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 * $FreeBSD$ 27 */ 28 29 #include <sys/param.h> 30 #include <sys/systm.h> 31 #include <sys/kernel.h> 32 #include <sys/lock.h> 33 #include <sys/mutex.h> 34 #include <sys/proc.h> 35 #include <sys/malloc.h> 36 #include <sys/unistd.h> 37 #include <sys/file.h> 38 #include <sys/fcntl.h> 39 #include <sys/selinfo.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 52 #include <vm/vm_zone.h> 53 54 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 55 56 static int kqueue_scan(struct file *fp, int maxevents, 57 struct kevent *ulistp, const struct timespec *timeout, 58 struct thread *td); 59 static int kqueue_read(struct file *fp, struct uio *uio, 60 struct ucred *cred, int flags, struct thread *td); 61 static int kqueue_write(struct file *fp, struct uio *uio, 62 struct ucred *cred, int flags, struct thread *td); 63 static int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 64 struct thread *td); 65 static int kqueue_poll(struct file *fp, int events, struct ucred *cred, 66 struct thread *td); 67 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 68 static int kqueue_stat(struct file *fp, struct stat *st, struct thread *td); 69 static int kqueue_close(struct file *fp, struct thread *td); 70 static void kqueue_wakeup(struct kqueue *kq); 71 72 static struct fileops kqueueops = { 73 kqueue_read, 74 kqueue_write, 75 kqueue_ioctl, 76 kqueue_poll, 77 kqueue_kqfilter, 78 kqueue_stat, 79 kqueue_close 80 }; 81 82 static void knote_attach(struct knote *kn, struct filedesc *fdp); 83 static void knote_drop(struct knote *kn, struct thread *td); 84 static void knote_enqueue(struct knote *kn); 85 static void knote_dequeue(struct knote *kn); 86 static void knote_init(void); 87 static struct knote *knote_alloc(void); 88 static void knote_free(struct knote *kn); 89 90 static void filt_kqdetach(struct knote *kn); 91 static int filt_kqueue(struct knote *kn, long hint); 92 static int filt_procattach(struct knote *kn); 93 static void filt_procdetach(struct knote *kn); 94 static int filt_proc(struct knote *kn, long hint); 95 static int filt_fileattach(struct knote *kn); 96 static void filt_timerexpire(void *knx); 97 static int filt_timerattach(struct knote *kn); 98 static void filt_timerdetach(struct knote *kn); 99 static int filt_timer(struct knote *kn, long hint); 100 101 static struct filterops file_filtops = 102 { 1, filt_fileattach, NULL, NULL }; 103 static struct filterops kqread_filtops = 104 { 1, NULL, filt_kqdetach, filt_kqueue }; 105 static struct filterops proc_filtops = 106 { 0, filt_procattach, filt_procdetach, filt_proc }; 107 static struct filterops timer_filtops = 108 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 109 110 static vm_zone_t knote_zone; 111 static int kq_ncallouts = 0; 112 static int kq_calloutmax = (4 * 1024); 113 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 114 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 115 116 #define KNOTE_ACTIVATE(kn) do { \ 117 kn->kn_status |= KN_ACTIVE; \ 118 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 119 knote_enqueue(kn); \ 120 } while(0) 121 122 #define KN_HASHSIZE 64 /* XXX should be tunable */ 123 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 124 125 extern struct filterops aio_filtops; 126 extern struct filterops sig_filtops; 127 128 /* 129 * Table for for all system-defined filters. 