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 *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, void *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 uma_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 static int 126 filt_nullattach(struct knote *kn) 127 { 128 129 return (ENXIO); 130 }; 131 132 struct filterops null_filtops = 133 { 0, filt_nullattach, NULL, NULL }; 134 135 extern struct filterops sig_filtops; 136 137 /* 138 * Table for for all system-defined filters. 139 */ 140 static struct filterops *sysfilt_ops[] = { 141 &file_filtops, /* EVFILT_READ */ 142 &file_filtops, /* EVFILT_WRITE */ 143 &null_filtops, /* EVFILT_AIO */ 144 &file_filtops, /* EVFILT_VNODE */ 145 &proc_filtops, /* EVFILT_PROC */ 146 &sig_filtops, /* EVFILT_SIGNAL */ 147 &timer_filtops, /* EVFILT_TIMER */ 148 &file_filtops, /* EVFILT_NETDEV */ 149 }; 150 151 static int 152 filt_fileattach(struct knote *kn) 153 { 154 155 return (fo_kqfilter(kn->kn_fp, kn)); 156 } 157 158 /*ARGSUSED*/ 159 static int 160 kqueue_kqfilter(struct file *fp, struct knote *kn) 161 { 162 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 163 164 if (kn->kn_filter != EVFILT_READ) 165 return (1); 166 167 kn->kn_fop = &kqread_filtops; 168 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 169 return (0); 170 } 171 172 static void 173 filt_kqdetach(struct knote *kn) 174 { 175 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 176 177 SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext); 178 } 179 180 /*ARGSUSED*/ 181 static int 182 filt_kqueue(struct knote *kn, long hint) 183 { 184 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 185 186 kn->kn_data = kq->kq_count; 187 return (kn->kn_data > 0); 188 } 189 190 static int 191 filt_procattach(struct knote *kn) 192 { 193 struct proc *p; 194 int error; 195 196 p = pfind(kn->kn_id); 197 if (p == NULL) 198 return (ESRCH); 199 if ((error = p_cansee(curthread, p))) { 200 PROC_UNLOCK(p); 201 return (error); 202 } 203 204 kn->kn_ptr.p_proc = p; 205 kn->kn_flags |= EV_CLEAR; /* automatically set */ 206 207 /* 208 * internal flag indicating registration done by kernel 209 */ 210 if (kn->kn_flags & EV_FLAG1) { 211 kn->kn_data = kn->kn_sdata; /* ppid */ 212 kn->kn_fflags = NOTE_CHILD; 213 kn->kn_flags &= ~EV_FLAG1; 214 } 215 216 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 217 PROC_UNLOCK(p); 218 219 return (0); 220 } 221 222 /* 223 * The knote may be attached to a different process, which may exit, 224 * leaving nothing for the knote to be attached to. So when the process 225 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 226 * it will be deleted when read out. However, as part of the knote deletion, 227 * this routine is called, so a check is needed to avoid actually performing 228 * a detach, because the original process does not exist any more. 229 */ 230 static void 231 filt_procdetach(struct knote *kn) 232 { 233 struct proc *p = kn->kn_ptr.p_proc; 234 235 if (kn->kn_status & KN_DETACHED) 236 return; 237 238 PROC_LOCK(p); 239 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 240 PROC_UNLOCK(p); 241 } 242 243 static int 244 filt_proc(struct knote *kn, long hint) 245 { 246 u_int event; 247 248 /* 249 * mask off extra data 250 */ 251 event = (u_int)hint & NOTE_PCTRLMASK; 252 253 /* 254 * if the user is interested in this event, record it. 255 */ 256 if (kn->kn_sfflags & event) 257 kn->kn_fflags |= event; 258 259 /* 260 * process is gone, so flag the event as finished. 