lib/netdev-linux.c

   1 /*
   2  * Copyright (c) 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 <assert.h>
  19 #include <errno.h>
  20 #include <fcntl.h>
  21 #include <arpa/inet.h>
  22 #include <inttypes.h>
  23 #include <linux/if_tun.h>
  24 #include <linux/ip.h>
  25 #include <linux/types.h>
  26 #include <linux/ethtool.h>
  27 #include <linux/rtnetlink.h>
  28 #include <linux/sockios.h>
  29 #include <linux/version.h>
  30 #include <sys/types.h>
  31 #include <sys/ioctl.h>
  32 #include <sys/socket.h>
  33 #include <netpacket/packet.h>
  34 #include <net/ethernet.h>
  35 #include <net/if.h>
  36 #include <linux/if_tunnel.h>
  37 #include <net/if_arp.h>
  38 #include <net/if_packet.h>
  39 #include <net/route.h>
  40 #include <netinet/in.h>
  41 #include <poll.h>
  42 #include <stdlib.h>
  43 #include <string.h>
  44 #include <unistd.h>
  45
  46 #include "coverage.h"
  47 #include "dynamic-string.h"
  48 #include "fatal-signal.h"
  49 #include "netdev-provider.h"
  50 #include "netlink.h"
  51 #include "ofpbuf.h"
  52 #include "openflow/openflow.h"
  53 #include "openvswitch/gre.h"
  54 #include "packets.h"
  55 #include "poll-loop.h"
  56 #include "rtnetlink.h"
  57 #include "socket-util.h"
  58 #include "shash.h"
  59 #include "svec.h"
  60
  61 #ifndef GRE_IOCTL_ONLY
  62 #include <linux/if_link.h>
  63 #endif
  64
  65 #define THIS_MODULE VLM_netdev_linux
  66 #include "vlog.h"
  67 \f
  68 /* These were introduced in Linux 2.6.14, so they might be missing if we have
  69  * old headers. */
  70 #ifndef ADVERTISED_Pause
  71 #define ADVERTISED_Pause                (1 << 13)
  72 #endif
  73 #ifndef ADVERTISED_Asym_Pause
  74 #define ADVERTISED_Asym_Pause           (1 << 14)
  75 #endif
  76
  77 static struct rtnetlink_notifier netdev_linux_cache_notifier;
  78 static int cache_notifier_refcount;
  79
  80 enum {
  81     VALID_IFINDEX = 1 << 0,
  82     VALID_ETHERADDR = 1 << 1,
  83     VALID_IN4 = 1 << 2,
  84     VALID_IN6 = 1 << 3,
  85     VALID_MTU = 1 << 4,
  86     VALID_CARRIER = 1 << 5,
  87     VALID_IS_INTERNAL = 1 << 6
  88 };
  89
  90 struct tap_state {
  91     int fd;
  92 };
  93
  94 struct netdev_dev_linux {
  95     struct netdev_dev netdev_dev;
  96
  97     struct shash_node *shash_node;
  98     unsigned int cache_valid;
  99
 100     int ifindex;
 101     uint8_t etheraddr[ETH_ADDR_LEN];
 102     struct in_addr address, netmask;
 103     struct in6_addr in6;
 104     int mtu;
 105     int carrier;
 106     bool is_internal;
 107
 108     union {
 109         struct tap_state tap;
 110     } state;
 111 };
 112
 113 struct netdev_linux {
 114     struct netdev netdev;
 115     int fd;
 116 };
 117
 118 /* An AF_INET socket (used for ioctl operations). */
 119 static int af_inet_sock = -1;
 120
 121 struct gre_config {
 122     uint32_t local_ip;
 123     uint32_t remote_ip;
 124     uint32_t in_key;
 125     uint32_t out_key;
 126     bool have_in_key;
 127     bool have_out_key;
 128     bool in_csum;
 129     bool out_csum;
 130 };
 131
 132 static struct {
 133     union {
 134         struct nl_sock *nl_sock;
 135         int ioctl_fd;
 136     };
 137     bool use_ioctl;
 138 } gre_descriptors;
 139
 140 struct netdev_linux_notifier {
 141     struct netdev_notifier notifier;
 142     struct list node;
 143 };
 144
 145 static struct shash netdev_linux_notifiers =
 146     SHASH_INITIALIZER(&netdev_linux_notifiers);
 147 static struct rtnetlink_notifier netdev_linux_poll_notifier;
 148
 149 /* This is set pretty low because we probably won't learn anything from the
 150  * additional log messages. */
 151 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
 152
 153 static int destroy_gre(const char *name);
 154 static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *,
 155                                    int cmd, const char *cmd_name);
 156 static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd,
 157                                  const char *cmd_name);
 158 static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *,
 159                                  int cmd, const char *cmd_name);
 160 static int get_flags(const struct netdev *, int *flagsp);
 161 static int set_flags(struct netdev *, int flags);
 162 static int do_get_ifindex(const char *netdev_name);
 163 static int get_ifindex(const struct netdev *, int *ifindexp);
 164 static int do_set_addr(struct netdev *netdev,
 165                        int ioctl_nr, const char *ioctl_name,
 166                        struct in_addr addr);
 167 static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]);
 168 static int set_etheraddr(const char *netdev_name, int hwaddr_family,
 169                          const uint8_t[ETH_ADDR_LEN]);
 170 static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats);
 171 static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats);
 172
 173 static struct netdev_dev_linux *
 174 netdev_dev_linux_cast(const struct netdev_dev *netdev_dev)
 175 {
 176     const char *type = netdev_dev_get_type(netdev_dev);
 177     assert(!strcmp(type, "system") || !strcmp(type, "tap")
 178             || !strcmp(type, "gre"));
 179     return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev);
 180 }
 181
 182 static struct netdev_linux *
 183 netdev_linux_cast(const struct netdev *netdev)
 184 {
 185     const char *type = netdev_get_type(netdev);
 186     assert(!strcmp(type, "system") || !strcmp(type, "tap")
 187             || !strcmp(type, "gre"));
 188     return CONTAINER_OF(netdev, struct netdev_linux, netdev);
 189 }
 190
 191 static int
 192 netdev_linux_init(void)
 193 {
 194     static int status = -1;
 195     if (status < 0) {
 196         af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0);
 197         status = af_inet_sock >= 0 ? 0 : errno;
 198         if (status) {
 199             VLOG_ERR("failed to create inet socket: %s", strerror(status));
 200         }
 201     }
 202     return status;
 203 }
 204
 205 static void
 206 netdev_linux_run(void)
 207 {
 208     rtnetlink_notifier_run();
 209 }
 210
 211 static void
 212 netdev_linux_wait(void)
 213 {
 214     rtnetlink_notifier_wait();
 215 }
 216
 217 static void
 218 netdev_linux_cache_cb(const struct rtnetlink_change *change,
 219                       void *aux OVS_UNUSED)
 220 {
 221     struct netdev_dev_linux *dev;
 222     if (change) {
 223         struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname);
 224         if (base_dev) {
 225             dev = netdev_dev_linux_cast(base_dev);
 226             dev->cache_valid = 0;
 227         }
 228     } else {
 229         struct shash device_shash;
 230         struct shash_node *node;
 231
 232         shash_init(&device_shash);
 233         netdev_dev_get_devices(&netdev_linux_class, &device_shash);
 234         SHASH_FOR_EACH (node, &device_shash) {
 235             dev = node->data;
 236             dev->cache_valid = 0;
 237         }
 238         shash_destroy(&device_shash);
 239     }
 240 }
 241
 242 /* The arguments are marked as unused to prevent warnings on platforms where
 243  * the Netlink interface isn't supported. */
 244 static int
 245 setup_gre_netlink(const char *name OVS_UNUSED,
 246                   struct gre_config *config OVS_UNUSED, bool create OVS_UNUSED)
 247 {
 248 #ifdef GRE_IOCTL_ONLY
 249     return EOPNOTSUPP;
 250 #else
 251     int error;
 252     struct ofpbuf request, *reply;
 253     unsigned int nl_flags;
 254     struct ifinfomsg ifinfomsg;
 255     struct nlattr *linkinfo_hdr;
 256     struct nlattr *info_data_hdr;
 257     uint16_t iflags = 0;
 258     uint16_t oflags = 0;
 259     uint8_t pmtudisc = 0;
 260
 261     VLOG_DBG("%s: attempting to create gre device using netlink", name);
 262
 263     if (!gre_descriptors.nl_sock) {
 264         error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0,
 265                                &gre_descriptors.nl_sock);
 266         if (error) {
 267             VLOG_WARN("couldn't create netlink socket: %s", strerror(error));
 268             goto error;
 269         }
 270     }
 271
 272     ofpbuf_init(&request, 0);
 273
 274     nl_flags = NLM_F_REQUEST;
 275     if (create) {
 276         nl_flags |= NLM_F_CREATE|NLM_F_EXCL;
 277     }
 278
 279     /* We over-reserve space, because we do some pointer arithmetic
 280      * and don't want the buffer address shifting under us. */
 281     nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 2048, RTM_NEWLINK,
 282                         nl_flags);
 283
 284     memset(&ifinfomsg, 0, sizeof ifinfomsg);
 285     ifinfomsg.ifi_family = AF_UNSPEC;
 286     nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
 287
 288     linkinfo_hdr = ofpbuf_tail(&request);
 289     nl_msg_put_unspec(&request, IFLA_LINKINFO, NULL, 0);
 290
 291     nl_msg_put_unspec(&request, IFLA_INFO_KIND, "gretap", 6);
 292
 293     info_data_hdr = ofpbuf_tail(&request);
 294     nl_msg_put_unspec(&request, IFLA_INFO_DATA, NULL, 0);
 295
 296     /* Set flags */
 297     if (config->have_in_key) {
 298         iflags |= GRE_KEY;
 299     }
 300     if (config->have_out_key) {
 301         oflags |= GRE_KEY;
 302     }
 303
 304     if (config->in_csum) {
 305         iflags |= GRE_CSUM;
 306     }
 307     if (config->out_csum) {
 308         oflags |= GRE_CSUM;
 309     }
 310
 311     /* Add options */
 312     nl_msg_put_u32(&request, IFLA_GRE_IKEY, config->in_key);
 313     nl_msg_put_u32(&request, IFLA_GRE_OKEY, config->out_key);
 314     nl_msg_put_u16(&request, IFLA_GRE_IFLAGS, iflags);
 315     nl_msg_put_u16(&request, IFLA_GRE_OFLAGS, oflags);
 316     nl_msg_put_u32(&request, IFLA_GRE_LOCAL, config->local_ip);
 317     nl_msg_put_u32(&request, IFLA_GRE_REMOTE, config->remote_ip);
 318     nl_msg_put_u8(&request, IFLA_GRE_PMTUDISC, pmtudisc);
 319     nl_msg_put_u8(&request, IFLA_GRE_TTL, 0);
 320     nl_msg_put_u8(&request, IFLA_GRE_TOS, 0);
 321
 322     info_data_hdr->nla_len = (char *)ofpbuf_tail(&request)
 323                                 - (char *)info_data_hdr;
 324     linkinfo_hdr->nla_len = (char *)ofpbuf_tail(&request)
 325                                 - (char *)linkinfo_hdr;
 326
 327     nl_msg_put_string(&request, IFLA_IFNAME, name);
 328
 329     error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
 330     ofpbuf_uninit(&request);
 331     if (error) {
 332         VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
 333         goto error;
 334     }
 335     ofpbuf_delete(reply);
 336
 337 error:
 338     return error;
 339 #endif
 340 }
 341
 342 static int
