1 /* $OpenBSD: kern_event.c,v 1.85 2018/04/03 09:10:02 mpi 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 return (EBADF); 484 FREF(fp); 485 486 if (fp->f_type != DTYPE_KQUEUE) { 487 error = EBADF; 488 goto done; 489 } 490 491 if (SCARG(uap, timeout) != NULL) { 492 error = copyin(SCARG(uap, timeout), &ts, sizeof(ts)); 493 if (error) 494 goto done; 495 #ifdef KTRACE 496 if (KTRPOINT(p, KTR_STRUCT)) 497 ktrreltimespec(p, &ts); 498 #endif 499 SCARG(uap, timeout) = &ts; 500 } 501 502 kq = fp->f_data; 503 nerrors = 0; 504 505 while (SCARG(uap, nchanges) > 0) { 506 n = SCARG(uap, nchanges) > KQ_NEVENTS ? 507 KQ_NEVENTS : SCARG(uap, nchanges); 508 error = copyin(SCARG(uap, changelist), kev, 509 n * sizeof(struct kevent)); 510 if (error) 511 goto done; 512 #ifdef KTRACE 513 if (KTRPOINT(p, KTR_STRUCT)) 514 ktrevent(p, kev, n); 515 #endif 516 for (i = 0; i < n; i++) { 517 kevp = &kev[i]; 518 kevp->flags &= ~EV_SYSFLAGS; 519 error = kqueue_register(kq, kevp, p); 520 if (error || (kevp->flags & EV_RECEIPT)) { 521 if (SCARG(uap, nevents) != 0) { 522 kevp->flags = EV_ERROR; 523 kevp->data = error; 524 copyout(kevp, SCARG(uap, eventlist), 525 sizeof(*kevp)); 526 SCARG(uap, eventlist)++; 527 SCARG(uap, nevents)--; 528 nerrors++; 529 } else { 530 goto done; 531 } 532 } 533 } 534 SCARG(uap, nchanges) -= n; 535 SCARG(uap, changelist) += n; 536 } 537 if (nerrors) { 538 *retval = nerrors; 539 error = 0; 540 goto done; 541 } 542 543 KQREF(kq); 544 FRELE(fp, p); 545 error = kqueue_scan(kq, SCARG(uap, nevents), SCARG(uap, eventlist), 546 SCARG(uap, timeout), p, &n); 547 KQRELE(kq); 548 *retval = n; 549 return (error); 550 551 done: 552 FRELE(fp, p); 553 return (error); 554 } 555 556 int 557 kqueue_register(struct kqueue *kq, struct kevent *kev, struct proc *p) 558 { 559 struct filedesc *fdp = kq->kq_fdp; 560 struct filterops *fops = NULL; 561 struct file *fp = NULL; 562 struct knote *kn = NULL; 563 int s, error = 0; 564 565 if (kev->filter < 0) { 566 if (kev->filter + EVFILT_SYSCOUNT < 0) 567 return (EINVAL); 568 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 569 } 570 571 if (fops == NULL) { 572 /* 573 * XXX 574 * filter attach routine is responsible for ensuring that 575 * the identifier can be attached to it. 576 */ 577 return (EINVAL); 578 } 579 580 if (fops->f_isfd) { 581 /* validate descriptor */ 582 if (kev->ident > INT_MAX) 583 return (EBADF); 584 if ((fp = fd_getfile(fdp, kev->ident)) == NULL) 585 return (EBADF); 586 FREF(fp); 587 588 if (kev->ident < fdp->fd_knlistsize) { 589 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) { 590 if (kq == kn->kn_kq && 591 kev->filter == kn->kn_filter) 592 break; 593 } 594 } 595 } else { 596 if (fdp->fd_knhashmask != 0) { 597 struct klist *list; 598 599 list = &fdp->fd_knhash[ 600 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 601 SLIST_FOREACH(kn, list, kn_link) { 602 if (kev->ident == kn->kn_id && 603 kq == kn->kn_kq && 604 kev->filter == kn->kn_filter) 605 break; 606 } 607 } 608 } 609 610 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 611 error = ENOENT; 612 goto done; 613 } 614 615 /* 616 * kn now contains the matching knote, or NULL if no match 617 */ 618 if (kev->flags & EV_ADD) { 619 620 if (kn == NULL) { 621 kn = knote_alloc(); 622 if (kn == NULL) { 623 error = ENOMEM; 624 goto done; 625 } 626 kn->kn_fp = fp; 627 kn->kn_kq = kq; 628 kn->kn_fop = fops; 629 630 /* 631 * apply reference count to knote structure, and 632 * do not release it at the end of this routine. 633 */ 634 fp = NULL; 635 636 kn->kn_sfflags = kev->fflags; 637 kn->kn_sdata = kev->data; 638 kev->fflags = 0; 639 kev->data = 0; 640 kn->kn_kevent = *kev; 641 642 knote_attach(kn, fdp); 643 if ((error = fops->f_attach(kn)) != 0) { 644 knote_drop(kn, p, fdp); 645 goto done; 646 } 647 } else { 648 /* 649 * The user may change some filter values after the 650 * initial EV_ADD, but doing so will not reset any 651 * filters which have already been triggered. 652 */ 653 kn->kn_sfflags = kev->fflags; 654 kn->kn_sdata = kev->data; 655 kn->kn_kevent.udata = kev->udata; 656 } 657 658 s = splhigh(); 659 if (kn->kn_fop->f_event(kn, 0)) 660 KNOTE_ACTIVATE(kn); 661 splx(s); 662 663 } else if (kev->flags & EV_DELETE) { 664 kn->kn_fop->f_detach(kn); 665 knote_drop(kn, p, p->p_fd); 666 goto done; 667 } 668 669 if ((kev->flags & EV_DISABLE) && 670 ((kn->kn_status & KN_DISABLED) == 0)) { 671 s = splhigh(); 672 kn->kn_status |= KN_DISABLED; 673 splx(s); 674 } 675 676 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 677 s = splhigh(); 678 kn->kn_status &= ~KN_DISABLED; 679 if ((kn->kn_status & KN_ACTIVE) && 680 ((kn->kn_status & KN_QUEUED) == 0)) 681 knote_enqueue(kn); 682 splx(s); 683 } 684 685 done: 686 if (fp != NULL) 687 FRELE(fp, p); 688 return (error); 689 } 690 691 int 692 kqueue_scan(struct kqueue *kq, int maxevents, struct kevent *ulistp, 693 const struct timespec *tsp, struct proc *p, int *retval) 694 { 695 struct kevent *kevp; 696 struct timeval atv, rtv, ttv; 697 struct knote *kn, marker; 698 int s, count, timeout, nkev = 0, error = 0; 699 struct kevent kev[KQ_NEVENTS]; 700 701 count = maxevents; 702 if (count == 0) 703 goto done; 704 705 if (tsp != NULL) { 706 TIMESPEC_TO_TIMEVAL(&atv, tsp); 707 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 708 /* No timeout, just poll */ 709 timeout = -1; 710 goto start; 711 } 712 if (itimerfix(&atv)) { 713 error = EINVAL; 714 goto done; 715 } 716 717 timeout = atv.tv_sec > 24 * 60 * 60 ? 718 24 * 60 * 60 * hz : tvtohz(&atv); 719 720 getmicrouptime(&rtv); 721 timeradd(&atv, &rtv, &atv); 722 } else { 723 atv.tv_sec = 0; 724 atv.tv_usec = 0; 725 timeout = 0; 726 } 727 goto start; 728 729 retry: 730 if (atv.tv_sec || atv.tv_usec) { 731 getmicrouptime(&rtv); 732 if (timercmp(&rtv, &atv, >=)) 733 goto done; 734 ttv = atv; 735 timersub(&ttv, &rtv, &ttv); 736 timeout = ttv.tv_sec > 24 * 60 * 60 ? 