1 /*- 2 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD: src/sys/kern/kern_event.c,v 1.2.2.10 2004/04/04 07:03:14 cperciva Exp $ 27 * $DragonFly: src/sys/kern/kern_event.c,v 1.18 2005/10/19 07:51:42 sephe Exp $ 28 */ 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/proc.h> 34 #include <sys/malloc.h> 35 #include <sys/unistd.h> 36 #include <sys/file.h> 37 #include <sys/fcntl.h> 38 #include <sys/select.h> 39 #include <sys/queue.h> 40 #include <sys/event.h> 41 #include <sys/eventvar.h> 42 #include <sys/poll.h> 43 #include <sys/protosw.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/stat.h> 47 #include <sys/sysctl.h> 48 #include <sys/sysproto.h> 49 #include <sys/uio.h> 50 #include <sys/thread2.h> 51 #include <sys/file2.h> 52 53 #include <vm/vm_zone.h> 54 55 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 56 57 static int kqueue_scan(struct file *fp, int maxevents, 58 struct kevent *ulistp, const struct timespec *timeout, 59 struct thread *td, int *res); 60 static int kqueue_read(struct file *fp, struct uio *uio, 61 struct ucred *cred, int flags, struct thread *td); 62 static int kqueue_write(struct file *fp, struct uio *uio, 63 struct ucred *cred, int flags, struct thread *td); 64 static int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 65 struct thread *td); 66 static int kqueue_poll(struct file *fp, int events, struct ucred *cred, 67 struct thread *td); 68 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 69 static int kqueue_stat(struct file *fp, struct stat *st, struct thread *td); 70 static int kqueue_close(struct file *fp, struct thread *td); 71 static void kqueue_wakeup(struct kqueue *kq); 72 73 static struct fileops kqueueops = { 74 NULL, /* port */ 75 NULL, /* clone */ 76 kqueue_read, 77 kqueue_write, 78 kqueue_ioctl, 79 kqueue_poll, 80 kqueue_kqfilter, 81 kqueue_stat, 82 kqueue_close, 83 nofo_shutdown 84 }; 85 86 static void knote_attach(struct knote *kn, struct filedesc *fdp); 87 static void knote_drop(struct knote *kn, struct thread *td); 88 static void knote_enqueue(struct knote *kn); 89 static void knote_dequeue(struct knote *kn); 90 static void knote_init(void); 91 static struct knote *knote_alloc(void); 92 static void knote_free(struct knote *kn); 93 94 static void filt_kqdetach(struct knote *kn); 95 static int filt_kqueue(struct knote *kn, long hint); 96 static int filt_procattach(struct knote *kn); 97 static void filt_procdetach(struct knote *kn); 98 static int filt_proc(struct knote *kn, long hint); 99 static int filt_fileattach(struct knote *kn); 100 static void filt_timerexpire(void *knx); 101 static int filt_timerattach(struct knote *kn); 102 static void filt_timerdetach(struct knote *kn); 103 static int filt_timer(struct knote *kn, long hint); 104 105 static struct filterops file_filtops = 106 { 1, filt_fileattach, NULL, NULL }; 107 static struct filterops kqread_filtops = 108 { 1, NULL, filt_kqdetach, filt_kqueue }; 109 static struct filterops proc_filtops = 110 { 0, filt_procattach, filt_procdetach, filt_proc }; 111 static struct filterops timer_filtops = 112 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 113 114 static vm_zone_t knote_zone; 115 static int kq_ncallouts = 0; 116 static int kq_calloutmax = (4 * 1024); 117 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 118 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 119 120 #define KNOTE_ACTIVATE(kn) do { \ 121 kn->kn_status |= KN_ACTIVE; \ 122 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 123 knote_enqueue(kn); \ 124 } while(0) 125 126 #define KN_HASHSIZE 64 /* XXX should be tunable */ 127 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 128 129 extern struct filterops aio_filtops; 130 extern struct filterops sig_filtops; 131 132 /* 133 * Table for for all system-defined filters. 