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 const struct l3mdev_ops ipvl_l3mdev_ops = {
  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;
  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         ida_init(&port->ida);
 123         port->dev_id_start = 1;
 124
 125         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
 126         if (err)
 127                 goto err;
 128
 129         dev->priv_flags |= IFF_IPVLAN_MASTER;
 130         return 0;
 131
 132 err:
 133         kfree(port);
 134         return err;
 135 }
 136
 137 static void ipvlan_port_destroy(struct net_device *dev)
 138 {
 139         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
 140         struct sk_buff *skb;
 141
 142         dev->priv_flags &= ~IFF_IPVLAN_MASTER;
 143         if (port->mode == IPVLAN_MODE_L3S) {
 144                 dev->priv_flags &= ~IFF_L3MDEV_MASTER;
 145                 ipvlan_unregister_nf_hook();
 146                 dev->l3mdev_ops = NULL;
 147         }
 148         netdev_rx_handler_unregister(dev);
 149         cancel_work_sync(&port->wq);
 150         while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
 151                 if (skb->dev)
 152                         dev_put(skb->dev);
 153                 kfree_skb(skb);
 154         }
 155         ida_destroy(&port->ida);
 156         kfree(port);
 157 }
 158
 159 #define IPVLAN_FEATURES \
 160         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 161          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 162          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 163          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 164
 165 #define IPVLAN_STATE_MASK \
 166         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 167
 168 static int ipvlan_init(struct net_device *dev)
 169 {
 170         struct ipvl_dev *ipvlan = netdev_priv(dev);
 171         const struct net_device *phy_dev = ipvlan->phy_dev;
 172         struct ipvl_port *port = ipvlan->port;
 173
 174         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 175                      (phy_dev->state & IPVLAN_STATE_MASK);
 176         dev->features = phy_dev->features & IPVLAN_FEATURES;
 177         dev->features |= NETIF_F_LLTX;
 178         dev->gso_max_size = phy_dev->gso_max_size;
 179         dev->gso_max_segs = phy_dev->gso_max_segs;
 180         dev->hard_header_len = phy_dev->hard_header_len;
 181
 182         netdev_lockdep_set_classes(dev);
 183
 184         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 185         if (!ipvlan->pcpu_stats)
 186                 return -ENOMEM;
 187
 188         port->count += 1;
 189
 190         return 0;
 191 }
 192
 193 static void ipvlan_uninit(struct net_device *dev)
 194 {
 195         struct ipvl_dev *ipvlan = netdev_priv(dev);
 196         struct ipvl_port *port = ipvlan->port;
 197
 198         free_percpu(ipvlan->pcpu_stats);
 199
 200         port->count -= 1;
 201         if (!port->count)
 202                 ipvlan_port_destroy(port->dev);
 203 }
 204
 205 static int ipvlan_open(struct net_device *dev)
 206 {
 207         struct ipvl_dev *ipvlan = netdev_priv(dev);
 208         struct net_device *phy_dev = ipvlan->phy_dev;
 209         struct ipvl_addr *addr;
 210
 211         if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
 212             ipvlan->port->mode == IPVLAN_MODE_L3S)
 213                 dev->flags |= IFF_NOARP;
 214         else
 215                 dev->flags &= ~IFF_NOARP;
 216
 217         list_for_each_entry(addr, &ipvlan->addrs, anode)
 218                 ipvlan_ht_addr_add(ipvlan, addr);
 219
 220         return dev_uc_add(phy_dev, phy_dev->dev_addr);
 221 }
 222
 223 static int ipvlan_stop(struct net_device *dev)
 224 {
 225         struct ipvl_dev *ipvlan = netdev_priv(dev);
 226         struct net_device *phy_dev = ipvlan->phy_dev;
 227         struct ipvl_addr *addr;
 228
 229         dev_uc_unsync(phy_dev, dev);
 230         dev_mc_unsync(phy_dev, dev);
 231
 232         dev_uc_del(phy_dev, phy_dev->dev_addr);
 233
 234         list_for_each_entry(addr, &ipvlan->addrs, anode)
 235                 ipvlan_ht_addr_del(addr);
 236
 237         return 0;
 238 }
 239
 240 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 241                                      struct net_device *dev)
 242 {
 243         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 244         int skblen = skb->len;
 245         int ret;
 246
 247         ret = ipvlan_queue_xmit(skb, dev);
 248         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 249                 struct ipvl_pcpu_stats *pcptr;