737 24 * 60 * 60 * hz : tvtohz(&ttv); 738 } 739 740 start: 741 if (kq->kq_state & KQ_DYING) { 742 error = EBADF; 743 goto done; 744 } 745 746 kevp = &kev[0]; 747 s = splhigh(); 748 if (kq->kq_count == 0) { 749 if (timeout < 0) { 750 error = EWOULDBLOCK; 751 } else { 752 kq->kq_state |= KQ_SLEEP; 753 error = tsleep(kq, PSOCK | PCATCH, "kqread", timeout); 754 } 755 splx(s); 756 if (error == 0) 757 goto retry; 758 /* don't restart after signals... */ 759 if (error == ERESTART) 760 error = EINTR; 761 else if (error == EWOULDBLOCK) 762 error = 0; 763 goto done; 764 } 765 766 TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe); 767 while (count) { 768 kn = TAILQ_FIRST(&kq->kq_head); 769 if (kn == &marker) { 770 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 771 splx(s); 772 if (count == maxevents) 773 goto retry; 774 goto done; 775 } 776 777 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 778 kq->kq_count--; 779 780 if (kn->kn_status & KN_DISABLED) { 781 kn->kn_status &= ~KN_QUEUED; 782 continue; 783 } 784 if ((kn->kn_flags & EV_ONESHOT) == 0 && 785 kn->kn_fop->f_event(kn, 0) == 0) { 786 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 787 continue; 788 } 789 *kevp = kn->kn_kevent; 790 kevp++; 791 nkev++; 792 if (kn->kn_flags & EV_ONESHOT) { 793 kn->kn_status &= ~KN_QUEUED; 794 splx(s); 795 kn->kn_fop->f_detach(kn); 796 knote_drop(kn, p, p->p_fd); 797 s = splhigh(); 798 } else if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) { 799 if (kn->kn_flags & EV_CLEAR) { 800 kn->kn_data = 0; 801 kn->kn_fflags = 0; 802 } 803 if (kn->kn_flags & EV_DISPATCH) 804 kn->kn_status |= KN_DISABLED; 805 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 806 } else { 807 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 808 kq->kq_count++; 809 } 810 count--; 811 if (nkev == KQ_NEVENTS) { 812 splx(s); 813 #ifdef KTRACE 814 if (KTRPOINT(p, KTR_STRUCT)) 815 ktrevent(p, kev, nkev); 816 #endif 817 error = copyout(kev, ulistp, 818 sizeof(struct kevent) * nkev); 819 ulistp += nkev; 820 nkev = 0; 821 kevp = &kev[0]; 822 s = splhigh(); 823 if (error) 824 break; 825 } 826 } 827 TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe); 828 splx(s); 829 done: 830 if (nkev != 0) { 831 #ifdef KTRACE 832 if (KTRPOINT(p, KTR_STRUCT)) 833 ktrevent(p, kev, nkev); 834 #endif 835 error = copyout(kev, ulistp, 836 sizeof(struct kevent) * nkev); 837 } 838 *retval = maxevents - count; 839 return (error); 840 } 841 842 /* 843 * XXX 844 * This could be expanded to call kqueue_scan, if desired. 