261 */ 262 if (event == NOTE_EXIT) { 263 kn->kn_status |= KN_DETACHED; 264 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 265 return (1); 266 } 267 268 /* 269 * process forked, and user wants to track the new process, 270 * so attach a new knote to it, and immediately report an 271 * event with the parent's pid. 272 */ 273 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 274 struct kevent kev; 275 int error; 276 277 /* 278 * register knote with new process. 279 */ 280 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 281 kev.filter = kn->kn_filter; 282 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 283 kev.fflags = kn->kn_sfflags; 284 kev.data = kn->kn_id; /* parent */ 285 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 286 error = kqueue_register(kn->kn_kq, &kev, NULL); 287 if (error) 288 kn->kn_fflags |= NOTE_TRACKERR; 289 } 290 291 return (kn->kn_fflags != 0); 292 } 293 294 static void 295 filt_timerexpire(void *knx) 296 { 297 struct knote *kn = knx; 298 struct callout *calloutp; 299 struct timeval tv; 300 int tticks; 301 302 kn->kn_data++; 303 KNOTE_ACTIVATE(kn); 304 305 if ((kn->kn_flags & EV_ONESHOT) == 0) { 306 tv.tv_sec = kn->kn_sdata / 1000; 307 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 308 tticks = tvtohz(&tv); 309 calloutp = (struct callout *)kn->kn_hook; 310 callout_reset(calloutp, tticks, filt_timerexpire, kn); 311 } 312 } 313 314 /* 315 * data contains amount of time to sleep, in milliseconds 316 */ 317 static int 318 filt_timerattach(struct knote *kn) 319 { 320 struct callout *calloutp; 321 struct timeval tv; 322 int tticks; 323 324 if (kq_ncallouts >= kq_calloutmax) 325 return (ENOMEM); 326 kq_ncallouts++; 327 328 tv.tv_sec = kn->kn_sdata / 1000; 329 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 330 tticks = tvtohz(&tv); 331 332 kn->kn_flags |= EV_CLEAR; /* automatically set */ 333 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 334 M_KQUEUE, M_WAITOK); 335 callout_init(calloutp, 0); 336 callout_reset(calloutp, tticks, filt_timerexpire, kn); 337 kn->kn_hook = calloutp; 338 339 return (0); 340 } 341 342 static void 343 filt_timerdetach(struct knote *kn) 344 { 345 struct callout *calloutp; 346 347 calloutp = (struct callout *)kn->kn_hook; 348 callout_stop(calloutp); 349 FREE(calloutp, M_KQUEUE); 350 kq_ncallouts--; 351 } 352 353 static int 354 filt_timer(struct knote *kn, long hint) 355 { 356 357 return (kn->kn_data != 0); 358 } 359 360 /* 361 * MPSAFE 362 */ 363 int 364 kqueue(struct thread *td, struct kqueue_args *uap) 365 { 366 struct filedesc *fdp; 367 struct kqueue *kq; 368 struct file *fp; 369 int fd, error; 370 371 mtx_lock(&Giant); 372 fdp = td->td_proc->p_fd; 373 error = falloc(td, &fp, &fd); 374 if (error) 375 goto done2; 376 kq = malloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 377 TAILQ_INIT(&kq->kq_head); 378 FILE_LOCK(fp); 379 fp->f_flag = FREAD | FWRITE; 380 fp->f_type = DTYPE_KQUEUE; 381 fp->f_ops = &kqueueops; 382 TAILQ_INIT(&kq->kq_head); 383 fp->f_data = kq; 384 FILE_UNLOCK(fp); 385 FILEDESC_LOCK(fdp); 386 