kern_event.c revision 197242
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: head/sys/kern/kern_event.c 197242 2009-09-16 03:37:39Z sson $"); 31 32#include "opt_ktrace.h" 33 34#include <sys/param.h> 35#include <sys/systm.h> 36#include <sys/kernel.h> 37#include <sys/lock.h> 38#include <sys/mutex.h> 39#include <sys/proc.h> 40#include <sys/malloc.h> 41#include <sys/unistd.h> 42#include <sys/file.h> 43#include <sys/filedesc.h> 44#include <sys/filio.h> 45#include <sys/fcntl.h> 46#include <sys/kthread.h> 47#include <sys/selinfo.h> 48#include <sys/queue.h> 49#include <sys/event.h> 50#include <sys/eventvar.h> 51#include <sys/poll.h> 52#include <sys/protosw.h> 53#include <sys/sigio.h> 54#include <sys/signalvar.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57#include <sys/stat.h> 58#include <sys/sysctl.h> 59#include <sys/sysproto.h> 60#include <sys/syscallsubr.h> 61#include <sys/taskqueue.h> 62#include <sys/uio.h> 63#ifdef KTRACE 64#include <sys/ktrace.h> 65#endif 66 67#include <vm/uma.h> 68 69static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 70 71/* 72 * This lock is used if multiple kq locks are required. This possibly 73 * should be made into a per proc lock. 74 */ 75static struct mtx kq_global; 76MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 77#define KQ_GLOBAL_LOCK(lck, haslck) do { \ 78 if (!haslck) \ 79 mtx_lock(lck); \ 80 haslck = 1; \ 81} while (0) 82#define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 83 if (haslck) \ 84 mtx_unlock(lck); \ 85 haslck = 0; \ 86} while (0) 87 88TASKQUEUE_DEFINE_THREAD(kqueue); 89 90static int kevent_copyout(void *arg, struct kevent *kevp, int count); 91static int kevent_copyin(void *arg, struct kevent *kevp, int count); 92static int kqueue_register(struct kqueue *kq, struct kevent *kev, 93 struct thread *td, int waitok); 94static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 95static void kqueue_release(struct kqueue *kq, int locked); 96static int kqueue_expand(struct kqueue *kq, struct filterops *fops, 97 uintptr_t ident, int waitok); 98static void kqueue_task(void *arg, int pending); 99static int kqueue_scan(struct kqueue *kq, int maxevents, 100 struct kevent_copyops *k_ops, 101 const struct timespec *timeout, 102 struct kevent *keva, struct thread *td); 103static void kqueue_wakeup(struct kqueue *kq); 104static struct filterops *kqueue_fo_find(int filt); 105static void kqueue_fo_release(int filt); 106 107static fo_rdwr_t kqueue_read; 108static fo_rdwr_t kqueue_write; 109static fo_truncate_t kqueue_truncate; 110static fo_ioctl_t kqueue_ioctl; 111static fo_poll_t kqueue_poll; 112static fo_kqfilter_t kqueue_kqfilter; 113static fo_stat_t kqueue_stat; 114static fo_close_t kqueue_close; 115 116static struct fileops kqueueops = { 117 .fo_read = kqueue_read, 118 .fo_write = kqueue_write, 119 .fo_truncate = kqueue_truncate, 120 .fo_ioctl = kqueue_ioctl, 121 .fo_poll = kqueue_poll, 122 .fo_kqfilter = kqueue_kqfilter, 123 .fo_stat = kqueue_stat, 124 .fo_close = kqueue_close, 125}; 126 127static int knote_attach(struct knote *kn, struct kqueue *kq); 128static void knote_drop(struct knote *kn, struct thread *td); 129static void knote_enqueue(struct knote *kn); 130static void knote_dequeue(struct knote *kn); 131static void knote_init(void); 132static struct knote *knote_alloc(int waitok); 133static void knote_free(struct knote *kn); 134 135static void filt_kqdetach(struct knote *kn); 136static int filt_kqueue(struct knote *kn, long hint); 137static int filt_procattach(struct knote *kn); 138static void filt_procdetach(struct knote *kn); 139static int filt_proc(struct knote *kn, long hint); 140static int filt_fileattach(struct knote *kn); 141static void filt_timerexpire(void *knx); 142static int filt_timerattach(struct knote *kn); 143static void filt_timerdetach(struct knote *kn); 144static int filt_timer(struct knote *kn, long hint); 145static int filt_userattach(struct knote *kn); 146static void filt_userdetach(struct knote *kn); 147static int filt_user(struct knote *kn, long hint); 148static void filt_usertouch(struct knote *kn, struct kevent *kev, long type); 149 150static struct filterops file_filtops = { 151 .f_isfd = 1, 152 .f_attach = filt_fileattach, 153}; 154static struct filterops kqread_filtops = { 155 .f_isfd = 1, 156 .f_detach = filt_kqdetach, 157 .f_event = filt_kqueue, 158}; 159/* XXX - move to kern_proc.c? */ 160static struct filterops proc_filtops = { 161 .f_isfd = 0, 162 .f_attach = filt_procattach, 163 .f_detach = filt_procdetach, 164 .f_event = filt_proc, 165}; 166static struct filterops timer_filtops = { 167 .f_isfd = 0, 168 .f_attach = filt_timerattach, 169 .f_detach = filt_timerdetach, 170 .f_event = filt_timer, 171}; 172static struct filterops user_filtops = { 173 .f_attach = filt_userattach, 174 .f_detach = filt_userdetach, 175 .f_event = filt_user, 176 .f_touch = filt_usertouch, 177}; 178 179static uma_zone_t knote_zone; 180static int kq_ncallouts = 0; 181static int kq_calloutmax = (4 * 1024); 182SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 183 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 184 185/* XXX - ensure not KN_INFLUX?? */ 186#define KNOTE_ACTIVATE(kn, islock) do { \ 187 if ((islock)) \ 188 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 189 else \ 190 KQ_LOCK((kn)->kn_kq); \ 191 (kn)->kn_status |= KN_ACTIVE; \ 192 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 193 knote_enqueue((kn)); \ 194 if (!(islock)) \ 195 KQ_UNLOCK((kn)->kn_kq); \ 196} while(0) 197#define KQ_LOCK(kq) do { \ 198 mtx_lock(&(kq)->kq_lock); \ 199} while (0) 200#define KQ_FLUX_WAKEUP(kq) do { \ 201 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 202 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 203 wakeup((kq)); \ 204 } \ 205} while (0) 206#define KQ_UNLOCK_FLUX(kq) do { \ 207 KQ_FLUX_WAKEUP(kq); \ 208 mtx_unlock(&(kq)->kq_lock); \ 209} while (0) 210#define KQ_UNLOCK(kq) do { \ 211 mtx_unlock(&(kq)->kq_lock); \ 212} while (0) 213#define KQ_OWNED(kq) do { \ 214 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 215} while (0) 216#define KQ_NOTOWNED(kq) do { \ 217 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 218} while (0) 219#define KN_LIST_LOCK(kn) do { \ 220 if (kn->kn_knlist != NULL) \ 221 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \ 222} while (0) 223#define KN_LIST_UNLOCK(kn) do { \ 224 if (kn->kn_knlist != NULL) \ 225 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \ 226} while (0) 227#define KNL_ASSERT_LOCK(knl, islocked) do { \ 228 if (islocked) \ 229 KNL_ASSERT_LOCKED(knl); \ 230 else \ 231 KNL_ASSERT_UNLOCKED(knl); \ 232} while (0) 233#ifdef INVARIANTS 234#define KNL_ASSERT_LOCKED(knl) do { \ 235 knl->kl_assert_locked((knl)->kl_lockarg); \ 236} while (0) 237#define KNL_ASSERT_UNLOCKED(knl) do { \ 238 knl->kl_assert_unlocked((knl)->kl_lockarg); \ 239} while (0) 240#else /* !INVARIANTS */ 241#define KNL_ASSERT_LOCKED(knl) do {} while(0) 242#define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 243#endif /* INVARIANTS */ 244 245#define KN_HASHSIZE 64 /* XXX should be tunable */ 246#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 247 248static int 249filt_nullattach(struct knote *kn) 250{ 251 252 return (ENXIO); 253}; 254 255struct filterops null_filtops = { 256 .f_isfd = 0, 257 .f_attach = filt_nullattach, 258}; 259 260/* XXX - make SYSINIT to add these, and move into respective modules. */ 261extern struct filterops sig_filtops; 262extern struct filterops fs_filtops; 263 264/* 265 * Table for for all system-defined filters. 