drivers/net/ipvlan/ipvlan_main.c

   1 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of the GNU General Public License as
   5  * published by the Free Software Foundation; either version 2 of
   6  * the License, or (at your option) any later version.
   7  *
   8  */
   9
  10 #include "ipvlan.h"
  11
  12 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
  13 {
  14         ipvlan->dev->mtu = dev->mtu - ipvlan->mtu_adj;
  15 }
  16
  17 static void ipvlan_set_port_mode(struct ipvl_port *port, u16 nval)
  18 {
  19         struct ipvl_dev *ipvlan;
  20
  21         if (port->mode != nval) {
  22                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
  23                         if (nval == IPVLAN_MODE_L3)
  24                                 ipvlan->dev->flags |= IFF_NOARP;
  25                         else
  26                                 ipvlan->dev->flags &= ~IFF_NOARP;
  27                 }
  28                 port->mode = nval;
  29         }
  30 }
  31
  32 static int ipvlan_port_create(struct net_device *dev)
  33 {
  34         struct ipvl_port *port;
  35         int err, idx;
  36
  37         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) {
  38                 netdev_err(dev, "Master is either lo or non-ether device\n");
  39                 return -EINVAL;
  40         }
  41
  42         if (netif_is_macvlan_port(dev)) {
  43                 netdev_err(dev, "Master is a macvlan port.\n");
  44                 return -EBUSY;
  45         }
  46
  47         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
  48         if (!port)
  49                 return -ENOMEM;
  50
  51         port->dev = dev;
  52         port->mode = IPVLAN_MODE_L3;
  53         INIT_LIST_HEAD(&port->ipvlans);
  54         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
  55                 INIT_HLIST_HEAD(&port->hlhead[idx]);
  56
  57         skb_queue_head_init(&port->backlog);
  58         INIT_WORK(&port->wq, ipvlan_process_multicast);
  59
  60         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
  61         if (err)
  62                 goto err;
  63
  64         dev->priv_flags |= IFF_IPVLAN_MASTER;
  65         return 0;
  66
  67 err:
  68         kfree_rcu(port, rcu);
  69         return err;
  70 }
  71
  72 static void ipvlan_port_destroy(struct net_device *dev)
  73 {
  74         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
  75
  76         dev->priv_flags &= ~IFF_IPVLAN_MASTER;
  77         netdev_rx_handler_unregister(dev);
  78         cancel_work_sync(&port->wq);
  79         __skb_queue_purge(&port->backlog);
  80         kfree_rcu(port, rcu);
  81 }
  82
  83 /* ipvlan network devices have devices nesting below it and are a special
  84  * "super class" of normal network devices; split their locks off into a
  85  * separate class since they always nest.
  86  */
  87 static struct lock_class_key ipvlan_netdev_xmit_lock_key;
  88 static struct lock_class_key ipvlan_netdev_addr_lock_key;
  89
  90 #define IPVLAN_FEATURES \
  91         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
  92          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
  93          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
  94          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
  95
  96 #define IPVLAN_STATE_MASK \
  97         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
  98
  99 static void ipvlan_set_lockdep_class_one(struct net_device *dev,
 100                                          struct netdev_queue *txq,
 101                                          void *_unused)
 102 {
 103         lockdep_set_class(&txq->_xmit_lock, &ipvlan_netdev_xmit_lock_key);
 104 }
 105
 106 static void ipvlan_set_lockdep_class(struct net_device *dev)
 107 {
 108         lockdep_set_class(&dev->addr_list_lock, &ipvlan_netdev_addr_lock_key);
 109         netdev_for_each_tx_queue(dev, ipvlan_set_lockdep_class_one, NULL);
 110 }
 111
 112 static int ipvlan_init(struct net_device *dev)
 113 {
 114         struct ipvl_dev *ipvlan = netdev_priv(dev);
 115         const struct net_device *phy_dev = ipvlan->phy_dev;
 116         struct ipvl_port *port = ipvlan->port;
 117
 118         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 119                      (phy_dev->state & IPVLAN_STATE_MASK);
