1 /* $OpenBSD: kern_event.c,v 1.76 2016/09/15 02:00:16 dlg 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_HASHSIZE 64 /* XXX should be tunable */ 123 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 124 125 extern struct filterops sig_filtops; 126 #ifdef notyet 127 extern struct filterops aio_filtops; 128 #endif 129 130 /* 131 * Table for for all system-defined filters. 132 */ 133 struct filterops *sysfilt_ops[] = { 134 &file_filtops, /* EVFILT_READ */ 135 &file_filtops, /* EVFILT_WRITE */ 136 NULL, /*&aio_filtops,*/ /* EVFILT_AIO */ 137 &file_filtops, /* EVFILT_VNODE */ 138 &proc_filtops, /* EVFILT_PROC */ 139 &sig_filtops, /* EVFILT_SIGNAL */ 140 &timer_filtops, /* EVFILT_TIMER */ 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, &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; 480 481 if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL || 482 (fp->f_type != DTYPE_KQUEUE)) 483 return (EBADF); 484 485 FREF(fp); 486 487 if (SCARG(uap, timeout) != NULL) { 488 error = copyin(SCARG(uap, timeout), &ts, sizeof(ts)); 489 if (error) 490 goto done; 491 #ifdef KTRACE 492 if (KTRPOINT(p, KTR_STRUCT)) 493 ktrreltimespec(p, &ts); 494 #endif 495 SCARG(uap, timeout) = &ts; 496 } 497 498 kq = fp->f_data; 499 nerrors = 0; 500 501 while (SCARG(uap, nchanges) > 0) { 502 n = SCARG(uap, nchanges) > KQ_NEVENTS ? 503 KQ_NEVENTS : SCARG(uap, nchanges); 504 error = copyin(SCARG(uap, changelist), kq->kq_kev, 505 n * sizeof(struct kevent)); 506 if (error) 507 goto done; 508 #ifdef KTRACE 509 if (KTRPOINT(p, KTR_STRUCT)) 510 ktrevent(p, kq->kq_kev, n); 511 #endif 512 for (i = 0; i < n; i++) { 513 kevp = &kq->kq_kev[i]; 514 kevp->flags &= ~EV_SYSFLAGS; 515 error = kqueue_register(kq, kevp, p); 516 if (error) { 517 if (SCARG(uap, nevents) != 0) { 518 kevp->flags = EV_ERROR; 519 kevp->data = error; 520 copyout(kevp, SCARG(uap, eventlist), 521 sizeof(*kevp)); 522 SCARG(uap, eventlist)++; 523 SCARG(uap, nevents)--; 524 nerrors++; 525 } else { 526 goto done; 527 } 528 } 529 } 530 SCARG(uap, nchanges) -= n; 531 SCARG(uap, changelist) += n; 532 } 533 if (nerrors) { 534 *retval = nerrors; 535 error = 0; 536 goto done; 537 } 538 539 KQREF(kq); 540 FRELE(fp, p); 541 error = kqueue_scan(kq, SCARG(uap, nevents), SCARG(uap, eventlist), 542 SCARG(uap, timeout), p, &n); 543 KQRELE(kq); 544 *retval = n; 545 return (error); 546 547 done: 548 FRELE(fp, p); 549 return (error); 550 } 551 552 int 553 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p) 554 { 555 struct filedesc *fdp = kq->kq_fdp; 556 struct filterops *fops = NULL; 557 struct file *fp = NULL; 558 struct knote *kn = NULL; 559 int s, error = 0; 560 561 if (kev->filter < 0) { 562 if (kev->filter + EVFILT_SYSCOUNT < 0) 563 return (EINVAL); 564 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 565 } 566 567 if (fops == NULL) { 568 /* 569 * XXX 570 * filter attach routine is responsible for ensuring that 571 * the identifier can be attached to it. 572 */ 573 return (EINVAL); 574 } 575 576 if (fops->f_isfd) { 577 /* validate descriptor */ 578 