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 /* XXX locking? take proc_token here? */ 271 p = kn->kn_ptr.p_proc; 272 knote_remove(&p->p_klist, kn); 273 } 274 275 static int 276 filt_proc(struct knote *kn, long hint) 277 { 278 u_int event; 279 280 /* 281 * mask off extra data 282 */ 283 event = (u_int)hint & NOTE_PCTRLMASK; 284 285 /* 286 * if the user is interested in this event, record it. 287 */ 288 if (kn->kn_sfflags & event) 289 kn->kn_fflags |= event; 290 291 /* 292 * Process is gone, so flag the event as finished. Detach the 293 * knote from the process now because the process will be poof, 294 * gone later on. 295 */ 296 if (event == NOTE_EXIT) { 297 struct proc *p = kn->kn_ptr.p_proc; 298 if ((kn->kn_status & KN_DETACHED) == 0) { 299 PHOLD(p); 300 knote_remove(&p->p_klist, kn); 301 kn->kn_status |= KN_DETACHED; 302 kn->kn_data = p->p_xstat; 303 kn->kn_ptr.p_proc = NULL; 304 PRELE(p); 305 } 306 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 307 return (1); 308 } 309 310 /* 311 * process forked, and user wants to track the new process, 312 * so attach a new knote to it, and immediately report an 313 * event with the parent's pid. 314 */ 315 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 316 struct kevent kev; 317 int error; 318 319 /* 320 * register knote with new process. 321 */ 322 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 323 kev.filter = kn->kn_filter; 324 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 325 kev.fflags = kn->kn_sfflags; 326 kev.data = kn->kn_id; /* parent */ 327 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 328 error = kqueue_register(kn->kn_kq, &kev); 329 if (error) 330 kn->kn_fflags |= NOTE_TRACKERR; 331 } 332 333 return (kn->kn_fflags != 0); 334 } 335 336 /* 337 * The callout interlocks with callout_terminate() but can still 338 * race a deletion so if KN_DELETING is set we just don't touch 339 * the knote. 340 */ 341 static void 342 filt_timerexpire(void *knx) 343 { 344 struct lwkt_token *tok; 345 struct knote *kn = knx; 346 struct callout *calloutp; 347 struct timeval tv; 348 int tticks; 349 350 tok = lwkt_token_pool_lookup(kn->kn_kq); 351 lwkt_gettoken(tok); 352 if ((kn->kn_status & KN_DELETING) == 0) { 353 kn->kn_data++; 354 KNOTE_ACTIVATE(kn); 355 356 if ((kn->kn_flags & EV_ONESHOT) == 0) { 357 tv.tv_sec = kn->kn_sdata / 1000; 358 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 359 tticks = tvtohz_high(&tv); 360 calloutp = (struct callout *)kn->kn_hook; 361 callout_reset(calloutp, tticks, filt_timerexpire, kn); 362 } 363 } 364 lwkt_reltoken(tok); 365 } 366 367 /* 368 * data contains amount of time to sleep, in milliseconds 369 */ 370 static int 371 filt_timerattach(struct knote *kn) 372 { 373 struct callout *calloutp; 374 struct timeval tv; 375 int tticks; 376 377 if (kq_ncallouts >= kq_calloutmax) { 378 kn->kn_hook = NULL; 379 return (ENOMEM); 380 } 381 kq_ncallouts++; 382 383 tv.tv_sec = kn->kn_sdata / 1000; 384 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 385 tticks = tvtohz_high(&tv); 386 387 kn->kn_flags |= EV_CLEAR; /* automatically set */ 388 calloutp = kmalloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK); 389 callout_init(calloutp); 390 kn->kn_hook = (caddr_t)calloutp; 391 callout_reset(calloutp, tticks, filt_timerexpire, kn); 392 393 return (0); 394 } 395 396 /* 397 * This function is called with the knote flagged locked but it is 398 * still possible to race a callout event due to the callback blocking. 399 * We must call callout_terminate() instead of callout_stop() to deal 400 * with the race. 401 */ 402 static void 403 filt_timerdetach(struct knote *kn) 404 { 405 struct callout *calloutp; 406 407 calloutp = (struct callout *)kn->kn_hook; 408 callout_terminate(calloutp); 409 kfree(calloutp, M_KQUEUE); 410 kq_ncallouts--; 411 } 412 413 static int 414 filt_timer(struct knote *kn, long hint) 415 { 416 417 return (kn->kn_data != 0); 418 } 419 420 /* 421 * Acquire a knote, return non-zero on success, 0 on failure. 422 * 423 * If we cannot acquire the knote we sleep and return 0. The knote 424 * may be stale on return in this case and the caller must restart 425 * whatever loop they are in. 426 * 427 * Related kq token must be held. 428 */ 429 static __inline 430 int 431 knote_acquire(struct knote *kn) 432 { 433 if (kn->kn_status & KN_PROCESSING) { 434 kn->kn_status |= KN_WAITING | KN_REPROCESS; 435 tsleep(kn, 0, "kqepts", hz); 436 /* knote may be stale now */ 437 return(0); 438 } 439 kn->kn_status |= KN_PROCESSING; 440 return(1); 441 } 442 443 /* 444 * Release an acquired knote, clearing KN_PROCESSING and handling any 445 * KN_REPROCESS events. 