1/*- 2 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 3 * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org> 4 * Copyright (c) 2009 Apple, Inc. 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 29#include <sys/cdefs.h> 30__FBSDID("$FreeBSD$"); 31 32#include "opt_ktrace.h" 33 34#include <sys/param.h> 35#include <sys/systm.h> 36#include <sys/capability.h> 37#include <sys/kernel.h> 38#include <sys/lock.h> 39#include <sys/mutex.h> 40#include <sys/proc.h> 41#include <sys/malloc.h> 42#include <sys/unistd.h> 43#include <sys/file.h> 44#include <sys/filedesc.h> 45#include <sys/filio.h> 46#include <sys/fcntl.h> 47#include <sys/kthread.h> 48#include <sys/selinfo.h> 49#include <sys/queue.h> 50#include <sys/event.h> 51#include <sys/eventvar.h> 52#include <sys/poll.h> 53#include <sys/protosw.h> 54#include <sys/sigio.h> 55#include <sys/signalvar.h> 56#include <sys/socket.h> 57#include <sys/socketvar.h> 58#include <sys/stat.h> 59#include <sys/sysctl.h> 60#include <sys/sysproto.h> 61#include <sys/syscallsubr.h> 62#include <sys/taskqueue.h> 63#include <sys/uio.h> 64#ifdef KTRACE 65#include <sys/ktrace.h> 66#endif 67 68#include <vm/uma.h> 69 70static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 71 72/* 73 * This lock is used if multiple kq locks are required. This possibly 74 * should be made into a per proc lock. 75 */ 76static struct mtx kq_global; 77MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 78#define KQ_GLOBAL_LOCK(lck, haslck) do { \ 79 if (!haslck) \ 80 mtx_lock(lck); \ 81 haslck = 1; \ 82} while (0) 83#define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 84 if (haslck) \ 85 mtx_unlock(lck); \ 86 haslck = 0; \ 87} while (0) 88 89TASKQUEUE_DEFINE_THREAD(kqueue); 90 91static int kevent_copyout(void *arg, struct kevent *kevp, int count); 92static int kevent_copyin(void *arg, struct kevent *kevp, int count); 93static int kqueue_register(struct kqueue *kq, struct kevent *kev, 94 struct thread *td, int waitok); 95static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 96static void kqueue_release(struct kqueue *kq, int locked); 97static int kqueue_expand(struct kqueue *kq, struct filterops *fops, 98 uintptr_t ident, int waitok); 99static void kqueue_task(void *arg, int pending); 100static int kqueue_scan(struct kqueue *kq, int maxevents, 101 struct kevent_copyops *k_ops, 102 const struct timespec *timeout, 103 struct kevent *keva, struct thread *td); 104static void kqueue_wakeup(struct kqueue *kq); 105static struct filterops *kqueue_fo_find(int filt); 106static void kqueue_fo_release(int filt); 107 108static fo_rdwr_t kqueue_read; 109static fo_rdwr_t kqueue_write; 110static fo_truncate_t kqueue_truncate; 111static fo_ioctl_t kqueue_ioctl; 112static fo_poll_t kqueue_poll; 113static fo_kqfilter_t kqueue_kqfilter; 114static fo_stat_t kqueue_stat; 115static fo_close_t kqueue_close; 116 117static struct fileops kqueueops = { 118 .fo_read = kqueue_read, 119 .fo_write = kqueue_write, 120 .fo_truncate = kqueue_truncate, 121 .fo_ioctl = kqueue_ioctl, 122 .fo_poll = kqueue_poll, 123 .fo_kqfilter = kqueue_kqfilter, 124 .fo_stat = kqueue_stat, 125 .fo_close = kqueue_close, 126 .fo_chmod = invfo_chmod, 127 .fo_chown = invfo_chown, 128}; 129 130static int knote_attach(struct knote *kn, struct kqueue *kq); 131static void knote_drop(struct knote *kn, struct thread *td); 132static void knote_enqueue(struct knote *kn); 133static void knote_dequeue(struct knote *kn); 134static void knote_init(void); 135static struct knote *knote_alloc(int waitok); 136static void knote_free(struct knote *kn); 137 138static void filt_kqdetach(struct knote *kn); 139static int filt_kqueue(struct knote *kn, long hint); 140static int filt_procattach(struct knote *kn); 141static void filt_procdetach(struct knote *kn); 142static int filt_proc(struct knote *kn, long hint); 143static int filt_fileattach(struct knote *kn); 144static void filt_timerexpire(void *knx); 145static int filt_timerattach(struct knote *kn); 146static void filt_timerdetach(struct knote *kn); 147static int filt_timer(struct knote *kn, long hint); 148static int filt_userattach(struct knote *kn); 149static void filt_userdetach(struct knote *kn); 150static int filt_user(struct knote *kn, long hint); 151static void filt_usertouch(struct knote *kn, struct kevent *kev, 152 u_long type); 153 154static struct filterops file_filtops = { 155 .f_isfd = 1, 156 .f_attach = filt_fileattach, 157}; 158static struct filterops kqread_filtops = { 159 .f_isfd = 1, 160 .f_detach = filt_kqdetach, 161 .f_event = filt_kqueue, 162}; 163/* XXX - move to kern_proc.c? */ 164static struct filterops proc_filtops = { 165 .f_isfd = 0, 166 .f_attach = filt_procattach, 167 .f_detach = filt_procdetach, 168 .f_event = filt_proc, 169}; 170static struct filterops timer_filtops = { 171 .f_isfd = 0, 172 .f_attach = filt_timerattach, 173 .f_detach = filt_timerdetach, 174 .f_event = filt_timer, 175}; 176static struct filterops user_filtops = { 177 .f_attach = filt_userattach, 178 .f_detach = filt_userdetach, 179 .f_event = filt_user, 180 .f_touch = filt_usertouch, 181}; 182 183static uma_zone_t knote_zone; 184static int kq_ncallouts = 0; 185static int kq_calloutmax = (4 * 1024); 186SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 187 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 188 189/* XXX - ensure not KN_INFLUX?? */ 190#define KNOTE_ACTIVATE(kn, islock) do { \ 191 if ((islock)) \ 192 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 193 else \ 194 KQ_LOCK((kn)->kn_kq); \ 195 (kn)->kn_status |= KN_ACTIVE; \ 196 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 197 knote_enqueue((kn)); \ 198 if (!(islock)) \ 199 KQ_UNLOCK((kn)->kn_kq); \ 200} while(0) 201#define KQ_LOCK(kq) do { \ 202 mtx_lock(&(kq)->kq_lock); \ 203} while (0) 204#define KQ_FLUX_WAKEUP(kq) do { \ 205 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 206 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 207 wakeup((kq)); \ 208 } \ 209} while (0) 210#define KQ_UNLOCK_FLUX(kq) do { \ 211 KQ_FLUX_WAKEUP(kq); \ 212 mtx_unlock(&(kq)->kq_lock); \ 213} while (0) 214#define KQ_UNLOCK(kq) do { \ 215 mtx_unlock(&(kq)->kq_lock); \ 216} while (0) 217#define KQ_OWNED(kq) do { \ 218 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 219} while (0) 220#define KQ_NOTOWNED(kq) do { \ 221 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 222} while (0) 223#define KN_LIST_LOCK(kn) do { \ 224 if (kn->kn_knlist != NULL) \ 225 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \ 226} while (0) 227#define KN_LIST_UNLOCK(kn) do { \ 228 if (kn->kn_knlist != NULL) \ 229 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \ 230} while (0) 231#define KNL_ASSERT_LOCK(knl, islocked) do { \ 232 if (islocked) \ 233 KNL_ASSERT_LOCKED(knl); \ 234 else \ 235 KNL_ASSERT_UNLOCKED(knl); \ 236} while (0) 237#ifdef INVARIANTS 238#define KNL_ASSERT_LOCKED(knl) do { \ 239 knl->kl_assert_locked((knl)->kl_lockarg); \ 240} while (0) 241#define KNL_ASSERT_UNLOCKED(knl) do { \ 242 knl->kl_assert_unlocked((knl)->kl_lockarg); \ 243} while (0) 244#else /* !INVARIANTS */ 245#define KNL_ASSERT_LOCKED(knl) do {} while(0) 246#define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 247#endif /* INVARIANTS */ 248 249#define KN_HASHSIZE 64 /* XXX should be tunable */ 250#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 251 252static int 253filt_nullattach(struct knote *kn) 254{ 255 256 return (ENXIO); 257}; 258 259struct filterops null_filtops = { 260 .f_isfd = 0, 261 .f_attach = filt_nullattach, 262}; 263 264/* XXX - make SYSINIT to add these, and move into respective modules. */ 265extern struct filterops sig_filtops; 266extern struct filterops fs_filtops; 267 268/* 269 * Table for for all system-defined filters. 