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 u32 ipvl_nf_hook_refcnt = 0;
  13
  14 static struct nf_hook_ops ipvl_nfops[] __read_mostly = {
  15         {
  16                 .hook     = ipvlan_nf_input,
  17                 .pf       = NFPROTO_IPV4,
  18                 .hooknum  = NF_INET_LOCAL_IN,
  19                 .priority = INT_MAX,
  20         },
  21         {
  22                 .hook     = ipvlan_nf_input,
  23                 .pf       = NFPROTO_IPV6,
  24                 .hooknum  = NF_INET_LOCAL_IN,
  25                 .priority = INT_MAX,
  26         },
  27 };
  28
  29 static struct l3mdev_ops ipvl_l3mdev_ops __read_mostly = {
  30         .l3mdev_l3_rcv = ipvlan_l3_rcv,
  31 };
  32
  33 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
  34 {
  35         ipvlan->dev->mtu = dev->mtu - ipvlan->mtu_adj;
  36 }
  37
  38 static int ipvlan_register_nf_hook(void)
  39 {
  40         int err = 0;
  41
  42         if (!ipvl_nf_hook_refcnt) {
  43                 err = _nf_register_hooks(ipvl_nfops, ARRAY_SIZE(ipvl_nfops));
  44                 if (!err)
  45                         ipvl_nf_hook_refcnt = 1;
  46         } else {
  47                 ipvl_nf_hook_refcnt++;
  48         }
  49
  50         return err;
  51 }
  52
  53 static void ipvlan_unregister_nf_hook(void)
  54 {
  55         WARN_ON(!ipvl_nf_hook_refcnt);
  56
  57         ipvl_nf_hook_refcnt--;
  58         if (!ipvl_nf_hook_refcnt)
  59                 _nf_unregister_hooks(ipvl_nfops, ARRAY_SIZE(ipvl_nfops));
  60 }
  61
  62 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval)
  63 {
  64         struct ipvl_dev *ipvlan;
  65         struct net_device *mdev = port->dev;
  66         int err = 0;
  67
  68         ASSERT_RTNL();
  69         if (port->mode != nval) {
  70                 if (nval == IPVLAN_MODE_L3S) {
  71                         /* New mode is L3S */
  72                         err = ipvlan_register_nf_hook();
  73                         if (!err) {
  74                                 mdev->l3mdev_ops = &ipvl_l3mdev_ops;
  75                                 mdev->priv_flags |= IFF_L3MDEV_MASTER;
  76                         } else
  77                                 return err;
  78                 } else if (port->mode == IPVLAN_MODE_L3S) {
  79                         /* Old mode was L3S */
  80                         mdev->priv_flags &= ~IFF_L3MDEV_MASTER;
  81                         ipvlan_unregister_nf_hook();
  82                         mdev->l3mdev_ops = NULL;
  83                 }
  84                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
  85                         if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S)
  86                                 ipvlan->dev->flags |= IFF_NOARP;
  87                         else
  88                                 ipvlan->dev->flags &= ~IFF_NOARP;
  89                 }
  90                 port->mode = nval;
  91         }
  92         return err;
  93 }
  94
  95 static int ipvlan_port_create(struct net_device *dev)
  96 {
  97         struct ipvl_port *port;
  98         int err, idx;
  99
 100         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) {
 101                 netdev_err(dev, "Master is either lo or non-ether device\n");
 102                 return -EINVAL;
 103         }
 104
 105         if (netif_is_macvlan_port(dev)) {
 106                 netdev_err(dev, "Master is a macvlan port.\n");
 107                 return -EBUSY;
 108         }
 109
 110         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
 111         if (!port)
 112                 return -ENOMEM;
 113
 114         port->dev = dev;
 115         port->mode = IPVLAN_MODE_L3;
 116         INIT_LIST_HEAD(&port->ipvlans);
 117         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
 118                 INIT_HLIST_HEAD(&port->hlhead[idx]);
 119
 120         skb_queue_head_init(&port->backlog);
 121         INIT_WORK(&port->wq, ipvlan_process_multicast);
 122
 123         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
 124         if (err)
 125                 goto err;
 126
 127         dev->priv_flags |= IFF_IPVLAN_MASTER;
 128         return 0;
 129
 130 err:
 131         kfree_rcu(port, rcu);
 132         return err;
 133 }
 134
 135 static void ipvlan_port_destroy(struct net_device *dev)
 136 {
 137         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
 138
 139         dev->priv_flags &= ~IFF_IPVLAN_MASTER;
 140         if (port->mode == IPVLAN_MODE_L3S) {
 141                 dev->priv_flags &= ~IFF_L3MDEV_MASTER;
 142                 ipvlan_unregister_nf_hook();
 143                 dev->l3mdev_ops = NULL;
 144         }
 145         netdev_rx_handler_unregister(dev);
 146         cancel_work_sync(&port->wq);
 147         __skb_queue_purge(&port->backlog);
 148         kfree_rcu(port, rcu);
 149 }
 150
 151 #define IPVLAN_FEATURES \