134 */ 135 static struct filterops *sysfilt_ops[] = { 136 &file_filtops, /* EVFILT_READ */ 137 &file_filtops, /* EVFILT_WRITE */ 138 &aio_filtops, /* EVFILT_AIO */ 139 &file_filtops, /* EVFILT_VNODE */ 140 &proc_filtops, /* EVFILT_PROC */ 141 &sig_filtops, /* EVFILT_SIGNAL */ 142 &timer_filtops, /* EVFILT_TIMER */ 143 }; 144 145 static int 146 filt_fileattach(struct knote *kn) 147 { 148 return (fo_kqfilter(kn->kn_fp, kn)); 149 } 150 151 /*ARGSUSED*/ 152 static int 153 kqueue_kqfilter(struct file *fp, struct knote *kn) 154 { 155 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 156 157 if (kn->kn_filter != EVFILT_READ) 158 return (1); 159 160 kn->kn_fop = &kqread_filtops; 161 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 162 return (0); 163 } 164 165 static void 166 filt_kqdetach(struct knote *kn) 167 { 168 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 169 170 SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext); 171 } 172 173 /*ARGSUSED*/ 174 static int 175 filt_kqueue(struct knote *kn, long hint) 176 { 177 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 178 179 kn->kn_data = kq->kq_count; 180 return (kn->kn_data > 0); 181 } 182 183 static int 184 filt_procattach(struct knote *kn) 185 { 186 struct proc *p; 187 int immediate; 188 189 immediate = 0; 190 p = pfind(kn->kn_id); 191 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 192 p = zpfind(kn->kn_id); 193 immediate = 1; 194 } 195 if (p == NULL) 196 return (ESRCH); 197 if (! PRISON_CHECK(curproc->p_ucred, p->p_ucred)) 198 return (EACCES); 199 200 kn->kn_ptr.p_proc = p; 201 kn->kn_flags |= EV_CLEAR; /* automatically set */ 202 203 /* 204 * internal flag indicating registration done by kernel 205 */ 206 if (kn->kn_flags & EV_FLAG1) { 207 kn->kn_data = kn->kn_sdata; /* ppid */ 208 kn->kn_fflags = NOTE_CHILD; 209 kn->kn_flags &= ~EV_FLAG1; 210 } 211 212 /* XXX lock the proc here while adding to the list? */ 213 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 214 215 /* 216 * Immediately activate any exit notes if the target process is a 217 * zombie. This is necessary to handle the case where the target 218 * process, e.g. a child, dies before the kevent is registered. 219 */ 220 if (immediate && filt_proc(kn, NOTE_EXIT)) 221 KNOTE_ACTIVATE(kn); 222 223 return (0); 224 } 225 226 /* 227 * The knote may be attached to a different process, which may exit, 228 * leaving nothing for the knote to be attached to. So when the process 229 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 230 * it will be deleted when read out. However, as part of the knote deletion, 231 * this routine is called, so a check is needed to avoid actually performing 232 * a detach, because the original process does not exist any more. 233 */ 234 static void 235 filt_procdetach(struct knote *kn) 236 { 237 struct proc *p = kn->kn_ptr.p_proc; 238 239 if (kn->kn_status & KN_DETACHED) 240 return; 241 242 /* XXX locking? this might modify another process. */ 243 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 244 } 245 246 static int 247 filt_proc(struct knote *kn, long hint) 248 { 249 u_int event; 250 251 /* 252 * mask off extra data 253 */ 254 event = (u_int)hint & NOTE_PCTRLMASK; 255 256 /* 257 * if the user is interested in this event, record it. 258 */ 259 if (kn->kn_sfflags & event) 260 kn->kn_fflags |= event; 261 262 /* 263 * process is gone, so flag the event as finished. 