270 */ 271static struct mtx filterops_lock; 272MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", 273 MTX_DEF); 274static struct { 275 struct filterops *for_fop; 276 int for_refcnt; 277} sysfilt_ops[EVFILT_SYSCOUNT] = { 278 { &file_filtops }, /* EVFILT_READ */ 279 { &file_filtops }, /* EVFILT_WRITE */ 280 { &null_filtops }, /* EVFILT_AIO */ 281 { &file_filtops }, /* EVFILT_VNODE */ 282 { &proc_filtops }, /* EVFILT_PROC */ 283 { &sig_filtops }, /* EVFILT_SIGNAL */ 284 { &timer_filtops }, /* EVFILT_TIMER */ 285 { &null_filtops }, /* former EVFILT_NETDEV */ 286 { &fs_filtops }, /* EVFILT_FS */ 287 { &null_filtops }, /* EVFILT_LIO */ 288 { &user_filtops }, /* EVFILT_USER */ 289}; 290 291/* 292 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 293 * method. 294 */ 295static int 296filt_fileattach(struct knote *kn) 297{ 298 299 return (fo_kqfilter(kn->kn_fp, kn)); 300} 301 302/*ARGSUSED*/ 303static int 304kqueue_kqfilter(struct file *fp, struct knote *kn) 305{ 306 struct kqueue *kq = kn->kn_fp->f_data; 307 308 if (kn->kn_filter != EVFILT_READ) 309 return (EINVAL); 310 311 kn->kn_status |= KN_KQUEUE; 312 kn->kn_fop = &kqread_filtops; 313 knlist_add(&kq->kq_sel.si_note, kn, 0); 314 315 return (0); 316} 317 318static void 319filt_kqdetach(struct knote *kn) 320{ 321 struct kqueue *kq = kn->kn_fp->f_data; 322 323 knlist_remove(&kq->kq_sel.si_note, kn, 0); 324} 325 326/*ARGSUSED*/ 327static int 328filt_kqueue(struct knote *kn, long hint) 329{ 330 struct kqueue *kq = kn->kn_fp->f_data; 331 332 kn->kn_data = kq->kq_count; 333 return (kn->kn_data > 0); 334} 335 336/* XXX - move to kern_proc.c? */ 337static int 338filt_procattach(struct knote *kn) 339{ 340 struct proc *p; 341 int immediate; 342 int error; 343 344 immediate = 0; 345 p = pfind(kn->kn_id); 346 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 347 p = zpfind(kn->kn_id); 348 immediate = 1; 349 } else if (p != NULL && (p->p_flag & P_WEXIT)) { 350 immediate = 1; 351 } 352 353 if (p == NULL) 354 return (ESRCH); 355 if ((error = p_cansee(curthread, p))) { 356 PROC_UNLOCK(p); 357 return (error); 358 } 359 360 kn->kn_ptr.p_proc = p; 361 kn->kn_flags |= EV_CLEAR; /* automatically set */ 362 363 /* 364 * internal flag indicating registration done by kernel 365 */ 366 if (kn->kn_flags & EV_FLAG1) { 367 kn->kn_data = kn->kn_sdata; /* ppid */ 368 kn->kn_fflags = NOTE_CHILD; 369 kn->kn_flags &= ~EV_FLAG1; 370 } 371 372 if (immediate == 0) 373 knlist_add(&p->p_klist, kn, 1); 374 375 /* 376 * Immediately activate any exit notes if the target process is a 377 * zombie. This is necessary to handle the case where the target 378 * process, e.g. a child, dies before the kevent is registered. 379 */ 380 if (immediate && filt_proc(kn, NOTE_EXIT)) 381 KNOTE_ACTIVATE(kn, 0); 382 383 PROC_UNLOCK(p); 384 385 return (0); 386} 387 388/* 389 * The knote may be attached to a different process, which may exit, 390 * leaving nothing for the knote to be attached to. So when the process 391 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 392 * it will be deleted when read out. However, as part of the knote deletion, 393 * this routine is called, so a check is needed to avoid actually performing 394 * a detach, because the original process does not exist any more. 395 */ 396/* XXX - move to kern_proc.c? */ 397static void 398filt_procdetach(struct knote *kn) 399{ 400 struct proc *p; 401 402 p = kn->kn_ptr.p_proc; 403 knlist_remove(&p->p_klist, kn, 0); 404 kn->kn_ptr.p_proc = NULL; 405} 406 407/* XXX - move to kern_proc.c? */ 408static int 409filt_proc(struct knote *kn, long hint) 410{ 411 struct proc *p = kn->kn_ptr.p_proc; 412 u_int event; 413 414 /* 415 * mask off extra data 416 */ 417 event = (u_int)hint & NOTE_PCTRLMASK; 418 419 /* 420 * if the user is interested in this event, record it. 421 */ 422 if (kn->kn_sfflags & event) 423 kn->kn_fflags |= event; 424 425 /* 426 * process is gone, so flag the event as finished. 427 */ 428 if (event == NOTE_EXIT) { 429 if (!(kn->kn_status & KN_DETACHED)) 430 knlist_remove_inevent(&p->p_klist, kn); 431 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 432 kn->kn_ptr.p_proc = NULL; 433 if (kn->kn_fflags & NOTE_EXIT) 434 kn->kn_data = p->p_xstat; 435 if (kn->kn_fflags == 0) 436 kn->kn_flags |= EV_DROP; 437 return (1); 438 } 439 440 return (kn->kn_fflags != 0); 441} 442 443/* 444 * Called when the process forked. It mostly does the same as the 445 * knote(), activating all knotes registered to be activated when the 446 * process forked. Additionally, for each knote attached to the 447 * parent, check whether user wants to track the new process. If so 448 * attach a new knote to it, and immediately report an event with the 449 * child's pid. 450 */ 451void 452knote_fork(struct knlist *list, int pid) 453{ 454 struct kqueue *kq; 455 struct knote *kn; 456 struct kevent kev; 457 int error; 458 459 if (list == NULL) 460 return; 461 list->kl_lock(list->kl_lockarg); 462 463 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 464 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) 465 continue; 466 kq = kn->kn_kq; 467 KQ_LOCK(kq); 468 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { 469 KQ_UNLOCK(kq); 470 continue; 471 } 472 473 /* 474 * The same as knote(), activate the event. 475 */ 476 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 477 kn->kn_status |= KN_HASKQLOCK; 478 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 479 KNOTE_ACTIVATE(kn, 1); 480 kn->kn_status &= ~KN_HASKQLOCK; 481 KQ_UNLOCK(kq); 482 continue; 483 } 484 485 /* 486 * The NOTE_TRACK case. In addition to the activation 487 * of the event, we need to register new event to 488 * track the child. Drop the locks in preparation for 489 * the call to kqueue_register(). 490 */ 491 kn->kn_status |= KN_INFLUX; 492 KQ_UNLOCK(kq); 493 list->kl_unlock(list->kl_lockarg); 494 495 /* 496 * Activate existing knote and register a knote with 497 * new process. 498 */ 499 kev.ident = pid; 500 kev.filter = kn->kn_filter; 501 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 502 kev.fflags = kn->kn_sfflags; 503 kev.data = kn->kn_id; /* parent */ 504 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 505 error = kqueue_register(kq, &kev, NULL, 0); 506 if (error) 507 kn->kn_fflags |= NOTE_TRACKERR; 508 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 509 KNOTE_ACTIVATE(kn, 0); 510 KQ_LOCK(kq); 511 kn->kn_status &= ~KN_INFLUX; 512 KQ_UNLOCK_FLUX(kq); 513 list->kl_lock(list->kl_lockarg); 514 } 515 list->kl_unlock(list->kl_lockarg); 516} 517 518/* 519 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the 520 * interval timer support code. 521 */ 522static int 523timertoticks(intptr_t data) 524{ 525 struct timeval tv; 526 int tticks; 527 528 tv.tv_sec = data / 1000; 529 tv.tv_usec = (data % 1000) * 1000; 530 tticks = tvtohz(&tv); 531 532 return tticks; 533} 534 535static void 536filt_timerexpire(void *knx) 537{ 538 struct knote *kn = knx; 539 struct callout *calloutp; 540 541 kn->kn_data++; 542 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 543 544 /* 545 * timertoticks() uses tvtohz() which always adds 1 to allow 546 * for the time until the next clock interrupt being strictly 547 * less than 1 clock tick. We don't want that here since we 548 * want to appear to be in sync with the clock interrupt even 549 * when we're delayed. 