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/signalvar.h>
  52 #include <sys/filio.h>
  53
  54 #include <sys/thread2.h>
  55 #include <sys/file2.h>
  56 #include <sys/mplock2.h>
  57
  58 #include <vm/vm_zone.h>
  59
  60 MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
  61
  62 static int      kqueue_scan(struct file *fp, int maxevents,
  63                     struct kevent *ulistp, const struct timespec *timeout,
  64                     struct thread *td, int *res);
  65 static int      kqueue_read(struct file *fp, struct uio *uio,
  66                     struct ucred *cred, int flags);
  67 static int      kqueue_write(struct file *fp, struct uio *uio,
  68                     struct ucred *cred, int flags);
  69 static int      kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
  70                     struct ucred *cred, struct sysmsg *msg);
  71 static int      kqueue_poll(struct file *fp, int events, struct ucred *cred);
  72 static int      kqueue_kqfilter(struct file *fp, struct knote *kn);
  73 static int      kqueue_stat(struct file *fp, struct stat *st,
  74                     struct ucred *cred);
  75 static int      kqueue_close(struct file *fp);
  76 static void     kqueue_wakeup(struct kqueue *kq);
  77
  78 /*
  79  * MPSAFE
  80  */
  81 static struct fileops kqueueops = {
  82         .fo_read = kqueue_read,
  83         .fo_write = kqueue_write,
  84         .fo_ioctl = kqueue_ioctl,
  85         .fo_poll = kqueue_poll,
  86         .fo_kqfilter = kqueue_kqfilter,
  87         .fo_stat = kqueue_stat,
  88         .fo_close = kqueue_close,
  89         .fo_shutdown = nofo_shutdown
  90 };
  91
  92 static void     knote_attach(struct knote *kn, struct filedesc *fdp);
  93 static void     knote_drop(struct knote *kn, struct thread *td);
  94 static void     knote_enqueue(struct knote *kn);
  95 static void     knote_dequeue(struct knote *kn);
  96 static void     knote_init(void);
  97 static struct   knote *knote_alloc(void);
  98 static void     knote_free(struct knote *kn);
  99
 100 static void     filt_kqdetach(struct knote *kn);
 101 static int      filt_kqueue(struct knote *kn, long hint);
 102 static int      filt_procattach(struct knote *kn);
 103 static void     filt_procdetach(struct knote *kn);
 104 static int      filt_proc(struct knote *kn, long hint);
 105 static int      filt_fileattach(struct knote *kn);
 106 static void     filt_timerexpire(void *knx);
 107 static int      filt_timerattach(struct knote *kn);
 108 static void     filt_timerdetach(struct knote *kn);
 109 static int      filt_timer(struct knote *kn, long hint);
 110
 111 static struct filterops file_filtops =
 112         { 1, filt_fileattach, NULL, NULL };
 113 static struct filterops kqread_filtops =
 114         { 1, NULL, filt_kqdetach, filt_kqueue };
 115 static struct filterops proc_filtops =
 116         { 0, filt_procattach, filt_procdetach, filt_proc };
 117 static struct filterops timer_filtops =
 118         { 0, filt_timerattach, filt_timerdetach, filt_timer };
 119
 120 static vm_zone_t        knote_zone;
 121 static int              kq_ncallouts = 0;
 122 static int              kq_calloutmax = (4 * 1024);
 123 SYSCTL_INT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
 124     &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");
 125
 126 #define KNOTE_ACTIVATE(kn) do {                                         \
 127         kn->kn_status |= KN_ACTIVE;                                     \
 128         if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)           \
 129                 knote_enqueue(kn);                                      \
 130 } while(0)
 131
 132 #define KN_HASHSIZE             64              /* XXX should be tunable */
 133 #define KN_HASH(val, mask)      (((val) ^ (val >> 8)) & (mask))
 134
 135 extern struct filterops aio_filtops;
 136 extern struct filterops sig_filtops;
 137
 138 /*
 139  * Table for for all system-defined filters.