 250
 251                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 252
 253                 u64_stats_update_begin(&pcptr->syncp);
 254                 pcptr->tx_pkts++;
 255                 pcptr->tx_bytes += skblen;
 256                 u64_stats_update_end(&pcptr->syncp);
 257         } else {
 258                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 259         }
 260         return ret;
 261 }
 262
 263 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 264                                              netdev_features_t features)
 265 {
 266         struct ipvl_dev *ipvlan = netdev_priv(dev);
 267
 268         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 269 }
 270
 271 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 272 {
 273         struct ipvl_dev *ipvlan = netdev_priv(dev);
 274         struct net_device *phy_dev = ipvlan->phy_dev;
 275
 276         if (change & IFF_ALLMULTI)
 277                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 278 }
 279
 280 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 281 {
 282         struct ipvl_dev *ipvlan = netdev_priv(dev);
 283
 284         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 285                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 286         } else {
 287                 struct netdev_hw_addr *ha;
 288                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 289
 290                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 291                 netdev_for_each_mc_addr(ha, dev)
 292                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 293
 294                 /* Turn-on broadcast bit irrespective of address family,
 295                  * since broadcast is deferred to a work-queue, hence no
 296                  * impact on fast-path processing.
 297                  */
 298                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 299
 300                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 301                             IPVLAN_MAC_FILTER_SIZE);
 302         }
 303         dev_uc_sync(ipvlan->phy_dev, dev);
 304         dev_mc_sync(ipvlan->phy_dev, dev);
 305 }
 306
 307 static void ipvlan_get_stats64(struct net_device *dev,
 308                                struct rtnl_link_stats64 *s)
 309 {
 310         struct ipvl_dev *ipvlan = netdev_priv(dev);
 311
 312         if (ipvlan->pcpu_stats) {
 313                 struct ipvl_pcpu_stats *pcptr;
 314                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 315                 u32 rx_errs = 0, tx_drps = 0;
 316                 u32 strt;
 317                 int idx;
 318
 319                 for_each_possible_cpu(idx) {
 320                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 321                         do {
 322                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 323                                 rx_pkts = pcptr->rx_pkts;
 324                                 rx_bytes = pcptr->rx_bytes;
 325                                 rx_mcast = pcptr->rx_mcast;
 326                                 tx_pkts = pcptr->tx_pkts;
 327                                 tx_bytes = pcptr->tx_bytes;
 328                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 329                                                            strt));
 330
 331                         s->rx_packets += rx_pkts;
 332                         s->rx_bytes += rx_bytes;
 333                         s->multicast += rx_mcast;
 334                         s->tx_packets += tx_pkts;
 335                         s->tx_bytes += tx_bytes;
 336
 337                         /* u32 values are updated without syncp protection. */
 338                         rx_errs += pcptr->rx_errs;
 339                         tx_drps += pcptr->tx_drps;
 340                 }
 341                 s->rx_errors = rx_errs;
 342                 s->rx_dropped = rx_errs;
 343                 s->tx_dropped = tx_drps;
 344         }
 345 }
 346
 347 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 348 {
 349         struct ipvl_dev *ipvlan = netdev_priv(dev);
 350         struct net_device *phy_dev = ipvlan->phy_dev;
 351
 352         return vlan_vid_add(phy_dev, proto, vid);
 353 }
 354
 355 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 356                                    u16 vid)
 357 {
 358         struct ipvl_dev *ipvlan = netdev_priv(dev);
 359         struct net_device *phy_dev = ipvlan->phy_dev;
 360