130 */ 131 static struct filterops *sysfilt_ops[] = { 132 &file_filtops, /* EVFILT_READ */ 133 &file_filtops, /* EVFILT_WRITE */ 134 &aio_filtops, /* EVFILT_AIO */ 135 &file_filtops, /* EVFILT_VNODE */ 136 &proc_filtops, /* EVFILT_PROC */ 137 &sig_filtops, /* EVFILT_SIGNAL */ 138 &timer_filtops, /* EVFILT_TIMER */ 139 }; 140 141 static int 142 filt_fileattach(struct knote *kn) 143 { 144 145 return (fo_kqfilter(kn->kn_fp, kn)); 146 } 147 148 /*ARGSUSED*/ 149 static int 150 kqueue_kqfilter(struct file *fp, struct knote *kn) 151 { 152 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 153 154 if (kn->kn_filter != EVFILT_READ) 155 return (1); 156 157 kn->kn_fop = &kqread_filtops; 158 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 159 return (0); 160 } 161 162 static void 163 filt_kqdetach(struct knote *kn) 164 { 165 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 166 167 SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext); 168 } 169 170 /*ARGSUSED*/ 171 static int 172 filt_kqueue(struct knote *kn, long hint) 173 { 174 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 175 176 kn->kn_data = kq->kq_count; 177 return (kn->kn_data > 0); 178 } 179 180 static int 181 filt_procattach(struct knote *kn) 182 { 183 struct proc *p; 184 int error; 185 186 p = pfind(kn->kn_id); 187 if (p == NULL) 188 return (ESRCH); 189 if ((error = p_cansee(curproc, p))) { 190 PROC_UNLOCK(p); 191 return (error); 192 } 193 194 kn->kn_ptr.p_proc = p; 195 kn->kn_flags |= EV_CLEAR; /* automatically set */ 196 197 /* 198 * internal flag indicating registration done by kernel 199 */ 200 if (kn->kn_flags & EV_FLAG1) { 201 kn->kn_data = kn->kn_sdata; /* ppid */ 202 kn->kn_fflags = NOTE_CHILD; 203 kn->kn_flags &= ~EV_FLAG1; 204 } 205 206 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 207 PROC_UNLOCK(p); 208 209 return (0); 210 } 211 212 /* 213 * The knote may be attached to a different process, which may exit, 214 * leaving nothing for the knote to be attached to. So when the process 215 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 216 * it will be deleted when read out. However, as part of the knote deletion, 217 * this routine is called, so a check is needed to avoid actually performing 218 * a detach, because the original process does not exist any more. 219 */ 220 static void 221 filt_procdetach(struct knote *kn) 222 { 223 struct proc *p = kn->kn_ptr.p_proc; 224 225 if (kn->kn_status & KN_DETACHED) 226 return; 227 228 PROC_LOCK(p); 229 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 230 PROC_UNLOCK(p); 231 } 232 233 static int 234 filt_proc(struct knote *kn, long hint) 235 { 236 u_int event; 237 238 /* 239 * mask off extra data 240 */ 241 event = (u_int)hint & NOTE_PCTRLMASK; 242 243 /* 244 * if the user is interested in this event, record it. 245 */ 246 if (kn->kn_sfflags & event) 247 kn->kn_fflags |= event; 248 249 /* 250 * process is gone, so flag the event as finished. 251 */ 252 if (event == NOTE_EXIT) { 253 kn->kn_status |= KN_DETACHED; 254 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 255 return (1); 256 } 257 258 /* 259 * process forked, and user wants to track the new process, 260 * so attach a new knote to it, and immediately report an 261 * event with the parent's pid. 262 */ 263 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 264 struct kevent kev; 265 int error; 266 267 /* 268 * register knote with new process. 