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 unsigned int ipvlan_netid __read_mostly;
  13
  14 struct ipvlan_netns {
  15         unsigned int ipvl_nf_hook_refcnt;
  16 };
  17
  18 static const struct nf_hook_ops ipvl_nfops[] = {
  19         {
  20                 .hook     = ipvlan_nf_input,
  21                 .pf       = NFPROTO_IPV4,
  22                 .hooknum  = NF_INET_LOCAL_IN,
  23                 .priority = INT_MAX,
  24         },
  25         {
  26                 .hook     = ipvlan_nf_input,
  27                 .pf       = NFPROTO_IPV6,
  28                 .hooknum  = NF_INET_LOCAL_IN,
  29                 .priority = INT_MAX,
  30         },
  31 };
  32
  33 static const struct l3mdev_ops ipvl_l3mdev_ops = {
  34         .l3mdev_l3_rcv = ipvlan_l3_rcv,
  35 };
  36
  37 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
  38 {
  39         ipvlan->dev->mtu = dev->mtu;
  40 }
  41
  42 static int ipvlan_register_nf_hook(struct net *net)
  43 {
  44         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
  45         int err = 0;
  46
  47         if (!vnet->ipvl_nf_hook_refcnt) {
  48                 err = nf_register_net_hooks(net, ipvl_nfops,
  49                                             ARRAY_SIZE(ipvl_nfops));
  50                 if (!err)
  51                         vnet->ipvl_nf_hook_refcnt = 1;
  52         } else {
  53                 vnet->ipvl_nf_hook_refcnt++;
  54         }
  55
  56         return err;
  57 }
  58
  59 static void ipvlan_unregister_nf_hook(struct net *net)
  60 {
  61         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
  62
  63         if (WARN_ON(!vnet->ipvl_nf_hook_refcnt))
  64                 return;
  65
  66         vnet->ipvl_nf_hook_refcnt--;
  67         if (!vnet->ipvl_nf_hook_refcnt)
  68                 nf_unregister_net_hooks(net, ipvl_nfops,
  69                                         ARRAY_SIZE(ipvl_nfops));
  70 }
  71
  72 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval)
  73 {
  74         struct ipvl_dev *ipvlan;
  75         struct net_device *mdev = port->dev;
  76         unsigned int flags;
  77         int err;
  78
  79         ASSERT_RTNL();
  80         if (port->mode != nval) {
  81                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
  82                         flags = ipvlan->dev->flags;
  83                         if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
  84                                 err = dev_change_flags(ipvlan->dev,
  85                                                        flags | IFF_NOARP);
  86                         } else {
  87                                 err = dev_change_flags(ipvlan->dev,
  88                                                        flags & ~IFF_NOARP);
  89                         }
  90                         if (unlikely(err))
  91                                 goto fail;
  92                 }
  93                 if (nval == IPVLAN_MODE_L3S) {
  94                         /* New mode is L3S */
  95                         err = ipvlan_register_nf_hook(read_pnet(&port->pnet));
  96                         if (!err) {
  97                                 mdev->l3mdev_ops = &ipvl_l3mdev_ops;
  98                                 mdev->priv_flags |= IFF_L3MDEV_RX_HANDLER;
  99                         } else
 100                                 goto fail;
 101                 } else if (port->mode == IPVLAN_MODE_L3S) {
 102                         /* Old mode was L3S */
 103                         mdev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
 104                         ipvlan_unregister_nf_hook(read_pnet(&port->pnet));
 105                         mdev->l3mdev_ops = NULL;
 106                 }
 107                 port->mode = nval;
 108         }
 109         return 0;
 110
 111 fail:
 112         /* Undo the flags changes that have been done so far. */
 113         list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
 114                 flags = ipvlan->dev->flags;
 115                 if (port->mode == IPVLAN_MODE_L3 ||
 116                     port->mode == IPVLAN_MODE_L3S)
 117                         dev_change_flags(ipvlan->dev, flags | IFF_NOARP);
 118                 else
 119                         dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP);
 120         }
 121
 122         return err;
 123 }
 124
 125 static int ipvlan_port_create(struct net_device *dev)
 126 {
 127         struct ipvl_port *port;
 128         int err, idx;
 129
 130         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) {
 131                 netdev_err(dev, "Master is either lo or non-ether device\n");