446 * 447 * Caller must be holding the related kq token 448 * 449 * Non-zero is returned if the knote is destroyed or detached. 450 */ 451 static __inline 452 int 453 knote_release(struct knote *kn) 454 { 455 while (kn->kn_status & KN_REPROCESS) { 456 kn->kn_status &= ~KN_REPROCESS; 457 if (kn->kn_status & KN_WAITING) { 458 kn->kn_status &= ~KN_WAITING; 459 wakeup(kn); 460 } 461 if (kn->kn_status & KN_DELETING) { 462 knote_detach_and_drop(kn); 463 return(1); 464 /* NOT REACHED */ 465 } 466 if (filter_event(kn, 0)) 467 KNOTE_ACTIVATE(kn); 468 } 469 if (kn->kn_status & KN_DETACHED) { 470 kn->kn_status &= ~KN_PROCESSING; 471 return(1); 472 } else { 473 kn->kn_status &= ~KN_PROCESSING; 474 return(0); 475 } 476 } 477 478 /* 479 * Initialize a kqueue. 480 * 481 * NOTE: The lwp/proc code initializes a kqueue for select/poll ops. 482 * 483 * MPSAFE 484 */ 485 void 486 kqueue_init(struct kqueue *kq, struct filedesc *fdp) 487 { 488 TAILQ_INIT(&kq->kq_knpend); 489 TAILQ_INIT(&kq->kq_knlist); 490 kq->kq_count = 0; 491 kq->kq_fdp = fdp; 492 SLIST_INIT(&kq->kq_kqinfo.ki_note); 493 } 494 495 /* 496 * Terminate a kqueue. Freeing the actual kq itself is left up to the 497 * caller (it might be embedded in a lwp so we don't do it here). 498 * 499 * The kq's knlist must be completely eradicated so block on any 500 * processing races. 501 */ 502 void 503 kqueue_terminate(struct kqueue *kq) 504 { 505 struct lwkt_token *tok; 506 struct knote *kn; 507 508 tok = lwkt_token_pool_lookup(kq); 509 lwkt_gettoken(tok); 510 while ((kn = TAILQ_FIRST(&kq->kq_knlist)) != NULL) { 511 if (knote_acquire(kn)) 512 knote_detach_and_drop(kn); 513 } 514 if (kq->kq_knhash) { 515 hashdestroy(kq->kq_knhash, M_KQUEUE, kq->kq_knhashmask); 516 kq->kq_knhash = NULL; 517 kq->kq_knhashmask = 0; 518 } 519 lwkt_reltoken(tok); 520 } 521 522 /* 523 * MPSAFE 524 */ 525 int 526 sys_kqueue(struct kqueue_args *uap) 527 { 528 struct thread *td = curthread; 529 struct kqueue *kq; 530 struct file *fp; 531 int fd, error; 532 533 error = falloc(td->td_lwp, &fp, &fd); 534 if (error) 535 return (error); 536 fp->f_flag = FREAD | FWRITE; 537 fp->f_type = DTYPE_KQUEUE; 538 fp->f_ops = &kqueueops; 539 540 kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 541 kqueue_init(kq, td->td_proc->p_fd); 542 fp->f_data = kq; 543 544 fsetfd(kq->kq_fdp, fp, fd); 545 uap->sysmsg_result = fd; 546 fdrop(fp); 547 return (error); 548 } 549 550 /* 551 * Copy 'count' items into the destination list pointed to by uap->eventlist. 552 */ 553 static int 554 kevent_copyout(void *arg, struct kevent *kevp, int count, int *res) 555 { 556 struct kevent_copyin_args *kap; 557 int error; 558 559 kap = (struct kevent_copyin_args *)arg; 560 561 error = copyout(kevp, kap->ka->eventlist, count * sizeof(*kevp)); 562 if (error == 0) { 563 kap->ka->eventlist += count; 564 *res += count; 565 } else { 566 *res = -1; 567 } 568 569 return (error); 570 } 571 572 /* 573 * Copy at most 'max' items from the list pointed to by kap->changelist, 574 * return number of items in 'events'. 575 */ 576 static int 577 kevent_copyin(void *arg, struct kevent *kevp, int max, int *events) 578 { 579 struct kevent_copyin_args *kap; 580 int error, count; 581 582 kap = (struct kevent_copyin_args *)arg; 583 584 count = min(kap->ka->nchanges - kap->pchanges, max); 585 error = copyin(kap->ka->changelist, kevp, count * sizeof *kevp); 586 if (error == 0) { 587 kap->ka->changelist += count; 588 kap->pchanges += count; 589 *events = count; 590 } 591 592 return (error); 593 } 594 595 /* 596 * MPSAFE 597 */ 598 int 599 kern_kevent(struct kqueue *kq, int nevents, int *res, void *uap, 600 k_copyin_fn kevent_copyinfn, k_copyout_fn kevent_copyoutfn, 601 struct timespec *tsp_in) 602 { 603 struct kevent *kevp; 604 struct timespec *tsp; 605 int i, n, total, error, nerrors = 0; 606 int lres; 607 int limit = kq_checkloop; 608 struct kevent kev[KQ_NEVENTS]; 609 struct knote marker; 610 struct lwkt_token *tok; 611 612 if (tsp_in == NULL || tsp_in->tv_sec || tsp_in->tv_nsec) 613 atomic_set_int(&curthread->td_mpflags, TDF_MP_BATCH_DEMARC); 614 615 616 tsp = tsp_in; 617 *res = 0; 618 619 tok = lwkt_token_pool_lookup(kq); 620 lwkt_gettoken(tok); 621 for ( ;; ) { 622 n = 0; 623 error = kevent_copyinfn(uap, kev, KQ_NEVENTS, &n); 624 if (error) 625 goto done; 626 if (n == 0) 627 break; 628 for (i = 0; i < n; i++) { 629 kevp = &kev[i]; 630 kevp->flags &= ~EV_SYSFLAGS; 631 error = kqueue_register(kq, kevp); 632 633 /* 634 * If a registration returns an error we 635 * immediately post the error. The kevent() 636 * call itself will fail with the error if 637 * no space is available for posting. 