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 #define EVENT_REGISTER 1 58 #define EVENT_PROCESS 2 59 60 /* 61 * Global token for kqueue subsystem 62 */ 63 #if 0 64 struct lwkt_token kq_token = LWKT_TOKEN_INITIALIZER(kq_token); 65 SYSCTL_LONG(_lwkt, OID_AUTO, kq_collisions, 66 CTLFLAG_RW, &kq_token.t_collisions, 0, 67 "Collision counter of kq_token"); 68 #endif 69 70 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 71 72 struct kevent_copyin_args { 73 struct kevent_args *ka; 74 int pchanges; 75 }; 76 77 static int kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 78 struct knote *marker); 79 static int kqueue_read(struct file *fp, struct uio *uio, 80 struct ucred *cred, int flags); 81 static int kqueue_write(struct file *fp, struct uio *uio, 82 struct ucred *cred, int flags); 83 static int kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 84 struct ucred *cred, struct sysmsg *msg); 85 static int kqueue_kqfilter(struct file *fp, struct knote *kn); 86 static int kqueue_stat(struct file *fp, struct stat *st, 87 struct ucred *cred); 88 static int kqueue_close(struct file *fp); 89 static void kqueue_wakeup(struct kqueue *kq); 90 static int filter_attach(struct knote *kn); 91 static int filter_event(struct knote *kn, long hint); 92 93 /* 94 * MPSAFE 95 */ 96 static struct fileops kqueueops = { 97 .fo_read = kqueue_read, 98 .fo_write = kqueue_write, 99 .fo_ioctl = kqueue_ioctl, 100 .fo_kqfilter = kqueue_kqfilter, 101 .fo_stat = kqueue_stat, 102 .fo_close = kqueue_close, 103 .fo_shutdown = nofo_shutdown 104 }; 105 106 static void knote_attach(struct knote *kn); 107 static void knote_drop(struct knote *kn); 108 static void knote_detach_and_drop(struct knote *kn); 109 static void knote_enqueue(struct knote *kn); 110 static void knote_dequeue(struct knote *kn); 111 static struct knote *knote_alloc(void); 112 static void knote_free(struct knote *kn); 113 114 static void filt_kqdetach(struct knote *kn); 115 static int filt_kqueue(struct knote *kn, long hint); 116 static int filt_procattach(struct knote *kn); 117 static void filt_procdetach(struct knote *kn); 118 static int filt_proc(struct knote *kn, long hint); 119 static int filt_fileattach(struct knote *kn); 120 static void filt_timerexpire(void *knx); 121 static int filt_timerattach(struct knote *kn); 122 static void filt_timerdetach(struct knote *kn); 123 static int filt_timer(struct knote *kn, long hint); 124 static int filt_userattach(struct knote *kn); 125 static void filt_userdetach(struct knote *kn); 126 static int filt_user(struct knote *kn, long hint); 127 static void filt_usertouch(struct knote *kn, struct kevent *kev, 128 u_long type); 129 130 static struct filterops file_filtops = 131 { FILTEROP_ISFD | FILTEROP_MPSAFE, filt_fileattach, NULL, NULL }; 132 static struct filterops kqread_filtops = 133 { FILTEROP_ISFD | FILTEROP_MPSAFE, NULL, filt_kqdetach, filt_kqueue }; 134 static struct filterops proc_filtops = 135 { 0, filt_procattach, filt_procdetach, filt_proc }; 136 static struct filterops timer_filtops = 137 { 0, filt_timerattach, filt_timerdetach, filt_timer }; 138 static struct filterops user_filtops = 139 { 0, filt_userattach, filt_userdetach, filt_user }; 140 141 static int kq_ncallouts = 0; 142 static int kq_calloutmax = (4 * 1024); 143 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW, 144 &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue"); 145 static int kq_checkloop = 1000000; 146 SYSCTL_INT(_kern, OID_AUTO, kq_checkloop, CTLFLAG_RW, 147 &kq_checkloop, 0, "Maximum number of callouts allocated for kqueue"); 148 149 #define KNOTE_ACTIVATE(kn) do { \ 150 kn->kn_status |= KN_ACTIVE; \ 151 if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) \ 152 knote_enqueue(kn); \ 153 } while(0) 154 155 #define KN_HASHSIZE 64 /* XXX should be tunable */ 156 #define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 157 158 extern struct filterops aio_filtops; 159 extern struct filterops sig_filtops; 160 161 /* 162 * Table for for all system-defined filters. 163 */ 164 static struct filterops *sysfilt_ops[] = { 165 &file_filtops, /* EVFILT_READ */ 166 &file_filtops, /* EVFILT_WRITE */ 167 &aio_filtops, /* EVFILT_AIO */ 168 &file_filtops, /* EVFILT_VNODE */ 169 &proc_filtops, /* EVFILT_PROC */ 170 &sig_filtops, /* EVFILT_SIGNAL */ 171 &timer_filtops, /* EVFILT_TIMER */ 172 &file_filtops, /* EVFILT_EXCEPT */ 173 &user_filtops, /* EVFILT_USER */ 174 }; 175 176 static int 177 filt_fileattach(struct knote *kn) 178 { 179 return (fo_kqfilter(kn->kn_fp, kn)); 180 } 181 182 /* 183 * MPSAFE 184 */ 185 static int 186 kqueue_kqfilter(struct file *fp, struct knote *kn) 187 { 188 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 189 190 if (kn->kn_filter != EVFILT_READ) 191 return (EOPNOTSUPP); 192 193 kn->kn_fop = &kqread_filtops; 194 knote_insert(&kq->kq_kqinfo.ki_note, kn); 195 return (0); 196 } 197 198 static void 199 filt_kqdetach(struct knote *kn) 200 { 201 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 202 203 knote_remove(&kq->kq_kqinfo.ki_note, kn); 204 } 205 206 /*ARGSUSED*/ 207 static int 208 filt_kqueue(struct knote *kn, long hint) 209 { 210 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 211 212 kn->kn_data = kq->kq_count; 213 return (kn->kn_data > 0); 214 } 215 216 static int 217 filt_procattach(struct knote *kn) 218 { 219 struct proc *p; 220 int immediate; 221 222 immediate = 0; 223 p = pfind(kn->kn_id); 224 if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) { 225 p = zpfind(kn->kn_id); 226 immediate = 1; 227 } 228 if (p == NULL) { 229 return (ESRCH); 230 } 231 if (!PRISON_CHECK(curthread->td_ucred, p->p_ucred)) { 232 if (p) 233 PRELE(p); 234 return (EACCES); 235 } 236 237 lwkt_gettoken(&p->p_token); 238 kn->kn_ptr.p_proc = p; 239 kn->kn_flags |= EV_CLEAR; /* automatically set */ 240 241 /* 242 * internal flag indicating registration done by kernel 243 */ 244 if (kn->kn_flags & EV_FLAG1) { 245 kn->kn_data = kn->kn_sdata; /* ppid */ 246 kn->kn_fflags = NOTE_CHILD; 247 kn->kn_flags &= ~EV_FLAG1; 248 } 249 250 knote_insert(&p->p_klist, kn); 251 252 /* 253 * Immediately activate any exit notes if the target process is a 254 * zombie. This is necessary to handle the case where the target 255 * process, e.g. a child, dies before the kevent is negistered. 