 132                 return -EINVAL;
 133         }
 134
 135         if (netdev_is_rx_handler_busy(dev)) {
 136                 netdev_err(dev, "Device is already in use.\n");
 137                 return -EBUSY;
 138         }
 139
 140         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
 141         if (!port)
 142                 return -ENOMEM;
 143
 144         write_pnet(&port->pnet, dev_net(dev));
 145         port->dev = dev;
 146         port->mode = IPVLAN_MODE_L3;
 147         INIT_LIST_HEAD(&port->ipvlans);
 148         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
 149                 INIT_HLIST_HEAD(&port->hlhead[idx]);
 150
 151         skb_queue_head_init(&port->backlog);
 152         INIT_WORK(&port->wq, ipvlan_process_multicast);
 153         ida_init(&port->ida);
 154         port->dev_id_start = 1;
 155
 156         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
 157         if (err)
 158                 goto err;
 159
 160         dev->priv_flags |= IFF_IPVLAN_MASTER;
 161         return 0;
 162
 163 err:
 164         kfree(port);
 165         return err;
 166 }
 167
 168 static void ipvlan_port_destroy(struct net_device *dev)
 169 {
 170         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
 171         struct sk_buff *skb;
 172
 173         dev->priv_flags &= ~IFF_IPVLAN_MASTER;
 174         if (port->mode == IPVLAN_MODE_L3S) {
 175                 dev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
 176                 ipvlan_unregister_nf_hook(dev_net(dev));
 177                 dev->l3mdev_ops = NULL;
 178         }
 179         netdev_rx_handler_unregister(dev);
 180         cancel_work_sync(&port->wq);
 181         while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
 182                 if (skb->dev)
 183                         dev_put(skb->dev);
 184                 kfree_skb(skb);
 185         }
 186         ida_destroy(&port->ida);
 187         kfree(port);
 188 }
 189
 190 #define IPVLAN_ALWAYS_ON_OFLOADS \
 191         (NETIF_F_SG | NETIF_F_HW_CSUM | \
 192          NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
 193
 194 #define IPVLAN_ALWAYS_ON \
 195         (IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED)
 196
 197 #define IPVLAN_FEATURES \
 198         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 199          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
 200          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 201          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 202
 203         /* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
 204
 205 #define IPVLAN_STATE_MASK \
 206         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 207
 208 static int ipvlan_init(struct net_device *dev)
 209 {
 210         struct ipvl_dev *ipvlan = netdev_priv(dev);
 211         const struct net_device *phy_dev = ipvlan->phy_dev;
 212         struct ipvl_port *port = ipvlan->port;
 213
 214         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 215                      (phy_dev->state & IPVLAN_STATE_MASK);
 216         dev->features = phy_dev->features & IPVLAN_FEATURES;
 217         dev->features |= IPVLAN_ALWAYS_ON;
 218         dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
 219         dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
 220         dev->gso_max_size = phy_dev->gso_max_size;
 221         dev->gso_max_segs = phy_dev->gso_max_segs;
 222         dev->hard_header_len = phy_dev->hard_header_len;
 223
 224         netdev_lockdep_set_classes(dev);
 225
 226         ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
 227         if (!ipvlan->pcpu_stats)
 228                 return -ENOMEM;
 229
 230         port->count += 1;
 231
 232         return 0;
 233 }
 234
 235 static void ipvlan_uninit(struct net_device *dev)
 236 {
 237         struct ipvl_dev *ipvlan = netdev_priv(dev);
 238         struct ipvl_port *port = ipvlan->port;
 239
 240         free_percpu(ipvlan->pcpu_stats);
 241
 242         port->count -= 1;
 243         if (!port->count)
 244                 ipvlan_port_destroy(port->dev);
 245 }
 246
 247 static int ipvlan_open(struct net_device *dev)
 248 {
 249         struct ipvl_dev *ipvlan = netdev_priv(dev);
 250         struct ipvl_addr *addr;
 251
 252         if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
 253             ipvlan->port->mode == IPVLAN_MODE_L3S)
 254                 dev->flags |= IFF_NOARP;
 255         else
 256                 dev->flags &= ~IFF_NOARP;
 257
 258         list_for_each_entry(addr, &ipvlan->addrs, anode)
 259                 ipvlan_ht_addr_add(ipvlan, addr);
 260
 261         return 0;
