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/uma.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 *active_cred, int flags, struct thread *td); 61 static int kqueue_write(struct file *fp, struct uio *uio, 62 struct ucred *active_cred, int flags, struct thread *td); 63 static int kqueue_ioctl(struct file *fp, u_long com, void *data, 64 struct ucred *active_cred, struct thread *td); 65 static int kqueue_poll(struct file *fp, int events, 66 struct ucred *active_cred, 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, 69 struct ucred *active_cred, 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 kqueue_read, 75 kqueue_write, 76 kqueue_ioctl, 77 kqueue_poll, 78 kqueue_kqfilter, 79 kqueue_stat, 80 kqueue_close 81 }; 82 83 static void knote_attach(struct knote *kn, struct filedesc *fdp); 84 static void knote_drop(struct knote *kn, struct thread *td); 85 static void knote_enqueue(struct knote *kn); 86 static void knote_dequeue(struct knote *kn); 87 static void knote_init(void); 88 static struct knote *knote_alloc(void); 89 static void knote_free(struct knote *kn); 90 91 static void filt_kqdetach(struct knote *kn); 92 static int filt_kqueue(struct knote *kn, long hint); 93 static int filt_procattach(struct knote *kn); 94 static void filt_procdetach(struct knote *kn); 95 static int filt_proc(struct knote *kn, long hint); 96 static int filt_fileattach(struct knote *kn); 97 static void filt_timerexpire(void *knx); 98 static int filt_timerattach(struct knote *kn); 99 static void filt_timerdetach(struct knote *kn); 100 static int filt_timer(struct knote *kn, long hint); 101 102 static struct filterops file_filtops = 103 { 1, filt_fileattach, NULL, NULL }; 104 static struct filterops kqread_filtops = 105 { 1, NULL, filt_kqdetach, filt_kqueue }; 106 static struct filterops proc_filtops = 107 { 0, filt_procattach, filt_procdetach, filt_proc }; 108 static struct filterops timer_filtops = 109 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 110 111 static uma_zone_t knote_zone; 112 static int kq_ncallouts = 0; 113 static int kq_calloutmax = (4 * 1024); 114 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 115 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 116 117 #define KNOTE_ACTIVATE(kn) do { \ 118 kn->kn_status |= KN_ACTIVE; \ 119 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 120 knote_enqueue(kn); \ 121 } while(0) 122 123 #define KN_HASHSIZE 64 /* XXX should be tunable */ 124 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 125 126 static int 127 filt_nullattach(struct knote *kn) 128 { 129 130 return (ENXIO); 131 }; 132 133 struct filterops null_filtops = 134 { 0, filt_nullattach, NULL, NULL }; 135 136 extern struct filterops sig_filtops; 137 138 /* 139 * Table for for all system-defined filters. 