if (kev->ident > INT_MAX) 579 return (EBADF); 580 if ((fp = fd_getfile(fdp, kev->ident)) == NULL) 581 return (EBADF); 582 FREF(fp); 583 584 if (kev->ident < fdp->fd_knlistsize) { 585 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) { 586 if (kq == kn->kn_kq && 587 kev->filter == kn->kn_filter) 588 break; 589 } 590 } 591 } else { 592 if (fdp->fd_knhashmask != 0) { 593 struct klist *list; 594 595 list = &fdp->fd_knhash[ 596 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 597 SLIST_FOREACH(kn, list, kn_link) { 598 if (kev->ident == kn->kn_id && 599 kq == kn->kn_kq && 600 kev->filter == kn->kn_filter) 601 break; 602 } 603 } 604 } 605 606 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 607 error = ENOENT; 608 goto done; 609 } 610 611 /* 612 * kn now contains the matching knote, or NULL if no match 613 */ 614 if (kev->flags & EV_ADD) { 615 616 if (kn == NULL) { 617 kn = knote_alloc(); 618 if (kn == NULL) { 619 error = ENOMEM; 620 goto done; 621 } 622 kn->kn_fp = fp; 623 kn->kn_kq = kq; 624 kn->kn_fop = fops; 625 626 /* 627 * apply reference count to knote structure, and 628 * do not release it at the end of this routine. 629 */ 630 fp = NULL; 631 632 kn->kn_sfflags = kev->fflags; 633 kn->kn_sdata = kev->data; 634 kev->fflags = 0; 635 kev->data = 0; 636 kn->kn_kevent = *kev; 637 638 knote_attach(kn, fdp); 639 if ((error = fops->f_attach(kn)) != 0) { 640 knote_drop(kn, p, fdp); 641 goto done; 642 } 643 } else { 644 /* 645 * The user may change some filter values after the 646 * initial EV_ADD, but doing so will not reset any 647 * filters which have already been triggered. 648 */ 649 kn->kn_sfflags = kev->fflags; 650 kn->kn_sdata = kev->data; 651 kn->kn_kevent.udata = kev->udata; 652 } 653 654 s = splhigh(); 655 if (kn->kn_fop->f_event(kn, 0)) 656 KNOTE_ACTIVATE(kn); 657 splx(s); 658 659 } else if (kev->flags & EV_DELETE) { 660 kn->kn_fop->f_detach(kn); 661 knote_drop(kn, p, p->p_fd); 662 goto done; 663 } 664 665 if ((kev->flags & EV_DISABLE) && 666 ((kn->kn_status & KN_DISABLED) == 0)) { 667 s = splhigh(); 668 kn->kn_status |= KN_DISABLED; 669 splx(s); 670 } 671 672 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 673 s = splhigh(); 674 kn->kn_status &= ~KN_DISABLED; 675 if ((kn->kn_status & KN_ACTIVE) && 676 ((kn->kn_status & KN_QUEUED) == 0)) 677 knote_enqueue(kn); 678 splx(s); 679 } 680 681 done: 682 if (fp != NULL) 683 FRELE(fp, p); 684 return (error); 685 } 686 687 int 688 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent *ulistp, 689 const struct timespec *tsp, struct proc *p, int *retval) 690 { 691 struct kevent *kevp; 692 struct timeval atv, rtv, ttv; 693 struct knote *kn, marker; 694 int s, count, timeout, nkev = 0, error = 0; 695 696 count = maxevents; 697 if (count == 0) 698 goto done; 699 700 if (tsp != NULL) { 701 TIMESPEC_TO_TIMEVAL(&atv, tsp); 702 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 703 /* No timeout, just poll */ 704 timeout = -1; 705 goto start; 706 } 707 if (itimerfix(&atv)) { 708 error = EINVAL; 709 goto done; 710 } 711 712 timeout = atv.tv_sec > 24 * 60 * 60 ? 