1 /* 2 * Copyright 2000-2009 Niels Provos <[email protected]> 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 3. The name of the author may not be used to endorse or promote products 13 * derived from this software without specific prior written permission. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 #include "event-config.h" 27 28 #include <sys/types.h> 29 #include <sys/resource.h> 30 #ifdef _EVENT_HAVE_SYS_TIME_H 31 #include <sys/time.h> 32 #endif 33 #include <sys/queue.h> 34 #include <sys/devpoll.h> 35 #include <signal.h> 36 #include <stdio.h> 37 #include <stdlib.h> 38 #include <string.h> 39 #include <unistd.h> 40 #include <fcntl.h> 41 #include <errno.h> 42 43 #include "event2/event.h" 44 #include "event2/event_struct.h" 45 #include "event-internal.h" 46 #include "evsignal-internal.h" 47 #include "log-internal.h" 48 #include "evmap-internal.h" 49 50 struct devpollop { 51 struct pollfd *events; 52 int nevents; 53 int dpfd; 54 struct pollfd *changes; 55 int nchanges; 56 }; 57 58 static void *devpoll_init (struct event_base *); 59 static int devpoll_add(struct event_base *, int fd, short old, short events, void *); 60 static int devpoll_del(struct event_base *, int fd, short old, short events, void *); 61 static int devpoll_dispatch (struct event_base *, struct timeval *); 62 static void devpoll_dealloc (struct event_base *); 63 64 const struct eventop devpollops = { 65 "devpoll", 66 devpoll_init, 67 devpoll_add, 68 devpoll_del, 69 devpoll_dispatch, 70 devpoll_dealloc, 71 1, /* need reinit */ 72 EV_FEATURE_FDS|EV_FEATURE_O1, 73 0 74 }; 75 76 #define NEVENT 32000 77 78 static int 79 devpoll_commit(struct devpollop *devpollop) 80 { 81 /* 82 * Due to a bug in Solaris, we have to use pwrite with an offset of 0. 83 * Write is limited to 2GB of data, until it will fail. 84 */ 85 if (pwrite(devpollop->dpfd, devpollop->changes, 86 sizeof(struct pollfd) * devpollop->nchanges, 0) == -1) 87 return(-1); 88 89 devpollop->nchanges = 0; 90 return(0); 91 } 92 93 static int 94 devpoll_queue(struct devpollop *devpollop, int fd, int events) { 95 struct pollfd *pfd; 96 97 if (devpollop->nchanges >= devpollop->nevents) { 98 /* 99 * Change buffer is full, must commit it to /dev/poll before 100 * adding more 101 */ 102 if (devpoll_commit(devpollop) != 0) 103 return(-1); 104 } 105 106 pfd = &devpollop->changes[devpollop->nchanges++]; 107 pfd->fd = fd; 108 pfd->events = events; 109 pfd->revents = 0; 110 111 return(0); 112 } 113 114 static void * 115 devpoll_init(struct event_base *base) 116 { 117 int dpfd, nfiles = NEVENT; 118 struct rlimit rl; 119 struct devpollop *devpollop; 120 121 if (!(devpollop = mm_calloc(1, sizeof(struct devpollop)))) 122 return (NULL); 123 124 if (getrlimit(RLIMIT_NOFILE, &rl) == 0 && 125 rl.rlim_cur != RLIM_INFINITY) 126 nfiles = rl.rlim_cur; 127 128 /* Initialize the kernel queue */ 129 if ((dpfd = open("/dev/poll", O_RDWR)) == -1) { 130 event_warn("open: /dev/poll"); 131 mm_free(devpollop); 132 return (NULL); 133 } 134 135 devpollop->dpfd = dpfd; 136 137 /* Initialize fields */ 138 /* FIXME: allocating 'nfiles' worth of space here can be 139 * expensive and unnecessary. See how epoll.c does it instead. */ 140 devpollop->events = mm_calloc(nfiles, sizeof(struct pollfd)); 141 if (devpollop->events == NULL) { 142 mm_free(devpollop); 143 close(dpfd); 144 return (NULL); 145 } 146 devpollop->nevents = nfiles; 147 148 devpollop->changes = mm_calloc(nfiles, sizeof(struct pollfd)); 149 if (devpollop->changes == NULL) { 150 mm_free(devpollop->events); 151 mm_free(devpollop); 152 close(dpfd); 153 return (NULL); 154 } 155 156 evsig_init(base); 157 158 return (devpollop); 159 } 