1 /*- 2 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD: src/sys/kern/kern_event.c,v 1.2.2.10 2004/04/04 07:03:14 cperciva Exp $ 27 */ 28 29 #include <sys/param.h> 30 #include <sys/systm.h> 31 #include <sys/kernel.h> 32 #include <sys/proc.h> 33 #include <sys/malloc.h> 34 #include <sys/unistd.h> 35 #include <sys/file.h> 36 #include <sys/lock.h> 37 #include <sys/fcntl.h> 38 #include <sys/queue.h> 39 #include <sys/event.h> 40 #include <sys/eventvar.h> 41 #include <sys/protosw.h> 42 #include <sys/socket.h> 43 #include <sys/socketvar.h> 44 #include <sys/stat.h> 45 #include <sys/sysctl.h> 46 #include <sys/sysproto.h> 47 #include <sys/thread.h> 48 #include <sys/uio.h> 49 #include <sys/signalvar.h> 50 #include <sys/filio.h> 51 #include <sys/ktr.h> 52 53 #include <sys/thread2.h> 54 #include <sys/file2.h> 55 #include <sys/mplock2.h> 56 57 /* 58 * Global token for kqueue subsystem 59 */ 60 #if 0 61 struct lwkt_token kq_token = LWKT_TOKEN_INITIALIZER(kq_token); 62 SYSCTL_LONG(_lwkt, OID_AUTO, kq_collisions, 63 CTLFLAG_RW, &kq_token.t_collisions, 0, 64 "Collision counter of kq_token"); 65 #endif 66 67 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 68 69 struct kevent_copyin_args { 70 struct kevent_args *ka; 71 int pchanges; 72 }; 73 74 static int kqueue_sleep(struct kqueue *kq, struct timespec *tsp); 75 static int kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 76 struct knote *marker); 77 static int kqueue_read(struct file *fp, struct uio *uio, 78 struct ucred *cred, int flags); 79 static int kqueue_write(struct file *fp, struct uio *uio, 80 struct ucred *cred, int flags); 81 static int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 82 struct ucred *cred, struct sysmsg *msg); 83 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 84 static int kqueue_stat(struct file *fp, struct stat *st, 85 struct ucred *cred); 86 static int kqueue_close(struct file *fp); 87 static void kqueue_wakeup(struct kqueue *kq); 88 static int filter_attach(struct knote *kn); 89 static int filter_event(struct knote *kn, long hint); 90 91 /* 92 * MPSAFE 93 */ 94 static struct fileops kqueueops = { 95 .fo_read = kqueue_read, 96 .fo_write = kqueue_write, 97 .fo_ioctl = kqueue_ioctl, 98 .fo_kqfilter = kqueue_kqfilter, 99 .fo_stat = kqueue_stat, 100 .fo_close = kqueue_close, 101 .fo_shutdown = nofo_shutdown 102 }; 103 104 static void knote_attach(struct knote *kn); 105 static void knote_drop(struct knote *kn); 106 static void knote_detach_and_drop(struct knote *kn); 107 static void knote_enqueue(struct knote *kn); 108 static void knote_dequeue(struct knote *kn); 109 static struct knote *knote_alloc(void); 110 static void knote_free(struct knote *kn); 111 112 static void filt_kqdetach(struct knote *kn); 113 static int filt_kqueue(struct knote *kn, long hint); 114 static int filt_procattach(struct knote *kn); 115 static void filt_procdetach(struct knote *kn); 116 static int filt_proc(struct knote *kn, long hint); 117 static int filt_fileattach(struct knote *kn); 118 static void filt_timerexpire(void *knx); 119 static int filt_timerattach(struct knote *kn); 120 static void filt_timerdetach(struct knote *kn); 121 static int filt_timer(struct knote *kn, long hint); 122 123 static struct filterops file_filtops = 124 { FILTEROP_ISFD, filt_fileattach, NULL, NULL }; 125 static struct filterops kqread_filtops = 126 { FILTEROP_ISFD, NULL, filt_kqdetach, filt_kqueue }; 127 static struct filterops proc_filtops = 128 { 0, filt_procattach, filt_procdetach, filt_proc }; 129 static struct filterops timer_filtops = 130 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 131 132 static int kq_ncallouts = 0; 133 static int kq_calloutmax = (4 * 1024); 134 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 135 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 136 static int kq_checkloop = 1000000; 137 SYSCTL_INT(_kern, OID_AUTO, kq_checkloop, CTLFLAG_RW, 138 &kq_checkloop, 0, "Maximum number of callouts allocated for kqueue"); 139 140 #define KNOTE_ACTIVATE(kn) do { \ 141 kn->kn_status |= KN_ACTIVE; \ 142 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 143 knote_enqueue(kn); \ 144 } while(0) 145 146 #define KN_HASHSIZE 64 /* XXX should be tunable */ 147 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 148 149 extern struct filterops aio_filtops; 150 extern struct filterops sig_filtops; 151 152 /* 153 * Table for for all system-defined filters. 154 */ 155 static struct filterops *sysfilt_ops[] = { 156 &file_filtops, /* EVFILT_READ */ 157 &file_filtops, /* EVFILT_WRITE */ 158 &aio_filtops, /* EVFILT_AIO */ 159 &file_filtops, /* EVFILT_VNODE */ 160 &proc_filtops, /* EVFILT_PROC */ 161 &sig_filtops, /* EVFILT_SIGNAL */ 162 &timer_filtops, /* EVFILT_TIMER */ 163 &file_filtops, /* EVFILT_EXCEPT */ 164 }; 165 166 static int 167 filt_fileattach(struct knote *kn) 168 { 169 return (fo_kqfilter(kn->kn_fp, kn)); 170 } 171 172 /* 173 * MPSAFE 174 */ 175 static int 176 kqueue_kqfilter(struct file *fp, struct knote *kn) 177 { 178 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 179 180 if (kn->kn_filter != EVFILT_READ) 181 return (EOPNOTSUPP); 182 183 kn->kn_fop = &kqread_filtops; 184 knote_insert(&kq->kq_kqinfo.ki_note, kn); 185 return (0); 186 } 187 188 static void 189 filt_kqdetach(struct