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
 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->gso_max_segs = phy_dev->gso_max_segs;
 123         dev->hard_header_len = phy_dev->hard_header_len;
 124
 125         ipvlan_set_lockdep_class(dev);
 126
 127         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 128         if (!ipvlan->pcpu_stats)
 129                 return -ENOMEM;
 130
 131         return 0;
 132 }
 133
 134 static void ipvlan_uninit(struct net_device *dev)
 135 {
 136         struct ipvl_dev *ipvlan = netdev_priv(dev);
 137         struct ipvl_port *port = ipvlan->port;
 138
 139         free_percpu(ipvlan->pcpu_stats);
 140
 141         port->count -= 1;
 142         if (!port->count)
 143                 ipvlan_port_destroy(port->dev);
 144 }
 145
 146 static int ipvlan_open(struct net_device *dev)
 147 {
 148         struct ipvl_dev *ipvlan = netdev_priv(dev);
 149         struct net_device *phy_dev = ipvlan->phy_dev;
 150         struct ipvl_addr *addr;
 151
 152         if (ipvlan->port->mode == IPVLAN_MODE_L3)
 153                 dev->flags |= IFF_NOARP;
 154         else
 155                 dev->flags &= ~IFF_NOARP;
 156
 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         list_for_each_entry(addr, &ipvlan->addrs, anode)
 175                 ipvlan_ht_addr_del(addr);
 176
 177         return 0;
 178 }
 179
 180 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 181                                      struct net_device *dev)
 182 {
 183         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 184         int skblen = skb->len;
 185         int ret;
 186
 187         ret = ipvlan_queue_xmit(skb, dev);
 188         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 189                 struct ipvl_pcpu_stats *pcptr;
 190
 191                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 192
 193                 u64_stats_update_begin(&pcptr->syncp);
 194                 pcptr->tx_pkts++;
 195                 pcptr->tx_bytes += skblen;
 196                 u64_stats_update_end(&pcptr->syncp);
 197         } else {
 198                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 199         }
 200         return ret;
 201 }
 202
 203 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 204                                              netdev_features_t features)
 205 {
 206         struct ipvl_dev *ipvlan = netdev_priv(dev);
 207
 208         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 209 }
 210
 211 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 212 {
 213         struct ipvl_dev *ipvlan = netdev_priv(dev);
 214         struct net_device *phy_dev = ipvlan->phy_dev;
 215
 216         if (change & IFF_ALLMULTI)
 217                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 218 }
 219
 220 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 221 {
 222         struct ipvl_dev *ipvlan = netdev_priv(dev);
 223
 224         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 225                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 226         } else {
 227                 struct netdev_hw_addr *ha;
 228                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 229
 230                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 231                 netdev_for_each_mc_addr(ha, dev)
 232                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 233
 234                 /* Turn-on broadcast bit irrespective of address family,
 235                  * since broadcast is deferred to a work-queue, hence no
 236                  * impact on fast-path processing.
 237                  */
 238                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 239
 240                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 241                             IPVLAN_MAC_FILTER_SIZE);
 242         }
 243         dev_uc_sync(ipvlan->phy_dev, dev);
 244         dev_mc_sync(ipvlan->phy_dev, dev);
 245 }
 246
 247 static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
 248                                                     struct rtnl_link_stats64 *s)
 249 {
 250         struct ipvl_dev *ipvlan = netdev_priv(dev);
 251
 252         if (ipvlan->pcpu_stats) {
 253                 struct ipvl_pcpu_stats *pcptr;
 254                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 255                 u32 rx_errs = 0, tx_drps = 0;
 256                 u32 strt;
 257                 int idx;
 258
 259                 for_each_possible_cpu(idx) {
 260                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 261                         do {
 262                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 263                                 rx_pkts = pcptr->rx_pkts;
 264                                 rx_bytes = pcptr->rx_bytes;
 265                                 rx_mcast = pcptr->rx_mcast;
 266                                 tx_pkts = pcptr->tx_pkts;
 267                                 tx_bytes = pcptr->tx_bytes;
 268                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 269                                                            strt));
 270
 271                         s->rx_packets += rx_pkts;
 272                         s->rx_bytes += rx_bytes;
 273                         s->multicast += rx_mcast;
 274                         s->tx_packets += tx_pkts;
 275                         s->tx_bytes += tx_bytes;
 276
 277                         /* u32 values are updated without syncp protection. */
 278                         rx_errs += pcptr->rx_errs;
 279                         tx_drps += pcptr->tx_drps;
 280                 }
 281                 s->rx_errors = rx_errs;
 282                 s->rx_dropped = rx_errs;
 283                 s->tx_dropped = tx_drps;
 284         }
 285         return s;
 286 }
 287
 288 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 289 {
 290         struct ipvl_dev *ipvlan = netdev_priv(dev);
 291         struct net_device *phy_dev = ipvlan->phy_dev;
 292
 293         return vlan_vid_add(phy_dev, proto, vid);
 294 }
 295