550 */ 551 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) { 552 calloutp = (struct callout *)kn->kn_hook; 553 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata) - 1, 554 filt_timerexpire, kn); 555 } 556} 557 558/* 559 * data contains amount of time to sleep, in milliseconds 560 */ 561static int 562filt_timerattach(struct knote *kn) 563{ 564 struct callout *calloutp; 565 566 atomic_add_int(&kq_ncallouts, 1); 567 568 if (kq_ncallouts >= kq_calloutmax) { 569 atomic_add_int(&kq_ncallouts, -1); 570 return (ENOMEM); 571 } 572 573 kn->kn_flags |= EV_CLEAR; /* automatically set */ 574 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */ 575 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK); 576 callout_init(calloutp, CALLOUT_MPSAFE); 577 kn->kn_hook = calloutp; 578 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata), 579 filt_timerexpire, kn); 580 581 return (0); 582} 583 584static void 585filt_timerdetach(struct knote *kn) 586{ 587 struct callout *calloutp; 588 589 calloutp = (struct callout *)kn->kn_hook; 590 callout_drain(calloutp); 591 free(calloutp, M_KQUEUE); 592 atomic_add_int(&kq_ncallouts, -1); 593 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */ 594} 595 596static int 597filt_timer(struct knote *kn, long hint) 598{ 599 600 return (kn->kn_data != 0); 601} 602 603static int 604filt_userattach(struct knote *kn) 605{ 606 607 /* 608 * EVFILT_USER knotes are not attached to anything in the kernel. 609 */ 610 kn->kn_hook = NULL; 611 if (kn->kn_fflags & NOTE_TRIGGER) 612 kn->kn_hookid = 1; 613 else 614 kn->kn_hookid = 0; 615 return (0); 616} 617 618static void 619filt_userdetach(__unused struct knote *kn) 620{ 621 622 /* 623 * EVFILT_USER knotes are not attached to anything in the kernel. 624 */ 625} 626 627static int 628filt_user(struct knote *kn, __unused long hint) 629{ 630 631 return (kn->kn_hookid); 632} 633 634static void 635filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 636{ 637 u_int ffctrl; 638 639 switch (type) { 640 case EVENT_REGISTER: 641 if (kev->fflags & NOTE_TRIGGER) 642 kn->kn_hookid = 1; 643 644 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 645 kev->fflags &= NOTE_FFLAGSMASK; 646 switch (ffctrl) { 647 case NOTE_FFNOP: 648 break; 649 650 case NOTE_FFAND: 651 kn->kn_sfflags &= kev->fflags; 652 break; 653 654 case NOTE_FFOR: 655 kn->kn_sfflags |= kev->fflags; 656 break; 657 658 case NOTE_FFCOPY: 659 kn->kn_sfflags = kev->fflags; 660 break; 661 662 default: 663 /* XXX Return error? */ 664 break; 665 } 666 kn->kn_sdata = kev->data; 667 if (kev->flags & EV_CLEAR) { 668 kn->kn_hookid = 0; 669 kn->kn_data = 0; 670 kn->kn_fflags = 0; 671 } 672 break; 673 674 case EVENT_PROCESS: 675 *kev = kn->kn_kevent; 676 kev->fflags = kn->kn_sfflags; 677 kev->data = kn->kn_sdata; 678 if (kn->kn_flags & EV_CLEAR) { 679 kn->kn_hookid = 0; 680 kn->kn_data = 0; 681 kn->kn_fflags = 0; 682 } 683 break; 684 685 default: 686 panic("filt_usertouch() - invalid type (%ld)", type); 687 break; 688 } 689} 690 691int 692sys_kqueue(struct thread *td, struct kqueue_args *uap) 693{ 694 struct filedesc *fdp; 695 struct kqueue *kq; 696 struct file *fp; 697 int fd, error; 698 699 fdp = td->td_proc->p_fd; 700 error = falloc(td, &fp, &fd, 0); 701 if (error) 702 goto done2; 703 704 /* An extra reference on `nfp' has been held for us by falloc(). */ 705 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 706 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK); 707 TAILQ_INIT(&kq->kq_head); 708 kq->kq_fdp = fdp; 709 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 710 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 711 712 FILEDESC_XLOCK(fdp); 713 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 714 FILEDESC_XUNLOCK(fdp); 715 716 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 717 fdrop(fp, td); 718 719 td->td_retval[0] = fd; 720done2: 721 return (error); 722} 723 724#ifndef _SYS_SYSPROTO_H_ 725struct kevent_args { 726 int fd; 727 const struct kevent *changelist; 728 int nchanges; 729 struct kevent *eventlist; 730 int nevents; 731 const struct timespec *timeout; 732}; 733#endif 734int 735sys_kevent(struct thread *td, struct kevent_args *uap) 736{ 737 struct timespec ts, *tsp; 738 struct kevent_copyops k_ops = { uap, 739 kevent_copyout, 740 kevent_copyin}; 741 int error; 742#ifdef KTRACE 743 struct uio ktruio; 744 struct iovec ktriov; 745 struct uio *ktruioin = NULL; 746 struct uio *ktruioout = NULL; 747#endif 748 749 if (uap->timeout != NULL) { 750 error = copyin(uap->timeout, &ts, sizeof(ts)); 751 if (error) 752 return (error); 753 tsp = &ts; 754 } else 755 tsp = NULL; 756 757#ifdef KTRACE 758 if (KTRPOINT(td, KTR_GENIO)) { 759 ktriov.iov_base = uap->changelist; 760 ktriov.iov_len = uap->nchanges * sizeof(struct kevent); 761 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1, 762 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ, 763 .uio_td = td }; 764 ktruioin = cloneuio(&ktruio); 765 ktriov.iov_base = uap->eventlist; 766 ktriov.iov_len = uap->nevents * sizeof(struct kevent); 767 ktruioout = cloneuio(&ktruio); 768 } 769#endif 770 771 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 772 &k_ops, tsp); 773 774#ifdef KTRACE 775 if (ktruioin != NULL) { 776 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent); 777 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0); 778 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent); 779 ktrgenio(uap->fd, UIO_READ, ktruioout, error); 780 } 781#endif 782 783 return (error); 784} 785 786/* 787 * Copy 'count' items into the destination list pointed to by uap->eventlist. 788 */ 789static int 790kevent_copyout(void *arg, struct kevent *kevp, int count) 791{ 792 struct kevent_args *uap; 793 int error; 794 795 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 796 uap = (struct kevent_args *)arg; 797 798 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 799 if (error == 0) 800 uap->eventlist += count; 801 return (error); 802} 803 804/* 805 * Copy 'count' items from the list pointed to by uap->changelist. 