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 1.28 2006/08/02 01:25:25 dillon 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/lock.h> 38 #include <sys/fcntl.h> 39 #include <sys/select.h> 40 #include <sys/queue.h> 41 #include <sys/event.h> 42 #include <sys/eventvar.h> 43 #include <sys/poll.h> 44 #include <sys/protosw.h> 45 #include <sys/socket.h> 46 #include <sys/socketvar.h> 47 #include <sys/stat.h> 48 #include <sys/sysctl.h> 49 #include <sys/sysproto.h> 50 #include <sys/uio.h> 51 #include <sys/thread2.h> 52 #include <sys/file2.h> 53 54 #include <vm/vm_zone.h> 55 56 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 57 58 static int kqueue_scan(struct file *fp, int maxevents, 59 struct kevent *ulistp, const struct timespec *timeout, 60 struct thread *td, int *res); 61 static int kqueue_read(struct file *fp, struct uio *uio, 62 struct ucred *cred, int flags); 63 static int kqueue_write(struct file *fp, struct uio *uio, 64 struct ucred *cred, int flags); 65 static int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 66 struct ucred *cred); 67 static int kqueue_poll(struct file *fp, int events, struct ucred *cred); 68 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 69 static int kqueue_stat(struct file *fp, struct stat *st, 70 struct ucred *cred); 71 static int kqueue_close(struct file *fp); 72 static void kqueue_wakeup(struct kqueue *kq); 73 74 /* 75 * MPSAFE 76 */ 77 static struct fileops kqueueops = { 78 .fo_read = kqueue_read, 79 .fo_write = kqueue_write, 80 .fo_ioctl = kqueue_ioctl, 81 .fo_poll = kqueue_poll, 82 .fo_kqfilter = kqueue_kqfilter, 83 .fo_stat = kqueue_stat, 84 .fo_close = kqueue_close, 85 .fo_shutdown = nofo_shutdown 86 }; 87 88 static void knote_attach(struct knote *kn, struct filedesc *fdp); 89 static void knote_drop(struct knote *kn, struct thread *td); 90 static void knote_enqueue(struct knote *kn); 91 static void knote_dequeue(struct knote *kn); 92 static void knote_init(void); 93 static struct knote *knote_alloc(void); 94 static void knote_free(struct knote *kn); 95 96 static void filt_kqdetach(struct knote *kn); 97 static int filt_kqueue(struct knote *kn, long hint); 98 static int filt_procattach(struct knote *kn); 99 static void filt_procdetach(struct knote *kn); 100 static int filt_proc(struct knote *kn, long hint); 101 static int filt_fileattach(struct knote *kn); 102 static void filt_timerexpire(void *knx); 103 static int filt_timerattach(struct knote *kn); 104 static void filt_timerdetach(struct knote *kn); 105 static int filt_timer(struct knote *kn, long hint); 106 107 static struct filterops file_filtops = 108 { 1, filt_fileattach, NULL, NULL }; 109 static struct filterops kqread_filtops = 110 { 1, NULL, filt_kqdetach, filt_kqueue }; 111 static struct filterops proc_filtops = 112 { 0, filt_procattach, filt_procdetach, filt_proc }; 113 static struct filterops timer_filtops = 114 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 115 116 static vm_zone_t knote_zone; 117 static int kq_ncallouts = 0; 118 static int kq_calloutmax = (4 * 1024); 119 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 120 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 121 122 #define KNOTE_ACTIVATE(kn) do { \ 123 kn->kn_status |= KN_ACTIVE; \ 124 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 125 knote_enqueue(kn); \ 126 } while(0) 127 128 #define KN_HASHSIZE 64 /* XXX should be tunable */ 129 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 130 131 extern struct filterops aio_filtops; 132 extern struct filterops sig_filtops; 133 134 /* 135 * Table for for all system-defined filters. 