if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT) { 305 rw_exit(&kqueue_filter_lock); 306 return (EINVAL); /* too many */ 307 } 308 309 for (i = 0; i < user_kfilterc; i++) { 310 kfilter = &user_kfilters[i]; 311 if (kfilter->name == NULL) { 312 /* Previously deregistered slot. Reuse. */ 313 goto reuse; 314 } 315 } 316 317 /* check if need to grow user_kfilters */ 318 if (user_kfilterc + 1 > user_kfiltermaxc) { 319 /* Grow in KFILTER_EXTENT chunks. */ 320 user_kfiltermaxc += KFILTER_EXTENT; 321 len = user_kfiltermaxc * sizeof(*kfilter); 322 kfilter = kmem_alloc(len, KM_SLEEP); 323 memset((char *)kfilter + user_kfiltersz, 0, len - user_kfiltersz); 324 if (user_kfilters != NULL) { 325 memcpy(kfilter, user_kfilters, user_kfiltersz); 326 kmem_free(user_kfilters, user_kfiltersz); 327 } 328 user_kfiltersz = len; 329 user_kfilters = kfilter; 330 } 331 /* Adding new slot */ 332 kfilter = &user_kfilters[user_kfilterc++]; 333reuse: 334 kfilter->namelen = strlen(name) + 1; 335 kfilter->name = kmem_alloc(kfilter->namelen, KM_SLEEP); 336 memcpy(__UNCONST(kfilter->name), name, kfilter->namelen); 337 338 kfilter->filter = (kfilter - user_kfilters) + EVFILT_SYSCOUNT; 339 340 kfilter->filtops = kmem_alloc(sizeof(*filtops), KM_SLEEP); 341 memcpy(__UNCONST(kfilter->filtops), filtops, sizeof(*filtops)); 342 343 if (retfilter != NULL) 344 *retfilter = kfilter->filter; 345 rw_exit(&kqueue_filter_lock); 346 347 return (0); 348} 349 350/* 351 * Unregister a kfilter previously registered with kfilter_register. 352 * This retains the filter id, but clears the name and frees filtops (filter 353 * operations), so that the number isn't reused during a boot. 354 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 355 */ 356int 357kfilter_unregister(const char *name) 358{ 359 struct kfilter *kfilter; 360 361 if (name == NULL || name[0] == '\0') 362 return (EINVAL); /* invalid name */ 363 364 rw_enter(&kqueue_filter_lock, RW_WRITER); 365 if (kfilter_byname_sys(name) != NULL) { 366 rw_exit(&kqueue_filter_lock); 367 return (EINVAL); /* can't detach system filters */ 368 } 369 370 kfilter = kfilter_byname_user(name); 371 if (kfilter == NULL) { 372 rw_exit(&kqueue_filter_lock); 373 return (ENOENT); 374 } 375 if (kfilter->refcnt != 0) { 376 rw_exit(&kqueue_filter_lock); 377 return (EBUSY); 378 } 379 380 /* Cast away const (but we know it's safe. */ 381 kmem_free(__UNCONST(kfilter->name), kfilter->namelen); 382 kfilter->name = NULL; /* mark as `not implemented' */ 383 384 if (kfilter->filtops != NULL) { 385 /* Cast away const (but we know it's safe. */ 386 kmem_free(__UNCONST(kfilter->filtops), 387 sizeof(*kfilter->filtops)); 388 kfilter->filtops = NULL; /* mark as `not implemented' */ 389 } 390 rw_exit(&kqueue_filter_lock); 391 392 return (0); 393} 394 395 396/* 397 * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file 398 * descriptors. Calls fileops kqfilter method for given file descriptor. 399 */ 400static int 401filt_fileattach(struct knote *kn) 402{ 403 file_t *fp; 404 405 fp = kn->kn_obj; 406 407 return (*fp->f_ops->fo_kqfilter)(fp, kn); 408} 409 410/* 411 * Filter detach method for EVFILT_READ on kqueue descriptor. 412 */ 413static void 414filt_kqdetach(struct knote *kn) 415{ 416 struct kqueue *kq; 417 418 kq = ((file_t *)kn->kn_obj)->f_data; 419 420 mutex_spin_enter(&kq->kq_lock); 421 SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext); 422 mutex_spin_exit(&kq->kq_lock); 423} 424 425/* 426 * Filter event method for EVFILT_READ on kqueue descriptor. 427 */ 428/*ARGSUSED*/ 429static int 430filt_kqueue(struct knote *kn, long hint) 431{ 432 struct kqueue *kq; 433 int rv; 434 435 kq = ((file_t *)kn->kn_obj)->f_data; 436 437 if (hint != NOTE_SUBMIT) 438 mutex_spin_enter(&kq->kq_lock); 439 kn->kn_data = kq->kq_count; 440 rv = (kn->kn_data > 0); 441 if (hint != NOTE_SUBMIT) 442 mutex_spin_exit(&kq->kq_lock); 443 444 return rv; 445} 446 447/* 448 * Filter attach method for EVFILT_PROC. 449 */ 450static int 451filt_procattach(struct knote *kn) 452{ 453 struct proc *p, *curp; 454 struct lwp *curl; 455 456 curl = curlwp; 457 curp = curl->l_proc; 458 459 mutex_enter(proc_lock); 460 p = proc_find(kn->kn_id); 461 if (p == NULL) { 462 mutex_exit(proc_lock); 463 return ESRCH; 464 } 465 466 /* 467 * Fail if it's not owned by you, or the last exec gave us 468 * setuid/setgid privs (unless you're root). 