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/jail.h> 54 #include <sys/kthread.h> 55 #include <sys/selinfo.h> 56 #include <sys/queue.h> 57 #include <sys/event.h> 58 #include <sys/eventvar.h> 59 #include <sys/poll.h> 60 #include <sys/protosw.h> 61 #include <sys/resourcevar.h> 62 #include <sys/sbuf.h> 63 #include <sys/sigio.h> 64 #include <sys/signalvar.h> 65 #include <sys/socket.h> 66 #include <sys/socketvar.h> 67 #include <sys/stat.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/sysproto.h> 71 #include <sys/syscallsubr.h> 72 #include <sys/taskqueue.h> 73 #include <sys/uio.h> 74 #include <sys/user.h> 75 #ifdef KTRACE 76 #include <sys/ktrace.h> 77 #endif 78 #include <machine/atomic.h> 79 #ifdef COMPAT_FREEBSD32 80 #include <compat/freebsd32/freebsd32.h> 81 #include <compat/freebsd32/freebsd32_util.h> 82 #endif 83 84 #include <vm/uma.h> 85 86 static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 87 88 /* 89 * This lock is used if multiple kq locks are required. This possibly 90 * should be made into a per proc lock. 91 */ 92 static struct mtx kq_global; 93 MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF); 94 #define KQ_GLOBAL_LOCK(lck, haslck) do { \ 95 if (!haslck) \ 96 mtx_lock(lck); \ 97 haslck = 1; \ 98 } while (0) 99 #define KQ_GLOBAL_UNLOCK(lck, haslck) do { \ 100 if (haslck) \ 101 mtx_unlock(lck); \ 102 haslck = 0; \ 103 } while (0) 104 105 TASKQUEUE_DEFINE_THREAD(kqueue_ctx); 106 107 static int kevent_copyout(void *arg, struct kevent *kevp, int count); 108 static int kevent_copyin(void *arg, struct kevent *kevp, int count); 109 static int kqueue_register(struct kqueue *kq, struct kevent *kev, 110 struct thread *td, int mflag); 111 static int kqueue_acquire(struct file *fp, struct kqueue **kqp); 112 static void kqueue_release(struct kqueue *kq, int locked); 113 static void kqueue_destroy(struct kqueue *kq); 114 static void kqueue_drain(struct kqueue *kq, struct thread *td); 115 static int kqueue_expand(struct kqueue *kq, const struct filterops *fops, 116 uintptr_t ident, int mflag); 117 static void kqueue_task(void *arg, int pending); 118 static int kqueue_scan(struct kqueue *kq, int maxevents, 119 struct kevent_copyops *k_ops, 120 const struct timespec *timeout, 121 struct kevent *keva, struct thread *td); 122 static void kqueue_wakeup(struct kqueue *kq); 123 static const struct filterops *kqueue_fo_find(int filt); 124 static void kqueue_fo_release(int filt); 125 struct g_kevent_args; 126 static int kern_kevent_generic(struct thread *td, 127 struct g_kevent_args *uap, 128 struct kevent_copyops *k_ops, const char *struct_name); 129 130 static fo_ioctl_t kqueue_ioctl; 131 static fo_poll_t kqueue_poll; 132 static fo_kqfilter_t kqueue_kqfilter; 133 static fo_stat_t kqueue_stat; 134 static fo_close_t kqueue_close; 135 static fo_fill_kinfo_t kqueue_fill_kinfo; 136 137 static const struct fileops kqueueops = { 138 .fo_read = invfo_rdwr, 139 .fo_write = invfo_rdwr, 140 .fo_truncate = invfo_truncate, 141 .fo_ioctl = kqueue_ioctl, 142 .fo_poll = kqueue_poll, 143 .fo_kqfilter = kqueue_kqfilter, 144 .fo_stat = kqueue_stat, 145 .fo_close = kqueue_close, 146 .fo_chmod = invfo_chmod, 147 .fo_chown = invfo_chown, 148 .fo_sendfile = invfo_sendfile, 149 .fo_cmp = file_kcmp_generic, 150 .fo_fill_kinfo = kqueue_fill_kinfo, 151 }; 152 153 static int knote_attach(struct knote *kn, struct kqueue *kq); 154 static void knote_drop(struct knote *kn, struct thread *td); 155 static void knote_drop_detached(struct knote *kn, struct thread *td); 156 static void knote_enqueue(struct knote *kn); 157 static void knote_dequeue(struct knote *kn); 158 static void knote_init(void); 159 static struct knote *knote_alloc(int mflag); 160 static void knote_free(struct knote *kn); 161 162 static void filt_kqdetach(struct knote *kn); 163 static int filt_kqueue(struct knote *kn, long hint); 164 static int filt_procattach(struct knote *kn); 165 static void filt_procdetach(struct knote *kn); 166 static int filt_proc(struct knote *kn, long hint); 167 static int filt_jailattach(struct knote *kn); 168 static void filt_jaildetach(struct knote *kn); 169 static int filt_jail(struct knote *kn, long hint); 170 static int filt_fileattach(struct knote *kn); 171 static void filt_timerexpire(void *knx); 172 static void filt_timerexpire_l(struct knote *kn, bool proc_locked); 173 static int filt_timerattach(struct knote *kn); 174 static void filt_timerdetach(struct knote *kn); 175 static void filt_timerstart(struct knote *kn, sbintime_t to); 176 static void filt_timertouch(struct knote *kn, struct kevent *kev, 177 u_long type); 178 static int filt_timervalidate(struct knote *kn, sbintime_t *to); 179 static int filt_timer(struct knote *kn, long hint); 180 static int filt_userattach(struct knote *kn); 181 static void filt_userdetach(struct knote *kn); 182 static int filt_user(struct knote *kn, long hint); 183 static void filt_usertouch(struct knote *kn, struct kevent *kev, 184 u_long type); 185 186 static const struct filterops file_filtops = { 187 .f_isfd = 1, 188 .f_attach = filt_fileattach, 189 }; 190 static const struct filterops kqread_filtops = { 191 .f_isfd = 1, 192 .f_detach = filt_kqdetach, 193 .f_event = filt_kqueue, 194 }; 195 /* XXX - move to kern_proc.c? */ 196 static const struct filterops proc_filtops = { 197 .f_isfd = 0, 198 .f_attach = filt_procattach, 199 .f_detach = filt_procdetach, 200 .f_event = filt_proc, 201 }; 202 static const struct filterops jail_filtops = { 203 .f_isfd = 0, 204 .f_attach = filt_jailattach, 205 .f_detach = filt_jaildetach, 206 .f_event = filt_jail, 207 }; 208 static const struct filterops timer_filtops = { 209 .f_isfd = 0, 210 .f_attach = filt_timerattach, 211 .f_detach = filt_timerdetach, 212 .f_event = filt_timer, 213 .f_touch = filt_timertouch, 214 }; 215 static const struct filterops user_filtops = { 216 .f_attach = filt_userattach, 217 .f_detach = filt_userdetach, 218 .f_event = filt_user, 219 .f_touch = filt_usertouch, 220 }; 221 222 static uma_zone_t knote_zone; 223 static unsigned int __exclusive_cache_line kq_ncallouts; 224 static unsigned int kq_calloutmax = 4 * 1024; 225 SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 226 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 227 228 /* XXX - ensure not influx ? */ 229 #define KNOTE_ACTIVATE(kn, islock) do { \ 230 if ((islock)) \ 231 mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED); \ 232 else \ 233 KQ_LOCK((kn)->kn_kq); \ 234 (kn)->kn_status |= KN_ACTIVE; \ 235 if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 236 knote_enqueue((kn)); \ 237 if (!(islock)) \ 238 KQ_UNLOCK((kn)->kn_kq); \ 239 } while (0) 240 #define KQ_LOCK(kq) do { \ 241 mtx_lock(&(kq)->kq_lock); \ 242 } while (0) 243 #define KQ_FLUX_WAKEUP(kq) do { \ 244 if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) { \ 245 (kq)->kq_state &= ~KQ_FLUXWAIT; \ 246 wakeup((kq)); \ 247 } \ 248 } while (0) 249 #define KQ_UNLOCK_FLUX(kq) do { \ 250 KQ_FLUX_WAKEUP(kq); \ 251 mtx_unlock(&(kq)->kq_lock); \ 252 } while (0) 253 #define KQ_UNLOCK(kq) do { \ 254 mtx_unlock(&(kq)->kq_lock); \ 255 } while (0) 256 #define KQ_OWNED(kq) do { \ 257 mtx_assert(&(kq)->kq_lock, MA_OWNED); \ 258 } while (0) 259 #define KQ_NOTOWNED(kq) do { \ 260 mtx_assert(&(kq)->kq_lock, MA_NOTOWNED); \ 261 } while (0) 262 263 static struct knlist * 264 kn_list_lock(struct knote *kn) 265 { 266 struct knlist *knl; 267 268 knl = kn->kn_knlist; 269 if (knl != NULL) 270 knl->kl_lock(knl->kl_lockarg); 271 return (knl); 272 } 273 274 static void 275 kn_list_unlock(struct knlist *knl) 276 { 277 bool do_free; 278 279 if (knl == NULL) 280 return; 281 do_free = knl->kl_autodestroy && knlist_empty(knl); 282 knl->kl_unlock(knl->kl_lockarg); 283 if (do_free) { 284 knlist_destroy(knl); 285 free(knl, M_KQUEUE); 286 } 287 } 288 289 static bool 290 kn_in_flux(struct knote *kn) 291 { 292 293 return (kn->kn_influx > 0); 294 } 295 296 static void 297 kn_enter_flux(struct knote *kn) 298 { 299 300 KQ_OWNED(kn->kn_kq); 301 MPASS(kn->kn_influx < INT_MAX); 302 kn->kn_influx++; 303 } 304 305 static bool 306 kn_leave_flux(struct knote *kn) 307 { 308 309 KQ_OWNED(kn->kn_kq); 310 MPASS(kn->kn_influx > 0); 311 kn->kn_influx--; 312 return (kn->kn_influx == 0); 313 } 314 315 #define KNL_ASSERT_LOCK(knl, islocked) do { \ 316 if (islocked) \ 317 KNL_ASSERT_LOCKED(knl); \ 318 else \ 319 KNL_ASSERT_UNLOCKED(knl); \ 320 } while (0) 321 #ifdef INVARIANTS 322 #define KNL_ASSERT_LOCKED(knl) do { \ 323 knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED); \ 324 } while (0) 325 #define KNL_ASSERT_UNLOCKED(knl) do { \ 326 knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED); \ 327 } while (0) 328 #else /* !INVARIANTS */ 329 #define KNL_ASSERT_LOCKED(knl) do {} while (0) 330 #define KNL_ASSERT_UNLOCKED(knl) do {} while (0) 331 #endif /* INVARIANTS */ 332 333 #ifndef KN_HASHSIZE 334 #define KN_HASHSIZE 64 /* XXX should be tunable */ 335 #endif 336 337 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 338 339 static int 340 filt_nullattach(struct knote *kn) 341 { 342 343 return (ENXIO); 344 }; 345 346 static const struct filterops null_filtops = { 347 .f_isfd = 0, 348 .f_attach = filt_nullattach, 349 }; 350 351 /* XXX - make SYSINIT to add these, and move into respective modules. */ 352 extern const struct filterops sig_filtops; 353 extern const struct filterops fs_filtops; 354 355 /* 356 * Table for all system-defined filters. 357 */ 358 static struct mtx filterops_lock; 359 MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF); 360 static struct { 361 const struct filterops *for_fop; 362 int for_nolock; 363 int for_refcnt; 364 } sysfilt_ops[EVFILT_SYSCOUNT] = { 365 [~EVFILT_READ] = { &file_filtops, 1 }, 366 [~EVFILT_WRITE] = { &file_filtops, 1 }, 367 [~EVFILT_AIO] = { &null_filtops }, 368 [~EVFILT_VNODE] = { &file_filtops, 1 }, 369 [~EVFILT_PROC] = { &proc_filtops, 1 }, 370 [~EVFILT_SIGNAL] = { &sig_filtops, 1 }, 371 [~EVFILT_TIMER] = { &timer_filtops, 1 }, 372 [~EVFILT_PROCDESC] = { &file_filtops, 1 }, 373 [~EVFILT_FS] = { &fs_filtops, 1 }, 374 [~EVFILT_LIO] = { &null_filtops }, 375 [~EVFILT_USER] = { &user_filtops, 1 }, 376 [~EVFILT_SENDFILE] = { &null_filtops }, 377 [~EVFILT_EMPTY] = { &file_filtops, 1 }, 378 [~EVFILT_JAIL] = { &jail_filtops, 1 }, 379 }; 380 381 /* 382 * Simple redirection for all cdevsw style objects to call their fo_kqfilter 383 * method. 