1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 5 * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org> 6 * Copyright (c) 2009 Apple, Inc. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31 #include <sys/cdefs.h> 32 #include "opt_ktrace.h" 33 #include "opt_kqueue.h" 34 35 #ifdef COMPAT_FREEBSD11 36 #define _WANT_FREEBSD11_KEVENT 37 #endif 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/capsicum.h> 42 #include <sys/kernel.h> 43 #include <sys/limits.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/proc.h> 47 #include <sys/malloc.h> 48 #include <sys/unistd.h> 49 #include <sys/file.h> 50 #include <sys/filedesc.h> 51 #include <sys/filio.h> 52 #include <sys/fcntl.h> 53 #include <sys/kthread.h> 54 #include <sys/selinfo.h> 55 #include <sys/queue.h> 56 #include <sys/event.h> 57 #include <sys/eventvar.h> 58 #include <sys/poll.h> 59 #include <sys/protosw.h> 60 #include <sys/resourcevar.h> 61 #include <sys/sbuf.h> 62 #include <sys/sigio.h> 63 #include <sys/signalvar.h> 64 #include <sys/socket.h> 65 #include <sys/socketvar.h> 66 #include <sys/stat.h> 67 #include <sys/sysctl.h> 68 #include <sys/sysent.h> 69 #include <sys/sysproto.h> 70 #include <sys/syscallsubr.h> 71 #include <sys/taskqueue.h> 72 #include <sys/uio.h> 73 #include <sys/user.h> 74 #ifdef KTRACE 75 #include <sys/ktrace.h> 76 #endif 77 #include <machine/atomic.h> 78 #ifdef COMPAT_FREEBSD32 79 #include <compat/freebsd32/freebsd32.h> 80 #include <compat/freebsd32/freebsd32_util.h> 81 #endif 82 83 #include <vm/uma.h> 84 85 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 86 87 /* 88 * This lock is used if multiple kq locks are required. This possibly 89 * should be made into a per proc lock. 90 */ 91 static struct mtx kq_global; 92 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 93 #define KQ_GLOBAL_LOCK(lck, haslck) do { \ 94 if (!haslck) \ 95 mtx_lock(lck); \ 96 haslck = 1; \ 97 } while (0) 98 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 99 if (haslck) \ 100 mtx_unlock(lck); \ 101 haslck = 0; \ 102 } while (0) 103 104 TASKQUEUE_DEFINE_THREAD(kqueue_ctx); 105 106 static int kevent_copyout(void *arg, struct kevent *kevp, int count); 107 static int kevent_copyin(void *arg, struct kevent *kevp, int count); 108 static int kqueue_register(struct kqueue *kq, struct kevent *kev, 109 struct thread *td, int mflag); 110 static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 111 static void kqueue_release(struct kqueue *kq, int locked); 112 static void kqueue_destroy(struct kqueue *kq); 113 static void kqueue_drain(struct kqueue *kq, struct thread *td); 114 static int kqueue_expand(struct kqueue *kq, const struct filterops *fops, 115 uintptr_t ident, int mflag); 116 static void kqueue_task(void *arg, int pending); 117 static int kqueue_scan(struct kqueue *kq, int maxevents, 118 struct kevent_copyops *k_ops, 119 const struct timespec *timeout, 120 struct kevent *keva, struct thread *td); 121 static void kqueue_wakeup(struct kqueue *kq); 122 static const struct filterops *kqueue_fo_find(int filt); 123 static void kqueue_fo_release(int filt); 124 struct g_kevent_args; 125 static int kern_kevent_generic(struct thread *td, 126 struct g_kevent_args *uap, 127 struct kevent_copyops *k_ops, const char *struct_name); 128 129 static fo_ioctl_t kqueue_ioctl; 130 static fo_poll_t kqueue_poll; 131 static fo_kqfilter_t kqueue_kqfilter; 132 static fo_stat_t kqueue_stat; 133 static fo_close_t kqueue_close; 134 static fo_fill_kinfo_t kqueue_fill_kinfo; 135 136 static const struct fileops kqueueops = { 137 .fo_read = invfo_rdwr, 138 .fo_write = invfo_rdwr, 139 .fo_truncate = invfo_truncate, 140 .fo_ioctl = kqueue_ioctl, 141 .fo_poll = kqueue_poll, 142 .fo_kqfilter = kqueue_kqfilter, 143 .fo_stat = kqueue_stat, 144 .fo_close = kqueue_close, 145 .fo_chmod = invfo_chmod, 146 .fo_chown = invfo_chown, 147 .fo_sendfile = invfo_sendfile, 148 .fo_cmp = file_kcmp_generic, 149 .fo_fill_kinfo = kqueue_fill_kinfo, 150 }; 151 152 static int knote_attach(struct knote *kn, struct kqueue *kq); 153 static void knote_drop(struct knote *kn, struct thread *td); 154 static void knote_drop_detached(struct knote *kn, struct thread *td); 155 static void knote_enqueue(struct knote *kn); 156 static void knote_dequeue(struct knote *kn); 157 static void knote_init(void); 158 static struct knote *knote_alloc(int mflag); 159 static void knote_free(struct knote *kn); 160 161 static void filt_kqdetach(struct knote *kn); 162 static int filt_kqueue(struct knote *kn, long hint); 163 static int filt_procattach(struct knote *kn); 164 static void filt_procdetach(struct knote *kn); 165 static int filt_proc(struct knote *kn, long hint); 166 static int filt_fileattach(struct knote *kn); 167 static void filt_timerexpire(void *knx); 168 static void filt_timerexpire_l(struct knote *kn, bool proc_locked); 169 static int filt_timerattach(struct knote *kn); 170 static void filt_timerdetach(struct knote *kn); 171 static void filt_timerstart(struct knote *kn, sbintime_t to); 172 static void filt_timertouch(struct knote *kn, struct kevent *kev, 173 u_long type); 174 static int filt_timervalidate(struct knote *kn, sbintime_t *to); 175 static int filt_timer(struct knote *kn, long hint); 176 static int filt_userattach(struct knote *kn); 177 static void filt_userdetach(struct knote *kn); 178 static int filt_user(struct knote *kn, long hint); 179 static void filt_usertouch(struct knote *kn, struct kevent *kev, 180 u_long type); 181 182 static const struct filterops file_filtops = { 183 .f_isfd = 1, 184 .f_attach = filt_fileattach, 185 }; 186 static const struct filterops kqread_filtops = { 187 .f_isfd = 1, 188 .f_detach = filt_kqdetach, 189 .f_event = filt_kqueue, 190 }; 191 /* XXX - move to kern_proc.c? */ 192 static const struct filterops proc_filtops = { 193 .f_isfd = 0, 194 .f_attach = filt_procattach, 195 .f_detach = filt_procdetach, 196 .f_event = filt_proc, 197 }; 198 static const struct filterops timer_filtops = { 199 .f_isfd = 0, 200 .f_attach = filt_timerattach, 201 .f_detach = filt_timerdetach, 202 .f_event = filt_timer, 203 .f_touch = filt_timertouch, 204 }; 205 static const struct filterops user_filtops = { 206 .f_attach = filt_userattach, 207 .f_detach = filt_userdetach, 208 .f_event = filt_user, 209 .f_touch = filt_usertouch, 210 }; 211 212 static uma_zone_t knote_zone; 213 static unsigned int __exclusive_cache_line kq_ncallouts; 214 static unsigned int kq_calloutmax = 4 * 1024; 215 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 216 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 217 218 /* XXX - ensure not influx ? */ 219 #define KNOTE_ACTIVATE(kn, islock) do { \ 220 if ((islock)) \ 221 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 222 else \ 223 KQ_LOCK((kn)->kn_kq); \ 224 (kn)->kn_status |= KN_ACTIVE; \ 225 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 226 knote_enqueue((kn)); \ 227 if (!(islock)) \ 228 KQ_UNLOCK((kn)->kn_kq); \ 229 } while (0) 230 #define KQ_LOCK(kq) do { \ 231 mtx_lock(&(kq)->kq_lock); \ 232 } while (0) 233 #define KQ_FLUX_WAKEUP(kq) do { \ 234 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 235 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 236 wakeup((kq)); \ 237 } \ 238 } while (0) 239 #define KQ_UNLOCK_FLUX(kq) do { \ 240 KQ_FLUX_WAKEUP(kq); \ 241 mtx_unlock(&(kq)->kq_lock); \ 242 } while (0) 243 #define KQ_UNLOCK(kq) do { \ 244 mtx_unlock(&(kq)->kq_lock); \ 245 } while (0) 246 #define KQ_OWNED(kq) do { \ 247 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 248 } while (0) 249 #define KQ_NOTOWNED(kq) do { \ 250 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 251 } while (0) 252 253 static struct knlist * 254 kn_list_lock(struct knote *kn) 255 { 256 struct knlist *knl; 257 258 knl = kn->kn_knlist; 259 if (knl != NULL) 260 knl->kl_lock(knl->kl_lockarg); 261 return (knl); 262 } 263 264 static void 265 kn_list_unlock(struct knlist *knl) 266 { 267 bool do_free; 268 269 if (knl == NULL) 270 return; 271 do_free = knl->kl_autodestroy && knlist_empty(knl); 272 knl->kl_unlock(knl->kl_lockarg); 273 if (do_free) { 274 knlist_destroy(knl); 275 free(knl, M_KQUEUE); 276 } 277 } 278 279 static bool 280 kn_in_flux(struct knote *kn) 281 { 282 283 return (kn->kn_influx > 0); 284 } 285 286 static void 287 kn_enter_flux(struct knote *kn) 288 { 289 290 KQ_OWNED(kn->kn_kq); 291 MPASS(kn->kn_influx < INT_MAX); 292 kn->kn_influx++; 293 } 294 295 static bool 296 kn_leave_flux(struct knote *kn) 297 { 298 299 KQ_OWNED(kn->kn_kq); 300 MPASS(kn->kn_influx > 0); 301 kn->kn_influx--; 302 return (kn->kn_influx == 0); 303 } 304 305 #define KNL_ASSERT_LOCK(knl, islocked) do { \ 306 if (islocked) \ 307 KNL_ASSERT_LOCKED(knl); \ 308 else \ 309 KNL_ASSERT_UNLOCKED(knl); \ 310 } while (0) 311 #ifdef INVARIANTS 312 #define KNL_ASSERT_LOCKED(knl) do { \ 313 knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED); \ 314 } while (0) 315 #define KNL_ASSERT_UNLOCKED(knl) do { \ 316 knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED); \ 317 } while (0) 318 #else /* !INVARIANTS */ 319 #define KNL_ASSERT_LOCKED(knl) do {} while (0) 320 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 321 #endif /* INVARIANTS */ 322 323 #ifndef KN_HASHSIZE 324 #define KN_HASHSIZE 64 /* XXX should be tunable */ 325 #endif 326 327 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 328 329 static int 330 filt_nullattach(struct knote *kn) 331 { 332 333 return (ENXIO); 334 }; 335 336 static const struct filterops null_filtops = { 337 .f_isfd = 0, 338 .f_attach = filt_nullattach, 339 }; 340 341 /* XXX - make SYSINIT to add these, and move into respective modules. */ 342 extern const struct filterops sig_filtops; 343 extern const struct filterops fs_filtops; 344 345 /* 346 * Table for all system-defined filters. 