469 */ 470 mutex_enter(p->p_lock); 471 mutex_exit(proc_lock); 472 if (kauth_authorize_process(curl->l_cred, KAUTH_PROCESS_KEVENT_FILTER, 473 p, NULL, NULL, NULL) != 0) { 474 mutex_exit(p->p_lock); 475 return EACCES; 476 } 477 478 kn->kn_obj = p; 479 kn->kn_flags |= EV_CLEAR; /* automatically set */ 480 481 /* 482 * internal flag indicating registration done by kernel 483 */ 484 if (kn->kn_flags & EV_FLAG1) { 485 kn->kn_data = kn->kn_sdata; /* ppid */ 486 kn->kn_fflags = NOTE_CHILD; 487 kn->kn_flags &= ~EV_FLAG1; 488 } 489 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 490 mutex_exit(p->p_lock); 491 492 return 0; 493} 494 495/* 496 * Filter detach method for EVFILT_PROC. 497 * 498 * The knote may be attached to a different process, which may exit, 499 * leaving nothing for the knote to be attached to. So when the process 500 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 501 * it will be deleted when read out. However, as part of the knote deletion, 502 * this routine is called, so a check is needed to avoid actually performing 503 * a detach, because the original process might not exist any more. 504 */ 505static void 506filt_procdetach(struct knote *kn) 507{ 508 struct proc *p; 509 510 if (kn->kn_status & KN_DETACHED) 511 return; 512 513 p = kn->kn_obj; 514 515 mutex_enter(p->p_lock); 516 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 517 mutex_exit(p->p_lock); 518} 519 520/* 521 * Filter event method for EVFILT_PROC. 522 */ 523static int 524filt_proc(struct knote *kn, long hint) 525{ 526 u_int event, fflag; 527 struct kevent kev; 528 struct kqueue *kq; 529 int error; 530 531 event = (u_int)hint & NOTE_PCTRLMASK; 532 kq = kn->kn_kq; 533 fflag = 0; 534 535 /* If the user is interested in this event, record it. */ 536 if (kn->kn_sfflags & event) 537 fflag |= event; 538 539 if (event == NOTE_EXIT) { 540 /* 541 * Process is gone, so flag the event as finished. 542 * 543 * Detach the knote from watched process and mark 544 * it as such. We can't leave this to kqueue_scan(), 545 * since the process might not exist by then. And we 546 * have to do this now, since psignal KNOTE() is called 547 * also for zombies and we might end up reading freed 548 * memory if the kevent would already be picked up 549 * and knote g/c'ed. 550 */ 551 filt_procdetach(kn); 552 553 mutex_spin_enter(&kq->kq_lock); 554 kn->kn_status |= KN_DETACHED; 555 /* Mark as ONESHOT, so that the knote it g/c'ed when read */ 556 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 557 kn->kn_fflags |= fflag; 558 mutex_spin_exit(&kq->kq_lock); 559 560 return 1; 561 } 562 563 mutex_spin_enter(&kq->kq_lock); 564 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 565 /* 566 * Process forked, and user wants to track the new process, 567 * so attach a new knote to it, and immediately report an 568 * event with the parent's pid. Register knote with new 569 * process. 570 */ 571 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 572 kev.filter = kn->kn_filter; 573 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 574 kev.fflags = kn->kn_sfflags; 575 kev.data = kn->kn_id; /* parent */ 576 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 577 mutex_spin_exit(&kq->kq_lock); 578 error = kqueue_register(kq, &kev); 579 mutex_spin_enter(&kq->kq_lock); 580 if (error != 0) 581 kn->kn_fflags |= NOTE_TRACKERR; 582 } 583 kn->kn_fflags |= fflag; 584 fflag = kn->kn_fflags; 585 mutex_spin_exit(&kq->kq_lock); 586 587 return fflag != 0; 588} 589 590static void 591filt_timerexpire(void *knx) 592{ 593 struct knote *kn = knx; 594 int tticks; 595 596 mutex_enter(&kqueue_misc_lock); 597 kn->kn_data++; 598 knote_activate(kn); 599 if ((kn->kn_flags & EV_ONESHOT) == 0) { 600 tticks = mstohz(kn->kn_sdata); 601 if (tticks <= 0) 602 tticks = 1; 603 callout_schedule((callout_t *)kn->kn_hook, tticks); 604 } 605 mutex_exit(&kqueue_misc_lock); 606} 607 608/* 609 * data contains amount of time to sleep, in milliseconds 610 */ 611static int 612filt_timerattach(struct knote *kn) 613{ 614 callout_t *calloutp; 615 struct kqueue *kq; 616 int tticks; 617 618 tticks = mstohz(kn->kn_sdata); 619 620 /* if the supplied value is under