384 */ 385 static int 386 filt_fileattach(struct knote *kn) 387 { 388 389 return (fo_kqfilter(kn->kn_fp, kn)); 390 } 391 392 /*ARGSUSED*/ 393 static int 394 kqueue_kqfilter(struct file *fp, struct knote *kn) 395 { 396 struct kqueue *kq = kn->kn_fp->f_data; 397 398 if (kn->kn_filter != EVFILT_READ) 399 return (EINVAL); 400 401 kn->kn_status |= KN_KQUEUE; 402 kn->kn_fop = &kqread_filtops; 403 knlist_add(&kq->kq_sel.si_note, kn, 0); 404 405 return (0); 406 } 407 408 static void 409 filt_kqdetach(struct knote *kn) 410 { 411 struct kqueue *kq = kn->kn_fp->f_data; 412 413 knlist_remove(&kq->kq_sel.si_note, kn, 0); 414 } 415 416 /*ARGSUSED*/ 417 static int 418 filt_kqueue(struct knote *kn, long hint) 419 { 420 struct kqueue *kq = kn->kn_fp->f_data; 421 422 kn->kn_data = kq->kq_count; 423 return (kn->kn_data > 0); 424 } 425 426 /* XXX - move to kern_proc.c? */ 427 static int 428 filt_procattach(struct knote *kn) 429 { 430 struct proc *p; 431 int error; 432 bool exiting, immediate; 433 434 exiting = immediate = false; 435 if (kn->kn_sfflags & NOTE_EXIT) 436 p = pfind_any(kn->kn_id); 437 else 438 p = pfind(kn->kn_id); 439 if (p == NULL) 440 return (ESRCH); 441 if (p->p_flag & P_WEXIT) 442 exiting = true; 443 444 if ((error = p_cansee(curthread, p))) { 445 PROC_UNLOCK(p); 446 return (error); 447 } 448 449 kn->kn_ptr.p_proc = p; 450 kn->kn_flags |= EV_CLEAR; /* automatically set */ 451 452 /* 453 * Internal flag indicating registration done by kernel for the 454 * purposes of getting a NOTE_CHILD notification. 455 */ 456 if (kn->kn_flags & EV_FLAG2) { 457 kn->kn_flags &= ~EV_FLAG2; 458 kn->kn_data = kn->kn_sdata; /* ppid */ 459 kn->kn_fflags = NOTE_CHILD; 460 kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK); 461 immediate = true; /* Force immediate activation of child note. */ 462 } 463 /* 464 * Internal flag indicating registration done by kernel (for other than 465 * NOTE_CHILD). 466 */ 467 if (kn->kn_flags & EV_FLAG1) { 468 kn->kn_flags &= ~EV_FLAG1; 469 } 470 471 knlist_add(p->p_klist, kn, 1); 472 473 /* 474 * Immediately activate any child notes or, in the case of a zombie 475 * target process, exit notes. The latter is necessary to handle the 476 * case where the target process, e.g. a child, dies before the kevent 477 * is registered. 478 */ 479 if (immediate || (exiting && filt_proc(kn, NOTE_EXIT))) 480 KNOTE_ACTIVATE(kn, 0); 481 482 PROC_UNLOCK(p); 483 484 return (0); 485 } 486 487 /* 488 * The knote may be attached to a different process, which may exit, 489 * leaving nothing for the knote to be attached to. So when the process 490 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 491 * it will be deleted when read out. However, as part of the knote deletion, 492 * this routine is called, so a check is needed to avoid actually performing 493 * a detach, because the original process does not exist any more. 494 */ 495 /* XXX - move to kern_proc.c? */ 496 static void 497 filt_procdetach(struct knote *kn) 498 { 499 500 knlist_remove(kn->kn_knlist, kn, 0); 501 kn->kn_ptr.p_proc = NULL; 502 } 503 504 /* XXX - move to kern_proc.c? */ 505 static int 506 filt_proc(struct knote *kn, long hint) 507 { 508 struct proc *p; 509 u_int event; 510 511 p = kn->kn_ptr.p_proc; 512 if (p == NULL) /* already activated, from attach filter */ 513 return (0); 514 515 /* Mask off extra data. */ 516 event = (u_int)hint & NOTE_PCTRLMASK; 517 518 /* If the user is interested in this event, record it. */ 519 if (kn->kn_sfflags & event) 520 kn->kn_fflags |= event; 521 522 /* Process is gone, so flag the event as finished. */ 523 if (event == NOTE_EXIT) { 524 kn->kn_flags |= EV_EOF | EV_ONESHOT; 525 kn->kn_ptr.p_proc = NULL; 526 if (kn->kn_fflags & NOTE_EXIT) 527 kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig); 528 if (kn->kn_fflags == 0) 529 kn->kn_flags |= EV_DROP; 530 return (1); 531 } 532 533 return (kn->kn_fflags != 0); 534 } 535 536 /* 537 * Called when the process forked. It mostly does the same as the 538 * knote(), activating all knotes registered to be activated when the 539 * process forked. Additionally, for each knote attached to the 540 * parent, check whether user wants to track the new process. If so 541 * attach a new knote to it, and immediately report an event with the 542 * child's pid. This is also called on jail creation, which is treated 543 * the same way by jail events. 544 */ 545 void 546 knote_fork(struct knlist *list, int pid) 547 { 548 struct kqueue *kq; 549 struct knote *kn; 550 struct kevent kev; 551 int error; 552 553 MPASS(list != NULL); 554 KNL_ASSERT_LOCKED(list); 555 if (SLIST_EMPTY(&list->kl_list)) 556 return; 557 558 memset(&kev, 0, sizeof(kev)); 559 SLIST_FOREACH(kn, &list->kl_list, kn_selnext) { 560 kq = kn->kn_kq; 561 KQ_LOCK(kq); 562 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 563 KQ_UNLOCK(kq); 564 continue; 565 } 566 567 /* 568 * The same as knote(), activate the event. 569 */ 570 _Static_assert(NOTE_JAIL_CHILD == NOTE_FORK, 571 "NOTE_JAIL_CHILD should be the same as NOTE_FORK"); 572 if ((kn->kn_sfflags & NOTE_TRACK) == 0) { 573 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 574 KNOTE_ACTIVATE(kn, 1); 575 KQ_UNLOCK(kq); 576 continue; 577 } 578 579 /* 580 * The NOTE_TRACK case. In addition to the activation 581 * of the event, we need to register new events to 582 * track the child. Drop the locks in preparation for 583 * the call to kqueue_register(). 584 */ 585 kn_enter_flux(kn); 586 KQ_UNLOCK(kq); 587 list->kl_unlock(list->kl_lockarg); 588 589 /* 590 * Activate existing knote and register tracking knotes with 591 * new process. 592 * 593 * First register a knote to get just the child notice. This 594 * must be a separate note from a potential NOTE_EXIT 595 * notification since both NOTE_CHILD and NOTE_EXIT are defined 596 * to use the data field (in conflicting ways). 597 */ 598 kev.ident = pid; 599 kev.filter = kn->kn_filter; 600 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT | 601 EV_FLAG2; 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; 608 609 /* 610 * Then register another knote to track other potential events 611 * from the new process. 612 */ 613 kev.ident = pid; 614 kev.filter = kn->kn_filter; 615 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 616 kev.fflags = kn->kn_sfflags; 617 kev.data = kn->kn_id; /* parent */ 618 kev.udata = kn->kn_kevent.udata;/* preserve udata */ 619 error = kqueue_register(kq, &kev, NULL, M_NOWAIT); 620 if (error) 621 kn->kn_fflags |= NOTE_TRACKERR; 622 if (kn->kn_fop->f_event(kn, NOTE_FORK)) 623 KNOTE_ACTIVATE(kn, 0); 624 list->kl_lock(list->kl_lockarg); 625 KQ_LOCK(kq); 626 kn_leave_flux(kn); 627 KQ_UNLOCK_FLUX(kq); 628 } 629 } 630 631 int 632 filt_jailattach(struct knote *kn) 633 { 634 struct prison *pr; 635 bool immediate; 636 637 immediate = false; 638 if (kn->kn_id == 0) { 639 /* Let jid=0 watch the current prison (including prison0). */ 640 pr = curthread->td_ucred->cr_prison; 641 mtx_lock(&pr->pr_mtx); 642 } else if (kn->kn_flags & (EV_FLAG1 | EV_FLAG2)) { 643 /* 644 * The kernel registers prisons before they are valid, 645 * so prison_find_child will fail. 646 */ 647 TAILQ_FOREACH(pr, &allprison, pr_list) { 648 if (pr->pr_id < kn->kn_id) 649 continue; 650 if (pr->pr_id > kn->kn_id) { 651 pr = NULL; 652 break; 653 } 654 mtx_lock(&pr->pr_mtx); 655 break; 656 } 657 if (pr == NULL) 658 return (ENOENT); 659 } else { 660 sx_slock(&allprison_lock); 661 pr = prison_find_child(curthread->td_ucred->cr_prison, 662 kn->kn_id); 663 sx_sunlock(&allprison_lock); 664 if (pr == NULL) 665 return (ENOENT); 666 if (!prison_isalive(pr)) { 667 mtx_unlock(&pr->pr_mtx); 668 return (ENOENT); 669 } 670 } 671 kn->kn_ptr.p_prison = pr; 672 kn->kn_flags |= EV_CLEAR; 673 674 /* 675 * Internal flag indicating registration done by kernel for the 676 * purposes of getting a NOTE_CHILD notification. 677 */ 678 if (kn->kn_flags & EV_FLAG2) { 679 kn->kn_flags &= ~EV_FLAG2; 680 kn->kn_data = kn->kn_sdata; /* parent id */ 681 kn->kn_fflags = NOTE_CHILD; 682 kn->kn_sfflags &= ~NOTE_JAIL_CTRLMASK; 683 immediate = true; /* Force immediate activation of child note. */ 684 } 685 /* 686 * Internal flag indicating registration done by kernel (for other than 687 * NOTE_CHILD). 688 */ 689 if (kn->kn_flags & EV_FLAG1) { 690 kn->kn_flags &= ~EV_FLAG1; 691 } 692 693 knlist_add(pr->pr_klist, kn, 1); 694 695 /* Immediately activate any child notes. */ 696 if (immediate) 697 KNOTE_ACTIVATE(kn, 0); 698 699 mtx_unlock(&pr->pr_mtx); 700 return (0); 701 } 702 703 void 704 filt_jaildetach(struct knote *kn) 705 { 706 if (kn->kn_ptr.p_prison != NULL) { 707 knlist_remove(kn->kn_knlist, kn, 0); 708 kn->kn_ptr.p_prison = NULL; 709 } else 710 kn->kn_status |= KN_DETACHED; 711 } 712 713 int 714 filt_jail(struct knote *kn, long hint) 715 { 716 struct prison *pr; 717 u_int event; 718 719 pr = kn->kn_ptr.p_prison; 720 if (pr == NULL) /* already activated, from attach filter */ 721 return (0); 722 723 /* Mask off extra data. */ 724 event = (u_int)hint & NOTE_JAIL_CTRLMASK; 725 726 /* If the user is interested in this event, record it. */ 727 if (kn->kn_sfflags & event) 728 kn->kn_fflags |= event; 729 730 /* Report the attached process id. */ 731 if (event == NOTE_JAIL_ATTACH) { 732 if (kn->kn_data != 0) 733 kn->kn_fflags |= NOTE_JAIL_ATTACH_MULTI; 734 kn->kn_data = hint & NOTE_JAIL_DATAMASK; 735 } 736 737 /* Prison is gone, so flag the event as finished. */ 738 if (event == NOTE_JAIL_REMOVE) { 739 kn->kn_flags |= EV_EOF | EV_ONESHOT; 740 kn->kn_ptr.p_prison = NULL; 741 if (kn->kn_fflags == 0) 742 kn->kn_flags |= EV_DROP; 743 return (1); 744 } 745 746 return (kn->kn_fflags != 0); 747 } 748 749 /* 750 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the 751 * interval timer support code. 752 */ 753 754 #define NOTE_TIMER_PRECMASK \ 755 (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS) 756 757 static sbintime_t 758 timer2sbintime(int64_t data, int flags) 759 { 760 int64_t secs; 761 762 /* 763 * Macros for converting to the fractional second portion of an 764 * sbintime_t using 64bit multiplication to improve precision. 