1 /* $OpenBSD: kern_event.c,v 1.48 2012/07/08 17:21:08 guenther Exp $ */ 2 3 /*- 4 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD: src/sys/kern/kern_event.c,v 1.22 2001/02/23 20:32:42 jlemon Exp $ 29 */ 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/kernel.h> 34 #include <sys/proc.h> 35 #include <sys/malloc.h> 36 #include <sys/unistd.h> 37 #include <sys/file.h> 38 #include <sys/filedesc.h> 39 #include <sys/fcntl.h> 40 #include <sys/selinfo.h> 41 #include <sys/queue.h> 42 #include <sys/event.h> 43 #include <sys/eventvar.h> 44 #include <sys/ktrace.h> 45 #include <sys/pool.h> 46 #include <sys/protosw.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/stat.h> 50 #include <sys/uio.h> 51 #include <sys/mount.h> 52 #include <sys/poll.h> 53 #include <sys/syscallargs.h> 54 #include <sys/timeout.h> 55 56 int kqueue_scan(struct kqueue *kq, int maxevents, 57 struct kevent *ulistp, const struct timespec *timeout, 58 struct proc *p, int *retval); 59 60 int kqueue_read(struct file *fp, off_t *poff, struct uio *uio, 61 struct ucred *cred); 62 int kqueue_write(struct file *fp, off_t *poff, struct uio *uio, 63 struct ucred *cred); 64 int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 65 struct proc *p); 66 int kqueue_poll(struct file *fp, int events, struct proc *p); 67 int kqueue_kqfilter(struct file *fp, struct knote *kn); 68 int kqueue_stat(struct file *fp, struct stat *st, struct proc *p); 69 int kqueue_close(struct file *fp, struct proc *p); 70 void kqueue_wakeup(struct kqueue *kq); 71 72 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 void knote_attach(struct knote *kn, struct filedesc *fdp); 83 void knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp); 84 void knote_enqueue(struct knote *kn); 85 void knote_dequeue(struct knote *kn); 86 #define knote_alloc() ((struct knote *)pool_get(&knote_pool, PR_WAITOK)) 87 #define knote_free(kn) pool_put(&knote_pool, (kn)) 88 89 void filt_kqdetach(struct knote *kn); 90 int filt_kqueue(struct knote *kn, long hint); 91 int filt_procattach(struct knote *kn); 92 void filt_procdetach(struct knote *kn); 93 int filt_proc(struct knote *kn, long hint); 94 int filt_fileattach(struct knote *kn); 95 void filt_timerexpire(void *knx); 96 int filt_timerattach(struct knote *kn); 97 void filt_timerdetach(struct knote *kn); 98 int filt_timer(struct knote *kn, long hint); 99 void filt_seltruedetach(struct knote *kn); 100 101 struct filterops kqread_filtops = 102 { 1, NULL, filt_kqdetach, filt_kqueue }; 103 struct filterops proc_filtops = 104 { 0, filt_procattach, filt_procdetach, filt_proc }; 105 struct filterops file_filtops = 106 { 1, filt_fileattach, NULL, NULL }; 107 struct filterops timer_filtops = 108 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 109 110 struct pool knote_pool; 111 struct pool kqueue_pool; 112 int kq_ntimeouts = 0; 113 int kq_timeoutmax = (4 * 1024); 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 sig_filtops; 125 #ifdef notyet 126 extern struct filterops aio_filtops; 127 #endif 128 129 /* 130 * Table for for all system-defined filters. 