kern_event.c revision 315470
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/11/sys/kern/kern_event.c 315470 2017-03-18 03:49:50Z kib $"); 31 32#include "opt_ktrace.h" 33#include "opt_kqueue.h" 34 35#include <sys/param.h> 36#include <sys/systm.h> 37#include <sys/capsicum.h> 38#include <sys/kernel.h> 39#include <sys/lock.h> 40#include <sys/mutex.h> 41#include <sys/rwlock.h> 42#include <sys/proc.h> 43#include <sys/malloc.h> 44#include <sys/unistd.h> 45#include <sys/file.h> 46#include <sys/filedesc.h> 47#include <sys/filio.h> 48#include <sys/fcntl.h> 49#include <sys/kthread.h> 50#include <sys/selinfo.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/resourcevar.h> 57#include <sys/sigio.h> 58#include <sys/signalvar.h> 59#include <sys/socket.h> 60#include <sys/socketvar.h> 61#include <sys/stat.h> 62#include <sys/sysctl.h> 63#include <sys/sysproto.h> 64#include <sys/syscallsubr.h> 65#include <sys/taskqueue.h> 66#include <sys/uio.h> 67#include <sys/user.h> 68#ifdef KTRACE 69#include <sys/ktrace.h> 70#endif 71#include <machine/atomic.h> 72 73#include <vm/uma.h> 74 75static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 76 77/* 78 * This lock is used if multiple kq locks are required. This possibly 79 * should be made into a per proc lock. 80 */ 81static struct mtx kq_global; 82MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 83#define KQ_GLOBAL_LOCK(lck, haslck) do { \ 84 if (!haslck) \ 85 mtx_lock(lck); \ 86 haslck = 1; \ 87} while (0) 88#define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 89 if (haslck) \ 90 mtx_unlock(lck); \ 91 haslck = 0; \ 92} while (0) 93 94TASKQUEUE_DEFINE_THREAD(kqueue_ctx); 95 96static int kevent_copyout(void *arg, struct kevent *kevp, int count); 97static int kevent_copyin(void *arg, struct kevent *kevp, int count); 98static int kqueue_register(struct kqueue *kq, struct kevent *kev, 99 struct thread *td, int waitok); 100static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 101static void kqueue_release(struct kqueue *kq, int locked); 102static void kqueue_destroy(struct kqueue *kq); 103static void kqueue_drain(struct kqueue *kq, struct thread *td); 104static int kqueue_expand(struct kqueue *kq, struct filterops *fops, 105 uintptr_t ident, int waitok); 106static void kqueue_task(void *arg, int pending); 107static int kqueue_scan(struct kqueue *kq, int maxevents, 108 struct kevent_copyops *k_ops, 109 const struct timespec *timeout, 110 struct kevent *keva, struct thread *td); 111static void kqueue_wakeup(struct kqueue *kq); 112static struct filterops *kqueue_fo_find(int filt); 113static void kqueue_fo_release(int filt); 114 115static fo_ioctl_t kqueue_ioctl; 116static fo_poll_t kqueue_poll; 117static fo_kqfilter_t kqueue_kqfilter; 118static fo_stat_t kqueue_stat; 119static fo_close_t kqueue_close; 120static fo_fill_kinfo_t kqueue_fill_kinfo; 121 122static struct fileops kqueueops = { 123 .fo_read = invfo_rdwr, 124 .fo_write = invfo_rdwr, 125 .fo_truncate = invfo_truncate, 126 .fo_ioctl = kqueue_ioctl, 127 .fo_poll = kqueue_poll, 128 .fo_kqfilter = kqueue_kqfilter, 129 .fo_stat = kqueue_stat, 130 .fo_close = kqueue_close, 131 .fo_chmod = invfo_chmod, 132 .fo_chown = invfo_chown, 133 .fo_sendfile = invfo_sendfile, 134 .fo_fill_kinfo = kqueue_fill_kinfo, 135}; 136 137static int knote_attach(struct knote *kn, struct kqueue *kq); 138static void knote_drop(struct knote *kn, struct thread *td); 139static void knote_enqueue(struct knote *kn); 140static void knote_dequeue(struct knote *kn); 141static void knote_init(void); 142static struct knote *knote_alloc(int waitok); 143static void knote_free(struct knote *kn); 144 145static void filt_kqdetach(struct knote *kn); 146static int filt_kqueue(struct knote *kn, long hint); 147static int filt_procattach(struct knote *kn); 148static void filt_procdetach(struct knote *kn); 149static int filt_proc(struct knote *kn, long hint); 150static int filt_fileattach(struct knote *kn); 151static void filt_timerexpire(void *knx); 152static int filt_timerattach(struct knote *kn); 153static void filt_timerdetach(struct knote *kn); 154static int filt_timer(struct knote *kn, long hint); 155static int filt_userattach(struct knote *kn); 156static void filt_userdetach(struct knote *kn); 157static int filt_user(struct knote *kn, long hint); 158static void filt_usertouch(struct knote *kn, struct kevent *kev, 159 u_long type); 160 161static struct filterops file_filtops = { 162 .f_isfd = 1, 163 .f_attach = filt_fileattach, 164}; 165static struct filterops kqread_filtops = { 166 .f_isfd = 1, 167 .f_detach = filt_kqdetach, 168 .f_event = filt_kqueue, 169}; 170/* XXX - move to kern_proc.c? */ 171static struct filterops proc_filtops = { 172 .f_isfd = 0, 173 .f_attach = filt_procattach, 174 .f_detach = filt_procdetach, 175 .f_event = filt_proc, 176}; 177static struct filterops timer_filtops = { 178 .f_isfd = 0, 179 .f_attach = filt_timerattach, 180 .f_detach = filt_timerdetach, 181 .f_event = filt_timer, 182}; 183static struct filterops user_filtops = { 184 .f_attach = filt_userattach, 185 .f_detach = filt_userdetach, 186 .f_event = filt_user, 187 .f_touch = filt_usertouch, 188}; 189 190static uma_zone_t knote_zone; 191static unsigned int kq_ncallouts = 0; 192static unsigned int kq_calloutmax = 4 * 1024; 193SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 194 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 195 196/* XXX - ensure not KN_INFLUX?? */ 197#define KNOTE_ACTIVATE(kn, islock) do { \ 198 if ((islock)) \ 199 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 200 else \ 201 KQ_LOCK((kn)->kn_kq); \ 202 (kn)->kn_status |= KN_ACTIVE; \ 203 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 204 knote_enqueue((kn)); \ 205 if (!(islock)) \ 206 KQ_UNLOCK((kn)->kn_kq); \ 207} while(0) 208#define KQ_LOCK(kq) do { \ 209 mtx_lock(&(kq)->kq_lock); \ 210} while (0) 211#define KQ_FLUX_WAKEUP(kq) do { \ 212 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 213 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 214 wakeup((kq)); \ 215 } \ 216} while (0) 217#define KQ_UNLOCK_FLUX(kq) do { \ 218 KQ_FLUX_WAKEUP(kq); \ 219 mtx_unlock(&(kq)->kq_lock); \ 220} while (0) 221#define KQ_UNLOCK(kq) do { \ 222 mtx_unlock(&(kq)->kq_lock); \ 223} while (0) 224#define KQ_OWNED(kq) do { \ 225 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 226} while (0) 227#define KQ_NOTOWNED(kq) do { \ 228 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 229} while (0) 230 231static struct knlist * 232kn_list_lock(struct knote *kn) 233{ 234 struct knlist *knl; 235 236 knl = kn->kn_knlist; 237 if (knl != NULL) 238 knl->kl_lock(knl->kl_lockarg); 239 return (knl); 240} 241 242static void 243kn_list_unlock(struct knlist *knl) 244{ 245 bool do_free; 246 247 if (knl == NULL) 248 return; 249 do_free = knl->kl_autodestroy && knlist_empty(knl); 250 knl->kl_unlock(knl->kl_lockarg); 251 if (do_free) { 252 knlist_destroy(knl); 253 free(knl, M_KQUEUE); 254 } 255} 256 257#define KNL_ASSERT_LOCK(knl, islocked) do { \ 258 if (islocked) \ 259 KNL_ASSERT_LOCKED(knl); \ 260 else \ 261 KNL_ASSERT_UNLOCKED(knl); \ 262} while (0) 263#ifdef INVARIANTS 264#define KNL_ASSERT_LOCKED(knl) do { \ 265 knl->kl_assert_locked((knl)->kl_lockarg); \ 266} while (0) 267#define KNL_ASSERT_UNLOCKED(knl) do { \ 268 knl->kl_assert_unlocked((knl)->kl_lockarg); \ 269} while (0) 270#else /* !INVARIANTS */ 271#define KNL_ASSERT_LOCKED(knl) do {} while(0) 272#define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 273#endif /* INVARIANTS */ 274 275#ifndef KN_HASHSIZE 276#define KN_HASHSIZE 64 /* XXX should be tunable */ 277#endif 278 279#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 280 281static int 282filt_nullattach(struct knote *kn) 283{ 284 285 return (ENXIO); 286}; 287 288struct filterops null_filtops = { 289 .f_isfd = 0, 290 .f_attach = filt_nullattach, 291}; 292 293/* XXX - make SYSINIT to add these, and move into respective modules. */ 294extern struct filterops sig_filtops; 295extern struct filterops fs_filtops; 296 297/* 298 * Table for for all system-defined filters. 