release/src/router/dnscrypt/src/libevent/kqueue.c

   1 /*      $OpenBSD: kqueue.c,v 1.5 2002/07/10 14:41:31 art Exp $  */
   2
   3 /*
   4  * Copyright 2000-2007 Niels Provos <provos@citi.umich.edu>
   5  * Copyright 2007-2012 Niels Provos and Nick Mathewson
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. The name of the author may not be used to endorse or promote products
  16  *    derived from this software without specific prior written permission.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28  */
  29 #include "event2/event-config.h"
  30
  31 #ifndef _GNU_SOURCE
  32 # define _GNU_SOURCE
  33 #endif
  34
  35 #include <sys/types.h>
  36 #ifdef _EVENT_HAVE_SYS_TIME_H
  37 #include <sys/time.h>
  38 #endif
  39 #include <sys/queue.h>
  40 #include <sys/event.h>
  41 #include <signal.h>
  42 #include <stdio.h>
  43 #include <stdlib.h>
  44 #include <string.h>
  45 #include <unistd.h>
  46 #include <errno.h>
  47 #ifdef _EVENT_HAVE_INTTYPES_H
  48 #include <inttypes.h>
  49 #endif
  50
  51 /* Some platforms apparently define the udata field of struct kevent as
  52  * intptr_t, whereas others define it as void*.  There doesn't seem to be an
  53  * easy way to tell them apart via autoconf, so we need to use OS macros. */
  54 #if defined(_EVENT_HAVE_INTTYPES_H) && !defined(__OpenBSD__) && !defined(__FreeBSD__) && !defined(__darwin__) && !defined(__APPLE__)
  55 #define PTR_TO_UDATA(x) ((intptr_t)(x))
  56 #define INT_TO_UDATA(x) ((intptr_t)(x))
  57 #else
  58 #define PTR_TO_UDATA(x) (x)
  59 #define INT_TO_UDATA(x) ((void*)(x))
  60 #endif
  61
  62 #include "event-internal.h"
  63 #include "log-internal.h"
  64 #include "evmap-internal.h"
  65 #include "event2/thread.h"
  66 #include "evthread-internal.h"
  67 #include "changelist-internal.h"
  68
  69 #define NEVENT          64
  70
  71 struct kqop {
  72         struct kevent *changes;
  73         int changes_size;
  74
  75         struct kevent *events;
  76         int events_size;
  77         int kq;
  78         pid_t pid;
  79 };
  80
  81 static void kqop_free(struct kqop *kqop);
  82
  83 static void *kq_init(struct event_base *);
  84 static int kq_sig_add(struct event_base *, int, short, short, void *);
  85 static int kq_sig_del(struct event_base *, int, short, short, void *);
  86 static int kq_dispatch(struct event_base *, struct timeval *);
  87 static void kq_dealloc(struct event_base *);
  88
  89 const struct eventop kqops = {
  90         "kqueue",
  91         kq_init,
  92         event_changelist_add,
  93         event_changelist_del,
  94         kq_dispatch,
  95         kq_dealloc,
  96         1 /* need reinit */,
  97     EV_FEATURE_ET|EV_FEATURE_O1|EV_FEATURE_FDS,
  98         EVENT_CHANGELIST_FDINFO_SIZE
  99 };
 100
 101 static const struct eventop kqsigops = {
 102         "kqueue_signal",
 103         NULL,
 104         kq_sig_add,
 105         kq_sig_del,
 106         NULL,
 107         NULL,
 108         1 /* need reinit */,
 109         0,
 110         0
 111 };
 112
 113 static void *
 114 kq_init(struct event_base *base)
 115 {
 116         int kq = -1;
 117         struct kqop *kqueueop = NULL;
 118
 119         if (!(kqueueop = mm_calloc(1, sizeof(struct kqop))))
 120                 return (NULL);
 121
 122 /* Initialize the kernel queue */
 123
 124         if ((kq = kqueue()) == -1) {
 125                 event_warn("kqueue");
 126                 goto err;
 127         }
 128
 129         kqueueop->kq = kq;
 130
 131         kqueueop->pid = getpid();
 132
 133         /* Initialize fields */
 134         kqueueop->changes = mm_calloc(NEVENT, sizeof(struct kevent));
 135         if (kqueueop->changes == NULL)
 136                 goto err;
 137         kqueueop->events = mm_calloc(NEVENT, sizeof(struct kevent));
 138         if (kqueueop->events == NULL)
 139                 goto err;
 140         kqueueop->events_size = kqueueop->changes_size = NEVENT;
 141
 142         /* Check for Mac OS X kqueue bug. */
 143         memset(&kqueueop->changes[0], 0, sizeof kqueueop->changes[0]);
 144         kqueueop->changes[0].ident = -1;
 145         kqueueop->changes[0].filter = EVFILT_READ;
 146         kqueueop->changes[0].flags = EV_ADD;
 147         /*
 148          * If kqueue works, then kevent will succeed, and it will
 149          * stick an error in events[0].  If kqueue is broken, then
 150          * kevent will fail.
