kern_event.c revision 280258
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: stable/10/sys/kern/kern_event.c 280258 2015-03-19 13:37:36Z rwatson $"); 31 32#include "opt_ktrace.h" 33 34#include <sys/param.h> 35#include <sys/systm.h> 36#include <sys/capsicum.h> 37#include <sys/kernel.h> 38#include <sys/lock.h> 39#include <sys/mutex.h> 40#include <sys/rwlock.h> 41#include <sys/proc.h> 42#include <sys/malloc.h> 43#include <sys/unistd.h> 44#include <sys/file.h> 45#include <sys/filedesc.h> 46#include <sys/filio.h> 47#include <sys/fcntl.h> 48#include <sys/kthread.h> 49#include <sys/selinfo.h> 50#include <sys/stdatomic.h> 51#include <sys/queue.h> 52#include <sys/event.h> 53#include <sys/eventvar.h> 54#include <sys/poll.h> 55#include <sys/protosw.h> 56#include <sys/sigio.h> 57#include <sys/signalvar.h> 58#include <sys/socket.h> 59#include <sys/socketvar.h> 60#include <sys/stat.h> 61#include <sys/sysctl.h> 62#include <sys/sysproto.h> 63#include <sys/syscallsubr.h> 64#include <sys/taskqueue.h> 65#include <sys/uio.h> 66#ifdef KTRACE 67#include <sys/ktrace.h> 68#endif 69 70#include <vm/uma.h> 71 72static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 73 74/* 75 * This lock is used if multiple kq locks are required. This possibly 76 * should be made into a per proc lock. 77 */ 78static struct mtx kq_global; 79MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 80#define KQ_GLOBAL_LOCK(lck, haslck) do { \ 81 if (!haslck) \ 82 mtx_lock(lck); \ 83 haslck = 1; \ 84} while (0) 85#define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 86 if (haslck) \ 87 mtx_unlock(lck); \ 88 haslck = 0; \ 89} while (0) 90 91TASKQUEUE_DEFINE_THREAD(kqueue); 92 93static int kevent_copyout(void *arg, struct kevent *kevp, int count); 94static int kevent_copyin(void *arg, struct kevent *kevp, int count); 95static int kqueue_register(struct kqueue *kq, struct kevent *kev, 96 struct thread *td, int waitok); 97static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 98static void kqueue_release(struct kqueue *kq, int locked); 99static int kqueue_expand(struct kqueue *kq, struct filterops *fops, 100 uintptr_t ident, int waitok); 101static void kqueue_task(void *arg, int pending); 102static int kqueue_scan(struct kqueue *kq, int maxevents, 103 struct kevent_copyops *k_ops, 104 const struct timespec *timeout, 105 struct kevent *keva, struct thread *td); 106static void kqueue_wakeup(struct kqueue *kq); 107static struct filterops *kqueue_fo_find(int filt); 108static void kqueue_fo_release(int filt); 109 110static fo_rdwr_t kqueue_read; 111static fo_rdwr_t kqueue_write; 112static fo_truncate_t kqueue_truncate; 113static fo_ioctl_t kqueue_ioctl; 114static fo_poll_t kqueue_poll; 115static fo_kqfilter_t kqueue_kqfilter; 116static fo_stat_t kqueue_stat; 117static fo_close_t kqueue_close; 118 119static struct fileops kqueueops = { 120 .fo_read = kqueue_read, 121 .fo_write = kqueue_write, 122 .fo_truncate = kqueue_truncate, 123 .fo_ioctl = kqueue_ioctl, 124 .fo_poll = kqueue_poll, 125 .fo_kqfilter = kqueue_kqfilter, 126 .fo_stat = kqueue_stat, 127 .fo_close = kqueue_close, 128 .fo_chmod = invfo_chmod, 129 .fo_chown = invfo_chown, 130 .fo_sendfile = invfo_sendfile, 131}; 132 133static int knote_attach(struct knote *kn, struct kqueue *kq); 134static void knote_drop(struct knote *kn, struct thread *td); 135static void knote_enqueue(struct knote *kn); 136static void knote_dequeue(struct knote *kn); 137static void knote_init(void); 138static struct knote *knote_alloc(int waitok); 139static void knote_free(struct knote *kn); 140 141static void filt_kqdetach(struct knote *kn); 142static int filt_kqueue(struct knote *kn, long hint); 143static int filt_procattach(struct knote *kn); 144static void filt_procdetach(struct knote *kn); 145static int filt_proc(struct knote *kn, long hint); 146static int filt_fileattach(struct knote *kn); 147static void filt_timerexpire(void *knx); 148static int filt_timerattach(struct knote *kn); 149static void filt_timerdetach(struct knote *kn); 150static int filt_timer(struct knote *kn, long hint); 151static int filt_userattach(struct knote *kn); 152static void filt_userdetach(struct knote *kn); 153static int filt_user(struct knote *kn, long hint); 154static void filt_usertouch(struct knote *kn, struct kevent *kev, 155 u_long type); 156 157static struct filterops file_filtops = { 158 .f_isfd = 1, 159 .f_attach = filt_fileattach, 160}; 161static struct filterops kqread_filtops = { 162 .f_isfd = 1, 163 .f_detach = filt_kqdetach, 164 .f_event = filt_kqueue, 165}; 166/* XXX - move to kern_proc.c? */ 167static struct filterops proc_filtops = { 168 .f_isfd = 0, 169 .f_attach = filt_procattach, 170 .f_detach = filt_procdetach, 171 .f_event = filt_proc, 172}; 173static struct filterops timer_filtops = { 174 .f_isfd = 0, 175 .f_attach = filt_timerattach, 176 .f_detach = filt_timerdetach, 177 .f_event = filt_timer, 178}; 179static struct filterops user_filtops = { 180 .f_attach = filt_userattach, 181 .f_detach = filt_userdetach, 182 .f_event = filt_user, 183 .f_touch = filt_usertouch, 184}; 185 186static uma_zone_t knote_zone; 187static atomic_uint kq_ncallouts = ATOMIC_VAR_INIT(0); 188static unsigned int kq_calloutmax = 4 * 1024; 189SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 190 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 191 192/* XXX - ensure not KN_INFLUX?? */ 193#define KNOTE_ACTIVATE(kn, islock) do { \ 194 if ((islock)) \ 195 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 196 else \ 197 KQ_LOCK((kn)->kn_kq); \ 198 (kn)->kn_status |= KN_ACTIVE; \ 199 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 200 knote_enqueue((kn)); \ 201 if (!(islock)) \ 202 KQ_UNLOCK((kn)->kn_kq); \ 203} while(0) 204#define KQ_LOCK(kq) do { \ 205 mtx_lock(&(kq)->kq_lock); \ 206} while (0) 207#define KQ_FLUX_WAKEUP(kq) do { \ 208 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 209 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 210 wakeup((kq)); \ 211 } \ 212} while (0) 213#define KQ_UNLOCK_FLUX(kq) do { \ 214 KQ_FLUX_WAKEUP(kq); \ 215 mtx_unlock(&(kq)->kq_lock); \ 216} while (0) 217#define KQ_UNLOCK(kq) do { \ 218 mtx_unlock(&(kq)->kq_lock); \ 219} while (0) 220#define KQ_OWNED(kq) do { \ 221 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 222} while (0) 223#define KQ_NOTOWNED(kq) do { \ 224 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 225} while (0) 226#define KN_LIST_LOCK(kn) do { \ 227 if (kn->kn_knlist != NULL) \ 228 kn->kn_knlist->kl_lock(kn->kn_knlist->kl_lockarg); \ 229} while (0) 230#define KN_LIST_UNLOCK(kn) do { \ 231 if (kn->kn_knlist != NULL) \ 232 kn->kn_knlist->kl_unlock(kn->kn_knlist->kl_lockarg); \ 233} while (0) 234#define KNL_ASSERT_LOCK(knl, islocked) do { \ 235 if (islocked) \ 236 KNL_ASSERT_LOCKED(knl); \ 237 else \ 238 KNL_ASSERT_UNLOCKED(knl); \ 239} while (0) 240#ifdef INVARIANTS 241#define KNL_ASSERT_LOCKED(knl) do { \ 242 knl->kl_assert_locked((knl)->kl_lockarg); \ 243} while (0) 244#define KNL_ASSERT_UNLOCKED(knl) do { \ 245 knl->kl_assert_unlocked((knl)->kl_lockarg); \ 246} while (0) 247#else /* !INVARIANTS */ 248#define KNL_ASSERT_LOCKED(knl) do {} while(0) 249#define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 250#endif /* INVARIANTS */ 251 252#define KN_HASHSIZE 64 /* XXX should be tunable */ 253#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 254 255static int 256filt_nullattach(struct knote *kn) 257{ 258 259 return (ENXIO); 260}; 261 262struct filterops null_filtops = { 263 .f_isfd = 0, 264 .f_attach = filt_nullattach, 265}; 266 267/* XXX - make SYSINIT to add these, and move into respective modules. */ 268extern struct filterops sig_filtops; 269extern struct filterops fs_filtops; 270 271/* 272 * Table for for all system-defined filters. 