lib/dpif-linux.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010 Nicira Networks.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18 #include "dpif.h"
  19
  20 #include <assert.h>
  21 #include <ctype.h>
  22 #include <errno.h>
  23 #include <fcntl.h>
  24 #include <inttypes.h>
  25 #include <net/if.h>
  26 #include <linux/types.h>
  27 #include <linux/ethtool.h>
  28 #include <linux/pkt_sched.h>
  29 #include <linux/rtnetlink.h>
  30 #include <linux/sockios.h>
  31 #include <stdlib.h>
  32 #include <sys/ioctl.h>
  33 #include <sys/stat.h>
  34 #include <unistd.h>
  35
  36 #include "dpif-provider.h"
  37 #include "netdev.h"
  38 #include "ofpbuf.h"
  39 #include "poll-loop.h"
  40 #include "rtnetlink.h"
  41 #include "shash.h"
  42 #include "svec.h"
  43 #include "util.h"
  44 #include "vlog.h"
  45
  46 VLOG_DEFINE_THIS_MODULE(dpif_linux)
  47
  48 /* Datapath interface for the openvswitch Linux kernel module. */
  49 struct dpif_linux {
  50     struct dpif dpif;
  51     int fd;
  52
  53     /* Used by dpif_linux_get_all_names(). */
  54     char *local_ifname;
  55     int minor;
  56
  57     /* Change notification. */
  58     int local_ifindex;          /* Ifindex of local port. */
  59     struct shash changed_ports;  /* Ports that have changed. */
  60     struct rtnetlink_notifier port_notifier;
  61     bool change_error;
  62 };
  63
  64 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
  65
  66 static int do_ioctl(const struct dpif *, int cmd, const void *arg);
  67 static int lookup_minor(const char *name, int *minor);
  68 static int finish_open(struct dpif *, const char *local_ifname);
  69 static int get_openvswitch_major(void);
  70 static int create_minor(const char *name, int minor, struct dpif **dpifp);
  71 static int open_minor(int minor, struct dpif **dpifp);
  72 static int make_openvswitch_device(int minor, char **fnp);
  73 static void dpif_linux_port_changed(const struct rtnetlink_change *,
  74                                     void *dpif);
  75
  76 static struct dpif_linux *
  77 dpif_linux_cast(const struct dpif *dpif)
  78 {
  79     dpif_assert_class(dpif, &dpif_linux_class);
  80     return CONTAINER_OF(dpif, struct dpif_linux, dpif);
  81 }
  82
  83 static int
  84 dpif_linux_enumerate(struct svec *all_dps)
  85 {
  86     int major;
  87     int error;
  88     int i;
  89
  90     /* Check that the Open vSwitch module is loaded. */
  91     major = get_openvswitch_major();
  92     if (major < 0) {
  93         return -major;
  94     }
  95
  96     error = 0;
  97     for (i = 0; i < ODP_MAX; i++) {
  98         struct dpif *dpif;
  99         char devname[16];
 100         int retval;
 101
 102         sprintf(devname, "dp%d", i);
 103         retval = dpif_open(devname, "system", &dpif);
 104         if (!retval) {
 105             svec_add(all_dps, devname);
 106             dpif_uninit(dpif, true);
 107         } else if (retval != ENODEV && !error) {
 108             error = retval;
 109         }
 110     }
 111     return error;
 112 }
 113
 114 static int
 115 dpif_linux_open(const char *name, const char *type OVS_UNUSED, bool create,
 116                 struct dpif **dpifp)
 117 {
 118     int minor;
 119
 120     minor = !strncmp(name, "dp", 2)
 121             && isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1;
 122     if (create) {
 123         if (minor >= 0) {
 124             return create_minor(name, minor, dpifp);
 125         } else {
 126             /* Scan for unused minor number. */
 127             for (minor = 0; minor < ODP_MAX; minor++) {
 128                 int error = create_minor(name, minor, dpifp);
 129                 if (error != EBUSY) {
 130                     return error;
 131                 }
 132             }
 133
 134             /* All datapath numbers in use. */
 135             return ENOBUFS;
 136         }
 137     } else {
 138         struct dpif_linux *dpif;
 139         struct odp_port port;
 140         int error;
 141
 142         if (minor < 0) {
 143             error = lookup_minor(name, &minor);
 144             if (error) {
 145                 return error;
 146             }
 147         }
 148
 149         error = open_minor(minor, dpifp);
 150         if (error) {
 151             return error;