140 */ 141 static struct filterops *sysfilt_ops[] = { 142 &file_filtops, /* EVFILT_READ */ 143 &file_filtops, /* EVFILT_WRITE */ 144 &null_filtops, /* EVFILT_AIO */ 145 &file_filtops, /* EVFILT_VNODE */ 146 &proc_filtops, /* EVFILT_PROC */ 147 &sig_filtops, /* EVFILT_SIGNAL */ 148 &timer_filtops, /* EVFILT_TIMER */ 149 &file_filtops, /* EVFILT_NETDEV */ 150 }; 151 152 static int 153 filt_fileattach(struct knote *kn) 154 { 155 156 return (fo_kqfilter(kn->kn_fp, kn)); 157 } 158 159 /*ARGSUSED*/ 160 static int 161 kqueue_kqfilter(struct file *fp, struct knote *kn) 162 { 163 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 164 165 if (kn->kn_filter != EVFILT_READ) 166 return (1); 167 168 kn->kn_fop = &kqread_filtops; 169 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 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 error; 196 197 p = pfind(kn->kn_id); 198 if (p == NULL) 199 return (ESRCH); 200 if ((error = p_cansee(curthread, p))) { 201 PROC_UNLOCK(p); 202 return (error); 203 } 204 205 kn->kn_ptr.p_proc = p; 206 kn->kn_flags |= EV_CLEAR; /* automatically set */ 207 208 /* 209 * internal flag indicating registration done by kernel 210 */ 211 if (kn->kn_flags & EV_FLAG1) { 212 kn->kn_data = kn->kn_sdata; /* ppid */ 213 kn->kn_fflags = NOTE_CHILD; 214 kn->kn_flags &= ~EV_FLAG1; 215 } 216 217 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 218 PROC_UNLOCK(p); 219 220 return (0); 221 } 222 223 /* 224 * The knote may be attached to a different process, which may exit, 225 * leaving nothing for the knote to be attached to. So when the process 226 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 227 * it will be deleted when read out. However, as part of the knote deletion, 228 * this routine is called, so a check is needed to avoid actually performing 229 * a detach, because the original process does not exist any more. 230 */ 231 static void 232 filt_procdetach(struct knote *kn) 233 { 234 struct proc *p = kn->kn_ptr.p_proc; 235 236 if (kn->kn_status & KN_DETACHED) 237 return; 238 239 PROC_LOCK(p); 240 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 241 PROC_UNLOCK(p); 242 } 243 244 static int 245 filt_proc(struct knote *kn, long hint) 246 { 247 u_int event; 248 249 /* 250 * mask off extra data 251 */ 252 event = (u_int)hint & NOTE_PCTRLMASK; 253 254 /* 255 * if the user is interested in this event, record it. 256 */ 257 if (kn->kn_sfflags & event) 258 kn->kn_fflags |= event; 259 260 /* 261 * process is gone, so flag the event as finished. 262 */ 263 if (event == NOTE_EXIT) { 264 kn->kn_status |= KN_DETACHED; 265 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 266 return (1); 267 } 268 269 /* 270 * process forked, and user wants to track the new process, 271 * so attach a new knote to it, and immediately report an 272 * event with the parent's pid. 273 */ 274 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 275 struct kevent kev; 276 int error; 277 278 /* 279 * register knote with new process. 280 */ 281 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 282 kev.filter = kn->kn_filter; 283 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 284 kev.fflags = kn->kn_sfflags; 285 kev.data = kn->kn_id; /* parent */ 286 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 287 error = kqueue_register(kn->kn_kq, &kev, NULL); 288 if (error) 289 kn->kn_fflags |= NOTE_TRACKERR; 290 } 291 292 return (kn->kn_fflags != 0); 293 } 294 295 static void 296 filt_timerexpire(void *knx) 297 { 298 struct