347 */ 348 static struct mtx filterops_lock; 349 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF); 350 static struct { 351 const struct filterops *for_fop; 352 int for_nolock; 353 int for_refcnt; 354 } sysfilt_ops[EVFILT_SYSCOUNT] = { 355 [~EVFILT_READ] = { &file_filtops, 1 }, 356 [~EVFILT_WRITE] = { &file_filtops, 1 }, 357 [~EVFILT_AIO] = { &null_filtops }, 358 [~EVFILT_VNODE] = { &file_filtops, 1 }, 359 [~EVFILT_PROC] = { &proc_filtops, 1 }, 360 [~EVFILT_SIGNAL] = { &sig_filtops, 1 }, 361 [~EVFILT_TIMER] = { &timer_filtops, 1 }, 362 [~EVFILT_PROCDESC] = { &file_filtops, 1 }, 363 [~EVFILT_FS] = { &fs_filtops, 1 }, 364 [~EVFILT_LIO] = { &null_filtops }, 365 [~EVFILT_USER] = { &user_filtops, 1 }, 366 [~EVFILT_SENDFILE] = { &null_filtops }, 367 [~EVFILT_EMPTY] = { &file_filtops, 1 }, 368 }; 369 370 /* 371 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 372 * method. 373 */ 374 static int 375 filt_fileattach(struct knote *kn) 376 { 377 378 return (fo_kqfilter(kn->kn_fp, kn)); 379 } 380 381 /*ARGSUSED*/ 382 static int 383 kqueue_kqfilter(struct file *fp, struct knote *kn) 384 { 385 struct kqueue *kq = kn->kn_fp->f_data; 386 387 if (kn->kn_filter != EVFILT_READ) 388 return (EINVAL); 389 390 kn->kn_status |= KN_KQUEUE; 391 kn->kn_fop = &kqread_filtops; 392 knlist_add(&kq->kq_sel.si_note, kn, 0); 393 394 return (0); 395 } 396 397 static void 398 filt_kqdetach(struct knote *kn) 399 { 400 struct kqueue *kq = kn->kn_fp->f_data; 401 402 knlist_remove(&kq->kq_sel.si_note, kn, 0); 403 } 404 405 /*ARGSUSED*/ 406 static int 407 filt_kqueue(struct knote *kn, long hint) 408 { 409 struct kqueue *kq = kn->kn_fp->f_data; 410 411 kn->kn_data = kq->kq_count; 412 return (kn->kn_data > 0); 413 } 414 415 /* XXX - move to kern_proc.c? */ 416 static int 417 filt_procattach(struct knote *kn) 418 { 419 struct proc *p; 420 int error; 421 bool exiting, immediate; 422 423 exiting = immediate = false; 424 if (kn->kn_sfflags & NOTE_EXIT) 425 p = pfind_any(kn->kn_id); 426 else 427 p = pfind(kn->kn_id); 428 if (p == NULL) 429 return (ESRCH); 430 if (p->p_flag & P_WEXIT) 431 exiting = true; 432 433 if ((error = p_cansee(curthread, p))) { 434 PROC_UNLOCK(p); 435 return (error); 436 } 437 438 kn->kn_ptr.p_proc = p; 439 kn->kn_flags |= EV_CLEAR; /* automatically set */ 440 441 /* 442 * Internal flag indicating registration done by kernel for the 443 * purposes of getting a NOTE_CHILD notification. 444 */ 445 if (kn->kn_flags & EV_FLAG2) { 446 kn->kn_flags &= ~EV_FLAG2; 447 kn->kn_data = kn->kn_sdata; /* ppid */ 448 kn->kn_fflags = NOTE_CHILD; 449 kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK); 450 immediate = true; /* Force immediate activation of child note. */ 451 } 452 /* 453 * Internal flag indicating registration done by kernel (for other than 454 * NOTE_CHILD). 455 */ 456 if (kn->kn_flags & EV_FLAG1) { 457 kn->kn_flags &= ~EV_FLAG1; 458 } 459 460 knlist_add(p->p_klist, kn, 1); 461 462 /* 463 * Immediately activate any child notes or, in the case of a zombie 464 * target process, exit notes. The latter is necessary to handle the 465 * case where the target process, e.g. a child, dies before the kevent 466 * is registered. 467 */ 468 if (immediate || (exiting && filt_proc(kn, NOTE_EXIT))) 469 KNOTE_ACTIVATE(kn, 0); 470 471 PROC_UNLOCK(p); 472 473 return (0); 474 } 475 476 /* 477 * The knote may be attached to a different process, which may exit, 478 * leaving nothing for the knote to be attached to. So when the process 479 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 480 * it will be deleted when read out. However, as part of the knote deletion, 481 * this routine is called, so a check is needed to avoid actually performing 482 * a detach, because the original process does not exist any more. 483 */ 484 /* XXX - move to kern_proc.c? */ 485 static void 486 filt_procdetach(struct knote *kn) 487 { 488 489 knlist_remove(kn->kn_knlist, kn, 0); 490 kn->kn_ptr.p_proc = NULL; 491 } 492 493 /* XXX - move to kern_proc.c? */ 494 static int 495 filt_proc(struct knote *kn, long hint) 496 { 497 struct proc *p; 498 u_int event; 499 500 p = kn->kn_ptr.p_proc; 501 if (p == NULL) /* already activated, from attach filter */ 502 return (0); 503 504 /* Mask off extra data. */ 505 event = (u_int)hint & NOTE_PCTRLMASK; 506 507 /* If the user is interested in this event, record it. */ 508 if (kn->kn_sfflags & event) 509 kn->kn_fflags |= event; 510 511 /* Process is gone, so flag the event as finished. */ 512 if (event == NOTE_EXIT) { 513 kn->kn_flags |= EV_EOF | EV_ONESHOT; 514 kn->kn_ptr.p_proc = NULL; 515 if (kn->kn_fflags & NOTE_EXIT) 516 kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig); 517 if (kn->kn_fflags == 0) 518 kn->kn_flags |= EV_DROP; 519 return (1); 520 } 521 522 return (kn->kn_fflags != 0); 523 } 524 525 /* 526 * Called when the process forked. It mostly does the same as the 527 * knote(), activating all knotes registered to be activated when the 528 * process forked. Additionally, for each knote attached to the 529 * parent, check whether user wants to track the new process. If so 530 * attach a new knote to it, and immediately report an event with the 531 * child's pid. 532 */ 533 void 534 knote_fork(struct knlist *list, int pid) 535 { 536 struct kqueue *kq; 537 struct knote *kn; 538 struct kevent kev; 539 int error; 540 541 MPASS(list != NULL); 542 KNL_ASSERT_LOCKED(list); 543 if (SLIST_EMPTY(&list->kl_list)) 544 return; 545 546 memset(&kev, 0, sizeof(kev)); 547 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 548 kq = kn->kn_kq; 549 KQ_LOCK(kq); 550 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 551 KQ_UNLOCK(kq); 552 continue; 553 } 554 555 /* 556 * The same as knote(), activate the event. 557 */ 558 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 559 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 560 KNOTE_ACTIVATE(kn, 1); 561 KQ_UNLOCK(kq); 562 continue; 563 } 564 565 /* 566 * The NOTE_TRACK case. In addition to the activation 567 * of the event, we need to register new events to 568 * track the child. Drop the locks in preparation for 569 * the call to kqueue_register(). 570 */ 571 kn_enter_flux(kn); 572 KQ_UNLOCK(kq); 573 list->kl_unlock(list->kl_lockarg); 574 575 /* 576 * Activate existing knote and register tracking knotes with 577 * new process. 578 * 579 * First register a knote to get just the child notice. This 580 * must be a separate note from a potential NOTE_EXIT 581 * notification since both NOTE_CHILD and NOTE_EXIT are defined 582 * to use the data field (in conflicting ways). 583 */ 584 kev.ident = pid; 585 kev.filter = kn->kn_filter; 586 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT | 587 EV_FLAG2; 588 kev.fflags = kn->kn_sfflags; 589 kev.data = kn->kn_id; /* parent */ 590 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 591 error = kqueue_register(kq, &kev, NULL, M_NOWAIT); 592 if (error) 593 kn->kn_fflags |= NOTE_TRACKERR; 594 595 /* 596 * Then register another knote to track other potential events 597 * from the new process. 598 */ 599 kev.ident = pid; 600 kev.filter = kn->kn_filter; 601 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 602 kev.fflags = kn->kn_sfflags; 603 kev.data = kn->kn_id; /* parent */ 604 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 605 error = kqueue_register(kq, &kev, NULL, M_NOWAIT); 606 if (error) 607 kn->kn_fflags |= NOTE_TRACKERR; 608 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 609 KNOTE_ACTIVATE(kn, 0); 610 list->kl_lock(list->kl_lockarg); 611 KQ_LOCK(kq); 612 kn_leave_flux(kn); 613 KQ_UNLOCK_FLUX(kq); 614 } 615 } 616 617 /* 618 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the 619 * interval timer support code. 620 */ 621 622 #define NOTE_TIMER_PRECMASK \ 623 (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS) 624 625 static sbintime_t 626 timer2sbintime(int64_t data, int flags) 627 { 628 int64_t secs; 629 630 /* 631 * Macros for converting to the fractional second portion of an 632 * sbintime_t using 64bit multiplication to improve precision. 633 */ 634 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32) 635 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32) 636 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32) 637 switch (flags & NOTE_TIMER_PRECMASK) { 638 case NOTE_SECONDS: 639 #ifdef __LP64__ 640 if (data > (SBT_MAX / SBT_1S)) 641 return (SBT_MAX); 642 #endif 643 return ((sbintime_t)data << 32); 644 case NOTE_MSECONDS: /* FALLTHROUGH */ 645 case 0: 646 if (data >= 1000) { 647 secs = data / 1000; 648 #ifdef __LP64__ 649 if (secs > (SBT_MAX / SBT_1S)) 650 return (SBT_MAX); 651 #endif 652 return (secs << 32 | MS_TO_SBT(data % 1000)); 653 } 654 return (MS_TO_SBT(data)); 655 case NOTE_USECONDS: 656 if (data >= 1000000) { 657 secs = data / 1000000; 658 #ifdef __LP64__ 659 if (secs > (SBT_MAX / SBT_1S)) 660 return (SBT_MAX); 661 #endif 662 return (secs << 32 | US_TO_SBT(data % 1000000)); 663 } 664 return (US_TO_SBT(data)); 665 case NOTE_NSECONDS: 666 if (data >= 1000000000) { 667 secs = data / 1000000000; 668 #ifdef __LP64__ 669 if (secs > (SBT_MAX / SBT_1S)) 670 return (SBT_MAX); 671 #endif 672 return (secs << 32 | NS_TO_SBT(data % 1000000000)); 673 } 674 return (NS_TO_SBT(data)); 675 default: 676 break; 677 } 678 return (-1); 679 } 680 681 struct kq_timer_cb_data { 682 struct callout c; 683 struct proc *p; 684 struct knote *kn; 685 int cpuid; 686 int flags; 687 TAILQ_ENTRY(kq_timer_cb_data) link; 688 sbintime_t next; /* next timer event fires at */ 689 sbintime_t to; /* precalculated timer period, 0 for abs */ 690 }; 691 692 #define KQ_TIMER_CB_ENQUEUED 0x01 693 694 static void 695 kqtimer_sched_callout(struct kq_timer_cb_data *kc) 696 { 697 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn, 698 kc->cpuid, C_ABSOLUTE); 699 } 700 701 void 702 kqtimer_proc_continue(struct proc *p) 703 { 704 struct kq_timer_cb_data *kc, *kc1; 705 struct bintime bt; 706 sbintime_t now; 707 708 PROC_LOCK_ASSERT(p, MA_OWNED); 709 710 getboottimebin(&bt); 711 now = bttosbt(bt); 712 713 TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) { 714 TAILQ_REMOVE(&p->p_kqtim_stop, kc, link); 715 kc->flags &= ~KQ_TIMER_CB_ENQUEUED; 716 if (kc->next <= now) 717 filt_timerexpire_l(kc->kn, true); 718 else 719 kqtimer_sched_callout(kc); 720 } 721 } 722 723 static void 724 filt_timerexpire_l(struct knote *kn, bool proc_locked) 725 { 726 struct kq_timer_cb_data *kc; 727 struct proc *p; 728 uint64_t delta; 729 sbintime_t now; 730 731 kc = kn->kn_ptr.p_v; 732 733 if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) { 734 kn->kn_data++; 735 KNOTE_ACTIVATE(kn, 0); 736 return; 737 } 738 739 now = sbinuptime(); 740 if (now >= kc->next) { 741 delta = (now - kc->next) / kc->to; 742 if (delta == 0) 743 delta = 1; 744 kn->kn_data += delta; 745 kc->next += delta * kc->to; 746 if (now >= kc->next) /* overflow */ 747 kc->next = now + kc->to; 748 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 749 } 750 751 /* 752 * Initial check for stopped kc->p is racy. It is fine to 753 * miss the set of the stop flags, at worst we would schedule 754 * one more callout. On the other hand, it is not fine to not 755 * schedule when we we missed clearing of the flags, we 756 * recheck them under the lock and observe consistent state. 