136 */ 137 static struct filterops *sysfilt_ops[] = { 138 &file_filtops, /* EVFILT_READ */ 139 &file_filtops, /* EVFILT_WRITE */ 140 &aio_filtops, /* EVFILT_AIO */ 141 &file_filtops, /* EVFILT_VNODE */ 142 &proc_filtops, /* EVFILT_PROC */ 143 &sig_filtops, /* EVFILT_SIGNAL */ 144 &timer_filtops, /* EVFILT_TIMER */ 145 }; 146 147 static int 148 filt_fileattach(struct knote *kn) 149 { 150 return (fo_kqfilter(kn->kn_fp, kn)); 151 } 152 153 /* 154 * MPALMOSTSAFE - acquires mplock 155 */ 156 static int 157 kqueue_kqfilter(struct file *fp, struct knote *kn) 158 { 159 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 160 161 get_mplock(); 162 if (kn->kn_filter != EVFILT_READ) { 163 rel_mplock(); 164 return (1); 165 } 166 167 kn->kn_fop = &kqread_filtops; 168 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 169 rel_mplock(); 170 return (0); 171 } 172 173 static void 174 filt_kqdetach(struct knote *kn) 175 { 176 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 177 178 SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext); 179 } 180 181 /*ARGSUSED*/ 182 static int 183 filt_kqueue(struct knote *kn, long hint) 184 { 185 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 186 187 kn->kn_data = kq->kq_count; 188 return (kn->kn_data > 0); 189 } 190 191 static int 192 filt_procattach(struct knote *kn) 193 { 194 struct proc *p; 195 int immediate; 196 197 immediate = 0; 198 p = pfind(kn->kn_id); 199 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 200 p = zpfind(kn->kn_id); 201 immediate = 1; 202 } 203 if (p == NULL) 204 return (ESRCH); 205 if (! PRISON_CHECK(curproc->p_ucred, p->p_ucred)) 206 return (EACCES); 207 208 kn->kn_ptr.p_proc = p; 209 kn->kn_flags |= EV_CLEAR; /* automatically set */ 210 211 /* 212 * internal flag indicating registration done by kernel 213 */ 214 if (kn->kn_flags & EV_FLAG1) { 215 kn->kn_data = kn->kn_sdata; /* ppid */ 216 kn->kn_fflags = NOTE_CHILD; 217 kn->kn_flags &= ~EV_FLAG1; 218 } 219 220 /* XXX lock the proc here while adding to the list? */ 221 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 222 223 /* 224 * Immediately activate any exit notes if the target process is a 225 * zombie. This is necessary to handle the case where the target 226 * process, e.g. a child, dies before the kevent is registered. 227 */ 228 if (immediate && filt_proc(kn, NOTE_EXIT)) 229 KNOTE_ACTIVATE(kn); 230 231 return (0); 232 } 233 234 /* 235 * The knote may be attached to a different process, which may exit, 236 * leaving nothing for the knote to be attached to. So when the process 237 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 238 * it will be deleted when read out. However, as part of the knote deletion, 239 * this routine is called, so a check is needed to avoid actually performing 240 * a detach, because the original process does not exist any more. 241 */ 242 static void 243 filt_procdetach(struct knote *kn) 244 { 245 struct proc *p = kn->kn_ptr.p_proc; 246 247 if (kn->kn_status & KN_DETACHED) 248 return; 249 250 /* XXX locking? this might modify another process. */ 251 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 252 } 253 254 static int 255 filt_proc(struct knote *kn, long hint) 256 { 257 u_int event; 258 259 /* 260 * mask off extra data 261 */ 262 event = (u_int)hint & NOTE_PCTRLMASK; 263 264 /* 265 * if the user is interested in this event, record it. 266 */ 267 if (kn->kn_sfflags & event) 268 kn->kn_fflags |= event; 269 270 /* 271 * process is gone, so flag the event as finished. 