sys/kern/kern_event.c

   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  * $DragonFly: src/sys/kern/kern_event.c,v 1.33 2007/02/03 17:05:57 corecode Exp $
  28  */
  29
  30 #include <sys/param.h>
  31 #include <sys/systm.h>
  32 #include <sys/kernel.h>
  33 #include <sys/proc.h>
  34 #include <sys/malloc.h>
  35 #include <sys/unistd.h>
  36 #include <sys/file.h>
  37 #include <sys/lock.h>
  38 #include <sys/fcntl.h>
  39 #include <sys/select.h>
  40 #include <sys/queue.h>
  41 #include <sys/event.h>
  42 #include <sys/eventvar.h>
  43 #include <sys/poll.h>
  44 #include <sys/protosw.h>
  45 #include <sys/socket.h>
  46 #include <sys/socketvar.h>
  47 #include <sys/stat.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/sysproto.h>
  50 #include <sys/uio.h>
  51 #include <sys/thread2.h>
  52 #include <sys/signalvar.h>
  53 #include <sys/filio.h>
  54 #include <sys/file2.h>
  55
  56 #include <vm/vm_zone.h>
  57
  58 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
  59
  60 static int      kqueue_scan(struct file *fp, int maxevents,
  61                     struct kevent *ulistp, const struct timespec *timeout,
  62                     struct thread *td, int *res);
  63 static int      kqueue_read(struct file *fp, struct uio *uio,
  64                     struct ucred *cred, int flags);
  65 static int      kqueue_write(struct file *fp, struct uio *uio,
  66                     struct ucred *cred, int flags);
  67 static int      kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
  68                     struct ucred *cred, struct sysmsg *msg);
  69 static int      kqueue_poll(struct file *fp, int events, struct ucred *cred);
  70 static int      kqueue_kqfilter(struct file *fp, struct knote *kn);
  71 static int      kqueue_stat(struct file *fp, struct stat *st,
  72                     struct ucred *cred);
  73 static int      kqueue_close(struct file *fp);
  74 static void     kqueue_wakeup(struct kqueue *kq);
  75
  76 /*
  77  * MPSAFE
  78  */
  79 static struct fileops kqueueops = {
  80         .fo_read = kqueue_read,
  81         .fo_write = kqueue_write,
  82         .fo_ioctl = kqueue_ioctl,
  83         .fo_poll = kqueue_poll,
  84         .fo_kqfilter = kqueue_kqfilter,
  85         .fo_stat = kqueue_stat,
  86         .fo_close = kqueue_close,
  87         .fo_shutdown = nofo_shutdown
  88 };
  89
  90 static void     knote_attach(struct knote *kn, struct filedesc *fdp);
  91 static void     knote_drop(struct knote *kn, struct thread *td);
  92 static void     knote_enqueue(struct knote *kn);
  93 static void     knote_dequeue(struct knote *kn);
  94 static void     knote_init(void);
  95 static struct   knote *knote_alloc(void);
  96 static void     knote_free(struct knote *kn);
  97
  98 static void     filt_kqdetach(struct knote *kn);
  99 static int      filt_kqueue(struct knote *kn, long hint);
 100 static int      filt_procattach(struct knote *kn);
 101 static void     filt_procdetach(struct knote *kn);
 102 static int      filt_proc(struct knote *kn, long hint);
 103 static int      filt_fileattach(struct knote *kn);
 104 static void     filt_timerexpire(void *knx);
 105 static int      filt_timerattach(struct knote *kn);
 106 static void     filt_timerdetach(struct knote *kn);
 107 static int      filt_timer(struct knote *kn, long hint);
 108
 109 static struct filterops file_filtops =
 110         { 1, filt_fileattach, NULL, NULL };
 111 static struct filterops kqread_filtops =
 112         { 1, NULL, filt_kqdetach, filt_kqueue };
 113 static struct filterops proc_filtops =
 114         { 0, filt_procattach, filt_procdetach, filt_proc };
 115 static struct filterops timer_filtops =
 116         { 0, filt_timerattach, filt_timerdetach, filt_timer };
 117
 118 static vm_zone_t        knote_zone;
 119 static int              kq_ncallouts = 0;
 120 static int              kq_calloutmax = (4 * 1024);
 121 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
 122     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
 123
 124 #define KNOTE_ACTIVATE(kn) do {                                         \
 125         kn->kn_status |= KN_ACTIVE;                                     \
 126         if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)           \
 127                 knote_enqueue(kn);                                      \
 128 } while(0)
 129
 130 #define KN_HASHSIZE             64              /* XXX should be tunable */
 131 #define KN_HASH(val, mask)      (((val) ^ (val >> 8)) & (mask))
 132
 133 extern struct filterops aio_filtops;
 134 extern struct filterops sig_filtops;
 135
 136 /*
 137  * Table for for all system-defined filters.