266 */ 267static struct mtx filterops_lock; 268MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", 269 MTX_DEF); 270static struct { 271 struct filterops *for_fop; 272 int for_refcnt; 273} sysfilt_ops[EVFILT_SYSCOUNT] = { 274 { &file_filtops }, /* EVFILT_READ */ 275 { &file_filtops }, /* EVFILT_WRITE */ 276 { &null_filtops }, /* EVFILT_AIO */ 277 { &file_filtops }, /* EVFILT_VNODE */ 278 { &proc_filtops }, /* EVFILT_PROC */ 279 { &sig_filtops }, /* EVFILT_SIGNAL */ 280 { &timer_filtops }, /* EVFILT_TIMER */ 281 { &file_filtops }, /* EVFILT_NETDEV */ 282 { &fs_filtops }, /* EVFILT_FS */ 283 { &null_filtops }, /* EVFILT_LIO */ 284 { &user_filtops }, /* EVFILT_USER */ 285}; 286 287/* 288 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 289 * method. 290 */ 291static int 292filt_fileattach(struct knote *kn) 293{ 294 295 return (fo_kqfilter(kn->kn_fp, kn)); 296} 297 298/*ARGSUSED*/ 299static int 300kqueue_kqfilter(struct file *fp, struct knote *kn) 301{ 302 struct kqueue *kq = kn->kn_fp->f_data; 303 304 if (kn->kn_filter != EVFILT_READ) 305 return (EINVAL); 306 307 kn->kn_status |= KN_KQUEUE; 308 kn->kn_fop = &kqread_filtops; 309 knlist_add(&kq->kq_sel.si_note, kn, 0); 310 311 return (0); 312} 313 314static void 315filt_kqdetach(struct knote *kn) 316{ 317 struct kqueue *kq = kn->kn_fp->f_data; 318 319 knlist_remove(&kq->kq_sel.si_note, kn, 0); 320} 321 322/*ARGSUSED*/ 323static int 324filt_kqueue(struct knote *kn, long hint) 325{ 326 struct kqueue *kq = kn->kn_fp->f_data; 327 328 kn->kn_data = kq->kq_count; 329 return (kn->kn_data > 0); 330} 331 332/* XXX - move to kern_proc.c? */ 333static int 334filt_procattach(struct knote *kn) 335{ 336 struct proc *p; 337 int immediate; 338 int error; 339 340 immediate = 0; 341 p = pfind(kn->kn_id); 342 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 343 p = zpfind(kn->kn_id); 344 immediate = 1; 345 } else if (p != NULL && (p->p_flag & P_WEXIT)) { 346 immediate = 1; 347 } 348 349 if (p == NULL) 350 return (ESRCH); 351 if ((error = p_cansee(curthread, p))) 352 return (error); 353 354 kn->kn_ptr.p_proc = p; 355 kn->kn_flags |= EV_CLEAR; /* automatically set */ 356 357 /* 358 * internal flag indicating registration done by kernel 359 */ 360 if (kn->kn_flags & EV_FLAG1) { 361 kn->kn_data = kn->kn_sdata; /* ppid */ 362 kn->kn_fflags = NOTE_CHILD; 363 kn->kn_flags &= ~EV_FLAG1; 364 } 365 366 if (immediate == 0) 367 knlist_add(&p->p_klist, kn, 1); 368 369 /* 370 * Immediately activate any exit notes if the target process is a 371 * zombie. This is necessary to handle the case where the target 372 * process, e.g. a child, dies before the kevent is registered. 373 */ 374 if (immediate && filt_proc(kn, NOTE_EXIT)) 375 KNOTE_ACTIVATE(kn, 0); 376 377 PROC_UNLOCK(p); 378 379 return (0); 380} 381 382/* 383 * The knote may be attached to a different process, which may exit, 384 * leaving nothing for the knote to be attached to. So when the process 385 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 386 * it will be deleted when read out. However, as part of the knote deletion, 387 * this routine is called, so a check is needed to avoid actually performing 388 * a detach, because the original process does not exist any more. 389 */ 390/* XXX - move to kern_proc.c? */ 391static void 392filt_procdetach(struct knote *kn) 393{ 394 struct proc *p; 395 396 p = kn->kn_ptr.p_proc; 397 knlist_remove(&p->p_klist, kn, 0); 398 kn->kn_ptr.p_proc = NULL; 399} 400 401/* XXX - move to kern_proc.c? */ 402static int 403filt_proc(struct knote *kn, long hint) 404{ 405 struct proc *p = kn->kn_ptr.p_proc; 406 u_int event; 407 408 /* 409 * mask off extra data 410 */ 411 event = (u_int)hint & NOTE_PCTRLMASK; 412 413 /* 414 * if the user is interested in this event, record it. 415 */ 416 if (kn->kn_sfflags & event) 417 kn->kn_fflags |= event; 418 419 /* 420 * process is gone, so flag the event as finished. 421 */ 422 if (event == NOTE_EXIT) { 423 if (!(kn->kn_status & KN_DETACHED)) 424 knlist_remove_inevent(&p->p_klist, kn); 425 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 426 kn->kn_data = p->p_xstat; 427 kn->kn_ptr.p_proc = NULL; 428 return (1); 429 } 430 431 return (kn->kn_fflags != 0); 432} 433 434/* 435 * Called when the process forked. It mostly does the same as the 436 * knote(), activating all knotes registered to be activated when the 437 * process forked. Additionally, for each knote attached to the 438 * parent, check whether user wants to track the new process. If so 439 * attach a new knote to it, and immediately report an event with the 440 * child's pid. 441 */ 442void 443knote_fork(struct knlist *list, int pid) 444{ 445 struct kqueue *kq; 446 struct knote *kn; 447 struct kevent kev; 448 int error; 449 450 if (list == NULL) 451 return; 452 list->kl_lock(list->kl_lockarg); 453 454 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 455 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) 456 continue; 457 kq = kn->kn_kq; 458 KQ_LOCK(kq); 459 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 460 KQ_UNLOCK(kq); 461 continue; 462 } 463 464 /* 465 * The same as knote(), activate the event. 466 */ 467 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 468 kn->kn_status |= KN_HASKQLOCK; 469 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid)) 470 KNOTE_ACTIVATE(kn, 1); 471 kn->kn_status &= ~KN_HASKQLOCK; 472 KQ_UNLOCK(kq); 473 continue; 474 } 475 476 /* 477 * The NOTE_TRACK case. In addition to the activation 478 * of the event, we need to register new event to 479 * track the child. Drop the locks in preparation for 480 * the call to kqueue_register(). 481 */ 482 kn->kn_status |= KN_INFLUX; 483 KQ_UNLOCK(kq); 484 list->kl_unlock(list->kl_lockarg); 485 486 /* 487 * Activate existing knote and register a knote with 488 * new process. 