1 /* $OpenBSD: kern_event.c,v 1.84 2018/01/13 12:58:40 robert 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/pledge.h> 36 #include <sys/malloc.h> 37 #include <sys/unistd.h> 38 #include <sys/file.h> 39 #include <sys/filedesc.h> 40 #include <sys/fcntl.h> 41 #include <sys/selinfo.h> 42 #include <sys/queue.h> 43 #include <sys/event.h> 44 #include <sys/eventvar.h> 45 #include <sys/ktrace.h> 46 #include <sys/pool.h> 47 #include <sys/protosw.h> 48 #include <sys/socket.h> 49 #include <sys/socketvar.h> 50 #include <sys/stat.h> 51 #include <sys/uio.h> 52 #include <sys/mount.h> 53 #include <sys/poll.h> 54 #include <sys/syscallargs.h> 55 #include <sys/timeout.h> 56 57 int kqueue_scan(struct kqueue *kq, int maxevents, 58 struct kevent *ulistp, const struct timespec *timeout, 59 struct proc *p, int *retval); 60 61 int kqueue_read(struct file *fp, off_t *poff, struct uio *uio, 62 struct ucred *cred); 63 int kqueue_write(struct file *fp, off_t *poff, struct uio *uio, 64 struct ucred *cred); 65 int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 66 struct proc *p); 67 int kqueue_poll(struct file *fp, int events, struct proc *p); 68 int kqueue_kqfilter(struct file *fp, struct knote *kn); 69 int kqueue_stat(struct file *fp, struct stat *st, struct proc *p); 70 int kqueue_close(struct file *fp, struct proc *p); 71 void kqueue_wakeup(struct kqueue *kq); 72 73 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 void knote_attach(struct knote *kn, struct filedesc *fdp); 84 void knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp); 85 void knote_enqueue(struct knote *kn); 86 void knote_dequeue(struct knote *kn); 87 #define knote_alloc() ((struct knote *)pool_get(&knote_pool, PR_WAITOK)) 88 #define knote_free(kn) pool_put(&knote_pool, (kn)) 89 90 void filt_kqdetach(struct knote *kn); 91 int filt_kqueue(struct knote *kn, long hint); 92 int filt_procattach(struct knote *kn); 93 void filt_procdetach(struct knote *kn); 94 int filt_proc(struct knote *kn, long hint); 95 int filt_fileattach(struct knote *kn); 96 void filt_timerexpire(void *knx); 97 int filt_timerattach(struct knote *kn); 98 void filt_timerdetach(struct knote *kn); 99 int filt_timer(struct knote *kn, long hint); 100 void filt_seltruedetach(struct knote *kn); 101 102 struct filterops kqread_filtops = 103 { 1, NULL, filt_kqdetach, filt_kqueue }; 104 struct filterops proc_filtops = 105 { 0, filt_procattach, filt_procdetach, filt_proc }; 106 struct filterops file_filtops = 107 { 1, filt_fileattach, NULL, NULL }; 108 struct filterops timer_filtops = 109 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 110 111 struct pool knote_pool; 112 struct pool kqueue_pool; 113 int kq_ntimeouts = 0; 114 int kq_timeoutmax = (4 * 1024); 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_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 &file_filtops, /* EVFILT_DEVICE */ 141 }; 142 143 void KQREF(struct kqueue *); 144 void KQRELE(struct kqueue *); 145 146 void 147 KQREF(struct kqueue *kq) 148 { 149 ++kq->kq_refs; 150 } 151 152 void 153 KQRELE(struct kqueue *kq) 154 { 155 if (--kq->kq_refs == 0) { 156 