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 1.2.2.9 2003/05/08 07:47:16 kbyanc Exp $ 27 * $DragonFly: src/sys/kern/kern_event.c,v 1.2 2003/06/17 04:28:41 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/fcntl.h> 38 #include <sys/select.h> 39 #include <sys/queue.h> 40 #include <sys/event.h> 41 #include <sys/eventvar.h> 42 #include <sys/poll.h> 43 #include <sys/protosw.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/stat.h> 47 #include <sys/sysctl.h> 48 #include <sys/sysproto.h> 49 #include <sys/uio.h> 50 51 #include <vm/vm_zone.h> 52 53 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 54 55 static int kqueue_scan(struct file *fp, int maxevents, 56 struct kevent *ulistp, const struct timespec *timeout, 57 struct proc *p); 58 static int kqueue_read(struct file *fp, struct uio *uio, 59 struct ucred *cred, int flags, struct proc *p); 60 static int kqueue_write(struct file *fp, struct uio *uio, 61 struct ucred *cred, int flags, struct proc *p); 62 static int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 63 struct proc *p); 64 static int kqueue_poll(struct file *fp, int events, struct ucred *cred, 65 struct proc *p); 66 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 67 static int kqueue_stat(struct file *fp, struct stat *st, struct proc *p); 68 static int kqueue_close(struct file *fp, struct proc *p); 69 static void kqueue_wakeup(struct kqueue *kq); 70 71 static struct fileops kqueueops = { 72 kqueue_read, 73 kqueue_write, 74 kqueue_ioctl, 75 kqueue_poll, 76 kqueue_kqfilter, 77 kqueue_stat, 78 kqueue_close 79 }; 80 81 static void knote_attach(struct knote *kn, struct filedesc *fdp); 82 static void knote_drop(struct knote *kn, struct proc *p); 83 static void knote_enqueue(struct knote *kn); 84 static void knote_dequeue(struct knote *kn); 85 static void knote_init(void); 86 static struct knote *knote_alloc(void); 87 static void knote_free(struct knote *kn); 88 89 static void filt_kqdetach(struct knote *kn); 90 static int filt_kqueue(struct knote *kn, long hint); 91 static int filt_procattach(struct knote *kn); 92 static void filt_procdetach(struct knote *kn); 93 static int filt_proc(struct knote *kn, long hint); 94 static int filt_fileattach(struct knote *kn); 95 static void filt_timerexpire(void *knx); 96 static int filt_timerattach(struct knote *kn); 97 static void filt_timerdetach(struct knote *kn); 98 static int filt_timer(struct knote *kn, long hint); 99 100 static struct filterops file_filtops = 101 { 1, filt_fileattach, NULL, NULL }; 102 static struct filterops kqread_filtops = 103 { 1, NULL, filt_kqdetach, filt_kqueue }; 104 static struct filterops proc_filtops = 105 { 0, filt_procattach, filt_procdetach, filt_proc }; 106 static struct filterops timer_filtops = 107 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 108 109 static vm_zone_t knote_zone; 110 static int kq_ncallouts = 0; 111 static int kq_calloutmax = (4 * 1024); 112 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 113 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 114 115 #define KNOTE_ACTIVATE(kn) do { \ 116 kn->kn_status |= KN_ACTIVE; \ 117 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 118 knote_enqueue(kn); \ 119 } while(0) 120 121 #define KN_HASHSIZE 64 /* XXX should be tunable */ 122 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 123 124 extern struct filterops aio_filtops; 125 extern struct filterops sig_filtops; 126 127 /* 128 * Table for for all system-defined filters. 