489 */ 490 kev.ident = pid; 491 kev.filter = kn->kn_filter; 492 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 493 kev.fflags = kn->kn_sfflags; 494 kev.data = kn->kn_id; /* parent */ 495 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 496 error = kqueue_register(kq, &kev, NULL, 0); 497 if (kn->kn_fop->f_event(kn, NOTE_FORK | pid)) 498 KNOTE_ACTIVATE(kn, 0); 499 if (error) 500 kn->kn_fflags |= NOTE_TRACKERR; 501 KQ_LOCK(kq); 502 kn->kn_status &= ~KN_INFLUX; 503 KQ_UNLOCK_FLUX(kq); 504 list->kl_lock(list->kl_lockarg); 505 } 506 list->kl_unlock(list->kl_lockarg); 507} 508 509static int 510timertoticks(intptr_t data) 511{ 512 struct timeval tv; 513 int tticks; 514 515 tv.tv_sec = data / 1000; 516 tv.tv_usec = (data % 1000) * 1000; 517 tticks = tvtohz(&tv); 518 519 return tticks; 520} 521 522/* XXX - move to kern_timeout.c? */ 523static void 524filt_timerexpire(void *knx) 525{ 526 struct knote *kn = knx; 527 struct callout *calloutp; 528 529 kn->kn_data++; 530 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 531 532 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) { 533 calloutp = (struct callout *)kn->kn_hook; 534 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata), 535 filt_timerexpire, kn); 536 } 537} 538 539/* 540 * data contains amount of time to sleep, in milliseconds 541 */ 542/* XXX - move to kern_timeout.c? */ 543static int 544filt_timerattach(struct knote *kn) 545{ 546 struct callout *calloutp; 547 548 atomic_add_int(&kq_ncallouts, 1); 549 550 if (kq_ncallouts >= kq_calloutmax) { 551 atomic_add_int(&kq_ncallouts, -1); 552 return (ENOMEM); 553 } 554 555 kn->kn_flags |= EV_CLEAR; /* automatically set */ 556 kn->kn_status &= ~KN_DETACHED; /* knlist_add usually sets it */ 557 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK); 558 callout_init(calloutp, CALLOUT_MPSAFE); 559 kn->kn_hook = calloutp; 560 callout_reset_curcpu(calloutp, timertoticks(kn->kn_sdata), 561 filt_timerexpire, kn); 562 563 return (0); 564} 565 566/* XXX - move to kern_timeout.c? */ 567static void 568filt_timerdetach(struct knote *kn) 569{ 570 struct callout *calloutp; 571 572 calloutp = (struct callout *)kn->kn_hook; 573 callout_drain(calloutp); 574 free(calloutp, M_KQUEUE); 575 atomic_add_int(&kq_ncallouts, -1); 576 kn->kn_status |= KN_DETACHED; /* knlist_remove usually clears it */ 577} 578 579/* XXX - move to kern_timeout.c? */ 580static int 581filt_timer(struct knote *kn, long hint) 582{ 583 584 return (kn->kn_data != 0); 585} 586 587static int 588filt_userattach(struct knote *kn) 589{ 590 591 /* 592 * EVFILT_USER knotes are not attached to anything in the kernel. 593 */ 594 kn->kn_hook = NULL; 595 if (kn->kn_fflags & NOTE_TRIGGER) 596 kn->kn_hookid = 1; 597 else 598 kn->kn_hookid = 0; 599 return (0); 600} 601 602static void 603filt_userdetach(__unused struct knote *kn) 604{ 605 606 /* 607 * EVFILT_USER knotes are not attached to anything in the kernel. 608 */ 609} 610 611static int 612filt_user(struct knote *kn, __unused long hint) 613{ 614 615 return (kn->kn_hookid); 616} 617 618static void 619filt_usertouch(struct knote *kn, struct kevent *kev, long type) 620{ 621 int ffctrl; 622 623 switch (type) { 624 case EVENT_REGISTER: 625 if (kev->fflags & NOTE_TRIGGER) 626 kn->kn_hookid = 1; 627 628 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 629 kev->fflags &= NOTE_FFLAGSMASK; 630 switch (ffctrl) { 631 case NOTE_FFNOP: 632 break; 633 634 case NOTE_FFAND: 635 kn->kn_sfflags &= kev->fflags; 636 break; 637 638 case NOTE_FFOR: 639 kn->kn_sfflags |= kev->fflags; 640 break; 641 642 case NOTE_FFCOPY: 643 kn->kn_sfflags = kev->fflags; 644 break; 645 646 default: 647 /* XXX Return error? */ 648 break; 649 } 650 kn->kn_sdata = kev->data; 651 if (kev->flags & EV_CLEAR) { 652 kn->kn_hookid = 0; 653 kn->kn_data = 0; 654 kn->kn_fflags = 0; 655 } 656 break; 657 658 case EVENT_PROCESS: 659 *kev = kn->kn_kevent; 660 kev->fflags = kn->kn_sfflags; 661 kev->data = kn->kn_sdata; 662 if (kn->kn_flags & EV_CLEAR) { 663 kn->kn_hookid = 0; 664 kn->kn_data = 0; 665 kn->kn_fflags = 0; 666 } 667 break; 668 669 default: 670 panic("filt_usertouch() - invalid type (%ld)", type); 671 break; 672 } 673} 674 675int 676kqueue(struct thread *td, struct kqueue_args *uap) 677{ 678 struct filedesc *fdp; 679 struct kqueue *kq; 680 struct file *fp; 681 int fd, error; 682 683 fdp = td->td_proc->p_fd; 684 error = falloc(td, &fp, &fd); 685 if (error) 686 goto done2; 687 688 /* An extra reference on `nfp' has been held for us by falloc(). */ 689 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 690 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK); 691 TAILQ_INIT(&kq->kq_head); 692 kq->kq_fdp = fdp; 693 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 694 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 695 696 FILEDESC_XLOCK(fdp); 697 SLIST_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 698 FILEDESC_XUNLOCK(fdp); 699 700 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 701 fdrop(fp, td); 702 703 td->td_retval[0] = fd; 704done2: 705 return (error); 706} 707 708#ifndef _SYS_SYSPROTO_H_ 709struct kevent_args { 710 int fd; 711 const struct kevent *changelist; 712 int nchanges; 713 struct kevent *eventlist; 714 int nevents; 715 const struct timespec *timeout; 716}; 717#endif 718int 719kevent(struct thread *td, struct kevent_args *uap) 720{ 721 struct timespec ts, *tsp; 722 struct kevent_copyops k_ops = { uap, 723 kevent_copyout, 724 kevent_copyin}; 725 int error; 726#ifdef KTRACE 727 struct uio ktruio; 728 struct iovec ktriov; 729 struct uio *ktruioin = NULL; 730 struct uio *ktruioout = NULL; 731#endif 732 733 if (uap->timeout != NULL) { 734 error = copyin(uap->timeout, &ts, sizeof(ts)); 735 if (error) 736 return (error); 737 tsp = &ts; 738 } else 739 tsp = NULL; 740 741#ifdef KTRACE 742 if (KTRPOINT(td, KTR_GENIO)) { 743 ktriov.iov_base = uap->changelist; 744 ktriov.iov_len = uap->nchanges * sizeof(struct kevent); 745 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1, 746 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ, 747 .uio_td = td }; 748 ktruioin = cloneuio(&ktruio); 749 ktriov.iov_base = uap->eventlist; 750 ktriov.iov_len = uap->nevents * sizeof(struct kevent); 751 ktruioout = cloneuio(&ktruio); 752 } 753#endif 754 755 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 756 &k_ops, tsp); 757 758#ifdef KTRACE 759 if (ktruioin != NULL) { 760 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent); 761 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0); 762 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent); 763 ktrgenio(uap->fd, UIO_READ, ktruioout, error); 764 } 765#endif 766 767 return (error); 768} 769 770/* 771 * Copy 'count' items into the destination list pointed to by uap->eventlist. 772 */ 773static int 774kevent_copyout(void *arg, struct kevent *kevp, int count) 775{ 776 struct kevent_args *uap; 777 int error; 778 779 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 780 uap = (struct kevent_args *)arg; 781 782 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 783 if (error == 0) 784 uap->eventlist += count; 785 return (error); 786} 787 788/* 789 * Copy 'count' items from the list pointed to by uap->changelist. 