knote *kn = knx; 299 struct callout *calloutp; 300 struct timeval tv; 301 int tticks; 302 303 kn->kn_data++; 304 KNOTE_ACTIVATE(kn); 305 306 if ((kn->kn_flags & EV_ONESHOT) == 0) { 307 tv.tv_sec = kn->kn_sdata / 1000; 308 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 309 tticks = tvtohz(&tv); 310 calloutp = (struct callout *)kn->kn_hook; 311 callout_reset(calloutp, tticks, filt_timerexpire, kn); 312 } 313 } 314 315 /* 316 * data contains amount of time to sleep, in milliseconds 317 */ 318 static int 319 filt_timerattach(struct knote *kn) 320 { 321 struct callout *calloutp; 322 struct timeval tv; 323 int tticks; 324 325 if (kq_ncallouts >= kq_calloutmax) 326 return (ENOMEM); 327 kq_ncallouts++; 328 329 tv.tv_sec = kn->kn_sdata / 1000; 330 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 331 tticks = tvtohz(&tv); 332 333 kn->kn_flags |= EV_CLEAR; /* automatically set */ 334 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 335 M_KQUEUE, M_WAITOK); 336 callout_init(calloutp, 0); 337 callout_reset(calloutp, tticks, filt_timerexpire, kn); 338 kn->kn_hook = calloutp; 339 340 return (0); 341 } 342 343 static void 344 filt_timerdetach(struct knote *kn) 345 { 346 struct callout *calloutp; 347 348 calloutp = (struct callout *)kn->kn_hook; 349 callout_stop(calloutp); 350 FREE(calloutp, M_KQUEUE); 351 kq_ncallouts--; 352 } 353 354 static int 355 filt_timer(struct knote *kn, long hint) 356 { 357 358 return (kn->kn_data != 0); 359 } 360 361 /* 362 * MPSAFE 363 */ 364 int 365 kqueue(struct thread *td, struct kqueue_args *uap) 366 { 367 struct filedesc *fdp; 368 struct kqueue *kq; 369 struct file *fp; 370 int fd, error; 371 372 mtx_lock(&Giant); 373 fdp = td->td_proc->p_fd; 374 error = falloc(td, &fp, &fd); 375 if (error) 376 goto done2; 377 kq = malloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 378 TAILQ_INIT(&kq->kq_head); 379 FILE_LOCK(fp); 380 fp->f_flag = FREAD | FWRITE; 381 fp->f_type = DTYPE_KQUEUE; 382 fp->f_ops = &kqueueops; 383 TAILQ_INIT(&kq->kq_head); 384 fp->f_data = kq; 385 FILE_UNLOCK(fp); 386 FILEDESC_LOCK(fdp); 387 td->td_retval[0] = fd; 388 if (fdp->fd_knlistsize < 0) 389 fdp->fd_knlistsize = 0; /* this process has a kq */ 390 FILEDESC_UNLOCK(fdp); 391 kq->kq_fdp = fdp; 392 done2: 393 mtx_unlock(&Giant); 394 return (error); 395 } 396 397 #ifndef _SYS_SYSPROTO_H_ 398 struct kevent_args { 399 int fd; 400 const struct kevent *changelist; 401 int nchanges; 402 struct kevent *eventlist; 403 int nevents; 404 const struct timespec *timeout; 405 }; 406 #endif 407 /* 408 * MPSAFE 409 */ 410 int 411 kevent(struct thread *td, struct kevent_args *uap) 412 { 413 struct kevent *kevp; 414 struct kqueue *kq; 415 struct file *fp; 416 struct timespec ts; 417 int i, n, nerrors, error; 418 419 if ((error = fget(td, uap->fd, &fp)) != 0) 420 return (error); 421 if (fp->f_type != DTYPE_KQUEUE) { 422 fdrop(fp, td); 423 return (EBADF); 424 } 425 if (uap->timeout != NULL) { 426 error = copyin(uap->timeout, &ts, sizeof(ts)); 427 if (error) 428 goto done_nogiant; 429 uap->timeout = &ts; 430 } 431 mtx_lock(&Giant); 432 433 kq = (struct kqueue *)fp->f_data; 434 nerrors = 0; 435 436 