713 24 * 60 * 60 * hz : tvtohz(&atv); 714 715 getmicrouptime(&rtv); 716 timeradd(&atv, &rtv, &atv); 717 } else { 718 atv.tv_sec = 0; 719 atv.tv_usec = 0; 720 timeout = 0; 721 } 722 goto start; 723 724 retry: 725 if (atv.tv_sec || atv.tv_usec) { 726 getmicrouptime(&rtv); 727 if (timercmp(&rtv, &atv, >=)) 728 goto done; 729 ttv = atv; 730 timersub(&ttv, &rtv, &ttv); 731 timeout = ttv.tv_sec > 24 * 60 * 60 ? 732 24 * 60 * 60 * hz : tvtohz(&ttv); 733 } 734 735 start: 736 if (kq->kq_state & KQ_DYING) { 737 error = EBADF; 738 goto done; 739 } 740 741 kevp = kq->kq_kev; 742 s = splhigh(); 743 if (kq->kq_count == 0) { 744 if (timeout < 0) { 745 error = EWOULDBLOCK; 746 } else { 747 kq->kq_state |= KQ_SLEEP; 748 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 749 } 750 splx(s); 751 if (error == 0) 752 goto retry; 753 /* don't restart after signals... */ 754 if (error == ERESTART) 755 error = EINTR; 756 else if (error == EWOULDBLOCK) 757 error = 0; 758 goto done; 759 } 760 761 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 762 while (count) { 763 kn = TAILQ_FIRST(&kq->kq_head); 764 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 765 if (kn == &marker) { 766 splx(s); 767 if (count == maxevents) 768 goto retry; 769 goto done; 770 } 771 if (kn->kn_status & KN_DISABLED) { 772 kn->kn_status &= ~KN_QUEUED; 773 kq->kq_count--; 774 continue; 775 } 776 if ((kn->kn_flags & EV_ONESHOT) == 0 && 777 kn->kn_fop->f_event(kn, 0) == 0) { 778 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 779 kq->kq_count--; 780 continue; 781 } 782 *kevp = kn->kn_kevent; 783 kevp++; 784 nkev++; 785 if (kn->kn_flags & EV_ONESHOT) { 786 kn->kn_status &= ~KN_QUEUED; 787 kq->kq_count--; 788 splx(s); 789 kn->kn_fop->f_detach(kn); 790 knote_drop(kn, p, p->p_fd); 791 s = splhigh(); 792 } else if (kn->kn_flags & EV_CLEAR) { 793 kn->kn_data = 0; 794 kn->kn_fflags = 0; 795 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 796 kq->kq_count--; 797 } else { 798 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 799 } 800 count--; 801 if (nkev == KQ_NEVENTS) { 802 splx(s); 803 #ifdef KTRACE 804 if (KTRPOINT(p, KTR_STRUCT)) 805 ktrevent(p, kq->kq_kev, nkev); 806 #endif 807 error = copyout(kq->kq_kev, ulistp, 808 sizeof(struct kevent) * nkev); 809 ulistp += nkev; 810 nkev = 0; 811 kevp = kq->kq_kev; 812 s = splhigh(); 813 if (error) 814 break; 815 } 816 } 817 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 818 splx(s); 819 done: 820 if (nkev != 0) { 821 #ifdef KTRACE 822 if (KTRPOINT(p, KTR_STRUCT)) 823 ktrevent(p, kq->kq_kev, nkev); 824 #endif 825 error = copyout(kq->kq_kev, ulistp, 826 sizeof(struct kevent) * nkev); 827 } 828 *retval = maxevents - count; 829 return (error); 830 } 831 832 /* 833 * XXX 834 * This could be expanded to call kqueue_scan, if desired. 