td->td_retval[0] = fd; 387 if (fdp->fd_knlistsize < 0) 388 fdp->fd_knlistsize = 0; /* this process has a kq */ 389 FILEDESC_UNLOCK(fdp); 390 kq->kq_fdp = fdp; 391 done2: 392 mtx_unlock(&Giant); 393 return (error); 394 } 395 396 #ifndef _SYS_SYSPROTO_H_ 397 struct kevent_args { 398 int fd; 399 const struct kevent *changelist; 400 int nchanges; 401 struct kevent *eventlist; 402 int nevents; 403 const struct timespec *timeout; 404 }; 405 #endif 406 /* 407 * MPSAFE 408 */ 409 int 410 kevent(struct thread *td, struct kevent_args *uap) 411 { 412 struct kevent *kevp; 413 struct kqueue *kq; 414 struct file *fp; 415 struct timespec ts; 416 int i, n, nerrors, error; 417 418 if ((error = fget(td, uap->fd, &fp)) != 0) 419 return (error); 420 if (fp->f_type != DTYPE_KQUEUE) { 421 fdrop(fp, td); 422 return (EBADF); 423 } 424 if (uap->timeout != NULL) { 425 error = copyin(uap->timeout, &ts, sizeof(ts)); 426 if (error) 427 goto done_nogiant; 428 uap->timeout = &ts; 429 } 430 mtx_lock(&Giant); 431 432 kq = (struct kqueue *)fp->f_data; 433 nerrors = 0; 434 435 while (uap->nchanges > 0) { 436 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges; 437 error = copyin(uap->changelist, kq->kq_kev, 438 n * sizeof(struct kevent)); 439 if (error) 440 goto done; 441 for (i = 0; i < n; i++) { 442 kevp = &kq->kq_kev[i]; 443 kevp->flags &= ~EV_SYSFLAGS; 444 error = kqueue_register(kq, kevp, td); 445 if (error) { 446 if (uap->nevents != 0) { 447 kevp->flags = EV_ERROR; 448 kevp->data = error; 449 (void) copyout(kevp, 450 uap->eventlist, 451 sizeof(*kevp)); 452 uap->eventlist++; 453 uap->nevents--; 454 nerrors++; 455 } else { 456 goto done; 457 } 458 } 459 } 460 uap->nchanges -= n; 461 uap->changelist += n; 462 } 463 if (nerrors) { 464 td->td_retval[0] = nerrors; 465 error = 0; 466 goto done; 467 } 468 469 error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, td); 470 done: 471 mtx_unlock(&Giant); 472 done_nogiant: 473 if (fp != NULL) 474 fdrop(fp, td); 475 return (error); 476 } 477 478 int 479 kqueue_add_filteropts(int filt, struct filterops *filtops) 480 { 481 482 if (filt > 0) 483 panic("filt(%d) > 0", filt); 484 if (filt + EVFILT_SYSCOUNT < 0) 485 panic("filt(%d) + EVFILT_SYSCOUNT(%d) == %d < 0", 486 filt, EVFILT_SYSCOUNT, filt + EVFILT_SYSCOUNT); 487 if (sysfilt_ops[~filt] != &null_filtops) 488 panic("sysfilt_ops[~filt(%d)] != &null_filtops", filt); 489 sysfilt_ops[~filt] = filtops; 490 return (0); 491 } 492 493 int 494 kqueue_del_filteropts(int filt) 495 { 496 497 if (filt > 0) 498 panic("filt(%d) > 0", filt); 499 if (filt + EVFILT_SYSCOUNT < 0) 500 panic("filt(%d) + EVFILT_SYSCOUNT(%d) == %d < 0", 501 filt, EVFILT_SYSCOUNT, filt + EVFILT_SYSCOUNT); 502 if (sysfilt_ops[~filt] == &null_filtops) 503 panic("sysfilt_ops[~filt(%d)] != &null_filtops", filt); 504 sysfilt_ops[~filt] = &null_filtops; 505 return (0); 506 } 507 508 int 509 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td) 510 { 511 struct filedesc *fdp = kq->kq_fdp; 512 struct filterops *fops; 513 struct file *fp = NULL; 514 struct knote *kn = NULL; 515 int s, error = 0; 516 517 if (kev->filter < 0) { 518 if (kev->filter + EVFILT_SYSCOUNT < 0) 519 return (EINVAL); 520 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 521 } else { 522 /* 523 * XXX 524 * filter attach