638 * 639 * Such errors normally bypass the timeout/blocking 640 * code. However, if the copyoutfn function refuses 641 * to post the error (see sys_poll()), then we 642 * ignore it too. 643 */ 644 if (error) { 645 kevp->flags = EV_ERROR; 646 kevp->data = error; 647 lres = *res; 648 kevent_copyoutfn(uap, kevp, 1, res); 649 if (*res < 0) { 650 goto done; 651 } else if (lres != *res) { 652 nevents--; 653 nerrors++; 654 } 655 } 656 } 657 } 658 if (nerrors) { 659 error = 0; 660 goto done; 661 } 662 663 /* 664 * Acquire/wait for events - setup timeout 665 */ 666 if (tsp != NULL) { 667 struct timespec ats; 668 669 if (tsp->tv_sec || tsp->tv_nsec) { 670 getnanouptime(&ats); 671 timespecadd(tsp, &ats); /* tsp = target time */ 672 } 673 } 674 675 /* 676 * Loop as required. 677 * 678 * Collect as many events as we can. Sleeping on successive 679 * loops is disabled if copyoutfn has incremented (*res). 680 * 681 * The loop stops if an error occurs, all events have been 682 * scanned (the marker has been reached), or fewer than the 683 * maximum number of events is found. 684 * 685 * The copyoutfn function does not have to increment (*res) in 686 * order for the loop to continue. 687 * 688 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents. 689 */ 690 total = 0; 691 error = 0; 692 marker.kn_filter = EVFILT_MARKER; 693 marker.kn_status = KN_PROCESSING; 694 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 695 while ((n = nevents - total) > 0) { 696 if (n > KQ_NEVENTS) 697 n = KQ_NEVENTS; 698 699 /* 700 * If no events are pending sleep until timeout (if any) 701 * or an event occurs. 702 * 703 * After the sleep completes the marker is moved to the 704 * end of the list, making any received events available 705 * to our scan. 706 */ 707 if (kq->kq_count == 0 && *res == 0) { 708 error = kqueue_sleep(kq, tsp); 709 if (error) 710 break; 711 712 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 713 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 714 } 715 716 /* 717 * Process all received events 718 * Account for all non-spurious events in our total 719 */ 720 i = kqueue_scan(kq, kev, n, &marker); 721 if (i) { 722 lres = *res; 723 error = kevent_copyoutfn(uap, kev, i, res); 724 total += *res - lres; 725 if (error) 726 break; 727 } 728 if (limit && --limit == 0) 729 panic("kqueue: checkloop failed i=%d", i); 730 731 /* 732 * Normally when fewer events are returned than requested 733 * we can stop. However, if only spurious events were 734 * collected the copyout will not bump (*res) and we have 735 * to continue. 736 */ 737 if (i < n && *res) 738 break; 739 740 /* 741 * Deal with an edge case where spurious events can cause 742 * a loop to occur without moving the marker. This can 743 * prevent kqueue_scan() from picking up new events which 744 * race us. We must be sure to move the marker for this 745 * case. 746 * 747 * NOTE: We do not want to move the marker if events 748 * were scanned because normal kqueue operations 749 * may reactivate events. Moving the marker in 750 * that case could result in duplicates for the 751 * same event. 752 */ 753 if (i == 0) { 754 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 755 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 756 } 757 } 758 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 759 760 /* Timeouts do not return EWOULDBLOCK. */ 761 if (error == EWOULDBLOCK) 762 error = 0; 763 764 done: 765 lwkt_reltoken(tok); 766 return (error); 767 } 768 769 /* 770 * MPALMOSTSAFE 771 */ 772 int 773 sys_kevent(struct kevent_args *uap) 774 { 775 struct thread *td = curthread; 776 struct proc *p = td->td_proc; 777 struct timespec ts, *tsp; 778 struct kqueue *kq; 779 struct file *fp = NULL; 780 struct kevent_copyin_args *kap, ka; 781 int error; 782 783 if (uap->timeout) { 