 343 setup_gre_ioctl(const char *name, struct gre_config *config, bool create)
 344 {
 345     struct ip_tunnel_parm p;
 346     struct ifreq ifr;
 347
 348     VLOG_DBG("%s: attempting to create gre device using ioctl", name);
 349
 350     memset(&p, 0, sizeof p);
 351
 352     strncpy(p.name, name, IFNAMSIZ);
 353
 354     p.iph.version = 4;
 355     p.iph.ihl = 5;
 356     p.iph.protocol = IPPROTO_GRE;
 357     p.iph.saddr = config->local_ip;
 358     p.iph.daddr = config->remote_ip;
 359
 360     if (config->have_in_key) {
 361         p.i_flags |= GRE_KEY;
 362         p.i_key = config->in_key;
 363     }
 364     if (config->have_out_key) {
 365         p.o_flags |= GRE_KEY;
 366         p.o_key = config->out_key;
 367     }
 368
 369     if (config->in_csum) {
 370         p.i_flags |= GRE_CSUM;
 371     }
 372     if (config->out_csum) {
 373         p.o_flags |= GRE_CSUM;
 374     }
 375
 376     strncpy(ifr.ifr_name, create ? GRE_IOCTL_DEVICE : name, IFNAMSIZ);
 377     ifr.ifr_ifru.ifru_data = (void *)&p;
 378
 379     if (!gre_descriptors.ioctl_fd) {
 380         gre_descriptors.ioctl_fd = socket(AF_INET, SOCK_DGRAM, 0);
 381         if (gre_descriptors.ioctl_fd < 0) {
 382             VLOG_WARN("couldn't create gre ioctl socket: %s", strerror(errno));
 383             gre_descriptors.ioctl_fd = 0;
 384             return errno;
 385         }
 386     }
 387
 388     if (ioctl(gre_descriptors.ioctl_fd, create ? SIOCADDGRETAP : SIOCCHGGRETAP,
 389               &ifr) < 0) {
 390         VLOG_WARN("couldn't do gre ioctl: %s", strerror(errno));
 391         return errno;
 392     }
 393
 394     return 0;
 395 }
 396
 397 /* The arguments are marked as unused to prevent warnings on platforms where
 398  * the Netlink interface isn't supported. */
 399 static bool
 400 check_gre_device_netlink(const char *name OVS_UNUSED)
 401 {
 402 #ifdef GRE_IOCTL_ONLY
 403     return false;
 404 #else
 405     static const struct nl_policy getlink_policy[] = {
 406         [IFLA_LINKINFO] = { .type = NL_A_NESTED, .optional = false },
 407     };
 408
 409     static const struct nl_policy linkinfo_policy[] = {
 410         [IFLA_INFO_KIND] = { .type = NL_A_STRING, .optional = false },
 411     };
 412
 413     int error;
 414     bool ret = false;
 415     struct ofpbuf request, *reply;
 416     struct ifinfomsg ifinfomsg;
 417     struct nlattr *getlink_attrs[ARRAY_SIZE(getlink_policy)];
 418     struct nlattr *linkinfo_attrs[ARRAY_SIZE(linkinfo_policy)];
 419     struct ofpbuf linkinfo;
 420     const char *device_kind;
 421
 422     ofpbuf_init(&request, 0);
 423
 424     nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock,
 425                         NLMSG_LENGTH(sizeof ifinfomsg), RTM_GETLINK,
 426                         NLM_F_REQUEST);
 427
 428     memset(&ifinfomsg, 0, sizeof ifinfomsg);
 429     ifinfomsg.ifi_family = AF_UNSPEC;
 430     ifinfomsg.ifi_index =  do_get_ifindex(name);
 431     nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
 432
 433     error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
 434     ofpbuf_uninit(&request);
 435     if (error) {
 436         VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
 437         return false;
 438     }
 439
 440     if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
 441                          getlink_policy, getlink_attrs,
 442                          ARRAY_SIZE(getlink_policy))) {
 443         VLOG_WARN("received bad rtnl message (getlink policy)");
 444         goto error;
 445     }
 446
 447     linkinfo.data = (void *)nl_attr_get(getlink_attrs[IFLA_LINKINFO]);
 448     linkinfo.size = nl_attr_get_size(getlink_attrs[IFLA_LINKINFO]);
 449     if (!nl_policy_parse(&linkinfo, 0, linkinfo_policy,
 450                         linkinfo_attrs, ARRAY_SIZE(linkinfo_policy))) {
 451         VLOG_WARN("received bad rtnl message (linkinfo policy)");
 452         goto error;
 453     }
 454
 455     device_kind = nl_attr_get_string(linkinfo_attrs[IFLA_INFO_KIND]);
 456     ret = !strcmp(device_kind, "gretap");
 457
 458 error:
 459     ofpbuf_delete(reply);
 460     return ret;
 461 #endif
 462 }
 463
 464 static bool
 465 check_gre_device_ioctl(const char *name)
 466 {
 467     struct ethtool_drvinfo drvinfo;
 468     int error;
 469
 470     memset(&drvinfo, 0, sizeof drvinfo);
 471     error = netdev_linux_do_ethtool(name, (struct ethtool_cmd *)&drvinfo,
 472                                     ETHTOOL_GDRVINFO, "ETHTOOL_GDRVINFO");
 473
 474     return !error && !strcmp(drvinfo.driver, "ip_gre")
 475            && !strcmp(drvinfo.bus_info, "gretap");
 476 }
 477
 478 static int
 479 setup_gre(const char *name, const struct shash *args, bool create)
 480 {
 481     int error;
 482     struct in_addr in_addr;
 483     struct shash_node *node;
 484     struct gre_config config;
 485
 486     memset(&config, 0, sizeof config);
 487     config.in_csum = true;
 488     config.out_csum = true;
 489
 490     SHASH_FOR_EACH (node, args) {
 491         if (!strcmp(node->name, "remote_ip")) {
 492             if (lookup_ip(node->data, &in_addr)) {
 493                 VLOG_WARN("bad 'remote_ip' for gre device %s ", name);
 494             } else {
 495                 config.remote_ip = in_addr.s_addr;
 496             }
 497         } else if (!strcmp(node->name, "local_ip")) {
 498             if (lookup_ip(node->data, &in_addr)) {
 499                 VLOG_WARN("bad 'local_ip' for gre device %s ", name);
 500             } else {
 501                 config.local_ip = in_addr.s_addr;
 502             }
 503         } else if (!strcmp(node->name, "key")) {
 504             config.have_in_key = true;
 505             config.have_out_key = true;
 506             config.in_key = htonl(atoi(node->data));
 507             config.out_key = htonl(atoi(node->data));
 508         } else if (!strcmp(node->name, "in_key")) {
 509             config.have_in_key = true;
 510             config.in_key = htonl(atoi(node->data));
 511         } else if (!strcmp(node->name, "out_key")) {
 512             config.have_out_key = true;
 513             config.out_key = htonl(atoi(node->data));
 514         } else if (!strcmp(node->name, "csum")) {
 515             if (!strcmp(node->data, "false")) {
 516                 config.in_csum = false;
 517                 config.out_csum = false;
 518             }
 519         } else {
 520             VLOG_WARN("unknown gre argument '%s'", node->name);
 521         }
 522     }
 523
 524     if (!config.remote_ip) {
 525         VLOG_WARN("gre type requires valid 'remote_ip' argument");
 526         error = EINVAL;
 527         goto error;
 528     }
 529
 530     if (!gre_descriptors.use_ioctl) {
 531         error = setup_gre_netlink(name, &config, create);
 532         if (error == EOPNOTSUPP) {
 533             gre_descriptors.use_ioctl = true;
 534         }
 535     }
 536     if (gre_descriptors.use_ioctl) {
 537         error = setup_gre_ioctl(name, &config, create);
 538     }
 539
 540     if (create && error == EEXIST) {
 541         bool gre_device;
 542
 543         if (gre_descriptors.use_ioctl) {
 544             gre_device = check_gre_device_ioctl(name);
 545         } else {
 546             gre_device = check_gre_device_netlink(name);
 547         }
 548
 549         if (!gre_device) {
 550             goto error;
 551         }
 552
 553         VLOG_WARN("replacing existing gre device %s", name);
 554         error = destroy_gre(name);
 555         if (error) {
 556             goto error;
 557         }
 558
 559         if (gre_descriptors.use_ioctl) {
 560             error = setup_gre_ioctl(name, &config, create);
 561         } else {
 562             error = setup_gre_netlink(name, &config, create);
 563         }
 564     }
 565
 566 error:
 567     return error;
 568 }
 569
 570 /* Creates the netdev device of 'type' with 'name'. */
 571 static int
 572 netdev_linux_create_system(const char *name, const char *type OVS_UNUSED,
 573                     const struct shash *args, struct netdev_dev **netdev_devp)
 574 {
 575     struct netdev_dev_linux *netdev_dev;
 576     int error;
 577
 578     if (!shash_is_empty(args)) {
 579         VLOG_WARN("%s: arguments for system devices should be empty", name);
 580     }
 581
 582     if (!cache_notifier_refcount) {
 583         error = rtnetlink_notifier_register(&netdev_linux_cache_notifier,
 584                                             netdev_linux_cache_cb, NULL);
 585         if (error) {
 586             return error;
 587         }
 588     }
 589     cache_notifier_refcount++;
 590
 591     netdev_dev = xzalloc(sizeof *netdev_dev);
 592     netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class);
 593
 594     *netdev_devp = &netdev_dev->netdev_dev;
 595     return 0;
 596 }
 597
 598 /* For most types of netdevs we open the device for each call of
 599  * netdev_open().  However, this is not the case with tap devices,
 600  * since it is only possible to open the device once.  In this
 601  * situation we share a single file descriptor, and consequently
 602  * buffers, across all readers.  Therefore once data is read it will
 603  * be unavailable to other reads for tap devices. */
 604 static int
 605 netdev_linux_create_tap(const char *name, const char *type OVS_UNUSED,
 606                     const struct shash *args, struct netdev_dev **netdev_devp)
 607 {
 608     struct netdev_dev_linux *netdev_dev;
 609     struct tap_state *state;
 610     static const char tap_dev[] = "/dev/net/tun";
 611     struct ifreq ifr;
 612     int error;
 613
 614     if (!shash_is_empty(args)) {
 615         VLOG_WARN("%s: arguments for TAP devices should be empty", name);
 616     }
 617
 618     netdev_dev = xzalloc(sizeof *netdev_dev);
 619     state = &netdev_dev->state.tap;
 620
 621     /* Open tap device. */
 622     state->fd = open(tap_dev, O_RDWR);
 623     if (state->fd < 0) {
 624         error = errno;
 625         VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error));
 626         goto error;
 627     }
 628
 629     /* Create tap device. */
 630     ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 631     strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
 632     if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) {
 633         VLOG_WARN("%s: creating tap device failed: %s", name,
 634                   strerror(errno));
 635         error = errno;
 636         goto error;
 637     }
 638
 639     /* Make non-blocking. */
 640     error = set_nonblocking(state->fd);
 641     if (error) {