299 */ 300static struct mtx filterops_lock; 301MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", 302 MTX_DEF); 303static struct { 304 struct filterops *for_fop; 305 int for_nolock; 306 int for_refcnt; 307} sysfilt_ops[EVFILT_SYSCOUNT] = { 308 { &file_filtops, 1 }, /* EVFILT_READ */ 309 { &file_filtops, 1 }, /* EVFILT_WRITE */ 310 { &null_filtops }, /* EVFILT_AIO */ 311 { &file_filtops, 1 }, /* EVFILT_VNODE */ 312 { &proc_filtops, 1 }, /* EVFILT_PROC */ 313 { &sig_filtops, 1 }, /* EVFILT_SIGNAL */ 314 { &timer_filtops, 1 }, /* EVFILT_TIMER */ 315 { &file_filtops, 1 }, /* EVFILT_PROCDESC */ 316 { &fs_filtops, 1 }, /* EVFILT_FS */ 317 { &null_filtops }, /* EVFILT_LIO */ 318 { &user_filtops, 1 }, /* EVFILT_USER */ 319 { &null_filtops }, /* EVFILT_SENDFILE */ 320}; 321 322/* 323 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 324 * method. 325 */ 326static int 327filt_fileattach(struct knote *kn) 328{ 329 330 return (fo_kqfilter(kn->kn_fp, kn)); 331} 332 333/*ARGSUSED*/ 334static int 335kqueue_kqfilter(struct file *fp, struct knote *kn) 336{ 337 struct kqueue *kq = kn->kn_fp->f_data; 338 339 if (kn->kn_filter != EVFILT_READ) 340 return (EINVAL); 341 342 kn->kn_status |= KN_KQUEUE; 343 kn->kn_fop = &kqread_filtops; 344 knlist_add(&kq->kq_sel.si_note, kn, 0); 345 346 return (0); 347} 348 349static void 350filt_kqdetach(struct knote *kn) 351{ 352 struct kqueue *kq = kn->kn_fp->f_data; 353 354 knlist_remove(&kq->kq_sel.si_note, kn, 0); 355} 356 357/*ARGSUSED*/ 358static int 359filt_kqueue(struct knote *kn, long hint) 360{ 361 struct kqueue *kq = kn->kn_fp->f_data; 362 363 kn->kn_data = kq->kq_count; 364 return (kn->kn_data > 0); 365} 366 367/* XXX - move to kern_proc.c? */ 368static int 369filt_procattach(struct knote *kn) 370{ 371 struct proc *p; 372 int error; 373 bool exiting, immediate; 374 375 exiting = immediate = false; 376 p = pfind(kn->kn_id); 377 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 378 p = zpfind(kn->kn_id); 379 exiting = true; 380 } else if (p != NULL && (p->p_flag & P_WEXIT)) { 381 exiting = true; 382 } 383 384 if (p == NULL) 385 return (ESRCH); 386 if ((error = p_cansee(curthread, p))) { 387 PROC_UNLOCK(p); 388 return (error); 389 } 390 391 kn->kn_ptr.p_proc = p; 392 kn->kn_flags |= EV_CLEAR; /* automatically set */ 393 394 /* 395 * Internal flag indicating registration done by kernel for the 396 * purposes of getting a NOTE_CHILD notification. 397 */ 398 if (kn->kn_flags & EV_FLAG2) { 399 kn->kn_flags &= ~EV_FLAG2; 400 kn->kn_data = kn->kn_sdata; /* ppid */ 401 kn->kn_fflags = NOTE_CHILD; 402 kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK); 403 immediate = true; /* Force immediate activation of child note. */ 404 } 405 /* 406 * Internal flag indicating registration done by kernel (for other than 407 * NOTE_CHILD). 408 */ 409 if (kn->kn_flags & EV_FLAG1) { 410 kn->kn_flags &= ~EV_FLAG1; 411 } 412 413 knlist_add(p->p_klist, kn, 1); 414 415 /* 416 * Immediately activate any child notes or, in the case of a zombie 417 * target process, exit notes. The latter is necessary to handle the 418 * case where the target process, e.g. a child, dies before the kevent 419 * is registered. 420 */ 421 if (immediate || (exiting && filt_proc(kn, NOTE_EXIT))) 422 KNOTE_ACTIVATE(kn, 0); 423 424 PROC_UNLOCK(p); 425 426 return (0); 427} 428 429/* 430 * The knote may be attached to a different process, which may exit, 431 * leaving nothing for the knote to be attached to. So when the process 432 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 433 * it will be deleted when read out. However, as part of the knote deletion, 434 * this routine is called, so a check is needed to avoid actually performing 435 * a detach, because the original process does not exist any more. 436 */ 437/* XXX - move to kern_proc.c? */ 438static void 439filt_procdetach(struct knote *kn) 440{ 441 442 knlist_remove(kn->kn_knlist, kn, 0); 443 kn->kn_ptr.p_proc = NULL; 444} 445 446/* XXX - move to kern_proc.c? */ 447static int 448filt_proc(struct knote *kn, long hint) 449{ 450 struct proc *p; 451 u_int event; 452 453 p = kn->kn_ptr.p_proc; 454 if (p == NULL) /* already activated, from attach filter */ 455 return (0); 456 457 /* Mask off extra data. */ 458 event = (u_int)hint & NOTE_PCTRLMASK; 459 460 /* If the user is interested in this event, record it. */ 461 if (kn->kn_sfflags & event) 462 kn->kn_fflags |= event; 463 464 /* Process is gone, so flag the event as finished. */ 465 if (event == NOTE_EXIT) { 466 kn->kn_flags |= EV_EOF | EV_ONESHOT; 467 kn->kn_ptr.p_proc = NULL; 468 if (kn->kn_fflags & NOTE_EXIT) 469 kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig); 470 if (kn->kn_fflags == 0) 471 kn->kn_flags |= EV_DROP; 472 return (1); 473 } 474 475 return (kn->kn_fflags != 0); 476} 477 478/* 479 * Called when the process forked. It mostly does the same as the 480 * knote(), activating all knotes registered to be activated when the 481 * process forked. Additionally, for each knote attached to the 482 * parent, check whether user wants to track the new process. If so 483 * attach a new knote to it, and immediately report an event with the 484 * child's pid. 485 */ 486void 487knote_fork(struct knlist *list, int pid) 488{ 489 struct kqueue *kq; 490 struct knote *kn; 491 struct kevent kev; 492 int error; 493 494 if (list == NULL) 495 return; 496 list->kl_lock(list->kl_lockarg); 497 498 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 499 kq = kn->kn_kq; 500 KQ_LOCK(kq); 501 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { 502 KQ_UNLOCK(kq); 503 continue; 504 } 505 506 /* 507 * The same as knote(), activate the event. 508 */ 509 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 510 kn->kn_status |= KN_HASKQLOCK; 511 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 512 KNOTE_ACTIVATE(kn, 1); 513 kn->kn_status &= ~KN_HASKQLOCK; 514 KQ_UNLOCK(kq); 515 continue; 516 } 517 518 /* 519 * The NOTE_TRACK case. In addition to the activation 520 * of the event, we need to register new events to 521 * track the child. Drop the locks in preparation for 522 * the call to kqueue_register(). 523 */ 524 kn->kn_status |= KN_INFLUX; 525 KQ_UNLOCK(kq); 526 list->kl_unlock(list->kl_lockarg); 527 528 /* 529 * Activate existing knote and register tracking knotes with 530 * new process. 531 * 532 * First register a knote to get just the child notice. This 533 * must be a separate note from a potential NOTE_EXIT 534 * notification since both NOTE_CHILD and NOTE_EXIT are defined 535 * to use the data field (in conflicting ways). 536 */ 537 kev.ident = pid; 538 kev.filter = kn->kn_filter; 539 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT | 540 EV_FLAG2; 541 kev.fflags = kn->kn_sfflags; 542 kev.data = kn->kn_id; /* parent */ 543 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 544 error = kqueue_register(kq, &kev, NULL, 0); 545 if (error) 546 kn->kn_fflags |= NOTE_TRACKERR; 547 548 /* 549 * Then register another knote to track other potential events 550 * from the new process. 551 */ 552 kev.ident = pid; 553 kev.filter = kn->kn_filter; 554 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 555 kev.fflags = kn->kn_sfflags; 556 kev.data = kn->kn_id; /* parent */ 557 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 558 error = kqueue_register(kq, &kev, NULL, 0); 559 if (error) 560 kn->kn_fflags |= NOTE_TRACKERR; 561 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 562 KNOTE_ACTIVATE(kn, 0); 563 KQ_LOCK(kq); 564 kn->kn_status &= ~KN_INFLUX; 565 KQ_UNLOCK_FLUX(kq); 566 list->kl_lock(list->kl_lockarg); 567 } 568 list->kl_unlock(list->kl_lockarg); 569} 570 571/* 572 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the 573 * interval timer support code. 574 */ 575 576#define NOTE_TIMER_PRECMASK (NOTE_SECONDS|NOTE_MSECONDS|NOTE_USECONDS| \ 577 NOTE_NSECONDS) 578 579static sbintime_t 580timer2sbintime(intptr_t data, int flags) 581{ 582 583 /* 584 * Macros for converting to the fractional second portion of an 585 * sbintime_t using 64bit multiplication to improve precision. 