264 */ 265 if (event == NOTE_EXIT) { 266 kn->kn_status |= KN_DETACHED; 267 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 268 return (1); 269 } 270 271 /* 272 * process forked, and user wants to track the new process, 273 * so attach a new knote to it, and immediately report an 274 * event with the parent's pid. 275 */ 276 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 277 struct kevent kev; 278 int error; 279 280 /* 281 * register knote with new process. 282 */ 283 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 284 kev.filter = kn->kn_filter; 285 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 286 kev.fflags = kn->kn_sfflags; 287 kev.data = kn->kn_id; /* parent */ 288 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 289 error = kqueue_register(kn->kn_kq, &kev, NULL); 290 if (error) 291 kn->kn_fflags |= NOTE_TRACKERR; 292 } 293 294 return (kn->kn_fflags != 0); 295 } 296 297 static void 298 filt_timerexpire(void *knx) 299 { 300 struct knote *kn = knx; 301 struct callout *calloutp; 302 struct timeval tv; 303 int tticks; 304 305 kn->kn_data++; 306 KNOTE_ACTIVATE(kn); 307 308 if ((kn->kn_flags & EV_ONESHOT) == 0) { 309 tv.tv_sec = kn->kn_sdata / 1000; 310 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 311 tticks = tvtohz_high(&tv); 312 calloutp = (struct callout *)kn->kn_hook; 313 callout_reset(calloutp, tticks, filt_timerexpire, kn); 314 } 315 } 316 317 /* 318 * data contains amount of time to sleep, in milliseconds 319 */ 320 static int 321 filt_timerattach(struct knote *kn) 322 { 323 struct callout *calloutp; 324 struct timeval tv; 325 int tticks; 326 327 if (kq_ncallouts >= kq_calloutmax) 328 return (ENOMEM); 329 kq_ncallouts++; 330 331 tv.tv_sec = kn->kn_sdata / 1000; 332 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 333 tticks = tvtohz_high(&tv); 334 335 kn->kn_flags |= EV_CLEAR; /* automatically set */ 336 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 337 M_KQUEUE, M_WAITOK); 338 callout_init(calloutp); 339 kn->kn_hook = (caddr_t)calloutp; 340 callout_reset(calloutp, tticks, filt_timerexpire, kn); 341 342 return (0); 343 } 344 345 static void 346 filt_timerdetach(struct knote *kn) 347 { 348 struct callout *calloutp; 349 350 calloutp = (struct callout *)kn->kn_hook; 351 callout_stop(calloutp); 352 FREE(calloutp, M_KQUEUE); 353 kq_ncallouts--; 354 } 355 356 static int 357 filt_timer(struct knote *kn, long hint) 358 { 359 360 return (kn->kn_data != 0); 361 } 362 363 int 364 kqueue(struct kqueue_args *uap) 365 { 366 struct proc *p = curproc; 367 struct filedesc *fdp = p->p_fd; 368 struct kqueue *kq; 369 struct file *fp; 370 int fd, error; 371 372 error = falloc(p, &fp, &fd); 373 if (error) 374 return (error); 375 fp->f_flag = FREAD | FWRITE; 376 fp->f_type = DTYPE_KQUEUE; 377 fp->f_ops = &kqueueops; 378 kq = malloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 379 TAILQ_INIT(&kq->kq_head); 380 fp->f_data = kq; 381 uap->sysmsg_result = fd; 382 fdrop(fp, curthread); 383 if (fdp->fd_knlistsize < 0) 384 fdp->fd_knlistsize = 0; /* this process has a kq */ 385 kq->kq_fdp = fdp; 386 return (error); 387 } 388 389 int 390 kevent(struct kevent_args *uap) 391 { 392 struct thread *td = curthread; 393 struct proc *p = td->td_proc; 394 struct filedesc *fdp; 395 struct kevent *kevp; 396 struct kqueue *kq; 397 struct file *fp = NULL; 398 struct timespec ts; 399 int i, n, nerrors, error; 400 401 KKASSERT(p); 402 fdp = p->p_fd; 403 404 if (((u_int)uap->fd) >= fdp->fd_nfiles || 405 (fp = fdp->fd_files[uap->fd].fp) == NULL || 406 (fp->f_type != DTYPE_KQUEUE)) 407 return (EBADF); 408 409 fhold(fp); 410 411 if (uap->timeout != NULL) { 412 error = copyin(uap->timeout, &ts, sizeof(ts)); 413 if (error) 414 goto done; 415 uap->timeout = &ts; 416 } 417 418 kq = (struct kqueue *)fp->f_data; 419 nerrors = 0; 420 421 while (uap->nchanges > 0) { 422 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges; 423 error = copyin(uap->changelist, kq->kq_kev, 424 n * sizeof(struct kevent)); 425 if (error) 426 goto done; 427 for (i = 0; i < n; i++) { 428 kevp = &kq->kq_kev[i]; 429 kevp->flags &= ~EV_SYSFLAGS; 430 error = kqueue_register(kq, kevp, td); 431 if (error) { 432 if (uap->nevents != 0) { 433 kevp->flags = EV_ERROR; 434 kevp->data = error; 435 (void) copyout((caddr_t)kevp, 436 (caddr_t)uap->eventlist, 437 sizeof(*kevp)); 438 uap->eventlist++; 439 uap->nevents--; 440 nerrors++; 441 } else { 442 goto done; 443 } 444 } 445 } 446 uap->nchanges -= n; 447 uap->changelist += n; 448 } 449 if (nerrors) { 450 uap->sysmsg_result = nerrors; 451 error = 0; 452 goto done; 453 } 454 455 error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, td, &uap->sysmsg_result); 456 done: 457 if (fp != NULL) 458 fdrop(fp, p->p_thread); 459 return (error); 460 } 461 462 int 463 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td) 464 { 465 struct filedesc *fdp = kq->kq_fdp; 466 struct filterops *fops; 467 struct file *fp = NULL; 468 struct knote *kn = NULL; 469 int error = 0; 470 471 if (kev->filter < 0) { 472 if (kev->filter + EVFILT_SYSCOUNT < 0) 473 return (EINVAL); 474 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 475 } else { 476 /* 477 * XXX 478 * filter attach routine is responsible for insuring that 479 * the identifier can be attached to it. 480 */ 481 printf("unknown filter: %d\n", kev->filter); 482 return (EINVAL); 483 } 484 485 if (fops->f_isfd) { 486 /* validate descriptor */ 487 if ((u_int)kev->ident >= fdp->fd_nfiles || 488 (fp = fdp->fd_files[kev->ident].fp) == NULL) 489 return (EBADF); 490 fhold(fp); 491 492 if (kev->ident < fdp->fd_knlistsize) { 493 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 494 if (kq == kn->kn_kq && 495 kev->filter == kn->kn_filter) 496 break; 497 } 498 } else { 499 if (fdp->fd_knhashmask != 0) { 500 struct klist *list; 501 502 list = &fdp->fd_knhash[ 503 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 504 SLIST_FOREACH(kn, list, kn_link) 505 if (kev->ident == kn->kn_id && 506 kq == kn->kn_kq && 507 kev->filter == kn->kn_filter) 508 break; 509 } 510 } 511 512 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 513 error = ENOENT; 514 goto done; 515 } 516 517 /* 518 * kn now contains the matching knote, or NULL if no match 519 */ 520 if (kev->flags & EV_ADD) { 521 522 if (kn == NULL) { 523 kn = knote_alloc(); 524 if (kn == NULL) { 525 error = ENOMEM; 526 goto done; 527 } 528 kn->kn_fp = fp; 529 kn->kn_kq = kq; 530 kn->kn_fop = fops; 531 532 /* 533 * apply reference count to knote structure, and 534 * do not release it at the end of this routine. 535 */ 536 fp = NULL; 537 538 kn->kn_sfflags = kev->fflags; 539 kn->kn_sdata = kev->data; 540 kev->fflags = 0; 541 kev->data = 0; 542 kn->kn_kevent = *kev; 543 544 knote_attach(kn, fdp); 545 if ((error = fops->f_attach(kn)) != 0) { 546 knote_drop(kn, td); 547 goto done; 548 } 549 } else { 550 /* 551 * The user may change some filter values after the 552 * initial EV_ADD, but doing so will not reset any 553 * filter which have already been triggered. 