while (uap->nchanges > 0) { 437 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges; 438 error = copyin(uap->changelist, kq->kq_kev, 439 n * sizeof(struct kevent)); 440 if (error) 441 goto done; 442 for (i = 0; i < n; i++) { 443 kevp = &kq->kq_kev[i]; 444 kevp->flags &= ~EV_SYSFLAGS; 445 error = kqueue_register(kq, kevp, td); 446 if (error) { 447 if (uap->nevents != 0) { 448 kevp->flags = EV_ERROR; 449 kevp->data = error; 450 (void) copyout(kevp, 451 uap->eventlist, 452 sizeof(*kevp)); 453 uap->eventlist++; 454 uap->nevents--; 455 nerrors++; 456 } else { 457 goto done; 458 } 459 } 460 } 461 uap->nchanges -= n; 462 uap->changelist += n; 463 } 464 if (nerrors) { 465 td->td_retval[0] = nerrors; 466 error = 0; 467 goto done; 468 } 469 470 error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, td); 471 done: 472 mtx_unlock(&Giant); 473 done_nogiant: 474 if (fp != NULL) 475 fdrop(fp, td); 476 return (error); 477 } 478 479 int 480 kqueue_add_filteropts(int filt, struct filterops *filtops) 481 { 482 483 if (filt > 0) 484 panic("filt(%d) > 0", filt); 485 if (filt + EVFILT_SYSCOUNT < 0) 486 panic("filt(%d) + EVFILT_SYSCOUNT(%d) == %d < 0", 487 filt, EVFILT_SYSCOUNT, filt + EVFILT_SYSCOUNT); 488 if (sysfilt_ops[~filt] != &null_filtops) 489 panic("sysfilt_ops[~filt(%d)] != &null_filtops", filt); 490 sysfilt_ops[~filt] = filtops; 491 return (0); 492 } 493 494 int 495 kqueue_del_filteropts(int filt) 496 { 497 498 if (filt > 0) 499 panic("filt(%d) > 0", filt); 500 if (filt + EVFILT_SYSCOUNT < 0) 501 panic("filt(%d) + EVFILT_SYSCOUNT(%d) == %d < 0", 502 filt, EVFILT_SYSCOUNT, filt + EVFILT_SYSCOUNT); 503 if (sysfilt_ops[~filt] == &null_filtops) 504 panic("sysfilt_ops[~filt(%d)] != &null_filtops", filt); 505 sysfilt_ops[~filt] = &null_filtops; 506 return (0); 507 } 508 509 int 510 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td) 511 { 512 struct filedesc *fdp = kq->kq_fdp; 513 struct filterops *fops; 514 struct file *fp = NULL; 515 struct knote *kn = NULL; 516 int s, error = 0; 517 518 if (kev->filter < 0) { 519 if (kev->filter + EVFILT_SYSCOUNT < 0) 520 return (EINVAL); 521 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 522 } else { 523 /* 524 * XXX 525 * filter attach routine is responsible for insuring that 526 * the identifier can be attached to it. 527 */ 528 printf("unknown filter: %d\n", kev->filter); 529 return (EINVAL); 530 } 531 532 FILEDESC_LOCK(fdp); 533 if (fops->f_isfd) { 534 /* validate descriptor */ 535 if ((u_int)kev->ident >= fdp->fd_nfiles || 536 (fp = fdp->fd_ofiles[kev->ident]) == NULL) { 537 FILEDESC_UNLOCK(fdp); 538 return (EBADF); 539 } 540 fhold(fp); 541 542 if (kev->ident < fdp->fd_knlistsize) { 543 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 544 if (kq == kn->kn_kq && 545 kev->filter == kn->kn_filter) 546 break; 547 } 548 } else { 549 if (fdp->fd_knhashmask != 0) { 550 struct klist *list; 551 552 list = &fdp->fd_knhash[ 553 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 554 SLIST_FOREACH(kn, list, kn_link) 555 if (kev->ident == kn->kn_id && 556 kq == kn->kn_kq && 557 kev->filter == kn->kn_filter) 558 break; 559 } 560 } 561 