272 */ 273 if (event == NOTE_EXIT) { 274 kn->kn_status |= KN_DETACHED; 275 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 276 return (1); 277 } 278 279 /* 280 * process forked, and user wants to track the new process, 281 * so attach a new knote to it, and immediately report an 282 * event with the parent's pid. 283 */ 284 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 285 struct kevent kev; 286 int error; 287 288 /* 289 * register knote with new process. 290 */ 291 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 292 kev.filter = kn->kn_filter; 293 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 294 kev.fflags = kn->kn_sfflags; 295 kev.data = kn->kn_id; /* parent */ 296 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 297 error = kqueue_register(kn->kn_kq, &kev, NULL); 298 if (error) 299 kn->kn_fflags |= NOTE_TRACKERR; 300 } 301 302 return (kn->kn_fflags != 0); 303 } 304 305 static void 306 filt_timerexpire(void *knx) 307 { 308 struct knote *kn = knx; 309 struct callout *calloutp; 310 struct timeval tv; 311 int tticks; 312 313 kn->kn_data++; 314 KNOTE_ACTIVATE(kn); 315 316 if ((kn->kn_flags & EV_ONESHOT) == 0) { 317 tv.tv_sec = kn->kn_sdata / 1000; 318 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 319 tticks = tvtohz_high(&tv); 320 calloutp = (struct callout *)kn->kn_hook; 321 callout_reset(calloutp, tticks, filt_timerexpire, kn); 322 } 323 } 324 325 /* 326 * data contains amount of time to sleep, in milliseconds 327 */ 328 static int 329 filt_timerattach(struct knote *kn) 330 { 331 struct callout *calloutp; 332 struct timeval tv; 333 int tticks; 334 335 if (kq_ncallouts >= kq_calloutmax) 336 return (ENOMEM); 337 kq_ncallouts++; 338 339 tv.tv_sec = kn->kn_sdata / 1000; 340 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 341 tticks = tvtohz_high(&tv); 342 343 kn->kn_flags |= EV_CLEAR; /* automatically set */ 344 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 345 M_KQUEUE, M_WAITOK); 346 callout_init(calloutp); 347 kn->kn_hook = (caddr_t)calloutp; 348 callout_reset(calloutp, tticks, filt_timerexpire, kn); 349 350 return (0); 351 } 352 353 static void 354 filt_timerdetach(struct knote *kn) 355 { 356 struct callout *calloutp; 357 358 calloutp = (struct callout *)kn->kn_hook; 359 callout_stop(calloutp); 360 FREE(calloutp, M_KQUEUE); 361 kq_ncallouts--; 362 } 363 364 static int 365 filt_timer(struct knote *kn, long hint) 366 { 367 368 return (kn->kn_data != 0); 369 } 370 371 int 372 sys_kqueue(struct kqueue_args *uap) 373 { 374 struct proc *p = curproc; 375 struct filedesc *fdp = p->p_fd; 376 struct kqueue *kq; 377 struct file *fp; 378 int fd, error; 379 380 error = falloc(p, &fp, &fd); 381 if (error) 382 return (error); 383 fp->f_flag = FREAD | FWRITE; 384 fp->f_type = DTYPE_KQUEUE; 385 fp->f_ops = &kqueueops; 386 387 kq = malloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 388 TAILQ_INIT(&kq->kq_head); 389 kq->kq_fdp = fdp; 390 fp->f_data = kq; 391 392 fsetfd(p, fp, fd); 393 uap->sysmsg_result = fd; 394 fdrop(fp); 395 return (error); 396 } 397 398 int 399 sys_kevent(struct kevent_args *uap) 400 { 401 struct thread *td = curthread; 402 struct proc *p = td->td_proc; 403 struct kevent *kevp; 404 struct kqueue *kq; 405 struct file *fp = NULL; 406 struct timespec ts; 407 int i, n, nerrors, error; 408 409 KKASSERT(p); 410 