131 */ 132 struct filterops *sysfilt_ops[] = { 133 &file_filtops, /* EVFILT_READ */ 134 &file_filtops, /* EVFILT_WRITE */ 135 NULL, /*&aio_filtops,*/ /* EVFILT_AIO */ 136 &file_filtops, /* EVFILT_VNODE */ 137 &proc_filtops, /* EVFILT_PROC */ 138 &sig_filtops, /* EVFILT_SIGNAL */ 139 &timer_filtops, /* EVFILT_TIMER */ 140 }; 141 142 void KQREF(struct kqueue *); 143 void KQRELE(struct kqueue *); 144 145 void 146 KQREF(struct kqueue *kq) 147 { 148 ++kq->kq_refs; 149 } 150 151 void 152 KQRELE(struct kqueue *kq) 153 { 154 if (--kq->kq_refs == 0) { 155 pool_put(&kqueue_pool, kq); 156 } 157 } 158 159 void kqueue_init(void); 160 161 void 162 kqueue_init(void) 163 { 164 165 pool_init(&kqueue_pool, sizeof(struct kqueue), 0, 0, 0, "kqueuepl", 166 &pool_allocator_nointr); 167 pool_init(&knote_pool, sizeof(struct knote), 0, 0, 0, "knotepl", 168 &pool_allocator_nointr); 169 } 170 171 int 172 filt_fileattach(struct knote *kn) 173 { 174 struct file *fp = kn->kn_fp; 175 176 return ((*fp->f_ops->fo_kqfilter)(fp, kn)); 177 } 178 179 int 180 kqueue_kqfilter(struct file *fp, struct knote *kn) 181 { 182 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 183 184 if (kn->kn_filter != EVFILT_READ) 185 return (EINVAL); 186 187 kn->kn_fop = &kqread_filtops; 188 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 189 return (0); 190 } 191 192 void 193 filt_kqdetach(struct knote *kn) 194 { 195 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 196 197 SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext); 198 } 199 200 /*ARGSUSED*/ 201 int 202 filt_kqueue(struct knote *kn, long hint) 203 { 204 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 205 206 kn->kn_data = kq->kq_count; 207 return (kn->kn_data > 0); 208 } 209 210 int 211 filt_procattach(struct knote *kn) 212 { 213 struct process *pr; 214 215 pr = prfind(kn->kn_id); 216 if (pr == NULL) 217 return (ESRCH); 218 219 /* exiting processes can't be specified */ 220 if (pr->ps_flags & PS_EXITING) 221 return (ESRCH); 222 223 /* 224 * Fail if it's not owned by you, or the last exec gave us 225 * setuid/setgid privs (unless you're root). 226 */ 227 if (pr != curproc->p_p && 228 (pr->ps_cred->p_ruid != curproc->p_cred->p_ruid || 229 (pr->ps_flags & PS_SUGID)) && suser(curproc, 0) != 0) 230 return (EACCES); 231 232 kn->kn_ptr.p_process = pr; 233 kn->kn_flags |= EV_CLEAR; /* automatically set */ 234 235 /* 236 * internal flag indicating registration done by kernel 237 */ 238 if (kn->kn_flags & EV_FLAG1) { 239 kn->kn_data = kn->kn_sdata; /* ppid */ 240 kn->kn_fflags = NOTE_CHILD; 241 kn->kn_flags &= ~EV_FLAG1; 242 } 243 244 /* XXX lock the proc here while adding to the list? */ 245 SLIST_INSERT_HEAD(&pr->ps_klist, kn, kn_selnext); 246 247 return (0); 248 } 249 250 /* 251 * The knote may be attached to a different process, which may exit, 252 * leaving nothing for the knote to be attached to. So when the process 253 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 254 * it will be deleted when read out. However, as part of the knote deletion, 255 * this routine is called, so a check is needed to avoid actually performing 256 * a detach, because the original process does not exist any more. 257 */ 258 void 259 filt_procdetach(struct knote *kn) 260 { 261 struct process *pr = kn->kn_ptr.p_process; 262 263 if (kn->kn_status & KN_DETACHED) 264 return; 