 642         goto error;
 643     }
 644
 645     netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class);
 646     *netdev_devp = &netdev_dev->netdev_dev;
 647     return 0;
 648
 649 error:
 650     free(netdev_dev);
 651     return error;
 652 }
 653
 654 static int
 655 if_up(const char *name)
 656 {
 657     struct ifreq ifr;
 658
 659     strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
 660     ifr.ifr_flags = IFF_UP;
 661
 662     if (ioctl(af_inet_sock, SIOCSIFFLAGS, &ifr) == -1) {
 663         VLOG_DBG_RL(&rl, "%s: failed to bring device up: %s",
 664                     name, strerror(errno));
 665         return errno;
 666     }
 667
 668     return 0;
 669 }
 670
 671 static int
 672 netdev_linux_create_gre(const char *name, const char *type OVS_UNUSED,
 673                     const struct shash *args, struct netdev_dev **netdev_devp)
 674 {
 675     struct netdev_dev_linux *netdev_dev;
 676     int error;
 677
 678     netdev_dev = xzalloc(sizeof *netdev_dev);
 679
 680     error = setup_gre(name, args, true);
 681     if (error) {
 682         goto error;
 683     }
 684
 685     error = if_up(name);
 686     if (error) {
 687         goto error;
 688     }
 689
 690     netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_gre_class);
 691     *netdev_devp = &netdev_dev->netdev_dev;
 692     return 0;
 693
 694 error:
 695     free(netdev_dev);
 696     return error;
 697 }
 698
 699 static int
 700 netdev_linux_reconfigure_gre(struct netdev_dev *netdev_dev_,
 701                              const struct shash *args)
 702 {
 703     const char *name = netdev_dev_get_name(netdev_dev_);
 704
 705     return setup_gre(name, args, false);
 706 }
 707
 708 /* The arguments are marked as unused to prevent warnings on platforms where
 709  * the Netlink interface isn't supported. */
 710 static int
 711 destroy_gre_netlink(const char *name OVS_UNUSED)
 712 {
 713 #ifdef GRE_IOCTL_ONLY
 714     return EOPNOTSUPP;
 715 #else
 716     int error;
 717     struct ofpbuf request, *reply;
 718     struct ifinfomsg ifinfomsg;
 719     int ifindex;
 720
 721     ofpbuf_init(&request, 0);
 722
 723     nl_msg_put_nlmsghdr(&request, gre_descriptors.nl_sock, 0, RTM_DELLINK,
 724                         NLM_F_REQUEST);
 725
 726     memset(&ifinfomsg, 0, sizeof ifinfomsg);
 727     ifinfomsg.ifi_family = AF_UNSPEC;
 728     nl_msg_put(&request, &ifinfomsg, sizeof ifinfomsg);
 729
 730     ifindex = do_get_ifindex(name);
 731     nl_msg_put_u32(&request, IFLA_LINK, ifindex);
 732
 733     nl_msg_put_string(&request, IFLA_IFNAME, name);
 734
 735     error = nl_sock_transact(gre_descriptors.nl_sock, &request, &reply);
 736     ofpbuf_uninit(&request);
 737     if (error) {
 738         VLOG_WARN("couldn't transact netlink socket: %s", strerror(error));
 739         goto error;
 740     }
 741     ofpbuf_delete(reply);
 742
 743 error:
 744     return 0;
 745 #endif
 746 }
 747
 748 static int
 749 destroy_gre_ioctl(const char *name)
 750 {
 751     struct ip_tunnel_parm p;
 752     struct ifreq ifr;
 753
 754     memset(&p, 0, sizeof p);
 755     strncpy(p.name, name, IFNAMSIZ);
 756
 757     strncpy(ifr.ifr_name, name, IFNAMSIZ);
 758     ifr.ifr_ifru.ifru_data = (void *)&p;
 759
 760     if (ioctl(gre_descriptors.ioctl_fd, SIOCDELGRETAP, &ifr) < 0) {
 761         VLOG_WARN("couldn't do gre ioctl: %s\n", strerror(errno));
 762         return errno;
 763     }
 764
 765     return 0;
 766 }
 767
 768 static void
 769 destroy_tap(struct netdev_dev_linux *netdev_dev)
 770 {
 771     struct tap_state *state = &netdev_dev->state.tap;
 772
 773     if (state->fd >= 0) {
 774         close(state->fd);
 775     }
 776 }
 777
 778 static int
 779 destroy_gre(const char *name)
 780 {
 781     if (gre_descriptors.use_ioctl) {
 782         return destroy_gre_ioctl(name);
 783     } else {
 784         return destroy_gre_netlink(name);
 785     }
 786 }
 787
 788 /* Destroys the netdev device 'netdev_dev_'. */
 789 static void
 790 netdev_linux_destroy(struct netdev_dev *netdev_dev_)
 791 {
 792     struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
 793     const char *type = netdev_dev_get_type(netdev_dev_);
 794
 795     if (!strcmp(type, "system")) {
 796         cache_notifier_refcount--;
 797
 798         if (!cache_notifier_refcount) {
 799             rtnetlink_notifier_unregister(&netdev_linux_cache_notifier);
 800         }
 801     } else if (!strcmp(type, "tap")) {
 802         destroy_tap(netdev_dev);
 803     } else if (!strcmp(type, "gre")) {
 804         destroy_gre(netdev_dev_get_name(&netdev_dev->netdev_dev));
 805     }
 806
 807     free(netdev_dev_);
 808 }
 809
 810 static int
 811 netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype,
 812                   struct netdev **netdevp)
 813 {
 814     struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_);
 815     struct netdev_linux *netdev;
 816     enum netdev_flags flags;
 817     int error;
 818
 819     /* Allocate network device. */
 820     netdev = xzalloc(sizeof *netdev);
 821     netdev->fd = -1;
 822     netdev_init(&netdev->netdev, netdev_dev_);
 823
 824     error = netdev_get_flags(&netdev->netdev, &flags);
 825     if (error == ENODEV) {
 826         goto error;
 827     }
 828
 829     if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap")) {
 830         netdev->fd = netdev_dev->state.tap.fd;
 831     } else if (ethertype != NETDEV_ETH_TYPE_NONE) {
 832         struct sockaddr_ll sll;
 833         int protocol;
 834         int ifindex;
 835
 836         /* Create file descriptor. */
 837         protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL
 838                     : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2
 839                     : ethertype);
 840         netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol));
 841         if (netdev->fd < 0) {
 842             error = errno;
 843             goto error;
 844         }
 845
 846         /* Set non-blocking mode. */
 847         error = set_nonblocking(netdev->fd);
 848         if (error) {
 849             goto error;
 850         }
 851
 852         /* Get ethernet device index. */
 853         error = get_ifindex(&netdev->netdev, &ifindex);
 854         if (error) {
 855             goto error;
 856         }
 857
 858         /* Bind to specific ethernet device. */
 859         memset(&sll, 0, sizeof sll);
 860         sll.sll_family = AF_PACKET;
 861         sll.sll_ifindex = ifindex;
 862         if (bind(netdev->fd,
 863                  (struct sockaddr *) &sll, sizeof sll) < 0) {
 864             error = errno;
 865             VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_),
 866                      strerror(error));
 867             goto error;
 868         }
 869
 870         /* Between the socket() and bind() calls above, the socket receives all
 871          * packets of the requested type on all system interfaces.  We do not
 872          * want to receive that data, but there is no way to avoid it.  So we
 873          * must now drain out the receive queue. */
 874         error = drain_rcvbuf(netdev->fd);
 875         if (error) {
 876             goto error;
 877         }
 878     }
 879
 880     *netdevp = &netdev->netdev;
 881     return 0;
 882
 883 error:
 884     netdev_uninit(&netdev->netdev, true);
 885     return error;
 886 }
 887
 888 /* Closes and destroys 'netdev'. */
 889 static void
 890 netdev_linux_close(struct netdev *netdev_)
 891 {
 892     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
 893
 894     if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) {
 895         close(netdev->fd);
 896     }
 897     free(netdev);
 898 }
 899
 900 /* Initializes 'svec' with a list of the names of all known network devices. */
 901 static int
 902 netdev_linux_enumerate(struct svec *svec)
 903 {
 904     struct if_nameindex *names;
 905
 906     names = if_nameindex();
 907     if (names) {
 908         size_t i;
 909
 910         for (i = 0; names[i].if_name != NULL; i++) {
 911             svec_add(svec, names[i].if_name);
 912         }
 913         if_freenameindex(names);
 914         return 0;
 915     } else {
 916         VLOG_WARN("could not obtain list of network device names: %s",
 917                   strerror(errno));
 918         return errno;
 919     }
 920 }
 921
 922 static int
 923 netdev_linux_recv(struct netdev *netdev_, void *data, size_t size)
 924 {
 925     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
 926
 927     if (netdev->fd < 0) {
 928         /* Device was opened with NETDEV_ETH_TYPE_NONE. */
 929         return -EAGAIN;
 930     }
 931
 932     for (;;) {
 933         ssize_t retval = read(netdev->fd, data, size);
 934         if (retval >= 0) {
 935             return retval;
 936         } else if (errno != EINTR) {
 937             if (errno != EAGAIN) {
 938                 VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s",
 939                              strerror(errno), netdev_get_name(netdev_));
 940             }
 941             return -errno;
 942         }
 943     }
 944 }
 945
 946 /* Registers with the poll loop to wake up from the next call to poll_block()
 947  * when a packet is ready to be received with netdev_recv() on 'netdev'. */
 948 static void
 949 netdev_linux_recv_wait(struct netdev *netdev_)
 950 {
 951     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
 952     if (netdev->fd >= 0) {
 953         poll_fd_wait(netdev->fd, POLLIN);
 954     }
 955 }
 956
 957 /* Discards all packets waiting to be received from 'netdev'. */
 958 static int
 959 netdev_linux_drain(struct netdev *netdev_)
 960 {
 961     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
 962     if (netdev->fd < 0) {
 963         return 0;
 964     } else if (!strcmp(netdev_get_type(netdev_), "tap")) {
 965         struct ifreq ifr;
 966         int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
 967                                           SIOCGIFTXQLEN, "SIOCGIFTXQLEN");
 968         if (error) {
 969             return error;
 970         }
 971         drain_fd(netdev->fd, ifr.ifr_qlen);
 972         return 0;
 973     } else {
 974         return drain_rcvbuf(netdev->fd);
 975     }
 976 }
 977
 978 /* Sends 'buffer' on 'netdev'.  Returns 0 if successful, otherwise a positive
 979  * errno value.  Returns EAGAIN without blocking if the packet cannot be queued
 980  * immediately.  Returns EMSGSIZE if a partial packet was transmitted or if
 981  * the packet is too big or too small to transmit on the device.