765 */ 766 #define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32) 767 #define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32) 768 #define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32) 769 switch (flags & NOTE_TIMER_PRECMASK) { 770 case NOTE_SECONDS: 771 #ifdef __LP64__ 772 if (data > (SBT_MAX / SBT_1S)) 773 return (SBT_MAX); 774 #endif 775 return ((sbintime_t)data << 32); 776 case NOTE_MSECONDS: /* FALLTHROUGH */ 777 case 0: 778 if (data >= 1000) { 779 secs = data / 1000; 780 #ifdef __LP64__ 781 if (secs > (SBT_MAX / SBT_1S)) 782 return (SBT_MAX); 783 #endif 784 return (secs << 32 | MS_TO_SBT(data % 1000)); 785 } 786 return (MS_TO_SBT(data)); 787 case NOTE_USECONDS: 788 if (data >= 1000000) { 789 secs = data / 1000000; 790 #ifdef __LP64__ 791 if (secs > (SBT_MAX / SBT_1S)) 792 return (SBT_MAX); 793 #endif 794 return (secs << 32 | US_TO_SBT(data % 1000000)); 795 } 796 return (US_TO_SBT(data)); 797 case NOTE_NSECONDS: 798 if (data >= 1000000000) { 799 secs = data / 1000000000; 800 #ifdef __LP64__ 801 if (secs > (SBT_MAX / SBT_1S)) 802 return (SBT_MAX); 803 #endif 804 return (secs << 32 | NS_TO_SBT(data % 1000000000)); 805 } 806 return (NS_TO_SBT(data)); 807 default: 808 break; 809 } 810 return (-1); 811 } 812 813 struct kq_timer_cb_data { 814 struct callout c; 815 struct proc *p; 816 struct knote *kn; 817 int cpuid; 818 int flags; 819 TAILQ_ENTRY(kq_timer_cb_data) link; 820 sbintime_t next; /* next timer event fires at */ 821 sbintime_t to; /* precalculated timer period, 0 for abs */ 822 }; 823 824 #define KQ_TIMER_CB_ENQUEUED 0x01 825 826 static void 827 kqtimer_sched_callout(struct kq_timer_cb_data *kc) 828 { 829 callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn, 830 kc->cpuid, C_ABSOLUTE); 831 } 832 833 void 834 kqtimer_proc_continue(struct proc *p) 835 { 836 struct kq_timer_cb_data *kc, *kc1; 837 struct bintime bt; 838 sbintime_t now; 839 840 PROC_LOCK_ASSERT(p, MA_OWNED); 841 842 getboottimebin(&bt); 843 now = bttosbt(bt); 844 845 TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) { 846 TAILQ_REMOVE(&p->p_kqtim_stop, kc, link); 847 kc->flags &= ~KQ_TIMER_CB_ENQUEUED; 848 if (kc->next <= now) 849 filt_timerexpire_l(kc->kn, true); 850 else 851 kqtimer_sched_callout(kc); 852 } 853 } 854 855 static void 856 filt_timerexpire_l(struct knote *kn, bool proc_locked) 857 { 858 struct kq_timer_cb_data *kc; 859 struct proc *p; 860 uint64_t delta; 861 sbintime_t now; 862 863 kc = kn->kn_ptr.p_v; 864 865 if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) { 866 kn->kn_data++; 867 KNOTE_ACTIVATE(kn, 0); 868 return; 869 } 870 871 now = sbinuptime(); 872 if (now >= kc->next) { 873 delta = (now - kc->next) / kc->to; 874 if (delta == 0) 875 delta = 1; 876 kn->kn_data += delta; 877 kc->next += delta * kc->to; 878 if (now >= kc->next) /* overflow */ 879 kc->next = now + kc->to; 880 KNOTE_ACTIVATE(kn, 0); /* XXX - handle locking */ 881 } 882 883 /* 884 * Initial check for stopped kc->p is racy. It is fine to 885 * miss the set of the stop flags, at worst we would schedule 886 * one more callout. On the other hand, it is not fine to not 887 * schedule when we we missed clearing of the flags, we 888 * recheck them under the lock and observe consistent state. 889 */ 890 p = kc->p; 891 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 892 if (!proc_locked) 893 PROC_LOCK(p); 894 if (P_SHOULDSTOP(p) || P_KILLED(p)) { 895 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) { 896 kc->flags |= KQ_TIMER_CB_ENQUEUED; 897 TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link); 898 } 899 if (!proc_locked) 900 PROC_UNLOCK(p); 901 return; 902 } 903 if (!proc_locked) 904 PROC_UNLOCK(p); 905 } 906 kqtimer_sched_callout(kc); 907 } 908 909 static void 910 filt_timerexpire(void *knx) 911 { 912 filt_timerexpire_l(knx, false); 913 } 914 915 /* 916 * data contains amount of time to sleep 917 */ 918 static int 919 filt_timervalidate(struct knote *kn, sbintime_t *to) 920 { 921 struct bintime bt; 922 sbintime_t sbt; 923 924 if (kn->kn_sdata < 0) 925 return (EINVAL); 926 if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0) 927 kn->kn_sdata = 1; 928 /* 929 * The only fflags values supported are the timer unit 930 * (precision) and the absolute time indicator. 931 */ 932 if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0) 933 return (EINVAL); 934 935 *to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags); 936 if (*to < 0) 937 return (EINVAL); 938 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 939 getboottimebin(&bt); 940 sbt = bttosbt(bt); 941 *to = MAX(0, *to - sbt); 942 } 943 return (0); 944 } 945 946 static int 947 filt_timerattach(struct knote *kn) 948 { 949 struct kq_timer_cb_data *kc; 950 sbintime_t to; 951 int error; 952 953 to = -1; 954 error = filt_timervalidate(kn, &to); 955 if (error != 0) 956 return (error); 957 KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 || 958 (kn->kn_sfflags & NOTE_ABSTIME) != 0, 959 ("%s: periodic timer has a calculated zero timeout", __func__)); 960 KASSERT(to >= 0, 961 ("%s: timer has a calculated negative timeout", __func__)); 962 963 if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) { 964 atomic_subtract_int(&kq_ncallouts, 1); 965 return (ENOMEM); 966 } 967 968 if ((kn->kn_sfflags & NOTE_ABSTIME) == 0) 969 kn->kn_flags |= EV_CLEAR; /* automatically set */ 970 kn->kn_status &= ~KN_DETACHED; /* knlist_add clears it */ 971 kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK); 972 kc->kn = kn; 973 kc->p = curproc; 974 kc->cpuid = PCPU_GET(cpuid); 975 kc->flags = 0; 976 callout_init(&kc->c, 1); 977 filt_timerstart(kn, to); 978 979 return (0); 980 } 981 982 static void 983 filt_timerstart(struct knote *kn, sbintime_t to) 984 { 985 struct kq_timer_cb_data *kc; 986 987 kc = kn->kn_ptr.p_v; 988 if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) { 989 kc->next = to; 990 kc->to = 0; 991 } else { 992 kc->next = to + sbinuptime(); 993 kc->to = to; 994 } 995 kqtimer_sched_callout(kc); 996 } 997 998 static void 999 filt_timerdetach(struct knote *kn) 1000 { 1001 struct kq_timer_cb_data *kc; 1002 unsigned int old __unused; 1003 bool pending; 1004 1005 kc = kn->kn_ptr.p_v; 1006 do { 1007 callout_drain(&kc->c); 1008 1009 /* 1010 * kqtimer_proc_continue() might have rescheduled this callout. 1011 * Double-check, using the process mutex as an interlock. 1012 */ 1013 PROC_LOCK(kc->p); 1014 if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) { 1015 kc->flags &= ~KQ_TIMER_CB_ENQUEUED; 1016 TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link); 1017 } 1018 pending = callout_pending(&kc->c); 1019 PROC_UNLOCK(kc->p); 1020 } while (pending); 1021 free(kc, M_KQUEUE); 1022 old = atomic_fetchadd_int(&kq_ncallouts, -1); 1023 KASSERT(old > 0, ("Number of callouts cannot become negative")); 1024 kn->kn_status |= KN_DETACHED; /* knlist_remove sets it */ 1025 } 1026 1027 static void 1028 filt_timertouch(struct knote *kn, struct kevent *kev, u_long type) 1029 { 1030 struct kq_timer_cb_data *kc; 1031 struct kqueue *kq; 1032 sbintime_t to; 1033 int error; 1034 1035 switch (type) { 1036 case EVENT_REGISTER: 1037 /* Handle re-added timers that update data/fflags */ 1038 if (kev->flags & EV_ADD) { 1039 kc = kn->kn_ptr.p_v; 1040 1041 /* Drain any existing callout. */ 1042 callout_drain(&kc->c); 1043 1044 /* Throw away any existing undelivered record 1045 * of the timer expiration. This is done under 1046 * the presumption that if a process is 1047 * re-adding this timer with new parameters, 1048 * it is no longer interested in what may have 1049 * happened under the old parameters. If it is 1050 * interested, it can wait for the expiration, 1051 * delete the old timer definition, and then 1052 * add the new one. 1053 * 1054 * This has to be done while the kq is locked: 1055 * - if enqueued, dequeue 1056 * - make it no longer active 1057 * - clear the count of expiration events 1058 */ 1059 kq = kn->kn_kq; 1060 KQ_LOCK(kq); 1061 if (kn->kn_status & KN_QUEUED) 1062 knote_dequeue(kn); 1063 1064 kn->kn_status &= ~KN_ACTIVE; 1065 kn->kn_data = 0; 1066 KQ_UNLOCK(kq); 1067 1068 /* Reschedule timer based on new data/fflags */ 1069 kn->kn_sfflags = kev->fflags; 1070 kn->kn_sdata = kev->data; 1071 error = filt_timervalidate(kn, &to); 1072 if (error != 0) { 1073 kn->kn_flags |= EV_ERROR; 1074 kn->kn_data = error; 1075 } else 1076 filt_timerstart(kn, to); 1077 } 1078 break; 1079 1080 case EVENT_PROCESS: 1081 *kev = kn->kn_kevent; 1082 if (kn->kn_flags & EV_CLEAR) { 1083 kn->kn_data = 0; 1084 kn->kn_fflags = 0; 1085 } 1086 break; 1087 1088 default: 1089 panic("filt_timertouch() - invalid type (%ld)", type); 1090 break; 1091 } 1092 } 1093 1094 static int 1095 filt_timer(struct knote *kn, long hint) 1096 { 1097 1098 return (kn->kn_data != 0); 1099 } 1100 1101 static int 1102 filt_userattach(struct knote *kn) 1103 { 1104 1105 /* 1106 * EVFILT_USER knotes are not attached to anything in the kernel. 1107 */ 1108 kn->kn_hook = NULL; 1109 if (kn->kn_fflags & NOTE_TRIGGER) 1110 kn->kn_hookid = 1; 1111 else 1112 kn->kn_hookid = 0; 1113 return (0); 1114 } 1115 1116 static void 1117 filt_userdetach(__unused struct knote *kn) 1118 { 1119 1120 /* 1121 * EVFILT_USER knotes are not attached to anything in the kernel. 