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 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 void ipvlan_set_port_mode(struct ipvl_port *port, u32 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         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
  42         if (!port)
  43                 return -ENOMEM;
  44
  45         port->dev = dev;
  46         port->mode = IPVLAN_MODE_L3;
  47         INIT_LIST_HEAD(&port->ipvlans);
  48         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
  49                 INIT_HLIST_HEAD(&port->hlhead[idx]);
  50
  51         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
  52         if (err)
  53                 goto err;
  54
  55         dev->priv_flags |= IFF_IPVLAN_MASTER;
  56         return 0;
  57
  58 err:
  59         kfree_rcu(port, rcu);
  60         return err;
  61 }
  62
  63 static void ipvlan_port_destroy(struct net_device *dev)
  64 {
  65         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
  66
  67         dev->priv_flags &= ~IFF_IPVLAN_MASTER;
  68         netdev_rx_handler_unregister(dev);
  69         kfree_rcu(port, rcu);
  70 }
  71
  72 /* ipvlan network devices have devices nesting below it and are a special
  73  * "super class" of normal network devices; split their locks off into a
  74  * separate class since they always nest.
  75  */
  76 static struct lock_class_key ipvlan_netdev_xmit_lock_key;
  77 static struct lock_class_key ipvlan_netdev_addr_lock_key;
  78
  79 #define IPVLAN_FEATURES \
  80         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
  81          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
  82          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
  83          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
  84
  85 #define IPVLAN_STATE_MASK \
  86         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
  87
  88 static void ipvlan_set_lockdep_class_one(struct net_device *dev,
  89                                          struct netdev_queue *txq,
  90                                          void *_unused)
  91 {
  92         lockdep_set_class(&txq->_xmit_lock, &ipvlan_netdev_xmit_lock_key);
  93 }
  94
  95 static void ipvlan_set_lockdep_class(struct net_device *dev)
  96 {
  97         lockdep_set_class(&dev->addr_list_lock, &ipvlan_netdev_addr_lock_key);
  98         netdev_for_each_tx_queue(dev, ipvlan_set_lockdep_class_one, NULL);
  99 }
 100
 101 static int ipvlan_init(struct net_device *dev)
 102 {
 103         struct ipvl_dev *ipvlan = netdev_priv(dev);
 104         const struct net_device *phy_dev = ipvlan->phy_dev;
 105
 106         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 107                      (phy_dev->state & IPVLAN_STATE_MASK);
 108         dev->features = phy_dev->features & IPVLAN_FEATURES;
 109         dev->features |= NETIF_F_LLTX;
 110         dev->gso_max_size = phy_dev->gso_max_size;
 111         dev->iflink = phy_dev->ifindex;
 112         dev->hard_header_len = phy_dev->hard_header_len;
 113
 114         ipvlan_set_lockdep_class(dev);
 115
 116         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 117         if (!ipvlan->pcpu_stats)
 118                 return -ENOMEM;
 119
 120         return 0;
 121 }
 122
 123 static void ipvlan_uninit(struct net_device *dev)
 124 {
 125         struct ipvl_dev *ipvlan = netdev_priv(dev);
 126         struct ipvl_port *port = ipvlan->port;
 127
 128         if (ipvlan->pcpu_stats)
 129                 free_percpu(ipvlan->pcpu_stats);
 130
 131         port->count -= 1;
 132         if (!port->count)
 133                 ipvlan_port_destroy(port->dev);
 134 }
 135
 136 static int ipvlan_open(struct net_device *dev)
 137 {
 138         struct ipvl_dev *ipvlan = netdev_priv(dev);
 139         struct net_device *phy_dev = ipvlan->phy_dev;
 140         struct ipvl_addr *addr;
 141
 142         if (ipvlan->port->mode == IPVLAN_MODE_L3)
 143                 dev->flags |= IFF_NOARP;
 144         else
 145                 dev->flags &= ~IFF_NOARP;
 146
 147         if (ipvlan->ipv6cnt > 0 || ipvlan->ipv4cnt > 0) {
 148                 list_for_each_entry(addr, &ipvlan->addrs, anode)
 149                         ipvlan_ht_addr_add(ipvlan, addr);
 150         }
 151         return dev_uc_add(phy_dev, phy_dev->dev_addr);
 152 }
 153
 154 static int ipvlan_stop(struct net_device *dev)
 155 {
 156         struct ipvl_dev *ipvlan = netdev_priv(dev);
 157         struct net_device *phy_dev = ipvlan->phy_dev;
 158         struct ipvl_addr *addr;