273 */ 274static struct mtx filterops_lock; 275MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", 276 MTX_DEF); 277static struct { 278 struct filterops *for_fop; 279 int for_refcnt; 280} sysfilt_ops[EVFILT_SYSCOUNT] = { 281 { &file_filtops }, /* EVFILT_READ */ 282 { &file_filtops }, /* EVFILT_WRITE */ 283 { &null_filtops }, /* EVFILT_AIO */ 284 { &file_filtops }, /* EVFILT_VNODE */ 285 { &proc_filtops }, /* EVFILT_PROC */ 286 { &sig_filtops }, /* EVFILT_SIGNAL */ 287 { &timer_filtops }, /* EVFILT_TIMER */ 288 { &null_filtops }, /* former EVFILT_NETDEV */ 289 { &fs_filtops }, /* EVFILT_FS */ 290 { &null_filtops }, /* EVFILT_LIO */ 291 { &user_filtops }, /* EVFILT_USER */ 292}; 293 294/* 295 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 296 * method. 297 */ 298static int 299filt_fileattach(struct knote *kn) 300{ 301 302 return (fo_kqfilter(kn->kn_fp, kn)); 303} 304 305/*ARGSUSED*/ 306static int 307kqueue_kqfilter(struct file *fp, struct knote *kn) 308{ 309 struct kqueue *kq = kn->kn_fp->f_data; 310 311 if (kn->kn_filter != EVFILT_READ) 312 return (EINVAL); 313 314 kn->kn_status |= KN_KQUEUE; 315 kn->kn_fop = &kqread_filtops; 316 knlist_add(&kq->kq_sel.si_note, kn, 0); 317 318 return (0); 319} 320 321static void 322filt_kqdetach(struct knote *kn) 323{ 324 struct kqueue *kq = kn->kn_fp->f_data; 325 326 knlist_remove(&kq->kq_sel.si_note, kn, 0); 327} 328 329/*ARGSUSED*/ 330static int 331filt_kqueue(struct knote *kn, long hint) 332{ 333 struct kqueue *kq = kn->kn_fp->f_data; 334 335 kn->kn_data = kq->kq_count; 336 return (kn->kn_data > 0); 337} 338 339/* XXX - move to kern_proc.c? */ 340static int 341filt_procattach(struct knote *kn) 342{ 343 struct proc *p; 344 int immediate; 345 int error; 346 347 immediate = 0; 348 p = pfind(kn->kn_id); 349 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 350 p = zpfind(kn->kn_id); 351 immediate = 1; 352 } else if (p != NULL && (p->p_flag & P_WEXIT)) { 353 immediate = 1; 354 } 355 356 if (p == NULL) 357 return (ESRCH); 358 if ((error = p_cansee(curthread, p))) { 359 PROC_UNLOCK(p); 360 return (error); 361 } 362 363 kn->kn_ptr.p_proc = p; 364 kn->kn_flags |= EV_CLEAR; /* automatically set */ 365 366 /* 367 * internal flag indicating registration done by kernel 368 */ 369 if (kn->kn_flags & EV_FLAG1) { 370 kn->kn_data = kn->kn_sdata; /* ppid */ 371 kn->kn_fflags = NOTE_CHILD; 372 kn->kn_flags &= ~EV_FLAG1; 373 } 374 375 if (immediate == 0) 376 knlist_add(&p->p_klist, kn, 1); 377 378 /* 379 * Immediately activate any exit notes if the target process is a 380 * zombie. This is necessary to handle the case where the target 381 * process, e.g. a child, dies before the kevent is registered. 382 */ 383 if (immediate && filt_proc(kn, NOTE_EXIT)) 384 KNOTE_ACTIVATE(kn, 0); 385 386 PROC_UNLOCK(p); 387 388 return (0); 389} 390 391/* 392 * The knote may be attached to a different process, which may exit, 393 * leaving nothing for the knote to be attached to. So when the process 394 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 395 * it will be deleted when read out. However, as part of the knote deletion, 396 * this routine is called, so a check is needed to avoid actually performing 397 * a detach, because the original process does not exist any more. 398 */ 399/* XXX - move to kern_proc.c? */ 400static void 401filt_procdetach(struct knote *kn) 402{ 403 struct proc *p; 404 405 p = kn->kn_ptr.p_proc; 406 knlist_remove(&p->p_klist, kn, 0); 407 kn->kn_ptr.p_proc = NULL; 408} 409 410/* XXX - move to kern_proc.c? */ 411static int 412filt_proc(struct knote *kn, long hint) 413{ 414 struct proc *p = kn->kn_ptr.p_proc; 415 u_int event; 416 417 /* 418 * mask off extra data 419 */ 420 event = (u_int)hint & NOTE_PCTRLMASK; 421 422 /* 423 * if the user is interested in this event, record it. 424 */ 425 if (kn->kn_sfflags & event) 426 kn->kn_fflags |= event; 427 428 /* 429 * process is gone, so flag the event as finished. 430 */ 431 if (event == NOTE_EXIT) { 432 if (!(kn->kn_status & KN_DETACHED)) 433 knlist_remove_inevent(&p->p_klist, kn); 434 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 435 kn->kn_ptr.p_proc = NULL; 436 if (kn->kn_fflags & NOTE_EXIT) 437 kn->kn_data = p->p_xstat; 438 if (kn->kn_fflags == 0) 439 kn->kn_flags |= EV_DROP; 440 return (1); 441 } 442 443 return (kn->kn_fflags != 0); 444} 445 446/* 447 * Called when the process forked. It mostly does the same as the 448 * knote(), activating all knotes registered to be activated when the 449 * process forked. Additionally, for each knote attached to the 450 * parent, check whether user wants to track the new process. If so 451 * attach a new knote to it, and immediately report an event with the 452 * child's pid. 453 */ 454void 455knote_fork(struct knlist *list, int pid) 456{ 457 struct kqueue *kq; 458 struct knote *kn; 459 struct kevent kev; 460 int error; 461 462 if (list == NULL) 463 return; 464 list->kl_lock(list->kl_lockarg); 465 466 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 467 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) 468 continue; 469 kq = kn->kn_kq; 470 KQ_LOCK(kq); 471 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { 472 KQ_UNLOCK(kq); 473 continue; 474 } 475 476 /* 477 * The same as knote(), activate the event. 478 */ 479 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 480 kn->kn_status |= KN_HASKQLOCK; 481 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 482 KNOTE_ACTIVATE(kn, 1); 483 kn->kn_status &= ~KN_HASKQLOCK; 484 KQ_UNLOCK(kq); 485 continue; 486 } 487 488 /* 489 * The NOTE_TRACK case. In addition to the activation 490 * of the event, we need to register new event to 491 * track the child. Drop the locks in preparation for 492 * the call to kqueue_register(). 493 */ 494 kn->kn_status |= KN_INFLUX; 495 KQ_UNLOCK(kq); 496 list->kl_unlock(list->kl_lockarg); 497 498 /* 499 * Activate existing knote and register a knote with 500 * new process. 501 */ 502 kev.ident = pid; 503 kev.filter = kn->kn_filter; 504 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 505 kev.fflags = kn->kn_sfflags; 506 kev.data = kn->kn_id; /* parent */ 507 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 508 error = kqueue_register(kq, &kev, NULL, 0); 509 if (error) 510 kn->kn_fflags |= NOTE_TRACKERR; 511 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 512 KNOTE_ACTIVATE(kn, 0); 513 KQ_LOCK(kq); 514 kn->kn_status &= ~KN_INFLUX; 515 KQ_UNLOCK_FLUX(kq); 516 list->kl_lock(list->kl_lockarg); 517 } 518 list->kl_unlock(list->kl_lockarg); 519} 520 521/* 522 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the 523 * interval timer support code. 524 */ 525 526#define NOTE_TIMER_PRECMASK (NOTE_SECONDS|NOTE_MSECONDS|NOTE_USECONDS| \ 527 NOTE_NSECONDS) 528 529static __inline sbintime_t 530timer2sbintime(intptr_t data, int flags) 531{ 532 sbintime_t modifier; 533 534 switch (flags & NOTE_TIMER_PRECMASK) { 535 case NOTE_SECONDS: 536 modifier = SBT_1S; 537 break; 538 case NOTE_MSECONDS: /* FALLTHROUGH */ 539 case 0: 540 modifier = SBT_1MS; 541 break; 542 case NOTE_USECONDS: 543 modifier = SBT_1US; 544 break; 545 case NOTE_NSECONDS: 546 modifier = SBT_1NS; 547 break; 548 default: 549 return (-1); 550 } 551 552#ifdef __LP64__ 553 if (data > SBT_MAX / modifier) 554 return (SBT_MAX); 555#endif 556 return (modifier * data); 557} 558 559static void 560filt_timerexpire(void *knx) 561{ 562 struct callout *calloutp; 563 struct knote *kn; 