 152         }
 153         dpif = dpif_linux_cast(*dpifp);
 154
 155         /* We need the local port's ifindex for the poll function.  Start by
 156          * getting the local port's name. */
 157         memset(&port, 0, sizeof port);
 158         port.port = ODPP_LOCAL;
 159         if (ioctl(dpif->fd, ODP_PORT_QUERY, &port)) {
 160             error = errno;
 161             if (error != ENODEV) {
 162                 VLOG_WARN("%s: probe returned unexpected error: %s",
 163                           dpif_name(*dpifp), strerror(error));
 164             }
 165             dpif_uninit(*dpifp, true);
 166             return error;
 167         }
 168
 169         /* Then use that to finish up opening. */
 170         return finish_open(&dpif->dpif, port.devname);
 171     }
 172 }
 173
 174 static void
 175 dpif_linux_close(struct dpif *dpif_)
 176 {
 177     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 178     rtnetlink_notifier_unregister(&dpif->port_notifier);
 179     shash_destroy(&dpif->changed_ports);
 180     free(dpif->local_ifname);
 181     close(dpif->fd);
 182     free(dpif);
 183 }
 184
 185 static int
 186 dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names)
 187 {
 188     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 189
 190     svec_add_nocopy(all_names, xasprintf("dp%d", dpif->minor));
 191     svec_add(all_names, dpif->local_ifname);
 192     return 0;
 193 }
 194
 195 static int
 196 dpif_linux_destroy(struct dpif *dpif_)
 197 {
 198     struct odp_port *ports;
 199     size_t n_ports;
 200     int err;
 201     int i;
 202
 203     err = dpif_port_list(dpif_, &ports, &n_ports);
 204     if (err) {
 205         return err;
 206     }
 207
 208     for (i = 0; i < n_ports; i++) {
 209         if (ports[i].port != ODPP_LOCAL) {
 210             err = do_ioctl(dpif_, ODP_VPORT_DEL, ports[i].devname);
 211             if (err) {
 212                 VLOG_WARN_RL(&error_rl, "%s: error deleting port %s (%s)",
 213                              dpif_name(dpif_), ports[i].devname, strerror(err));
 214             }
 215         }
 216     }
 217
 218     free(ports);
 219
 220     return do_ioctl(dpif_, ODP_DP_DESTROY, NULL);
 221 }
 222
 223 static int
 224 dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
 225 {
 226     memset(stats, 0, sizeof *stats);
 227     return do_ioctl(dpif_, ODP_DP_STATS, stats);
 228 }
 229
 230 static int
 231 dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
 232 {
 233     int drop_frags;
 234     int error;
 235
 236     error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags);
 237     if (!error) {
 238         *drop_fragsp = drop_frags & 1;
 239     }
 240     return error;
 241 }
 242
 243 static int
 244 dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
 245 {
 246     int drop_frags_int = drop_frags;
 247     return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int);
 248 }
 249
 250 static int
 251 dpif_linux_port_add(struct dpif *dpif_, const char *devname, uint16_t flags,
 252                     uint16_t *port_no)
 253 {
 254     struct odp_port port;
 255     int error;
 256
 257     memset(&port, 0, sizeof port);
 258     strncpy(port.devname, devname, sizeof port.devname);
 259     port.flags = flags;
 260     error = do_ioctl(dpif_, ODP_PORT_ATTACH, &port);
 261     if (!error) {
 262         *port_no = port.port;
 263     }
 264     return error;
 265 }
 266
 267 static int
 268 dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
 269 {
 270     int tmp = port_no;
 271     int err;
 272     struct odp_port port;
 273
 274     err = dpif_port_query_by_number(dpif_, port_no, &port);
 275     if (err) {
 276         return err;
 277     }
 278
 279     err = do_ioctl(dpif_, ODP_PORT_DETACH, &tmp);
 280     if (err) {
 281         return err;
 282     }
 283
 284     if (!netdev_is_open(port.devname)) {
 285         /* Try deleting the port if no one has it open.  This shouldn't
 286          * actually be necessary unless the config changed while we weren't
 287          * running but it won't hurt anything if the port is already gone. */
 288         do_ioctl(dpif_, ODP_VPORT_DEL, port.devname);
 289     }
 290
 291     return 0;
 292 }
 293
 294 static int
 295 dpif_linux_port_query_by_number(const struct dpif *dpif_, uint16_t port_no,
