1 /* $NetBSD: kern_event.c,v 1.71 2010/09/10 10:23:46 drochner Exp $ */ 2 3 /*- 4 * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Doran. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 * 57 * FreeBSD: src/sys/kern/kern_event.c,v 1.27 2001/07/05 17:10:44 rwatson Exp 58 */ 59 60 #include <sys/cdefs.h> 61 __KERNEL_RCSID(0, "$NetBSD: kern_event.c,v 1.71 2010/09/10 10:23:46 drochner Exp $"); 62 63 #include <sys/param.h> 64 #include <sys/systm.h> 65 #include <sys/kernel.h> 66 #include <sys/proc.h> 67 #include <sys/file.h> 68 #include <sys/select.h> 69 #include <sys/queue.h> 70 #include <sys/event.h> 71 #include <sys/eventvar.h> 72 #include <sys/poll.h> 73 #include <sys/kmem.h> 74 #include <sys/stat.h> 75 #include <sys/filedesc.h> 76 #include <sys/syscallargs.h> 77 #include <sys/kauth.h> 78 #include <sys/conf.h> 79 #include <sys/atomic.h> 80 81 static int kqueue_scan(file_t *, size_t, struct kevent *, 82 const struct timespec *, register_t *, 83 const struct kevent_ops *, struct kevent *, 84 size_t); 85 static int kqueue_ioctl(file_t *, u_long, void *); 86 static int kqueue_fcntl(file_t *, u_int, void *); 87 static int kqueue_poll(file_t *, int); 88 static int kqueue_kqfilter(file_t *, struct knote *); 89 static int kqueue_stat(file_t *, struct stat *); 90 static int kqueue_close(file_t *); 91 static int kqueue_register(struct kqueue *, struct kevent *); 92 static void kqueue_doclose(struct kqueue *, struct klist *, int); 93 94 static void knote_detach(struct knote *, filedesc_t *fdp, bool); 95 static void knote_enqueue(struct knote *); 96 static void knote_activate(struct knote *); 97 98 static void filt_kqdetach(struct knote *); 99 static int filt_kqueue(struct knote *, long hint); 100 static int filt_procattach(struct knote *); 101 static void filt_procdetach(struct knote *); 102 static int filt_proc(struct knote *, long hint); 103 static int filt_fileattach(struct knote *); 104 static void filt_timerexpire(void *x); 105 static int filt_timerattach(struct knote *); 106 static void filt_timerdetach(struct knote *); 107 static int filt_timer(struct knote *, long hint); 108 109 static const struct fileops kqueueops = { 110 .fo_read = (void *)enxio, 111 .fo_write = (void *)enxio, 112 .fo_ioctl = kqueue_ioctl, 113 .fo_fcntl = kqueue_fcntl, 114 .fo_poll = kqueue_poll, 115 .fo_stat = kqueue_stat, 116 .fo_close = kqueue_close, 117 .fo_kqfilter = kqueue_kqfilter, 118 .fo_restart = fnullop_restart, 119 }; 120 121 static const struct filterops kqread_filtops = 122 { 1, NULL, filt_kqdetach, filt_kqueue }; 123 static const struct filterops proc_filtops = 124 { 0, filt_procattach, filt_procdetach, filt_proc }; 125 static const struct filterops file_filtops = 126 { 1, filt_fileattach, NULL, NULL }; 127 static const struct filterops timer_filtops = 128 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 129 130 static u_int kq_ncallouts = 0; 131 static int kq_calloutmax = (4 * 1024); 132 133 #define KN_HASHSIZE 64 /* XXX should be tunable */ 134 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 135 136 extern const struct filterops sig_filtops; 137 138 /* 139 * Table for for all system-defined filters. 140 * These should be listed in the numeric order of the EVFILT_* defines. 141 * If filtops is NULL, the filter isn't implemented in NetBSD. 142 * End of list is when name is NULL. 143 * 144 * Note that 'refcnt' is meaningless for built-in filters. 145 */ 146 struct kfilter { 147 const char *name; /* name of filter */ 148 uint32_t filter; /* id of filter */ 149 unsigned refcnt; /* reference count */ 150 const struct filterops *filtops;/* operations for filter */ 151 size_t namelen; /* length of name string */ 152 }; 153 154 /* System defined filters */ 155 static struct kfilter sys_kfilters[] = { 156 { "EVFILT_READ", EVFILT_READ, 0, &file_filtops, 0 }, 157 { "EVFILT_WRITE", EVFILT_WRITE, 0, &file_filtops, 0, }, 158 { "EVFILT_AIO", EVFILT_AIO, 0, NULL, 0 }, 159 { "EVFILT_VNODE", EVFILT_VNODE, 0, &file_filtops, 0 }, 160 { "EVFILT_PROC", EVFILT_PROC, 0, &proc_filtops, 0 }, 161 { "EVFILT_SIGNAL", EVFILT_SIGNAL, 0, &sig_filtops, 0 }, 162 { "EVFILT_TIMER", EVFILT_TIMER, 0, &timer_filtops, 0 }, 163 { NULL, 0, 0, NULL, 0 }, 164 }; 165 166 /* User defined kfilters */ 167 static struct kfilter *user_kfilters; /* array */ 168 static int user_kfilterc; /* current offset */ 169 static int user_kfiltermaxc; /* max size so far */ 170 static size_t user_kfiltersz; /* size of allocated memory */ 171 172 /* Locks */ 173 static krwlock_t kqueue_filter_lock; /* lock on filter lists */ 174 static kmutex_t kqueue_misc_lock; /* miscellaneous */ 175 176 static kauth_listener_t kqueue_listener; 177 178 static int 179 kqueue_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie, 180 void *arg0, void *arg1, void *arg2, void *arg3) 181 { 182 struct proc *p; 183 int result; 184 185 result = KAUTH_RESULT_DEFER; 186 p = arg0; 187 188 if (action != KAUTH_PROCESS_KEVENT_FILTER) 189 return result; 190 191 if ((kauth_cred_getuid(p->p_cred) != kauth_cred_getuid(cred) || 192 ISSET(p->p_flag, PK_SUGID))) 193 return result; 194 195 result = KAUTH_RESULT_ALLOW; 196 197 return result; 198 } 199 200 /* 201 * Initialize the kqueue subsystem. 