265 266 /* XXX locking? this might modify another process. */ 267 SLIST_REMOVE(&pr->ps_klist, kn, knote, kn_selnext); 268 } 269 270 int 271 filt_proc(struct knote *kn, long hint) 272 { 273 u_int event; 274 275 /* 276 * mask off extra data 277 */ 278 event = (u_int)hint & NOTE_PCTRLMASK; 279 280 /* 281 * if the user is interested in this event, record it. 282 */ 283 if (kn->kn_sfflags & event) 284 kn->kn_fflags |= event; 285 286 /* 287 * process is gone, so flag the event as finished and remove it 288 * from the process's klist 289 */ 290 if (event == NOTE_EXIT) { 291 struct process *pr = kn->kn_ptr.p_process; 292 293 kn->kn_status |= KN_DETACHED; 294 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 295 SLIST_REMOVE(&pr->ps_klist, kn, knote, kn_selnext); 296 return (1); 297 } 298 299 /* 300 * process forked, and user wants to track the new process, 301 * so attach a new knote to it, and immediately report an 302 * event with the parent's pid. 303 */ 304 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 305 struct kevent kev; 306 int error; 307 308 /* 309 * register knote with new process. 310 */ 311 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 312 kev.filter = kn->kn_filter; 313 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 314 kev.fflags = kn->kn_sfflags; 315 kev.data = kn->kn_id; /* parent */ 316 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 317 error = kqueue_register(kn->kn_kq, &kev, NULL); 318 if (error) 319 kn->kn_fflags |= NOTE_TRACKERR; 320 } 321 322 return (kn->kn_fflags != 0); 323 } 324 325 void 326 filt_timerexpire(void *knx) 327 { 328 struct knote *kn = knx; 329 struct timeval tv; 330 int tticks; 331 332 kn->kn_data++; 333 KNOTE_ACTIVATE(kn); 334 335 if ((kn->kn_flags & EV_ONESHOT) == 0) { 336 tv.tv_sec = kn->kn_sdata / 1000; 337 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 338 tticks = tvtohz(&tv); 339 timeout_add((struct timeout *)kn->kn_hook, tticks); 340 } 341 } 342 343 344 /* 345 * data contains amount of time to sleep, in milliseconds 346 */ 347 int 348 filt_timerattach(struct knote *kn) 349 { 350 struct timeout *to; 351 struct timeval tv; 352 int tticks; 353 354 if (kq_ntimeouts > kq_timeoutmax) 355 return (ENOMEM); 356 kq_ntimeouts++; 357 358 tv.tv_sec = kn->kn_sdata / 1000; 359 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 360 tticks = tvtohz(&tv); 361 362 kn->kn_flags |= EV_CLEAR; /* automatically set */ 363 to = malloc(sizeof(*to), M_KEVENT, M_WAITOK); 364 timeout_set(to, filt_timerexpire, kn); 365 timeout_add(to, tticks); 366 kn->kn_hook = to; 367 368 return (0); 369 } 370 371 void 372 filt_timerdetach(struct knote *kn) 373 { 374 struct timeout *to; 375 376 to = (struct timeout *)kn->kn_hook; 377 timeout_del(to); 378 free(to, M_KEVENT); 379 kq_ntimeouts--; 380 } 381 382 int 383 filt_timer(struct knote *kn, long hint) 384 { 385 return (kn->kn_data != 0); 386 } 387 388 389 /* 390 * filt_seltrue: 391 * 392 * This filter "event" routine simulates seltrue(). 393 */ 394 int 395 filt_seltrue(struct knote *kn, long hint) 396 { 397 398 /* 399 * We don't know how much data can be read/written, 400 * but we know that it *can* be. This is about as 401 * good as select/poll does as well. 402 */ 403 kn->kn_data = 0; 404 return (1); 405 } 406 407 /* 408 * This provides full kqfilter entry for device switch tables, which 409 * has same effect as filter using filt_seltrue() as filter method. 