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
  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_ALL_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
 117         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 118                      (phy_dev->state & IPVLAN_STATE_MASK);
 119         dev->features = phy_dev->features & IPVLAN_FEATURES;
 120         dev->features |= NETIF_F_LLTX;
 121         dev->gso_max_size = phy_dev->gso_max_size;
 122         dev->hard_header_len = phy_dev->hard_header_len;
 123
 124         ipvlan_set_lockdep_class(dev);
 125
 126         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 127         if (!ipvlan->pcpu_stats)
 128                 return -ENOMEM;
 129
 130         return 0;
 131 }
 132
 133 static void ipvlan_uninit(struct net_device *dev)
 134 {
 135         struct ipvl_dev *ipvlan = netdev_priv(dev);
 136         struct ipvl_port *port = ipvlan->port;
 137
 138         free_percpu(ipvlan->pcpu_stats);
 139
 140         port->count -= 1;
 141         if (!port->count)
 142                 ipvlan_port_destroy(port->dev);
 143 }
 144
 145 static int ipvlan_open(struct net_device *dev)
 146 {
 147         struct ipvl_dev *ipvlan = netdev_priv(dev);
 148         struct net_device *phy_dev = ipvlan->phy_dev;
 149         struct ipvl_addr *addr;
 150
 151         if (ipvlan->port->mode == IPVLAN_MODE_L3)
 152                 dev->flags |= IFF_NOARP;
 153         else
 154                 dev->flags &= ~IFF_NOARP;
 155
 156         if (ipvlan->ipv6cnt > 0 || ipvlan->ipv4cnt > 0) {
 157                 list_for_each_entry(addr, &ipvlan->addrs, anode)
 158                         ipvlan_ht_addr_add(ipvlan, addr);
 159         }
 160         return dev_uc_add(phy_dev, phy_dev->dev_addr);
 161 }
 162
 163 static int ipvlan_stop(struct net_device *dev)
 164 {
 165         struct ipvl_dev *ipvlan = netdev_priv(dev);
 166         struct net_device *phy_dev = ipvlan->phy_dev;
 167         struct ipvl_addr *addr;
 168
 169         dev_uc_unsync(phy_dev, dev);
 170         dev_mc_unsync(phy_dev, dev);
 171
 172         dev_uc_del(phy_dev, phy_dev->dev_addr);
 173
 174         if (ipvlan->ipv6cnt > 0 || ipvlan->ipv4cnt > 0) {
 175                 list_for_each_entry(addr, &ipvlan->addrs, anode)
 176                         ipvlan_ht_addr_del(addr, !dev->dismantle);
 177         }
 178         return 0;
 179 }
 180
 181 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 182                                      struct net_device *dev)
 183 {
 184         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 185         int skblen = skb->len;
 186         int ret;
 187
 188         ret = ipvlan_queue_xmit(skb, dev);
 189         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 190                 struct ipvl_pcpu_stats *pcptr;
 191
 192                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 193
 194                 u64_stats_update_begin(&pcptr->syncp);
 195                 pcptr->tx_pkts++;
 196                 pcptr->tx_bytes += skblen;
 197                 u64_stats_update_end(&pcptr->syncp);
 198         } else {
 199                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 200         }
 201         return ret;
 202 }
 203
 204 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 205                                              netdev_features_t features)
 206 {
 207         struct ipvl_dev *ipvlan = netdev_priv(dev);
 208
 209         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 210 }
 211
 212 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 213 {
 214         struct ipvl_dev *ipvlan = netdev_priv(dev);
 215         struct net_device *phy_dev = ipvlan->phy_dev;
 216
 217         if (change & IFF_ALLMULTI)
 218                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 219 }
 220
 221 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 222 {
 223         struct ipvl_dev *ipvlan = netdev_priv(dev);
 224
 225         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 226                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 227         } else {
 228                 struct netdev_hw_addr *ha;
 229                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 230
 231                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 232                 netdev_for_each_mc_addr(ha, dev)
 233                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 234
 235                 /* Turn-on broadcast bit irrespective of address family,
 236                  * since broadcast is deferred to a work-queue, hence no
 237                  * impact on fast-path processing.