784 error = copyin(uap->timeout, &ts, sizeof(ts)); 785 if (error) 786 return (error); 787 tsp = &ts; 788 } else { 789 tsp = NULL; 790 } 791 fp = holdfp(p->p_fd, uap->fd, -1); 792 if (fp == NULL) 793 return (EBADF); 794 if (fp->f_type != DTYPE_KQUEUE) { 795 fdrop(fp); 796 return (EBADF); 797 } 798 799 kq = (struct kqueue *)fp->f_data; 800 801 kap = &ka; 802 kap->ka = uap; 803 kap->pchanges = 0; 804 805 error = kern_kevent(kq, uap->nevents, &uap->sysmsg_result, kap, 806 kevent_copyin, kevent_copyout, tsp); 807 808 fdrop(fp); 809 810 return (error); 811 } 812 813 /* 814 * Caller must be holding the kq token 815 */ 816 int 817 kqueue_register(struct kqueue *kq, struct kevent *kev) 818 { 819 struct lwkt_token *tok; 820 struct filedesc *fdp = kq->kq_fdp; 821 struct filterops *fops; 822 struct file *fp = NULL; 823 struct knote *kn = NULL; 824 int error = 0; 825 826 if (kev->filter < 0) { 827 if (kev->filter + EVFILT_SYSCOUNT < 0) 828 return (EINVAL); 829 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 830 } else { 831 /* 832 * XXX 833 * filter attach routine is responsible for insuring that 834 * the identifier can be attached to it. 835 */ 836 kprintf("unknown filter: %d\n", kev->filter); 837 return (EINVAL); 838 } 839 840 tok = lwkt_token_pool_lookup(kq); 841 lwkt_gettoken(tok); 842 if (fops->f_flags & FILTEROP_ISFD) { 843 /* validate descriptor */ 844 fp = holdfp(fdp, kev->ident, -1); 845 if (fp == NULL) { 846 lwkt_reltoken(tok); 847 return (EBADF); 848 } 849 lwkt_getpooltoken(&fp->f_klist); 850 again1: 851 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 852 if (kn->kn_kq == kq && 853 kn->kn_filter == kev->filter && 854 kn->kn_id == kev->ident) { 855 if (knote_acquire(kn) == 0) 856 goto again1; 857 break; 858 } 859 } 860 lwkt_relpooltoken(&fp->f_klist); 861 } else { 862 if (kq->kq_knhashmask) { 863 struct klist *list; 864 865 list = &kq->kq_knhash[ 866 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 867 lwkt_getpooltoken(list); 868 again2: 869 SLIST_FOREACH(kn, list, kn_link) { 870 if (kn->kn_id == kev->ident && 871 kn->kn_filter == kev->filter) { 872 if (knote_acquire(kn) == 0) 873 goto again2; 874 break; 875 } 876 } 877 lwkt_relpooltoken(list); 878 } 879 } 880 881 /* 882 * NOTE: At this point if kn is non-NULL we will have acquired 883 * it and set KN_PROCESSING. 884 */ 885 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 886 error = ENOENT; 887 goto done; 888 } 889 890 /* 891 * kn now contains the matching knote, or NULL if no match 892 */ 893 if (kev->flags & EV_ADD) { 894 if (kn == NULL) { 895 kn = knote_alloc(); 896 if (kn == NULL) { 897 error = ENOMEM; 898 goto done; 899 } 900 kn->kn_fp = fp; 901 kn->kn_kq = kq; 902 kn->kn_fop = fops; 903 904 /* 905 * apply reference count to knote structure, and 906 * do not release it at the end of this routine. 907 */ 908 fp = NULL; 909 910 kn->kn_sfflags = kev->fflags; 911 kn->kn_sdata = kev->data; 912 kev->fflags = 0; 913 kev->data = 0; 914 kn->kn_kevent = *kev; 915 916 /* 917 * KN_PROCESSING prevents the knote from getting 918 * ripped out from under us while we are trying 919 * to attach it, in case the attach blocks. 920 */ 921 kn->kn_status = KN_PROCESSING; 922 knote_attach(kn); 923 if ((error = filter_attach(kn)) != 0) { 924 kn->kn_status |= KN_DELETING | KN_REPROCESS; 925 knote_drop(kn); 926 goto done; 927 } 928 929 /* 930 * Interlock against close races which either tried 931 * to remove our knote while we were blocked or missed 932 * it entirely prior to our attachment. We do not 933 * want to end up with a knote on a closed descriptor. 934 */ 935 if ((fops->f_flags & FILTEROP_ISFD) && 936 checkfdclosed(fdp, kev->ident, kn->kn_fp)) { 937 kn->kn_status |= KN_DELETING | KN_REPROCESS; 938 } 939 } else { 940 /* 941 * The user may change some filter values after the 942 * initial EV_ADD, but doing so will not reset any 943 * filter which have already been triggered. 944 */ 945 KKASSERT(kn->kn_status & KN_PROCESSING); 946 kn->kn_sfflags = kev->fflags; 947 kn->kn_sdata = kev->data; 948 kn->kn_kevent.udata = kev->udata; 949 } 950 951 /* 952 * Execute the filter event to immediately activate the 953 * knote if necessary. If reprocessing events are pending 954 * due to blocking above we do not run the filter here 955 * but instead let knote_release() do it. Otherwise we 956 * might run the filter on a deleted event. 