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);
  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 = kn->kn_ptr.p_proc;
 248
 249         if (kn->kn_status & KN_DETACHED)
 250                 return;
 251
 252         /* XXX locking?  this might modify another process. */
 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.
 274          */
 275         if (event == NOTE_EXIT) {
 276                 kn->kn_status |= KN_DETACHED;
 277                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 278                 return (1);
 279         }
 280
 281         /*
 282          * process forked, and user wants to track the new process,
 283          * so attach a new knote to it, and immediately report an
 284          * event with the parent's pid.
 285          */
 286         if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
 287                 struct kevent kev;
 288                 int error;
 289
 290                 /*
 291                  * register knote with new process.
 292                  */
 293                 kev.ident = hint & NOTE_PDATAMASK;      /* pid */
 294                 kev.filter = kn->kn_filter;
 295                 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
 296                 kev.fflags = kn->kn_sfflags;
 297                 kev.data = kn->kn_id;                   /* parent */
 298                 kev.udata = kn->kn_kevent.udata;        /* preserve udata */
 299                 error = kqueue_register(kn->kn_kq, &kev, NULL);
 300                 if (error)
 301                         kn->kn_fflags |= NOTE_TRACKERR;
 302         }
 303
 304         return (kn->kn_fflags != 0);
 305 }
 306
 307 static void
 308 filt_timerexpire(void *knx)
 309 {
 310         struct knote *kn = knx;
 311         struct callout *calloutp;
 312         struct timeval tv;
 313         int tticks;
 314
 315         kn->kn_data++;
 316         KNOTE_ACTIVATE(kn);
 317
 318         if ((kn->kn_flags & EV_ONESHOT) == 0) {
 319                 tv.tv_sec = kn->kn_sdata / 1000;
 320                 tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
 321                 tticks = tvtohz_high(&tv);
 322                 calloutp = (struct callout *)kn->kn_hook;
 323                 callout_reset(calloutp, tticks, filt_timerexpire, kn);
 324         }
 325 }
 326
 327 /*
 328  * data contains amount of time to sleep, in milliseconds
 329  */
 330 static int
 331 filt_timerattach(struct knote *kn)
 332 {
 333         struct callout *calloutp;
 334         struct timeval tv;
 335         int tticks;
 336
 337         if (kq_ncallouts >= kq_calloutmax)
 338                 return (ENOMEM);
 339         kq_ncallouts++;
 340
 341         tv.tv_sec = kn->kn_sdata / 1000;
 342         tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
 343         tticks = tvtohz_high(&tv);
 344
 345         kn->kn_flags |= EV_CLEAR;               /* automatically set */
 346         MALLOC(calloutp, struct callout *, sizeof(*calloutp),
 347             M_KQUEUE, M_WAITOK);
 348         callout_init(calloutp);
 349         kn->kn_hook = (caddr_t)calloutp;
 350         callout_reset(calloutp, tticks, filt_timerexpire, kn);
 351
 352         return (0);
 353 }
 354
 355 static void
 356 filt_timerdetach(struct knote *kn)
 357 {
 358         struct callout *calloutp;
 359
 360         calloutp = (struct callout *)kn->kn_hook;
 361         callout_stop(calloutp);
 362         FREE(calloutp, M_KQUEUE);
 363         kq_ncallouts--;
 364 }
 365
 366 static int
 367 filt_timer(struct knote *kn, long hint)
 368 {
 369
 370         return (kn->kn_data != 0);
 371 }
 372
 373 int
 374 sys_kqueue(struct kqueue_args *uap)
 375 {
 376         struct proc *p = curproc;
 377         struct filedesc *fdp = p->p_fd;
 378         struct kqueue *kq;
 379         struct file *fp;
 380         int fd, error;
 381
 382         error = falloc(p, &fp, &fd);
 383         if (error)
 384                 return (error);
 385         fp->f_flag = FREAD | FWRITE;
 386         fp->f_type = DTYPE_KQUEUE;
 387         fp->f_ops = &kqueueops;
 388
 389         kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO);
 390         TAILQ_INIT(&kq->kq_head);
 391         kq->kq_fdp = fdp;
 392         fp->f_data = kq;
 393
 394         fsetfd(p, fp, fd);
 395         uap->sysmsg_result = fd;
 396         fdrop(fp);
 397         return (error);
 398 }
 399
 400 int
 401 sys_kevent(struct kevent_args *uap)
 402 {
 403         struct thread *td = curthread;
 404         struct proc *p = td->td_proc;
 405         struct kevent *kevp;
 406         struct kqueue *kq;
 407         struct file *fp = NULL;
 408         struct timespec ts;
 409         int i, n, nerrors, error;
 410
 411         KKASSERT(p);
 412
 413         fp = holdfp(p->p_fd, uap->fd, -1);
 414         if (fp == NULL)
 415                 return (EBADF);
 416         if (fp->f_type != DTYPE_KQUEUE) {
 417                 fdrop(fp);
 418                 return (EBADF);
 419         }
 420
 421         if (uap->timeout != NULL) {