 152         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 153          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 154          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 155          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 156
 157 #define IPVLAN_STATE_MASK \
 158         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 159
 160 static int ipvlan_init(struct net_device *dev)
 161 {
 162         struct ipvl_dev *ipvlan = netdev_priv(dev);
 163         const struct net_device *phy_dev = ipvlan->phy_dev;
 164         struct ipvl_port *port = ipvlan->port;
 165
 166         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 167                      (phy_dev->state & IPVLAN_STATE_MASK);
 168         dev->features = phy_dev->features & IPVLAN_FEATURES;
 169         dev->features |= NETIF_F_LLTX;
 170         dev->gso_max_size = phy_dev->gso_max_size;
 171         dev->gso_max_segs = phy_dev->gso_max_segs;
 172         dev->hard_header_len = phy_dev->hard_header_len;
 173
 174         netdev_lockdep_set_classes(dev);
 175
 176         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 177         if (!ipvlan->pcpu_stats)
 178                 return -ENOMEM;
 179
 180         port->count += 1;
 181
 182         return 0;
 183 }
 184
 185 static void ipvlan_uninit(struct net_device *dev)
 186 {
 187         struct ipvl_dev *ipvlan = netdev_priv(dev);
 188         struct ipvl_port *port = ipvlan->port;
 189
 190         free_percpu(ipvlan->pcpu_stats);
 191
 192         port->count -= 1;
 193         if (!port->count)
 194                 ipvlan_port_destroy(port->dev);
 195 }
 196
 197 static int ipvlan_open(struct net_device *dev)
 198 {
 199         struct ipvl_dev *ipvlan = netdev_priv(dev);
 200         struct net_device *phy_dev = ipvlan->phy_dev;
 201         struct ipvl_addr *addr;
 202
 203         if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
 204             ipvlan->port->mode == IPVLAN_MODE_L3S)
 205                 dev->flags |= IFF_NOARP;
 206         else
 207                 dev->flags &= ~IFF_NOARP;
 208
 209         list_for_each_entry(addr, &ipvlan->addrs, anode)
 210                 ipvlan_ht_addr_add(ipvlan, addr);
 211
 212         return dev_uc_add(phy_dev, phy_dev->dev_addr);
 213 }
 214
 215 static int ipvlan_stop(struct net_device *dev)
 216 {
 217         struct ipvl_dev *ipvlan = netdev_priv(dev);
 218         struct net_device *phy_dev = ipvlan->phy_dev;
 219         struct ipvl_addr *addr;
 220
 221         dev_uc_unsync(phy_dev, dev);
 222         dev_mc_unsync(phy_dev, dev);
 223
 224         dev_uc_del(phy_dev, phy_dev->dev_addr);
 225
 226         list_for_each_entry(addr, &ipvlan->addrs, anode)
 227                 ipvlan_ht_addr_del(addr);
 228
 229         return 0;
 230 }
 231
 232 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 233                                      struct net_device *dev)
 234 {
 235         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 236         int skblen = skb->len;
 237         int ret;
 238
 239         ret = ipvlan_queue_xmit(skb, dev);
 240         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 241                 struct ipvl_pcpu_stats *pcptr;
 242
 243                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 244
 245                 u64_stats_update_begin(&pcptr->syncp);
 246                 pcptr->tx_pkts++;
 247                 pcptr->tx_bytes += skblen;
 248                 u64_stats_update_end(&pcptr->syncp);
 249         } else {
 250                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 251         }
 252         return ret;
 253 }
 254
 255 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 256                                              netdev_features_t features)
 257 {
 258         struct ipvl_dev *ipvlan = netdev_priv(dev);
 259
 260         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 261 }
 262
 263 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 264 {
 265         struct ipvl_dev *ipvlan = netdev_priv(dev);
 266         struct net_device *phy_dev = ipvlan->phy_dev;
 267
 268         if (change & IFF_ALLMULTI)
 269                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 270 }
 271
 272 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 273 {
 274         struct ipvl_dev *ipvlan = netdev_priv(dev);
 275
 276         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 277                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 278         } else {
 279                 struct netdev_hw_addr *ha;
 280                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 281
 282                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 283                 netdev_for_each_mc_addr(ha, dev)
 284                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 285
 286                 /* Turn-on broadcast bit irrespective of address family,
 287                  * since broadcast is deferred to a work-queue, hence no
 288                  * impact on fast-path processing.