957 */ 958 if ((kn->kn_status & KN_REPROCESS) == 0) { 959 if (filter_event(kn, 0)) 960 KNOTE_ACTIVATE(kn); 961 } 962 } else if (kev->flags & EV_DELETE) { 963 /* 964 * Delete the existing knote 965 */ 966 knote_detach_and_drop(kn); 967 goto done; 968 } 969 970 /* 971 * Disablement does not deactivate a knote here. 972 */ 973 if ((kev->flags & EV_DISABLE) && 974 ((kn->kn_status & KN_DISABLED) == 0)) { 975 kn->kn_status |= KN_DISABLED; 976 } 977 978 /* 979 * Re-enablement may have to immediately enqueue an active knote. 980 */ 981 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 982 kn->kn_status &= ~KN_DISABLED; 983 if ((kn->kn_status & KN_ACTIVE) && 984 ((kn->kn_status & KN_QUEUED) == 0)) { 985 knote_enqueue(kn); 986 } 987 } 988 989 /* 990 * Handle any required reprocessing 991 */ 992 knote_release(kn); 993 /* kn may be invalid now */ 994 995 done: 996 lwkt_reltoken(tok); 997 if (fp != NULL) 998 fdrop(fp); 999 return (error); 1000 } 1001 1002 /* 1003 * Block as necessary until the target time is reached. 1004 * If tsp is NULL we block indefinitely. If tsp->ts_secs/nsecs are both 1005 * 0 we do not block at all. 1006 * 1007 * Caller must be holding the kq token. 1008 */ 1009 static int 1010 kqueue_sleep(struct kqueue *kq, struct timespec *tsp) 1011 { 1012 int error = 0; 1013 1014 if (tsp == NULL) { 1015 kq->kq_state |= KQ_SLEEP; 1016 error = tsleep(kq, PCATCH, "kqread", 0); 1017 } else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 1018 error = EWOULDBLOCK; 1019 } else { 1020 struct timespec ats; 1021 struct timespec atx = *tsp; 1022 int timeout; 1023 1024 getnanouptime(&ats); 1025 timespecsub(&atx, &ats); 1026 if (ats.tv_sec < 0) { 1027 error = EWOULDBLOCK; 1028 } else { 1029 timeout = atx.tv_sec > 24 * 60 * 60 ? 1030 24 * 60 * 60 * hz : tstohz_high(&atx); 1031 kq->kq_state |= KQ_SLEEP; 1032 error = tsleep(kq, PCATCH, "kqread", timeout); 1033 } 1034 } 1035 1036 /* don't restart after signals... */ 1037 if (error == ERESTART) 1038 return (EINTR); 1039 1040 return (error); 1041 } 1042 1043 /* 1044 * Scan the kqueue, return the number of active events placed in kevp up 1045 * to count. 1046 * 1047 * Continuous mode events may get recycled, do not continue scanning past 1048 * marker unless no events have been collected. 1049 * 1050 * Caller must be holding the kq token 1051 */ 1052 static int 1053 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 1054 struct knote *marker) 1055 { 1056 struct knote *kn, local_marker; 1057 int total; 1058 1059 total = 0; 1060 local_marker.kn_filter = EVFILT_MARKER; 1061 local_marker.kn_status = KN_PROCESSING; 1062 1063 /* 1064 * Collect events. 1065 */ 1066 TAILQ_INSERT_HEAD(&kq->kq_knpend, &local_marker, kn_tqe); 1067 while (count) { 1068 kn = TAILQ_NEXT(&local_marker, kn_tqe); 1069 if (kn->kn_filter == EVFILT_MARKER) { 1070 /* Marker reached, we are done */ 1071 if (kn == marker) 1072 break; 1073 1074 /* Move local marker past some other threads marker */ 1075 kn = TAILQ_NEXT(kn, kn_tqe); 1076 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1077 TAILQ_INSERT_BEFORE(kn, &local_marker, kn_tqe); 1078 continue; 1079 } 1080 1081 /* 1082 * We can't skip a knote undergoing processing, otherwise 1083 * we risk not returning it when the user process expects 1084 * it should be returned. Sleep and retry. 1085 */ 1086 if (knote_acquire(kn) == 0) 1087 continue; 1088 1089 /* 1090 * Remove the event for processing. 1091 * 1092 * WARNING! We must leave KN_QUEUED set to prevent the 1093 * event from being KNOTE_ACTIVATE()d while 1094 * the queue state is in limbo, in case we 1095 * block. 1096 * 1097 * WARNING! We must set KN_PROCESSING to avoid races 1098 * against deletion or another thread's 1099 * processing. 1100 */ 1101 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1102 kq->kq_count--; 1103 1104 /* 1105 * We have to deal with an extremely important race against 1106 * file descriptor close()s here. The file descriptor can 1107 * disappear MPSAFE, and there is a small window of 1108 * opportunity between that and the call to knote_fdclose(). 1109 * 1110 * If we hit that window here while doselect or dopoll is 1111 * trying to delete a spurious event they will not be able 1112 * to match up the event against a knote and will go haywire. 