our resolution, use 1 tick */ 621 if (tticks == 0) { 622 if (kn->kn_sdata == 0) 623 return EINVAL; 624 tticks = 1; 625 } 626 627 if (atomic_inc_uint_nv(&kq_ncallouts) >= kq_calloutmax || 628 (calloutp = kmem_alloc(sizeof(*calloutp), KM_NOSLEEP)) == NULL) { 629 atomic_dec_uint(&kq_ncallouts); 630 return ENOMEM; 631 } 632 callout_init(calloutp, CALLOUT_MPSAFE); 633 634 kq = kn->kn_kq; 635 mutex_spin_enter(&kq->kq_lock); 636 kn->kn_flags |= EV_CLEAR; /* automatically set */ 637 kn->kn_hook = calloutp; 638 mutex_spin_exit(&kq->kq_lock); 639 640 callout_reset(calloutp, tticks, filt_timerexpire, kn); 641 642 return (0); 643} 644 645static void 646filt_timerdetach(struct knote *kn) 647{ 648 callout_t *calloutp; 649 650 calloutp = (callout_t *)kn->kn_hook; 651 callout_halt(calloutp, NULL); 652 callout_destroy(calloutp); 653 kmem_free(calloutp, sizeof(*calloutp)); 654 atomic_dec_uint(&kq_ncallouts); 655} 656 657static int 658filt_timer(struct knote *kn, long hint) 659{ 660 int rv; 661 662 mutex_enter(&kqueue_misc_lock); 663 rv = (kn->kn_data != 0); 664 mutex_exit(&kqueue_misc_lock); 665 666 return rv; 667} 668 669/* 670 * filt_seltrue: 671 * 672 * This filter "event" routine simulates seltrue(). 673 */ 674int 675filt_seltrue(struct knote *kn, long hint) 676{ 677 678 /* 679 * We don't know how much data can be read/written, 680 * but we know that it *can* be. This is about as 681 * good as select/poll does as well. 682 */ 683 kn->kn_data = 0; 684 return (1); 685} 686 687/* 688 * This provides full kqfilter entry for device switch tables, which 689 * has same effect as filter using filt_seltrue() as filter method. 690 */ 691static void 692filt_seltruedetach(struct knote *kn) 693{ 694 /* Nothing to do */ 695} 696 697const struct filterops seltrue_filtops = 698 { 1, NULL, filt_seltruedetach, filt_seltrue }; 699 700int 701seltrue_kqfilter(dev_t dev, struct knote *kn) 702{ 703 switch (kn->kn_filter) { 704 case EVFILT_READ: 705 case EVFILT_WRITE: 706 kn->kn_fop = &seltrue_filtops; 707 break; 708 default: 709 return (EINVAL); 710 } 711 712 /* Nothing more to do */ 713 return (0); 714} 715 716/* 717 * kqueue(2) system call. 718 */ 719static int 720kqueue1(struct lwp *l, int flags, register_t *retval) 721{ 722 struct kqueue *kq; 723 file_t *fp; 724 int fd, error; 725 726 if ((error = fd_allocfile(&fp, &fd)) != 0) 727 return error; 728 fp->f_flag = FREAD | FWRITE | (flags & (FNONBLOCK|FNOSIGPIPE)); 729 fp->f_type = DTYPE_KQUEUE; 730 fp->f_ops = &kqueueops; 731 kq = kmem_zalloc(sizeof(*kq), KM_SLEEP); 732 mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED); 733 cv_init(&kq->kq_cv, "kqueue"); 734 selinit(&kq->kq_sel); 735 TAILQ_INIT(&kq->kq_head); 736 fp->f_data = kq; 737 *retval = fd; 738 kq->kq_fdp = curlwp->l_fd; 739 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 740 fd_affix(curproc, fp, fd); 741 return error; 742} 743 744/* 745 * kqueue(2) system call. 746 */ 747int 748sys_kqueue(struct lwp *l, const void *v, register_t *retval) 749{ 750 return kqueue1(l, 0, retval); 751} 752 753int 754sys_kqueue1(struct lwp *l, const struct sys_kqueue1_args *uap, 755 register_t *retval) 756{ 757 /* { 758 syscallarg(int) flags; 759 } */ 760 return kqueue1(l, SCARG(uap, flags), retval); 761} 762 763/* 764 * kevent(2) system call. 765 */ 766int 767kevent_fetch_changes(void *private, const struct kevent *changelist, 768 struct kevent *changes, size_t index, int n) 769{ 770 771 return copyin(changelist + index, changes, n * sizeof(*changes)); 772} 773 774int 775kevent_put_events(void *private, struct kevent *events, 776 struct kevent *eventlist, size_t index, int n) 777{ 778 779 return copyout(events, eventlist + index, n * sizeof(*events)); 780} 781 782static const struct kevent_ops kevent_native_ops = { 783 .keo_private = NULL, 784 .keo_fetch_timeout = copyin, 785 .keo_fetch_changes = kevent_fetch_changes, 786 .keo_put_events = kevent_put_events, 787}; 788 789int 790sys___kevent50(struct lwp *l, const struct sys___kevent50_args *uap, 791 register_t *retval) 792{ 793 /* { 794 syscallarg(int) fd; 795 syscallarg(const struct kevent *) changelist; 796 syscallarg(size_t) nchanges; 797 syscallarg(struct kevent *) eventlist; 798 syscallarg(size_t) nevents; 799 syscallarg(const struct timespec *) timeout; 