 159
 160         dev_uc_unsync(phy_dev, dev);
 161         dev_mc_unsync(phy_dev, dev);
 162
 163         dev_uc_del(phy_dev, phy_dev->dev_addr);
 164
 165         if (ipvlan->ipv6cnt > 0 || ipvlan->ipv4cnt > 0) {
 166                 list_for_each_entry(addr, &ipvlan->addrs, anode)
 167                         ipvlan_ht_addr_del(addr, !dev->dismantle);
 168         }
 169         return 0;
 170 }
 171
 172 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 173                                      struct net_device *dev)
 174 {
 175         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 176         int skblen = skb->len;
 177         int ret;
 178
 179         ret = ipvlan_queue_xmit(skb, dev);
 180         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 181                 struct ipvl_pcpu_stats *pcptr;
 182
 183                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 184
 185                 u64_stats_update_begin(&pcptr->syncp);
 186                 pcptr->tx_pkts++;
 187                 pcptr->tx_bytes += skblen;
 188                 u64_stats_update_end(&pcptr->syncp);
 189         } else {
 190                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 191         }
 192         return ret;
 193 }
 194
 195 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 196                                              netdev_features_t features)
 197 {
 198         struct ipvl_dev *ipvlan = netdev_priv(dev);
 199
 200         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 201 }
 202
 203 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 204 {
 205         struct ipvl_dev *ipvlan = netdev_priv(dev);
 206         struct net_device *phy_dev = ipvlan->phy_dev;
 207
 208         if (change & IFF_ALLMULTI)
 209                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 210 }
 211
 212 static void ipvlan_set_broadcast_mac_filter(struct ipvl_dev *ipvlan, bool set)
 213 {
 214         struct net_device *dev = ipvlan->dev;
 215         unsigned int hashbit = ipvlan_mac_hash(dev->broadcast);
 216
 217         if (set && !test_bit(hashbit, ipvlan->mac_filters))
 218                 __set_bit(hashbit, ipvlan->mac_filters);
 219         else if (!set && test_bit(hashbit, ipvlan->mac_filters))
 220                 __clear_bit(hashbit, ipvlan->mac_filters);
 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                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 238                             IPVLAN_MAC_FILTER_SIZE);
 239         }
 240         dev_uc_sync(ipvlan->phy_dev, dev);
 241         dev_mc_sync(ipvlan->phy_dev, dev);
 242 }
 243
 244 static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
 245                                                     struct rtnl_link_stats64 *s)
 246 {
 247         struct ipvl_dev *ipvlan = netdev_priv(dev);
 248
 249         if (ipvlan->pcpu_stats) {
 250                 struct ipvl_pcpu_stats *pcptr;
 251                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 252                 u32 rx_errs = 0, tx_drps = 0;
 253                 u32 strt;
 254                 int idx;
 255
 256                 for_each_possible_cpu(idx) {
 257                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 258                         do {
 259                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 260                                 rx_pkts = pcptr->rx_pkts;
 261                                 rx_bytes = pcptr->rx_bytes;
 262                                 rx_mcast = pcptr->rx_mcast;
 263                                 tx_pkts = pcptr->tx_pkts;
 264                                 tx_bytes = pcptr->tx_bytes;
 265                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 266                                                            strt));
 267
 268                         s->rx_packets += rx_pkts;
 269                         s->rx_bytes += rx_bytes;
 270                         s->multicast += rx_mcast;
 271                         s->tx_packets += tx_pkts;
 272                         s->tx_bytes += tx_bytes;
 273
 274                         /* u32 values are updated without syncp protection. */
 275                         rx_errs += pcptr->rx_errs;
 276                         tx_drps += pcptr->tx_drps;
 277                 }
 278                 s->rx_errors = rx_errs;
 279                 s->rx_dropped = rx_errs;
 280                 s->tx_dropped = tx_drps;
 281         }
 282         return s;
 283 }
 284