790 */ 791static int 792kevent_copyin(void *arg, struct kevent *kevp, int count) 793{ 794 struct kevent_args *uap; 795 int error; 796 797 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 798 uap = (struct kevent_args *)arg; 799 800 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 801 if (error == 0) 802 uap->changelist += count; 803 return (error); 804} 805 806int 807kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 808 struct kevent_copyops *k_ops, const struct timespec *timeout) 809{ 810 struct kevent keva[KQ_NEVENTS]; 811 struct kevent *kevp, *changes; 812 struct kqueue *kq; 813 struct file *fp; 814 int i, n, nerrors, error; 815 816 if ((error = fget(td, fd, &fp)) != 0) 817 return (error); 818 if ((error = kqueue_acquire(fp, &kq)) != 0) 819 goto done_norel; 820 821 nerrors = 0; 822 823 while (nchanges > 0) { 824 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 825 error = k_ops->k_copyin(k_ops->arg, keva, n); 826 if (error) 827 goto done; 828 changes = keva; 829 for (i = 0; i < n; i++) { 830 kevp = &changes[i]; 831 if (!kevp->filter) 832 continue; 833 kevp->flags &= ~EV_SYSFLAGS; 834 error = kqueue_register(kq, kevp, td, 1); 835 if (error) { 836 if (nevents != 0) { 837 kevp->flags = EV_ERROR; 838 kevp->data = error; 839 (void) k_ops->k_copyout(k_ops->arg, 840 kevp, 1); 841 nevents--; 842 nerrors++; 843 } else { 844 goto done; 845 } 846 } 847 } 848 nchanges -= n; 849 } 850 if (nerrors) { 851 td->td_retval[0] = nerrors; 852 error = 0; 853 goto done; 854 } 855 856 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td); 857done: 858 kqueue_release(kq, 0); 859done_norel: 860 fdrop(fp, td); 861 return (error); 862} 863 864int 865kqueue_add_filteropts(int filt, struct filterops *filtops) 866{ 867 int error; 868 869 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 870 printf( 871"trying to add a filterop that is out of range: %d is beyond %d\n", 872 ~filt, EVFILT_SYSCOUNT); 873 return EINVAL; 874 } 875 mtx_lock(&filterops_lock); 876 if (sysfilt_ops[~filt].for_fop != &null_filtops && 877 sysfilt_ops[~filt].for_fop != NULL) 878 error = EEXIST; 879 else { 880 sysfilt_ops[~filt].for_fop = filtops; 881 sysfilt_ops[~filt].for_refcnt = 0; 882 } 883 mtx_unlock(&filterops_lock); 884 885 return (0); 886} 887 888int 889kqueue_del_filteropts(int filt) 890{ 891 int error; 892 893 error = 0; 894 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 895 return EINVAL; 896 897 mtx_lock(&filterops_lock); 898 if (sysfilt_ops[~filt].for_fop == &null_filtops || 899 sysfilt_ops[~filt].for_fop == NULL) 900 error = EINVAL; 901 else if (sysfilt_ops[~filt].for_refcnt != 0) 902 error = EBUSY; 903 else { 904 sysfilt_ops[~filt].for_fop = &null_filtops; 905 sysfilt_ops[~filt].for_refcnt = 0; 906 } 907 mtx_unlock(&filterops_lock); 908 909 return error; 910} 911 912static struct filterops * 913kqueue_fo_find(int filt) 914{ 915 916 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 917 return NULL; 918 919 mtx_lock(&filterops_lock); 920 sysfilt_ops[~filt].for_refcnt++; 921 if (sysfilt_ops[~filt].for_fop == NULL) 922 sysfilt_ops[~filt].for_fop = &null_filtops; 923 mtx_unlock(&filterops_lock); 924 925 return sysfilt_ops[~filt].for_fop; 926} 927 928static void 929kqueue_fo_release(int filt) 930{ 931 932 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 933 return; 934 935 mtx_lock(&filterops_lock); 936 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 937 ("filter object refcount not valid on release")); 938 sysfilt_ops[~filt].for_refcnt--; 939 mtx_unlock(&filterops_lock); 940} 941 942/* 943 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will 944 * influence if memory allocation should wait. Make sure it is 0 if you 945 * hold any mutexes. 946 */ 947static int 948kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok) 949{ 950 struct filterops *fops; 951 struct file *fp; 952 struct knote *kn, *tkn; 953 int error, filt, event; 954 int haskqglobal; 955 956 fp = NULL; 957 kn = NULL; 958 error = 0; 959 haskqglobal = 0; 960 961 filt = kev->filter; 962 fops = kqueue_fo_find(filt); 963 if (fops == NULL) 964 return EINVAL; 965 966 tkn = knote_alloc(waitok); /* prevent waiting with locks */ 967 968findkn: 969 if (fops->f_isfd) { 970 KASSERT(td != NULL, ("td is NULL")); 971 error = fget(td, kev->ident, &fp); 972 if (error) 973 goto done; 974 975 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 976 kev->ident, 0) != 0) { 977 /* try again */ 978 fdrop(fp, td); 979 fp = NULL; 980 error = kqueue_expand(kq, fops, kev->ident, waitok); 981 if (error) 982 goto done; 983 goto findkn; 984 } 985 986 if (fp->f_type == DTYPE_KQUEUE) { 987 /* 988 * if we add some inteligence about what we are doing, 989 * we should be able to support events on ourselves. 990 * We need to know when we are doing this to prevent 991 * getting both the knlist lock and the kq lock since 992 * they are the same thing. 993 */ 994 if (fp->f_data == kq) { 995 error = EINVAL; 996 goto done; 997 } 998 999 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1000 } 1001 1002 KQ_LOCK(kq); 1003 if (kev->ident < kq->kq_knlistsize) { 1004 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1005 if (kev->filter == kn->kn_filter) 1006 break; 1007 } 1008 } else { 1009 if ((kev->flags & EV_ADD) == EV_ADD) 1010 kqueue_expand(kq, fops, kev->ident, waitok); 1011 1012 KQ_LOCK(kq); 1013 if (kq->kq_knhashmask != 0) { 1014 struct klist *list; 1015 1016 list = &kq->kq_knhash[ 1017 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1018 SLIST_FOREACH(kn, list, kn_link) 1019 if (kev->ident == kn->kn_id && 1020 kev->filter == kn->kn_filter) 1021 break; 1022 } 1023 } 1024 1025 /* knote is in the process of changing, wait for it to stablize. */ 1026 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1027 if (fp != NULL) { 1028 fdrop(fp, td); 1029 fp = NULL; 1030 } 1031 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1032 kq->kq_state |= KQ_FLUXWAIT; 1033 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1034 goto findkn; 1035 } 1036 1037 /* 1038 * kn now contains the matching knote, or NULL if no match 1039 */ 1040 if (kn == NULL) { 1041 if (kev->flags & EV_ADD) { 1042 kn = tkn; 1043 tkn = NULL; 1044 if (kn == NULL) { 1045 KQ_UNLOCK(kq); 1046 error = ENOMEM; 1047 goto done; 1048 } 1049 kn->kn_fp = fp; 1050 kn->kn_kq = kq; 1051 kn->kn_fop = fops; 1052 /* 1053 * apply reference counts to knote structure, and 1054 * do not release it at the end of this routine. 1055 */ 1056 fops = NULL; 1057 fp = NULL; 1058 1059 kn->kn_sfflags = kev->fflags; 1060 kn->kn_sdata = kev->data; 1061 kev->fflags = 0; 1062 kev->data = 0; 1063 kn->kn_kevent = *kev; 1064 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1065 EV_ENABLE | EV_DISABLE); 1066 kn->kn_status = KN_INFLUX|KN_DETACHED; 1067 1068 error = knote_attach(kn, kq); 1069 KQ_UNLOCK(kq); 1070 if (error != 0) { 1071 tkn = kn; 1072 goto done; 1073 } 1074 1075 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1076 knote_drop(kn, td); 1077 goto done; 1078 } 1079 KN_LIST_LOCK(kn); 1080 goto done_ev_add; 1081 } else { 1082 /* No matching knote and the EV_ADD flag is not set. */ 1083 KQ_UNLOCK(kq); 1084 error = ENOENT; 1085 goto done; 1086 } 1087 } 1088 1089 if (kev->flags & EV_DELETE) { 1090 kn->kn_status |= KN_INFLUX; 1091 KQ_UNLOCK(kq); 1092 if (!(kn->kn_status & KN_DETACHED)) 1093 kn->kn_fop->f_detach(kn); 1094 knote_drop(kn, td); 1095 goto done; 1096 } 1097 1098 /* 1099 * The user may change some filter values after the initial EV_ADD, 1100 * but doing so will not reset any filter which has already been 1101 * triggered. 