routine is responsible for insuring that 525 * the identifier can be attached to it. 526 */ 527 printf("unknown filter: %d\n", kev->filter); 528 return (EINVAL); 529 } 530 531 FILEDESC_LOCK(fdp); 532 if (fops->f_isfd) { 533 /* validate descriptor */ 534 if ((u_int)kev->ident >= fdp->fd_nfiles || 535 (fp = fdp->fd_ofiles[kev->ident]) == NULL) { 536 FILEDESC_UNLOCK(fdp); 537 return (EBADF); 538 } 539 fhold(fp); 540 541 if (kev->ident < fdp->fd_knlistsize) { 542 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 543 if (kq == kn->kn_kq && 544 kev->filter == kn->kn_filter) 545 break; 546 } 547 } else { 548 if (fdp->fd_knhashmask != 0) { 549 struct klist *list; 550 551 list = &fdp->fd_knhash[ 552 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 553 SLIST_FOREACH(kn, list, kn_link) 554 if (kev->ident == kn->kn_id && 555 kq == kn->kn_kq && 556 kev->filter == kn->kn_filter) 557 break; 558 } 559 } 560 FILEDESC_UNLOCK(fdp); 561 562 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 563 error = ENOENT; 564 goto done; 565 } 566 567 /* 568 * kn now contains the matching knote, or NULL if no match 569 */ 570 if (kev->flags & EV_ADD) { 571 572 if (kn == NULL) { 573 kn = knote_alloc(); 574 if (kn == NULL) { 575 error = ENOMEM; 576 goto done; 577 } 578 kn->kn_fp = fp; 579 kn->kn_kq = kq; 580 kn->kn_fop = fops; 581 582 /* 583 * apply reference count to knote structure, and 584 * do not release it at the end of this routine. 585 */ 586 fp = NULL; 587 588 kn->kn_sfflags = kev->fflags; 589 kn->kn_sdata = kev->data; 590 kev->fflags = 0; 591 kev->data = 0; 592 kn->kn_kevent = *kev; 593 594 knote_attach(kn, fdp); 595 if ((error = fops->f_attach(kn)) != 0) { 596 knote_drop(kn, td); 597 goto done; 598 } 599 } else { 600 /* 601 * The user may change some filter values after the 602 * initial EV_ADD, but doing so will not reset any 603 * filter which have already been triggered. 604 */ 605 kn->kn_sfflags = kev->fflags; 606 kn->kn_sdata = kev->data; 607 kn->kn_kevent.udata = kev->udata; 608 } 609 610 s = splhigh(); 611 if (kn->kn_fop->f_event(kn, 0)) 612 KNOTE_ACTIVATE(kn); 613 splx(s); 614 615 } else if (kev->flags & EV_DELETE) { 616 kn->kn_fop->f_detach(kn); 617 knote_drop(kn, td); 618 goto done; 619 } 620 621 if ((kev->flags & EV_DISABLE) && 622 ((kn->kn_status & KN_DISABLED) == 0)) { 623 s = splhigh(); 624 kn->kn_status |= KN_DISABLED; 625 splx(s); 626 } 627 628 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 629 s = splhigh(); 630 kn->kn_status &= ~KN_DISABLED; 631 if ((kn->kn_status & KN_ACTIVE) && 632 ((kn->kn_status & KN_QUEUED) == 0)) 633 knote_enqueue(kn); 634 splx(s); 635 } 636 637 done: 638 if (fp != NULL) 639 fdrop(fp, td); 640 return (error); 641 } 642 643 static int 644 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp, 645 const struct timespec *tsp, struct thread *td) 646 { 647 struct kqueue *kq; 648 struct kevent *kevp; 649 struct timeval atv, rtv, ttv; 650 struct knote *kn, marker; 651 int s, count, timeout, nkev = 0, error = 0; 652 653 FILE_LOCK_ASSERT(fp, MA_NOTOWNED); 654 655 kq = (struct kqueue *)fp->f_data; 656 count = maxevents; 657 if (count == 0) 658 goto done; 659 660 if (tsp != NULL) { 661 TIMESPEC_TO_TIMEVAL(&atv, tsp); 662 if (itimerfix(&atv)) { 663 error = EINVAL; 664 goto done; 665 } 666 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 667 timeout = -1; 668 else 669 timeout = atv.tv_sec > 24 * 60 * 60 ? 