586 */ 587#define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32) 588#define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32) 589#define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32) 590 switch (flags & NOTE_TIMER_PRECMASK) { 591 case NOTE_SECONDS: 592#ifdef __LP64__ 593 if (data > (SBT_MAX / SBT_1S)) 594 return (SBT_MAX); 595#endif 596 return ((sbintime_t)data << 32); 597 case NOTE_MSECONDS: /* FALLTHROUGH */ 598 case 0: 599 if (data >= 1000) { 600 int64_t secs = data / 1000; 601#ifdef __LP64__ 602 if (secs > (SBT_MAX / SBT_1S)) 603 return (SBT_MAX); 604#endif 605 return (secs << 32 | MS_TO_SBT(data % 1000)); 606 } 607 return MS_TO_SBT(data); 608 case NOTE_USECONDS: 609 if (data >= 1000000) { 610 int64_t secs = data / 1000000; 611#ifdef __LP64__ 612 if (secs > (SBT_MAX / SBT_1S)) 613 return (SBT_MAX); 614#endif 615 return (secs << 32 | US_TO_SBT(data % 1000000)); 616 } 617 return US_TO_SBT(data); 618 case NOTE_NSECONDS: 619 if (data >= 1000000000) { 620 int64_t secs = data / 1000000000; 621#ifdef __LP64__ 622 if (secs > (SBT_MAX / SBT_1S)) 623 return (SBT_MAX); 624#endif 625 return (secs << 32 | US_TO_SBT(data % 1000000000)); 626 } 627 return (NS_TO_SBT(data)); 628 default: 629 break; 630 } 631 return (-1); 632} 633 634struct kq_timer_cb_data { 635 struct callout c; 636 sbintime_t next; /* next timer event fires at */ 637 sbintime_t to; /* precalculated timer period */ 638}; 639 640static void 641filt_timerexpire(void *knx) 642{ 643 struct knote *kn; 644 struct kq_timer_cb_data *kc; 645 646 kn = knx; 647 kn->kn_data++; 648 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 649 650 if ((kn->kn_flags & EV_ONESHOT) != 0) 651 return; 652 653 kc = kn->kn_ptr.p_v; 654 kc->next += kc->to; 655 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, 656 PCPU_GET(cpuid), C_ABSOLUTE); 657} 658 659/* 660 * data contains amount of time to sleep 661 */ 662static int 663filt_timerattach(struct knote *kn) 664{ 665 struct kq_timer_cb_data *kc; 666 sbintime_t to; 667 unsigned int ncallouts; 668 669 if (kn->kn_sdata < 0) 670 return (EINVAL); 671 if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) 672 kn->kn_sdata = 1; 673 /* Only precision unit are supported in flags so far */ 674 if ((kn->kn_sfflags & ~NOTE_TIMER_PRECMASK) != 0) 675 return (EINVAL); 676 677 to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); 678 if (to < 0) 679 return (EINVAL); 680 681 do { 682 ncallouts = kq_ncallouts; 683 if (ncallouts >= kq_calloutmax) 684 return (ENOMEM); 685 } while (!atomic_cmpset_int(&kq_ncallouts, ncallouts, ncallouts + 1)); 686 687 kn->kn_flags |= EV_CLEAR; /* automatically set */ 688 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ 689 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK); 690 callout_init(&kc->c, 1); 691 kc->next = to + sbinuptime(); 692 kc->to = to; 693 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kn, 694 PCPU_GET(cpuid), C_ABSOLUTE); 695 696 return (0); 697} 698 699static void 700filt_timerdetach(struct knote *kn) 701{ 702 struct kq_timer_cb_data *kc; 703 unsigned int old; 704 705 kc = kn->kn_ptr.p_v; 706 callout_drain(&kc->c); 707 free(kc, M_KQUEUE); 708 old = atomic_fetchadd_int(&kq_ncallouts, -1); 709 KASSERT(old > 0, ("Number of callouts cannot become negative")); 710 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ 711} 712 713static int 714filt_timer(struct knote *kn, long hint) 715{ 716 717 return (kn->kn_data != 0); 718} 719 720static int 721filt_userattach(struct knote *kn) 722{ 723 724 /* 725 * EVFILT_USER knotes are not attached to anything in the kernel. 726 */ 727 kn->kn_hook = NULL; 728 if (kn->kn_fflags & NOTE_TRIGGER) 729 kn->kn_hookid = 1; 730 else 731 kn->kn_hookid = 0; 732 return (0); 733} 734 735static void 736filt_userdetach(__unused struct knote *kn) 737{ 738 739 /* 740 * EVFILT_USER knotes are not attached to anything in the kernel. 741 */ 742} 743 744static int 745filt_user(struct knote *kn, __unused long hint) 746{ 747 748 return (kn->kn_hookid); 749} 750 751static void 752filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 753{ 754 u_int ffctrl; 755 756 switch (type) { 757 case EVENT_REGISTER: 758 if (kev->fflags & NOTE_TRIGGER) 759 kn->kn_hookid = 1; 760 761 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 762 kev->fflags &= NOTE_FFLAGSMASK; 763 switch (ffctrl) { 764 case NOTE_FFNOP: 765 break; 766 767 case NOTE_FFAND: 768 kn->kn_sfflags &= kev->fflags; 769 break; 770 771 case NOTE_FFOR: 772 kn->kn_sfflags |= kev->fflags; 773 break; 774 775 case NOTE_FFCOPY: 776 kn->kn_sfflags = kev->fflags; 777 break; 778 779 default: 780 /* XXX Return error? */ 781 break; 782 } 783 kn->kn_sdata = kev->data; 784 if (kev->flags & EV_CLEAR) { 785 kn->kn_hookid = 0; 786 kn->kn_data = 0; 787 kn->kn_fflags = 0; 788 } 789 break; 790 791 case EVENT_PROCESS: 792 *kev = kn->kn_kevent; 793 kev->fflags = kn->kn_sfflags; 794 kev->data = kn->kn_sdata; 795 if (kn->kn_flags & EV_CLEAR) { 796 kn->kn_hookid = 0; 797 kn->kn_data = 0; 798 kn->kn_fflags = 0; 799 } 800 break; 801 802 default: 803 panic("filt_usertouch() - invalid type (%ld)", type); 804 break; 805 } 806} 807 808int 809sys_kqueue(struct thread *td, struct kqueue_args *uap) 810{ 811 812 return (kern_kqueue(td, 0, NULL)); 813} 814 815static void 816kqueue_init(struct kqueue *kq) 817{ 818 819 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK); 820 TAILQ_INIT(&kq->kq_head); 821 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 822 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 823} 824 825int 826kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps) 827{ 828 struct filedesc *fdp; 829 struct kqueue *kq; 830 struct file *fp; 831 struct ucred *cred; 832 int fd, error; 833 834 fdp = td->td_proc->p_fd; 835 cred = td->td_ucred; 836 if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) 837 return (ENOMEM); 838 839 error = falloc_caps(td, &fp, &fd, flags, fcaps); 840 if (error != 0) { 841 chgkqcnt(cred->cr_ruidinfo, -1, 0); 842 return (error); 843 } 844 845 /* An extra reference on `fp' has been held for us by falloc(). */ 846 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 847 kqueue_init(kq); 848 kq->kq_fdp = fdp; 849 kq->kq_cred = crhold(cred); 850 851 FILEDESC_XLOCK(fdp); 852 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 853 FILEDESC_XUNLOCK(fdp); 854 855 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 856 fdrop(fp, td); 857 858 td->td_retval[0] = fd; 859 return (0); 860} 861 862#ifdef KTRACE 863static size_t 864kev_iovlen(int n, u_int kgio) 865{ 866 867 if (n < 0 || n >= kgio / sizeof(struct kevent)) 868 return (kgio); 869 return (n * sizeof(struct kevent)); 870} 871#endif 872 873#ifndef _SYS_SYSPROTO_H_ 874struct kevent_args { 875 int fd; 876 const struct kevent *changelist; 877 int nchanges; 878 struct kevent *eventlist; 879 int nevents; 880 const struct timespec *timeout; 881}; 882#endif 883int 884sys_kevent(struct thread *td, struct kevent_args *uap) 885{ 886 struct timespec ts, *tsp; 887 struct kevent_copyops k_ops = { uap, 888 kevent_copyout, 889 kevent_copyin}; 890 int error; 891#ifdef KTRACE 892 struct uio ktruio; 893 struct iovec ktriov; 894 struct uio *ktruioin = NULL; 895 struct uio *ktruioout = NULL; 896 u_int kgio; 897#endif 898 899 if (uap->timeout != NULL) { 900 error = copyin(uap->timeout, &ts, sizeof(ts)); 901 if (error) 902 return (error); 903 tsp = &ts; 904 } else 905 tsp = NULL; 906 907#ifdef KTRACE 908 if (KTRPOINT(td, KTR_GENIO)) { 909 kgio = ktr_geniosize; 910 ktriov.iov_base = uap->changelist; 911 ktriov.iov_len = kev_iovlen(uap->nchanges, kgio); 912 ktruio = (struct uio){ .uio_iov = &ktriov, .uio_iovcnt = 1, 913 .uio_segflg = UIO_USERSPACE, .uio_rw = UIO_READ, 914 .uio_td = td }; 915 ktruioin = cloneuio(&ktruio); 916 ktriov.iov_base = uap->eventlist; 917 ktriov.iov_len = kev_iovlen(uap->nevents, kgio); 918 ktriov.iov_len = uap->nevents * sizeof(struct kevent); 919 ktruioout = cloneuio(&ktruio); 920 } 921#endif 922 923 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 924 &k_ops, tsp); 925 926#ifdef KTRACE 927 if (ktruioin != NULL) { 928 ktruioin->uio_resid = kev_iovlen(uap->nchanges, kgio); 929 ktrgenio(uap->fd, UIO_WRITE, ktruioin, 0); 930 ktruioout->uio_resid = kev_iovlen(td->td_retval[0], kgio); 931 ktrgenio(uap->fd, UIO_READ, ktruioout, error); 932 } 933#endif 934 935 return (error); 936} 937 938/* 939 * Copy 'count' items into the destination list pointed to by uap->eventlist. 