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_ALWAYS_ON_OFLOADS \
  91         (NETIF_F_SG | NETIF_F_HW_CSUM | \
  92          NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
  93
  94 #define IPVLAN_ALWAYS_ON \
  95         (IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED)
  96
  97 #define IPVLAN_FEATURES \
  98         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
  99          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 100          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 101          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 102
 103         /* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
 104
 105 #define IPVLAN_STATE_MASK \
 106         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 107
 108 static void ipvlan_set_lockdep_class_one(struct net_device *dev,
 109                                          struct netdev_queue *txq,
 110                                          void *_unused)
 111 {
 112         lockdep_set_class(&txq->_xmit_lock, &ipvlan_netdev_xmit_lock_key);
 113 }
 114
 115 static void ipvlan_set_lockdep_class(struct net_device *dev)
 116 {
 117         lockdep_set_class(&dev->addr_list_lock, &ipvlan_netdev_addr_lock_key);
 118         netdev_for_each_tx_queue(dev, ipvlan_set_lockdep_class_one, NULL);
 119 }
 120
 121 static int ipvlan_init(struct net_device *dev)
 122 {
 123         struct ipvl_dev *ipvlan = netdev_priv(dev);
 124         const struct net_device *phy_dev = ipvlan->phy_dev;
 125
 126         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 127                      (phy_dev->state & IPVLAN_STATE_MASK);
 128         dev->features = phy_dev->features & IPVLAN_FEATURES;
 129         dev->features |= IPVLAN_ALWAYS_ON;
 130         dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
 131         dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
 132         dev->gso_max_size = phy_dev->gso_max_size;
 133         dev->hard_header_len = phy_dev->hard_header_len;
 134
 135         ipvlan_set_lockdep_class(dev);
 136
 137         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 138         if (!ipvlan->pcpu_stats)
 139                 return -ENOMEM;
 140
 141         return 0;
 142 }
 143
 144 static void ipvlan_uninit(struct net_device *dev)
 145 {
 146         struct ipvl_dev *ipvlan = netdev_priv(dev);
 147         struct ipvl_port *port = ipvlan->port;
 148
 149         free_percpu(ipvlan->pcpu_stats);
 150
 151         port->count -= 1;
 152         if (!port->count)
 153                 ipvlan_port_destroy(port->dev);
 154 }
 155
 156 static int ipvlan_open(struct net_device *dev)
 157 {
 158         struct ipvl_dev *ipvlan = netdev_priv(dev);
 159         struct ipvl_addr *addr;
 160
 161         if (ipvlan->port->mode == IPVLAN_MODE_L3)
 162                 dev->flags |= IFF_NOARP;
 163         else
 164                 dev->flags &= ~IFF_NOARP;
 165
 166         list_for_each_entry(addr, &ipvlan->addrs, anode)
 167                 ipvlan_ht_addr_add(ipvlan, addr);
 168
 169         return 0;
 170 }
 171
 172 static int ipvlan_stop(struct net_device *dev)
 173 {
 174         struct ipvl_dev *ipvlan = netdev_priv(dev);
 175         struct net_device *phy_dev = ipvlan->phy_dev;
 176         struct ipvl_addr *addr;
 177
 178         dev_uc_unsync(phy_dev, dev);
 179         dev_mc_unsync(phy_dev, dev);
 180
 181         list_for_each_entry(addr, &ipvlan->addrs, anode)
 182                 ipvlan_ht_addr_del(addr);
 183
 184         return 0;
 185 }
 186
 187 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 188                                      struct net_device *dev)
 189 {
 190         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 191         int skblen = skb->len;
 192         int ret;
 193
 194         ret = ipvlan_queue_xmit(skb, dev);
 195         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 196                 struct ipvl_pcpu_stats *pcptr;
 197
 198                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 199
 200                 u64_stats_update_begin(&pcptr->syncp);
 201                 pcptr->tx_pkts++;
 202                 pcptr->tx_bytes += skblen;
 203                 u64_stats_update_end(&pcptr->syncp);
 204         } else {
 205                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 206         }
 207         return ret;
 208 }
 209
 210 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 211                                              netdev_features_t features)
 212 {
 213         struct ipvl_dev *ipvlan = netdev_priv(dev);
 214
 215         features |= NETIF_F_ALL_FOR_ALL;
 216         features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 217         features = netdev_increment_features(ipvlan->phy_dev->features,
 218                                              features, features);
 219         features |= IPVLAN_ALWAYS_ON;
 220         features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
 221
 222         return features;
 223 }
 224
 225 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 226 {
 227         struct ipvl_dev *ipvlan = netdev_priv(dev);
 228         struct net_device *phy_dev = ipvlan->phy_dev;
 229
 230         if (change & IFF_ALLMULTI)
 231                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 232 }
 233
 234 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 235 {
 236         struct ipvl_dev *ipvlan = netdev_priv(dev);
 237
 238         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 239                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 240         } else {
 241                 struct netdev_hw_addr *ha;
 242                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 243
 244                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 245                 netdev_for_each_mc_addr(ha, dev)
 246                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 247
 248                 /* Turn-on broadcast bit irrespective of address family,
 249                  * since broadcast is deferred to a work-queue, hence no
 250                  * impact on fast-path processing.