 262 }
 263
 264 static int ipvlan_stop(struct net_device *dev)
 265 {
 266         struct ipvl_dev *ipvlan = netdev_priv(dev);
 267         struct net_device *phy_dev = ipvlan->phy_dev;
 268         struct ipvl_addr *addr;
 269
 270         dev_uc_unsync(phy_dev, dev);
 271         dev_mc_unsync(phy_dev, dev);
 272
 273         list_for_each_entry(addr, &ipvlan->addrs, anode)
 274                 ipvlan_ht_addr_del(addr);
 275
 276         return 0;
 277 }
 278
 279 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 280                                      struct net_device *dev)
 281 {
 282         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 283         int skblen = skb->len;
 284         int ret;
 285
 286         ret = ipvlan_queue_xmit(skb, dev);
 287         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 288                 struct ipvl_pcpu_stats *pcptr;
 289
 290                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 291
 292                 u64_stats_update_begin(&pcptr->syncp);
 293                 pcptr->tx_pkts++;
 294                 pcptr->tx_bytes += skblen;
 295                 u64_stats_update_end(&pcptr->syncp);
 296         } else {
 297                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 298         }
 299         return ret;
 300 }
 301
 302 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 303                                              netdev_features_t features)
 304 {
 305         struct ipvl_dev *ipvlan = netdev_priv(dev);
 306
 307         features |= NETIF_F_ALL_FOR_ALL;
 308         features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 309         features = netdev_increment_features(ipvlan->phy_dev->features,
 310                                              features, features);
 311         features |= IPVLAN_ALWAYS_ON;
 312         features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
 313
 314         return features;
 315 }
 316
 317 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 318 {
 319         struct ipvl_dev *ipvlan = netdev_priv(dev);
 320         struct net_device *phy_dev = ipvlan->phy_dev;
 321
 322         if (change & IFF_ALLMULTI)
 323                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 324 }
 325
 326 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 327 {
 328         struct ipvl_dev *ipvlan = netdev_priv(dev);
 329
 330         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 331                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 332         } else {
 333                 struct netdev_hw_addr *ha;
 334                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 335
 336                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 337                 netdev_for_each_mc_addr(ha, dev)
 338                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 339
 340                 /* Turn-on broadcast bit irrespective of address family,
 341                  * since broadcast is deferred to a work-queue, hence no
 342                  * impact on fast-path processing.
 343                  */
 344                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 345
 346                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 347                             IPVLAN_MAC_FILTER_SIZE);
 348         }
 349         dev_uc_sync(ipvlan->phy_dev, dev);
 350         dev_mc_sync(ipvlan->phy_dev, dev);
 351 }
 352
 353 static void ipvlan_get_stats64(struct net_device *dev,
 354                                struct rtnl_link_stats64 *s)
 355 {
 356         struct ipvl_dev *ipvlan = netdev_priv(dev);
 357
 358         if (ipvlan->pcpu_stats) {
 359                 struct ipvl_pcpu_stats *pcptr;
 360                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 361                 u32 rx_errs = 0, tx_drps = 0;
 362                 u32 strt;
 363                 int idx;
 364
 365                 for_each_possible_cpu(idx) {
 366                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 367                         do {
 368                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 369                                 rx_pkts = pcptr->rx_pkts;
 370                                 rx_bytes = pcptr->rx_bytes;
 371                                 rx_mcast = pcptr->rx_mcast;
 372                                 tx_pkts = pcptr->tx_pkts;
 373                                 tx_bytes = pcptr->tx_bytes;
 374                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 375                                                            strt));
 376
 377                         s->rx_packets += rx_pkts;