1102 */ 1103 kn->kn_status |= KN_INFLUX; 1104 KQ_UNLOCK(kq); 1105 KN_LIST_LOCK(kn); 1106 kn->kn_kevent.udata = kev->udata; 1107 if (!fops->f_isfd && fops->f_touch != NULL) { 1108 fops->f_touch(kn, kev, EVENT_REGISTER); 1109 } else { 1110 kn->kn_sfflags = kev->fflags; 1111 kn->kn_sdata = kev->data; 1112 } 1113 1114 /* 1115 * We can get here with kn->kn_knlist == NULL. This can happen when 1116 * the initial attach event decides that the event is "completed" 1117 * already. i.e. filt_procattach is called on a zombie process. It 1118 * will call filt_proc which will remove it from the list, and NULL 1119 * kn_knlist. 1120 */ 1121done_ev_add: 1122 event = kn->kn_fop->f_event(kn, 0); 1123 KQ_LOCK(kq); 1124 if (event) 1125 KNOTE_ACTIVATE(kn, 1); 1126 kn->kn_status &= ~KN_INFLUX; 1127 KN_LIST_UNLOCK(kn); 1128 1129 if ((kev->flags & EV_DISABLE) && 1130 ((kn->kn_status & KN_DISABLED) == 0)) { 1131 kn->kn_status |= KN_DISABLED; 1132 } 1133 1134 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 1135 kn->kn_status &= ~KN_DISABLED; 1136 if ((kn->kn_status & KN_ACTIVE) && 1137 ((kn->kn_status & KN_QUEUED) == 0)) 1138 knote_enqueue(kn); 1139 } 1140 KQ_UNLOCK_FLUX(kq); 1141 1142done: 1143 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1144 if (fp != NULL) 1145 fdrop(fp, td); 1146 if (tkn != NULL) 1147 knote_free(tkn); 1148 if (fops != NULL) 1149 kqueue_fo_release(filt); 1150 return (error); 1151} 1152 1153static int 1154kqueue_acquire(struct file *fp, struct kqueue **kqp) 1155{ 1156 int error; 1157 struct kqueue *kq; 1158 1159 error = 0; 1160 1161 kq = fp->f_data; 1162 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1163 return (EBADF); 1164 *kqp = kq; 1165 KQ_LOCK(kq); 1166 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1167 KQ_UNLOCK(kq); 1168 return (EBADF); 1169 } 1170 kq->kq_refcnt++; 1171 KQ_UNLOCK(kq); 1172 1173 return error; 1174} 1175 1176static void 1177kqueue_release(struct kqueue *kq, int locked) 1178{ 1179 if (locked) 1180 KQ_OWNED(kq); 1181 else 1182 KQ_LOCK(kq); 1183 kq->kq_refcnt--; 1184 if (kq->kq_refcnt == 1) 1185 wakeup(&kq->kq_refcnt); 1186 if (!locked) 1187 KQ_UNLOCK(kq); 1188} 1189 1190static void 1191kqueue_schedtask(struct kqueue *kq) 1192{ 1193 1194 KQ_OWNED(kq); 1195 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1196 ("scheduling kqueue task while draining")); 1197 1198 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1199 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task); 1200 kq->kq_state |= KQ_TASKSCHED; 1201 } 1202} 1203 1204/* 1205 * Expand the kq to make sure we have storage for fops/ident pair. 1206 * 1207 * Return 0 on success (or no work necessary), return errno on failure. 1208 * 1209 * Not calling hashinit w/ waitok (proper malloc flag) should be safe. 1210 * If kqueue_register is called from a non-fd context, there usually/should 1211 * be no locks held. 1212 */ 1213static int 1214kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, 1215 int waitok) 1216{ 1217 struct klist *list, *tmp_knhash; 1218 u_long tmp_knhashmask; 1219 int size; 1220 int fd; 1221 int mflag = waitok ? M_WAITOK : M_NOWAIT; 1222 1223 KQ_NOTOWNED(kq); 1224 1225 if (fops->f_isfd) { 1226 fd = ident; 1227 if (kq->kq_knlistsize <= fd) { 1228 size = kq->kq_knlistsize; 1229 while (size <= fd) 1230 size += KQEXTENT; 1231 list = malloc(size * sizeof list, M_KQUEUE, mflag); 1232 if (list == NULL) 1233 return ENOMEM; 1234 KQ_LOCK(kq); 1235 if (kq->kq_knlistsize > fd) { 1236 free(list, M_KQUEUE); 1237 list = NULL; 1238 } else { 1239 if (kq->kq_knlist != NULL) { 1240 bcopy(kq->kq_knlist, list, 1241 kq->kq_knlistsize * sizeof list); 1242 free(kq->kq_knlist, M_KQUEUE); 1243 kq->kq_knlist = NULL; 1244 } 1245 bzero((caddr_t)list + 1246 kq->kq_knlistsize * sizeof list, 1247 (size - kq->kq_knlistsize) * sizeof list); 1248 kq->kq_knlistsize = size; 1249 kq->kq_knlist = list; 1250 } 1251 KQ_UNLOCK(kq); 1252 } 1253 } else { 1254 if (kq->kq_knhashmask == 0) { 1255 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1256 &tmp_knhashmask); 1257 if (tmp_knhash == NULL) 1258 return ENOMEM; 1259 KQ_LOCK(kq); 1260 if (kq->kq_knhashmask == 0) { 1261 kq->kq_knhash = tmp_knhash; 1262 kq->kq_knhashmask = tmp_knhashmask; 1263 } else { 1264 free(tmp_knhash, M_KQUEUE); 1265 } 1266 KQ_UNLOCK(kq); 1267 } 1268 } 1269 1270 KQ_NOTOWNED(kq); 1271 return 0; 1272} 1273 1274static void 1275kqueue_task(void *arg, int pending) 1276{ 1277 struct kqueue *kq; 1278 int haskqglobal; 1279 1280 haskqglobal = 0; 1281 kq = arg; 1282 1283 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1284 KQ_LOCK(kq); 1285 1286 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 1287 1288 kq->kq_state &= ~KQ_TASKSCHED; 1289 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 1290 wakeup(&kq->kq_state); 1291 } 1292 KQ_UNLOCK(kq); 1293 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1294} 1295 1296/* 1297 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 1298 * We treat KN_MARKER knotes as if they are INFLUX. 1299 */ 1300static int 1301kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 1302 const struct timespec *tsp, struct kevent *keva, struct thread *td) 1303{ 1304 struct kevent *kevp; 1305 struct timeval atv, rtv, ttv; 1306 struct knote *kn, *marker; 1307 int count, timeout, nkev, error, influx; 1308 int haskqglobal, touch; 1309 1310 count = maxevents; 1311 nkev = 0; 1312 error = 0; 1313 haskqglobal = 0; 1314 1315 if (maxevents == 0) 1316 goto done_nl; 1317 1318 if (tsp != NULL) { 1319 TIMESPEC_TO_TIMEVAL(&atv, tsp); 1320 if (itimerfix(&atv)) { 1321 error = EINVAL; 1322 goto done_nl; 1323 } 1324 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 1325 timeout = -1; 1326 else 1327 timeout = atv.tv_sec > 24 * 60 * 60 ? 1328 24 * 60 * 60 * hz : tvtohz(&atv); 1329 getmicrouptime(&rtv); 1330 timevaladd(&atv, &rtv); 1331 } else { 1332 atv.tv_sec = 0; 1333 atv.tv_usec = 0; 1334 timeout = 0; 1335 } 1336 marker = knote_alloc(1); 1337 if (marker == NULL) { 1338 error = ENOMEM; 1339 goto done_nl; 1340 } 1341 marker->kn_status = KN_MARKER; 1342 KQ_LOCK(kq); 1343 goto start; 1344 1345retry: 1346 if (atv.tv_sec || atv.tv_usec) { 1347 getmicrouptime(&rtv); 1348 if (timevalcmp(&rtv, &atv, >=)) 1349 goto done; 1350 ttv = atv; 1351 timevalsub(&ttv, &rtv); 1352 timeout = ttv.tv_sec > 24 * 60 * 60 ? 