269 */ 270 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 271 kev.filter = kn->kn_filter; 272 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 273 kev.fflags = kn->kn_sfflags; 274 kev.data = kn->kn_id; /* parent */ 275 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 276 error = kqueue_register(kn->kn_kq, &kev, NULL); 277 if (error) 278 kn->kn_fflags |= NOTE_TRACKERR; 279 } 280 281 return (kn->kn_fflags != 0); 282 } 283 284 static void 285 filt_timerexpire(void *knx) 286 { 287 struct knote *kn = knx; 288 struct callout *calloutp; 289 struct timeval tv; 290 int tticks; 291 292 kn->kn_data++; 293 KNOTE_ACTIVATE(kn); 294 295 if ((kn->kn_flags & EV_ONESHOT) == 0) { 296 tv.tv_sec = kn->kn_sdata / 1000; 297 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 298 tticks = tvtohz(&tv); 299 calloutp = (struct callout *)kn->kn_hook; 300 callout_reset(calloutp, tticks, filt_timerexpire, kn); 301 } 302 } 303 304 /* 305 * data contains amount of time to sleep, in milliseconds 306 */ 307 static int 308 filt_timerattach(struct knote *kn) 309 { 310 struct callout *calloutp; 311 struct timeval tv; 312 int tticks; 313 314 if (kq_ncallouts >= kq_calloutmax) 315 return (ENOMEM); 316 kq_ncallouts++; 317 318 tv.tv_sec = kn->kn_sdata / 1000; 319 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 320 tticks = tvtohz(&tv); 321 322 kn->kn_flags |= EV_CLEAR; /* automatically set */ 323 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 324 M_KQUEUE, M_WAITOK); 325 callout_init(calloutp, 0); 326 callout_reset(calloutp, tticks, filt_timerexpire, kn); 327 kn->kn_hook = (caddr_t)calloutp; 328 329 return (0); 330 } 331 332 static void 333 filt_timerdetach(struct knote *kn) 334 { 335 struct callout *calloutp; 336 337 calloutp = (struct callout *)kn->kn_hook; 338 callout_stop(calloutp); 339 FREE(calloutp, M_KQUEUE); 340 kq_ncallouts--; 341 } 342 343 static int 344 filt_timer(struct knote *kn, long hint) 345 { 346 347 return (kn->kn_data != 0); 348 } 349 350 /* 351 * MPSAFE 352 */ 353 int 354 kqueue(struct thread *td, struct kqueue_args *uap) 355 { 356 struct filedesc *fdp; 357 struct kqueue *kq; 358 struct file *fp; 359 int fd, error; 360 361 mtx_lock(&Giant); 362 fdp = td->td_proc->p_fd; 363 error = falloc(td, &fp, &fd); 364 if (error) 365 goto done2; 366 fp->f_flag = FREAD | FWRITE; 367 fp->f_type = DTYPE_KQUEUE; 368 fp->f_ops = &kqueueops; 369 kq = malloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 370 TAILQ_INIT(&kq->kq_head); 371 fp->f_data = (caddr_t)kq; 372 td->td_retval[0] = fd; 373 if (fdp->fd_knlistsize < 0) 374 fdp->fd_knlistsize = 0; /* this process has a kq */ 375 kq->kq_fdp = fdp; 376 done2: 377 mtx_unlock(&Giant); 378 return (error); 379 } 380 381 #ifndef _SYS_SYSPROTO_H_ 382 struct kevent_args { 383 int fd; 384 const struct kevent *changelist; 385 int nchanges; 386 struct kevent *eventlist; 387 int nevents; 388 const struct timespec *timeout; 389 }; 390 #endif 391 /* 392 * MPSAFE 393 */ 394 int 395 kevent(struct thread *td, struct kevent_args *uap) 396 { 397 struct kevent *kevp; 398 struct kqueue *kq; 399 struct file *fp; 400 struct timespec ts; 401 int i, n, nerrors, error; 402 403 mtx_lock(&Giant); 404 if ((error = fget(td, uap->fd, &fp)) != 0) 405 goto done; 406 if (fp->f_type != DTYPE_KQUEUE) { 407 error = EBADF; 408 goto done; 409 } 410 if (uap->timeout != NULL) { 411 error = copyin(uap->timeout, &ts, sizeof(ts)); 412 if (error) 413 goto done; 414 uap->timeout = &ts; 415 } 416 417 kq = (struct kqueue *)fp->f_data; 418 nerrors = 0; 419 420 while (uap->nchanges > 0) { 421 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges; 422 error = copyin(uap->changelist, kq->kq_kev, 423 n * sizeof(struct kevent)); 424 if (error) 425 goto done; 426 for (i = 0; i < n; i++) { 427 kevp = &kq->kq_kev[i]; 428 kevp->flags &= ~EV_SYSFLAGS; 429 error = kqueue_register(kq, kevp, td); 430 if (error) { 431 if (uap->nevents != 0) { 432 kevp->flags = EV_ERROR; 433 kevp->data = error; 434 (void) copyout((caddr_t)kevp, 435 (caddr_t)uap->eventlist, 436 sizeof(*kevp)); 437 uap->eventlist++; 438 uap->nevents--; 439 nerrors++; 440 } else { 441 goto done; 442 } 443 } 444 } 445 uap->nchanges -= n; 446 uap->changelist += n; 447 } 448 if (nerrors) { 449 td->td_retval[0] = nerrors; 450 error = 0; 451 goto done; 452 } 453 454 error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, td); 455 done: 456 if (fp != NULL) 457 fdrop(fp, td); 458 mtx_unlock(&Giant); 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 s, 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_ofiles[kev->ident]) == 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 s = splhigh(); 561 if (kn->kn_fop->f_event(kn, 0)) 562 KNOTE_ACTIVATE(kn); 563 splx(s); 564 565 } else if (kev->flags & EV_DELETE) { 566 kn->kn_fop->f_detach(kn); 567 knote_drop(kn, td); 568 goto done; 569 } 570 571 if ((kev->flags & EV_DISABLE) && 572 ((kn->kn_status & KN_DISABLED) == 0)) { 573 s = splhigh(); 574 kn->kn_status |= KN_DISABLED; 575 splx(s); 576 } 577 578 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 579 s = splhigh(); 580 kn->kn_status &= ~KN_DISABLED; 581 if ((kn->kn_status & KN_ACTIVE) && 582 ((kn->kn_status & KN_QUEUED) == 0)) 583 knote_enqueue(kn); 584 splx(s); 585 } 586 587 done: 588 if (fp != NULL) 589 fdrop(fp, td); 590 return (error); 591 } 592 593 static int 594 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp, 595 const struct timespec *tsp, struct thread *td) 596 { 597 struct kqueue *kq = (struct kqueue *)fp->f_data; 598 struct kevent *kevp; 599 struct timeval atv, rtv, ttv; 600 struct knote *kn, marker; 601 int s, count, timeout, nkev = 0, error = 0; 602 603 count = maxevents; 604 if (count == 0) 605 goto done; 606 607 if (tsp != NULL) { 608 TIMESPEC_TO_TIMEVAL(&atv, tsp); 609 if (itimerfix(&atv)) { 610 error = EINVAL; 611 goto done; 612 } 613 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 614 timeout = -1; 615 else 616 timeout = atv.tv_sec > 24 * 60 * 60 ? 617 24 * 60 * 60 * hz : tvtohz(&atv); 618 getmicrouptime(&rtv); 619 timevaladd(&atv, &rtv); 620 } else { 621 atv.tv_sec = 0; 622 atv.tv_usec = 0; 623 timeout = 0; 624 } 625 goto start; 626 627 retry: 628 if (atv.tv_sec || atv.tv_usec) { 629 getmicrouptime(&rtv); 630 if (timevalcmp(&rtv, &atv, >=)) 631 goto done; 632 ttv = atv; 633 timevalsub(&ttv, &rtv); 634 timeout = ttv.tv_sec > 24 * 60 * 60 ? 635 24 * 60 * 60 * hz : tvtohz(&ttv); 636 } 637 638 start: 639 kevp = kq->kq_kev; 640 s = splhigh(); 641 if (kq->kq_count == 0) { 642 if (timeout < 0) { 643 error = EWOULDBLOCK; 644 } else { 645 kq->kq_state |= KQ_SLEEP; 646 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 647 } 648 splx(s); 649 if (error == 0) 650 goto retry; 651 /* don't restart after signals... */ 652 if (error == ERESTART) 653 error = EINTR; 654 else if (error == EWOULDBLOCK) 655 error = 0; 656 goto done; 657 } 658 659 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 660 while (count) { 661 kn = TAILQ_FIRST(&kq->kq_head); 662 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 663 if (kn == &marker) { 664 splx(s); 665 if (count == maxevents) 666 goto retry; 667 goto done; 668 } 669 if (kn->kn_status & KN_DISABLED) { 670 kn->kn_status &= ~KN_QUEUED; 671 kq->kq_count--; 672 continue; 673 } 674 if ((kn->kn_flags & EV_ONESHOT) == 0 && 675 kn->kn_fop->f_event(kn, 0) == 0) { 676 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 677 kq->kq_count--; 678 continue; 679 } 680 *kevp = kn->kn_kevent; 681 kevp++; 682 nkev++; 683 if (kn->kn_flags & EV_ONESHOT) { 684 kn->kn_status &= ~KN_QUEUED; 685 kq->kq_count--; 686 splx(s); 687 kn->kn_fop->f_detach(kn); 688 knote_drop(kn, td); 689 s = splhigh(); 690 } else if (kn->kn_flags & EV_CLEAR) { 691 kn->kn_data = 0; 692 kn->kn_fflags = 0; 693 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 694 kq->kq_count--; 695 } else { 696 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 697 } 698 count--; 699 if (nkev == KQ_NEVENTS) { 700 splx(s); 701 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 702 sizeof(struct kevent) * nkev); 703 ulistp += nkev; 704 nkev = 0; 705 kevp = kq->kq_kev; 706 s = splhigh(); 707 if (error) 708 break; 709 } 710 } 711 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 712 splx(s); 713 done: 714 if (nkev != 0) 715 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 716 sizeof(struct kevent) * nkev); 717 td->td_retval[0] = maxevents - count; 718 return (error); 719 } 720 721 /* 722 * XXX 723 * This could be expanded to call kqueue_scan, if desired. 724 */ 725 /*ARGSUSED*/ 726 static int 727 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, 728 int flags, struct thread *td) 729 { 730 return (ENXIO); 731 } 732 733 /*ARGSUSED*/ 734 static int 735 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, 736 int flags, struct thread *td) 737 { 738 return (ENXIO); 739 } 740 741 /*ARGSUSED*/ 742 static int 743 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td) 744 { 745 return (ENOTTY); 746 } 747 748 /*ARGSUSED*/ 749 static int 750 kqueue_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 751 { 752 struct kqueue *kq = (struct kqueue *)fp->f_data; 753 int revents = 0; 754 int s = splnet(); 755 756 if (events & (POLLIN | POLLRDNORM)) { 757 if (kq->kq_count) { 758 revents |= events & (POLLIN | POLLRDNORM); 759 } else { 760 selrecord(td, &kq->kq_sel); 761 kq->kq_state |= KQ_SEL; 762 } 763 } 764 splx(s); 765 return (revents); 766 } 767 768 /*ARGSUSED*/ 769 static int 770 kqueue_stat(struct file *fp, struct stat *st, struct thread *td) 771 { 772 struct kqueue *kq = (struct kqueue *)fp->f_data; 773 774 bzero((void *)st, sizeof(*st)); 775 st->st_size = kq->kq_count; 776 st->st_blksize = sizeof(struct kevent); 777 st->st_mode = S_IFIFO; 778 return (0); 779 } 780 781 /*ARGSUSED*/ 782 static int 783 kqueue_close(struct file *fp, struct thread *td) 784 { 785 struct kqueue *kq = (struct kqueue *)fp->f_data; 786 struct filedesc *fdp = td->td_proc->p_fd; 787 struct knote **knp, *kn, *kn0; 788 int i; 789 790 for (i = 0; i < fdp->fd_knlistsize; i++) { 791 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 792 kn = *knp; 793 while (kn != NULL) { 794 kn0 = SLIST_NEXT(kn, kn_link); 795 if (kq == kn->kn_kq) { 796 kn->kn_fop->f_detach(kn); 797 fdrop(kn->kn_fp, td); 798 knote_free(kn); 799 *knp = kn0; 800 } else { 801 knp = &SLIST_NEXT(kn, kn_link); 802 } 803 kn = kn0; 804 } 805 } 806 if (fdp->fd_knhashmask != 0) { 807 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 808 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 809 kn = *knp; 810 while (kn != NULL) { 811 kn0 = SLIST_NEXT(kn, kn_link); 812 if (kq == kn->kn_kq) { 813 kn->kn_fop->f_detach(kn); 814 /* XXX non-fd release of kn->kn_ptr */ 815 knote_free(kn); 816 *knp = kn0; 817 } else { 818 knp = &SLIST_NEXT(kn, kn_link); 819 } 820 kn = kn0; 821 } 822 } 823 } 824 free(kq, M_KQUEUE); 825 fp->f_data = NULL; 826 827 return (0); 828 } 829 830 static void 831 kqueue_wakeup(struct kqueue *kq) 832 { 833 834 if (kq->kq_state & KQ_SLEEP) { 835 kq->kq_state &= ~KQ_SLEEP; 836 wakeup(kq); 837 } 838 if (kq->kq_state & KQ_SEL) { 839 kq->kq_state &= ~KQ_SEL; 840 selwakeup(&kq->kq_sel); 841 } 842 KNOTE(&kq->kq_sel.si_note, 0); 843 } 844 845 /* 846 * walk down a list of knotes, activating them if their event has triggered. 