 140  */
 141 static struct filterops *sysfilt_ops[] = {
 142         &file_filtops,                  /* EVFILT_READ */
 143         &file_filtops,                  /* EVFILT_WRITE */
 144         &aio_filtops,                   /* EVFILT_AIO */
 145         &file_filtops,                  /* EVFILT_VNODE */
 146         &proc_filtops,                  /* EVFILT_PROC */
 147         &sig_filtops,                   /* EVFILT_SIGNAL */
 148         &timer_filtops,                 /* EVFILT_TIMER */
 149 };
 150
 151 static int
 152 filt_fileattach(struct knote *kn)
 153 {
 154         return (fo_kqfilter(kn->kn_fp, kn));
 155 }
 156
 157 /*
 158  * MPALMOSTSAFE - acquires mplock
 159  */
 160 static int
 161 kqueue_kqfilter(struct file *fp, struct knote *kn)
 162 {
 163         struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
 164
 165         get_mplock();
 166         if (kn->kn_filter != EVFILT_READ) {
 167                 rel_mplock();
 168                 return (1);
 169         }
 170
 171         kn->kn_fop = &kqread_filtops;
 172         SLIST_INSERT_HEAD(&kq->kq_sel.si_note, kn, kn_selnext);
 173         rel_mplock();
 174         return (0);
 175 }
 176
 177 static void
 178 filt_kqdetach(struct knote *kn)
 179 {
 180         struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
 181
 182         SLIST_REMOVE(&kq->kq_sel.si_note, kn, knote, kn_selnext);
 183 }
 184
 185 /*ARGSUSED*/
 186 static int
 187 filt_kqueue(struct knote *kn, long hint)
 188 {
 189         struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
 190
 191         kn->kn_data = kq->kq_count;
 192         return (kn->kn_data > 0);
 193 }
 194
 195 static int
 196 filt_procattach(struct knote *kn)
 197 {
 198         struct proc *p;
 199         int immediate;
 200
 201         immediate = 0;
 202         p = pfind(kn->kn_id);
 203         if (p == NULL && (kn->kn_sfflags & NOTE_EXIT)) {
 204                 p = zpfind(kn->kn_id);
 205                 immediate = 1;
 206         }
 207         if (p == NULL)
 208                 return (ESRCH);
 209         if (!PRISON_CHECK(curthread->td_ucred, p->p_ucred))
 210                 return (EACCES);
 211
 212         kn->kn_ptr.p_proc = p;
 213         kn->kn_flags |= EV_CLEAR;               /* automatically set */
 214
 215         /*
 216          * internal flag indicating registration done by kernel
 217          */
 218         if (kn->kn_flags & EV_FLAG1) {
 219                 kn->kn_data = kn->kn_sdata;             /* ppid */
 220                 kn->kn_fflags = NOTE_CHILD;
 221                 kn->kn_flags &= ~EV_FLAG1;
 222         }
 223
 224         /* XXX lock the proc here while adding to the list? */
 225         SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
 226
 227         /*
 228          * Immediately activate any exit notes if the target process is a
 229          * zombie.  This is necessary to handle the case where the target
 230          * process, e.g. a child, dies before the kevent is registered.
 231          */
 232         if (immediate && filt_proc(kn, NOTE_EXIT))
 233                 KNOTE_ACTIVATE(kn);
 234
 235         return (0);
 236 }
 237
 238 /*
 239  * The knote may be attached to a different process, which may exit,
 240  * leaving nothing for the knote to be attached to.  So when the process
 241  * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
 242  * it will be deleted when read out.  However, as part of the knote deletion,
 243  * this routine is called, so a check is needed to avoid actually performing
 244  * a detach, because the original process does not exist any more.
 245  */
 246 static void
 247 filt_procdetach(struct knote *kn)
 248 {
 249         struct proc *p;
 250
 251         if (kn->kn_status & KN_DETACHED)
 252                 return;
 253         /* XXX locking?  this might modify another process. */
 254         p = kn->kn_ptr.p_proc;
 255         SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
 256 }
 257
 258 static int
 259 filt_proc(struct knote *kn, long hint)
 260 {
 261         u_int event;
 262
 263         /*
 264          * mask off extra data
 265          */
 266         event = (u_int)hint & NOTE_PCTRLMASK;
 267
 268         /*
 269          * if the user is interested in this event, record it.
 270          */
 271         if (kn->kn_sfflags & event)
 272                 kn->kn_fflags |= event;
 273
 274         /*
 275          * Process is gone, so flag the event as finished.  Detach the
 276          * knote from the process now because the process will be poof,
 277          * gone later on.