800 } */ 801 802 return kevent1(retval, SCARG(uap, fd), SCARG(uap, changelist), 803 SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents), 804 SCARG(uap, timeout), &kevent_native_ops); 805} 806 807int 808kevent1(register_t *retval, int fd, 809 const struct kevent *changelist, size_t nchanges, 810 struct kevent *eventlist, size_t nevents, 811 const struct timespec *timeout, 812 const struct kevent_ops *keops) 813{ 814 struct kevent *kevp; 815 struct kqueue *kq; 816 struct timespec ts; 817 size_t i, n, ichange; 818 int nerrors, error; 819 struct kevent kevbuf[8]; /* approx 300 bytes on 64-bit */ 820 file_t *fp; 821 822 /* check that we're dealing with a kq */ 823 fp = fd_getfile(fd); 824 if (fp == NULL) 825 return (EBADF); 826 827 if (fp->f_type != DTYPE_KQUEUE) { 828 fd_putfile(fd); 829 return (EBADF); 830 } 831 832 if (timeout != NULL) { 833 error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts)); 834 if (error) 835 goto done; 836 timeout = &ts; 837 } 838 839 kq = (struct kqueue *)fp->f_data; 840 nerrors = 0; 841 ichange = 0; 842 843 /* traverse list of events to register */ 844 while (nchanges > 0) { 845 n = MIN(nchanges, __arraycount(kevbuf)); 846 error = (*keops->keo_fetch_changes)(keops->keo_private, 847 changelist, kevbuf, ichange, n); 848 if (error) 849 goto done; 850 for (i = 0; i < n; i++) { 851 kevp = &kevbuf[i]; 852 kevp->flags &= ~EV_SYSFLAGS; 853 /* register each knote */ 854 error = kqueue_register(kq, kevp); 855 if (error) { 856 if (nevents != 0) { 857 kevp->flags = EV_ERROR; 858 kevp->data = error; 859 error = (*keops->keo_put_events) 860 (keops->keo_private, kevp, 861 eventlist, nerrors, 1); 862 if (error) 863 goto done; 864 nevents--; 865 nerrors++; 866 } else { 867 goto done; 868 } 869 } 870 } 871 nchanges -= n; /* update the results */ 872 ichange += n; 873 } 874 if (nerrors) { 875 *retval = nerrors; 876 error = 0; 877 goto done; 878 } 879 880 /* actually scan through the events */ 881 error = kqueue_scan(fp, nevents, eventlist, timeout, retval, keops, 882 kevbuf, __arraycount(kevbuf)); 883 done: 884 fd_putfile(fd); 885 return (error); 886} 887 888/* 889 * Register a given kevent kev onto the kqueue 890 */ 891static int 892kqueue_register(struct kqueue *kq, struct kevent *kev) 893{ 894 struct kfilter *kfilter; 895 filedesc_t *fdp; 896 file_t *fp; 897 fdfile_t *ff; 898 struct knote *kn, *newkn; 899 struct klist *list; 900 int error, fd, rv; 901 902 fdp = kq->kq_fdp; 903 fp = NULL; 904 kn = NULL; 905 error = 0; 906 fd = 0; 907 908 newkn = kmem_zalloc(sizeof(*newkn), KM_SLEEP); 909 910 rw_enter(&kqueue_filter_lock, RW_READER); 911 kfilter = kfilter_byfilter(kev->filter); 912 if (kfilter == NULL || kfilter->filtops == NULL) { 913 /* filter not found nor implemented */ 914 rw_exit(&kqueue_filter_lock); 915 kmem_free(newkn, sizeof(*newkn)); 916 return (EINVAL); 917 } 918 919 /* search if knote already exists */ 920 if (kfilter->filtops->f_isfd) { 921 /* monitoring a file descriptor */ 922 fd = kev->ident; 923 if ((fp = fd_getfile(fd)) == NULL) { 924 rw_exit(&kqueue_filter_lock); 925 kmem_free(newkn, sizeof(*newkn)); 926 return EBADF; 927 } 928 mutex_enter(&fdp->fd_lock); 929 ff = fdp->fd_dt->dt_ff[fd]; 930 if (fd <= fdp->fd_lastkqfile) { 931 SLIST_FOREACH(kn, &ff->ff_knlist, kn_link) { 932 if (kq == kn->kn_kq && 933 kev->filter == kn->kn_filter) 934 break; 935 } 936 } 937 } else { 938 /* 939 * not monitoring a file descriptor, so 940 * lookup knotes in internal hash table 941 */ 942 mutex_enter(&fdp->fd_lock); 943 if (fdp->fd_knhashmask != 0) { 944 list = &fdp->fd_knhash[ 945 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 946 SLIST_FOREACH(kn, list, kn_link) { 947 if (kev->ident == kn->kn_id && 948 kq == kn->kn_kq && 949 kev->filter == kn->kn_filter) 950 break; 951 } 952 } 953 } 954 955 /* 956 * kn now contains the matching knote, or NULL if no match 957 */ 958 if (kev->flags & EV_ADD) { 959 if (kn == NULL) { 960 /* create new knote */ 961 kn = newkn; 962 newkn = NULL; 963 kn->kn_obj = fp; 964 kn->kn_id = kev->ident; 965 kn->kn_kq = kq; 966 kn->kn_fop = kfilter->filtops; 967 kn->kn_kfilter = kfilter; 968 kn->kn_sfflags = kev->fflags; 969 kn->kn_sdata = kev->data; 970 kev->fflags = 0; 971 kev->data = 0; 972 kn->kn_kevent = *kev; 973 974 /* 975 * apply reference count to knote structure, and 976 * do not release it at the end of this routine. 