835 */ 836 int 837 kqueue_read(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 838 { 839 return (ENXIO); 840 } 841 842 int 843 kqueue_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 844 845 { 846 return (ENXIO); 847 } 848 849 int 850 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p) 851 { 852 return (ENOTTY); 853 } 854 855 int 856 kqueue_poll(struct file *fp, int events, struct proc *p) 857 { 858 struct kqueue *kq = (struct kqueue *)fp->f_data; 859 int revents = 0; 860 int s = splhigh(); 861 862 if (events & (POLLIN | POLLRDNORM)) { 863 if (kq->kq_count) { 864 revents |= events & (POLLIN | POLLRDNORM); 865 } else { 866 selrecord(p, &kq->kq_sel); 867 kq->kq_state |= KQ_SEL; 868 } 869 } 870 splx(s); 871 return (revents); 872 } 873 874 int 875 kqueue_stat(struct file *fp, struct stat *st, struct proc *p) 876 { 877 struct kqueue *kq = fp->f_data; 878 879 memset(st, 0, sizeof(*st)); 880 st->st_size = kq->kq_count; 881 st->st_blksize = sizeof(struct kevent); 882 st->st_mode = S_IFIFO; 883 return (0); 884 } 885 886 int 887 kqueue_close(struct file *fp, struct proc *p) 888 { 889 struct kqueue *kq = fp->f_data; 890 struct filedesc *fdp = p->p_fd; 891 struct knote **knp, *kn, *kn0; 892 int i; 893 894 for (i = 0; i < fdp->fd_knlistsize; i++) { 895 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 896 kn = *knp; 897 while (kn != NULL) { 898 kn0 = SLIST_NEXT(kn, kn_link); 899 if (kq == kn->kn_kq) { 900 kn->kn_fop->f_detach(kn); 901 FRELE(kn->kn_fp, p); 902 knote_free(kn); 903 *knp = kn0; 904 } else { 905 knp = &SLIST_NEXT(kn, kn_link); 906 } 907 kn = kn0; 908 } 909 } 910 if (fdp->fd_knhashmask != 0) { 911 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 912 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 913 kn = *knp; 914 while (kn != NULL) { 915 kn0 = SLIST_NEXT(kn, kn_link); 916 if (kq == kn->kn_kq) { 917 kn->kn_fop->f_detach(kn); 918 /* XXX non-fd release of kn->kn_ptr */ 919 knote_free(kn); 920 *knp = kn0; 921 } else { 922 knp = &SLIST_NEXT(kn, kn_link); 923 } 924 kn = kn0; 925 } 926 } 927 } 928 fp->f_data = NULL; 929 930 kq->kq_state |= KQ_DYING; 931 kqueue_wakeup(kq); 932 KQRELE(kq); 933 934 return (0); 935 } 936 937 void 938 kqueue_wakeup(struct kqueue *kq) 939 { 940 941 if (kq->kq_state & KQ_SLEEP) { 942 kq->kq_state &= ~KQ_SLEEP; 943 wakeup(kq); 944 } 945 if (kq->kq_state & KQ_SEL) { 946 kq->kq_state &= ~KQ_SEL; 947 selwakeup(&kq->kq_sel); 948 } else 949 KNOTE(&kq->kq_sel.si_note, 0); 950 } 951 952 /* 953 * activate one knote. 954 */ 955 void 956 knote_activate(struct knote *kn) 957 { 958 KNOTE_ACTIVATE(kn); 959 } 960 961 /* 962 * walk down a list of knotes, activating them if their event has triggered. 963 */ 964 void 965 knote(struct klist *list, long hint) 966 { 967 struct knote *kn, *kn0; 968 969 SLIST_FOREACH_SAFE(kn, list, kn_selnext, kn0) 970 if (kn->kn_fop->f_event(kn, hint)) 971 KNOTE_ACTIVATE(kn); 972 } 973 974 /* 975 * remove all knotes from a specified klist 976 */ 977 void 978 knote_remove(struct proc *p, struct klist *list) 979 { 980 struct knote *kn; 981 982 while ((kn = SLIST_FIRST(list)) != NULL) { 983 kn->kn_fop->f_detach(kn); 984 knote_drop(kn, p, p->p_fd); 985 } 986 } 987 988 /* 989 * remove all knotes referencing a specified fd 990 */ 991 void 992 knote_fdclose(struct