256 */ 257 if (immediate && filt_proc(kn, NOTE_EXIT)) 258 KNOTE_ACTIVATE(kn); 259 lwkt_reltoken(&p->p_token); 260 PRELE(p); 261 262 return (0); 263 } 264 265 /* 266 * The knote may be attached to a different process, which may exit, 267 * leaving nothing for the knote to be attached to. So when the process 268 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so 269 * it will be deleted when read out. However, as part of the knote deletion, 270 * this routine is called, so a check is needed to avoid actually performing 271 * a detach, because the original process does not exist any more. 272 */ 273 static void 274 filt_procdetach(struct knote *kn) 275 { 276 struct proc *p; 277 278 if (kn->kn_status & KN_DETACHED) 279 return; 280 p = kn->kn_ptr.p_proc; 281 knote_remove(&p->p_klist, kn); 282 } 283 284 static int 285 filt_proc(struct knote *kn, long hint) 286 { 287 u_int event; 288 289 /* 290 * mask off extra data 291 */ 292 event = (u_int)hint & NOTE_PCTRLMASK; 293 294 /* 295 * if the user is interested in this event, record it. 296 */ 297 if (kn->kn_sfflags & event) 298 kn->kn_fflags |= event; 299 300 /* 301 * Process is gone, so flag the event as finished. Detach the 302 * knote from the process now because the process will be poof, 303 * gone later on. 304 */ 305 if (event == NOTE_EXIT) { 306 struct proc *p = kn->kn_ptr.p_proc; 307 if ((kn->kn_status & KN_DETACHED) == 0) { 308 PHOLD(p); 309 knote_remove(&p->p_klist, kn); 310 kn->kn_status |= KN_DETACHED; 311 kn->kn_data = p->p_xstat; 312 kn->kn_ptr.p_proc = NULL; 313 PRELE(p); 314 } 315 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 316 return (1); 317 } 318 319 /* 320 * process forked, and user wants to track the new process, 321 * so attach a new knote to it, and immediately report an 322 * event with the parent's pid. 323 */ 324 if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) { 325 struct kevent kev; 326 int error; 327 328 /* 329 * register knote with new process. 330 */ 331 kev.ident = hint & NOTE_PDATAMASK; /* pid */ 332 kev.filter = kn->kn_filter; 333 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1; 334 kev.fflags = kn->kn_sfflags; 335 kev.data = kn->kn_id; /* parent */ 336 kev.udata = kn->kn_kevent.udata; /* preserve udata */ 337 error = kqueue_register(kn->kn_kq, &kev); 338 if (error) 339 kn->kn_fflags |= NOTE_TRACKERR; 340 } 341 342 return (kn->kn_fflags != 0); 343 } 344 345 /* 346 * The callout interlocks with callout_terminate() but can still 347 * race a deletion so if KN_DELETING is set we just don't touch 348 * the knote. 349 */ 350 static void 351 filt_timerexpire(void *knx) 352 { 353 struct lwkt_token *tok; 354 struct knote *kn = knx; 355 struct callout *calloutp; 356 struct timeval tv; 357 int tticks; 358 359 tok = lwkt_token_pool_lookup(kn->kn_kq); 360 lwkt_gettoken(tok); 361 if ((kn->kn_status & KN_DELETING) == 0) { 362 kn->kn_data++; 363 KNOTE_ACTIVATE(kn); 364 365 if ((kn->kn_flags & EV_ONESHOT) == 0) { 366 tv.tv_sec = kn->kn_sdata / 1000; 367 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 368 tticks = tvtohz_high(&tv); 369 calloutp = (struct callout *)kn->kn_hook; 370 callout_reset(calloutp, tticks, filt_timerexpire, kn); 371 } 372 } 373 lwkt_reltoken(tok); 374 } 375 376 /* 377 * data contains amount of time to sleep, in milliseconds 378 */ 379 static int 380 filt_timerattach(struct knote *kn) 381 { 382 struct callout *calloutp; 383 struct timeval tv; 384 int tticks; 385 386 if (kq_ncallouts >= kq_calloutmax) { 387 kn->kn_hook = NULL; 388 return (ENOMEM); 389 } 390 kq_ncallouts++; 391 392 tv.tv_sec = kn->kn_sdata / 1000; 393 tv.tv_usec = (kn->kn_sdata % 1000) * 1000; 394 tticks = tvtohz_high(&tv); 395 396 kn->kn_flags |= EV_CLEAR; /* automatically set */ 397 calloutp = kmalloc(sizeof(*calloutp), M_KQUEUE, M_WAITOK); 398 callout_init(calloutp); 399 kn->kn_hook = (caddr_t)calloutp; 400 callout_reset(calloutp, tticks, filt_timerexpire, kn); 401 402 return (0); 403 } 404 405 /* 406 * This function is called with the knote flagged locked but it is 407 * still possible to race a callout event due to the callback blocking. 408 * We must call callout_terminate() instead of callout_stop() to deal 409 * with the race. 410 */ 411 static void 412 filt_timerdetach(struct knote *kn) 413 { 414 struct callout *calloutp; 415 416 calloutp = (struct callout *)kn->kn_hook; 417 callout_terminate(calloutp); 418 kfree(calloutp, M_KQUEUE); 419 kq_ncallouts--; 420 } 421 422 static int 423 filt_timer(struct knote *kn, long hint) 424 { 425 426 return (kn->kn_data != 0); 427 } 428 429 /* 430 * EVFILT_USER 431 */ 432 static int 433 filt_userattach(struct knote *kn) 434 { 435 kn->kn_hook = NULL; 436 if (kn->kn_fflags & NOTE_TRIGGER) 437 kn->kn_ptr.hookid = 1; 438 else 439 kn->kn_ptr.hookid = 0; 440 return 0; 441 } 442 443 /* 444 * This function is called with the knote flagged locked but it is 445 * still possible to race a callout event due to the callback blocking. 