 278          */
 279         if (event == NOTE_EXIT) {
 280                 struct proc *p = kn->kn_ptr.p_proc;
 281                 if ((kn->kn_status & KN_DETACHED) == 0) {
 282                         SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
 283                         kn->kn_status |= KN_DETACHED;
 284                         kn->kn_data = p->p_xstat;
 285                         kn->kn_ptr.p_proc = NULL;
 286                 }
 287                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 288                 return (1);
 289         }
 290
 291         /*
 292          * process forked, and user wants to track the new process,
 293          * so attach a new knote to it, and immediately report an
 294          * event with the parent's pid.
 295          */
 296         if ((event == NOTE_FORK) && (kn->kn_sfflags & NOTE_TRACK)) {
 297                 struct kevent kev;
 298                 int error;
 299
 300                 /*
 301                  * register knote with new process.
 302                  */
 303                 kev.ident = hint & NOTE_PDATAMASK;      /* pid */
 304                 kev.filter = kn->kn_filter;
 305                 kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
 306                 kev.fflags = kn->kn_sfflags;
 307                 kev.data = kn->kn_id;                   /* parent */
 308                 kev.udata = kn->kn_kevent.udata;        /* preserve udata */
 309                 error = kqueue_register(kn->kn_kq, &kev, NULL);
 310                 if (error)
 311                         kn->kn_fflags |= NOTE_TRACKERR;
 312         }
 313
 314         return (kn->kn_fflags != 0);
 315 }
 316
 317 static void
 318 filt_timerexpire(void *knx)
 319 {
 320         struct knote *kn = knx;
 321         struct callout *calloutp;
 322         struct timeval tv;
 323         int tticks;
 324
 325         kn->kn_data++;
 326         KNOTE_ACTIVATE(kn);
 327
 328         if ((kn->kn_flags & EV_ONESHOT) == 0) {
 329                 tv.tv_sec = kn->kn_sdata / 1000;
 330                 tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
 331                 tticks = tvtohz_high(&tv);
 332                 calloutp = (struct callout *)kn->kn_hook;
 333                 callout_reset(calloutp, tticks, filt_timerexpire, kn);
 334         }
 335 }
 336
 337 /*
 338  * data contains amount of time to sleep, in milliseconds
 339  */
 340 static int
 341 filt_timerattach(struct knote *kn)
 342 {
 343         struct callout *calloutp;
 344         struct timeval tv;
 345         int tticks;
 346
 347         if (kq_ncallouts >= kq_calloutmax)
 348                 return (ENOMEM);
 349         kq_ncallouts++;
 350
 351         tv.tv_sec = kn->kn_sdata / 1000;
 352         tv.tv_usec = (kn->kn_sdata % 1000) * 1000;
 353         tticks = tvtohz_high(&tv);
 354
 355         kn->kn_flags |= EV_CLEAR;               /* automatically set */
 356         MALLOC(calloutp, struct callout *, sizeof(*calloutp),
 357             M_KQUEUE, M_WAITOK);
 358         callout_init(calloutp);
 359         kn->kn_hook = (caddr_t)calloutp;
 360         callout_reset(calloutp, tticks, filt_timerexpire, kn);
 361
 362         return (0);
 363 }
 364
 365 static void
 366 filt_timerdetach(struct knote *kn)
 367 {
 368         struct callout *calloutp;
 369
 370         calloutp = (struct callout *)kn->kn_hook;
 371         callout_stop(calloutp);
 372         FREE(calloutp, M_KQUEUE);
 373         kq_ncallouts--;
 374 }
 375
 376 static int
 377 filt_timer(struct knote *kn, long hint)
 378 {
 379
 380         return (kn->kn_data != 0);
 381 }
 382
 383 /*
 384  * MPSAFE
 385  */
 386 int
 387 sys_kqueue(struct kqueue_args *uap)
 388 {
 389         struct thread *td = curthread;
 390         struct kqueue *kq;
 391         struct file *fp;
 392         int fd, error;
 393
 394         error = falloc(td->td_lwp, &fp, &fd);
 395         if (error)
 396                 return (error);
 397         fp->f_flag = FREAD | FWRITE;