410 */ 411 void 412 filt_seltruedetach(struct knote *kn) 413 { 414 /* Nothing to do */ 415 } 416 417 const struct filterops seltrue_filtops = 418 { 1, NULL, filt_seltruedetach, filt_seltrue }; 419 420 int 421 seltrue_kqfilter(dev_t dev, struct knote *kn) 422 { 423 switch (kn->kn_filter) { 424 case EVFILT_READ: 425 case EVFILT_WRITE: 426 kn->kn_fop = &seltrue_filtops; 427 break; 428 default: 429 return (EINVAL); 430 } 431 432 /* Nothing more to do */ 433 return (0); 434 } 435 436 int 437 sys_kqueue(struct proc *p, void *v, register_t *retval) 438 { 439 struct filedesc *fdp = p->p_fd; 440 struct kqueue *kq; 441 struct file *fp; 442 int fd, error; 443 444 fdplock(fdp); 445 error = falloc(p, &fp, &fd); 446 fdpunlock(fdp); 447 if (error) 448 return (error); 449 fp->f_flag = FREAD | FWRITE; 450 fp->f_type = DTYPE_KQUEUE; 451 fp->f_ops = &kqueueops; 452 kq = pool_get(&kqueue_pool, PR_WAITOK|PR_ZERO); 453 TAILQ_INIT(&kq->kq_head); 454 fp->f_data = (caddr_t)kq; 455 KQREF(kq); 456 *retval = fd; 457 if (fdp->fd_knlistsize < 0) 458 fdp->fd_knlistsize = 0; /* this process has a kq */ 459 kq->kq_fdp = fdp; 460 FILE_SET_MATURE(fp, p); 461 return (0); 462 } 463 464 int 465 sys_kevent(struct proc *p, void *v, register_t *retval) 466 { 467 struct filedesc* fdp = p->p_fd; 468 struct sys_kevent_args /* { 469 syscallarg(int) fd; 470 syscallarg(const struct kevent *) changelist; 471 syscallarg(int) nchanges; 472 syscallarg(struct kevent *) eventlist; 473 syscallarg(int) nevents; 474 syscallarg(const struct timespec *) timeout; 475 } */ *uap = v; 476 struct kevent *kevp; 477 struct kqueue *kq; 478 struct file *fp; 479 struct timespec ts; 480 int i, n, nerrors, error; 481 482 if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL || 483 (fp->f_type != DTYPE_KQUEUE)) 484 return (EBADF); 485 486 FREF(fp); 487 488 if (SCARG(uap, timeout) != NULL) { 489 error = copyin(SCARG(uap, timeout), &ts, sizeof(ts)); 490 if (error) 491 goto done; 492 #ifdef KTRACE 493 if (KTRPOINT(p, KTR_STRUCT)) 494 ktrreltimespec(p, &ts); 495 #endif 496 SCARG(uap, timeout) = &ts; 497 } 498 499 kq = (struct kqueue *)fp->f_data; 500 nerrors = 0; 501 502 while (SCARG(uap, nchanges) > 0) { 503 n = SCARG(uap, nchanges) > KQ_NEVENTS 504 ? KQ_NEVENTS : SCARG(uap, nchanges); 505 error = copyin(SCARG(uap, changelist), kq->kq_kev, 506 n * sizeof(struct kevent)); 507 if (error) 508 goto done; 509 for (i = 0; i < n; i++) { 510 kevp = &kq->kq_kev[i]; 511 kevp->flags &= ~EV_SYSFLAGS; 512 error = kqueue_register(kq, kevp, p); 513 if (error) { 514 if (SCARG(uap, nevents) != 0) { 515 kevp->flags = EV_ERROR; 516 kevp->data = error; 517 (void) copyout((caddr_t)kevp, 518 (caddr_t)SCARG(uap, eventlist), 519 sizeof(*kevp)); 520 SCARG(uap, eventlist)++; 521 SCARG(uap, nevents)--; 522 nerrors++; 523 } else { 524 goto done; 525 } 526 } 527 } 528 SCARG(uap, nchanges) -= n; 529 SCARG(uap, changelist) += n; 530 } 531 if (nerrors) { 532 *retval = nerrors; 533 error = 0; 534 goto done; 535 } 536 537 KQREF(kq); 538 FRELE(fp, p); 