knote *kn) 190 { 191 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 192 193 knote_remove(&kq->kq_kqinfo.ki_note, kn); 194 } 195 196 /*ARGSUSED*/ 197 static int 198 filt_kqueue(struct knote *kn, long hint) 199 { 200 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 201 202 kn->kn_data = kq->kq_count; 203 return (kn->kn_data > 0); 204 } 205 206 static int 207 filt_procattach(struct knote *kn) 208 { 209 struct proc *p; 210 int immediate; 211 212 immediate = 0; 213 p = pfind(kn->kn_id); 214 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 215 p = zpfind(kn->kn_id); 216 immediate = 1; 217 } 218 if (p == NULL) { 219 return (ESRCH); 220 } 221 if (!PRISON_CHECK(curthread->td_ucred, p->p_ucred)) { 222 if (p) 223 PRELE(p); 224 return (EACCES); 225 } 226 227 lwkt_gettoken(&p->p_token); 228 kn->kn_ptr.p_proc = p; 229 kn->kn_flags |= EV_CLEAR; /* automatically set */ 230 231 /* 232 * internal flag indicating registration done by kernel 233 */ 234 if (kn->kn_flags & EV_FLAG1) { 235 kn->kn_data = kn->kn_sdata; /* ppid */ 236 kn->kn_fflags = NOTE_CHILD; 237 kn->kn_flags &= ~EV_FLAG1; 238 } 239 240 knote_insert(&p->p_klist, kn); 241 242 /* 243 * Immediately activate any exit notes if the target process is a 244 * zombie. This is necessary to handle the case where the target 245 * process, e.g. a child, dies before the kevent is negistered. 246 */ 247 if (immediate && filt_proc(kn, NOTE_EXIT)) 248 KNOTE_ACTIVATE(kn); 249 lwkt_reltoken(&p->p_token); 250 PRELE(p); 251 252 return (0); 253 } 254 255 /* 256 * The knote may be attached to a different process, which may exit, 257 * leaving nothing for the knote to be attached to. So when the process 258 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 259 * it will be deleted when read out. However, as part of the knote deletion, 260 * this routine is called, so a check is needed to avoid actually performing 261 * a detach, because the original process does not exist any more. 262 */ 263 static void 264 filt_procdetach(struct knote *kn) 265 { 266 struct proc *p; 267 268 if (kn->kn_status & KN_DETACHED) 269 return; 270 p = kn->kn_ptr.p_proc; 271 knote_remove(&p->p_klist, kn); 272 } 273 274 static int 275 filt_proc(struct knote *kn, long hint) 276 { 277 u_int event; 278 279 /* 280 * mask off extra data 281 */ 282 event = (u_int)hint & NOTE_PCTRLMASK; 283 284 /* 285 * if the user is interested in this event, record it. 286 */ 287 if (kn->kn_sfflags & event) 288 kn->kn_fflags |= event; 289 290 /* 291 * Process is gone, so flag the event as finished. Detach the 292 * knote from the process now because the process will be poof, 293 * gone later on. 294 */ 295 if (event == NOTE_EXIT) { 296 struct proc *p = kn->kn_ptr.p_proc; 297 if ((kn->kn_status & KN_DETACHED) == 0) { 298 PHOLD(p); 299 knote_remove(&p->p_klist, kn); 300 kn->kn_status |= KN_DETACHED; 301 kn->kn_data = p->p_xstat; 302 kn->kn_ptr.p_proc = NULL; 303 PRELE(p); 304 } 305 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 306 return (1); 307 } 308 309 /* 310 * process forked, and user wants to track the new process, 311 * so attach a new knote to it, and immediately report an 312 * event with the parent's pid. 313 */ 314 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 315 struct kevent kev; 316 int error; 317 318 /* 319 * register knote with new process. 320 */ 321 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 322 kev.filter = kn->kn_filter; 323 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 324 kev.fflags = kn->kn_sfflags; 325 kev.data = kn->kn_id; /* parent */ 326 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 327 error = kqueue_register(kn->kn_kq, &kev); 328 if (error) 329 kn->kn_fflags |= NOTE_TRACKERR; 330 } 331 332 return (kn->kn_fflags != 0); 333 } 334 335 /* 336 * The callout interlocks with callout_terminate() but can still 337 * race a deletion so if KN_DELETING is set we just don't touch 338 * the knote. 339 */ 340 static void 341 filt_timerexpire(void *knx) 342 { 343 struct lwkt_token *tok; 344 struct knote *kn = knx; 345 struct callout *calloutp; 346 struct timeval tv; 347 int tticks; 348 349 tok = lwkt_token_pool_lookup(kn->kn_kq); 350 lwkt_gettoken(tok); 351 if ((kn->kn_status & KN_DELETING) == 0) { 352 kn->kn_data++; 353 KNOTE_ACTIVATE(kn); 354 355 if ((kn->kn_flags & EV_ONESHOT) == 0) { 356 tv.tv_sec = kn->kn_sdata / 1000; 357 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 358 tticks = tvtohz_high(&tv); 359 calloutp = (struct callout *)kn->kn_hook; 360 callout_reset(calloutp, tticks, filt_timerexpire, kn); 361 } 362 } 363 lwkt_reltoken(tok); 364 } 365 366 /* 367 * data contains amount of time to sleep, in milliseconds 368 */ 369 static int 370 filt_timerattach(struct knote *kn) 371 { 372 struct callout *calloutp; 373 struct timeval tv; 374 int tticks; 375 376 if (kq_ncallouts >= kq_calloutmax) { 377 kn->kn_hook = NULL; 378 return (ENOMEM); 379 } 380 kq_ncallouts++; 381 382 tv.tv_sec = kn->kn_sdata / 1000; 383 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 384 tticks = tvtohz_high(&tv); 385 386 kn->kn_flags |= EV_CLEAR; /* automatically set */ 387 calloutp = kmalloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK); 388 callout_init(calloutp); 389 kn->kn_hook = (caddr_t)calloutp; 390 callout_reset(calloutp, tticks, filt_timerexpire, kn); 391 392 return (0); 393 } 394 395 /* 396 * This function is called with the knote flagged locked but it is 397 * still possible to race a callout event due to the callback blocking. 