 982  *
 983  * The caller retains ownership of 'buffer' in all cases.
 984  *
 985  * The kernel maintains a packet transmission queue, so the caller is not
 986  * expected to do additional queuing of packets. */
 987 static int
 988 netdev_linux_send(struct netdev *netdev_, const void *data, size_t size)
 989 {
 990     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
 991
 992     /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE.
 993      */
 994     if (netdev->fd < 0) {
 995         return EPIPE;
 996     }
 997
 998     for (;;) {
 999         ssize_t retval = write(netdev->fd, data, size);
1000         if (retval < 0) {
1001             /* The Linux AF_PACKET implementation never blocks waiting for room
1002              * for packets, instead returning ENOBUFS.  Translate this into
1003              * EAGAIN for the caller. */
1004             if (errno == ENOBUFS) {
1005                 return EAGAIN;
1006             } else if (errno == EINTR) {
1007                 continue;
1008             } else if (errno != EAGAIN) {
1009                 VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
1010                              netdev_get_name(netdev_), strerror(errno));
1011             }
1012             return errno;
1013         } else if (retval != size) {
1014             VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of "
1015                          "%zu) on %s", retval, size, netdev_get_name(netdev_));
1016             return EMSGSIZE;
1017         } else {
1018             return 0;
1019         }
1020     }
1021 }
1022
1023 /* Registers with the poll loop to wake up from the next call to poll_block()
1024  * when the packet transmission queue has sufficient room to transmit a packet
1025  * with netdev_send().
1026  *
1027  * The kernel maintains a packet transmission queue, so the client is not
1028  * expected to do additional queuing of packets.  Thus, this function is
1029  * unlikely to ever be used.  It is included for completeness. */
1030 static void
1031 netdev_linux_send_wait(struct netdev *netdev_)
1032 {
1033     struct netdev_linux *netdev = netdev_linux_cast(netdev_);
1034     if (netdev->fd < 0) {
1035         /* Nothing to do. */
1036     } else if (strcmp(netdev_get_type(netdev_), "tap")) {
1037         poll_fd_wait(netdev->fd, POLLOUT);
1038     } else {
1039         /* TAP device always accepts packets.*/
1040         poll_immediate_wake();
1041     }
1042 }
1043
1044 /* Attempts to set 'netdev''s MAC address to 'mac'.  Returns 0 if successful,
1045  * otherwise a positive errno value. */
1046 static int
1047 netdev_linux_set_etheraddr(struct netdev *netdev_,
1048                            const uint8_t mac[ETH_ADDR_LEN])
1049 {
1050     struct netdev_dev_linux *netdev_dev =
1051                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1052     int error;
1053
1054     if (!(netdev_dev->cache_valid & VALID_ETHERADDR)
1055         || !eth_addr_equals(netdev_dev->etheraddr, mac)) {
1056         error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac);
1057         if (!error) {
1058             netdev_dev->cache_valid |= VALID_ETHERADDR;
1059             memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN);
1060         }
1061     } else {
1062         error = 0;
1063     }
1064     return error;
1065 }
1066
1067 /* Returns a pointer to 'netdev''s MAC address.  The caller must not modify or
1068  * free the returned buffer. */
1069 static int
1070 netdev_linux_get_etheraddr(const struct netdev *netdev_,
1071                            uint8_t mac[ETH_ADDR_LEN])
1072 {
1073     struct netdev_dev_linux *netdev_dev =
1074                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1075     if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) {
1076         int error = get_etheraddr(netdev_get_name(netdev_),
1077                                   netdev_dev->etheraddr);
1078         if (error) {
1079             return error;
1080         }
1081         netdev_dev->cache_valid |= VALID_ETHERADDR;
1082     }
1083     memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN);
1084     return 0;
1085 }
1086
1087 /* Returns the maximum size of transmitted (and received) packets on 'netdev',
1088  * in bytes, not including the hardware header; thus, this is typically 1500
1089  * bytes for Ethernet devices. */
1090 static int
1091 netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup)
1092 {
1093     struct netdev_dev_linux *netdev_dev =
1094                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1095     if (!(netdev_dev->cache_valid & VALID_MTU)) {
1096         struct ifreq ifr;
1097         int error;
1098
1099         error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr,
1100                                       SIOCGIFMTU, "SIOCGIFMTU");
1101         if (error) {
1102             return error;
1103         }
1104         netdev_dev->mtu = ifr.ifr_mtu;
1105         netdev_dev->cache_valid |= VALID_MTU;
1106     }
1107     *mtup = netdev_dev->mtu;
1108     return 0;
1109 }
1110
1111 /* Returns the ifindex of 'netdev', if successful, as a positive number.
1112  * On failure, returns a negative errno value. */
1113 static int
1114 netdev_linux_get_ifindex(const struct netdev *netdev)
1115 {
1116     int ifindex, error;
1117
1118     error = get_ifindex(netdev, &ifindex);
1119     return error ? -error : ifindex;
1120 }
1121
1122 static int
1123 netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier)
1124 {
1125     struct netdev_dev_linux *netdev_dev =
1126                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1127     int error = 0;
1128     char *fn = NULL;
1129     int fd = -1;
1130
1131     if (!(netdev_dev->cache_valid & VALID_CARRIER)) {
1132         char line[8];
1133         int retval;
1134
1135         fn = xasprintf("/sys/class/net/%s/carrier",
1136                        netdev_get_name(netdev_));
1137         fd = open(fn, O_RDONLY);
1138         if (fd < 0) {
1139             error = errno;
1140             VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error));
1141             goto exit;
1142         }
1143
1144         retval = read(fd, line, sizeof line);
1145         if (retval < 0) {
1146             error = errno;
1147             if (error == EINVAL) {
1148                 /* This is the normal return value when we try to check carrier
1149                  * if the network device is not up. */
1150             } else {
1151                 VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error));
1152             }
1153             goto exit;
1154         } else if (retval == 0) {
1155             error = EPROTO;
1156             VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn);
1157             goto exit;
1158         }
1159
1160         if (line[0] != '0' && line[0] != '1') {
1161             error = EPROTO;
1162             VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)",
1163                          fn, line[0]);
1164             goto exit;
1165         }
1166         netdev_dev->carrier = line[0] != '0';
1167         netdev_dev->cache_valid |= VALID_CARRIER;
1168     }
1169     *carrier = netdev_dev->carrier;
1170     error = 0;
1171
1172 exit:
1173     if (fd >= 0) {
1174         close(fd);
1175     }
1176     free(fn);
1177     return error;
1178 }
1179
1180 /* Check whether we can we use RTM_GETLINK to get network device statistics.
1181  * In pre-2.6.19 kernels, this was only available if wireless extensions were
1182  * enabled. */
1183 static bool
1184 check_for_working_netlink_stats(void)
1185 {
1186     /* Decide on the netdev_get_stats() implementation to use.  Netlink is
1187      * preferable, so if that works, we'll use it. */
1188     int ifindex = do_get_ifindex("lo");
1189     if (ifindex < 0) {
1190         VLOG_WARN("failed to get ifindex for lo, "
1191                   "obtaining netdev stats from proc");
1192         return false;
1193     } else {
1194         struct netdev_stats stats;
1195         int error = get_stats_via_netlink(ifindex, &stats);
1196         if (!error) {
1197             VLOG_DBG("obtaining netdev stats via rtnetlink");
1198             return true;
1199         } else {
1200             VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats "
1201                       "via proc (you are probably running a pre-2.6.19 "
1202                       "kernel)", strerror(error));
1203             return false;
1204         }
1205     }
1206 }
1207
1208 /* Retrieves current device stats for 'netdev'.