 285 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 286 {
 287         struct ipvl_dev *ipvlan = netdev_priv(dev);
 288         struct net_device *phy_dev = ipvlan->phy_dev;
 289
 290         return vlan_vid_add(phy_dev, proto, vid);
 291 }
 292
 293 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 294                                    u16 vid)
 295 {
 296         struct ipvl_dev *ipvlan = netdev_priv(dev);
 297         struct net_device *phy_dev = ipvlan->phy_dev;
 298
 299         vlan_vid_del(phy_dev, proto, vid);
 300         return 0;
 301 }
 302
 303 static const struct net_device_ops ipvlan_netdev_ops = {
 304         .ndo_init               = ipvlan_init,
 305         .ndo_uninit             = ipvlan_uninit,
 306         .ndo_open               = ipvlan_open,
 307         .ndo_stop               = ipvlan_stop,
 308         .ndo_start_xmit         = ipvlan_start_xmit,
 309         .ndo_fix_features       = ipvlan_fix_features,
 310         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 311         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 312         .ndo_get_stats64        = ipvlan_get_stats64,
 313         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 314         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 315 };
 316
 317 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 318                               unsigned short type, const void *daddr,
 319                               const void *saddr, unsigned len)
 320 {
 321         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 322         struct net_device *phy_dev = ipvlan->phy_dev;
 323
 324         /* TODO Probably use a different field than dev_addr so that the
 325          * mac-address on the virtual device is portable and can be carried
 326          * while the packets use the mac-addr on the physical device.
 327          */
 328         return dev_hard_header(skb, phy_dev, type, daddr,
 329                                saddr ? : dev->dev_addr, len);
 330 }
 331
 332 static const struct header_ops ipvlan_header_ops = {
 333         .create         = ipvlan_hard_header,
 334         .rebuild        = eth_rebuild_header,
 335         .parse          = eth_header_parse,
 336         .cache          = eth_header_cache,
 337         .cache_update   = eth_header_cache_update,
 338 };
 339
 340 static int ipvlan_ethtool_get_settings(struct net_device *dev,
 341                                        struct ethtool_cmd *cmd)
 342 {
 343         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 344
 345         return __ethtool_get_settings(ipvlan->phy_dev, cmd);
 346 }
 347
 348 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 349                                        struct ethtool_drvinfo *drvinfo)
 350 {
 351         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 352         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 353 }
 354
 355 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 356 {
 357         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 358
 359         return ipvlan->msg_enable;
 360 }
 361
 362 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 363 {
 364         struct ipvl_dev *ipvlan = netdev_priv(dev);
 365
 366         ipvlan->msg_enable = value;
 367 }
 368
 369 static const struct ethtool_ops ipvlan_ethtool_ops = {
 370         .get_link       = ethtool_op_get_link,
 371         .get_settings   = ipvlan_ethtool_get_settings,
 372         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 373         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 374         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 375 };
 376
 377 static int ipvlan_nl_changelink(struct net_device *dev,
 378                                 struct nlattr *tb[], struct nlattr *data[])
 379 {
 380         struct ipvl_dev *ipvlan = netdev_priv(dev);
 381         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 382
 383         if (data && data[IFLA_IPVLAN_MODE]) {
 384                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 385
 386                 ipvlan_set_port_mode(port, nmode);
 387         }
 388         return 0;
 389 }
 390
 391 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 392 {
 393         return (0
 394                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 395                 );
 396 }
 397
 398 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 399 {
 400         if (data && data[IFLA_IPVLAN_MODE]) {
 401                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 402