 138  */
 139 static struct filterops *sysfilt_ops[] = {
 140         &file_filtops,                  /* EVFILT_READ */
 141         &file_filtops,                  /* EVFILT_WRITE */
 142         &aio_filtops,                   /* EVFILT_AIO */
 143         &file_filtops,                  /* EVFILT_VNODE */
 144         &proc_filtops,                  /* EVFILT_PROC */
 145         &sig_filtops,                   /* EVFILT_SIGNAL */
 146         &timer_filtops,                 /* EVFILT_TIMER */
 147 };
 148
 149 static int
 150 filt_fileattach(struct knote *kn)
 151 {
 152         return (fo_kqfilter(kn->kn_fp, kn));
 153 }
 154
 155 /*
 156  * MPALMOSTSAFE - acquires mplock
 157  */
 158 static int
 159 kqueue_kqfilter(struct file *fp, struct knote *kn)
 160 {
 161         struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
 162
 163         get_mplock();
 164         if (kn->kn_filter != EVFILT_READ) {
 165                 rel_mplock();
 166                 return (1);
 167         }
 168
 169         kn->kn_fop = &kqread_filtops;
 170         SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext);
 171         rel_mplock();
 172         return (0);
 173 }
 174
 175 static void
 176 filt_kqdetach(struct knote *kn)
 177 {
 178         struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
 179
 180         SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext);
 181 }
 182
 183 /*ARGSUSED*/
 184 static int
 185 filt_kqueue(struct knote *kn, long hint)
 186 {
 187         struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
 188
 189         kn->kn_data = kq->kq_count;
 190         return (kn->kn_data > 0);
 191 }
 192
 193 static int
 194 filt_procattach(struct knote *kn)
 195 {
 196         struct proc *p;
 197         int immediate;
 198
 199         immediate = 0;
 200         p = pfind(kn->kn_id);
 201         if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
 202                 p = zpfind(kn->kn_id);
 203                 immediate = 1;
 204         }
 205         if (p == NULL)
 206                 return (ESRCH);
 207         if (! PRISON_CHECK(curproc->p_ucred, p->p_ucred))
 208                 return (EACCES);
 209
 210         kn->kn_ptr.p_proc = p;
 211         kn->kn_flags |= EV_CLEAR;               /* automatically set */
 212
 213         /*
 214          * internal flag indicating registration done by kernel
 215          */
 216         if (kn->kn_flags & EV_FLAG1) {
 217                 kn->kn_data = kn->kn_sdata;             /* ppid */
 218                 kn->kn_fflags = NOTE_CHILD;
 219                 kn->kn_flags &= ~EV_FLAG1;
 220         }
 221
 222         /* XXX lock the proc here while adding to the list? */
 223         SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
 224
 225         /*
 226          * Immediately activate any exit notes if the target process is a
 227          * zombie.  This is necessary to handle the case where the target
 228          * process, e.g. a child, dies before the kevent is registered.
 229          */
 230         if (immediate && filt_proc(kn, NOTE_EXIT))
 231                 KNOTE_ACTIVATE(kn);
 232
 233         return (0);
 234 }
 235
 236 /*
 237  * The knote may be attached to a different process, which may exit,
 238  * leaving nothing for the knote to be attached to.  So when the process
 239  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
 240  * it will be deleted when read out.  However, as part of the knote deletion,
 241  * this routine is called, so a check is needed to avoid actually performing
 242  * a detach, because the original process does not exist any more.
 243  */
 244 static void
 245 filt_procdetach(struct knote *kn)
 246 {
 247         struct proc *p;
 248
 249         if (kn->kn_status & KN_DETACHED)
 250                 return;
 251         /* XXX locking?  this might modify another process. */
 252         p = kn->kn_ptr.p_proc;
 253         SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
 254 }
 255
 256 static int
 257 filt_proc(struct knote *kn, long hint)
 258 {
 259         u_int event;
 260
 261         /*
 262          * mask off extra data
 263          */
 264         event = (u_int)hint & NOTE_PCTRLMASK;
 265
 266         /*
 267          * if the user is interested in this event, record it.