1113 */ 1114 if ((kn->kn_fop->f_flags & FILTEROP_ISFD) && 1115 checkfdclosed(kq->kq_fdp, kn->kn_kevent.ident, kn->kn_fp)) { 1116 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1117 } 1118 1119 if (kn->kn_status & KN_DISABLED) { 1120 /* 1121 * If disabled we ensure the event is not queued 1122 * but leave its active bit set. On re-enablement 1123 * the event may be immediately triggered. 1124 */ 1125 kn->kn_status &= ~KN_QUEUED; 1126 } else if ((kn->kn_flags & EV_ONESHOT) == 0 && 1127 (kn->kn_status & KN_DELETING) == 0 && 1128 filter_event(kn, 0) == 0) { 1129 /* 1130 * If not running in one-shot mode and the event 1131 * is no longer present we ensure it is removed 1132 * from the queue and ignore it. 1133 */ 1134 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1135 } else { 1136 /* 1137 * Post the event 1138 */ 1139 *kevp++ = kn->kn_kevent; 1140 ++total; 1141 --count; 1142 1143 if (kn->kn_flags & EV_ONESHOT) { 1144 kn->kn_status &= ~KN_QUEUED; 1145 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1146 } else if (kn->kn_flags & EV_CLEAR) { 1147 kn->kn_data = 0; 1148 kn->kn_fflags = 0; 1149 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1150 } else { 1151 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1152 kq->kq_count++; 1153 } 1154 } 1155 1156 /* 1157 * Handle any post-processing states 1158 */ 1159 knote_release(kn); 1160 } 1161 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1162 1163 return (total); 1164 } 1165 1166 /* 1167 * XXX 1168 * This could be expanded to call kqueue_scan, if desired. 1169 * 1170 * MPSAFE 1171 */ 1172 static int 1173 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1174 { 1175 return (ENXIO); 1176 } 1177 1178 /* 1179 * MPSAFE 1180 */ 1181 static int 1182 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1183 { 1184 return (ENXIO); 1185 } 1186 1187 /* 1188 * MPALMOSTSAFE 1189 */ 1190 static int 1191 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 1192 struct ucred *cred, struct sysmsg *msg) 1193 { 1194 struct lwkt_token *tok; 1195 struct kqueue *kq; 1196 int error; 1197 1198 kq = (struct kqueue *)fp->f_data; 1199 tok = lwkt_token_pool_lookup(kq); 1200 lwkt_gettoken(tok); 1201 1202 switch(com) { 1203 case FIOASYNC: 1204 if (*(int *)data) 1205 kq->kq_state |= KQ_ASYNC; 1206 else 1207 kq->kq_state &= ~KQ_ASYNC; 1208 error = 0; 1209 break; 1210 case FIOSETOWN: 1211 error = fsetown(*(int *)data, &kq->kq_sigio); 1212 break; 1213 default: 1214 error = ENOTTY; 1215 break; 1216 } 1217 lwkt_reltoken(tok); 1218 return (error); 1219 } 1220 1221 /* 1222 * MPSAFE 1223 */ 1224 static int 1225 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred) 1226 { 1227 struct kqueue *kq = (struct kqueue *)fp->f_data; 1228 1229 bzero((void *)st, sizeof(*st)); 1230 st->st_size = kq->kq_count; 1231 st->st_blksize = sizeof(struct kevent); 1232 st->st_mode = S_IFIFO; 1233 return (0); 1234 } 1235 1236 /* 1237 * MPSAFE 1238 */ 1239 static int 1240 kqueue_close(struct file *fp) 1241 { 1242 struct kqueue *kq = (struct kqueue *)fp->f_data; 1243 1244 kqueue_terminate(kq); 1245 1246 fp->f_data = NULL; 1247 funsetown(&kq->kq_sigio); 1248 1249 kfree(kq, M_KQUEUE); 1250 return (0); 1251 } 1252 1253 static void 1254 kqueue_wakeup(struct kqueue *kq) 1255 { 1256 if (kq->kq_state & KQ_SLEEP) { 1257 kq->kq_state &= ~KQ_SLEEP; 1258 wakeup(kq); 1259 } 1260 KNOTE(&kq->kq_kqinfo.ki_note, 0); 1261 } 1262 1263 /* 1264 * Calls filterops f_attach function, acquiring mplock if filter is not 1265 * marked as FILTEROP_MPSAFE. 1266 * 1267 * Caller must be holding the related kq token 1268 */ 1269 static int 1270 filter_attach(struct knote *kn) 1271 { 1272 int ret; 1273 1274 if (!(kn->kn_fop->f_flags & FILTEROP_MPSAFE)) { 1275 get_mplock(); 1276 ret = kn->kn_fop->f_attach(kn); 1277 rel_mplock(); 1278 } else { 1279 ret = kn->kn_fop->f_attach(kn); 1280 } 1281 1282 return (ret); 1283 } 1284 1285 /* 1286 * Detach the knote and drop it, destroying the knote. 1287 * 1288 * Calls filterops f_detach function, acquiring mplock if filter is not 1289 * marked as FILTEROP_MPSAFE. 1290 * 1291 * Caller must be holding the related kq token 1292 */ 1293 static void 1294 knote_detach_and_drop(struct knote *kn) 1295 { 1296 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1297 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1298 kn->kn_fop->f_detach(kn); 1299 } else { 1300 get_mplock(); 1301 kn->kn_fop->f_detach(kn); 1302 rel_mplock(); 1303 } 1304 knote_drop(kn); 1305 } 1306 1307 /* 1308 * Calls filterops f_event function, acquiring mplock if filter is not 1309 * marked as FILTEROP_MPSAFE. 