pool_put(&kqueue_pool, kq); 157 } 158 } 159 160 void kqueue_init(void); 161 162 void 163 kqueue_init(void) 164 { 165 166 pool_init(&kqueue_pool, sizeof(struct kqueue), 0, IPL_NONE, PR_WAITOK, 167 "kqueuepl", NULL); 168 pool_init(&knote_pool, sizeof(struct knote), 0, IPL_NONE, PR_WAITOK, 169 "knotepl", NULL); 170 } 171 172 int 173 filt_fileattach(struct knote *kn) 174 { 175 struct file *fp = kn->kn_fp; 176 177 return fp->f_ops->fo_kqfilter(fp, kn); 178 } 179 180 int 181 kqueue_kqfilter(struct file *fp, struct knote *kn) 182 { 183 struct kqueue *kq = kn->kn_fp->f_data; 184 185 if (kn->kn_filter != EVFILT_READ) 186 return (EINVAL); 187 188 kn->kn_fop = &kqread_filtops; 189 SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext); 190 return (0); 191 } 192 193 void 194 filt_kqdetach(struct knote *kn) 195 { 196 struct kqueue *kq = kn->kn_fp->f_data; 197 198 SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext); 199 } 200 201 int 202 filt_kqueue(struct knote *kn, long hint) 203 { 204 struct kqueue *kq = 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 if ((curproc->p_p->ps_flags & PS_PLEDGE) && 216 (curproc->p_p->ps_pledge & PLEDGE_PROC) == 0) 217 return pledge_fail(curproc, EPERM, PLEDGE_PROC); 218 219 if (kn->kn_id > PID_MAX) 220 return ESRCH; 221 222 pr = prfind(kn->kn_id); 223 if (pr == NULL) 224 return (ESRCH); 225 226 /* exiting processes can't be specified */ 227 if (pr->ps_flags & PS_EXITING) 228 return (ESRCH); 229 230 kn->kn_ptr.p_process = pr; 231 kn->kn_flags |= EV_CLEAR; /* automatically set */ 232 233 /* 234 * internal flag indicating registration done by kernel 235 */ 236 if (kn->kn_flags & EV_FLAG1) { 237 kn->kn_data = kn->kn_sdata; /* ppid */ 238 kn->kn_fflags = NOTE_CHILD; 239 kn->kn_flags &= ~EV_FLAG1; 240 } 241 242 /* XXX lock the proc here while adding to the list? */ 243 SLIST_INSERT_HEAD(&pr->ps_klist, kn, kn_selnext); 244 245 return (0); 246 } 247 248 /* 249 * The knote may be attached to a different process, which may exit, 250 * leaving nothing for the knote to be attached to. So when the process 251 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 252 * it will be deleted when read out. However, as part of the knote deletion, 253 * this routine is called, so a check is needed to avoid actually performing 254 * a detach, because the original process does not exist any more. 255 */ 256 void 257 filt_procdetach(struct knote *kn) 258 { 259 struct process *pr = kn->kn_ptr.p_process; 260 261 if (kn->kn_status & KN_DETACHED) 262 return; 263 264 /* XXX locking? this might modify another process. */ 265 SLIST_REMOVE(&pr->ps_klist, kn, knote, kn_selnext); 266 } 267 268 int 269 filt_proc(struct knote *kn, long hint) 270 { 271 u_int event; 272 273 /* 274 * mask off extra data 275 */ 276 event = (u_int)hint & NOTE_PCTRLMASK; 277 278 /* 279 * if the user is interested in this event, record it. 280 */ 281 if (kn->kn_sfflags & event) 282 kn->kn_fflags |= event; 283 284 /* 285 * process is gone, so flag the event as finished and remove it 286 * from the process's klist 287 */ 288 if (event == NOTE_EXIT) { 289 struct process *pr = kn->kn_ptr.p_process; 290 291 kn->kn_status |= KN_DETACHED; 292 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 293 kn->kn_data = pr->ps_mainproc->p_xstat; 294 SLIST_REMOVE(&pr->ps_klist, kn, knote, kn_selnext); 295 return (1); 296 } 297 298 /* 299 * process forked, and user wants to track the new process, 300 * so attach a new knote to it, and immediately report an 301 * event with the parent's pid. 302 */ 303 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 304 struct kevent kev; 305 int error; 306 307 /* 308 * register knote with new process. 309 */ 310 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 311 kev.filter = kn->kn_filter; 312 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 313 kev.fflags = kn->kn_sfflags; 314 kev.data = kn->kn_id; /* parent */ 315 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 316 error = kqueue_register(kn->kn_kq, &kev, NULL); 317 if (error) 318 kn->kn_fflags |= NOTE_TRACKERR; 319 } 320 321 return (kn->kn_fflags != 0); 322 } 323 324 static void 325 filt_timer_timeout_add(struct knote *kn) 326 { 327 struct timeval tv; 328 int tticks; 329 330 tv.tv_sec = kn->kn_sdata / 1000; 331 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 332 tticks = tvtohz(&tv); 333 timeout_add(kn->kn_hook, tticks ? tticks : 1); 334 } 335 336 void 337 filt_timerexpire(void *knx) 338 { 339 struct knote *kn = knx; 340 341 kn->kn_data++; 342 KNOTE_ACTIVATE(kn); 343 344 if ((kn->kn_flags & EV_ONESHOT) == 0) 345 filt_timer_timeout_add(kn); 346 } 347 348 349 /* 350 * data contains amount of time to sleep, in milliseconds 351 */ 352 int 353 filt_timerattach(struct knote *kn) 354 { 355 struct timeout *to; 356 357 if (kq_ntimeouts > kq_timeoutmax) 358 return (ENOMEM); 359 kq_ntimeouts++; 360 361 kn->kn_flags |= EV_CLEAR; /* automatically set */ 362 to = malloc(sizeof(*to), M_KEVENT, M_WAITOK); 363 timeout_set(to, filt_timerexpire, kn); 364 kn->kn_hook = to; 365 filt_timer_timeout_add(kn); 366 367 return (0); 368 } 369 370 void 371 filt_timerdetach(struct knote *kn) 372 { 373 struct timeout *to; 374 375 to = (struct timeout *)kn->kn_hook; 376 timeout_del(to); 377 free(to, M_KEVENT, sizeof(*to)); 378 kq_ntimeouts--; 379 } 380 381 int 382 filt_timer(struct knote *kn, long hint) 383 { 384 return (kn->kn_data != 0); 385 } 386 387 388 /* 389 * filt_seltrue: 390 * 391 * This filter "event" routine simulates seltrue(). 392 */ 393 int 394 filt_seltrue(struct knote *kn, long hint) 395 { 396 397 /* 398 * We don't know how much data can be read/written, 399 * but we know that it *can* be. This is about as 400 * good as select/poll does as well. 401 */ 402 kn->kn_data = 0; 403 return (1); 404 } 405 406 /* 407 * This provides full kqfilter entry for device switch tables, which 408 * has same effect as filter using filt_seltrue() as filter method. 