977 */ 978 fp = NULL; 979 980 if (!kn->kn_fop->f_isfd) { 981 /* 982 * If knote is not on an fd, store on 983 * internal hash table. 984 */ 985 if (fdp->fd_knhashmask == 0) { 986 /* XXXAD can block with fd_lock held */ 987 fdp->fd_knhash = hashinit(KN_HASHSIZE, 988 HASH_LIST, true, 989 &fdp->fd_knhashmask); 990 } 991 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, 992 fdp->fd_knhashmask)]; 993 } else { 994 /* Otherwise, knote is on an fd. */ 995 list = (struct klist *) 996 &fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist; 997 if ((int)kn->kn_id > fdp->fd_lastkqfile) 998 fdp->fd_lastkqfile = kn->kn_id; 999 } 1000 SLIST_INSERT_HEAD(list, kn, kn_link); 1001 1002 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1003 error = (*kfilter->filtops->f_attach)(kn); 1004 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1005 if (error != 0) { 1006#ifdef DIAGNOSTIC 1007 printf("%s: event not supported for file type" 1008 " %d\n", __func__, fp ? fp->f_type : -1); 1009#endif 1010 /* knote_detach() drops fdp->fd_lock */ 1011 knote_detach(kn, fdp, false); 1012 goto done; 1013 } 1014 atomic_inc_uint(&kfilter->refcnt); 1015 } else { 1016 /* 1017 * The user may change some filter values after the 1018 * initial EV_ADD, but doing so will not reset any 1019 * filter which have already been triggered. 1020 */ 1021 kn->kn_sfflags = kev->fflags; 1022 kn->kn_sdata = kev->data; 1023 kn->kn_kevent.udata = kev->udata; 1024 } 1025 /* 1026 * We can get here if we are trying to attach 1027 * an event to a file descriptor that does not 1028 * support events, and the attach routine is 1029 * broken and does not return an error. 1030 */ 1031 KASSERT(kn->kn_fop->f_event != NULL); 1032 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1033 rv = (*kn->kn_fop->f_event)(kn, 0); 1034 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1035 if (rv) 1036 knote_activate(kn); 1037 } else { 1038 if (kn == NULL) { 1039 error = ENOENT; 1040 mutex_exit(&fdp->fd_lock); 1041 goto done; 1042 } 1043 if (kev->flags & EV_DELETE) { 1044 /* knote_detach() drops fdp->fd_lock */ 1045 knote_detach(kn, fdp, true); 1046 goto done; 1047 } 1048 } 1049 1050 /* disable knote */ 1051 if ((kev->flags & EV_DISABLE)) { 1052 mutex_spin_enter(&kq->kq_lock); 1053 if ((kn->kn_status & KN_DISABLED) == 0) 1054 kn->kn_status |= KN_DISABLED; 1055 mutex_spin_exit(&kq->kq_lock); 1056 } 1057 1058 /* enable knote */ 1059 if ((kev->flags & EV_ENABLE)) { 1060 knote_enqueue(kn); 1061 } 1062 mutex_exit(&fdp->fd_lock); 1063 done: 1064 rw_exit(&kqueue_filter_lock); 1065 if (newkn != NULL) 1066 kmem_free(newkn, sizeof(*newkn)); 1067 if (fp != NULL) 1068 fd_putfile(fd); 1069 return (error); 1070} 1071 1072#if defined(DEBUG) 1073static void 1074kq_check(struct kqueue *kq) 1075{ 1076 const struct knote *kn; 1077 int count; 1078 int nmarker; 1079 1080 KASSERT(mutex_owned(&kq->kq_lock)); 1081 KASSERT(kq->kq_count >= 0); 1082 1083 count = 0; 1084 nmarker = 0; 1085 TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) { 1086 if ((kn->kn_status & (KN_MARKER | KN_QUEUED)) == 0) { 1087 panic("%s: kq=%p kn=%p inconsist 1", __func__, kq, kn); 1088 } 1089 if ((kn->kn_status & KN_MARKER) == 0) { 1090 if (kn->kn_kq != kq) { 1091 panic("%s: kq=%p kn=%p inconsist 2", 1092 __func__, kq, kn); 1093 } 1094 if ((kn->kn_status & KN_ACTIVE) == 0) { 1095 panic("%s: kq=%p kn=%p: not active", 1096 __func__, kq, kn); 1097 } 1098 count++; 1099 if (count > kq->kq_count) { 1100 goto bad; 1101 } 1102 } else { 1103 nmarker++; 1104#if 0 1105 if (nmarker > 10000) { 1106 panic("%s: kq=%p too many markers: %d != %d, " 1107 "nmarker=%d", 1108 __func__, kq, kq->kq_count, count, nmarker); 1109 } 1110#endif 1111 } 1112 } 1113 if (kq->kq_count != count) { 1114bad: 1115 panic("%s: kq=%p inconsist 3: %d != %d, nmarker=%d", 1116 __func__, kq, kq->kq_count, count, nmarker); 1117 } 1118} 1119#else /* defined(DEBUG) */ 1120#define kq_check(a) /* nothing */ 1121#endif /* defined(DEBUG) */ 1122 1123/* 1124 * Scan through the list of events on fp (for a maximum of maxevents), 1125 * returning the results in to ulistp. Timeout is determined by tsp; if 1126 * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait 1127 * as appropriate. 