757 */ 758 p = kc->p; 759 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 760 if (!proc_locked) 761 PROC_LOCK(p); 762 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 763 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) { 764 kc->flags |= KQ_TIMER_CB_ENQUEUED; 765 TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link); 766 } 767 if (!proc_locked) 768 PROC_UNLOCK(p); 769 return; 770 } 771 if (!proc_locked) 772 PROC_UNLOCK(p); 773 } 774 kqtimer_sched_callout(kc); 775 } 776 777 static void 778 filt_timerexpire(void *knx) 779 { 780 filt_timerexpire_l(knx, false); 781 } 782 783 /* 784 * data contains amount of time to sleep 785 */ 786 static int 787 filt_timervalidate(struct knote *kn, sbintime_t *to) 788 { 789 struct bintime bt; 790 sbintime_t sbt; 791 792 if (kn->kn_sdata < 0) 793 return (EINVAL); 794 if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) 795 kn->kn_sdata = 1; 796 /* 797 * The only fflags values supported are the timer unit 798 * (precision) and the absolute time indicator. 799 */ 800 if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0) 801 return (EINVAL); 802 803 *to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); 804 if (*to < 0) 805 return (EINVAL); 806 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 807 getboottimebin(&bt); 808 sbt = bttosbt(bt); 809 *to = MAX(0, *to - sbt); 810 } 811 return (0); 812 } 813 814 static int 815 filt_timerattach(struct knote *kn) 816 { 817 struct kq_timer_cb_data *kc; 818 sbintime_t to; 819 int error; 820 821 to = -1; 822 error = filt_timervalidate(kn, &to); 823 if (error != 0) 824 return (error); 825 KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 || 826 (kn->kn_sfflags & NOTE_ABSTIME) != 0, 827 ("%s: periodic timer has a calculated zero timeout", __func__)); 828 KASSERT(to >= 0, 829 ("%s: timer has a calculated negative timeout", __func__)); 830 831 if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) { 832 atomic_subtract_int(&kq_ncallouts, 1); 833 return (ENOMEM); 834 } 835 836 if ((kn->kn_sfflags & NOTE_ABSTIME) == 0) 837 kn->kn_flags |= EV_CLEAR; /* automatically set */ 838 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ 839 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK); 840 kc->kn = kn; 841 kc->p = curproc; 842 kc->cpuid = PCPU_GET(cpuid); 843 kc->flags = 0; 844 callout_init(&kc->c, 1); 845 filt_timerstart(kn, to); 846 847 return (0); 848 } 849 850 static void 851 filt_timerstart(struct knote *kn, sbintime_t to) 852 { 853 struct kq_timer_cb_data *kc; 854 855 kc = kn->kn_ptr.p_v; 856 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 857 kc->next = to; 858 kc->to = 0; 859 } else { 860 kc->next = to + sbinuptime(); 861 kc->to = to; 862 } 863 kqtimer_sched_callout(kc); 864 } 865 866 static void 867 filt_timerdetach(struct knote *kn) 868 { 869 struct kq_timer_cb_data *kc; 870 unsigned int old __unused; 871 bool pending; 872 873 kc = kn->kn_ptr.p_v; 874 do { 875 callout_drain(&kc->c); 876 877 /* 878 * kqtimer_proc_continue() might have rescheduled this callout. 879 * Double-check, using the process mutex as an interlock. 880 */ 881 PROC_LOCK(kc->p); 882 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) { 883 kc->flags &= ~KQ_TIMER_CB_ENQUEUED; 884 TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link); 885 } 886 pending = callout_pending(&kc->c); 887 PROC_UNLOCK(kc->p); 888 } while (pending); 889 free(kc, M_KQUEUE); 890 old = atomic_fetchadd_int(&kq_ncallouts, -1); 891 KASSERT(old > 0, ("Number of callouts cannot become negative")); 892 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ 893 } 894 895 static void 896 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type) 897 { 898 struct kq_timer_cb_data *kc; 899 struct kqueue *kq; 900 sbintime_t to; 901 int error; 902 903 switch (type) { 904 case EVENT_REGISTER: 905 /* Handle re-added timers that update data/fflags */ 906 if (kev->flags & EV_ADD) { 907 kc = kn->kn_ptr.p_v; 908 909 /* Drain any existing callout. */ 910 callout_drain(&kc->c); 911 912 /* Throw away any existing undelivered record 913 * of the timer expiration. This is done under 914 * the presumption that if a process is 915 * re-adding this timer with new parameters, 916 * it is no longer interested in what may have 917 * happened under the old parameters. If it is 918 * interested, it can wait for the expiration, 919 * delete the old timer definition, and then 920 * add the new one. 921 * 922 * This has to be done while the kq is locked: 923 * - if enqueued, dequeue 924 * - make it no longer active 925 * - clear the count of expiration events 926 */ 927 kq = kn->kn_kq; 928 KQ_LOCK(kq); 929 if (kn->kn_status & KN_QUEUED) 930 knote_dequeue(kn); 931 932 kn->kn_status &= ~KN_ACTIVE; 933 kn->kn_data = 0; 934 KQ_UNLOCK(kq); 935 936 /* Reschedule timer based on new data/fflags */ 937 kn->kn_sfflags = kev->fflags; 938 kn->kn_sdata = kev->data; 939 error = filt_timervalidate(kn, &to); 940 if (error != 0) { 941 kn->kn_flags |= EV_ERROR; 942 kn->kn_data = error; 943 } else 944 filt_timerstart(kn, to); 945 } 946 break; 947 948 case EVENT_PROCESS: 949 *kev = kn->kn_kevent; 950 if (kn->kn_flags & EV_CLEAR) { 951 kn->kn_data = 0; 952 kn->kn_fflags = 0; 953 } 954 break; 955 956 default: 957 panic("filt_timertouch() - invalid type (%ld)", type); 958 break; 959 } 960 } 961 962 static int 963 filt_timer(struct knote *kn, long hint) 964 { 965 966 return (kn->kn_data != 0); 967 } 968 969 static int 970 filt_userattach(struct knote *kn) 971 { 972 973 /* 974 * EVFILT_USER knotes are not attached to anything in the kernel. 975 */ 976 kn->kn_hook = NULL; 977 if (kn->kn_fflags & NOTE_TRIGGER) 978 kn->kn_hookid = 1; 979 else 980 kn->kn_hookid = 0; 981 return (0); 982 } 983 984 static void 985 filt_userdetach(__unused struct knote *kn) 986 { 987 988 /* 989 * EVFILT_USER knotes are not attached to anything in the kernel. 990 */ 991 } 992 993 static int 994 filt_user(struct knote *kn, __unused long hint) 995 { 996 997 return (kn->kn_hookid); 998 } 999 1000 static void 1001 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 1002 { 1003 u_int ffctrl; 1004 1005 switch (type) { 1006 case EVENT_REGISTER: 1007 if (kev->fflags & NOTE_TRIGGER) 1008 kn->kn_hookid = 1; 1009 1010 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 1011 kev->fflags &= NOTE_FFLAGSMASK; 1012 switch (ffctrl) { 1013 case NOTE_FFNOP: 1014 break; 1015 1016 case NOTE_FFAND: 1017 kn->kn_sfflags &= kev->fflags; 1018 break; 1019 1020 case NOTE_FFOR: 1021 kn->kn_sfflags |= kev->fflags; 1022 break; 1023 1024 case NOTE_FFCOPY: 1025 kn->kn_sfflags = kev->fflags; 1026 break; 1027 1028 default: 1029 /* XXX Return error? */ 1030 break; 1031 } 1032 kn->kn_sdata = kev->data; 1033 if (kev->flags & EV_CLEAR) { 1034 kn->kn_hookid = 0; 1035 kn->kn_data = 0; 1036 kn->kn_fflags = 0; 1037 } 1038 break; 1039 1040 case EVENT_PROCESS: 1041 *kev = kn->kn_kevent; 1042 kev->fflags = kn->kn_sfflags; 1043 kev->data = kn->kn_sdata; 1044 if (kn->kn_flags & EV_CLEAR) { 1045 kn->kn_hookid = 0; 1046 kn->kn_data = 0; 1047 kn->kn_fflags = 0; 1048 } 1049 break; 1050 1051 default: 1052 panic("filt_usertouch() - invalid type (%ld)", type); 1053 break; 1054 } 1055 } 1056 1057 int 1058 sys_kqueue(struct thread *td, struct kqueue_args *uap) 1059 { 1060 1061 return (kern_kqueue(td, 0, NULL)); 1062 } 1063 1064 int 1065 sys_kqueuex(struct thread *td, struct kqueuex_args *uap) 1066 { 1067 int flags; 1068 1069 if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0) 1070 return (EINVAL); 1071 flags = 0; 1072 if ((uap->flags & KQUEUE_CLOEXEC) != 0) 1073 flags |= O_CLOEXEC; 1074 return (kern_kqueue(td, flags, NULL)); 1075 } 1076 1077 static void 1078 kqueue_init(struct kqueue *kq) 1079 { 1080 1081 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK); 1082 TAILQ_INIT(&kq->kq_head); 1083 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 1084 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 1085 } 1086 1087 int 1088 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps) 1089 { 1090 struct filedesc *fdp; 1091 struct kqueue *kq; 1092 struct file *fp; 1093 struct ucred *cred; 1094 int fd, error; 1095 1096 fdp = td->td_proc->p_fd; 1097 cred = td->td_ucred; 1098 if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) 1099 return (ENOMEM); 1100 1101 error = falloc_caps(td, &fp, &fd, flags, fcaps); 1102 if (error != 0) { 1103 chgkqcnt(cred->cr_ruidinfo, -1, 0); 1104 return (error); 1105 } 1106 1107 /* An extra reference on `fp' has been held for us by falloc(). */ 1108 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 1109 kqueue_init(kq); 1110 kq->kq_fdp = fdp; 1111 kq->kq_cred = crhold(cred); 1112 1113 FILEDESC_XLOCK(fdp); 1114 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 1115 FILEDESC_XUNLOCK(fdp); 1116 1117 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 1118 fdrop(fp, td); 1119 1120 td->td_retval[0] = fd; 1121 return (0); 1122 } 1123 1124 struct g_kevent_args { 1125 int fd; 1126 const void *changelist; 1127 int nchanges; 1128 void *eventlist; 1129 int nevents; 