539 error = kqueue_scan(kq, SCARG(uap, nevents), SCARG(uap, eventlist), 540 SCARG(uap, timeout), p, &n); 541 KQRELE(kq); 542 *retval = n; 543 return (error); 544 545 done: 546 FRELE(fp, p); 547 return (error); 548 } 549 550 int 551 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p) 552 { 553 struct filedesc *fdp = kq->kq_fdp; 554 struct filterops *fops = NULL; 555 struct file *fp = NULL; 556 struct knote *kn = NULL; 557 int s, error = 0; 558 559 if (kev->filter < 0) { 560 if (kev->filter + EVFILT_SYSCOUNT < 0) 561 return (EINVAL); 562 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 563 } 564 565 if (fops == NULL) { 566 /* 567 * XXX 568 * filter attach routine is responsible for ensuring that 569 * the identifier can be attached to it. 570 */ 571 return (EINVAL); 572 } 573 574 if (fops->f_isfd) { 575 /* validate descriptor */ 576 if ((fp = fd_getfile(fdp, kev->ident)) == NULL) 577 return (EBADF); 578 FREF(fp); 579 580 if (kev->ident < fdp->fd_knlistsize) { 581 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 582 if (kq == kn->kn_kq && 583 kev->filter == kn->kn_filter) 584 break; 585 } 586 } else { 587 if (fdp->fd_knhashmask != 0) { 588 struct klist *list; 589 590 list = &fdp->fd_knhash[ 591 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 592 SLIST_FOREACH(kn, list, kn_link) 593 if (kev->ident == kn->kn_id && 594 kq == kn->kn_kq && 595 kev->filter == kn->kn_filter) 596 break; 597 } 598 } 599 600 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 601 error = ENOENT; 602 goto done; 603 } 604 605 /* 606 * kn now contains the matching knote, or NULL if no match 607 */ 608 if (kev->flags & EV_ADD) { 609 610 if (kn == NULL) { 611 kn = knote_alloc(); 612 if (kn == NULL) { 613 error = ENOMEM; 614 goto done; 615 } 616 kn->kn_fp = fp; 617 kn->kn_kq = kq; 618 kn->kn_fop = fops; 619 620 /* 621 * apply reference count to knote structure, and 622 * do not release it at the end of this routine. 623 */ 624 fp = NULL; 625 626 kn->kn_sfflags = kev->fflags; 627 kn->kn_sdata = kev->data; 628 kev->fflags = 0; 629 kev->data = 0; 630 kn->kn_kevent = *kev; 631 632 knote_attach(kn, fdp); 633 if ((error = fops->f_attach(kn)) != 0) { 634 knote_drop(kn, p, fdp); 635 goto done; 636 } 637 } else { 638 /* 639 * The user may change some filter values after the 640 * initial EV_ADD, but doing so will not reset any 641 * filters which have already been triggered. 642 */ 643 kn->kn_sfflags = kev->fflags; 644 kn->kn_sdata = kev->data; 645 kn->kn_kevent.udata = kev->udata; 646 } 647 648 s = splhigh(); 649 if (kn->kn_fop->f_event(kn, 0)) 650 KNOTE_ACTIVATE(kn); 651 splx(s); 652 653 } else if (kev->flags & EV_DELETE) { 654 kn->kn_fop->f_detach(kn); 655 knote_drop(kn, p, p->p_fd); 656 goto done; 657 } 658 659 if ((kev->flags & EV_DISABLE) && 660 ((kn->kn_status & KN_DISABLED) == 0)) { 661 s = splhigh(); 662 kn->kn_status |= KN_DISABLED; 663 splx(s); 664 } 665 666 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 667 s = splhigh(); 668 kn->kn_status &= ~KN_DISABLED; 669 if ((kn->kn_status & KN_ACTIVE) && 670 ((kn->kn_status & KN_QUEUED) == 0)) 671 knote_enqueue(kn); 672 splx(s); 673 } 674 675 done: 676 if (fp != NULL) 677 FRELE(fp, p); 678 return (error); 