670 24 * 60 * 60 * hz : tvtohz(&atv); 671 getmicrouptime(&rtv); 672 timevaladd(&atv, &rtv); 673 } else { 674 atv.tv_sec = 0; 675 atv.tv_usec = 0; 676 timeout = 0; 677 } 678 goto start; 679 680 retry: 681 if (atv.tv_sec || atv.tv_usec) { 682 getmicrouptime(&rtv); 683 if (timevalcmp(&rtv, &atv, >=)) 684 goto done; 685 ttv = atv; 686 timevalsub(&ttv, &rtv); 687 timeout = ttv.tv_sec > 24 * 60 * 60 ? 688 24 * 60 * 60 * hz : tvtohz(&ttv); 689 } 690 691 start: 692 kevp = kq->kq_kev; 693 s = splhigh(); 694 if (kq->kq_count == 0) { 695 if (timeout < 0) { 696 error = EWOULDBLOCK; 697 } else { 698 kq->kq_state |= KQ_SLEEP; 699 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 700 } 701 splx(s); 702 if (error == 0) 703 goto retry; 704 /* don't restart after signals... */ 705 if (error == ERESTART) 706 error = EINTR; 707 else if (error == EWOULDBLOCK) 708 error = 0; 709 goto done; 710 } 711 712 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 713 while (count) { 714 kn = TAILQ_FIRST(&kq->kq_head); 715 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 716 if (kn == &marker) { 717 splx(s); 718 if (count == maxevents) 719 goto retry; 720 goto done; 721 } 722 if (kn->kn_status & KN_DISABLED) { 723 kn->kn_status &= ~KN_QUEUED; 724 kq->kq_count--; 725 continue; 726 } 727 if ((kn->kn_flags & EV_ONESHOT) == 0 && 728 kn->kn_fop->f_event(kn, 0) == 0) { 729 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 730 kq->kq_count--; 731 continue; 732 } 733 *kevp = kn->kn_kevent; 734 kevp++; 735 nkev++; 736 if (kn->kn_flags & EV_ONESHOT) { 737 kn->kn_status &= ~KN_QUEUED; 738 kq->kq_count--; 739 splx(s); 740 kn->kn_fop->f_detach(kn); 741 knote_drop(kn, td); 742 s = splhigh(); 743 } else if (kn->kn_flags & EV_CLEAR) { 744 kn->kn_data = 0; 745 kn->kn_fflags = 0; 746 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 747 kq->kq_count--; 748 } else { 749 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 750 } 751 count--; 752 if (nkev == KQ_NEVENTS) { 753 splx(s); 754 error = copyout(&kq->kq_kev, ulistp, 755 sizeof(struct kevent) * nkev); 756 ulistp += nkev; 757 nkev = 0; 758 kevp = kq->kq_kev; 759 s = splhigh(); 760 if (error) 761 break; 762 } 763 } 764 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 765 splx(s); 766 done: 767 if (nkev != 0) 768 error = copyout(&kq->kq_kev, ulistp, 769 sizeof(struct kevent) * nkev); 770 td->td_retval[0] = maxevents - count; 771 return (error); 772 } 773 774 /* 775 * XXX 776 * This could be expanded to call kqueue_scan, if desired. 777 */ 778 /*ARGSUSED*/ 779 static int 780 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, 781 int flags, struct thread *td) 782 { 783 return (ENXIO); 784 } 785 786 /*ARGSUSED*/ 787 static int 788 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, 789 int flags, struct thread *td) 790 { 791 return (ENXIO); 792 } 793 794 /*ARGSUSED*/ 795 static int 796 kqueue_ioctl(struct file *fp, u_long com, void *data, struct thread *td) 797 { 798 return (ENOTTY); 799 } 800 801 /*ARGSUSED*/ 802 static int 803 kqueue_poll(struct file *fp, int events, struct ucred *cred, struct thread *td) 804 { 805 struct kqueue *kq; 806 int revents = 