 151          */
 152         if (kevent(kq,
 153                 kqueueop->changes, 1, kqueueop->events, NEVENT, NULL) != 1 ||
 154             (int)kqueueop->events[0].ident != -1 ||
 155             kqueueop->events[0].flags != EV_ERROR) {
 156                 event_warn("%s: detected broken kqueue; not using.", __func__);
 157                 goto err;
 158         }
 159
 160         base->evsigsel = &kqsigops;
 161
 162         return (kqueueop);
 163 err:
 164         if (kqueueop)
 165                 kqop_free(kqueueop);
 166
 167         return (NULL);
 168 }
 169
 170 static void
 171 kq_sighandler(int sig)
 172 {
 173         /* Do nothing here */
 174 }
 175
 176 #define ADD_UDATA 0x30303
 177
 178 static void
 179 kq_setup_kevent(struct kevent *out, evutil_socket_t fd, int filter, short change)
 180 {
 181         memset(out, 0, sizeof(struct kevent));
 182         out->ident = fd;
 183         out->filter = filter;
 184
 185         if (change & EV_CHANGE_ADD) {
 186                 out->flags = EV_ADD;
 187                 /* We set a magic number here so that we can tell 'add'
 188                  * errors from 'del' errors. */
 189                 out->udata = INT_TO_UDATA(ADD_UDATA);
 190                 if (change & EV_ET)
 191                         out->flags |= EV_CLEAR;
 192 #ifdef NOTE_EOF
 193                 /* Make it behave like select() and poll() */
 194                 if (filter == EVFILT_READ)
 195                         out->fflags = NOTE_EOF;
 196 #endif
 197         } else {
 198                 EVUTIL_ASSERT(change & EV_CHANGE_DEL);
 199                 out->flags = EV_DELETE;
 200         }
 201 }
 202
 203 static int
 204 kq_build_changes_list(const struct event_changelist *changelist,
 205     struct kqop *kqop)
 206 {
 207         int i;
 208         int n_changes = 0;
 209
 210         for (i = 0; i < changelist->n_changes; ++i) {
 211                 struct event_change *in_ch = &changelist->changes[i];
 212                 struct kevent *out_ch;
 213                 if (n_changes >= kqop->changes_size - 1) {
 214                         int newsize = kqop->changes_size * 2;
 215                         struct kevent *newchanges;
 216
 217                         newchanges = mm_realloc(kqop->changes,
 218                             newsize * sizeof(struct kevent));
 219                         if (newchanges == NULL) {
 220                                 event_warn("%s: realloc", __func__);
 221                                 return (-1);
 222                         }
 223                         kqop->changes = newchanges;
 224                         kqop->changes_size = newsize;
 225                 }
 226                 if (in_ch->read_change) {
 227                         out_ch = &kqop->changes[n_changes++];
 228                         kq_setup_kevent(out_ch, in_ch->fd, EVFILT_READ,
 229                             in_ch->read_change);
 230                 }
 231                 if (in_ch->write_change) {
 232                         out_ch = &kqop->changes[n_changes++];
 233                         kq_setup_kevent(out_ch, in_ch->fd, EVFILT_WRITE,
 234                             in_ch->write_change);
 235                 }
 236         }
 237         return n_changes;
 238 }
 239
 240 static int
 241 kq_grow_events(struct kqop *kqop, size_t new_size)
 242 {
 243         struct kevent *newresult;
 244
 245         newresult = mm_realloc(kqop->events,
 246             new_size * sizeof(struct kevent));
 247
 248         if (newresult) {
 249                 kqop->events = newresult;
 250                 kqop->events_size = new_size;