1128 */ 1129static int 1130kqueue_scan(file_t *fp, size_t maxevents, struct kevent *ulistp, 1131 const struct timespec *tsp, register_t *retval, 1132 const struct kevent_ops *keops, struct kevent *kevbuf, 1133 size_t kevcnt) 1134{ 1135 struct kqueue *kq; 1136 struct kevent *kevp; 1137 struct timespec ats, sleepts; 1138 struct knote *kn, *marker; 1139 size_t count, nkev, nevents; 1140 int timeout, error, rv; 1141 filedesc_t *fdp; 1142 1143 fdp = curlwp->l_fd; 1144 kq = fp->f_data; 1145 count = maxevents; 1146 nkev = nevents = error = 0; 1147 if (count == 0) { 1148 *retval = 0; 1149 return 0; 1150 } 1151 1152 if (tsp) { /* timeout supplied */ 1153 ats = *tsp; 1154 if (inittimeleft(&ats, &sleepts) == -1) { 1155 *retval = maxevents; 1156 return EINVAL; 1157 } 1158 timeout = tstohz(&ats); 1159 if (timeout <= 0) 1160 timeout = -1; /* do poll */ 1161 } else { 1162 /* no timeout, wait forever */ 1163 timeout = 0; 1164 } 1165 1166 marker = kmem_zalloc(sizeof(*marker), KM_SLEEP); 1167 marker->kn_status = KN_MARKER; 1168 mutex_spin_enter(&kq->kq_lock); 1169 retry: 1170 kevp = kevbuf; 1171 if (kq->kq_count == 0) { 1172 if (timeout >= 0) { 1173 error = cv_timedwait_sig(&kq->kq_cv, 1174 &kq->kq_lock, timeout); 1175 if (error == 0) { 1176 if (tsp == NULL || (timeout = 1177 gettimeleft(&ats, &sleepts)) > 0) 1178 goto retry; 1179 } else { 1180 /* don't restart after signals... */ 1181 if (error == ERESTART) 1182 error = EINTR; 1183 if (error == EWOULDBLOCK) 1184 error = 0; 1185 } 1186 } 1187 } else { 1188 /* mark end of knote list */ 1189 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1190 1191 while (count != 0) { 1192 kn = TAILQ_FIRST(&kq->kq_head); /* get next knote */ 1193 while ((kn->kn_status & KN_MARKER) != 0) { 1194 if (kn == marker) { 1195 /* it's our marker, stop */ 1196 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1197 if (count < maxevents || (tsp != NULL && 1198 (timeout = gettimeleft(&ats, 1199 &sleepts)) <= 0)) 1200 goto done; 1201 goto retry; 1202 } 1203 /* someone else's marker. */ 1204 kn = TAILQ_NEXT(kn, kn_tqe); 1205 } 1206 kq_check(kq); 1207 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1208 kq->kq_count--; 1209 kn->kn_status &= ~KN_QUEUED; 1210 kq_check(kq); 1211 if (kn->kn_status & KN_DISABLED) { 1212 /* don't want disabled events */ 1213 continue; 1214 } 1215 if ((kn->kn_flags & EV_ONESHOT) == 0) { 1216 mutex_spin_exit(&kq->kq_lock); 1217 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1218 rv = (*kn->kn_fop->f_event)(kn, 0); 1219 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1220 mutex_spin_enter(&kq->kq_lock); 1221 /* Re-poll if note was re-enqueued. */ 1222 if ((kn->kn_status & KN_QUEUED) != 0) 1223 continue; 1224 if (rv == 0) { 1225 /* 1226 * non-ONESHOT event that hasn't 1227 * triggered again, so de-queue. 1228 */ 1229 kn->kn_status &= ~KN_ACTIVE; 1230 continue; 1231 } 1232 } 1233 /* XXXAD should be got from f_event if !oneshot. */ 1234 *kevp++ = kn->kn_kevent; 1235 nkev++; 1236 if (kn->kn_flags & EV_ONESHOT) { 1237 /* delete ONESHOT events after retrieval */ 1238 mutex_spin_exit(&kq->kq_lock); 1239 mutex_enter(&fdp->fd_lock); 1240 knote_detach(kn, fdp, true); 1241 mutex_spin_enter(&kq->kq_lock); 1242 } else if (kn->kn_flags & EV_CLEAR) { 1243 /* clear state after retrieval */ 1244 kn->kn_data = 0; 1245 kn->kn_fflags = 0; 1246 kn->kn_status &= ~KN_ACTIVE; 1247 } else { 1248 /* add event back on list */ 1249 kq_check(kq); 1250 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1251 kq->kq_count++; 1252 kn->kn_status |= KN_QUEUED; 1253 kq_check(kq); 1254 } 1255 if (nkev == kevcnt) { 1256 /* do copyouts in kevcnt chunks */ 1257 mutex_spin_exit(&kq->kq_lock); 1258 error = (*keops->keo_put_events) 1259 (keops->keo_private, 1260 kevbuf, ulistp, nevents, nkev); 1261 mutex_spin_enter(&kq->kq_lock); 1262 nevents += nkev; 1263 nkev = 0; 1264 kevp = kevbuf; 1265 } 1266 count--; 1267 if (error != 0 || count == 0) { 1268 /* remove marker */ 1269 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1270 break; 1271 } 1272 } 1273 } 1274 done: 1275 mutex_spin_exit(&kq->kq_lock); 1276 if (marker != NULL) 1277 kmem_free(marker, sizeof(*marker)); 1278 if (nkev != 0) { 1279 /* copyout remaining events */ 1280 error = (*keops->keo_put_events)(keops->keo_private, 1281 kevbuf, ulistp, nevents, nkev); 1282 } 1283 *retval = maxevents - count; 1284 1285 return error; 1286} 1287 1288/* 1289 * fileops ioctl method for a kqueue descriptor. 