 251                  */
 252                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 253
 254                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 255                             IPVLAN_MAC_FILTER_SIZE);
 256         }
 257         dev_uc_sync(ipvlan->phy_dev, dev);
 258         dev_mc_sync(ipvlan->phy_dev, dev);
 259 }
 260
 261 static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
 262                                                     struct rtnl_link_stats64 *s)
 263 {
 264         struct ipvl_dev *ipvlan = netdev_priv(dev);
 265
 266         if (ipvlan->pcpu_stats) {
 267                 struct ipvl_pcpu_stats *pcptr;
 268                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 269                 u32 rx_errs = 0, tx_drps = 0;
 270                 u32 strt;
 271                 int idx;
 272
 273                 for_each_possible_cpu(idx) {
 274                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 275                         do {
 276                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 277                                 rx_pkts = pcptr->rx_pkts;
 278                                 rx_bytes = pcptr->rx_bytes;
 279                                 rx_mcast = pcptr->rx_mcast;
 280                                 tx_pkts = pcptr->tx_pkts;
 281                                 tx_bytes = pcptr->tx_bytes;
 282                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 283                                                            strt));
 284
 285                         s->rx_packets += rx_pkts;
 286                         s->rx_bytes += rx_bytes;
 287                         s->multicast += rx_mcast;
 288                         s->tx_packets += tx_pkts;
 289                         s->tx_bytes += tx_bytes;
 290
 291                         /* u32 values are updated without syncp protection. */
 292                         rx_errs += pcptr->rx_errs;
 293                         tx_drps += pcptr->tx_drps;
 294                 }
 295                 s->rx_errors = rx_errs;
 296                 s->rx_dropped = rx_errs;
 297                 s->tx_dropped = tx_drps;
 298         }
 299         return s;
 300 }
 301
 302 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 303 {
 304         struct ipvl_dev *ipvlan = netdev_priv(dev);
 305         struct net_device *phy_dev = ipvlan->phy_dev;
 306
 307         return vlan_vid_add(phy_dev, proto, vid);
 308 }
 309
 310 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 311                                    u16 vid)
 312 {
 313         struct ipvl_dev *ipvlan = netdev_priv(dev);
 314         struct net_device *phy_dev = ipvlan->phy_dev;
 315
 316         vlan_vid_del(phy_dev, proto, vid);
 317         return 0;
 318 }
 319
 320 static int ipvlan_get_iflink(const struct net_device *dev)
 321 {
 322         struct ipvl_dev *ipvlan = netdev_priv(dev);
 323
 324         return ipvlan->phy_dev->ifindex;
 325 }
 326
 327 static const struct net_device_ops ipvlan_netdev_ops = {
 328         .ndo_init               = ipvlan_init,
 329         .ndo_uninit             = ipvlan_uninit,
 330         .ndo_open               = ipvlan_open,
 331         .ndo_stop               = ipvlan_stop,
 332         .ndo_start_xmit         = ipvlan_start_xmit,
 333         .ndo_fix_features       = ipvlan_fix_features,
 334         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 335         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 336         .ndo_get_stats64        = ipvlan_get_stats64,
 337         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 338         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 339         .ndo_get_iflink         = ipvlan_get_iflink,
 340 };
 341
 342 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 343                               unsigned short type, const void *daddr,
 344                               const void *saddr, unsigned len)
 345 {
 346         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 347         struct net_device *phy_dev = ipvlan->phy_dev;
 348
 349         /* TODO Probably use a different field than dev_addr so that the
 350          * mac-address on the virtual device is portable and can be carried
 351          * while the packets use the mac-addr on the physical device.