 398         fp->f_type = DTYPE_KQUEUE;
 399         fp->f_ops = &kqueueops;
 400
 401         kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO);
 402         TAILQ_INIT(&kq->kq_head);
 403         kq->kq_fdp = td->td_proc->p_fd;
 404         fp->f_data = kq;
 405
 406         fsetfd(kq->kq_fdp, fp, fd);
 407         uap->sysmsg_result = fd;
 408         fdrop(fp);
 409         return (error);
 410 }
 411
 412 /*
 413  * MPALMOSTSAFE
 414  */
 415 int
 416 sys_kevent(struct kevent_args *uap)
 417 {
 418         struct thread *td = curthread;
 419         struct proc *p = td->td_proc;
 420         struct kevent *kevp;
 421         struct kqueue *kq;
 422         struct file *fp = NULL;
 423         struct timespec ts;
 424         int i, n, nerrors, error;
 425
 426         fp = holdfp(p->p_fd, uap->fd, -1);
 427         if (fp == NULL)
 428                 return (EBADF);
 429         if (fp->f_type != DTYPE_KQUEUE) {
 430                 fdrop(fp);
 431                 return (EBADF);
 432         }
 433
 434         if (uap->timeout != NULL) {
 435                 error = copyin(uap->timeout, &ts, sizeof(ts));
 436                 if (error)
 437                         goto done;
 438                 uap->timeout = &ts;
 439         }
 440
 441         kq = (struct kqueue *)fp->f_data;
 442         nerrors = 0;
 443
 444         get_mplock();
 445         while (uap->nchanges > 0) {
 446                 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges;
 447                 error = copyin(uap->changelist, kq->kq_kev,
 448                                n * sizeof(struct kevent));
 449                 if (error)
 450                         goto done;
 451                 for (i = 0; i < n; i++) {
 452                         kevp = &kq->kq_kev[i];
 453                         kevp->flags &= ~EV_SYSFLAGS;
 454                         error = kqueue_register(kq, kevp, td);
 455                         if (error) {
 456                                 if (uap->nevents != 0) {
 457                                         kevp->flags = EV_ERROR;
 458                                         kevp->data = error;
 459                                         (void) copyout((caddr_t)kevp,
 460                                             (caddr_t)uap->eventlist,
 461                                             sizeof(*kevp));
 462                                         uap->eventlist++;
 463                                         uap->nevents--;
 464                                         nerrors++;
 465                                 } else {
 466                                         goto done;
 467                                 }
 468                         }
 469                 }
 470                 uap->nchanges -= n;
 471                 uap->changelist += n;
 472         }
 473         if (nerrors) {
 474                 uap->sysmsg_result = nerrors;
 475                 error = 0;
 476                 goto done;
 477         }
 478
 479         error = kqueue_scan(fp, uap->nevents, uap->eventlist,
 480                             uap->timeout, td, &uap->sysmsg_result);
 481 done:
 482         rel_mplock();
 483         if (fp != NULL)
 484                 fdrop(fp);
 485         return (error);
 486 }
 487
 488 int
 489 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td)
 490 {
 491         struct filedesc *fdp = kq->kq_fdp;
 492         struct filterops *fops;
 493         struct file *fp = NULL;
 494         struct knote *kn = NULL;
 495         int error = 0;
 496
 497         if (kev->filter < 0) {
 498                 if (kev->filter + EVFILT_SYSCOUNT < 0)
 499                         return (EINVAL);
 500                 fops = sysfilt_ops[~kev->filter];       /* to 0-base index */
 501         } else {
 502                 /*
 503                  * XXX
 504                  * filter attach routine is responsible for insuring that
 505                  * the identifier can be attached to it.
