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 /*
 382  * MPSAFE
 383  */
 384 int
 385 sys_kqueue(struct kqueue_args *uap)
 386 {
 387         struct proc *p = curproc;
 388         struct filedesc *fdp = p->p_fd;
 389         struct kqueue *kq;
 390         struct file *fp;
 391         int fd, error;
 392
 393         error = falloc(p, &fp, &fd);
 394         if (error)
 395                 return (error);
 396         fp->f_flag = FREAD | FWRITE;
 397         fp->f_type = DTYPE_KQUEUE;
 398         fp->f_ops = &kqueueops;
 399
 400         kq = kmalloc(sizeof(struct kqueue), M_KQUEUE, M_WAITOK | M_ZERO);
 401         TAILQ_INIT(&kq->kq_head);
 402         kq->kq_fdp = fdp;
 403         fp->f_data = kq;
 404
 405         fsetfd(p, fp, fd);
 406         uap->sysmsg_result = fd;
 407         fdrop(fp);
 408         return (error);
 409 }
 410
 411 /*
 412  * MPALMOSTSAFE
 413  */
 414 int
 415 sys_kevent(struct kevent_args *uap)
 416 {
 417         struct thread *td = curthread;
 418         struct proc *p = td->td_proc;
 419         struct kevent *kevp;
 420         struct kqueue *kq;
 421         struct file *fp = NULL;
 422         struct timespec ts;
 423         int i, n, nerrors, error;
 424
 425         fp = holdfp(p->p_fd, uap->fd, -1);
 426         if (fp == NULL)
 427                 return (EBADF);
 428         if (fp->f_type != DTYPE_KQUEUE) {
 429                 fdrop(fp);
 430                 return (EBADF);
 431         }
 432
 433         if (uap->timeout != NULL) {
 434                 error = copyin(uap->timeout, &ts, sizeof(ts));
 435                 if (error)
 436                         goto done;
 437                 uap->timeout = &ts;
 438         }
 439
 440         kq = (struct kqueue *)fp->f_data;
 441         nerrors = 0;
 442
 443         get_mplock();
 444         while (uap->nchanges > 0) {
 445                 n = uap->nchanges > KQ_NEVENTS ? KQ_NEVENTS : uap->nchanges;
 446                 error = copyin(uap->changelist, kq->kq_kev,
 447                                n * sizeof(struct kevent));
 448                 if (error)
 449                         goto done;
 450                 for (i = 0; i < n; i++) {
 451                         kevp = &kq->kq_kev[i];
 452                         kevp->flags &= ~EV_SYSFLAGS;
 453                         error = kqueue_register(kq, kevp, td);
 454                         if (error) {
 455                                 if (uap->nevents != 0) {
 456                                         kevp->flags = EV_ERROR;
 457                                         kevp->data = error;
 458                                         (void) copyout((caddr_t)kevp,
 459                                             (caddr_t)uap->eventlist,
 460                                             sizeof(*kevp));
 461                                         uap->eventlist++;
 462                                         uap->nevents--;
 463                                         nerrors++;
 464                                 } else {
 465                                         goto done;
 466                                 }
 467                         }
 468                 }
 469                 uap->nchanges -= n;
 470                 uap->changelist += n;
 471         }
 472         if (nerrors) {
 473                 uap->sysmsg_result = nerrors;
 474                 error = 0;
 475                 goto done;
 476         }
 477
 478         error = kqueue_scan(fp, uap->nevents, uap->eventlist,
 479                             uap->timeout, td, &uap->sysmsg_result);
 480 done:
 481         rel_mplock();
 482         if (fp != NULL)
 483                 fdrop(fp);
 484         return (error);
 485 }
 486
 487 int
 488 kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td)
 489 {
 490         struct filedesc *fdp = kq->kq_fdp;
 491         struct filterops *fops;
 492         struct file *fp = NULL;
 493         struct knote *kn = NULL;
 494         int error = 0;
 495
 496         if (kev->filter < 0) {
 497                 if (kev->filter + EVFILT_SYSCOUNT < 0)
 498                         return (EINVAL);
 499                 fops = sysfilt_ops[~kev->filter];       /* to 0-base index */
 500         } else {
 501                 /*
 502                  * XXX
 503                  * filter attach routine is responsible for insuring that
 504                  * the identifier can be attached to it.