202 */ 203 void 204 kqueue_init(void) 205 { 206 207 rw_init(&kqueue_filter_lock); 208 mutex_init(&kqueue_misc_lock, MUTEX_DEFAULT, IPL_NONE); 209 210 kqueue_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS, 211 kqueue_listener_cb, NULL); 212 } 213 214 /* 215 * Find kfilter entry by name, or NULL if not found. 216 */ 217 static struct kfilter * 218 kfilter_byname_sys(const char *name) 219 { 220 int i; 221 222 KASSERT(rw_lock_held(&kqueue_filter_lock)); 223 224 for (i = 0; sys_kfilters[i].name != NULL; i++) { 225 if (strcmp(name, sys_kfilters[i].name) == 0) 226 return &sys_kfilters[i]; 227 } 228 return NULL; 229 } 230 231 static struct kfilter * 232 kfilter_byname_user(const char *name) 233 { 234 int i; 235 236 KASSERT(rw_lock_held(&kqueue_filter_lock)); 237 238 /* user filter slots have a NULL name if previously deregistered */ 239 for (i = 0; i < user_kfilterc ; i++) { 240 if (user_kfilters[i].name != NULL && 241 strcmp(name, user_kfilters[i].name) == 0) 242 return &user_kfilters[i]; 243 } 244 return NULL; 245 } 246 247 static struct kfilter * 248 kfilter_byname(const char *name) 249 { 250 struct kfilter *kfilter; 251 252 KASSERT(rw_lock_held(&kqueue_filter_lock)); 253 254 if ((kfilter = kfilter_byname_sys(name)) != NULL) 255 return kfilter; 256 257 return kfilter_byname_user(name); 258 } 259 260 /* 261 * Find kfilter entry by filter id, or NULL if not found. 262 * Assumes entries are indexed in filter id order, for speed. 263 */ 264 static struct kfilter * 265 kfilter_byfilter(uint32_t filter) 266 { 267 struct kfilter *kfilter; 268 269 KASSERT(rw_lock_held(&kqueue_filter_lock)); 270 271 if (filter < EVFILT_SYSCOUNT) /* it's a system filter */ 272 kfilter = &sys_kfilters[filter]; 273 else if (user_kfilters != NULL && 274 filter < EVFILT_SYSCOUNT + user_kfilterc) 275 /* it's a user filter */ 276 kfilter = &user_kfilters[filter - EVFILT_SYSCOUNT]; 277 else 278 return (NULL); /* out of range */ 279 KASSERT(kfilter->filter == filter); /* sanity check! */ 280 return (kfilter); 281 } 282 283 /* 284 * Register a new kfilter. Stores the entry in user_kfilters. 285 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 286 * If retfilter != NULL, the new filterid is returned in it. 287 */ 288 int 289 kfilter_register(const char *name, const struct filterops *filtops, 290 int *retfilter) 291 { 292 struct kfilter *kfilter; 293 size_t len; 294 int i; 295 296 if (name == NULL || name[0] == '\0' || filtops == NULL) 297 return (EINVAL); /* invalid args */ 298 299 rw_enter(&kqueue_filter_lock, RW_WRITER); 300 if (kfilter_byname(name) != NULL) { 301 rw_exit(&kqueue_filter_lock); 302 return (EEXIST); /* already exists */ 303 } 304 if (user_kfilterc > 0xffffffff - EVFILT_SYSCOUNT) { 305 rw_exit(&kqueue_filter_lock); 306 return (EINVAL); /* too many */ 307 } 308 309 for (i = 0; i < user_kfilterc; i++) { 310 kfilter = &user_kfilters[i]; 311 if (kfilter->name == NULL) { 312 /* Previously deregistered slot. Reuse. */ 313 goto reuse; 314 } 315 } 316 317 /* check if need to grow user_kfilters */ 318 if (user_kfilterc + 1 > user_kfiltermaxc) { 319 /* Grow in KFILTER_EXTENT chunks. */ 320 user_kfiltermaxc += KFILTER_EXTENT; 321 len = user_kfiltermaxc * sizeof(*kfilter); 322 kfilter = kmem_alloc(len, KM_SLEEP); 323 memset((char *)kfilter + user_kfiltersz, 0, len - user_kfiltersz); 324 if (user_kfilters != NULL) { 325 memcpy(kfilter, user_kfilters, user_kfiltersz); 326 kmem_free(user_kfilters, user_kfiltersz); 327 } 328 user_kfiltersz = len; 329 user_kfilters = kfilter; 330 } 331 /* Adding new slot */ 332 kfilter = &user_kfilters[user_kfilterc++]; 333 reuse: 334 kfilter->namelen = strlen(name) + 1; 335 kfilter->name = kmem_alloc(kfilter->namelen, KM_SLEEP); 336 memcpy(__UNCONST(kfilter->name), name, kfilter->namelen); 337 338 kfilter->filter = (kfilter - user_kfilters) + EVFILT_SYSCOUNT; 339 340 kfilter->filtops = kmem_alloc(sizeof(*filtops), KM_SLEEP); 341 memcpy(__UNCONST(kfilter->filtops), filtops, sizeof(*filtops)); 342 343 if (retfilter != NULL) 344 *retfilter = kfilter->filter; 345 rw_exit(&kqueue_filter_lock); 346 347 return (0); 348 } 349 350 /* 351 * Unregister a kfilter previously registered with kfilter_register. 352 * This retains the filter id, but clears the name and frees filtops (filter 353 * operations), so that the number isn't reused during a boot. 354 * Returns 0 if operation succeeded, or an appropriate errno(2) otherwise. 355 */ 356 int 357 kfilter_unregister(const char *name) 358 { 359 struct kfilter *kfilter; 360 361 if (name == NULL || name[0] == '\0') 362 return (EINVAL); /* invalid name */ 363 364 rw_enter(&kqueue_filter_lock, RW_WRITER); 365 if (kfilter_byname_sys(name) != NULL) { 366 rw_exit(&kqueue_filter_lock); 367 return (EINVAL); /* can't detach system filters */ 368 } 369 370 kfilter = kfilter_byname_user(name); 371 if (kfilter == NULL) { 372 rw_exit(&kqueue_filter_lock); 373 return (ENOENT); 374 } 375 if (kfilter->refcnt != 0) { 376 rw_exit(&kqueue_filter_lock); 377 return (EBUSY); 378 } 379 380 /* Cast away const (but we know it's safe. */ 381 kmem_free(__UNCONST(kfilter->name), kfilter->namelen); 382 kfilter->name = NULL; /* mark as `not implemented' */ 383 384 if (kfilter->filtops != NULL) { 385 /* Cast away const (but we know it's safe. */ 386 kmem_free(__UNCONST(kfilter->filtops), 387 sizeof(*kfilter->filtops)); 388 kfilter->filtops = NULL; /* mark as `not implemented' */ 389 } 390 rw_exit(&kqueue_filter_lock); 391 392 return (0); 393 } 394 395 396 /* 397 * Filter attach method for EVFILT_READ and EVFILT_WRITE on normal file 398 * descriptors. Calls fileops kqfilter method for given file descriptor. 