 120         dev->features = phy_dev->features & IPVLAN_FEATURES;
 121         dev->features |= NETIF_F_LLTX;
 122         dev->gso_max_size = phy_dev->gso_max_size;
 123         dev->gso_max_segs = phy_dev->gso_max_segs;
 124         dev->hard_header_len = phy_dev->hard_header_len;
 125
 126         ipvlan_set_lockdep_class(dev);
 127
 128         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 129         if (!ipvlan->pcpu_stats)
 130                 return -ENOMEM;
 131
 132         port->count += 1;
 133
 134         return 0;
 135 }
 136
 137 static void ipvlan_uninit(struct net_device *dev)
 138 {
 139         struct ipvl_dev *ipvlan = netdev_priv(dev);
 140         struct ipvl_port *port = ipvlan->port;
 141
 142         free_percpu(ipvlan->pcpu_stats);
 143
 144         port->count -= 1;
 145         if (!port->count)
 146                 ipvlan_port_destroy(port->dev);
 147 }
 148
 149 static int ipvlan_open(struct net_device *dev)
 150 {
 151         struct ipvl_dev *ipvlan = netdev_priv(dev);
 152         struct net_device *phy_dev = ipvlan->phy_dev;
 153         struct ipvl_addr *addr;
 154
 155         if (ipvlan->port->mode == IPVLAN_MODE_L3)
 156                 dev->flags |= IFF_NOARP;
 157         else
 158                 dev->flags &= ~IFF_NOARP;
 159
 160         list_for_each_entry(addr, &ipvlan->addrs, anode)
 161                 ipvlan_ht_addr_add(ipvlan, addr);
 162
 163         return dev_uc_add(phy_dev, phy_dev->dev_addr);
 164 }
 165
 166 static int ipvlan_stop(struct net_device *dev)
 167 {
 168         struct ipvl_dev *ipvlan = netdev_priv(dev);
 169         struct net_device *phy_dev = ipvlan->phy_dev;
 170         struct ipvl_addr *addr;
 171
 172         dev_uc_unsync(phy_dev, dev);
 173         dev_mc_unsync(phy_dev, dev);
 174
 175         dev_uc_del(phy_dev, phy_dev->dev_addr);
 176
 177         list_for_each_entry(addr, &ipvlan->addrs, anode)
 178                 ipvlan_ht_addr_del(addr);
 179
 180         return 0;
 181 }
 182
 183 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 184                                      struct net_device *dev)
 185 {
 186         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 187         int skblen = skb->len;
 188         int ret;
 189
 190         ret = ipvlan_queue_xmit(skb, dev);
 191         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 192                 struct ipvl_pcpu_stats *pcptr;
 193
 194                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 195
 196                 u64_stats_update_begin(&pcptr->syncp);
 197                 pcptr->tx_pkts++;
 198                 pcptr->tx_bytes += skblen;
 199                 u64_stats_update_end(&pcptr->syncp);
 200         } else {
 201                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 202         }
 203         return ret;
 204 }
 205
 206 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 207                                              netdev_features_t features)
 208 {
 209         struct ipvl_dev *ipvlan = netdev_priv(dev);
 210
 211         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 212 }
 213
 214 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 215 {
 216         struct ipvl_dev *ipvlan = netdev_priv(dev);
 217         struct net_device *phy_dev = ipvlan->phy_dev;
 218
 219         if (change & IFF_ALLMULTI)
 220                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 221 }
 222
 223 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 224 {
 225         struct ipvl_dev *ipvlan = netdev_priv(dev);
 226
 227         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 228                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 229         } else {
 230                 struct netdev_hw_addr *ha;
 231                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 232
 233                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 234                 netdev_for_each_mc_addr(ha, dev)
 235                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 236
 237                 /* Turn-on broadcast bit irrespective of address family,
 238                  * since broadcast is deferred to a work-queue, hence no
 239                  * impact on fast-path processing.