 422                 error = copyin(uap->timeout, &ts, sizeof(ts));
 423                 if (error)
 424                         goto done;
 425                 uap->timeout = &ts;
 426         }
 427
 428         kq = (struct kqueue *)fp->f_data;
 429         nerrors = 0;
 430
 431         while (uap->nchanges > 0) {
 432                 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges;
 433                 error = copyin(uap->changelist, kq->kq_kev,
 434                     n * sizeof(struct kevent));
 435                 if (error)
 436                         goto done;
 437                 for (i = 0; i < n; i++) {
 438                         kevp = &kq->kq_kev[i];
 439                         kevp->flags &= ~EV_SYSFLAGS;
 440                         error = kqueue_register(kq, kevp, td);
 441                         if (error) {
 442                                 if (uap->nevents != 0) {
 443                                         kevp->flags = EV_ERROR;
 444                                         kevp->data = error;
 445                                         (void) copyout((caddr_t)kevp,
 446                                             (caddr_t)uap->eventlist,
 447                                             sizeof(*kevp));
 448                                         uap->eventlist++;
 449                                         uap->nevents--;
 450                                         nerrors++;
 451                                 } else {
 452                                         goto done;
 453                                 }
 454                         }
 455                 }
 456                 uap->nchanges -= n;
 457                 uap->changelist += n;
 458         }
 459         if (nerrors) {
 460                 uap->sysmsg_result = nerrors;
 461                 error = 0;
 462                 goto done;
 463         }
 464
 465         error = kqueue_scan(fp, uap->nevents, uap->eventlist, uap->timeout, td, &uap->sysmsg_result);
 466 done:
 467         if (fp != NULL)
 468                 fdrop(fp);
 469         return (error);
 470 }
 471
 472 int
 473 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td)
 474 {
 475         struct filedesc *fdp = kq->kq_fdp;
 476         struct filterops *fops;
 477         struct file *fp = NULL;
 478         struct knote *kn = NULL;
 479         int error = 0;
 480
 481         if (kev->filter < 0) {
 482                 if (kev->filter + EVFILT_SYSCOUNT < 0)
 483                         return (EINVAL);
 484                 fops = sysfilt_ops[~kev->filter];       /* to 0-base index */
 485         } else {
 486                 /*
 487                  * XXX
 488                  * filter attach routine is responsible for insuring that
 489                  * the identifier can be attached to it.
 490                  */
 491                 kprintf("unknown filter: %d\n", kev->filter);
 492                 return (EINVAL);
 493         }
 494
 495         if (fops->f_isfd) {
 496                 /* validate descriptor */
 497                 fp = holdfp(fdp, kev->ident, -1);
 498                 if (fp == NULL)
 499                         return (EBADF);
 500
 501                 if (kev->ident < fdp->fd_knlistsize) {
 502                         SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
 503                                 if (kq == kn->kn_kq &&
 504                                     kev->filter == kn->kn_filter)
 505                                         break;
 506                 }
 507         } else {
 508                 if (fdp->fd_knhashmask != 0) {
 509                         struct klist *list;
 510
 511                         list = &fdp->fd_knhash[
 512                             KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
 513                         SLIST_FOREACH(kn, list, kn_link)
 514                                 if (kev->ident == kn->kn_id &&
 515                                     kq == kn->kn_kq &&
 516                                     kev->filter == kn->kn_filter)
 517                                         break;
 518                 }
 519         }
 520
 521         if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
 522                 error = ENOENT;
 523                 goto done;
 524         }
 525
 526         /*
 527          * kn now contains the matching knote, or NULL if no match
 528          */
 529         if (kev->flags & EV_ADD) {
 530
 531                 if (kn == NULL) {
 532                         kn = knote_alloc();
 533                         if (kn == NULL) {
 534                                 error = ENOMEM;
 535                                 goto done;
 536                         }
 537                         kn->kn_fp = fp;
 538                         kn->kn_kq = kq;
 539                         kn->kn_fop = fops;
 540
 541                         /*
 542                          * apply reference count to knote structure, and
 543                          * do not release it at the end of this routine.