 268          */
 269         if (kn->kn_sfflags & event)
 270                 kn->kn_fflags |= event;
 271
 272         /*
 273          * Process is gone, so flag the event as finished.  Detach the
 274          * knote from the process now because the process will be poof,
 275          * gone later on.
 276          */
 277         if (event == NOTE_EXIT) {
 278                 struct proc *p = kn->kn_ptr.p_proc;
 279                 if ((kn->kn_status & KN_DETACHED) == 0) {
 280                         SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
 281                         kn->kn_status |= KN_DETACHED;
 282                         kn->kn_data = p->p_xstat;
 283                         kn->kn_ptr.p_proc = NULL;
 284                 }
 285                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 286                 return (1);
 287         }
 288
 289         /*
 290          * process forked, and user wants to track the new process,
 291          * so attach a new knote to it, and immediately report an
 292          * event with the parent's pid.
 293          */
 294         if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
 295                 struct kevent kev;
 296                 int error;
 297
 298                 /*
 299                  * register knote with new process.
 300                  */
 301                 kev.ident = hint & NOTE_PDATAMASK;      /* pid */
 302                 kev.filter = kn->kn_filter;
 303                 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
 304                 kev.fflags = kn->kn_sfflags;
 305                 kev.data = kn->kn_id;                   /* parent */
 306                 kev.udata = kn->kn_kevent.udata;        /* preserve udata */
 307                 error = kqueue_register(kn->kn_kq, &kev, NULL);
 308                 if (error)
 309                         kn->kn_fflags |= NOTE_TRACKERR;
 310         }
 311
 312         return (kn->kn_fflags != 0);
 313 }
 314
 315 static void
 316 filt_timerexpire(void *knx)
 317 {
 318         struct knote *kn = knx;
 319         struct callout *calloutp;
 320         struct timeval tv;
 321         int tticks;
 322
 323         kn->kn_data++;
 324         KNOTE_ACTIVATE(kn);
 325
 326         if ((kn->kn_flags & EV_ONESHOT) == 0) {
 327                 tv.tv_sec = kn->kn_sdata / 1000;
 328                 tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
 329                 tticks = tvtohz_high(&tv);
 330                 calloutp = (struct callout *)kn->kn_hook;
 331                 callout_reset(calloutp, tticks, filt_timerexpire, kn);
 332         }
 333 }
 334
 335 /*
 336  * data contains amount of time to sleep, in milliseconds
 337  */
 338 static int
 339 filt_timerattach(struct knote *kn)
 340 {
 341         struct callout *calloutp;
 342         struct timeval tv;
 343         int tticks;
 344
 345         if (kq_ncallouts >= kq_calloutmax)
 346                 return (ENOMEM);
 347         kq_ncallouts++;
 348
 349         tv.tv_sec = kn->kn_sdata / 1000;
 350         tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
 351         tticks = tvtohz_high(&tv);
 352
 353         kn->kn_flags |= EV_CLEAR;               /* automatically set */
 354         MALLOC(calloutp, struct callout *, sizeof(*calloutp),
 355             M_KQUEUE, M_WAITOK);
 356         callout_init(calloutp);
 357         kn->kn_hook = (caddr_t)calloutp;
 358         callout_reset(calloutp, tticks, filt_timerexpire, kn);
 359
 360         return (0);
 361 }
 362
 363 static void
 364 filt_timerdetach(struct knote *kn)
 365 {
 366         struct callout *calloutp;
 367
 368         calloutp = (struct callout *)kn->kn_hook;
 369         callout_stop(calloutp);
 370         FREE(calloutp, M_KQUEUE);
 371         kq_ncallouts--;
 372 }
 373
 374 static int
 375 filt_timer(struct knote *kn, long hint)
 376 {
 377
 378         return (kn->kn_data != 0);
 379 }
 380
 381 int
 382 sys_kqueue(struct kqueue_args *uap)
 383 {
 384         struct proc *p = curproc;
 385         struct filedesc *fdp = p->p_fd;