FILEDESC_UNLOCK(fdp); 562 563 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 564 error = ENOENT; 565 goto done; 566 } 567 568 /* 569 * kn now contains the matching knote, or NULL if no match 570 */ 571 if (kev->flags & EV_ADD) { 572 573 if (kn == NULL) { 574 kn = knote_alloc(); 575 if (kn == NULL) { 576 error = ENOMEM; 577 goto done; 578 } 579 kn->kn_fp = fp; 580 kn->kn_kq = kq; 581 kn->kn_fop = fops; 582 583 /* 584 * apply reference count to knote structure, and 585 * do not release it at the end of this routine. 586 */ 587 fp = NULL; 588 589 kn->kn_sfflags = kev->fflags; 590 kn->kn_sdata = kev->data; 591 kev->fflags = 0; 592 kev->data = 0; 593 kn->kn_kevent = *kev; 594 595 knote_attach(kn, fdp); 596 if ((error = fops->f_attach(kn)) != 0) { 597 knote_drop(kn, td); 598 goto done; 599 } 600 } else { 601 /* 602 * The user may change some filter values after the 603 * initial EV_ADD, but doing so will not reset any 604 * filter which have already been triggered. 605 */ 606 kn->kn_sfflags = kev->fflags; 607 kn->kn_sdata = kev->data; 608 kn->kn_kevent.udata = kev->udata; 609 } 610 611 s = splhigh(); 612 if (kn->kn_fop->f_event(kn, 0)) 613 KNOTE_ACTIVATE(kn); 614 splx(s); 615 616 } else if (kev->flags & EV_DELETE) { 617 kn->kn_fop->f_detach(kn); 618 knote_drop(kn, td); 619 goto done; 620 } 621 622 if ((kev->flags & EV_DISABLE) && 623 ((kn->kn_status & KN_DISABLED) == 0)) { 624 s = splhigh(); 625 kn->kn_status |= KN_DISABLED; 626 splx(s); 627 } 628 629 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 630 s = splhigh(); 631 kn->kn_status &= ~KN_DISABLED; 632 if ((kn->kn_status & KN_ACTIVE) && 633 ((kn->kn_status & KN_QUEUED) == 0)) 634 knote_enqueue(kn); 635 splx(s); 636 } 637 638 done: 639 if (fp != NULL) 640 fdrop(fp, td); 641 return (error); 642 } 643 644 static int 645 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp, 646 const struct timespec *tsp, struct thread *td) 647 { 648 struct kqueue *kq; 649 struct kevent *kevp; 650 struct timeval atv, rtv, ttv; 651 struct knote *kn, marker; 652 int s, count, timeout, nkev = 0, error = 0; 653 654 FILE_LOCK_ASSERT(fp, MA_NOTOWNED); 655 656 kq = (struct kqueue *)fp->f_data; 657 count = maxevents; 658 if (count == 0) 659 goto done; 660 661 if (tsp != NULL) { 662 TIMESPEC_TO_TIMEVAL(&atv, tsp); 663 if (itimerfix(&atv)) { 664 error = EINVAL; 665 goto done; 666 } 667 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 668 timeout = -1; 669 else 670 timeout = atv.tv_sec > 24 * 60 * 60 ? 671 24 * 60 * 60 * hz : tvtohz(&atv); 672 getmicrouptime(&rtv); 673 timevaladd(&atv, &rtv); 674 } else { 675 atv.tv_sec = 0; 676 atv.tv_usec = 0; 677 timeout = 0; 678 } 679 goto start; 680 681 retry: 682 if (atv.tv_sec || atv.tv_usec) { 683 getmicrouptime(&rtv); 684 if (timevalcmp(&rtv, &atv, >=)) 685 goto done; 686 ttv = atv; 687 timevalsub(&ttv, &rtv); 688 timeout = ttv.tv_sec > 24 * 60 * 60 ? 