399 */ 400 static int 401 filt_fileattach(struct knote *kn) 402 { 403 file_t *fp; 404 405 fp = kn->kn_obj; 406 407 return (*fp->f_ops->fo_kqfilter)(fp, kn); 408 } 409 410 /* 411 * Filter detach method for EVFILT_READ on kqueue descriptor. 412 */ 413 static void 414 filt_kqdetach(struct knote *kn) 415 { 416 struct kqueue *kq; 417 418 kq = ((file_t *)kn->kn_obj)->f_data; 419 420 mutex_spin_enter(&kq->kq_lock); 421 SLIST_REMOVE(&kq->kq_sel.sel_klist, kn, knote, kn_selnext); 422 mutex_spin_exit(&kq->kq_lock); 423 } 424 425 /* 426 * Filter event method for EVFILT_READ on kqueue descriptor. 427 */ 428 /*ARGSUSED*/ 429 static int 430 filt_kqueue(struct knote *kn, long hint) 431 { 432 struct kqueue *kq; 433 int rv; 434 435 kq = ((file_t *)kn->kn_obj)->f_data; 436 437 if (hint != NOTE_SUBMIT) 438 mutex_spin_enter(&kq->kq_lock); 439 kn->kn_data = kq->kq_count; 440 rv = (kn->kn_data > 0); 441 if (hint != NOTE_SUBMIT) 442 mutex_spin_exit(&kq->kq_lock); 443 444 return rv; 445 } 446 447 /* 448 * Filter attach method for EVFILT_PROC. 449 */ 450 static int 451 filt_procattach(struct knote *kn) 452 { 453 struct proc *p, *curp; 454 struct lwp *curl; 455 456 curl = curlwp; 457 curp = curl->l_proc; 458 459 mutex_enter(proc_lock); 460 p = proc_find(kn->kn_id); 461 if (p == NULL) { 462 mutex_exit(proc_lock); 463 return ESRCH; 464 } 465 466 /* 467 * Fail if it's not owned by you, or the last exec gave us 468 * setuid/setgid privs (unless you're root). 469 */ 470 mutex_enter(p->p_lock); 471 mutex_exit(proc_lock); 472 if (kauth_authorize_process(curl->l_cred, KAUTH_PROCESS_KEVENT_FILTER, 473 p, NULL, NULL, NULL) != 0) { 474 mutex_exit(p->p_lock); 475 return EACCES; 476 } 477 478 kn->kn_obj = p; 479 kn->kn_flags |= EV_CLEAR; /* automatically set */ 480 481 /* 482 * internal flag indicating registration done by kernel 483 */ 484 if (kn->kn_flags & EV_FLAG1) { 485 kn->kn_data = kn->kn_sdata; /* ppid */ 486 kn->kn_fflags = NOTE_CHILD; 487 kn->kn_flags &= ~EV_FLAG1; 488 } 489 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 490 mutex_exit(p->p_lock); 491 492 return 0; 493 } 494 495 /* 496 * Filter detach method for EVFILT_PROC. 497 * 498 * The knote may be attached to a different process, which may exit, 499 * leaving nothing for the knote to be attached to. So when the process 500 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 501 * it will be deleted when read out. However, as part of the knote deletion, 502 * this routine is called, so a check is needed to avoid actually performing 503 * a detach, because the original process might not exist any more. 504 */ 505 static void 506 filt_procdetach(struct knote *kn) 507 { 508 struct proc *p; 509 510 if (kn->kn_status & KN_DETACHED) 511 return; 512 513 p = kn->kn_obj; 514 515 mutex_enter(p->p_lock); 516 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 517 mutex_exit(p->p_lock); 518 } 519 520 /* 521 * Filter event method for EVFILT_PROC. 522 */ 523 static int 524 filt_proc(struct knote *kn, long hint) 525 { 526 u_int event, fflag; 527 struct kevent kev; 528 struct kqueue *kq; 529 int error; 530 531 event = (u_int)hint & NOTE_PCTRLMASK; 532 kq = kn->kn_kq; 533 fflag = 0; 534 535 /* If the user is interested in this event, record it. */ 536 if (kn->kn_sfflags & event) 537 fflag |= event; 538 539 if (event == NOTE_EXIT) { 540 /* 541 * Process is gone, so flag the event as finished. 542 * 543 * Detach the knote from watched process and mark 544 * it as such. We can't leave this to kqueue_scan(), 545 * since the process might not exist by then. And we 546 * have to do this now, since psignal KNOTE() is called 547 * also for zombies and we might end up reading freed 548 * memory if the kevent would already be picked up 549 * and knote g/c'ed. 550 */ 551 filt_procdetach(kn); 552 553 mutex_spin_enter(&kq->kq_lock); 554 kn->kn_status |= KN_DETACHED; 555 /* Mark as ONESHOT, so that the knote it g/c'ed when read */ 556 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 557 kn->kn_fflags |= fflag; 558 mutex_spin_exit(&kq->kq_lock); 559 560 return 1; 561 } 562 563 mutex_spin_enter(&kq->kq_lock); 564 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 565 /* 566 * Process forked, and user wants to track the new process, 567 * so attach a new knote to it, and immediately report an 568 * event with the parent's pid. Register knote with new 569 * process. 