 386         struct kqueue *kq;
 387         struct file *fp;
 388         int fd, error;
 389
 390         error = falloc(p, &fp, &fd);
 391         if (error)
 392                 return (error);
 393         fp->f_flag = FREAD | FWRITE;
 394         fp->f_type = DTYPE_KQUEUE;
 395         fp->f_ops = &kqueueops;
 396
 397         kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO);
 398         TAILQ_INIT(&kq->kq_head);
 399         kq->kq_fdp = fdp;
 400         fp->f_data = kq;
 401
 402         fsetfd(p, fp, fd);
 403         uap->sysmsg_result = fd;
 404         fdrop(fp);
 405         return (error);
 406 }
 407
 408 int
 409 sys_kevent(struct kevent_args *uap)
 410 {
 411         struct thread *td = curthread;
 412         struct proc *p = td->td_proc;
 413         struct kevent *kevp;
 414         struct kqueue *kq;
 415         struct file *fp = NULL;
 416         struct timespec ts;
 417         int i, n, nerrors, error;
 418
 419         KKASSERT(p);
 420
 421         fp = holdfp(p->p_fd, uap->fd, -1);
 422         if (fp == NULL)
 423                 return (EBADF);
 424         if (fp->f_type != DTYPE_KQUEUE) {
 425                 fdrop(fp);
 426                 return (EBADF);
 427         }
 428
 429         if (uap->timeout != NULL) {
 430                 error = copyin(uap->timeout, &ts, sizeof(ts));
 431                 if (error)
 432                         goto done;
 433                 uap->timeout = &ts;
 434         }
 435
 436         kq = (struct kqueue *)fp->f_data;
 437         nerrors = 0;
 438
 439         while (uap->nchanges > 0) {
 440                 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges;
 441                 error = copyin(uap->changelist, kq->kq_kev,
 442                     n * sizeof(struct kevent));
 443                 if (error)
 444                         goto done;
 445                 for (i = 0; i < n; i++) {
 446                         kevp = &kq->kq_kev[i];
 447                         kevp->flags &= ~EV_SYSFLAGS;
 448                         error = kqueue_register(kq, kevp, td);
 449                         if (error) {
 450                                 if (uap->nevents != 0) {
 451                                         kevp->flags = EV_ERROR;
 452                                         kevp->data = error;
 453                                         (void) copyout((caddr_t)kevp,
 454                                             (caddr_t)uap->eventlist,
 455                                             sizeof(*kevp));
 456                                         uap->eventlist++;
 457                                         uap->nevents--;
 458                                         nerrors++;
 459                                 } else {
 460                                         goto done;
 461                                 }
 462                         }
 463                 }
 464                 uap->nchanges -= n;
 465                 uap->changelist += n;
 466         }
 467         if (nerrors) {
 468                 uap->sysmsg_result = nerrors;
 469                 error = 0;
 470                 goto done;
 471         }
 472
 473         error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, td, &uap->sysmsg_result);
 474 done:
 475         if (fp != NULL)
 476                 fdrop(fp);
 477         return (error);
 478 }
 479
 480 int
 481 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td)
 482 {
 483         struct filedesc *fdp = kq->kq_fdp;
 484         struct filterops *fops;
 485         struct file *fp = NULL;
 486         struct knote *kn = NULL;
 487         int error = 0;
 488
 489         if (kev->filter < 0) {
 490                 if (kev->filter + EVFILT_SYSCOUNT < 0)
 491                         return (EINVAL);
 492                 fops = sysfilt_ops[~kev->filter];       /* to 0-base index */
 493         } else {
 494                 /*
 495                  * XXX
 496                  * filter attach routine is responsible for insuring that
 497                  * the identifier can be attached to it.