1130 const struct timespec *timeout; 1131 }; 1132 1133 int 1134 sys_kevent(struct thread *td, struct kevent_args *uap) 1135 { 1136 struct kevent_copyops k_ops = { 1137 .arg = uap, 1138 .k_copyout = kevent_copyout, 1139 .k_copyin = kevent_copyin, 1140 .kevent_size = sizeof(struct kevent), 1141 }; 1142 struct g_kevent_args gk_args = { 1143 .fd = uap->fd, 1144 .changelist = uap->changelist, 1145 .nchanges = uap->nchanges, 1146 .eventlist = uap->eventlist, 1147 .nevents = uap->nevents, 1148 .timeout = uap->timeout, 1149 }; 1150 1151 return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent")); 1152 } 1153 1154 static int 1155 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, 1156 struct kevent_copyops *k_ops, const char *struct_name) 1157 { 1158 struct timespec ts, *tsp; 1159 #ifdef KTRACE 1160 struct kevent *eventlist = uap->eventlist; 1161 #endif 1162 int error; 1163 1164 if (uap->timeout != NULL) { 1165 error = copyin(uap->timeout, &ts, sizeof(ts)); 1166 if (error) 1167 return (error); 1168 tsp = &ts; 1169 } else 1170 tsp = NULL; 1171 1172 #ifdef KTRACE 1173 if (KTRPOINT(td, KTR_STRUCT_ARRAY)) 1174 ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist, 1175 uap->nchanges, k_ops->kevent_size); 1176 #endif 1177 1178 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 1179 k_ops, tsp); 1180 1181 #ifdef KTRACE 1182 if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) 1183 ktrstructarray(struct_name, UIO_USERSPACE, eventlist, 1184 td->td_retval[0], k_ops->kevent_size); 1185 #endif 1186 1187 return (error); 1188 } 1189 1190 /* 1191 * Copy 'count' items into the destination list pointed to by uap->eventlist. 1192 */ 1193 static int 1194 kevent_copyout(void *arg, struct kevent *kevp, int count) 1195 { 1196 struct kevent_args *uap; 1197 int error; 1198 1199 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1200 uap = (struct kevent_args *)arg; 1201 1202 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 1203 if (error == 0) 1204 uap->eventlist += count; 1205 return (error); 1206 } 1207 1208 /* 1209 * Copy 'count' items from the list pointed to by uap->changelist. 1210 */ 1211 static int 1212 kevent_copyin(void *arg, struct kevent *kevp, int count) 1213 { 1214 struct kevent_args *uap; 1215 int error; 1216 1217 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1218 uap = (struct kevent_args *)arg; 1219 1220 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 1221 if (error == 0) 1222 uap->changelist += count; 1223 return (error); 1224 } 1225 1226 #ifdef COMPAT_FREEBSD11 1227 static int 1228 kevent11_copyout(void *arg, struct kevent *kevp, int count) 1229 { 1230 struct freebsd11_kevent_args *uap; 1231 struct freebsd11_kevent kev11; 1232 int error, i; 1233 1234 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1235 uap = (struct freebsd11_kevent_args *)arg; 1236 1237 for (i = 0; i < count; i++) { 1238 kev11.ident = kevp->ident; 1239 kev11.filter = kevp->filter; 1240 kev11.flags = kevp->flags; 1241 kev11.fflags = kevp->fflags; 1242 kev11.data = kevp->data; 1243 kev11.udata = kevp->udata; 1244 error = copyout(&kev11, uap->eventlist, sizeof(kev11)); 1245 if (error != 0) 1246 break; 1247 uap->eventlist++; 1248 kevp++; 1249 } 1250 return (error); 1251 } 1252 1253 /* 1254 * Copy 'count' items from the list pointed to by uap->changelist. 1255 */ 1256 static int 1257 kevent11_copyin(void *arg, struct kevent *kevp, int count) 1258 { 1259 struct freebsd11_kevent_args *uap; 1260 struct freebsd11_kevent kev11; 1261 int error, i; 1262 1263 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1264 uap = (struct freebsd11_kevent_args *)arg; 1265 1266 for (i = 0; i < count; i++) { 1267 error = copyin(uap->changelist, &kev11, sizeof(kev11)); 1268 if (error != 0) 1269 break; 1270 kevp->ident = kev11.ident; 1271 kevp->filter = kev11.filter; 1272 kevp->flags = kev11.flags; 1273 kevp->fflags = kev11.fflags; 1274 kevp->data = (uintptr_t)kev11.data; 1275 kevp->udata = kev11.udata; 1276 bzero(&kevp->ext, sizeof(kevp->ext)); 1277 uap->changelist++; 1278 kevp++; 1279 } 1280 return (error); 1281 } 1282 1283 int 1284 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap) 1285 { 1286 struct kevent_copyops k_ops = { 1287 .arg = uap, 1288 .k_copyout = kevent11_copyout, 1289 .k_copyin = kevent11_copyin, 1290 .kevent_size = sizeof(struct freebsd11_kevent), 1291 }; 1292 struct g_kevent_args gk_args = { 1293 .fd = uap->fd, 1294 .changelist = uap->changelist, 1295 .nchanges = uap->nchanges, 1296 .eventlist = uap->eventlist, 1297 .nevents = uap->nevents, 1298 .timeout = uap->timeout, 1299 }; 1300 1301 return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent")); 1302 } 1303 #endif 1304 1305 int 1306 kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 1307 struct kevent_copyops *k_ops, const struct timespec *timeout) 1308 { 1309 cap_rights_t rights; 1310 struct file *fp; 1311 int error; 1312 1313 cap_rights_init_zero(&rights); 1314 if (nchanges > 0) 1315 cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE); 1316 if (nevents > 0) 1317 cap_rights_set_one(&rights, CAP_KQUEUE_EVENT); 1318 error = fget(td, fd, &rights, &fp); 1319 if (error != 0) 1320 return (error); 1321 1322 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); 1323 fdrop(fp, td); 1324 1325 return (error); 1326 } 1327 1328 static int 1329 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents, 1330 struct kevent_copyops *k_ops, const struct timespec *timeout) 1331 { 1332 struct kevent keva[KQ_NEVENTS]; 1333 struct kevent *kevp, *changes; 1334 int i, n, nerrors, error; 1335 1336 if (nchanges < 0) 1337 return (EINVAL); 1338 1339 nerrors = 0; 1340 while (nchanges > 0) { 1341 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 1342 error = k_ops->k_copyin(k_ops->arg, keva, n); 1343 if (error) 1344 return (error); 1345 changes = keva; 1346 for (i = 0; i < n; i++) { 1347 kevp = &changes[i]; 1348 if (!kevp->filter) 1349 continue; 1350 kevp->flags &= ~EV_SYSFLAGS; 1351 error = kqueue_register(kq, kevp, td, M_WAITOK); 1352 if (error || (kevp->flags & EV_RECEIPT)) { 1353 if (nevents == 0) 1354 return (error); 1355 kevp->flags = EV_ERROR; 1356 kevp->data = error; 1357 (void)k_ops->k_copyout(k_ops->arg, kevp, 1); 1358 nevents--; 1359 nerrors++; 1360 } 1361 } 1362 nchanges -= n; 1363 } 1364 if (nerrors) { 1365 td->td_retval[0] = nerrors; 1366 return (0); 1367 } 1368 1369 return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td)); 1370 } 1371 1372 int 1373 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, 1374 struct kevent_copyops *k_ops, const struct timespec *timeout) 1375 { 1376 struct kqueue *kq; 1377 int error; 1378 1379 error = kqueue_acquire(fp, &kq); 1380 if (error != 0) 1381 return (error); 1382 error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout); 1383 kqueue_release(kq, 0); 1384 return (error); 1385 } 1386 1387 /* 1388 * Performs a kevent() call on a temporarily created kqueue. This can be 1389 * used to perform one-shot polling, similar to poll() and select(). 1390 */ 1391 int 1392 kern_kevent_anonymous(struct thread *td, int nevents, 1393 struct kevent_copyops *k_ops) 1394 { 1395 struct kqueue kq = {}; 1396 int error; 1397 1398 kqueue_init(&kq); 1399 kq.kq_refcnt = 1; 1400 error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL); 1401 kqueue_drain(&kq, td); 1402 kqueue_destroy(&kq); 1403 return (error); 1404 } 1405 1406 int 1407 kqueue_add_filteropts(int filt, const struct filterops *filtops) 1408 { 1409 int error; 1410 1411 error = 0; 1412 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 1413 printf( 1414 "trying to add a filterop that is out of range: %d is beyond %d\n", 1415 ~filt, EVFILT_SYSCOUNT); 1416 return EINVAL; 1417 } 1418 mtx_lock(&filterops_lock); 1419 if (sysfilt_ops[~filt].for_fop != &null_filtops && 1420 sysfilt_ops[~filt].for_fop != NULL) 1421 error = EEXIST; 1422 else { 1423 sysfilt_ops[~filt].for_fop = filtops; 1424 sysfilt_ops[~filt].for_refcnt = 0; 1425 } 1426 mtx_unlock(&filterops_lock); 1427 1428 return (error); 1429 } 1430 1431 int 1432 kqueue_del_filteropts(int filt) 1433 { 1434 int error; 1435 1436 error = 0; 1437 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1438 return EINVAL; 1439 1440 mtx_lock(&filterops_lock); 1441 if (sysfilt_ops[~filt].for_fop == &null_filtops || 1442 sysfilt_ops[~filt].for_fop == NULL) 1443 error = EINVAL; 1444 else if (sysfilt_ops[~filt].for_refcnt != 0) 1445 error = EBUSY; 1446 else { 1447 sysfilt_ops[~filt].for_fop = &null_filtops; 1448 sysfilt_ops[~filt].for_refcnt = 0; 1449 } 1450 mtx_unlock(&filterops_lock); 1451 1452 return error; 1453 } 1454 1455 static const struct filterops * 1456 kqueue_fo_find(int filt) 1457 { 1458 1459 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1460 return NULL; 1461 1462 if (sysfilt_ops[~filt].for_nolock) 1463 return sysfilt_ops[~filt].for_fop; 1464 1465 mtx_lock(&filterops_lock); 1466 sysfilt_ops[~filt].for_refcnt++; 1467 if (sysfilt_ops[~filt].for_fop == NULL) 1468 sysfilt_ops[~filt].for_fop = &null_filtops; 1469 mtx_unlock(&filterops_lock); 1470 1471 return sysfilt_ops[~filt].for_fop; 1472 } 1473 1474 static void 1475 kqueue_fo_release(int filt) 1476 { 1477 1478 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1479 return; 1480 1481 if (sysfilt_ops[~filt].for_nolock) 1482 return; 1483 1484 mtx_lock(&filterops_lock); 1485 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 1486 ("filter object refcount not valid on release")); 1487 sysfilt_ops[~filt].for_refcnt--; 1488 mtx_unlock(&filterops_lock); 1489 } 1490 1491 /* 1492 * A ref to kq (obtained via kqueue_acquire) must be held. 1493 */ 1494 static int 1495 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, 1496 int mflag) 1497 { 1498 const struct filterops *fops; 1499 struct file *fp; 1500 struct knote *kn, *tkn; 1501 struct knlist *knl; 1502 int error, filt, event; 1503 int haskqglobal, filedesc_unlock; 1504 1505 if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE)) 1506 return (EINVAL); 1507 1508 fp = NULL; 1509 kn = NULL; 1510 knl = NULL; 1511 error = 0; 1512 haskqglobal = 0; 1513 filedesc_unlock = 0; 1514 1515 filt = kev->filter; 1516 fops = kqueue_fo_find(filt); 1517 if (fops == NULL) 1518 return EINVAL; 1519 1520 if (kev->flags & EV_ADD) { 1521 /* Reject an invalid flag pair early */ 1522 if (kev->flags & EV_KEEPUDATA) { 1523 tkn = NULL; 1524 error = EINVAL; 1525 goto done; 1526 } 1527 1528 /* 1529 * Prevent waiting with locks. Non-sleepable 1530 * allocation failures are handled in the loop, only 1531 * if the spare knote appears to be actually required. 