 352          */
 353         return dev_hard_header(skb, phy_dev, type, daddr,
 354                                saddr ? : dev->dev_addr, len);
 355 }
 356
 357 static const struct header_ops ipvlan_header_ops = {
 358         .create         = ipvlan_hard_header,
 359         .parse          = eth_header_parse,
 360         .cache          = eth_header_cache,
 361         .cache_update   = eth_header_cache_update,
 362 };
 363
 364 static int ipvlan_ethtool_get_settings(struct net_device *dev,
 365                                        struct ethtool_cmd *cmd)
 366 {
 367         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 368
 369         return __ethtool_get_settings(ipvlan->phy_dev, cmd);
 370 }
 371
 372 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 373                                        struct ethtool_drvinfo *drvinfo)
 374 {
 375         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 376         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 377 }
 378
 379 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 380 {
 381         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 382
 383         return ipvlan->msg_enable;
 384 }
 385
 386 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 387 {
 388         struct ipvl_dev *ipvlan = netdev_priv(dev);
 389
 390         ipvlan->msg_enable = value;
 391 }
 392
 393 static const struct ethtool_ops ipvlan_ethtool_ops = {
 394         .get_link       = ethtool_op_get_link,
 395         .get_settings   = ipvlan_ethtool_get_settings,
 396         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 397         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 398         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 399 };
 400
 401 static int ipvlan_nl_changelink(struct net_device *dev,
 402                                 struct nlattr *tb[], struct nlattr *data[])
 403 {
 404         struct ipvl_dev *ipvlan = netdev_priv(dev);
 405         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 406
 407         if (!data)
 408                 return 0;
 409         if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
 410                 return -EPERM;
 411
 412         if (data[IFLA_IPVLAN_MODE]) {
 413                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 414
 415                 ipvlan_set_port_mode(port, nmode);
 416         }
 417         return 0;
 418 }
 419
 420 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 421 {
 422         return (0
 423                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 424                 );
 425 }
 426
 427 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 428 {
 429         if (data && data[IFLA_IPVLAN_MODE]) {
 430                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 431
 432                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 433                         return -EINVAL;
 434         }
 435         return 0;
 436 }
 437
 438 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 439                               const struct net_device *dev)
 440 {
 441         struct ipvl_dev *ipvlan = netdev_priv(dev);
 442         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 443         int ret = -EINVAL;
 444
 445         if (!port)
 446                 goto err;
 447
 448         ret = -EMSGSIZE;
 449         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 450                 goto err;
 451
 452         return 0;
 453
 454 err:
 455         return ret;
 456 }
 457
 458 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 459                            struct nlattr *tb[], struct nlattr *data[])
 460 {
 461         struct ipvl_dev *ipvlan = netdev_priv(dev);
 462         struct ipvl_port *port;
 463         struct net_device *phy_dev;
 464         int err;
 465
 466         if (!tb[IFLA_LINK])
 467                 return -EINVAL;
 468
 469         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 470         if (!phy_dev)
 471                 return -ENODEV;
 472
 473         if (netif_is_ipvlan(phy_dev)) {
 474                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 475
 476                 phy_dev = tmp->phy_dev;
 477                 if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
 478                         return -EPERM;
 479         } else if (!netif_is_ipvlan_port(phy_dev)) {
 480                 err = ipvlan_port_create(phy_dev);
 481                 if (err < 0)
 482                         return err;
 483         }
 484
 485         port = ipvlan_port_get_rtnl(phy_dev);
 486         if (data && data[IFLA_IPVLAN_MODE])
 487                 port->mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 488
 489         ipvlan->phy_dev = phy_dev;
 490         ipvlan->dev = dev;
 491         ipvlan->port = port;
 492         ipvlan->sfeatures = IPVLAN_FEATURES;
 493         INIT_LIST_HEAD(&ipvlan->addrs);
 494
 495         /* TODO Probably put random address here to be presented to the
 496          * world but keep using the physical-dev address for the outgoing
 497          * packets.