940 */ 941static int 942kevent_copyout(void *arg, struct kevent *kevp, int count) 943{ 944 struct kevent_args *uap; 945 int error; 946 947 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 948 uap = (struct kevent_args *)arg; 949 950 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 951 if (error == 0) 952 uap->eventlist += count; 953 return (error); 954} 955 956/* 957 * Copy 'count' items from the list pointed to by uap->changelist. 958 */ 959static int 960kevent_copyin(void *arg, struct kevent *kevp, int count) 961{ 962 struct kevent_args *uap; 963 int error; 964 965 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 966 uap = (struct kevent_args *)arg; 967 968 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 969 if (error == 0) 970 uap->changelist += count; 971 return (error); 972} 973 974int 975kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 976 struct kevent_copyops *k_ops, const struct timespec *timeout) 977{ 978 cap_rights_t rights; 979 struct file *fp; 980 int error; 981 982 cap_rights_init(&rights); 983 if (nchanges > 0) 984 cap_rights_set(&rights, CAP_KQUEUE_CHANGE); 985 if (nevents > 0) 986 cap_rights_set(&rights, CAP_KQUEUE_EVENT); 987 error = fget(td, fd, &rights, &fp); 988 if (error != 0) 989 return (error); 990 991 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); 992 fdrop(fp, td); 993 994 return (error); 995} 996 997static int 998kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents, 999 struct kevent_copyops *k_ops, const struct timespec *timeout) 1000{ 1001 struct kevent keva[KQ_NEVENTS]; 1002 struct kevent *kevp, *changes; 1003 int i, n, nerrors, error; 1004 1005 nerrors = 0; 1006 while (nchanges > 0) { 1007 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 1008 error = k_ops->k_copyin(k_ops->arg, keva, n); 1009 if (error) 1010 return (error); 1011 changes = keva; 1012 for (i = 0; i < n; i++) { 1013 kevp = &changes[i]; 1014 if (!kevp->filter) 1015 continue; 1016 kevp->flags &= ~EV_SYSFLAGS; 1017 error = kqueue_register(kq, kevp, td, 1); 1018 if (error || (kevp->flags & EV_RECEIPT)) { 1019 if (nevents == 0) 1020 return (error); 1021 kevp->flags = EV_ERROR; 1022 kevp->data = error; 1023 (void)k_ops->k_copyout(k_ops->arg, kevp, 1); 1024 nevents--; 1025 nerrors++; 1026 } 1027 } 1028 nchanges -= n; 1029 } 1030 if (nerrors) { 1031 td->td_retval[0] = nerrors; 1032 return (0); 1033 } 1034 1035 return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td)); 1036} 1037 1038int 1039kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, 1040 struct kevent_copyops *k_ops, const struct timespec *timeout) 1041{ 1042 struct kqueue *kq; 1043 int error; 1044 1045 error = kqueue_acquire(fp, &kq); 1046 if (error != 0) 1047 return (error); 1048 error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout); 1049 kqueue_release(kq, 0); 1050 return (error); 1051} 1052 1053/* 1054 * Performs a kevent() call on a temporarily created kqueue. This can be 1055 * used to perform one-shot polling, similar to poll() and select(). 1056 */ 1057int 1058kern_kevent_anonymous(struct thread *td, int nevents, 1059 struct kevent_copyops *k_ops) 1060{ 1061 struct kqueue kq = {}; 1062 int error; 1063 1064 kqueue_init(&kq); 1065 kq.kq_refcnt = 1; 1066 error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL); 1067 kqueue_drain(&kq, td); 1068 kqueue_destroy(&kq); 1069 return (error); 1070} 1071 1072int 1073kqueue_add_filteropts(int filt, struct filterops *filtops) 1074{ 1075 int error; 1076 1077 error = 0; 1078 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 1079 printf( 1080"trying to add a filterop that is out of range: %d is beyond %d\n", 1081 ~filt, EVFILT_SYSCOUNT); 1082 return EINVAL; 1083 } 1084 mtx_lock(&filterops_lock); 1085 if (sysfilt_ops[~filt].for_fop != &null_filtops && 1086 sysfilt_ops[~filt].for_fop != NULL) 1087 error = EEXIST; 1088 else { 1089 sysfilt_ops[~filt].for_fop = filtops; 1090 sysfilt_ops[~filt].for_refcnt = 0; 1091 } 1092 mtx_unlock(&filterops_lock); 1093 1094 return (error); 1095} 1096 1097int 1098kqueue_del_filteropts(int filt) 1099{ 1100 int error; 1101 1102 error = 0; 1103 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1104 return EINVAL; 1105 1106 mtx_lock(&filterops_lock); 1107 if (sysfilt_ops[~filt].for_fop == &null_filtops || 1108 sysfilt_ops[~filt].for_fop == NULL) 1109 error = EINVAL; 1110 else if (sysfilt_ops[~filt].for_refcnt != 0) 1111 error = EBUSY; 1112 else { 1113 sysfilt_ops[~filt].for_fop = &null_filtops; 1114 sysfilt_ops[~filt].for_refcnt = 0; 1115 } 1116 mtx_unlock(&filterops_lock); 1117 1118 return error; 1119} 1120 1121static struct filterops * 1122kqueue_fo_find(int filt) 1123{ 1124 1125 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1126 return NULL; 1127 1128 if (sysfilt_ops[~filt].for_nolock) 1129 return sysfilt_ops[~filt].for_fop; 1130 1131 mtx_lock(&filterops_lock); 1132 sysfilt_ops[~filt].for_refcnt++; 1133 if (sysfilt_ops[~filt].for_fop == NULL) 1134 sysfilt_ops[~filt].for_fop = &null_filtops; 1135 mtx_unlock(&filterops_lock); 1136 1137 return sysfilt_ops[~filt].for_fop; 1138} 1139 1140static void 1141kqueue_fo_release(int filt) 1142{ 1143 1144 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1145 return; 1146 1147 if (sysfilt_ops[~filt].for_nolock) 1148 return; 1149 1150 mtx_lock(&filterops_lock); 1151 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 1152 ("filter object refcount not valid on release")); 1153 sysfilt_ops[~filt].for_refcnt--; 1154 mtx_unlock(&filterops_lock); 1155} 1156 1157/* 1158 * A ref to kq (obtained via kqueue_acquire) must be held. waitok will 1159 * influence if memory allocation should wait. Make sure it is 0 if you 1160 * hold any mutexes. 1161 */ 1162static int 1163kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, int waitok) 1164{ 1165 struct filterops *fops; 1166 struct file *fp; 1167 struct knote *kn, *tkn; 1168 struct knlist *knl; 1169 cap_rights_t rights; 1170 int error, filt, event; 1171 int haskqglobal, filedesc_unlock; 1172 1173 if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE)) 1174 return (EINVAL); 1175 1176 fp = NULL; 1177 kn = NULL; 1178 knl = NULL; 1179 error = 0; 1180 haskqglobal = 0; 1181 filedesc_unlock = 0; 1182 1183 filt = kev->filter; 1184 fops = kqueue_fo_find(filt); 1185 if (fops == NULL) 1186 return EINVAL; 1187 1188 if (kev->flags & EV_ADD) { 1189 /* 1190 * Prevent waiting with locks. Non-sleepable 1191 * allocation failures are handled in the loop, only 1192 * if the spare knote appears to be actually required. 1193 */ 1194 tkn = knote_alloc(waitok); 1195 } else { 1196 tkn = NULL; 1197 } 1198 1199findkn: 1200 if (fops->f_isfd) { 1201 KASSERT(td != NULL, ("td is NULL")); 1202 if (kev->ident > INT_MAX) 1203 error = EBADF; 1204 else 1205 error = fget(td, kev->ident, 1206 cap_rights_init(&rights, CAP_EVENT), &fp); 1207 if (error) 1208 goto done; 1209 1210 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 1211 kev->ident, 0) != 0) { 1212 /* try again */ 1213 fdrop(fp, td); 1214 fp = NULL; 1215 error = kqueue_expand(kq, fops, kev->ident, waitok); 1216 if (error) 1217 goto done; 1218 goto findkn; 1219 } 1220 1221 if (fp->f_type == DTYPE_KQUEUE) { 1222 /* 1223 * If we add some intelligence about what we are doing, 1224 * we should be able to support events on ourselves. 1225 * We need to know when we are doing this to prevent 1226 * getting both the knlist lock and the kq lock since 1227 * they are the same thing. 1228 */ 1229 if (fp->f_data == kq) { 1230 error = EINVAL; 1231 goto done; 1232 } 1233 1234 /* 1235 * Pre-lock the filedesc before the global 1236 * lock mutex, see the comment in 1237 * kqueue_close(). 1238 */ 1239 FILEDESC_XLOCK(td->td_proc->p_fd); 1240 filedesc_unlock = 1; 1241 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1242 } 1243 1244 KQ_LOCK(kq); 1245 if (kev->ident < kq->kq_knlistsize) { 1246 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1247 if (kev->filter == kn->kn_filter) 1248 break; 1249 } 1250 } else { 1251 if ((kev->flags & EV_ADD) == EV_ADD) 1252 kqueue_expand(kq, fops, kev->ident, waitok); 1253 1254 KQ_LOCK(kq); 1255 1256 /* 1257 * If possible, find an existing knote to use for this kevent. 