 238                  */
 239                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 240
 241                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 242                             IPVLAN_MAC_FILTER_SIZE);
 243         }
 244         dev_uc_sync(ipvlan->phy_dev, dev);
 245         dev_mc_sync(ipvlan->phy_dev, dev);
 246 }
 247
 248 static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
 249                                                     struct rtnl_link_stats64 *s)
 250 {
 251         struct ipvl_dev *ipvlan = netdev_priv(dev);
 252
 253         if (ipvlan->pcpu_stats) {
 254                 struct ipvl_pcpu_stats *pcptr;
 255                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 256                 u32 rx_errs = 0, tx_drps = 0;
 257                 u32 strt;
 258                 int idx;
 259
 260                 for_each_possible_cpu(idx) {
 261                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 262                         do {
 263                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 264                                 rx_pkts = pcptr->rx_pkts;
 265                                 rx_bytes = pcptr->rx_bytes;
 266                                 rx_mcast = pcptr->rx_mcast;
 267                                 tx_pkts = pcptr->tx_pkts;
 268                                 tx_bytes = pcptr->tx_bytes;
 269                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 270                                                            strt));
 271
 272                         s->rx_packets += rx_pkts;
 273                         s->rx_bytes += rx_bytes;
 274                         s->multicast += rx_mcast;
 275                         s->tx_packets += tx_pkts;
 276                         s->tx_bytes += tx_bytes;
 277
 278                         /* u32 values are updated without syncp protection. */
 279                         rx_errs += pcptr->rx_errs;
 280                         tx_drps += pcptr->tx_drps;
 281                 }
 282                 s->rx_errors = rx_errs;
 283                 s->rx_dropped = rx_errs;
 284                 s->tx_dropped = tx_drps;
 285         }
 286         return s;
 287 }
 288
 289 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 290 {
 291         struct ipvl_dev *ipvlan = netdev_priv(dev);
 292         struct net_device *phy_dev = ipvlan->phy_dev;
 293
 294         return vlan_vid_add(phy_dev, proto, vid);
 295 }
 296
 297 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 298                                    u16 vid)
 299 {
 300         struct ipvl_dev *ipvlan = netdev_priv(dev);
 301         struct net_device *phy_dev = ipvlan->phy_dev;
 302
 303         vlan_vid_del(phy_dev, proto, vid);
 304         return 0;
 305 }
 306
 307 static int ipvlan_get_iflink(const struct net_device *dev)
 308 {
 309         struct ipvl_dev *ipvlan = netdev_priv(dev);
 310
 311         return ipvlan->phy_dev->ifindex;
 312 }
 313
 314 static const struct net_device_ops ipvlan_netdev_ops = {
 315         .ndo_init               = ipvlan_init,
 316         .ndo_uninit             = ipvlan_uninit,
 317         .ndo_open               = ipvlan_open,
 318         .ndo_stop               = ipvlan_stop,
 319         .ndo_start_xmit         = ipvlan_start_xmit,
 320         .ndo_fix_features       = ipvlan_fix_features,
 321         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 322         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 323         .ndo_get_stats64        = ipvlan_get_stats64,
 324         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 325         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 326         .ndo_get_iflink         = ipvlan_get_iflink,
 327 };
 328
 329 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 330                               unsigned short type, const void *daddr,
 331                               const void *saddr, unsigned len)
 332 {
 333         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 334         struct net_device *phy_dev = ipvlan->phy_dev;
 335
 336         /* TODO Probably use a different field than dev_addr so that the
 337          * mac-address on the virtual device is portable and can be carried
 338          * while the packets use the mac-addr on the physical device.