 544                          */
 545                         fp = NULL;
 546
 547                         kn->kn_sfflags = kev->fflags;
 548                         kn->kn_sdata = kev->data;
 549                         kev->fflags = 0;
 550                         kev->data = 0;
 551                         kn->kn_kevent = *kev;
 552
 553                         knote_attach(kn, fdp);
 554                         if ((error = fops->f_attach(kn)) != 0) {
 555                                 knote_drop(kn, td);
 556                                 goto done;
 557                         }
 558                 } else {
 559                         /*
 560                          * The user may change some filter values after the
 561                          * initial EV_ADD, but doing so will not reset any
 562                          * filter which have already been triggered.
 563                          */
 564                         kn->kn_sfflags = kev->fflags;
 565                         kn->kn_sdata = kev->data;
 566                         kn->kn_kevent.udata = kev->udata;
 567                 }
 568
 569                 crit_enter();
 570                 if (kn->kn_fop->f_event(kn, 0))
 571                         KNOTE_ACTIVATE(kn);
 572                 crit_exit();
 573         } else if (kev->flags & EV_DELETE) {
 574                 kn->kn_fop->f_detach(kn);
 575                 knote_drop(kn, td);
 576                 goto done;
 577         }
 578
 579         if ((kev->flags & EV_DISABLE) &&
 580             ((kn->kn_status & KN_DISABLED) == 0)) {
 581                 crit_enter();
 582                 kn->kn_status |= KN_DISABLED;
 583                 crit_exit();
 584         }
 585
 586         if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
 587                 crit_enter();
 588                 kn->kn_status &= ~KN_DISABLED;
 589                 if ((kn->kn_status & KN_ACTIVE) &&
 590                     ((kn->kn_status & KN_QUEUED) == 0))
 591                         knote_enqueue(kn);
 592                 crit_exit();
 593         }
 594
 595 done:
 596         if (fp != NULL)
 597                 fdrop(fp);
 598         return (error);
 599 }
 600
 601 static int
 602 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
 603         const struct timespec *tsp, struct thread *td, int *res)
 604 {
 605         struct kqueue *kq = (struct kqueue *)fp->f_data;
 606         struct kevent *kevp;
 607         struct timeval atv, rtv, ttv;
 608         struct knote *kn, marker;
 609         int count, timeout, nkev = 0, error = 0;
 610
 611         count = maxevents;
 612         if (count == 0)
 613                 goto done;
 614
 615         if (tsp != NULL) {
 616                 TIMESPEC_TO_TIMEVAL(&atv, tsp);
 617                 if (itimerfix(&atv)) {
 618                         error = EINVAL;
 619                         goto done;
 620                 }
 621                 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
 622                         timeout = -1;
 623                 else
 624                         timeout = atv.tv_sec > 24 * 60 * 60 ?
 625                             24 * 60 * 60 * hz : tvtohz_high(&atv);
 626                 getmicrouptime(&rtv);
 627                 timevaladd(&atv, &rtv);
 628         } else {
 629                 atv.tv_sec = 0;
 630                 atv.tv_usec = 0;
 631                 timeout = 0;
 632         }
 633         goto start;
 634
 635 retry:
 636         if (atv.tv_sec || atv.tv_usec) {
 637                 getmicrouptime(&rtv);
 638                 if (timevalcmp(&rtv, &atv, >=))
 639                         goto done;
 640                 ttv = atv;
 641                 timevalsub(&ttv, &rtv);
 642                 timeout = ttv.tv_sec > 24 * 60 * 60 ?