160 161 static int 162 devpoll_dispatch(struct event_base *base, struct timeval *tv) 163 { 164 struct devpollop *devpollop = base->evbase; 165 struct pollfd *events = devpollop->events; 166 struct dvpoll dvp; 167 int i, res, timeout = -1; 168 169 if (devpollop->nchanges) 170 devpoll_commit(devpollop); 171 172 if (tv != NULL) 173 timeout = tv->tv_sec * 1000 + (tv->tv_usec + 999) / 1000; 174 175 dvp.dp_fds = devpollop->events; 176 dvp.dp_nfds = devpollop->nevents; 177 dvp.dp_timeout = timeout; 178 179 EVBASE_RELEASE_LOCK(base, th_base_lock); 180 181 res = ioctl(devpollop->dpfd, DP_POLL, &dvp); 182 183 EVBASE_ACQUIRE_LOCK(base, th_base_lock); 184 185 if (res == -1) { 186 if (errno != EINTR) { 187 event_warn("ioctl: DP_POLL"); 188 return (-1); 189 } 190 191 evsig_process(base); 192 return (0); 193 } else if (base->sig.evsig_caught) { 194 evsig_process(base); 195 } 196 197 event_debug(("%s: devpoll_wait reports %d", __func__, res)); 198 199 for (i = 0; i < res; i++) { 200 int which = 0; 201 int what = events[i].revents; 202 203 if (what & POLLHUP) 204 what |= POLLIN | POLLOUT; 205 else if (what & POLLERR) 206 what |= POLLIN | POLLOUT; 207 208 if (what & POLLIN) 209 which |= EV_READ; 210 if (what & POLLOUT) 211 which |= EV_WRITE; 212 213 if (!which) 214 continue; 215 216 /* XXX(niels): not sure if this works for devpoll */ 217 evmap_io_active(base, events[i].fd, which); 218 } 219 220 return (0); 221 } 222 223 224 static int 225 devpoll_add(struct event_base *base, int fd, short old, short events, void *p) 226 { 227 struct devpollop *devpollop = base->evbase; 228 int res; 229 (void)p; 230 231 /* 232 * It's not necessary to OR the existing read/write events that we 233 * are currently interested in with the new event we are adding. 234 * The /dev/poll driver ORs any new events with the existing events 235 * that it has cached for the fd. 236 */ 237 238 res = 0; 239 if (events & EV_READ) 240 res |= POLLIN; 241 if (events & EV_WRITE) 242 res |= POLLOUT; 243 244 if (devpoll_queue(devpollop, fd, res) != 0) 245 return(-1); 246 247 return (0); 248 } 249 250 static int 251 devpoll_del(struct event_base *base, int fd, short old, short events, void *p) 252 { 253 struct devpollop *devpollop = base->evbase; 254 int res; 255 (void)p; 256 257 res = 0; 258 if (events & EV_READ) 259 res |= POLLIN; 260 if (events & EV_WRITE) 261 res |= POLLOUT; 262 263 /* 264 * The only way to remove an fd from the /dev/poll monitored set is 265 * to use POLLREMOVE by itself. This removes ALL events for the fd 266 * provided so if we care about two events and are only removing one 267 * we must re-add the other event after POLLREMOVE. 268 */ 269 270 if (devpoll_queue(devpollop, fd, POLLREMOVE) != 0) 271 return(-1); 272 273 if ((res & (POLLIN|POLLOUT)) != (POLLIN|POLLOUT)) { 274 /* 275 * We're not deleting all events, so we must resubmit the 276 * event that we are still interested in if one exists. 277 */ 278 279 if ((res & POLLIN) && (old & EV_WRITE)) { 280 /* Deleting read, still care about write */ 281 devpoll_queue(devpollop, fd, POLLOUT); 282 } else if ((res & POLLOUT) && (old & EV_READ)) { 283 /* Deleting write, still care about read */ 284 devpoll_queue(devpollop, fd, POLLIN); 285 } 286 } 287 288 return (0); 289 } 290 291 static void 292 devpoll_dealloc(struct event_base *base) 293 { 294 struct devpollop *devpollop = base->evbase; 295 296 evsig_dealloc(base); 297 if (devpollop->events) 298 mm_free(devpollop->events); 299 if (devpollop->changes) 300 mm_free(devpollop->changes); 301 if (devpollop->dpfd >= 0) 302 close(devpollop->dpfd); 303 304 memset(devpollop, 0, sizeof(struct devpollop)); 305 mm_free(devpollop); 306 } 307