 403                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 404                         return -EINVAL;
 405         }
 406         return 0;
 407 }
 408
 409 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 410                               const struct net_device *dev)
 411 {
 412         struct ipvl_dev *ipvlan = netdev_priv(dev);
 413         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 414         int ret = -EINVAL;
 415
 416         if (!port)
 417                 goto err;
 418
 419         ret = -EMSGSIZE;
 420         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 421                 goto err;
 422
 423         return 0;
 424
 425 err:
 426         return ret;
 427 }
 428
 429 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 430                            struct nlattr *tb[], struct nlattr *data[])
 431 {
 432         struct ipvl_dev *ipvlan = netdev_priv(dev);
 433         struct ipvl_port *port;
 434         struct net_device *phy_dev;
 435         int err;
 436
 437         if (!tb[IFLA_LINK])
 438                 return -EINVAL;
 439
 440         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 441         if (!phy_dev)
 442                 return -ENODEV;
 443
 444         if (ipvlan_dev_slave(phy_dev)) {
 445                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 446
 447                 phy_dev = tmp->phy_dev;
 448         } else if (!ipvlan_dev_master(phy_dev)) {
 449                 err = ipvlan_port_create(phy_dev);
 450                 if (err < 0)
 451                         return err;
 452         }
 453
 454         port = ipvlan_port_get_rtnl(phy_dev);
 455         if (data && data[IFLA_IPVLAN_MODE])
 456                 port->mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 457
 458         ipvlan->phy_dev = phy_dev;
 459         ipvlan->dev = dev;
 460         ipvlan->port = port;
 461         ipvlan->sfeatures = IPVLAN_FEATURES;
 462         INIT_LIST_HEAD(&ipvlan->addrs);
 463         ipvlan->ipv4cnt = 0;
 464         ipvlan->ipv6cnt = 0;
 465
 466         /* TODO Probably put random address here to be presented to the
 467          * world but keep using the physical-dev address for the outgoing
 468          * packets.
 469          */
 470         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 471
 472         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 473
 474         port->count += 1;
 475         err = register_netdevice(dev);
 476         if (err < 0)
 477                 goto ipvlan_destroy_port;
 478
 479         err = netdev_upper_dev_link(phy_dev, dev);
 480         if (err)
 481                 goto ipvlan_destroy_port;
 482
 483         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 484         netif_stacked_transfer_operstate(phy_dev, dev);
 485         return 0;
 486
 487 ipvlan_destroy_port:
 488         port->count -= 1;
 489         if (!port->count)
 490                 ipvlan_port_destroy(phy_dev);
 491
 492         return err;
 493 }
 494
 495 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 496 {
 497         struct ipvl_dev *ipvlan = netdev_priv(dev);
 498         struct ipvl_addr *addr, *next;
 499
 500         if (ipvlan->ipv6cnt > 0 || ipvlan->ipv4cnt > 0) {
 501                 list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 502                         ipvlan_ht_addr_del(addr, !dev->dismantle);
 503                         list_del_rcu(&addr->anode);
 504                 }
 505         }
 506         list_del_rcu(&ipvlan->pnode);
 507         unregister_netdevice_queue(dev, head);
 508         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 509 }
 510
 511 static void ipvlan_link_setup(struct net_device *dev)
 512 {
 513         ether_setup(dev);
 514
 515         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 516         dev->priv_flags |= IFF_UNICAST_FLT;
 517         dev->netdev_ops = &ipvlan_netdev_ops;
 518         dev->destructor = free_netdev;
 519         dev->header_ops = &ipvlan_header_ops;
 520         dev->ethtool_ops = &ipvlan_ethtool_ops;
 521         dev->tx_queue_len = 0;
 522 }
 523
 524 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 525 {
 526         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 527 };
 528
 529 static struct rtnl_link_ops ipvlan_link_ops = {
 530         .kind           = "ipvlan",
 531         .priv_size      = sizeof(struct ipvl_dev),
 532
 533         .get_size       = ipvlan_nl_getsize,
 534         .policy         = ipvlan_nl_policy,