806 */ 807static int 808kevent_copyin(void *arg, struct kevent *kevp, int count) 809{ 810 struct kevent_args *uap; 811 int error; 812 813 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 814 uap = (struct kevent_args *)arg; 815 816 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 817 if (error == 0) 818 uap->changelist += count; 819 return (error); 820} 821 822int 823kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 824 struct kevent_copyops *k_ops, const struct timespec *timeout) 825{ 826 struct kevent keva[KQ_NEVENTS]; 827 struct kevent *kevp, *changes; 828 struct kqueue *kq; 829 struct file *fp; 830 int i, n, nerrors, error; 831 832 if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0) 833 return (error); 834 if ((error = kqueue_acquire(fp, &kq)) != 0) 835 goto done_norel; 836 837 nerrors = 0; 838 839 while (nchanges > 0) { 840 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 841 error = k_ops->k_copyin(k_ops->arg, keva, n); 842 if (error) 843 goto done; 844 changes = keva; 845 for (i = 0; i < n; i++) { 846 kevp = &changes[i]; 847 if (!kevp->filter) 848 continue; 849 kevp->flags &= ~EV_SYSFLAGS; 850 error = kqueue_register(kq, kevp, td, 1); 851 if (error || (kevp->flags & EV_RECEIPT)) { 852 if (nevents != 0) { 853 kevp->flags = EV_ERROR; 854 kevp->data = error; 855 (void) k_ops->k_copyout(k_ops->arg, 856 kevp, 1); 857 nevents--; 858 nerrors++; 859 } else { 860 goto done; 861 } 862 } 863 } 864 nchanges -= n; 865 } 866 if (nerrors) { 867 td->td_retval[0] = nerrors; 868 error = 0; 869 goto done; 870 } 871 872 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td); 873done: 874 kqueue_release(kq, 0); 875done_norel: 876 fdrop(fp, td); 877 return (error); 878} 879 880int 881kqueue_add_filteropts(int filt, struct filterops *filtops) 882{ 883 int error; 884 885 error = 0; 886 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 887 printf( 888"trying to add a filterop that is out of range: %d is beyond %d\n", 889 ~filt, EVFILT_SYSCOUNT); 890 return EINVAL; 891 } 892 mtx_lock(&filterops_lock); 893 if (sysfilt_ops[~filt].for_fop != &null_filtops && 894 sysfilt_ops[~filt].for_fop != NULL) 895 error = EEXIST; 896 else { 897 sysfilt_ops[~filt].for_fop = filtops; 898 sysfilt_ops[~filt].for_refcnt = 0; 899 } 900 mtx_unlock(&filterops_lock); 901 902 return (error); 903} 904 905int 906kqueue_del_filteropts(int filt) 907{ 908 int error; 909 910 error = 0; 911 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 912 return EINVAL; 913 914 mtx_lock(&filterops_lock); 915 if (sysfilt_ops[~filt].for_fop == &null_filtops || 916 sysfilt_ops[~filt].for_fop == NULL) 917 error = EINVAL; 918 else if (sysfilt_ops[~filt].for_refcnt != 0) 919 error = EBUSY; 920 else { 921 sysfilt_ops[~filt].for_fop = &null_filtops; 922 sysfilt_ops[~filt].for_refcnt = 0; 923 } 924 mtx_unlock(&filterops_lock); 925 926 return error; 927} 928 929static struct filterops * 930kqueue_fo_find(int filt) 931{ 932 933 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 934 return NULL; 935 936 mtx_lock(&filterops_lock); 937 sysfilt_ops[~filt].for_refcnt++; 938 if (sysfilt_ops[~filt].for_fop == NULL) 939 sysfilt_ops[~filt].for_fop = &null_filtops; 940 mtx_unlock(&filterops_lock); 941 942 return sysfilt_ops[~filt].for_fop; 943} 944 945static void 946kqueue_fo_release(int filt) 947{ 948 949 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 950 return; 951 952 mtx_lock(&filterops_lock); 953 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 954 ("filter object refcount not valid on release")); 955 sysfilt_ops[~filt].for_refcnt--; 956 mtx_unlock(&filterops_lock); 957} 958 959/* 960 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will 961 * influence if memory allocation should wait. Make sure it is 0 if you 962 * hold any mutexes. 963 */ 964static int 965kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok) 966{ 967 struct filterops *fops; 968 struct file *fp; 969 struct knote *kn, *tkn; 970 int error, filt, event; 971 int haskqglobal, filedesc_unlock; 972 973 fp = NULL; 974 kn = NULL; 975 error = 0; 976 haskqglobal = 0; 977 filedesc_unlock = 0; 978 979 filt = kev->filter; 980 fops = kqueue_fo_find(filt); 981 if (fops == NULL) 982 return EINVAL; 983 984 tkn = knote_alloc(waitok); /* prevent waiting with locks */ 985 986findkn: 987 if (fops->f_isfd) { 988 KASSERT(td != NULL, ("td is NULL")); 989 error = fget(td, kev->ident, CAP_POLL_EVENT, &fp); 990 if (error) 991 goto done; 992 993 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 994 kev->ident, 0) != 0) { 995 /* try again */ 996 fdrop(fp, td); 997 fp = NULL; 998 error = kqueue_expand(kq, fops, kev->ident, waitok); 999 if (error) 1000 goto done; 1001 goto findkn; 1002 } 1003 1004 if (fp->f_type == DTYPE_KQUEUE) { 1005 /* 1006 * if we add some inteligence about what we are doing, 1007 * we should be able to support events on ourselves. 1008 * We need to know when we are doing this to prevent 1009 * getting both the knlist lock and the kq lock since 1010 * they are the same thing. 1011 */ 1012 if (fp->f_data == kq) { 1013 error = EINVAL; 1014 goto done; 1015 } 1016 1017 /* 1018 * Pre-lock the filedesc before the global 1019 * lock mutex, see the comment in 1020 * kqueue_close(). 1021 */ 1022 FILEDESC_XLOCK(td->td_proc->p_fd); 1023 filedesc_unlock = 1; 1024 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1025 } 1026 1027 KQ_LOCK(kq); 1028 if (kev->ident < kq->kq_knlistsize) { 1029 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1030 if (kev->filter == kn->kn_filter) 1031 break; 1032 } 1033 } else { 1034 if ((kev->flags & EV_ADD) == EV_ADD) 1035 kqueue_expand(kq, fops, kev->ident, waitok); 1036 1037 KQ_LOCK(kq); 1038 if (kq->kq_knhashmask != 0) { 1039 struct klist *list; 1040 1041 list = &kq->kq_knhash[ 1042 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1043 SLIST_FOREACH(kn, list, kn_link) 1044 if (kev->ident == kn->kn_id && 1045 kev->filter == kn->kn_filter) 1046 break; 1047 } 1048 } 1049 1050 /* knote is in the process of changing, wait for it to stablize. */ 1051 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1052 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1053 if (filedesc_unlock) { 1054 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1055 filedesc_unlock = 0; 1056 } 1057 kq->kq_state |= KQ_FLUXWAIT; 1058 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1059 if (fp != NULL) { 1060 fdrop(fp, td); 1061 fp = NULL; 1062 } 1063 goto findkn; 1064 } 1065 1066 /* 1067 * kn now contains the matching knote, or NULL if no match 1068 */ 1069 if (kn == NULL) { 1070 if (kev->flags & EV_ADD) { 1071 kn = tkn; 1072 tkn = NULL; 1073 if (kn == NULL) { 1074 KQ_UNLOCK(kq); 1075 error = ENOMEM; 1076 goto done; 1077 } 1078 kn->kn_fp = fp; 1079 kn->kn_kq = kq; 1080 kn->kn_fop = fops; 1081 /* 1082 * apply reference counts to knote structure, and 1083 * do not release it at the end of this routine. 1084 */ 1085 fops = NULL; 1086 fp = NULL; 1087 1088 kn->kn_sfflags = kev->fflags; 1089 kn->kn_sdata = kev->data; 1090 kev->fflags = 0; 1091 kev->data = 0; 1092 kn->kn_kevent = *kev; 1093 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1094 EV_ENABLE | EV_DISABLE); 1095 kn->kn_status = KN_INFLUX|KN_DETACHED; 1096 1097 error = knote_attach(kn, kq); 1098 KQ_UNLOCK(kq); 1099 if (error != 0) { 1100 tkn = kn; 1101 goto done; 1102 } 1103 1104 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1105 knote_drop(kn, td); 1106 goto done; 1107 } 1108 KN_LIST_LOCK(kn); 1109 goto done_ev_add; 1110 } else { 1111 /* No matching knote and the EV_ADD flag is not set. */ 1112 KQ_UNLOCK(kq); 1113 error = ENOENT; 1114 goto done; 1115 } 1116 } 1117 1118 if (kev->flags & EV_DELETE) { 1119 kn->kn_status |= KN_INFLUX; 1120 KQ_UNLOCK(kq); 1121 if (!(kn->kn_status & KN_DETACHED)) 1122 kn->kn_fop->f_detach(kn); 1123 knote_drop(kn, td); 1124 goto done; 1125 } 1126 1127 /* 1128 * The user may change some filter values after the initial EV_ADD, 1129 * but doing so will not reset any filter which has already been 1130 * triggered. 