 643                         24 * 60 * 60 * hz : tvtohz_high(&ttv);
 644         }
 645
 646 start:
 647         kevp = kq->kq_kev;
 648         crit_enter();
 649         if (kq->kq_count == 0) {
 650                 if (timeout < 0) {
 651                         error = EWOULDBLOCK;
 652                 } else {
 653                         kq->kq_state |= KQ_SLEEP;
 654                         error = tsleep(kq, PCATCH, "kqread", timeout);
 655                 }
 656                 crit_exit();
 657                 if (error == 0)
 658                         goto retry;
 659                 /* don't restart after signals... */
 660                 if (error == ERESTART)
 661                         error = EINTR;
 662                 else if (error == EWOULDBLOCK)
 663                         error = 0;
 664                 goto done;
 665         }
 666
 667         TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
 668         while (count) {
 669                 kn = TAILQ_FIRST(&kq->kq_head);
 670                 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
 671                 if (kn == &marker) {
 672                         crit_exit();
 673                         if (count == maxevents)
 674                                 goto retry;
 675                         goto done;
 676                 }
 677                 if (kn->kn_status & KN_DISABLED) {
 678                         kn->kn_status &= ~KN_QUEUED;
 679                         kq->kq_count--;
 680                         continue;
 681                 }
 682                 if ((kn->kn_flags & EV_ONESHOT) == 0 &&
 683                     kn->kn_fop->f_event(kn, 0) == 0) {
 684                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 685                         kq->kq_count--;
 686                         continue;
 687                 }
 688                 *kevp = kn->kn_kevent;
 689                 kevp++;
 690                 nkev++;
 691                 if (kn->kn_flags & EV_ONESHOT) {
 692                         kn->kn_status &= ~KN_QUEUED;
 693                         kq->kq_count--;
 694                         crit_exit();
 695                         kn->kn_fop->f_detach(kn);
 696                         knote_drop(kn, td);
 697                         crit_enter();
 698                 } else if (kn->kn_flags & EV_CLEAR) {
 699                         kn->kn_data = 0;
 700                         kn->kn_fflags = 0;
 701                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 702                         kq->kq_count--;
 703                 } else {
 704                         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
 705                 }
 706                 count--;
 707                 if (nkev == KQ_NEVENTS) {
 708                         crit_exit();
 709                         error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 710                             sizeof(struct kevent) * nkev);
 711                         ulistp += nkev;
 712                         nkev = 0;
 713                         kevp = kq->kq_kev;
 714                         crit_enter();
 715                         if (error)
 716                                 break;
 717                 }
 718         }
 719         TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
 720         crit_exit();
 721 done:
 722         if (nkev != 0)
 723                 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 724                     sizeof(struct kevent) * nkev);
 725         *res = maxevents - count;
 726         return (error);
 727 }
 728
 729 /*
 730  * XXX
 731  * This could be expanded to call kqueue_scan, if desired.
 732  *
 733  * MPSAFE
 734  */
 735 static int
 736 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 737 {
 738         return (ENXIO);
 739 }
 740
 741 /*
 742  * MPSAFE
 743  */
 744 static int
 745 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 746 {
 747         return (ENXIO);
 748 }
 749
 750 /*
 751  * MPSAFE
 752  */
 753 static int
 754 kqueue_ioctl(struct file *fp, u_long com, caddr_t data, struct ucred *cred)
 755 {
 756         struct kqueue *kq;
 757         int error;
 758
 759         get_mplock();
 760         kq = (struct kqueue *)fp->f_data;
 761
 762         switch(com) {
 763         case FIOASYNC:
 764                 if (*(int *)data)