 378                         s->rx_bytes += rx_bytes;
 379                         s->multicast += rx_mcast;
 380                         s->tx_packets += tx_pkts;
 381                         s->tx_bytes += tx_bytes;
 382
 383                         /* u32 values are updated without syncp protection. */
 384                         rx_errs += pcptr->rx_errs;
 385                         tx_drps += pcptr->tx_drps;
 386                 }
 387                 s->rx_errors = rx_errs;
 388                 s->rx_dropped = rx_errs;
 389                 s->tx_dropped = tx_drps;
 390         }
 391 }
 392
 393 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 394 {
 395         struct ipvl_dev *ipvlan = netdev_priv(dev);
 396         struct net_device *phy_dev = ipvlan->phy_dev;
 397
 398         return vlan_vid_add(phy_dev, proto, vid);
 399 }
 400
 401 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 402                                    u16 vid)
 403 {
 404         struct ipvl_dev *ipvlan = netdev_priv(dev);
 405         struct net_device *phy_dev = ipvlan->phy_dev;
 406
 407         vlan_vid_del(phy_dev, proto, vid);
 408         return 0;
 409 }
 410
 411 static int ipvlan_get_iflink(const struct net_device *dev)
 412 {
 413         struct ipvl_dev *ipvlan = netdev_priv(dev);
 414
 415         return ipvlan->phy_dev->ifindex;
 416 }
 417
 418 static const struct net_device_ops ipvlan_netdev_ops = {
 419         .ndo_init               = ipvlan_init,
 420         .ndo_uninit             = ipvlan_uninit,
 421         .ndo_open               = ipvlan_open,
 422         .ndo_stop               = ipvlan_stop,
 423         .ndo_start_xmit         = ipvlan_start_xmit,
 424         .ndo_fix_features       = ipvlan_fix_features,
 425         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 426         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 427         .ndo_get_stats64        = ipvlan_get_stats64,
 428         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 429         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 430         .ndo_get_iflink         = ipvlan_get_iflink,
 431 };
 432
 433 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 434                               unsigned short type, const void *daddr,
 435                               const void *saddr, unsigned len)
 436 {
 437         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 438         struct net_device *phy_dev = ipvlan->phy_dev;
 439
 440         /* TODO Probably use a different field than dev_addr so that the
 441          * mac-address on the virtual device is portable and can be carried
 442          * while the packets use the mac-addr on the physical device.
 443          */
 444         return dev_hard_header(skb, phy_dev, type, daddr,
 445                                saddr ? : dev->dev_addr, len);
 446 }
 447
 448 static const struct header_ops ipvlan_header_ops = {
 449         .create         = ipvlan_hard_header,
 450         .parse          = eth_header_parse,
 451         .cache          = eth_header_cache,
 452         .cache_update   = eth_header_cache_update,
 453 };
 454
 455 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 456                                              struct ethtool_link_ksettings *cmd)
 457 {
 458         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 459
 460         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 461 }
 462
 463 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 464                                        struct ethtool_drvinfo *drvinfo)
 465 {
 466         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 467         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 468 }
 469
 470 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 471 {
 472         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 473
 474         return ipvlan->msg_enable;
 475 }
 476
 477 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 478 {
 479         struct ipvl_dev *ipvlan = netdev_priv(dev);
 480
 481         ipvlan->msg_enable = value;
 482 }
 483
 484 static const struct ethtool_ops ipvlan_ethtool_ops = {
 485         .get_link       = ethtool_op_get_link,
 486         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 487         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 488         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 489         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 490 };
 491
 492 static int ipvlan_nl_changelink(struct net_device *dev,
 493                                 struct nlattr *tb[], struct nlattr *data[],
 494                                 struct netlink_ext_ack *extack)
 495 {