 240                  */
 241                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 242
 243                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 244                             IPVLAN_MAC_FILTER_SIZE);
 245         }
 246         dev_uc_sync(ipvlan->phy_dev, dev);
 247         dev_mc_sync(ipvlan->phy_dev, dev);
 248 }
 249
 250 static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
 251                                                     struct rtnl_link_stats64 *s)
 252 {
 253         struct ipvl_dev *ipvlan = netdev_priv(dev);
 254
 255         if (ipvlan->pcpu_stats) {
 256                 struct ipvl_pcpu_stats *pcptr;
 257                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 258                 u32 rx_errs = 0, tx_drps = 0;
 259                 u32 strt;
 260                 int idx;
 261
 262                 for_each_possible_cpu(idx) {
 263                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 264                         do {
 265                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 266                                 rx_pkts = pcptr->rx_pkts;
 267                                 rx_bytes = pcptr->rx_bytes;
 268                                 rx_mcast = pcptr->rx_mcast;
 269                                 tx_pkts = pcptr->tx_pkts;
 270                                 tx_bytes = pcptr->tx_bytes;
 271                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 272                                                            strt));
 273
 274                         s->rx_packets += rx_pkts;
 275                         s->rx_bytes += rx_bytes;
 276                         s->multicast += rx_mcast;
 277                         s->tx_packets += tx_pkts;
 278                         s->tx_bytes += tx_bytes;
 279
 280                         /* u32 values are updated without syncp protection. */
 281                         rx_errs += pcptr->rx_errs;
 282                         tx_drps += pcptr->tx_drps;
 283                 }
 284                 s->rx_errors = rx_errs;
 285                 s->rx_dropped = rx_errs;
 286                 s->tx_dropped = tx_drps;
 287         }
 288         return s;
 289 }
 290
 291 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 292 {
 293         struct ipvl_dev *ipvlan = netdev_priv(dev);
 294         struct net_device *phy_dev = ipvlan->phy_dev;
 295
 296         return vlan_vid_add(phy_dev, proto, vid);
 297 }
 298
 299 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 300                                    u16 vid)
 301 {
 302         struct ipvl_dev *ipvlan = netdev_priv(dev);
 303         struct net_device *phy_dev = ipvlan->phy_dev;
 304
 305         vlan_vid_del(phy_dev, proto, vid);
 306         return 0;
 307 }
 308
 309 static int ipvlan_get_iflink(const struct net_device *dev)
 310 {
 311         struct ipvl_dev *ipvlan = netdev_priv(dev);
 312
 313         return ipvlan->phy_dev->ifindex;
 314 }
 315
 316 static const struct net_device_ops ipvlan_netdev_ops = {
 317         .ndo_init               = ipvlan_init,
 318         .ndo_uninit             = ipvlan_uninit,
 319         .ndo_open               = ipvlan_open,
 320         .ndo_stop               = ipvlan_stop,
 321         .ndo_start_xmit         = ipvlan_start_xmit,
 322         .ndo_fix_features       = ipvlan_fix_features,
 323         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 324         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 325         .ndo_get_stats64        = ipvlan_get_stats64,
 326         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 327         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 328         .ndo_get_iflink         = ipvlan_get_iflink,
 329 };
 330
 331 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 332                               unsigned short type, const void *daddr,
 333                               const void *saddr, unsigned len)
 334 {
 335         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 336         struct net_device *phy_dev = ipvlan->phy_dev;
 337
 338         /* TODO Probably use a different field than dev_addr so that the
 339          * mac-address on the virtual device is portable and can be carried
 340          * while the packets use the mac-addr on the physical device.