1290 * 1291 * Two ioctls are currently supported. They both use struct kfilter_mapping: 1292 * KFILTER_BYNAME find name for filter, and return result in 1293 * name, which is of size len. 1294 * KFILTER_BYFILTER find filter for name. len is ignored. 1295 */ 1296/*ARGSUSED*/ 1297static int 1298kqueue_ioctl(file_t *fp, u_long com, void *data) 1299{ 1300 struct kfilter_mapping *km; 1301 const struct kfilter *kfilter; 1302 char *name; 1303 int error; 1304 1305 km = data; 1306 error = 0; 1307 name = kmem_alloc(KFILTER_MAXNAME, KM_SLEEP); 1308 1309 switch (com) { 1310 case KFILTER_BYFILTER: /* convert filter -> name */ 1311 rw_enter(&kqueue_filter_lock, RW_READER); 1312 kfilter = kfilter_byfilter(km->filter); 1313 if (kfilter != NULL) { 1314 strlcpy(name, kfilter->name, KFILTER_MAXNAME); 1315 rw_exit(&kqueue_filter_lock); 1316 error = copyoutstr(name, km->name, km->len, NULL); 1317 } else { 1318 rw_exit(&kqueue_filter_lock); 1319 error = ENOENT; 1320 } 1321 break; 1322 1323 case KFILTER_BYNAME: /* convert name -> filter */ 1324 error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL); 1325 if (error) { 1326 break; 1327 } 1328 rw_enter(&kqueue_filter_lock, RW_READER); 1329 kfilter = kfilter_byname(name); 1330 if (kfilter != NULL) 1331 km->filter = kfilter->filter; 1332 else 1333 error = ENOENT; 1334 rw_exit(&kqueue_filter_lock); 1335 break; 1336 1337 default: 1338 error = ENOTTY; 1339 break; 1340 1341 } 1342 kmem_free(name, KFILTER_MAXNAME); 1343 return (error); 1344} 1345 1346/* 1347 * fileops fcntl method for a kqueue descriptor. 1348 */ 1349static int 1350kqueue_fcntl(file_t *fp, u_int com, void *data) 1351{ 1352 1353 return (ENOTTY); 1354} 1355 1356/* 1357 * fileops poll method for a kqueue descriptor. 1358 * Determine if kqueue has events pending. 1359 */ 1360static int 1361kqueue_poll(file_t *fp, int events) 1362{ 1363 struct kqueue *kq; 1364 int revents; 1365 1366 kq = fp->f_data; 1367 1368 revents = 0; 1369 if (events & (POLLIN | POLLRDNORM)) { 1370 mutex_spin_enter(&kq->kq_lock); 1371 if (kq->kq_count != 0) { 1372 revents |= events & (POLLIN | POLLRDNORM); 1373 } else { 1374 selrecord(curlwp, &kq->kq_sel); 1375 } 1376 kq_check(kq); 1377 mutex_spin_exit(&kq->kq_lock); 1378 } 1379 1380 return revents; 1381} 1382 1383/* 1384 * fileops stat method for a kqueue descriptor. 1385 * Returns dummy info, with st_size being number of events pending. 1386 */ 1387static int 1388kqueue_stat(file_t *fp, struct stat *st) 1389{ 1390 struct kqueue *kq; 1391 1392 kq = fp->f_data; 1393 1394 memset(st, 0, sizeof(*st)); 1395 st->st_size = kq->kq_count; 1396 st->st_blksize = sizeof(struct kevent); 1397 st->st_mode = S_IFIFO; 1398 1399 return 0; 1400} 1401 1402static void 1403kqueue_doclose(struct kqueue *kq, struct klist *list, int fd) 1404{ 1405 struct knote *kn; 1406 filedesc_t *fdp; 1407 1408 fdp = kq->kq_fdp; 1409 1410 KASSERT(mutex_owned(&fdp->fd_lock)); 1411 1412 for (kn = SLIST_FIRST(list); kn != NULL;) { 1413 if (kq != kn->kn_kq) { 1414 kn = SLIST_NEXT(kn, kn_link); 1415 continue; 1416 } 1417 knote_detach(kn, fdp, true); 1418 mutex_enter(&fdp->fd_lock); 1419 kn = SLIST_FIRST(list); 1420 } 1421} 1422 1423 1424/* 1425 * fileops close method for a kqueue descriptor. 1426 */ 1427static int 1428kqueue_close(file_t *fp) 1429{ 1430 struct kqueue *kq; 1431 filedesc_t *fdp; 1432 fdfile_t *ff; 1433 int i; 1434 1435 kq = fp->f_data; 1436 fp->f_data = NULL; 1437 fp->f_type = 0; 1438 fdp = curlwp->l_fd; 1439 1440 mutex_enter(&fdp->fd_lock); 1441 for (i = 0; i <= fdp->fd_lastkqfile; i++) { 1442 if ((ff = fdp->fd_dt->dt_ff[i]) == NULL) 1443 continue; 1444 kqueue_doclose(kq, (struct klist *)&ff->ff_knlist, i); 1445 } 1446 if (fdp->fd_knhashmask != 0) { 1447 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 1448 kqueue_doclose(kq, &fdp->fd_knhash[i], -1); 1449 } 1450 } 1451 mutex_exit(&fdp->fd_lock); 1452 1453 KASSERT(kq->kq_count == 0); 1454 mutex_destroy(&kq->kq_lock); 1455 cv_destroy(&kq->kq_cv); 1456 seldestroy(&kq->kq_sel); 1457 kmem_free(kq, sizeof(*kq)); 1458 1459 return (0); 1460} 1461 1462/* 1463 * struct fileops kqfilter method for a kqueue descriptor. 