 498          */
 499         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 500
 501         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 502
 503         port->count += 1;
 504         err = register_netdevice(dev);
 505         if (err < 0)
 506                 goto ipvlan_destroy_port;
 507
 508         err = netdev_upper_dev_link(phy_dev, dev);
 509         if (err)
 510                 goto ipvlan_destroy_port;
 511
 512         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 513         netif_stacked_transfer_operstate(phy_dev, dev);
 514         return 0;
 515
 516 ipvlan_destroy_port:
 517         port->count -= 1;
 518         if (!port->count)
 519                 ipvlan_port_destroy(phy_dev);
 520
 521         return err;
 522 }
 523
 524 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 525 {
 526         struct ipvl_dev *ipvlan = netdev_priv(dev);
 527         struct ipvl_addr *addr, *next;
 528
 529         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 530                 ipvlan_ht_addr_del(addr);
 531                 list_del(&addr->anode);
 532                 kfree_rcu(addr, rcu);
 533         }
 534
 535         list_del_rcu(&ipvlan->pnode);
 536         unregister_netdevice_queue(dev, head);
 537         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 538 }
 539
 540 static void ipvlan_link_setup(struct net_device *dev)
 541 {
 542         ether_setup(dev);
 543
 544         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 545         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 546         dev->netdev_ops = &ipvlan_netdev_ops;
 547         dev->destructor = free_netdev;
 548         dev->header_ops = &ipvlan_header_ops;
 549         dev->ethtool_ops = &ipvlan_ethtool_ops;
 550 }
 551
 552 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 553 {
 554         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 555 };
 556
 557 static struct rtnl_link_ops ipvlan_link_ops = {
 558         .kind           = "ipvlan",
 559         .priv_size      = sizeof(struct ipvl_dev),
 560
 561         .get_size       = ipvlan_nl_getsize,
 562         .policy         = ipvlan_nl_policy,
 563         .validate       = ipvlan_nl_validate,
 564         .fill_info      = ipvlan_nl_fillinfo,
 565         .changelink     = ipvlan_nl_changelink,
 566         .maxtype        = IFLA_IPVLAN_MAX,
 567
 568         .setup          = ipvlan_link_setup,
 569         .newlink        = ipvlan_link_new,
 570         .dellink        = ipvlan_link_delete,
 571 };
 572
 573 static int ipvlan_link_register(struct rtnl_link_ops *ops)
 574 {
 575         return rtnl_link_register(ops);
 576 }
 577
 578 static int ipvlan_device_event(struct notifier_block *unused,
 579                                unsigned long event, void *ptr)
 580 {
 581         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 582         struct ipvl_dev *ipvlan, *next;
 583         struct ipvl_port *port;
 584         LIST_HEAD(lst_kill);
 585
 586         if (!netif_is_ipvlan_port(dev))
 587                 return NOTIFY_DONE;
 588
 589         port = ipvlan_port_get_rtnl(dev);
 590
 591         switch (event) {
 592         case NETDEV_CHANGE:
 593                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 594                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 595                                                          ipvlan->dev);
 596                 break;
 597
 598         case NETDEV_UNREGISTER:
 599                 if (dev->reg_state != NETREG_UNREGISTERING)
 600                         break;
 601
 602                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 603                                          pnode)
 604                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 605                                                             &lst_kill);
 606                 unregister_netdevice_many(&lst_kill);
 607                 break;
 608
 609         case NETDEV_FEAT_CHANGE:
 610                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 611                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 612                         netdev_update_features(ipvlan->dev);
 613                 }
 614                 break;
 615
 616         case NETDEV_CHANGEMTU:
 617                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 618                         ipvlan_adjust_mtu(ipvlan, dev);
 619                 break;
 620
 621         case NETDEV_PRE_TYPE_CHANGE:
 622                 /* Forbid underlying device to change its type. */
 623                 return NOTIFY_BAD;
 624         }
 625         return NOTIFY_DONE;
 626 }
 627
 628 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 629 {
 630         struct ipvl_addr *addr;
 631
 632         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 633                 netif_err(ipvlan, ifup, ipvlan->dev,
 634                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 635                           ip6_addr, ipvlan->dev->name);
 636                 return -EINVAL;
 637         }
 638         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 639         if (!addr)
 640                 return -ENOMEM;
 641
 642         addr->master = ipvlan;
 643         memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
 644         addr->atype = IPVL_IPV6;
 645         list_add_tail(&addr->anode, &ipvlan->addrs);
 646
 647         /* If the interface is not up, the address will be added to the hash
 648          * list by ipvlan_open.