845 */ 846 int 847 kqueue_read(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 848 { 849 return (ENXIO); 850 } 851 852 int 853 kqueue_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred) 854 855 { 856 return (ENXIO); 857 } 858 859 int 860 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct proc *p) 861 { 862 return (ENOTTY); 863 } 864 865 int 866 kqueue_poll(struct file *fp, int events, struct proc *p) 867 { 868 struct kqueue *kq = (struct kqueue *)fp->f_data; 869 int revents = 0; 870 int s = splhigh(); 871 872 if (events & (POLLIN | POLLRDNORM)) { 873 if (kq->kq_count) { 874 revents |= events & (POLLIN | POLLRDNORM); 875 } else { 876 selrecord(p, &kq->kq_sel); 877 kq->kq_state |= KQ_SEL; 878 } 879 } 880 splx(s); 881 return (revents); 882 } 883 884 int 885 kqueue_stat(struct file *fp, struct stat *st, struct proc *p) 886 { 887 struct kqueue *kq = fp->f_data; 888 889 memset(st, 0, sizeof(*st)); 890 st->st_size = kq->kq_count; 891 st->st_blksize = sizeof(struct kevent); 892 st->st_mode = S_IFIFO; 893 return (0); 894 } 895 896 int 897 kqueue_close(struct file *fp, struct proc *p) 898 { 899 struct kqueue *kq = fp->f_data; 900 struct filedesc *fdp = p->p_fd; 901 struct knote **knp, *kn, *kn0; 902 int i; 903 904 for (i = 0; i < fdp->fd_knlistsize; i++) { 905 knp = &SLIST_FIRST(&fdp->fd_knlist[i]); 906 kn = *knp; 907 while (kn != NULL) { 908 kn0 = SLIST_NEXT(kn, kn_link); 909 if (kq == kn->kn_kq) { 910 kn->kn_fop->f_detach(kn); 911 FRELE(kn->kn_fp, p); 912 knote_free(kn); 913 *knp = kn0; 914 } else { 915 knp = &SLIST_NEXT(kn, kn_link); 916 } 917 kn = kn0; 918 } 919 } 920 if (fdp->fd_knhashmask != 0) { 921 for (i = 0; i < fdp->fd_knhashmask + 1; i++) { 922 knp = &SLIST_FIRST(&fdp->fd_knhash[i]); 923 kn = *knp; 924 while (kn != NULL) { 925 kn0 = SLIST_NEXT(kn, kn_link); 926 if (kq == kn->kn_kq) { 927 kn->kn_fop->f_detach(kn); 928 /* XXX non-fd release of kn->kn_ptr */ 929 knote_free(kn); 930 *knp = kn0; 931 } else { 932 knp = &SLIST_NEXT(kn, kn_link); 933 } 934 kn = kn0; 935 } 936 } 937 } 938 fp->f_data = NULL; 939 940 kq->kq_state |= KQ_DYING; 941 kqueue_wakeup(kq); 942 KQRELE(kq); 943 944 return (0); 945 } 946 947 void 948 kqueue_wakeup(struct kqueue *kq) 949 { 950 951 if (kq->kq_state & KQ_SLEEP) { 952 kq->kq_state &= ~KQ_SLEEP; 953 wakeup(kq); 954 } 955 if (kq->kq_state & KQ_SEL) { 956 kq->kq_state &= ~KQ_SEL; 957 selwakeup(&kq->kq_sel); 958 } else 959 KNOTE(&kq->kq_sel.si_note, 0); 960 } 961 962 /* 963 * activate one knote. 964 */ 965 void 966 knote_activate(struct knote *kn) 967 { 968 KNOTE_ACTIVATE(kn); 969 } 970 971 /* 972 * walk down a list of knotes, activating them if their event has triggered. 973 */ 974 void 975 knote(struct klist *list, long hint) 976 { 977 struct knote *kn, *kn0; 978 979 SLIST_FOREACH_SAFE(kn, list, kn_selnext, kn0) 980 if (kn->kn_fop->f_event(kn, hint)) 981 KNOTE_ACTIVATE(kn); 982 } 983 984 /* 985 * remove all knotes from a specified klist 986 */ 987 void 988 knote_remove(struct proc *p, struct klist *list) 989 { 990 struct knote *kn; 991 992 while ((kn = SLIST_FIRST(list)) != NULL) { 993 kn->kn_fop->f_detach(kn); 994 knote_drop(kn, p, p->p_fd); 995 } 996 } 997 998 /* 999 * remove all knotes referencing a specified fd 1000 */ 1001 void 1002 knote_fdclose(struct proc *p, int fd) 1003 { 1004 struct filedesc *fdp = p->p_fd; 1005 struct klist *list = &fdp->fd_knlist[fd]; 