398 * We must call callout_terminate() instead of callout_stop() to deal 399 * with the race. 400 */ 401 static void 402 filt_timerdetach(struct knote *kn) 403 { 404 struct callout *calloutp; 405 406 calloutp = (struct callout *)kn->kn_hook; 407 callout_terminate(calloutp); 408 kfree(calloutp, M_KQUEUE); 409 kq_ncallouts--; 410 } 411 412 static int 413 filt_timer(struct knote *kn, long hint) 414 { 415 416 return (kn->kn_data != 0); 417 } 418 419 /* 420 * Acquire a knote, return non-zero on success, 0 on failure. 421 * 422 * If we cannot acquire the knote we sleep and return 0. The knote 423 * may be stale on return in this case and the caller must restart 424 * whatever loop they are in. 425 * 426 * Related kq token must be held. 427 */ 428 static __inline 429 int 430 knote_acquire(struct knote *kn) 431 { 432 if (kn->kn_status & KN_PROCESSING) { 433 kn->kn_status |= KN_WAITING | KN_REPROCESS; 434 tsleep(kn, 0, "kqepts", hz); 435 /* knote may be stale now */ 436 return(0); 437 } 438 kn->kn_status |= KN_PROCESSING; 439 return(1); 440 } 441 442 /* 443 * Release an acquired knote, clearing KN_PROCESSING and handling any 444 * KN_REPROCESS events. 445 * 446 * Caller must be holding the related kq token 447 * 448 * Non-zero is returned if the knote is destroyed or detached. 449 */ 450 static __inline 451 int 452 knote_release(struct knote *kn) 453 { 454 while (kn->kn_status & KN_REPROCESS) { 455 kn->kn_status &= ~KN_REPROCESS; 456 if (kn->kn_status & KN_WAITING) { 457 kn->kn_status &= ~KN_WAITING; 458 wakeup(kn); 459 } 460 if (kn->kn_status & KN_DELETING) { 461 knote_detach_and_drop(kn); 462 return(1); 463 /* NOT REACHED */ 464 } 465 if (filter_event(kn, 0)) 466 KNOTE_ACTIVATE(kn); 467 } 468 if (kn->kn_status & KN_DETACHED) { 469 kn->kn_status &= ~KN_PROCESSING; 470 return(1); 471 } else { 472 kn->kn_status &= ~KN_PROCESSING; 473 return(0); 474 } 475 } 476 477 /* 478 * Initialize a kqueue. 479 * 480 * NOTE: The lwp/proc code initializes a kqueue for select/poll ops. 481 * 482 * MPSAFE 483 */ 484 void 485 kqueue_init(struct kqueue *kq, struct filedesc *fdp) 486 { 487 TAILQ_INIT(&kq->kq_knpend); 488 TAILQ_INIT(&kq->kq_knlist); 489 kq->kq_count = 0; 490 kq->kq_fdp = fdp; 491 SLIST_INIT(&kq->kq_kqinfo.ki_note); 492 } 493 494 /* 495 * Terminate a kqueue. Freeing the actual kq itself is left up to the 496 * caller (it might be embedded in a lwp so we don't do it here). 497 * 498 * The kq's knlist must be completely eradicated so block on any 499 * processing races. 500 */ 501 void 502 kqueue_terminate(struct kqueue *kq) 503 { 504 struct lwkt_token *tok; 505 struct knote *kn; 506 507 tok = lwkt_token_pool_lookup(kq); 508 lwkt_gettoken(tok); 509 while ((kn = TAILQ_FIRST(&kq->kq_knlist)) != NULL) { 510 if (knote_acquire(kn)) 511 knote_detach_and_drop(kn); 512 } 513 if (kq->kq_knhash) { 514 hashdestroy(kq->kq_knhash, M_KQUEUE, kq->kq_knhashmask); 515 kq->kq_knhash = NULL; 516 kq->kq_knhashmask = 0; 517 } 518 lwkt_reltoken(tok); 519 } 520 521 /* 522 * MPSAFE 523 */ 524 int 525 sys_kqueue(struct kqueue_args *uap) 526 { 527 struct thread *td = curthread; 528 struct kqueue *kq; 529 struct file *fp; 530 int fd, error; 531 532 error = falloc(td->td_lwp, &fp, &fd); 533 if (error) 534 return (error); 535 fp->f_flag = FREAD | FWRITE; 536 fp->f_type = DTYPE_KQUEUE; 537 fp->f_ops = &kqueueops; 538 539 kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 540 kqueue_init(kq, td->td_proc->p_fd); 541 fp->f_data = kq; 542 543 fsetfd(kq->kq_fdp, fp, fd); 544 uap->sysmsg_result = fd; 545 fdrop(fp); 546 return (error); 547 } 548 549 /* 550 * Copy 'count' items into the destination list pointed to by uap->eventlist. 551 */ 552 static int 553 kevent_copyout(void *arg, struct kevent *kevp, int count, int *res) 554 { 555 struct kevent_copyin_args *kap; 556 int error; 557 558 kap = (struct kevent_copyin_args *)arg; 559 560 error = copyout(kevp, kap->ka->eventlist, count * sizeof(*kevp)); 561 if (error == 0) { 562 kap->ka->eventlist += count; 563 *res += count; 564 } else { 565 *res = -1; 566 } 567 568 return (error); 569 } 570 571 /* 572 * Copy at most 'max' items from the list pointed to by kap->changelist, 573 * return number of items in 'events'. 574 */ 575 static int 576 kevent_copyin(void *arg, struct kevent *kevp, int max, int *events) 577 { 578 struct kevent_copyin_args *kap; 579 int error, count; 580 581 kap = (struct kevent_copyin_args *)arg; 582 583 count = min(kap->ka->nchanges - kap->pchanges, max); 584 error = copyin(kap->ka->changelist, kevp, count * sizeof *kevp); 585 if (error == 0) { 586 kap->ka->changelist += count; 587 kap->pchanges += count; 588 *events = count; 589 } 590 591 return (error); 592 } 593 594 /* 595 * MPSAFE 596 */ 597 int 598 kern_kevent(struct kqueue *kq, int nevents, int *res, void *uap, 599 k_copyin_fn kevent_copyinfn, k_copyout_fn kevent_copyoutfn, 600 struct timespec *tsp_in) 601 { 602 struct kevent *kevp; 603 struct timespec *tsp; 604 int i, n, total, error, nerrors = 0; 605 int lres; 606 int limit = kq_checkloop; 607 struct kevent kev[KQ_NEVENTS]; 608 struct knote marker; 609 struct lwkt_token *tok; 610 611 if (tsp_in == NULL || tsp_in->tv_sec || tsp_in->tv_nsec) 612 atomic_set_int(&curthread->td_mpflags, TDF_MP_BATCH_DEMARC); 613 614 615 tsp = tsp_in; 616 *res = 0; 617 618 tok = lwkt_token_pool_lookup(kq); 619 lwkt_gettoken(tok); 620 for ( ;; ) { 621 n = 0; 622 error = kevent_copyinfn(uap, kev, KQ_NEVENTS, &n); 623 if (error) 624 goto done; 625 if (n == 0) 626 break; 627 for (i = 0; i < n; i++) { 628 kevp = &kev[i]; 629 kevp->flags &= ~EV_SYSFLAGS; 630 error = kqueue_register(kq, kevp); 631 632 /* 633 * If a registration returns an error we 634 * immediately post the error. The kevent() 635 * call itself will fail with the error if 636 * no space is available for posting. 