679 } 680 681 int 682 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent *ulistp, 683 const struct timespec *tsp, struct proc *p, int *retval) 684 { 685 struct kevent *kevp; 686 struct timeval atv, rtv, ttv; 687 struct knote *kn, marker; 688 int s, count, timeout, nkev = 0, error = 0; 689 690 count = maxevents; 691 if (count == 0) 692 goto done; 693 694 if (tsp != NULL) { 695 TIMESPEC_TO_TIMEVAL(&atv, tsp); 696 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 697 /* No timeout, just poll */ 698 timeout = -1; 699 goto start; 700 } 701 if (itimerfix(&atv)) { 702 error = EINVAL; 703 goto done; 704 } 705 706 timeout = atv.tv_sec > 24 * 60 * 60 ? 707 24 * 60 * 60 * hz : tvtohz(&atv); 708 709 getmicrouptime(&rtv); 710 timeradd(&atv, &rtv, &atv); 711 } else { 712 atv.tv_sec = 0; 713 atv.tv_usec = 0; 714 timeout = 0; 715 } 716 goto start; 717 718 retry: 719 if (atv.tv_sec || atv.tv_usec) { 720 getmicrouptime(&rtv); 721 if (timercmp(&rtv, &atv, >=)) 722 goto done; 723 ttv = atv; 724 timersub(&ttv, &rtv, &ttv); 725 timeout = ttv.tv_sec > 24 * 60 * 60 ? 726 24 * 60 * 60 * hz : tvtohz(&ttv); 727 } 728 729 start: 730 if (kq->kq_state & KQ_DYING) { 731 error = EBADF; 732 goto done; 733 } 734 735 kevp = kq->kq_kev; 736 s = splhigh(); 737 if (kq->kq_count == 0) { 738 if (timeout < 0) { 739 error = EWOULDBLOCK; 740 } else { 741 kq->kq_state |= KQ_SLEEP; 742 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 743 } 744 splx(s); 745 if (error == 0) 746 goto retry; 747 /* don't restart after signals... */ 748 if (error == ERESTART) 749 error = EINTR; 750 else if (error == EWOULDBLOCK) 751 error = 0; 752 goto done; 753 } 754 755 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 756 while (count) { 757 kn = TAILQ_FIRST(&kq->kq_head); 758 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 759 if (kn == &marker) { 760 splx(s); 761 if (count == maxevents) 762 goto retry; 763 goto done; 764 } 765 if (kn->kn_status & KN_DISABLED) { 766 kn->kn_status &= ~KN_QUEUED; 767 kq->kq_count--; 768 continue; 769 } 770 if ((kn->kn_flags & EV_ONESHOT) == 0 && 771 kn->kn_fop->f_event(kn, 0) == 0) { 772 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 773 kq->kq_count--; 774 continue; 775 } 776 *kevp = kn->kn_kevent; 777 kevp++; 778 nkev++; 779 if (kn->kn_flags & EV_ONESHOT) { 780 kn->kn_status &= ~KN_QUEUED; 781 kq->kq_count--; 782 splx(s); 783 kn->kn_fop->f_detach(kn); 784 knote_drop(kn, p, p->p_fd); 785 s = splhigh(); 786 } else if (kn->kn_flags & EV_CLEAR) { 787 kn->kn_data = 0; 788 kn->kn_fflags = 0; 789 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 790 kq->kq_count--; 791 } else { 792 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 793 } 794 count--; 795 if (nkev == KQ_NEVENTS) { 796 splx(s); 797 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 798 sizeof(struct kevent) * nkev); 799 ulistp += nkev; 800 nkev = 0; 801 kevp = kq->kq_kev; 802 s = splhigh(); 803 if (error) 804 break; 805 } 806 } 807 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 808 splx(s); 809 done: 810 if (nkev != 0) 811 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp, 812 sizeof(struct kevent) * nkev); 813 *retval = maxevents - count; 814 return (error); 815 } 816 817 /* 818 * XXX 819 * This could be expanded to call kqueue_scan, if desired. 