1209  *
1210  * XXX All of the members of struct netdev_stats are 64 bits wide, but on
1211  * 32-bit architectures the Linux network stats are only 32 bits. */
1212 static int
1213 netdev_linux_get_stats(const struct netdev *netdev_,
1214                        struct netdev_stats *stats)
1215 {
1216     struct netdev_dev_linux *netdev_dev =
1217                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1218     static int use_netlink_stats = -1;
1219     int error;
1220     struct netdev_stats raw_stats;
1221     struct netdev_stats *collect_stats = stats;
1222
1223     COVERAGE_INC(netdev_get_stats);
1224
1225     if (!(netdev_dev->cache_valid & VALID_IS_INTERNAL)) {
1226         netdev_dev->is_internal = !strcmp(netdev_get_type(netdev_), "tap");
1227         if (!netdev_dev->is_internal) {
1228             struct ethtool_drvinfo drvinfo;
1229
1230             memset(&drvinfo, 0, sizeof drvinfo);
1231             error = netdev_linux_do_ethtool(netdev_get_name(netdev_),
1232                                             (struct ethtool_cmd *)&drvinfo,
1233                                             ETHTOOL_GDRVINFO,
1234                                             "ETHTOOL_GDRVINFO");
1235
1236             if (!error) {
1237                 netdev_dev->is_internal = !strcmp(drvinfo.driver,
1238                                                         "openvswitch");
1239             }
1240         }
1241
1242         netdev_dev->cache_valid |= VALID_IS_INTERNAL;
1243     }
1244
1245     if (netdev_dev->is_internal) {
1246         collect_stats = &raw_stats;
1247     }
1248
1249     if (use_netlink_stats < 0) {
1250         use_netlink_stats = check_for_working_netlink_stats();
1251     }
1252     if (use_netlink_stats) {
1253         int ifindex;
1254
1255         error = get_ifindex(netdev_, &ifindex);
1256         if (!error) {
1257             error = get_stats_via_netlink(ifindex, collect_stats);
1258         }
1259     } else {
1260         error = get_stats_via_proc(netdev_get_name(netdev_), collect_stats);
1261     }
1262
1263     /* If this port is an internal port then the transmit and receive stats
1264      * will appear to be swapped relative to the other ports since we are the
1265      * one sending the data, not a remote computer.  For consistency, we swap
1266      * them back here. */
1267     if (!error && netdev_dev->is_internal) {
1268         stats->rx_packets = raw_stats.tx_packets;
1269         stats->tx_packets = raw_stats.rx_packets;
1270         stats->rx_bytes = raw_stats.tx_bytes;
1271         stats->tx_bytes = raw_stats.rx_bytes;
1272         stats->rx_errors = raw_stats.tx_errors;
1273         stats->tx_errors = raw_stats.rx_errors;
1274         stats->rx_dropped = raw_stats.tx_dropped;
1275         stats->tx_dropped = raw_stats.rx_dropped;
1276         stats->multicast = raw_stats.multicast;
1277         stats->collisions = raw_stats.collisions;
1278         stats->rx_length_errors = 0;
1279         stats->rx_over_errors = 0;
1280         stats->rx_crc_errors = 0;
1281         stats->rx_frame_errors = 0;
1282         stats->rx_fifo_errors = 0;
1283         stats->rx_missed_errors = 0;
1284         stats->tx_aborted_errors = 0;
1285         stats->tx_carrier_errors = 0;
1286         stats->tx_fifo_errors = 0;
1287         stats->tx_heartbeat_errors = 0;
1288         stats->tx_window_errors = 0;
1289     }
1290
1291     return error;
1292 }
1293
1294 /* Stores the features supported by 'netdev' into each of '*current',
1295  * '*advertised', '*supported', and '*peer' that are non-null.  Each value is a
1296  * bitmap of "enum ofp_port_features" bits, in host byte order.  Returns 0 if
1297  * successful, otherwise a positive errno value. */
1298 static int
1299 netdev_linux_get_features(struct netdev *netdev,
1300                           uint32_t *current, uint32_t *advertised,
1301                           uint32_t *supported, uint32_t *peer)
1302 {
1303     struct ethtool_cmd ecmd;
1304     int error;
1305
1306     memset(&ecmd, 0, sizeof ecmd);
1307     error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1308                                     ETHTOOL_GSET, "ETHTOOL_GSET");
1309     if (error) {
1310         return error;
1311     }
1312
1313     /* Supported features. */
1314     *supported = 0;
1315     if (ecmd.supported & SUPPORTED_10baseT_Half) {
1316         *supported |= OFPPF_10MB_HD;
1317     }
1318     if (ecmd.supported & SUPPORTED_10baseT_Full) {
1319         *supported |= OFPPF_10MB_FD;
1320     }
1321     if (ecmd.supported & SUPPORTED_100baseT_Half)  {
1322         *supported |= OFPPF_100MB_HD;
1323     }
1324     if (ecmd.supported & SUPPORTED_100baseT_Full) {
1325         *supported |= OFPPF_100MB_FD;
1326     }
1327     if (ecmd.supported & SUPPORTED_1000baseT_Half) {
1328         *supported |= OFPPF_1GB_HD;
1329     }
1330     if (ecmd.supported & SUPPORTED_1000baseT_Full) {
1331         *supported |= OFPPF_1GB_FD;
1332     }
1333     if (ecmd.supported & SUPPORTED_10000baseT_Full) {
1334         *supported |= OFPPF_10GB_FD;
1335     }
1336     if (ecmd.supported & SUPPORTED_TP) {
1337         *supported |= OFPPF_COPPER;
1338     }
1339     if (ecmd.supported & SUPPORTED_FIBRE) {
1340         *supported |= OFPPF_FIBER;
1341     }
1342     if (ecmd.supported & SUPPORTED_Autoneg) {
1343         *supported |= OFPPF_AUTONEG;
1344     }
1345     if (ecmd.supported & SUPPORTED_Pause) {
1346         *supported |= OFPPF_PAUSE;
1347     }
1348     if (ecmd.supported & SUPPORTED_Asym_Pause) {
1349         *supported |= OFPPF_PAUSE_ASYM;
1350     }
1351
1352     /* Advertised features. */
1353     *advertised = 0;
1354     if (ecmd.advertising & ADVERTISED_10baseT_Half) {
1355         *advertised |= OFPPF_10MB_HD;
1356     }
1357     if (ecmd.advertising & ADVERTISED_10baseT_Full) {
1358         *advertised |= OFPPF_10MB_FD;
1359     }
1360     if (ecmd.advertising & ADVERTISED_100baseT_Half) {
1361         *advertised |= OFPPF_100MB_HD;
1362     }
1363     if (ecmd.advertising & ADVERTISED_100baseT_Full) {
1364         *advertised |= OFPPF_100MB_FD;
1365     }
1366     if (ecmd.advertising & ADVERTISED_1000baseT_Half) {
1367         *advertised |= OFPPF_1GB_HD;
1368     }
1369     if (ecmd.advertising & ADVERTISED_1000baseT_Full) {
1370         *advertised |= OFPPF_1GB_FD;
1371     }
1372     if (ecmd.advertising & ADVERTISED_10000baseT_Full) {
1373         *advertised |= OFPPF_10GB_FD;
1374     }
1375     if (ecmd.advertising & ADVERTISED_TP) {
1376         *advertised |= OFPPF_COPPER;
1377     }
1378     if (ecmd.advertising & ADVERTISED_FIBRE) {
1379         *advertised |= OFPPF_FIBER;
1380     }
1381     if (ecmd.advertising & ADVERTISED_Autoneg) {
1382         *advertised |= OFPPF_AUTONEG;
1383     }
1384     if (ecmd.advertising & ADVERTISED_Pause) {
1385         *advertised |= OFPPF_PAUSE;
1386     }
1387     if (ecmd.advertising & ADVERTISED_Asym_Pause) {
1388         *advertised |= OFPPF_PAUSE_ASYM;
1389     }
1390
1391     /* Current settings. */
1392     if (ecmd.speed == SPEED_10) {
1393         *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD;
1394     } else if (ecmd.speed == SPEED_100) {
1395         *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD;
1396     } else if (ecmd.speed == SPEED_1000) {
1397         *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD;
1398     } else if (ecmd.speed == SPEED_10000) {
1399         *current = OFPPF_10GB_FD;
1400     } else {
1401         *current = 0;
1402     }
1403
1404     if (ecmd.port == PORT_TP) {
1405         *current |= OFPPF_COPPER;
1406     } else if (ecmd.port == PORT_FIBRE) {
1407         *current |= OFPPF_FIBER;
1408     }
1409
1410     if (ecmd.autoneg) {
1411         *current |= OFPPF_AUTONEG;
1412     }
1413
1414     /* Peer advertisements. */
1415     *peer = 0;                  /* XXX */
1416
1417     return 0;
1418 }
1419
1420 /* Set the features advertised by 'netdev' to 'advertise'. */
1421 static int
1422 netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise)
1423 {
1424     struct ethtool_cmd ecmd;
1425     int error;
1426
1427     memset(&ecmd, 0, sizeof ecmd);
1428     error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1429                                     ETHTOOL_GSET, "ETHTOOL_GSET");
1430     if (error) {
1431         return error;
1432     }
1433
1434     ecmd.advertising = 0;
1435     if (advertise & OFPPF_10MB_HD) {
1436         ecmd.advertising |= ADVERTISED_10baseT_Half;
1437     }
1438     if (advertise & OFPPF_10MB_FD) {
1439         ecmd.advertising |= ADVERTISED_10baseT_Full;
1440     }
1441     if (advertise & OFPPF_100MB_HD) {
1442         ecmd.advertising |= ADVERTISED_100baseT_Half;
1443     }
1444     if (advertise & OFPPF_100MB_FD) {
1445         ecmd.advertising |= ADVERTISED_100baseT_Full;
1446     }
1447     if (advertise & OFPPF_1GB_HD) {
1448         ecmd.advertising |= ADVERTISED_1000baseT_Half;
1449     }
1450     if (advertise & OFPPF_1GB_FD) {
1451         ecmd.advertising |= ADVERTISED_1000baseT_Full;
1452     }
1453     if (advertise & OFPPF_10GB_FD) {
1454         ecmd.advertising |= ADVERTISED_10000baseT_Full;
1455     }
1456     if (advertise & OFPPF_COPPER) {
1457         ecmd.advertising |= ADVERTISED_TP;
1458     }
1459     if (advertise & OFPPF_FIBER) {
1460         ecmd.advertising |= ADVERTISED_FIBRE;
1461     }
1462     if (advertise & OFPPF_AUTONEG) {
1463         ecmd.advertising |= ADVERTISED_Autoneg;
1464     }
1465     if (advertise & OFPPF_PAUSE) {
1466         ecmd.advertising |= ADVERTISED_Pause;
1467     }
1468     if (advertise & OFPPF_PAUSE_ASYM) {
1469         ecmd.advertising |= ADVERTISED_Asym_Pause;
1470     }
1471     return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd,
1472                                    ETHTOOL_SSET, "ETHTOOL_SSET");
1473 }
1474
1475 /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with
1476  * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device
1477  * and returns 0.  Otherwise returns a errno value (specifically ENOENT if
1478  * 'netdev_name' is the name of a network device that is not a VLAN device) and
1479  * sets '*vlan_vid' to -1. */
1480 static int
1481 netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid)
1482 {
1483     const char *netdev_name = netdev_get_name(netdev);
1484     struct ds line = DS_EMPTY_INITIALIZER;
1485     FILE *stream = NULL;
1486     int error;
1487     char *fn;
1488
1489     COVERAGE_INC(netdev_get_vlan_vid);
1490     fn = xasprintf("/proc/net/vlan/%s", netdev_name);
1491     stream = fopen(fn, "r");
1492     if (!stream) {
1493         error = errno;
1494         goto done;
1495     }
1496
1497     if (ds_get_line(&line, stream)) {
1498         if (ferror(stream)) {
1499             error = errno;
1500             VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno));
1501         } else {
1502             error = EPROTO;
1503             VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn);
1504         }
1505         goto done;
1506     }
1507
1508     if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) {
1509         error = EPROTO;
1510         VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"",
1511                     fn, ds_cstr(&line));
1512         goto done;
1513     }
1514
1515     error = 0;
1516
1517 done:
1518     free(fn);
1519     if (stream) {
1520         fclose(stream);
1521     }
1522     ds_destroy(&line);
1523     if (error) {
1524         *vlan_vid = -1;
1525     }
1526     return error;
1527 }
1528
1529 #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress"
1530 #define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1"
1531 /* We redirect stderr to /dev/null because we often want to remove all
1532  * traffic control configuration on a port so its in a known state.  If
1533  * this done when there is no such configuration, tc complains, so we just
1534  * always ignore it.