570 */ 571 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 572 kev.filter = kn->kn_filter; 573 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 574 kev.fflags = kn->kn_sfflags; 575 kev.data = kn->kn_id; /* parent */ 576 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 577 mutex_spin_exit(&kq->kq_lock); 578 error = kqueue_register(kq, &kev); 579 mutex_spin_enter(&kq->kq_lock); 580 if (error != 0) 581 kn->kn_fflags |= NOTE_TRACKERR; 582 } 583 kn->kn_fflags |= fflag; 584 fflag = kn->kn_fflags; 585 mutex_spin_exit(&kq->kq_lock); 586 587 return fflag != 0; 588 } 589 590 static void 591 filt_timerexpire(void *knx) 592 { 593 struct knote *kn = knx; 594 int tticks; 595 596 mutex_enter(&kqueue_misc_lock); 597 kn->kn_data++; 598 knote_activate(kn); 599 if ((kn->kn_flags & EV_ONESHOT) == 0) { 600 tticks = mstohz(kn->kn_sdata); 601 callout_schedule((callout_t *)kn->kn_hook, tticks); 602 } 603 mutex_exit(&kqueue_misc_lock); 604 } 605 606 /* 607 * data contains amount of time to sleep, in milliseconds 608 */ 609 static int 610 filt_timerattach(struct knote *kn) 611 { 612 callout_t *calloutp; 613 struct kqueue *kq; 614 int tticks; 615 616 tticks = mstohz(kn->kn_sdata); 617 618 /* if the supplied value is under our resolution, use 1 tick */ 619 if (tticks == 0) { 620 if (kn->kn_sdata == 0) 621 return EINVAL; 622 tticks = 1; 623 } 624 625 if (atomic_inc_uint_nv(&kq_ncallouts) >= kq_calloutmax || 626 (calloutp = kmem_alloc(sizeof(*calloutp), KM_NOSLEEP)) == NULL) { 627 atomic_dec_uint(&kq_ncallouts); 628 return ENOMEM; 629 } 630 callout_init(calloutp, CALLOUT_MPSAFE); 631 632 kq = kn->kn_kq; 633 mutex_spin_enter(&kq->kq_lock); 634 kn->kn_flags |= EV_CLEAR; /* automatically set */ 635 kn->kn_hook = calloutp; 636 mutex_spin_exit(&kq->kq_lock); 637 638 callout_reset(calloutp, tticks, filt_timerexpire, kn); 639 640 return (0); 641 } 642 643 static void 644 filt_timerdetach(struct knote *kn) 645 { 646 callout_t *calloutp; 647 648 calloutp = (callout_t *)kn->kn_hook; 649 callout_halt(calloutp, NULL); 650 callout_destroy(calloutp); 651 kmem_free(calloutp, sizeof(*calloutp)); 652 atomic_dec_uint(&kq_ncallouts); 653 } 654 655 static int 656 filt_timer(struct knote *kn, long hint) 657 { 658 int rv; 659 660 mutex_enter(&kqueue_misc_lock); 661 rv = (kn->kn_data != 0); 662 mutex_exit(&kqueue_misc_lock); 663 664 return rv; 665 } 666 667 /* 668 * filt_seltrue: 669 * 670 * This filter "event" routine simulates seltrue(). 671 */ 672 int 673 filt_seltrue(struct knote *kn, long hint) 674 { 675 676 /* 677 * We don't know how much data can be read/written, 678 * but we know that it *can* be. This is about as 679 * good as select/poll does as well. 680 */ 681 kn->kn_data = 0; 682 return (1); 683 } 684 685 /* 686 * This provides full kqfilter entry for device switch tables, which 687 * has same effect as filter using filt_seltrue() as filter method. 688 */ 689 static void 690 filt_seltruedetach(struct knote *kn) 691 { 692 /* Nothing to do */ 693 } 694 695 const struct filterops seltrue_filtops = 696 { 1, NULL, filt_seltruedetach, filt_seltrue }; 697 698 int 699 seltrue_kqfilter(dev_t dev, struct knote *kn) 700 { 701 switch (kn->kn_filter) { 702 case EVFILT_READ: 703 case EVFILT_WRITE: 704 kn->kn_fop = &seltrue_filtops; 705 break; 706 default: 707 return (EINVAL); 708 } 709 710 /* Nothing more to do */ 711 return (0); 712 } 713 714 /* 715 * kqueue(2) system call. 716 */ 717 int 718 sys_kqueue(struct lwp *l, const void *v, register_t *retval) 719 { 720 struct kqueue *kq; 721 file_t *fp; 722 int fd, error; 723 724 if ((error = fd_allocfile(&fp, &fd)) != 0) 725 return error; 726 fp->f_flag = FREAD | FWRITE; 727 fp->f_type = DTYPE_KQUEUE; 728 fp->f_ops = &kqueueops; 729 kq = kmem_zalloc(sizeof(*kq), KM_SLEEP); 730 mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED); 731 cv_init(&kq->kq_cv, "kqueue"); 732 selinit(&kq->kq_sel); 733 TAILQ_INIT(&kq->kq_head); 734 fp->f_data = kq; 735 *retval = fd; 736 kq->kq_fdp = curlwp->l_fd; 737 fd_affix(curproc, fp, fd); 738 return error; 739 } 740 741 /* 742 * kevent(2) system call. 743 */ 744 int 745 kevent_fetch_changes(void *private, const struct kevent *changelist, 746 struct kevent *changes, size_t index, int n) 747 { 748 749 return copyin(changelist + index, changes, n * sizeof(*changes)); 750 } 751 752 int 753 kevent_put_events(void *private, struct kevent *events, 754 struct kevent *eventlist, size_t index, int n) 755 { 756 757 return copyout(events, eventlist + index, n * sizeof(*events)); 758 } 759 760 static const struct kevent_ops kevent_native_ops = { 761 .keo_private = NULL, 762 .keo_fetch_timeout = copyin, 763 .keo_fetch_changes = kevent_fetch_changes, 764 .keo_put_events = kevent_put_events, 765 }; 766 767 int 768 sys___kevent50(struct lwp *l, const struct sys___kevent50_args *uap, 769 register_t *retval) 770 { 771 /* { 772 syscallarg(int) fd; 773 syscallarg(const struct kevent *) changelist; 774 syscallarg(size_t) nchanges; 775 syscallarg(struct kevent *) eventlist; 776 syscallarg(size_t) nevents; 777 syscallarg(const struct timespec *) timeout; 778 } */ 779 780 return kevent1(retval, SCARG(uap, fd), SCARG(uap, changelist), 781 SCARG(uap, nchanges), SCARG(uap, eventlist), SCARG(uap, nevents), 782 SCARG(uap, timeout), &kevent_native_ops); 783 } 784 785 int 786 kevent1(register_t *retval, int fd, 787 const struct kevent *changelist, size_t nchanges, 788 struct