 505                  */
 506                 kprintf("unknown filter: %d\n", kev->filter);
 507                 return (EINVAL);
 508         }
 509
 510         if (fops->f_isfd) {
 511                 /* validate descriptor */
 512                 fp = holdfp(fdp, kev->ident, -1);
 513                 if (fp == NULL)
 514                         return (EBADF);
 515
 516                 if (kev->ident < fdp->fd_knlistsize) {
 517                         SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
 518                                 if (kq == kn->kn_kq &&
 519                                     kev->filter == kn->kn_filter)
 520                                         break;
 521                 }
 522         } else {
 523                 if (fdp->fd_knhashmask != 0) {
 524                         struct klist *list;
 525
 526                         list = &fdp->fd_knhash[
 527                             KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
 528                         SLIST_FOREACH(kn, list, kn_link)
 529                                 if (kev->ident == kn->kn_id &&
 530                                     kq == kn->kn_kq &&
 531                                     kev->filter == kn->kn_filter)
 532                                         break;
 533                 }
 534         }
 535
 536         if (kn == NULL && ((kev->flags & EV_ADD) == 0)) {
 537                 error = ENOENT;
 538                 goto done;
 539         }
 540
 541         /*
 542          * kn now contains the matching knote, or NULL if no match
 543          */
 544         if (kev->flags & EV_ADD) {
 545
 546                 if (kn == NULL) {
 547                         kn = knote_alloc();
 548                         if (kn == NULL) {
 549                                 error = ENOMEM;
 550                                 goto done;
 551                         }
 552                         kn->kn_fp = fp;
 553                         kn->kn_kq = kq;
 554                         kn->kn_fop = fops;
 555
 556                         /*
 557                          * apply reference count to knote structure, and
 558                          * do not release it at the end of this routine.
 559                          */
 560                         fp = NULL;
 561
 562                         kn->kn_sfflags = kev->fflags;
 563                         kn->kn_sdata = kev->data;
 564                         kev->fflags = 0;
 565                         kev->data = 0;
 566                         kn->kn_kevent = *kev;
 567
 568                         knote_attach(kn, fdp);
 569                         if ((error = fops->f_attach(kn)) != 0) {
 570                                 knote_drop(kn, td);
 571                                 goto done;
 572                         }
 573                 } else {
 574                         /*
 575                          * The user may change some filter values after the
 576                          * initial EV_ADD, but doing so will not reset any
 577                          * filter which have already been triggered.