1258 */ 1259 if (kev->filter == EVFILT_PROC && 1260 (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) { 1261 /* This is an internal creation of a process tracking 1262 * note. Don't attempt to coalesce this with an 1263 * existing note. 1264 */ 1265 ; 1266 } else if (kq->kq_knhashmask != 0) { 1267 struct klist *list; 1268 1269 list = &kq->kq_knhash[ 1270 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1271 SLIST_FOREACH(kn, list, kn_link) 1272 if (kev->ident == kn->kn_id && 1273 kev->filter == kn->kn_filter) 1274 break; 1275 } 1276 } 1277 1278 /* knote is in the process of changing, wait for it to stabilize. */ 1279 if (kn != NULL && (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1280 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1281 if (filedesc_unlock) { 1282 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1283 filedesc_unlock = 0; 1284 } 1285 kq->kq_state |= KQ_FLUXWAIT; 1286 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1287 if (fp != NULL) { 1288 fdrop(fp, td); 1289 fp = NULL; 1290 } 1291 goto findkn; 1292 } 1293 1294 /* 1295 * kn now contains the matching knote, or NULL if no match 1296 */ 1297 if (kn == NULL) { 1298 if (kev->flags & EV_ADD) { 1299 kn = tkn; 1300 tkn = NULL; 1301 if (kn == NULL) { 1302 KQ_UNLOCK(kq); 1303 error = ENOMEM; 1304 goto done; 1305 } 1306 kn->kn_fp = fp; 1307 kn->kn_kq = kq; 1308 kn->kn_fop = fops; 1309 /* 1310 * apply reference counts to knote structure, and 1311 * do not release it at the end of this routine. 1312 */ 1313 fops = NULL; 1314 fp = NULL; 1315 1316 kn->kn_sfflags = kev->fflags; 1317 kn->kn_sdata = kev->data; 1318 kev->fflags = 0; 1319 kev->data = 0; 1320 kn->kn_kevent = *kev; 1321 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1322 EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); 1323 kn->kn_status = KN_INFLUX|KN_DETACHED; 1324 1325 error = knote_attach(kn, kq); 1326 KQ_UNLOCK(kq); 1327 if (error != 0) { 1328 tkn = kn; 1329 goto done; 1330 } 1331 1332 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1333 knote_drop(kn, td); 1334 goto done; 1335 } 1336 knl = kn_list_lock(kn); 1337 goto done_ev_add; 1338 } else { 1339 /* No matching knote and the EV_ADD flag is not set. */ 1340 KQ_UNLOCK(kq); 1341 error = ENOENT; 1342 goto done; 1343 } 1344 } 1345 1346 if (kev->flags & EV_DELETE) { 1347 kn->kn_status |= KN_INFLUX; 1348 KQ_UNLOCK(kq); 1349 if (!(kn->kn_status & KN_DETACHED)) 1350 kn->kn_fop->f_detach(kn); 1351 knote_drop(kn, td); 1352 goto done; 1353 } 1354 1355 if (kev->flags & EV_FORCEONESHOT) { 1356 kn->kn_flags |= EV_ONESHOT; 1357 KNOTE_ACTIVATE(kn, 1); 1358 } 1359 1360 /* 1361 * The user may change some filter values after the initial EV_ADD, 1362 * but doing so will not reset any filter which has already been 1363 * triggered. 1364 */ 1365 kn->kn_status |= KN_INFLUX | KN_SCAN; 1366 KQ_UNLOCK(kq); 1367 knl = kn_list_lock(kn); 1368 kn->kn_kevent.udata = kev->udata; 1369 if (!fops->f_isfd && fops->f_touch != NULL) { 1370 fops->f_touch(kn, kev, EVENT_REGISTER); 1371 } else { 1372 kn->kn_sfflags = kev->fflags; 1373 kn->kn_sdata = kev->data; 1374 } 1375 1376 /* 1377 * We can get here with kn->kn_knlist == NULL. This can happen when 1378 * the initial attach event decides that the event is "completed" 1379 * already. i.e. filt_procattach is called on a zombie process. It 1380 * will call filt_proc which will remove it from the list, and NULL 1381 * kn_knlist. 1382 */ 1383done_ev_add: 1384 if ((kev->flags & EV_ENABLE) != 0) 1385 kn->kn_status &= ~KN_DISABLED; 1386 else if ((kev->flags & EV_DISABLE) != 0) 1387 kn->kn_status |= KN_DISABLED; 1388 1389 if ((kn->kn_status & KN_DISABLED) == 0) 1390 event = kn->kn_fop->f_event(kn, 0); 1391 else 1392 event = 0; 1393 1394 KQ_LOCK(kq); 1395 if (event) 1396 kn->kn_status |= KN_ACTIVE; 1397 if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) == 1398 KN_ACTIVE) 1399 knote_enqueue(kn); 1400 kn->kn_status &= ~(KN_INFLUX | KN_SCAN); 1401 kn_list_unlock(knl); 1402 KQ_UNLOCK_FLUX(kq); 1403 1404done: 1405 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1406 if (filedesc_unlock) 1407 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1408 if (fp != NULL) 1409 fdrop(fp, td); 1410 knote_free(tkn); 1411 if (fops != NULL) 1412 kqueue_fo_release(filt); 1413 return (error); 1414} 1415 1416static int 1417kqueue_acquire(struct file *fp, struct kqueue **kqp) 1418{ 1419 int error; 1420 struct kqueue *kq; 1421 1422 error = 0; 1423 1424 kq = fp->f_data; 1425 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1426 return (EBADF); 1427 *kqp = kq; 1428 KQ_LOCK(kq); 1429 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1430 KQ_UNLOCK(kq); 1431 return (EBADF); 1432 } 1433 kq->kq_refcnt++; 1434 KQ_UNLOCK(kq); 1435 1436 return error; 1437} 1438 1439static void 1440kqueue_release(struct kqueue *kq, int locked) 1441{ 1442 if (locked) 1443 KQ_OWNED(kq); 1444 else 1445 KQ_LOCK(kq); 1446 kq->kq_refcnt--; 1447 if (kq->kq_refcnt == 1) 1448 wakeup(&kq->kq_refcnt); 1449 if (!locked) 1450 KQ_UNLOCK(kq); 1451} 1452 1453static void 1454kqueue_schedtask(struct kqueue *kq) 1455{ 1456 1457 KQ_OWNED(kq); 1458 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1459 ("scheduling kqueue task while draining")); 1460 1461 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1462 taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task); 1463 kq->kq_state |= KQ_TASKSCHED; 1464 } 1465} 1466 1467/* 1468 * Expand the kq to make sure we have storage for fops/ident pair. 1469 * 1470 * Return 0 on success (or no work necessary), return errno on failure. 1471 * 1472 * Not calling hashinit w/ waitok (proper malloc flag) should be safe. 1473 * If kqueue_register is called from a non-fd context, there usually/should 1474 * be no locks held. 1475 */ 1476static int 1477kqueue_expand(struct kqueue *kq, struct filterops *fops, uintptr_t ident, 1478 int waitok) 1479{ 1480 struct klist *list, *tmp_knhash, *to_free; 1481 u_long tmp_knhashmask; 1482 int size; 1483 int fd; 1484 int mflag = waitok ? M_WAITOK : M_NOWAIT; 1485 1486 KQ_NOTOWNED(kq); 1487 1488 to_free = NULL; 1489 if (fops->f_isfd) { 1490 fd = ident; 1491 if (kq->kq_knlistsize <= fd) { 1492 size = kq->kq_knlistsize; 1493 while (size <= fd) 1494 size += KQEXTENT; 1495 list = malloc(size * sizeof(*list), M_KQUEUE, mflag); 1496 if (list == NULL) 1497 return ENOMEM; 1498 KQ_LOCK(kq); 1499 if (kq->kq_knlistsize > fd) { 1500 to_free = list; 1501 list = NULL; 1502 } else { 1503 if (kq->kq_knlist != NULL) { 1504 bcopy(kq->kq_knlist, list, 1505 kq->kq_knlistsize * sizeof(*list)); 1506 to_free = kq->kq_knlist; 1507 kq->kq_knlist = NULL; 1508 } 1509 bzero((caddr_t)list + 1510 kq->kq_knlistsize * sizeof(*list), 1511 (size - kq->kq_knlistsize) * sizeof(*list)); 1512 kq->kq_knlistsize = size; 1513 kq->kq_knlist = list; 1514 } 1515 KQ_UNLOCK(kq); 1516 } 1517 } else { 1518 if (kq->kq_knhashmask == 0) { 1519 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1520 &tmp_knhashmask); 1521 if (tmp_knhash == NULL) 1522 return ENOMEM; 1523 KQ_LOCK(kq); 1524 if (kq->kq_knhashmask == 0) { 1525 kq->kq_knhash = tmp_knhash; 1526 kq->kq_knhashmask = tmp_knhashmask; 1527 } else { 1528 to_free = tmp_knhash; 1529 } 1530 KQ_UNLOCK(kq); 1531 } 1532 } 1533 free(to_free, M_KQUEUE); 1534 1535 KQ_NOTOWNED(kq); 1536 return 0; 1537} 1538 1539static void 1540kqueue_task(void *arg, int pending) 1541{ 1542 struct kqueue *kq; 1543 int haskqglobal; 1544 1545 haskqglobal = 0; 1546 kq = arg; 1547 1548 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1549 KQ_LOCK(kq); 1550 1551 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 1552 1553 kq->kq_state &= ~KQ_TASKSCHED; 1554 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 1555 wakeup(&kq->kq_state); 1556 } 1557 KQ_UNLOCK(kq); 1558 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1559} 1560 1561/* 1562 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 1563 * We treat KN_MARKER knotes as if they are INFLUX. 