1532 */ 1533 tkn = knote_alloc(mflag); 1534 } else { 1535 tkn = NULL; 1536 } 1537 1538 findkn: 1539 if (fops->f_isfd) { 1540 KASSERT(td != NULL, ("td is NULL")); 1541 if (kev->ident > INT_MAX) 1542 error = EBADF; 1543 else 1544 error = fget(td, kev->ident, &cap_event_rights, &fp); 1545 if (error) 1546 goto done; 1547 1548 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 1549 kev->ident, M_NOWAIT) != 0) { 1550 /* try again */ 1551 fdrop(fp, td); 1552 fp = NULL; 1553 error = kqueue_expand(kq, fops, kev->ident, mflag); 1554 if (error) 1555 goto done; 1556 goto findkn; 1557 } 1558 1559 if (fp->f_type == DTYPE_KQUEUE) { 1560 /* 1561 * If we add some intelligence about what we are doing, 1562 * we should be able to support events on ourselves. 1563 * We need to know when we are doing this to prevent 1564 * getting both the knlist lock and the kq lock since 1565 * they are the same thing. 1566 */ 1567 if (fp->f_data == kq) { 1568 error = EINVAL; 1569 goto done; 1570 } 1571 1572 /* 1573 * Pre-lock the filedesc before the global 1574 * lock mutex, see the comment in 1575 * kqueue_close(). 1576 */ 1577 FILEDESC_XLOCK(td->td_proc->p_fd); 1578 filedesc_unlock = 1; 1579 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1580 } 1581 1582 KQ_LOCK(kq); 1583 if (kev->ident < kq->kq_knlistsize) { 1584 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1585 if (kev->filter == kn->kn_filter) 1586 break; 1587 } 1588 } else { 1589 if ((kev->flags & EV_ADD) == EV_ADD) { 1590 error = kqueue_expand(kq, fops, kev->ident, mflag); 1591 if (error != 0) 1592 goto done; 1593 } 1594 1595 KQ_LOCK(kq); 1596 1597 /* 1598 * If possible, find an existing knote to use for this kevent. 1599 */ 1600 if (kev->filter == EVFILT_PROC && 1601 (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) { 1602 /* This is an internal creation of a process tracking 1603 * note. Don't attempt to coalesce this with an 1604 * existing note. 1605 */ 1606 ; 1607 } else if (kq->kq_knhashmask != 0) { 1608 struct klist *list; 1609 1610 list = &kq->kq_knhash[ 1611 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1612 SLIST_FOREACH(kn, list, kn_link) 1613 if (kev->ident == kn->kn_id && 1614 kev->filter == kn->kn_filter) 1615 break; 1616 } 1617 } 1618 1619 /* knote is in the process of changing, wait for it to stabilize. */ 1620 if (kn != NULL && kn_in_flux(kn)) { 1621 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1622 if (filedesc_unlock) { 1623 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1624 filedesc_unlock = 0; 1625 } 1626 kq->kq_state |= KQ_FLUXWAIT; 1627 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1628 if (fp != NULL) { 1629 fdrop(fp, td); 1630 fp = NULL; 1631 } 1632 goto findkn; 1633 } 1634 1635 /* 1636 * kn now contains the matching knote, or NULL if no match 1637 */ 1638 if (kn == NULL) { 1639 if (kev->flags & EV_ADD) { 1640 kn = tkn; 1641 tkn = NULL; 1642 if (kn == NULL) { 1643 KQ_UNLOCK(kq); 1644 error = ENOMEM; 1645 goto done; 1646 } 1647 kn->kn_fp = fp; 1648 kn->kn_kq = kq; 1649 kn->kn_fop = fops; 1650 /* 1651 * apply reference counts to knote structure, and 1652 * do not release it at the end of this routine. 1653 */ 1654 fops = NULL; 1655 fp = NULL; 1656 1657 kn->kn_sfflags = kev->fflags; 1658 kn->kn_sdata = kev->data; 1659 kev->fflags = 0; 1660 kev->data = 0; 1661 kn->kn_kevent = *kev; 1662 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1663 EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); 1664 kn->kn_status = KN_DETACHED; 1665 if ((kev->flags & EV_DISABLE) != 0) 1666 kn->kn_status |= KN_DISABLED; 1667 kn_enter_flux(kn); 1668 1669 error = knote_attach(kn, kq); 1670 KQ_UNLOCK(kq); 1671 if (error != 0) { 1672 tkn = kn; 1673 goto done; 1674 } 1675 1676 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1677 knote_drop_detached(kn, td); 1678 goto done; 1679 } 1680 knl = kn_list_lock(kn); 1681 goto done_ev_add; 1682 } else { 1683 /* No matching knote and the EV_ADD flag is not set. */ 1684 KQ_UNLOCK(kq); 1685 error = ENOENT; 1686 goto done; 1687 } 1688 } 1689 1690 if (kev->flags & EV_DELETE) { 1691 kn_enter_flux(kn); 1692 KQ_UNLOCK(kq); 1693 knote_drop(kn, td); 1694 goto done; 1695 } 1696 1697 if (kev->flags & EV_FORCEONESHOT) { 1698 kn->kn_flags |= EV_ONESHOT; 1699 KNOTE_ACTIVATE(kn, 1); 1700 } 1701 1702 if ((kev->flags & EV_ENABLE) != 0) 1703 kn->kn_status &= ~KN_DISABLED; 1704 else if ((kev->flags & EV_DISABLE) != 0) 1705 kn->kn_status |= KN_DISABLED; 1706 1707 /* 1708 * The user may change some filter values after the initial EV_ADD, 1709 * but doing so will not reset any filter which has already been 1710 * triggered. 1711 */ 1712 kn->kn_status |= KN_SCAN; 1713 kn_enter_flux(kn); 1714 KQ_UNLOCK(kq); 1715 knl = kn_list_lock(kn); 1716 if ((kev->flags & EV_KEEPUDATA) == 0) 1717 kn->kn_kevent.udata = kev->udata; 1718 if (!fops->f_isfd && fops->f_touch != NULL) { 1719 fops->f_touch(kn, kev, EVENT_REGISTER); 1720 } else { 1721 kn->kn_sfflags = kev->fflags; 1722 kn->kn_sdata = kev->data; 1723 } 1724 1725 done_ev_add: 1726 /* 1727 * We can get here with kn->kn_knlist == NULL. This can happen when 1728 * the initial attach event decides that the event is "completed" 1729 * already, e.g., filt_procattach() is called on a zombie process. It 1730 * will call filt_proc() which will remove it from the list, and NULL 1731 * kn_knlist. 1732 * 1733 * KN_DISABLED will be stable while the knote is in flux, so the 1734 * unlocked read will not race with an update. 1735 */ 1736 if ((kn->kn_status & KN_DISABLED) == 0) 1737 event = kn->kn_fop->f_event(kn, 0); 1738 else 1739 event = 0; 1740 1741 KQ_LOCK(kq); 1742 if (event) 1743 kn->kn_status |= KN_ACTIVE; 1744 if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) == 1745 KN_ACTIVE) 1746 knote_enqueue(kn); 1747 kn->kn_status &= ~KN_SCAN; 1748 kn_leave_flux(kn); 1749 kn_list_unlock(knl); 1750 KQ_UNLOCK_FLUX(kq); 1751 1752 done: 1753 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1754 if (filedesc_unlock) 1755 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1756 if (fp != NULL) 1757 fdrop(fp, td); 1758 knote_free(tkn); 1759 if (fops != NULL) 1760 kqueue_fo_release(filt); 1761 return (error); 1762 } 1763 1764 static int 1765 kqueue_acquire(struct file *fp, struct kqueue **kqp) 1766 { 1767 int error; 1768 struct kqueue *kq; 1769 1770 error = 0; 1771 1772 kq = fp->f_data; 1773 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1774 return (EBADF); 1775 *kqp = kq; 1776 KQ_LOCK(kq); 1777 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1778 KQ_UNLOCK(kq); 1779 return (EBADF); 1780 } 1781 kq->kq_refcnt++; 1782 KQ_UNLOCK(kq); 1783 1784 return error; 1785 } 1786 1787 static void 1788 kqueue_release(struct kqueue *kq, int locked) 1789 { 1790 if (locked) 1791 KQ_OWNED(kq); 1792 else 1793 KQ_LOCK(kq); 1794 kq->kq_refcnt--; 1795 if (kq->kq_refcnt == 1) 1796 wakeup(&kq->kq_refcnt); 1797 if (!locked) 1798 KQ_UNLOCK(kq); 1799 } 1800 1801 static void 1802 ast_kqueue(struct thread *td, int tda __unused) 1803 { 1804 taskqueue_quiesce(taskqueue_kqueue_ctx); 1805 } 1806 1807 static void 1808 kqueue_schedtask(struct kqueue *kq) 1809 { 1810 KQ_OWNED(kq); 1811 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1812 ("scheduling kqueue task while draining")); 1813 1814 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1815 taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task); 1816 kq->kq_state |= KQ_TASKSCHED; 1817 ast_sched(curthread, TDA_KQUEUE); 1818 } 1819 } 1820 1821 /* 1822 * Expand the kq to make sure we have storage for fops/ident pair. 1823 * 1824 * Return 0 on success (or no work necessary), return errno on failure. 1825 */ 1826 static int 1827 kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident, 1828 int mflag) 1829 { 1830 struct klist *list, *tmp_knhash, *to_free; 1831 u_long tmp_knhashmask; 1832 int error, fd, size; 1833 1834 KQ_NOTOWNED(kq); 1835 1836 error = 0; 1837 to_free = NULL; 1838 if (fops->f_isfd) { 1839 fd = ident; 1840 if (kq->kq_knlistsize <= fd) { 1841 size = kq->kq_knlistsize; 1842 while (size <= fd) 1843 size += KQEXTENT; 1844 list = malloc(size * sizeof(*list), M_KQUEUE, mflag); 1845 if (list == NULL) 1846 return ENOMEM; 1847 KQ_LOCK(kq); 1848 if ((kq->kq_state & KQ_CLOSING) != 0) { 1849 to_free = list; 1850 error = EBADF; 1851 } else if (kq->kq_knlistsize > fd) { 1852 to_free = list; 1853 } else { 1854 if (kq->kq_knlist != NULL) { 1855 bcopy(kq->kq_knlist, list, 1856 kq->kq_knlistsize * sizeof(*list)); 1857 to_free = kq->kq_knlist; 1858 kq->kq_knlist = NULL; 1859 } 1860 bzero((caddr_t)list + 1861 kq->kq_knlistsize * sizeof(*list), 1862 (size - kq->kq_knlistsize) * sizeof(*list)); 1863 kq->kq_knlistsize = size; 1864 kq->kq_knlist = list; 1865 } 1866 KQ_UNLOCK(kq); 1867 } 1868 } else { 1869 if (kq->kq_knhashmask == 0) { 1870 tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE, 1871 &tmp_knhashmask, (mflag & M_WAITOK) != 0 ? 