1122 */ 1123 } 1124 1125 static int 1126 filt_user(struct knote *kn, __unused long hint) 1127 { 1128 1129 return (kn->kn_hookid); 1130 } 1131 1132 static void 1133 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 1134 { 1135 u_int ffctrl; 1136 1137 switch (type) { 1138 case EVENT_REGISTER: 1139 if (kev->fflags & NOTE_TRIGGER) 1140 kn->kn_hookid = 1; 1141 1142 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 1143 kev->fflags &= NOTE_FFLAGSMASK; 1144 switch (ffctrl) { 1145 case NOTE_FFNOP: 1146 break; 1147 1148 case NOTE_FFAND: 1149 kn->kn_sfflags &= kev->fflags; 1150 break; 1151 1152 case NOTE_FFOR: 1153 kn->kn_sfflags |= kev->fflags; 1154 break; 1155 1156 case NOTE_FFCOPY: 1157 kn->kn_sfflags = kev->fflags; 1158 break; 1159 1160 default: 1161 /* XXX Return error? */ 1162 break; 1163 } 1164 kn->kn_sdata = kev->data; 1165 if (kev->flags & EV_CLEAR) { 1166 kn->kn_hookid = 0; 1167 kn->kn_data = 0; 1168 kn->kn_fflags = 0; 1169 } 1170 break; 1171 1172 case EVENT_PROCESS: 1173 *kev = kn->kn_kevent; 1174 kev->fflags = kn->kn_sfflags; 1175 kev->data = kn->kn_sdata; 1176 if (kn->kn_flags & EV_CLEAR) { 1177 kn->kn_hookid = 0; 1178 kn->kn_data = 0; 1179 kn->kn_fflags = 0; 1180 } 1181 break; 1182 1183 default: 1184 panic("filt_usertouch() - invalid type (%ld)", type); 1185 break; 1186 } 1187 } 1188 1189 int 1190 sys_kqueue(struct thread *td, struct kqueue_args *uap) 1191 { 1192 1193 return (kern_kqueue(td, 0, NULL)); 1194 } 1195 1196 int 1197 sys_kqueuex(struct thread *td, struct kqueuex_args *uap) 1198 { 1199 int flags; 1200 1201 if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0) 1202 return (EINVAL); 1203 flags = 0; 1204 if ((uap->flags & KQUEUE_CLOEXEC) != 0) 1205 flags |= O_CLOEXEC; 1206 return (kern_kqueue(td, flags, NULL)); 1207 } 1208 1209 static void 1210 kqueue_init(struct kqueue *kq) 1211 { 1212 1213 mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK); 1214 TAILQ_INIT(&kq->kq_head); 1215 knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock); 1216 TASK_INIT(&kq->kq_task, 0, kqueue_task, kq); 1217 } 1218 1219 int 1220 kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps) 1221 { 1222 struct filedesc *fdp; 1223 struct kqueue *kq; 1224 struct file *fp; 1225 struct ucred *cred; 1226 int fd, error; 1227 1228 fdp = td->td_proc->p_fd; 1229 cred = td->td_ucred; 1230 if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES))) 1231 return (ENOMEM); 1232 1233 error = falloc_caps(td, &fp, &fd, flags, fcaps); 1234 if (error != 0) { 1235 chgkqcnt(cred->cr_ruidinfo, -1, 0); 1236 return (error); 1237 } 1238 1239 /* An extra reference on `fp' has been held for us by falloc(). */ 1240 kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO); 1241 kqueue_init(kq); 1242 kq->kq_fdp = fdp; 1243 kq->kq_cred = crhold(cred); 1244 1245 FILEDESC_XLOCK(fdp); 1246 TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list); 1247 FILEDESC_XUNLOCK(fdp); 1248 1249 finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops); 1250 fdrop(fp, td); 1251 1252 td->td_retval[0] = fd; 1253 return (0); 1254 } 1255 1256 struct g_kevent_args { 1257 int fd; 1258 const void *changelist; 1259 int nchanges; 1260 void *eventlist; 1261 int nevents; 1262 const struct timespec *timeout; 1263 }; 1264 1265 int 1266 sys_kevent(struct thread *td, struct kevent_args *uap) 1267 { 1268 struct kevent_copyops k_ops = { 1269 .arg = uap, 1270 .k_copyout = kevent_copyout, 1271 .k_copyin = kevent_copyin, 1272 .kevent_size = sizeof(struct kevent), 1273 }; 1274 struct g_kevent_args gk_args = { 1275 .fd = uap->fd, 1276 .changelist = uap->changelist, 1277 .nchanges = uap->nchanges, 1278 .eventlist = uap->eventlist, 1279 .nevents = uap->nevents, 1280 .timeout = uap->timeout, 1281 }; 1282 1283 return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent")); 1284 } 1285 1286 static int 1287 kern_kevent_generic(struct thread *td, struct g_kevent_args *uap, 1288 struct kevent_copyops *k_ops, const char *struct_name) 1289 { 1290 struct timespec ts, *tsp; 1291 #ifdef KTRACE 1292 struct kevent *eventlist = uap->eventlist; 1293 #endif 1294 int error; 1295 1296 if (uap->timeout != NULL) { 1297 error = copyin(uap->timeout, &ts, sizeof(ts)); 1298 if (error) 1299 return (error); 1300 tsp = &ts; 1301 } else 1302 tsp = NULL; 1303 1304 #ifdef KTRACE 1305 if (KTRPOINT(td, KTR_STRUCT_ARRAY)) 1306 ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist, 1307 uap->nchanges, k_ops->kevent_size); 1308 #endif 1309 1310 error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents, 1311 k_ops, tsp); 1312 1313 #ifdef KTRACE 1314 if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY)) 1315 ktrstructarray(struct_name, UIO_USERSPACE, eventlist, 1316 td->td_retval[0], k_ops->kevent_size); 1317 #endif 1318 1319 return (error); 1320 } 1321 1322 /* 1323 * Copy 'count' items into the destination list pointed to by uap->eventlist. 1324 */ 1325 static int 1326 kevent_copyout(void *arg, struct kevent *kevp, int count) 1327 { 1328 struct kevent_args *uap; 1329 int error; 1330 1331 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1332 uap = (struct kevent_args *)arg; 1333 1334 error = copyout(kevp, uap->eventlist, count * sizeof *kevp); 1335 if (error == 0) 1336 uap->eventlist += count; 1337 return (error); 1338 } 1339 1340 /* 1341 * Copy 'count' items from the list pointed to by uap->changelist. 1342 */ 1343 static int 1344 kevent_copyin(void *arg, struct kevent *kevp, int count) 1345 { 1346 struct kevent_args *uap; 1347 int error; 1348 1349 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1350 uap = (struct kevent_args *)arg; 1351 1352 error = copyin(uap->changelist, kevp, count * sizeof *kevp); 1353 if (error == 0) 1354 uap->changelist += count; 1355 return (error); 1356 } 1357 1358 #ifdef COMPAT_FREEBSD11 1359 static int 1360 kevent11_copyout(void *arg, struct kevent *kevp, int count) 1361 { 1362 struct freebsd11_kevent_args *uap; 1363 struct freebsd11_kevent kev11; 1364 int error, i; 1365 1366 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1367 uap = (struct freebsd11_kevent_args *)arg; 1368 1369 for (i = 0; i < count; i++) { 1370 kev11.ident = kevp->ident; 1371 kev11.filter = kevp->filter; 1372 kev11.flags = kevp->flags; 1373 kev11.fflags = kevp->fflags; 1374 kev11.data = kevp->data; 1375 kev11.udata = kevp->udata; 1376 error = copyout(&kev11, uap->eventlist, sizeof(kev11)); 1377 if (error != 0) 1378 break; 1379 uap->eventlist++; 1380 kevp++; 1381 } 1382 return (error); 1383 } 1384 1385 /* 1386 * Copy 'count' items from the list pointed to by uap->changelist. 1387 */ 1388 static int 1389 kevent11_copyin(void *arg, struct kevent *kevp, int count) 1390 { 1391 struct freebsd11_kevent_args *uap; 1392 struct freebsd11_kevent kev11; 1393 int error, i; 1394 1395 KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count)); 1396 uap = (struct freebsd11_kevent_args *)arg; 1397 1398 for (i = 0; i < count; i++) { 1399 error = copyin(uap->changelist, &kev11, sizeof(kev11)); 1400 if (error != 0) 1401 break; 1402 kevp->ident = kev11.ident; 1403 kevp->filter = kev11.filter; 1404 kevp->flags = kev11.flags; 1405 kevp->fflags = kev11.fflags; 1406 kevp->data = (uintptr_t)kev11.data; 1407 kevp->udata = kev11.udata; 1408 bzero(&kevp->ext, sizeof(kevp->ext)); 1409 uap->changelist++; 1410 kevp++; 1411 } 1412 return (error); 1413 } 1414 1415 int 1416 freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap) 1417 { 1418 struct kevent_copyops k_ops = { 1419 .arg = uap, 1420 .k_copyout = kevent11_copyout, 1421 .k_copyin = kevent11_copyin, 1422 .kevent_size = sizeof(struct freebsd11_kevent), 1423 }; 1424 struct g_kevent_args gk_args = { 1425 .fd = uap->fd, 1426 .changelist = uap->changelist, 1427 .nchanges = uap->nchanges, 1428 .eventlist = uap->eventlist, 1429 .nevents = uap->nevents, 1430 .timeout = uap->timeout, 1431 }; 1432 1433 return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent")); 1434 } 1435 #endif 1436 1437 int 1438 kern_kevent(struct thread *td, int fd, int nchanges, int nevents, 1439 struct kevent_copyops *k_ops, const struct timespec *timeout) 1440 { 1441 cap_rights_t rights; 1442 struct file *fp; 1443 int error; 1444 1445 cap_rights_init_zero(&rights); 1446 if (nchanges > 0) 1447 cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE); 1448 if (nevents > 0) 1449 cap_rights_set_one(&rights, CAP_KQUEUE_EVENT); 1450 error = fget(td, fd, &rights, &fp); 1451 if (error != 0) 1452 return (error); 1453 1454 error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout); 1455 fdrop(fp, td); 1456 1457 return (error); 1458 } 1459 1460 static int 1461 kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents, 1462 struct kevent_copyops *k_ops, const struct timespec *timeout) 1463 { 1464 struct kevent keva[KQ_NEVENTS]; 1465 struct kevent *kevp, *changes; 1466 int i, n, nerrors, error; 1467 1468 if (nchanges < 0) 1469 return (EINVAL); 1470 1471 nerrors = 0; 1472 while (nchanges > 0) { 1473 n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges; 1474 error = k_ops->k_copyin(k_ops->arg, keva, n); 1475 if (error) 1476 return (error); 1477 changes = keva; 1478 for (i = 0; i < n; i++) { 1479 kevp = &changes[i]; 1480 if (!kevp->filter) 1481 continue; 1482 kevp->flags &= ~EV_SYSFLAGS; 1483 error = kqueue_register(kq, kevp, td, M_WAITOK); 1484 if (error || (kevp->flags & EV_RECEIPT)) { 1485 if (nevents == 0) 1486 return (error); 1487 kevp->flags = EV_ERROR; 1488 kevp->data = error; 1489 (void)k_ops->k_copyout(k_ops->arg, kevp, 1); 1490 nevents--; 1491 nerrors++; 1492 } 1493 } 1494 nchanges -= n; 1495 } 1496 if (nerrors) { 1497 td->td_retval[0] = nerrors; 1498 return (0); 1499 } 1500 1501 return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td)); 1502 } 1503 1504 int 1505 kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents, 1506 struct kevent_copyops *k_ops, const struct timespec *timeout) 1507 { 1508 struct kqueue *kq; 1509 int error; 1510 1511 error = kqueue_acquire(fp, &kq); 1512 if (error != 0) 1513 return (error); 1514 error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout); 1515 kqueue_release(kq, 0); 1516 return (error); 1517 } 1518 1519 /* 1520 * Performs a kevent() call on a temporarily created kqueue. This can be 1521 * used to perform one-shot polling, similar to poll() and select(). 1522 */ 1523 int 1524 kern_kevent_anonymous(struct thread *td, int nevents, 1525 struct kevent_copyops *k_ops) 1526 { 1527 struct kqueue kq = {}; 1528 int error; 1529 1530 kqueue_init(&kq); 1531 kq.kq_refcnt = 1; 1532 error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL); 1533 kqueue_drain(&kq, td); 1534 kqueue_destroy(&kq); 1535 return (error); 1536 } 1537 1538 int 1539 kqueue_add_filteropts(int filt, const struct filterops *filtops) 1540 { 1541 int error; 1542 1543 error = 0; 1544 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) { 1545 printf( 1546 "trying to add a filterop that is out of range: %d is beyond %d\n", 1547 ~filt, EVFILT_SYSCOUNT); 1548 return EINVAL; 1549 } 1550 mtx_lock(&filterops_lock); 1551 if (sysfilt_ops[~filt].for_fop != &null_filtops && 1552 sysfilt_ops[~filt].for_fop != NULL) 1553 error = EEXIST; 1554 else { 1555 sysfilt_ops[~filt].for_fop = filtops; 1556 sysfilt_ops[~filt].for_refcnt = 0; 1557 } 1558 mtx_unlock(&filterops_lock); 1559 1560 return (error); 1561 } 1562 1563 int 1564 kqueue_del_filteropts(int filt) 1565 { 1566 int error; 1567 1568 error = 0; 1569 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1570 return EINVAL; 1571 1572 mtx_lock(&filterops_lock); 1573 if (sysfilt_ops[~filt].for_fop == &null_filtops || 1574 sysfilt_ops[~filt].for_fop == NULL) 1575 error = EINVAL; 1576 else if (sysfilt_ops[~filt].for_refcnt != 0) 1577 error = EBUSY; 1578 else { 1579 sysfilt_ops[~filt].for_fop = &null_filtops; 1580 sysfilt_ops[~filt].for_refcnt = 0; 1581 } 1582 mtx_unlock(&filterops_lock); 1583 1584 return error; 1585 } 1586 1587 static const struct filterops * 1588 kqueue_fo_find(int filt) 1589 { 1590 1591 