 341          */
 342         return dev_hard_header(skb, phy_dev, type, daddr,
 343                                saddr ? : dev->dev_addr, len);
 344 }
 345
 346 static const struct header_ops ipvlan_header_ops = {
 347         .create         = ipvlan_hard_header,
 348         .parse          = eth_header_parse,
 349         .cache          = eth_header_cache,
 350         .cache_update   = eth_header_cache_update,
 351 };
 352
 353 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 354                                              struct ethtool_link_ksettings *cmd)
 355 {
 356         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 357
 358         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 359 }
 360
 361 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 362                                        struct ethtool_drvinfo *drvinfo)
 363 {
 364         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 365         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 366 }
 367
 368 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 369 {
 370         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 371
 372         return ipvlan->msg_enable;
 373 }
 374
 375 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 376 {
 377         struct ipvl_dev *ipvlan = netdev_priv(dev);
 378
 379         ipvlan->msg_enable = value;
 380 }
 381
 382 static const struct ethtool_ops ipvlan_ethtool_ops = {
 383         .get_link       = ethtool_op_get_link,
 384         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 385         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 386         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 387         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 388 };
 389
 390 static int ipvlan_nl_changelink(struct net_device *dev,
 391                                 struct nlattr *tb[], struct nlattr *data[])
 392 {
 393         struct ipvl_dev *ipvlan = netdev_priv(dev);
 394         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 395
 396         if (data && data[IFLA_IPVLAN_MODE]) {
 397                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 398
 399                 ipvlan_set_port_mode(port, nmode);
 400         }
 401         return 0;
 402 }
 403
 404 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 405 {
 406         return (0
 407                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 408                 );
 409 }
 410
 411 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 412 {
 413         if (data && data[IFLA_IPVLAN_MODE]) {
 414                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 415
 416                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 417                         return -EINVAL;
 418         }
 419         return 0;
 420 }
 421
 422 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 423                               const struct net_device *dev)
 424 {
 425         struct ipvl_dev *ipvlan = netdev_priv(dev);
 426         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 427         int ret = -EINVAL;
 428
 429         if (!port)
 430                 goto err;
 431
 432         ret = -EMSGSIZE;
 433         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 434                 goto err;
 435
 436         return 0;
 437
 438 err:
 439         return ret;
 440 }
 441
 442 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 443                            struct nlattr *tb[], struct nlattr *data[])
 444 {
 445         struct ipvl_dev *ipvlan = netdev_priv(dev);
 446         struct ipvl_port *port;
 447         struct net_device *phy_dev;
 448         int err;
 449         u16 mode = IPVLAN_MODE_L3;
 450
 451         if (!tb[IFLA_LINK])
 452                 return -EINVAL;
 453
 454         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 455         if (!phy_dev)
 456                 return -ENODEV;
 457
 458         if (netif_is_ipvlan(phy_dev)) {
 459                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 460
 461                 phy_dev = tmp->phy_dev;
 462         } else if (!netif_is_ipvlan_port(phy_dev)) {
 463                 err = ipvlan_port_create(phy_dev);
 464                 if (err < 0)
 465                         return err;
 466         }
 467
 468         if (data && data[IFLA_IPVLAN_MODE])
 469                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 470
 471         port = ipvlan_port_get_rtnl(phy_dev);
 472         ipvlan->phy_dev = phy_dev;
 473         ipvlan->dev = dev;
 474         ipvlan->port = port;
 475         ipvlan->sfeatures = IPVLAN_FEATURES;
 476         ipvlan_adjust_mtu(ipvlan, phy_dev);
 477         INIT_LIST_HEAD(&ipvlan->addrs);
 478
 479         /* TODO Probably put random address here to be presented to the
 480          * world but keep using the physical-dev address for the outgoing
 481          * packets.
 482          */
 483         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 484
 485         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 486
 487         err = register_netdevice(dev);
 488         if (err < 0)
 489                 return err;
 490
 491         err = netdev_upper_dev_link(phy_dev, dev);
 492         if (err) {