 765                         kq->kq_state |= KQ_ASYNC;
 766                 else
 767                         kq->kq_state &= ~KQ_ASYNC;
 768                 error = 0;
 769                 break;
 770         case FIOSETOWN:
 771                 error = fsetown(*(int *)data, &kq->kq_sigio);
 772                 break;
 773         default:
 774                 error = ENOTTY;
 775                 break;
 776         }
 777         rel_mplock();
 778         return (error);
 779 }
 780
 781 /*
 782  * MPALMOSTSAFE - acquires mplock
 783  */
 784 static int
 785 kqueue_poll(struct file *fp, int events, struct ucred *cred)
 786 {
 787         struct kqueue *kq = (struct kqueue *)fp->f_data;
 788         int revents = 0;
 789
 790         get_mplock();
 791         crit_enter();
 792         if (events & (POLLIN | POLLRDNORM)) {
 793                 if (kq->kq_count) {
 794                         revents |= events & (POLLIN | POLLRDNORM);
 795                 } else {
 796                         selrecord(curthread, &kq->kq_sel);
 797                         kq->kq_state |= KQ_SEL;
 798                 }
 799         }
 800         crit_exit();
 801         rel_mplock();
 802         return (revents);
 803 }
 804
 805 /*
 806  * MPSAFE
 807  */
 808 static int
 809 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
 810 {
 811         struct kqueue *kq = (struct kqueue *)fp->f_data;
 812
 813         bzero((void *)st, sizeof(*st));
 814         st->st_size = kq->kq_count;
 815         st->st_blksize = sizeof(struct kevent);
 816         st->st_mode = S_IFIFO;
 817         return (0);
 818 }
 819
 820 /*
 821  * MPALMOSTSAFE - acquires mplock
 822  */
 823 static int
 824 kqueue_close(struct file *fp)
 825 {
 826         struct thread *td = curthread;
 827         struct proc *p = td->td_proc;
 828         struct kqueue *kq = (struct kqueue *)fp->f_data;
 829         struct filedesc *fdp;
 830         struct knote **knp, *kn, *kn0;
 831         int i;
 832
 833         KKASSERT(p);
 834         get_mplock();
 835         fdp = p->p_fd;
 836         for (i = 0; i < fdp->fd_knlistsize; i++) {
 837                 knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
 838                 kn = *knp;
 839                 while (kn != NULL) {
 840                         kn0 = SLIST_NEXT(kn, kn_link);
 841                         if (kq == kn->kn_kq) {
 842                                 kn->kn_fop->f_detach(kn);
 843                                 fdrop(kn->kn_fp);
 844                                 knote_free(kn);
 845                                 *knp = kn0;
 846                         } else {
 847                                 knp = &SLIST_NEXT(kn, kn_link);
 848                         }
 849                         kn = kn0;
 850                 }
 851         }
 852         if (fdp->fd_knhashmask != 0) {
 853                 for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
 854                         knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
 855                         kn = *knp;
 856                         while (kn != NULL) {
 857                                 kn0 = SLIST_NEXT(kn, kn_link);
 858                                 if (kq == kn->kn_kq) {
 859                                         kn->kn_fop->f_detach(kn);
 860                 /* XXX non-fd release of kn->kn_ptr */
 861                                         knote_free(kn);
 862                                         *knp = kn0;
 863                                 } else {
 864                                         knp = &SLIST_NEXT(kn, kn_link);
 865                                 }
 866                                 kn = kn0;
 867                         }
 868                 }
 869         }
 870         fp->f_data = NULL;
 871         funsetown(kq->kq_sigio);
 872         rel_mplock();
 873
 874         kfree(kq, M_KQUEUE);
 875         return (0);
 876 }
 877
 878 static void
 879 kqueue_wakeup(struct kqueue *kq)
 880 {
 881         if (kq->kq_state & KQ_SLEEP) {
 882                 kq->kq_state &= ~KQ_SLEEP;
 883                 wakeup(kq);
 884         }
 885         if (kq->kq_state & KQ_SEL) {
 886                 kq->kq_state &= ~KQ_SEL;
 887                 selwakeup(&kq->kq_sel);
 888         }
 889         KNOTE(&kq->kq_sel.si_note, 0);
 890 }
 891
 892 /*
 893  * walk down a list of knotes, activating them if their event has triggered.