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1592 return NULL; 1593 1594 if (sysfilt_ops[~filt].for_nolock) 1595 return sysfilt_ops[~filt].for_fop; 1596 1597 mtx_lock(&filterops_lock); 1598 sysfilt_ops[~filt].for_refcnt++; 1599 if (sysfilt_ops[~filt].for_fop == NULL) 1600 sysfilt_ops[~filt].for_fop = &null_filtops; 1601 mtx_unlock(&filterops_lock); 1602 1603 return sysfilt_ops[~filt].for_fop; 1604 } 1605 1606 static void 1607 kqueue_fo_release(int filt) 1608 { 1609 1610 if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) 1611 return; 1612 1613 if (sysfilt_ops[~filt].for_nolock) 1614 return; 1615 1616 mtx_lock(&filterops_lock); 1617 KASSERT(sysfilt_ops[~filt].for_refcnt > 0, 1618 ("filter object refcount not valid on release")); 1619 sysfilt_ops[~filt].for_refcnt--; 1620 mtx_unlock(&filterops_lock); 1621 } 1622 1623 /* 1624 * A ref to kq (obtained via kqueue_acquire) must be held. 1625 */ 1626 static int 1627 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td, 1628 int mflag) 1629 { 1630 const struct filterops *fops; 1631 struct file *fp; 1632 struct knote *kn, *tkn; 1633 struct knlist *knl; 1634 int error, filt, event; 1635 int haskqglobal, filedesc_unlock; 1636 1637 if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE)) 1638 return (EINVAL); 1639 1640 fp = NULL; 1641 kn = NULL; 1642 knl = NULL; 1643 error = 0; 1644 haskqglobal = 0; 1645 filedesc_unlock = 0; 1646 1647 filt = kev->filter; 1648 fops = kqueue_fo_find(filt); 1649 if (fops == NULL) 1650 return EINVAL; 1651 1652 if (kev->flags & EV_ADD) { 1653 /* Reject an invalid flag pair early */ 1654 if (kev->flags & EV_KEEPUDATA) { 1655 tkn = NULL; 1656 error = EINVAL; 1657 goto done; 1658 } 1659 1660 /* 1661 * Prevent waiting with locks. Non-sleepable 1662 * allocation failures are handled in the loop, only 1663 * if the spare knote appears to be actually required. 1664 */ 1665 tkn = knote_alloc(mflag); 1666 } else { 1667 tkn = NULL; 1668 } 1669 1670 findkn: 1671 if (fops->f_isfd) { 1672 KASSERT(td != NULL, ("td is NULL")); 1673 if (kev->ident > INT_MAX) 1674 error = EBADF; 1675 else 1676 error = fget(td, kev->ident, &cap_event_rights, &fp); 1677 if (error) 1678 goto done; 1679 1680 if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops, 1681 kev->ident, M_NOWAIT) != 0) { 1682 /* try again */ 1683 fdrop(fp, td); 1684 fp = NULL; 1685 error = kqueue_expand(kq, fops, kev->ident, mflag); 1686 if (error) 1687 goto done; 1688 goto findkn; 1689 } 1690 1691 if (fp->f_type == DTYPE_KQUEUE) { 1692 /* 1693 * If we add some intelligence about what we are doing, 1694 * we should be able to support events on ourselves. 1695 * We need to know when we are doing this to prevent 1696 * getting both the knlist lock and the kq lock since 1697 * they are the same thing. 1698 */ 1699 if (fp->f_data == kq) { 1700 error = EINVAL; 1701 goto done; 1702 } 1703 1704 /* 1705 * Pre-lock the filedesc before the global 1706 * lock mutex, see the comment in 1707 * kqueue_close(). 1708 */ 1709 FILEDESC_XLOCK(td->td_proc->p_fd); 1710 filedesc_unlock = 1; 1711 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 1712 } 1713 1714 KQ_LOCK(kq); 1715 if (kev->ident < kq->kq_knlistsize) { 1716 SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link) 1717 if (kev->filter == kn->kn_filter) 1718 break; 1719 } 1720 } else { 1721 if ((kev->flags & EV_ADD) == EV_ADD) { 1722 error = kqueue_expand(kq, fops, kev->ident, mflag); 1723 if (error != 0) 1724 goto done; 1725 } 1726 1727 KQ_LOCK(kq); 1728 1729 /* 1730 * If possible, find an existing knote to use for this kevent. 1731 */ 1732 if ((kev->filter == EVFILT_PROC || kev->filter == EVFILT_JAIL) 1733 && (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) { 1734 /* This is an internal creation of a process tracking 1735 * note. Don't attempt to coalesce this with an 1736 * existing note. 1737 */ 1738 ; 1739 } else if (kq->kq_knhashmask != 0) { 1740 struct klist *list; 1741 1742 list = &kq->kq_knhash[ 1743 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1744 SLIST_FOREACH(kn, list, kn_link) 1745 if (kev->ident == kn->kn_id && 1746 kev->filter == kn->kn_filter) 1747 break; 1748 } 1749 } 1750 1751 /* knote is in the process of changing, wait for it to stabilize. */ 1752 if (kn != NULL && kn_in_flux(kn)) { 1753 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1754 if (filedesc_unlock) { 1755 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1756 filedesc_unlock = 0; 1757 } 1758 kq->kq_state |= KQ_FLUXWAIT; 1759 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0); 1760 if (fp != NULL) { 1761 fdrop(fp, td); 1762 fp = NULL; 1763 } 1764 goto findkn; 1765 } 1766 1767 /* 1768 * kn now contains the matching knote, or NULL if no match 1769 */ 1770 if (kn == NULL) { 1771 if (kev->flags & EV_ADD) { 1772 kn = tkn; 1773 tkn = NULL; 1774 if (kn == NULL) { 1775 KQ_UNLOCK(kq); 1776 error = ENOMEM; 1777 goto done; 1778 } 1779 kn->kn_fp = fp; 1780 kn->kn_kq = kq; 1781 kn->kn_fop = fops; 1782 /* 1783 * apply reference counts to knote structure, and 1784 * do not release it at the end of this routine. 1785 */ 1786 fops = NULL; 1787 fp = NULL; 1788 1789 kn->kn_sfflags = kev->fflags; 1790 kn->kn_sdata = kev->data; 1791 kev->fflags = 0; 1792 kev->data = 0; 1793 kn->kn_kevent = *kev; 1794 kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE | 1795 EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT); 1796 kn->kn_status = KN_DETACHED; 1797 if ((kev->flags & EV_DISABLE) != 0) 1798 kn->kn_status |= KN_DISABLED; 1799 kn_enter_flux(kn); 1800 1801 error = knote_attach(kn, kq); 1802 KQ_UNLOCK(kq); 1803 if (error != 0) { 1804 tkn = kn; 1805 goto done; 1806 } 1807 1808 if ((error = kn->kn_fop->f_attach(kn)) != 0) { 1809 knote_drop_detached(kn, td); 1810 goto done; 1811 } 1812 knl = kn_list_lock(kn); 1813 goto done_ev_add; 1814 } else { 1815 /* No matching knote and the EV_ADD flag is not set. */ 1816 KQ_UNLOCK(kq); 1817 error = ENOENT; 1818 goto done; 1819 } 1820 } 1821 1822 if (kev->flags & EV_DELETE) { 1823 kn_enter_flux(kn); 1824 KQ_UNLOCK(kq); 1825 knote_drop(kn, td); 1826 goto done; 1827 } 1828 1829 if (kev->flags & EV_FORCEONESHOT) { 1830 kn->kn_flags |= EV_ONESHOT; 1831 KNOTE_ACTIVATE(kn, 1); 1832 } 1833 1834 if ((kev->flags & EV_ENABLE) != 0) 1835 kn->kn_status &= ~KN_DISABLED; 1836 else if ((kev->flags & EV_DISABLE) != 0) 1837 kn->kn_status |= KN_DISABLED; 1838 1839 /* 1840 * The user may change some filter values after the initial EV_ADD, 1841 * but doing so will not reset any filter which has already been 1842 * triggered. 1843 */ 1844 kn->kn_status |= KN_SCAN; 1845 kn_enter_flux(kn); 1846 KQ_UNLOCK(kq); 1847 knl = kn_list_lock(kn); 1848 if ((kev->flags & EV_KEEPUDATA) == 0) 1849 kn->kn_kevent.udata = kev->udata; 1850 if (!fops->f_isfd && fops->f_touch != NULL) { 1851 fops->f_touch(kn, kev, EVENT_REGISTER); 1852 } else { 1853 kn->kn_sfflags = kev->fflags; 1854 kn->kn_sdata = kev->data; 1855 } 1856 1857 done_ev_add: 1858 /* 1859 * We can get here with kn->kn_knlist == NULL. This can happen when 1860 * the initial attach event decides that the event is "completed" 1861 * already, e.g., filt_procattach() is called on a zombie process. It 1862 * will call filt_proc() which will remove it from the list, and NULL 1863 * kn_knlist. 1864 * 1865 * KN_DISABLED will be stable while the knote is in flux, so the 1866 * unlocked read will not race with an update. 1867 */ 1868 if ((kn->kn_status & KN_DISABLED) == 0) 1869 event = kn->kn_fop->f_event(kn, 0); 1870 else 1871 event = 0; 1872 1873 KQ_LOCK(kq); 1874 if (event) 1875 kn->kn_status |= KN_ACTIVE; 1876 if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) == 1877 KN_ACTIVE) 1878 knote_enqueue(kn); 1879 kn->kn_status &= ~KN_SCAN; 1880 kn_leave_flux(kn); 1881 kn_list_unlock(knl); 1882 KQ_UNLOCK_FLUX(kq); 1883 1884 done: 1885 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 1886 if (filedesc_unlock) 1887 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1888 if (fp != NULL) 1889 fdrop(fp, td); 1890 knote_free(tkn); 1891 if (fops != NULL) 1892 kqueue_fo_release(filt); 1893 return (error); 1894 } 1895 1896 static int 1897 kqueue_acquire(struct file *fp, struct kqueue **kqp) 1898 { 1899 int error; 1900 struct kqueue *kq; 1901 1902 error = 0; 1903 1904 kq = fp->f_data; 1905 if (fp->f_type != DTYPE_KQUEUE || kq == NULL) 1906 return (EINVAL); 1907 *kqp = kq; 1908 KQ_LOCK(kq); 1909 if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) { 1910 KQ_UNLOCK(kq); 1911 return (EBADF); 1912 } 1913 kq->kq_refcnt++; 1914 KQ_UNLOCK(kq); 1915 1916 return error; 1917 } 1918 1919 static void 1920 kqueue_release(struct kqueue *kq, int locked) 1921 { 1922 if (locked) 1923 KQ_OWNED(kq); 1924 else 1925 KQ_LOCK(kq); 1926 kq->kq_refcnt--; 1927 if (kq->kq_refcnt == 1) 1928 wakeup(&kq->kq_refcnt); 1929 if (!locked) 1930 KQ_UNLOCK(kq); 1931 } 1932 1933 static void 1934 ast_kqueue(struct thread *td, int tda __unused) 1935 { 1936 taskqueue_quiesce(taskqueue_kqueue_ctx); 1937 } 1938 1939 static void 1940 kqueue_schedtask(struct kqueue *kq) 1941 { 1942 KQ_OWNED(kq); 1943 KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN), 1944 ("scheduling kqueue task while draining")); 1945 1946 if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) { 1947 taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task); 1948 kq->kq_state |= KQ_TASKSCHED; 1949 ast_sched(curthread, TDA_KQUEUE); 1950 } 1951 } 1952 1953 /* 1954 * Expand the kq to make sure we have storage for fops/ident pair. 1955 * 1956 * Return 0 on success (or no work necessary), return errno on failure. 1957 */ 1958 static int 1959 kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident, 1960 int mflag) 1961 { 1962 struct klist *list, *tmp_knhash, *to_free; 1963 u_long tmp_knhashmask; 1964 int error, fd, size; 1965 1966 KQ_NOTOWNED(kq); 1967 1968 error = 0; 1969 to_free = NULL; 1970 if (fops->f_isfd) { 1971 fd = ident; 1972 if (kq->kq_knlistsize <= fd) { 1973 size = kq->kq_knlistsize; 1974 while (size <= fd) 1975 size += KQEXTENT; 1976 list = malloc(size * sizeof(*list), M_KQUEUE, mflag); 1977 if (list == NULL) 1978 return ENOMEM; 1979 KQ_LOCK(kq); 1980 if ((kq->kq_state & KQ_CLOSING) != 0) { 1981 to_free = list; 1982 error = EBADF; 1983 } else if (kq->kq_knlistsize > fd) { 1984 to_free = list; 1985 } else { 1986 if (kq->kq_knlist != NULL) { 1987 bcopy(kq->kq_knlist, list, 1988 kq->kq_knlistsize * sizeof(*list)); 1989 to_free = kq->kq_knlist; 1990 kq->kq_knlist = NULL; 1991 } 1992 bzero((caddr_t)list + 1993 kq->kq_knlistsize * sizeof(*list), 1994 (size - kq->kq_knlistsize) * sizeof(*list)); 1995 kq->kq_knlistsize = size; 1996 kq->kq_knlist = list; 1997 } 1998 KQ_UNLOCK(kq); 1999 } 2000 } else { 2001 if (kq->kq_knhashmask == 0) { 2002 tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE, 2003 &tmp_knhashmask, (mflag & M_WAITOK) != 0 ? 