564 565 kn = knx; 566 kn->kn_data++; 567 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 568 569 if ((kn->kn_flags & EV_ONESHOT) != EV_ONESHOT) { 570 calloutp = (struct callout *)kn->kn_hook; 571 *kn->kn_ptr.p_nexttime += timer2sbintime(kn->kn_sdata, 572 kn->kn_sfflags); 573 callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0, 574 filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); 575 } 576} 577 578/* 579 * data contains amount of time to sleep 580 */ 581static int 582filt_timerattach(struct knote *kn) 583{ 584 struct callout *calloutp; 585 sbintime_t to; 586 unsigned int ncallouts; 587 588 if ((intptr_t)kn->kn_sdata < 0) 589 return (EINVAL); 590 if ((intptr_t)kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) 591 kn->kn_sdata = 1; 592 /* Only precision unit are supported in flags so far */ 593 if (kn->kn_sfflags & ~NOTE_TIMER_PRECMASK) 594 return (EINVAL); 595 596 to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); 597 if (to < 0) 598 return (EINVAL); 599 600 ncallouts = atomic_load_explicit(&kq_ncallouts, memory_order_relaxed); 601 do { 602 if (ncallouts >= kq_calloutmax) 603 return (ENOMEM); 604 } while (!atomic_compare_exchange_weak_explicit(&kq_ncallouts, 605 &ncallouts, ncallouts + 1, memory_order_relaxed, 606 memory_order_relaxed)); 607 608 kn->kn_flags |= EV_CLEAR; /* automatically set */ 609 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ 610 kn->kn_ptr.p_nexttime = malloc(sizeof(sbintime_t), M_KQUEUE, M_WAITOK); 611 calloutp = malloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK); 612 callout_init(calloutp, CALLOUT_MPSAFE); 613 kn->kn_hook = calloutp; 614 *kn->kn_ptr.p_nexttime = to + sbinuptime(); 615 callout_reset_sbt_on(calloutp, *kn->kn_ptr.p_nexttime, 0, 616 filt_timerexpire, kn, PCPU_GET(cpuid), C_ABSOLUTE); 617 618 return (0); 619} 620 621static void 622filt_timerdetach(struct knote *kn) 623{ 624 struct callout *calloutp; 625 unsigned int old; 626 627 calloutp = (struct callout *)kn->kn_hook; 628 callout_drain(calloutp); 629 free(calloutp, M_KQUEUE); 630 free(kn->kn_ptr.p_nexttime, M_KQUEUE); 631 old = atomic_fetch_sub_explicit(&kq_ncallouts, 1, memory_order_relaxed); 632 KASSERT(old > 0, ("Number of callouts cannot become negative")); 633 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ 634} 635 636static int 637filt_timer(struct knote *kn, long hint) 638{ 639 640 return (kn->kn_data != 0); 641} 642 643static int 644filt_userattach(struct knote *kn) 645{ 646 647 /* 648 * EVFILT_USER knotes are not attached to anything in the kernel. 649 */ 650 kn->kn_hook = NULL; 651 if (kn->kn_fflags & NOTE_TRIGGER) 652 kn->kn_hookid = 1; 653 else 654 kn->kn_hookid = 0; 655 return (0); 656} 657 658static void 659filt_userdetach(__unused struct knote *kn) 660{ 661 662 /* 663 * EVFILT_USER knotes are not attached to anything in the kernel. 664 */ 665} 666 667static int 668filt_user(struct knote *kn, __unused long hint) 669{ 670 671 return (kn->kn_hookid); 672} 673 674static void 675filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 676{ 677 u_int ffctrl; 678 679 switch (type) { 680 case EVENT_REGISTER: 681 if (kev->fflags & NOTE_TRIGGER) 682 kn->kn_hookid = 1; 683 684 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 685 kev->fflags &= NOTE_FFLAGSMASK; 686 switch (ffctrl) { 687 case NOTE_FFNOP: 688 break; 689 690 case NOTE_FFAND: 691 kn->kn_sfflags &= kev->fflags; 692 break; 693 694 case NOTE_FFOR: 695 kn->kn_sfflags |= kev->fflags; 696 break; 697 698 case NOTE_FFCOPY: 699 kn->kn_sfflags = kev->fflags; 700 break; 701 702 default: 703 /* XXX Return error? */ 704 break; 705 } 706 kn->kn_sdata = kev->data; 707 if (kev->flags & EV_CLEAR) { 708 kn->kn_hookid = 0; 709 kn->kn_data = 0; 710 kn->kn_fflags = 0; 711 } 712 break; 713 714 case EVENT_PROCESS: 715 *kev = kn->kn_kevent; 716 kev->fflags = kn->kn_sfflags; 717 kev->data = kn->kn_sdata; 718 if (kn->kn_flags & EV_CLEAR) { 719 kn->kn_hookid = 0; 720 kn->kn_data = 0; 721 kn->kn_fflags = 0; 722 } 723 break; 724 725 default: 726 panic("filt_usertouch() - invalid type (%ld)", type); 727 break; 728 } 729} 730 731int 732sys_kqueue(struct thread *td, struct kqueue_args *uap) 733{ 734 struct filedesc *fdp; 735 struct kqueue *kq; 736 struct file *fp; 737 int fd, error; 738 739 fdp = td->td_proc->p_fd; 740 error = falloc(td, &fp, &fd, 0); 741 if (error) 742 goto done2; 743 744 /* An extra reference on `fp' has been held for us by falloc(). */ 745 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 746 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF|MTX_DUPOK); 747 TAILQ_INIT(&kq->kq_head); 748 kq->kq_fdp = fdp; 749 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 750 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 751 752 FILEDESC_XLOCK(fdp); 753 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 754 FILEDESC_XUNLOCK(fdp); 755 756 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 757 fdrop(fp, td); 758 759 td->td_retval[0] = fd; 760done2: 761 return (error); 762} 763 764#ifndef _SYS_SYSPROTO_H_ 765struct kevent_args { 766 int fd; 767 const struct kevent *changelist; 768 int nchanges; 769 struct kevent *eventlist; 770 int nevents; 771 const struct timespec *timeout; 772}; 773#endif 774int 775sys_kevent(struct thread *td, struct kevent_args *uap) 776{ 777 struct timespec ts, *tsp; 778 struct kevent_copyops k_ops = { uap, 779 kevent_copyout, 780 kevent_copyin}; 781 int error; 782#ifdef KTRACE 783 struct uio ktruio; 784 struct iovec ktriov; 785 struct uio *ktruioin = NULL; 786 struct uio *ktruioout = NULL; 787#endif 788 789 if (uap->timeout != NULL) { 790 error = copyin(uap->timeout, &ts, sizeof(ts)); 791 if (error) 792 return (error); 793 tsp = &ts; 794 } else 795 tsp = NULL; 796 797#ifdef KTRACE 798 if (KTRPOINT(td, KTR_GENIO)) { 799 ktriov.iov_base = uap->changelist; 800 ktriov.iov_len = uap->nchanges * sizeof(struct kevent); 801 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1, 802 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ, 803 .uio_td = td }; 804 ktruioin = cloneuio(&ktruio); 805 ktriov.iov_base = uap->eventlist; 806 ktriov.iov_len = uap->nevents * sizeof(struct kevent); 807 ktruioout = cloneuio(&ktruio); 808 } 809#endif 810 811 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 812 &k_ops, tsp); 813 814#ifdef KTRACE 815 if (ktruioin != NULL) { 816 ktruioin->uio_resid = uap->nchanges * sizeof(struct kevent); 817 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0); 818 ktruioout->uio_resid = td->td_retval[0] * sizeof(struct kevent); 819 ktrgenio(uap->fd, UIO_READ, ktruioout, error); 820 } 821#endif 822 823 return (error); 824} 825 826/* 827 * Copy 'count' items into the destination list pointed to by uap->eventlist. 828 */ 829static int 830kevent_copyout(void *arg, struct kevent *kevp, int count) 831{ 832 struct kevent_args *uap; 833 int error; 834 835 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 836 uap = (struct kevent_args *)arg; 837 838 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 839 if (error == 0) 840 uap->eventlist += count; 841 return (error); 842} 843 844/* 845 * Copy 'count' items from the list pointed to by uap->changelist. 