 493                 unregister_netdevice(dev);
 494                 return err;
 495         }
 496
 497         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 498         ipvlan_set_port_mode(port, mode);
 499
 500         netif_stacked_transfer_operstate(phy_dev, dev);
 501         return 0;
 502 }
 503
 504 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 505 {
 506         struct ipvl_dev *ipvlan = netdev_priv(dev);
 507         struct ipvl_addr *addr, *next;
 508
 509         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 510                 ipvlan_ht_addr_del(addr);
 511                 list_del(&addr->anode);
 512                 kfree_rcu(addr, rcu);
 513         }
 514
 515         list_del_rcu(&ipvlan->pnode);
 516         unregister_netdevice_queue(dev, head);
 517         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 518 }
 519
 520 static void ipvlan_link_setup(struct net_device *dev)
 521 {
 522         ether_setup(dev);
 523
 524         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 525         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 526         dev->netdev_ops = &ipvlan_netdev_ops;
 527         dev->destructor = free_netdev;
 528         dev->header_ops = &ipvlan_header_ops;
 529         dev->ethtool_ops = &ipvlan_ethtool_ops;
 530 }
 531
 532 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 533 {
 534         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 535 };
 536
 537 static struct rtnl_link_ops ipvlan_link_ops = {
 538         .kind           = "ipvlan",
 539         .priv_size      = sizeof(struct ipvl_dev),
 540
 541         .get_size       = ipvlan_nl_getsize,
 542         .policy         = ipvlan_nl_policy,
 543         .validate       = ipvlan_nl_validate,
 544         .fill_info      = ipvlan_nl_fillinfo,
 545         .changelink     = ipvlan_nl_changelink,
 546         .maxtype        = IFLA_IPVLAN_MAX,
 547
 548         .setup          = ipvlan_link_setup,
 549         .newlink        = ipvlan_link_new,
 550         .dellink        = ipvlan_link_delete,
 551 };
 552
 553 static int ipvlan_link_register(struct rtnl_link_ops *ops)
 554 {
 555         return rtnl_link_register(ops);
 556 }
 557
 558 static int ipvlan_device_event(struct notifier_block *unused,
 559                                unsigned long event, void *ptr)
 560 {
 561         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 562         struct ipvl_dev *ipvlan, *next;
 563         struct ipvl_port *port;
 564         LIST_HEAD(lst_kill);
 565
 566         if (!netif_is_ipvlan_port(dev))
 567                 return NOTIFY_DONE;
 568
 569         port = ipvlan_port_get_rtnl(dev);
 570
 571         switch (event) {
 572         case NETDEV_CHANGE:
 573                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 574                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 575                                                          ipvlan->dev);
 576                 break;
 577
 578         case NETDEV_UNREGISTER:
 579                 if (dev->reg_state != NETREG_UNREGISTERING)
 580                         break;
 581
 582                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 583                                          pnode)
 584                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 585                                                             &lst_kill);
 586                 unregister_netdevice_many(&lst_kill);
 587                 break;
 588
 589         case NETDEV_FEAT_CHANGE:
 590                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 591                         ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
 592                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 593                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 594                         netdev_features_change(ipvlan->dev);
 595                 }
 596                 break;
 597
 598         case NETDEV_CHANGEMTU:
 599                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 600                         ipvlan_adjust_mtu(ipvlan, dev);
 601                 break;
 602
 603         case NETDEV_PRE_TYPE_CHANGE:
 604                 /* Forbid underlying device to change its type. */
 605                 return NOTIFY_BAD;
 606         }
 607         return NOTIFY_DONE;
 608 }
 609
 610 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 611 {
 612         struct ipvl_addr *addr;
 613
 614         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 615                 netif_err(ipvlan, ifup, ipvlan->dev,
 616                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 617                           ip6_addr, ipvlan->dev->name);
 618                 return -EINVAL;
 619         }
 620         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 621         if (!addr)
 622                 return -ENOMEM;
 623
 624         addr->master = ipvlan;
 625         memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
 626         addr->atype = IPVL_IPV6;
 627         list_add_tail(&addr->anode, &ipvlan->addrs);
 628
 629         /* If the interface is not up, the address will be added to the hash
 630          * list by ipvlan_open.