446 * We must call callout_terminate() instead of callout_stop() to deal 447 * with the race. 448 */ 449 static void 450 filt_userdetach(struct knote *kn) 451 { 452 /* nothing to do */ 453 } 454 455 static int 456 filt_user(struct knote *kn, long hint) 457 { 458 return (kn->kn_ptr.hookid); 459 } 460 461 static void 462 filt_usertouch(struct knote *kn, struct kevent *kev, u_long type) 463 { 464 u_int ffctrl; 465 466 switch (type) { 467 case EVENT_REGISTER: 468 if (kev->fflags & NOTE_TRIGGER) 469 kn->kn_ptr.hookid = 1; 470 471 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 472 kev->fflags &= NOTE_FFLAGSMASK; 473 switch (ffctrl) { 474 case NOTE_FFNOP: 475 break; 476 477 case NOTE_FFAND: 478 kn->kn_sfflags &= kev->fflags; 479 break; 480 481 case NOTE_FFOR: 482 kn->kn_sfflags |= kev->fflags; 483 break; 484 485 case NOTE_FFCOPY: 486 kn->kn_sfflags = kev->fflags; 487 break; 488 489 default: 490 /* XXX Return error? */ 491 break; 492 } 493 kn->kn_sdata = kev->data; 494 495 /* 496 * This is not the correct use of EV_CLEAR in an event 497 * modification, it should have been passed as a NOTE instead. 498 * But we need to maintain compatibility with Apple & FreeBSD. 499 * 500 * Note however that EV_CLEAR can still be used when doing 501 * the initial registration of the event and works as expected 502 * (clears the event on reception). 503 */ 504 if (kev->flags & EV_CLEAR) { 505 kn->kn_ptr.hookid = 0; 506 kn->kn_data = 0; 507 kn->kn_fflags = 0; 508 } 509 break; 510 511 case EVENT_PROCESS: 512 *kev = kn->kn_kevent; 513 kev->fflags = kn->kn_sfflags; 514 kev->data = kn->kn_sdata; 515 if (kn->kn_flags & EV_CLEAR) { 516 kn->kn_ptr.hookid = 0; 517 /* kn_data, kn_fflags handled by parent */ 518 } 519 break; 520 521 default: 522 panic("filt_usertouch() - invalid type (%ld)", type); 523 break; 524 } 525 } 526 527 /* 528 * Acquire a knote, return non-zero on success, 0 on failure. 529 * 530 * If we cannot acquire the knote we sleep and return 0. The knote 531 * may be stale on return in this case and the caller must restart 532 * whatever loop they are in. 533 * 534 * Related kq token must be held. 535 */ 536 static __inline 537 int 538 knote_acquire(struct knote *kn) 539 { 540 if (kn->kn_status & KN_PROCESSING) { 541 kn->kn_status |= KN_WAITING | KN_REPROCESS; 542 tsleep(kn, 0, "kqepts", hz); 543 /* knote may be stale now */ 544 return(0); 545 } 546 kn->kn_status |= KN_PROCESSING; 547 return(1); 548 } 549 550 /* 551 * Release an acquired knote, clearing KN_PROCESSING and handling any 552 * KN_REPROCESS events. 553 * 554 * Caller must be holding the related kq token 555 * 556 * Non-zero is returned if the knote is destroyed or detached. 557 */ 558 static __inline 559 int 560 knote_release(struct knote *kn) 561 { 562 while (kn->kn_status & KN_REPROCESS) { 563 kn->kn_status &= ~KN_REPROCESS; 564 if (kn->kn_status & KN_WAITING) { 565 kn->kn_status &= ~KN_WAITING; 566 wakeup(kn); 567 } 568 if (kn->kn_status & KN_DELETING) { 569 knote_detach_and_drop(kn); 570 return(1); 571 /* NOT REACHED */ 572 } 573 if (filter_event(kn, 0)) 574 KNOTE_ACTIVATE(kn); 575 } 576 if (kn->kn_status & KN_DETACHED) { 577 kn->kn_status &= ~KN_PROCESSING; 578 return(1); 579 } else { 580 kn->kn_status &= ~KN_PROCESSING; 581 return(0); 582 } 583 } 584 585 /* 586 * Initialize a kqueue. 587 * 588 * NOTE: The lwp/proc code initializes a kqueue for select/poll ops. 589 * 590 * MPSAFE 591 */ 592 void 593 kqueue_init(struct kqueue *kq, struct filedesc *fdp) 594 { 595 TAILQ_INIT(&kq->kq_knpend); 596 TAILQ_INIT(&kq->kq_knlist); 597 kq->kq_count = 0; 598 kq->kq_fdp = fdp; 599 SLIST_INIT(&kq->kq_kqinfo.ki_note); 600 } 601 602 /* 603 * Terminate a kqueue. Freeing the actual kq itself is left up to the 604 * caller (it might be embedded in a lwp so we don't do it here). 605 * 606 * The kq's knlist must be completely eradicated so block on any 607 * processing races. 608 */ 609 void 610 kqueue_terminate(struct kqueue *kq) 611 { 612 struct lwkt_token *tok; 613 struct knote *kn; 614 615 tok = lwkt_token_pool_lookup(kq); 616 lwkt_gettoken(tok); 617 while ((kn = TAILQ_FIRST(&kq->kq_knlist)) != NULL) { 618 if (knote_acquire(kn)) 619 knote_detach_and_drop(kn); 620 } 621 if (kq->kq_knhash) { 622 hashdestroy(kq->kq_knhash, M_KQUEUE, kq->kq_knhashmask); 623 kq->kq_knhash = NULL; 624 kq->kq_knhashmask = 0; 625 } 626 lwkt_reltoken(tok); 627 } 628 629 /* 630 * MPSAFE 631 */ 632 int 633 sys_kqueue(struct kqueue_args *uap) 634 { 635 struct thread *td = curthread; 636 struct kqueue *kq; 637 struct file *fp; 638 int fd, error; 639 640 error = falloc(td->td_lwp, &fp, &fd); 641 if (error) 642 return (error); 643 fp->f_flag = FREAD | FWRITE; 644 fp->f_type = DTYPE_KQUEUE; 645 fp->f_ops = &kqueueops; 646 647 kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO); 648 kqueue_init(kq, td->td_proc->p_fd); 649 fp->f_data = kq; 650 651 fsetfd(kq->kq_fdp, fp, fd); 652 uap->sysmsg_result = fd; 653 fdrop(fp); 654 return (error); 655 } 656 657 /* 658 * Copy 'count' items into the destination list pointed to by uap->eventlist. 659 */ 660 static int 661 kevent_copyout(void *arg, struct kevent *kevp, int count, int *res) 662 { 663 struct kevent_copyin_args *kap; 664 int error; 665 666 kap = (struct kevent_copyin_args *)arg; 667 668 error = copyout(kevp, kap->ka->eventlist, count * sizeof(*kevp)); 669 if (error == 0) { 670 kap->ka->eventlist += count; 671 *res += count; 672 } else { 673 *res = -1; 674 } 675 676 return (error); 677 } 678 679 /* 680 * Copy at most 'max' items from the list pointed to by kap->changelist, 681 * return number of items in 'events'. 