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         unsigned int flags;
  67         int err;
  68
  69         ASSERT_RTNL();
  70         if (port->mode != nval) {
  71                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
  72                         flags = ipvlan->dev->flags;
  73                         if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
  74                                 err = dev_change_flags(ipvlan->dev,
  75                                                        flags | IFF_NOARP);
  76                         } else {
  77                                 err = dev_change_flags(ipvlan->dev,
  78                                                        flags & ~IFF_NOARP);
  79                         }
  80                         if (unlikely(err))
  81                                 goto fail;
  82                 }
  83                 if (nval == IPVLAN_MODE_L3S) {
  84                         /* New mode is L3S */
  85                         err = ipvlan_register_nf_hook();
  86                         if (!err) {
  87                                 mdev->l3mdev_ops = &ipvl_l3mdev_ops;
  88                                 mdev->priv_flags |= IFF_L3MDEV_RX_HANDLER;
  89                         } else
  90                                 goto fail;
  91                 } else if (port->mode == IPVLAN_MODE_L3S) {
  92                         /* Old mode was L3S */
  93                         mdev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
  94                         ipvlan_unregister_nf_hook();
  95                         mdev->l3mdev_ops = NULL;
  96                 }
  97                 port->mode = nval;
  98         }
  99         return 0;
 100
 101 fail:
 102         /* Undo the flags changes that have been done so far. */
 103         list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
 104                 flags = ipvlan->dev->flags;
 105                 if (port->mode == IPVLAN_MODE_L3 ||
 106                     port->mode == IPVLAN_MODE_L3S)
 107                         dev_change_flags(ipvlan->dev, flags | IFF_NOARP);
 108                 else
 109                         dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP);
 110         }
 111
 112         return err;
 113 }
 114
 115 static int ipvlan_port_create(struct net_device *dev)
 116 {
 117         struct ipvl_port *port;
 118         int err, idx;
 119
 120         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) {
 121                 netdev_err(dev, "Master is either lo or non-ether device\n");
 122                 return -EINVAL;
 123         }
 124
 125         if (netif_is_macvlan_port(dev)) {
 126                 netdev_err(dev, "Master is a macvlan port.\n");
 127                 return -EBUSY;
 128         }
 129
 130         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
 131         if (!port)
 132                 return -ENOMEM;
 133
 134         port->dev = dev;
 135         port->mode = IPVLAN_MODE_L3;
 136         INIT_LIST_HEAD(&port->ipvlans);
 137         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
 138                 INIT_HLIST_HEAD(&port->hlhead[idx]);
 139
 140         skb_queue_head_init(&port->backlog);
 141         INIT_WORK(&port->wq, ipvlan_process_multicast);
 142
 143         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
 144         if (err)
 145                 goto err;
 146
 147         dev->priv_flags |= IFF_IPVLAN_MASTER;
 148         return 0;
 149
 150 err:
 151         kfree_rcu(port, rcu);
 152         return err;
 153 }
 154
 155 static void ipvlan_port_destroy(struct net_device *dev)
 156 {
 157         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
 158
 159         dev->priv_flags &= ~IFF_IPVLAN_MASTER;
 160         if (port->mode == IPVLAN_MODE_L3S) {
 161                 dev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
 162                 ipvlan_unregister_nf_hook();
 163                 dev->l3mdev_ops = NULL;
 164         }
 165         netdev_rx_handler_unregister(dev);
 166         cancel_work_sync(&port->wq);
 167         __skb_queue_purge(&port->backlog);
 168         kfree_rcu(port, rcu);
 169 }
 170
 171 #define IPVLAN_ALWAYS_ON_OFLOADS \
 172         (NETIF_F_SG | NETIF_F_HW_CSUM | \
 173          NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
 174
 175 #define IPVLAN_ALWAYS_ON \
 176         (IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED)
 177
 178 #define IPVLAN_FEATURES \
 179         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 180          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 181          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 182          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 183
 184         /* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
 185
 186 #define IPVLAN_STATE_MASK \
 187         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 188
 189 static int ipvlan_init(struct net_device *dev)