1353 24 * 60 * 60 * hz : tvtohz(&ttv); 1354 } 1355 1356start: 1357 kevp = keva; 1358 if (kq->kq_count == 0) { 1359 if (timeout < 0) { 1360 error = EWOULDBLOCK; 1361 } else { 1362 kq->kq_state |= KQ_SLEEP; 1363 error = msleep(kq, &kq->kq_lock, PSOCK | PCATCH, 1364 "kqread", timeout); 1365 } 1366 if (error == 0) 1367 goto retry; 1368 /* don't restart after signals... */ 1369 if (error == ERESTART) 1370 error = EINTR; 1371 else if (error == EWOULDBLOCK) 1372 error = 0; 1373 goto done; 1374 } 1375 1376 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1377 influx = 0; 1378 while (count) { 1379 KQ_OWNED(kq); 1380 kn = TAILQ_FIRST(&kq->kq_head); 1381 1382 if ((kn->kn_status == KN_MARKER && kn != marker) || 1383 (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1384 if (influx) { 1385 influx = 0; 1386 KQ_FLUX_WAKEUP(kq); 1387 } 1388 kq->kq_state |= KQ_FLUXWAIT; 1389 error = msleep(kq, &kq->kq_lock, PSOCK, 1390 "kqflxwt", 0); 1391 continue; 1392 } 1393 1394 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1395 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 1396 kn->kn_status &= ~KN_QUEUED; 1397 kq->kq_count--; 1398 continue; 1399 } 1400 if (kn == marker) { 1401 KQ_FLUX_WAKEUP(kq); 1402 if (count == maxevents) 1403 goto retry; 1404 goto done; 1405 } 1406 KASSERT((kn->kn_status & KN_INFLUX) == 0, 1407 ("KN_INFLUX set when not suppose to be")); 1408 1409 if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 1410 kn->kn_status &= ~KN_QUEUED; 1411 kn->kn_status |= KN_INFLUX; 1412 kq->kq_count--; 1413 KQ_UNLOCK(kq); 1414 /* 1415 * We don't need to lock the list since we've marked 1416 * it _INFLUX. 1417 */ 1418 *kevp = kn->kn_kevent; 1419 if (!(kn->kn_status & KN_DETACHED)) 1420 kn->kn_fop->f_detach(kn); 1421 knote_drop(kn, td); 1422 KQ_LOCK(kq); 1423 kn = NULL; 1424 } else { 1425 kn->kn_status |= KN_INFLUX; 1426 KQ_UNLOCK(kq); 1427 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 1428 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1429 KN_LIST_LOCK(kn); 1430 if (kn->kn_fop->f_event(kn, 0) == 0) { 1431 KQ_LOCK(kq); 1432 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1433 kn->kn_status &= 1434 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX); 1435 kq->kq_count--; 1436 KN_LIST_UNLOCK(kn); 1437 influx = 1; 1438 continue; 1439 } 1440 touch = (!kn->kn_fop->f_isfd && 1441 kn->kn_fop->f_touch != NULL); 1442 if (touch) 1443 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 1444 else 1445 *kevp = kn->kn_kevent; 1446 KQ_LOCK(kq); 1447 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1448 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 1449 /* 1450 * Manually clear knotes who weren't 1451 * 'touch'ed. 1452 */ 1453 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 1454 kn->kn_data = 0; 1455 kn->kn_fflags = 0; 1456 } 1457 if (kn->kn_flags & EV_DISPATCH) 1458 kn->kn_status |= KN_DISABLED; 1459 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1460 kq->kq_count--; 1461 } else 1462 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1463 1464 kn->kn_status &= ~(KN_INFLUX); 1465 KN_LIST_UNLOCK(kn); 1466 influx = 1; 1467 } 1468 1469 /* we are returning a copy to the user */ 1470 kevp++; 1471 nkev++; 1472 count--; 1473 1474 if (nkev == KQ_NEVENTS) { 1475 influx = 0; 1476 KQ_UNLOCK_FLUX(kq); 1477 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1478 nkev = 0; 1479 kevp = keva; 1480 KQ_LOCK(kq); 1481 if (error) 1482 break; 1483 } 1484 } 1485 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1486done: 1487 KQ_OWNED(kq); 1488 KQ_UNLOCK_FLUX(kq); 1489 knote_free(marker); 1490done_nl: 1491 KQ_NOTOWNED(kq); 1492 if (nkev != 0) 1493 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1494 td->td_retval[0] = maxevents - count; 1495 return (error); 1496} 1497 1498/* 1499 * XXX 1500 * This could be expanded to call kqueue_scan, if desired. 1501 */ 1502/*ARGSUSED*/ 1503static int 1504kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 1505 int flags, struct thread *td) 1506{ 1507 return (ENXIO); 1508} 1509 1510/*ARGSUSED*/ 1511static int 1512kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 1513 int flags, struct thread *td) 1514{ 1515 return (ENXIO); 1516} 1517 1518/*ARGSUSED*/ 1519static int 1520kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1521 struct thread *td) 1522{ 1523 1524 return (EINVAL); 1525} 1526 1527/*ARGSUSED*/ 1528static int 1529kqueue_ioctl(struct file *fp, u_long cmd, void *data, 1530 struct ucred *active_cred, struct thread *td) 1531{ 1532 /* 1533 * Enabling sigio causes two major problems: 1534 * 1) infinite recursion: 1535 * Synopsys: kevent is being used to track signals and have FIOASYNC 1536 * set. On receipt of a signal this will cause a kqueue to recurse 1537 * into itself over and over. Sending the sigio causes the kqueue 1538 * to become ready, which in turn posts sigio again, forever. 1539 * Solution: this can be solved by setting a flag in the kqueue that 1540 * we have a SIGIO in progress. 1541 * 2) locking problems: 1542 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 1543 * us above the proc and pgrp locks. 1544 * Solution: Post a signal using an async mechanism, being sure to 1545 * record a generation count in the delivery so that we do not deliver 1546 * a signal to the wrong process. 1547 * 1548 * Note, these two mechanisms are somewhat mutually exclusive! 1549 */ 1550#if 0 1551 struct kqueue *kq; 1552 1553 kq = fp->f_data; 1554 switch (cmd) { 1555 case FIOASYNC: 1556 if (*(int *)data) { 1557 kq->kq_state |= KQ_ASYNC; 1558 } else { 1559 kq->kq_state &= ~KQ_ASYNC; 1560 } 1561 return (0); 1562 1563 case FIOSETOWN: 1564 return (fsetown(*(int *)data, &kq->kq_sigio)); 1565 1566 case FIOGETOWN: 1567 *(int *)data = fgetown(&kq->kq_sigio); 1568 return (0); 1569 } 1570#endif 1571 1572 return (ENOTTY); 1573} 1574 1575/*ARGSUSED*/ 1576static int 1577kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 1578 struct thread *td) 1579{ 1580 struct kqueue *kq; 1581 int revents = 0; 1582 int error; 1583 1584 if ((error = kqueue_acquire(fp, &kq))) 1585 return POLLERR; 1586 1587 KQ_LOCK(kq); 1588 if (events & (POLLIN | POLLRDNORM)) { 1589 if (kq->kq_count) { 1590 revents |= events & (POLLIN | POLLRDNORM); 1591 } else { 1592 selrecord(td, &kq->kq_sel); 1593 if (SEL_WAITING(&kq->kq_sel)) 1594 kq->kq_state |= KQ_SEL; 1595 } 1596 } 1597 kqueue_release(kq, 1); 1598 KQ_UNLOCK(kq); 1599 return (revents); 1600} 1601 1602/*ARGSUSED*/ 1603static int 1604kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 1605 struct thread *td) 1606{ 1607 1608 bzero((void *)st, sizeof *st); 1609 /* 1610 * We no longer return kq_count because the unlocked value is useless. 1611 * If you spent all this time getting the count, why not spend your 1612 * syscall better by calling kevent? 1613 * 1614 * XXX - This is needed for libc_r. 1615 */ 1616 st->st_mode = S_IFIFO; 1617 return (0); 1618} 1619 1620/*ARGSUSED*/ 1621static int 1622kqueue_close(struct file *fp, struct thread *td) 1623{ 1624 struct kqueue *kq = fp->f_data; 1625 struct filedesc *fdp; 1626 struct knote *kn; 1627 int i; 1628 int error; 1629 1630 if ((error = kqueue_acquire(fp, &kq))) 1631 return error; 1632 1633 KQ_LOCK(kq); 1634 1635 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 1636 ("kqueue already closing")); 1637 kq->kq_state |= KQ_CLOSING; 1638 if (kq->kq_refcnt > 1) 1639 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 1640 1641 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 1642 fdp = kq->kq_fdp; 1643 1644 KASSERT(knlist_empty(&kq->kq_sel.si_note), 1645 ("kqueue's knlist not empty")); 1646 1647 for (i = 0; i < kq->kq_knlistsize; i++) { 1648 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 1649 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1650 kq->kq_state |= KQ_FLUXWAIT; 1651 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 1652 continue; 1653 } 1654 kn->kn_status |= KN_INFLUX; 1655 KQ_UNLOCK(kq); 1656 if (!