682 */ 683 static int 684 kevent_copyin(void *arg, struct kevent *kevp, int max, int *events) 685 { 686 struct kevent_copyin_args *kap; 687 int error, count; 688 689 kap = (struct kevent_copyin_args *)arg; 690 691 count = min(kap->ka->nchanges - kap->pchanges, max); 692 error = copyin(kap->ka->changelist, kevp, count * sizeof *kevp); 693 if (error == 0) { 694 kap->ka->changelist += count; 695 kap->pchanges += count; 696 *events = count; 697 } 698 699 return (error); 700 } 701 702 /* 703 * MPSAFE 704 */ 705 int 706 kern_kevent(struct kqueue *kq, int nevents, int *res, void *uap, 707 k_copyin_fn kevent_copyinfn, k_copyout_fn kevent_copyoutfn, 708 struct timespec *tsp_in) 709 { 710 struct kevent *kevp; 711 struct timespec *tsp, ats; 712 int i, n, total, error, nerrors = 0; 713 int lres; 714 int limit = kq_checkloop; 715 struct kevent kev[KQ_NEVENTS]; 716 struct knote marker; 717 struct lwkt_token *tok; 718 719 if (tsp_in == NULL || tsp_in->tv_sec || tsp_in->tv_nsec) 720 atomic_set_int(&curthread->td_mpflags, TDF_MP_BATCH_DEMARC); 721 722 tsp = tsp_in; 723 *res = 0; 724 725 for (;;) { 726 n = 0; 727 error = kevent_copyinfn(uap, kev, KQ_NEVENTS, &n); 728 if (error) 729 return error; 730 if (n == 0) 731 break; 732 for (i = 0; i < n; i++) { 733 kevp = &kev[i]; 734 kevp->flags &= ~EV_SYSFLAGS; 735 error = kqueue_register(kq, kevp); 736 737 /* 738 * If a registration returns an error we 739 * immediately post the error. The kevent() 740 * call itself will fail with the error if 741 * no space is available for posting. 742 * 743 * Such errors normally bypass the timeout/blocking 744 * code. However, if the copyoutfn function refuses 745 * to post the error (see sys_poll()), then we 746 * ignore it too. 747 */ 748 if (error) { 749 kevp->flags = EV_ERROR; 750 kevp->data = error; 751 lres = *res; 752 kevent_copyoutfn(uap, kevp, 1, res); 753 if (*res < 0) { 754 return error; 755 } else if (lres != *res) { 756 nevents--; 757 nerrors++; 758 } 759 } 760 } 761 } 762 if (nerrors) 763 return 0; 764 765 /* 766 * Acquire/wait for events - setup timeout 767 */ 768 if (tsp != NULL) { 769 if (tsp->tv_sec || tsp->tv_nsec) { 770 getnanouptime(&ats); 771 timespecadd(tsp, &ats); /* tsp = target time */ 772 } 773 } 774 775 /* 776 * Loop as required. 777 * 778 * Collect as many events as we can. Sleeping on successive 779 * loops is disabled if copyoutfn has incremented (*res). 780 * 781 * The loop stops if an error occurs, all events have been 782 * scanned (the marker has been reached), or fewer than the 783 * maximum number of events is found. 784 * 785 * The copyoutfn function does not have to increment (*res) in 786 * order for the loop to continue. 787 * 788 * NOTE: doselect() usually passes 0x7FFFFFFF for nevents. 789 */ 790 total = 0; 791 error = 0; 792 marker.kn_filter = EVFILT_MARKER; 793 marker.kn_status = KN_PROCESSING; 794 tok = lwkt_token_pool_lookup(kq); 795 lwkt_gettoken(tok); 796 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 797 lwkt_reltoken(tok); 798 while ((n = nevents - total) > 0) { 799 if (n > KQ_NEVENTS) 800 n = KQ_NEVENTS; 801 802 /* 803 * If no events are pending sleep until timeout (if any) 804 * or an event occurs. 805 * 806 * After the sleep completes the marker is moved to the 807 * end of the list, making any received events available 808 * to our scan. 809 */ 810 if (kq->kq_count == 0 && *res == 0) { 811 int timeout; 812 813 if (tsp == NULL) { 814 timeout = 0; 815 } else if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) { 816 error = EWOULDBLOCK; 817 break; 818 } else { 819 struct timespec atx = *tsp; 820 821 getnanouptime(&ats); 822 timespecsub(&atx, &ats); 823 if (atx.tv_sec < 0) { 824 error = EWOULDBLOCK; 825 break; 826 } else { 827 timeout = atx.tv_sec > 24 * 60 * 60 ? 828 24 * 60 * 60 * hz : 829 tstohz_high(&atx); 830 } 831 } 832 833 lwkt_gettoken(tok); 834 if (kq->kq_count == 0) { 835 kq->kq_state |= KQ_SLEEP; 836 error = tsleep(kq, PCATCH, "kqread", timeout); 837 838 /* don't restart after signals... */ 839 if (error == ERESTART) 840 error = EINTR; 841 if (error) { 842 lwkt_reltoken(tok); 843 break; 844 } 845 846 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 847 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, 848 kn_tqe); 849 } 850 lwkt_reltoken(tok); 851 } 852 853 /* 854 * Process all received events 855 * Account for all non-spurious events in our total 856 */ 857 i = kqueue_scan(kq, kev, n, &marker); 858 if (i) { 859 lres = *res; 860 error = kevent_copyoutfn(uap, kev, i, res); 861 total += *res - lres; 862 if (error) 863 break; 864 } 865 if (limit && --limit == 0) 866 panic("kqueue: checkloop failed i=%d", i); 867 868 /* 869 * Normally when fewer events are returned than requested 870 * we can stop. However, if only spurious events were 871 * collected the copyout will not bump (*res) and we have 872 * to continue. 873 */ 874 if (i < n && *res) 875 break; 876 877 /* 878 * Deal with an edge case where spurious events can cause 879 * a loop to occur without moving the marker. This can 880 * prevent kqueue_scan() from picking up new events which 881 * race us. We must be sure to move the marker for this 882 * case. 883 * 884 * NOTE: We do not want to move the marker if events 885 * were scanned because normal kqueue operations 886 * may reactivate events. Moving the marker in 887 * that case could result in duplicates for the 888 * same event. 