1310 * 1311 * If the knote is in the middle of being created or deleted we cannot 1312 * safely call the filter op. 1313 * 1314 * Caller must be holding the related kq token 1315 */ 1316 static int 1317 filter_event(struct knote *kn, long hint) 1318 { 1319 int ret; 1320 1321 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1322 ret = kn->kn_fop->f_event(kn, hint); 1323 } else { 1324 get_mplock(); 1325 ret = kn->kn_fop->f_event(kn, hint); 1326 rel_mplock(); 1327 } 1328 return (ret); 1329 } 1330 1331 /* 1332 * Walk down a list of knotes, activating them if their event has triggered. 1333 * 1334 * If we encounter any knotes which are undergoing processing we just mark 1335 * them for reprocessing and do not try to [re]activate the knote. However, 1336 * if a hint is being passed we have to wait and that makes things a bit 1337 * sticky. 1338 */ 1339 void 1340 knote(struct klist *list, long hint) 1341 { 1342 struct kqueue *kq; 1343 struct knote *kn; 1344 struct knote *kntmp; 1345 1346 lwkt_getpooltoken(list); 1347 restart: 1348 SLIST_FOREACH(kn, list, kn_next) { 1349 kq = kn->kn_kq; 1350 lwkt_getpooltoken(kq); 1351 1352 /* temporary verification hack */ 1353 SLIST_FOREACH(kntmp, list, kn_next) { 1354 if (kn == kntmp) 1355 break; 1356 } 1357 if (kn != kntmp || kn->kn_kq != kq) { 1358 lwkt_relpooltoken(kq); 1359 goto restart; 1360 } 1361 1362 if (kn->kn_status & KN_PROCESSING) { 1363 /* 1364 * Someone else is processing the knote, ask the 1365 * other thread to reprocess it and don't mess 1366 * with it otherwise. 1367 */ 1368 if (hint == 0) { 1369 kn->kn_status |= KN_REPROCESS; 1370 lwkt_relpooltoken(kq); 1371 continue; 1372 } 1373 1374 /* 1375 * If the hint is non-zero we have to wait or risk 1376 * losing the state the caller is trying to update. 1377 * 1378 * XXX This is a real problem, certain process 1379 * and signal filters will bump kn_data for 1380 * already-processed notes more than once if 1381 * we restart the list scan. FIXME. 1382 */ 1383 kn->kn_status |= KN_WAITING | KN_REPROCESS; 1384 tsleep(kn, 0, "knotec", hz); 1385 lwkt_relpooltoken(kq); 1386 goto restart; 1387 } 1388 1389 /* 1390 * Become the reprocessing master ourselves. 1391 * 1392 * If hint is non-zer running the event is mandatory 1393 * when not deleting so do it whether reprocessing is 1394 * set or not. 1395 */ 1396 kn->kn_status |= KN_PROCESSING; 1397 if ((kn->kn_status & KN_DELETING) == 0) { 1398 if (filter_event(kn, hint)) 1399 KNOTE_ACTIVATE(kn); 1400 } 1401 if (knote_release(kn)) { 1402 lwkt_relpooltoken(kq); 1403 goto restart; 1404 } 1405 lwkt_relpooltoken(kq); 1406 } 1407 lwkt_relpooltoken(list); 1408 } 1409 1410 /* 1411 * Insert knote at head of klist. 1412 * 1413 * This function may only be called via a filter function and thus 1414 * kq_token should already be held and marked for processing. 1415 */ 1416 void 1417 knote_insert(struct klist *klist, struct knote *kn) 1418 { 1419 lwkt_getpooltoken(klist); 1420 KKASSERT(kn->kn_status & KN_PROCESSING); 1421 SLIST_INSERT_HEAD(klist, kn, kn_next); 1422 lwkt_relpooltoken(klist); 1423 } 1424 1425 /* 1426 * Remove knote from a klist 1427 * 1428 * This function may only be called via a filter function and thus 1429 * kq_token should already be held and marked for processing. 1430 */ 1431 void 1432 knote_remove(struct klist *klist, struct knote *kn) 1433 { 1434 lwkt_getpooltoken(klist); 1435 KKASSERT(kn->kn_status & KN_PROCESSING); 1436 SLIST_REMOVE(klist, kn, knote, kn_next); 1437 lwkt_relpooltoken(klist); 1438 } 1439 1440 #if 0 1441 /* 1442 * Remove all knotes from a specified klist 1443 * 1444 * Only called from aio. 1445 */ 1446 void 1447 knote_empty(struct klist *list) 1448 { 1449 struct knote *kn; 1450 1451 lwkt_gettoken(&kq_token); 1452 while ((kn = SLIST_FIRST(list)) != NULL) { 1453 if (knote_acquire(kn)) 1454 knote_detach_and_drop(kn); 1455 } 1456 lwkt_reltoken(&kq_token); 1457 } 1458 #endif 1459 1460 void 1461 knote_assume_knotes(struct kqinfo *src, struct kqinfo *dst, 1462 struct filterops *ops, void *hook) 1463 { 1464 struct kqueue *kq; 1465 struct knote *kn; 1466 1467 lwkt_getpooltoken(&src->ki_note); 1468 lwkt_getpooltoken(&dst->ki_note); 1469 while ((kn = SLIST_FIRST(&src->ki_note)) != NULL) { 