1 /* $NetBSD: devpoll.c,v 1.5 2020/05/25 20:47:33 christos Exp $ */ 2 3 /* 4 * Copyright 2000-2009 Niels Provos <provos@citi.umich.edu> 5 * Copyright 2009-2012 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 #include "evconfig-private.h" 31 32 #ifdef EVENT__HAVE_DEVPOLL 33 34 #include <sys/types.h> 35 #include <sys/resource.h> 36 #ifdef EVENT__HAVE_SYS_TIME_H 37 #include <sys/time.h> 38 #endif 39 #include <sys/queue.h> 40 #include <sys/devpoll.h> 41 #include <signal.h> 42 #include <stdio.h> 43 #include <stdlib.h> 44 #include <string.h> 45 #include <unistd.h> 46 #include <fcntl.h> 47 #include <errno.h> 48 49 #include "event2/event.h" 50 #include "event2/event_struct.h" 51 #include "event2/thread.h" 52 #include "event-internal.h" 53 #include "evsignal-internal.h" 54 #include "log-internal.h" 55 #include "evmap-internal.h" 56 #include "evthread-internal.h" 57 58 struct devpollop { 59 struct pollfd *events; 60 int nevents; 61 int dpfd; 62 struct pollfd *changes; 63 int nchanges; 64 }; 65 66 static void *devpoll_init(struct event_base *); 67 static int devpoll_add(struct event_base *, int fd, short old, short events, void *); 68 static int devpoll_del(struct event_base *, int fd, short old, short events, void *); 69 static int devpoll_dispatch(struct event_base *, struct timeval *); 70 static void devpoll_dealloc(struct event_base *); 71 72 const struct eventop devpollops = { 73 "devpoll", 74 devpoll_init, 75 devpoll_add, 76 devpoll_del, 77 devpoll_dispatch, 78 devpoll_dealloc, 79 1, /* need reinit */ 80 EV_FEATURE_FDS|EV_FEATURE_O1, 81 0 82 }; 83 84 #define NEVENT 32000 85 86 static int 87 devpoll_commit(struct devpollop *devpollop) 88 { 89 /* 90 * Due to a bug in Solaris, we have to use pwrite with an offset of 0. 91 * Write is limited to 2GB of data, until it will fail. 92 */ 93 if (pwrite(devpollop->dpfd, devpollop->changes, 94 sizeof(struct pollfd) * devpollop->nchanges, 0) == -1) 95 return (-1); 96 97 devpollop->nchanges = 0; 98 return (0); 99 } 100 101 static int 102 devpoll_queue(struct devpollop *devpollop, int fd, int events) { 103 struct pollfd *pfd; 104 105 if (devpollop->nchanges >= devpollop->nevents) { 106 /* 107 * Change buffer is full, must commit it to /dev/poll before 108 * adding more 109 */ 110 if (devpoll_commit(devpollop) != 0) 111 return (-1); 112 } 113 114 pfd = &devpollop->changes[devpollop->nchanges++]; 115 pfd->fd = fd; 116 pfd->events = events; 117 pfd->revents = 0; 118 119 return (0); 120 } 121 122 static void * 123 devpoll_init(struct event_base *base) 124 { 125 int dpfd, nfiles = NEVENT; 126 struct rlimit rl; 127 struct devpollop *devpollop; 128 129 if (!(devpollop = mm_calloc(1, sizeof(struct devpollop)))) 130 return (NULL); 131 132 if (getrlimit(RLIMIT_NOFILE, &rl) == 0 && 133 rl.rlim_cur != RLIM_INFINITY) 134 nfiles = rl.rlim_cur; 135 136 /* Initialize the kernel queue */ 137 if ((dpfd = evutil_open_closeonexec_("/dev/poll", O_RDWR, 0)) == -1) { 138 event_warn("open: /dev/poll"); 139 mm_free(devpollop); 140 return (NULL); 141 } 142 143 devpollop->dpfd = dpfd; 144 145 /* Initialize fields */ 146 /* FIXME: allocating 'nfiles' worth of space here can be 147 * expensive and unnecessary. See how epoll.c does it instead. */ 148 devpollop->events = mm_calloc(nfiles, sizeof(struct pollfd)); 149 if (devpollop->events == NULL) { 150 mm_free(devpollop); 151 close(dpfd); 152 return (NULL); 153 } 154 devpollop->nevents = nfiles; 155 156 devpollop->changes = mm_calloc(nfiles, sizeof(struct pollfd)); 157 if (devpollop->changes == NULL) { 158 mm_free(devpollop->events); 159 mm_free(devpollop); 160 close(dpfd); 161 return (NULL); 162 } 163 164 evsig_init_(base); 