1564 */ 1565static int 1566kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 1567 const struct timespec *tsp, struct kevent *keva, struct thread *td) 1568{ 1569 struct kevent *kevp; 1570 struct knote *kn, *marker; 1571 struct knlist *knl; 1572 sbintime_t asbt, rsbt; 1573 int count, error, haskqglobal, influx, nkev, touch; 1574 1575 count = maxevents; 1576 nkev = 0; 1577 error = 0; 1578 haskqglobal = 0; 1579 1580 if (maxevents == 0) 1581 goto done_nl; 1582 1583 rsbt = 0; 1584 if (tsp != NULL) { 1585 if (tsp->tv_sec < 0 || tsp->tv_nsec < 0 || 1586 tsp->tv_nsec >= 1000000000) { 1587 error = EINVAL; 1588 goto done_nl; 1589 } 1590 if (timespecisset(tsp)) { 1591 if (tsp->tv_sec <= INT32_MAX) { 1592 rsbt = tstosbt(*tsp); 1593 if (TIMESEL(&asbt, rsbt)) 1594 asbt += tc_tick_sbt; 1595 if (asbt <= SBT_MAX - rsbt) 1596 asbt += rsbt; 1597 else 1598 asbt = 0; 1599 rsbt >>= tc_precexp; 1600 } else 1601 asbt = 0; 1602 } else 1603 asbt = -1; 1604 } else 1605 asbt = 0; 1606 marker = knote_alloc(1); 1607 marker->kn_status = KN_MARKER; 1608 KQ_LOCK(kq); 1609 1610retry: 1611 kevp = keva; 1612 if (kq->kq_count == 0) { 1613 if (asbt == -1) { 1614 error = EWOULDBLOCK; 1615 } else { 1616 kq->kq_state |= KQ_SLEEP; 1617 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, 1618 "kqread", asbt, rsbt, C_ABSOLUTE); 1619 } 1620 if (error == 0) 1621 goto retry; 1622 /* don't restart after signals... */ 1623 if (error == ERESTART) 1624 error = EINTR; 1625 else if (error == EWOULDBLOCK) 1626 error = 0; 1627 goto done; 1628 } 1629 1630 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1631 influx = 0; 1632 while (count) { 1633 KQ_OWNED(kq); 1634 kn = TAILQ_FIRST(&kq->kq_head); 1635 1636 if ((kn->kn_status == KN_MARKER && kn != marker) || 1637 (kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1638 if (influx) { 1639 influx = 0; 1640 KQ_FLUX_WAKEUP(kq); 1641 } 1642 kq->kq_state |= KQ_FLUXWAIT; 1643 error = msleep(kq, &kq->kq_lock, PSOCK, 1644 "kqflxwt", 0); 1645 continue; 1646 } 1647 1648 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1649 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 1650 kn->kn_status &= ~KN_QUEUED; 1651 kq->kq_count--; 1652 continue; 1653 } 1654 if (kn == marker) { 1655 KQ_FLUX_WAKEUP(kq); 1656 if (count == maxevents) 1657 goto retry; 1658 goto done; 1659 } 1660 KASSERT((kn->kn_status & KN_INFLUX) == 0, 1661 ("KN_INFLUX set when not suppose to be")); 1662 1663 if ((kn->kn_flags & EV_DROP) == EV_DROP) { 1664 kn->kn_status &= ~KN_QUEUED; 1665 kn->kn_status |= KN_INFLUX; 1666 kq->kq_count--; 1667 KQ_UNLOCK(kq); 1668 /* 1669 * We don't need to lock the list since we've marked 1670 * it _INFLUX. 1671 */ 1672 if (!(kn->kn_status & KN_DETACHED)) 1673 kn->kn_fop->f_detach(kn); 1674 knote_drop(kn, td); 1675 KQ_LOCK(kq); 1676 continue; 1677 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 1678 kn->kn_status &= ~KN_QUEUED; 1679 kn->kn_status |= KN_INFLUX; 1680 kq->kq_count--; 1681 KQ_UNLOCK(kq); 1682 /* 1683 * We don't need to lock the list since we've marked 1684 * it _INFLUX. 1685 */ 1686 *kevp = kn->kn_kevent; 1687 if (!(kn->kn_status & KN_DETACHED)) 1688 kn->kn_fop->f_detach(kn); 1689 knote_drop(kn, td); 1690 KQ_LOCK(kq); 1691 kn = NULL; 1692 } else { 1693 kn->kn_status |= KN_INFLUX | KN_SCAN; 1694 KQ_UNLOCK(kq); 1695 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 1696 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1697 knl = kn_list_lock(kn); 1698 if (kn->kn_fop->f_event(kn, 0) == 0) { 1699 KQ_LOCK(kq); 1700 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1701 kn->kn_status &= 1702 ~(KN_QUEUED | KN_ACTIVE | KN_INFLUX | 1703 KN_SCAN); 1704 kq->kq_count--; 1705 kn_list_unlock(knl); 1706 influx = 1; 1707 continue; 1708 } 1709 touch = (!kn->kn_fop->f_isfd && 1710 kn->kn_fop->f_touch != NULL); 1711 if (touch) 1712 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 1713 else 1714 *kevp = kn->kn_kevent; 1715 KQ_LOCK(kq); 1716 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1717 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 1718 /* 1719 * Manually clear knotes who weren't 1720 * 'touch'ed. 1721 */ 1722 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 1723 kn->kn_data = 0; 1724 kn->kn_fflags = 0; 1725 } 1726 if (kn->kn_flags & EV_DISPATCH) 1727 kn->kn_status |= KN_DISABLED; 1728 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1729 kq->kq_count--; 1730 } else 1731 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1732 1733 kn->kn_status &= ~(KN_INFLUX | KN_SCAN); 1734 kn_list_unlock(knl); 1735 influx = 1; 1736 } 1737 1738 /* we are returning a copy to the user */ 1739 kevp++; 1740 nkev++; 1741 count--; 1742 1743 if (nkev == KQ_NEVENTS) { 1744 influx = 0; 1745 KQ_UNLOCK_FLUX(kq); 1746 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1747 nkev = 0; 1748 kevp = keva; 1749 KQ_LOCK(kq); 1750 if (error) 1751 break; 1752 } 1753 } 1754 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1755done: 1756 KQ_OWNED(kq); 1757 KQ_UNLOCK_FLUX(kq); 1758 knote_free(marker); 1759done_nl: 1760 KQ_NOTOWNED(kq); 1761 if (nkev != 0) 1762 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 1763 td->td_retval[0] = maxevents - count; 1764 return (error); 1765} 1766 1767/*ARGSUSED*/ 1768static int 1769kqueue_ioctl(struct file *fp, u_long cmd, void *data, 1770 struct ucred *active_cred, struct thread *td) 1771{ 1772 /* 1773 * Enabling sigio causes two major problems: 1774 * 1) infinite recursion: 1775 * Synopsys: kevent is being used to track signals and have FIOASYNC 1776 * set. On receipt of a signal this will cause a kqueue to recurse 1777 * into itself over and over. Sending the sigio causes the kqueue 1778 * to become ready, which in turn posts sigio again, forever. 1779 * Solution: this can be solved by setting a flag in the kqueue that 1780 * we have a SIGIO in progress. 1781 * 2) locking problems: 1782 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 1783 * us above the proc and pgrp locks. 1784 * Solution: Post a signal using an async mechanism, being sure to 1785 * record a generation count in the delivery so that we do not deliver 1786 * a signal to the wrong process. 1787 * 1788 * Note, these two mechanisms are somewhat mutually exclusive! 1789 */ 1790#if 0 1791 struct kqueue *kq; 1792 1793 kq = fp->f_data; 1794 switch (cmd) { 1795 case FIOASYNC: 1796 if (*(int *)data) { 1797 kq->kq_state |= KQ_ASYNC; 1798 } else { 1799 kq->kq_state &= ~KQ_ASYNC; 1800 } 1801 return (0); 1802 1803 case FIOSETOWN: 1804 return (fsetown(*(int *)data, &kq->kq_sigio)); 1805 1806 case FIOGETOWN: 1807 *(int *)data = fgetown(&kq->kq_sigio); 1808 return (0); 1809 } 1810#endif 1811 1812 return (ENOTTY); 1813} 1814 1815/*ARGSUSED*/ 1816static int 1817kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 1818 struct thread *td) 1819{ 1820 struct kqueue *kq; 1821 int revents = 0; 1822 int error; 1823 1824 if ((error = kqueue_acquire(fp, &kq))) 1825 return POLLERR; 1826 1827 KQ_LOCK(kq); 1828 if (events & (POLLIN | POLLRDNORM)) { 1829 if (kq->kq_count) { 1830 revents |= events & (POLLIN | POLLRDNORM); 1831 } else { 1832 selrecord(td, &kq->kq_sel); 1833 if (SEL_WAITING(&kq->kq_sel)) 1834 kq->kq_state |= KQ_SEL; 1835 } 1836 } 1837 kqueue_release(kq, 1); 1838 KQ_UNLOCK(kq); 1839 return (revents); 1840} 1841 1842/*ARGSUSED*/ 1843static int 1844kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 1845 struct thread *td) 1846{ 1847 1848 bzero((void *)st, sizeof *st); 1849 /* 1850 * We no longer return kq_count because the unlocked value is useless. 1851 * If you spent all this time getting the count, why not spend your 1852 * syscall better by calling kevent? 1853 * 1854 * XXX - This is needed for libc_r. 1855 */ 1856 st->st_mode = S_IFIFO; 1857 return (0); 1858} 1859 1860static void 1861kqueue_drain(struct kqueue *kq, struct thread *td) 1862{ 1863 struct knote *kn; 1864 int i; 1865 1866 KQ_LOCK(kq); 1867 1868 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 1869 ("kqueue already closing")); 1870 kq->kq_state |= KQ_CLOSING; 1871 if (kq->kq_refcnt > 1) 1872 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 1873 1874 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 1875 1876 KASSERT(knlist_empty(&kq->kq_sel.si_note), 1877 ("kqueue's knlist not empty")); 1878 1879 for (i = 0; i < kq->kq_knlistsize; i++) { 1880 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 1881 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1882 kq->kq_state |= KQ_FLUXWAIT; 1883 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 1884 continue; 1885 } 1886 kn->kn_status |= KN_INFLUX; 1887 KQ_UNLOCK(kq); 1888 if (!