1470 kq = kn->kn_kq; 1471 lwkt_getpooltoken(kq); 1472 if (SLIST_FIRST(&src->ki_note) != kn || kn->kn_kq != kq) { 1473 lwkt_relpooltoken(kq); 1474 continue; 1475 } 1476 if (knote_acquire(kn)) { 1477 knote_remove(&src->ki_note, kn); 1478 kn->kn_fop = ops; 1479 kn->kn_hook = hook; 1480 knote_insert(&dst->ki_note, kn); 1481 knote_release(kn); 1482 /* kn may be invalid now */ 1483 } 1484 lwkt_relpooltoken(kq); 1485 } 1486 lwkt_relpooltoken(&dst->ki_note); 1487 lwkt_relpooltoken(&src->ki_note); 1488 } 1489 1490 /* 1491 * Remove all knotes referencing a specified fd 1492 */ 1493 void 1494 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd) 1495 { 1496 struct kqueue *kq; 1497 struct knote *kn; 1498 struct knote *kntmp; 1499 1500 lwkt_getpooltoken(&fp->f_klist); 1501 restart: 1502 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 1503 if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) { 1504 kq = kn->kn_kq; 1505 lwkt_getpooltoken(kq); 1506 1507 /* temporary verification hack */ 1508 SLIST_FOREACH(kntmp, &fp->f_klist, kn_link) { 1509 if (kn == kntmp) 1510 break; 1511 } 1512 if (kn != kntmp || kn->kn_kq->kq_fdp != fdp || 1513 kn->kn_id != fd || kn->kn_kq != kq) { 1514 lwkt_relpooltoken(kq); 1515 goto restart; 1516 } 1517 if (knote_acquire(kn)) 1518 knote_detach_and_drop(kn); 1519 lwkt_relpooltoken(kq); 1520 goto restart; 1521 } 1522 } 1523 lwkt_relpooltoken(&fp->f_klist); 1524 } 1525 1526 /* 1527 * Low level attach function. 1528 * 1529 * The knote should already be marked for processing. 1530 * Caller must hold the related kq token. 1531 */ 1532 static void 1533 knote_attach(struct knote *kn) 1534 { 1535 struct klist *list; 1536 struct kqueue *kq = kn->kn_kq; 1537 1538 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1539 KKASSERT(kn->kn_fp); 1540 list = &kn->kn_fp->f_klist; 1541 } else { 1542 if (kq->kq_knhashmask == 0) 1543 kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1544 &kq->kq_knhashmask); 1545 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1546 } 1547 lwkt_getpooltoken(list); 1548 SLIST_INSERT_HEAD(list, kn, kn_link); 1549 TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink); 1550 lwkt_relpooltoken(list); 1551 } 1552 1553 /* 1554 * Low level drop function. 1555 * 1556 * The knote should already be marked for processing. 1557 * Caller must hold the related kq token. 1558 */ 1559 static void 1560 knote_drop(struct knote *kn) 1561 { 1562 struct kqueue *kq; 1563 struct klist *list; 1564 1565 kq = kn->kn_kq; 1566 1567 if (kn->kn_fop->f_flags & FILTEROP_ISFD) 1568 list = &kn->kn_fp->f_klist; 1569 else 1570 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1571 1572 lwkt_getpooltoken(list); 1573 SLIST_REMOVE(list, kn, knote, kn_link); 1574 TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink); 1575 if (kn->kn_status & KN_QUEUED) 1576 knote_dequeue(kn); 1577 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1578 fdrop(kn->kn_fp); 1579 kn->kn_fp = NULL; 1580 } 1581 knote_free(kn); 1582 lwkt_relpooltoken(list); 1583 } 1584 1585 /* 1586 * Low level enqueue function. 1587 * 1588 * The knote should already be marked for processing. 1589 * Caller must be holding the kq token 1590 */ 1591 static void 1592 knote_enqueue(struct knote *kn) 1593 { 1594 struct kqueue *kq = kn->kn_kq; 1595 1596 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1597 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1598 kn->kn_status |= KN_QUEUED; 1599 ++kq->kq_count; 1600 1601 /* 1602 * Send SIGIO on request (typically set up as a mailbox signal) 1603 */ 1604 if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1) 1605 pgsigio(kq->kq_sigio, SIGIO, 0); 1606 1607 kqueue_wakeup(kq); 1608 } 1609 1610 /* 1611 * Low level dequeue function. 1612 * 1613 * The knote should already be marked for processing. 1614 * Caller must be holding the kq token 1615 */ 1616 static void 1617 knote_dequeue(struct knote *kn) 1618 { 1619 struct kqueue *kq = kn->kn_kq; 1620 1621 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1622 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1623 kn->kn_status &= ~KN_QUEUED; 1624 kq->kq_count--; 1625 } 1626 1627 static struct knote * 1628 knote_alloc(void) 1629 { 1630 return kmalloc(sizeof(struct knote), M_KQUEUE, M_WAITOK); 1631 } 1632 1633 static void 1634 knote_free(struct knote *kn) 1635 { 1636 kfree(kn, M_KQUEUE); 1637 } 1638