1006 1007 knote_remove(p, list); 1008 } 1009 1010 /* 1011 * handle a process exiting, including the triggering of NOTE_EXIT notes 1012 * XXX this could be more efficient, doing a single pass down the klist 1013 */ 1014 void 1015 knote_processexit(struct proc *p) 1016 { 1017 struct process *pr = p->p_p; 1018 1019 KNOTE(&pr->ps_klist, NOTE_EXIT); 1020 1021 /* remove other knotes hanging off the process */ 1022 knote_remove(p, &pr->ps_klist); 1023 } 1024 1025 void 1026 knote_attach(struct knote *kn, struct filedesc *fdp) 1027 { 1028 struct klist *list; 1029 int size; 1030 1031 if (!kn->kn_fop->f_isfd) { 1032 if (fdp->fd_knhashmask == 0) 1033 fdp->fd_knhash = hashinit(KN_HASHSIZE, M_TEMP, 1034 M_WAITOK, &fdp->fd_knhashmask); 1035 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1036 goto done; 1037 } 1038 1039 if (fdp->fd_knlistsize <= kn->kn_id) { 1040 size = fdp->fd_knlistsize; 1041 while (size <= kn->kn_id) 1042 size += KQEXTENT; 1043 list = mallocarray(size, sizeof(struct klist), M_TEMP, 1044 M_WAITOK); 1045 memcpy(list, fdp->fd_knlist, 1046 fdp->fd_knlistsize * sizeof(struct klist)); 1047 memset(&list[fdp->fd_knlistsize], 0, 1048 (size - fdp->fd_knlistsize) * sizeof(struct klist)); 1049 free(fdp->fd_knlist, M_TEMP, 1050 fdp->fd_knlistsize * sizeof(struct klist)); 1051 fdp->fd_knlistsize = size; 1052 fdp->fd_knlist = list; 1053 } 1054 list = &fdp->fd_knlist[kn->kn_id]; 1055 done: 1056 SLIST_INSERT_HEAD(list, kn, kn_link); 1057 kn->kn_status = 0; 1058 } 1059 1060 /* 1061 * should be called at spl == 0, since we don't want to hold spl 1062 * while calling FRELE and knote_free. 1063 */ 1064 void 1065 knote_drop(struct knote *kn, struct proc *p, struct filedesc *fdp) 1066 { 1067 struct klist *list; 1068 1069 if (kn->kn_fop->f_isfd) 1070 list = &fdp->fd_knlist[kn->kn_id]; 1071 else 1072 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 1073 1074 SLIST_REMOVE(list, kn, knote, kn_link); 1075 if (kn->kn_status & KN_QUEUED) 1076 knote_dequeue(kn); 1077 if (kn->kn_fop->f_isfd) 1078 FRELE(kn->kn_fp, p); 1079 knote_free(kn); 1080 } 1081 1082 1083 void 1084 knote_enqueue(struct knote *kn) 1085 { 1086 struct kqueue *kq = kn->kn_kq; 1087 int s = splhigh(); 1088 1089 KASSERT((kn->kn_status & KN_QUEUED) == 0); 1090 1091 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1092 kn->kn_status |= KN_QUEUED; 1093 kq->kq_count++; 1094 splx(s); 1095 kqueue_wakeup(kq); 1096 } 1097 1098 void 1099 knote_dequeue(struct knote *kn) 1100 { 1101 struct kqueue *kq = kn->kn_kq; 1102 int s = splhigh(); 1103 1104 KASSERT(kn->kn_status & KN_QUEUED); 1105 1106 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1107 kn->kn_status &= ~KN_QUEUED; 1108 kq->kq_count--; 1109 splx(s); 1110 } 1111 1112 void 1113 klist_invalidate(struct klist *list) 1114 { 1115 struct knote *kn; 1116 1117 SLIST_FOREACH(kn, list, kn_selnext) { 1118 kn->kn_status |= KN_DETACHED; 1119 kn->kn_flags |= EV_EOF | EV_ONESHOT; 1120 } 1121 } 1122