 289                  */
 290                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 291
 292                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 293                             IPVLAN_MAC_FILTER_SIZE);
 294         }
 295         dev_uc_sync(ipvlan->phy_dev, dev);
 296         dev_mc_sync(ipvlan->phy_dev, dev);
 297 }
 298
 299 static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
 300                                                     struct rtnl_link_stats64 *s)
 301 {
 302         struct ipvl_dev *ipvlan = netdev_priv(dev);
 303
 304         if (ipvlan->pcpu_stats) {
 305                 struct ipvl_pcpu_stats *pcptr;
 306                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 307                 u32 rx_errs = 0, tx_drps = 0;
 308                 u32 strt;
 309                 int idx;
 310
 311                 for_each_possible_cpu(idx) {
 312                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 313                         do {
 314                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 315                                 rx_pkts = pcptr->rx_pkts;
 316                                 rx_bytes = pcptr->rx_bytes;
 317                                 rx_mcast = pcptr->rx_mcast;
 318                                 tx_pkts = pcptr->tx_pkts;
 319                                 tx_bytes = pcptr->tx_bytes;
 320                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 321                                                            strt));
 322
 323                         s->rx_packets += rx_pkts;
 324                         s->rx_bytes += rx_bytes;
 325                         s->multicast += rx_mcast;
 326                         s->tx_packets += tx_pkts;
 327                         s->tx_bytes += tx_bytes;
 328
 329                         /* u32 values are updated without syncp protection. */
 330                         rx_errs += pcptr->rx_errs;
 331                         tx_drps += pcptr->tx_drps;
 332                 }
 333                 s->rx_errors = rx_errs;
 334                 s->rx_dropped = rx_errs;
 335                 s->tx_dropped = tx_drps;
 336         }
 337         return s;
 338 }
 339
 340 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 341 {
 342         struct ipvl_dev *ipvlan = netdev_priv(dev);
 343         struct net_device *phy_dev = ipvlan->phy_dev;
 344
 345         return vlan_vid_add(phy_dev, proto, vid);
 346 }
 347
 348 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 349                                    u16 vid)
 350 {
 351         struct ipvl_dev *ipvlan = netdev_priv(dev);
 352         struct net_device *phy_dev = ipvlan->phy_dev;
 353
 354         vlan_vid_del(phy_dev, proto, vid);
 355         return 0;
 356 }
 357
 358 static int ipvlan_get_iflink(const struct net_device *dev)
 359 {
 360         struct ipvl_dev *ipvlan = netdev_priv(dev);
 361
 362         return ipvlan->phy_dev->ifindex;
 363 }
 364
 365 static const struct net_device_ops ipvlan_netdev_ops = {
 366         .ndo_init               = ipvlan_init,
 367         .ndo_uninit             = ipvlan_uninit,
 368         .ndo_open               = ipvlan_open,
 369         .ndo_stop               = ipvlan_stop,
 370         .ndo_start_xmit         = ipvlan_start_xmit,
 371         .ndo_fix_features       = ipvlan_fix_features,
 372         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 373         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 374         .ndo_get_stats64        = ipvlan_get_stats64,
 375         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 376         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 377         .ndo_get_iflink         = ipvlan_get_iflink,
 378 };
 379
 380 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 381                               unsigned short type, const void *daddr,
 382                               const void *saddr, unsigned len)
 383 {
 384         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 385         struct net_device *phy_dev = ipvlan->phy_dev;
 386
 387         /* TODO Probably use a different field than dev_addr so that the
 388          * mac-address on the virtual device is portable and can be carried
 389          * while the packets use the mac-addr on the physical device.