1535  */
1536 #define POLICE_DEL_CMD "/sbin/tc qdisc del dev %s handle ffff: ingress 2>/dev/null"
1537
1538 /* Attempts to set input rate limiting (policing) policy. */
1539 static int
1540 netdev_linux_set_policing(struct netdev *netdev,
1541                           uint32_t kbits_rate, uint32_t kbits_burst)
1542 {
1543     const char *netdev_name = netdev_get_name(netdev);
1544     char command[1024];
1545
1546     COVERAGE_INC(netdev_set_policing);
1547     if (kbits_rate) {
1548         if (!kbits_burst) {
1549             /* Default to 1000 kilobits if not specified. */
1550             kbits_burst = 1000;
1551         }
1552
1553         /* xxx This should be more careful about only adding if it
1554          * xxx actually exists, as opposed to always deleting it. */
1555         snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1556         if (system(command) == -1) {
1557             VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1558         }
1559
1560         snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name);
1561         if (system(command) != 0) {
1562             VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name);
1563             return -1;
1564         }
1565
1566         snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name,
1567                 kbits_rate, kbits_burst);
1568         if (system(command) != 0) {
1569             VLOG_WARN_RL(&rl, "%s: problem configuring policing",
1570                     netdev_name);
1571             return -1;
1572         }
1573     } else {
1574         snprintf(command, sizeof(command), POLICE_DEL_CMD, netdev_name);
1575         if (system(command) == -1) {
1576             VLOG_WARN_RL(&rl, "%s: problem removing policing", netdev_name);
1577         }
1578     }
1579
1580     return 0;
1581 }
1582
1583 static int
1584 netdev_linux_get_in4(const struct netdev *netdev_,
1585                      struct in_addr *address, struct in_addr *netmask)
1586 {
1587     struct netdev_dev_linux *netdev_dev =
1588                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1589
1590     if (!(netdev_dev->cache_valid & VALID_IN4)) {
1591         int error;
1592
1593         error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address,
1594                                       SIOCGIFADDR, "SIOCGIFADDR");
1595         if (error) {
1596             return error;
1597         }
1598
1599         error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask,
1600                                       SIOCGIFNETMASK, "SIOCGIFNETMASK");
1601         if (error) {
1602             return error;
1603         }
1604
1605         netdev_dev->cache_valid |= VALID_IN4;
1606     }
1607     *address = netdev_dev->address;
1608     *netmask = netdev_dev->netmask;
1609     return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0;
1610 }
1611
1612 static int
1613 netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address,
1614                      struct in_addr netmask)
1615 {
1616     struct netdev_dev_linux *netdev_dev =
1617                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1618     int error;
1619
1620     error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address);
1621     if (!error) {
1622         netdev_dev->cache_valid |= VALID_IN4;
1623         netdev_dev->address = address;
1624         netdev_dev->netmask = netmask;
1625         if (address.s_addr != INADDR_ANY) {
1626             error = do_set_addr(netdev_, SIOCSIFNETMASK,
1627                                 "SIOCSIFNETMASK", netmask);
1628         }
1629     }
1630     return error;
1631 }
1632
1633 static bool
1634 parse_if_inet6_line(const char *line,
1635                     struct in6_addr *in6, char ifname[16 + 1])
1636 {
1637     uint8_t *s6 = in6->s6_addr;
1638 #define X8 "%2"SCNx8
1639     return sscanf(line,
1640                   " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8
1641                   "%*x %*x %*x %*x %16s\n",
1642                   &s6[0], &s6[1], &s6[2], &s6[3],
1643                   &s6[4], &s6[5], &s6[6], &s6[7],
1644                   &s6[8], &s6[9], &s6[10], &s6[11],
1645                   &s6[12], &s6[13], &s6[14], &s6[15],
1646                   ifname) == 17;
1647 }
1648
1649 /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if
1650  * 'in6' is non-null) and returns true.  Otherwise, returns false. */
1651 static int
1652 netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6)
1653 {
1654     struct netdev_dev_linux *netdev_dev =
1655                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
1656     if (!(netdev_dev->cache_valid & VALID_IN6)) {
1657         FILE *file;
1658         char line[128];
1659
1660         netdev_dev->in6 = in6addr_any;
1661
1662         file = fopen("/proc/net/if_inet6", "r");
1663         if (file != NULL) {
1664             const char *name = netdev_get_name(netdev_);
1665             while (fgets(line, sizeof line, file)) {
1666                 struct in6_addr in6;
1667                 char ifname[16 + 1];
1668                 if (parse_if_inet6_line(line, &in6, ifname)
1669                     && !strcmp(name, ifname))
1670                 {
1671                     netdev_dev->in6 = in6;
1672                     break;
1673                 }
1674             }
1675             fclose(file);
1676         }
1677         netdev_dev->cache_valid |= VALID_IN6;
1678     }
1679     *in6 = netdev_dev->in6;
1680     return 0;
1681 }
1682
1683 static void
1684 make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr)
1685 {
1686     struct sockaddr_in sin;
1687     memset(&sin, 0, sizeof sin);
1688     sin.sin_family = AF_INET;
1689     sin.sin_addr = addr;
1690     sin.sin_port = 0;
1691
1692     memset(sa, 0, sizeof *sa);
1693     memcpy(sa, &sin, sizeof sin);
1694 }
1695
1696 static int
1697 do_set_addr(struct netdev *netdev,
1698             int ioctl_nr, const char *ioctl_name, struct in_addr addr)
1699 {
1700     struct ifreq ifr;
1701     strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name);
1702     make_in4_sockaddr(&ifr.ifr_addr, addr);
1703
1704     return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr,
1705                                  ioctl_name);
1706 }
1707
1708 /* Adds 'router' as a default IP gateway. */
1709 static int
1710 netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router)
1711 {
1712     struct in_addr any = { INADDR_ANY };
1713     struct rtentry rt;
1714     int error;
1715
1716     memset(&rt, 0, sizeof rt);
1717     make_in4_sockaddr(&rt.rt_dst, any);
1718     make_in4_sockaddr(&rt.rt_gateway, router);
1719     make_in4_sockaddr(&rt.rt_genmask, any);
1720     rt.rt_flags = RTF_UP | RTF_GATEWAY;
1721     COVERAGE_INC(netdev_add_router);
1722     error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0;
1723     if (error) {
1724         VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error));
1725     }
1726     return error;
1727 }
1728
1729 static int
1730 netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop,
1731                           char **netdev_name)
1732 {
1733     static const char fn[] = "/proc/net/route";
1734     FILE *stream;
1735     char line[256];
1736     int ln;
1737
1738     *netdev_name = NULL;
1739     stream = fopen(fn, "r");
1740     if (stream == NULL) {
1741         VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
1742         return errno;
1743     }
1744
1745     ln = 0;
1746     while (fgets(line, sizeof line, stream)) {
1747         if (++ln >= 2) {
1748             char iface[17];
1749             uint32_t dest, gateway, mask;
1750             int refcnt, metric, mtu;
1751             unsigned int flags, use, window, irtt;
1752
1753             if (sscanf(line,
1754                        "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32
1755                        " %d %u %u\n",
1756                        iface, &dest, &gateway, &flags, &refcnt,
1757                        &use, &metric, &mask, &mtu, &window, &irtt) != 11) {
1758
1759                 VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s",
1760                         fn, ln, line);
1761                 continue;
1762             }
1763             if (!(flags & RTF_UP)) {
1764                 /* Skip routes that aren't up. */
1765                 continue;
1766             }
1767
1768             /* The output of 'dest', 'mask', and 'gateway' were given in
1769              * network byte order, so we don't need need any endian
1770              * conversions here. */
1771             if ((dest & mask) == (host->s_addr & mask)) {
1772                 if (!gateway) {
1773                     /* The host is directly reachable. */
1774                     next_hop->s_addr = 0;
1775                 } else {
1776                     /* To reach the host, we must go through a gateway. */
1777                     next_hop->s_addr = gateway;
1778                 }
1779                 *netdev_name = xstrdup(iface);
1780                 fclose(stream);
1781                 return 0;
1782             }
1783         }
1784     }
1785
1786     fclose(stream);
1787     return ENXIO;
1788 }
1789