 296 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 297                                    u16 vid)
 298 {
 299         struct ipvl_dev *ipvlan = netdev_priv(dev);
 300         struct net_device *phy_dev = ipvlan->phy_dev;
 301
 302         vlan_vid_del(phy_dev, proto, vid);
 303         return 0;
 304 }
 305
 306 static int ipvlan_get_iflink(const struct net_device *dev)
 307 {
 308         struct ipvl_dev *ipvlan = netdev_priv(dev);
 309
 310         return ipvlan->phy_dev->ifindex;
 311 }
 312
 313 static const struct net_device_ops ipvlan_netdev_ops = {
 314         .ndo_init               = ipvlan_init,
 315         .ndo_uninit             = ipvlan_uninit,
 316         .ndo_open               = ipvlan_open,
 317         .ndo_stop               = ipvlan_stop,
 318         .ndo_start_xmit         = ipvlan_start_xmit,
 319         .ndo_fix_features       = ipvlan_fix_features,
 320         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 321         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 322         .ndo_get_stats64        = ipvlan_get_stats64,
 323         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 324         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 325         .ndo_get_iflink         = ipvlan_get_iflink,
 326 };
 327
 328 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 329                               unsigned short type, const void *daddr,
 330                               const void *saddr, unsigned len)
 331 {
 332         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 333         struct net_device *phy_dev = ipvlan->phy_dev;
 334
 335         /* TODO Probably use a different field than dev_addr so that the
 336          * mac-address on the virtual device is portable and can be carried
 337          * while the packets use the mac-addr on the physical device.
 338          */
 339         return dev_hard_header(skb, phy_dev, type, daddr,
 340                                saddr ? : dev->dev_addr, len);
 341 }
 342
 343 static const struct header_ops ipvlan_header_ops = {
 344         .create         = ipvlan_hard_header,
 345         .parse          = eth_header_parse,
 346         .cache          = eth_header_cache,
 347         .cache_update   = eth_header_cache_update,
 348 };
 349
 350 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 351                                              struct ethtool_link_ksettings *cmd)
 352 {
 353         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 354
 355         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 356 }
 357
 358 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 359                                        struct ethtool_drvinfo *drvinfo)
 360 {
 361         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 362         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 363 }
 364
 365 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 366 {
 367         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 368
 369         return ipvlan->msg_enable;
 370 }
 371
 372 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 373 {
 374         struct ipvl_dev *ipvlan = netdev_priv(dev);
 375
 376         ipvlan->msg_enable = value;
 377 }
 378
 379 static const struct ethtool_ops ipvlan_ethtool_ops = {
 380         .get_link       = ethtool_op_get_link,
 381         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 382         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 383         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 384         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 385 };
 386
 387 static int ipvlan_nl_changelink(struct net_device *dev,
 388                                 struct nlattr *tb[], struct nlattr *data[])
 389 {
 390         struct ipvl_dev *ipvlan = netdev_priv(dev);
 391         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 392
 393         if (data && data[IFLA_IPVLAN_MODE]) {
 394                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 395
 396                 ipvlan_set_port_mode(port, nmode);
 397         }
 398         return 0;
 399 }
 400
 401 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 402 {
 403         return (0
 404                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 405                 );
 406 }
 407
 408 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 409 {
 410         if (data && data[IFLA_IPVLAN_MODE]) {
 411                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 412
 413                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 414                         return -EINVAL;
 415         }
 416         return 0;
 417 }
 418
 419 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 420                               const struct net_device *dev)
 421 {
 422         struct ipvl_dev *ipvlan = netdev_priv(dev);
 423         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 424         int ret = -EINVAL;
 425
 426         if (!port)
 427                 goto err;
 428
 429         ret = -EMSGSIZE;
 430         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 431                 goto err;
 432
 433         return 0;
 434
 435 err:
 436         return ret;
 437 }
 438
 439 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 440                            struct nlattr *tb[], struct nlattr *data[])
 441 {
 442         struct ipvl_dev *ipvlan = netdev_priv(dev);
 443         struct ipvl_port *port;
 444         struct net_device *phy_dev;
 445         int err;
 446         u16 mode = IPVLAN_MODE_L3;
 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         if (data && data[IFLA_IPVLAN_MODE])
 466                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 467
 468         port = ipvlan_port_get_rtnl(phy_dev);
 469         ipvlan->phy_dev = phy_dev;
 470         ipvlan->dev = dev;
 471         ipvlan->port = port;
 472         ipvlan->sfeatures = IPVLAN_FEATURES;
 473         ipvlan_adjust_mtu(ipvlan, phy_dev);
 474         INIT_LIST_HEAD(&ipvlan->addrs);
 475
 476         /* TODO Probably put random address here to be presented to the
 477          * world but keep using the physical-dev address for the outgoing
 478          * packets.