554 */ 555 kn->kn_sfflags = kev->fflags; 556 kn->kn_sdata = kev->data; 557 kn->kn_kevent.udata = kev->udata; 558 } 559 560 crit_enter(); 561 if (kn->kn_fop->f_event(kn, 0)) 562 KNOTE_ACTIVATE(kn); 563 crit_exit(); 564 } else if (kev->flags & EV_DELETE) { 565 kn->kn_fop->f_detach(kn); 566 knote_drop(kn, td); 567 goto done; 568 } 569 570 if ((kev->flags & EV_DISABLE) && 571 ((kn->kn_status & KN_DISABLED) == 0)) { 572 crit_enter(); 573 kn->kn_status |= KN_DISABLED; 574 crit_exit(); 575 } 576 577 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 578 crit_enter(); 579 kn->kn_status &= ~KN_DISABLED; 580 if ((kn->kn_status & KN_ACTIVE) && 581 ((kn->kn_status & KN_QUEUED) == 0)) 582 knote_enqueue(kn); 583 crit_exit(); 584 } 585 586 done: 587 if (fp != NULL) 588 fdrop(fp, td); 589 return (error); 590 } 591 592 static int 593 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp, 594 const struct timespec *tsp, struct thread *td, int *res) 595 { 596 struct kqueue *kq = (struct kqueue *)fp->f_data; 597 struct kevent *kevp; 598 struct timeval atv, rtv, ttv; 599 struct knote *kn, marker; 600 int count, timeout, nkev = 0, error = 0; 601 602 count = maxevents; 603 if (count == 0) 604 goto done; 605 606 if (tsp != NULL) { 607 TIMESPEC_TO_TIMEVAL(&atv, tsp); 608 if (itimerfix(&atv)) { 609 error = EINVAL; 610 goto done; 611 } 612 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 613 timeout = -1; 614 else 615 timeout = atv.tv_sec > 24 * 60 * 60 ? 616 24 * 60 * 60 * hz : tvtohz_high(&atv); 617 getmicrouptime(&rtv); 618 timevaladd(&atv, &rtv); 619 } else { 620 atv.tv_sec = 0; 621 atv.tv_usec = 0; 622 timeout = 0; 623 } 624 goto start; 625 626 retry: 627 if (atv.tv_sec || atv.tv_usec) { 628 getmicrouptime(&rtv); 629 if (timevalcmp(&rtv, &atv, >=)) 630 goto done; 631 ttv = atv; 632 timevalsub(&ttv, &rtv); 633 timeout = ttv.tv_sec > 24 * 60 * 60 ? 634 24 * 60 * 60 * hz : tvtohz_high(&ttv); 635 } 636 637 start: 638 kevp = kq->kq_kev; 639 crit_enter(); 640 if (kq->kq_count == 0) { 641 if (timeout < 0) { 642 error = EWOULDBLOCK; 643 } else { 644 kq->kq_state |= KQ_SLEEP; 645 error = tsleep(kq, PCATCH, "kqread", timeout); 646 } 647 crit_exit(); 648 if (error == 0) 649 goto retry; 650 /* don't restart after signals... */ 651 if (error == ERESTART) 652 error = EINTR; 653 else if (error == EWOULDBLOCK) 654 error = 0; 655 goto done; 656 } 657 658 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 659 while (count) { 660 kn = TAILQ_FIRST(&kq->kq_head); 661 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 662 if (kn == &marker) { 663 crit_exit(); 664 if (count == maxevents) 665 goto retry; 666 goto done; 667 } 668 if (kn->kn_status & KN_DISABLED) { 669 kn->kn_status &= ~KN_QUEUED; 670 kq->kq_count--; 671 continue; 672 } 673 if ((kn->kn_flags & EV_ONESHOT) == 0 && 674 kn->kn_fop->f_event(kn, 0) == 0) { 675 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 676 kq->kq_count--; 677 continue; 678 } 679 *kevp = kn->kn_kevent; 680 kevp++; 681 nkev++; 682 if (kn->kn_flags & EV_ONESHOT) { 683 kn->kn_status &= ~KN_QUEUED; 684 kq->kq_count--; 685 crit_exit(); 686 kn->kn_fop->f_detach(kn); 687 knote_drop(kn, td); 688 crit_enter(); 689 } else if (kn->kn_flags & EV_CLEAR) { 690 kn->kn_data = 0; 691 kn->kn_fflags = 0; 692 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 693 