 496         struct ipvl_dev *ipvlan = netdev_priv(dev);
 497         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 498         int err = 0;
 499
 500         if (!data)
 501                 return 0;
 502         if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
 503                 return -EPERM;
 504
 505         if (data[IFLA_IPVLAN_MODE]) {
 506                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 507
 508                 err = ipvlan_set_port_mode(port, nmode);
 509         }
 510         return err;
 511 }
 512
 513 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 514 {
 515         return (0
 516                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 517                 );
 518 }
 519
 520 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
 521                               struct netlink_ext_ack *extack)
 522 {
 523         if (data && data[IFLA_IPVLAN_MODE]) {
 524                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 525
 526                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 527                         return -EINVAL;
 528         }
 529         return 0;
 530 }
 531
 532 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 533                               const struct net_device *dev)
 534 {
 535         struct ipvl_dev *ipvlan = netdev_priv(dev);
 536         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 537         int ret = -EINVAL;
 538
 539         if (!port)
 540                 goto err;
 541
 542         ret = -EMSGSIZE;
 543         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 544                 goto err;
 545
 546         return 0;
 547
 548 err:
 549         return ret;
 550 }
 551
 552 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 553                     struct nlattr *tb[], struct nlattr *data[],
 554                     struct netlink_ext_ack *extack)
 555 {
 556         struct ipvl_dev *ipvlan = netdev_priv(dev);
 557         struct ipvl_port *port;
 558         struct net_device *phy_dev;
 559         int err;
 560         u16 mode = IPVLAN_MODE_L3;
 561         bool create = false;
 562
 563         if (!tb[IFLA_LINK])
 564                 return -EINVAL;
 565
 566         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 567         if (!phy_dev)
 568                 return -ENODEV;
 569
 570         if (netif_is_ipvlan(phy_dev)) {
 571                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 572
 573                 phy_dev = tmp->phy_dev;
 574                 if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
 575                         return -EPERM;
 576         } else if (!netif_is_ipvlan_port(phy_dev)) {
 577                 err = ipvlan_port_create(phy_dev);
 578                 if (err < 0)
 579                         return err;
 580                 create = true;
 581         }
 582
 583         if (data && data[IFLA_IPVLAN_MODE])
 584                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 585
 586         port = ipvlan_port_get_rtnl(phy_dev);
 587         ipvlan->phy_dev = phy_dev;
 588         ipvlan->dev = dev;
 589         ipvlan->port = port;
 590         ipvlan->sfeatures = IPVLAN_FEATURES;
 591         if (!tb[IFLA_MTU])
 592                 ipvlan_adjust_mtu(ipvlan, phy_dev);
 593         INIT_LIST_HEAD(&ipvlan->addrs);
 594
 595         /* If the port-id base is at the MAX value, then wrap it around and
 596          * begin from 0x1 again. This may be due to a busy system where lots
 597          * of slaves are getting created and deleted.
 598          */
 599         if (port->dev_id_start == 0xFFFE)
 600                 port->dev_id_start = 0x1;
 601
 602         /* Since L2 address is shared among all IPvlan slaves including
 603          * master, use unique 16 bit dev-ids to diffentiate among them.
 604          * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
 605          * slave link [see addrconf_ifid_eui48()].
 606          */
 607         err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
 608                              GFP_KERNEL);
 609         if (err < 0)
 610                 err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
 611                                      GFP_KERNEL);
 612         if (err < 0)
 613                 goto destroy_ipvlan_port;
 614         dev->dev_id = err;
 615         /* Increment id-base to the next slot for the future assignment */
 616         port->dev_id_start = err + 1;
 617
 618         /* TODO Probably put random address here to be presented to the
 619          * world but keep using the physical-dev address for the outgoing
 620          * packets.