 578                          */
 579                         kn->kn_sfflags = kev->fflags;
 580                         kn->kn_sdata = kev->data;
 581                         kn->kn_kevent.udata = kev->udata;
 582                 }
 583
 584                 crit_enter();
 585                 if (kn->kn_fop->f_event(kn, 0))
 586                         KNOTE_ACTIVATE(kn);
 587                 crit_exit();
 588         } else if (kev->flags & EV_DELETE) {
 589                 kn->kn_fop->f_detach(kn);
 590                 knote_drop(kn, td);
 591                 goto done;
 592         }
 593
 594         if ((kev->flags & EV_DISABLE) &&
 595             ((kn->kn_status & KN_DISABLED) == 0)) {
 596                 crit_enter();
 597                 kn->kn_status |= KN_DISABLED;
 598                 crit_exit();
 599         }
 600
 601         if ((kev->flags & EV_ENABLE) && (kn->kn_status & KN_DISABLED)) {
 602                 crit_enter();
 603                 kn->kn_status &= ~KN_DISABLED;
 604                 if ((kn->kn_status & KN_ACTIVE) &&
 605                     ((kn->kn_status & KN_QUEUED) == 0))
 606                         knote_enqueue(kn);
 607                 crit_exit();
 608         }
 609
 610 done:
 611         if (fp != NULL)
 612                 fdrop(fp);
 613         return (error);
 614 }
 615
 616 static int
 617 kqueue_scan(struct file *fp, int maxevents, struct kevent *ulistp,
 618         const struct timespec *tsp, struct thread *td, int *res)
 619 {
 620         struct kqueue *kq = (struct kqueue *)fp->f_data;
 621         struct kevent *kevp;
 622         struct timeval atv, rtv, ttv;
 623         struct knote *kn, marker;
 624         int count, timeout, nkev = 0, error = 0;
 625
 626         count = maxevents;
 627         if (count == 0)
 628                 goto done;
 629
 630         if (tsp != NULL) {
 631                 TIMESPEC_TO_TIMEVAL(&atv, tsp);
 632                 if (itimerfix(&atv)) {
 633                         error = EINVAL;
 634                         goto done;
 635                 }
 636                 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
 637                         timeout = -1;
 638                 else
 639                         timeout = atv.tv_sec > 24 * 60 * 60 ?
 640                             24 * 60 * 60 * hz : tvtohz_high(&atv);
 641                 getmicrouptime(&rtv);
 642                 timevaladd(&atv, &rtv);
 643         } else {
 644                 atv.tv_sec = 0;
 645                 atv.tv_usec = 0;
 646                 timeout = 0;
 647         }
 648         goto start;
 649
 650 retry:
 651         if (atv.tv_sec || atv.tv_usec) {
 652                 getmicrouptime(&rtv);
 653                 if (timevalcmp(&rtv, &atv, >=))
 654                         goto done;
 655                 ttv = atv;
 656                 timevalsub(&ttv, &rtv);
 657                 timeout = ttv.tv_sec > 24 * 60 * 60 ?
 658                         24 * 60 * 60 * hz : tvtohz_high(&ttv);
 659         }
 660
 661 start:
 662         kevp = kq->kq_kev;
 663         crit_enter();
 664         if (kq->kq_count == 0) {
 665                 if (timeout < 0) {
 666                         error = EWOULDBLOCK;
 667                 } else {
 668                         kq->kq_state |= KQ_SLEEP;
 669                         error = tsleep(kq, PCATCH, "kqread", timeout);
 670                 }
 671                 crit_exit();
 672                 if (error == 0)
 673                         goto retry;
 674                 /* don't restart after signals... */
 675                 if (error == ERESTART)
 676                         error = EINTR;
 677                 else if (error == EWOULDBLOCK)
 678                         error = 0;
 679                 goto done;
 680         }
 681
 682         TAILQ_INSERT_TAIL(&kq->kq_head, &marker, kn_tqe);
 683         while (count) {
 684                 kn = TAILQ_FIRST(&kq->kq_head);
 685                 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
 686                 if (kn == &marker) {
 687                         crit_exit();
 688                         if (count == maxevents)
 689                                 goto retry;
 690                         goto done;
 691                 }
 692                 if (kn->kn_status & KN_DISABLED) {
 693                         kn->kn_status &= ~KN_QUEUED;
 694                         kq->kq_count--;
 695                         continue;
 696                 }
 697                 if ((kn->kn_flags & EV_ONESHOT) == 0 &&
 698                     kn->kn_fop->f_event(kn, 0) == 0) {
 699                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 700                         kq->kq_count--;
 701                         continue;
 702                 }
 703                 *kevp = kn->kn_kevent;
 704                 kevp++;
 705                 nkev++;
 706                 if (kn->kn_flags & EV_ONESHOT) {
 707                         kn->kn_status &= ~KN_QUEUED;
 708                         kq->kq_count--;
 709                         crit_exit();
 710                         kn->kn_fop->f_detach(kn);
 711                         knote_drop(kn, td);
 712                         crit_enter();
 713                 } else if (kn->kn_flags & EV_CLEAR) {
 714                         kn->kn_data = 0;
 715                         kn->kn_fflags = 0;
 716                         kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
 717                         kq->kq_count--;
 718                 } else {
 719                         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
 720                 }
 721                 count--;
 722                 if (nkev == KQ_NEVENTS) {
 723                         crit_exit();
 724                         error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 725                             sizeof(struct kevent) * nkev);
 726                         ulistp += nkev;
 727                         nkev = 0;
 728                         kevp = kq->kq_kev;
 729                         crit_enter();
 730                         if (error)
 731                                 break;
 732                 }
 733         }
 734         TAILQ_REMOVE(&kq->kq_head, &marker, kn_tqe);
 735         crit_exit();
 736 done:
 737         if (nkev != 0)
 738                 error = copyout((caddr_t)&kq->kq_kev, (caddr_t)ulistp,
 739                     sizeof(struct kevent) * nkev);
 740         *res = maxevents - count;
 741         return (error);
 742 }
 743
 744 /*
 745  * XXX
 746  * This could be expanded to call kqueue_scan, if desired.