 296                           struct odp_port *port)
 297 {
 298     memset(port, 0, sizeof *port);
 299     port->port = port_no;
 300     return do_ioctl(dpif_, ODP_PORT_QUERY, port);
 301 }
 302
 303 static int
 304 dpif_linux_port_query_by_name(const struct dpif *dpif_, const char *devname,
 305                               struct odp_port *port)
 306 {
 307     memset(port, 0, sizeof *port);
 308     strncpy(port->devname, devname, sizeof port->devname);
 309     return do_ioctl(dpif_, ODP_PORT_QUERY, port);
 310 }
 311
 312 static int
 313 dpif_linux_flow_flush(struct dpif *dpif_)
 314 {
 315     return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL);
 316 }
 317
 318 static int
 319 dpif_linux_port_list(const struct dpif *dpif_, struct odp_port *ports, int n)
 320 {
 321     struct odp_portvec pv;
 322     int error;
 323
 324     pv.ports = ports;
 325     pv.n_ports = n;
 326     error = do_ioctl(dpif_, ODP_PORT_LIST, &pv);
 327     return error ? -error : pv.n_ports;
 328 }
 329
 330 static int
 331 dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep)
 332 {
 333     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 334
 335     if (dpif->change_error) {
 336         dpif->change_error = false;
 337         shash_clear(&dpif->changed_ports);
 338         return ENOBUFS;
 339     } else if (!shash_is_empty(&dpif->changed_ports)) {
 340         struct shash_node *node = shash_first(&dpif->changed_ports);
 341         *devnamep = shash_steal(&dpif->changed_ports, node);
 342         return 0;
 343     } else {
 344         return EAGAIN;
 345     }
 346 }
 347
 348 static void
 349 dpif_linux_port_poll_wait(const struct dpif *dpif_)
 350 {
 351     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 352     if (!shash_is_empty(&dpif->changed_ports) || dpif->change_error) {
 353         poll_immediate_wake();
 354     } else {
 355         rtnetlink_notifier_wait();
 356     }
 357 }
 358
 359 static int
 360 dpif_linux_port_group_get(const struct dpif *dpif_, int group,
 361                           uint16_t ports[], int n)
 362 {
 363     struct odp_port_group pg;
 364     int error;
 365
 366     assert(n <= UINT16_MAX);
 367     pg.group = group;
 368     pg.ports = ports;
 369     pg.n_ports = n;
 370     error = do_ioctl(dpif_, ODP_PORT_GROUP_GET, &pg);
 371     return error ? -error : pg.n_ports;
 372 }
 373
 374 static int
 375 dpif_linux_port_group_set(struct dpif *dpif_, int group,
 376                           const uint16_t ports[], int n)
 377 {
 378     struct odp_port_group pg;
 379
 380     assert(n <= UINT16_MAX);
 381     pg.group = group;
 382     pg.ports = (uint16_t *) ports;
 383     pg.n_ports = n;
 384     return do_ioctl(dpif_, ODP_PORT_GROUP_SET, &pg);
 385 }
 386
 387 static int
 388 dpif_linux_flow_get(const struct dpif *dpif_, struct odp_flow flows[], int n)
 389 {
 390     struct odp_flowvec fv;
 391     fv.flows = flows;
 392     fv.n_flows = n;
 393     return do_ioctl(dpif_, ODP_FLOW_GET, &fv);
 394 }
 395
 396 static int
 397 dpif_linux_flow_put(struct dpif *dpif_, struct odp_flow_put *put)
 398 {
 399     return do_ioctl(dpif_, ODP_FLOW_PUT, put);
 400 }
 401
 402 static int
 403 dpif_linux_flow_del(struct dpif *dpif_, struct odp_flow *flow)
 404 {
 405     return do_ioctl(dpif_, ODP_FLOW_DEL, flow);
 406 }
 407
 408 static int
 409 dpif_linux_flow_list(const struct dpif *dpif_, struct odp_flow flows[], int n)
 410 {
 411     struct odp_flowvec fv;
 412     int error;
 413
 414     fv.flows = flows;
 415     fv.n_flows = n;
 416     error = do_ioctl(dpif_, ODP_FLOW_LIST, &fv);
 417     return error ? -error : fv.n_flows;
 418 }
 419
 420 static int
 421 dpif_linux_execute(struct dpif *dpif_, uint16_t in_port,
 422                    const union odp_action actions[], int n_actions,
 423                    const struct ofpbuf *buf)
 424 {
 425     struct odp_execute execute;
 426     memset(&execute, 0, sizeof execute);
 427     execute.in_port = in_port;
 428     execute.actions = (union odp_action *) actions;
 429     execute.n_actions = n_actions;
 430     execute.data = buf->data;
 431     execute.length = buf->size;
 432     return do_ioctl(dpif_, ODP_EXECUTE, &execute);
 433 }
 434
 435 static int
 436 dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