689 24 * 60 * 60 * hz : tvtohz(&ttv); 690 } 691 692 start: 693 kevp = kq->kq_kev; 694 s = splhigh(); 695 if (kq->kq_count == 0) { 696 if (timeout < 0) { 697 error = EWOULDBLOCK; 698 } else { 699 kq->kq_state |= KQ_SLEEP; 700 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 701 } 702 splx(s); 703 if (error == 0) 704 goto retry; 705 /* don't restart after signals... */ 706 if (error == ERESTART) 707 error = EINTR; 708 else if (error == EWOULDBLOCK) 709 error = 0; 710 goto done; 711 } 712 713 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 714 while (count) { 715 kn = TAILQ_FIRST(&kq->kq_head); 716 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 717 if (kn == &marker) { 718 splx(s); 719 if (count == maxevents) 720 goto retry; 721 goto done; 722 } 723 if (kn->kn_status & KN_DISABLED) { 724 kn->kn_status &= ~KN_QUEUED; 725 kq->kq_count--; 726 continue; 727 } 728 if ((kn->kn_flags & EV_ONESHOT) == 0 && 729 kn->kn_fop->f_event(kn, 0) == 0) { 730 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 731 kq->kq_count--; 732 continue; 733 } 734 *kevp = kn->kn_kevent; 735 kevp++; 736 nkev++; 737 if (kn->kn_flags & EV_ONESHOT) { 738 kn->kn_status &= ~KN_QUEUED; 739 kq->kq_count--; 740 splx(s); 741 kn->kn_fop->f_detach(kn); 742 knote_drop(kn, td); 743 s = splhigh(); 744 } else if (kn->kn_flags & EV_CLEAR) { 745 kn->kn_data = 0; 746 kn->kn_fflags = 0; 747 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 748 kq->kq_count--; 749 } else { 750 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 751 } 752 count--; 753 if (nkev == KQ_NEVENTS) { 754 splx(s); 755 error = copyout(&kq->kq_kev, ulistp, 756 sizeof(struct kevent) * nkev); 757 ulistp += nkev; 758 nkev = 0; 759 kevp = kq->kq_kev; 760 s = splhigh(); 761 if (error) 762 break; 763 } 764 } 765 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 766 splx(s); 767 done: 768 if (nkev != 0) 769 error = copyout(&kq->kq_kev, ulistp, 770 sizeof(struct kevent) * nkev); 771 td->td_retval[0] = maxevents - count; 772 return (error); 773 } 774 775 /* 776 * XXX 777 * This could be expanded to call kqueue_scan, if desired. 778 */ 779 /*ARGSUSED*/ 780 static int 781 kqueue_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 782 int flags, struct thread *td) 783 { 784 return (ENXIO); 785 } 786 787 /*ARGSUSED*/ 788 static int 789 kqueue_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 790 int flags, struct thread *td) 791 { 792 return (ENXIO); 793 } 794 795 /*ARGSUSED*/ 796 static int 797 kqueue_ioctl(struct file *fp, u_long com, void *data, 798 struct ucred *active_cred, struct thread *td) 799 { 800 return (ENOTTY); 801 } 802 803 /*ARGSUSED*/ 804 static int 805 kqueue_poll(struct file *fp, int events, struct ucred *active_cred, 806 struct thread *td) 807 { 808 struct kqueue *kq; 809 int revents = 0; 810 int s = splnet(); 811 812 kq = (struct kqueue *)fp->f_data; 813 if (events & (POLLIN | POLLRDNORM)) { 814 if (kq->kq_count) { 815 revents |= events & (POLLIN | POLLRDNORM); 816 } else { 817 selrecord(td, &kq->kq_sel); 818 kq->kq_state |= KQ_SEL; 819 } 820 } 821 splx(s); 822 return (revents); 823 } 824 825 /*ARGSUSED*/ 826 static int 827 kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 828 struct thread *td) 829 { 830 struct kqueue *kq; 831 832 kq = (struct kqueue *)fp->f_data; 833 bzero((void *)st, sizeof(*st)); 834 st->st_size = kq->kq_count; 835 st->st_blksize = sizeof(struct