409 */ 410 void 411 filt_seltruedetach(struct knote *kn) 412 { 413 /* Nothing to do */ 414 } 415 416 const struct filterops seltrue_filtops = 417 { 1, NULL, filt_seltruedetach, filt_seltrue }; 418 419 int 420 seltrue_kqfilter(dev_t dev, struct knote *kn) 421 { 422 switch (kn->kn_filter) { 423 case EVFILT_READ: 424 case EVFILT_WRITE: 425 kn->kn_fop = &seltrue_filtops; 426 break; 427 default: 428 return (EINVAL); 429 } 430 431 /* Nothing more to do */ 432 return (0); 433 } 434 435 int 436 sys_kqueue(struct proc *p, void *v, register_t *retval) 437 { 438 struct filedesc *fdp = p->p_fd; 439 struct kqueue *kq; 440 struct file *fp; 441 int fd, error; 442 443 fdplock(fdp); 444 error = falloc(p, 0, &fp, &fd); 445 fdpunlock(fdp); 446 if (error) 447 return (error); 448 fp->f_flag = FREAD | FWRITE; 449 fp->f_type = DTYPE_KQUEUE; 450 fp->f_ops = &kqueueops; 451 kq = pool_get(&kqueue_pool, PR_WAITOK|PR_ZERO); 452 TAILQ_INIT(&kq->kq_head); 453 fp->f_data = kq; 454 KQREF(kq); 455 *retval = fd; 456 if (fdp->fd_knlistsize < 0) 457 fdp->fd_knlistsize = 0; /* this process has a kq */ 458 kq->kq_fdp = fdp; 459 FILE_SET_MATURE(fp, p); 460 return (0); 461 } 462 463 int 464 sys_kevent(struct proc *p, void *v, register_t *retval) 465 { 466 struct filedesc* fdp = p->p_fd; 467 struct sys_kevent_args /* { 468 syscallarg(int) fd; 469 syscallarg(const struct kevent *) changelist; 470 syscallarg(int) nchanges; 471 syscallarg(struct kevent *) eventlist; 472 syscallarg(int) nevents; 473 syscallarg(const struct timespec *) timeout; 474 } */ *uap = v; 475 struct kevent *kevp; 476 struct kqueue *kq; 477 struct file *fp; 478 struct timespec ts; 479 int i, n, nerrors, error; 480 struct kevent kev[KQ_NEVENTS]; 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 = 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), kev, 506 n * sizeof(struct kevent)); 507 if (error) 508 goto done; 509 #ifdef KTRACE 510 if (KTRPOINT(p, KTR_STRUCT)) 511 ktrevent(p, kev, n); 512 #endif 513 for (i = 0; i < n; i++) { 514 kevp = &kev[i]; 515 kevp->flags &= ~EV_SYSFLAGS; 516 error = kqueue_register(kq, kevp, p); 517 if (error || (kevp->flags & EV_RECEIPT)) { 518 if (SCARG(uap, nevents) != 0) { 519 kevp->flags = EV_ERROR; 520 kevp->data = error; 521 copyout(kevp, SCARG(uap, eventlist), 522 sizeof(*kevp)); 523 SCARG(uap, eventlist)++; 524 SCARG(uap, nevents)--; 525 nerrors++; 526 } else { 527 goto done; 528 } 529 } 530 } 531 SCARG(uap, nchanges) -= n; 532 SCARG(uap, changelist) += n; 533 } 534 if (nerrors) { 535 *retval = nerrors; 536 error = 0; 537 goto done; 538 } 539 540 KQREF(kq); 541 FRELE(fp, p); 542 error = kqueue_scan(kq, SCARG(uap, nevents), SCARG(uap, eventlist), 543 SCARG(uap, timeout), p, &n); 544 KQRELE(kq); 545 *retval = n; 546 return (error); 547 548 done: 549 FRELE(fp, p); 550 return (error); 551 } 552 553 int 554 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p) 555 { 556 struct filedesc *fdp = kq->kq_fdp; 557 struct filterops *fops = NULL; 558 struct file *fp = NULL; 559 struct knote *kn = NULL; 560 int s, error = 0; 561 562 if (kev->filter < 0) { 563 if (kev->filter + EVFILT_SYSCOUNT < 0) 564 return (EINVAL); 565 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 566 } 567 568 if (fops == NULL) { 569 /* 570 * XXX 571 * filter attach routine is responsible for ensuring that 572 * the identifier can be attached to it. 573 */ 574 return (EINVAL); 575 } 576 577 