846 */ 847static int 848kevent_copyin(void *arg, struct kevent *kevp, int count) 849{ 850 struct kevent_args *uap; 851 int error; 852 853 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 854 uap = (struct kevent_args *)arg; 855 856 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 857 if (error == 0) 858 uap->changelist += count; 859 return (error); 860} 861 862int 863kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 864 struct kevent_copyops *k_ops, const struct timespec *timeout) 865{ 866 struct kevent keva[KQ_NEVENTS]; 867 struct kevent *kevp, *changes; 868 struct kqueue *kq; 869 struct file *fp; 870 cap_rights_t rights; 871 int i, n, nerrors, error; 872 873 cap_rights_init(&rights); 874 if (nchanges > 0) 875 cap_rights_set(&rights, CAP_KQUEUE_CHANGE); 876 if (nevents > 0) 877 cap_rights_set(&rights, CAP_KQUEUE_EVENT); 878 error = fget(td, fd, &rights, &fp); 879 if (error != 0) 880 return (error); 881 882 error = kqueue_acquire(fp, &kq); 883 if (error != 0) 884 goto done_norel; 885 886 nerrors = 0; 887 888 while (nchanges > 0) { 889 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 890 error = k_ops->k_copyin(k_ops->arg, keva, n); 891 if (error) 892 goto done; 893 changes = keva; 894 for (i = 0; i < n; i++) { 895 kevp = &changes[i]; 896 if (!kevp->filter) 897 continue; 898 kevp->flags &= ~EV_SYSFLAGS; 899 error = kqueue_register(kq, kevp, td, 1); 900 if (error || (kevp->flags & EV_RECEIPT)) { 901 if (nevents != 0) { 902 kevp->flags = EV_ERROR; 903 kevp->data = error; 904 (void) k_ops->k_copyout(k_ops->arg, 905 kevp, 1); 906 nevents--; 907 nerrors++; 908 } else { 909 goto done; 910 } 911 } 912 } 913 nchanges -= n; 914 } 915 if (nerrors) { 916 td->td_retval[0] = nerrors; 917 error = 0; 918 goto done; 919 } 920 921 error = kqueue_scan(kq, nevents, k_ops, timeout, keva, td); 922done: 923 kqueue_release(kq, 0); 924done_norel: 925 fdrop(fp, td); 926 return (error); 927} 928 929int 930kqueue_add_filteropts(int filt, struct filterops *filtops) 931{ 932 int error; 933 934 error = 0; 935 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 936 printf( 937"trying to add a filterop that is out of range: %d is beyond %d\n", 938 ~filt, EVFILT_SYSCOUNT); 939 return EINVAL; 940 } 941 mtx_lock(&filterops_lock); 942 if (sysfilt_ops[~filt].for_fop != &null_filtops && 943 sysfilt_ops[~filt].for_fop != NULL) 944 error = EEXIST; 945 else { 946 sysfilt_ops[~filt].for_fop = filtops; 947 sysfilt_ops[~filt].for_refcnt = 0; 948 } 949 mtx_unlock(&filterops_lock); 950 951 return (error); 952} 953 954int 955kqueue_del_filteropts(int filt) 956{ 957 int error; 958 959 error = 0; 960 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 961 return EINVAL; 962 963 mtx_lock(&filterops_lock); 964 if (sysfilt_ops[~filt].for_fop == &null_filtops || 965 sysfilt_ops[~filt].for_fop == NULL) 966 error = EINVAL; 967 else if (sysfilt_ops[~filt].for_refcnt != 0) 968 error = EBUSY; 969 else { 970 sysfilt_ops[~filt].for_fop = &null_filtops; 971 sysfilt_ops[~filt].for_refcnt = 0; 972 } 973 mtx_unlock(&filterops_lock); 974 975 return error; 976} 977 978static struct filterops * 979kqueue_fo_find(int filt) 980{ 981 982 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 983 return NULL; 984 985 mtx_lock(&filterops_lock); 986 sysfilt_ops[~filt].for_refcnt++; 987 if (sysfilt_ops[~filt].for_fop == NULL) 988 sysfilt_ops[~filt].for_fop = &null_filtops; 989 mtx_unlock(&filterops_lock); 990 991 return sysfilt_ops[~filt].for_fop; 992} 993 994static void 995kqueue_fo_release(int filt) 996{ 997 998 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 999 return; 1000 1001 mtx_lock(&filterops_lock); 1002 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 1003 ("filter object refcount not valid on release")); 1004 sysfilt_ops[~filt].for_refcnt--; 1005 mtx_unlock(&filterops_lock); 1006} 1007 1008/* 1009 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will 1010 * influence if memory allocation should wait. Make sure it is 0 if you 1011 * hold any mutexes. 1012 */ 1013static int 1014kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok) 1015{ 1016 struct filterops *fops; 1017 struct file *fp; 1018 struct knote *kn, *tkn; 1019 cap_rights_t rights; 1020 int error, filt, event; 1021 int haskqglobal, filedesc_unlock; 1022 1023 fp = NULL; 1024 kn = NULL; 1025 error = 0; 1026 haskqglobal = 0; 1027 filedesc_unlock = 0; 1028 1029 filt = kev->filter; 1030 fops = kqueue_fo_find(filt); 1031 if (fops == NULL) 1032 return EINVAL; 1033 1034 tkn = knote_alloc(waitok); /* prevent waiting with locks */ 1035 1036findkn: 1037 if (fops->f_isfd) { 1038 KASSERT(td != NULL, ("td is NULL")); 1039 error = fget(td, kev->ident, 1040 cap_rights_init(&rights, CAP_EVENT), &fp); 1041 if (error) 1042 goto done; 1043 1044 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 1045 kev->ident, 0) != 0) { 1046 /* try again */ 1047 fdrop(fp, td); 1048 fp = NULL; 1049 error = kqueue_expand(kq, fops, kev->ident, waitok); 1050 if (error) 1051 goto done; 1052 goto findkn; 1053 } 1054 1055 if (fp->f_type == DTYPE_KQUEUE) { 1056 /* 1057 * if we add some inteligence about what we are doing, 1058 * we should be able to support events on ourselves. 1059 * We need to know when we are doing this to prevent 1060 * getting both the knlist lock and the kq lock since 1061 * they are the same thing. 1062 */ 1063 if (fp->f_data == kq) { 1064 error = EINVAL; 1065 goto done; 1066 } 1067 1068 /* 1069 * Pre-lock the filedesc before the global 1070 * lock mutex, see the comment in 1071 * kqueue_close(). 1072 */ 1073 FILEDESC_XLOCK(td->td_proc->p_fd); 1074 filedesc_unlock = 1; 1075 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1076 } 1077 1078 KQ_LOCK(kq); 1079 if (kev->ident < kq->kq_knlistsize) { 1080 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1081 if (kev->filter == kn->kn_filter) 1082 break; 1083 } 1084 } else { 1085 if ((kev->flags & EV_ADD) == EV_ADD) 1086 kqueue_expand(kq, fops, kev->ident, waitok); 1087 1088 KQ_LOCK(kq); 1089 if (kq->kq_knhashmask != 0) { 1090 struct klist *list; 1091 1092 list = &kq->kq_knhash[ 1093 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1094 SLIST_FOREACH(kn, list, kn_link) 1095 if (kev->ident == kn->kn_id && 1096 kev->filter == kn->kn_filter) 1097 break; 1098 } 1099 } 1100 1101 /* knote is in the process of changing, wait for it to stablize. */ 1102 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1103 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1104 if (filedesc_unlock) { 1105 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1106 filedesc_unlock = 0; 1107 } 1108 kq->kq_state |= KQ_FLUXWAIT; 1109 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1110 if (fp != NULL) { 1111 fdrop(fp, td); 1112 fp = NULL; 1113 } 1114 goto findkn; 1115 } 1116 1117 /* 1118 * kn now contains the matching knote, or NULL if no match 1119 */ 1120 if (kn == NULL) { 1121 if (kev->flags & EV_ADD) { 1122 kn = tkn; 1123 tkn = NULL; 1124 if (kn == NULL) { 1125 KQ_UNLOCK(kq); 1126 error = ENOMEM; 1127 goto done; 1128 } 1129 kn->kn_fp = fp; 1130 kn->kn_kq = kq; 1131 kn->kn_fop = fops; 1132 /* 1133 * apply reference counts to knote structure, and 1134 * do not release it at the end of this routine. 1135 */ 1136 fops = NULL; 1137 fp = NULL; 1138 1139 kn->kn_sfflags = kev->fflags; 1140 kn->kn_sdata = kev->data; 1141 kev->fflags = 0; 1142 kev->data = 0; 1143 kn->kn_kevent = *kev; 1144 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1145 EV_ENABLE | EV_DISABLE); 1146 kn->kn_status = KN_INFLUX|KN_DETACHED; 1147 1148 error = knote_attach(kn, kq); 1149 KQ_UNLOCK(kq); 1150 if (error != 0) { 1151 tkn = kn; 1152 goto done; 1153 } 1154 1155 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1156 knote_drop(kn, td); 1157 goto done; 1158 } 1159 KN_LIST_LOCK(kn); 1160 goto done_ev_add; 1161 } else { 1162 /* No matching knote and the EV_ADD flag is not set. */ 1163 KQ_UNLOCK(kq); 1164 error = ENOENT; 1165 goto done; 1166 } 1167 } 1168 1169 if (kev->flags & EV_DELETE) { 1170 kn->kn_status |= KN_INFLUX; 1171 KQ_UNLOCK(kq); 1172 if (!