 339          */
 340         return dev_hard_header(skb, phy_dev, type, daddr,
 341                                saddr ? : dev->dev_addr, len);
 342 }
 343
 344 static const struct header_ops ipvlan_header_ops = {
 345         .create         = ipvlan_hard_header,
 346         .parse          = eth_header_parse,
 347         .cache          = eth_header_cache,
 348         .cache_update   = eth_header_cache_update,
 349 };
 350
 351 static int ipvlan_ethtool_get_settings(struct net_device *dev,
 352                                        struct ethtool_cmd *cmd)
 353 {
 354         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 355
 356         return __ethtool_get_settings(ipvlan->phy_dev, cmd);
 357 }
 358
 359 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 360                                        struct ethtool_drvinfo *drvinfo)
 361 {
 362         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 363         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 364 }
 365
 366 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 367 {
 368         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 369
 370         return ipvlan->msg_enable;
 371 }
 372
 373 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 374 {
 375         struct ipvl_dev *ipvlan = netdev_priv(dev);
 376
 377         ipvlan->msg_enable = value;
 378 }
 379
 380 static const struct ethtool_ops ipvlan_ethtool_ops = {
 381         .get_link       = ethtool_op_get_link,
 382         .get_settings   = ipvlan_ethtool_get_settings,
 383         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 384         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 385         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 386 };
 387
 388 static int ipvlan_nl_changelink(struct net_device *dev,
 389                                 struct nlattr *tb[], struct nlattr *data[])
 390 {
 391         struct ipvl_dev *ipvlan = netdev_priv(dev);
 392         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 393
 394         if (data && data[IFLA_IPVLAN_MODE]) {
 395                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 396
 397                 ipvlan_set_port_mode(port, nmode);
 398         }
 399         return 0;
 400 }
 401
 402 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 403 {
 404         return (0
 405                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 406                 );
 407 }
 408
 409 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 410 {
 411         if (data && data[IFLA_IPVLAN_MODE]) {
 412                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 413
 414                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 415                         return -EINVAL;
 416         }
 417         return 0;
 418 }
 419
 420 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 421                               const struct net_device *dev)
 422 {
 423         struct ipvl_dev *ipvlan = netdev_priv(dev);
 424         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 425         int ret = -EINVAL;
 426
 427         if (!port)
 428                 goto err;
 429
 430         ret = -EMSGSIZE;
 431         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 432                 goto err;
 433
 434         return 0;
 435
 436 err:
 437         return ret;
 438 }
 439
 440 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 441                            struct nlattr *tb[], struct nlattr *data[])
 442 {
 443         struct ipvl_dev *ipvlan = netdev_priv(dev);
 444         struct ipvl_port *port;
 445         struct net_device *phy_dev;
 446         int err;
 447
 448         if (!tb[IFLA_LINK])
 449                 return -EINVAL;
 450
 451         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 452         if (!phy_dev)
 453                 return -ENODEV;
 454
 455         if (netif_is_ipvlan(phy_dev)) {
 456                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 457
 458                 phy_dev = tmp->phy_dev;
 459         } else if (!netif_is_ipvlan_port(phy_dev)) {
 460                 err = ipvlan_port_create(phy_dev);
 461                 if (err < 0)
 462                         return err;
 463         }
 464
 465         port = ipvlan_port_get_rtnl(phy_dev);
 466         if (data && data[IFLA_IPVLAN_MODE])
 467                 port->mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 468
 469         ipvlan->phy_dev = phy_dev;
 470         ipvlan->dev = dev;
 471         ipvlan->port = port;
 472         ipvlan->sfeatures = IPVLAN_FEATURES;
 473         INIT_LIST_HEAD(&ipvlan->addrs);
 474         ipvlan->ipv4cnt = 0;
 475         ipvlan->ipv6cnt = 0;
 476
 477         /* TODO Probably put random address here to be presented to the
 478          * world but keep using the physical-dev address for the outgoing
 479          * packets.