if (fops->f_isfd) { 578 /* validate descriptor */ 579 if (kev->ident > INT_MAX) 580 return (EBADF); 581 if ((fp = fd_getfile(fdp, kev->ident)) == NULL) 582 return (EBADF); 583 FREF(fp); 584 585 if (kev->ident < fdp->fd_knlistsize) { 586 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) { 587 if (kq == kn->kn_kq && 588 kev->filter == kn->kn_filter) 589 break; 590 } 591 } 592 } else { 593 if (fdp->fd_knhashmask != 0) { 594 struct klist *list; 595 596 list = &fdp->fd_knhash[ 597 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 598 SLIST_FOREACH(kn, list, kn_link) { 599 if (kev->ident == kn->kn_id && 600 kq == kn->kn_kq && 601 kev->filter == kn->kn_filter) 602 break; 603 } 604 } 605 } 606 607 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 608 error = ENOENT; 609 goto done; 610 } 611 612 /* 613 * kn now contains the matching knote, or NULL if no match 614 */ 615 if (kev->flags & EV_ADD) { 616 617 if (kn == NULL) { 618 kn = knote_alloc(); 619 if (kn == NULL) { 620 error = ENOMEM; 621 goto done; 622 } 623 kn->kn_fp = fp; 624 kn->kn_kq = kq; 625 kn->kn_fop = fops; 626 627 /* 628 * apply reference count to knote structure, and 629 * do not release it at the end of this routine. 630 */ 631 fp = NULL; 632 633 kn->kn_sfflags = kev->fflags; 634 kn->kn_sdata = kev->data; 635 kev->fflags = 0; 636 kev->data = 0; 637 kn->kn_kevent = *kev; 638 639 knote_attach(kn, fdp); 640 if ((error = fops->f_attach(kn)) != 0) { 641 knote_drop(kn, p, fdp); 642 goto done; 643 } 644 } else { 645 /* 646 * The user may change some filter values after the 647 * initial EV_ADD, but doing so will not reset any 648 * filters which have already been triggered. 649 */ 650 kn->kn_sfflags = kev->fflags; 651 kn->kn_sdata = kev->data; 652 kn->kn_kevent.udata = kev->udata; 653 } 654 655 s = splhigh(); 656 if (kn->kn_fop->f_event(kn, 0)) 657 KNOTE_ACTIVATE(kn); 658 splx(s); 659 660 } else if (kev->flags & EV_DELETE) { 661 kn->kn_fop->f_detach(kn); 662 knote_drop(kn, p, p->p_fd); 663 goto done; 664 } 665 666 if ((kev->flags & EV_DISABLE) && 667 ((kn->kn_status & KN_DISABLED) == 0)) { 668 s = splhigh(); 669 kn->kn_status |= KN_DISABLED; 670 splx(s); 671 } 672 673 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 674 s = splhigh(); 675 kn->kn_status &= ~KN_DISABLED; 676 if ((kn->kn_status & KN_ACTIVE) && 677 ((kn->kn_status & KN_QUEUED) == 0)) 678 knote_enqueue(kn); 679 splx(s); 680 } 681 682 done: 683 if (fp != NULL) 684 FRELE(fp, p); 685 return (error); 686 } 687 688 int 689 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent *ulistp, 690 const struct timespec *tsp, struct proc *p, int *retval) 691 { 692 struct kevent *kevp; 693 struct timeval atv, rtv, ttv; 694 struct knote *kn, marker; 695 int s, count, timeout, nkev = 0, error = 0; 696 struct kevent kev[KQ_NEVENTS]; 697 698 count = maxevents; 699 if (count == 0) 700 goto done; 701 702 if (tsp != NULL) { 703 TIMESPEC_TO_TIMEVAL(&atv, tsp); 704 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 705 /* No timeout, just poll */ 706 timeout = -1; 707 goto start; 708 } 709 if (itimerfix(&atv)) { 710 error = EINVAL; 711 goto done; 712 } 713 714 timeout = atv.tv_sec > 24 * 60 * 60 ? 