2004 HASH_WAITOK : HASH_NOWAIT); 2005 if (tmp_knhash == NULL) 2006 return (ENOMEM); 2007 KQ_LOCK(kq); 2008 if ((kq->kq_state & KQ_CLOSING) != 0) { 2009 to_free = tmp_knhash; 2010 error = EBADF; 2011 } else if (kq->kq_knhashmask == 0) { 2012 kq->kq_knhash = tmp_knhash; 2013 kq->kq_knhashmask = tmp_knhashmask; 2014 } else { 2015 to_free = tmp_knhash; 2016 } 2017 KQ_UNLOCK(kq); 2018 } 2019 } 2020 free(to_free, M_KQUEUE); 2021 2022 KQ_NOTOWNED(kq); 2023 return (error); 2024 } 2025 2026 static void 2027 kqueue_task(void *arg, int pending) 2028 { 2029 struct kqueue *kq; 2030 int haskqglobal; 2031 2032 haskqglobal = 0; 2033 kq = arg; 2034 2035 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 2036 KQ_LOCK(kq); 2037 2038 KNOTE_LOCKED(&kq->kq_sel.si_note, 0); 2039 2040 kq->kq_state &= ~KQ_TASKSCHED; 2041 if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) { 2042 wakeup(&kq->kq_state); 2043 } 2044 KQ_UNLOCK(kq); 2045 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2046 } 2047 2048 /* 2049 * Scan, update kn_data (if not ONESHOT), and copyout triggered events. 2050 * We treat KN_MARKER knotes as if they are in flux. 2051 */ 2052 static int 2053 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops, 2054 const struct timespec *tsp, struct kevent *keva, struct thread *td) 2055 { 2056 struct kevent *kevp; 2057 struct knote *kn, *marker; 2058 struct knlist *knl; 2059 sbintime_t asbt, rsbt; 2060 int count, error, haskqglobal, influx, nkev, touch; 2061 2062 count = maxevents; 2063 nkev = 0; 2064 error = 0; 2065 haskqglobal = 0; 2066 2067 if (maxevents == 0) 2068 goto done_nl; 2069 if (maxevents < 0) { 2070 error = EINVAL; 2071 goto done_nl; 2072 } 2073 2074 rsbt = 0; 2075 if (tsp != NULL) { 2076 if (!timespecvalid_interval(tsp)) { 2077 error = EINVAL; 2078 goto done_nl; 2079 } 2080 if (timespecisset(tsp)) { 2081 if (tsp->tv_sec <= INT32_MAX) { 2082 rsbt = tstosbt(*tsp); 2083 if (TIMESEL(&asbt, rsbt)) 2084 asbt += tc_tick_sbt; 2085 if (asbt <= SBT_MAX - rsbt) 2086 asbt += rsbt; 2087 else 2088 asbt = 0; 2089 rsbt >>= tc_precexp; 2090 } else 2091 asbt = 0; 2092 } else 2093 asbt = -1; 2094 } else 2095 asbt = 0; 2096 marker = knote_alloc(M_WAITOK); 2097 marker->kn_status = KN_MARKER; 2098 KQ_LOCK(kq); 2099 2100 retry: 2101 kevp = keva; 2102 if (kq->kq_count == 0) { 2103 if (asbt == -1) { 2104 error = EWOULDBLOCK; 2105 } else { 2106 kq->kq_state |= KQ_SLEEP; 2107 error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH, 2108 "kqread", asbt, rsbt, C_ABSOLUTE); 2109 } 2110 if (error == 0) 2111 goto retry; 2112 /* don't restart after signals... */ 2113 if (error == ERESTART) 2114 error = EINTR; 2115 else if (error == EWOULDBLOCK) 2116 error = 0; 2117 goto done; 2118 } 2119 2120 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 2121 influx = 0; 2122 while (count) { 2123 KQ_OWNED(kq); 2124 kn = TAILQ_FIRST(&kq->kq_head); 2125 2126 if ((kn->kn_status == KN_MARKER && kn != marker) || 2127 kn_in_flux(kn)) { 2128 if (influx) { 2129 influx = 0; 2130 KQ_FLUX_WAKEUP(kq); 2131 } 2132 kq->kq_state |= KQ_FLUXWAIT; 2133 error = msleep(kq, &kq->kq_lock, PSOCK, 2134 "kqflxwt", 0); 2135 continue; 2136 } 2137 2138 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2139 if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) { 2140 kn->kn_status &= ~KN_QUEUED; 2141 kq->kq_count--; 2142 continue; 2143 } 2144 if (kn == marker) { 2145 KQ_FLUX_WAKEUP(kq); 2146 if (count == maxevents) 2147 goto retry; 2148 goto done; 2149 } 2150 KASSERT(!kn_in_flux(kn), 2151 ("knote %p is unexpectedly in flux", kn)); 2152 2153 if ((kn->kn_flags & EV_DROP) == EV_DROP) { 2154 kn->kn_status &= ~KN_QUEUED; 2155 kn_enter_flux(kn); 2156 kq->kq_count--; 2157 KQ_UNLOCK(kq); 2158 /* 2159 * We don't need to lock the list since we've 2160 * marked it as in flux. 2161 */ 2162 knote_drop(kn, td); 2163 KQ_LOCK(kq); 2164 continue; 2165 } else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) { 2166 kn->kn_status &= ~KN_QUEUED; 2167 kn_enter_flux(kn); 2168 kq->kq_count--; 2169 KQ_UNLOCK(kq); 2170 /* 2171 * We don't need to lock the list since we've 2172 * marked the knote as being in flux. 2173 */ 2174 *kevp = kn->kn_kevent; 2175 knote_drop(kn, td); 2176 KQ_LOCK(kq); 2177 kn = NULL; 2178 } else { 2179 kn->kn_status |= KN_SCAN; 2180 kn_enter_flux(kn); 2181 KQ_UNLOCK(kq); 2182 if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE) 2183 KQ_GLOBAL_LOCK(&kq_global, haskqglobal); 2184 knl = kn_list_lock(kn); 2185 if (kn->kn_fop->f_event(kn, 0) == 0) { 2186 KQ_LOCK(kq); 2187 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2188 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE | 2189 KN_SCAN); 2190 kn_leave_flux(kn); 2191 kq->kq_count--; 2192 kn_list_unlock(knl); 2193 influx = 1; 2194 continue; 2195 } 2196 touch = (!kn->kn_fop->f_isfd && 2197 kn->kn_fop->f_touch != NULL); 2198 if (touch) 2199 kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS); 2200 else 2201 *kevp = kn->kn_kevent; 2202 KQ_LOCK(kq); 2203 KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal); 2204 if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 2205 /* 2206 * Manually clear knotes who weren't 2207 * 'touch'ed. 2208 */ 2209 if (touch == 0 && kn->kn_flags & EV_CLEAR) { 2210 kn->kn_data = 0; 2211 kn->kn_fflags = 0; 2212 } 2213 if (kn->kn_flags & EV_DISPATCH) 2214 kn->kn_status |= KN_DISABLED; 2215 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 2216 kq->kq_count--; 2217 } else 2218 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2219 2220 kn->kn_status &= ~KN_SCAN; 2221 kn_leave_flux(kn); 2222 kn_list_unlock(knl); 2223 influx = 1; 2224 } 2225 2226 /* we are returning a copy to the user */ 2227 kevp++; 2228 nkev++; 2229 count--; 2230 2231 if (nkev == KQ_NEVENTS) { 2232 influx = 0; 2233 KQ_UNLOCK_FLUX(kq); 2234 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 2235 nkev = 0; 2236 kevp = keva; 2237 KQ_LOCK(kq); 2238 if (error) 2239 break; 2240 } 2241 } 2242 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 2243 done: 2244 KQ_OWNED(kq); 2245 KQ_UNLOCK_FLUX(kq); 2246 knote_free(marker); 2247 done_nl: 2248 KQ_NOTOWNED(kq); 2249 if (nkev != 0) 2250 error = k_ops->k_copyout(k_ops->arg, keva, nkev); 2251 td->td_retval[0] = maxevents - count; 2252 return (error); 2253 } 2254 2255 /*ARGSUSED*/ 2256 static int 2257 kqueue_ioctl(struct file *fp, u_long cmd, void *data, 2258 struct ucred *active_cred, struct thread *td) 2259 { 2260 /* 2261 * Enabling sigio causes two major problems: 2262 * 1) infinite recursion: 2263 * Synopsys: kevent is being used to track signals and have FIOASYNC 2264 * set. On receipt of a signal this will cause a kqueue to recurse 2265 * into itself over and over. Sending the sigio causes the kqueue 2266 * to become ready, which in turn posts sigio again, forever. 2267 * Solution: this can be solved by setting a flag in the kqueue that 2268 * we have a SIGIO in progress. 2269 * 2) locking problems: 2270 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts 2271 * us above the proc and pgrp locks. 2272 * Solution: Post a signal using an async mechanism, being sure to 2273 * record a generation count in the delivery so that we do not deliver 2274 * a signal to the wrong process. 2275 * 2276 * Note, these two mechanisms are somewhat mutually exclusive! 2277 */ 2278 #if 0 2279 struct kqueue *kq; 2280 2281 kq = fp->f_data; 2282 switch (cmd) { 2283 case FIOASYNC: 2284 if (*(int *)data) { 2285 kq->kq_state |= KQ_ASYNC; 2286 } else { 2287 kq->kq_state &= ~KQ_ASYNC; 2288 } 2289 return (0); 2290 2291 case FIOSETOWN: 2292 return (fsetown(*(int *)data, &kq->kq_sigio)); 2293 2294 case FIOGETOWN: 2295 *(int *)data = fgetown(&kq->kq_sigio); 2296 return (0); 2297 } 2298 #endif 2299 2300 return (ENOTTY); 2301 } 2302 2303 /*ARGSUSED*/ 2304 static int 2305 kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 2306 struct thread *td) 2307 { 2308 struct kqueue *kq; 2309 int revents = 0; 2310 int error; 2311 2312 if ((error = kqueue_acquire(fp, &kq))) 2313 return POLLERR; 2314 2315 KQ_LOCK(kq); 2316 if (events & (POLLIN | POLLRDNORM)) { 2317 if (kq->kq_count) { 2318 revents |= events & (POLLIN | POLLRDNORM); 2319 } else { 2320 selrecord(td, &kq->kq_sel); 2321 if (SEL_WAITING(&kq->kq_sel)) 2322 kq->kq_state |= KQ_SEL; 2323 } 2324 } 2325 kqueue_release(kq, 1); 2326 KQ_UNLOCK(kq); 2327 return (revents); 2328 } 2329 2330 /*ARGSUSED*/ 2331 static int 2332 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred) 2333 { 2334 2335 bzero((void *)st, sizeof *st); 2336 /* 2337 * We no longer return kq_count because the unlocked value is useless. 2338 * If you spent all this time getting the count, why not spend your 2339 * syscall better by calling kevent? 