 480          */
 481         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 482
 483         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 484
 485         port->count += 1;
 486         err = register_netdevice(dev);
 487         if (err < 0)
 488                 goto ipvlan_destroy_port;
 489
 490         err = netdev_upper_dev_link(phy_dev, dev);
 491         if (err)
 492                 goto ipvlan_destroy_port;
 493
 494         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 495         netif_stacked_transfer_operstate(phy_dev, dev);
 496         return 0;
 497
 498 ipvlan_destroy_port:
 499         port->count -= 1;
 500         if (!port->count)
 501                 ipvlan_port_destroy(phy_dev);
 502
 503         return err;
 504 }
 505
 506 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 507 {
 508         struct ipvl_dev *ipvlan = netdev_priv(dev);
 509         struct ipvl_addr *addr, *next;
 510
 511         if (ipvlan->ipv6cnt > 0 || ipvlan->ipv4cnt > 0) {
 512                 list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 513                         ipvlan_ht_addr_del(addr, !dev->dismantle);
 514                         list_del(&addr->anode);
 515                 }
 516         }
 517         list_del_rcu(&ipvlan->pnode);
 518         unregister_netdevice_queue(dev, head);
 519         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 520 }
 521
 522 static void ipvlan_link_setup(struct net_device *dev)
 523 {
 524         ether_setup(dev);
 525
 526         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 527         dev->priv_flags |= IFF_UNICAST_FLT;
 528         dev->netdev_ops = &ipvlan_netdev_ops;
 529         dev->destructor = free_netdev;
 530         dev->header_ops = &ipvlan_header_ops;
 531         dev->ethtool_ops = &ipvlan_ethtool_ops;
 532         dev->tx_queue_len = 0;
 533 }
 534
 535 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 536 {
 537         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 538 };
 539
 540 static struct rtnl_link_ops ipvlan_link_ops = {
 541         .kind           = "ipvlan",
 542         .priv_size      = sizeof(struct ipvl_dev),
 543
 544         .get_size       = ipvlan_nl_getsize,
 545         .policy         = ipvlan_nl_policy,
 546         .validate       = ipvlan_nl_validate,
 547         .fill_info      = ipvlan_nl_fillinfo,
 548         .changelink     = ipvlan_nl_changelink,
 549         .maxtype        = IFLA_IPVLAN_MAX,
 550
 551         .setup          = ipvlan_link_setup,
 552         .newlink        = ipvlan_link_new,
 553         .dellink        = ipvlan_link_delete,
 554 };
 555
 556 static int ipvlan_link_register(struct rtnl_link_ops *ops)
 557 {
 558         return rtnl_link_register(ops);
 559 }
 560
 561 static int ipvlan_device_event(struct notifier_block *unused,
 562                                unsigned long event, void *ptr)
 563 {
 564         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 565         struct ipvl_dev *ipvlan, *next;
 566         struct ipvl_port *port;
 567         LIST_HEAD(lst_kill);
 568
 569         if (!netif_is_ipvlan_port(dev))
 570                 return NOTIFY_DONE;
 571
 572         port = ipvlan_port_get_rtnl(dev);
 573
 574         switch (event) {
 575         case NETDEV_CHANGE:
 576                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 577                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 578                                                          ipvlan->dev);
 579                 break;
 580
 581         case NETDEV_UNREGISTER:
 582                 if (dev->reg_state != NETREG_UNREGISTERING)
 583                         break;
 584
 585                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 586                                          pnode)
 587                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 588                                                             &lst_kill);
 589                 unregister_netdevice_many(&lst_kill);
 590                 break;
 591
 592         case NETDEV_FEAT_CHANGE:
 593                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 594                         ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
 595                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 596                         netdev_features_change(ipvlan->dev);
 597                 }
 598                 break;
 599
 600         case NETDEV_CHANGEMTU:
 601                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 602                         ipvlan_adjust_mtu(ipvlan, dev);
 603                 break;
 604
 605         case NETDEV_PRE_TYPE_CHANGE:
 606                 /* Forbid underlying device to change its type. */
 607                 return NOTIFY_BAD;
 608         }
 609         return NOTIFY_DONE;
 610 }
 611
 612 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 613 {
 614         struct ipvl_addr *addr;
 615
 616         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 617                 netif_err(ipvlan, ifup, ipvlan->dev,
 618                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 619                           ip6_addr, ipvlan->dev->name);
 620                 return -EINVAL;
 621         }
 622         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 623         if (!addr)
 624                 return -ENOMEM;
 625
 626         addr->master = ipvlan;
 627         memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
 628         addr->atype = IPVL_IPV6;
 629         list_add_tail(&addr->anode, &ipvlan->addrs);
 630         ipvlan->ipv6cnt++;
 631         /* If the interface is not up, the address will be added to the hash
 632          * list by ipvlan_open.