 437 {
 438     return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask);
 439 }
 440
 441 static int
 442 dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
 443 {
 444     return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask);
 445 }
 446
 447 static int
 448 dpif_linux_get_sflow_probability(const struct dpif *dpif_,
 449                                  uint32_t *probability)
 450 {
 451     return do_ioctl(dpif_, ODP_GET_SFLOW_PROBABILITY, probability);
 452 }
 453
 454 static int
 455 dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
 456 {
 457     return do_ioctl(dpif_, ODP_SET_SFLOW_PROBABILITY, &probability);
 458 }
 459
 460 static int
 461 dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
 462                              uint32_t queue_id, uint32_t *priority)
 463 {
 464     if (queue_id < 0xf000) {
 465         *priority = TC_H_MAKE(1 << 16, queue_id + 1);
 466         return 0;
 467     } else {
 468         return EINVAL;
 469     }
 470 }
 471
 472 static int
 473 dpif_linux_recv(struct dpif *dpif_, struct ofpbuf **bufp)
 474 {
 475     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 476     struct ofpbuf *buf;
 477     int retval;
 478     int error;
 479
 480     buf = ofpbuf_new_with_headroom(65536, DPIF_RECV_MSG_PADDING);
 481     retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
 482     if (retval < 0) {
 483         error = errno;
 484         if (error != EAGAIN) {
 485             VLOG_WARN_RL(&error_rl, "%s: read failed: %s",
 486                          dpif_name(dpif_), strerror(error));
 487         }
 488     } else if (retval >= sizeof(struct odp_msg)) {
 489         struct odp_msg *msg = buf->data;
 490         if (msg->length <= retval) {
 491             buf->size += retval;
 492             *bufp = buf;
 493             return 0;
 494         } else {
 495             VLOG_WARN_RL(&error_rl, "%s: discarding message truncated "
 496                          "from %"PRIu32" bytes to %d",
 497                          dpif_name(dpif_), msg->length, retval);
 498             error = ERANGE;
 499         }
 500     } else if (!retval) {
 501         VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif_));
 502         error = EPROTO;
 503     } else {
 504         VLOG_WARN_RL(&error_rl,
 505                      "%s: discarding too-short message (%d bytes)",
 506                      dpif_name(dpif_), retval);
 507         error = ERANGE;
 508     }
 509
 510     *bufp = NULL;
 511     ofpbuf_delete(buf);
 512     return error;
 513 }
 514
 515 static void
 516 dpif_linux_recv_wait(struct dpif *dpif_)
 517 {
 518     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 519     poll_fd_wait(dpif->fd, POLLIN);
 520 }
 521
 522 const struct dpif_class dpif_linux_class = {
 523     "system",
 524     NULL,
 525     NULL,
 526     dpif_linux_enumerate,
 527     dpif_linux_open,
 528     dpif_linux_close,
 529     dpif_linux_get_all_names,
 530     dpif_linux_destroy,
 531     dpif_linux_get_stats,
 532     dpif_linux_get_drop_frags,
 533     dpif_linux_set_drop_frags,
 534     dpif_linux_port_add,
 535     dpif_linux_port_del,
 536     dpif_linux_port_query_by_number,
 537     dpif_linux_port_query_by_name,
 538     dpif_linux_port_list,
 539     dpif_linux_port_poll,
 540     dpif_linux_port_poll_wait,
 541     dpif_linux_port_group_get,
 542     dpif_linux_port_group_set,
 543     dpif_linux_flow_get,
 544     dpif_linux_flow_put,
 545     dpif_linux_flow_del,
 546     dpif_linux_flow_flush,
 547     dpif_linux_flow_list,
 548     dpif_linux_execute,
 549     dpif_linux_recv_get_mask,
 550     dpif_linux_recv_set_mask,
 551     dpif_linux_get_sflow_probability,
 552     dpif_linux_set_sflow_probability,
 553     dpif_linux_queue_to_priority,
 554     dpif_linux_recv,
 555     dpif_linux_recv_wait,
 556 };
 557 \f
 558 static int get_openvswitch_major(void);
 559 static int get_major(const char *target);
 560
 561 static int
 562 do_ioctl(const struct dpif *dpif_, int cmd, const void *arg)
 563 {
 564     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 565     return ioctl(dpif->fd, cmd, arg) ? errno : 0;
 566 }
 567
 568 static int
 569 lookup_minor(const char *name, int *minorp)
 570 {
 571     struct ethtool_drvinfo drvinfo;