1872 HASH_WAITOK : HASH_NOWAIT); 1873 if (tmp_knhash == NULL) 1874 return (ENOMEM); 1875 KQ_LOCK(kq); 1876 if ((kq->kq_state & KQ_CLOSING) != 0) { 1877 to_free = tmp_knhash; 1878 error = EBADF; 1879 } else if (kq->kq_knhashmask == 0) { 1880 kq->kq_knhash = tmp_knhash; 1881 kq->kq_knhashmask = tmp_knhashmask; 1882 } else { 1883 to_free = tmp_knhash; 1884 } 1885 KQ_UNLOCK(kq); 1886 } 1887 } 1888 free(to_free, M_KQUEUE); 1889 1890 KQ_NOTOWNED(kq); 1891 return (error); 1892 } 1893 1894 static void 1895 kqueue_task(void *arg, int pending) 1896 { 1897 struct kqueue *kq; 1898 int haskqglobal; 1899 1900 haskqglobal = 0; 1901 kq = arg; 1902 1903 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1904 KQ_LOCK(kq); 1905 1906 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 1907 1908 kq->kq_state &= ~KQ_TASKSCHED; 1909 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 1910 wakeup(&kq->kq_state); 1911 } 1912 KQ_UNLOCK(kq); 1913 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1914 } 1915 1916 /* 1917 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 1918 * We treat KN_MARKER knotes as if they are in flux. 1919 */ 1920 static int 1921 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 1922 const struct timespec *tsp, struct kevent *keva, struct thread *td) 1923 { 1924 struct kevent *kevp; 1925 struct knote *kn, *marker; 1926 struct knlist *knl; 1927 sbintime_t asbt, rsbt; 1928 int count, error, haskqglobal, influx, nkev, touch; 1929 1930 count = maxevents; 1931 nkev = 0; 1932 error = 0; 1933 haskqglobal = 0; 1934 1935 if (maxevents == 0) 1936 goto done_nl; 1937 if (maxevents < 0) { 1938 error = EINVAL; 1939 goto done_nl; 1940 } 1941 1942 rsbt = 0; 1943 if (tsp != NULL) { 1944 if (!timespecvalid_interval(tsp)) { 1945 error = EINVAL; 1946 goto done_nl; 1947 } 1948 if (timespecisset(tsp)) { 1949 if (tsp->tv_sec <= INT32_MAX) { 1950 rsbt = tstosbt(*tsp); 1951 if (TIMESEL(&asbt, rsbt)) 1952 asbt += tc_tick_sbt; 1953 if (asbt <= SBT_MAX - rsbt) 1954 asbt += rsbt; 1955 else 1956 asbt = 0; 1957 rsbt >>= tc_precexp; 1958 } else 1959 asbt = 0; 1960 } else 1961 asbt = -1; 1962 } else 1963 asbt = 0; 1964 marker = knote_alloc(M_WAITOK); 1965 marker->kn_status = KN_MARKER; 1966 KQ_LOCK(kq); 1967 1968 retry: 1969 kevp = keva; 1970 if (kq->kq_count == 0) { 1971 if (asbt == -1) { 1972 error = EWOULDBLOCK; 1973 } else { 1974 kq->kq_state |= KQ_SLEEP; 1975 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, 1976 "kqread", asbt, rsbt, C_ABSOLUTE); 1977 } 1978 if (error == 0) 1979 goto retry; 1980 /* don't restart after signals... */ 1981 if (error == ERESTART) 1982 error = EINTR; 1983 else if (error == EWOULDBLOCK) 1984 error = 0; 1985 goto done; 1986 } 1987 1988 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1989 influx = 0; 1990 while (count) { 1991 KQ_OWNED(kq); 1992 kn = TAILQ_FIRST(&kq->kq_head); 1993 1994 if ((kn->kn_status == KN_MARKER && kn != marker) || 1995 kn_in_flux(kn)) { 1996 if (influx) { 1997 influx = 0; 1998 KQ_FLUX_WAKEUP(kq); 1999 } 2000 kq->kq_state |= KQ_FLUXWAIT; 2001 error = msleep(kq, &kq->kq_lock, PSOCK, 2002 "kqflxwt", 0); 2003 continue; 2004 } 2005 2006 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2007 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 2008 kn->kn_status &= ~KN_QUEUED; 2009 kq->kq_count--; 2010 continue; 2011 } 2012 if (kn == marker) { 2013 KQ_FLUX_WAKEUP(kq); 2014 if (count == maxevents) 2015 goto retry; 2016 goto done; 2017 } 2018 KASSERT(!kn_in_flux(kn), 2019 ("knote %p is unexpectedly in flux", kn)); 2020 2021 if ((kn->kn_flags & EV_DROP) == EV_DROP) { 2022 kn->kn_status &= ~KN_QUEUED; 2023 kn_enter_flux(kn); 2024 kq->kq_count--; 2025 KQ_UNLOCK(kq); 2026 /* 2027 * We don't need to lock the list since we've 2028 * marked it as in flux. 2029 */ 2030 knote_drop(kn, td); 2031 KQ_LOCK(kq); 2032 continue; 2033 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 2034 kn->kn_status &= ~KN_QUEUED; 2035 kn_enter_flux(kn); 2036 kq->kq_count--; 2037 KQ_UNLOCK(kq); 2038 /* 2039 * We don't need to lock the list since we've 2040 * marked the knote as being in flux. 2041 */ 2042 *kevp = kn->kn_kevent; 2043 knote_drop(kn, td); 2044 KQ_LOCK(kq); 2045 kn = NULL; 2046 } else { 2047 kn->kn_status |= KN_SCAN; 2048 kn_enter_flux(kn); 2049 KQ_UNLOCK(kq); 2050 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 2051 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 2052 knl = kn_list_lock(kn); 2053 if (kn->kn_fop->f_event(kn, 0) == 0) { 2054 KQ_LOCK(kq); 2055 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2056 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE | 2057 KN_SCAN); 2058 kn_leave_flux(kn); 2059 kq->kq_count--; 2060 kn_list_unlock(knl); 2061 influx = 1; 2062 continue; 2063 } 2064 touch = (!kn->kn_fop->f_isfd && 2065 kn->kn_fop->f_touch != NULL); 2066 if (touch) 2067 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 2068 else 2069 *kevp = kn->kn_kevent; 2070 KQ_LOCK(kq); 2071 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2072 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 2073 /* 2074 * Manually clear knotes who weren't 2075 * 'touch'ed. 2076 */ 2077 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 2078 kn->kn_data = 0; 2079 kn->kn_fflags = 0; 2080 } 2081 if (kn->kn_flags & EV_DISPATCH) 2082 kn->kn_status |= KN_DISABLED; 2083 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 2084 kq->kq_count--; 2085 } else 2086 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2087 2088 kn->kn_status &= ~KN_SCAN; 2089 kn_leave_flux(kn); 2090 kn_list_unlock(knl); 2091 influx = 1; 2092 } 2093 2094 /* we are returning a copy to the user */ 2095 kevp++; 2096 nkev++; 2097 count--; 2098 2099 if (nkev == KQ_NEVENTS) { 2100 influx = 0; 2101 KQ_UNLOCK_FLUX(kq); 2102 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 2103 nkev = 0; 2104 kevp = keva; 2105 KQ_LOCK(kq); 2106 if (error) 2107 break; 2108 } 2109 } 2110 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 2111 done: 2112 KQ_OWNED(kq); 2113 KQ_UNLOCK_FLUX(kq); 2114 knote_free(marker); 2115 done_nl: 2116 KQ_NOTOWNED(kq); 2117 if (nkev != 0) 2118 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 2119 td->td_retval[0] = maxevents - count; 2120 return (error); 2121 } 2122 2123 /*ARGSUSED*/ 2124 static int 2125 kqueue_ioctl(struct file *fp, u_long cmd, void *data, 2126 struct ucred *active_cred, struct thread *td) 2127 { 2128 /* 2129 * Enabling sigio causes two major problems: 2130 * 1) infinite recursion: 2131 * Synopsys: kevent is being used to track signals and have FIOASYNC 2132 * set. On receipt of a signal this will cause a kqueue to recurse 2133 * into itself over and over. Sending the sigio causes the kqueue 2134 * to become ready, which in turn posts sigio again, forever. 2135 * Solution: this can be solved by setting a flag in the kqueue that 2136 * we have a SIGIO in progress. 2137 * 2) locking problems: 2138 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 2139 * us above the proc and pgrp locks. 2140 * Solution: Post a signal using an async mechanism, being sure to 2141 * record a generation count in the delivery so that we do not deliver 2142 * a signal to the wrong process. 2143 * 2144 * Note, these two mechanisms are somewhat mutually exclusive! 2145 */ 2146 #if 0 2147 struct kqueue *kq; 2148 2149 kq = fp->f_data; 2150 switch (cmd) { 2151 case FIOASYNC: 2152 if (*(int *)data) { 2153 kq->kq_state |= KQ_ASYNC; 2154 } else { 2155 kq->kq_state &= ~KQ_ASYNC; 2156 } 2157 return (0); 2158 2159 case FIOSETOWN: 2160 return (fsetown(*(int *)data, &kq->kq_sigio)); 2161 2162 case FIOGETOWN: 2163 *(int *)data = fgetown(&kq->kq_sigio); 2164 return (0); 2165 } 2166 #endif 2167 2168 return (ENOTTY); 2169 } 2170 2171 /*ARGSUSED*/ 2172 static int 2173 kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 2174 struct thread *td) 2175 { 2176 struct kqueue *kq; 2177 int revents = 0; 2178 int error; 2179 2180 if ((error = kqueue_acquire(fp, &kq))) 2181 return POLLERR; 2182 2183 KQ_LOCK(kq); 2184 if (events & (POLLIN | POLLRDNORM)) { 2185 if (kq->kq_count) { 2186 revents |= events & (POLLIN | POLLRDNORM); 2187 } else { 2188 selrecord(td, &kq->kq_sel); 2189 if (SEL_WAITING(&kq->kq_sel)) 2190 kq->kq_state |= KQ_SEL; 2191 } 2192 } 2193 kqueue_release(kq, 1); 2194 KQ_UNLOCK(kq); 2195 return (revents); 2196 } 2197 2198 /*ARGSUSED*/ 2199 static int 2200 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred) 2201 { 2202 2203 bzero((void *)st, sizeof *st); 2204 /* 2205 * We no longer return kq_count because the unlocked value is useless. 2206 * If you spent all this time getting the count, why not spend your 2207 * syscall better by calling kevent? 2208 * 2209 * XXX - This is needed for libc_r. 2210 */ 2211 st->st_mode = S_IFIFO; 2212 return (0); 2213 } 2214 2215 static void 2216 kqueue_drain(struct kqueue *kq, struct thread *td) 2217 { 2218 struct knote *kn; 2219 int i; 2220 2221 KQ_LOCK(kq); 2222 2223 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 2224 ("kqueue already closing")); 2225 kq->kq_state |= KQ_CLOSING; 2226 if (kq->kq_refcnt > 1) 2227 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 2228 2229 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 2230 2231 KASSERT(knlist_empty(&kq->kq_sel.si_note), 2232 ("kqueue's knlist not empty")); 2233 2234 for (i = 0; i < kq->kq_knlistsize; i++) { 2235 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 2236 if (kn_in_flux(kn)) { 2237 kq->kq_state |= KQ_FLUXWAIT; 2238 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 2239 continue; 2240 } 2241 kn_enter_flux(kn); 2242 KQ_UNLOCK(kq); 2243 knote_drop(kn, td); 2244 KQ_LOCK(kq); 2245 } 2246 } 2247 if (kq->kq_knhashmask != 0) { 2248 for (i = 0; i <= kq->kq_knhashmask; i++) { 2249 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 2250 if (kn_in_flux(kn)) { 2251 kq->kq_state |= KQ_FLUXWAIT; 2252 msleep(kq, &kq->kq_lock, PSOCK, 2253 "kqclo2", 0); 2254 continue; 2255 } 2256 kn_enter_flux(kn); 2257 KQ_UNLOCK(kq); 2258 knote_drop(kn, td); 2259 KQ_LOCK(kq); 2260 } 2261 } 2262 } 2263 2264 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 2265 kq->kq_state |= KQ_TASKDRAIN; 2266 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 2267 } 2268 2269 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2270 selwakeuppri(&kq->kq_sel, PSOCK); 2271 if (!SEL_WAITING(&kq->kq_sel)) 2272 kq->kq_state &= ~KQ_SEL; 2273 } 2274 2275 KQ_UNLOCK(kq); 2276 } 2277 2278 static void 2279 kqueue_destroy(struct kqueue *kq) 2280 { 2281 2282 KASSERT(kq->kq_fdp == NULL, 2283 ("kqueue still attached to a file descriptor")); 2284 seldrain(&kq->kq_sel); 2285 knlist_destroy(&kq->kq_sel.si_note); 2286 mtx_destroy(&kq->kq_lock); 2287 2288 if (kq->kq_knhash != NULL) 2289 free(kq->kq_knhash, M_KQUEUE); 2290 if (kq->kq_knlist != NULL) 2291 free(kq->kq_knlist, M_KQUEUE); 2292 2293 funsetown(&kq->kq_sigio); 2294 } 2295 2296 /*ARGSUSED*/ 2297 static int 2298 kqueue_close(struct file *fp, struct thread *td) 2299 { 2300 struct kqueue *kq = fp->f_data; 2301 struct filedesc *fdp; 2302 int error; 2303 int filedesc_unlock; 2304 2305 if ((error = kqueue_acquire(fp, &kq))) 2306 return error; 2307 kqueue_drain(kq, td); 2308 2309 /* 2310 * We could be called due to the knote_drop() doing fdrop(), 2311 * called from kqueue_register(). In this case the global 2312 * lock is owned, and filedesc sx is locked before, to not 2313 * take the sleepable lock after non-sleepable. 