 633          */
 634         if (netif_running(ipvlan->dev))
 635                 ipvlan_ht_addr_add(ipvlan, addr);
 636
 637         return 0;
 638 }
 639
 640 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 641 {
 642         struct ipvl_addr *addr;
 643
 644         addr = ipvlan_find_addr(ipvlan, ip6_addr, true);
 645         if (!addr)
 646                 return;
 647
 648         ipvlan_ht_addr_del(addr, true);
 649         list_del(&addr->anode);
 650         ipvlan->ipv6cnt--;
 651         WARN_ON(ipvlan->ipv6cnt < 0);
 652         kfree_rcu(addr, rcu);
 653
 654         return;
 655 }
 656
 657 static int ipvlan_addr6_event(struct notifier_block *unused,
 658                               unsigned long event, void *ptr)
 659 {
 660         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 661         struct net_device *dev = (struct net_device *)if6->idev->dev;
 662         struct ipvl_dev *ipvlan = netdev_priv(dev);
 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         ipvlan->ipv4cnt++;
 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, true);
 721         list_del(&addr->anode);
 722         ipvlan->ipv4cnt--;
 723         WARN_ON(ipvlan->ipv4cnt < 0);
 724         kfree_rcu(addr, rcu);
 725
 726         return;
 727 }
 728
 729 static int ipvlan_addr4_event(struct notifier_block *unused,
 730                               unsigned long event, void *ptr)
 731 {
 732         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 733         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 734         struct ipvl_dev *ipvlan = netdev_priv(dev);
 735         struct in_addr ip4_addr;
 736
 737         if (!netif_is_ipvlan(dev))
 738                 return NOTIFY_DONE;
 739
 740         if (!ipvlan || !ipvlan->port)
 741                 return NOTIFY_DONE;
 742
 743         switch (event) {
 744         case NETDEV_UP:
 745                 ip4_addr.s_addr = if4->ifa_address;
 746                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 747                         return NOTIFY_BAD;
 748                 break;
 749
 750         case NETDEV_DOWN:
 751                 ip4_addr.s_addr = if4->ifa_address;
 752                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 753                 break;
 754         }
 755
 756         return NOTIFY_OK;
 757 }
 758
 759 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 760         .notifier_call = ipvlan_addr4_event,
 761 };
 762
 763 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 764         .notifier_call = ipvlan_device_event,
 765 };
 766
 767 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 768         .notifier_call = ipvlan_addr6_event,
 769 };
 770
 771 static int __init ipvlan_init_module(void)
 772 {
 773         int err;
 774
 775         ipvlan_init_secret();
 776         register_netdevice_notifier(&ipvlan_notifier_block);
 777         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 778         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 779
 780         err = ipvlan_link_register(&ipvlan_link_ops);
 781         if (err < 0)
 782                 goto error;
 783
 784         return 0;
 785 error:
 786         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 787         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 788         unregister_netdevice_notifier(&ipvlan_notifier_block);
 789         return err;
 790 }
 791
 792 static void __exit ipvlan_cleanup_module(void)
 793 {
 794         rtnl_link_unregister(&ipvlan_link_ops);
 795         unregister_netdevice_notifier(&ipvlan_notifier_block);
 796         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 797         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 798 }
 799
 800 module_init(ipvlan_init_module);
 801 module_exit(ipvlan_cleanup_module);
 802
 803 MODULE_LICENSE("GPL");
 804 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 805 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 806 MODULE_ALIAS_RTNL_LINK("ipvlan");