 251                 return 0;
 252         } else {
 253                 return -1;
 254         }
 255 }
 256
 257 static int
 258 kq_dispatch(struct event_base *base, struct timeval *tv)
 259 {
 260         struct kqop *kqop = base->evbase;
 261         struct kevent *events = kqop->events;
 262         struct kevent *changes;
 263         struct timespec ts, *ts_p = NULL;
 264         int i, n_changes, res;
 265
 266         if (tv != NULL) {
 267                 TIMEVAL_TO_TIMESPEC(tv, &ts);
 268                 ts_p = &ts;
 269         }
 270
 271         /* Build "changes" from "base->changes" */
 272         EVUTIL_ASSERT(kqop->changes);
 273         n_changes = kq_build_changes_list(&base->changelist, kqop);
 274         if (n_changes < 0)
 275                 return -1;
 276
 277         event_changelist_remove_all(&base->changelist, base);
 278
 279         /* steal the changes array in case some broken code tries to call
 280          * dispatch twice at once. */
 281         changes = kqop->changes;
 282         kqop->changes = NULL;
 283
 284         /* Make sure that 'events' is at least as long as the list of changes:
 285          * otherwise errors in the changes can get reported as a -1 return
 286          * value from kevent() rather than as EV_ERROR events in the events
 287          * array.
 288          *
 289          * (We could instead handle -1 return values from kevent() by
 290          * retrying with a smaller changes array or a larger events array,
 291          * but this approach seems less risky for now.)
 292          */
 293         if (kqop->events_size < n_changes) {
 294                 int new_size = kqop->events_size;
 295                 do {
 296                         new_size *= 2;
 297                 } while (new_size < n_changes);
 298
 299                 kq_grow_events(kqop, new_size);
 300                 events = kqop->events;
 301         }
 302
 303         EVBASE_RELEASE_LOCK(base, th_base_lock);
 304
 305         res = kevent(kqop->kq, changes, n_changes,
 306             events, kqop->events_size, ts_p);
 307
 308         EVBASE_ACQUIRE_LOCK(base, th_base_lock);
 309
 310         EVUTIL_ASSERT(kqop->changes == NULL);
 311         kqop->changes = changes;
 312
 313         if (res == -1) {
 314                 if (errno != EINTR) {
 315                         event_warn("kevent");
 316                         return (-1);
 317                 }
 318
 319                 return (0);
 320         }
 321
 322         event_debug(("%s: kevent reports %d", __func__, res));
 323
 324         for (i = 0; i < res; i++) {
 325                 int which = 0;
 326
 327                 if (events[i].flags & EV_ERROR) {
 328                         switch (events[i].data) {
 329
 330                         /* Can occur on delete if we are not currently
 331                          * watching any events on this fd.  That can
 332                          * happen when the fd was closed and another
 333                          * file was opened with that fd. */
 334                         case ENOENT:
 335                         /* Can occur for reasons not fully understood
 336                          * on FreeBSD. */
 337                         case EINVAL:
 338                                 continue;
 339
 340                         /* Can occur on a delete if the fd is closed. */
 341                         case EBADF:
 342                                 /* XXXX On NetBSD, we can also get EBADF if we
 343                                  * try to add the write side of a pipe, but
 344                                  * the read side has already been closed.