 498                  */
 499                 kprintf("unknown filter: %d\n", kev->filter);
 500                 return (EINVAL);
 501         }
 502
 503         if (fops->f_isfd) {
 504                 /* validate descriptor */
 505                 fp = holdfp(fdp, kev->ident, -1);
 506                 if (fp == NULL)
 507                         return (EBADF);
 508
 509                 if (kev->ident < fdp->fd_knlistsize) {
 510                         SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
 511                                 if (kq == kn->kn_kq &&
 512                                     kev->filter == kn->kn_filter)
 513                                         break;
 514                 }
 515         } else {
 516                 if (fdp->fd_knhashmask != 0) {
 517                         struct klist *list;
 518
 519                         list = &fdp->fd_knhash[
 520                             KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
 521                         SLIST_FOREACH(kn, list, kn_link)
 522                                 if (kev->ident == kn->kn_id &&
 523                                     kq == kn->kn_kq &&
 524                                     kev->filter == kn->kn_filter)
 525                                         break;
 526                 }
 527         }
 528
 529         if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
 530                 error = ENOENT;
 531                 goto done;
 532         }
 533
 534         /*
 535          * kn now contains the matching knote, or NULL if no match
 536          */
 537         if (kev->flags & EV_ADD) {
 538
 539                 if (kn == NULL) {
 540                         kn = knote_alloc();
 541                         if (kn == NULL) {
 542                                 error = ENOMEM;
 543                                 goto done;
 544                         }
 545                         kn->kn_fp = fp;
 546                         kn->kn_kq = kq;
 547                         kn->kn_fop = fops;
 548
 549                         /*
 550                          * apply reference count to knote structure, and
 551                          * do not release it at the end of this routine.
 552                          */
 553                         fp = NULL;
 554
 555                         kn->kn_sfflags = kev->fflags;
 556                         kn->kn_sdata = kev->data;
 557                         kev->fflags = 0;
 558                         kev->data = 0;
 559                         kn->kn_kevent = *kev;
 560
 561                         knote_attach(kn, fdp);
 562                         if ((error = fops->f_attach(kn)) != 0) {
 563                                 knote_drop(kn, td);
 564                                 goto done;
 565                         }
 566                 } else {
 567                         /*
 568                          * The user may change some filter values after the
 569                          * initial EV_ADD, but doing so will not reset any
 570                          * filter which have already been triggered.
 571                          */
 572                         kn->kn_sfflags = kev->fflags;
 573                         kn->kn_sdata = kev->data;
 574                         kn->kn_kevent.udata = kev->udata;
 575                 }
 576
 577                 crit_enter();
 578                 if (kn->kn_fop->f_event(kn, 0))
 579                         KNOTE_ACTIVATE(kn);
 580                 crit_exit();
 581         } else if (kev->flags & EV_DELETE) {
 582                 kn->kn_fop->f_detach(kn);
 583                 knote_drop(kn, td);
 584                 goto done;
 585         }
 586
 587         if ((kev->flags & EV_DISABLE) &&
 588             ((kn->kn_status & KN_DISABLED) == 0)) {
 589                 crit_enter();
 590                 kn->kn_status |= KN_DISABLED;
 591                 crit_exit();
 592         }
 593
 594         if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
 595                 crit_enter();
 596                 kn->kn_status &= ~KN_DISABLED;
 597                 if ((kn->kn_status & KN_ACTIVE) &&
 598                     ((kn->kn_status & KN_QUEUED) == 0))
 599                         knote_enqueue(kn);
 600                 crit_exit();
 601         }
 602
 603 done:
 604         if (fp != NULL)
 605                 fdrop(fp);
 606         return (error);
 607 }
 608
 609 static int
 610 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
 611         const struct timespec *tsp, struct thread *td, int *res)
 612 {
 613         struct kqueue *kq = (struct kqueue *)fp->f_data;
 614         struct kevent *kevp;
 615         struct timeval atv, rtv, ttv;
 616         struct knote *kn, marker;
 617         int count, timeout, nkev = 0, error = 0;
 618
 619         count = maxevents;
 620         if (count == 0)
 621                 goto done;
 622
 623         if (tsp != NULL) {
 624                 TIMESPEC_TO_TIMEVAL(&atv, tsp);
 625                 if (itimerfix(&atv)) {
 626                         error = EINVAL;
 627                         goto done;
 628                 }
 629                 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
 630                         timeout = -1;
 631                 else
 632                         timeout = atv.tv_sec > 24 * 60 * 60 ?
 633                             24 * 60 * 60 * hz : tvtohz_high(&atv);
 634                 getmicrouptime(&rtv);
 635                 timevaladd(&atv, &rtv);
 636         } else {
 637                 atv.tv_sec = 0;
 638                 atv.tv_usec = 0;
 639                 timeout = 0;
 640         }
 641         goto start;
 642
 643 retry:
 644         if (atv.tv_sec || atv.tv_usec) {
 645                 getmicrouptime(&rtv);
 646                 if (timevalcmp(&rtv, &atv, >=))
 647                         goto done;
 648                 ttv = atv;
 649                 timevalsub(&ttv, &rtv);
 650                 timeout = ttv.tv_sec > 24 * 60 * 60 ?