715 24 * 60 * 60 * hz : tvtohz(&atv); 716 717 getmicrouptime(&rtv); 718 timeradd(&atv, &rtv, &atv); 719 } else { 720 atv.tv_sec = 0; 721 atv.tv_usec = 0; 722 timeout = 0; 723 } 724 goto start; 725 726 retry: 727 if (atv.tv_sec || atv.tv_usec) { 728 getmicrouptime(&rtv); 729 if (timercmp(&rtv, &atv, >=)) 730 goto done; 731 ttv = atv; 732 timersub(&ttv, &rtv, &ttv); 733 timeout = ttv.tv_sec > 24 * 60 * 60 ? 734 24 * 60 * 60 * hz : tvtohz(&ttv); 735 } 736 737 start: 738 if (kq->kq_state & KQ_DYING) { 739 error = EBADF; 740 goto done; 741 } 742 743 kevp = &kev[0]; 744 s = splhigh(); 745 if (kq->kq_count == 0) { 746 if (timeout < 0) { 747 error = EWOULDBLOCK; 748 } else { 749 kq->kq_state |= KQ_SLEEP; 750 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 751 } 752 splx(s); 753 if (error == 0) 754 goto retry; 755 /* don't restart after signals... */ 756 if (error == ERESTART) 757 error = EINTR; 758 else if (error == EWOULDBLOCK) 759 error = 0; 760 goto done; 761 } 762 763 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 764 while (count) { 765 kn = TAILQ_FIRST(&kq->kq_head); 766 if (kn == &marker) { 767 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 768 splx(s); 769 if (count == maxevents) 770 goto retry; 771 goto done; 772 } 773 774 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 775 kq->kq_count--; 776 777 if (kn->kn_status & KN_DISABLED) { 778 kn->kn_status &= ~KN_QUEUED; 779 continue; 780 } 781 if ((kn->kn_flags & EV_ONESHOT) == 0 && 782 kn->kn_fop->f_event(kn, 0) == 0) { 783 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 784 continue; 785 } 786 *kevp = kn->kn_kevent; 787 kevp++; 788 nkev++; 789 if (kn->kn_flags & EV_ONESHOT) { 790 kn->kn_status &= ~KN_QUEUED; 791 splx(s); 792 kn->kn_fop->f_detach(kn); 793 knote_drop(kn, p, p->p_fd); 794 s = splhigh(); 795 } else if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 796 if (kn->kn_flags & EV_CLEAR) { 797 kn->kn_data = 0; 798 kn->kn_fflags = 0; 799 } 800 if (kn->kn_flags & EV_DISPATCH) 801 kn->kn_status |= KN_DISABLED; 802 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 803 } else { 804 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 805 kq->kq_count++; 806 } 807 count--; 808 if (nkev == KQ_NEVENTS) { 809 splx(s); 810 #ifdef KTRACE 811 if (KTRPOINT(p, KTR_STRUCT)) 812 ktrevent(p, kev, nkev); 813 #endif 814 error = copyout(kev, ulistp, 815 sizeof(struct kevent) * nkev); 816 ulistp += nkev; 817 nkev = 0; 818 kevp = &kev[0]; 819 s = splhigh(); 820 if (error) 821 break; 822 } 823 } 824 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 825 splx(s); 826 done: 827 if (nkev != 0) { 828 #ifdef KTRACE 829 if (KTRPOINT(p, KTR_STRUCT)) 830 ktrevent(p, kev, nkev); 831 #endif 832 error = copyout(kev, ulistp, 833 sizeof(struct kevent) * nkev); 834 } 835 *retval = maxevents - count; 836 return (error); 837 } 838 839 /* 840 * XXX 841 * This could be expanded to call kqueue_scan, if desired. 