 190 {
 191         struct ipvl_dev *ipvlan = netdev_priv(dev);
 192         const struct net_device *phy_dev = ipvlan->phy_dev;
 193         struct ipvl_port *port = ipvlan->port;
 194
 195         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 196                      (phy_dev->state & IPVLAN_STATE_MASK);
 197         dev->features = phy_dev->features & IPVLAN_FEATURES;
 198         dev->features |= IPVLAN_ALWAYS_ON;
 199         dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
 200         dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
 201         dev->gso_max_size = phy_dev->gso_max_size;
 202         dev->gso_max_segs = phy_dev->gso_max_segs;
 203         dev->hard_header_len = phy_dev->hard_header_len;
 204
 205         netdev_lockdep_set_classes(dev);
 206
 207         ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
 208         if (!ipvlan->pcpu_stats)
 209                 return -ENOMEM;
 210
 211         port->count += 1;
 212
 213         return 0;
 214 }
 215
 216 static void ipvlan_uninit(struct net_device *dev)
 217 {
 218         struct ipvl_dev *ipvlan = netdev_priv(dev);
 219         struct ipvl_port *port = ipvlan->port;
 220
 221         free_percpu(ipvlan->pcpu_stats);
 222
 223         port->count -= 1;
 224         if (!port->count)
 225                 ipvlan_port_destroy(port->dev);
 226 }
 227
 228 static int ipvlan_open(struct net_device *dev)
 229 {
 230         struct ipvl_dev *ipvlan = netdev_priv(dev);
 231         struct ipvl_addr *addr;
 232
 233         if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
 234             ipvlan->port->mode == IPVLAN_MODE_L3S)
 235                 dev->flags |= IFF_NOARP;
 236         else
 237                 dev->flags &= ~IFF_NOARP;
 238
 239         list_for_each_entry(addr, &ipvlan->addrs, anode)
 240                 ipvlan_ht_addr_add(ipvlan, addr);
 241
 242         return 0;
 243 }
 244
 245 static int ipvlan_stop(struct net_device *dev)
 246 {
 247         struct ipvl_dev *ipvlan = netdev_priv(dev);
 248         struct net_device *phy_dev = ipvlan->phy_dev;
 249         struct ipvl_addr *addr;
 250
 251         dev_uc_unsync(phy_dev, dev);
 252         dev_mc_unsync(phy_dev, dev);
 253
 254         list_for_each_entry(addr, &ipvlan->addrs, anode)
 255                 ipvlan_ht_addr_del(addr);
 256
 257         return 0;
 258 }
 259
 260 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 261                                      struct net_device *dev)
 262 {
 263         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 264         int skblen = skb->len;
 265         int ret;
 266
 267         ret = ipvlan_queue_xmit(skb, dev);
 268         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 269                 struct ipvl_pcpu_stats *pcptr;
 270
 271                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 272
 273                 u64_stats_update_begin(&pcptr->syncp);
 274                 pcptr->tx_pkts++;
 275                 pcptr->tx_bytes += skblen;
 276                 u64_stats_update_end(&pcptr->syncp);
 277         } else {
 278                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 279         }
 280         return ret;
 281 }
 282
 283 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 284                                              netdev_features_t features)
 285 {
 286         struct ipvl_dev *ipvlan = netdev_priv(dev);
 287
 288         features |= NETIF_F_ALL_FOR_ALL;
 289         features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 290         features = netdev_increment_features(ipvlan->phy_dev->features,
 291                                              features, features);
 292         features |= IPVLAN_ALWAYS_ON;
 293         features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
 294
 295         return features;
 296 }
 297
 298 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 299 {
 300         struct ipvl_dev *ipvlan = netdev_priv(dev);
 301         struct net_device *phy_dev = ipvlan->phy_dev;
 302
 303         if (change & IFF_ALLMULTI)
 304                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 305 }
 306
 307 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 308 {
 309         struct ipvl_dev *ipvlan = netdev_priv(dev);
 310
 311         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 312                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 313         } else {
 314                 struct netdev_hw_addr *ha;
 315                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 316
 317                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 318                 netdev_for_each_mc_addr(ha, dev)
 319                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 320
 321                 /* Turn-on broadcast bit irrespective of address family,
 322                  * since broadcast is deferred to a work-queue, hence no
 323                  * impact on fast-path processing.