889 */ 890 if (i == 0) { 891 lwkt_gettoken(tok); 892 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 893 TAILQ_INSERT_TAIL(&kq->kq_knpend, &marker, kn_tqe); 894 lwkt_reltoken(tok); 895 } 896 } 897 lwkt_gettoken(tok); 898 TAILQ_REMOVE(&kq->kq_knpend, &marker, kn_tqe); 899 lwkt_reltoken(tok); 900 901 /* Timeouts do not return EWOULDBLOCK. */ 902 if (error == EWOULDBLOCK) 903 error = 0; 904 return error; 905 } 906 907 /* 908 * MPALMOSTSAFE 909 */ 910 int 911 sys_kevent(struct kevent_args *uap) 912 { 913 struct thread *td = curthread; 914 struct proc *p = td->td_proc; 915 struct timespec ts, *tsp; 916 struct kqueue *kq; 917 struct file *fp = NULL; 918 struct kevent_copyin_args *kap, ka; 919 int error; 920 921 if (uap->timeout) { 922 error = copyin(uap->timeout, &ts, sizeof(ts)); 923 if (error) 924 return (error); 925 tsp = &ts; 926 } else { 927 tsp = NULL; 928 } 929 fp = holdfp(p->p_fd, uap->fd, -1); 930 if (fp == NULL) 931 return (EBADF); 932 if (fp->f_type != DTYPE_KQUEUE) { 933 fdrop(fp); 934 return (EBADF); 935 } 936 937 kq = (struct kqueue *)fp->f_data; 938 939 kap = &ka; 940 kap->ka = uap; 941 kap->pchanges = 0; 942 943 error = kern_kevent(kq, uap->nevents, &uap->sysmsg_result, kap, 944 kevent_copyin, kevent_copyout, tsp); 945 946 fdrop(fp); 947 948 return (error); 949 } 950 951 int 952 kqueue_register(struct kqueue *kq, struct kevent *kev) 953 { 954 struct lwkt_token *tok; 955 struct filedesc *fdp = kq->kq_fdp; 956 struct filterops *fops; 957 struct file *fp = NULL; 958 struct knote *kn = NULL; 959 int error = 0; 960 961 if (kev->filter < 0) { 962 if (kev->filter + EVFILT_SYSCOUNT < 0) 963 return (EINVAL); 964 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 965 } else { 966 /* 967 * XXX 968 * filter attach routine is responsible for insuring that 969 * the identifier can be attached to it. 970 */ 971 return (EINVAL); 972 } 973 974 tok = lwkt_token_pool_lookup(kq); 975 lwkt_gettoken(tok); 976 if (fops->f_flags & FILTEROP_ISFD) { 977 /* validate descriptor */ 978 fp = holdfp(fdp, kev->ident, -1); 979 if (fp == NULL) { 980 lwkt_reltoken(tok); 981 return (EBADF); 982 } 983 lwkt_getpooltoken(&fp->f_klist); 984 again1: 985 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 986 if (kn->kn_kq == kq && 987 kn->kn_filter == kev->filter && 988 kn->kn_id == kev->ident) { 989 if (knote_acquire(kn) == 0) 990 goto again1; 991 break; 992 } 993 } 994 lwkt_relpooltoken(&fp->f_klist); 995 } else { 996 if (kq->kq_knhashmask) { 997 struct klist *list; 998 999 list = &kq->kq_knhash[ 1000 KN_HASH((u_long)kev->ident, kq->kq_knhashmask)]; 1001 lwkt_getpooltoken(list); 1002 again2: 1003 SLIST_FOREACH(kn, list, kn_link) { 1004 if (kn->kn_id == kev->ident && 1005 kn->kn_filter == kev->filter) { 1006 if (knote_acquire(kn) == 0) 1007 goto again2; 1008 break; 1009 } 1010 } 1011 lwkt_relpooltoken(list); 1012 } 1013 } 1014 1015 /* 1016 * NOTE: At this point if kn is non-NULL we will have acquired 1017 * it and set KN_PROCESSING. 1018 */ 1019 if (kn == NULL && ((kev->flags & EV_ADD) == 0)) { 1020 error = ENOENT; 1021 goto done; 1022 } 1023 1024 /* 1025 * kn now contains the matching knote, or NULL if no match 1026 */ 1027 if (kev->flags & EV_ADD) { 1028 if (kn == NULL) { 1029 kn = knote_alloc(); 1030 if (kn == NULL) { 1031 error = ENOMEM; 1032 goto done; 1033 } 1034 kn->kn_fp = fp; 1035 kn->kn_kq = kq; 1036 kn->kn_fop = fops; 1037 1038 /* 1039 * apply reference count to knote structure, and 1040 * do not release it at the end of this routine. 1041 */ 1042 fp = NULL; 1043 1044 kn->kn_sfflags = kev->fflags; 1045 kn->kn_sdata = kev->data; 1046 kev->fflags = 0; 1047 kev->data = 0; 1048 kn->kn_kevent = *kev; 1049 1050 /* 1051 * KN_PROCESSING prevents the knote from getting 1052 * ripped out from under us while we are trying 1053 * to attach it, in case the attach blocks. 1054 */ 1055 kn->kn_status = KN_PROCESSING; 1056 knote_attach(kn); 1057 if ((error = filter_attach(kn)) != 0) { 1058 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1059 knote_drop(kn); 1060 goto done; 1061 } 1062 1063 /* 1064 * Interlock against close races which either tried 1065 * to remove our knote while we were blocked or missed 1066 * it entirely prior to our attachment. We do not 1067 * want to end up with a knote on a closed descriptor. 1068 */ 1069 if ((fops->f_flags & FILTEROP_ISFD) && 1070 checkfdclosed(fdp, kev->ident, kn->kn_fp)) { 1071 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1072 } 1073 } else { 1074 /* 1075 * The user may change some filter values after the 1076 * initial EV_ADD, but doing so will not reset any 1077 * filter which have already been triggered. 1078 */ 1079 KKASSERT(kn->kn_status & KN_PROCESSING); 1080 if (fops == &user_filtops) { 1081 filt_usertouch(kn, kev, EVENT_REGISTER); 1082 } else { 1083 kn->kn_sfflags = kev->fflags; 1084 kn->kn_sdata = kev->data; 1085 kn->kn_kevent.udata = kev->udata; 1086 } 1087 } 1088 1089 /* 1090 * Execute the filter event to immediately activate the 1091 * knote if necessary. If reprocessing events are pending 1092 * due to blocking above we do not run the filter here 1093 * but instead let knote_release() do it. Otherwise we 1094 * might run the filter on a deleted event. 1095 */ 1096 if ((kn->kn_status & KN_REPROCESS) == 0) { 1097 if (filter_event(kn, 0)) 1098 KNOTE_ACTIVATE(kn); 1099 } 1100 } else if (kev->flags & EV_DELETE) { 1101 /* 1102 * Delete the existing knote 1103 */ 1104 knote_detach_and_drop(kn); 1105 goto done; 1106 } else { 1107 /* 1108 * Modify an existing event. 1109 * 1110 * The user may change some filter values after the 1111 * initial EV_ADD, but doing so will not reset any 1112 * filter which have already been triggered. 