kq->kq_count--; 694 } else { 695 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 696 } 697 count--; 698 if (nkev == KQ_NEVENTS) { 699 crit_exit(); 700 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 701 sizeof(struct kevent) * nkev); 702 ulistp += nkev; 703 nkev = 0; 704 kevp = kq->kq_kev; 705 crit_enter(); 706 if (error) 707 break; 708 } 709 } 710 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 711 crit_exit(); 712 done: 713 if (nkev != 0) 714 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 715 sizeof(struct kevent) * nkev); 716 *res = maxevents - count; 717 return (error); 718 } 719 720 /* 721 * XXX 722 * This could be expanded to call kqueue_scan, if desired. 723 */ 724 /*ARGSUSED*/ 725 static int 726 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, 727 int flags, struct thread *td) 728 { 729 return (ENXIO); 730 } 731 732 /*ARGSUSED*/ 733 static int 734 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, 735 int flags, struct thread *td) 736 { 737 return (ENXIO); 738 } 739 740 /*ARGSUSED*/ 741 static int 742 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td) 743 { 744 return (ENOTTY); 745 } 746 747 /*ARGSUSED*/ 748 static int 749 kqueue_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 750 { 751 struct kqueue *kq = (struct kqueue *)fp->f_data; 752 int revents = 0; 753 754 crit_enter(); 755 if (events & (POLLIN | POLLRDNORM)) { 756 if (kq->kq_count) { 757 revents |= events & (POLLIN | POLLRDNORM); 758 } else { 759 selrecord(td, &kq->kq_sel); 760 kq->kq_state |= KQ_SEL; 761 } 762 } 763 crit_exit(); 764 return (revents); 765 } 766 767 /*ARGSUSED*/ 768 static int 769 kqueue_stat(struct file *fp, struct stat *st, struct thread *td) 770 { 771 struct kqueue *kq = (struct kqueue *)fp->f_data; 772 773 bzero((void *)st, sizeof(*st)); 774 st->st_size = kq->kq_count; 775 st->st_blksize = sizeof(struct kevent); 776 st->st_mode = S_IFIFO; 777 return (0); 778 } 779 780 /*ARGSUSED*/ 781 static int 782 kqueue_close(struct file *fp, struct thread *td) 783 { 784 struct proc *p = td->td_proc; 785 struct kqueue *kq = (struct kqueue *)fp->f_data; 786 struct filedesc *fdp; 787 struct knote **knp, *kn, *kn0; 788 int i; 789 790 KKASSERT(p); 791 fdp = p->p_fd; 792 for (i = 0; i < fdp->fd_knlistsize; i++) { 793 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 794 kn = *knp; 795 while (kn != NULL) { 796 kn0 = SLIST_NEXT(kn, kn_link); 797 if (kq == kn->kn_kq) { 798 kn->kn_fop->f_detach(kn); 799 fdrop(kn->kn_fp, td); 800 knote_free(kn); 801 *knp = kn0; 802 } else { 803 knp = &SLIST_NEXT(kn, kn_link); 804 } 805 kn = kn0; 806 } 807 } 808 if (fdp->fd_knhashmask != 0) { 809 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 810 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 811 kn = *knp; 812 while (kn != NULL) { 813 kn0 = SLIST_NEXT(kn, kn_link); 814 if (kq == kn->kn_kq) { 815 kn->kn_fop->f_detach(kn); 816 /* XXX non-fd release of kn->kn_ptr */ 817 knote_free(kn); 818 *knp = kn0; 819 } else { 820 knp = &SLIST_NEXT(kn, kn_link); 821 } 822 kn = kn0; 823 } 824 } 825 } 826 free(kq, M_KQUEUE); 827 fp->f_data = NULL; 828 829 return (0); 830 } 831 832 static void 833 kqueue_wakeup(struct kqueue *kq) 834 { 835 836 if (kq->kq_state & KQ_SLEEP) { 837 kq->kq_state &= ~KQ_SLEEP; 838 wakeup(kq); 839 } 840 if (kq->kq_state & KQ_SEL) { 841 kq->kq_state &= ~KQ_SEL; 842 selwakeup(&kq->kq_sel); 843 } 844 KNOTE(&kq->kq_sel.si_note, 0); 845 } 846 847 /* 848 * walk down a list of knotes, activating them if their event has triggered. 