842 */ 843 int 844 kqueue_read(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 845 { 846 return (ENXIO); 847 } 848 849 int 850 kqueue_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 851 852 { 853 return (ENXIO); 854 } 855 856 int 857 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p) 858 { 859 return (ENOTTY); 860 } 861 862 int 863 kqueue_poll(struct file *fp, int events, struct proc *p) 864 { 865 struct kqueue *kq = (struct kqueue *)fp->f_data; 866 int revents = 0; 867 int s = splhigh(); 868 869 if (events & (POLLIN | POLLRDNORM)) { 870 if (kq->kq_count) { 871 revents |= events & (POLLIN | POLLRDNORM); 872 } else { 873 selrecord(p, &kq->kq_sel); 874 kq->kq_state |= KQ_SEL; 875 } 876 } 877 splx(s); 878 return (revents); 879 } 880 881 int 882 kqueue_stat(struct file *fp, struct stat *st, struct proc *p) 883 { 884 struct kqueue *kq = fp->f_data; 885 886 memset(st, 0, sizeof(*st)); 887 st->st_size = kq->kq_count; 888 st->st_blksize = sizeof(struct kevent); 889 st->st_mode = S_IFIFO; 890 return (0); 891 } 892 893 int 894 kqueue_close(struct file *fp, struct proc *p) 895 { 896 struct kqueue *kq = fp->f_data; 897 struct filedesc *fdp = p->p_fd; 898 struct knote **knp, *kn, *kn0; 899 int i; 900 901 for (i = 0; i < fdp->fd_knlistsize; i++) { 902 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 903 kn = *knp; 904 while (kn != NULL) { 905 kn0 = SLIST_NEXT(kn, kn_link); 906 if (kq == kn->kn_kq) { 907 kn->kn_fop->f_detach(kn); 908 FRELE(kn->kn_fp, p); 909 knote_free(kn); 910 *knp = kn0; 911 } else { 912 knp = &SLIST_NEXT(kn, kn_link); 913 } 914 kn = kn0; 915 } 916 } 917 if (fdp->fd_knhashmask != 0) { 918 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 919 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 920 kn = *knp; 921 while (kn != NULL) { 922 kn0 = SLIST_NEXT(kn, kn_link); 923 if (kq == kn->kn_kq) { 924 kn->kn_fop->f_detach(kn); 925 /* XXX non-fd release of kn->kn_ptr */ 926 knote_free(kn); 927 *knp = kn0; 928 } else { 929 knp = &SLIST_NEXT(kn, kn_link); 930 } 931 kn = kn0; 932 } 933 } 934 } 935 fp->f_data = NULL; 936 937 kq->kq_state |= KQ_DYING; 938 kqueue_wakeup(kq); 939 KQRELE(kq); 940 941 return (0); 942 } 943 944 void 945 kqueue_wakeup(struct kqueue *kq) 946 { 947 948 if (kq->kq_state & KQ_SLEEP) { 949 kq->kq_state &= ~KQ_SLEEP; 950 wakeup(kq); 951 } 952 if (kq->kq_state & KQ_SEL) { 953 kq->kq_state &= ~KQ_SEL; 954 selwakeup(&kq->kq_sel); 955 } else 956 KNOTE(&kq->kq_sel.si_note, 0); 957 } 958 959 /* 960 * activate one knote. 961 */ 962 void 963 knote_activate(struct knote *kn) 964 { 965 KNOTE_ACTIVATE(kn); 966 } 967 968 /* 969 * walk down a list of knotes, activating them if their event has triggered. 970 */ 971 void 972 knote(struct klist *list, long hint) 973 { 974 struct knote *kn, *kn0; 975 976 SLIST_FOREACH_SAFE(kn, list, kn_selnext, kn0) 977 if (kn->kn_fop->f_event(kn, hint)) 978 KNOTE_ACTIVATE(kn); 979 } 980 981 /* 982 * remove all knotes from a specified klist 983 */ 984 void 985 knote_remove(struct proc *p, struct klist *list) 986 { 987 struct knote *kn; 988 989 while ((kn = SLIST_FIRST(list)) != NULL) { 990 kn->kn_fop->f_detach(kn); 991 knote_drop(kn, p, p->p_fd); 992 } 993 } 994 995 /* 996 * remove all knotes referencing