(kn->kn_status & KN_DETACHED)) 1889 kn->kn_fop->f_detach(kn); 1890 knote_drop(kn, td); 1891 KQ_LOCK(kq); 1892 } 1893 } 1894 if (kq->kq_knhashmask != 0) { 1895 for (i = 0; i <= kq->kq_knhashmask; i++) { 1896 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 1897 if ((kn->kn_status & KN_INFLUX) == KN_INFLUX) { 1898 kq->kq_state |= KQ_FLUXWAIT; 1899 msleep(kq, &kq->kq_lock, PSOCK, 1900 "kqclo2", 0); 1901 continue; 1902 } 1903 kn->kn_status |= KN_INFLUX; 1904 KQ_UNLOCK(kq); 1905 if (!(kn->kn_status & KN_DETACHED)) 1906 kn->kn_fop->f_detach(kn); 1907 knote_drop(kn, td); 1908 KQ_LOCK(kq); 1909 } 1910 } 1911 } 1912 1913 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 1914 kq->kq_state |= KQ_TASKDRAIN; 1915 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 1916 } 1917 1918 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 1919 selwakeuppri(&kq->kq_sel, PSOCK); 1920 if (!SEL_WAITING(&kq->kq_sel)) 1921 kq->kq_state &= ~KQ_SEL; 1922 } 1923 1924 KQ_UNLOCK(kq); 1925} 1926 1927static void 1928kqueue_destroy(struct kqueue *kq) 1929{ 1930 1931 KASSERT(kq->kq_fdp == NULL, 1932 ("kqueue still attached to a file descriptor")); 1933 seldrain(&kq->kq_sel); 1934 knlist_destroy(&kq->kq_sel.si_note); 1935 mtx_destroy(&kq->kq_lock); 1936 1937 if (kq->kq_knhash != NULL) 1938 free(kq->kq_knhash, M_KQUEUE); 1939 if (kq->kq_knlist != NULL) 1940 free(kq->kq_knlist, M_KQUEUE); 1941 1942 funsetown(&kq->kq_sigio); 1943} 1944 1945/*ARGSUSED*/ 1946static int 1947kqueue_close(struct file *fp, struct thread *td) 1948{ 1949 struct kqueue *kq = fp->f_data; 1950 struct filedesc *fdp; 1951 int error; 1952 int filedesc_unlock; 1953 1954 if ((error = kqueue_acquire(fp, &kq))) 1955 return error; 1956 kqueue_drain(kq, td); 1957 1958 /* 1959 * We could be called due to the knote_drop() doing fdrop(), 1960 * called from kqueue_register(). In this case the global 1961 * lock is owned, and filedesc sx is locked before, to not 1962 * take the sleepable lock after non-sleepable. 1963 */ 1964 fdp = kq->kq_fdp; 1965 kq->kq_fdp = NULL; 1966 if (!sx_xlocked(FILEDESC_LOCK(fdp))) { 1967 FILEDESC_XLOCK(fdp); 1968 filedesc_unlock = 1; 1969 } else 1970 filedesc_unlock = 0; 1971 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); 1972 if (filedesc_unlock) 1973 FILEDESC_XUNLOCK(fdp); 1974 1975 kqueue_destroy(kq); 1976 chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); 1977 crfree(kq->kq_cred); 1978 free(kq, M_KQUEUE); 1979 fp->f_data = NULL; 1980 1981 return (0); 1982} 1983 1984static int 1985kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 1986{ 1987 1988 kif->kf_type = KF_TYPE_KQUEUE; 1989 return (0); 1990} 1991 1992static void 1993kqueue_wakeup(struct kqueue *kq) 1994{ 1995 KQ_OWNED(kq); 1996 1997 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 1998 kq->kq_state &= ~KQ_SLEEP; 1999 wakeup(kq); 2000 } 2001 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2002 selwakeuppri(&kq->kq_sel, PSOCK); 2003 if (!SEL_WAITING(&kq->kq_sel)) 2004 kq->kq_state &= ~KQ_SEL; 2005 } 2006 if (!knlist_empty(&kq->kq_sel.si_note)) 2007 kqueue_schedtask(kq); 2008 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 2009 pgsigio(&kq->kq_sigio, SIGIO, 0); 2010 } 2011} 2012 2013/* 2014 * Walk down a list of knotes, activating them if their event has triggered. 2015 * 2016 * There is a possibility to optimize in the case of one kq watching another. 2017 * Instead of scheduling a task to wake it up, you could pass enough state 2018 * down the chain to make up the parent kqueue. Make this code functional 2019 * first. 2020 */ 2021void 2022knote(struct knlist *list, long hint, int lockflags) 2023{ 2024 struct kqueue *kq; 2025 struct knote *kn, *tkn; 2026 int error; 2027 bool own_influx; 2028 2029 if (list == NULL) 2030 return; 2031 2032 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 2033 2034 if ((lockflags & KNF_LISTLOCKED) == 0) 2035 list->kl_lock(list->kl_lockarg); 2036 2037 /* 2038 * If we unlock the list lock (and set KN_INFLUX), we can 2039 * eliminate the kqueue scheduling, but this will introduce 2040 * four lock/unlock's for each knote to test. Also, marker 2041 * would be needed to keep iteration position, since filters 2042 * or other threads could remove events. 2043 */ 2044 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) { 2045 kq = kn->kn_kq; 2046 KQ_LOCK(kq); 2047 if ((kn->kn_status & (KN_INFLUX | KN_SCAN)) == KN_INFLUX) { 2048 /* 2049 * Do not process the influx notes, except for 2050 * the influx coming from the kq unlock in the 2051 * kqueue_scan(). In the later case, we do 2052 * not interfere with the scan, since the code 2053 * fragment in kqueue_scan() locks the knlist, 2054 * and cannot proceed until we finished. 2055 */ 2056 KQ_UNLOCK(kq); 2057 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 2058 own_influx = (kn->kn_status & KN_INFLUX) == 0; 2059 if (own_influx) 2060 kn->kn_status |= KN_INFLUX; 2061 KQ_UNLOCK(kq); 2062 error = kn->kn_fop->f_event(kn, hint); 2063 KQ_LOCK(kq); 2064 if (own_influx) 2065 kn->kn_status &= ~KN_INFLUX; 2066 if (error) 2067 KNOTE_ACTIVATE(kn, 1); 2068 KQ_UNLOCK_FLUX(kq); 2069 } else { 2070 kn->kn_status |= KN_HASKQLOCK; 2071 if (kn->kn_fop->f_event(kn, hint)) 2072 KNOTE_ACTIVATE(kn, 1); 2073 kn->kn_status &= ~KN_HASKQLOCK; 2074 KQ_UNLOCK(kq); 2075 } 2076 } 2077 if ((lockflags & KNF_LISTLOCKED) == 0) 2078 list->kl_unlock(list->kl_lockarg); 2079} 2080 2081/* 2082 * add a knote to a knlist 2083 */ 2084void 2085knlist_add(struct knlist *knl, struct knote *kn, int islocked) 2086{ 2087 KNL_ASSERT_LOCK(knl, islocked); 2088 KQ_NOTOWNED(kn->kn_kq); 2089 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == 2090 (KN_INFLUX|KN_DETACHED), ("knote not KN_INFLUX and KN_DETACHED")); 2091 if (!islocked) 2092 knl->kl_lock(knl->kl_lockarg); 2093 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 2094 if (!islocked) 2095 knl->kl_unlock(knl->kl_lockarg); 2096 KQ_LOCK(kn->kn_kq); 2097 kn->kn_knlist = knl; 2098 kn->kn_status &= ~KN_DETACHED; 2099 KQ_UNLOCK(kn->kn_kq); 2100} 2101 2102static void 2103knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, 2104 int kqislocked) 2105{ 2106 KASSERT(!(!!kqislocked && !knlislocked), ("kq locked w/o knl locked")); 2107 KNL_ASSERT_LOCK(knl, knlislocked); 2108 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 2109 if (!kqislocked) 2110 KASSERT((kn->kn_status & (KN_INFLUX|KN_DETACHED)) == KN_INFLUX, 2111 ("knlist_remove called w/o knote being KN_INFLUX or already removed")); 2112 if (!knlislocked) 2113 knl->kl_lock(knl->kl_lockarg); 2114 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 2115 kn->kn_knlist = NULL; 2116 if (!knlislocked) 2117 kn_list_unlock(knl); 2118 if (!kqislocked) 2119 KQ_LOCK(kn->kn_kq); 2120 kn->kn_status |= KN_DETACHED; 2121 if (!kqislocked) 2122 KQ_UNLOCK(kn->kn_kq); 2123} 2124 2125/* 2126 * remove knote from the specified knlist 2127 */ 2128void 2129knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 2130{ 2131 2132 knlist_remove_kq(knl, kn, islocked, 0); 2133} 2134 2135int 2136knlist_empty(struct knlist *knl) 2137{ 2138 2139 KNL_ASSERT_LOCKED(knl); 2140 return (SLIST_EMPTY(&knl->kl_list)); 2141} 2142 2143static struct mtx knlist_lock; 2144MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 2145 MTX_DEF); 2146static void knlist_mtx_lock(void *arg); 2147static void knlist_mtx_unlock(void *arg); 2148 2149static void 2150knlist_mtx_lock(void *arg) 2151{ 2152 2153 mtx_lock((struct mtx *)arg); 2154} 2155 2156static void 2157knlist_mtx_unlock(void *arg) 2158{ 2159 2160 mtx_unlock((struct mtx *)arg); 2161} 2162 2163static void 2164knlist_mtx_assert_locked(void *arg) 2165{ 2166 2167 mtx_assert((struct mtx *)arg, MA_OWNED); 2168} 2169 2170static void 2171knlist_mtx_assert_unlocked(void *arg) 2172{ 2173 2174 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 2175} 2176 2177static void 2178knlist_rw_rlock(void *arg) 2179{ 2180 2181 rw_rlock((struct rwlock *)arg); 2182} 2183 2184static void 2185knlist_rw_runlock(void *arg) 2186{ 2187 2188 rw_runlock((struct rwlock *)arg); 2189} 2190 2191static void 