 651                         24 * 60 * 60 * hz : tvtohz_high(&ttv);
 652         }
 653
 654 start:
 655         kevp = kq->kq_kev;
 656         crit_enter();
 657         if (kq->kq_count == 0) {
 658                 if (timeout < 0) {
 659                         error = EWOULDBLOCK;
 660                 } else {
 661                         kq->kq_state |= KQ_SLEEP;
 662                         error = tsleep(kq, PCATCH, "kqread", timeout);
 663                 }
 664                 crit_exit();
 665                 if (error == 0)
 666                         goto retry;
 667                 /* don't restart after signals... */
 668                 if (error == ERESTART)
 669                         error = EINTR;
 670                 else if (error == EWOULDBLOCK)
 671                         error = 0;
 672                 goto done;
 673         }
 674
 675         TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
 676         while (count) {
 677                 kn = TAILQ_FIRST(&kq->kq_head);
 678                 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
 679                 if (kn == &marker) {
 680                         crit_exit();
 681                         if (count == maxevents)
 682                                 goto retry;
 683                         goto done;
 684                 }
 685                 if (kn->kn_status & KN_DISABLED) {
 686                         kn->kn_status &= ~KN_QUEUED;
 687                         kq->kq_count--;
 688                         continue;
 689                 }
 690                 if ((kn->kn_flags & EV_ONESHOT) == 0 &&
 691                     kn->kn_fop->f_event(kn, 0) == 0) {
 692                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 693                         kq->kq_count--;
 694                         continue;
 695                 }
 696                 *kevp = kn->kn_kevent;
 697                 kevp++;
 698                 nkev++;
 699                 if (kn->kn_flags & EV_ONESHOT) {
 700                         kn->kn_status &= ~KN_QUEUED;
 701                         kq->kq_count--;
 702                         crit_exit();
 703                         kn->kn_fop->f_detach(kn);
 704                         knote_drop(kn, td);
 705                         crit_enter();
 706                 } else if (kn->kn_flags & EV_CLEAR) {
 707                         kn->kn_data = 0;
 708                         kn->kn_fflags = 0;
 709                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 710                         kq->kq_count--;
 711                 } else {
 712                         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
 713                 }
 714                 count--;
 715                 if (nkev == KQ_NEVENTS) {
 716                         crit_exit();
 717                         error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 718                             sizeof(struct kevent) * nkev);
 719                         ulistp += nkev;
 720                         nkev = 0;
 721                         kevp = kq->kq_kev;
 722                         crit_enter();
 723                         if (error)
 724                                 break;
 725                 }
 726         }
 727         TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
 728         crit_exit();
 729 done:
 730         if (nkev != 0)
 731                 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 732                     sizeof(struct kevent) * nkev);
 733         *res = maxevents - count;
 734         return (error);
 735 }
 736
 737 /*
 738  * XXX
 739  * This could be expanded to call kqueue_scan, if desired.