847 */ 848 void 849 knote(struct klist *list, long hint) 850 { 851 struct knote *kn; 852 853 SLIST_FOREACH(kn, list, kn_selnext) 854 if (kn->kn_fop->f_event(kn, hint)) 855 KNOTE_ACTIVATE(kn); 856 } 857 858 /* 859 * remove all knotes from a specified klist 860 */ 861 void 862 knote_remove(struct thread *td, struct klist *list) 863 { 864 struct knote *kn; 865 866 while ((kn = SLIST_FIRST(list)) != NULL) { 867 kn->kn_fop->f_detach(kn); 868 knote_drop(kn, td); 869 } 870 } 871 872 /* 873 * remove all knotes referencing a specified fd 874 */ 875 void 876 knote_fdclose(struct thread *td, int fd) 877 { 878 struct filedesc *fdp = td->td_proc->p_fd; 879 struct klist *list = &fdp->fd_knlist[fd]; 880 881 knote_remove(td, list); 882 } 883 884 static void 885 knote_attach(struct knote *kn, struct filedesc *fdp) 886 { 887 struct klist *list; 888 int size; 889 890 if (! kn->kn_fop->f_isfd) { 891 if (fdp->fd_knhashmask == 0) 892 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 893 &fdp->fd_knhashmask); 894 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 895 goto done; 896 } 897 898 if (fdp->fd_knlistsize <= kn->kn_id) { 899 size = fdp->fd_knlistsize; 900 while (size <= kn->kn_id) 901 size += KQEXTENT; 902 MALLOC(list, struct klist *, 903 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 904 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, 905 fdp->fd_knlistsize * sizeof(struct klist *)); 906 bzero((caddr_t)list + 907 fdp->fd_knlistsize * sizeof(struct klist *), 908 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 909 if (fdp->fd_knlist != NULL) 910 FREE(fdp->fd_knlist, M_KQUEUE); 911 fdp->fd_knlistsize = size; 912 fdp->fd_knlist = list; 913 } 914 list = &fdp->fd_knlist[kn->kn_id]; 915 done: 916 SLIST_INSERT_HEAD(list, kn, kn_link); 917 kn->kn_status = 0; 918 } 919 920 /* 921 * should be called at spl == 0, since we don't want to hold spl 922 * while calling fdrop and free. 923 */ 924 static void 925 knote_drop(struct knote *kn, struct thread *td) 926 { 927 struct filedesc *fdp = td->td_proc->p_fd; 928 struct klist *list; 929 930 if (kn->kn_fop->f_isfd) 931 list = &fdp->fd_knlist[kn->kn_id]; 932 else 933 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 934 935 SLIST_REMOVE(list, kn, knote, kn_link); 936 if (kn->kn_status & KN_QUEUED) 937 knote_dequeue(kn); 938 if (kn->kn_fop->f_isfd) 939 fdrop(kn->kn_fp, td); 940 knote_free(kn); 941 } 942 943 944 static void 945 knote_enqueue(struct knote *kn) 946 { 947 struct kqueue *kq = kn->kn_kq; 948 int s = splhigh(); 949 950 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 951 952 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 953 kn->kn_status |= KN_QUEUED; 954 kq->kq_count++; 955 splx(s); 956 kqueue_wakeup(kq); 957 } 958 959 static void 960 knote_dequeue(struct knote *kn) 961 { 962 struct kqueue *kq = kn->kn_kq; 963 int s = splhigh(); 964 965 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 966 967 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 968 kn->kn_status &= ~KN_QUEUED; 969 kq->kq_count--; 970 splx(s); 971 } 972 973 static void 974 knote_init(void) 975 { 976 knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1); 977 } 978 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 979 980 static struct knote * 981 knote_alloc(void) 982 { 983 return ((struct knote *)zalloc(knote_zone)); 984 } 985 986 static void 987 knote_free(struct knote *kn) 988 { 989 zfree(knote_zone, kn); 990 } 991