1790 /* Looks up the ARP table entry for 'ip' on 'netdev'.  If one exists and can be
1791  * successfully retrieved, it stores the corresponding MAC address in 'mac' and
1792  * returns 0.  Otherwise, it returns a positive errno value; in particular,
1793  * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */
1794 static int
1795 netdev_linux_arp_lookup(const struct netdev *netdev,
1796                         uint32_t ip, uint8_t mac[ETH_ADDR_LEN])
1797 {
1798     struct arpreq r;
1799     struct sockaddr_in sin;
1800     int retval;
1801
1802     memset(&r, 0, sizeof r);
1803     sin.sin_family = AF_INET;
1804     sin.sin_addr.s_addr = ip;
1805     sin.sin_port = 0;
1806     memcpy(&r.arp_pa, &sin, sizeof sin);
1807     r.arp_ha.sa_family = ARPHRD_ETHER;
1808     r.arp_flags = 0;
1809     strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev);
1810     COVERAGE_INC(netdev_arp_lookup);
1811     retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0;
1812     if (!retval) {
1813         memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN);
1814     } else if (retval != ENXIO) {
1815         VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s",
1816                      netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval));
1817     }
1818     return retval;
1819 }
1820
1821 static int
1822 nd_to_iff_flags(enum netdev_flags nd)
1823 {
1824     int iff = 0;
1825     if (nd & NETDEV_UP) {
1826         iff |= IFF_UP;
1827     }
1828     if (nd & NETDEV_PROMISC) {
1829         iff |= IFF_PROMISC;
1830     }
1831     return iff;
1832 }
1833
1834 static int
1835 iff_to_nd_flags(int iff)
1836 {
1837     enum netdev_flags nd = 0;
1838     if (iff & IFF_UP) {
1839         nd |= NETDEV_UP;
1840     }
1841     if (iff & IFF_PROMISC) {
1842         nd |= NETDEV_PROMISC;
1843     }
1844     return nd;
1845 }
1846
1847 static int
1848 netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off,
1849                           enum netdev_flags on, enum netdev_flags *old_flagsp)
1850 {
1851     int old_flags, new_flags;
1852     int error;
1853
1854     error = get_flags(netdev, &old_flags);
1855     if (!error) {
1856         *old_flagsp = iff_to_nd_flags(old_flags);
1857         new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on);
1858         if (new_flags != old_flags) {
1859             error = set_flags(netdev, new_flags);
1860         }
1861     }
1862     return error;
1863 }
1864
1865 static void
1866 poll_notify(struct list *list)
1867 {
1868     struct netdev_linux_notifier *notifier;
1869     LIST_FOR_EACH (notifier, struct netdev_linux_notifier, node, list) {
1870         struct netdev_notifier *n = &notifier->notifier;
1871         n->cb(n);
1872     }
1873 }
1874
1875 static void
1876 netdev_linux_poll_cb(const struct rtnetlink_change *change,
1877                      void *aux OVS_UNUSED)
1878 {
1879     if (change) {
1880         struct list *list = shash_find_data(&netdev_linux_notifiers,
1881                                             change->ifname);
1882         if (list) {
1883             poll_notify(list);
1884         }
1885     } else {
1886         struct shash_node *node;
1887         SHASH_FOR_EACH (node, &netdev_linux_notifiers) {
1888             poll_notify(node->data);
1889         }
1890     }
1891 }
1892
1893 static int
1894 netdev_linux_poll_add(struct netdev *netdev,
1895                       void (*cb)(struct netdev_notifier *), void *aux,
1896                       struct netdev_notifier **notifierp)
1897 {
1898     const char *netdev_name = netdev_get_name(netdev);
1899     struct netdev_linux_notifier *notifier;
1900     struct list *list;
1901
1902     if (shash_is_empty(&netdev_linux_notifiers)) {
1903         int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier,
1904                                                    netdev_linux_poll_cb, NULL);
1905         if (error) {
1906             return error;
1907         }
1908     }
1909
1910     list = shash_find_data(&netdev_linux_notifiers, netdev_name);
1911     if (!list) {
1912         list = xmalloc(sizeof *list);
1913         list_init(list);
1914         shash_add(&netdev_linux_notifiers, netdev_name, list);
1915     }
1916
1917     notifier = xmalloc(sizeof *notifier);
1918     netdev_notifier_init(&notifier->notifier, netdev, cb, aux);
1919     list_push_back(list, &notifier->node);
1920     *notifierp = &notifier->notifier;
1921     return 0;
1922 }
1923
1924 static void
1925 netdev_linux_poll_remove(struct netdev_notifier *notifier_)
1926 {
1927     struct netdev_linux_notifier *notifier =
1928         CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier);
1929     struct list *list;
1930
1931     /* Remove 'notifier' from its list. */
1932     list = list_remove(&notifier->node);
1933     if (list_is_empty(list)) {
1934         /* The list is now empty.  Remove it from the hash and free it. */
1935         const char *netdev_name = netdev_get_name(notifier->notifier.netdev);
1936         shash_delete(&netdev_linux_notifiers,
1937                      shash_find(&netdev_linux_notifiers, netdev_name));
1938         free(list);
1939     }
1940     free(notifier);
1941
1942     /* If that was the last notifier, unregister. */
1943     if (shash_is_empty(&netdev_linux_notifiers)) {
1944         rtnetlink_notifier_unregister(&netdev_linux_poll_notifier);
1945     }
1946 }
1947
1948 const struct netdev_class netdev_linux_class = {
1949     "system",
1950
1951     netdev_linux_init,
1952     netdev_linux_run,
1953     netdev_linux_wait,
1954
1955     netdev_linux_create_system,
1956     netdev_linux_destroy,
1957     NULL,                       /* reconfigure */
1958
1959     netdev_linux_open,
1960     netdev_linux_close,
1961
1962     netdev_linux_enumerate,
1963
1964     netdev_linux_recv,
1965     netdev_linux_recv_wait,
1966     netdev_linux_drain,
1967
1968     netdev_linux_send,
1969     netdev_linux_send_wait,
1970
1971     netdev_linux_set_etheraddr,
1972     netdev_linux_get_etheraddr,
1973     netdev_linux_get_mtu,
1974     netdev_linux_get_ifindex,
1975     netdev_linux_get_carrier,
1976     netdev_linux_get_stats,
1977
1978     netdev_linux_get_features,
1979     netdev_linux_set_advertisements,
1980     netdev_linux_get_vlan_vid,
1981     netdev_linux_set_policing,
1982
1983     netdev_linux_get_in4,
1984     netdev_linux_set_in4,
1985     netdev_linux_get_in6,
1986     netdev_linux_add_router,
1987     netdev_linux_get_next_hop,
1988     netdev_linux_arp_lookup,
1989
1990     netdev_linux_update_flags,
1991
1992     netdev_linux_poll_add,
1993     netdev_linux_poll_remove,
1994 };
1995
1996 const struct netdev_class netdev_tap_class = {
1997     "tap",
1998
1999     netdev_linux_init,
2000     netdev_linux_run,
2001     netdev_linux_wait,
2002
2003     netdev_linux_create_tap,
2004     netdev_linux_destroy,
2005     NULL,                       /* reconfigure */
2006
2007     netdev_linux_open,
2008     netdev_linux_close,
2009
2010     NULL,                       /* enumerate */
2011
2012     netdev_linux_recv,
2013     netdev_linux_recv_wait,
2014     netdev_linux_drain,
2015
2016     netdev_linux_send,
2017     netdev_linux_send_wait,
2018
2019     netdev_linux_set_etheraddr,
2020     netdev_linux_get_etheraddr,
2021     netdev_linux_get_mtu,
2022     netdev_linux_get_ifindex,
2023     netdev_linux_get_carrier,
2024     netdev_linux_get_stats,
2025
2026     netdev_linux_get_features,
2027     netdev_linux_set_advertisements,
2028     netdev_linux_get_vlan_vid,
2029     netdev_linux_set_policing,
2030
2031     netdev_linux_get_in4,
2032     netdev_linux_set_in4,
2033     netdev_linux_get_in6,
2034     netdev_linux_add_router,
2035     netdev_linux_get_next_hop,
2036     netdev_linux_arp_lookup,
2037
2038     netdev_linux_update_flags,
2039
2040     netdev_linux_poll_add,
2041     netdev_linux_poll_remove,
2042 };
2043
2044 const struct netdev_class netdev_gre_class = {
2045     "gre",
2046
2047     netdev_linux_init,
2048     netdev_linux_run,
2049     netdev_linux_wait,
2050
2051     netdev_linux_create_gre,
2052     netdev_linux_destroy,
2053     netdev_linux_reconfigure_gre,
2054
2055     netdev_linux_open,
2056     netdev_linux_close,
2057
2058     NULL,                       /* enumerate */
2059
2060     netdev_linux_recv,
2061     netdev_linux_recv_wait,
2062     netdev_linux_drain,
2063
2064     netdev_linux_send,
2065     netdev_linux_send_wait,
2066
2067     netdev_linux_set_etheraddr,
2068     netdev_linux_get_etheraddr,
2069     netdev_linux_get_mtu,
2070     netdev_linux_get_ifindex,
2071     netdev_linux_get_carrier,
2072     netdev_linux_get_stats,
2073
2074     netdev_linux_get_features,
2075     netdev_linux_set_advertisements,
2076     netdev_linux_get_vlan_vid,
2077     netdev_linux_set_policing,
2078
2079     netdev_linux_get_in4,
2080     netdev_linux_set_in4,
2081     netdev_linux_get_in6,
2082     netdev_linux_add_router,
2083     netdev_linux_get_next_hop,
2084     netdev_linux_arp_lookup,
2085
2086     netdev_linux_update_flags,
2087
2088     netdev_linux_poll_add,
2089     netdev_linux_poll_remove,
2090 };
2091 \f
2092 static int
2093 get_stats_via_netlink(int ifindex, struct netdev_stats *stats)
2094 {
2095     /* Policy for RTNLGRP_LINK messages.