411 fp = holdfp(p->p_fd, uap->fd, -1); 412 if (fp == NULL) 413 return (EBADF); 414 if (fp->f_type != DTYPE_KQUEUE) { 415 fdrop(fp); 416 return (EBADF); 417 } 418 419 if (uap->timeout != NULL) { 420 error = copyin(uap->timeout, &ts, sizeof(ts)); 421 if (error) 422 goto done; 423 uap->timeout = &ts; 424 } 425 426 kq = (struct kqueue *)fp->f_data; 427 nerrors = 0; 428 429 while (uap->nchanges > 0) { 430 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges; 431 error = copyin(uap->changelist, kq->kq_kev, 432 n * sizeof(struct kevent)); 433 if (error) 434 goto done; 435 for (i = 0; i < n; i++) { 436 kevp = &kq->kq_kev[i]; 437 kevp->flags &= ~EV_SYSFLAGS; 438 error = kqueue_register(kq, kevp, td); 439 if (error) { 440 if (uap->nevents != 0) { 441 kevp->flags = EV_ERROR; 442 kevp->data = error; 443 (void) copyout((caddr_t)kevp, 444 (caddr_t)uap->eventlist, 445 sizeof(*kevp)); 446 uap->eventlist++; 447 uap->nevents--; 448 nerrors++; 449 } else { 450 goto done; 451 } 452 } 453 } 454 uap->nchanges -= n; 455 uap->changelist += n; 456 } 457 if (nerrors) { 458 uap->sysmsg_result = nerrors; 459 error = 0; 460 goto done; 461 } 462 463 error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, td, &uap->sysmsg_result); 464 done: 465 if (fp != NULL) 466 fdrop(fp); 467 return (error); 468 } 469 470 int 471 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td) 472 { 473 struct filedesc *fdp = kq->kq_fdp; 474 struct filterops *fops; 475 struct file *fp = NULL; 476 struct knote *kn = NULL; 477 int error = 0; 478 479 if (kev->filter < 0) { 480 if (kev->filter + EVFILT_SYSCOUNT < 0) 481 return (EINVAL); 482 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 483 } else { 484 /* 485 * XXX 486 * filter attach routine is responsible for insuring that 487 * the identifier can be attached to it. 488 */ 489 printf("unknown filter: %d\n", kev->filter); 490 return (EINVAL); 491 } 492 493 if (fops->f_isfd) { 494 /* validate descriptor */ 495 fp = holdfp(fdp, kev->ident, -1); 496 if (fp == NULL) 497 return (EBADF); 498 499 if (kev->ident < fdp->fd_knlistsize) { 500 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 501 if (kq == kn->kn_kq && 502 kev->filter == kn->kn_filter) 503 break; 504 } 505 } else { 506 if (fdp->fd_knhashmask != 0) { 507 struct klist *list; 508 509 list = &fdp->fd_knhash[ 510 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 511 SLIST_FOREACH(kn, list, kn_link) 512 if (kev->ident == kn->kn_id && 513 kq == kn->kn_kq && 514 kev->filter == kn->kn_filter) 515 break; 516 } 517 } 518 519 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 520 error = ENOENT; 521 goto done; 522 } 523 524 /* 525 * kn now contains the matching knote, or NULL if no match 526 */ 527 if (kev->flags & EV_ADD) { 528 529 if (kn == NULL) { 530 kn = knote_alloc(); 531 if (kn == NULL) { 532 error = ENOMEM; 533 goto done; 534 } 535 kn->kn_fp = fp; 536 kn->kn_kq = kq; 537 kn->kn_fop = fops; 538 539 /* 540 * apply reference count to knote structure, and 541 * do not release it at the end of this routine. 542 */ 543 fp = NULL; 544 545 kn->kn_sfflags = kev->fflags; 546 kn->kn_sdata = kev->data; 547 kev->fflags = 0; 548 kev->data = 0; 549 kn->kn_kevent = *kev; 550 551 knote_attach(kn, fdp); 552 if ((error = fops->f_attach(kn)) != 0) { 553 knote_drop(kn, td); 554 goto done; 555 } 556 } else { 557 /* 558 * The user may change some filter values after the 559 * initial EV_ADD, but doing so will not reset any 560 * filter which have already been triggered. 