 390          */
 391         return dev_hard_header(skb, phy_dev, type, daddr,
 392                                saddr ? : dev->dev_addr, len);
 393 }
 394
 395 static const struct header_ops ipvlan_header_ops = {
 396         .create         = ipvlan_hard_header,
 397         .parse          = eth_header_parse,
 398         .cache          = eth_header_cache,
 399         .cache_update   = eth_header_cache_update,
 400 };
 401
 402 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 403                                              struct ethtool_link_ksettings *cmd)
 404 {
 405         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 406
 407         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 408 }
 409
 410 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 411                                        struct ethtool_drvinfo *drvinfo)
 412 {
 413         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 414         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 415 }
 416
 417 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 418 {
 419         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 420
 421         return ipvlan->msg_enable;
 422 }
 423
 424 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 425 {
 426         struct ipvl_dev *ipvlan = netdev_priv(dev);
 427
 428         ipvlan->msg_enable = value;
 429 }
 430
 431 static const struct ethtool_ops ipvlan_ethtool_ops = {
 432         .get_link       = ethtool_op_get_link,
 433         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 434         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 435         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 436         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 437 };
 438
 439 static int ipvlan_nl_changelink(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 = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 444         int err = 0;
 445
 446         if (data && data[IFLA_IPVLAN_MODE]) {
 447                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 448
 449                 err = ipvlan_set_port_mode(port, nmode);
 450         }
 451         return err;
 452 }
 453
 454 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 455 {
 456         return (0
 457                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 458                 );
 459 }
 460
 461 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 462 {
 463         if (data && data[IFLA_IPVLAN_MODE]) {
 464                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 465
 466                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 467                         return -EINVAL;
 468         }
 469         return 0;
 470 }
 471
 472 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 473                               const struct net_device *dev)
 474 {
 475         struct ipvl_dev *ipvlan = netdev_priv(dev);
 476         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 477         int ret = -EINVAL;
 478
 479         if (!port)
 480                 goto err;
 481
 482         ret = -EMSGSIZE;
 483         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 484                 goto err;
 485
 486         return 0;
 487
 488 err:
 489         return ret;
 490 }
 491
 492 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 493                            struct nlattr *tb[], struct nlattr *data[])
 494 {
 495         struct ipvl_dev *ipvlan = netdev_priv(dev);
 496         struct ipvl_port *port;
 497         struct net_device *phy_dev;
 498         int err;
 499         u16 mode = IPVLAN_MODE_L3;
 500
 501         if (!tb[IFLA_LINK])
 502                 return -EINVAL;
 503
 504         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 505         if (!phy_dev)
 506                 return -ENODEV;
 507
 508         if (netif_is_ipvlan(phy_dev)) {
 509                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 510
 511                 phy_dev = tmp->phy_dev;
 512         } else if (!netif_is_ipvlan_port(phy_dev)) {
 513                 err = ipvlan_port_create(phy_dev);
 514                 if (err < 0)
 515                         return err;
 516         }
 517
 518         if (data && data[IFLA_IPVLAN_MODE])
 519                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 520
 521         port = ipvlan_port_get_rtnl(phy_dev);
 522         ipvlan->phy_dev = phy_dev;
 523         ipvlan->dev = dev;
 524         ipvlan->port = port;
 525         ipvlan->sfeatures = IPVLAN_FEATURES;
 526         ipvlan_adjust_mtu(ipvlan, phy_dev);
 527         INIT_LIST_HEAD(&ipvlan->addrs);
 528
 529         /* TODO Probably put random address here to be presented to the
 530          * world but keep using the physical-dev address for the outgoing
 531          * packets.