637 * 638 * Such errors normally bypass the timeout/blocking 639 * code. However, if the copyoutfn function refuses 640 * to post the error (see sys_poll()), then we 641 * ignore it too. 642 */ 643 if (error) { 644 kevp->flags = EV_ERROR; 645 kevp->data = error; 646 lres = *res; 647 kevent_copyoutfn(uap, kevp, 1, res); 648 if (*res < 0) { 649 goto done; 650 } else if (lres != *res) { 651 nevents--; 652 nerrors++; 653 } 654 } 655 } 656 } 657 if (nerrors) { 658 error = 0; 659 goto done; 660 } 661 662 /* 663 * Acquire/wait for events - setup timeout 664 */ 665 if (tsp != NULL) { 666 struct timespec ats; 667 668 if (tsp->tv_sec || tsp->tv_nsec) { 669 getnanouptime(&ats); 670 timespecadd(tsp, &ats); /* tsp = target time */ 671 } 672 } 673 674 /* 675 * Loop as required. 676 * 677 * Collect as many events as we can. Sleeping on successive 678 * loops is disabled if copyoutfn has incremented (*res). 679 * 680 * The loop stops if an error occurs, all events have been 681 * scanned (the marker has been reached), or fewer than the 682 * maximum number of events is found. 683 * 684 * The copyoutfn function does not have to increment (*res) in 685 * order for the loop to continue. 686 * 687 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents. 688 */ 689 total = 0; 690 error = 0; 691 marker.kn_filter = EVFILT_MARKER; 692 marker.kn_status = KN_PROCESSING; 693 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 694 while ((n = nevents - total) > 0) { 695 if (n > KQ_NEVENTS) 696 n = KQ_NEVENTS; 697 698 /* 699 * If no events are pending sleep until timeout (if any) 700 * or an event occurs. 701 * 702 * After the sleep completes the marker is moved to the 703 * end of the list, making any received events available 704 * to our scan. 705 */ 706 if (kq->kq_count == 0 && *res == 0) { 707 error = kqueue_sleep(kq, tsp); 708 if (error) 709 break; 710 711 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 712 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 713 } 714 715 /* 716 * Process all received events 717 * Account for all non-spurious events in our total 718 */ 719 i = kqueue_scan(kq, kev, n, &marker); 720 if (i) { 721 lres = *res; 722 error = kevent_copyoutfn(uap, kev, i, res); 723 total += *res - lres; 724 if (error) 725 break; 726 } 727 if (limit && --limit == 0) 728 panic("kqueue: checkloop failed i=%d", i); 729 730 /* 731 * Normally when fewer events are returned than requested 732 * we can stop. However, if only spurious events were 733 * collected the copyout will not bump (*res) and we have 734 * to continue. 735 */ 736 if (i < n && *res) 737 break; 738 739 /* 740 * Deal with an edge case where spurious events can cause 741 * a loop to occur without moving the marker. This can 742 * prevent kqueue_scan() from picking up new events which 743 * race us. We must be sure to move the marker for this 744 * case. 745 * 746 * NOTE: We do not want to move the marker if events 747 * were scanned because normal kqueue operations 748 * may reactivate events. Moving the marker in 749 * that case could result in duplicates for the 750 * same event. 751 */ 752 if (i == 0) { 753 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 754 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 755 } 756 } 757 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 758 759 /* Timeouts do not return EWOULDBLOCK. */ 760 if (error == EWOULDBLOCK) 761 error = 0; 762 763 done: 764 lwkt_reltoken(tok); 765 return (error); 766 } 767 768 /* 769 * MPALMOSTSAFE 770 */ 771 int 772 sys_kevent(struct kevent_args *uap) 773 { 774 struct thread *td = curthread; 775 struct proc *p = td->td_proc; 776 struct timespec ts, *tsp; 777 struct kqueue *kq; 778 struct file *fp = NULL; 779 struct kevent_copyin_args *kap, ka; 780 int error; 781 782 if (uap->timeout) { 783 error = copyin(uap->timeout, &ts, sizeof(ts)); 784 if (error) 785 return (error); 786 tsp = &ts; 787 } else { 788 tsp = NULL; 789 } 790 fp = holdfp(p->p_fd, uap->fd, -1); 791 if (fp == NULL) 792 return (EBADF); 793 if (fp->f_type != DTYPE_KQUEUE) { 794 fdrop(fp); 795 return (EBADF); 796 } 797 798 kq = (struct kqueue *)fp->f_data; 799 800 kap = &ka; 801 kap->ka = uap; 802 kap->pchanges = 0; 803 804 error = kern_kevent(kq, uap->nevents, &uap->sysmsg_result, kap, 805 kevent_copyin, kevent_copyout, tsp); 806 807 fdrop(fp); 808 809 return (error); 810 } 811 812 /* 813 * Caller must be holding the kq token 814 */ 815 int 816 kqueue_register(struct kqueue *kq, struct kevent *kev) 817 { 818 struct lwkt_token *tok; 819 struct filedesc *fdp = kq->kq_fdp; 820 struct filterops *fops; 821 struct file *fp = NULL; 822 struct knote *kn = NULL; 823 int error = 0; 824 825 if (kev->filter < 0) { 826 if (kev->filter + EVFILT_SYSCOUNT < 0) 827 return (EINVAL); 828 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 829 } else { 830 /* 831 * XXX 832 * filter attach routine is responsible for insuring that 833 * the identifier can be attached to it. 834 */ 835 kprintf("unknown filter: %d\n", kev->filter); 836 return (EINVAL); 837 } 838 839 tok = lwkt_token_pool_lookup(kq); 840 lwkt_gettoken(tok); 841 if (fops->f_flags & FILTEROP_ISFD) { 842 /* validate descriptor */ 843 fp = holdfp(fdp, kev->ident, -1); 844 if (fp == NULL) { 845 lwkt_reltoken(tok); 846 return (EBADF); 847 } 848 lwkt_getpooltoken(&fp->f_klist); 849 again1: 850 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 851 if (kn->kn_kq == kq && 852 kn->kn_filter == kev->filter && 853 kn->kn_id == kev->ident) { 854 if (knote_acquire(kn) == 0) 855 goto again1; 856 break; 857 } 858 } 859 lwkt_relpooltoken(&fp->f_klist); 860 } else { 861 if (kq->kq_knhashmask) { 862 struct klist *list; 863 864 list = &kq->kq_knhash[ 865 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 866 lwkt_getpooltoken(list); 867 again2: 868 SLIST_FOREACH(kn, list, kn_link) { 869 if (kn->kn_id == kev->ident && 870 kn->kn_filter == kev->filter) { 871 if (knote_acquire(kn) == 0) 872 goto again2; 873 break; 874 } 875 } 876 lwkt_relpooltoken(list); 877 } 878 } 879 880 /* 881 * NOTE: At this point if kn is non-NULL we will have acquired 882 * it and set KN_PROCESSING. 883 */ 884 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 885 error = ENOENT; 886 goto done; 887 } 888 889 /* 890 * kn now contains the matching knote, or NULL if no match 891 */ 892 if (kev->flags & EV_ADD) { 893 if (kn == NULL) { 894 kn = knote_alloc(); 895 if (kn == NULL) { 896 error = ENOMEM; 897 goto done; 898 } 899 kn->kn_fp = fp; 900 kn->kn_kq = kq; 901 kn->kn_fop = fops; 902 903 /* 904 * apply reference count to knote structure, and 905 * do not release it at the end of this routine. 906 */ 907 fp = NULL; 908 909 kn->kn_sfflags = kev->fflags; 910 kn->kn_sdata = kev->data; 911 kev->fflags = 0; 912 kev->data = 0; 913 kn->kn_kevent = *kev; 914 915 /* 916 * KN_PROCESSING prevents the knote from getting 917 * ripped out from under us while we are trying 918 * to attach it, in case the attach blocks. 919 */ 920 kn->kn_status = KN_PROCESSING; 921 knote_attach(kn); 922 if ((error = filter_attach(kn)) != 0) { 923 kn->kn_status |= KN_DELETING | KN_REPROCESS; 924 knote_drop(kn); 925 goto done; 926 } 927 928 /* 929 * Interlock against close races which either tried 930 * to remove our knote while we were blocked or missed 931 * it entirely prior to our attachment. We do not 932 * want to end up with a knote on a closed descriptor. 933 */ 934 if ((fops->f_flags & FILTEROP_ISFD) && 935 checkfdclosed(fdp, kev->ident, kn->kn_fp)) { 936 kn->kn_status |= KN_DELETING | KN_REPROCESS; 937 } 938 } else { 939 /* 940 * The user may change some filter values after the 941 * initial EV_ADD, but doing so will not reset any 942 * filter which have already been triggered. 943 */ 944 KKASSERT(kn->kn_status & KN_PROCESSING); 945 kn->kn_sfflags = kev->fflags; 946 kn->kn_sdata = kev->data; 947 kn->kn_kevent.udata = kev->udata; 948 } 949 950 /* 951 * Execute the filter event to immediately activate the 952 * knote if necessary. If reprocessing events are pending 953 * due to blocking above we do not run the filter here 954 * but instead let knote_release() do it. Otherwise we 955 * might run the filter on a deleted event. 956 */ 957 if ((kn->kn_status & KN_REPROCESS) == 0) { 958 if (filter_event(kn, 0)) 959 KNOTE_ACTIVATE(kn); 960 } 961 } else if (kev->flags & EV_DELETE) { 962 /* 963 * Delete the existing knote 964 */ 965 knote_detach_and_drop(kn); 966 goto done; 967 } 968 969 /* 970 * Disablement does not deactivate a knote here. 971 */ 972 if ((kev->flags & EV_DISABLE) && 973 ((kn->kn_status & KN_DISABLED) == 0)) { 974 kn->kn_status |= KN_DISABLED; 975 } 976 977 /* 978 * Re-enablement may have to immediately enqueue an active knote. 979 */ 980 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 981 kn->kn_status &= ~KN_DISABLED; 982 if ((kn->kn_status & KN_ACTIVE) && 983 ((kn->kn_status & KN_QUEUED) == 0)) { 984 knote_enqueue(kn); 985 } 986 } 987 988 /* 989 * Handle any required reprocessing 990 */ 991 knote_release(kn); 992 /* kn may be invalid now */ 993 994 done: 995 lwkt_reltoken(tok); 996 if (fp != NULL) 997 fdrop(fp); 998 return (error); 999 } 1000 1001 /* 1002 * Block as necessary until the target time is reached. 1003 * If tsp is NULL we block indefinitely. If tsp->ts_secs/nsecs are both 1004 * 0 we do not block at all. 1005 * 1006 * Caller must be holding the kq token. 1007 */ 1008 static int 1009 kqueue_sleep(struct kqueue *kq, struct timespec *tsp) 1010 { 1011 int error = 0; 1012 1013 if (tsp == NULL) { 1014 kq->kq_state |= KQ_SLEEP; 1015 error = tsleep(kq, PCATCH, "kqread", 0); 1016 } else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 1017 error = EWOULDBLOCK; 1018 } else { 1019 struct timespec ats; 1020 struct timespec atx = *tsp; 1021 int timeout; 1022 1023 getnanouptime(&ats); 1024 timespecsub(&atx, &ats); 1025 if (ats.tv_sec < 0) { 1026 error = EWOULDBLOCK; 1027 } else { 1028 timeout = atx.tv_sec > 24 * 60 * 60 ? 