 621          */
 622         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 623
 624         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 625
 626         err = register_netdevice(dev);
 627         if (err < 0)
 628                 goto remove_ida;
 629
 630         err = netdev_upper_dev_link(phy_dev, dev);
 631         if (err) {
 632                 goto unregister_netdev;
 633         }
 634         err = ipvlan_set_port_mode(port, mode);
 635         if (err) {
 636                 goto unlink_netdev;
 637         }
 638
 639         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 640         netif_stacked_transfer_operstate(phy_dev, dev);
 641         return 0;
 642
 643 unlink_netdev:
 644         netdev_upper_dev_unlink(phy_dev, dev);
 645 unregister_netdev:
 646         unregister_netdevice(dev);
 647 remove_ida:
 648         ida_simple_remove(&port->ida, dev->dev_id);
 649 destroy_ipvlan_port:
 650         if (create)
 651                 ipvlan_port_destroy(phy_dev);
 652         return err;
 653 }
 654 EXPORT_SYMBOL_GPL(ipvlan_link_new);
 655
 656 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 657 {
 658         struct ipvl_dev *ipvlan = netdev_priv(dev);
 659         struct ipvl_addr *addr, *next;
 660
 661         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 662                 ipvlan_ht_addr_del(addr);
 663                 list_del(&addr->anode);
 664                 kfree_rcu(addr, rcu);
 665         }
 666
 667         ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
 668         list_del_rcu(&ipvlan->pnode);
 669         unregister_netdevice_queue(dev, head);
 670         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 671 }
 672 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
 673
 674 void ipvlan_link_setup(struct net_device *dev)
 675 {
 676         ether_setup(dev);
 677
 678         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 679         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 680         dev->netdev_ops = &ipvlan_netdev_ops;
 681         dev->needs_free_netdev = true;
 682         dev->header_ops = &ipvlan_header_ops;
 683         dev->ethtool_ops = &ipvlan_ethtool_ops;
 684 }
 685 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
 686
 687 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 688 {
 689         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 690 };
 691
 692 static struct rtnl_link_ops ipvlan_link_ops = {
 693         .kind           = "ipvlan",
 694         .priv_size      = sizeof(struct ipvl_dev),
 695
 696         .setup          = ipvlan_link_setup,
 697         .newlink        = ipvlan_link_new,
 698         .dellink        = ipvlan_link_delete,
 699 };
 700
 701 int ipvlan_link_register(struct rtnl_link_ops *ops)
 702 {
 703         ops->get_size   = ipvlan_nl_getsize;
 704         ops->policy     = ipvlan_nl_policy;
 705         ops->validate   = ipvlan_nl_validate;
 706         ops->fill_info  = ipvlan_nl_fillinfo;
 707         ops->changelink = ipvlan_nl_changelink;
 708         ops->maxtype    = IFLA_IPVLAN_MAX;
 709         return rtnl_link_register(ops);
 710 }
 711 EXPORT_SYMBOL_GPL(ipvlan_link_register);
 712
 713 static int ipvlan_device_event(struct notifier_block *unused,
 714                                unsigned long event, void *ptr)
 715 {
 716         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 717         struct ipvl_dev *ipvlan, *next;
 718         struct ipvl_port *port;
 719         LIST_HEAD(lst_kill);
 720
 721         if (!netif_is_ipvlan_port(dev))
 722                 return NOTIFY_DONE;
 723
 724         port = ipvlan_port_get_rtnl(dev);
 725
 726         switch (event) {
 727         case NETDEV_CHANGE:
 728                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 729                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 730                                                          ipvlan->dev);
 731                 break;
 732
 733         case NETDEV_REGISTER: {
 734                 struct net *oldnet, *newnet = dev_net(dev);
 735                 struct ipvlan_netns *old_vnet;
 736
 737                 oldnet = read_pnet(&port->pnet);
 738                 if (net_eq(newnet, oldnet))
 739                         break;
 740
 741                 write_pnet(&port->pnet, newnet);
 742
 743                 old_vnet = net_generic(oldnet, ipvlan_netid);
 744                 if (!old_vnet->ipvl_nf_hook_refcnt)
 745                         break;
 746
 747                 ipvlan_register_nf_hook(newnet);
 748                 ipvlan_unregister_nf_hook(oldnet);
 749                 break;
 750         }
 751         case NETDEV_UNREGISTER:
 752                 if (dev->reg_state != NETREG_UNREGISTERING)
 753                         break;
 754
 755                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 756                                          pnode)
 757                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 758                                                             &lst_kill);
 759                 unregister_netdevice_many(&lst_kill);
 760                 break;
 761
 762         case NETDEV_FEAT_CHANGE:
 763                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 764                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 765                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 766                         netdev_update_features(ipvlan->dev);
 767                 }
 768                 break;
 769
 770         case NETDEV_CHANGEMTU:
 771                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 772                         ipvlan_adjust_mtu(ipvlan, dev);
 773                 break;
 774
 775         case NETDEV_PRE_TYPE_CHANGE:
 776                 /* Forbid underlying device to change its type. */
 777                 return NOTIFY_BAD;
 778         }
 779         return NOTIFY_DONE;
 780 }
 781
 782 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 783 {
 784         struct ipvl_addr *addr;
 785
 786         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 787         if (!addr)
 788                 return -ENOMEM;
 789
 790         addr->master = ipvlan;
 791         if (is_v6) {
 792                 memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
 793                 addr->atype = IPVL_IPV6;
 794         } else {
 795                 memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
 796                 addr->atype = IPVL_IPV4;
 797         }
 798         list_add_tail(&addr->anode, &ipvlan->addrs);
 799
 800         /* If the interface is not up, the address will be added to the hash
 801          * list by ipvlan_open.