(kn->kn_status & KN_DETACHED)) 1657 kn->kn_fop->f_detach(kn); 1658 knote_drop(kn, td); 1659 KQ_LOCK(kq); 1660 } 1661 } 1662 if (kq->kq_knhashmask != 0) { 1663 for (i = 0; i <= kq->kq_knhashmask; i++) { 1664 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 1665 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1666 kq->kq_state |= KQ_FLUXWAIT; 1667 msleep(kq, &kq->kq_lock, PSOCK, 1668 "kqclo2", 0); 1669 continue; 1670 } 1671 kn->kn_status |= KN_INFLUX; 1672 KQ_UNLOCK(kq); 1673 if (!(kn->kn_status & KN_DETACHED)) 1674 kn->kn_fop->f_detach(kn); 1675 knote_drop(kn, td); 1676 KQ_LOCK(kq); 1677 } 1678 } 1679 } 1680 1681 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 1682 kq->kq_state |= KQ_TASKDRAIN; 1683 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 1684 } 1685 1686 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1687 selwakeuppri(&kq->kq_sel, PSOCK); 1688 if (!SEL_WAITING(&kq->kq_sel)) 1689 kq->kq_state &= ~KQ_SEL; 1690 } 1691 1692 KQ_UNLOCK(kq); 1693 1694 FILEDESC_XLOCK(fdp); 1695 SLIST_REMOVE(&fdp->fd_kqlist, kq, kqueue, kq_list); 1696 FILEDESC_XUNLOCK(fdp); 1697 1698 knlist_destroy(&kq->kq_sel.si_note); 1699 mtx_destroy(&kq->kq_lock); 1700 kq->kq_fdp = NULL; 1701 1702 if (kq->kq_knhash != NULL) 1703 free(kq->kq_knhash, M_KQUEUE); 1704 if (kq->kq_knlist != NULL) 1705 free(kq->kq_knlist, M_KQUEUE); 1706 1707 funsetown(&kq->kq_sigio); 1708 free(kq, M_KQUEUE); 1709 fp->f_data = NULL; 1710 1711 return (0); 1712} 1713 1714static void 1715kqueue_wakeup(struct kqueue *kq) 1716{ 1717 KQ_OWNED(kq); 1718 1719 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 1720 kq->kq_state &= ~KQ_SLEEP; 1721 wakeup(kq); 1722 } 1723 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1724 selwakeuppri(&kq->kq_sel, PSOCK); 1725 if (!SEL_WAITING(&kq->kq_sel)) 1726 kq->kq_state &= ~KQ_SEL; 1727 } 1728 if (!knlist_empty(&kq->kq_sel.si_note)) 1729 kqueue_schedtask(kq); 1730 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 1731 pgsigio(&kq->kq_sigio, SIGIO, 0); 1732 } 1733} 1734 1735/* 1736 * Walk down a list of knotes, activating them if their event has triggered. 1737 * 1738 * There is a possibility to optimize in the case of one kq watching another. 1739 * Instead of scheduling a task to wake it up, you could pass enough state 1740 * down the chain to make up the parent kqueue. Make this code functional 1741 * first. 1742 */ 1743void 1744knote(struct knlist *list, long hint, int lockflags) 1745{ 1746 struct kqueue *kq; 1747 struct knote *kn; 1748 int error; 1749 1750 if (list == NULL) 1751 return; 1752 1753 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 1754 1755 if ((lockflags & KNF_LISTLOCKED) == 0) 1756 list->kl_lock(list->kl_lockarg); 1757 1758 /* 1759 * If we unlock the list lock (and set KN_INFLUX), we can eliminate 1760 * the kqueue scheduling, but this will introduce four 1761 * lock/unlock's for each knote to test. If we do, continue to use 1762 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is 1763 * only safe if you want to remove the current item, which we are 1764 * not doing. 1765 */ 1766 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 1767 kq = kn->kn_kq; 1768 if ((kn->kn_status & KN_INFLUX) != KN_INFLUX) { 1769 KQ_LOCK(kq); 1770 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1771 KQ_UNLOCK(kq); 1772 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 1773 kn->kn_status |= KN_INFLUX; 1774 KQ_UNLOCK(kq); 1775 error = kn->kn_fop->f_event(kn, hint); 1776 KQ_LOCK(kq); 1777 kn->kn_status &= ~KN_INFLUX; 1778 if (error) 1779 KNOTE_ACTIVATE(kn, 1); 1780 KQ_UNLOCK_FLUX(kq); 1781 } else { 1782 kn->kn_status |= KN_HASKQLOCK; 1783 if (kn->kn_fop->f_event(kn, hint)) 1784 KNOTE_ACTIVATE(kn, 1); 1785 kn->kn_status &= ~KN_HASKQLOCK; 1786 KQ_UNLOCK(kq); 1787 } 1788 } 1789 kq = NULL; 1790 } 1791 if ((lockflags & KNF_LISTLOCKED) == 0) 1792 list->kl_unlock(list->kl_lockarg); 1793} 1794 1795/* 1796 * add a knote to a knlist 1797 */ 1798void 1799knlist_add(struct knlist *knl, struct knote *kn, int islocked) 1800{ 1801 KNL_ASSERT_LOCK(knl, islocked); 1802 KQ_NOTOWNED(kn->kn_kq); 1803 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == 1804 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED")); 1805 if (!islocked) 1806 knl->kl_lock(knl->kl_lockarg); 1807 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 1808 if (!islocked) 1809 knl->kl_unlock(knl->kl_lockarg); 1810 KQ_LOCK(kn->kn_kq); 1811 kn->kn_knlist = knl; 1812 kn->kn_status &= ~KN_DETACHED; 1813 KQ_UNLOCK(kn->kn_kq); 1814} 1815 1816static void 1817knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked) 1818{ 1819 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked")); 1820 KNL_ASSERT_LOCK(knl, knlislocked); 1821 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 1822 if (!kqislocked) 1823 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX, 1824 ("knlist_remove called w/o knote being KN_INFLUX or already removed")); 1825 if (!knlislocked) 1826 knl->kl_lock(knl->kl_lockarg); 1827 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 1828 kn->kn_knlist = NULL; 1829 if (!knlislocked) 1830 knl->kl_unlock(knl->kl_lockarg); 1831 if (!kqislocked) 1832 KQ_LOCK(kn->kn_kq); 1833 kn->kn_status |= KN_DETACHED; 1834 if (!kqislocked) 1835 KQ_UNLOCK(kn->kn_kq); 1836} 1837 1838/* 1839 * remove all knotes from a specified klist 1840 */ 1841void 1842knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 1843{ 1844 1845 knlist_remove_kq(knl, kn, islocked, 0); 1846} 1847 1848/* 1849 * remove knote from a specified klist while in f_event handler. 1850 */ 1851void 1852knlist_remove_inevent(struct knlist *knl, struct knote *kn) 1853{ 1854 1855 knlist_remove_kq(knl, kn, 1, 1856 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK); 1857} 1858 1859int 1860knlist_empty(struct knlist *knl) 1861{ 1862 KNL_ASSERT_LOCKED(knl); 1863 return SLIST_EMPTY(&knl->kl_list); 1864} 1865 1866static struct mtx knlist_lock; 1867MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 1868 MTX_DEF); 1869static void knlist_mtx_lock(void *arg); 1870static void knlist_mtx_unlock(void *arg); 1871 1872static void 1873knlist_mtx_lock(void *arg) 1874{ 1875 mtx_lock((struct mtx *)arg); 1876} 1877 1878static void 1879knlist_mtx_unlock(void *arg) 1880{ 1881 mtx_unlock((struct mtx *)arg); 1882} 1883 1884static void 1885knlist_mtx_assert_locked(void *arg) 1886{ 1887 mtx_assert((struct mtx *)arg, MA_OWNED); 1888} 1889 1890static void 1891knlist_mtx_assert_unlocked(void *arg) 1892{ 1893 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 1894} 1895 1896void 1897knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 1898 void (*kl_unlock)(void *), 1899 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *)) 1900{ 1901 1902 if (lock == NULL) 1903 knl->kl_lockarg = &knlist_lock; 1904 else 1905 knl->kl_lockarg = lock; 1906 1907 if (kl_lock == NULL) 1908 knl->kl_lock = knlist_mtx_lock; 1909 else 1910 knl->kl_lock = kl_lock; 1911 if (kl_unlock == NULL) 1912 knl->kl_unlock = knlist_mtx_unlock; 1913 else 1914 knl->kl_unlock = kl_unlock; 1915 if (kl_assert_locked == NULL) 1916 knl->kl_assert_locked = knlist_mtx_assert_locked; 1917 else 1918 knl->kl_assert_locked = kl_assert_locked; 1919 if (kl_assert_unlocked == NULL) 1920 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked; 1921 else 1922 knl->kl_assert_unlocked = kl_assert_unlocked; 1923 1924 SLIST_INIT(&knl->kl_list); 1925} 1926 1927void 1928knlist_init_mtx(struct knlist *knl, struct mtx *lock) 1929{ 1930 1931 knlist_init(knl, lock, NULL, NULL, NULL, NULL); 1932} 1933 1934void 1935knlist_destroy(struct knlist *knl) 1936{ 1937 1938#ifdef INVARIANTS 1939 /* 1940 * if we run across this error, we need to find the offending 1941 * driver and have it call knlist_clear. 