kevent *eventlist, size_t nevents, 789 const struct timespec *timeout, 790 const struct kevent_ops *keops) 791 { 792 struct kevent *kevp; 793 struct kqueue *kq; 794 struct timespec ts; 795 size_t i, n, ichange; 796 int nerrors, error; 797 struct kevent kevbuf[8]; /* approx 300 bytes on 64-bit */ 798 file_t *fp; 799 800 /* check that we're dealing with a kq */ 801 fp = fd_getfile(fd); 802 if (fp == NULL) 803 return (EBADF); 804 805 if (fp->f_type != DTYPE_KQUEUE) { 806 fd_putfile(fd); 807 return (EBADF); 808 } 809 810 if (timeout != NULL) { 811 error = (*keops->keo_fetch_timeout)(timeout, &ts, sizeof(ts)); 812 if (error) 813 goto done; 814 timeout = &ts; 815 } 816 817 kq = (struct kqueue *)fp->f_data; 818 nerrors = 0; 819 ichange = 0; 820 821 /* traverse list of events to register */ 822 while (nchanges > 0) { 823 n = MIN(nchanges, __arraycount(kevbuf)); 824 error = (*keops->keo_fetch_changes)(keops->keo_private, 825 changelist, kevbuf, ichange, n); 826 if (error) 827 goto done; 828 for (i = 0; i < n; i++) { 829 kevp = &kevbuf[i]; 830 kevp->flags &= ~EV_SYSFLAGS; 831 /* register each knote */ 832 error = kqueue_register(kq, kevp); 833 if (error) { 834 if (nevents != 0) { 835 kevp->flags = EV_ERROR; 836 kevp->data = error; 837 error = (*keops->keo_put_events) 838 (keops->keo_private, kevp, 839 eventlist, nerrors, 1); 840 if (error) 841 goto done; 842 nevents--; 843 nerrors++; 844 } else { 845 goto done; 846 } 847 } 848 } 849 nchanges -= n; /* update the results */ 850 ichange += n; 851 } 852 if (nerrors) { 853 *retval = nerrors; 854 error = 0; 855 goto done; 856 } 857 858 /* actually scan through the events */ 859 error = kqueue_scan(fp, nevents, eventlist, timeout, retval, keops, 860 kevbuf, __arraycount(kevbuf)); 861 done: 862 fd_putfile(fd); 863 return (error); 864 } 865 866 /* 867 * Register a given kevent kev onto the kqueue 868 */ 869 static int 870 kqueue_register(struct kqueue *kq, struct kevent *kev) 871 { 872 struct kfilter *kfilter; 873 filedesc_t *fdp; 874 file_t *fp; 875 fdfile_t *ff; 876 struct knote *kn, *newkn; 877 struct klist *list; 878 int error, fd, rv; 879 880 fdp = kq->kq_fdp; 881 fp = NULL; 882 kn = NULL; 883 error = 0; 884 fd = 0; 885 886 newkn = kmem_zalloc(sizeof(*newkn), KM_SLEEP); 887 888 rw_enter(&kqueue_filter_lock, RW_READER); 889 kfilter = kfilter_byfilter(kev->filter); 890 if (kfilter == NULL || kfilter->filtops == NULL) { 891 /* filter not found nor implemented */ 892 rw_exit(&kqueue_filter_lock); 893 kmem_free(newkn, sizeof(*newkn)); 894 return (EINVAL); 895 } 896 897 mutex_enter(&fdp->fd_lock); 898 899 /* search if knote already exists */ 900 if (kfilter->filtops->f_isfd) { 901 /* monitoring a file descriptor */ 902 fd = kev->ident; 903 if ((fp = fd_getfile(fd)) == NULL) { 904 mutex_exit(&fdp->fd_lock); 905 rw_exit(&kqueue_filter_lock); 906 kmem_free(newkn, sizeof(*newkn)); 907 return EBADF; 908 } 909 ff = fdp->fd_dt->dt_ff[fd]; 910 if (fd <= fdp->fd_lastkqfile) { 911 SLIST_FOREACH(kn, &ff->ff_knlist, kn_link) { 912 if (kq == kn->kn_kq && 913 kev->filter == kn->kn_filter) 914 break; 915 } 916 } 917 } else { 918 /* 919 * not monitoring a file descriptor, so 920 * lookup knotes in internal hash table 921 */ 922 if (fdp->fd_knhashmask != 0) { 923 list = &fdp->fd_knhash[ 924 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 925 SLIST_FOREACH(kn, list, kn_link) { 926 if (kev->ident == kn->kn_id && 927 kq == kn->kn_kq && 928 kev->filter == kn->kn_filter) 929 break; 930 } 931 } 932 } 933 934 /* 935 * kn now contains the matching knote, or NULL if no match 936 */ 937 if (kev->flags & EV_ADD) { 938 if (kn == NULL) { 939 /* create new knote */ 940 kn = newkn; 941 newkn = NULL; 942 kn->kn_obj = fp; 943 kn->kn_kq = kq; 944 kn->kn_fop = kfilter->filtops; 945 kn->kn_kfilter = kfilter; 946 kn->kn_sfflags = kev->fflags; 947 kn->kn_sdata = kev->data; 948 kev->fflags = 0; 949 kev->data = 0; 950 kn->kn_kevent = *kev; 951 952 /* 953 * apply reference count to knote structure, and 954 * do not release it at the end of this routine. 955 */ 956 fp = NULL; 957 958 if (!kn->kn_fop->f_isfd) { 959 /* 960 * If knote is not on an fd, store on 961 * internal hash table. 962 */ 963 if (fdp->fd_knhashmask == 0) { 964 /* XXXAD can block with fd_lock held */ 965 fdp->fd_knhash = hashinit(KN_HASHSIZE, 966 HASH_LIST, true, 967 &fdp->fd_knhashmask); 968 } 969 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, 970 fdp->fd_knhashmask)]; 971 } else { 972 /* Otherwise, knote is on an fd. */ 973 list = (struct klist *) 974 &fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist; 975 if ((int)kn->kn_id > fdp->fd_lastkqfile) 976 fdp->fd_lastkqfile = kn->kn_id; 977 } 978 SLIST_INSERT_HEAD(list, kn, kn_link); 979 980 KERNEL_LOCK(1, NULL); /* XXXSMP */ 981 error = (*kfilter->filtops->f_attach)(kn); 982 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 983 if (error != 0) { 984 /* knote_detach() drops fdp->fd_lock */ 985 knote_detach(kn, fdp, false); 986 goto done; 987 } 988 atomic_inc_uint(&kfilter->refcnt); 989 } else { 990 /* 991 * The user may change some filter values after the 992 * initial EV_ADD, but doing so will not reset any 993 * filter which have already been triggered. 