 802          */
 803         if (netif_running(ipvlan->dev))
 804                 ipvlan_ht_addr_add(ipvlan, addr);
 805
 806         return 0;
 807 }
 808
 809 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 810 {
 811         struct ipvl_addr *addr;
 812
 813         addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
 814         if (!addr)
 815                 return;
 816
 817         ipvlan_ht_addr_del(addr);
 818         list_del(&addr->anode);
 819         kfree_rcu(addr, rcu);
 820
 821         return;
 822 }
 823
 824 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 825 {
 826         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 827                 netif_err(ipvlan, ifup, ipvlan->dev,
 828                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 829                           ip6_addr, ipvlan->dev->name);
 830                 return -EINVAL;
 831         }
 832
 833         return ipvlan_add_addr(ipvlan, ip6_addr, true);
 834 }
 835
 836 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 837 {
 838         return ipvlan_del_addr(ipvlan, ip6_addr, true);
 839 }
 840
 841 static int ipvlan_addr6_event(struct notifier_block *unused,
 842                               unsigned long event, void *ptr)
 843 {
 844         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 845         struct net_device *dev = (struct net_device *)if6->idev->dev;
 846         struct ipvl_dev *ipvlan = netdev_priv(dev);
 847
 848         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 849         if (in_softirq())
 850                 return NOTIFY_DONE;
 851
 852         if (!netif_is_ipvlan(dev))
 853                 return NOTIFY_DONE;
 854
 855         if (!ipvlan || !ipvlan->port)
 856                 return NOTIFY_DONE;
 857
 858         switch (event) {
 859         case NETDEV_UP:
 860                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 861                         return NOTIFY_BAD;
 862                 break;
 863
 864         case NETDEV_DOWN:
 865                 ipvlan_del_addr6(ipvlan, &if6->addr);
 866                 break;
 867         }
 868
 869         return NOTIFY_OK;
 870 }
 871
 872 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
 873                                         unsigned long event, void *ptr)
 874 {
 875         struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
 876         struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
 877         struct ipvl_dev *ipvlan = netdev_priv(dev);
 878
 879         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 880         if (in_softirq())
 881                 return NOTIFY_DONE;
 882
 883         if (!netif_is_ipvlan(dev))
 884                 return NOTIFY_DONE;
 885
 886         if (!ipvlan || !ipvlan->port)
 887                 return NOTIFY_DONE;
 888
 889         switch (event) {
 890         case NETDEV_UP:
 891                 if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true))
 892                         return notifier_from_errno(-EADDRINUSE);
 893                 break;
 894         }
 895
 896         return NOTIFY_OK;
 897 }
 898
 899 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 900 {
 901         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 902                 netif_err(ipvlan, ifup, ipvlan->dev,
 903                           "Failed to add IPv4=%pI4 on %s intf.\n",
 904                           ip4_addr, ipvlan->dev->name);
 905                 return -EINVAL;
 906         }
 907
 908         return ipvlan_add_addr(ipvlan, ip4_addr, false);
 909 }
 910
 911 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 912 {
 913         return ipvlan_del_addr(ipvlan, ip4_addr, false);
 914 }
 915
 916 static int ipvlan_addr4_event(struct notifier_block *unused,
 917                               unsigned long event, void *ptr)
 918 {
 919         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 920         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 921         struct ipvl_dev *ipvlan = netdev_priv(dev);
 922         struct in_addr ip4_addr;
 923
 924         if (!netif_is_ipvlan(dev))
 925                 return NOTIFY_DONE;
 926
 927         if (!ipvlan || !ipvlan->port)
 928                 return NOTIFY_DONE;
 929
 930         switch (event) {
 931         case NETDEV_UP:
 932                 ip4_addr.s_addr = if4->ifa_address;
 933                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 934                         return NOTIFY_BAD;