kevent); 836 st->st_mode = S_IFIFO; 837 return (0); 838 } 839 840 /*ARGSUSED*/ 841 static int 842 kqueue_close(struct file *fp, struct thread *td) 843 { 844 struct kqueue *kq = (struct kqueue *)fp->f_data; 845 struct filedesc *fdp = td->td_proc->p_fd; 846 struct knote **knp, *kn, *kn0; 847 int i; 848 849 FILEDESC_LOCK(fdp); 850 for (i = 0; i < fdp->fd_knlistsize; i++) { 851 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 852 kn = *knp; 853 while (kn != NULL) { 854 kn0 = SLIST_NEXT(kn, kn_link); 855 if (kq == kn->kn_kq) { 856 kn->kn_fop->f_detach(kn); 857 *knp = kn0; 858 FILE_LOCK(kn->kn_fp); 859 FILEDESC_UNLOCK(fdp); 860 fdrop_locked(kn->kn_fp, td); 861 knote_free(kn); 862 FILEDESC_LOCK(fdp); 863 } else { 864 knp = &SLIST_NEXT(kn, kn_link); 865 } 866 kn = kn0; 867 } 868 } 869 if (fdp->fd_knhashmask != 0) { 870 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 871 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 872 kn = *knp; 873 while (kn != NULL) { 874 kn0 = SLIST_NEXT(kn, kn_link); 875 if (kq == kn->kn_kq) { 876 kn->kn_fop->f_detach(kn); 877 *knp = kn0; 878 /* XXX non-fd release of kn->kn_ptr */ 879 FILEDESC_UNLOCK(fdp); 880 knote_free(kn); 881 FILEDESC_LOCK(fdp); 882 } else { 883 knp = &SLIST_NEXT(kn, kn_link); 884 } 885 kn = kn0; 886 } 887 } 888 } 889 FILEDESC_UNLOCK(fdp); 890 free(kq, M_KQUEUE); 891 fp->f_data = NULL; 892 893 return (0); 894 } 895 896 static void 897 kqueue_wakeup(struct kqueue *kq) 898 { 899 900 if (kq->kq_state & KQ_SLEEP) { 901 kq->kq_state &= ~KQ_SLEEP; 902 wakeup(kq); 903 } 904 if (kq->kq_state & KQ_SEL) { 905 kq->kq_state &= ~KQ_SEL; 906 selwakeup(&kq->kq_sel); 907 } 908 KNOTE(&kq->kq_sel.si_note, 0); 909 } 910 911 /* 912 * walk down a list of knotes, activating them if their event has triggered. 913 */ 914 void 915 knote(struct klist *list, long hint) 916 { 917 struct knote *kn; 918 919 SLIST_FOREACH(kn, list, kn_selnext) 920 if (kn->kn_fop->f_event(kn, hint)) 921 KNOTE_ACTIVATE(kn); 922 } 923 924 /* 925 * remove all knotes from a specified klist 926 */ 927 void 928 knote_remove(struct thread *td, struct klist *list) 929 { 930 struct knote *kn; 931 932 while ((kn = SLIST_FIRST(list)) != NULL) { 933 kn->kn_fop->f_detach(kn); 934 knote_drop(kn, td); 935 } 936 } 937 938 /* 939 * remove all knotes referencing a specified fd 940 */ 941 void 942 knote_fdclose(struct thread *td, int fd) 943 { 944 struct filedesc *fdp = td->td_proc->p_fd; 945 struct klist *list; 946 947 FILEDESC_LOCK(fdp); 948 list = &fdp->fd_knlist[fd]; 949 FILEDESC_UNLOCK(fdp); 950 knote_remove(td, list); 951 } 952 953 static void 954 knote_attach(struct knote *kn, struct filedesc *fdp) 955 { 956 struct klist *list, *oldlist, *tmp_knhash; 957 u_long tmp_knhashmask; 958 int size, newsize; 959 960 FILEDESC_LOCK(fdp); 961 962 if (! kn->kn_fop->f_isfd) { 963 if (fdp->fd_knhashmask == 0) { 964 FILEDESC_UNLOCK(fdp); 965 tmp_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 966 &tmp_knhashmask); 967 FILEDESC_LOCK(fdp); 968 if (fdp->fd_knhashmask == 0) { 969 fdp->fd_knhash = tmp_knhash; 970 fdp->fd_knhashmask = tmp_knhashmask; 971 } else { 972 FILEDESC_UNLOCK(fdp); 973 