(kn->kn_status & KN_DETACHED)) 1173 kn->kn_fop->f_detach(kn); 1174 knote_drop(kn, td); 1175 goto done; 1176 } 1177 1178 /* 1179 * The user may change some filter values after the initial EV_ADD, 1180 * but doing so will not reset any filter which has already been 1181 * triggered. 1182 */ 1183 kn->kn_status |= KN_INFLUX | KN_SCAN; 1184 KQ_UNLOCK(kq); 1185 KN_LIST_LOCK(kn); 1186 kn->kn_kevent.udata = kev->udata; 1187 if (!fops->f_isfd && fops->f_touch != NULL) { 1188 fops->f_touch(kn, kev, EVENT_REGISTER); 1189 } else { 1190 kn->kn_sfflags = kev->fflags; 1191 kn->kn_sdata = kev->data; 1192 } 1193 1194 /* 1195 * We can get here with kn->kn_knlist == NULL. This can happen when 1196 * the initial attach event decides that the event is "completed" 1197 * already. i.e. filt_procattach is called on a zombie process. It 1198 * will call filt_proc which will remove it from the list, and NULL 1199 * kn_knlist. 1200 */ 1201done_ev_add: 1202 event = kn->kn_fop->f_event(kn, 0); 1203 KQ_LOCK(kq); 1204 if (event) 1205 KNOTE_ACTIVATE(kn, 1); 1206 kn->kn_status &= ~(KN_INFLUX | KN_SCAN); 1207 KN_LIST_UNLOCK(kn); 1208 1209 if ((kev->flags & EV_DISABLE) && 1210 ((kn->kn_status & KN_DISABLED) == 0)) { 1211 kn->kn_status |= KN_DISABLED; 1212 } 1213 1214 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 1215 kn->kn_status &= ~KN_DISABLED; 1216 if ((kn->kn_status & KN_ACTIVE) && 1217 ((kn->kn_status & KN_QUEUED) == 0)) 1218 knote_enqueue(kn); 1219 } 1220 KQ_UNLOCK_FLUX(kq); 1221 1222done: 1223 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1224 if (filedesc_unlock) 1225 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1226 if (fp != NULL) 1227 fdrop(fp, td); 1228 if (tkn != NULL) 1229 knote_free(tkn); 1230 if (fops != NULL) 1231 kqueue_fo_release(filt); 1232 return (error); 1233} 1234 1235static int 1236kqueue_acquire(struct file *fp, struct kqueue **kqp) 1237{ 1238 int error; 1239 struct kqueue *kq; 1240 1241 error = 0; 1242 1243 kq = fp->f_data; 1244 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1245 return (EBADF); 1246 *kqp = kq; 1247 KQ_LOCK(kq); 1248 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1249 KQ_UNLOCK(kq); 1250 return (EBADF); 1251 } 1252 kq->kq_refcnt++; 1253 KQ_UNLOCK(kq); 1254 1255 return error; 1256} 1257 1258static void 1259kqueue_release(struct kqueue *kq, int locked) 1260{ 1261 if (locked) 1262 KQ_OWNED(kq); 1263 else 1264 KQ_LOCK(kq); 1265 kq->kq_refcnt--; 1266 if (kq->kq_refcnt == 1) 1267 wakeup(&kq->kq_refcnt); 1268 if (!locked) 1269 KQ_UNLOCK(kq); 1270} 1271 1272static void 1273kqueue_schedtask(struct kqueue *kq) 1274{ 1275 1276 KQ_OWNED(kq); 1277 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1278 ("scheduling kqueue task while draining")); 1279 1280 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1281 taskqueue_enqueue(taskqueue_kqueue, &kq->kq_task); 1282 kq->kq_state |= KQ_TASKSCHED; 1283 } 1284} 1285 1286/* 1287 * Expand the kq to make sure we have storage for fops/ident pair. 1288 * 1289 * Return 0 on success (or no work necessary), return errno on failure. 1290 * 1291 * Not calling hashinit w/ waitok (proper malloc flag) should be safe. 1292 * If kqueue_register is called from a non-fd context, there usually/should 1293 * be no locks held. 1294 */ 1295static int 1296kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, 1297 int waitok) 1298{ 1299 struct klist *list, *tmp_knhash, *to_free; 1300 u_long tmp_knhashmask; 1301 int size; 1302 int fd; 1303 int mflag = waitok ? M_WAITOK : M_NOWAIT; 1304 1305 KQ_NOTOWNED(kq); 1306 1307 to_free = NULL; 1308 if (fops->f_isfd) { 1309 fd = ident; 1310 if (kq->kq_knlistsize <= fd) { 1311 size = kq->kq_knlistsize; 1312 while (size <= fd) 1313 size += KQEXTENT; 1314 list = malloc(size * sizeof(*list), M_KQUEUE, mflag); 1315 if (list == NULL) 1316 return ENOMEM; 1317 KQ_LOCK(kq); 1318 if (kq->kq_knlistsize > fd) { 1319 to_free = list; 1320 list = NULL; 1321 } else { 1322 if (kq->kq_knlist != NULL) { 1323 bcopy(kq->kq_knlist, list, 1324 kq->kq_knlistsize * sizeof(*list)); 1325 to_free = kq->kq_knlist; 1326 kq->kq_knlist = NULL; 1327 } 1328 bzero((caddr_t)list + 1329 kq->kq_knlistsize * sizeof(*list), 1330 (size - kq->kq_knlistsize) * sizeof(*list)); 1331 kq->kq_knlistsize = size; 1332 kq->kq_knlist = list; 1333 } 1334 KQ_UNLOCK(kq); 1335 } 1336 } else { 1337 if (kq->kq_knhashmask == 0) { 1338 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1339 &tmp_knhashmask); 1340 if (tmp_knhash == NULL) 1341 return ENOMEM; 1342 KQ_LOCK(kq); 1343 if (kq->kq_knhashmask == 0) { 1344 kq->kq_knhash = tmp_knhash; 1345 kq->kq_knhashmask = tmp_knhashmask; 1346 } else { 1347 to_free = tmp_knhash; 1348 } 1349 KQ_UNLOCK(kq); 1350 } 1351 } 1352 free(to_free, M_KQUEUE); 1353 1354 KQ_NOTOWNED(kq); 1355 return 0; 1356} 1357 1358static void 1359kqueue_task(void *arg, int pending) 1360{ 1361 struct kqueue *kq; 1362 int haskqglobal; 1363 1364 haskqglobal = 0; 1365 kq = arg; 1366 1367 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1368 KQ_LOCK(kq); 1369 1370 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 1371 1372 kq->kq_state &= ~KQ_TASKSCHED; 1373 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 1374 wakeup(&kq->kq_state); 1375 } 1376 KQ_UNLOCK(kq); 1377 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1378} 1379 1380/* 1381 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 1382 * We treat KN_MARKER knotes as if they are INFLUX. 1383 */ 1384static int 1385kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 1386 const struct timespec *tsp, struct kevent *keva, struct thread *td) 1387{ 1388 struct kevent *kevp; 1389 struct knote *kn, *marker; 1390 sbintime_t asbt, rsbt; 1391 int count, error, haskqglobal, influx, nkev, touch; 1392 1393 count = maxevents; 1394 nkev = 0; 1395 error = 0; 1396 haskqglobal = 0; 1397 1398 if (maxevents == 0) 1399 goto done_nl; 1400 1401 rsbt = 0; 1402 if (tsp != NULL) { 1403 if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 || 1404 tsp->tv_nsec >= 1000000000) { 1405 error = EINVAL; 1406 goto done_nl; 1407 } 1408 if (timespecisset(tsp)) { 1409 if (tsp->tv_sec <= INT32_MAX) { 1410 rsbt = tstosbt(*tsp); 1411 if (TIMESEL(&asbt, rsbt)) 1412 asbt += tc_tick_sbt; 1413 if (asbt <= INT64_MAX - rsbt) 1414 asbt += rsbt; 1415 else 1416 asbt = 0; 1417 rsbt >>= tc_precexp; 1418 } else 1419 asbt = 0; 1420 } else 1421 asbt = -1; 1422 } else 1423 asbt = 0; 1424 marker = knote_alloc(1); 1425 if (marker == NULL) { 1426 error = ENOMEM; 1427 goto done_nl; 1428 } 1429 marker->kn_status = KN_MARKER; 1430 KQ_LOCK(kq); 1431 1432retry: 1433 kevp = keva; 1434 if (kq->kq_count == 0) { 1435 if (asbt == -1) { 1436 error = EWOULDBLOCK; 1437 } else { 1438 kq->kq_state |= KQ_SLEEP; 1439 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, 1440 "kqread", asbt, rsbt, C_ABSOLUTE); 1441 } 1442 if (error == 0) 1443 goto retry; 1444 /* don't restart after signals... */ 1445 if (error == ERESTART) 1446 error = EINTR; 1447 else if (error == EWOULDBLOCK) 1448 error = 0; 1449 goto done; 1450 } 1451 1452 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1453 influx = 0; 1454 while (count) { 1455 KQ_OWNED(kq); 1456 kn = TAILQ_FIRST(&kq->kq_head); 1457 1458 if ((kn->kn_status == KN_MARKER && kn != marker) || 1459 (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1460 if (influx) { 1461 influx = 0; 1462 KQ_FLUX_WAKEUP(kq); 1463 } 1464 kq->kq_state |= KQ_FLUXWAIT; 1465 error = msleep(kq, &kq->kq_lock, PSOCK, 1466 "kqflxwt", 0); 1467 continue; 1468 } 1469 1470 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1471 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 1472 kn->kn_status &= ~KN_QUEUED; 1473 kq->kq_count--; 1474 continue; 1475 } 1476 if (kn == marker) { 1477 KQ_FLUX_WAKEUP(kq); 1478 if (count == maxevents) 1479 goto retry; 1480 goto done; 1481 } 1482 KASSERT((kn->kn_status & KN_INFLUX) == 0, 1483 ("KN_INFLUX set when not suppose to be")); 1484 1485 if ((kn->kn_flags & EV_DROP) == EV_DROP) { 1486 kn->kn_status &= ~KN_QUEUED; 1487 kn->kn_status |= KN_INFLUX; 1488 kq->kq_count--; 1489 KQ_UNLOCK(kq); 1490 /* 1491 * We don't need to lock the list since we've marked 1492 * it _INFLUX. 