 361         vlan_vid_del(phy_dev, proto, vid);
 362         return 0;
 363 }
 364
 365 static int ipvlan_get_iflink(const struct net_device *dev)
 366 {
 367         struct ipvl_dev *ipvlan = netdev_priv(dev);
 368
 369         return ipvlan->phy_dev->ifindex;
 370 }
 371
 372 static const struct net_device_ops ipvlan_netdev_ops = {
 373         .ndo_init               = ipvlan_init,
 374         .ndo_uninit             = ipvlan_uninit,
 375         .ndo_open               = ipvlan_open,
 376         .ndo_stop               = ipvlan_stop,
 377         .ndo_start_xmit         = ipvlan_start_xmit,
 378         .ndo_fix_features       = ipvlan_fix_features,
 379         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 380         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 381         .ndo_get_stats64        = ipvlan_get_stats64,
 382         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 383         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 384         .ndo_get_iflink         = ipvlan_get_iflink,
 385 };
 386
 387 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 388                               unsigned short type, const void *daddr,
 389                               const void *saddr, unsigned len)
 390 {
 391         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 392         struct net_device *phy_dev = ipvlan->phy_dev;
 393
 394         /* TODO Probably use a different field than dev_addr so that the
 395          * mac-address on the virtual device is portable and can be carried
 396          * while the packets use the mac-addr on the physical device.
 397          */
 398         return dev_hard_header(skb, phy_dev, type, daddr,
 399                                saddr ? : dev->dev_addr, len);
 400 }
 401
 402 static const struct header_ops ipvlan_header_ops = {
 403         .create         = ipvlan_hard_header,
 404         .parse          = eth_header_parse,
 405         .cache          = eth_header_cache,
 406         .cache_update   = eth_header_cache_update,
 407 };
 408
 409 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 410                                              struct ethtool_link_ksettings *cmd)
 411 {
 412         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 413
 414         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 415 }
 416
 417 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 418                                        struct ethtool_drvinfo *drvinfo)
 419 {
 420         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 421         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 422 }
 423
 424 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 425 {
 426         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 427
 428         return ipvlan->msg_enable;
 429 }
 430
 431 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 432 {
 433         struct ipvl_dev *ipvlan = netdev_priv(dev);
 434
 435         ipvlan->msg_enable = value;
 436 }
 437
 438 static const struct ethtool_ops ipvlan_ethtool_ops = {
 439         .get_link       = ethtool_op_get_link,
 440         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 441         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 442         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 443         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 444 };
 445
 446 static int ipvlan_nl_changelink(struct net_device *dev,
 447                                 struct nlattr *tb[], struct nlattr *data[])
 448 {
 449         struct ipvl_dev *ipvlan = netdev_priv(dev);
 450         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 451         int err = 0;
 452
 453         if (data && data[IFLA_IPVLAN_MODE]) {
 454                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 455
 456                 err = ipvlan_set_port_mode(port, nmode);
 457         }
 458         return err;
 459 }
 460
 461 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 462 {
 463         return (0
 464                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 465                 );
 466 }
 467
 468 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 469 {
 470         if (data && data[IFLA_IPVLAN_MODE]) {
 471                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 472
 473                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 474                         return -EINVAL;
 475         }
 476         return 0;
 477 }
 478
 479 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 480                               const struct net_device *dev)