1131 */ 1132 kn->kn_status |= KN_INFLUX | KN_SCAN; 1133 KQ_UNLOCK(kq); 1134 KN_LIST_LOCK(kn); 1135 kn->kn_kevent.udata = kev->udata; 1136 if (!fops->f_isfd && fops->f_touch != NULL) { 1137 fops->f_touch(kn, kev, EVENT_REGISTER); 1138 } else { 1139 kn->kn_sfflags = kev->fflags; 1140 kn->kn_sdata = kev->data; 1141 } 1142 1143 /* 1144 * We can get here with kn->kn_knlist == NULL. This can happen when 1145 * the initial attach event decides that the event is "completed" 1146 * already. i.e. filt_procattach is called on a zombie process. It 1147 * will call filt_proc which will remove it from the list, and NULL 1148 * kn_knlist. 1149 */ 1150done_ev_add: 1151 event = kn->kn_fop->f_event(kn, 0); 1152 KQ_LOCK(kq); 1153 if (event) 1154 KNOTE_ACTIVATE(kn, 1); 1155 kn->kn_status &= ~(KN_INFLUX | KN_SCAN); 1156 KN_LIST_UNLOCK(kn); 1157 1158 if ((kev->flags & EV_DISABLE) && 1159 ((kn->kn_status & KN_DISABLED) == 0)) { 1160 kn->kn_status |= KN_DISABLED; 1161 } 1162 1163 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 1164 kn->kn_status &= ~KN_DISABLED; 1165 if ((kn->kn_status & KN_ACTIVE) && 1166 ((kn->kn_status & KN_QUEUED) == 0)) 1167 knote_enqueue(kn); 1168 } 1169 KQ_UNLOCK_FLUX(kq); 1170 1171done: 1172 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1173 if (filedesc_unlock) 1174 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1175 if (fp != NULL) 1176 fdrop(fp, td); 1177 if (tkn != NULL) 1178 knote_free(tkn); 1179 if (fops != NULL) 1180 kqueue_fo_release(filt); 1181 return (error); 1182} 1183 1184static int 1185kqueue_acquire(struct file *fp, struct kqueue **kqp) 1186{ 1187 int error; 1188 struct kqueue *kq; 1189 1190 error = 0; 1191 1192 kq = fp->f_data; 1193 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1194 return (EBADF); 1195 *kqp = kq; 1196 KQ_LOCK(kq); 1197 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1198 KQ_UNLOCK(kq); 1199 return (EBADF); 1200 } 1201 kq->kq_refcnt++; 1202 KQ_UNLOCK(kq); 1203 1204 return error; 1205} 1206 1207static void 1208kqueue_release(struct kqueue *kq, int locked) 1209{ 1210 if (locked) 1211 KQ_OWNED(kq); 1212 else 1213 KQ_LOCK(kq); 1214 kq->kq_refcnt--; 1215 if (kq->kq_refcnt == 1) 1216 wakeup(&kq->kq_refcnt); 1217 if (!locked) 1218 KQ_UNLOCK(kq); 1219} 1220 1221static void 1222kqueue_schedtask(struct kqueue *kq) 1223{ 1224 1225 KQ_OWNED(kq); 1226 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1227 ("scheduling kqueue task while draining")); 1228 1229 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1230 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task); 1231 kq->kq_state |= KQ_TASKSCHED; 1232 } 1233} 1234 1235/* 1236 * Expand the kq to make sure we have storage for fops/ident pair. 1237 * 1238 * Return 0 on success (or no work necessary), return errno on failure. 1239 * 1240 * Not calling hashinit w/ waitok (proper malloc flag) should be safe. 1241 * If kqueue_register is called from a non-fd context, there usually/should 1242 * be no locks held. 1243 */ 1244static int 1245kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, 1246 int waitok) 1247{ 1248 struct klist *list, *tmp_knhash, *to_free; 1249 u_long tmp_knhashmask; 1250 int size; 1251 int fd; 1252 int mflag = waitok ? M_WAITOK : M_NOWAIT; 1253 1254 KQ_NOTOWNED(kq); 1255 1256 to_free = NULL; 1257 if (fops->f_isfd) { 1258 fd = ident; 1259 if (kq->kq_knlistsize <= fd) { 1260 size = kq->kq_knlistsize; 1261 while (size <= fd) 1262 size += KQEXTENT; 1263 list = malloc(size * sizeof(*list), M_KQUEUE, mflag); 1264 if (list == NULL) 1265 return ENOMEM; 1266 KQ_LOCK(kq); 1267 if (kq->kq_knlistsize > fd) { 1268 to_free = list; 1269 list = NULL; 1270 } else { 1271 if (kq->kq_knlist != NULL) { 1272 bcopy(kq->kq_knlist, list, 1273 kq->kq_knlistsize * sizeof(*list)); 1274 to_free = kq->kq_knlist; 1275 kq->kq_knlist = NULL; 1276 } 1277 bzero((caddr_t)list + 1278 kq->kq_knlistsize * sizeof(*list), 1279 (size - kq->kq_knlistsize) * sizeof(*list)); 1280 kq->kq_knlistsize = size; 1281 kq->kq_knlist = list; 1282 } 1283 KQ_UNLOCK(kq); 1284 } 1285 } else { 1286 if (kq->kq_knhashmask == 0) { 1287 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1288 &tmp_knhashmask); 1289 if (tmp_knhash == NULL) 1290 return ENOMEM; 1291 KQ_LOCK(kq); 1292 if (kq->kq_knhashmask == 0) { 1293 kq->kq_knhash = tmp_knhash; 1294 kq->kq_knhashmask = tmp_knhashmask; 1295 } else { 1296 to_free = tmp_knhash; 1297 } 1298 KQ_UNLOCK(kq); 1299 } 1300 } 1301 free(to_free, M_KQUEUE); 1302 1303 KQ_NOTOWNED(kq); 1304 return 0; 1305} 1306 1307static void 1308kqueue_task(void *arg, int pending) 1309{ 1310 struct kqueue *kq; 1311 int haskqglobal; 1312 1313 haskqglobal = 0; 1314 kq = arg; 1315 1316 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1317 KQ_LOCK(kq); 1318 1319 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 1320 1321 kq->kq_state &= ~KQ_TASKSCHED; 1322 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 1323 wakeup(&kq->kq_state); 1324 } 1325 KQ_UNLOCK(kq); 1326 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1327} 1328 1329/* 1330 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 1331 * We treat KN_MARKER knotes as if they are INFLUX. 1332 */ 1333static int 1334kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 1335 const struct timespec *tsp, struct kevent *keva, struct thread *td) 1336{ 1337 struct kevent *kevp; 1338 struct timeval atv, rtv, ttv; 1339 struct knote *kn, *marker; 1340 int count, timeout, nkev, error, influx; 1341 int haskqglobal, touch; 1342 1343 count = maxevents; 1344 nkev = 0; 1345 error = 0; 1346 haskqglobal = 0; 1347 1348 if (maxevents == 0) 1349 goto done_nl; 1350 1351 if (tsp != NULL) { 1352 TIMESPEC_TO_TIMEVAL(&atv, tsp); 1353 if (itimerfix(&atv)) { 1354 error = EINVAL; 1355 goto done_nl; 1356 } 1357 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 1358 timeout = -1; 1359 else 1360 timeout = atv.tv_sec > 24 * 60 * 60 ? 1361 24 * 60 * 60 * hz : tvtohz(&atv); 1362 getmicrouptime(&rtv); 1363 timevaladd(&atv, &rtv); 1364 } else { 1365 atv.tv_sec = 0; 1366 atv.tv_usec = 0; 1367 timeout = 0; 1368 } 1369 marker = knote_alloc(1); 1370 if (marker == NULL) { 1371 error = ENOMEM; 1372 goto done_nl; 1373 } 1374 marker->kn_status = KN_MARKER; 1375 KQ_LOCK(kq); 1376 goto start; 1377 1378retry: 1379 if (atv.tv_sec || atv.tv_usec) { 1380 getmicrouptime(&rtv); 1381 if (timevalcmp(&rtv, &atv, >=)) 1382 goto done; 1383 ttv = atv; 1384 timevalsub(&ttv, &rtv); 1385 timeout = ttv.tv_sec > 24 * 60 * 60 ? 1386 24 * 60 * 60 * hz : tvtohz(&ttv); 1387 } 1388 1389start: 1390 kevp = keva; 1391 if (kq->kq_count == 0) { 1392 if (timeout < 0) { 1393 error = EWOULDBLOCK; 1394 } else { 1395 kq->kq_state |= KQ_SLEEP; 1396 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH, 1397 "kqread", timeout); 1398 } 1399 if (error == 0) 1400 goto retry; 1401 /* don't restart after signals... */ 1402 if (error == ERESTART) 1403 error = EINTR; 1404 else if (error == EWOULDBLOCK) 1405 error = 0; 1406 goto done; 1407 } 1408 1409 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1410 influx = 0; 1411 while (count) { 1412 KQ_OWNED(kq); 1413 kn = TAILQ_FIRST(&kq->kq_head); 1414 1415 if ((kn->kn_status == KN_MARKER && kn != marker) || 1416 (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1417 if (influx) { 1418 influx = 0; 1419 KQ_FLUX_WAKEUP(kq); 1420 } 1421 kq->kq_state |= KQ_FLUXWAIT; 1422 error = msleep(kq, &kq->kq_lock, PSOCK, 1423 "kqflxwt", 0); 1424 continue; 1425 } 1426 1427 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1428 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 1429 kn->kn_status &= ~KN_QUEUED; 1430 kq->kq_count--; 1431 continue; 1432 } 1433 if (kn == marker) { 1434 KQ_FLUX_WAKEUP(kq); 1435 if (count == maxevents) 1436 goto retry; 1437 goto done; 1438 } 1439 KASSERT((kn->kn_status & KN_INFLUX) == 0, 1440 ("KN_INFLUX set when not suppose to be")); 1441 1442 if ((kn->kn_flags & EV_DROP) == EV_DROP) { 1443 kn->kn_status &= ~KN_QUEUED; 1444 kn->kn_status |= KN_INFLUX; 1445 kq->kq_count--; 1446 KQ_UNLOCK(kq); 1447 /* 1448 * We don't need to lock the list since we've marked 1449 * it _INFLUX. 