 479          */
 480         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 481
 482         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 483
 484         port->count += 1;
 485         err = register_netdevice(dev);
 486         if (err < 0)
 487                 goto ipvlan_destroy_port;
 488
 489         err = netdev_upper_dev_link(phy_dev, dev);
 490         if (err)
 491                 goto ipvlan_destroy_port;
 492
 493         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 494         ipvlan_set_port_mode(port, mode);
 495
 496         netif_stacked_transfer_operstate(phy_dev, dev);
 497         return 0;
 498
 499 ipvlan_destroy_port:
 500         port->count -= 1;
 501         if (!port->count)
 502                 ipvlan_port_destroy(phy_dev);
 503
 504         return err;
 505 }
 506
 507 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 508 {
 509         struct ipvl_dev *ipvlan = netdev_priv(dev);
 510         struct ipvl_addr *addr, *next;
 511
 512         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 513                 ipvlan_ht_addr_del(addr);
 514                 list_del(&addr->anode);
 515                 kfree_rcu(addr, rcu);
 516         }
 517
 518         list_del_rcu(&ipvlan->pnode);
 519         unregister_netdevice_queue(dev, head);
 520         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 521 }
 522
 523 static void ipvlan_link_setup(struct net_device *dev)
 524 {
 525         ether_setup(dev);
 526
 527         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 528         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 529         dev->netdev_ops = &ipvlan_netdev_ops;
 530         dev->destructor = free_netdev;
 531         dev->header_ops = &ipvlan_header_ops;
 532         dev->ethtool_ops = &ipvlan_ethtool_ops;
 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                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 597                         netdev_features_change(ipvlan->dev);
 598                 }
 599                 break;
 600
 601         case NETDEV_CHANGEMTU:
 602                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 603                         ipvlan_adjust_mtu(ipvlan, dev);
 604                 break;
 605
 606         case NETDEV_PRE_TYPE_CHANGE:
 607                 /* Forbid underlying device to change its type. */
 608                 return NOTIFY_BAD;
 609         }
 610         return NOTIFY_DONE;
 611 }
 612
 613 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 614 {
 615         struct ipvl_addr *addr;
 616
 617         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 618                 netif_err(ipvlan, ifup, ipvlan->dev,
 619                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 620                           ip6_addr, ipvlan->dev->name);
 621                 return -EINVAL;
 622         }
 623         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 624         if (!addr)
 625                 return -ENOMEM;
 626
 627         addr->master = ipvlan;
 628         memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
 629         addr->atype = IPVL_IPV6;
 630         list_add_tail(&addr->anode, &ipvlan->addrs);
 631
 632         /* If the interface is not up, the address will be added to the hash
 633          * list by ipvlan_open.