 740  *
 741  * MPSAFE
 742  */
 743 static int
 744 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 745 {
 746         return (ENXIO);
 747 }
 748
 749 /*
 750  * MPSAFE
 751  */
 752 static int
 753 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 754 {
 755         return (ENXIO);
 756 }
 757
 758 /*
 759  * MPSAFE
 760  */
 761 static int
 762 kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
 763              struct ucred *cred, struct sysmsg *msg)
 764 {
 765         struct kqueue *kq;
 766         int error;
 767
 768         get_mplock();
 769         kq = (struct kqueue *)fp->f_data;
 770
 771         switch(com) {
 772         case FIOASYNC:
 773                 if (*(int *)data)
 774                         kq->kq_state |= KQ_ASYNC;
 775                 else
 776                         kq->kq_state &= ~KQ_ASYNC;
 777                 error = 0;
 778                 break;
 779         case FIOSETOWN:
 780                 error = fsetown(*(int *)data, &kq->kq_sigio);
 781                 break;
 782         default:
 783                 error = ENOTTY;
 784                 break;
 785         }
 786         rel_mplock();
 787         return (error);
 788 }
 789
 790 /*
 791  * MPALMOSTSAFE - acquires mplock
 792  */
 793 static int
 794 kqueue_poll(struct file *fp, int events, struct ucred *cred)
 795 {
 796         struct kqueue *kq = (struct kqueue *)fp->f_data;
 797         int revents = 0;
 798
 799         get_mplock();
 800         crit_enter();
 801         if (events & (POLLIN | POLLRDNORM)) {
 802                 if (kq->kq_count) {
 803                         revents |= events & (POLLIN | POLLRDNORM);
 804                 } else {
 805                         selrecord(curthread, &kq->kq_sel);
 806                         kq->kq_state |= KQ_SEL;
 807                 }
 808         }
 809         crit_exit();
 810         rel_mplock();
 811         return (revents);
 812 }
 813
 814 /*
 815  * MPSAFE
 816  */
 817 static int
 818 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
 819 {
 820         struct kqueue *kq = (struct kqueue *)fp->f_data;
 821
 822         bzero((void *)st, sizeof(*st));
 823         st->st_size = kq->kq_count;
 824         st->st_blksize = sizeof(struct kevent);
 825         st->st_mode = S_IFIFO;
 826         return (0);
 827 }
 828
 829 /*
 830  * MPALMOSTSAFE - acquires mplock
 831  */
 832 static int
 833 kqueue_close(struct file *fp)
 834 {
 835         struct thread *td = curthread;
 836         struct proc *p = td->td_proc;
 837         struct kqueue *kq = (struct kqueue *)fp->f_data;
 838         struct filedesc *fdp;
 839         struct knote **knp, *kn, *kn0;
 840         int i;
 841
 842         KKASSERT(p);
 843         get_mplock();
 844         fdp = p->p_fd;
 845         for (i = 0; i < fdp->fd_knlistsize; i++) {
 846                 knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
 847                 kn = *knp;
 848                 while (kn != NULL) {
 849                         kn0 = SLIST_NEXT(kn, kn_link);
 850                         if (kq == kn->kn_kq) {
 851                                 kn->kn_fop->f_detach(kn);
 852                                 fdrop(kn->kn_fp);
 853                                 knote_free(kn);
 854                                 *knp = kn0;
 855                         } else {
 856                                 knp = &SLIST_NEXT(kn, kn_link);
 857                         }
 858                         kn = kn0;
 859                 }
 860         }
 861         if (fdp->fd_knhashmask != 0) {
 862                 for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
 863                         knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
 864                         kn = *knp;
 865                         while (kn != NULL) {
 866                                 kn0 = SLIST_NEXT(kn, kn_link);
 867                                 if (kq == kn->kn_kq) {
 868                                         kn->kn_fop->f_detach(kn);
 869                 /* XXX non-fd release of kn->kn_ptr */
 870                                         knote_free(kn);
 871                                         *knp = kn0;
 872                                 } else {
 873                                         knp = &SLIST_NEXT(kn, kn_link);
 874                                 }
 875                                 kn = kn0;
 876                         }
 877                 }
 878         }
 879         fp->f_data = NULL;
 880         funsetown(kq->kq_sigio);
 881         rel_mplock();
 882
 883         kfree(kq, M_KQUEUE);
 884         return (0);
 885 }
 886
 887 static void
 888 kqueue_wakeup(struct kqueue *kq)
 889 {
 890         if (kq->kq_state & KQ_SLEEP) {
 891                 kq->kq_state &= ~KQ_SLEEP;
 892                 wakeup(kq);
 893         }
 894         if (kq->kq_state & KQ_SEL) {
 895                 kq->kq_state &= ~KQ_SEL;
 896                 selwakeup(&kq->kq_sel);
 897         }
 898         KNOTE(&kq->kq_sel.si_note, 0);
 899 }
 900
 901 /*
 902  * walk down a list of knotes, activating them if their event has triggered.