 935                 break;
 936
 937         case NETDEV_DOWN:
 938                 ip4_addr.s_addr = if4->ifa_address;
 939                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 940                 break;
 941         }
 942
 943         return NOTIFY_OK;
 944 }
 945
 946 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
 947                                         unsigned long event, void *ptr)
 948 {
 949         struct in_validator_info *ivi = (struct in_validator_info *)ptr;
 950         struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
 951         struct ipvl_dev *ipvlan = netdev_priv(dev);
 952
 953         if (!netif_is_ipvlan(dev))
 954                 return NOTIFY_DONE;
 955
 956         if (!ipvlan || !ipvlan->port)
 957                 return NOTIFY_DONE;
 958
 959         switch (event) {
 960         case NETDEV_UP:
 961                 if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false))
 962                         return notifier_from_errno(-EADDRINUSE);
 963                 break;
 964         }
 965
 966         return NOTIFY_OK;
 967 }
 968
 969 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 970         .notifier_call = ipvlan_addr4_event,
 971 };
 972
 973 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
 974         .notifier_call = ipvlan_addr4_validator_event,
 975 };
 976
 977 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 978         .notifier_call = ipvlan_device_event,
 979 };
 980
 981 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 982         .notifier_call = ipvlan_addr6_event,
 983 };
 984
 985 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
 986         .notifier_call = ipvlan_addr6_validator_event,
 987 };
 988
 989 static void ipvlan_ns_exit(struct net *net)
 990 {
 991         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
 992
 993         if (WARN_ON_ONCE(vnet->ipvl_nf_hook_refcnt)) {
 994                 vnet->ipvl_nf_hook_refcnt = 0;
 995                 nf_unregister_net_hooks(net, ipvl_nfops,
 996                                         ARRAY_SIZE(ipvl_nfops));
 997         }
 998 }
 999
1000 static struct pernet_operations ipvlan_net_ops = {
1001         .id = &ipvlan_netid,
1002         .size = sizeof(struct ipvlan_netns),
1003         .exit = ipvlan_ns_exit,
1004 };
1005
1006 static int __init ipvlan_init_module(void)
1007 {
1008         int err;
1009
1010         ipvlan_init_secret();
1011         register_netdevice_notifier(&ipvlan_notifier_block);
1012         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1013         register_inet6addr_validator_notifier(
1014             &ipvlan_addr6_vtor_notifier_block);
1015         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1016         register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1017
1018         err = register_pernet_subsys(&ipvlan_net_ops);
1019         if (err < 0)
1020                 goto error;
1021
1022         err = ipvlan_link_register(&ipvlan_link_ops);
1023         if (err < 0) {
1024                 unregister_pernet_subsys(&ipvlan_net_ops);
1025                 goto error;
1026         }
1027
1028         return 0;
1029 error:
1030         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1031         unregister_inetaddr_validator_notifier(
1032             &ipvlan_addr4_vtor_notifier_block);
1033         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1034         unregister_inet6addr_validator_notifier(
1035             &ipvlan_addr6_vtor_notifier_block);
1036         unregister_netdevice_notifier(&ipvlan_notifier_block);
1037         return err;
1038 }
1039
1040 static void __exit ipvlan_cleanup_module(void)
1041 {
1042         rtnl_link_unregister(&ipvlan_link_ops);
1043         unregister_pernet_subsys(&ipvlan_net_ops);
1044         unregister_netdevice_notifier(&ipvlan_notifier_block);
1045         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1046         unregister_inetaddr_validator_notifier(
1047             &ipvlan_addr4_vtor_notifier_block);
1048         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1049         unregister_inet6addr_validator_notifier(
1050             &ipvlan_addr6_vtor_notifier_block);
1051 }
1052
1053 module_init(ipvlan_init_module);
1054 module_exit(ipvlan_cleanup_module);
1055
1056 MODULE_LICENSE("GPL");
1057 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1058 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1059 MODULE_ALIAS_RTNL_LINK("ipvlan");