 535         .validate       = ipvlan_nl_validate,
 536         .fill_info      = ipvlan_nl_fillinfo,
 537         .changelink     = ipvlan_nl_changelink,
 538         .maxtype        = IFLA_IPVLAN_MAX,
 539
 540         .setup          = ipvlan_link_setup,
 541         .newlink        = ipvlan_link_new,
 542         .dellink        = ipvlan_link_delete,
 543 };
 544
 545 static int ipvlan_link_register(struct rtnl_link_ops *ops)
 546 {
 547         return rtnl_link_register(ops);
 548 }
 549
 550 static int ipvlan_device_event(struct notifier_block *unused,
 551                                unsigned long event, void *ptr)
 552 {
 553         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 554         struct ipvl_dev *ipvlan, *next;
 555         struct ipvl_port *port;
 556         LIST_HEAD(lst_kill);
 557
 558         if (!ipvlan_dev_master(dev))
 559                 return NOTIFY_DONE;
 560
 561         port = ipvlan_port_get_rtnl(dev);
 562
 563         switch (event) {
 564         case NETDEV_CHANGE:
 565                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 566                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 567                                                          ipvlan->dev);
 568                 break;
 569
 570         case NETDEV_UNREGISTER:
 571                 if (dev->reg_state != NETREG_UNREGISTERING)
 572                         break;
 573
 574                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 575                                          pnode)
 576                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 577                                                             &lst_kill);
 578                 unregister_netdevice_many(&lst_kill);
 579                 break;
 580
 581         case NETDEV_FEAT_CHANGE:
 582                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 583                         ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
 584                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 585                         netdev_features_change(ipvlan->dev);
 586                 }
 587                 break;
 588
 589         case NETDEV_CHANGEMTU:
 590                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 591                         ipvlan_adjust_mtu(ipvlan, dev);
 592                 break;
 593
 594         case NETDEV_PRE_TYPE_CHANGE:
 595                 /* Forbid underlying device to change its type. */
 596                 return NOTIFY_BAD;
 597         }
 598         return NOTIFY_DONE;
 599 }
 600
 601 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 602 {
 603         struct ipvl_addr *addr;
 604
 605         if (ipvlan_addr_busy(ipvlan, ip6_addr, true)) {
 606                 netif_err(ipvlan, ifup, ipvlan->dev,
 607                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 608                           ip6_addr, ipvlan->dev->name);
 609                 return -EINVAL;
 610         }
 611         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 612         if (!addr)
 613                 return -ENOMEM;
 614
 615         addr->master = ipvlan;
 616         memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
 617         addr->atype = IPVL_IPV6;
 618         list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
 619         ipvlan->ipv6cnt++;
 620         ipvlan_ht_addr_add(ipvlan, addr);
 621
 622         return 0;
 623 }
 624
 625 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 626 {
 627         struct ipvl_addr *addr;
 628
 629         addr = ipvlan_ht_addr_lookup(ipvlan->port, ip6_addr, true);
 630         if (!addr)
 631                 return;
 632
 633         ipvlan_ht_addr_del(addr, true);
 634         list_del_rcu(&addr->anode);
 635         ipvlan->ipv6cnt--;
 636         WARN_ON(ipvlan->ipv6cnt < 0);
 637         kfree_rcu(addr, rcu);
 638
 639         return;
 640 }
 641
 642 static int ipvlan_addr6_event(struct notifier_block *unused,
 643                               unsigned long event, void *ptr)
 644 {
 645         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 646         struct net_device *dev = (struct net_device *)if6->idev->dev;
 647         struct ipvl_dev *ipvlan = netdev_priv(dev);
 648
 649         if (!ipvlan_dev_slave(dev))
 650                 return NOTIFY_DONE;
 651
 652         if (!ipvlan || !ipvlan->port)
 653                 return NOTIFY_DONE;
 654
 655         switch (event) {
 656         case NETDEV_UP:
 657                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 658                         return NOTIFY_BAD;
 659                 break;
 660
 661         case NETDEV_DOWN:
 662                 ipvlan_del_addr6(ipvlan, &if6->addr);
 663                 break;
 664         }
 665
 666         return NOTIFY_OK;
 667 }
 668
 669 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 670 {
 671         struct ipvl_addr *addr;
 672
 673         if (ipvlan_addr_busy(ipvlan, ip4_addr, false)) {
 674                 netif_err(ipvlan, ifup, ipvlan->dev,
 675                           "Failed to add IPv4=%pI4 on %s intf.\n",
 676                           ip4_addr, ipvlan->dev->name);
 677                 return -EINVAL;
 678         }
 679         addr = kzalloc(sizeof(struct ipvl_addr), GFP_KERNEL);
 680         if (!addr)
 681                 return -ENOMEM;
 682
 683         addr->master = ipvlan;
 684         memcpy(&addr->ip4addr, ip4_addr, sizeof(struct in_addr));
 685         addr->atype = IPVL_IPV4;
 686         list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
 687         ipvlan->ipv4cnt++;
 688         ipvlan_ht_addr_add(ipvlan, addr);
 689         ipvlan_set_broadcast_mac_filter(ipvlan, true);
 690
 691         return 0;
 692 }
 693
 694 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 695 {
 696         struct ipvl_addr *addr;
 697
 698         addr = ipvlan_ht_addr_lookup(ipvlan->port, ip4_addr, false);
 699         if (!addr)
 700                 return;
 701
 702         ipvlan_ht_addr_del(addr, true);
 703         list_del_rcu(&addr->anode);
 704         ipvlan->ipv4cnt--;
 705         WARN_ON(ipvlan->ipv4cnt < 0);
 706         if (!ipvlan->ipv4cnt)
 707             ipvlan_set_broadcast_mac_filter(ipvlan, false);
 708         kfree_rcu(addr, rcu);
 709
 710         return;
 711 }
 712
 713 static int ipvlan_addr4_event(struct notifier_block *unused,
 714                               unsigned long event, void *ptr)
 715 {
 716         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 717         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 718         struct ipvl_dev *ipvlan = netdev_priv(dev);
 719         struct in_addr ip4_addr;
 720
 721         if (!ipvlan_dev_slave(dev))
 722                 return NOTIFY_DONE;
 723
 724         if (!ipvlan || !ipvlan->port)
 725                 return NOTIFY_DONE;
 726
 727         switch (event) {
 728         case NETDEV_UP:
 729                 ip4_addr.s_addr = if4->ifa_address;
 730                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 731                         return NOTIFY_BAD;
 732                 break;
 733
 734         case NETDEV_DOWN:
 735                 ip4_addr.s_addr = if4->ifa_address;
 736                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 737                 break;
 738         }
 739
 740         return NOTIFY_OK;
 741 }
 742
 743 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 744         .notifier_call = ipvlan_addr4_event,
 745 };
 746
 747 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 748         .notifier_call = ipvlan_device_event,
 749 };
 750
 751 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 752         .notifier_call = ipvlan_addr6_event,
 753 };
 754
 755 static int __init ipvlan_init_module(void)
 756 {
 757         int err;
 758
 759         ipvlan_init_secret();
 760         register_netdevice_notifier(&ipvlan_notifier_block);
 761         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 762         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 763
 764         err = ipvlan_link_register(&ipvlan_link_ops);
 765         if (err < 0)
 766                 goto error;
 767
 768         return 0;
 769 error:
 770         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 771         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 772         unregister_netdevice_notifier(&ipvlan_notifier_block);
 773         return err;
 774 }
 775
 776 static void __exit ipvlan_cleanup_module(void)
 777 {
 778         rtnl_link_unregister(&ipvlan_link_ops);
 779         unregister_netdevice_notifier(&ipvlan_notifier_block);
 780         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 781         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 782 }
 783
 784 module_init(ipvlan_init_module);
 785 module_exit(ipvlan_cleanup_module);
 786
 787 MODULE_LICENSE("GPL");
 788 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 789 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 790 MODULE_ALIAS_RTNL_LINK("ipvlan");