 572     int minor, port_no;
 573     struct ifreq ifr;
 574     int error;
 575     int sock;
 576
 577     sock = socket(AF_INET, SOCK_DGRAM, 0);
 578     if (sock < 0) {
 579         VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno));
 580         error = errno;
 581         goto error;
 582     }
 583
 584     memset(&ifr, 0, sizeof ifr);
 585     strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
 586     ifr.ifr_data = (caddr_t) &drvinfo;
 587
 588     memset(&drvinfo, 0, sizeof drvinfo);
 589     drvinfo.cmd = ETHTOOL_GDRVINFO;
 590     if (ioctl(sock, SIOCETHTOOL, &ifr)) {
 591         VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno));
 592         error = errno;
 593         goto error_close_sock;
 594     }
 595
 596     if (strcmp(drvinfo.driver, "openvswitch")) {
 597         VLOG_WARN("%s is not an openvswitch device", name);
 598         error = EOPNOTSUPP;
 599         goto error_close_sock;
 600     }
 601
 602     if (sscanf(drvinfo.bus_info, "%d.%d", &minor, &port_no) != 2) {
 603         VLOG_WARN("%s ethtool bus_info has unexpected format", name);
 604         error = EPROTOTYPE;
 605         goto error_close_sock;
 606     } else if (port_no != ODPP_LOCAL) {
 607         /* This is an Open vSwitch device but not the local port.  We
 608          * intentionally support only using the name of the local port as the
 609          * name of a datapath; otherwise, it would be too difficult to
 610          * enumerate all the names of a datapath. */
 611         error = EOPNOTSUPP;
 612         goto error_close_sock;
 613     }
 614
 615     *minorp = minor;
 616     close(sock);
 617     return 0;
 618
 619 error_close_sock:
 620     close(sock);
 621 error:
 622     return error;
 623 }
 624
 625 static int
 626 make_openvswitch_device(int minor, char **fnp)
 627 {
 628     const char dirname[] = "/dev/net";
 629     int major;
 630     dev_t dev;
 631     struct stat s;
 632     char fn[128];
 633
 634     *fnp = NULL;
 635
 636     major = get_openvswitch_major();
 637     if (major < 0) {
 638         return -major;
 639     }
 640     dev = makedev(major, minor);
 641
 642     sprintf(fn, "%s/dp%d", dirname, minor);
 643     if (!stat(fn, &s)) {
 644         if (!S_ISCHR(s.st_mode)) {
 645             VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing",
 646                          fn);
 647         } else if (s.st_rdev != dev) {
 648             VLOG_WARN_RL(&error_rl,
 649                          "%s is device %u:%u but should be %u:%u, fixing",
 650                          fn, major(s.st_rdev), minor(s.st_rdev),
 651                          major(dev), minor(dev));
 652         } else {
 653             goto success;
 654         }
 655         if (unlink(fn)) {
 656             VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)",
 657                          fn, strerror(errno));
 658             return errno;
 659         }
 660     } else if (errno == ENOENT) {
 661         if (stat(dirname, &s)) {
 662             if (errno == ENOENT) {
 663                 if (mkdir(dirname, 0755)) {
 664                     VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)",
 665                                  dirname, strerror(errno));
 666                     return errno;
 667                 }
 668             } else {
 669                 VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)",
 670                              dirname, strerror(errno));
 671                 return errno;
 672             }
 673         }
 674     } else {
 675         VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno));
 676         return errno;
 677     }
 678
 679     /* The device needs to be created. */
 680     if (mknod(fn, S_IFCHR | 0700, dev)) {
 681         VLOG_WARN_RL(&error_rl,
 682                      "%s: creating character device %u:%u failed (%s)",
 683                      fn, major(dev), minor(dev), strerror(errno));
 684         return errno;
 685     }
 686
 687 success:
 688     *fnp = xstrdup(fn);
 689     return 0;
 690 }
 691
 692 /* Return the major device number of the Open vSwitch device.  If it
 693  * cannot be determined, a negative errno is returned. */
 694 static int
 695 get_openvswitch_major(void)
 696 {