 747  *
 748  * MPSAFE
 749  */
 750 static int
 751 kqueue_read(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 752 {
 753         return (ENXIO);
 754 }
 755
 756 /*
 757  * MPSAFE
 758  */
 759 static int
 760 kqueue_write(struct file *fp, struct uio *uio, struct ucred *cred, int flags)
 761 {
 762         return (ENXIO);
 763 }
 764
 765 /*
 766  * MPSAFE
 767  */
 768 static int
 769 kqueue_ioctl(struct file *fp, u_long com, caddr_t data,
 770              struct ucred *cred, struct sysmsg *msg)
 771 {
 772         struct kqueue *kq;
 773         int error;
 774
 775         get_mplock();
 776         kq = (struct kqueue *)fp->f_data;
 777
 778         switch(com) {
 779         case FIOASYNC:
 780                 if (*(int *)data)
 781                         kq->kq_state |= KQ_ASYNC;
 782                 else
 783                         kq->kq_state &= ~KQ_ASYNC;
 784                 error = 0;
 785                 break;
 786         case FIOSETOWN:
 787                 error = fsetown(*(int *)data, &kq->kq_sigio);
 788                 break;
 789         default:
 790                 error = ENOTTY;
 791                 break;
 792         }
 793         rel_mplock();
 794         return (error);
 795 }
 796
 797 /*
 798  * MPALMOSTSAFE - acquires mplock
 799  */
 800 static int
 801 kqueue_poll(struct file *fp, int events, struct ucred *cred)
 802 {
 803         struct kqueue *kq = (struct kqueue *)fp->f_data;
 804         int revents = 0;
 805
 806         get_mplock();
 807         crit_enter();
 808         if (events & (POLLIN | POLLRDNORM)) {
 809                 if (kq->kq_count) {
 810                         revents |= events & (POLLIN | POLLRDNORM);
 811                 } else {
 812                         selrecord(curthread, &kq->kq_sel);
 813                         kq->kq_state |= KQ_SEL;
 814                 }
 815         }
 816         crit_exit();
 817         rel_mplock();
 818         return (revents);
 819 }
 820
 821 /*
 822  * MPSAFE
 823  */
 824 static int
 825 kqueue_stat(struct file *fp, struct stat *st, struct ucred *cred)
 826 {
 827         struct kqueue *kq = (struct kqueue *)fp->f_data;
 828
 829         bzero((void *)st, sizeof(*st));
 830         st->st_size = kq->kq_count;
 831         st->st_blksize = sizeof(struct kevent);
 832         st->st_mode = S_IFIFO;
 833         return (0);
 834 }
 835
 836 /*
 837  * MPALMOSTSAFE - acquires mplock
 838  */
 839 static int
 840 kqueue_close(struct file *fp)
 841 {
 842         struct thread *td = curthread;
 843         struct proc *p = td->td_proc;
 844         struct kqueue *kq = (struct kqueue *)fp->f_data;
 845         struct filedesc *fdp;
 846         struct knote **knp, *kn, *kn0;
 847         int i;
 848
 849         KKASSERT(p);
 850         get_mplock();
 851         fdp = p->p_fd;
 852         for (i = 0; i < fdp->fd_knlistsize; i++) {
 853                 knp = &SLIST_FIRST(&fdp->fd_knlist[i]);
 854                 kn = *knp;
 855                 while (kn != NULL) {
 856                         kn0 = SLIST_NEXT(kn, kn_link);
 857                         if (kq == kn->kn_kq) {
 858                                 kn->kn_fop->f_detach(kn);
 859                                 fdrop(kn->kn_fp);
 860                                 knote_free(kn);
 861                                 *knp = kn0;