820 */ 821 /*ARGSUSED*/ 822 int 823 kqueue_read(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 824 { 825 return (ENXIO); 826 } 827 828 /*ARGSUSED*/ 829 int 830 kqueue_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 831 832 { 833 return (ENXIO); 834 } 835 836 /*ARGSUSED*/ 837 int 838 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p) 839 { 840 return (ENOTTY); 841 } 842 843 /*ARGSUSED*/ 844 int 845 kqueue_poll(struct file *fp, int events, struct proc *p) 846 { 847 struct kqueue *kq = (struct kqueue *)fp->f_data; 848 int revents = 0; 849 int s = splhigh(); 850 851 if (events & (POLLIN | POLLRDNORM)) { 852 if (kq->kq_count) { 853 revents |= events & (POLLIN | POLLRDNORM); 854 } else { 855 selrecord(p, &kq->kq_sel); 856 kq->kq_state |= KQ_SEL; 857 } 858 } 859 splx(s); 860 return (revents); 861 } 862 863 /*ARGSUSED*/ 864 int 865 kqueue_stat(struct file *fp, struct stat *st, struct proc *p) 866 { 867 struct kqueue *kq = (struct kqueue *)fp->f_data; 868 869 bzero((void *)st, sizeof(*st)); 870 st->st_size = kq->kq_count; 871 st->st_blksize = sizeof(struct kevent); 872 st->st_mode = S_IFIFO; 873 return (0); 874 } 875 876 /*ARGSUSED*/ 877 int 878 kqueue_close(struct file *fp, struct proc *p) 879 { 880 struct kqueue *kq = (struct kqueue *)fp->f_data; 881 struct filedesc *fdp = p->p_fd; 882 struct knote **knp, *kn, *kn0; 883 int i; 884 885 for (i = 0; i < fdp->fd_knlistsize; i++) { 886 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 887 kn = *knp; 888 while (kn != NULL) { 889 kn0 = SLIST_NEXT(kn, kn_link); 890 if (kq == kn->kn_kq) { 891 kn->kn_fop->f_detach(kn); 892 FRELE(kn->kn_fp, p); 893 knote_free(kn); 894 *knp = kn0; 895 } else { 896 knp = &SLIST_NEXT(kn, kn_link); 897 } 898 kn = kn0; 899 } 900 } 901 if (fdp->fd_knhashmask != 0) { 902 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 903 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 904 kn = *knp; 905 while (kn != NULL) { 906 kn0 = SLIST_NEXT(kn, kn_link); 907 if (kq == kn->kn_kq) { 908 kn->kn_fop->f_detach(kn); 909 /* XXX non-fd release of kn->kn_ptr */ 910 knote_free(kn); 911 *knp = kn0; 912 } else { 913 knp = &SLIST_NEXT(kn, kn_link); 914 } 915 kn = kn0; 916 } 917 } 918 } 919 fp->f_data = NULL; 920 921 kq->kq_state |= KQ_DYING; 922 kqueue_wakeup(kq); 923 KQRELE(kq); 924 925 return (0); 926 } 927 928 void 929 kqueue_wakeup(struct kqueue *kq) 930 { 931 932 if (kq->kq_state & KQ_SLEEP) { 933 kq->kq_state &= ~KQ_SLEEP; 934 wakeup(kq); 935 } 936 if (kq->kq_state & KQ_SEL) { 937 kq->kq_state &= ~KQ_SEL; 938 selwakeup(&kq->kq_sel); 939 } else 940 KNOTE(&kq->kq_sel.si_note, 0); 941 } 942 943 /* 944 * walk down a list of knotes, activating them if their event has triggered. 