 697     static int openvswitch_major = -1;
 698     if (openvswitch_major < 0) {
 699         openvswitch_major = get_major("openvswitch");
 700     }
 701     return openvswitch_major;
 702 }
 703
 704 static int
 705 get_major(const char *target)
 706 {
 707     const char fn[] = "/proc/devices";
 708     char line[128];
 709     FILE *file;
 710     int ln;
 711
 712     file = fopen(fn, "r");
 713     if (!file) {
 714         VLOG_ERR("opening %s failed (%s)", fn, strerror(errno));
 715         return -errno;
 716     }
 717
 718     for (ln = 1; fgets(line, sizeof line, file); ln++) {
 719         char name[64];
 720         int major;
 721
 722         if (!strncmp(line, "Character", 9) || line[0] == '\0') {
 723             /* Nothing to do. */
 724         } else if (!strncmp(line, "Block", 5)) {
 725             /* We only want character devices, so skip the rest of the file. */
 726             break;
 727         } else if (sscanf(line, "%d %63s", &major, name)) {
 728             if (!strcmp(name, target)) {
 729                 fclose(file);
 730                 return major;
 731             }
 732         } else {
 733             VLOG_WARN_ONCE("%s:%d: syntax error", fn, ln);
 734         }
 735     }
 736
 737     fclose(file);
 738
 739     VLOG_ERR("%s: %s major not found (is the module loaded?)", fn, target);
 740     return -ENODEV;
 741 }
 742
 743 static int
 744 finish_open(struct dpif *dpif_, const char *local_ifname)
 745 {
 746     struct dpif_linux *dpif = dpif_linux_cast(dpif_);
 747     dpif->local_ifname = xstrdup(local_ifname);
 748     dpif->local_ifindex = if_nametoindex(local_ifname);
 749     if (!dpif->local_ifindex) {
 750         int error = errno;
 751         dpif_uninit(dpif_, true);
 752         VLOG_WARN("could not get ifindex of %s device: %s",
 753                   local_ifname, strerror(errno));
 754         return error;
 755     }
 756     return 0;
 757 }
 758
 759 static int
 760 create_minor(const char *name, int minor, struct dpif **dpifp)
 761 {
 762     int error = open_minor(minor, dpifp);
 763     if (!error) {
 764         error = do_ioctl(*dpifp, ODP_DP_CREATE, name);
 765         if (!error) {
 766             error = finish_open(*dpifp, name);
 767         } else {
 768             dpif_uninit(*dpifp, true);
 769         }
 770     }
 771     return error;
 772 }
 773
 774 static int
 775 open_minor(int minor, struct dpif **dpifp)
 776 {
 777     int error;
 778     char *fn;
 779     int fd;
 780
 781     error = make_openvswitch_device(minor, &fn);
 782     if (error) {
 783         return error;
 784     }
 785
 786     fd = open(fn, O_RDONLY | O_NONBLOCK);
 787     if (fd >= 0) {
 788         struct dpif_linux *dpif = xmalloc(sizeof *dpif);
 789         error = rtnetlink_notifier_register(&dpif->port_notifier,
 790                                            dpif_linux_port_changed, dpif);
 791         if (!error) {
 792             char *name;
 793
 794             name = xasprintf("dp%d", minor);
 795             dpif_init(&dpif->dpif, &dpif_linux_class, name, minor, minor);
 796             free(name);
 797
 798             dpif->fd = fd;
 799             dpif->local_ifname = NULL;
 800             dpif->minor = minor;
 801             dpif->local_ifindex = 0;
 802             shash_init(&dpif->changed_ports);
 803             dpif->change_error = false;
 804             *dpifp = &dpif->dpif;
 805         } else {
 806             free(dpif);
 807         }
 808     } else {
 809         error = errno;
 810         VLOG_WARN("%s: open failed (%s)", fn, strerror(error));
 811     }
 812     free(fn);
 813
 814     return error;
 815 }
 816
 817 static void
 818 dpif_linux_port_changed(const struct rtnetlink_change *change, void *dpif_)
 819 {
 820     struct dpif_linux *dpif = dpif_;
 821
 822     if (change) {
 823         if (change->master_ifindex == dpif->local_ifindex
 824             && (change->nlmsg_type == RTM_NEWLINK
 825                 || change->nlmsg_type == RTM_DELLINK))
 826         {
 827             /* Our datapath changed, either adding a new port or deleting an
 828              * existing one. */
 829             shash_add_once(&dpif->changed_ports, change->ifname, NULL);
 830         }
 831     } else {
 832         dpif->change_error = true;
 833     }
 834 }