1450 */ 1451 if (!(kn->kn_status & KN_DETACHED)) 1452 kn->kn_fop->f_detach(kn); 1453 knote_drop(kn, td); 1454 KQ_LOCK(kq); 1455 continue; 1456 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 1457 kn->kn_status &= ~KN_QUEUED; 1458 kn->kn_status |= KN_INFLUX; 1459 kq->kq_count--; 1460 KQ_UNLOCK(kq); 1461 /* 1462 * We don't need to lock the list since we've marked 1463 * it _INFLUX. 1464 */ 1465 *kevp = kn->kn_kevent; 1466 if (!(kn->kn_status & KN_DETACHED)) 1467 kn->kn_fop->f_detach(kn); 1468 knote_drop(kn, td); 1469 KQ_LOCK(kq); 1470 kn = NULL; 1471 } else { 1472 kn->kn_status |= KN_INFLUX | KN_SCAN; 1473 KQ_UNLOCK(kq); 1474 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 1475 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1476 KN_LIST_LOCK(kn); 1477 if (kn->kn_fop->f_event(kn, 0) == 0) { 1478 KQ_LOCK(kq); 1479 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1480 kn->kn_status &= 1481 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX | 1482 KN_SCAN); 1483 kq->kq_count--; 1484 KN_LIST_UNLOCK(kn); 1485 influx = 1; 1486 continue; 1487 } 1488 touch = (!kn->kn_fop->f_isfd && 1489 kn->kn_fop->f_touch != NULL); 1490 if (touch) 1491 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 1492 else 1493 *kevp = kn->kn_kevent; 1494 KQ_LOCK(kq); 1495 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1496 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 1497 /* 1498 * Manually clear knotes who weren't 1499 * 'touch'ed. 1500 */ 1501 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 1502 kn->kn_data = 0; 1503 kn->kn_fflags = 0; 1504 } 1505 if (kn->kn_flags & EV_DISPATCH) 1506 kn->kn_status |= KN_DISABLED; 1507 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1508 kq->kq_count--; 1509 } else 1510 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1511 1512 kn->kn_status &= ~(KN_INFLUX | KN_SCAN); 1513 KN_LIST_UNLOCK(kn); 1514 influx = 1; 1515 } 1516 1517 /* we are returning a copy to the user */ 1518 kevp++; 1519 nkev++; 1520 count--; 1521 1522 if (nkev == KQ_NEVENTS) { 1523 influx = 0; 1524 KQ_UNLOCK_FLUX(kq); 1525 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1526 nkev = 0; 1527 kevp = keva; 1528 KQ_LOCK(kq); 1529 if (error) 1530 break; 1531 } 1532 } 1533 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1534done: 1535 KQ_OWNED(kq); 1536 KQ_UNLOCK_FLUX(kq); 1537 knote_free(marker); 1538done_nl: 1539 KQ_NOTOWNED(kq); 1540 if (nkev != 0) 1541 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1542 td->td_retval[0] = maxevents - count; 1543 return (error); 1544} 1545 1546/* 1547 * XXX 1548 * This could be expanded to call kqueue_scan, if desired. 1549 */ 1550/*ARGSUSED*/ 1551static int 1552kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 1553 int flags, struct thread *td) 1554{ 1555 return (ENXIO); 1556} 1557 1558/*ARGSUSED*/ 1559static int 1560kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 1561 int flags, struct thread *td) 1562{ 1563 return (ENXIO); 1564} 1565 1566/*ARGSUSED*/ 1567static int 1568kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1569 struct thread *td) 1570{ 1571 1572 return (EINVAL); 1573} 1574 1575/*ARGSUSED*/ 1576static int 1577kqueue_ioctl(struct file *fp, u_long cmd, void *data, 1578 struct ucred *active_cred, struct thread *td) 1579{ 1580 /* 1581 * Enabling sigio causes two major problems: 1582 * 1) infinite recursion: 1583 * Synopsys: kevent is being used to track signals and have FIOASYNC 1584 * set. On receipt of a signal this will cause a kqueue to recurse 1585 * into itself over and over. Sending the sigio causes the kqueue 1586 * to become ready, which in turn posts sigio again, forever. 1587 * Solution: this can be solved by setting a flag in the kqueue that 1588 * we have a SIGIO in progress. 1589 * 2) locking problems: 1590 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 1591 * us above the proc and pgrp locks. 1592 * Solution: Post a signal using an async mechanism, being sure to 1593 * record a generation count in the delivery so that we do not deliver 1594 * a signal to the wrong process. 1595 * 1596 * Note, these two mechanisms are somewhat mutually exclusive! 1597 */ 1598#if 0 1599 struct kqueue *kq; 1600 1601 kq = fp->f_data; 1602 switch (cmd) { 1603 case FIOASYNC: 1604 if (*(int *)data) { 1605 kq->kq_state |= KQ_ASYNC; 1606 } else { 1607 kq->kq_state &= ~KQ_ASYNC; 1608 } 1609 return (0); 1610 1611 case FIOSETOWN: 1612 return (fsetown(*(int *)data, &kq->kq_sigio)); 1613 1614 case FIOGETOWN: 1615 *(int *)data = fgetown(&kq->kq_sigio); 1616 return (0); 1617 } 1618#endif 1619 1620 return (ENOTTY); 1621} 1622 1623/*ARGSUSED*/ 1624static int 1625kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 1626 struct thread *td) 1627{ 1628 struct kqueue *kq; 1629 int revents = 0; 1630 int error; 1631 1632 if ((error = kqueue_acquire(fp, &kq))) 1633 return POLLERR; 1634 1635 KQ_LOCK(kq); 1636 if (events & (POLLIN | POLLRDNORM)) { 1637 if (kq->kq_count) { 1638 revents |= events & (POLLIN | POLLRDNORM); 1639 } else { 1640 selrecord(td, &kq->kq_sel); 1641 if (SEL_WAITING(&kq->kq_sel)) 1642 kq->kq_state |= KQ_SEL; 1643 } 1644 } 1645 kqueue_release(kq, 1); 1646 KQ_UNLOCK(kq); 1647 return (revents); 1648} 1649 1650/*ARGSUSED*/ 1651static int 1652kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 1653 struct thread *td) 1654{ 1655 1656 bzero((void *)st, sizeof *st); 1657 /* 1658 * We no longer return kq_count because the unlocked value is useless. 1659 * If you spent all this time getting the count, why not spend your 1660 * syscall better by calling kevent? 1661 * 1662 * XXX - This is needed for libc_r. 1663 */ 1664 st->st_mode = S_IFIFO; 1665 return (0); 1666} 1667 1668/*ARGSUSED*/ 1669static int 1670kqueue_close(struct file *fp, struct thread *td) 1671{ 1672 struct kqueue *kq = fp->f_data; 1673 struct filedesc *fdp; 1674 struct knote *kn; 1675 int i; 1676 int error; 1677 int filedesc_unlock; 1678 1679 if ((error = kqueue_acquire(fp, &kq))) 1680 return error; 1681 1682 filedesc_unlock = 0; 1683 KQ_LOCK(kq); 1684 1685 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 1686 ("kqueue already closing")); 1687 kq->kq_state |= KQ_CLOSING; 1688 if (kq->kq_refcnt > 1) 1689 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 1690 1691 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 1692 fdp = kq->kq_fdp; 1693 1694 KASSERT(knlist_empty(&kq->kq_sel.si_note), 1695 ("kqueue's knlist not empty")); 1696 1697 for (i = 0; i < kq->kq_knlistsize; i++) { 1698 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 1699 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1700 kq->kq_state |= KQ_FLUXWAIT; 1701 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 1702 continue; 1703 } 1704 kn->kn_status |= KN_INFLUX; 1705 KQ_UNLOCK(kq); 1706 if (!