 481 {
 482         struct ipvl_dev *ipvlan = netdev_priv(dev);
 483         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 484         int ret = -EINVAL;
 485
 486         if (!port)
 487                 goto err;
 488
 489         ret = -EMSGSIZE;
 490         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 491                 goto err;
 492
 493         return 0;
 494
 495 err:
 496         return ret;
 497 }
 498
 499 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 500                            struct nlattr *tb[], struct nlattr *data[])
 501 {
 502         struct ipvl_dev *ipvlan = netdev_priv(dev);
 503         struct ipvl_port *port;
 504         struct net_device *phy_dev;
 505         int err;
 506         u16 mode = IPVLAN_MODE_L3;
 507         bool create = false;
 508
 509         if (!tb[IFLA_LINK])
 510                 return -EINVAL;
 511
 512         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 513         if (!phy_dev)
 514                 return -ENODEV;
 515
 516         if (netif_is_ipvlan(phy_dev)) {
 517                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 518
 519                 phy_dev = tmp->phy_dev;
 520         } else if (!netif_is_ipvlan_port(phy_dev)) {
 521                 err = ipvlan_port_create(phy_dev);
 522                 if (err < 0)
 523                         return err;
 524                 create = true;
 525         }
 526
 527         if (data && data[IFLA_IPVLAN_MODE])
 528                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 529
 530         port = ipvlan_port_get_rtnl(phy_dev);
 531         ipvlan->phy_dev = phy_dev;
 532         ipvlan->dev = dev;
 533         ipvlan->port = port;
 534         ipvlan->sfeatures = IPVLAN_FEATURES;
 535         ipvlan_adjust_mtu(ipvlan, phy_dev);
 536         INIT_LIST_HEAD(&ipvlan->addrs);
 537
 538         /* If the port-id base is at the MAX value, then wrap it around and
 539          * begin from 0x1 again. This may be due to a busy system where lots
 540          * of slaves are getting created and deleted.
 541          */
 542         if (port->dev_id_start == 0xFFFE)
 543                 port->dev_id_start = 0x1;
 544
 545         /* Since L2 address is shared among all IPvlan slaves including
 546          * master, use unique 16 bit dev-ids to diffentiate among them.
 547          * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
 548          * slave link [see addrconf_ifid_eui48()].
 549          */
 550         err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
 551                              GFP_KERNEL);
 552         if (err < 0)
 553                 err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
 554                                      GFP_KERNEL);
 555         if (err < 0)
 556                 goto destroy_ipvlan_port;
 557         dev->dev_id = err;
 558         /* Increment id-base to the next slot for the future assignment */
 559         port->dev_id_start = err + 1;
 560
 561         /* TODO Probably put random address here to be presented to the
 562          * world but keep using the physical-dev address for the outgoing
 563          * packets.
 564          */
 565         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 566
 567         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 568
 569         err = register_netdevice(dev);
 570         if (err < 0)
 571                 goto remove_ida;
 572
 573         err = netdev_upper_dev_link(phy_dev, dev);
 574         if (err) {
 575                 goto unregister_netdev;
 576         }
 577         err = ipvlan_set_port_mode(port, mode);
 578         if (err) {
 579                 goto unlink_netdev;
 580         }
 581
 582         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 583         netif_stacked_transfer_operstate(phy_dev, dev);
 584         return 0;
 585
 586 unlink_netdev:
 587         netdev_upper_dev_unlink(phy_dev, dev);
 588 unregister_netdev:
 589         unregister_netdevice(dev);
 590 remove_ida:
 591         ida_simple_remove(&port->ida, dev->dev_id);
 592 destroy_ipvlan_port:
 593         if (create)
 594                 ipvlan_port_destroy(phy_dev);
 595         return err;
 596 }
 597
 598 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 599 {
 600         struct ipvl_dev *ipvlan = netdev_priv(dev);
 601         struct ipvl_addr *addr, *next;
 602
 603         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 604                 ipvlan_ht_addr_del(addr);
 605                 list_del(&addr->anode);