2192knlist_rw_assert_locked(void *arg) 2193{ 2194 2195 rw_assert((struct rwlock *)arg, RA_LOCKED); 2196} 2197 2198static void 2199knlist_rw_assert_unlocked(void *arg) 2200{ 2201 2202 rw_assert((struct rwlock *)arg, RA_UNLOCKED); 2203} 2204 2205void 2206knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 2207 void (*kl_unlock)(void *), 2208 void (*kl_assert_locked)(void *), void (*kl_assert_unlocked)(void *)) 2209{ 2210 2211 if (lock == NULL) 2212 knl->kl_lockarg = &knlist_lock; 2213 else 2214 knl->kl_lockarg = lock; 2215 2216 if (kl_lock == NULL) 2217 knl->kl_lock = knlist_mtx_lock; 2218 else 2219 knl->kl_lock = kl_lock; 2220 if (kl_unlock == NULL) 2221 knl->kl_unlock = knlist_mtx_unlock; 2222 else 2223 knl->kl_unlock = kl_unlock; 2224 if (kl_assert_locked == NULL) 2225 knl->kl_assert_locked = knlist_mtx_assert_locked; 2226 else 2227 knl->kl_assert_locked = kl_assert_locked; 2228 if (kl_assert_unlocked == NULL) 2229 knl->kl_assert_unlocked = knlist_mtx_assert_unlocked; 2230 else 2231 knl->kl_assert_unlocked = kl_assert_unlocked; 2232 2233 knl->kl_autodestroy = 0; 2234 SLIST_INIT(&knl->kl_list); 2235} 2236 2237void 2238knlist_init_mtx(struct knlist *knl, struct mtx *lock) 2239{ 2240 2241 knlist_init(knl, lock, NULL, NULL, NULL, NULL); 2242} 2243 2244struct knlist * 2245knlist_alloc(struct mtx *lock) 2246{ 2247 struct knlist *knl; 2248 2249 knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK); 2250 knlist_init_mtx(knl, lock); 2251 return (knl); 2252} 2253 2254void 2255knlist_init_rw_reader(struct knlist *knl, struct rwlock *lock) 2256{ 2257 2258 knlist_init(knl, lock, knlist_rw_rlock, knlist_rw_runlock, 2259 knlist_rw_assert_locked, knlist_rw_assert_unlocked); 2260} 2261 2262void 2263knlist_destroy(struct knlist *knl) 2264{ 2265 2266 KASSERT(KNLIST_EMPTY(knl), 2267 ("destroying knlist %p with knotes on it", knl)); 2268} 2269 2270void 2271knlist_detach(struct knlist *knl) 2272{ 2273 2274 KNL_ASSERT_LOCKED(knl); 2275 knl->kl_autodestroy = 1; 2276 if (knlist_empty(knl)) { 2277 knlist_destroy(knl); 2278 free(knl, M_KQUEUE); 2279 } 2280} 2281 2282/* 2283 * Even if we are locked, we may need to drop the lock to allow any influx 2284 * knotes time to "settle". 2285 */ 2286void 2287knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 2288{ 2289 struct knote *kn, *kn2; 2290 struct kqueue *kq; 2291 2292 KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl)); 2293 if (islocked) 2294 KNL_ASSERT_LOCKED(knl); 2295 else { 2296 KNL_ASSERT_UNLOCKED(knl); 2297again: /* need to reacquire lock since we have dropped it */ 2298 knl->kl_lock(knl->kl_lockarg); 2299 } 2300 2301 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 2302 kq = kn->kn_kq; 2303 KQ_LOCK(kq); 2304 if ((kn->kn_status & KN_INFLUX)) { 2305 KQ_UNLOCK(kq); 2306 continue; 2307 } 2308 knlist_remove_kq(knl, kn, 1, 1); 2309 if (killkn) { 2310 kn->kn_status |= KN_INFLUX | KN_DETACHED; 2311 KQ_UNLOCK(kq); 2312 knote_drop(kn, td); 2313 } else { 2314 /* Make sure cleared knotes disappear soon */ 2315 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 2316 KQ_UNLOCK(kq); 2317 } 2318 kq = NULL; 2319 } 2320 2321 if (!SLIST_EMPTY(&knl->kl_list)) { 2322 /* there are still KN_INFLUX remaining */ 2323 kn = SLIST_FIRST(&knl->kl_list); 2324 kq = kn->kn_kq; 2325 KQ_LOCK(kq); 2326 KASSERT(kn->kn_status & KN_INFLUX, 2327 ("knote removed w/o list lock")); 2328 knl->kl_unlock(knl->kl_lockarg); 2329 kq->kq_state |= KQ_FLUXWAIT; 2330 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 2331 kq = NULL; 2332 goto again; 2333 } 2334 2335 if (islocked) 2336 KNL_ASSERT_LOCKED(knl); 2337 else { 2338 knl->kl_unlock(knl->kl_lockarg); 2339 KNL_ASSERT_UNLOCKED(knl); 2340 } 2341} 2342 2343/* 2344 * Remove all knotes referencing a specified fd must be called with FILEDESC 2345 * lock. This prevents a race where a new fd comes along and occupies the 2346 * entry and we attach a knote to the fd. 2347 */ 2348void 2349knote_fdclose(struct thread *td, int fd) 2350{ 2351 struct filedesc *fdp = td->td_proc->p_fd; 2352 struct kqueue *kq; 2353 struct knote *kn; 2354 int influx; 2355 2356 FILEDESC_XLOCK_ASSERT(fdp); 2357 2358 /* 2359 * We shouldn't have to worry about new kevents appearing on fd 2360 * since filedesc is locked. 2361 */ 2362 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2363 KQ_LOCK(kq); 2364 2365again: 2366 influx = 0; 2367 while (kq->kq_knlistsize > fd && 2368 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2369 if (kn->kn_status & KN_INFLUX) { 2370 /* someone else might be waiting on our knote */ 2371 if (influx) 2372 wakeup(kq); 2373 kq->kq_state |= KQ_FLUXWAIT; 2374 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2375 goto again; 2376 } 2377 kn->kn_status |= KN_INFLUX; 2378 KQ_UNLOCK(kq); 2379 if (!(kn->kn_status & KN_DETACHED)) 2380 kn->kn_fop->f_detach(kn); 2381 knote_drop(kn, td); 2382 influx = 1; 2383 KQ_LOCK(kq); 2384 } 2385 KQ_UNLOCK_FLUX(kq); 2386 } 2387} 2388 2389static int 2390knote_attach(struct knote *kn, struct kqueue *kq) 2391{ 2392 struct klist *list; 2393 2394 KASSERT(kn->kn_status & KN_INFLUX, ("knote not marked INFLUX")); 2395 KQ_OWNED(kq); 2396 2397 if (kn->kn_fop->f_isfd) { 2398 if (kn->kn_id >= kq->kq_knlistsize) 2399 return (ENOMEM); 2400 list = &kq->kq_knlist[kn->kn_id]; 2401 } else { 2402 if (kq->kq_knhash == NULL) 2403 return (ENOMEM); 2404 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2405 } 2406 SLIST_INSERT_HEAD(list, kn, kn_link); 2407 return (0); 2408} 2409 2410/* 2411 * knote must already have been detached using the f_detach method. 2412 * no lock need to be held, it is assumed that the KN_INFLUX flag is set 2413 * to prevent other removal. 2414 */ 2415static void 2416knote_drop(struct knote *kn, struct thread *td) 2417{ 2418 struct kqueue *kq; 2419 struct klist *list; 2420 2421 kq = kn->kn_kq; 2422 2423 KQ_NOTOWNED(kq); 2424 KASSERT((kn->kn_status & KN_INFLUX) == KN_INFLUX, 2425 ("knote_drop called without KN_INFLUX set in kn_status")); 2426 2427 KQ_LOCK(kq); 2428 if (kn->kn_fop->f_isfd) 2429 list = &kq->kq_knlist[kn->kn_id]; 2430 else 2431 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2432 2433 if (!SLIST_EMPTY(list)) 2434 SLIST_REMOVE(list, kn, knote, kn_link); 2435 if (kn->kn_status & KN_QUEUED) 2436 knote_dequeue(kn); 2437 KQ_UNLOCK_FLUX(kq); 2438 2439 if (kn->kn_fop->f_isfd) { 2440 fdrop(kn->kn_fp, td); 2441 kn->kn_fp = NULL; 2442 } 2443 kqueue_fo_release(kn->kn_kevent.filter); 2444 kn->kn_fop = NULL; 2445 knote_free(kn); 2446} 2447 2448static void 2449knote_enqueue(struct knote *kn) 2450{ 2451 struct kqueue *kq = kn->kn_kq; 2452 2453 KQ_OWNED(kn->kn_kq); 2454 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2455 2456 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2457 kn->kn_status |= KN_QUEUED; 2458 kq->kq_count++; 2459 kqueue_wakeup(kq); 2460} 2461 2462static void 2463knote_dequeue(struct knote *kn) 2464{ 2465 struct kqueue *kq = kn->kn_kq; 2466 2467 KQ_OWNED(kn->kn_kq); 2468 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2469 2470 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2471 kn->kn_status &= ~KN_QUEUED; 2472 kq->kq_count--; 2473} 2474 2475static void 2476knote_init(void) 2477{ 2478 2479 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2480 NULL, NULL, UMA_ALIGN_PTR, 0); 2481} 2482SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2483 2484static struct knote * 2485knote_alloc(int waitok) 2486{ 2487 2488 return (uma_zalloc(knote_zone, (waitok ? M_WAITOK : M_NOWAIT) | 2489 M_ZERO)); 2490} 2491 2492static void 2493knote_free(struct knote *kn) 2494{ 2495 2496 uma_zfree(knote_zone, kn); 2497} 2498 2499/* 2500 * Register the kev w/ the kq specified by fd. 2501 */ 2502int 2503kqfd_register(int fd, struct kevent *kev, struct thread *td, int waitok) 2504{ 2505 struct kqueue *kq; 2506 struct file *fp; 2507 cap_rights_t rights; 2508 int error; 2509 2510 error = fget(td, fd, cap_rights_init(&rights, CAP_KQUEUE_CHANGE), &fp); 2511 if (error != 0) 2512 return (error); 2513 if ((error = kqueue_acquire(fp, &kq)) != 0) 2514 goto noacquire; 2515 2516 error = kqueue_register(kq, kev, td, waitok); 2517 kqueue_release(kq, 0); 2518 2519noacquire: 2520 fdrop(fp, td); 2521 return (error); 2522} 2523