proc *p, int fd) 993 { 994 struct filedesc *fdp = p->p_fd; 995 struct klist *list = &fdp->fd_knlist[fd]; 996 997 knote_remove(p, list); 998 } 999 1000 /* 1001 * handle a process exiting, including the triggering of NOTE_EXIT notes 1002 * XXX this could be more efficient, doing a single pass down the klist 1003 */ 1004 void 1005 knote_processexit(struct proc *p) 1006 { 1007 struct process *pr = p->p_p; 1008 1009 KNOTE(&pr->ps_klist, NOTE_EXIT); 1010 1011 /* remove other knotes hanging off the process */ 1012 knote_remove(p, &pr->ps_klist); 1013 } 1014 1015 void 1016 knote_attach(struct knote *kn, struct filedesc *fdp) 1017 { 1018 struct klist *list; 1019 int size; 1020 1021 if (!kn->kn_fop->f_isfd) { 1022 if (fdp->fd_knhashmask == 0) 1023 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP, 1024 M_WAITOK, &fdp->fd_knhashmask); 1025 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1026 goto done; 1027 } 1028 1029 if (fdp->fd_knlistsize <= kn->kn_id) { 1030 size = fdp->fd_knlistsize; 1031 while (size <= kn->kn_id) 1032 size += KQEXTENT; 1033 list = mallocarray(size, sizeof(struct klist), M_TEMP, 1034 M_WAITOK); 1035 memcpy(list, fdp->fd_knlist, 1036 fdp->fd_knlistsize * sizeof(struct klist)); 1037 memset(&list[fdp->fd_knlistsize], 0, 1038 (size - fdp->fd_knlistsize) * sizeof(struct klist)); 1039 free(fdp->fd_knlist, M_TEMP, 1040 fdp->fd_knlistsize * sizeof(struct klist)); 1041 fdp->fd_knlistsize = size; 1042 fdp->fd_knlist = list; 1043 } 1044 list = &fdp->fd_knlist[kn->kn_id]; 1045 done: 1046 SLIST_INSERT_HEAD(list, kn, kn_link); 1047 kn->kn_status = 0; 1048 } 1049 1050 /* 1051 * should be called at spl == 0, since we don't want to hold spl 1052 * while calling FRELE and knote_free. 1053 */ 1054 void 1055 knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp) 1056 { 1057 struct klist *list; 1058 1059 if (kn->kn_fop->f_isfd) 1060 list = &fdp->fd_knlist[kn->kn_id]; 1061 else 1062 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1063 1064 SLIST_REMOVE(list, kn, knote, kn_link); 1065 if (kn->kn_status & KN_QUEUED) 1066 knote_dequeue(kn); 1067 if (kn->kn_fop->f_isfd) 1068 FRELE(kn->kn_fp, p); 1069 knote_free(kn); 1070 } 1071 1072 1073 void 1074 knote_enqueue(struct knote *kn) 1075 { 1076 struct kqueue *kq = kn->kn_kq; 1077 int s = splhigh(); 1078 1079 KASSERT((kn->kn_status & KN_QUEUED) == 0); 1080 1081 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1082 kn->kn_status |= KN_QUEUED; 1083 kq->kq_count++; 1084 splx(s); 1085 kqueue_wakeup(kq); 1086 } 1087 1088 void 1089 knote_dequeue(struct knote *kn) 1090 { 1091 struct kqueue *kq = kn->kn_kq; 1092 int s = splhigh(); 1093 1094 KASSERT(kn->kn_status & KN_QUEUED); 1095 1096 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1097 kn->kn_status &= ~KN_QUEUED; 1098 kq->kq_count--; 1099 splx(s); 1100 } 1101 1102 void 1103 klist_invalidate(struct klist *list) 1104 { 1105 struct knote *kn; 1106 1107 SLIST_FOREACH(kn, list, kn_selnext) { 1108 kn->kn_status |= KN_DETACHED; 1109 kn->kn_flags |= EV_EOF | EV_ONESHOT; 1110 } 1111 } 1112