945 */ 946 void 947 knote(struct klist *list, long hint) 948 { 949 struct knote *kn; 950 951 SLIST_FOREACH(kn, list, kn_selnext) 952 if (kn->kn_fop->f_event(kn, hint)) 953 KNOTE_ACTIVATE(kn); 954 } 955 956 /* 957 * remove all knotes from a specified klist 958 */ 959 void 960 knote_remove(struct proc *p, struct klist *list) 961 { 962 struct knote *kn; 963 964 while ((kn = SLIST_FIRST(list)) != NULL) { 965 kn->kn_fop->f_detach(kn); 966 knote_drop(kn, p, p->p_fd); 967 } 968 } 969 970 /* 971 * remove all knotes referencing a specified fd 972 */ 973 void 974 knote_fdclose(struct proc *p, int fd) 975 { 976 struct filedesc *fdp = p->p_fd; 977 struct klist *list = &fdp->fd_knlist[fd]; 978 979 knote_remove(p, list); 980 } 981 982 /* 983 * handle a process exiting, including the triggering of NOTE_EXIT notes 984 * XXX this could be more efficient, doing a single pass down the klist 985 */ 986 void 987 knote_processexit(struct process *pr) 988 { 989 KNOTE(&pr->ps_klist, NOTE_EXIT); 990 991 /* remove other knotes hanging off the process */ 992 knote_remove(pr->ps_mainproc, &pr->ps_klist); 993 } 994 995 void 996 knote_attach(struct knote *kn, struct filedesc *fdp) 997 { 998 struct klist *list; 999 int size; 1000 1001 if (! kn->kn_fop->f_isfd) { 1002 if (fdp->fd_knhashmask == 0) 1003 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP, 1004 M_WAITOK, &fdp->fd_knhashmask); 1005 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1006 goto done; 1007 } 1008 1009 if (fdp->fd_knlistsize <= kn->kn_id) { 1010 size = fdp->fd_knlistsize; 1011 while (size <= kn->kn_id) 1012 size += KQEXTENT; 1013 list = malloc(size * sizeof(struct klist *), M_TEMP, M_WAITOK); 1014 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, 1015 fdp->fd_knlistsize * sizeof(struct klist *)); 1016 bzero((caddr_t)list + 1017 fdp->fd_knlistsize * sizeof(struct klist *), 1018 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 1019 if (fdp->fd_knlist != NULL) 1020 free(fdp->fd_knlist, M_TEMP); 1021 fdp->fd_knlistsize = size; 1022 fdp->fd_knlist = list; 1023 } 1024 list = &fdp->fd_knlist[kn->kn_id]; 1025 done: 1026 SLIST_INSERT_HEAD(list, kn, kn_link); 1027 kn->kn_status = 0; 1028 } 1029 1030 /* 1031 * should be called at spl == 0, since we don't want to hold spl 1032 * while calling FRELE and knote_free. 1033 */ 1034 void 1035 knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp) 1036 { 1037 struct klist *list; 1038 1039 if (kn->kn_fop->f_isfd) 1040 list = &fdp->fd_knlist[kn->kn_id]; 1041 else 1042 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1043 1044 SLIST_REMOVE(list, kn, knote, kn_link); 1045 if (kn->kn_status & KN_QUEUED) 1046 knote_dequeue(kn); 1047 if (kn->kn_fop->f_isfd) { 1048 FRELE(kn->kn_fp, p); 1049 } 1050 knote_free(kn); 1051 } 1052 1053 1054 void 1055 knote_enqueue(struct knote *kn) 1056 { 1057 struct kqueue *kq = kn->kn_kq; 1058 int s = splhigh(); 1059 1060 KASSERT((kn->kn_status & KN_QUEUED) == 0); 1061 1062 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1063 kn->kn_status |= KN_QUEUED; 1064 kq->kq_count++; 1065 splx(s); 1066 kqueue_wakeup(kq); 1067 } 1068 1069 void 1070 knote_dequeue(struct knote *kn) 1071 { 1072 struct kqueue *kq = kn->kn_kq; 1073 int s = splhigh(); 1074 1075 KASSERT(kn->kn_status & KN_QUEUED); 1076 1077 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1078 kn->kn_status &= ~KN_QUEUED; 1079 kq->kq_count--; 1080 splx(s); 1081 } 1082 1083 void 1084 klist_invalidate(struct klist *list) 1085 { 1086 struct knote *kn; 1087 1088 SLIST_FOREACH(kn, list, kn_selnext) { 1089 kn->kn_status |= KN_DETACHED; 1090 kn->kn_flags |= EV_EOF | EV_ONESHOT; 1091 } 1092 } 1093