 634          */
 635         if (netif_running(ipvlan->dev))
 636                 ipvlan_ht_addr_add(ipvlan, addr);
 637
 638         return 0;
 639 }
 640
 641 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 642 {
 643         struct ipvl_addr *addr;
 644
 645         addr = ipvlan_find_addr(ipvlan, ip6_addr, true);
 646         if (!addr)
 647                 return;
 648
 649         ipvlan_ht_addr_del(addr);
 650         list_del(&addr->anode);
 651         kfree_rcu(addr, rcu);
 652
 653         return;
 654 }
 655
 656 static int ipvlan_addr6_event(struct notifier_block *unused,
 657                               unsigned long event, void *ptr)
 658 {
 659         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 660         struct net_device *dev = (struct net_device *)if6->idev->dev;
 661         struct ipvl_dev *ipvlan = netdev_priv(dev);
 662
 663         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 664         if (in_softirq())
 665                 return NOTIFY_DONE;
 666
 667         if (!netif_is_ipvlan(dev))
 668                 return NOTIFY_DONE;
 669
 670         if (!ipvlan || !ipvlan->port)
 671                 return NOTIFY_DONE;
 672
 673         switch (event) {
 674         case NETDEV_UP:
 675                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 676                         return NOTIFY_BAD;
 677                 break;
 678
 679         case NETDEV_DOWN:
 680                 ipvlan_del_addr6(ipvlan, &if6->addr);
 681                 break;
 682         }
 683
 684         return NOTIFY_OK;
 685 }
 686
 687 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 688 {
 689         struct ipvl_addr *addr;
 690
 691         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 692                 netif_err(ipvlan, ifup, ipvlan->dev,
 693                           "Failed to add IPv4=%pI4 on %s intf.\n",
 694                           ip4_addr, ipvlan->dev->name);
 695                 return -EINVAL;
 696         }
 697         addr = kzalloc(sizeof(struct ipvl_addr), GFP_KERNEL);
 698         if (!addr)
 699                 return -ENOMEM;
 700
 701         addr->master = ipvlan;
 702         memcpy(&addr->ip4addr, ip4_addr, sizeof(struct in_addr));
 703         addr->atype = IPVL_IPV4;
 704         list_add_tail(&addr->anode, &ipvlan->addrs);
 705
 706         /* If the interface is not up, the address will be added to the hash
 707          * list by ipvlan_open.
 708          */
 709         if (netif_running(ipvlan->dev))
 710                 ipvlan_ht_addr_add(ipvlan, addr);
 711
 712         return 0;
 713 }
 714
 715 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 716 {
 717         struct ipvl_addr *addr;
 718
 719         addr = ipvlan_find_addr(ipvlan, ip4_addr, false);
 720         if (!addr)
 721                 return;
 722
 723         ipvlan_ht_addr_del(addr);
 724         list_del(&addr->anode);
 725         kfree_rcu(addr, rcu);
 726
 727         return;
 728 }
 729
 730 static int ipvlan_addr4_event(struct notifier_block *unused,
 731                               unsigned long event, void *ptr)
 732 {
 733         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 734         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 735         struct ipvl_dev *ipvlan = netdev_priv(dev);
 736         struct in_addr ip4_addr;
 737
 738         if (!netif_is_ipvlan(dev))
 739                 return NOTIFY_DONE;
 740
 741         if (!ipvlan || !ipvlan->port)
 742                 return NOTIFY_DONE;
 743
 744         switch (event) {
 745         case NETDEV_UP:
 746                 ip4_addr.s_addr = if4->ifa_address;
 747                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 748                         return NOTIFY_BAD;
 749                 break;
 750
 751         case NETDEV_DOWN:
 752                 ip4_addr.s_addr = if4->ifa_address;
 753                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 754                 break;
 755         }
 756
 757         return NOTIFY_OK;
 758 }
 759
 760 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 761         .notifier_call = ipvlan_addr4_event,
 762 };
 763
 764 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 765         .notifier_call = ipvlan_device_event,
 766 };
 767
 768 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 769         .notifier_call = ipvlan_addr6_event,
 770 };
 771
 772 static int __init ipvlan_init_module(void)
 773 {
 774         int err;
 775
 776         ipvlan_init_secret();
 777         register_netdevice_notifier(&ipvlan_notifier_block);
 778         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 779         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 780
 781         err = ipvlan_link_register(&ipvlan_link_ops);
 782         if (err < 0)
 783                 goto error;
 784
 785         return 0;
 786 error:
 787         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 788         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 789         unregister_netdevice_notifier(&ipvlan_notifier_block);
 790         return err;
 791 }
 792
 793 static void __exit ipvlan_cleanup_module(void)
 794 {
 795         rtnl_link_unregister(&ipvlan_link_ops);
 796         unregister_netdevice_notifier(&ipvlan_notifier_block);
 797         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 798         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 799 }
 800
 801 module_init(ipvlan_init_module);
 802 module_exit(ipvlan_cleanup_module);
 803
 804 MODULE_LICENSE("GPL");
 805 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 806 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 807 MODULE_ALIAS_RTNL_LINK("ipvlan");