free(tmp_knhash, M_KQUEUE); 974 FILEDESC_LOCK(fdp); 975 } 976 } 977 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 978 goto done; 979 } 980 981 if (fdp->fd_knlistsize <= kn->kn_id) { 982 retry: 983 size = fdp->fd_knlistsize; 984 while (size <= kn->kn_id) 985 size += KQEXTENT; 986 FILEDESC_UNLOCK(fdp); 987 MALLOC(list, struct klist *, 988 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 989 FILEDESC_LOCK(fdp); 990 newsize = fdp->fd_knlistsize; 991 while (newsize <= kn->kn_id) 992 newsize += KQEXTENT; 993 if (newsize != size) { 994 FILEDESC_UNLOCK(fdp); 995 free(list, M_TEMP); 996 FILEDESC_LOCK(fdp); 997 goto retry; 998 } 999 bcopy(fdp->fd_knlist, list, 1000 fdp->fd_knlistsize * sizeof(struct klist *)); 1001 bzero((caddr_t)list + 1002 fdp->fd_knlistsize * sizeof(struct klist *), 1003 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 1004 if (fdp->fd_knlist != NULL) 1005 oldlist = fdp->fd_knlist; 1006 else 1007 oldlist = NULL; 1008 fdp->fd_knlistsize = size; 1009 fdp->fd_knlist = list; 1010 FILEDESC_UNLOCK(fdp); 1011 if (oldlist != NULL) 1012 FREE(oldlist, M_KQUEUE); 1013 FILEDESC_LOCK(fdp); 1014 } 1015 list = &fdp->fd_knlist[kn->kn_id]; 1016 done: 1017 FILEDESC_UNLOCK(fdp); 1018 SLIST_INSERT_HEAD(list, kn, kn_link); 1019 kn->kn_status = 0; 1020 } 1021 1022 /* 1023 * should be called at spl == 0, since we don't want to hold spl 1024 * while calling fdrop and free. 1025 */ 1026 static void 1027 knote_drop(struct knote *kn, struct thread *td) 1028 { 1029 struct filedesc *fdp = td->td_proc->p_fd; 1030 struct klist *list; 1031 1032 FILEDESC_LOCK(fdp); 1033 if (kn->kn_fop->f_isfd) 1034 list = &fdp->fd_knlist[kn->kn_id]; 1035 else 1036 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1037 if (kn->kn_fop->f_isfd) 1038 FILE_LOCK(kn->kn_fp); 1039 FILEDESC_UNLOCK(fdp); 1040 1041 SLIST_REMOVE(list, kn, knote, kn_link); 1042 if (kn->kn_status & KN_QUEUED) 1043 knote_dequeue(kn); 1044 if (kn->kn_fop->f_isfd) 1045 fdrop_locked(kn->kn_fp, td); 1046 knote_free(kn); 1047 } 1048 1049 1050 static void 1051 knote_enqueue(struct knote *kn) 1052 { 1053 struct kqueue *kq = kn->kn_kq; 1054 int s = splhigh(); 1055 1056 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1057 1058 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1059 kn->kn_status |= KN_QUEUED; 1060 kq->kq_count++; 1061 splx(s); 1062 kqueue_wakeup(kq); 1063 } 1064 1065 static void 1066 knote_dequeue(struct knote *kn) 1067 { 1068 struct kqueue *kq = kn->kn_kq; 1069 int s = splhigh(); 1070 1071 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1072 1073 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1074 kn->kn_status &= ~KN_QUEUED; 1075 kq->kq_count--; 1076 splx(s); 1077 } 1078 1079 static void 1080 knote_init(void) 1081 { 1082 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 1083 NULL, NULL, UMA_ALIGN_PTR, 0); 1084 1085 } 1086 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 1087 1088 static struct knote * 1089 knote_alloc(void) 1090 { 1091 return ((struct knote *)uma_zalloc(knote_zone, M_WAITOK)); 1092 } 1093 1094 static void 1095 knote_free(struct knote *kn) 1096 { 1097 uma_zfree(knote_zone, kn); 1098 } 1099