1942 */ 1943 if (!SLIST_EMPTY(&knl->kl_list)) 1944 printf("WARNING: destroying knlist w/ knotes on it!\n"); 1945#endif 1946 1947 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL; 1948 SLIST_INIT(&knl->kl_list); 1949} 1950 1951/* 1952 * Even if we are locked, we may need to drop the lock to allow any influx 1953 * knotes time to "settle". 1954 */ 1955void 1956knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 1957{ 1958 struct knote *kn, *kn2; 1959 struct kqueue *kq; 1960 1961 if (islocked) 1962 KNL_ASSERT_LOCKED(knl); 1963 else { 1964 KNL_ASSERT_UNLOCKED(knl); 1965again: /* need to reacquire lock since we have dropped it */ 1966 knl->kl_lock(knl->kl_lockarg); 1967 } 1968 1969 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 1970 kq = kn->kn_kq; 1971 KQ_LOCK(kq); 1972 if ((kn->kn_status & KN_INFLUX)) { 1973 KQ_UNLOCK(kq); 1974 continue; 1975 } 1976 knlist_remove_kq(knl, kn, 1, 1); 1977 if (killkn) { 1978 kn->kn_status |= KN_INFLUX | KN_DETACHED; 1979 KQ_UNLOCK(kq); 1980 knote_drop(kn, td); 1981 } else { 1982 /* Make sure cleared knotes disappear soon */ 1983 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 1984 KQ_UNLOCK(kq); 1985 } 1986 kq = NULL; 1987 } 1988 1989 if (!SLIST_EMPTY(&knl->kl_list)) { 1990 /* there are still KN_INFLUX remaining */ 1991 kn = SLIST_FIRST(&knl->kl_list); 1992 kq = kn->kn_kq; 1993 KQ_LOCK(kq); 1994 KASSERT(kn->kn_status & KN_INFLUX, 1995 ("knote removed w/o list lock")); 1996 knl->kl_unlock(knl->kl_lockarg); 1997 kq->kq_state |= KQ_FLUXWAIT; 1998 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 1999 kq = NULL; 2000 goto again; 2001 } 2002 2003 if (islocked) 2004 KNL_ASSERT_LOCKED(knl); 2005 else { 2006 knl->kl_unlock(knl->kl_lockarg); 2007 KNL_ASSERT_UNLOCKED(knl); 2008 } 2009} 2010 2011/* 2012 * Remove all knotes referencing a specified fd must be called with FILEDESC 2013 * lock. This prevents a race where a new fd comes along and occupies the 2014 * entry and we attach a knote to the fd. 2015 */ 2016void 2017knote_fdclose(struct thread *td, int fd) 2018{ 2019 struct filedesc *fdp = td->td_proc->p_fd; 2020 struct kqueue *kq; 2021 struct knote *kn; 2022 int influx; 2023 2024 FILEDESC_XLOCK_ASSERT(fdp); 2025 2026 /* 2027 * We shouldn't have to worry about new kevents appearing on fd 2028 * since filedesc is locked. 2029 */ 2030 SLIST_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2031 KQ_LOCK(kq); 2032 2033again: 2034 influx = 0; 2035 while (kq->kq_knlistsize > fd && 2036 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2037 if (kn->kn_status & KN_INFLUX) { 2038 /* someone else might be waiting on our knote */ 2039 if (influx) 2040 wakeup(kq); 2041 kq->kq_state |= KQ_FLUXWAIT; 2042 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2043 goto again; 2044 } 2045 kn->kn_status |= KN_INFLUX; 2046 KQ_UNLOCK(kq); 2047 if (!(kn->kn_status & KN_DETACHED)) 2048 kn->kn_fop->f_detach(kn); 2049 knote_drop(kn, td); 2050 influx = 1; 2051 KQ_LOCK(kq); 2052 } 2053 KQ_UNLOCK_FLUX(kq); 2054 } 2055} 2056 2057static int 2058knote_attach(struct knote *kn, struct kqueue *kq) 2059{ 2060 struct klist *list; 2061 2062 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX")); 2063 KQ_OWNED(kq); 2064 2065 if (kn->kn_fop->f_isfd) { 2066 if (kn->kn_id >= kq->kq_knlistsize) 2067 return ENOMEM; 2068 list = &kq->kq_knlist[kn->kn_id]; 2069 } else { 2070 if (kq->kq_knhash == NULL) 2071 return ENOMEM; 2072 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2073 } 2074 2075 SLIST_INSERT_HEAD(list, kn, kn_link); 2076 2077 return 0; 2078} 2079 2080/* 2081 * knote must already have been detached using the f_detach method. 2082 * no lock need to be held, it is assumed that the KN_INFLUX flag is set 2083 * to prevent other removal. 2084 */ 2085static void 2086knote_drop(struct knote *kn, struct thread *td) 2087{ 2088 struct kqueue *kq; 2089 struct klist *list; 2090 2091 kq = kn->kn_kq; 2092 2093 KQ_NOTOWNED(kq); 2094 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX, 2095 ("knote_drop called without KN_INFLUX set in kn_status")); 2096 2097 KQ_LOCK(kq); 2098 if (kn->kn_fop->f_isfd) 2099 list = &kq->kq_knlist[kn->kn_id]; 2100 else 2101 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2102 2103 if (!SLIST_EMPTY(list)) 2104 SLIST_REMOVE(list, kn, knote, kn_link); 2105 if (kn->kn_status & KN_QUEUED) 2106 knote_dequeue(kn); 2107 KQ_UNLOCK_FLUX(kq); 2108 2109 if (kn->kn_fop->f_isfd) { 2110 fdrop(kn->kn_fp, td); 2111 kn->kn_fp = NULL; 2112 } 2113 kqueue_fo_release(kn->kn_kevent.filter); 2114 kn->kn_fop = NULL; 2115 knote_free(kn); 2116} 2117 2118static void 2119knote_enqueue(struct knote *kn) 2120{ 2121 struct kqueue *kq = kn->kn_kq; 2122 2123 KQ_OWNED(kn->kn_kq); 2124 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2125 2126 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2127 kn->kn_status |= KN_QUEUED; 2128 kq->kq_count++; 2129 kqueue_wakeup(kq); 2130} 2131 2132static void 2133knote_dequeue(struct knote *kn) 2134{ 2135 struct kqueue *kq = kn->kn_kq; 2136 2137 KQ_OWNED(kn->kn_kq); 2138 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2139 2140 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2141 kn->kn_status &= ~KN_QUEUED; 2142 kq->kq_count--; 2143} 2144 2145static void 2146knote_init(void) 2147{ 2148 2149 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2150 NULL, NULL, UMA_ALIGN_PTR, 0); 2151} 2152SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2153 2154static struct knote * 2155knote_alloc(int waitok) 2156{ 2157 return ((struct knote *)uma_zalloc(knote_zone, 2158 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO)); 2159} 2160 2161static void 2162knote_free(struct knote *kn) 2163{ 2164 if (kn != NULL) 2165 uma_zfree(knote_zone, kn); 2166} 2167 2168/* 2169 * Register the kev w/ the kq specified by fd. 2170 */ 2171int 2172kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok) 2173{ 2174 struct kqueue *kq; 2175 struct file *fp; 2176 int error; 2177 2178 if ((error = fget(td, fd, &fp)) != 0) 2179 return (error); 2180 if ((error = kqueue_acquire(fp, &kq)) != 0) 2181 goto noacquire; 2182 2183 error = kqueue_register(kq, kev, td, waitok); 2184 2185 kqueue_release(kq, 0); 2186 2187noacquire: 2188 fdrop(fp, td); 2189 2190 return error; 2191} 2192