849 */ 850 void 851 knote(struct klist *list, long hint) 852 { 853 struct knote *kn; 854 855 SLIST_FOREACH(kn, list, kn_selnext) 856 if (kn->kn_fop->f_event(kn, hint)) 857 KNOTE_ACTIVATE(kn); 858 } 859 860 /* 861 * remove all knotes from a specified klist 862 */ 863 void 864 knote_remove(struct thread *td, struct klist *list) 865 { 866 struct knote *kn; 867 868 while ((kn = SLIST_FIRST(list)) != NULL) { 869 kn->kn_fop->f_detach(kn); 870 knote_drop(kn, td); 871 } 872 } 873 874 /* 875 * remove all knotes referencing a specified fd 876 */ 877 void 878 knote_fdclose(struct proc *p, int fd) 879 { 880 struct filedesc *fdp = p->p_fd; 881 struct klist *list = &fdp->fd_knlist[fd]; 882 883 knote_remove(p->p_thread, list); 884 } 885 886 static void 887 knote_attach(struct knote *kn, struct filedesc *fdp) 888 { 889 struct klist *list; 890 int size; 891 892 if (! kn->kn_fop->f_isfd) { 893 if (fdp->fd_knhashmask == 0) 894 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 895 &fdp->fd_knhashmask); 896 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 897 goto done; 898 } 899 900 if (fdp->fd_knlistsize <= kn->kn_id) { 901 size = fdp->fd_knlistsize; 902 while (size <= kn->kn_id) 903 size += KQEXTENT; 904 MALLOC(list, struct klist *, 905 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 906 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, 907 fdp->fd_knlistsize * sizeof(struct klist *)); 908 bzero((caddr_t)list + 909 fdp->fd_knlistsize * sizeof(struct klist *), 910 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 911 if (fdp->fd_knlist != NULL) 912 FREE(fdp->fd_knlist, M_KQUEUE); 913 fdp->fd_knlistsize = size; 914 fdp->fd_knlist = list; 915 } 916 list = &fdp->fd_knlist[kn->kn_id]; 917 done: 918 SLIST_INSERT_HEAD(list, kn, kn_link); 919 kn->kn_status = 0; 920 } 921 922 /* 923 * should be called outside of a critical section, since we don't want to 924 * hold a critical section while calling fdrop and free. 925 */ 926 static void 927 knote_drop(struct knote *kn, struct thread *td) 928 { 929 struct filedesc *fdp; 930 struct klist *list; 931 932 KKASSERT(td->td_proc); 933 fdp = td->td_proc->p_fd; 934 if (kn->kn_fop->f_isfd) 935 list = &fdp->fd_knlist[kn->kn_id]; 936 else 937 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 938 939 SLIST_REMOVE(list, kn, knote, kn_link); 940 if (kn->kn_status & KN_QUEUED) 941 knote_dequeue(kn); 942 if (kn->kn_fop->f_isfd) 943 fdrop(kn->kn_fp, td); 944 knote_free(kn); 945 } 946 947 948 static void 949 knote_enqueue(struct knote *kn) 950 { 951 struct kqueue *kq = kn->kn_kq; 952 953 crit_enter(); 954 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 955 956 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 957 kn->kn_status |= KN_QUEUED; 958 kq->kq_count++; 959 crit_exit(); 960 kqueue_wakeup(kq); 961 } 962 963 static void 964 knote_dequeue(struct knote *kn) 965 { 966 struct kqueue *kq = kn->kn_kq; 967 968 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 969 crit_enter(); 970 971 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 972 kn->kn_status &= ~KN_QUEUED; 973 kq->kq_count--; 974 crit_exit(); 975 } 976 977 static void 978 knote_init(void) 979 { 980 knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1); 981 } 982 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 983 984 static struct knote * 985 knote_alloc(void) 986 { 987 return ((struct knote *)zalloc(knote_zone)); 988 } 989 990 static void 991 knote_free(struct knote *kn) 992 { 993 zfree(knote_zone, kn); 994 } 995