129 */ 130 static struct filterops *sysfilt_ops[] = { 131 &file_filtops, /* EVFILT_READ */ 132 &file_filtops, /* EVFILT_WRITE */ 133 &aio_filtops, /* EVFILT_AIO */ 134 &file_filtops, /* EVFILT_VNODE */ 135 &proc_filtops, /* EVFILT_PROC */ 136 &sig_filtops, /* EVFILT_SIGNAL */ 137 &timer_filtops, /* EVFILT_TIMER */ 138 }; 139 140 static int 141 filt_fileattach(struct knote *kn) 142 { 143 144 return (fo_kqfilter(kn->kn_fp, kn)); 145 } 146 147 /*ARGSUSED*/ 148 static int 149 kqueue_kqfilter(struct file *fp, struct knote *kn) 150 { 151 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 152 153 if (kn->kn_filter != EVFILT_READ) 154 return (1); 155 156 kn->kn_fop = &kqread_filtops; 157 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 158 return (0); 159 } 160 161 static void 162 filt_kqdetach(struct knote *kn) 163 { 164 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 165 166 SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext); 167 } 168 169 /*ARGSUSED*/ 170 static int 171 filt_kqueue(struct knote *kn, long hint) 172 { 173 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 174 175 kn->kn_data = kq->kq_count; 176 return (kn->kn_data > 0); 177 } 178 179 static int 180 filt_procattach(struct knote *kn) 181 { 182 struct proc *p; 183 int immediate; 184 185 immediate = 0; 186 p = pfind(kn->kn_id); 187 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 188 p = zpfind(kn->kn_id); 189 immediate = 1; 190 } 191 if (p == NULL) 192 return (ESRCH); 193 if (! PRISON_CHECK(curproc, p)) 194 return (EACCES); 195 196 kn->kn_ptr.p_proc = p; 197 kn->kn_flags |= EV_CLEAR; /* automatically set */ 198 199 /* 200 * internal flag indicating registration done by kernel 201 */ 202 if (kn->kn_flags & EV_FLAG1) { 203 kn->kn_data = kn->kn_sdata; /* ppid */ 204 kn->kn_fflags = NOTE_CHILD; 205 kn->kn_flags &= ~EV_FLAG1; 206 } 207 208 /* XXX lock the proc here while adding to the list? */ 209 SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext); 210 211 /* 212 * Immediately activate any exit notes if the target process is a 213 * zombie. This is necessary to handle the case where the target 214 * process, e.g. a child, dies before the kevent is registered. 215 */ 216 if (immediate && filt_proc(kn, NOTE_EXIT)) 217 KNOTE_ACTIVATE(kn); 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 /* XXX locking? this might modify another process. */ 239 SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext); 240 } 241 242 static int 243 filt_proc(struct knote *kn, long hint) 244 { 245 u_int event; 246 247 /* 248 * mask off extra data 249 */ 250 event = (u_int)hint & NOTE_PCTRLMASK; 251 252 /* 253 * if the user is interested in this event, record it. 254 */ 255 if (kn->kn_sfflags & event) 256 kn->kn_fflags |= event; 257 258 /* 259 * process is gone, so flag the event as finished. 260 */ 261 if (event == NOTE_EXIT) { 262 kn->kn_status |= KN_DETACHED; 263 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 264 return (1); 265 } 266 267 /* 268 * process forked, and user wants to track the new process, 269 * so attach a new knote to it, and immediately report an 270 * event with the parent's pid. 271 */ 272 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 273 struct kevent kev; 274 int error; 275 276 /* 277 * register knote with new process. 278 */ 279 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 280 kev.filter = kn->kn_filter; 281 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 282 kev.fflags = kn->kn_sfflags; 283 kev.data = kn->kn_id; /* parent */ 284 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 285 error = kqueue_register(kn->kn_kq, &kev, NULL); 286 if (error) 287 kn->kn_fflags |= NOTE_TRACKERR; 288 } 289 290 return (kn->kn_fflags != 0); 291 } 292 293 static void 294 filt_timerexpire(void *knx) 295 { 296 struct knote *kn = knx; 297 struct callout *calloutp; 298 struct timeval tv; 299 int tticks; 300 301 kn->kn_data++; 302 KNOTE_ACTIVATE(kn); 303 304 if ((kn->kn_flags & EV_ONESHOT) == 0) { 305 tv.tv_sec = kn->kn_sdata / 1000; 306 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 307 tticks = tvtohz(&tv); 308 calloutp = (struct callout *)kn->kn_hook; 309 callout_reset(calloutp, tticks, filt_timerexpire, kn); 310 } 311 } 312 313 /* 314 * data contains amount of time to sleep, in milliseconds 315 */ 316 static int 317 filt_timerattach(struct knote *kn) 318 { 319 struct callout *calloutp; 320 struct timeval tv; 321 int tticks; 322 323 if (kq_ncallouts >= kq_calloutmax) 324 return (ENOMEM); 325 kq_ncallouts++; 326 327 tv.tv_sec = kn->kn_sdata / 1000; 328 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 329 tticks = tvtohz(&tv); 330 331 kn->kn_flags |= EV_CLEAR; /* automatically set */ 332 MALLOC(calloutp, struct callout *, sizeof(*calloutp), 333 M_KQUEUE, M_WAITOK); 334 callout_init(calloutp); 335 callout_reset(calloutp, tticks, filt_timerexpire, kn); 336 kn->kn_hook = (caddr_t)calloutp; 337 338 return (0); 339 } 340 341 static void 342 filt_timerdetach(struct knote *kn) 343 { 344 struct callout *calloutp; 345 346 calloutp = (struct callout *)kn->kn_hook; 347 callout_stop(calloutp); 348 FREE(calloutp, M_KQUEUE); 349 kq_ncallouts--; 350 } 351 352 static int 353 filt_timer(struct knote *kn, long hint) 354 { 355 356 return (kn->kn_data != 0); 357 } 358 359 int 360 kqueue(struct proc *p, struct kqueue_args *uap) 361 { 362 struct filedesc *fdp = p->p_fd; 363 struct kqueue *kq; 364 struct file *fp; 365 int fd, error; 366 367 error = falloc(p, &fp, &fd); 368 if (error) 369 return (error); 370 fp->f_flag = FREAD | FWRITE; 371 fp->f_type = DTYPE_KQUEUE; 372 fp->f_ops = &kqueueops; 373 kq = malloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 374 TAILQ_INIT(&kq->kq_head); 375 fp->f_data = (caddr_t)kq; 376 p->p_retval[0] = fd; 377 if (fdp->fd_knlistsize < 0) 378 fdp->fd_knlistsize = 0; /* this process has a kq */ 379 kq->kq_fdp = fdp; 380 return (error); 381 } 382 383 #ifndef _SYS_SYSPROTO_H_ 384 struct kevent_args { 385 int fd; 386 const struct kevent *changelist; 387 int nchanges; 388 struct kevent *eventlist; 389 int nevents; 390 const struct timespec *timeout; 391 }; 392 #endif 393 int 394 kevent(struct proc *p, struct kevent_args *uap) 395 { 396 struct filedesc* fdp = p->p_fd; 397 struct kevent *kevp; 398 struct kqueue *kq; 399 struct file *fp = NULL; 400 struct timespec ts; 401 int i, n, nerrors, error; 402 403 if (((u_int)uap->fd) >= fdp->fd_nfiles || 404 (fp = fdp->fd_ofiles[uap->fd]) == NULL || 405 (fp->f_type != DTYPE_KQUEUE)) 406 return (EBADF); 407 408 fhold(fp); 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, p); 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 p->p_retval[0] = nerrors; 450 error = 0; 451 goto done; 452 } 453 454 error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, p); 455 done: 456 if (fp != NULL) 457 fdrop(fp, p); 458 return (error); 459 } 460 461 int 462 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p) 463 { 464 struct filedesc *fdp = kq->kq_fdp; 465 struct filterops *fops; 466 struct file *fp = NULL; 467 struct knote *kn = NULL; 468 int s, error = 0; 469 470 if (kev->filter < 0) { 471 if (kev->filter + EVFILT_SYSCOUNT < 0) 472 return (EINVAL); 473 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 474 } else { 475 /* 476 * XXX 477 * filter attach routine is responsible for insuring that 478 * the identifier can be attached to it. 479 */ 480 printf("unknown filter: %d\n", kev->filter); 481 return (EINVAL); 482 } 483 484 if (fops->f_isfd) { 485 /* validate descriptor */ 486 if ((u_int)kev->ident >= fdp->fd_nfiles || 487 (fp = fdp->fd_ofiles[kev->ident]) == NULL) 488 return (EBADF); 489 fhold(fp); 490 491 if (kev->ident < fdp->fd_knlistsize) { 492 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 493 if (kq == kn->kn_kq && 494 kev->filter == kn->kn_filter) 495 break; 496 } 497 } else { 498 if (fdp->fd_knhashmask != 0) { 499 struct klist *list; 500 501 list = &fdp->fd_knhash[ 502 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 503 SLIST_FOREACH(kn, list, kn_link) 504 if (kev->ident == kn->kn_id && 505 kq == kn->kn_kq && 506 kev->filter == kn->kn_filter) 507 break; 508 } 509 } 510 511 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 512 error = ENOENT; 513 goto done; 514 } 515 516 /* 517 * kn now contains the matching knote, or NULL if no match 518 */ 519 if (kev->flags & EV_ADD) { 520 521 if (kn == NULL) { 522 kn = knote_alloc(); 523 if (kn == NULL) { 524 error = ENOMEM; 525 goto done; 526 } 527 kn->kn_fp = fp; 528 kn->kn_kq = kq; 529 kn->kn_fop = fops; 530 531 /* 532 * apply reference count to knote structure, and 533 * do not release it at the end of this routine. 534 */ 535 fp = NULL; 536 537 kn->kn_sfflags = kev->fflags; 538 kn->kn_sdata = kev->data; 539 kev->fflags = 0; 540 kev->data = 0; 541 kn->kn_kevent = *kev; 542 543 knote_attach(kn, fdp); 544 if ((error = fops->f_attach(kn)) != 0) { 545 knote_drop(kn, p); 546 goto done; 547 } 548 } else { 549 /* 550 * The user may change some filter values after the 551 * initial EV_ADD, but doing so will not reset any 552 * filter which have already been triggered. 553 */ 554 kn->kn_sfflags = kev->fflags; 555 kn->kn_sdata = kev->data; 556 kn->kn_kevent.udata = kev->udata; 557 } 558 559 s = splhigh(); 560 if (kn->kn_fop->f_event(kn, 0)) 561 KNOTE_ACTIVATE(kn); 562 splx(s); 563 564 } else if (kev->flags & EV_DELETE) { 565 kn->kn_fop->f_detach(kn); 566 knote_drop(kn, p); 567 goto done; 568 } 569 570 if ((kev->flags & EV_DISABLE) && 571 ((kn->kn_status & KN_DISABLED) == 0)) { 572 s = splhigh(); 573 kn->kn_status |= KN_DISABLED; 574 splx(s); 575 } 576 577 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 578 s = splhigh(); 579 kn->kn_status &= ~KN_DISABLED; 580 if ((kn->kn_status & KN_ACTIVE) && 581 ((kn->kn_status & KN_QUEUED) == 0)) 582 knote_enqueue(kn); 583 splx(s); 584 } 585 586 done: 587 if (fp != NULL) 588 fdrop(fp, p); 589 return (error); 590 } 591 592 static int 593 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp, 594 const struct timespec *tsp, struct proc *p) 595 { 596 struct kqueue *kq = (struct kqueue *)fp->f_data; 597 struct kevent *kevp; 598 struct timeval atv, rtv, ttv; 599 struct knote *kn, marker; 600 int s, count, timeout, nkev = 0, error = 0; 601 602 count = maxevents; 603 if (count == 0) 604 goto done; 605 606 if (tsp != NULL) { 607 TIMESPEC_TO_TIMEVAL(&atv, tsp); 608 if (itimerfix(&atv)) { 609 error = EINVAL; 610 goto done; 611 } 612 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 613 timeout = -1; 614 else 615 timeout = atv.tv_sec > 24 * 60 * 60 ? 616 24 * 60 * 60 * hz : tvtohz(&atv); 617 getmicrouptime(&rtv); 618 timevaladd(&atv, &rtv); 619 } else { 620 atv.tv_sec = 0; 621 atv.tv_usec = 0; 622 timeout = 0; 623 } 624 goto start; 625 626 retry: 627 if (atv.tv_sec || atv.tv_usec) { 628 getmicrouptime(&rtv); 629 if (timevalcmp(&rtv, &atv, >=)) 630 goto done; 631 ttv = atv; 632 timevalsub(&ttv, &rtv); 633 timeout = ttv.tv_sec > 24 * 60 * 60 ? 634 24 * 60 * 60 * hz : tvtohz(&ttv); 635 } 636 637 start: 638 kevp = kq->kq_kev; 639 s = splhigh(); 640 if (kq->kq_count == 0) { 641 if (timeout < 0) { 642 error = EWOULDBLOCK; 643 } else { 644 kq->kq_state |= KQ_SLEEP; 645 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 646 } 647 splx(s); 648 if (error == 0) 649 goto retry; 650 /* don't restart after signals... */ 651 if (error == ERESTART) 652 error = EINTR; 653 else if (error == EWOULDBLOCK) 654 error = 0; 655 goto done; 656 } 657 658 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 659 while (count) { 660 kn = TAILQ_FIRST(&kq->kq_head); 661 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 662 if (kn == &marker) { 663 splx(s); 664 if (count == maxevents) 