1493 */ 1494 if (!(kn->kn_status & KN_DETACHED)) 1495 kn->kn_fop->f_detach(kn); 1496 knote_drop(kn, td); 1497 KQ_LOCK(kq); 1498 continue; 1499 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 1500 kn->kn_status &= ~KN_QUEUED; 1501 kn->kn_status |= KN_INFLUX; 1502 kq->kq_count--; 1503 KQ_UNLOCK(kq); 1504 /* 1505 * We don't need to lock the list since we've marked 1506 * it _INFLUX. 1507 */ 1508 *kevp = kn->kn_kevent; 1509 if (!(kn->kn_status & KN_DETACHED)) 1510 kn->kn_fop->f_detach(kn); 1511 knote_drop(kn, td); 1512 KQ_LOCK(kq); 1513 kn = NULL; 1514 } else { 1515 kn->kn_status |= KN_INFLUX | KN_SCAN; 1516 KQ_UNLOCK(kq); 1517 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 1518 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1519 KN_LIST_LOCK(kn); 1520 if (kn->kn_fop->f_event(kn, 0) == 0) { 1521 KQ_LOCK(kq); 1522 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1523 kn->kn_status &= 1524 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX | 1525 KN_SCAN); 1526 kq->kq_count--; 1527 KN_LIST_UNLOCK(kn); 1528 influx = 1; 1529 continue; 1530 } 1531 touch = (!kn->kn_fop->f_isfd && 1532 kn->kn_fop->f_touch != NULL); 1533 if (touch) 1534 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 1535 else 1536 *kevp = kn->kn_kevent; 1537 KQ_LOCK(kq); 1538 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1539 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 1540 /* 1541 * Manually clear knotes who weren't 1542 * 'touch'ed. 1543 */ 1544 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 1545 kn->kn_data = 0; 1546 kn->kn_fflags = 0; 1547 } 1548 if (kn->kn_flags & EV_DISPATCH) 1549 kn->kn_status |= KN_DISABLED; 1550 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1551 kq->kq_count--; 1552 } else 1553 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1554 1555 kn->kn_status &= ~(KN_INFLUX | KN_SCAN); 1556 KN_LIST_UNLOCK(kn); 1557 influx = 1; 1558 } 1559 1560 /* we are returning a copy to the user */ 1561 kevp++; 1562 nkev++; 1563 count--; 1564 1565 if (nkev == KQ_NEVENTS) { 1566 influx = 0; 1567 KQ_UNLOCK_FLUX(kq); 1568 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1569 nkev = 0; 1570 kevp = keva; 1571 KQ_LOCK(kq); 1572 if (error) 1573 break; 1574 } 1575 } 1576 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1577done: 1578 KQ_OWNED(kq); 1579 KQ_UNLOCK_FLUX(kq); 1580 knote_free(marker); 1581done_nl: 1582 KQ_NOTOWNED(kq); 1583 if (nkev != 0) 1584 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1585 td->td_retval[0] = maxevents - count; 1586 return (error); 1587} 1588 1589/* 1590 * XXX 1591 * This could be expanded to call kqueue_scan, if desired. 1592 */ 1593/*ARGSUSED*/ 1594static int 1595kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 1596 int flags, struct thread *td) 1597{ 1598 return (ENXIO); 1599} 1600 1601/*ARGSUSED*/ 1602static int 1603kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 1604 int flags, struct thread *td) 1605{ 1606 return (ENXIO); 1607} 1608 1609/*ARGSUSED*/ 1610static int 1611kqueue_truncate(struct file *fp, off_t length, struct ucred *active_cred, 1612 struct thread *td) 1613{ 1614 1615 return (EINVAL); 1616} 1617 1618/*ARGSUSED*/ 1619static int 1620kqueue_ioctl(struct file *fp, u_long cmd, void *data, 1621 struct ucred *active_cred, struct thread *td) 1622{ 1623 /* 1624 * Enabling sigio causes two major problems: 1625 * 1) infinite recursion: 1626 * Synopsys: kevent is being used to track signals and have FIOASYNC 1627 * set. On receipt of a signal this will cause a kqueue to recurse 1628 * into itself over and over. Sending the sigio causes the kqueue 1629 * to become ready, which in turn posts sigio again, forever. 1630 * Solution: this can be solved by setting a flag in the kqueue that 1631 * we have a SIGIO in progress. 1632 * 2) locking problems: 1633 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 1634 * us above the proc and pgrp locks. 1635 * Solution: Post a signal using an async mechanism, being sure to 1636 * record a generation count in the delivery so that we do not deliver 1637 * a signal to the wrong process. 1638 * 1639 * Note, these two mechanisms are somewhat mutually exclusive! 1640 */ 1641#if 0 1642 struct kqueue *kq; 1643 1644 kq = fp->f_data; 1645 switch (cmd) { 1646 case FIOASYNC: 1647 if (*(int *)data) { 1648 kq->kq_state |= KQ_ASYNC; 1649 } else { 1650 kq->kq_state &= ~KQ_ASYNC; 1651 } 1652 return (0); 1653 1654 case FIOSETOWN: 1655 return (fsetown(*(int *)data, &kq->kq_sigio)); 1656 1657 case FIOGETOWN: 1658 *(int *)data = fgetown(&kq->kq_sigio); 1659 return (0); 1660 } 1661#endif 1662 1663 return (ENOTTY); 1664} 1665 1666/*ARGSUSED*/ 1667static int 1668kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 1669 struct thread *td) 1670{ 1671 struct kqueue *kq; 1672 int revents = 0; 1673 int error; 1674 1675 if ((error = kqueue_acquire(fp, &kq))) 1676 return POLLERR; 1677 1678 KQ_LOCK(kq); 1679 if (events & (POLLIN | POLLRDNORM)) { 1680 if (kq->kq_count) { 1681 revents |= events & (POLLIN | POLLRDNORM); 1682 } else { 1683 selrecord(td, &kq->kq_sel); 1684 if (SEL_WAITING(&kq->kq_sel)) 1685 kq->kq_state |= KQ_SEL; 1686 } 1687 } 1688 kqueue_release(kq, 1); 1689 KQ_UNLOCK(kq); 1690 return (revents); 1691} 1692 1693/*ARGSUSED*/ 1694static int 1695kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 1696 struct thread *td) 1697{ 1698 1699 bzero((void *)st, sizeof *st); 1700 /* 1701 * We no longer return kq_count because the unlocked value is useless. 1702 * If you spent all this time getting the count, why not spend your 1703 * syscall better by calling kevent? 1704 * 1705 * XXX - This is needed for libc_r. 1706 */ 1707 st->st_mode = S_IFIFO; 1708 return (0); 1709} 1710 1711/*ARGSUSED*/ 1712static int 1713kqueue_close(struct file *fp, struct thread *td) 1714{ 1715 struct kqueue *kq = fp->f_data; 1716 struct filedesc *fdp; 1717 struct knote *kn; 1718 int i; 1719 int error; 1720 int filedesc_unlock; 1721 1722 if ((error = kqueue_acquire(fp, &kq))) 1723 return error; 1724 1725 filedesc_unlock = 0; 1726 KQ_LOCK(kq); 1727 1728 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 1729 ("kqueue already closing")); 1730 kq->kq_state |= KQ_CLOSING; 1731 if (kq->kq_refcnt > 1) 1732 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 1733 1734 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 1735 fdp = kq->kq_fdp; 1736 1737 KASSERT(knlist_empty(&kq->kq_sel.si_note), 1738 ("kqueue's knlist not empty")); 1739 1740 for (i = 0; i < kq->kq_knlistsize; i++) { 1741 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 1742 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1743 kq->kq_state |= KQ_FLUXWAIT; 1744 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 1745 continue; 1746 } 1747 kn->kn_status |= KN_INFLUX; 1748 KQ_UNLOCK(kq); 1749 if (!