1113 */ 1114 KKASSERT(kn->kn_status & KN_PROCESSING); 1115 if (fops == &user_filtops) { 1116 filt_usertouch(kn, kev, EVENT_REGISTER); 1117 } else { 1118 kn->kn_sfflags = kev->fflags; 1119 kn->kn_sdata = kev->data; 1120 kn->kn_kevent.udata = kev->udata; 1121 } 1122 1123 /* 1124 * Execute the filter event to immediately activate the 1125 * knote if necessary. If reprocessing events are pending 1126 * due to blocking above we do not run the filter here 1127 * but instead let knote_release() do it. Otherwise we 1128 * might run the filter on a deleted event. 1129 */ 1130 if ((kn->kn_status & KN_REPROCESS) == 0) { 1131 if (filter_event(kn, 0)) 1132 KNOTE_ACTIVATE(kn); 1133 } 1134 } 1135 1136 /* 1137 * Disablement does not deactivate a knote here. 1138 */ 1139 if ((kev->flags & EV_DISABLE) && 1140 ((kn->kn_status & KN_DISABLED) == 0)) { 1141 kn->kn_status |= KN_DISABLED; 1142 } 1143 1144 /* 1145 * Re-enablement may have to immediately enqueue an active knote. 1146 */ 1147 if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) { 1148 kn->kn_status &= ~KN_DISABLED; 1149 if ((kn->kn_status & KN_ACTIVE) && 1150 ((kn->kn_status & KN_QUEUED) == 0)) { 1151 knote_enqueue(kn); 1152 } 1153 } 1154 1155 /* 1156 * Handle any required reprocessing 1157 */ 1158 knote_release(kn); 1159 /* kn may be invalid now */ 1160 1161 done: 1162 lwkt_reltoken(tok); 1163 if (fp != NULL) 1164 fdrop(fp); 1165 return (error); 1166 } 1167 1168 /* 1169 * Scan the kqueue, return the number of active events placed in kevp up 1170 * to count. 1171 * 1172 * Continuous mode events may get recycled, do not continue scanning past 1173 * marker unless no events have been collected. 1174 */ 1175 static int 1176 kqueue_scan(struct kqueue *kq, struct kevent *kevp, int count, 1177 struct knote *marker) 1178 { 1179 struct knote *kn, local_marker; 1180 int total; 1181 1182 total = 0; 1183 local_marker.kn_filter = EVFILT_MARKER; 1184 local_marker.kn_status = KN_PROCESSING; 1185 1186 lwkt_getpooltoken(kq); 1187 1188 /* 1189 * Collect events. 1190 */ 1191 TAILQ_INSERT_HEAD(&kq->kq_knpend, &local_marker, kn_tqe); 1192 while (count) { 1193 kn = TAILQ_NEXT(&local_marker, kn_tqe); 1194 if (kn->kn_filter == EVFILT_MARKER) { 1195 /* Marker reached, we are done */ 1196 if (kn == marker) 1197 break; 1198 1199 /* Move local marker past some other threads marker */ 1200 kn = TAILQ_NEXT(kn, kn_tqe); 1201 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1202 TAILQ_INSERT_BEFORE(kn, &local_marker, kn_tqe); 1203 continue; 1204 } 1205 1206 /* 1207 * We can't skip a knote undergoing processing, otherwise 1208 * we risk not returning it when the user process expects 1209 * it should be returned. Sleep and retry. 1210 */ 1211 if (knote_acquire(kn) == 0) 1212 continue; 1213 1214 /* 1215 * Remove the event for processing. 1216 * 1217 * WARNING! We must leave KN_QUEUED set to prevent the 1218 * event from being KNOTE_ACTIVATE()d while 1219 * the queue state is in limbo, in case we 1220 * block. 1221 * 1222 * WARNING! We must set KN_PROCESSING to avoid races 1223 * against deletion or another thread's 1224 * processing. 1225 */ 1226 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1227 kq->kq_count--; 1228 1229 /* 1230 * We have to deal with an extremely important race against 1231 * file descriptor close()s here. The file descriptor can 1232 * disappear MPSAFE, and there is a small window of 1233 * opportunity between that and the call to knote_fdclose(). 1234 * 1235 * If we hit that window here while doselect or dopoll is 1236 * trying to delete a spurious event they will not be able 1237 * to match up the event against a knote and will go haywire. 1238 */ 1239 if ((kn->kn_fop->f_flags & FILTEROP_ISFD) && 1240 checkfdclosed(kq->kq_fdp, kn->kn_kevent.ident, kn->kn_fp)) { 1241 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1242 } 1243 1244 if (kn->kn_status & KN_DISABLED) { 1245 /* 1246 * If disabled we ensure the event is not queued 1247 * but leave its active bit set. On re-enablement 1248 * the event may be immediately triggered. 1249 */ 1250 kn->kn_status &= ~KN_QUEUED; 1251 } else if ((kn->kn_flags & EV_ONESHOT) == 0 && 1252 (kn->kn_status & KN_DELETING) == 0 && 1253 filter_event(kn, 0) == 0) { 1254 /* 1255 * If not running in one-shot mode and the event 1256 * is no longer present we ensure it is removed 1257 * from the queue and ignore it. 1258 */ 1259 kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE); 1260 } else { 1261 /* 1262 * Post the event 1263 */ 1264 if (kn->kn_fop == &user_filtops) 1265 filt_usertouch(kn, kevp, EVENT_PROCESS); 1266 else 1267 *kevp = kn->kn_kevent; 1268 ++kevp; 1269 ++total; 1270 --count; 1271 1272 if (kn->kn_flags & EV_ONESHOT) { 1273 kn->kn_status &= ~KN_QUEUED; 1274 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1275 } else { 1276 if (kn->kn_flags & EV_CLEAR) { 1277 kn->kn_data = 0; 1278 kn->kn_fflags = 0; 1279 kn->kn_status &= ~(KN_QUEUED | 1280 KN_ACTIVE); 1281 } else { 1282 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1283 kq->kq_count++; 1284 } 1285 } 1286 } 1287 1288 /* 1289 * Handle any post-processing states 1290 */ 1291 knote_release(kn); 1292 } 1293 TAILQ_REMOVE(&kq->kq_knpend, &local_marker, kn_tqe); 1294 1295 lwkt_relpooltoken(kq); 1296 return (total); 1297 } 1298 1299 /* 1300 * XXX 1301 * This could be expanded to call kqueue_scan, if desired. 