 606                 kfree_rcu(addr, rcu);
 607         }
 608
 609         ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
 610         list_del_rcu(&ipvlan->pnode);
 611         unregister_netdevice_queue(dev, head);
 612         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 613 }
 614
 615 static void ipvlan_link_setup(struct net_device *dev)
 616 {
 617         ether_setup(dev);
 618
 619         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 620         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 621         dev->netdev_ops = &ipvlan_netdev_ops;
 622         dev->destructor = free_netdev;
 623         dev->header_ops = &ipvlan_header_ops;
 624         dev->ethtool_ops = &ipvlan_ethtool_ops;
 625 }
 626
 627 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 628 {
 629         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 630 };
 631
 632 static struct rtnl_link_ops ipvlan_link_ops = {
 633         .kind           = "ipvlan",
 634         .priv_size      = sizeof(struct ipvl_dev),
 635
 636         .get_size       = ipvlan_nl_getsize,
 637         .policy         = ipvlan_nl_policy,
 638         .validate       = ipvlan_nl_validate,
 639         .fill_info      = ipvlan_nl_fillinfo,
 640         .changelink     = ipvlan_nl_changelink,
 641         .maxtype        = IFLA_IPVLAN_MAX,
 642
 643         .setup          = ipvlan_link_setup,
 644         .newlink        = ipvlan_link_new,
 645         .dellink        = ipvlan_link_delete,
 646 };
 647
 648 static int ipvlan_link_register(struct rtnl_link_ops *ops)
 649 {
 650         return rtnl_link_register(ops);
 651 }
 652
 653 static int ipvlan_device_event(struct notifier_block *unused,
 654                                unsigned long event, void *ptr)
 655 {
 656         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 657         struct ipvl_dev *ipvlan, *next;
 658         struct ipvl_port *port;
 659         LIST_HEAD(lst_kill);
 660
 661         if (!netif_is_ipvlan_port(dev))
 662                 return NOTIFY_DONE;
 663
 664         port = ipvlan_port_get_rtnl(dev);
 665
 666         switch (event) {
 667         case NETDEV_CHANGE:
 668                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 669                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 670                                                          ipvlan->dev);
 671                 break;
 672
 673         case NETDEV_UNREGISTER:
 674                 if (dev->reg_state != NETREG_UNREGISTERING)
 675                         break;
 676
 677                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 678                                          pnode)
 679                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 680                                                             &lst_kill);
 681                 unregister_netdevice_many(&lst_kill);
 682                 break;
 683
 684         case NETDEV_FEAT_CHANGE:
 685                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 686                         ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
 687                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 688                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 689                         netdev_features_change(ipvlan->dev);
 690                 }
 691                 break;
 692
 693         case NETDEV_CHANGEMTU:
 694                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 695                         ipvlan_adjust_mtu(ipvlan, dev);
 696                 break;
 697
 698         case NETDEV_PRE_TYPE_CHANGE:
 699                 /* Forbid underlying device to change its type. */
 700                 return NOTIFY_BAD;
 701         }
 702         return NOTIFY_DONE;
 703 }
 704
 705 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 706 {
 707         struct ipvl_addr *addr;
 708
 709         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 710         if (!addr)
 711                 return -ENOMEM;
 712
 713         addr->master = ipvlan;
 714         if (is_v6) {
 715                 memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
 716                 addr->atype = IPVL_IPV6;
 717         } else {
 718                 memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
 719                 addr->atype = IPVL_IPV4;
 720         }
 721         list_add_tail(&addr->anode, &ipvlan->addrs);
 722
 723         /* If the interface is not up, the address will be added to the hash
 724          * list by ipvlan_open.