665 goto retry; 666 goto done; 667 } 668 if (kn->kn_status & KN_DISABLED) { 669 kn->kn_status &= ~KN_QUEUED; 670 kq->kq_count--; 671 continue; 672 } 673 if ((kn->kn_flags & EV_ONESHOT) == 0 && 674 kn->kn_fop->f_event(kn, 0) == 0) { 675 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 676 kq->kq_count--; 677 continue; 678 } 679 *kevp = kn->kn_kevent; 680 kevp++; 681 nkev++; 682 if (kn->kn_flags & EV_ONESHOT) { 683 kn->kn_status &= ~KN_QUEUED; 684 kq->kq_count--; 685 splx(s); 686 kn->kn_fop->f_detach(kn); 687 knote_drop(kn, p); 688 s = splhigh(); 689 } else if (kn->kn_flags & EV_CLEAR) { 690 kn->kn_data = 0; 691 kn->kn_fflags = 0; 692 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 693 kq->kq_count--; 694 } else { 695 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 696 } 697 count--; 698 if (nkev == KQ_NEVENTS) { 699 splx(s); 700 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 701 sizeof(struct kevent) * nkev); 702 ulistp += nkev; 703 nkev = 0; 704 kevp = kq->kq_kev; 705 s = splhigh(); 706 if (error) 707 break; 708 } 709 } 710 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 711 splx(s); 712 done: 713 if (nkev != 0) 714 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 715 sizeof(struct kevent) * nkev); 716 p->p_retval[0] = maxevents - count; 717 return (error); 718 } 719 720 /* 721 * XXX 722 * This could be expanded to call kqueue_scan, if desired. 723 */ 724 /*ARGSUSED*/ 725 static int 726 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, 727 int flags, struct proc *p) 728 { 729 return (ENXIO); 730 } 731 732 /*ARGSUSED*/ 733 static int 734 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, 735 int flags, struct proc *p) 736 { 737 return (ENXIO); 738 } 739 740 /*ARGSUSED*/ 741 static int 742 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p) 743 { 744 return (ENOTTY); 745 } 746 747 /*ARGSUSED*/ 748 static int 749 kqueue_poll(struct file *fp, int events, struct ucred *cred, struct proc *p) 750 { 751 struct kqueue *kq = (struct kqueue *)fp->f_data; 752 int revents = 0; 753 int s = splnet(); 754 755 if (events & (POLLIN | POLLRDNORM)) { 756 if (kq->kq_count) { 757 revents |= events & (POLLIN | POLLRDNORM); 758 } else { 759 selrecord(p, &kq->kq_sel); 760 kq->kq_state |= KQ_SEL; 761 } 762 } 763 splx(s); 764 return (revents); 765 } 766 767 /*ARGSUSED*/ 768 static int 769 kqueue_stat(struct file *fp, struct stat *st, struct proc *p) 770 { 771 struct kqueue *kq = (struct kqueue *)fp->f_data; 772 773 bzero((void *)st, sizeof(*st)); 774 st->st_size = kq->kq_count; 775 st->st_blksize = sizeof(struct kevent); 776 st->st_mode = S_IFIFO; 777 return (0); 778 } 779 780 /*ARGSUSED*/ 781 static int 782 kqueue_close(struct file *fp, struct proc *p) 783 { 784 struct kqueue *kq = (struct kqueue *)fp->f_data; 785 struct filedesc *fdp = p->p_fd; 786 struct knote **knp, *kn, *kn0; 787 int i; 788 789 for (i = 0; i < fdp->fd_knlistsize; i++) { 790 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 791 kn = *knp; 792 while (kn != NULL) { 793 kn0 = SLIST_NEXT(kn, kn_link); 794 if (kq == kn->kn_kq) { 795 kn->kn_fop->f_detach(kn); 796 fdrop(kn->kn_fp, p); 797 knote_free(kn); 798 *knp = kn0; 799 } else { 800 knp = &SLIST_NEXT(kn, kn_link); 801 } 802 kn = kn0; 803 } 804 } 805 if (fdp->fd_knhashmask != 0) { 806 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 807 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 808 kn = *knp; 809 while (kn != NULL) { 810 kn0 = SLIST_NEXT(kn, kn_link); 811 if (kq == kn->kn_kq) { 812 kn->kn_fop->f_detach(kn); 813 /* XXX non-fd release of kn->kn_ptr */ 814 knote_free(kn); 815 *knp = kn0; 816 } else { 817 knp = &SLIST_NEXT(kn, kn_link); 818 } 819 kn = kn0; 820 } 821 } 822 } 823 free(kq, M_KQUEUE); 824 fp->f_data = NULL; 825 826 return (0); 827 } 828 829 static void 830 kqueue_wakeup(struct kqueue *kq) 831 { 832 833 if (kq->kq_state & KQ_SLEEP) { 834 kq->kq_state &= ~KQ_SLEEP; 835 wakeup(kq); 836 } 837 if (kq->kq_state & KQ_SEL) { 838 kq->kq_state &= ~KQ_SEL; 839 selwakeup(&kq->kq_sel); 840 } 841 KNOTE(&kq->kq_sel.si_note, 0); 842 } 843 844 /* 845 * walk down a list of knotes, activating them if their event has triggered. 