 649          */
 650         if (netif_running(ipvlan->dev))
 651                 ipvlan_ht_addr_add(ipvlan, addr);
 652
 653         return 0;
 654 }
 655
 656 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 657 {
 658         struct ipvl_addr *addr;
 659
 660         addr = ipvlan_find_addr(ipvlan, ip6_addr, true);
 661         if (!addr)
 662                 return;
 663
 664         ipvlan_ht_addr_del(addr);
 665         list_del(&addr->anode);
 666         kfree_rcu(addr, rcu);
 667
 668         return;
 669 }
 670
 671 static int ipvlan_addr6_event(struct notifier_block *unused,
 672                               unsigned long event, void *ptr)
 673 {
 674         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 675         struct net_device *dev = (struct net_device *)if6->idev->dev;
 676         struct ipvl_dev *ipvlan = netdev_priv(dev);
 677
 678         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 679         if (in_softirq())
 680                 return NOTIFY_DONE;
 681
 682         if (!netif_is_ipvlan(dev))
 683                 return NOTIFY_DONE;
 684
 685         if (!ipvlan || !ipvlan->port)
 686                 return NOTIFY_DONE;
 687
 688         switch (event) {
 689         case NETDEV_UP:
 690                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 691                         return NOTIFY_BAD;
 692                 break;
 693
 694         case NETDEV_DOWN:
 695                 ipvlan_del_addr6(ipvlan, &if6->addr);
 696                 break;
 697         }
 698
 699         return NOTIFY_OK;
 700 }
 701
 702 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 703 {
 704         struct ipvl_addr *addr;
 705
 706         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 707                 netif_err(ipvlan, ifup, ipvlan->dev,
 708                           "Failed to add IPv4=%pI4 on %s intf.\n",
 709                           ip4_addr, ipvlan->dev->name);
 710                 return -EINVAL;
 711         }
 712         addr = kzalloc(sizeof(struct ipvl_addr), GFP_KERNEL);
 713         if (!addr)
 714                 return -ENOMEM;
 715
 716         addr->master = ipvlan;
 717         memcpy(&addr->ip4addr, ip4_addr, sizeof(struct in_addr));
 718         addr->atype = IPVL_IPV4;
 719         list_add_tail(&addr->anode, &ipvlan->addrs);
 720
 721         /* If the interface is not up, the address will be added to the hash
 722          * list by ipvlan_open.
 723          */
 724         if (netif_running(ipvlan->dev))
 725                 ipvlan_ht_addr_add(ipvlan, addr);
 726
 727         return 0;
 728 }
 729
 730 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 731 {
 732         struct ipvl_addr *addr;
 733
 734         addr = ipvlan_find_addr(ipvlan, ip4_addr, false);
 735         if (!addr)
 736                 return;
 737
 738         ipvlan_ht_addr_del(addr);
 739         list_del(&addr->anode);
 740         kfree_rcu(addr, rcu);
 741
 742         return;
 743 }
 744
 745 static int ipvlan_addr4_event(struct notifier_block *unused,
 746                               unsigned long event, void *ptr)
 747 {
 748         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 749         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 750         struct ipvl_dev *ipvlan = netdev_priv(dev);
 751         struct in_addr ip4_addr;
 752
 753         if (!netif_is_ipvlan(dev))
 754                 return NOTIFY_DONE;
 755
 756         if (!ipvlan || !ipvlan->port)
 757                 return NOTIFY_DONE;
 758
 759         switch (event) {
 760         case NETDEV_UP:
 761                 ip4_addr.s_addr = if4->ifa_address;
 762                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 763                         return NOTIFY_BAD;
 764                 break;
 765
 766         case NETDEV_DOWN:
 767                 ip4_addr.s_addr = if4->ifa_address;
 768                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 769                 break;
 770         }
 771
 772         return NOTIFY_OK;
 773 }
 774
 775 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 776         .notifier_call = ipvlan_addr4_event,
 777 };
 778
 779 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 780         .notifier_call = ipvlan_device_event,
 781 };
 782
 783 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 784         .notifier_call = ipvlan_addr6_event,
 785 };
 786
 787 static int __init ipvlan_init_module(void)
 788 {
 789         int err;
 790
 791         ipvlan_init_secret();
 792         register_netdevice_notifier(&ipvlan_notifier_block);
 793         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 794         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 795
 796         err = ipvlan_link_register(&ipvlan_link_ops);
 797         if (err < 0)
 798                 goto error;
 799
 800         return 0;
 801 error:
 802         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 803         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 804         unregister_netdevice_notifier(&ipvlan_notifier_block);
 805         return err;
 806 }
 807
 808 static void __exit ipvlan_cleanup_module(void)
 809 {
 810         rtnl_link_unregister(&ipvlan_link_ops);
 811         unregister_netdevice_notifier(&ipvlan_notifier_block);
 812         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 813         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 814 }
 815
 816 module_init(ipvlan_init_module);
 817 module_exit(ipvlan_cleanup_module);
 818
 819 MODULE_LICENSE("GPL");
 820 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 821 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 822 MODULE_ALIAS_RTNL_LINK("ipvlan");