 506                  */
 507                 kprintf("unknown filter: %d\n", kev->filter);
 508                 return (EINVAL);
 509         }
 510
 511         if (fops->f_isfd) {
 512                 /* validate descriptor */
 513                 fp = holdfp(fdp, kev->ident, -1);
 514                 if (fp == NULL)
 515                         return (EBADF);
 516
 517                 if (kev->ident < fdp->fd_knlistsize) {
 518                         SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
 519                                 if (kq == kn->kn_kq &&
 520                                     kev->filter == kn->kn_filter)
 521                                         break;
 522                 }
 523         } else {
 524                 if (fdp->fd_knhashmask != 0) {
 525                         struct klist *list;
 526
 527                         list = &fdp->fd_knhash[
 528                             KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
 529                         SLIST_FOREACH(kn, list, kn_link)
 530                                 if (kev->ident == kn->kn_id &&
 531                                     kq == kn->kn_kq &&
 532                                     kev->filter == kn->kn_filter)
 533                                         break;
 534                 }
 535         }
 536
 537         if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
 538                 error = ENOENT;
 539                 goto done;
 540         }
 541
 542         /*
 543          * kn now contains the matching knote, or NULL if no match
 544          */
 545         if (kev->flags & EV_ADD) {
 546
 547                 if (kn == NULL) {
 548                         kn = knote_alloc();
 549                         if (kn == NULL) {
 550                                 error = ENOMEM;
 551                                 goto done;
 552                         }
 553                         kn->kn_fp = fp;
 554                         kn->kn_kq = kq;
 555                         kn->kn_fop = fops;
 556
 557                         /*
 558                          * apply reference count to knote structure, and
 559                          * do not release it at the end of this routine.
 560                          */
 561                         fp = NULL;
 562
 563                         kn->kn_sfflags = kev->fflags;
 564                         kn->kn_sdata = kev->data;
 565                         kev->fflags = 0;
 566                         kev->data = 0;
 567                         kn->kn_kevent = *kev;
 568
 569                         knote_attach(kn, fdp);
 570                         if ((error = fops->f_attach(kn)) != 0) {
 571                                 knote_drop(kn, td);
 572                                 goto done;
 573                         }
 574                 } else {
 575                         /*
 576                          * The user may change some filter values after the
 577                          * initial EV_ADD, but doing so will not reset any
 578                          * filter which have already been triggered.
 579                          */
 580                         kn->kn_sfflags = kev->fflags;
 581                         kn->kn_sdata = kev->data;
 582                         kn->kn_kevent.udata = kev->udata;
 583                 }
 584
 585                 crit_enter();
 586                 if (kn->kn_fop->f_event(kn, 0))
 587                         KNOTE_ACTIVATE(kn);
 588                 crit_exit();
 589         } else if (kev->flags & EV_DELETE) {
 590                 kn->kn_fop->f_detach(kn);
 591                 knote_drop(kn, td);
 592                 goto done;
 593         }
 594
 595         if ((kev->flags & EV_DISABLE) &&
 596             ((kn->kn_status & KN_DISABLED) == 0)) {
 597                 crit_enter();
 598                 kn->kn_status |= KN_DISABLED;
 599                 crit_exit();
 600         }
 601
 602         if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
 603                 crit_enter();
 604                 kn->kn_status &= ~KN_DISABLED;
 605                 if ((kn->kn_status & KN_ACTIVE) &&
 606                     ((kn->kn_status & KN_QUEUED) == 0))
 607                         knote_enqueue(kn);
 608                 crit_exit();
 609         }
 610
 611 done:
 612         if (fp != NULL)
 613                 fdrop(fp);
 614         return (error);
 615 }
 616
 617 static int
 618 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
 619         const struct timespec *tsp, struct thread *td, int *res)
 620 {
 621         struct kqueue *kq = (struct kqueue *)fp->f_data;
 622         struct kevent *kevp;
 623         struct timeval atv, rtv, ttv;
 624         struct knote *kn, marker;
 625         int count, timeout, nkev = 0, error = 0;
 626
 627         count = maxevents;
 628         if (count == 0)
 629                 goto done;
 630
 631         if (tsp != NULL) {
 632                 TIMESPEC_TO_TIMEVAL(&atv, tsp);
 633                 if (itimerfix(&atv)) {
 634                         error = EINVAL;
 635                         goto done;
 636                 }
 637                 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
 638                         timeout = -1;
 639                 else
 640                         timeout = atv.tv_sec > 24 * 60 * 60 ?
 641                             24 * 60 * 60 * hz : tvtohz_high(&atv);
 642                 getmicrouptime(&rtv);
 643                 timevaladd(&atv, &rtv);
 644         } else {
 645                 atv.tv_sec = 0;
 646                 atv.tv_usec = 0;
 647                 timeout = 0;
 648         }
 649         goto start;
 650
 651 retry:
 652         if (atv.tv_sec || atv.tv_usec) {
 653                 getmicrouptime(&rtv);
 654                 if (timevalcmp(&rtv, &atv, >=))
 655                         goto done;
 656                 ttv = atv;
 657                 timevalsub(&ttv, &rtv);
 658                 timeout = ttv.tv_sec > 24 * 60 * 60 ?