 532          */
 533         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 534
 535         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 536
 537         err = register_netdevice(dev);
 538         if (err < 0)
 539                 return err;
 540
 541         err = netdev_upper_dev_link(phy_dev, dev);
 542         if (err) {
 543                 unregister_netdevice(dev);
 544                 return err;
 545         }
 546         err = ipvlan_set_port_mode(port, mode);
 547         if (err) {
 548                 unregister_netdevice(dev);
 549                 return err;
 550         }
 551
 552         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 553         netif_stacked_transfer_operstate(phy_dev, dev);
 554         return 0;
 555 }
 556
 557 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 558 {
 559         struct ipvl_dev *ipvlan = netdev_priv(dev);
 560         struct ipvl_addr *addr, *next;
 561
 562         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 563                 ipvlan_ht_addr_del(addr);
 564                 list_del(&addr->anode);
 565                 kfree_rcu(addr, rcu);
 566         }
 567
 568         list_del_rcu(&ipvlan->pnode);
 569         unregister_netdevice_queue(dev, head);
 570         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 571 }
 572
 573 static void ipvlan_link_setup(struct net_device *dev)
 574 {
 575         ether_setup(dev);
 576
 577         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 578         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 579         dev->netdev_ops = &ipvlan_netdev_ops;
 580         dev->destructor = free_netdev;
 581         dev->header_ops = &ipvlan_header_ops;
 582         dev->ethtool_ops = &ipvlan_ethtool_ops;
 583 }
 584
 585 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 586 {
 587         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 588 };
 589
 590 static struct rtnl_link_ops ipvlan_link_ops = {
 591         .kind           = "ipvlan",
 592         .priv_size      = sizeof(struct ipvl_dev),
 593
 594         .get_size       = ipvlan_nl_getsize,
 595         .policy         = ipvlan_nl_policy,
 596         .validate       = ipvlan_nl_validate,
 597         .fill_info      = ipvlan_nl_fillinfo,
 598         .changelink     = ipvlan_nl_changelink,
 599         .maxtype        = IFLA_IPVLAN_MAX,
 600
 601         .setup          = ipvlan_link_setup,
 602         .newlink        = ipvlan_link_new,
 603         .dellink        = ipvlan_link_delete,
 604 };
 605
 606 static int ipvlan_link_register(struct rtnl_link_ops *ops)
 607 {
 608         return rtnl_link_register(ops);
 609 }
 610
 611 static int ipvlan_device_event(struct notifier_block *unused,
 612                                unsigned long event, void *ptr)
 613 {
 614         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 615         struct ipvl_dev *ipvlan, *next;
 616         struct ipvl_port *port;
 617         LIST_HEAD(lst_kill);
 618
 619         if (!netif_is_ipvlan_port(dev))
 620                 return NOTIFY_DONE;
 621
 622         port = ipvlan_port_get_rtnl(dev);
 623
 624         switch (event) {
 625         case NETDEV_CHANGE:
 626                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 627                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 628                                                          ipvlan->dev);
 629                 break;
 630
 631         case NETDEV_UNREGISTER:
 632                 if (dev->reg_state != NETREG_UNREGISTERING)
 633                         break;
 634
 635                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 636                                          pnode)
 637                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 638                                                             &lst_kill);
 639                 unregister_netdevice_many(&lst_kill);
 640                 break;
 641
 642         case NETDEV_FEAT_CHANGE:
 643                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 644                         ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
 645                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 646                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 647                         netdev_features_change(ipvlan->dev);
 648                 }
 649                 break;
 650
 651         case NETDEV_CHANGEMTU:
 652                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 653                         ipvlan_adjust_mtu(ipvlan, dev);
 654                 break;
 655
 656         case NETDEV_PRE_TYPE_CHANGE:
 657                 /* Forbid underlying device to change its type. */
 658                 return NOTIFY_BAD;
 659         }
 660         return NOTIFY_DONE;
 661 }
 662
 663 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 664 {
 665         struct ipvl_addr *addr;
 666
 667         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 668                 netif_err(ipvlan, ifup, ipvlan->dev,
 669                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 670                           ip6_addr, ipvlan->dev->name);
 671                 return -EINVAL;
 672         }
 673         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 674         if (!addr)
 675                 return -ENOMEM;
 676
 677         addr->master = ipvlan;
 678         memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
 679         addr->atype = IPVL_IPV6;
 680         list_add_tail(&addr->anode, &ipvlan->addrs);
 681
 682         /* If the interface is not up, the address will be added to the hash
 683          * list by ipvlan_open.