 324                  */
 325                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 326
 327                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 328                             IPVLAN_MAC_FILTER_SIZE);
 329         }
 330         dev_uc_sync(ipvlan->phy_dev, dev);
 331         dev_mc_sync(ipvlan->phy_dev, dev);
 332 }
 333
 334 static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
 335                                                     struct rtnl_link_stats64 *s)
 336 {
 337         struct ipvl_dev *ipvlan = netdev_priv(dev);
 338
 339         if (ipvlan->pcpu_stats) {
 340                 struct ipvl_pcpu_stats *pcptr;
 341                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 342                 u32 rx_errs = 0, tx_drps = 0;
 343                 u32 strt;
 344                 int idx;
 345
 346                 for_each_possible_cpu(idx) {
 347                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 348                         do {
 349                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 350                                 rx_pkts = pcptr->rx_pkts;
 351                                 rx_bytes = pcptr->rx_bytes;
 352                                 rx_mcast = pcptr->rx_mcast;
 353                                 tx_pkts = pcptr->tx_pkts;
 354                                 tx_bytes = pcptr->tx_bytes;
 355                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 356                                                            strt));
 357
 358                         s->rx_packets += rx_pkts;
 359                         s->rx_bytes += rx_bytes;
 360                         s->multicast += rx_mcast;
 361                         s->tx_packets += tx_pkts;
 362                         s->tx_bytes += tx_bytes;
 363
 364                         /* u32 values are updated without syncp protection. */
 365                         rx_errs += pcptr->rx_errs;
 366                         tx_drps += pcptr->tx_drps;
 367                 }
 368                 s->rx_errors = rx_errs;
 369                 s->rx_dropped = rx_errs;
 370                 s->tx_dropped = tx_drps;
 371         }
 372         return s;
 373 }
 374
 375 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 376 {
 377         struct ipvl_dev *ipvlan = netdev_priv(dev);
 378         struct net_device *phy_dev = ipvlan->phy_dev;
 379
 380         return vlan_vid_add(phy_dev, proto, vid);
 381 }
 382
 383 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 384                                    u16 vid)
 385 {
 386         struct ipvl_dev *ipvlan = netdev_priv(dev);
 387         struct net_device *phy_dev = ipvlan->phy_dev;
 388
 389         vlan_vid_del(phy_dev, proto, vid);
 390         return 0;
 391 }
 392
 393 static int ipvlan_get_iflink(const struct net_device *dev)
 394 {
 395         struct ipvl_dev *ipvlan = netdev_priv(dev);
 396
 397         return ipvlan->phy_dev->ifindex;
 398 }
 399
 400 static const struct net_device_ops ipvlan_netdev_ops = {
 401         .ndo_init               = ipvlan_init,
 402         .ndo_uninit             = ipvlan_uninit,
 403         .ndo_open               = ipvlan_open,
 404         .ndo_stop               = ipvlan_stop,
 405         .ndo_start_xmit         = ipvlan_start_xmit,
 406         .ndo_fix_features       = ipvlan_fix_features,
 407         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 408         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 409         .ndo_get_stats64        = ipvlan_get_stats64,
 410         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 411         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 412         .ndo_get_iflink         = ipvlan_get_iflink,
 413 };
 414
 415 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 416                               unsigned short type, const void *daddr,
 417                               const void *saddr, unsigned len)
 418 {
 419         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 420         struct net_device *phy_dev = ipvlan->phy_dev;
 421
 422         /* TODO Probably use a different field than dev_addr so that the
 423          * mac-address on the virtual device is portable and can be carried
 424          * while the packets use the mac-addr on the physical device.