1302 * 1303 * MPSAFE 1304 */ 1305 static int 1306 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1307 { 1308 return (ENXIO); 1309 } 1310 1311 /* 1312 * MPSAFE 1313 */ 1314 static int 1315 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags) 1316 { 1317 return (ENXIO); 1318 } 1319 1320 /* 1321 * MPALMOSTSAFE 1322 */ 1323 static int 1324 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, 1325 struct ucred *cred, struct sysmsg *msg) 1326 { 1327 struct lwkt_token *tok; 1328 struct kqueue *kq; 1329 int error; 1330 1331 kq = (struct kqueue *)fp->f_data; 1332 tok = lwkt_token_pool_lookup(kq); 1333 lwkt_gettoken(tok); 1334 1335 switch(com) { 1336 case FIOASYNC: 1337 if (*(int *)data) 1338 kq->kq_state |= KQ_ASYNC; 1339 else 1340 kq->kq_state &= ~KQ_ASYNC; 1341 error = 0; 1342 break; 1343 case FIOSETOWN: 1344 error = fsetown(*(int *)data, &kq->kq_sigio); 1345 break; 1346 default: 1347 error = ENOTTY; 1348 break; 1349 } 1350 lwkt_reltoken(tok); 1351 return (error); 1352 } 1353 1354 /* 1355 * MPSAFE 1356 */ 1357 static int 1358 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred) 1359 { 1360 struct kqueue *kq = (struct kqueue *)fp->f_data; 1361 1362 bzero((void *)st, sizeof(*st)); 1363 st->st_size = kq->kq_count; 1364 st->st_blksize = sizeof(struct kevent); 1365 st->st_mode = S_IFIFO; 1366 return (0); 1367 } 1368 1369 /* 1370 * MPSAFE 1371 */ 1372 static int 1373 kqueue_close(struct file *fp) 1374 { 1375 struct kqueue *kq = (struct kqueue *)fp->f_data; 1376 1377 kqueue_terminate(kq); 1378 1379 fp->f_data = NULL; 1380 funsetown(&kq->kq_sigio); 1381 1382 kfree(kq, M_KQUEUE); 1383 return (0); 1384 } 1385 1386 static void 1387 kqueue_wakeup(struct kqueue *kq) 1388 { 1389 if (kq->kq_state & KQ_SLEEP) { 1390 kq->kq_state &= ~KQ_SLEEP; 1391 wakeup(kq); 1392 } 1393 KNOTE(&kq->kq_kqinfo.ki_note, 0); 1394 } 1395 1396 /* 1397 * Calls filterops f_attach function, acquiring mplock if filter is not 1398 * marked as FILTEROP_MPSAFE. 1399 * 1400 * Caller must be holding the related kq token 1401 */ 1402 static int 1403 filter_attach(struct knote *kn) 1404 { 1405 int ret; 1406 1407 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1408 ret = kn->kn_fop->f_attach(kn); 1409 } else { 1410 get_mplock(); 1411 ret = kn->kn_fop->f_attach(kn); 1412 rel_mplock(); 1413 } 1414 return (ret); 1415 } 1416 1417 /* 1418 * Detach the knote and drop it, destroying the knote. 1419 * 1420 * Calls filterops f_detach function, acquiring mplock if filter is not 1421 * marked as FILTEROP_MPSAFE. 1422 * 1423 * Caller must be holding the related kq token 1424 */ 1425 static void 1426 knote_detach_and_drop(struct knote *kn) 1427 { 1428 kn->kn_status |= KN_DELETING | KN_REPROCESS; 1429 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1430 kn->kn_fop->f_detach(kn); 1431 } else { 1432 get_mplock(); 1433 kn->kn_fop->f_detach(kn); 1434 rel_mplock(); 1435 } 1436 knote_drop(kn); 1437 } 1438 1439 /* 1440 * Calls filterops f_event function, acquiring mplock if filter is not 1441 * marked as FILTEROP_MPSAFE. 1442 * 1443 * If the knote is in the middle of being created or deleted we cannot 1444 * safely call the filter op. 1445 * 1446 * Caller must be holding the related kq token 1447 */ 1448 static int 1449 filter_event(struct knote *kn, long hint) 1450 { 1451 int ret; 1452 1453 if (kn->kn_fop->f_flags & FILTEROP_MPSAFE) { 1454 ret = kn->kn_fop->f_event(kn, hint); 1455 } else { 1456 get_mplock(); 1457 ret = kn->kn_fop->f_event(kn, hint); 1458 rel_mplock(); 1459 } 1460 return (ret); 1461 } 1462 1463 /* 1464 * Walk down a list of knotes, activating them if their event has triggered. 1465 * 1466 * If we encounter any knotes which are undergoing processing we just mark 1467 * them for reprocessing and do not try to [re]activate the knote. However, 1468 * if a hint is being passed we have to wait and that makes things a bit 1469 * sticky. 1470 */ 1471 void 1472 knote(struct klist *list, long hint) 1473 { 1474 struct kqueue *kq; 1475 struct knote *kn; 1476 struct knote *kntmp; 1477 1478 lwkt_getpooltoken(list); 1479 restart: 1480 SLIST_FOREACH(kn, list, kn_next) { 1481 kq = kn->kn_kq; 1482 lwkt_getpooltoken(kq); 1483 1484 /* temporary verification hack */ 1485 SLIST_FOREACH(kntmp, list, kn_next) { 1486 if (kn == kntmp) 1487 break; 1488 } 1489 if (kn != kntmp || kn->kn_kq != kq) { 1490 lwkt_relpooltoken(kq); 1491 goto restart; 1492 } 1493 1494 if (kn->kn_status & KN_PROCESSING) { 1495 /* 1496 * Someone else is processing the knote, ask the 1497 * other thread to reprocess it and don't mess 1498 * with it otherwise. 1499 */ 1500 if (hint == 0) { 1501 kn->kn_status |= KN_REPROCESS; 1502 lwkt_relpooltoken(kq); 1503 continue; 1504 } 1505 1506 /* 1507 * If the hint is non-zero we have to wait or risk 1508 * losing the state the caller is trying to update. 1509 * 1510 * XXX This is a real problem, certain process 1511 * and signal filters will bump kn_data for 1512 * already-processed notes more than once if 1513 * we restart the list scan. FIXME. 1514 */ 1515 kn->kn_status |= KN_WAITING | KN_REPROCESS; 1516 tsleep(kn, 0, "knotec", hz); 1517 lwkt_relpooltoken(kq); 1518 goto restart; 1519 } 1520 1521 /* 1522 * Become the reprocessing master ourselves. 1523 * 1524 * If hint is non-zero running the event is mandatory 1525 * when not deleting so do it whether reprocessing is 1526 * set or not. 1527 */ 1528 kn->kn_status |= KN_PROCESSING; 1529 if ((kn->kn_status & KN_DELETING) == 0) { 1530 if (filter_event(kn, hint)) 1531 KNOTE_ACTIVATE(kn); 1532 } 1533 if (knote_release(kn)) { 1534 lwkt_relpooltoken(kq); 1535 goto restart; 1536 } 1537 lwkt_relpooltoken(kq); 1538 } 1539 lwkt_relpooltoken(list); 1540 } 1541 1542 /* 1543 * Insert knote at head of klist. 1544 * 1545 * This function may only be called via a filter function and thus 1546 * kq_token should already be held and marked for processing. 