0; 807 int s = splnet(); 808 809 kq = (struct kqueue *)fp->f_data; 810 if (events & (POLLIN | POLLRDNORM)) { 811 if (kq->kq_count) { 812 revents |= events & (POLLIN | POLLRDNORM); 813 } else { 814 selrecord(td, &kq->kq_sel); 815 kq->kq_state |= KQ_SEL; 816 } 817 } 818 splx(s); 819 return (revents); 820 } 821 822 /*ARGSUSED*/ 823 static int 824 kqueue_stat(struct file *fp, struct stat *st, struct thread *td) 825 { 826 struct kqueue *kq; 827 828 kq = (struct kqueue *)fp->f_data; 829 bzero((void *)st, sizeof(*st)); 830 st->st_size = kq->kq_count; 831 st->st_blksize = sizeof(struct kevent); 832 st->st_mode = S_IFIFO; 833 return (0); 834 } 835 836 /*ARGSUSED*/ 837 static int 838 kqueue_close(struct file *fp, struct thread *td) 839 { 840 struct kqueue *kq = (struct kqueue *)fp->f_data; 841 struct filedesc *fdp = td->td_proc->p_fd; 842 struct knote **knp, *kn, *kn0; 843 int i; 844 845 FILEDESC_LOCK(fdp); 846 for (i = 0; i < fdp->fd_knlistsize; i++) { 847 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 848 kn = *knp; 849 while (kn != NULL) { 850 kn0 = SLIST_NEXT(kn, kn_link); 851 if (kq == kn->kn_kq) { 852 kn->kn_fop->f_detach(kn); 853 *knp = kn0; 854 FILE_LOCK(kn->kn_fp); 855 FILEDESC_UNLOCK(fdp); 856 fdrop_locked(kn->kn_fp, td); 857 knote_free(kn); 858 FILEDESC_LOCK(fdp); 859 } else { 860 knp = &SLIST_NEXT(kn, kn_link); 861 } 862 kn = kn0; 863 } 864 } 865 if (fdp->fd_knhashmask != 0) { 866 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 867 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 868 kn = *knp; 869 while (kn != NULL) { 870 kn0 = SLIST_NEXT(kn, kn_link); 871 if (kq == kn->kn_kq) { 872 kn->kn_fop->f_detach(kn); 873 *knp = kn0; 874 /* XXX non-fd release of kn->kn_ptr */ 875 FILEDESC_UNLOCK(fdp); 876 knote_free(kn); 877 FILEDESC_LOCK(fdp); 878 } else { 879 knp = &SLIST_NEXT(kn, kn_link); 880 } 881 kn = kn0; 882 } 883 } 884 } 885 FILEDESC_UNLOCK(fdp); 886 free(kq, M_KQUEUE); 887 fp->f_data = NULL; 888 889 return (0); 890 } 891 892 static void 893 kqueue_wakeup(struct kqueue *kq) 894 { 895 896 if (kq->kq_state & KQ_SLEEP) { 897 kq->kq_state &= ~KQ_SLEEP; 898 wakeup(kq); 899 } 900 if (kq->kq_state & KQ_SEL) { 901 kq->kq_state &= ~KQ_SEL; 902 selwakeup(&kq->kq_sel); 903 } 904 KNOTE(&kq->kq_sel.si_note, 0); 905 } 906 907 /* 908 * walk down a list of knotes, activating them if their event has triggered. 909 */ 910 void 911 knote(struct klist *list, long hint) 912 { 913 struct knote *kn; 914 915 SLIST_FOREACH(kn, list, kn_selnext) 916 if (kn->kn_fop->f_event(kn, hint)) 917 KNOTE_ACTIVATE(kn); 918 } 919 920 /* 921 * remove all knotes from a specified klist 922 */ 923 void 924 knote_remove(struct thread *td, struct klist *list) 925 { 926 struct knote *kn; 927 928 while ((kn = SLIST_FIRST(list)) != NULL) { 929 kn->kn_fop->f_detach(kn); 930 knote_drop(kn, td); 931 } 932 } 933 934 /* 935 * remove all knotes referencing a specified fd 936 */ 937 void 938 knote_fdclose(struct thread *td, int fd) 939 { 940 struct filedesc *fdp = td->td_proc->p_fd; 941 struct klist *list; 942 943 FILEDESC_LOCK(fdp); 944 list = &fdp->fd_knlist[fd]; 945 FILEDESC_UNLOCK(fdp); 946 knote_remove(td, list); 947 } 948 949 static void 950 knote_attach(struct knote *kn, struct filedesc *fdp) 951 { 952 struct klist *list, *oldlist; 953 int