 425          */
 426         return dev_hard_header(skb, phy_dev, type, daddr,
 427                                saddr ? : dev->dev_addr, len);
 428 }
 429
 430 static const struct header_ops ipvlan_header_ops = {
 431         .create         = ipvlan_hard_header,
 432         .parse          = eth_header_parse,
 433         .cache          = eth_header_cache,
 434         .cache_update   = eth_header_cache_update,
 435 };
 436
 437 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 438                                              struct ethtool_link_ksettings *cmd)
 439 {
 440         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 441
 442         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 443 }
 444
 445 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 446                                        struct ethtool_drvinfo *drvinfo)
 447 {
 448         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 449         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 450 }
 451
 452 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 453 {
 454         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 455
 456         return ipvlan->msg_enable;
 457 }
 458
 459 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 460 {
 461         struct ipvl_dev *ipvlan = netdev_priv(dev);
 462
 463         ipvlan->msg_enable = value;
 464 }
 465
 466 static const struct ethtool_ops ipvlan_ethtool_ops = {
 467         .get_link       = ethtool_op_get_link,
 468         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 469         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 470         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 471         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 472 };
 473
 474 static int ipvlan_nl_changelink(struct net_device *dev,
 475                                 struct nlattr *tb[], struct nlattr *data[])
 476 {
 477         struct ipvl_dev *ipvlan = netdev_priv(dev);
 478         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 479         int err = 0;
 480
 481         if (!data)
 482                 return 0;
 483         if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
 484                 return -EPERM;
 485
 486         if (data[IFLA_IPVLAN_MODE]) {
 487                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 488
 489                 err = ipvlan_set_port_mode(port, nmode);
 490         }
 491         return err;
 492 }
 493
 494 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 495 {
 496         return (0
 497                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 498                 );
 499 }
 500
 501 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
 502 {
 503         if (data && data[IFLA_IPVLAN_MODE]) {
 504                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 505
 506                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 507                         return -EINVAL;
 508         }
 509         return 0;
 510 }
 511
 512 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 513                               const struct net_device *dev)
 514 {
 515         struct ipvl_dev *ipvlan = netdev_priv(dev);
 516         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 517         int ret = -EINVAL;
 518
 519         if (!port)
 520                 goto err;
 521
 522         ret = -EMSGSIZE;
 523         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 524                 goto err;
 525
 526         return 0;
 527
 528 err:
 529         return ret;
 530 }
 531
 532 static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 533                            struct nlattr *tb[], struct nlattr *data[])
 534 {
 535         struct ipvl_dev *ipvlan = netdev_priv(dev);
 536         struct ipvl_port *port;
 537         struct net_device *phy_dev;
 538         int err;
 539         u16 mode = IPVLAN_MODE_L3;
 540         bool create = false;
 541
 542         if (!tb[IFLA_LINK])
 543                 return -EINVAL;
 544
 545         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 546         if (!phy_dev)
 547                 return -ENODEV;
 548
 549         if (netif_is_ipvlan(phy_dev)) {
 550                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 551
 552                 phy_dev = tmp->phy_dev;
 553                 if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
 554                         return -EPERM;
 555         } else if (!netif_is_ipvlan_port(phy_dev)) {
 556                 err = ipvlan_port_create(phy_dev);
 557                 if (err < 0)
 558                         return err;
 559                 create = true;
 560         }
 561
 562         if (data && data[IFLA_IPVLAN_MODE])
 563                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 564
 565         port = ipvlan_port_get_rtnl(phy_dev);
 566         ipvlan->phy_dev = phy_dev;
 567         ipvlan->dev = dev;
 568         ipvlan->port = port;
 569         ipvlan->sfeatures = IPVLAN_FEATURES;
 570         if (!tb[IFLA_MTU])
 571                 ipvlan_adjust_mtu(ipvlan, phy_dev);
 572         INIT_LIST_HEAD(&ipvlan->addrs);
 573
 574         /* TODO Probably put random address here to be presented to the
 575          * world but keep using the physical-dev address for the outgoing
 576          * packets.