2314 */ 2315 fdp = kq->kq_fdp; 2316 kq->kq_fdp = NULL; 2317 if (!sx_xlocked(FILEDESC_LOCK(fdp))) { 2318 FILEDESC_XLOCK(fdp); 2319 filedesc_unlock = 1; 2320 } else 2321 filedesc_unlock = 0; 2322 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); 2323 if (filedesc_unlock) 2324 FILEDESC_XUNLOCK(fdp); 2325 2326 kqueue_destroy(kq); 2327 chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); 2328 crfree(kq->kq_cred); 2329 free(kq, M_KQUEUE); 2330 fp->f_data = NULL; 2331 2332 return (0); 2333 } 2334 2335 static int 2336 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2337 { 2338 struct kqueue *kq = fp->f_data; 2339 2340 kif->kf_type = KF_TYPE_KQUEUE; 2341 kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq; 2342 kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count; 2343 kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state; 2344 return (0); 2345 } 2346 2347 static void 2348 kqueue_wakeup(struct kqueue *kq) 2349 { 2350 KQ_OWNED(kq); 2351 2352 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 2353 kq->kq_state &= ~KQ_SLEEP; 2354 wakeup(kq); 2355 } 2356 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2357 selwakeuppri(&kq->kq_sel, PSOCK); 2358 if (!SEL_WAITING(&kq->kq_sel)) 2359 kq->kq_state &= ~KQ_SEL; 2360 } 2361 if (!knlist_empty(&kq->kq_sel.si_note)) 2362 kqueue_schedtask(kq); 2363 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 2364 pgsigio(&kq->kq_sigio, SIGIO, 0); 2365 } 2366 } 2367 2368 /* 2369 * Walk down a list of knotes, activating them if their event has triggered. 2370 * 2371 * There is a possibility to optimize in the case of one kq watching another. 2372 * Instead of scheduling a task to wake it up, you could pass enough state 2373 * down the chain to make up the parent kqueue. Make this code functional 2374 * first. 2375 */ 2376 void 2377 knote(struct knlist *list, long hint, int lockflags) 2378 { 2379 struct kqueue *kq; 2380 struct knote *kn, *tkn; 2381 int error; 2382 2383 if (list == NULL) 2384 return; 2385 2386 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 2387 2388 if ((lockflags & KNF_LISTLOCKED) == 0) 2389 list->kl_lock(list->kl_lockarg); 2390 2391 /* 2392 * If we unlock the list lock (and enter influx), we can 2393 * eliminate the kqueue scheduling, but this will introduce 2394 * four lock/unlock's for each knote to test. Also, marker 2395 * would be needed to keep iteration position, since filters 2396 * or other threads could remove events. 2397 */ 2398 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) { 2399 kq = kn->kn_kq; 2400 KQ_LOCK(kq); 2401 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 2402 /* 2403 * Do not process the influx notes, except for 2404 * the influx coming from the kq unlock in the 2405 * kqueue_scan(). In the later case, we do 2406 * not interfere with the scan, since the code 2407 * fragment in kqueue_scan() locks the knlist, 2408 * and cannot proceed until we finished. 2409 */ 2410 KQ_UNLOCK(kq); 2411 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 2412 kn_enter_flux(kn); 2413 KQ_UNLOCK(kq); 2414 error = kn->kn_fop->f_event(kn, hint); 2415 KQ_LOCK(kq); 2416 kn_leave_flux(kn); 2417 if (error) 2418 KNOTE_ACTIVATE(kn, 1); 2419 KQ_UNLOCK_FLUX(kq); 2420 } else { 2421 if (kn->kn_fop->f_event(kn, hint)) 2422 KNOTE_ACTIVATE(kn, 1); 2423 KQ_UNLOCK(kq); 2424 } 2425 } 2426 if ((lockflags & KNF_LISTLOCKED) == 0) 2427 list->kl_unlock(list->kl_lockarg); 2428 } 2429 2430 /* 2431 * add a knote to a knlist 2432 */ 2433 void 2434 knlist_add(struct knlist *knl, struct knote *kn, int islocked) 2435 { 2436 2437 KNL_ASSERT_LOCK(knl, islocked); 2438 KQ_NOTOWNED(kn->kn_kq); 2439 KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn)); 2440 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2441 ("knote %p was not detached", kn)); 2442 if (!islocked) 2443 knl->kl_lock(knl->kl_lockarg); 2444 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 2445 if (!islocked) 2446 knl->kl_unlock(knl->kl_lockarg); 2447 KQ_LOCK(kn->kn_kq); 2448 kn->kn_knlist = knl; 2449 kn->kn_status &= ~KN_DETACHED; 2450 KQ_UNLOCK(kn->kn_kq); 2451 } 2452 2453 static void 2454 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, 2455 int kqislocked) 2456 { 2457 2458 KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked")); 2459 KNL_ASSERT_LOCK(knl, knlislocked); 2460 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 2461 KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn)); 2462 KASSERT((kn->kn_status & KN_DETACHED) == 0, 2463 ("knote %p was already detached", kn)); 2464 if (!knlislocked) 2465 knl->kl_lock(knl->kl_lockarg); 2466 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 2467 kn->kn_knlist = NULL; 2468 if (!knlislocked) 2469 kn_list_unlock(knl); 2470 if (!kqislocked) 2471 KQ_LOCK(kn->kn_kq); 2472 kn->kn_status |= KN_DETACHED; 2473 if (!kqislocked) 2474 KQ_UNLOCK(kn->kn_kq); 2475 } 2476 2477 /* 2478 * remove knote from the specified knlist 2479 */ 2480 void 2481 knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 2482 { 2483 2484 knlist_remove_kq(knl, kn, islocked, 0); 2485 } 2486 2487 int 2488 knlist_empty(struct knlist *knl) 2489 { 2490 2491 KNL_ASSERT_LOCKED(knl); 2492 return (SLIST_EMPTY(&knl->kl_list)); 2493 } 2494 2495 static struct mtx knlist_lock; 2496 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 2497 MTX_DEF); 2498 static void knlist_mtx_lock(void *arg); 2499 static void knlist_mtx_unlock(void *arg); 2500 2501 static void 2502 knlist_mtx_lock(void *arg) 2503 { 2504 2505 mtx_lock((struct mtx *)arg); 2506 } 2507 2508 static void 2509 knlist_mtx_unlock(void *arg) 2510 { 2511 2512 mtx_unlock((struct mtx *)arg); 2513 } 2514 2515 static void 2516 knlist_mtx_assert_lock(void *arg, int what) 2517 { 2518 2519 if (what == LA_LOCKED) 2520 mtx_assert((struct mtx *)arg, MA_OWNED); 2521 else 2522 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 2523 } 2524 2525 void 2526 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 2527 void (*kl_unlock)(void *), 2528 void (*kl_assert_lock)(void *, int)) 2529 { 2530 2531 if (lock == NULL) 2532 knl->kl_lockarg = &knlist_lock; 2533 else 2534 knl->kl_lockarg = lock; 2535 2536 if (kl_lock == NULL) 2537 knl->kl_lock = knlist_mtx_lock; 2538 else 2539 knl->kl_lock = kl_lock; 2540 if (kl_unlock == NULL) 2541 knl->kl_unlock = knlist_mtx_unlock; 2542 else 2543 knl->kl_unlock = kl_unlock; 2544 if (kl_assert_lock == NULL) 2545 knl->kl_assert_lock = knlist_mtx_assert_lock; 2546 else 2547 knl->kl_assert_lock = kl_assert_lock; 2548 2549 knl->kl_autodestroy = 0; 2550 SLIST_INIT(&knl->kl_list); 2551 } 2552 2553 void 2554 knlist_init_mtx(struct knlist *knl, struct mtx *lock) 2555 { 2556 2557 knlist_init(knl, lock, NULL, NULL, NULL); 2558 } 2559 2560 struct knlist * 2561 knlist_alloc(struct mtx *lock) 2562 { 2563 struct knlist *knl; 2564 2565 knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK); 2566 knlist_init_mtx(knl, lock); 2567 return (knl); 2568 } 2569 2570 void 2571 knlist_destroy(struct knlist *knl) 2572 { 2573 2574 KASSERT(KNLIST_EMPTY(knl), 2575 ("destroying knlist %p with knotes on it", knl)); 2576 } 2577 2578 void 2579 knlist_detach(struct knlist *knl) 2580 { 2581 2582 KNL_ASSERT_LOCKED(knl); 2583 knl->kl_autodestroy = 1; 2584 if (knlist_empty(knl)) { 2585 knlist_destroy(knl); 2586 free(knl, M_KQUEUE); 2587 } 2588 } 2589 2590 /* 2591 * Even if we are locked, we may need to drop the lock to allow any influx 2592 * knotes time to "settle". 2593 */ 2594 void 2595 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 2596 { 2597 struct knote *kn, *kn2; 2598 struct kqueue *kq; 2599 2600 KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl)); 2601 if (islocked) 2602 KNL_ASSERT_LOCKED(knl); 2603 else { 2604 KNL_ASSERT_UNLOCKED(knl); 2605 again: /* need to reacquire lock since we have dropped it */ 2606 knl->kl_lock(knl->kl_lockarg); 2607 } 2608 2609 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 2610 kq = kn->kn_kq; 2611 KQ_LOCK(kq); 2612 if (kn_in_flux(kn)) { 2613 KQ_UNLOCK(kq); 2614 continue; 2615 } 2616 knlist_remove_kq(knl, kn, 1, 1); 2617 if (killkn) { 2618 kn_enter_flux(kn); 2619 KQ_UNLOCK(kq); 2620 knote_drop_detached(kn, td); 2621 } else { 2622 /* Make sure cleared knotes disappear soon */ 2623 kn->kn_flags |= EV_EOF | EV_ONESHOT; 2624 KQ_UNLOCK(kq); 2625 } 2626 kq = NULL; 2627 } 2628 2629 if (!SLIST_EMPTY(&knl->kl_list)) { 2630 /* there are still in flux knotes remaining */ 2631 kn = SLIST_FIRST(&knl->kl_list); 2632 kq = kn->kn_kq; 2633 KQ_LOCK(kq); 2634 KASSERT(kn_in_flux(kn), ("knote removed w/o list lock")); 2635 knl->kl_unlock(knl->kl_lockarg); 2636 kq->kq_state |= KQ_FLUXWAIT; 2637 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 2638 kq = NULL; 2639 goto again; 2640 } 2641 2642 if (islocked) 2643 KNL_ASSERT_LOCKED(knl); 2644 else { 2645 knl->kl_unlock(knl->kl_lockarg); 2646 KNL_ASSERT_UNLOCKED(knl); 2647 } 2648 } 2649 2650 /* 2651 * Remove all knotes referencing a specified fd must be called with FILEDESC 2652 * lock. This prevents a race where a new fd comes along and occupies the 2653 * entry and we attach a knote to the fd. 2654 */ 2655 void 2656 knote_fdclose(struct thread *td, int fd) 2657 { 2658 struct filedesc *fdp = td->td_proc->p_fd; 2659 struct kqueue *kq; 2660 struct knote *kn; 2661 int influx; 2662 2663 FILEDESC_XLOCK_ASSERT(fdp); 2664 2665 /* 2666 * We shouldn't have to worry about new kevents appearing on fd 2667 * since filedesc is locked. 