2340 * 2341 * XXX - This is needed for libc_r. 2342 */ 2343 st->st_mode = S_IFIFO; 2344 return (0); 2345 } 2346 2347 static void 2348 kqueue_drain(struct kqueue *kq, struct thread *td) 2349 { 2350 struct knote *kn; 2351 int i; 2352 2353 KQ_LOCK(kq); 2354 2355 KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING, 2356 ("kqueue already closing")); 2357 kq->kq_state |= KQ_CLOSING; 2358 if (kq->kq_refcnt > 1) 2359 msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0); 2360 2361 KASSERT(kq->kq_refcnt == 1, ("other refs are out there!")); 2362 2363 KASSERT(knlist_empty(&kq->kq_sel.si_note), 2364 ("kqueue's knlist not empty")); 2365 2366 for (i = 0; i < kq->kq_knlistsize; i++) { 2367 while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) { 2368 if (kn_in_flux(kn)) { 2369 kq->kq_state |= KQ_FLUXWAIT; 2370 msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0); 2371 continue; 2372 } 2373 kn_enter_flux(kn); 2374 KQ_UNLOCK(kq); 2375 knote_drop(kn, td); 2376 KQ_LOCK(kq); 2377 } 2378 } 2379 if (kq->kq_knhashmask != 0) { 2380 for (i = 0; i <= kq->kq_knhashmask; i++) { 2381 while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) { 2382 if (kn_in_flux(kn)) { 2383 kq->kq_state |= KQ_FLUXWAIT; 2384 msleep(kq, &kq->kq_lock, PSOCK, 2385 "kqclo2", 0); 2386 continue; 2387 } 2388 kn_enter_flux(kn); 2389 KQ_UNLOCK(kq); 2390 knote_drop(kn, td); 2391 KQ_LOCK(kq); 2392 } 2393 } 2394 } 2395 2396 if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) { 2397 kq->kq_state |= KQ_TASKDRAIN; 2398 msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0); 2399 } 2400 2401 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2402 selwakeuppri(&kq->kq_sel, PSOCK); 2403 if (!SEL_WAITING(&kq->kq_sel)) 2404 kq->kq_state &= ~KQ_SEL; 2405 } 2406 2407 KQ_UNLOCK(kq); 2408 } 2409 2410 static void 2411 kqueue_destroy(struct kqueue *kq) 2412 { 2413 2414 KASSERT(kq->kq_fdp == NULL, 2415 ("kqueue still attached to a file descriptor")); 2416 seldrain(&kq->kq_sel); 2417 knlist_destroy(&kq->kq_sel.si_note); 2418 mtx_destroy(&kq->kq_lock); 2419 2420 if (kq->kq_knhash != NULL) 2421 free(kq->kq_knhash, M_KQUEUE); 2422 if (kq->kq_knlist != NULL) 2423 free(kq->kq_knlist, M_KQUEUE); 2424 2425 funsetown(&kq->kq_sigio); 2426 } 2427 2428 /*ARGSUSED*/ 2429 static int 2430 kqueue_close(struct file *fp, struct thread *td) 2431 { 2432 struct kqueue *kq = fp->f_data; 2433 struct filedesc *fdp; 2434 int error; 2435 int filedesc_unlock; 2436 2437 if ((error = kqueue_acquire(fp, &kq))) 2438 return error; 2439 kqueue_drain(kq, td); 2440 2441 /* 2442 * We could be called due to the knote_drop() doing fdrop(), 2443 * called from kqueue_register(). In this case the global 2444 * lock is owned, and filedesc sx is locked before, to not 2445 * take the sleepable lock after non-sleepable. 2446 */ 2447 fdp = kq->kq_fdp; 2448 kq->kq_fdp = NULL; 2449 if (!sx_xlocked(FILEDESC_LOCK(fdp))) { 2450 FILEDESC_XLOCK(fdp); 2451 filedesc_unlock = 1; 2452 } else 2453 filedesc_unlock = 0; 2454 TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list); 2455 if (filedesc_unlock) 2456 FILEDESC_XUNLOCK(fdp); 2457 2458 kqueue_destroy(kq); 2459 chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0); 2460 crfree(kq->kq_cred); 2461 free(kq, M_KQUEUE); 2462 fp->f_data = NULL; 2463 2464 return (0); 2465 } 2466 2467 static int 2468 kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 2469 { 2470 struct kqueue *kq = fp->f_data; 2471 2472 kif->kf_type = KF_TYPE_KQUEUE; 2473 kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq; 2474 kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count; 2475 kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state; 2476 return (0); 2477 } 2478 2479 static void 2480 kqueue_wakeup(struct kqueue *kq) 2481 { 2482 KQ_OWNED(kq); 2483 2484 if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) { 2485 kq->kq_state &= ~KQ_SLEEP; 2486 wakeup(kq); 2487 } 2488 if ((kq->kq_state & KQ_SEL) == KQ_SEL) { 2489 selwakeuppri(&kq->kq_sel, PSOCK); 2490 if (!SEL_WAITING(&kq->kq_sel)) 2491 kq->kq_state &= ~KQ_SEL; 2492 } 2493 if (!knlist_empty(&kq->kq_sel.si_note)) 2494 kqueue_schedtask(kq); 2495 if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) { 2496 pgsigio(&kq->kq_sigio, SIGIO, 0); 2497 } 2498 } 2499 2500 /* 2501 * Walk down a list of knotes, activating them if their event has triggered. 2502 * 2503 * There is a possibility to optimize in the case of one kq watching another. 2504 * Instead of scheduling a task to wake it up, you could pass enough state 2505 * down the chain to make up the parent kqueue. Make this code functional 2506 * first. 2507 */ 2508 void 2509 knote(struct knlist *list, long hint, int lockflags) 2510 { 2511 struct kqueue *kq; 2512 struct knote *kn, *tkn; 2513 int error; 2514 2515 if (list == NULL) 2516 return; 2517 2518 KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED); 2519 2520 if ((lockflags & KNF_LISTLOCKED) == 0) 2521 list->kl_lock(list->kl_lockarg); 2522 2523 /* 2524 * If we unlock the list lock (and enter influx), we can 2525 * eliminate the kqueue scheduling, but this will introduce 2526 * four lock/unlock's for each knote to test. Also, marker 2527 * would be needed to keep iteration position, since filters 2528 * or other threads could remove events. 2529 */ 2530 SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) { 2531 kq = kn->kn_kq; 2532 KQ_LOCK(kq); 2533 if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) { 2534 /* 2535 * Do not process the influx notes, except for 2536 * the influx coming from the kq unlock in the 2537 * kqueue_scan(). In the later case, we do 2538 * not interfere with the scan, since the code 2539 * fragment in kqueue_scan() locks the knlist, 2540 * and cannot proceed until we finished. 2541 */ 2542 KQ_UNLOCK(kq); 2543 } else if ((lockflags & KNF_NOKQLOCK) != 0) { 2544 kn_enter_flux(kn); 2545 KQ_UNLOCK(kq); 2546 error = kn->kn_fop->f_event(kn, hint); 2547 KQ_LOCK(kq); 2548 kn_leave_flux(kn); 2549 if (error) 2550 KNOTE_ACTIVATE(kn, 1); 2551 KQ_UNLOCK_FLUX(kq); 2552 } else { 2553 if (kn->kn_fop->f_event(kn, hint)) 2554 KNOTE_ACTIVATE(kn, 1); 2555 KQ_UNLOCK(kq); 2556 } 2557 } 2558 if ((lockflags & KNF_LISTLOCKED) == 0) 2559 list->kl_unlock(list->kl_lockarg); 2560 } 2561 2562 /* 2563 * add a knote to a knlist 2564 */ 2565 void 2566 knlist_add(struct knlist *knl, struct knote *kn, int islocked) 2567 { 2568 2569 KNL_ASSERT_LOCK(knl, islocked); 2570 KQ_NOTOWNED(kn->kn_kq); 2571 KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn)); 2572 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2573 ("knote %p was not detached", kn)); 2574 if (!islocked) 2575 knl->kl_lock(knl->kl_lockarg); 2576 SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext); 2577 if (!islocked) 2578 knl->kl_unlock(knl->kl_lockarg); 2579 KQ_LOCK(kn->kn_kq); 2580 kn->kn_knlist = knl; 2581 kn->kn_status &= ~KN_DETACHED; 2582 KQ_UNLOCK(kn->kn_kq); 2583 } 2584 2585 static void 2586 knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked, 2587 int kqislocked) 2588 { 2589 2590 KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked")); 2591 KNL_ASSERT_LOCK(knl, knlislocked); 2592 mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED); 2593 KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn)); 2594 KASSERT((kn->kn_status & KN_DETACHED) == 0, 2595 ("knote %p was already detached", kn)); 2596 if (!knlislocked) 2597 knl->kl_lock(knl->kl_lockarg); 2598 SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext); 2599 kn->kn_knlist = NULL; 2600 if (!knlislocked) 2601 kn_list_unlock(knl); 2602 if (!kqislocked) 2603 KQ_LOCK(kn->kn_kq); 2604 kn->kn_status |= KN_DETACHED; 2605 if (!kqislocked) 2606 KQ_UNLOCK(kn->kn_kq); 2607 } 2608 2609 /* 2610 * remove knote from the specified knlist 2611 */ 2612 void 2613 knlist_remove(struct knlist *knl, struct knote *kn, int islocked) 2614 { 2615 2616 knlist_remove_kq(knl, kn, islocked, 0); 2617 } 2618 2619 int 2620 knlist_empty(struct knlist *knl) 2621 { 2622 2623 KNL_ASSERT_LOCKED(knl); 2624 return (SLIST_EMPTY(&knl->kl_list)); 2625 } 2626 2627 static struct mtx knlist_lock; 2628 MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects", 2629 MTX_DEF); 2630 static void knlist_mtx_lock(void *arg); 2631 static void knlist_mtx_unlock(void *arg); 2632 2633 static void 2634 knlist_mtx_lock(void *arg) 2635 { 2636 2637 mtx_lock((struct mtx *)arg); 2638 } 2639 2640 static void 2641 knlist_mtx_unlock(void *arg) 2642 { 2643 2644 mtx_unlock((struct mtx *)arg); 2645 } 2646 2647 static void 2648 knlist_mtx_assert_lock(void *arg, int what) 2649 { 2650 2651 if (what == LA_LOCKED) 2652 mtx_assert((struct mtx *)arg, MA_OWNED); 2653 else 2654 mtx_assert((struct mtx *)arg, MA_NOTOWNED); 2655 } 2656 2657 void 2658 knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *), 2659 void (*kl_unlock)(void *), 2660 void (*kl_assert_lock)(void *, int)) 2661 { 2662 2663 if (lock == NULL) 2664 knl->kl_lockarg = &knlist_lock; 2665 else 2666 knl->kl_lockarg = lock; 2667 2668 if (kl_lock == NULL) 2669 knl->kl_lock = knlist_mtx_lock; 2670 else 2671 knl->kl_lock = kl_lock; 2672 if (kl_unlock == NULL) 2673 knl->kl_unlock = knlist_mtx_unlock; 2674 else 2675 knl->kl_unlock = kl_unlock; 2676 if (kl_assert_lock == NULL) 2677 knl->kl_assert_lock = knlist_mtx_assert_lock; 2678 else 2679 knl->kl_assert_lock = kl_assert_lock; 2680 2681 knl->kl_autodestroy = 0; 2682 SLIST_INIT(&knl->kl_list); 2683 } 2684 2685 void 2686 knlist_init_mtx(struct knlist *knl, struct mtx *lock) 2687 { 2688 2689 knlist_init(knl, lock, NULL, NULL, NULL); 2690 } 2691 2692 struct knlist * 2693 knlist_alloc(struct mtx *lock) 2694 { 2695 struct knlist *knl; 2696 2697 knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK); 2698 knlist_init_mtx(knl, lock); 2699 return (knl); 2700 } 2701 2702 void 2703 knlist_destroy(struct knlist *knl) 2704 { 2705 2706 KASSERT(KNLIST_EMPTY(knl), 2707 ("destroying knlist %p with knotes on it", knl)); 2708 } 2709 2710 void 2711 knlist_detach(struct knlist *knl) 2712 { 2713 2714 KNL_ASSERT_LOCKED(knl); 2715 knl->kl_autodestroy = 1; 2716 if (knlist_empty(knl)) { 2717 knlist_destroy(knl); 2718 free(knl, M_KQUEUE); 2719 } 2720 } 2721 2722 /* 2723 * Even if we are locked, we may need to drop the lock to allow any influx 2724 * knotes time to "settle". 2725 */ 2726 void 2727 knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn) 2728 { 2729 struct knote *kn, *kn2; 2730 struct kqueue *kq; 2731 2732 KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl)); 2733 if (islocked) 2734 KNL_ASSERT_LOCKED(knl); 2735 else { 2736 KNL_ASSERT_UNLOCKED(knl); 2737 again: /* need to reacquire lock since we have dropped it */ 2738 knl->kl_lock(knl->kl_lockarg); 2739 } 2740 2741 SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) { 2742 kq = kn->kn_kq; 2743 KQ_LOCK(kq); 2744 if (kn_in_flux(kn)) { 2745 KQ_UNLOCK(kq); 2746 continue; 2747 } 2748 knlist_remove_kq(knl, kn, 1, 1); 2749 if (killkn) { 2750 kn_enter_flux(kn); 2751 KQ_UNLOCK(kq); 2752 knote_drop_detached(kn, td); 2753 } else { 2754 /* Make sure cleared knotes disappear soon */ 2755 kn->kn_flags |= EV_EOF | EV_ONESHOT; 2756 KQ_UNLOCK(kq); 2757 } 2758 kq = NULL; 2759 } 2760 2761 if (!SLIST_EMPTY(&knl->kl_list)) { 2762 /* there are still in flux knotes remaining */ 2763 kn = SLIST_FIRST(&knl->kl_list); 2764 kq = kn->kn_kq; 2765 KQ_LOCK(kq); 2766 KASSERT(kn_in_flux(kn), ("knote removed w/o list lock")); 2767 knl->kl_unlock(knl->kl_lockarg); 2768 kq->kq_state |= KQ_FLUXWAIT; 2769 msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0); 2770 kq = NULL; 2771 goto again; 2772 } 2773 2774 if (islocked) 2775 KNL_ASSERT_LOCKED(knl); 2776 else { 2777 knl->kl_unlock(knl->kl_lockarg); 2778 KNL_ASSERT_UNLOCKED(knl); 2779 } 2780 } 2781 2782 /* 2783 * Remove all knotes referencing a specified fd must be called with FILEDESC 2784 * lock. This prevents a race where a new fd comes along and occupies the 2785 * entry and we attach a knote to the fd. 2786 */ 2787 void 2788 knote_fdclose(struct thread *td, int fd) 2789 { 2790 struct filedesc *fdp = td->td_proc->p_fd; 2791 struct kqueue *kq; 2792 struct knote *kn; 2793 int influx; 2794 2795 FILEDESC_XLOCK_ASSERT(fdp); 2796 2797 /* 2798 * We shouldn't have to worry about new kevents appearing on fd 2799 * since filedesc is locked. 