994 */ 995 kn->kn_sfflags = kev->fflags; 996 kn->kn_sdata = kev->data; 997 kn->kn_kevent.udata = kev->udata; 998 } 999 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1000 rv = (*kn->kn_fop->f_event)(kn, 0); 1001 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1002 if (rv) 1003 knote_activate(kn); 1004 } else { 1005 if (kn == NULL) { 1006 error = ENOENT; 1007 mutex_exit(&fdp->fd_lock); 1008 goto done; 1009 } 1010 if (kev->flags & EV_DELETE) { 1011 /* knote_detach() drops fdp->fd_lock */ 1012 knote_detach(kn, fdp, true); 1013 goto done; 1014 } 1015 } 1016 1017 /* disable knote */ 1018 if ((kev->flags & EV_DISABLE)) { 1019 mutex_spin_enter(&kq->kq_lock); 1020 if ((kn->kn_status & KN_DISABLED) == 0) 1021 kn->kn_status |= KN_DISABLED; 1022 mutex_spin_exit(&kq->kq_lock); 1023 } 1024 1025 /* enable knote */ 1026 if ((kev->flags & EV_ENABLE)) { 1027 knote_enqueue(kn); 1028 } 1029 mutex_exit(&fdp->fd_lock); 1030 done: 1031 rw_exit(&kqueue_filter_lock); 1032 if (newkn != NULL) 1033 kmem_free(newkn, sizeof(*newkn)); 1034 if (fp != NULL) 1035 fd_putfile(fd); 1036 return (error); 1037 } 1038 1039 #if defined(DEBUG) 1040 static void 1041 kq_check(struct kqueue *kq) 1042 { 1043 const struct knote *kn; 1044 int count; 1045 int nmarker; 1046 1047 KASSERT(mutex_owned(&kq->kq_lock)); 1048 KASSERT(kq->kq_count >= 0); 1049 1050 count = 0; 1051 nmarker = 0; 1052 TAILQ_FOREACH(kn, &kq->kq_head, kn_tqe) { 1053 if ((kn->kn_status & (KN_MARKER | KN_QUEUED)) == 0) { 1054 panic("%s: kq=%p kn=%p inconsist 1", __func__, kq, kn); 1055 } 1056 if ((kn->kn_status & KN_MARKER) == 0) { 1057 if (kn->kn_kq != kq) { 1058 panic("%s: kq=%p kn=%p inconsist 2", 1059 __func__, kq, kn); 1060 } 1061 if ((kn->kn_status & KN_ACTIVE) == 0) { 1062 panic("%s: kq=%p kn=%p: not active", 1063 __func__, kq, kn); 1064 } 1065 count++; 1066 if (count > kq->kq_count) { 1067 goto bad; 1068 } 1069 } else { 1070 nmarker++; 1071 #if 0 1072 if (nmarker > 10000) { 1073 panic("%s: kq=%p too many markers: %d != %d, " 1074 "nmarker=%d", 1075 __func__, kq, kq->kq_count, count, nmarker); 1076 } 1077 #endif 1078 } 1079 } 1080 if (kq->kq_count != count) { 1081 bad: 1082 panic("%s: kq=%p inconsist 3: %d != %d, nmarker=%d", 1083 __func__, kq, kq->kq_count, count, nmarker); 1084 } 1085 } 1086 #else /* defined(DEBUG) */ 1087 #define kq_check(a) /* nothing */ 1088 #endif /* defined(DEBUG) */ 1089 1090 /* 1091 * Scan through the list of events on fp (for a maximum of maxevents), 1092 * returning the results in to ulistp. Timeout is determined by tsp; if 1093 * NULL, wait indefinitely, if 0 valued, perform a poll, otherwise wait 1094 * as appropriate. 1095 */ 1096 static int 1097 kqueue_scan(file_t *fp, size_t maxevents, struct kevent *ulistp, 1098 const struct timespec *tsp, register_t *retval, 1099 const struct kevent_ops *keops, struct kevent *kevbuf, 1100 size_t kevcnt) 1101 { 1102 struct kqueue *kq; 1103 struct kevent *kevp; 1104 struct timespec ats, sleepts; 1105 struct knote *kn, *marker; 1106 size_t count, nkev, nevents; 1107 int timeout, error, rv; 1108 filedesc_t *fdp; 1109 1110 fdp = curlwp->l_fd; 1111 kq = fp->f_data; 1112 count = maxevents; 1113 nkev = nevents = error = 0; 1114 if (count == 0) { 1115 *retval = 0; 1116 return 0; 1117 } 1118 1119 if (tsp) { /* timeout supplied */ 1120 ats = *tsp; 1121 if (inittimeleft(&ats, &sleepts) == -1) { 1122 *retval = maxevents; 1123 return EINVAL; 1124 } 1125 timeout = tstohz(&ats); 1126 if (timeout <= 0) 1127 timeout = -1; /* do poll */ 1128 } else { 1129 /* no timeout, wait forever */ 1130 timeout = 0; 1131 } 1132 1133 marker = kmem_zalloc(sizeof(*marker), KM_SLEEP); 1134 marker->kn_status = KN_MARKER; 1135 mutex_spin_enter(&kq->kq_lock); 1136 retry: 1137 kevp = kevbuf; 1138 if (kq->kq_count == 0) { 1139 if (timeout >= 0) { 1140 error = cv_timedwait_sig(&kq->kq_cv, 1141 &kq->kq_lock, timeout); 1142 if (error == 0) { 1143 if (tsp == NULL || (timeout = 1144 gettimeleft(&ats, &sleepts)) > 0) 1145 goto retry; 1146 } else { 1147 /* don't restart after signals... */ 1148 if (error == ERESTART) 1149 error = EINTR; 1150 if (error == EWOULDBLOCK) 1151 error = 0; 1152 } 1153 } 1154 } else { 1155 /* mark end of knote list */ 1156 TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe); 1157 1158 while (count != 0) { 1159 kn = TAILQ_FIRST(&kq->kq_head); /* get next knote */ 1160 while ((kn->kn_status & KN_MARKER) != 0) { 1161 if (kn == marker) { 1162 /* it's our marker, stop */ 1163 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1164 if (count < maxevents || (tsp != NULL && 1165 (timeout = gettimeleft(&ats, 1166 &sleepts)) <= 0)) 1167 goto done; 1168 goto retry; 1169 } 1170 /* someone else's marker. */ 1171 kn = TAILQ_NEXT(kn, kn_tqe); 1172 } 1173 kq_check(kq); 1174 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1175 kq->kq_count--; 1176 kn->kn_status &= ~KN_QUEUED; 1177 kq_check(kq); 1178 if (kn->kn_status & KN_DISABLED) { 1179 /* don't want disabled events */ 1180 continue; 1181 } 1182 if ((kn->kn_flags & EV_ONESHOT) == 0) { 1183 mutex_spin_exit(&kq->kq_lock); 1184 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1185 rv = (*kn->kn_fop->f_event)(kn, 0); 1186 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1187 mutex_spin_enter(&kq->kq_lock); 1188 /* Re-poll if note was re-enqueued. */ 1189 if ((kn->kn_status & KN_QUEUED) != 0) 1190 continue; 1191 if (rv == 0) { 1192 /* 1193 * non-ONESHOT event that hasn't 1194 * triggered again, so de-queue. 