561 */ 562 kn->kn_sfflags = kev->fflags; 563 kn->kn_sdata = kev->data; 564 kn->kn_kevent.udata = kev->udata; 565 } 566 567 crit_enter(); 568 if (kn->kn_fop->f_event(kn, 0)) 569 KNOTE_ACTIVATE(kn); 570 crit_exit(); 571 } else if (kev->flags & EV_DELETE) { 572 kn->kn_fop->f_detach(kn); 573 knote_drop(kn, td); 574 goto done; 575 } 576 577 if ((kev->flags & EV_DISABLE) && 578 ((kn->kn_status & KN_DISABLED) == 0)) { 579 crit_enter(); 580 kn->kn_status |= KN_DISABLED; 581 crit_exit(); 582 } 583 584 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 585 crit_enter(); 586 kn->kn_status &= ~KN_DISABLED; 587 if ((kn->kn_status & KN_ACTIVE) && 588 ((kn->kn_status & KN_QUEUED) == 0)) 589 knote_enqueue(kn); 590 crit_exit(); 591 } 592 593 done: 594 if (fp != NULL) 595 fdrop(fp); 596 return (error); 597 } 598 599 static int 600 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp, 601 const struct timespec *tsp, struct thread *td, int *res) 602 { 603 struct kqueue *kq = (struct kqueue *)fp->f_data; 604 struct kevent *kevp; 605 struct timeval atv, rtv, ttv; 606 struct knote *kn, marker; 607 int count, timeout, nkev = 0, error = 0; 608 609 count = maxevents; 610 if (count == 0) 611 goto done; 612 613 if (tsp != NULL) { 614 TIMESPEC_TO_TIMEVAL(&atv, tsp); 615 if (itimerfix(&atv)) { 616 error = EINVAL; 617 goto done; 618 } 619 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 620 timeout = -1; 621 else 622 timeout = atv.tv_sec > 24 * 60 * 60 ? 623 24 * 60 * 60 * hz : tvtohz_high(&atv); 624 getmicrouptime(&rtv); 625 timevaladd(&atv, &rtv); 626 } else { 627 atv.tv_sec = 0; 628 atv.tv_usec = 0; 629 timeout = 0; 630 } 631 goto start; 632 633 retry: 634 if (atv.tv_sec || atv.tv_usec) { 635 getmicrouptime(&rtv); 636 if (timevalcmp(&rtv, &atv, >=)) 637 goto done; 638 ttv = atv; 639 timevalsub(&ttv, &rtv); 640 timeout = ttv.tv_sec > 24 * 60 * 60 ? 641 24 * 60 * 60 * hz : tvtohz_high(&ttv); 642 } 643 644 start: 645 kevp = kq->kq_kev; 646 crit_enter(); 647 if (kq->kq_count == 0) { 648 if (timeout < 0) { 649 error = EWOULDBLOCK; 650 } else { 651 kq->kq_state |= KQ_SLEEP; 652 error = tsleep(kq, PCATCH, "kqread", timeout); 653 } 654 crit_exit(); 655 if (error == 0) 656 goto retry; 657 /* don't restart after signals... */ 658 if (error == ERESTART) 659 error = EINTR; 660 else if (error == EWOULDBLOCK) 661 error = 0; 662 goto done; 663 } 664 665 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 666 while (count) { 667 kn = TAILQ_FIRST(&kq->kq_head); 668 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 669 if (kn == &marker) { 670 crit_exit(); 671 if (count == maxevents) 672 goto retry; 673 goto done; 674 } 675 if (kn->kn_status & KN_DISABLED) { 676 kn->kn_status &= ~KN_QUEUED; 677 kq->kq_count--; 678 continue; 679 } 680 if ((kn->kn_flags & EV_ONESHOT) == 0 && 681 kn->kn_fop->f_event(kn, 0) == 0) { 682 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 683 kq->kq_count--; 684 continue; 685 } 686 *kevp = kn->kn_kevent; 687 kevp++; 688 nkev++; 689 if (kn->kn_flags & EV_ONESHOT) { 690 kn->kn_status &= ~KN_QUEUED; 691 kq->kq_count--; 692 crit_exit(); 693 kn->kn_fop->f_detach(kn); 694 knote_drop(kn, td); 695 crit_enter(); 696 } else if (kn->kn_flags & EV_CLEAR) { 697 kn->kn_data = 0; 698 kn->kn_fflags = 0; 699 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 700 kq->kq_count--; 701 } else { 702 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 703 } 704 count--; 705 if (nkev == KQ_NEVENTS) { 706 crit_exit(); 707 