1547 */ 1548 void 1549 knote_insert(struct klist *klist, struct knote *kn) 1550 { 1551 lwkt_getpooltoken(klist); 1552 KKASSERT(kn->kn_status & KN_PROCESSING); 1553 SLIST_INSERT_HEAD(klist, kn, kn_next); 1554 lwkt_relpooltoken(klist); 1555 } 1556 1557 /* 1558 * Remove knote from a klist 1559 * 1560 * This function may only be called via a filter function and thus 1561 * kq_token should already be held and marked for processing. 1562 */ 1563 void 1564 knote_remove(struct klist *klist, struct knote *kn) 1565 { 1566 lwkt_getpooltoken(klist); 1567 KKASSERT(kn->kn_status & KN_PROCESSING); 1568 SLIST_REMOVE(klist, kn, knote, kn_next); 1569 lwkt_relpooltoken(klist); 1570 } 1571 1572 #if 0 1573 /* 1574 * Remove all knotes from a specified klist 1575 * 1576 * Only called from aio. 1577 */ 1578 void 1579 knote_empty(struct klist *list) 1580 { 1581 struct knote *kn; 1582 1583 lwkt_gettoken(&kq_token); 1584 while ((kn = SLIST_FIRST(list)) != NULL) { 1585 if (knote_acquire(kn)) 1586 knote_detach_and_drop(kn); 1587 } 1588 lwkt_reltoken(&kq_token); 1589 } 1590 #endif 1591 1592 void 1593 knote_assume_knotes(struct kqinfo *src, struct kqinfo *dst, 1594 struct filterops *ops, void *hook) 1595 { 1596 struct kqueue *kq; 1597 struct knote *kn; 1598 1599 lwkt_getpooltoken(&src->ki_note); 1600 lwkt_getpooltoken(&dst->ki_note); 1601 while ((kn = SLIST_FIRST(&src->ki_note)) != NULL) { 1602 kq = kn->kn_kq; 1603 lwkt_getpooltoken(kq); 1604 if (SLIST_FIRST(&src->ki_note) != kn || kn->kn_kq != kq) { 1605 lwkt_relpooltoken(kq); 1606 continue; 1607 } 1608 if (knote_acquire(kn)) { 1609 knote_remove(&src->ki_note, kn); 1610 kn->kn_fop = ops; 1611 kn->kn_hook = hook; 1612 knote_insert(&dst->ki_note, kn); 1613 knote_release(kn); 1614 /* kn may be invalid now */ 1615 } 1616 lwkt_relpooltoken(kq); 1617 } 1618 lwkt_relpooltoken(&dst->ki_note); 1619 lwkt_relpooltoken(&src->ki_note); 1620 } 1621 1622 /* 1623 * Remove all knotes referencing a specified fd 1624 */ 1625 void 1626 knote_fdclose(struct file *fp, struct filedesc *fdp, int fd) 1627 { 1628 struct kqueue *kq; 1629 struct knote *kn; 1630 struct knote *kntmp; 1631 1632 lwkt_getpooltoken(&fp->f_klist); 1633 restart: 1634 SLIST_FOREACH(kn, &fp->f_klist, kn_link) { 1635 if (kn->kn_kq->kq_fdp == fdp && kn->kn_id == fd) { 1636 kq = kn->kn_kq; 1637 lwkt_getpooltoken(kq); 1638 1639 /* temporary verification hack */ 1640 SLIST_FOREACH(kntmp, &fp->f_klist, kn_link) { 1641 if (kn == kntmp) 1642 break; 1643 } 1644 if (kn != kntmp || kn->kn_kq->kq_fdp != fdp || 1645 kn->kn_id != fd || kn->kn_kq != kq) { 1646 lwkt_relpooltoken(kq); 1647 goto restart; 1648 } 1649 if (knote_acquire(kn)) 1650 knote_detach_and_drop(kn); 1651 lwkt_relpooltoken(kq); 1652 goto restart; 1653 } 1654 } 1655 lwkt_relpooltoken(&fp->f_klist); 1656 } 1657 1658 /* 1659 * Low level attach function. 1660 * 1661 * The knote should already be marked for processing. 1662 * Caller must hold the related kq token. 1663 */ 1664 static void 1665 knote_attach(struct knote *kn) 1666 { 1667 struct klist *list; 1668 struct kqueue *kq = kn->kn_kq; 1669 1670 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1671 KKASSERT(kn->kn_fp); 1672 list = &kn->kn_fp->f_klist; 1673 } else { 1674 if (kq->kq_knhashmask == 0) 1675 kq->kq_knhash = hashinit(KN_HASHSIZE, M_KQUEUE, 1676 &kq->kq_knhashmask); 1677 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1678 } 1679 lwkt_getpooltoken(list); 1680 SLIST_INSERT_HEAD(list, kn, kn_link); 1681 TAILQ_INSERT_HEAD(&kq->kq_knlist, kn, kn_kqlink); 1682 lwkt_relpooltoken(list); 1683 } 1684 1685 /* 1686 * Low level drop function. 1687 * 1688 * The knote should already be marked for processing. 1689 * Caller must hold the related kq token. 1690 */ 1691 static void 1692 knote_drop(struct knote *kn) 1693 { 1694 struct kqueue *kq; 1695 struct klist *list; 1696 1697 kq = kn->kn_kq; 1698 1699 if (kn->kn_fop->f_flags & FILTEROP_ISFD) 1700 list = &kn->kn_fp->f_klist; 1701 else 1702 list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)]; 1703 1704 lwkt_getpooltoken(list); 1705 SLIST_REMOVE(list, kn, knote, kn_link); 1706 TAILQ_REMOVE(&kq->kq_knlist, kn, kn_kqlink); 1707 if (kn->kn_status & KN_QUEUED) 1708 knote_dequeue(kn); 1709 if (kn->kn_fop->f_flags & FILTEROP_ISFD) { 1710 fdrop(kn->kn_fp); 1711 kn->kn_fp = NULL; 1712 } 1713 knote_free(kn); 1714 lwkt_relpooltoken(list); 1715 } 1716 1717 /* 1718 * Low level enqueue function. 1719 * 1720 * The knote should already be marked for processing. 1721 * Caller must be holding the kq token 1722 */ 1723 static void 1724 knote_enqueue(struct knote *kn) 1725 { 1726 struct kqueue *kq = kn->kn_kq; 1727 1728 KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued")); 1729 TAILQ_INSERT_TAIL(&kq->kq_knpend, kn, kn_tqe); 1730 kn->kn_status |= KN_QUEUED; 1731 ++kq->kq_count; 1732 1733 /* 1734 * Send SIGIO on request (typically set up as a mailbox signal) 1735 */ 1736 if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1) 1737 pgsigio(kq->kq_sigio, SIGIO, 0); 1738 1739 kqueue_wakeup(kq); 1740 } 1741 1742 /* 1743 * Low level dequeue function. 1744 * 1745 * The knote should already be marked for processing. 1746 * Caller must be holding the kq token 1747 */ 1748 static void 1749 knote_dequeue(struct knote *kn) 1750 { 1751 struct kqueue *kq = kn->kn_kq; 1752 1753 KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued")); 1754 TAILQ_REMOVE(&kq->kq_knpend, kn, kn_tqe); 1755 kn->kn_status &= ~KN_QUEUED; 1756 kq->kq_count--; 1757 } 1758 1759 static struct knote * 1760 knote_alloc(void) 1761 { 1762 return kmalloc(sizeof(struct knote), M_KQUEUE, M_WAITOK); 1763 } 1764 1765 static void 1766 knote_free(struct knote *kn) 1767 { 1768 kfree(kn, M_KQUEUE); 1769 } 1770