(kn->kn_status & KN_DETACHED)) 1707 kn->kn_fop->f_detach(kn); 1708 knote_drop(kn, td); 1709 KQ_LOCK(kq); 1710 } 1711 } 1712 if (kq->kq_knhashmask != 0) { 1713 for (i = 0; i <= kq->kq_knhashmask; i++) { 1714 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 1715 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1716 kq->kq_state |= KQ_FLUXWAIT; 1717 msleep(kq, &kq->kq_lock, PSOCK, 1718 "kqclo2", 0); 1719 continue; 1720 } 1721 kn->kn_status |= KN_INFLUX; 1722 KQ_UNLOCK(kq); 1723 if (!(kn->kn_status & KN_DETACHED)) 1724 kn->kn_fop->f_detach(kn); 1725 knote_drop(kn, td); 1726 KQ_LOCK(kq); 1727 } 1728 } 1729 } 1730 1731 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 1732 kq->kq_state |= KQ_TASKDRAIN; 1733 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 1734 } 1735 1736 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1737 selwakeuppri(&kq->kq_sel, PSOCK); 1738 if (!SEL_WAITING(&kq->kq_sel)) 1739 kq->kq_state &= ~KQ_SEL; 1740 } 1741 1742 KQ_UNLOCK(kq); 1743 1744 /* 1745 * We could be called due to the knote_drop() doing fdrop(), 1746 * called from kqueue_register(). In this case the global 1747 * lock is owned, and filedesc sx is locked before, to not 1748 * take the sleepable lock after non-sleepable. 1749 */ 1750 if (!sx_xlocked(FILEDESC_LOCK(fdp))) { 1751 FILEDESC_XLOCK(fdp); 1752 filedesc_unlock = 1; 1753 } else 1754 filedesc_unlock = 0; 1755 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); 1756 if (filedesc_unlock) 1757 FILEDESC_XUNLOCK(fdp); 1758 1759 seldrain(&kq->kq_sel); 1760 knlist_destroy(&kq->kq_sel.si_note); 1761 mtx_destroy(&kq->kq_lock); 1762 kq->kq_fdp = NULL; 1763 1764 if (kq->kq_knhash != NULL) 1765 free(kq->kq_knhash, M_KQUEUE); 1766 if (kq->kq_knlist != NULL) 1767 free(kq->kq_knlist, M_KQUEUE); 1768 1769 funsetown(&kq->kq_sigio); 1770 free(kq, M_KQUEUE); 1771 fp->f_data = NULL; 1772 1773 return (0); 1774} 1775 1776static void 1777kqueue_wakeup(struct kqueue *kq) 1778{ 1779 KQ_OWNED(kq); 1780 1781 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 1782 kq->kq_state &= ~KQ_SLEEP; 1783 wakeup(kq); 1784 } 1785 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1786 selwakeuppri(&kq->kq_sel, PSOCK); 1787 if (!SEL_WAITING(&kq->kq_sel)) 1788 kq->kq_state &= ~KQ_SEL; 1789 } 1790 if (!knlist_empty(&kq->kq_sel.si_note)) 1791 kqueue_schedtask(kq); 1792 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 1793 pgsigio(&kq->kq_sigio, SIGIO, 0); 1794 } 1795} 1796 1797/* 1798 * Walk down a list of knotes, activating them if their event has triggered. 1799 * 1800 * There is a possibility to optimize in the case of one kq watching another. 1801 * Instead of scheduling a task to wake it up, you could pass enough state 1802 * down the chain to make up the parent kqueue. Make this code functional 1803 * first. 1804 */ 1805void 1806knote(struct knlist *list, long hint, int lockflags) 1807{ 1808 struct kqueue *kq; 1809 struct knote *kn; 1810 int error; 1811 1812 if (list == NULL) 1813 return; 1814 1815 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 1816 1817 if ((lockflags & KNF_LISTLOCKED) == 0) 1818 list->kl_lock(list->kl_lockarg); 1819 1820 /* 1821 * If we unlock the list lock (and set KN_INFLUX), we can eliminate 1822 * the kqueue scheduling, but this will introduce four 1823 * lock/unlock's for each knote to test. If we do, continue to use 1824 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is 1825 * only safe if you want to remove the current item, which we are 1826 * not doing. 1827 */ 1828 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 1829 kq = kn->kn_kq; 1830 KQ_LOCK(kq); 1831 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { 1832 /* 1833 * Do not process the influx notes, except for 1834 * the influx coming from the kq unlock in the 1835 * kqueue_scan(). In the later case, we do 1836 * not interfere with the scan, since the code 1837 * fragment in kqueue_scan() locks the knlist, 1838 * and cannot proceed until we finished. 1839 */ 1840 KQ_UNLOCK(kq); 1841 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 1842 kn->kn_status |= KN_INFLUX; 1843 KQ_UNLOCK(kq); 1844 error = kn->kn_fop->f_event(kn, hint); 1845 KQ_LOCK(kq); 1846 kn->kn_status &= ~KN_INFLUX; 1847 if (error) 1848 KNOTE_ACTIVATE(kn, 1); 1849 KQ_UNLOCK_FLUX(kq); 1850 } else { 1851 kn->kn_status |= KN_HASKQLOCK; 1852 if (kn->kn_fop->f_event(kn, hint)) 1853 KNOTE_ACTIVATE(kn, 1); 1854 kn->kn_status &= ~KN_HASKQLOCK; 1855 KQ_UNLOCK(kq); 1856 } 1857 } 1858 if ((lockflags & KNF_LISTLOCKED) == 0) 1859 list->kl_unlock(list->kl_lockarg); 1860} 1861 1862/* 1863 * add a knote to a knlist 1864 */ 1865void 1866knlist_add(struct knlist *knl, struct knote *kn, int islocked) 1867{ 1868 KNL_ASSERT_LOCK(knl, islocked); 1869 KQ_NOTOWNED(kn->kn_kq); 1870 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == 1871 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED")); 1872 if (!islocked) 1873 knl->kl_lock(knl->kl_lockarg); 1874 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 1875 if (!islocked) 1876 knl->kl_unlock(knl->kl_lockarg); 1877 KQ_LOCK(kn->kn_kq); 1878 kn->kn_knlist = knl; 1879 kn->kn_status &= ~KN_DETACHED; 1880 KQ_UNLOCK(kn->kn_kq); 1881} 1882 1883static void 1884knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked) 1885{ 1886 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked")); 1887 KNL_ASSERT_LOCK(knl, knlislocked); 1888 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 1889 if (!kqislocked) 1890 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX, 1891 ("knlist_remove called w/o knote being KN_INFLUX or already removed")); 1892 if (!knlislocked) 1893 knl->kl_lock(knl->kl_lockarg); 1894 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 1895 kn->kn_knlist = NULL; 1896 if (!knlislocked) 1897 knl->kl_unlock(knl->kl_lockarg); 1898 if (!kqislocked) 1899 KQ_LOCK(kn->kn_kq); 1900 kn->kn_status |= KN_DETACHED; 1901 if (!kqislocked) 1902 KQ_UNLOCK(kn->kn_kq); 1903} 1904 1905/* 1906 * remove all knotes from a specified klist 1907 */ 1908void 1909knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 1910{ 1911 1912 knlist_remove_kq(knl, kn, islocked, 0); 1913} 1914 1915/* 1916 * remove knote from a specified klist while in f_event handler. 1917 */ 1918void 1919knlist_remove_inevent(struct knlist *knl, struct knote *kn) 1920{ 1921 1922 knlist_remove_kq(knl, kn, 1, 1923 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK); 1924} 1925 1926int 1927knlist_empty(struct knlist *knl) 1928{ 1929 KNL_ASSERT_LOCKED(knl); 1930 return SLIST_EMPTY(&knl->kl_list); 1931} 1932 1933static struct mtx knlist_lock; 1934MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 1935 MTX_DEF); 1936static void knlist_mtx_lock(void *arg); 1937static void knlist_mtx_unlock(void *arg); 1938 1939static void 1940knlist_mtx_lock(void *arg) 1941{ 1942 mtx_lock((struct mtx *)arg); 1943} 1944 1945static void 1946knlist_mtx_unlock(void *arg) 1947{ 1948 mtx_unlock((struct mtx *)arg); 1949} 1950 1951static void 1952knlist_mtx_assert_locked(void *arg) 1953{ 1954 mtx_assert((struct mtx *)arg, MA_OWNED); 1955} 1956 1957static void 1958knlist_mtx_assert_unlocked(void *arg) 1959{ 1960 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 1961} 1962 1963void 1964knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 1965 void (*kl_unlock)(void *), 1966 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *)) 1967{ 1968 1969 if (lock == NULL) 1970 knl->kl_lockarg = &knlist_lock; 1971 else 1972 knl->kl_lockarg = lock; 1973 1974 if (kl_lock == NULL) 1975 knl->kl_lock = knlist_mtx_lock; 1976 else 1977 knl->kl_lock = kl_lock; 1978 if (kl_unlock == NULL) 1979 knl->kl_unlock = knlist_mtx_unlock; 1980 else 1981 knl->kl_unlock = kl_unlock; 1982 if (kl_assert_locked == NULL) 1983 knl->kl_assert_locked = knlist_mtx_assert_locked; 1984 else 1985 knl->kl_assert_locked = kl_assert_locked; 1986 if (kl_assert_unlocked == NULL) 1987 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked; 1988 else 1989 knl->kl_assert_unlocked = kl_assert_unlocked; 1990 1991 SLIST_INIT(&knl->kl_list); 1992} 1993 1994void 1995knlist_init_mtx(struct knlist *knl, struct mtx *lock) 1996{ 1997 1998 knlist_init(knl, lock, NULL, NULL, NULL, NULL); 1999} 2000 2001void 2002knlist_destroy(struct knlist *knl) 2003{ 2004 2005#ifdef INVARIANTS 2006 /* 2007 * if we run across this error, we need to find the offending 2008 * driver and have it call knlist_clear. 