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 708 sizeof(struct kevent) * nkev); 709 ulistp += nkev; 710 nkev = 0; 711 kevp = kq->kq_kev; 712 crit_enter(); 713 if (error) 714 break; 715 } 716 } 717 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 718 crit_exit(); 719 done: 720 if (nkev != 0) 721 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 722 sizeof(struct kevent) * nkev); 723 *res = maxevents - count; 724 return (error); 725 } 726 727 /* 728 * XXX 729 * This could be expanded to call kqueue_scan, if desired. 730 * 731 * MPSAFE 732 */ 733 static int 734 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 735 { 736 return (ENXIO); 737 } 738 739 /* 740 * MPSAFE 741 */ 742 static int 743 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 744 { 745 return (ENXIO); 746 } 747 748 /* 749 * MPSAFE 750 */ 751 static int 752 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred) 753 { 754 return (ENOTTY); 755 } 756 757 /* 758 * MPALMOSTSAFE - acquires mplock 759 */ 760 static int 761 kqueue_poll(struct file *fp, int events, struct ucred *cred) 762 { 763 struct kqueue *kq = (struct kqueue *)fp->f_data; 764 int revents = 0; 765 766 get_mplock(); 767 crit_enter(); 768 if (events & (POLLIN | POLLRDNORM)) { 769 if (kq->kq_count) { 770 revents |= events & (POLLIN | POLLRDNORM); 771 } else { 772 selrecord(curthread, &kq->kq_sel); 773 kq->kq_state |= KQ_SEL; 774 } 775 } 776 crit_exit(); 777 rel_mplock(); 778 return (revents); 779 } 780 781 /* 782 * MPSAFE 783 */ 784 static int 785 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred) 786 { 787 struct kqueue *kq = (struct kqueue *)fp->f_data; 788 789 bzero((void *)st, sizeof(*st)); 790 st->st_size = kq->kq_count; 791 st->st_blksize = sizeof(struct kevent); 792 st->st_mode = S_IFIFO; 793 return (0); 794 } 795 796 /* 797 * MPALMOSTSAFE - acquires mplock 798 */ 799 static int 800 kqueue_close(struct file *fp) 801 { 802 struct thread *td = curthread; 803 struct proc *p = td->td_proc; 804 struct kqueue *kq = (struct kqueue *)fp->f_data; 805 struct filedesc *fdp; 806 struct knote **knp, *kn, *kn0; 807 int i; 808 809 KKASSERT(p); 810 get_mplock(); 811 fdp = p->p_fd; 812 for (i = 0; i < fdp->fd_knlistsize; i++) { 813 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 814 kn = *knp; 815 while (kn != NULL) { 816 kn0 = SLIST_NEXT(kn, kn_link); 817 if (kq == kn->kn_kq) { 818 kn->kn_fop->f_detach(kn); 819 fdrop(kn->kn_fp); 820 knote_free(kn); 821 *knp = kn0; 822 } else { 823 knp = &SLIST_NEXT(kn, kn_link); 824 } 825 kn = kn0; 826 } 827 } 828 if (fdp->fd_knhashmask != 0) { 829 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 830 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 831 kn = *knp; 832 while (kn != NULL) { 833 kn0 = SLIST_NEXT(kn, kn_link); 834 if (kq == kn->kn_kq) { 835 kn->kn_fop->f_detach(kn); 836 /* XXX non-fd release of kn->kn_ptr */ 837 knote_free(kn); 838 *knp = kn0; 839 } else { 840 knp = &SLIST_NEXT(kn, kn_link); 841 } 842 kn = kn0; 843 } 844 } 845 } 846 fp->f_data = NULL; 847 rel_mplock(); 848 849 free(kq, M_KQUEUE); 850 return (0); 851 } 852 853 static void 854 kqueue_wakeup(struct kqueue *kq) 855 { 856 if (kq->kq_state & KQ_SLEEP) { 857 kq->kq_state &= ~KQ_SLEEP; 858 wakeup(kq); 859 } 860 if (kq->kq_state & KQ_SEL) { 861 kq->kq_state &= ~KQ_SEL; 862 selwakeup(&kq->kq_sel); 863 } 864 KNOTE(&kq->kq_sel.si_note, 0); 865 } 866 867 /* 868 * walk down a list of knotes, activating them if their event has triggered. 