a specified fd 997 */ 998 void 999 knote_fdclose(struct proc *p, int fd) 1000 { 1001 struct filedesc *fdp = p->p_fd; 1002 struct klist *list = &fdp->fd_knlist[fd]; 1003 1004 knote_remove(p, list); 1005 } 1006 1007 /* 1008 * handle a process exiting, including the triggering of NOTE_EXIT notes 1009 * XXX this could be more efficient, doing a single pass down the klist 1010 */ 1011 void 1012 knote_processexit(struct proc *p) 1013 { 1014 struct process *pr = p->p_p; 1015 1016 KNOTE(&pr->ps_klist, NOTE_EXIT); 1017 1018 /* remove other knotes hanging off the process */ 1019 knote_remove(p, &pr->ps_klist); 1020 } 1021 1022 void 1023 knote_attach(struct knote *kn, struct filedesc *fdp) 1024 { 1025 struct klist *list; 1026 int size; 1027 1028 if (!kn->kn_fop->f_isfd) { 1029 if (fdp->fd_knhashmask == 0) 1030 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP, 1031 M_WAITOK, &fdp->fd_knhashmask); 1032 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1033 goto done; 1034 } 1035 1036 if (fdp->fd_knlistsize <= kn->kn_id) { 1037 size = fdp->fd_knlistsize; 1038 while (size <= kn->kn_id) 1039 size += KQEXTENT; 1040 list = mallocarray(size, sizeof(struct klist), M_TEMP, 1041 M_WAITOK); 1042 memcpy(list, fdp->fd_knlist, 1043 fdp->fd_knlistsize * sizeof(struct klist)); 1044 memset(&list[fdp->fd_knlistsize], 0, 1045 (size - fdp->fd_knlistsize) * sizeof(struct klist)); 1046 free(fdp->fd_knlist, M_TEMP, 1047 fdp->fd_knlistsize * sizeof(struct klist)); 1048 fdp->fd_knlistsize = size; 1049 fdp->fd_knlist = list; 1050 } 1051 list = &fdp->fd_knlist[kn->kn_id]; 1052 done: 1053 SLIST_INSERT_HEAD(list, kn, kn_link); 1054 kn->kn_status = 0; 1055 } 1056 1057 /* 1058 * should be called at spl == 0, since we don't want to hold spl 1059 * while calling FRELE and knote_free. 1060 */ 1061 void 1062 knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp) 1063 { 1064 struct klist *list; 1065 1066 if (kn->kn_fop->f_isfd) 1067 list = &fdp->fd_knlist[kn->kn_id]; 1068 else 1069 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1070 1071 SLIST_REMOVE(list, kn, knote, kn_link); 1072 if (kn->kn_status & KN_QUEUED) 1073 knote_dequeue(kn); 1074 if (kn->kn_fop->f_isfd) 1075 FRELE(kn->kn_fp, p); 1076 knote_free(kn); 1077 } 1078 1079 1080 void 1081 knote_enqueue(struct knote *kn) 1082 { 1083 struct kqueue *kq = kn->kn_kq; 1084 int s = splhigh(); 1085 1086 KASSERT((kn->kn_status & KN_QUEUED) == 0); 1087 1088 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1089 kn->kn_status |= KN_QUEUED; 1090 kq->kq_count++; 1091 splx(s); 1092 kqueue_wakeup(kq); 1093 } 1094 1095 void 1096 knote_dequeue(struct knote *kn) 1097 { 1098 struct kqueue *kq = kn->kn_kq; 1099 int s = splhigh(); 1100 1101 KASSERT(kn->kn_status & KN_QUEUED); 1102 1103 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1104 kn->kn_status &= ~KN_QUEUED; 1105 kq->kq_count--; 1106 splx(s); 1107 } 1108 1109 void 1110 klist_invalidate(struct klist *list) 1111 { 1112 struct knote *kn; 1113 1114 SLIST_FOREACH(kn, list, kn_selnext) { 1115 kn->kn_status |= KN_DETACHED; 1116 kn->kn_flags |= EV_EOF | EV_ONESHOT; 1117 } 1118 } 1119