2668 */ 2669 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2670 KQ_LOCK(kq); 2671 2672 again: 2673 influx = 0; 2674 while (kq->kq_knlistsize > fd && 2675 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2676 if (kn_in_flux(kn)) { 2677 /* someone else might be waiting on our knote */ 2678 if (influx) 2679 wakeup(kq); 2680 kq->kq_state |= KQ_FLUXWAIT; 2681 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2682 goto again; 2683 } 2684 kn_enter_flux(kn); 2685 KQ_UNLOCK(kq); 2686 influx = 1; 2687 knote_drop(kn, td); 2688 KQ_LOCK(kq); 2689 } 2690 KQ_UNLOCK_FLUX(kq); 2691 } 2692 } 2693 2694 static int 2695 knote_attach(struct knote *kn, struct kqueue *kq) 2696 { 2697 struct klist *list; 2698 2699 KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn)); 2700 KQ_OWNED(kq); 2701 2702 if ((kq->kq_state & KQ_CLOSING) != 0) 2703 return (EBADF); 2704 if (kn->kn_fop->f_isfd) { 2705 if (kn->kn_id >= kq->kq_knlistsize) 2706 return (ENOMEM); 2707 list = &kq->kq_knlist[kn->kn_id]; 2708 } else { 2709 if (kq->kq_knhash == NULL) 2710 return (ENOMEM); 2711 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2712 } 2713 SLIST_INSERT_HEAD(list, kn, kn_link); 2714 return (0); 2715 } 2716 2717 static void 2718 knote_drop(struct knote *kn, struct thread *td) 2719 { 2720 2721 if ((kn->kn_status & KN_DETACHED) == 0) 2722 kn->kn_fop->f_detach(kn); 2723 knote_drop_detached(kn, td); 2724 } 2725 2726 static void 2727 knote_drop_detached(struct knote *kn, struct thread *td) 2728 { 2729 struct kqueue *kq; 2730 struct klist *list; 2731 2732 kq = kn->kn_kq; 2733 2734 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2735 ("knote %p still attached", kn)); 2736 KQ_NOTOWNED(kq); 2737 2738 KQ_LOCK(kq); 2739 for (;;) { 2740 KASSERT(kn->kn_influx >= 1, 2741 ("knote_drop called on %p with influx %d", 2742 kn, kn->kn_influx)); 2743 if (kn->kn_influx == 1) 2744 break; 2745 kq->kq_state |= KQ_FLUXWAIT; 2746 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2747 } 2748 2749 if (kn->kn_fop->f_isfd) 2750 list = &kq->kq_knlist[kn->kn_id]; 2751 else 2752 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2753 2754 if (!SLIST_EMPTY(list)) 2755 SLIST_REMOVE(list, kn, knote, kn_link); 2756 if (kn->kn_status & KN_QUEUED) 2757 knote_dequeue(kn); 2758 KQ_UNLOCK_FLUX(kq); 2759 2760 if (kn->kn_fop->f_isfd) { 2761 fdrop(kn->kn_fp, td); 2762 kn->kn_fp = NULL; 2763 } 2764 kqueue_fo_release(kn->kn_kevent.filter); 2765 kn->kn_fop = NULL; 2766 knote_free(kn); 2767 } 2768 2769 static void 2770 knote_enqueue(struct knote *kn) 2771 { 2772 struct kqueue *kq = kn->kn_kq; 2773 2774 KQ_OWNED(kn->kn_kq); 2775 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2776 2777 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2778 kn->kn_status |= KN_QUEUED; 2779 kq->kq_count++; 2780 kqueue_wakeup(kq); 2781 } 2782 2783 static void 2784 knote_dequeue(struct knote *kn) 2785 { 2786 struct kqueue *kq = kn->kn_kq; 2787 2788 KQ_OWNED(kn->kn_kq); 2789 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2790 2791 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2792 kn->kn_status &= ~KN_QUEUED; 2793 kq->kq_count--; 2794 } 2795 2796 static void 2797 knote_init(void) 2798 { 2799 2800 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2801 NULL, NULL, UMA_ALIGN_PTR, 0); 2802 ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue); 2803 } 2804 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2805 2806 static struct knote * 2807 knote_alloc(int mflag) 2808 { 2809 2810 return (uma_zalloc(knote_zone, mflag | M_ZERO)); 2811 } 2812 2813 static void 2814 knote_free(struct knote *kn) 2815 { 2816 2817 uma_zfree(knote_zone, kn); 2818 } 2819 2820 /* 2821 * Register the kev w/ the kq specified by fd. 2822 */ 2823 int 2824 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag) 2825 { 2826 struct kqueue *kq; 2827 struct file *fp; 2828 cap_rights_t rights; 2829 int error; 2830 2831 error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE), 2832 &fp); 2833 if (error != 0) 2834 return (error); 2835 if ((error = kqueue_acquire(fp, &kq)) != 0) 2836 goto noacquire; 2837 2838 error = kqueue_register(kq, kev, td, mflag); 2839 kqueue_release(kq, 0); 2840 2841 noacquire: 2842 fdrop(fp, td); 2843 return (error); 2844 } 2845 2846 struct knote_status_export_bit { 2847 int kn_status_bit; 2848 int knt_status_bit; 2849 }; 2850 2851 #define ST(name) \ 2852 { .kn_status_bit = KN_##name, .knt_status_bit = KNOTE_STATUS_##name } 2853 static const struct knote_status_export_bit knote_status_export_bits[] = { 2854 ST(ACTIVE), 2855 ST(QUEUED), 2856 ST(DISABLED), 2857 ST(DETACHED), 2858 ST(KQUEUE), 2859 }; 2860 #undef ST 2861 2862 static int 2863 knote_status_export(int kn_status) 2864 { 2865 const struct knote_status_export_bit *b; 2866 unsigned i; 2867 int res; 2868 2869 res = 0; 2870 for (i = 0; i < nitems(knote_status_export_bits); i++) { 2871 b = &knote_status_export_bits[i]; 2872 if ((kn_status & b->kn_status_bit) != 0) 2873 res |= b->knt_status_bit; 2874 } 2875 return (res); 2876 } 2877 2878 static int 2879 kern_proc_kqueue_report_one(struct sbuf *s, struct proc *p, 2880 int kq_fd, struct kqueue *kq, struct knote *kn, bool compat32 __unused) 2881 { 2882 struct kinfo_knote kin; 2883 #ifdef COMPAT_FREEBSD32 2884 struct kinfo_knote32 kin32; 2885 #endif 2886 int error; 2887 2888 if (kn->kn_status == KN_MARKER) 2889 return (0); 2890 2891 memset(&kin, 0, sizeof(kin)); 2892 kin.knt_kq_fd = kq_fd; 2893 memcpy(&kin.knt_event, &kn->kn_kevent, sizeof(struct kevent)); 2894 kin.knt_status = knote_status_export(kn->kn_status); 2895 kn_enter_flux(kn); 2896 KQ_UNLOCK_FLUX(kq); 2897 if (kn->kn_fop->f_userdump != NULL) 2898 (void)kn->kn_fop->f_userdump(p, kn, &kin); 2899 #ifdef COMPAT_FREEBSD32 2900 if (compat32) { 2901 freebsd32_kinfo_knote_to_32(&kin, &kin32); 2902 error = sbuf_bcat(s, &kin32, sizeof(kin32)); 2903 } else 2904 #endif 2905 error = sbuf_bcat(s, &kin, sizeof(kin)); 2906 KQ_LOCK(kq); 2907 kn_leave_flux(kn); 2908 return (error); 2909 } 2910 2911 static int 2912 kern_proc_kqueue_report(struct sbuf *s, struct proc *p, int kq_fd, 2913 struct kqueue *kq, bool compat32) 2914 { 2915 struct knote *kn; 2916 int error, i; 2917 2918 error = 0; 2919 KQ_LOCK(kq); 2920 for (i = 0; i < kq->kq_knlistsize; i++) { 2921 SLIST_FOREACH(kn, &kq->kq_knlist[i], kn_link) { 2922 error = kern_proc_kqueue_report_one(s, p, kq_fd, 2923 kq, kn, compat32); 2924 if (error != 0) 2925 goto out; 2926 } 2927 } 2928 if (kq->kq_knhashmask == 0) 2929 goto out; 2930 for (i = 0; i <= kq->kq_knhashmask; i++) { 2931 SLIST_FOREACH(kn, &kq->kq_knhash[i], kn_link) { 2932 error = kern_proc_kqueue_report_one(s, p, kq_fd, 2933 kq, kn, compat32); 2934 if (error != 0) 2935 goto out; 2936 } 2937 } 2938 out: 2939 KQ_UNLOCK_FLUX(kq); 2940 return (error); 2941 } 2942 2943 struct kern_proc_kqueues_out1_cb_args { 2944 struct sbuf *s; 2945 bool compat32; 2946 }; 2947 2948 static int 2949 kern_proc_kqueues_out1_cb(struct proc *p, int fd, struct file *fp, void *arg) 2950 { 2951 struct kqueue *kq; 2952 struct kern_proc_kqueues_out1_cb_args *a; 2953 2954 if (fp->f_type != DTYPE_KQUEUE) 2955 return (0); 2956 a = arg; 2957 kq = fp->f_data; 2958 return (kern_proc_kqueue_report(a->s, p, fd, kq, a->compat32)); 2959 } 2960 2961 static int 2962 kern_proc_kqueues_out1(struct thread *td, struct proc *p, struct sbuf *s, 2963 bool compat32) 2964 { 2965 struct kern_proc_kqueues_out1_cb_args a; 2966 2967 a.s = s; 2968 a.compat32 = compat32; 2969 return (fget_remote_foreach(td, p, kern_proc_kqueues_out1_cb, &a)); 2970 } 2971 2972 int 2973 kern_proc_kqueues_out(struct proc *p, struct sbuf *sb, size_t maxlen, 2974 bool compat32) 2975 { 2976 struct sbuf *s, sm; 2977 size_t sb_len; 2978 int error; 2979 2980 if (maxlen == -1 || maxlen == 0) 2981 sb_len = 128; 2982 else 2983 sb_len = maxlen; 2984 s = sbuf_new(&sm, NULL, sb_len, maxlen == -1 ? SBUF_AUTOEXTEND : 2985 SBUF_FIXEDLEN); 2986 error = kern_proc_kqueues_out1(curthread, p, s, compat32); 2987 sbuf_finish(s); 2988 if (error == 0) { 2989 sbuf_bcat(sb, sbuf_data(s), MIN(sbuf_len(s), maxlen == -1 ? 2990 SIZE_T_MAX : maxlen)); 2991 } 2992 sbuf_delete(s); 2993 return (error); 2994 } 2995 2996 static int 2997 sysctl_kern_proc_kqueue_one(struct thread *td, struct sbuf *s, struct proc *p, 2998 int kq_fd, bool compat32) 2999 { 3000 struct file *fp; 3001 struct kqueue *kq; 3002 int error; 3003 3004 error = fget_remote(td, p, kq_fd, &fp); 3005 if (error == 0) { 3006 if (fp->f_type != DTYPE_KQUEUE) { 3007 error = EINVAL; 3008 } else { 3009 kq = fp->f_data; 3010 error = kern_proc_kqueue_report(s, p, kq_fd, kq, 3011 compat32); 3012 } 3013 fdrop(fp, td); 3014 } 3015 return (error); 3016 } 3017 3018 static int 3019 sysctl_kern_proc_kqueue(SYSCTL_HANDLER_ARGS) 3020 { 3021 struct thread *td; 3022 struct proc *p; 3023 struct sbuf *s, sm; 3024 int error, error1, *name; 3025 bool compat32; 3026 3027 name = (int *)arg1; 3028 if ((u_int)arg2 > 2 || (u_int)arg2 == 0) 3029 return (EINVAL); 3030 3031 error = pget((pid_t)name[0], PGET_HOLD | PGET_CANDEBUG, &p); 3032 if (error != 0) 3033 return (error); 3034 3035 td = curthread; 3036 #ifdef FREEBSD_COMPAT32 3037 compat32 = SV_CURPROC_FLAG(SV_ILP32); 3038 #else 3039 compat32 = false; 3040 #endif 3041 3042 s = sbuf_new_for_sysctl(&sm, NULL, 0, req); 3043 if (s == NULL) { 3044 error = ENOMEM; 3045 goto out; 3046 } 3047 sbuf_clear_flags(s, SBUF_INCLUDENUL); 3048 3049 if ((u_int)arg2 == 1) { 3050 error = kern_proc_kqueues_out1(td, p, s, compat32); 3051 } else { 3052 error = sysctl_kern_proc_kqueue_one(td, s, p, 3053 name[1] /* kq_fd */, compat32); 3054 } 3055 3056 error1 = sbuf_finish(s); 3057 if (error == 0) 3058 error = error1; 3059 sbuf_delete(s); 3060 3061 out: 3062 PRELE(p); 3063 return (error); 3064 } 3065 3066 static SYSCTL_NODE(_kern_proc, KERN_PROC_KQUEUE, kq, 3067 CTLFLAG_RD | CTLFLAG_MPSAFE, 3068 sysctl_kern_proc_kqueue, "KQueue events"); 3069