 903  */
 904 void
 905 knote(struct klist *list, long hint)
 906 {
 907         struct knote *kn;
 908
 909         SLIST_FOREACH(kn, list, kn_selnext)
 910                 if (kn->kn_fop->f_event(kn, hint))
 911                         KNOTE_ACTIVATE(kn);
 912 }
 913
 914 /*
 915  * remove all knotes from a specified klist
 916  */
 917 void
 918 knote_remove(struct thread *td, struct klist *list)
 919 {
 920         struct knote *kn;
 921
 922         while ((kn = SLIST_FIRST(list)) != NULL) {
 923                 kn->kn_fop->f_detach(kn);
 924                 knote_drop(kn, td);
 925         }
 926 }
 927
 928 /*
 929  * remove all knotes referencing a specified fd
 930  */
 931 void
 932 knote_fdclose(struct proc *p, int fd)
 933 {
 934         struct filedesc *fdp = p->p_fd;
 935         struct klist *list = &fdp->fd_knlist[fd];
 936         /* Take any thread of p */
 937         struct thread *td = FIRST_LWP_IN_PROC(p)->lwp_thread;
 938
 939         knote_remove(td, list);
 940 }
 941
 942 static void
 943 knote_attach(struct knote *kn, struct filedesc *fdp)
 944 {
 945         struct klist *list;
 946         int size;
 947
 948         if (! kn->kn_fop->f_isfd) {
 949                 if (fdp->fd_knhashmask == 0)
 950                         fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
 951                             &fdp->fd_knhashmask);
 952                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
 953                 goto done;
 954         }
 955
 956         if (fdp->fd_knlistsize <= kn->kn_id) {
 957                 size = fdp->fd_knlistsize;
 958                 while (size <= kn->kn_id)
 959                         size += KQEXTENT;
 960                 MALLOC(list, struct klist *,
 961                     size * sizeof(struct klist *), M_KQUEUE, M_WAITOK);
 962                 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
 963                     fdp->fd_knlistsize * sizeof(struct klist *));
 964                 bzero((caddr_t)list +
 965                     fdp->fd_knlistsize * sizeof(struct klist *),
 966                     (size - fdp->fd_knlistsize) * sizeof(struct klist *));
 967                 if (fdp->fd_knlist != NULL)
 968                         FREE(fdp->fd_knlist, M_KQUEUE);
 969                 fdp->fd_knlistsize = size;
 970                 fdp->fd_knlist = list;
 971         }
 972         list = &fdp->fd_knlist[kn->kn_id];
 973 done:
 974         SLIST_INSERT_HEAD(list, kn, kn_link);
 975         kn->kn_status = 0;
 976 }
 977
 978 /*
 979  * should be called outside of a critical section, since we don't want to
 980  * hold a critical section while calling fdrop and free.
 981  */
 982 static void
 983 knote_drop(struct knote *kn, struct thread *td)
 984 {
 985         struct filedesc *fdp;
 986         struct klist *list;
 987
 988         KKASSERT(td->td_proc);
 989         fdp = td->td_proc->p_fd;
 990         if (kn->kn_fop->f_isfd)
 991                 list = &fdp->fd_knlist[kn->kn_id];
 992         else
 993                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
 994
 995         SLIST_REMOVE(list, kn, knote, kn_link);
 996         if (kn->kn_status & KN_QUEUED)
 997                 knote_dequeue(kn);
 998         if (kn->kn_fop->f_isfd)
 999                 fdrop(kn->kn_fp);
1000         knote_free(kn);
1001 }
1002
1003
1004 static void
1005 knote_enqueue(struct knote *kn)
1006 {
1007         struct kqueue *kq = kn->kn_kq;
1008
1009         crit_enter();
1010         KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1011
1012         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1013         kn->kn_status |= KN_QUEUED;
1014         ++kq->kq_count;
1015
1016         /*
1017          * Send SIGIO on request (typically set up as a mailbox signal)
1018          */
1019         if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1020                 pgsigio(kq->kq_sigio, SIGIO, 0);
1021         crit_exit();
1022         kqueue_wakeup(kq);
1023 }
1024
1025 static void
1026 knote_dequeue(struct knote *kn)
1027 {
1028         struct kqueue *kq = kn->kn_kq;
1029
1030         KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1031         crit_enter();
1032
1033         TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1034         kn->kn_status &= ~KN_QUEUED;
1035         kq->kq_count--;
1036         crit_exit();
1037 }
1038
1039 static void
1040 knote_init(void)
1041 {
1042         knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
1043 }
1044 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1045
1046 static struct knote *
1047 knote_alloc(void)
1048 {
1049         return ((struct knote *)zalloc(knote_zone));
1050 }
1051
1052 static void
1053 knote_free(struct knote *kn)
1054 {
1055         zfree(knote_zone, kn);
1056 }