1195 */ 1196 kn->kn_status &= ~KN_ACTIVE; 1197 continue; 1198 } 1199 } 1200 /* XXXAD should be got from f_event if !oneshot. */ 1201 *kevp++ = kn->kn_kevent; 1202 nkev++; 1203 if (kn->kn_flags & EV_ONESHOT) { 1204 /* delete ONESHOT events after retrieval */ 1205 mutex_spin_exit(&kq->kq_lock); 1206 mutex_enter(&fdp->fd_lock); 1207 knote_detach(kn, fdp, true); 1208 mutex_spin_enter(&kq->kq_lock); 1209 } else if (kn->kn_flags & EV_CLEAR) { 1210 /* clear state after retrieval */ 1211 kn->kn_data = 0; 1212 kn->kn_fflags = 0; 1213 kn->kn_status &= ~KN_ACTIVE; 1214 } else { 1215 /* add event back on list */ 1216 kq_check(kq); 1217 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1218 kq->kq_count++; 1219 kn->kn_status |= KN_QUEUED; 1220 kq_check(kq); 1221 } 1222 if (nkev == kevcnt) { 1223 /* do copyouts in kevcnt chunks */ 1224 mutex_spin_exit(&kq->kq_lock); 1225 error = (*keops->keo_put_events) 1226 (keops->keo_private, 1227 kevbuf, ulistp, nevents, nkev); 1228 mutex_spin_enter(&kq->kq_lock); 1229 nevents += nkev; 1230 nkev = 0; 1231 kevp = kevbuf; 1232 } 1233 count--; 1234 if (error != 0 || count == 0) { 1235 /* remove marker */ 1236 TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe); 1237 break; 1238 } 1239 } 1240 } 1241 done: 1242 mutex_spin_exit(&kq->kq_lock); 1243 if (marker != NULL) 1244 kmem_free(marker, sizeof(*marker)); 1245 if (nkev != 0) { 1246 /* copyout remaining events */ 1247 error = (*keops->keo_put_events)(keops->keo_private, 1248 kevbuf, ulistp, nevents, nkev); 1249 } 1250 *retval = maxevents - count; 1251 1252 return error; 1253 } 1254 1255 /* 1256 * fileops ioctl method for a kqueue descriptor. 1257 * 1258 * Two ioctls are currently supported. They both use struct kfilter_mapping: 1259 * KFILTER_BYNAME find name for filter, and return result in 1260 * name, which is of size len. 1261 * KFILTER_BYFILTER find filter for name. len is ignored. 1262 */ 1263 /*ARGSUSED*/ 1264 static int 1265 kqueue_ioctl(file_t *fp, u_long com, void *data) 1266 { 1267 struct kfilter_mapping *km; 1268 const struct kfilter *kfilter; 1269 char *name; 1270 int error; 1271 1272 km = data; 1273 error = 0; 1274 name = kmem_alloc(KFILTER_MAXNAME, KM_SLEEP); 1275 1276 switch (com) { 1277 case KFILTER_BYFILTER: /* convert filter -> name */ 1278 rw_enter(&kqueue_filter_lock, RW_READER); 1279 kfilter = kfilter_byfilter(km->filter); 1280 if (kfilter != NULL) { 1281 strlcpy(name, kfilter->name, KFILTER_MAXNAME); 1282 rw_exit(&kqueue_filter_lock); 1283 error = copyoutstr(name, km->name, km->len, NULL); 1284 } else { 1285 rw_exit(&kqueue_filter_lock); 1286 error = ENOENT; 1287 } 1288 break; 1289 1290 case KFILTER_BYNAME: /* convert name -> filter */ 1291 error = copyinstr(km->name, name, KFILTER_MAXNAME, NULL); 1292 if (error) { 1293 break; 1294 } 1295 rw_enter(&kqueue_filter_lock, RW_READER); 1296 kfilter = kfilter_byname(name); 1297 if (kfilter != NULL) 1298 km->filter = kfilter->filter; 1299 else 1300 error = ENOENT; 1301 rw_exit(&kqueue_filter_lock); 1302 break; 1303 1304 default: 1305 error = ENOTTY; 1306 break; 1307 1308 } 1309 kmem_free(name, KFILTER_MAXNAME); 1310 return (error); 1311 } 1312 1313 /* 1314 * fileops fcntl method for a kqueue descriptor. 1315 */ 1316 static int 1317 kqueue_fcntl(file_t *fp, u_int com, void *data) 1318 { 1319 1320 return (ENOTTY); 1321 } 1322 1323 /* 1324 * fileops poll method for a kqueue descriptor. 1325 * Determine if kqueue has events pending. 1326 */ 1327 static int 1328 kqueue_poll(file_t *fp, int events) 1329 { 1330 struct kqueue *kq; 1331 int revents; 1332 1333 kq = fp->f_data; 1334 1335 revents = 0; 1336 if (events & (POLLIN | POLLRDNORM)) { 1337 mutex_spin_enter(&kq->kq_lock); 1338 if (kq->kq_count != 0) { 1339 revents |= events & (POLLIN | POLLRDNORM); 1340 } else { 1341 selrecord(curlwp, &kq->kq_sel); 1342 } 1343 kq_check(kq); 1344 mutex_spin_exit(&kq->kq_lock); 1345 } 1346 1347 return revents; 1348 } 1349 1350 /* 1351 * fileops stat method for a kqueue descriptor. 1352 * Returns dummy info, with st_size being number of events pending. 1353 */ 1354 static int 1355 kqueue_stat(file_t *fp, struct stat *st) 1356 { 1357 struct kqueue *kq; 1358 1359 kq = fp->f_data; 1360 1361 memset(st, 0, sizeof(*st)); 1362 st->st_size = kq->kq_count; 1363 st->st_blksize = sizeof(struct kevent); 1364 st->st_mode = S_IFIFO; 1365 1366 return 0; 1367 } 1368 1369 static void 1370 kqueue_doclose(struct kqueue *kq, struct klist *list, int fd) 1371 { 1372 struct knote *kn; 1373 filedesc_t *fdp; 1374 1375 fdp = kq->kq_fdp; 1376 1377 KASSERT(mutex_owned(&fdp->fd_lock)); 1378 1379 for (kn = SLIST_FIRST(list); kn != NULL;) { 1380 if (kq != kn->kn_kq) { 1381 kn = SLIST_NEXT(kn, kn_link); 1382 continue; 1383 } 1384 knote_detach(kn, fdp, true); 1385 mutex_enter(&fdp->fd_lock); 1386 kn = SLIST_FIRST(list); 1387 } 1388 } 1389 1390 1391 /* 1392 * fileops close method for a kqueue descriptor. 