size, newsize; 954 955 FILEDESC_LOCK(fdp); 956 957 if (! kn->kn_fop->f_isfd) { 958 if (fdp->fd_knhashmask == 0) 959 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 960 &fdp->fd_knhashmask); 961 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 962 goto done; 963 } 964 965 if (fdp->fd_knlistsize <= kn->kn_id) { 966 retry: 967 size = fdp->fd_knlistsize; 968 while (size <= kn->kn_id) 969 size += KQEXTENT; 970 FILEDESC_UNLOCK(fdp); 971 MALLOC(list, struct klist *, 972 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 973 FILEDESC_LOCK(fdp); 974 newsize = fdp->fd_knlistsize; 975 while (newsize <= kn->kn_id) 976 newsize += KQEXTENT; 977 if (newsize != size) { 978 FILEDESC_UNLOCK(fdp); 979 free(list, M_TEMP); 980 FILEDESC_LOCK(fdp); 981 goto retry; 982 } 983 bcopy(fdp->fd_knlist, list, 984 fdp->fd_knlistsize * sizeof(struct klist *)); 985 bzero((caddr_t)list + 986 fdp->fd_knlistsize * sizeof(struct klist *), 987 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 988 if (fdp->fd_knlist != NULL) 989 oldlist = fdp->fd_knlist; 990 else 991 oldlist = NULL; 992 fdp->fd_knlistsize = size; 993 fdp->fd_knlist = list; 994 FILEDESC_UNLOCK(fdp); 995 if (oldlist != NULL) 996 FREE(oldlist, M_KQUEUE); 997 FILEDESC_LOCK(fdp); 998 } 999 list = &fdp->fd_knlist[kn->kn_id]; 1000 done: 1001 FILEDESC_UNLOCK(fdp); 1002 SLIST_INSERT_HEAD(list, kn, kn_link); 1003 kn->kn_status = 0; 1004 } 1005 1006 /* 1007 * should be called at spl == 0, since we don't want to hold spl 1008 * while calling fdrop and free. 1009 */ 1010 static void 1011 knote_drop(struct knote *kn, struct thread *td) 1012 { 1013 struct filedesc *fdp = td->td_proc->p_fd; 1014 struct klist *list; 1015 1016 FILEDESC_LOCK(fdp); 1017 if (kn->kn_fop->f_isfd) 1018 list = &fdp->fd_knlist[kn->kn_id]; 1019 else 1020 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1021 if (kn->kn_fop->f_isfd) 1022 FILE_LOCK(kn->kn_fp); 1023 FILEDESC_UNLOCK(fdp); 1024 1025 SLIST_REMOVE(list, kn, knote, kn_link); 1026 if (kn->kn_status & KN_QUEUED) 1027 knote_dequeue(kn); 1028 if (kn->kn_fop->f_isfd) 1029 fdrop_locked(kn->kn_fp, td); 1030 knote_free(kn); 1031 } 1032 1033 1034 static void 1035 knote_enqueue(struct knote *kn) 1036 { 1037 struct kqueue *kq = kn->kn_kq; 1038 int s = splhigh(); 1039 1040 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1041 1042 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1043 kn->kn_status |= KN_QUEUED; 1044 kq->kq_count++; 1045 splx(s); 1046 kqueue_wakeup(kq); 1047 } 1048 1049 static void 1050 knote_dequeue(struct knote *kn) 1051 { 1052 struct kqueue *kq = kn->kn_kq; 1053 int s = splhigh(); 1054 1055 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1056 1057 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1058 kn->kn_status &= ~KN_QUEUED; 1059 kq->kq_count--; 1060 splx(s); 1061 } 1062 1063 static void 1064 knote_init(void) 1065 { 1066 knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL, 1067 NULL, NULL, UMA_ALIGN_PTR, 0); 1068 1069 } 1070 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 1071 1072 static struct knote * 1073 knote_alloc(void) 1074 { 1075 return ((struct knote *)uma_zalloc(knote_zone, M_WAITOK)); 1076 } 1077 1078 static void 1079 knote_free(struct knote *kn) 1080 { 1081 uma_zfree(knote_zone, kn); 1082 } 1083