(kn->kn_status & KN_DETACHED)) 1750 kn->kn_fop->f_detach(kn); 1751 knote_drop(kn, td); 1752 KQ_LOCK(kq); 1753 } 1754 } 1755 if (kq->kq_knhashmask != 0) { 1756 for (i = 0; i <= kq->kq_knhashmask; i++) { 1757 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 1758 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1759 kq->kq_state |= KQ_FLUXWAIT; 1760 msleep(kq, &kq->kq_lock, PSOCK, 1761 "kqclo2", 0); 1762 continue; 1763 } 1764 kn->kn_status |= KN_INFLUX; 1765 KQ_UNLOCK(kq); 1766 if (!(kn->kn_status & KN_DETACHED)) 1767 kn->kn_fop->f_detach(kn); 1768 knote_drop(kn, td); 1769 KQ_LOCK(kq); 1770 } 1771 } 1772 } 1773 1774 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 1775 kq->kq_state |= KQ_TASKDRAIN; 1776 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 1777 } 1778 1779 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1780 selwakeuppri(&kq->kq_sel, PSOCK); 1781 if (!SEL_WAITING(&kq->kq_sel)) 1782 kq->kq_state &= ~KQ_SEL; 1783 } 1784 1785 KQ_UNLOCK(kq); 1786 1787 /* 1788 * We could be called due to the knote_drop() doing fdrop(), 1789 * called from kqueue_register(). In this case the global 1790 * lock is owned, and filedesc sx is locked before, to not 1791 * take the sleepable lock after non-sleepable. 1792 */ 1793 if (!sx_xlocked(FILEDESC_LOCK(fdp))) { 1794 FILEDESC_XLOCK(fdp); 1795 filedesc_unlock = 1; 1796 } else 1797 filedesc_unlock = 0; 1798 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); 1799 if (filedesc_unlock) 1800 FILEDESC_XUNLOCK(fdp); 1801 1802 seldrain(&kq->kq_sel); 1803 knlist_destroy(&kq->kq_sel.si_note); 1804 mtx_destroy(&kq->kq_lock); 1805 kq->kq_fdp = NULL; 1806 1807 if (kq->kq_knhash != NULL) 1808 free(kq->kq_knhash, M_KQUEUE); 1809 if (kq->kq_knlist != NULL) 1810 free(kq->kq_knlist, M_KQUEUE); 1811 1812 funsetown(&kq->kq_sigio); 1813 free(kq, M_KQUEUE); 1814 fp->f_data = NULL; 1815 1816 return (0); 1817} 1818 1819static void 1820kqueue_wakeup(struct kqueue *kq) 1821{ 1822 KQ_OWNED(kq); 1823 1824 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 1825 kq->kq_state &= ~KQ_SLEEP; 1826 wakeup(kq); 1827 } 1828 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1829 selwakeuppri(&kq->kq_sel, PSOCK); 1830 if (!SEL_WAITING(&kq->kq_sel)) 1831 kq->kq_state &= ~KQ_SEL; 1832 } 1833 if (!knlist_empty(&kq->kq_sel.si_note)) 1834 kqueue_schedtask(kq); 1835 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 1836 pgsigio(&kq->kq_sigio, SIGIO, 0); 1837 } 1838} 1839 1840/* 1841 * Walk down a list of knotes, activating them if their event has triggered. 1842 * 1843 * There is a possibility to optimize in the case of one kq watching another. 1844 * Instead of scheduling a task to wake it up, you could pass enough state 1845 * down the chain to make up the parent kqueue. Make this code functional 1846 * first. 1847 */ 1848void 1849knote(struct knlist *list, long hint, int lockflags) 1850{ 1851 struct kqueue *kq; 1852 struct knote *kn; 1853 int error; 1854 1855 if (list == NULL) 1856 return; 1857 1858 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 1859 1860 if ((lockflags & KNF_LISTLOCKED) == 0) 1861 list->kl_lock(list->kl_lockarg); 1862 1863 /* 1864 * If we unlock the list lock (and set KN_INFLUX), we can eliminate 1865 * the kqueue scheduling, but this will introduce four 1866 * lock/unlock's for each knote to test. If we do, continue to use 1867 * SLIST_FOREACH, SLIST_FOREACH_SAFE is not safe in our case, it is 1868 * only safe if you want to remove the current item, which we are 1869 * not doing. 1870 */ 1871 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 1872 kq = kn->kn_kq; 1873 KQ_LOCK(kq); 1874 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { 1875 /* 1876 * Do not process the influx notes, except for 1877 * the influx coming from the kq unlock in the 1878 * kqueue_scan(). In the later case, we do 1879 * not interfere with the scan, since the code 1880 * fragment in kqueue_scan() locks the knlist, 1881 * and cannot proceed until we finished. 1882 */ 1883 KQ_UNLOCK(kq); 1884 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 1885 kn->kn_status |= KN_INFLUX; 1886 KQ_UNLOCK(kq); 1887 error = kn->kn_fop->f_event(kn, hint); 1888 KQ_LOCK(kq); 1889 kn->kn_status &= ~KN_INFLUX; 1890 if (error) 1891 KNOTE_ACTIVATE(kn, 1); 1892 KQ_UNLOCK_FLUX(kq); 1893 } else { 1894 kn->kn_status |= KN_HASKQLOCK; 1895 if (kn->kn_fop->f_event(kn, hint)) 1896 KNOTE_ACTIVATE(kn, 1); 1897 kn->kn_status &= ~KN_HASKQLOCK; 1898 KQ_UNLOCK(kq); 1899 } 1900 } 1901 if ((lockflags & KNF_LISTLOCKED) == 0) 1902 list->kl_unlock(list->kl_lockarg); 1903} 1904 1905/* 1906 * add a knote to a knlist 1907 */ 1908void 1909knlist_add(struct knlist *knl, struct knote *kn, int islocked) 1910{ 1911 KNL_ASSERT_LOCK(knl, islocked); 1912 KQ_NOTOWNED(kn->kn_kq); 1913 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == 1914 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED")); 1915 if (!islocked) 1916 knl->kl_lock(knl->kl_lockarg); 1917 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 1918 if (!islocked) 1919 knl->kl_unlock(knl->kl_lockarg); 1920 KQ_LOCK(kn->kn_kq); 1921 kn->kn_knlist = knl; 1922 kn->kn_status &= ~KN_DETACHED; 1923 KQ_UNLOCK(kn->kn_kq); 1924} 1925 1926static void 1927knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, int kqislocked) 1928{ 1929 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked")); 1930 KNL_ASSERT_LOCK(knl, knlislocked); 1931 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 1932 if (!kqislocked) 1933 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX, 1934 ("knlist_remove called w/o knote being KN_INFLUX or already removed")); 1935 if (!knlislocked) 1936 knl->kl_lock(knl->kl_lockarg); 1937 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 1938 kn->kn_knlist = NULL; 1939 if (!knlislocked) 1940 knl->kl_unlock(knl->kl_lockarg); 1941 if (!kqislocked) 1942 KQ_LOCK(kn->kn_kq); 1943 kn->kn_status |= KN_DETACHED; 1944 if (!kqislocked) 1945 KQ_UNLOCK(kn->kn_kq); 1946} 1947 1948/* 1949 * remove knote from the specified knlist 1950 */ 1951void 1952knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 1953{ 1954 1955 knlist_remove_kq(knl, kn, islocked, 0); 1956} 1957 1958/* 1959 * remove knote from the specified knlist while in f_event handler. 1960 */ 1961void 1962knlist_remove_inevent(struct knlist *knl, struct knote *kn) 1963{ 1964 1965 knlist_remove_kq(knl, kn, 1, 1966 (kn->kn_status & KN_HASKQLOCK) == KN_HASKQLOCK); 1967} 1968 1969int 1970knlist_empty(struct knlist *knl) 1971{ 1972 1973 KNL_ASSERT_LOCKED(knl); 1974 return SLIST_EMPTY(&knl->kl_list); 1975} 1976 1977static struct mtx knlist_lock; 1978MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 1979 MTX_DEF); 1980static void knlist_mtx_lock(void *arg); 1981static void knlist_mtx_unlock(void *arg); 1982 1983static void 1984knlist_mtx_lock(void *arg) 1985{ 1986 1987 mtx_lock((struct mtx *)arg); 1988} 1989 1990static void 1991knlist_mtx_unlock(void *arg) 1992{ 1993 1994 mtx_unlock((struct mtx *)arg); 1995} 1996 1997static void 1998knlist_mtx_assert_locked(void *arg) 1999{ 2000 2001 mtx_assert((struct mtx *)arg, MA_OWNED); 2002} 2003 2004static void 2005knlist_mtx_assert_unlocked(void *arg) 2006{ 2007 2008 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 2009} 2010 2011static void 2012knlist_rw_rlock(void *arg) 2013{ 2014 2015 rw_rlock((struct rwlock *)arg); 2016} 2017 2018static void 2019knlist_rw_runlock(void *arg) 2020{ 2021 2022 rw_runlock((struct rwlock *)arg); 2023} 2024 2025static void 2026knlist_rw_assert_locked(void *arg) 2027{ 2028 2029 rw_assert((struct rwlock *)arg, RA_LOCKED); 2030} 2031 2032static void 2033knlist_rw_assert_unlocked(void *arg) 2034{ 2035 2036 rw_assert((struct rwlock *)arg, RA_UNLOCKED); 2037} 2038 2039void 2040knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 2041 void (*kl_unlock)(void *), 2042 