 659                         24 * 60 * 60 * hz : tvtohz_high(&ttv);
 660         }
 661
 662 start:
 663         kevp = kq->kq_kev;
 664         crit_enter();
 665         if (kq->kq_count == 0) {
 666                 if (timeout < 0) {
 667                         error = EWOULDBLOCK;
 668                 } else {
 669                         kq->kq_state |= KQ_SLEEP;
 670                         error = tsleep(kq, PCATCH, "kqread", timeout);
 671                 }
 672                 crit_exit();
 673                 if (error == 0)
 674                         goto retry;
 675                 /* don't restart after signals... */
 676                 if (error == ERESTART)
 677                         error = EINTR;
 678                 else if (error == EWOULDBLOCK)
 679                         error = 0;
 680                 goto done;
 681         }
 682
 683         TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
 684         while (count) {
 685                 kn = TAILQ_FIRST(&kq->kq_head);
 686                 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
 687                 if (kn == &marker) {
 688                         crit_exit();
 689                         if (count == maxevents)
 690                                 goto retry;
 691                         goto done;
 692                 }
 693                 if (kn->kn_status & KN_DISABLED) {
 694                         kn->kn_status &= ~KN_QUEUED;
 695                         kq->kq_count--;
 696                         continue;
 697                 }
 698                 if ((kn->kn_flags & EV_ONESHOT) == 0 &&
 699                     kn->kn_fop->f_event(kn, 0) == 0) {
 700                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 701                         kq->kq_count--;
 702                         continue;
 703                 }
 704                 *kevp = kn->kn_kevent;
 705                 kevp++;
 706                 nkev++;
 707                 if (kn->kn_flags & EV_ONESHOT) {
 708                         kn->kn_status &= ~KN_QUEUED;
 709                         kq->kq_count--;
 710                         crit_exit();
 711                         kn->kn_fop->f_detach(kn);
 712                         knote_drop(kn, td);
 713                         crit_enter();
 714                 } else if (kn->kn_flags & EV_CLEAR) {
 715                         kn->kn_data = 0;
 716                         kn->kn_fflags = 0;
 717                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 718                         kq->kq_count--;
 719                 } else {
 720                         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
 721                 }
 722                 count--;
 723                 if (nkev == KQ_NEVENTS) {
 724                         crit_exit();
 725                         error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 726                             sizeof(struct kevent) * nkev);
 727                         ulistp += nkev;
 728                         nkev = 0;
 729                         kevp = kq->kq_kev;
 730                         crit_enter();
 731                         if (error)
 732                                 break;
 733                 }
 734         }
 735         TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
 736         crit_exit();
 737 done:
 738         if (nkev != 0)
 739                 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 740                     sizeof(struct kevent) * nkev);
 741         *res = maxevents - count;
 742         return (error);
 743 }
 744
 745 /*
 746  * XXX
 747  * This could be expanded to call kqueue_scan, if desired.