846 */ 847 void 848 knote(struct klist *list, long hint) 849 { 850 struct knote *kn; 851 852 SLIST_FOREACH(kn, list, kn_selnext) 853 if (kn->kn_fop->f_event(kn, hint)) 854 KNOTE_ACTIVATE(kn); 855 } 856 857 /* 858 * remove all knotes from a specified klist 859 */ 860 void 861 knote_remove(struct proc *p, struct klist *list) 862 { 863 struct knote *kn; 864 865 while ((kn = SLIST_FIRST(list)) != NULL) { 866 kn->kn_fop->f_detach(kn); 867 knote_drop(kn, p); 868 } 869 } 870 871 /* 872 * remove all knotes referencing a specified fd 873 */ 874 void 875 knote_fdclose(struct proc *p, int fd) 876 { 877 struct filedesc *fdp = p->p_fd; 878 struct klist *list = &fdp->fd_knlist[fd]; 879 880 knote_remove(p, list); 881 } 882 883 static void 884 knote_attach(struct knote *kn, struct filedesc *fdp) 885 { 886 struct klist *list; 887 int size; 888 889 if (! kn->kn_fop->f_isfd) { 890 if (fdp->fd_knhashmask == 0) 891 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 892 &fdp->fd_knhashmask); 893 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 894 goto done; 895 } 896 897 if (fdp->fd_knlistsize <= kn->kn_id) { 898 size = fdp->fd_knlistsize; 899 while (size <= kn->kn_id) 900 size += KQEXTENT; 901 MALLOC(list, struct klist *, 902 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 903 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, 904 fdp->fd_knlistsize * sizeof(struct klist *)); 905 bzero((caddr_t)list + 906 fdp->fd_knlistsize * sizeof(struct klist *), 907 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 908 if (fdp->fd_knlist != NULL) 909 FREE(fdp->fd_knlist, M_KQUEUE); 910 fdp->fd_knlistsize = size; 911 fdp->fd_knlist = list; 912 } 913 list = &fdp->fd_knlist[kn->kn_id]; 914 done: 915 SLIST_INSERT_HEAD(list, kn, kn_link); 916 kn->kn_status = 0; 917 } 918 919 /* 920 * should be called at spl == 0, since we don't want to hold spl 921 * while calling fdrop and free. 922 */ 923 static void 924 knote_drop(struct knote *kn, struct proc *p) 925 { 926 struct filedesc *fdp = p->p_fd; 927 struct klist *list; 928 929 if (kn->kn_fop->f_isfd) 930 list = &fdp->fd_knlist[kn->kn_id]; 931 else 932 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 933 934 SLIST_REMOVE(list, kn, knote, kn_link); 935 if (kn->kn_status & KN_QUEUED) 936 knote_dequeue(kn); 937 if (kn->kn_fop->f_isfd) 938 fdrop(kn->kn_fp, p); 939 knote_free(kn); 940 } 941 942 943 static void 944 knote_enqueue(struct knote *kn) 945 { 946 struct kqueue *kq = kn->kn_kq; 947 int s = splhigh(); 948 949 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 950 951 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 952 kn->kn_status |= KN_QUEUED; 953 kq->kq_count++; 954 splx(s); 955 kqueue_wakeup(kq); 956 } 957 958 static void 959 knote_dequeue(struct knote *kn) 960 { 961 struct kqueue *kq = kn->kn_kq; 962 int s = splhigh(); 963 964 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 965 966 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 967 kn->kn_status &= ~KN_QUEUED; 968 kq->kq_count--; 969 splx(s); 970 } 971 972 static void 973 knote_init(void) 974 { 975 knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1); 976 } 977 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 978 979 static struct knote * 980 knote_alloc(void) 981 { 982 return ((struct knote *)zalloc(knote_zone)); 983 } 984 985 static void 986 knote_free(struct knote *kn) 987 { 988 zfree(knote_zone, kn); 989 } 990