devpoll.c

   1 /*
   2  * Copyright 2000-2009 Niels Provos <provos@citi.umich.edu>
   3  * Copyright 2009-2012 Niels Provos and Nick Mathewson
   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  * 3. The name of the author may not be used to endorse or promote products
  14  *    derived from this software without specific prior written permission.
  15  *
  16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  26  */
  27 #include "event2/event-config.h"
  28 #include "evconfig-private.h"
  29
  30 #ifdef EVENT__HAVE_DEVPOLL
  31
  32 #include <sys/types.h>
  33 #include <sys/resource.h>
  34 #ifdef EVENT__HAVE_SYS_TIME_H
  35 #include <sys/time.h>
  36 #endif
  37 #include <sys/queue.h>
  38 #include <sys/devpoll.h>
  39 #include <signal.h>
  40 #include <stdio.h>
  41 #include <stdlib.h>
  42 #include <string.h>
  43 #include <unistd.h>
  44 #include <fcntl.h>
  45 #include <errno.h>
  46
  47 #include "event2/event.h"
  48 #include "event2/event_struct.h"
  49 #include "event2/thread.h"
  50 #include "event-internal.h"
  51 #include "evsignal-internal.h"
  52 #include "log-internal.h"
  53 #include "evmap-internal.h"
  54 #include "evthread-internal.h"
  55
  56 struct devpollop {
  57         struct pollfd *events;
  58         int nevents;
  59         int dpfd;
  60         struct pollfd *changes;
  61         int nchanges;
  62 };
  63
  64 static void *devpoll_init(struct event_base *);
  65 static int devpoll_add(struct event_base *, int fd, short old, short events, void *);
  66 static int devpoll_del(struct event_base *, int fd, short old, short events, void *);
  67 static int devpoll_dispatch(struct event_base *, struct timeval *);
  68 static void devpoll_dealloc(struct event_base *);
  69
  70 const struct eventop devpollops = {
  71         "devpoll",
  72         devpoll_init,
  73         devpoll_add,
  74         devpoll_del,
  75         devpoll_dispatch,
  76         devpoll_dealloc,
  77         1, /* need reinit */
  78         EV_FEATURE_FDS|EV_FEATURE_O1,
  79         0
  80 };
  81
  82 #define NEVENT  32000
  83
  84 static int
  85 devpoll_commit(struct devpollop *devpollop)
  86 {
  87         /*
  88          * Due to a bug in Solaris, we have to use pwrite with an offset of 0.
  89          * Write is limited to 2GB of data, until it will fail.
  90          */
  91         if (pwrite(devpollop->dpfd, devpollop->changes,
  92                 sizeof(struct pollfd) * devpollop->nchanges, 0) == -1)
  93                 return (-1);
  94
  95         devpollop->nchanges = 0;
  96         return (0);
  97 }
  98
  99 static int
 100 devpoll_queue(struct devpollop *devpollop, int fd, int events) {
 101         struct pollfd *pfd;
 102
 103         if (devpollop->nchanges >= devpollop->nevents) {
 104                 /*
 105                  * Change buffer is full, must commit it to /dev/poll before
 106                  * adding more
 107                  */
 108                 if (devpoll_commit(devpollop) != 0)
 109                         return (-1);
 110         }
 111
 112         pfd = &devpollop->changes[devpollop->nchanges++];
 113         pfd->fd = fd;
 114         pfd->events = events;
 115         pfd->revents = 0;
 116
 117         return (0);
 118 }
 119
 120 static void *
 121 devpoll_init(struct event_base *base)
 122 {
 123         int dpfd, nfiles = NEVENT;
 124         struct rlimit rl;
 125         struct devpollop *devpollop;
 126
 127         if (!(devpollop = mm_calloc(1, sizeof(struct devpollop))))
 128                 return (NULL);
 129
 130         if (getrlimit(RLIMIT_NOFILE, &rl) == 0 &&
 131             rl.rlim_cur != RLIM_INFINITY)
 132                 nfiles = rl.rlim_cur;
 133
 134         /* Initialize the kernel queue */
 135         if ((dpfd = evutil_open_closeonexec_("/dev/poll", O_RDWR, 0)) == -1) {
 136                 event_warn("open: /dev/poll");
 137                 mm_free(devpollop);
 138                 return (NULL);
 139         }
 140
 141         devpollop->dpfd = dpfd;
 142
 143         /* Initialize fields */
 144         /* FIXME: allocating 'nfiles' worth of space here can be
 145          * expensive and unnecessary.  See how epoll.c does it instead. */
 146         devpollop->events = mm_calloc(nfiles, sizeof(struct pollfd));
 147         if (devpollop->events == NULL) {
 148                 mm_free(devpollop);
 149                 close(dpfd);
 150                 return (NULL);
 151         }
 152         devpollop->nevents = nfiles;
 153
 154         devpollop->changes = mm_calloc(nfiles, sizeof(struct pollfd));
 155         if (devpollop->changes == NULL) {
 156                 mm_free(devpollop->events);
 157                 mm_free(devpollop);
 158                 close(dpfd);
 159                 return (NULL);
 160         }
 161
 162         evsig_init_(base);
 163
 164         return (devpollop);
 165 }
 166
 167 static int