 862                         } else {
 863                                 knp = &SLIST_NEXT(kn, kn_link);
 864                         }
 865                         kn = kn0;
 866                 }
 867         }
 868         if (fdp->fd_knhashmask != 0) {
 869                 for (i = 0; i < fdp->fd_knhashmask + 1; i++) {
 870                         knp = &SLIST_FIRST(&fdp->fd_knhash[i]);
 871                         kn = *knp;
 872                         while (kn != NULL) {
 873                                 kn0 = SLIST_NEXT(kn, kn_link);
 874                                 if (kq == kn->kn_kq) {
 875                                         kn->kn_fop->f_detach(kn);
 876                 /* XXX non-fd release of kn->kn_ptr */
 877                                         knote_free(kn);
 878                                         *knp = kn0;
 879                                 } else {
 880                                         knp = &SLIST_NEXT(kn, kn_link);
 881                                 }
 882                                 kn = kn0;
 883                         }
 884                 }
 885         }
 886         fp->f_data = NULL;
 887         funsetown(kq->kq_sigio);
 888         rel_mplock();
 889
 890         kfree(kq, M_KQUEUE);
 891         return (0);
 892 }
 893
 894 static void
 895 kqueue_wakeup(struct kqueue *kq)
 896 {
 897         if (kq->kq_state & KQ_SLEEP) {
 898                 kq->kq_state &= ~KQ_SLEEP;
 899                 wakeup(kq);
 900         }
 901         if (kq->kq_state & KQ_SEL) {
 902                 kq->kq_state &= ~KQ_SEL;
 903                 selwakeup(&kq->kq_sel);
 904         }
 905         KNOTE(&kq->kq_sel.si_note, 0);
 906 }
 907
 908 /*
 909  * walk down a list of knotes, activating them if their event has triggered.
 910  */
 911 void
 912 knote(struct klist *list, long hint)
 913 {
 914         struct knote *kn;
 915
 916         SLIST_FOREACH(kn, list, kn_selnext)
 917                 if (kn->kn_fop->f_event(kn, hint))
 918                         KNOTE_ACTIVATE(kn);
 919 }
 920
 921 /*
 922  * remove all knotes from a specified klist
 923  */
 924 void
 925 knote_remove(struct thread *td, struct klist *list)
 926 {
 927         struct knote *kn;
 928
 929         while ((kn = SLIST_FIRST(list)) != NULL) {
 930                 kn->kn_fop->f_detach(kn);
 931                 knote_drop(kn, td);
 932         }
 933 }
 934
 935 /*
 936  * remove all knotes referencing a specified fd
 937  */
 938 void
 939 knote_fdclose(struct proc *p, int fd)
 940 {
 941         struct filedesc *fdp = p->p_fd;
 942         struct klist *list = &fdp->fd_knlist[fd];
 943         /* Take any thread of p */
 944         struct thread *td = FIRST_LWP_IN_PROC(p)->lwp_thread;
 945
 946         knote_remove(td, list);
 947 }
 948
 949 static void
 950 knote_attach(struct knote *kn, struct filedesc *fdp)
 951 {
 952         struct klist *list;
 953         int size;
 954
 955         if (! kn->kn_fop->f_isfd) {
 956                 if (fdp->fd_knhashmask == 0)
 957                         fdp->fd_knhash = hashinit(KN_HASHSIZE, M_KQUEUE,
 958                             &fdp->fd_knhashmask);
 959                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
 960                 goto done;
 961         }
 962