1029 24 * 60 * 60 * hz : tstohz_high(&atx); 1030 kq->kq_state |= KQ_SLEEP; 1031 error = tsleep(kq, PCATCH, "kqread", timeout); 1032 } 1033 } 1034 1035 /* don't restart after signals... */ 1036 if (error == ERESTART) 1037 return (EINTR); 1038 1039 return (error); 1040 } 1041 1042 /* 1043 * Scan the kqueue, return the number of active events placed in kevp up 1044 * to count. 1045 * 1046 * Continuous mode events may get recycled, do not continue scanning past 1047 * marker unless no events have been collected. 1048 * 1049 * Caller must be holding the kq token 1050 */ 1051 static int 1052 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 1053 struct knote *marker) 1054 { 1055 struct knote *kn, local_marker; 1056 int total; 1057 1058 total = 0; 1059 local_marker.kn_filter = EVFILT_MARKER; 1060 local_marker.kn_status = KN_PROCESSING; 1061 1062 /* 1063 * Collect events. 1064 */ 1065 TAILQ_INSERT_HEAD(&kq->kq_knpend, &local_marker, kn_tqe); 1066 while (count) { 1067 kn = TAILQ_NEXT(&local_marker, kn_tqe); 1068 if (kn->kn_filter == EVFILT_MARKER) { 1069 /* Marker reached, we are done */ 1070 if (kn == marker) 1071 break; 1072 1073 /* Move local marker past some other threads marker */ 1074 kn = TAILQ_NEXT(kn, kn_tqe); 1075 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1076 TAILQ_INSERT_BEFORE(kn, &local_marker, kn_tqe); 1077 continue; 1078 } 1079 1080 /* 1081 * We can't skip a knote undergoing processing, otherwise 1082 * we risk not returning it when the user process expects 1083 * it should be returned. Sleep and retry. 1084 */ 1085 if (knote_acquire(kn) == 0) 1086 continue; 1087 1088 /* 1089 * Remove the event for processing. 1090 * 1091 * WARNING! We must leave KN_QUEUED set to prevent the 1092 * event from being KNOTE_ACTIVATE()d while 1093 * the queue state is in limbo, in case we 1094 * block. 1095 * 1096 * WARNING! We must set KN_PROCESSING to avoid races 1097 * against deletion or another thread's 1098 * processing. 1099 */ 1100 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1101 kq->kq_count--; 1102 1103 /* 1104 * We have to deal with an extremely important race against 1105 * file descriptor close()s here. The file descriptor can 1106 * disappear MPSAFE, and there is a small window of 1107 * opportunity between that and the call to knote_fdclose(). 1108 * 1109 * If we hit that window here while doselect or dopoll is 1110 * trying to delete a spurious event they will not be able 1111 * to match up the event against a knote and will go haywire. 1112 */ 1113 if ((kn->kn_fop->f_flags & FILTEROP_ISFD) && 1114 checkfdclosed(kq->kq_fdp, kn->kn_kevent.ident, kn->kn_fp)) { 1115 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1116 } 1117 1118 if (kn->kn_status & KN_DISABLED) { 1119 /* 1120 * If disabled we ensure the event is not queued 1121 * but leave its active bit set. On re-enablement 1122 * the event may be immediately triggered. 1123 */ 1124 kn->kn_status &= ~KN_QUEUED; 1125 } else if ((kn->kn_flags & EV_ONESHOT) == 0 && 1126 (kn->kn_status & KN_DELETING) == 0 && 1127 filter_event(kn, 0) == 0) { 1128 /* 1129 * If not running in one-shot mode and the event 1130 * is no longer present we ensure it is removed 1131 * from the queue and ignore it. 1132 */ 1133 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1134 } else { 1135 /* 1136 * Post the event 1137 */ 1138 *kevp++ = kn->kn_kevent; 1139 ++total; 1140 --count; 1141 1142 if (kn->kn_flags & EV_ONESHOT) { 1143 kn->kn_status &= ~KN_QUEUED; 1144 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1145 } else if (kn->kn_flags & EV_CLEAR) { 1146 kn->kn_data = 0; 1147 kn->kn_fflags = 0; 1148 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1149 } else { 1150 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1151 kq->kq_count++; 1152 } 1153 } 1154 1155 /* 1156 * Handle any post-processing states 1157 */ 1158 knote_release(kn); 1159 } 1160 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1161 1162 return (total); 1163 } 1164 1165 /* 1166 * XXX 1167 * This could be expanded to call kqueue_scan, if desired. 1168 * 1169 * MPSAFE 1170 */ 1171 static int 1172 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1173 { 1174 return (ENXIO); 1175 } 1176 1177 /* 1178 * MPSAFE 1179 */ 1180 static int 1181 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1182 { 1183 return (ENXIO); 1184 } 1185 1186 /* 1187 * MPALMOSTSAFE 1188 */ 1189 static int 1190 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 1191 struct ucred *cred, struct sysmsg *msg) 1192 { 1193 struct lwkt_token *tok; 1194 struct kqueue *kq; 1195 int error; 1196 1197 kq = (struct kqueue *)fp->f_data; 1198 tok = lwkt_token_pool_lookup(kq); 1199 lwkt_gettoken(tok); 1200 1201 switch(com) { 1202 case FIOASYNC: 1203 if (*(int *)data) 1204 kq->kq_state |= KQ_ASYNC; 1205 else 1206 kq->kq_state &= ~KQ_ASYNC; 1207 error = 0; 1208 break; 1209 case FIOSETOWN: 1210 error = fsetown(*(int *)data, &kq->kq_sigio); 1211 break; 1212 default: 1213 error = ENOTTY; 1214 break; 1215 } 1216 lwkt_reltoken(tok); 1217 return (error); 1218 } 1219 1220 /* 1221 * MPSAFE 1222 */ 1223 static int 1224 