2096      *
2097      * There are *many* more fields in these messages, but currently we only
2098      * care about these fields. */
2099     static const struct nl_policy rtnlgrp_link_policy[] = {
2100         [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
2101         [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true,
2102                          .min_len = sizeof(struct rtnl_link_stats) },
2103     };
2104
2105
2106     static struct nl_sock *rtnl_sock;
2107     struct ofpbuf request;
2108     struct ofpbuf *reply;
2109     struct ifinfomsg *ifi;
2110     const struct rtnl_link_stats *rtnl_stats;
2111     struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
2112     int error;
2113
2114     if (!rtnl_sock) {
2115         error = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock);
2116         if (error) {
2117             VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s",
2118                         strerror(error));
2119             return error;
2120         }
2121     }
2122
2123     ofpbuf_init(&request, 0);
2124     nl_msg_put_nlmsghdr(&request, rtnl_sock, sizeof *ifi,
2125                         RTM_GETLINK, NLM_F_REQUEST);
2126     ifi = ofpbuf_put_zeros(&request, sizeof *ifi);
2127     ifi->ifi_family = PF_UNSPEC;
2128     ifi->ifi_index = ifindex;
2129     error = nl_sock_transact(rtnl_sock, &request, &reply);
2130     ofpbuf_uninit(&request);
2131     if (error) {
2132         return error;
2133     }
2134
2135     if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
2136                          rtnlgrp_link_policy,
2137                          attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
2138         ofpbuf_delete(reply);
2139         return EPROTO;
2140     }
2141
2142     if (!attrs[IFLA_STATS]) {
2143         VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats");
2144         ofpbuf_delete(reply);
2145         return EPROTO;
2146     }
2147
2148     rtnl_stats = nl_attr_get(attrs[IFLA_STATS]);
2149     stats->rx_packets = rtnl_stats->rx_packets;
2150     stats->tx_packets = rtnl_stats->tx_packets;
2151     stats->rx_bytes = rtnl_stats->rx_bytes;
2152     stats->tx_bytes = rtnl_stats->tx_bytes;
2153     stats->rx_errors = rtnl_stats->rx_errors;
2154     stats->tx_errors = rtnl_stats->tx_errors;
2155     stats->rx_dropped = rtnl_stats->rx_dropped;
2156     stats->tx_dropped = rtnl_stats->tx_dropped;
2157     stats->multicast = rtnl_stats->multicast;
2158     stats->collisions = rtnl_stats->collisions;
2159     stats->rx_length_errors = rtnl_stats->rx_length_errors;
2160     stats->rx_over_errors = rtnl_stats->rx_over_errors;
2161     stats->rx_crc_errors = rtnl_stats->rx_crc_errors;
2162     stats->rx_frame_errors = rtnl_stats->rx_frame_errors;
2163     stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors;
2164     stats->rx_missed_errors = rtnl_stats->rx_missed_errors;
2165     stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors;
2166     stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors;
2167     stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors;
2168     stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors;
2169     stats->tx_window_errors = rtnl_stats->tx_window_errors;
2170
2171     ofpbuf_delete(reply);
2172
2173     return 0;
2174 }
2175
2176 static int
2177 get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats)
2178 {
2179     static const char fn[] = "/proc/net/dev";
2180     char line[1024];
2181     FILE *stream;
2182     int ln;
2183
2184     stream = fopen(fn, "r");
2185     if (!stream) {
2186         VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno));
2187         return errno;
2188     }
2189
2190     ln = 0;
2191     while (fgets(line, sizeof line, stream)) {
2192         if (++ln >= 3) {
2193             char devname[16];
2194 #define X64 "%"SCNu64
2195             if (sscanf(line,
2196                        " %15[^:]:"
2197                        X64 X64 X64 X64 X64 X64 X64 "%*u"
2198                        X64 X64 X64 X64 X64 X64 X64 "%*u",
2199                        devname,
2200                        &stats->rx_bytes,
2201                        &stats->rx_packets,
2202                        &stats->rx_errors,
2203                        &stats->rx_dropped,
2204                        &stats->rx_fifo_errors,
2205                        &stats->rx_frame_errors,
2206                        &stats->multicast,
2207                        &stats->tx_bytes,
2208                        &stats->tx_packets,
2209                        &stats->tx_errors,
2210                        &stats->tx_dropped,
2211                        &stats->tx_fifo_errors,
2212                        &stats->collisions,
2213                        &stats->tx_carrier_errors) != 15) {
2214                 VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln);
2215             } else if (!strcmp(devname, netdev_name)) {
2216                 stats->rx_length_errors = UINT64_MAX;
2217                 stats->rx_over_errors = UINT64_MAX;
2218                 stats->rx_crc_errors = UINT64_MAX;
2219                 stats->rx_missed_errors = UINT64_MAX;
2220                 stats->tx_aborted_errors = UINT64_MAX;
2221                 stats->tx_heartbeat_errors = UINT64_MAX;
2222                 stats->tx_window_errors = UINT64_MAX;
2223                 fclose(stream);
2224                 return 0;
2225             }
2226         }
2227     }
2228     VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name);
2229     fclose(stream);
2230     return ENODEV;
2231 }
2232 \f
2233 static int
2234 get_flags(const struct netdev *netdev, int *flags)
2235 {
2236     struct ifreq ifr;
2237     int error;
2238
2239     error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS,
2240                                   "SIOCGIFFLAGS");
2241     *flags = ifr.ifr_flags;
2242     return error;
2243 }
2244
2245 static int
2246 set_flags(struct netdev *netdev, int flags)
2247 {
2248     struct ifreq ifr;
2249
2250     ifr.ifr_flags = flags;
2251     return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS,
2252                                  "SIOCSIFFLAGS");
2253 }
2254
2255 static int
2256 do_get_ifindex(const char *netdev_name)
2257 {
2258     struct ifreq ifr;
2259
2260     strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2261     COVERAGE_INC(netdev_get_ifindex);
2262     if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) {
2263         VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s",
2264                      netdev_name, strerror(errno));
2265         return -errno;
2266     }
2267     return ifr.ifr_ifindex;
2268 }
2269
2270 static int
2271 get_ifindex(const struct netdev *netdev_, int *ifindexp)
2272 {
2273     struct netdev_dev_linux *netdev_dev =
2274                                 netdev_dev_linux_cast(netdev_get_dev(netdev_));
2275     *ifindexp = 0;
2276     if (!(netdev_dev->cache_valid & VALID_IFINDEX)) {
2277         int ifindex = do_get_ifindex(netdev_get_name(netdev_));
2278         if (ifindex < 0) {
2279             return -ifindex;
2280         }
2281         netdev_dev->cache_valid |= VALID_IFINDEX;
2282         netdev_dev->ifindex = ifindex;
2283     }
2284     *ifindexp = netdev_dev->ifindex;
2285     return 0;
2286 }
2287
2288 static int
2289 get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN])
2290 {
2291     struct ifreq ifr;
2292     int hwaddr_family;
2293
2294     memset(&ifr, 0, sizeof ifr);
2295     strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2296     COVERAGE_INC(netdev_get_hwaddr);
2297     if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) {
2298         VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s",
2299                  netdev_name, strerror(errno));
2300         return errno;
2301     }
2302     hwaddr_family = ifr.ifr_hwaddr.sa_family;
2303     if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) {
2304         VLOG_WARN("%s device has unknown hardware address family %d",
2305                   netdev_name, hwaddr_family);
2306     }
2307     memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN);
2308     return 0;
2309 }
2310
2311 static int
2312 set_etheraddr(const char *netdev_name, int hwaddr_family,
2313               const uint8_t mac[ETH_ADDR_LEN])
2314 {
2315     struct ifreq ifr;
2316
2317     memset(&ifr, 0, sizeof ifr);
2318     strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name);
2319     ifr.ifr_hwaddr.sa_family = hwaddr_family;
2320     memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN);
2321     COVERAGE_INC(netdev_set_hwaddr);
2322     if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) {
2323         VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s",
2324                  netdev_name, strerror(errno));
2325         return errno;
2326     }
2327     return 0;
2328 }
2329
2330 static int
2331 netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd,
2332                         int cmd, const char *cmd_name)
2333 {
2334     struct ifreq ifr;
2335
2336     memset(&ifr, 0, sizeof ifr);
2337     strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name);
2338     ifr.ifr_data = (caddr_t) ecmd;
2339
2340     ecmd->cmd = cmd;
2341     COVERAGE_INC(netdev_ethtool);
2342     if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) {
2343         return 0;
2344     } else {
2345         if (errno != EOPNOTSUPP) {
2346             VLOG_WARN_RL(&rl, "ethtool command %s on network device %s "
2347                          "failed: %s", cmd_name, name, strerror(errno));
2348         } else {
2349             /* The device doesn't support this operation.  That's pretty
2350              * common, so there's no point in logging anything. */
2351         }
2352         return errno;
2353     }
2354 }
2355
2356 static int
2357 netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd,
2358                       const char *cmd_name)
2359 {
2360     strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
2361     if (ioctl(af_inet_sock, cmd, ifr) == -1) {
2362         VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
2363                      strerror(errno));
2364         return errno;
2365     }
2366     return 0;
2367 }
2368
2369 static int
2370 netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip,
2371                       int cmd, const char *cmd_name)
2372 {
2373     struct ifreq ifr;
2374     int error;
2375
2376     ifr.ifr_addr.sa_family = AF_INET;
2377     error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name);
2378     if (!error) {
2379         const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
2380         *ip = sin->sin_addr;
2381     }
2382     return error;
2383 }