 168 devpoll_dispatch(struct event_base *base, struct timeval *tv)
 169 {
 170         struct devpollop *devpollop = base->evbase;
 171         struct pollfd *events = devpollop->events;
 172         struct dvpoll dvp;
 173         int i, res, timeout = -1;
 174
 175         if (devpollop->nchanges)
 176                 devpoll_commit(devpollop);
 177
 178         if (tv != NULL)
 179                 timeout = tv->tv_sec * 1000 + (tv->tv_usec + 999) / 1000;
 180
 181         dvp.dp_fds = devpollop->events;
 182         dvp.dp_nfds = devpollop->nevents;
 183         dvp.dp_timeout = timeout;
 184
 185         EVBASE_RELEASE_LOCK(base, th_base_lock);
 186
 187         res = ioctl(devpollop->dpfd, DP_POLL, &dvp);
 188
 189         EVBASE_ACQUIRE_LOCK(base, th_base_lock);
 190
 191         if (res == -1) {
 192                 if (errno != EINTR) {
 193                         event_warn("ioctl: DP_POLL");
 194                         return (-1);
 195                 }
 196
 197                 return (0);
 198         }
 199
 200         event_debug(("%s: devpoll_wait reports %d", __func__, res));
 201
 202         for (i = 0; i < res; i++) {
 203                 int which = 0;
 204                 int what = events[i].revents;
 205
 206                 if (what & POLLHUP)
 207                         what |= POLLIN | POLLOUT;
 208                 else if (what & POLLERR)
 209                         what |= POLLIN | POLLOUT;
 210
 211                 if (what & POLLIN)
 212                         which |= EV_READ;
 213                 if (what & POLLOUT)
 214                         which |= EV_WRITE;
 215
 216                 if (!which)
 217                         continue;
 218
 219                 /* XXX(niels): not sure if this works for devpoll */
 220                 evmap_io_active_(base, events[i].fd, which);
 221         }
 222
 223         return (0);
 224 }
 225
 226
 227 static int
 228 devpoll_add(struct event_base *base, int fd, short old, short events, void *p)
 229 {
 230         struct devpollop *devpollop = base->evbase;
 231         int res;
 232         (void)p;
 233
 234         /*
 235          * It's not necessary to OR the existing read/write events that we
 236          * are currently interested in with the new event we are adding.
 237          * The /dev/poll driver ORs any new events with the existing events
 238          * that it has cached for the fd.
 239          */
 240
 241         res = 0;
 242         if (events & EV_READ)
 243                 res |= POLLIN;
 244         if (events & EV_WRITE)
 245                 res |= POLLOUT;
 246
 247         if (devpoll_queue(devpollop, fd, res) != 0)
 248                 return (-1);
 249
 250         return (0);
 251 }
 252
 253 static int
 254 devpoll_del(struct event_base *base, int fd, short old, short events, void *p)
 255 {
 256         struct devpollop *devpollop = base->evbase;
 257         int res;
 258         (void)p;
 259
 260         res = 0;
 261         if (events & EV_READ)
 262                 res |= POLLIN;
 263         if (events & EV_WRITE)
 264                 res |= POLLOUT;
 265
 266         /*
 267          * The only way to remove an fd from the /dev/poll monitored set is
 268          * to use POLLREMOVE by itself.  This removes ALL events for the fd
 269          * provided so if we care about two events and are only removing one
 270          * we must re-add the other event after POLLREMOVE.
 271          */
 272
 273         if (devpoll_queue(devpollop, fd, POLLREMOVE) != 0)
 274                 return (-1);
 275
 276         if ((res & (POLLIN|POLLOUT)) != (POLLIN|POLLOUT)) {
 277                 /*
 278                  * We're not deleting all events, so we must resubmit the
 279                  * event that we are still interested in if one exists.
 280                  */
 281
 282                 if ((res & POLLIN) && (old & EV_WRITE)) {
 283                         /* Deleting read, still care about write */
 284                         devpoll_queue(devpollop, fd, POLLOUT);
 285                 } else if ((res & POLLOUT) && (old & EV_READ)) {
 286                         /* Deleting write, still care about read */
 287                         devpoll_queue(devpollop, fd, POLLIN);
 288                 }
 289         }
 290
 291         return (0);
 292 }
 293
 294 static void
 295 devpoll_dealloc(struct event_base *base)
 296 {
 297         struct devpollop *devpollop = base->evbase;
 298
 299         evsig_dealloc_(base);
 300         if (devpollop->events)
 301                 mm_free(devpollop->events);
 302         if (devpollop->changes)
 303                 mm_free(devpollop->changes);
 304         if (devpollop->dpfd >= 0)
 305                 close(devpollop->dpfd);
 306
 307         memset(devpollop, 0, sizeof(struct devpollop));
 308         mm_free(devpollop);
 309 }
 310
 311 #endif /* EVENT__HAVE_DEVPOLL */