 963         if (fdp->fd_knlistsize <= kn->kn_id) {
 964                 size = fdp->fd_knlistsize;
 965                 while (size <= kn->kn_id)
 966                         size += KQEXTENT;
 967                 MALLOC(list, struct klist *,
 968                     size * sizeof(struct klist *), M_KQUEUE, M_WAITOK);
 969                 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
 970                     fdp->fd_knlistsize * sizeof(struct klist *));
 971                 bzero((caddr_t)list +
 972                     fdp->fd_knlistsize * sizeof(struct klist *),
 973                     (size - fdp->fd_knlistsize) * sizeof(struct klist *));
 974                 if (fdp->fd_knlist != NULL)
 975                         FREE(fdp->fd_knlist, M_KQUEUE);
 976                 fdp->fd_knlistsize = size;
 977                 fdp->fd_knlist = list;
 978         }
 979         list = &fdp->fd_knlist[kn->kn_id];
 980 done:
 981         SLIST_INSERT_HEAD(list, kn, kn_link);
 982         kn->kn_status = 0;
 983 }
 984
 985 /*
 986  * should be called outside of a critical section, since we don't want to
 987  * hold a critical section while calling fdrop and free.
 988  */
 989 static void
 990 knote_drop(struct knote *kn, struct thread *td)
 991 {
 992         struct filedesc *fdp;
 993         struct klist *list;
 994
 995         KKASSERT(td->td_proc);
 996         fdp = td->td_proc->p_fd;
 997         if (kn->kn_fop->f_isfd)
 998                 list = &fdp->fd_knlist[kn->kn_id];
 999         else
1000                 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
1001
1002         SLIST_REMOVE(list, kn, knote, kn_link);
1003         if (kn->kn_status & KN_QUEUED)
1004                 knote_dequeue(kn);
1005         if (kn->kn_fop->f_isfd)
1006                 fdrop(kn->kn_fp);
1007         knote_free(kn);
1008 }
1009
1010
1011 static void
1012 knote_enqueue(struct knote *kn)
1013 {
1014         struct kqueue *kq = kn->kn_kq;
1015
1016         crit_enter();
1017         KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));
1018
1019         TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
1020         kn->kn_status |= KN_QUEUED;
1021         ++kq->kq_count;
1022
1023         /*
1024          * Send SIGIO on request (typically set up as a mailbox signal)
1025          */
1026         if (kq->kq_sigio && (kq->kq_state & KQ_ASYNC) && kq->kq_count == 1)
1027                 pgsigio(kq->kq_sigio, SIGIO, 0);
1028         crit_exit();
1029         kqueue_wakeup(kq);
1030 }
1031
1032 static void
1033 knote_dequeue(struct knote *kn)
1034 {
1035         struct kqueue *kq = kn->kn_kq;
1036
1037         KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));
1038         crit_enter();
1039
1040         TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
1041         kn->kn_status &= ~KN_QUEUED;
1042         kq->kq_count--;
1043         crit_exit();
1044 }
1045
1046 static void
1047 knote_init(void)
1048 {
1049         knote_zone = zinit("KNOTE", sizeof(struct knote), 0, 0, 1);
1050 }
1051 SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
1052
1053 static struct knote *
1054 knote_alloc(void)
1055 {
1056         return ((struct knote *)zalloc(knote_zone));
1057 }
1058
1059 static void
1060 knote_free(struct knote *kn)
1061 {
1062         zfree(knote_zone, kn);
1063 }