165 166 return (devpollop); 167 } 168 169 static int 170 devpoll_dispatch(struct event_base *base, struct timeval *tv) 171 { 172 struct devpollop *devpollop = base->evbase; 173 struct pollfd *events = devpollop->events; 174 struct dvpoll dvp; 175 int i, res, timeout = -1; 176 177 if (devpollop->nchanges) 178 devpoll_commit(devpollop); 179 180 if (tv != NULL) 181 timeout = tv->tv_sec * 1000 + (tv->tv_usec + 999) / 1000; 182 183 dvp.dp_fds = devpollop->events; 184 dvp.dp_nfds = devpollop->nevents; 185 dvp.dp_timeout = timeout; 186 187 EVBASE_RELEASE_LOCK(base, th_base_lock); 188 189 res = ioctl(devpollop->dpfd, DP_POLL, &dvp); 190 191 EVBASE_ACQUIRE_LOCK(base, th_base_lock); 192 193 if (res == -1) { 194 if (errno != EINTR) { 195 event_warn("ioctl: DP_POLL"); 196 return (-1); 197 } 198 199 return (0); 200 } 201 202 event_debug(("%s: devpoll_wait reports %d", __func__, res)); 203 204 for (i = 0; i < res; i++) { 205 int which = 0; 206 int what = events[i].revents; 207 208 if (what & POLLHUP) 209 what |= POLLIN | POLLOUT; 210 else if (what & POLLERR) 211 what |= POLLIN | POLLOUT; 212 213 if (what & POLLIN) 214 which |= EV_READ; 215 if (what & POLLOUT) 216 which |= EV_WRITE; 217 218 if (!which) 219 continue; 220 221 /* XXX(niels): not sure if this works for devpoll */ 222 evmap_io_active_(base, events[i].fd, which); 223 } 224 225 return (0); 226 } 227 228 229 static int 230 devpoll_add(struct event_base *base, int fd, short old, short events, void *p) 231 { 232 struct devpollop *devpollop = base->evbase; 233 int res; 234 (void)p; 235 236 /* 237 * It's not necessary to OR the existing read/write events that we 238 * are currently interested in with the new event we are adding. 239 * The /dev/poll driver ORs any new events with the existing events 240 * that it has cached for the fd. 241 */ 242 243 res = 0; 244 if (events & EV_READ) 245 res |= POLLIN; 246 if (events & EV_WRITE) 247 res |= POLLOUT; 248 249 if (devpoll_queue(devpollop, fd, res) != 0) 250 return (-1); 251 252 return (0); 253 } 254 255 static int 256 devpoll_del(struct event_base *base, int fd, short old, short events, void *p) 257 { 258 struct devpollop *devpollop = base->evbase; 259 int res; 260 (void)p; 261 262 res = 0; 263 if (events & EV_READ) 264 res |= POLLIN; 265 if (events & EV_WRITE) 266 res |= POLLOUT; 267 268 /* 269 * The only way to remove an fd from the /dev/poll monitored set is 270 * to use POLLREMOVE by itself. This removes ALL events for the fd 271 * provided so if we care about two events and are only removing one 272 * we must re-add the other event after POLLREMOVE. 273 */ 274 275 if (devpoll_queue(devpollop, fd, POLLREMOVE) != 0) 276 return (-1); 277 278 if ((res & (POLLIN|POLLOUT)) != (POLLIN|POLLOUT)) { 279 /* 280 * We're not deleting all events, so we must resubmit the 281 * event that we are still interested in if one exists. 282 */ 283 284 if ((res & POLLIN) && (old & EV_WRITE)) { 285 /* Deleting read, still care about write */ 286 devpoll_queue(devpollop, fd, POLLOUT); 287 } else if ((res & POLLOUT) && (old & EV_READ)) { 288 /* Deleting write, still care about read */ 289 devpoll_queue(devpollop, fd, POLLIN); 290 } 291 } 292 293 return (0); 294 } 295 296 static void 297 devpoll_dealloc(struct event_base *base) 298 { 299 struct devpollop *devpollop = base->evbase; 300 301 evsig_dealloc_(base); 302 if (devpollop->events) 303 mm_free(devpollop->events); 304 if (devpollop->changes) 305 mm_free(devpollop->changes); 306 if (devpollop->dpfd >= 0) 307 close(devpollop->dpfd); 308 309 memset(devpollop, 0, sizeof(struct devpollop)); 310 mm_free(devpollop); 311 } 312 313 #endif /* EVENT__HAVE_DEVPOLL */ 314