 577          */
 578         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 579
 580         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 581
 582         err = register_netdevice(dev);
 583         if (err < 0)
 584                 goto destroy_ipvlan_port;
 585
 586         err = netdev_upper_dev_link(phy_dev, dev);
 587         if (err) {
 588                 goto unregister_netdev;
 589         }
 590         err = ipvlan_set_port_mode(port, mode);
 591         if (err) {
 592                 goto unlink_netdev;
 593         }
 594
 595         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 596         netif_stacked_transfer_operstate(phy_dev, dev);
 597         return 0;
 598
 599 unlink_netdev:
 600         netdev_upper_dev_unlink(phy_dev, dev);
 601 unregister_netdev:
 602         unregister_netdevice(dev);
 603 destroy_ipvlan_port:
 604         if (create)
 605                 ipvlan_port_destroy(phy_dev);
 606         return err;
 607 }
 608
 609 static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 610 {
 611         struct ipvl_dev *ipvlan = netdev_priv(dev);
 612         struct ipvl_addr *addr, *next;
 613
 614         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 615                 ipvlan_ht_addr_del(addr);
 616                 list_del(&addr->anode);
 617                 kfree_rcu(addr, rcu);
 618         }
 619
 620         list_del_rcu(&ipvlan->pnode);
 621         unregister_netdevice_queue(dev, head);
 622         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 623 }
 624
 625 static void ipvlan_link_setup(struct net_device *dev)
 626 {
 627         ether_setup(dev);
 628
 629         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 630         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 631         dev->netdev_ops = &ipvlan_netdev_ops;
 632         dev->destructor = free_netdev;
 633         dev->header_ops = &ipvlan_header_ops;
 634         dev->ethtool_ops = &ipvlan_ethtool_ops;
 635 }
 636
 637 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 638 {
 639         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 640 };
 641
 642 static struct rtnl_link_ops ipvlan_link_ops = {
 643         .kind           = "ipvlan",
 644         .priv_size      = sizeof(struct ipvl_dev),
 645
 646         .get_size       = ipvlan_nl_getsize,
 647         .policy         = ipvlan_nl_policy,
 648         .validate       = ipvlan_nl_validate,
 649         .fill_info      = ipvlan_nl_fillinfo,
 650         .changelink     = ipvlan_nl_changelink,
 651         .maxtype        = IFLA_IPVLAN_MAX,
 652
 653         .setup          = ipvlan_link_setup,
 654         .newlink        = ipvlan_link_new,
 655         .dellink        = ipvlan_link_delete,
 656 };
 657
 658 static int ipvlan_link_register(struct rtnl_link_ops *ops)
 659 {
 660         return rtnl_link_register(ops);
 661 }
 662
 663 static int ipvlan_device_event(struct notifier_block *unused,
 664                                unsigned long event, void *ptr)
 665 {
 666         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 667         struct ipvl_dev *ipvlan, *next;
 668         struct ipvl_port *port;
 669         LIST_HEAD(lst_kill);
 670
 671         if (!netif_is_ipvlan_port(dev))
 672                 return NOTIFY_DONE;
 673
 674         port = ipvlan_port_get_rtnl(dev);
 675
 676         switch (event) {
 677         case NETDEV_CHANGE:
 678                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 679                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 680                                                          ipvlan->dev);
 681                 break;
 682
 683         case NETDEV_UNREGISTER:
 684                 if (dev->reg_state != NETREG_UNREGISTERING)
 685                         break;
 686
 687                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 688                                          pnode)
 689                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 690                                                             &lst_kill);
 691                 unregister_netdevice_many(&lst_kill);
 692                 break;
 693
 694         case NETDEV_FEAT_CHANGE:
 695                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 696                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 697                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 698                         netdev_update_features(ipvlan->dev);
 699                 }
 700                 break;
 701
 702         case NETDEV_CHANGEMTU:
 703                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 704                         ipvlan_adjust_mtu(ipvlan, dev);
 705                 break;
 706
 707         case NETDEV_PRE_TYPE_CHANGE:
 708                 /* Forbid underlying device to change its type. */
 709                 return NOTIFY_BAD;
 710         }
 711         return NOTIFY_DONE;
 712 }
 713
 714 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 715 {
 716         struct ipvl_addr *addr;
 717
 718         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 719                 netif_err(ipvlan, ifup, ipvlan->dev,
 720                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 721                           ip6_addr, ipvlan->dev->name);
 722                 return -EINVAL;
 723         }
 724         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 725         if (!addr)
 726                 return -ENOMEM;
 727
 728         addr->master = ipvlan;
 729         memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
 730         addr->atype = IPVL_IPV6;
 731         list_add_tail(&addr->anode, &ipvlan->addrs);
 732
 733         /* If the interface is not up, the address will be added to the hash
 734          * list by ipvlan_open.