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred) 1225 { 1226 struct kqueue *kq = (struct kqueue *)fp->f_data; 1227 1228 bzero((void *)st, sizeof(*st)); 1229 st->st_size = kq->kq_count; 1230 st->st_blksize = sizeof(struct kevent); 1231 st->st_mode = S_IFIFO; 1232 return (0); 1233 } 1234 1235 /* 1236 * MPSAFE 1237 */ 1238 static int 1239 kqueue_close(struct file *fp) 1240 { 1241 struct kqueue *kq = (struct kqueue *)fp->f_data; 1242 1243 kqueue_terminate(kq); 1244 1245 fp->f_data = NULL; 1246 funsetown(&kq->kq_sigio); 1247 1248 kfree(kq, M_KQUEUE); 1249 return (0); 1250 } 1251 1252 static void 1253 kqueue_wakeup(struct kqueue *kq) 1254 { 1255 if (kq->kq_state & KQ_SLEEP) { 1256 kq->kq_state &= ~KQ_SLEEP; 1257 wakeup(kq); 1258 } 1259 KNOTE(&kq->kq_kqinfo.ki_note, 0); 1260 } 1261 1262 /* 1263 * Calls filterops f_attach function, acquiring mplock if filter is not 1264 * marked as FILTEROP_MPSAFE. 1265 * 1266 * Caller must be holding the related kq token 1267 */ 1268 static int 1269 filter_attach(struct knote *kn) 1270 { 1271 int ret; 1272 1273 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1274 ret = kn->kn_fop->f_attach(kn); 1275 } else { 1276 get_mplock(); 1277 ret = kn->kn_fop->f_attach(kn); 1278 rel_mplock(); 1279 } 1280 return (ret); 1281 } 1282 1283 /* 1284 * Detach the knote and drop it, destroying the knote. 1285 * 1286 * Calls filterops f_detach function, acquiring mplock if filter is not 1287 * marked as FILTEROP_MPSAFE. 1288 * 1289 * Caller must be holding the related kq token 1290 */ 1291 static void 1292 knote_detach_and_drop(struct knote *kn) 1293 { 1294 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1295 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1296 kn->kn_fop->f_detach(kn); 1297 } else { 1298 get_mplock(); 1299 kn->kn_fop->f_detach(kn); 1300 rel_mplock(); 1301 } 1302 knote_drop(kn); 1303 } 1304 1305 /* 1306 * Calls filterops f_event function, acquiring mplock if filter is not 1307 * marked as FILTEROP_MPSAFE. 1308 * 1309 * If the knote is in the middle of being created or deleted we cannot 1310 * safely call the filter op. 1311 * 1312 * Caller must be holding the related kq token 1313 */ 1314 static int 1315 filter_event(struct knote *kn, long hint) 1316 { 1317 int ret; 1318 1319 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1320 ret = kn->kn_fop->f_event(kn, hint); 1321 } else { 1322 get_mplock(); 1323 ret = kn->kn_fop->f_event(kn, hint); 1324 rel_mplock(); 1325 } 1326 return (ret); 1327 } 1328 1329 /* 1330 * Walk down a list of knotes, activating them if their event has triggered. 1331 * 1332 * If we encounter any knotes which are undergoing processing we just mark 1333 * them for reprocessing and do not try to [re]activate the knote. However, 1334 * if a hint is being passed we have to wait and that makes things a bit 1335 * sticky. 1336 */ 1337 void 1338 knote(struct klist *list, long hint) 1339 { 1340 struct kqueue *kq; 1341 struct knote *kn; 1342 struct knote *kntmp; 1343 1344 lwkt_getpooltoken(list); 1345 restart: 1346 SLIST_FOREACH(kn, list, kn_next) { 1347 kq = kn->kn_kq; 1348 lwkt_getpooltoken(kq); 1349 1350 /* temporary verification hack */ 1351 SLIST_FOREACH(kntmp, list, kn_next) { 1352 if (kn == kntmp) 1353 break; 1354 } 1355 if (kn != kntmp || kn->kn_kq != kq) { 1356 lwkt_relpooltoken(kq); 1357 goto restart; 1358 } 1359 1360 if (kn->kn_status & KN_PROCESSING) { 1361 /* 1362 * Someone else is processing the knote, ask the 1363 * other thread to reprocess it and don't mess 1364 * with it otherwise. 1365 */ 1366 if (hint == 0) { 1367 kn->kn_status |= KN_REPROCESS; 1368 lwkt_relpooltoken(kq); 1369 continue; 1370 } 1371 1372 /* 1373 * If the hint is non-zero we have to wait or risk 1374 * losing the state the caller is trying to update. 1375 * 1376 * XXX This is a real problem, certain process 1377 * and signal filters will bump kn_data for 1378 * already-processed notes more than once if 1379 * we restart the list scan. FIXME. 1380 */ 1381 kn->kn_status |= KN_WAITING | KN_REPROCESS; 1382 tsleep(kn, 0, "knotec", hz); 1383 lwkt_relpooltoken(kq); 1384 goto restart; 1385 } 1386 1387 /* 1388 * Become the reprocessing master ourselves. 1389 * 1390 * If hint is non-zer running the event is mandatory 1391 * when not deleting so do it whether reprocessing is 1392 * set or not. 1393 */ 1394 kn->kn_status |= KN_PROCESSING; 1395 if ((kn->kn_status & KN_DELETING) == 0) { 1396 if (filter_event(kn, hint)) 1397 KNOTE_ACTIVATE(kn); 1398 } 1399 if (knote_release(kn)) { 1400 lwkt_relpooltoken(kq); 1401 goto restart; 1402 } 1403 lwkt_relpooltoken(kq); 1404 } 1405 lwkt_relpooltoken(list); 1406 } 1407 1408 /* 1409 * Insert knote at head of klist. 1410 * 1411 * This function may only be called via a filter function and thus 1412 * kq_token should already be held and marked for processing. 1413 */ 1414 void 1415 knote_insert(struct klist *klist, struct knote *kn) 1416 { 1417 lwkt_getpooltoken(klist); 1418 KKASSERT(kn->kn_status & KN_PROCESSING); 1419 SLIST_INSERT_HEAD(klist, kn, kn_next); 1420 lwkt_relpooltoken(klist); 1421 } 1422 1423 /* 1424 * Remove knote from a klist 1425 * 1426 * This function may only be called via a filter function and thus 1427 * kq_token should already be held and marked for processing. 