 631          */
 632         if (netif_running(ipvlan->dev))
 633                 ipvlan_ht_addr_add(ipvlan, addr);
 634
 635         return 0;
 636 }
 637
 638 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 639 {
 640         struct ipvl_addr *addr;
 641
 642         addr = ipvlan_find_addr(ipvlan, ip6_addr, true);
 643         if (!addr)
 644                 return;
 645
 646         ipvlan_ht_addr_del(addr);
 647         list_del(&addr->anode);
 648         kfree_rcu(addr, rcu);
 649
 650         return;
 651 }
 652
 653 static int ipvlan_addr6_event(struct notifier_block *unused,
 654                               unsigned long event, void *ptr)
 655 {
 656         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 657         struct net_device *dev = (struct net_device *)if6->idev->dev;
 658         struct ipvl_dev *ipvlan = netdev_priv(dev);
 659
 660         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 661         if (in_softirq())
 662                 return NOTIFY_DONE;
 663
 664         if (!netif_is_ipvlan(dev))
 665                 return NOTIFY_DONE;
 666
 667         if (!ipvlan || !ipvlan->port)
 668                 return NOTIFY_DONE;
 669
 670         switch (event) {
 671         case NETDEV_UP:
 672                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 673                         return NOTIFY_BAD;
 674                 break;
 675
 676         case NETDEV_DOWN:
 677                 ipvlan_del_addr6(ipvlan, &if6->addr);
 678                 break;
 679         }
 680
 681         return NOTIFY_OK;
 682 }
 683
 684 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 685 {
 686         struct ipvl_addr *addr;
 687
 688         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 689                 netif_err(ipvlan, ifup, ipvlan->dev,
 690                           "Failed to add IPv4=%pI4 on %s intf.\n",
 691                           ip4_addr, ipvlan->dev->name);
 692                 return -EINVAL;
 693         }
 694         addr = kzalloc(sizeof(struct ipvl_addr), GFP_KERNEL);
 695         if (!addr)
 696                 return -ENOMEM;
 697
 698         addr->master = ipvlan;
 699         memcpy(&addr->ip4addr, ip4_addr, sizeof(struct in_addr));
 700         addr->atype = IPVL_IPV4;
 701         list_add_tail(&addr->anode, &ipvlan->addrs);
 702
 703         /* If the interface is not up, the address will be added to the hash
 704          * list by ipvlan_open.
 705          */
 706         if (netif_running(ipvlan->dev))
 707                 ipvlan_ht_addr_add(ipvlan, addr);
 708
 709         return 0;
 710 }
 711
 712 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 713 {
 714         struct ipvl_addr *addr;
 715
 716         addr = ipvlan_find_addr(ipvlan, ip4_addr, false);
 717         if (!addr)
 718                 return;
 719
 720         ipvlan_ht_addr_del(addr);
 721         list_del(&addr->anode);
 722         kfree_rcu(addr, rcu);
 723
 724         return;
 725 }
 726
 727 static int ipvlan_addr4_event(struct notifier_block *unused,
 728                               unsigned long event, void *ptr)
 729 {
 730         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 731         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 732         struct ipvl_dev *ipvlan = netdev_priv(dev);
 733         struct in_addr ip4_addr;
 734
 735         if (!netif_is_ipvlan(dev))
 736                 return NOTIFY_DONE;
 737
 738         if (!ipvlan || !ipvlan->port)
 739                 return NOTIFY_DONE;
 740
 741         switch (event) {
 742         case NETDEV_UP:
 743                 ip4_addr.s_addr = if4->ifa_address;
 744                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 745                         return NOTIFY_BAD;
 746                 break;
 747
 748         case NETDEV_DOWN:
 749                 ip4_addr.s_addr = if4->ifa_address;
 750                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 751                 break;
 752         }
 753
 754         return NOTIFY_OK;
 755 }
 756
 757 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 758         .notifier_call = ipvlan_addr4_event,
 759 };
 760
 761 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 762         .notifier_call = ipvlan_device_event,
 763 };
 764
 765 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 766         .notifier_call = ipvlan_addr6_event,
 767 };
 768
 769 static int __init ipvlan_init_module(void)
 770 {
 771         int err;
 772
 773         ipvlan_init_secret();
 774         register_netdevice_notifier(&ipvlan_notifier_block);
 775         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 776         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 777
 778         err = ipvlan_link_register(&ipvlan_link_ops);
 779         if (err < 0)
 780                 goto error;
 781
 782         return 0;
 783 error:
 784         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 785         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 786         unregister_netdevice_notifier(&ipvlan_notifier_block);
 787         return err;
 788 }
 789
 790 static void __exit ipvlan_cleanup_module(void)
 791 {
 792         rtnl_link_unregister(&ipvlan_link_ops);
 793         unregister_netdevice_notifier(&ipvlan_notifier_block);
 794         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 795         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 796 }
 797
 798 module_init(ipvlan_init_module);
 799 module_exit(ipvlan_cleanup_module);
 800
 801 MODULE_LICENSE("GPL");
 802 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 803 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 804 MODULE_ALIAS_RTNL_LINK("ipvlan");