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *)) 2043{ 2044 2045 if (lock == NULL) 2046 knl->kl_lockarg = &knlist_lock; 2047 else 2048 knl->kl_lockarg = lock; 2049 2050 if (kl_lock == NULL) 2051 knl->kl_lock = knlist_mtx_lock; 2052 else 2053 knl->kl_lock = kl_lock; 2054 if (kl_unlock == NULL) 2055 knl->kl_unlock = knlist_mtx_unlock; 2056 else 2057 knl->kl_unlock = kl_unlock; 2058 if (kl_assert_locked == NULL) 2059 knl->kl_assert_locked = knlist_mtx_assert_locked; 2060 else 2061 knl->kl_assert_locked = kl_assert_locked; 2062 if (kl_assert_unlocked == NULL) 2063 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked; 2064 else 2065 knl->kl_assert_unlocked = kl_assert_unlocked; 2066 2067 SLIST_INIT(&knl->kl_list); 2068} 2069 2070void 2071knlist_init_mtx(struct knlist *knl, struct mtx *lock) 2072{ 2073 2074 knlist_init(knl, lock, NULL, NULL, NULL, NULL); 2075} 2076 2077void 2078knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock) 2079{ 2080 2081 knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock, 2082 knlist_rw_assert_locked, knlist_rw_assert_unlocked); 2083} 2084 2085void 2086knlist_destroy(struct knlist *knl) 2087{ 2088 2089#ifdef INVARIANTS 2090 /* 2091 * if we run across this error, we need to find the offending 2092 * driver and have it call knlist_clear or knlist_delete. 2093 */ 2094 if (!SLIST_EMPTY(&knl->kl_list)) 2095 printf("WARNING: destroying knlist w/ knotes on it!\n"); 2096#endif 2097 2098 knl->kl_lockarg = knl->kl_lock = knl->kl_unlock = NULL; 2099 SLIST_INIT(&knl->kl_list); 2100} 2101 2102/* 2103 * Even if we are locked, we may need to drop the lock to allow any influx 2104 * knotes time to "settle". 2105 */ 2106void 2107knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 2108{ 2109 struct knote *kn, *kn2; 2110 struct kqueue *kq; 2111 2112 if (islocked) 2113 KNL_ASSERT_LOCKED(knl); 2114 else { 2115 KNL_ASSERT_UNLOCKED(knl); 2116again: /* need to reacquire lock since we have dropped it */ 2117 knl->kl_lock(knl->kl_lockarg); 2118 } 2119 2120 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 2121 kq = kn->kn_kq; 2122 KQ_LOCK(kq); 2123 if ((kn->kn_status & KN_INFLUX)) { 2124 KQ_UNLOCK(kq); 2125 continue; 2126 } 2127 knlist_remove_kq(knl, kn, 1, 1); 2128 if (killkn) { 2129 kn->kn_status |= KN_INFLUX | KN_DETACHED; 2130 KQ_UNLOCK(kq); 2131 knote_drop(kn, td); 2132 } else { 2133 /* Make sure cleared knotes disappear soon */ 2134 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 2135 KQ_UNLOCK(kq); 2136 } 2137 kq = NULL; 2138 } 2139 2140 if (!SLIST_EMPTY(&knl->kl_list)) { 2141 /* there are still KN_INFLUX remaining */ 2142 kn = SLIST_FIRST(&knl->kl_list); 2143 kq = kn->kn_kq; 2144 KQ_LOCK(kq); 2145 KASSERT(kn->kn_status & KN_INFLUX, 2146 ("knote removed w/o list lock")); 2147 knl->kl_unlock(knl->kl_lockarg); 2148 kq->kq_state |= KQ_FLUXWAIT; 2149 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 2150 kq = NULL; 2151 goto again; 2152 } 2153 2154 if (islocked) 2155 KNL_ASSERT_LOCKED(knl); 2156 else { 2157 knl->kl_unlock(knl->kl_lockarg); 2158 KNL_ASSERT_UNLOCKED(knl); 2159 } 2160} 2161 2162/* 2163 * Remove all knotes referencing a specified fd must be called with FILEDESC 2164 * lock. This prevents a race where a new fd comes along and occupies the 2165 * entry and we attach a knote to the fd. 2166 */ 2167void 2168knote_fdclose(struct thread *td, int fd) 2169{ 2170 struct filedesc *fdp = td->td_proc->p_fd; 2171 struct kqueue *kq; 2172 struct knote *kn; 2173 int influx; 2174 2175 FILEDESC_XLOCK_ASSERT(fdp); 2176 2177 /* 2178 * We shouldn't have to worry about new kevents appearing on fd 2179 * since filedesc is locked. 2180 */ 2181 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2182 KQ_LOCK(kq); 2183 2184again: 2185 influx = 0; 2186 while (kq->kq_knlistsize > fd && 2187 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2188 if (kn->kn_status & KN_INFLUX) { 2189 /* someone else might be waiting on our knote */ 2190 if (influx) 2191 wakeup(kq); 2192 kq->kq_state |= KQ_FLUXWAIT; 2193 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2194 goto again; 2195 } 2196 kn->kn_status |= KN_INFLUX; 2197 KQ_UNLOCK(kq); 2198 if (!(kn->kn_status & KN_DETACHED)) 2199 kn->kn_fop->f_detach(kn); 2200 knote_drop(kn, td); 2201 influx = 1; 2202 KQ_LOCK(kq); 2203 } 2204 KQ_UNLOCK_FLUX(kq); 2205 } 2206} 2207 2208static int 2209knote_attach(struct knote *kn, struct kqueue *kq) 2210{ 2211 struct klist *list; 2212 2213 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX")); 2214 KQ_OWNED(kq); 2215 2216 if (kn->kn_fop->f_isfd) { 2217 if (kn->kn_id >= kq->kq_knlistsize) 2218 return ENOMEM; 2219 list = &kq->kq_knlist[kn->kn_id]; 2220 } else { 2221 if (kq->kq_knhash == NULL) 2222 return ENOMEM; 2223 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2224 } 2225 2226 SLIST_INSERT_HEAD(list, kn, kn_link); 2227 2228 return 0; 2229} 2230 2231/* 2232 * knote must already have been detached using the f_detach method. 2233 * no lock need to be held, it is assumed that the KN_INFLUX flag is set 2234 * to prevent other removal. 2235 */ 2236static void 2237knote_drop(struct knote *kn, struct thread *td) 2238{ 2239 struct kqueue *kq; 2240 struct klist *list; 2241 2242 kq = kn->kn_kq; 2243 2244 KQ_NOTOWNED(kq); 2245 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX, 2246 ("knote_drop called without KN_INFLUX set in kn_status")); 2247 2248 KQ_LOCK(kq); 2249 if (kn->kn_fop->f_isfd) 2250 list = &kq->kq_knlist[kn->kn_id]; 2251 else 2252 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2253 2254 if (!SLIST_EMPTY(list)) 2255 SLIST_REMOVE(list, kn, knote, kn_link); 2256 if (kn->kn_status & KN_QUEUED) 2257 knote_dequeue(kn); 2258 KQ_UNLOCK_FLUX(kq); 2259 2260 if (kn->kn_fop->f_isfd) { 2261 fdrop(kn->kn_fp, td); 2262 kn->kn_fp = NULL; 2263 } 2264 kqueue_fo_release(kn->kn_kevent.filter); 2265 kn->kn_fop = NULL; 2266 knote_free(kn); 2267} 2268 2269static void 2270knote_enqueue(struct knote *kn) 2271{ 2272 struct kqueue *kq = kn->kn_kq; 2273 2274 KQ_OWNED(kn->kn_kq); 2275 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2276 2277 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2278 kn->kn_status |= KN_QUEUED; 2279 kq->kq_count++; 2280 kqueue_wakeup(kq); 2281} 2282 2283static void 2284knote_dequeue(struct knote *kn) 2285{ 2286 struct kqueue *kq = kn->kn_kq; 2287 2288 KQ_OWNED(kn->kn_kq); 2289 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2290 2291 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2292 kn->kn_status &= ~KN_QUEUED; 2293 kq->kq_count--; 2294} 2295 2296static void 2297knote_init(void) 2298{ 2299 2300 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2301 NULL, NULL, UMA_ALIGN_PTR, 0); 2302} 2303SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2304 2305static struct knote * 2306knote_alloc(int waitok) 2307{ 2308 return ((struct knote *)uma_zalloc(knote_zone, 2309 (waitok ? M_WAITOK : M_NOWAIT)|M_ZERO)); 2310} 2311 2312static void 2313knote_free(struct knote *kn) 2314{ 2315 if (kn != NULL) 2316 uma_zfree(knote_zone, kn); 2317} 2318 2319/* 2320 * Register the kev w/ the kq specified by fd. 2321 */ 2322int 2323kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok) 2324{ 2325 struct kqueue *kq; 2326 struct file *fp; 2327 cap_rights_t rights; 2328 int error; 2329 2330 error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp); 2331 if (error != 0) 2332 return (error); 2333 if ((error = kqueue_acquire(fp, &kq)) != 0) 2334 goto noacquire; 2335 2336 error = kqueue_register(kq, kev, td, waitok); 2337 2338 kqueue_release(kq, 0); 2339 2340noacquire: 2341 fdrop(fp, td); 2342 2343 return error; 2344} 2345