 748  *
 749  * MPSAFE
 750  */
 751 static int
 752 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 753 {
 754         return (ENXIO);
 755 }
 756
 757 /*
 758  * MPSAFE
 759  */
 760 static int
 761 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 762 {
 763         return (ENXIO);
 764 }
 765
 766 /*
 767  * MPSAFE
 768  */
 769 static int
 770 kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
 771              struct ucred *cred, struct sysmsg *msg)
 772 {
 773         struct kqueue *kq;
 774         int error;
 775
 776         get_mplock();
 777         kq = (struct kqueue *)fp->f_data;
 778
 779         switch(com) {
 780         case FIOASYNC:
 781                 if (*(int *)data)
 782                         kq->kq_state |= KQ_ASYNC;
 783                 else
 784                         kq->kq_state &= ~KQ_ASYNC;
 785                 error = 0;
 786                 break;
 787         case FIOSETOWN:
 788                 error = fsetown(*(int *)data, &kq->kq_sigio);
 789                 break;
 790         default:
 791                 error = ENOTTY;
 792                 break;
 793         }
 794         rel_mplock();
 795         return (error);
 796 }
 797
 798 /*
 799  * MPALMOSTSAFE - acquires mplock
 800  */
 801 static int
 802 kqueue_poll(struct file *fp, int events, struct ucred *cred)
 803 {
 804         struct kqueue *kq = (struct kqueue *)fp->f_data;
 805         int revents = 0;
 806
 807         get_mplock();
 808         crit_enter();
 809         if (events & (POLLIN | POLLRDNORM)) {
 810                 if (kq->kq_count) {
 811                         revents |= events & (POLLIN | POLLRDNORM);
 812                 } else {
 813                         selrecord(curthread, &kq->kq_sel);
 814                         kq->kq_state |= KQ_SEL;
 815                 }
 816         }
 817         crit_exit();
 818         rel_mplock();
 819         return (revents);
 820 }
 821
 822 /*
 823  * MPSAFE
 824  */
 825 static int
 826 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
 827 {
 828         struct kqueue *kq = (struct kqueue *)fp->f_data;
 829
 830         bzero((void *)st, sizeof(*st));
 831         st->st_size = kq->kq_count;
 832         st->st_blksize = sizeof(struct kevent);
 833         st->st_mode = S_IFIFO;
 834         return (0);
 835 }
 836
 837 /*
 838  * MPALMOSTSAFE - acquires mplock
 839  */
 840 static int
 841 kqueue_close(struct file *fp)
 842 {
 843         struct thread *td = curthread;
 844         struct proc *p = td->td_proc;
 845         struct kqueue *kq = (struct kqueue *)fp->f_data;
 846         struct filedesc *fdp;
 847         struct knote **knp, *kn, *kn0;
 848         int i;
 849
 850         KKASSERT(p);
 851         get_mplock();
 852         fdp = p->p_fd;
 853         for (i = 0; i < fdp->fd_knlistsize; i++) {
 854                 knp = &SLIST_FIRST(&fdp->fd_knlist[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                                 fdrop(kn->kn_fp);
 861                                 knote_free(kn);
 862                                 *knp = kn0;
 863                         } else {
 864                                 knp = &SLIST_NEXT(kn, kn_link);
 865                         }
 866                         kn = kn0;
 867                 }
 868         }
 869         if (fdp->fd_knhashmask != 0) {
 870                 for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
 871                         knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
 872                         kn = *knp;
 873                         while (kn != NULL) {
 874                                 kn0 = SLIST_NEXT(kn, kn_link);
 875                                 if (kq == kn->kn_kq) {
 876                                         kn->kn_fop->f_detach(kn);
 877                 /* XXX non-fd release of kn->kn_ptr */
 878                                         knote_free(kn);
 879                                         *knp = kn0;
 880                                 } else {
 881                                         knp = &SLIST_NEXT(kn, kn_link);
 882                                 }
 883                                 kn = kn0;
 884                         }
 885                 }
 886         }
 887         fp->f_data = NULL;
 888         funsetown(kq->kq_sigio);
 889         rel_mplock();
 890
 891         kfree(kq, M_KQUEUE);
 892         return (0);
 893 }
 894
 895 static void
 896 kqueue_wakeup(struct kqueue *kq)
 897 {
 898         if (kq->kq_state & KQ_SLEEP) {
 899                 kq->kq_state &= ~KQ_SLEEP;
 900                 wakeup(kq);
 901         }
 902         if (kq->kq_state & KQ_SEL) {
 903                 kq->kq_state &= ~KQ_SEL;
 904                 selwakeup(&kq->kq_sel);
 905         }
 906         KNOTE(&kq->kq_sel.si_note, 0);
 907 }
 908
 909 /*
 910  * walk down a list of knotes, activating them if their event has triggered.