 894  */
 895 void
 896 knote(struct klist *list, long hint)
 897 {
 898         struct knote *kn;
 899
 900         SLIST_FOREACH(kn, list, kn_selnext)
 901                 if (kn->kn_fop->f_event(kn, hint))
 902                         KNOTE_ACTIVATE(kn);
 903 }
 904
 905 /*
 906  * remove all knotes from a specified klist
 907  */
 908 void
 909 knote_remove(struct thread *td, struct klist *list)
 910 {
 911         struct knote *kn;
 912
 913         while ((kn = SLIST_FIRST(list)) != NULL) {
 914                 kn->kn_fop->f_detach(kn);
 915                 knote_drop(kn, td);
 916         }
 917 }
 918
 919 /*
 920  * remove all knotes referencing a specified fd
 921  */
 922 void
 923 knote_fdclose(struct proc *p, int fd)
 924 {
 925         struct filedesc *fdp = p->p_fd;
 926         struct klist *list = &fdp->fd_knlist[fd];
 927         /* Take any thread of p */
 928         struct thread *td = FIRST_LWP_IN_PROC(p)->lwp_thread;
 929
 930         knote_remove(td, list);
 931 }
 932
 933 static void
 934 knote_attach(struct knote *kn, struct filedesc *fdp)
 935 {
 936         struct klist *list;
 937         int size;
 938
 939         if (! kn->kn_fop->f_isfd) {
 940                 if (fdp->fd_knhashmask == 0)
 941                         fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
 942                             &fdp->fd_knhashmask);
 943                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
 944                 goto done;
 945         }
 946
 947         if (fdp->fd_knlistsize <= kn->kn_id) {
 948                 size = fdp->fd_knlistsize;
 949                 while (size <= kn->kn_id)
 950                         size += KQEXTENT;
 951                 MALLOC(list, struct klist *,
 952                     size * sizeof(struct klist *), M_KQUEUE, M_WAITOK);
 953                 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
 954                     fdp->fd_knlistsize * sizeof(struct klist *));
 955                 bzero((caddr_t)list +
 956                     fdp->fd_knlistsize * sizeof(struct klist *),
 957                     (size - fdp->fd_knlistsize) * sizeof(struct klist *));
 958                 if (fdp->fd_knlist != NULL)
 959                         FREE(fdp->fd_knlist, M_KQUEUE);
 960                 fdp->fd_knlistsize = size;
 961                 fdp->fd_knlist = list;
 962         }
 963         list = &fdp->fd_knlist[kn->kn_id];
 964 done:
 965         SLIST_INSERT_HEAD(list, kn, kn_link);
 966         kn->kn_status = 0;
 967 }
 968
 969 /*
 970  * should be called outside of a critical section, since we don't want to
 971  * hold a critical section while calling fdrop and free.
 972  */
 973 static void
 974 knote_drop(struct knote *kn, struct thread *td)
 975 {
 976         struct filedesc *fdp;
 977         struct klist *list;
 978
 979         KKASSERT(td->td_proc);
 980         fdp = td->td_proc->p_fd;
 981         if (kn->kn_fop->f_isfd)
 982                 list = &fdp->fd_knlist[kn->kn_id];
 983         else
 984                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
 985
 986         SLIST_REMOVE(list, kn, knote, kn_link);
 987         if (kn->kn_status & KN_QUEUED)
 988                 knote_dequeue(kn);
 989         if (kn->kn_fop->f_isfd)
 990                 fdrop(kn->kn_fp);
 991         knote_free(kn);
 992 }
 993
 994
 995 static void
 996 knote_enqueue(struct knote *kn)
 997 {
 998         struct kqueue *kq = kn->kn_kq;
 999
1000         crit_enter();
1001         KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1002
1003         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1004         kn->kn_status |= KN_QUEUED;
1005         ++kq->kq_count;
1006
1007         /*
1008          * Send SIGIO on request (typically set up as a mailbox signal)
1009          */
1010         if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1011                 pgsigio(kq->kq_sigio, SIGIO, 0);
1012         crit_exit();
1013         kqueue_wakeup(kq);
1014 }
1015
1016 static void
1017 knote_dequeue(struct knote *kn)
1018 {
1019         struct kqueue *kq = kn->kn_kq;
1020
1021         KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1022         crit_enter();
1023
1024         TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1025         kn->kn_status &= ~KN_QUEUED;
1026         kq->kq_count--;
1027         crit_exit();
1028 }
1029
1030 static void
1031 knote_init(void)
1032 {
1033         knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
1034 }
1035 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1036
1037 static struct knote *
1038 knote_alloc(void)
1039 {
1040         return ((struct knote *)zalloc(knote_zone));
1041 }
1042
1043 static void
1044 knote_free(struct knote *kn)
1045 {
1046         zfree(knote_zone, kn);
1047 }