2800 */ 2801 TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) { 2802 KQ_LOCK(kq); 2803 2804 again: 2805 influx = 0; 2806 while (kq->kq_knlistsize > fd && 2807 (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) { 2808 if (kn_in_flux(kn)) { 2809 /* someone else might be waiting on our knote */ 2810 if (influx) 2811 wakeup(kq); 2812 kq->kq_state |= KQ_FLUXWAIT; 2813 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2814 goto again; 2815 } 2816 kn_enter_flux(kn); 2817 KQ_UNLOCK(kq); 2818 influx = 1; 2819 knote_drop(kn, td); 2820 KQ_LOCK(kq); 2821 } 2822 KQ_UNLOCK_FLUX(kq); 2823 } 2824 } 2825 2826 static int 2827 knote_attach(struct knote *kn, struct kqueue *kq) 2828 { 2829 struct klist *list; 2830 2831 KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn)); 2832 KQ_OWNED(kq); 2833 2834 if ((kq->kq_state & KQ_CLOSING) != 0) 2835 return (EBADF); 2836 if (kn->kn_fop->f_isfd) { 2837 if (kn->kn_id >= kq->kq_knlistsize) 2838 return (ENOMEM); 2839 list = &kq->kq_knlist[kn->kn_id]; 2840 } else { 2841 if (kq->kq_knhash == NULL) 2842 return (ENOMEM); 2843 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2844 } 2845 SLIST_INSERT_HEAD(list, kn, kn_link); 2846 return (0); 2847 } 2848 2849 static void 2850 knote_drop(struct knote *kn, struct thread *td) 2851 { 2852 2853 if ((kn->kn_status & KN_DETACHED) == 0) 2854 kn->kn_fop->f_detach(kn); 2855 knote_drop_detached(kn, td); 2856 } 2857 2858 static void 2859 knote_drop_detached(struct knote *kn, struct thread *td) 2860 { 2861 struct kqueue *kq; 2862 struct klist *list; 2863 2864 kq = kn->kn_kq; 2865 2866 KASSERT((kn->kn_status & KN_DETACHED) != 0, 2867 ("knote %p still attached", kn)); 2868 KQ_NOTOWNED(kq); 2869 2870 KQ_LOCK(kq); 2871 for (;;) { 2872 KASSERT(kn->kn_influx >= 1, 2873 ("knote_drop called on %p with influx %d", 2874 kn, kn->kn_influx)); 2875 if (kn->kn_influx == 1) 2876 break; 2877 kq->kq_state |= KQ_FLUXWAIT; 2878 msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0); 2879 } 2880 2881 if (kn->kn_fop->f_isfd) 2882 list = &kq->kq_knlist[kn->kn_id]; 2883 else 2884 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 2885 2886 if (!SLIST_EMPTY(list)) 2887 SLIST_REMOVE(list, kn, knote, kn_link); 2888 if (kn->kn_status & KN_QUEUED) 2889 knote_dequeue(kn); 2890 KQ_UNLOCK_FLUX(kq); 2891 2892 if (kn->kn_fop->f_isfd) { 2893 fdrop(kn->kn_fp, td); 2894 kn->kn_fp = NULL; 2895 } 2896 kqueue_fo_release(kn->kn_kevent.filter); 2897 kn->kn_fop = NULL; 2898 knote_free(kn); 2899 } 2900 2901 static void 2902 knote_enqueue(struct knote *kn) 2903 { 2904 struct kqueue *kq = kn->kn_kq; 2905 2906 KQ_OWNED(kn->kn_kq); 2907 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 2908 2909 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2910 kn->kn_status |= KN_QUEUED; 2911 kq->kq_count++; 2912 kqueue_wakeup(kq); 2913 } 2914 2915 static void 2916 knote_dequeue(struct knote *kn) 2917 { 2918 struct kqueue *kq = kn->kn_kq; 2919 2920 KQ_OWNED(kn->kn_kq); 2921 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 2922 2923 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2924 kn->kn_status &= ~KN_QUEUED; 2925 kq->kq_count--; 2926 } 2927 2928 static void 2929 knote_init(void) 2930 { 2931 2932 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 2933 NULL, NULL, UMA_ALIGN_PTR, 0); 2934 ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue); 2935 prison0.pr_klist = knlist_alloc(&prison0.pr_mtx); 2936 } 2937 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL); 2938 2939 static struct knote * 2940 knote_alloc(int mflag) 2941 { 2942 2943 return (uma_zalloc(knote_zone, mflag | M_ZERO)); 2944 } 2945 2946 static void 2947 knote_free(struct knote *kn) 2948 { 2949 2950 uma_zfree(knote_zone, kn); 2951 } 2952 2953 /* 2954 * Register the kev w/ the kq specified by fd. 2955 */ 2956 int 2957 kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag) 2958 { 2959 struct kqueue *kq; 2960 struct file *fp; 2961 cap_rights_t rights; 2962 int error; 2963 2964 error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE), 2965 &fp); 2966 if (error != 0) 2967 return (error); 2968 if ((error = kqueue_acquire(fp, &kq)) != 0) 2969 goto noacquire; 2970 2971 error = kqueue_register(kq, kev, td, mflag); 2972 kqueue_release(kq, 0); 2973 2974 noacquire: 2975 fdrop(fp, td); 2976 return (error); 2977 } 2978 2979 struct knote_status_export_bit { 2980 int kn_status_bit; 2981 int knt_status_bit; 2982 }; 2983 2984 #define ST(name) \ 2985 { .kn_status_bit = KN_##name, .knt_status_bit = KNOTE_STATUS_##name } 2986 static const struct knote_status_export_bit knote_status_export_bits[] = { 2987 ST(ACTIVE), 2988 ST(QUEUED), 2989 ST(DISABLED), 2990 ST(DETACHED), 2991 ST(KQUEUE), 2992 }; 2993 #undef ST 2994 2995 static int 2996 knote_status_export(int kn_status) 2997 { 2998 const struct knote_status_export_bit *b; 2999 unsigned i; 3000 int res; 3001 3002 res = 0; 3003 for (i = 0; i < nitems(knote_status_export_bits); i++) { 3004 b = &knote_status_export_bits[i]; 3005 if ((kn_status & b->kn_status_bit) != 0) 3006 res |= b->knt_status_bit; 3007 } 3008 return (res); 3009 } 3010 3011 static int 3012 kern_proc_kqueue_report_one(struct sbuf *s, struct proc *p, 3013 int kq_fd, struct kqueue *kq, struct knote *kn, bool compat32 __unused) 3014 { 3015 struct kinfo_knote kin; 3016 #ifdef COMPAT_FREEBSD32 3017 struct kinfo_knote32 kin32; 3018 #endif 3019 int error; 3020 3021 if (kn->kn_status == KN_MARKER) 3022 return (0); 3023 3024 memset(&kin, 0, sizeof(kin)); 3025 kin.knt_kq_fd = kq_fd; 3026 memcpy(&kin.knt_event, &kn->kn_kevent, sizeof(struct kevent)); 3027 kin.knt_status = knote_status_export(kn->kn_status); 3028 kn_enter_flux(kn); 3029 KQ_UNLOCK_FLUX(kq); 3030 if (kn->kn_fop->f_userdump != NULL) 3031 (void)kn->kn_fop->f_userdump(p, kn, &kin); 3032 #ifdef COMPAT_FREEBSD32 3033 if (compat32) { 3034 freebsd32_kinfo_knote_to_32(&kin, &kin32); 3035 error = sbuf_bcat(s, &kin32, sizeof(kin32)); 3036 } else 3037 #endif 3038 error = sbuf_bcat(s, &kin, sizeof(kin)); 3039 KQ_LOCK(kq); 3040 kn_leave_flux(kn); 3041 return (error); 3042 } 3043 3044 static int 3045 kern_proc_kqueue_report(struct sbuf *s, struct proc *p, int kq_fd, 3046 struct kqueue *kq, bool compat32) 3047 { 3048 struct knote *kn; 3049 int error, i; 3050 3051 error = 0; 3052 KQ_LOCK(kq); 3053 for (i = 0; i < kq->kq_knlistsize; i++) { 3054 SLIST_FOREACH(kn, &kq->kq_knlist[i], kn_link) { 3055 error = kern_proc_kqueue_report_one(s, p, kq_fd, 3056 kq, kn, compat32); 3057 if (error != 0) 3058 goto out; 3059 } 3060 } 3061 if (kq->kq_knhashmask == 0) 3062 goto out; 3063 for (i = 0; i <= kq->kq_knhashmask; i++) { 3064 SLIST_FOREACH(kn, &kq->kq_knhash[i], kn_link) { 3065 error = kern_proc_kqueue_report_one(s, p, kq_fd, 3066 kq, kn, compat32); 3067 if (error != 0) 3068 goto out; 3069 } 3070 } 3071 out: 3072 KQ_UNLOCK_FLUX(kq); 3073 return (error); 3074 } 3075 3076 struct kern_proc_kqueues_out1_cb_args { 3077 struct sbuf *s; 3078 bool compat32; 3079 }; 3080 3081 static int 3082 kern_proc_kqueues_out1_cb(struct proc *p, int fd, struct file *fp, void *arg) 3083 { 3084 struct kqueue *kq; 3085 struct kern_proc_kqueues_out1_cb_args *a; 3086 3087 if (fp->f_type != DTYPE_KQUEUE) 3088 return (0); 3089 a = arg; 3090 kq = fp->f_data; 3091 return (kern_proc_kqueue_report(a->s, p, fd, kq, a->compat32)); 3092 } 3093 3094 static int 3095 kern_proc_kqueues_out1(struct thread *td, struct proc *p, struct sbuf *s, 3096 bool compat32) 3097 { 3098 struct kern_proc_kqueues_out1_cb_args a; 3099 3100 a.s = s; 3101 a.compat32 = compat32; 3102 return (fget_remote_foreach(td, p, kern_proc_kqueues_out1_cb, &a)); 3103 } 3104 3105 int 3106 kern_proc_kqueues_out(struct proc *p, struct sbuf *sb, size_t maxlen, 3107 bool compat32) 3108 { 3109 struct sbuf *s, sm; 3110 size_t sb_len; 3111 int error; 3112 3113 if (maxlen == -1 || maxlen == 0) 3114 sb_len = 128; 3115 else 3116 sb_len = maxlen; 3117 s = sbuf_new(&sm, NULL, sb_len, maxlen == -1 ? SBUF_AUTOEXTEND : 3118 SBUF_FIXEDLEN); 3119 error = kern_proc_kqueues_out1(curthread, p, s, compat32); 3120 sbuf_finish(s); 3121 if (error == 0) { 3122 sbuf_bcat(sb, sbuf_data(s), MIN(sbuf_len(s), maxlen == -1 ? 3123 SIZE_T_MAX : maxlen)); 3124 } 3125 sbuf_delete(s); 3126 return (error); 3127 } 3128 3129 static int 3130 sysctl_kern_proc_kqueue_one(struct thread *td, struct sbuf *s, struct proc *p, 3131 int kq_fd, bool compat32) 3132 { 3133 struct file *fp; 3134 struct kqueue *kq; 3135 int error; 3136 3137 error = fget_remote(td, p, kq_fd, &fp); 3138 if (error == 0) { 3139 if (fp->f_type != DTYPE_KQUEUE) { 3140 error = EINVAL; 3141 } else { 3142 kq = fp->f_data; 3143 error = kern_proc_kqueue_report(s, p, kq_fd, kq, 3144 compat32); 3145 } 3146 fdrop(fp, td); 3147 } 3148 return (error); 3149 } 3150 3151 static int 3152 sysctl_kern_proc_kqueue(SYSCTL_HANDLER_ARGS) 3153 { 3154 struct thread *td; 3155 struct proc *p; 3156 struct sbuf *s, sm; 3157 int error, error1, *name; 3158 bool compat32; 3159 3160 name = (int *)arg1; 3161 if ((u_int)arg2 > 2 || (u_int)arg2 == 0) 3162 return (EINVAL); 3163 3164 error = pget((pid_t)name[0], PGET_HOLD | PGET_CANDEBUG, &p); 3165 if (error != 0) 3166 return (error); 3167 3168 td = curthread; 3169 #ifdef FREEBSD_COMPAT32 3170 compat32 = SV_CURPROC_FLAG(SV_ILP32); 3171 #else 3172 compat32 = false; 3173 #endif 3174 3175 s = sbuf_new_for_sysctl(&sm, NULL, 0, req); 3176 if (s == NULL) { 3177 error = ENOMEM; 3178 goto out; 3179 } 3180 sbuf_clear_flags(s, SBUF_INCLUDENUL); 3181 3182 if ((u_int)arg2 == 1) { 3183 error = kern_proc_kqueues_out1(td, p, s, compat32); 3184 } else { 3185 error = sysctl_kern_proc_kqueue_one(td, s, p, 3186 name[1] /* kq_fd */, compat32); 3187 } 3188 3189 error1 = sbuf_finish(s); 3190 if (error == 0) 3191 error = error1; 3192 sbuf_delete(s); 3193 3194 out: 3195 PRELE(p); 3196 return (error); 3197 } 3198 3199 static SYSCTL_NODE(_kern_proc, KERN_PROC_KQUEUE, kq, 3200 CTLFLAG_RD | CTLFLAG_MPSAFE, 3201 sysctl_kern_proc_kqueue, "KQueue events"); 3202