1464 * Event triggered when monitored kqueue changes. 1465 */ 1466static int 1467kqueue_kqfilter(file_t *fp, struct knote *kn) 1468{ 1469 struct kqueue *kq; 1470 filedesc_t *fdp; 1471 1472 kq = ((file_t *)kn->kn_obj)->f_data; 1473 1474 KASSERT(fp == kn->kn_obj); 1475 1476 if (kn->kn_filter != EVFILT_READ) 1477 return 1; 1478 1479 kn->kn_fop = &kqread_filtops; 1480 fdp = curlwp->l_fd; 1481 mutex_enter(&kq->kq_lock); 1482 SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext); 1483 mutex_exit(&kq->kq_lock); 1484 1485 return 0; 1486} 1487 1488 1489/* 1490 * Walk down a list of knotes, activating them if their event has 1491 * triggered. The caller's object lock (e.g. device driver lock) 1492 * must be held. 1493 */ 1494void 1495knote(struct klist *list, long hint) 1496{ 1497 struct knote *kn, *tmpkn; 1498 1499 SLIST_FOREACH_SAFE(kn, list, kn_selnext, tmpkn) { 1500 if ((*kn->kn_fop->f_event)(kn, hint)) 1501 knote_activate(kn); 1502 } 1503} 1504 1505/* 1506 * Remove all knotes referencing a specified fd 1507 */ 1508void 1509knote_fdclose(int fd) 1510{ 1511 struct klist *list; 1512 struct knote *kn; 1513 filedesc_t *fdp; 1514 1515 fdp = curlwp->l_fd; 1516 list = (struct klist *)&fdp->fd_dt->dt_ff[fd]->ff_knlist; 1517 mutex_enter(&fdp->fd_lock); 1518 while ((kn = SLIST_FIRST(list)) != NULL) { 1519 knote_detach(kn, fdp, true); 1520 mutex_enter(&fdp->fd_lock); 1521 } 1522 mutex_exit(&fdp->fd_lock); 1523} 1524 1525/* 1526 * Drop knote. Called with fdp->fd_lock held, and will drop before 1527 * returning. 1528 */ 1529static void 1530knote_detach(struct knote *kn, filedesc_t *fdp, bool dofop) 1531{ 1532 struct klist *list; 1533 struct kqueue *kq; 1534 1535 kq = kn->kn_kq; 1536 1537 KASSERT((kn->kn_status & KN_MARKER) == 0); 1538 KASSERT(mutex_owned(&fdp->fd_lock)); 1539 1540 /* Remove from monitored object. */ 1541 if (dofop) { 1542 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1543 (*kn->kn_fop->f_detach)(kn); 1544 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1545 } 1546 1547 /* Remove from descriptor table. */ 1548 if (kn->kn_fop->f_isfd) 1549 list = (struct klist *)&fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist; 1550 else 1551 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1552 1553 SLIST_REMOVE(list, kn, knote, kn_link); 1554 1555 /* Remove from kqueue. */ 1556 /* XXXAD should verify not in use by kqueue_scan. */ 1557 mutex_spin_enter(&kq->kq_lock); 1558 if ((kn->kn_status & KN_QUEUED) != 0) { 1559 kq_check(kq); 1560 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1561 kn->kn_status &= ~KN_QUEUED; 1562 kq->kq_count--; 1563 kq_check(kq); 1564 } 1565 mutex_spin_exit(&kq->kq_lock); 1566 1567 mutex_exit(&fdp->fd_lock); 1568 if (kn->kn_fop->f_isfd) 1569 fd_putfile(kn->kn_id); 1570 atomic_dec_uint(&kn->kn_kfilter->refcnt); 1571 kmem_free(kn, sizeof(*kn)); 1572} 1573 1574/* 1575 * Queue new event for knote. 1576 */ 1577static void 1578knote_enqueue(struct knote *kn) 1579{ 1580 struct kqueue *kq; 1581 1582 KASSERT((kn->kn_status & KN_MARKER) == 0); 1583 1584 kq = kn->kn_kq; 1585 1586 mutex_spin_enter(&kq->kq_lock); 1587 if ((kn->kn_status & KN_DISABLED) != 0) { 1588 kn->kn_status &= ~KN_DISABLED; 1589 } 1590 if ((kn->kn_status & (KN_ACTIVE | KN_QUEUED)) == KN_ACTIVE) { 1591 kq_check(kq); 1592 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1593 kn->kn_status |= KN_QUEUED; 1594 kq->kq_count++; 1595 kq_check(kq); 1596 cv_broadcast(&kq->kq_cv); 1597 selnotify(&kq->kq_sel, 0, NOTE_SUBMIT); 1598 } 1599 mutex_spin_exit(&kq->kq_lock); 1600} 1601/* 1602 * Queue new event for knote. 1603 */ 1604static void 1605knote_activate(struct knote *kn) 1606{ 1607 struct kqueue *kq; 1608 1609 KASSERT((kn->kn_status & KN_MARKER) == 0); 1610 1611 kq = kn->kn_kq; 1612 1613 mutex_spin_enter(&kq->kq_lock); 1614 kn->kn_status |= KN_ACTIVE; 1615 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) { 1616 kq_check(kq); 1617 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1618 kn->kn_status |= KN_QUEUED; 1619 kq->kq_count++; 1620 kq_check(kq); 1621 cv_broadcast(&kq->kq_cv); 1622 selnotify(&kq->kq_sel, 0, NOTE_SUBMIT); 1623 } 1624 mutex_spin_exit(&kq->kq_lock); 1625} 1626