1428 */ 1429 void 1430 knote_remove(struct klist *klist, struct knote *kn) 1431 { 1432 lwkt_getpooltoken(klist); 1433 KKASSERT(kn->kn_status & KN_PROCESSING); 1434 SLIST_REMOVE(klist, kn, knote, kn_next); 1435 lwkt_relpooltoken(klist); 1436 } 1437 1438 #if 0 1439 /* 1440 * Remove all knotes from a specified klist 1441 * 1442 * Only called from aio. 1443 */ 1444 void 1445 knote_empty(struct klist *list) 1446 { 1447 struct knote *kn; 1448 1449 lwkt_gettoken(&kq_token); 1450 while ((kn = SLIST_FIRST(list)) != NULL) { 1451 if (knote_acquire(kn)) 1452 knote_detach_and_drop(kn); 1453 } 1454 lwkt_reltoken(&kq_token); 1455 } 1456 #endif 1457 1458 void 1459 knote_assume_knotes(struct kqinfo *src, struct kqinfo *dst, 1460 struct filterops *ops, void *hook) 1461 { 1462 struct kqueue *kq; 1463 struct knote *kn; 1464 1465 lwkt_getpooltoken(&src->ki_note); 1466 lwkt_getpooltoken(&dst->ki_note); 1467 while ((kn = SLIST_FIRST(&src->ki_note)) != NULL) { 1468 kq = kn->kn_kq; 1469 lwkt_getpooltoken(kq); 1470 if (SLIST_FIRST(&src->ki_note) != kn || kn->kn_kq != kq) { 1471 lwkt_relpooltoken(kq); 1472 continue; 1473 } 1474 if (knote_acquire(kn)) { 1475 knote_remove(&src->ki_note, kn); 1476 kn->kn_fop = ops; 1477 kn->kn_hook = hook; 1478 knote_insert(&dst->ki_note, kn); 1479 knote_release(kn); 1480 /* kn may be invalid now */ 1481 } 1482 lwkt_relpooltoken(kq); 1483 } 1484 lwkt_relpooltoken(&dst->ki_note); 1485 lwkt_relpooltoken(&src->ki_note); 1486 } 1487 1488 /* 1489 * Remove all knotes referencing a specified fd 1490 */ 1491 void 1492 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd) 1493 { 1494 struct kqueue *kq; 1495 struct knote *kn; 1496 struct knote *kntmp; 1497 1498 lwkt_getpooltoken(&fp->f_klist); 1499 restart: 1500 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 1501 if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) { 1502 kq = kn->kn_kq; 1503 lwkt_getpooltoken(kq); 1504 1505 /* temporary verification hack */ 1506 SLIST_FOREACH(kntmp, &fp->f_klist, kn_link) { 1507 if (kn == kntmp) 1508 break; 1509 } 1510 if (kn != kntmp || kn->kn_kq->kq_fdp != fdp || 1511 kn->kn_id != fd || kn->kn_kq != kq) { 1512 lwkt_relpooltoken(kq); 1513 goto restart; 1514 } 1515 if (knote_acquire(kn)) 1516 knote_detach_and_drop(kn); 1517 lwkt_relpooltoken(kq); 1518 goto restart; 1519 } 1520 } 1521 lwkt_relpooltoken(&fp->f_klist); 1522 } 1523 1524 /* 1525 * Low level attach function. 1526 * 1527 * The knote should already be marked for processing. 1528 * Caller must hold the related kq token. 1529 */ 1530 static void 1531 knote_attach(struct knote *kn) 1532 { 1533 struct klist *list; 1534 struct kqueue *kq = kn->kn_kq; 1535 1536 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1537 KKASSERT(kn->kn_fp); 1538 list = &kn->kn_fp->f_klist; 1539 } else { 1540 if (kq->kq_knhashmask == 0) 1541 kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1542 &kq->kq_knhashmask); 1543 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1544 } 1545 lwkt_getpooltoken(list); 1546 SLIST_INSERT_HEAD(list, kn, kn_link); 1547 TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink); 1548 lwkt_relpooltoken(list); 1549 } 1550 1551 /* 1552 * Low level drop function. 1553 * 1554 * The knote should already be marked for processing. 1555 * Caller must hold the related kq token. 1556 */ 1557 static void 1558 knote_drop(struct knote *kn) 1559 { 1560 struct kqueue *kq; 1561 struct klist *list; 1562 1563 kq = kn->kn_kq; 1564 1565 if (kn->kn_fop->f_flags & FILTEROP_ISFD) 1566 list = &kn->kn_fp->f_klist; 1567 else 1568 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1569 1570 lwkt_getpooltoken(list); 1571 SLIST_REMOVE(list, kn, knote, kn_link); 1572 TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink); 1573 if (kn->kn_status & KN_QUEUED) 1574 knote_dequeue(kn); 1575 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1576 fdrop(kn->kn_fp); 1577 kn->kn_fp = NULL; 1578 } 1579 knote_free(kn); 1580 lwkt_relpooltoken(list); 1581 } 1582 1583 /* 1584 * Low level enqueue function. 1585 * 1586 * The knote should already be marked for processing. 1587 * Caller must be holding the kq token 1588 */ 1589 static void 1590 knote_enqueue(struct knote *kn) 1591 { 1592 struct kqueue *kq = kn->kn_kq; 1593 1594 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1595 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1596 kn->kn_status |= KN_QUEUED; 1597 ++kq->kq_count; 1598 1599 /* 1600 * Send SIGIO on request (typically set up as a mailbox signal) 1601 */ 1602 if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1) 1603 pgsigio(kq->kq_sigio, SIGIO, 0); 1604 1605 kqueue_wakeup(kq); 1606 } 1607 1608 /* 1609 * Low level dequeue function. 1610 * 1611 * The knote should already be marked for processing. 1612 * Caller must be holding the kq token 1613 */ 1614 static void 1615 knote_dequeue(struct knote *kn) 1616 { 1617 struct kqueue *kq = kn->kn_kq; 1618 1619 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1620 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1621 kn->kn_status &= ~KN_QUEUED; 1622 kq->kq_count--; 1623 } 1624 1625 static struct knote * 1626 knote_alloc(void) 1627 { 1628 return kmalloc(sizeof(struct knote), M_KQUEUE, M_WAITOK); 1629 } 1630 1631 static void 1632 knote_free(struct knote *kn) 1633 { 1634 kfree(kn, M_KQUEUE); 1635 } 1636