2009 */ 2010 if (!SLIST_EMPTY(&knl->kl_list)) 2011 printf("WARNING: destroying knlist w/ knotes on it!\n"); 2012#endif 2013 2014 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL; 2015 SLIST_INIT(&knl->kl_list); 2016} 2017 2018/* 2019 * Even if we are locked, we may need to drop the lock to allow any influx 2020 * knotes time to "settle". 2021 */ 2022void 2023knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 2024{ 2025 struct knote *kn, *kn2; 2026 struct kqueue *kq; 2027 2028 if (islocked) 2029 KNL_ASSERT_LOCKED(knl); 2030 else { 2031 KNL_ASSERT_UNLOCKED(knl); 2032again: /* need to reacquire lock since we have dropped it */ 2033 knl->kl_lock(knl->kl_lockarg); 2034 } 2035 2036 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 2037 kq = kn->kn_kq; 2038 KQ_LOCK(kq); 2039 if ((kn->kn_status & KN_INFLUX)) { 2040 KQ_UNLOCK(kq); 2041 continue; 2042 } 2043 knlist_remove_kq(knl, kn, 1, 1); 2044 if (killkn) { 2045 kn->kn_status |= KN_INFLUX | KN_DETACHED; 2046 KQ_UNLOCK(kq); 2047 knote_drop(kn, td); 2048 } else { 2049 /* Make sure cleared knotes disappear soon */ 2050 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 2051 KQ_UNLOCK(kq); 2052 } 2053 kq = NULL; 2054 } 2055 2056 if (!SLIST_EMPTY(&knl->kl_list)) { 2057 /* there are still KN_INFLUX remaining */ 2058 kn = SLIST_FIRST(&knl->kl_list); 2059 kq = kn->kn_kq; 2060 KQ_LOCK(kq); 2061 KASSERT(kn->kn_status & KN_INFLUX, 2062 ("knote removed w/o list lock")); 2063 knl->kl_unlock(knl->kl_lockarg); 2064 kq->kq_state |= KQ_FLUXWAIT; 2065 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 2066 kq = NULL; 2067 goto again; 2068 } 2069 2070 if (islocked) 2071 KNL_ASSERT_LOCKED(knl); 2072 else { 2073 knl->kl_unlock(knl->kl_lockarg); 2074 KNL_ASSERT_UNLOCKED(knl); 2075 } 2076} 2077 2078/* 2079 * Remove all knotes referencing a specified fd must be called with FILEDESC 2080 * lock. This prevents a race where a new fd comes along and occupies the 2081 * entry and we attach a knote to the fd. 2082 */ 2083void 2084knote_fdclose(struct thread *td, int fd) 2085{ 2086 struct filedesc *fdp = td->td_proc->p_fd; 2087 struct kqueue *kq; 2088 struct knote *kn; 2089 int influx; 2090 2091 FILEDESC_XLOCK_ASSERT(fdp); 2092 2093 /* 2094 * We shouldn't have to worry about new kevents appearing on fd 2095 * since filedesc is locked. 2096 */ 2097 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2098 KQ_LOCK(kq); 2099 2100again: 2101 influx = 0; 2102 while (kq->kq_knlistsize > fd && 2103 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2104 if (kn->kn_status & KN_INFLUX) { 2105 /* someone else might be waiting on our knote */ 2106 if (influx) 2107 wakeup(kq); 2108 kq->kq_state |= KQ_FLUXWAIT; 2109 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2110 goto again; 2111 } 2112 kn->kn_status |= KN_INFLUX; 2113 KQ_UNLOCK(kq); 2114 if (!(kn->kn_status & KN_DETACHED)) 2115 kn->kn_fop->f_detach(kn); 2116 knote_drop(kn, td); 2117 influx = 1; 2118 KQ_LOCK(kq); 2119 } 2120 KQ_UNLOCK_FLUX(kq); 2121 } 2122} 2123 2124static int 2125knote_attach(struct knote *kn, struct kqueue *kq) 2126{ 2127 struct klist *list; 2128 2129 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX")); 2130 KQ_OWNED(kq); 2131 2132 if (kn->kn_fop->f_isfd) { 2133 if (kn->kn_id >= kq->kq_knlistsize) 2134 return ENOMEM; 2135 list = &kq->kq_knlist[kn->kn_id]; 2136 } else { 2137 if (kq->kq_knhash == NULL) 2138 return ENOMEM; 2139 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2140 } 2141 2142 SLIST_INSERT_HEAD(list, kn, kn_link); 2143 2144 return 0; 2145} 2146 2147/* 2148 * knote must already have been detached using the f_detach method. 2149 * no lock need to be held, it is assumed that the KN_INFLUX flag is set 2150 * to prevent other removal. 2151 */ 2152static void 2153knote_drop(struct knote *kn, struct thread *td) 2154{ 2155 struct kqueue *kq; 2156 struct klist *list; 2157 2158 kq = kn->kn_kq; 2159 2160 KQ_NOTOWNED(kq); 2161 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX, 2162 ("knote_drop called without KN_INFLUX set in kn_status")); 2163 2164 KQ_LOCK(kq); 2165 if (kn->kn_fop->f_isfd) 2166 list = &kq->kq_knlist[kn->kn_id]; 2167 else 2168 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2169 2170 if (!SLIST_EMPTY(list)) 2171 SLIST_REMOVE(list, kn, knote, kn_link); 2172 if (kn->kn_status & KN_QUEUED) 2173 knote_dequeue(kn); 2174 KQ_UNLOCK_FLUX(kq); 2175 2176 if (kn->kn_fop->f_isfd) { 2177 fdrop(kn->kn_fp, td); 2178 kn->kn_fp = NULL; 2179 } 2180 kqueue_fo_release(kn->kn_kevent.filter); 2181 kn->kn_fop = NULL; 2182 knote_free(kn); 2183} 2184 2185static void 2186knote_enqueue(struct knote *kn) 2187{ 2188 struct kqueue *kq = kn->kn_kq; 2189 2190 KQ_OWNED(kn->kn_kq); 2191 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2192 2193 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2194 kn->kn_status |= KN_QUEUED; 2195 kq->kq_count++; 2196 kqueue_wakeup(kq); 2197} 2198 2199static void 2200knote_dequeue(struct knote *kn) 2201{ 2202 struct kqueue *kq = kn->kn_kq; 2203 2204 KQ_OWNED(kn->kn_kq); 2205 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2206 2207 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2208 kn->kn_status &= ~KN_QUEUED; 2209 kq->kq_count--; 2210} 2211 2212static void 2213knote_init(void) 2214{ 2215 2216 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2217 NULL, NULL, UMA_ALIGN_PTR, 0); 2218} 2219SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2220 2221static struct knote * 2222knote_alloc(int waitok) 2223{ 2224 return ((struct knote *)uma_zalloc(knote_zone, 2225 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO)); 2226} 2227 2228static void 2229knote_free(struct knote *kn) 2230{ 2231 if (kn != NULL) 2232 uma_zfree(knote_zone, kn); 2233} 2234 2235/* 2236 * Register the kev w/ the kq specified by fd. 2237 */ 2238int 2239kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok) 2240{ 2241 struct kqueue *kq; 2242 struct file *fp; 2243 int error; 2244 2245 if ((error = fget(td, fd, CAP_POST_EVENT, &fp)) != 0) 2246 return (error); 2247 if ((error = kqueue_acquire(fp, &kq)) != 0) 2248 goto noacquire; 2249 2250 error = kqueue_register(kq, kev, td, waitok); 2251 2252 kqueue_release(kq, 0); 2253 2254noacquire: 2255 fdrop(fp, td); 2256 2257 return error; 2258} 2259