869 */ 870 void 871 knote(struct klist *list, long hint) 872 { 873 struct knote *kn; 874 875 SLIST_FOREACH(kn, list, kn_selnext) 876 if (kn->kn_fop->f_event(kn, hint)) 877 KNOTE_ACTIVATE(kn); 878 } 879 880 /* 881 * remove all knotes from a specified klist 882 */ 883 void 884 knote_remove(struct thread *td, struct klist *list) 885 { 886 struct knote *kn; 887 888 while ((kn = SLIST_FIRST(list)) != NULL) { 889 kn->kn_fop->f_detach(kn); 890 knote_drop(kn, td); 891 } 892 } 893 894 /* 895 * remove all knotes referencing a specified fd 896 */ 897 void 898 knote_fdclose(struct proc *p, int fd) 899 { 900 struct filedesc *fdp = p->p_fd; 901 struct klist *list = &fdp->fd_knlist[fd]; 902 903 knote_remove(p->p_thread, list); 904 } 905 906 static void 907 knote_attach(struct knote *kn, struct filedesc *fdp) 908 { 909 struct klist *list; 910 int size; 911 912 if (! kn->kn_fop->f_isfd) { 913 if (fdp->fd_knhashmask == 0) 914 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 915 &fdp->fd_knhashmask); 916 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 917 goto done; 918 } 919 920 if (fdp->fd_knlistsize <= kn->kn_id) { 921 size = fdp->fd_knlistsize; 922 while (size <= kn->kn_id) 923 size += KQEXTENT; 924 MALLOC(list, struct klist *, 925 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 926 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, 927 fdp->fd_knlistsize * sizeof(struct klist *)); 928 bzero((caddr_t)list + 929 fdp->fd_knlistsize * sizeof(struct klist *), 930 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 931 if (fdp->fd_knlist != NULL) 932 FREE(fdp->fd_knlist, M_KQUEUE); 933 fdp->fd_knlistsize = size; 934 fdp->fd_knlist = list; 935 } 936 list = &fdp->fd_knlist[kn->kn_id]; 937 done: 938 SLIST_INSERT_HEAD(list, kn, kn_link); 939 kn->kn_status = 0; 940 } 941 942 /* 943 * should be called outside of a critical section, since we don't want to 944 * hold a critical section while calling fdrop and free. 945 */ 946 static void 947 knote_drop(struct knote *kn, struct thread *td) 948 { 949 struct filedesc *fdp; 950 struct klist *list; 951 952 KKASSERT(td->td_proc); 953 fdp = td->td_proc->p_fd; 954 if (kn->kn_fop->f_isfd) 955 list = &fdp->fd_knlist[kn->kn_id]; 956 else 957 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 958 959 SLIST_REMOVE(list, kn, knote, kn_link); 960 if (kn->kn_status & KN_QUEUED) 961 knote_dequeue(kn); 962 if (kn->kn_fop->f_isfd) 963 fdrop(kn->kn_fp); 964 knote_free(kn); 965 } 966 967 968 static void 969 knote_enqueue(struct knote *kn) 970 { 971 struct kqueue *kq = kn->kn_kq; 972 973 crit_enter(); 974 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 975 976 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 977 kn->kn_status |= KN_QUEUED; 978 kq->kq_count++; 979 crit_exit(); 980 kqueue_wakeup(kq); 981 } 982 983 static void 984 knote_dequeue(struct knote *kn) 985 { 986 struct kqueue *kq = kn->kn_kq; 987 988 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 989 crit_enter(); 990 991 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 992 kn->kn_status &= ~KN_QUEUED; 993 kq->kq_count--; 994 crit_exit(); 995 } 996 997 static void 998 knote_init(void) 999 { 1000 knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1); 1001 } 1002 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 1003 1004 static struct knote * 1005 knote_alloc(void) 1006 { 1007 return ((struct knote *)zalloc(knote_zone)); 1008 } 1009 1010 static void 1011 knote_free(struct knote *kn) 1012 { 1013 zfree(knote_zone, kn); 1014 } 1015