1393 */ 1394 static int 1395 kqueue_close(file_t *fp) 1396 { 1397 struct kqueue *kq; 1398 filedesc_t *fdp; 1399 fdfile_t *ff; 1400 int i; 1401 1402 kq = fp->f_data; 1403 fdp = curlwp->l_fd; 1404 1405 mutex_enter(&fdp->fd_lock); 1406 for (i = 0; i <= fdp->fd_lastkqfile; i++) { 1407 if ((ff = fdp->fd_dt->dt_ff[i]) == NULL) 1408 continue; 1409 kqueue_doclose(kq, (struct klist *)&ff->ff_knlist, i); 1410 } 1411 if (fdp->fd_knhashmask != 0) { 1412 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 1413 kqueue_doclose(kq, &fdp->fd_knhash[i], -1); 1414 } 1415 } 1416 mutex_exit(&fdp->fd_lock); 1417 1418 KASSERT(kq->kq_count == 0); 1419 mutex_destroy(&kq->kq_lock); 1420 cv_destroy(&kq->kq_cv); 1421 seldestroy(&kq->kq_sel); 1422 kmem_free(kq, sizeof(*kq)); 1423 fp->f_data = NULL; 1424 1425 return (0); 1426 } 1427 1428 /* 1429 * struct fileops kqfilter method for a kqueue descriptor. 1430 * Event triggered when monitored kqueue changes. 1431 */ 1432 static int 1433 kqueue_kqfilter(file_t *fp, struct knote *kn) 1434 { 1435 struct kqueue *kq; 1436 filedesc_t *fdp; 1437 1438 kq = ((file_t *)kn->kn_obj)->f_data; 1439 1440 KASSERT(fp == kn->kn_obj); 1441 1442 if (kn->kn_filter != EVFILT_READ) 1443 return 1; 1444 1445 kn->kn_fop = &kqread_filtops; 1446 fdp = curlwp->l_fd; 1447 mutex_enter(&kq->kq_lock); 1448 SLIST_INSERT_HEAD(&kq->kq_sel.sel_klist, kn, kn_selnext); 1449 mutex_exit(&kq->kq_lock); 1450 1451 return 0; 1452 } 1453 1454 1455 /* 1456 * Walk down a list of knotes, activating them if their event has 1457 * triggered. The caller's object lock (e.g. device driver lock) 1458 * must be held. 1459 */ 1460 void 1461 knote(struct klist *list, long hint) 1462 { 1463 struct knote *kn, *tmpkn; 1464 1465 SLIST_FOREACH_SAFE(kn, list, kn_selnext, tmpkn) { 1466 if ((*kn->kn_fop->f_event)(kn, hint)) 1467 knote_activate(kn); 1468 } 1469 } 1470 1471 /* 1472 * Remove all knotes referencing a specified fd 1473 */ 1474 void 1475 knote_fdclose(int fd) 1476 { 1477 struct klist *list; 1478 struct knote *kn; 1479 filedesc_t *fdp; 1480 1481 fdp = curlwp->l_fd; 1482 list = (struct klist *)&fdp->fd_dt->dt_ff[fd]->ff_knlist; 1483 mutex_enter(&fdp->fd_lock); 1484 while ((kn = SLIST_FIRST(list)) != NULL) { 1485 knote_detach(kn, fdp, true); 1486 mutex_enter(&fdp->fd_lock); 1487 } 1488 mutex_exit(&fdp->fd_lock); 1489 } 1490 1491 /* 1492 * Drop knote. Called with fdp->fd_lock held, and will drop before 1493 * returning. 1494 */ 1495 static void 1496 knote_detach(struct knote *kn, filedesc_t *fdp, bool dofop) 1497 { 1498 struct klist *list; 1499 struct kqueue *kq; 1500 1501 kq = kn->kn_kq; 1502 1503 KASSERT((kn->kn_status & KN_MARKER) == 0); 1504 KASSERT(mutex_owned(&fdp->fd_lock)); 1505 1506 /* Remove from monitored object. */ 1507 if (dofop) { 1508 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1509 (*kn->kn_fop->f_detach)(kn); 1510 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1511 } 1512 1513 /* Remove from descriptor table. */ 1514 if (kn->kn_fop->f_isfd) 1515 list = (struct klist *)&fdp->fd_dt->dt_ff[kn->kn_id]->ff_knlist; 1516 else 1517 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1518 1519 SLIST_REMOVE(list, kn, knote, kn_link); 1520 1521 /* Remove from kqueue. */ 1522 /* XXXAD should verify not in use by kqueue_scan. */ 1523 mutex_spin_enter(&kq->kq_lock); 1524 if ((kn->kn_status & KN_QUEUED) != 0) { 1525 kq_check(kq); 1526 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1527 kn->kn_status &= ~KN_QUEUED; 1528 kq->kq_count--; 1529 kq_check(kq); 1530 } 1531 mutex_spin_exit(&kq->kq_lock); 1532 1533 mutex_exit(&fdp->fd_lock); 1534 if (kn->kn_fop->f_isfd) 1535 fd_putfile(kn->kn_id); 1536 atomic_dec_uint(&kn->kn_kfilter->refcnt); 1537 kmem_free(kn, sizeof(*kn)); 1538 } 1539 1540 /* 1541 * Queue new event for knote. 1542 */ 1543 static void 1544 knote_enqueue(struct knote *kn) 1545 { 1546 struct kqueue *kq; 1547 1548 KASSERT((kn->kn_status & KN_MARKER) == 0); 1549 1550 kq = kn->kn_kq; 1551 1552 mutex_spin_enter(&kq->kq_lock); 1553 if ((kn->kn_status & KN_DISABLED) != 0) { 1554 kn->kn_status &= ~KN_DISABLED; 1555 } 1556 if ((kn->kn_status & (KN_ACTIVE | KN_QUEUED)) == KN_ACTIVE) { 1557 kq_check(kq); 1558 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1559 kn->kn_status |= KN_QUEUED; 1560 kq->kq_count++; 1561 kq_check(kq); 1562 cv_broadcast(&kq->kq_cv); 1563 selnotify(&kq->kq_sel, 0, NOTE_SUBMIT); 1564 } 1565 mutex_spin_exit(&kq->kq_lock); 1566 } 1567 /* 1568 * Queue new event for knote. 1569 */ 1570 static void 1571 knote_activate(struct knote *kn) 1572 { 1573 struct kqueue *kq; 1574 1575 KASSERT((kn->kn_status & KN_MARKER) == 0); 1576 1577 kq = kn->kn_kq; 1578 1579 mutex_spin_enter(&kq->kq_lock); 1580 kn->kn_status |= KN_ACTIVE; 1581 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) { 1582 kq_check(kq); 1583 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1584 kn->kn_status |= KN_QUEUED; 1585 kq->kq_count++; 1586 kq_check(kq); 1587 cv_broadcast(&kq->kq_cv); 1588 selnotify(&kq->kq_sel, 0, NOTE_SUBMIT); 1589 } 1590 mutex_spin_exit(&kq->kq_lock); 1591 } 1592