 911  */
 912 void
 913 knote(struct klist *list, long hint)
 914 {
 915         struct knote *kn;
 916
 917         SLIST_FOREACH(kn, list, kn_selnext)
 918                 if (kn->kn_fop->f_event(kn, hint))
 919                         KNOTE_ACTIVATE(kn);
 920 }
 921
 922 /*
 923  * remove all knotes from a specified klist
 924  */
 925 void
 926 knote_remove(struct thread *td, struct klist *list)
 927 {
 928         struct knote *kn;
 929
 930         while ((kn = SLIST_FIRST(list)) != NULL) {
 931                 kn->kn_fop->f_detach(kn);
 932                 knote_drop(kn, td);
 933         }
 934 }
 935
 936 /*
 937  * remove all knotes referencing a specified fd
 938  */
 939 void
 940 knote_fdclose(struct proc *p, int fd)
 941 {
 942         struct filedesc *fdp = p->p_fd;
 943         struct klist *list = &fdp->fd_knlist[fd];
 944         /* Take any thread of p */
 945         struct thread *td = FIRST_LWP_IN_PROC(p)->lwp_thread;
 946
 947         knote_remove(td, list);
 948 }
 949
 950 static void
 951 knote_attach(struct knote *kn, struct filedesc *fdp)
 952 {
 953         struct klist *list;
 954         int size;
 955
 956         if (! kn->kn_fop->f_isfd) {
 957                 if (fdp->fd_knhashmask == 0)
 958                         fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
 959                             &fdp->fd_knhashmask);
 960                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
 961                 goto done;
 962         }
 963
 964         if (fdp->fd_knlistsize <= kn->kn_id) {
 965                 size = fdp->fd_knlistsize;
 966                 while (size <= kn->kn_id)
 967                         size += KQEXTENT;
 968                 MALLOC(list, struct klist *,
 969                     size * sizeof(struct klist *), M_KQUEUE, M_WAITOK);
 970                 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
 971                     fdp->fd_knlistsize * sizeof(struct klist *));
 972                 bzero((caddr_t)list +
 973                     fdp->fd_knlistsize * sizeof(struct klist *),
 974                     (size - fdp->fd_knlistsize) * sizeof(struct klist *));
 975                 if (fdp->fd_knlist != NULL)
 976                         FREE(fdp->fd_knlist, M_KQUEUE);
 977                 fdp->fd_knlistsize = size;
 978                 fdp->fd_knlist = list;
 979         }
 980         list = &fdp->fd_knlist[kn->kn_id];
 981 done:
 982         SLIST_INSERT_HEAD(list, kn, kn_link);
 983         kn->kn_status = 0;
 984 }
 985
 986 /*
 987  * should be called outside of a critical section, since we don't want to
 988  * hold a critical section while calling fdrop and free.
 989  */
 990 static void
 991 knote_drop(struct knote *kn, struct thread *td)
 992 {
 993         struct filedesc *fdp;
 994         struct klist *list;
 995
 996         KKASSERT(td->td_proc);
 997         fdp = td->td_proc->p_fd;
 998         if (kn->kn_fop->f_isfd)
 999                 list = &fdp->fd_knlist[kn->kn_id];
1000         else
1001                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1002
1003         SLIST_REMOVE(list, kn, knote, kn_link);
1004         if (kn->kn_status & KN_QUEUED)
1005                 knote_dequeue(kn);
1006         if (kn->kn_fop->f_isfd)
1007                 fdrop(kn->kn_fp);
1008         knote_free(kn);
1009 }
1010
1011
1012 static void
1013 knote_enqueue(struct knote *kn)
1014 {
1015         struct kqueue *kq = kn->kn_kq;
1016
1017         crit_enter();
1018         KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1019
1020         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1021         kn->kn_status |= KN_QUEUED;
1022         ++kq->kq_count;
1023
1024         /*
1025          * Send SIGIO on request (typically set up as a mailbox signal)
1026          */
1027         if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1028                 pgsigio(kq->kq_sigio, SIGIO, 0);
1029         crit_exit();
1030         kqueue_wakeup(kq);
1031 }
1032
1033 static void
1034 knote_dequeue(struct knote *kn)
1035 {
1036         struct kqueue *kq = kn->kn_kq;
1037
1038         KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1039         crit_enter();
1040
1041         TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1042         kn->kn_status &= ~KN_QUEUED;
1043         kq->kq_count--;
1044         crit_exit();
1045 }
1046
1047 static void
1048 knote_init(void)
1049 {
1050         knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
1051 }
1052 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1053
1054 static struct knote *
1055 knote_alloc(void)
1056 {
1057         return ((struct knote *)zalloc(knote_zone));
1058 }
1059
1060 static void
1061 knote_free(struct knote *kn)
1062 {
1063         zfree(knote_zone, kn);
1064 }