 345                                  * Other BSDs call this situation 'EPIPE'. It
 346                                  * would be good if we had a way to report
 347                                  * this situation. */
 348                                 continue;
 349                         /* These two can occur on an add if the fd was one side
 350                          * of a pipe, and the other side was closed. */
 351                         case EPERM:
 352                         case EPIPE:
 353                                 /* Report read events, if we're listening for
 354                                  * them, so that the user can learn about any
 355                                  * add errors.  (If the operation was a
 356                                  * delete, then udata should be cleared.) */
 357                                 if (events[i].udata) {
 358                                         /* The operation was an add:
 359                                          * report the error as a read. */
 360                                         which |= EV_READ;
 361                                         break;
 362                                 } else {
 363                                         /* The operation was a del:
 364                                          * report nothing. */
 365                                         continue;
 366                                 }
 367
 368                         /* Other errors shouldn't occur. */
 369                         default:
 370                                 errno = events[i].data;
 371                                 return (-1);
 372                         }
 373                 } else if (events[i].filter == EVFILT_READ) {
 374                         which |= EV_READ;
 375                 } else if (events[i].filter == EVFILT_WRITE) {
 376                         which |= EV_WRITE;
 377                 } else if (events[i].filter == EVFILT_SIGNAL) {
 378                         which |= EV_SIGNAL;
 379                 }
 380
 381                 if (!which)
 382                         continue;
 383
 384                 if (events[i].filter == EVFILT_SIGNAL) {
 385                         evmap_signal_active(base, events[i].ident, 1);
 386                 } else {
 387                         evmap_io_active(base, events[i].ident, which | EV_ET);
 388                 }
 389         }
 390
 391         if (res == kqop->events_size) {
 392                 /* We used all the events space that we have. Maybe we should
 393                    make it bigger. */
 394                 kq_grow_events(kqop, kqop->events_size * 2);
 395         }
 396
 397         return (0);
 398 }
 399
 400 static void
 401 kqop_free(struct kqop *kqop)
 402 {
 403         if (kqop->changes)
 404                 mm_free(kqop->changes);
 405         if (kqop->events)
 406                 mm_free(kqop->events);
 407         if (kqop->kq >= 0 && kqop->pid == getpid())
 408                 close(kqop->kq);
 409         memset(kqop, 0, sizeof(struct kqop));
 410         mm_free(kqop);
 411 }
 412
 413 static void
 414 kq_dealloc(struct event_base *base)
 415 {
 416         struct kqop *kqop = base->evbase;
 417         evsig_dealloc(base);
 418         kqop_free(kqop);
 419 }
 420
 421 /* signal handling */
 422 static int
 423 kq_sig_add(struct event_base *base, int nsignal, short old, short events, void *p)
 424 {
 425         struct kqop *kqop = base->evbase;
 426         struct kevent kev;
 427         struct timespec timeout = { 0, 0 };
 428         (void)p;
 429
 430         EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG);
 431
 432         memset(&kev, 0, sizeof(kev));
 433         kev.ident = nsignal;
 434         kev.filter = EVFILT_SIGNAL;
 435         kev.flags = EV_ADD;
 436
 437         /* Be ready for the signal if it is sent any
 438          * time between now and the next call to
 439          * kq_dispatch. */
 440         if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1)
 441                 return (-1);
 442
 443         /* XXXX The manpage suggest we could use SIG_IGN instead of a
 444          * do-nothing handler */
 445         if (_evsig_set_handler(base, nsignal, kq_sighandler) == -1)
 446                 return (-1);
 447
 448         return (0);
 449 }
 450
 451 static int
 452 kq_sig_del(struct event_base *base, int nsignal, short old, short events, void *p)
 453 {
 454         struct kqop *kqop = base->evbase;
 455         struct kevent kev;
 456
 457         struct timespec timeout = { 0, 0 };
 458         (void)p;
 459
 460         EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG);
 461
 462         memset(&kev, 0, sizeof(kev));
 463         kev.ident = nsignal;
 464         kev.filter = EVFILT_SIGNAL;
 465         kev.flags = EV_DELETE;
 466
 467         /* Because we insert signal events
 468          * immediately, we need to delete them
 469          * immediately, too */
 470         if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1)
 471                 return (-1);
 472
 473         if (_evsig_restore_handler(base, nsignal) == -1)
 474                 return (-1);
 475
 476         return (0);
 477 }