 684          */
 685         if (netif_running(ipvlan->dev))
 686                 ipvlan_ht_addr_add(ipvlan, addr);
 687
 688         return 0;
 689 }
 690
 691 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 692 {
 693         struct ipvl_addr *addr;
 694
 695         addr = ipvlan_find_addr(ipvlan, ip6_addr, true);
 696         if (!addr)
 697                 return;
 698
 699         ipvlan_ht_addr_del(addr);
 700         list_del(&addr->anode);
 701         kfree_rcu(addr, rcu);
 702
 703         return;
 704 }
 705
 706 static int ipvlan_addr6_event(struct notifier_block *unused,
 707                               unsigned long event, void *ptr)
 708 {
 709         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 710         struct net_device *dev = (struct net_device *)if6->idev->dev;
 711         struct ipvl_dev *ipvlan = netdev_priv(dev);
 712
 713         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 714         if (in_softirq())
 715                 return NOTIFY_DONE;
 716
 717         if (!netif_is_ipvlan(dev))
 718                 return NOTIFY_DONE;
 719
 720         if (!ipvlan || !ipvlan->port)
 721                 return NOTIFY_DONE;
 722
 723         switch (event) {
 724         case NETDEV_UP:
 725                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 726                         return NOTIFY_BAD;
 727                 break;
 728
 729         case NETDEV_DOWN:
 730                 ipvlan_del_addr6(ipvlan, &if6->addr);
 731                 break;
 732         }
 733
 734         return NOTIFY_OK;
 735 }
 736
 737 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 738 {
 739         struct ipvl_addr *addr;
 740
 741         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 742                 netif_err(ipvlan, ifup, ipvlan->dev,
 743                           "Failed to add IPv4=%pI4 on %s intf.\n",
 744                           ip4_addr, ipvlan->dev->name);
 745                 return -EINVAL;
 746         }
 747         addr = kzalloc(sizeof(struct ipvl_addr), GFP_KERNEL);
 748         if (!addr)
 749                 return -ENOMEM;
 750
 751         addr->master = ipvlan;
 752         memcpy(&addr->ip4addr, ip4_addr, sizeof(struct in_addr));
 753         addr->atype = IPVL_IPV4;
 754         list_add_tail(&addr->anode, &ipvlan->addrs);
 755
 756         /* If the interface is not up, the address will be added to the hash
 757          * list by ipvlan_open.
 758          */
 759         if (netif_running(ipvlan->dev))
 760                 ipvlan_ht_addr_add(ipvlan, addr);
 761
 762         return 0;
 763 }
 764
 765 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 766 {
 767         struct ipvl_addr *addr;
 768
 769         addr = ipvlan_find_addr(ipvlan, ip4_addr, false);
 770         if (!addr)
 771                 return;
 772
 773         ipvlan_ht_addr_del(addr);
 774         list_del(&addr->anode);
 775         kfree_rcu(addr, rcu);
 776
 777         return;
 778 }
 779
 780 static int ipvlan_addr4_event(struct notifier_block *unused,
 781                               unsigned long event, void *ptr)
 782 {
 783         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 784         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 785         struct ipvl_dev *ipvlan = netdev_priv(dev);
 786         struct in_addr ip4_addr;
 787
 788         if (!netif_is_ipvlan(dev))
 789                 return NOTIFY_DONE;
 790
 791         if (!ipvlan || !ipvlan->port)
 792                 return NOTIFY_DONE;
 793
 794         switch (event) {
 795         case NETDEV_UP:
 796                 ip4_addr.s_addr = if4->ifa_address;
 797                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 798                         return NOTIFY_BAD;
 799                 break;
 800
 801         case NETDEV_DOWN:
 802                 ip4_addr.s_addr = if4->ifa_address;
 803                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 804                 break;
 805         }
 806
 807         return NOTIFY_OK;
 808 }
 809
 810 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 811         .notifier_call = ipvlan_addr4_event,
 812 };
 813
 814 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 815         .notifier_call = ipvlan_device_event,
 816 };
 817
 818 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 819         .notifier_call = ipvlan_addr6_event,
 820 };
 821
 822 static int __init ipvlan_init_module(void)
 823 {
 824         int err;
 825
 826         ipvlan_init_secret();
 827         register_netdevice_notifier(&ipvlan_notifier_block);
 828         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 829         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 830
 831         err = ipvlan_link_register(&ipvlan_link_ops);
 832         if (err < 0)
 833                 goto error;
 834
 835         return 0;
 836 error:
 837         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 838         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 839         unregister_netdevice_notifier(&ipvlan_notifier_block);
 840         return err;
 841 }
 842
 843 static void __exit ipvlan_cleanup_module(void)
 844 {
 845         rtnl_link_unregister(&ipvlan_link_ops);
 846         unregister_netdevice_notifier(&ipvlan_notifier_block);
 847         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 848         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 849 }
 850
 851 module_init(ipvlan_init_module);
 852 module_exit(ipvlan_cleanup_module);
 853
 854 MODULE_LICENSE("GPL");
 855 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 856 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 857 MODULE_ALIAS_RTNL_LINK("ipvlan");