 725          */
 726         if (netif_running(ipvlan->dev))
 727                 ipvlan_ht_addr_add(ipvlan, addr);
 728
 729         return 0;
 730 }
 731
 732 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 733 {
 734         struct ipvl_addr *addr;
 735
 736         addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
 737         if (!addr)
 738                 return;
 739
 740         ipvlan_ht_addr_del(addr);
 741         list_del(&addr->anode);
 742         kfree_rcu(addr, rcu);
 743
 744         return;
 745 }
 746
 747 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 748 {
 749         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 750                 netif_err(ipvlan, ifup, ipvlan->dev,
 751                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 752                           ip6_addr, ipvlan->dev->name);
 753                 return -EINVAL;
 754         }
 755
 756         return ipvlan_add_addr(ipvlan, ip6_addr, true);
 757 }
 758
 759 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 760 {
 761         return ipvlan_del_addr(ipvlan, ip6_addr, true);
 762 }
 763
 764 static int ipvlan_addr6_event(struct notifier_block *unused,
 765                               unsigned long event, void *ptr)
 766 {
 767         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 768         struct net_device *dev = (struct net_device *)if6->idev->dev;
 769         struct ipvl_dev *ipvlan = netdev_priv(dev);
 770
 771         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 772         if (in_softirq())
 773                 return NOTIFY_DONE;
 774
 775         if (!netif_is_ipvlan(dev))
 776                 return NOTIFY_DONE;
 777
 778         if (!ipvlan || !ipvlan->port)
 779                 return NOTIFY_DONE;
 780
 781         switch (event) {
 782         case NETDEV_UP:
 783                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 784                         return NOTIFY_BAD;
 785                 break;
 786
 787         case NETDEV_DOWN:
 788                 ipvlan_del_addr6(ipvlan, &if6->addr);
 789                 break;
 790         }
 791
 792         return NOTIFY_OK;
 793 }
 794
 795 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 796 {
 797         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 798                 netif_err(ipvlan, ifup, ipvlan->dev,
 799                           "Failed to add IPv4=%pI4 on %s intf.\n",
 800                           ip4_addr, ipvlan->dev->name);
 801                 return -EINVAL;
 802         }
 803
 804         return ipvlan_add_addr(ipvlan, ip4_addr, false);
 805 }
 806
 807 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 808 {
 809         return ipvlan_del_addr(ipvlan, ip4_addr, false);
 810 }
 811
 812 static int ipvlan_addr4_event(struct notifier_block *unused,
 813                               unsigned long event, void *ptr)
 814 {
 815         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 816         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 817         struct ipvl_dev *ipvlan = netdev_priv(dev);
 818         struct in_addr ip4_addr;
 819
 820         if (!netif_is_ipvlan(dev))
 821                 return NOTIFY_DONE;
 822
 823         if (!ipvlan || !ipvlan->port)
 824                 return NOTIFY_DONE;
 825
 826         switch (event) {
 827         case NETDEV_UP:
 828                 ip4_addr.s_addr = if4->ifa_address;
 829                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 830                         return NOTIFY_BAD;
 831                 break;
 832
 833         case NETDEV_DOWN:
 834                 ip4_addr.s_addr = if4->ifa_address;
 835                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 836                 break;
 837         }
 838
 839         return NOTIFY_OK;
 840 }
 841
 842 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 843         .notifier_call = ipvlan_addr4_event,
 844 };
 845
 846 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 847         .notifier_call = ipvlan_device_event,
 848 };
 849
 850 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 851         .notifier_call = ipvlan_addr6_event,
 852 };
 853
 854 static int __init ipvlan_init_module(void)
 855 {
 856         int err;
 857
 858         ipvlan_init_secret();
 859         register_netdevice_notifier(&ipvlan_notifier_block);
 860         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 861         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 862
 863         err = ipvlan_link_register(&ipvlan_link_ops);
 864         if (err < 0)
 865                 goto error;
 866
 867         return 0;
 868 error:
 869         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 870         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 871         unregister_netdevice_notifier(&ipvlan_notifier_block);
 872         return err;
 873 }
 874
 875 static void __exit ipvlan_cleanup_module(void)
 876 {
 877         rtnl_link_unregister(&ipvlan_link_ops);
 878         unregister_netdevice_notifier(&ipvlan_notifier_block);
 879         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 880         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 881 }
 882
 883 module_init(ipvlan_init_module);
 884 module_exit(ipvlan_cleanup_module);
 885
 886 MODULE_LICENSE("GPL");
 887 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 888 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 889 MODULE_ALIAS_RTNL_LINK("ipvlan");