 735          */
 736         if (netif_running(ipvlan->dev))
 737                 ipvlan_ht_addr_add(ipvlan, addr);
 738
 739         return 0;
 740 }
 741
 742 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 743 {
 744         struct ipvl_addr *addr;
 745
 746         addr = ipvlan_find_addr(ipvlan, ip6_addr, true);
 747         if (!addr)
 748                 return;
 749
 750         ipvlan_ht_addr_del(addr);
 751         list_del(&addr->anode);
 752         kfree_rcu(addr, rcu);
 753
 754         return;
 755 }
 756
 757 static int ipvlan_addr6_event(struct notifier_block *unused,
 758                               unsigned long event, void *ptr)
 759 {
 760         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 761         struct net_device *dev = (struct net_device *)if6->idev->dev;
 762         struct ipvl_dev *ipvlan = netdev_priv(dev);
 763
 764         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 765         if (in_softirq())
 766                 return NOTIFY_DONE;
 767
 768         if (!netif_is_ipvlan(dev))
 769                 return NOTIFY_DONE;
 770
 771         if (!ipvlan || !ipvlan->port)
 772                 return NOTIFY_DONE;
 773
 774         switch (event) {
 775         case NETDEV_UP:
 776                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 777                         return NOTIFY_BAD;
 778                 break;
 779
 780         case NETDEV_DOWN:
 781                 ipvlan_del_addr6(ipvlan, &if6->addr);
 782                 break;
 783         }
 784
 785         return NOTIFY_OK;
 786 }
 787
 788 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 789 {
 790         struct ipvl_addr *addr;
 791
 792         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 793                 netif_err(ipvlan, ifup, ipvlan->dev,
 794                           "Failed to add IPv4=%pI4 on %s intf.\n",
 795                           ip4_addr, ipvlan->dev->name);
 796                 return -EINVAL;
 797         }
 798         addr = kzalloc(sizeof(struct ipvl_addr), GFP_KERNEL);
 799         if (!addr)
 800                 return -ENOMEM;
 801
 802         addr->master = ipvlan;
 803         memcpy(&addr->ip4addr, ip4_addr, sizeof(struct in_addr));
 804         addr->atype = IPVL_IPV4;
 805         list_add_tail(&addr->anode, &ipvlan->addrs);
 806
 807         /* If the interface is not up, the address will be added to the hash
 808          * list by ipvlan_open.
 809          */
 810         if (netif_running(ipvlan->dev))
 811                 ipvlan_ht_addr_add(ipvlan, addr);
 812
 813         return 0;
 814 }
 815
 816 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 817 {
 818         struct ipvl_addr *addr;
 819
 820         addr = ipvlan_find_addr(ipvlan, ip4_addr, false);
 821         if (!addr)
 822                 return;
 823
 824         ipvlan_ht_addr_del(addr);
 825         list_del(&addr->anode);
 826         kfree_rcu(addr, rcu);
 827
 828         return;
 829 }
 830
 831 static int ipvlan_addr4_event(struct notifier_block *unused,
 832                               unsigned long event, void *ptr)
 833 {
 834         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 835         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 836         struct ipvl_dev *ipvlan = netdev_priv(dev);
 837         struct in_addr ip4_addr;
 838
 839         if (!netif_is_ipvlan(dev))
 840                 return NOTIFY_DONE;
 841
 842         if (!ipvlan || !ipvlan->port)
 843                 return NOTIFY_DONE;
 844
 845         switch (event) {
 846         case NETDEV_UP:
 847                 ip4_addr.s_addr = if4->ifa_address;
 848                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 849                         return NOTIFY_BAD;
 850                 break;
 851
 852         case NETDEV_DOWN:
 853                 ip4_addr.s_addr = if4->ifa_address;
 854                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 855                 break;
 856         }
 857
 858         return NOTIFY_OK;
 859 }
 860
 861 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 862         .notifier_call = ipvlan_addr4_event,
 863 };
 864
 865 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 866         .notifier_call = ipvlan_device_event,
 867 };
 868
 869 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 870         .notifier_call = ipvlan_addr6_event,
 871 };
 872
 873 static int __init ipvlan_init_module(void)
 874 {
 875         int err;
 876
 877         ipvlan_init_secret();
 878         register_netdevice_notifier(&ipvlan_notifier_block);
 879         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 880         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 881
 882         err = ipvlan_link_register(&ipvlan_link_ops);
 883         if (err < 0)
 884                 goto error;
 885
 886         return 0;
 887 error:
 888         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 889         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 890         unregister_netdevice_notifier(&ipvlan_notifier_block);
 891         return err;
 892 }
 893
 894 static void __exit ipvlan_cleanup_module(void)
 895 {
 896         rtnl_link_unregister(&ipvlan_link_ops);
 897         unregister_netdevice_notifier(&ipvlan_notifier_block);
 898         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
 899         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
 900 }
 901
 902 module_init(ipvlan_init_module);
 903 module_exit(ipvlan_cleanup_module);
 904
 905 MODULE_LICENSE("GPL");
 906 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
 907 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
 908 MODULE_ALIAS_RTNL_LINK("ipvlan");