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         int err = 0;
  77
  78         ASSERT_RTNL();
  79         if (port->mode != nval) {
  80                 if (nval == IPVLAN_MODE_L3S) {
  81                         /* New mode is L3S */
  82                         err = ipvlan_register_nf_hook(read_pnet(&port->pnet));
  83                         if (!err) {
  84                                 mdev->l3mdev_ops = &ipvl_l3mdev_ops;
  85                                 mdev->priv_flags |= IFF_L3MDEV_MASTER;
  86                         } else
  87                                 return err;
  88                 } else if (port->mode == IPVLAN_MODE_L3S) {
  89                         /* Old mode was L3S */
  90                         mdev->priv_flags &= ~IFF_L3MDEV_MASTER;
  91                         ipvlan_unregister_nf_hook(read_pnet(&port->pnet));
  92                         mdev->l3mdev_ops = NULL;
  93                 }
  94                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
  95                         if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S)
  96                                 ipvlan->dev->flags |= IFF_NOARP;
  97                         else
  98                                 ipvlan->dev->flags &= ~IFF_NOARP;
  99                 }
 100                 port->mode = nval;
 101         }
 102         return err;
 103 }
 104
 105 static int ipvlan_port_create(struct net_device *dev)
 106 {
 107         struct ipvl_port *port;
 108         int err, idx;
 109
 110         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
 111         if (!port)
 112                 return -ENOMEM;
 113
 114         write_pnet(&port->pnet, dev_net(dev));
 115         port->dev = dev;
 116         port->mode = IPVLAN_MODE_L3;
 117         INIT_LIST_HEAD(&port->ipvlans);
 118         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
 119                 INIT_HLIST_HEAD(&port->hlhead[idx]);
 120
 121         skb_queue_head_init(&port->backlog);
 122         INIT_WORK(&port->wq, ipvlan_process_multicast);
 123         ida_init(&port->ida);
 124         port->dev_id_start = 1;
 125
 126         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
 127         if (err)
 128                 goto err;
 129
 130         dev->priv_flags |= IFF_IPVLAN_MASTER;
 131         return 0;
 132
 133 err:
 134         kfree(port);
 135         return err;
 136 }
 137
 138 static void ipvlan_port_destroy(struct net_device *dev)
 139 {
 140         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
 141         struct sk_buff *skb;
 142
 143         dev->priv_flags &= ~IFF_IPVLAN_MASTER;
 144         if (port->mode == IPVLAN_MODE_L3S) {
 145                 dev->priv_flags &= ~IFF_L3MDEV_MASTER;
 146                 ipvlan_unregister_nf_hook(dev_net(dev));
 147                 dev->l3mdev_ops = NULL;
 148         }
 149         netdev_rx_handler_unregister(dev);
 150         cancel_work_sync(&port->wq);
 151         while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
 152                 if (skb->dev)
 153                         dev_put(skb->dev);
 154                 kfree_skb(skb);
 155         }
 156         ida_destroy(&port->ida);
 157         kfree(port);
 158 }
 159
 160 #define IPVLAN_FEATURES \
 161         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 162          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
 163          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 164          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 165
 166 #define IPVLAN_STATE_MASK \
 167         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 168
 169 static int ipvlan_init(struct net_device *dev)
 170 {
 171         struct ipvl_dev *ipvlan = netdev_priv(dev);
 172         struct net_device *phy_dev = ipvlan->phy_dev;
 173         struct ipvl_port *port;
 174         int err;
 175
 176         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 177                      (phy_dev->state & IPVLAN_STATE_MASK);
 178         dev->features = phy_dev->features & IPVLAN_FEATURES;
 179         dev->features |= NETIF_F_LLTX;
 180         dev->gso_max_size = phy_dev->gso_max_size;
 181         dev->gso_max_segs = phy_dev->gso_max_segs;
 182         dev->hard_header_len = phy_dev->hard_header_len;
 183
 184         netdev_lockdep_set_classes(dev);
 185
 186         ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
 187         if (!ipvlan->pcpu_stats)
 188                 return -ENOMEM;
 189
 190         if (!netif_is_ipvlan_port(phy_dev)) {
 191                 err = ipvlan_port_create(phy_dev);
 192                 if (err < 0) {
 193                         free_percpu(ipvlan->pcpu_stats);
 194                         return err;
 195                 }
 196         }
 197         port = ipvlan_port_get_rtnl(phy_dev);
 198         port->count += 1;
 199         return 0;
 200 }
 201
 202 static void ipvlan_uninit(struct net_device *dev)
 203 {
 204         struct ipvl_dev *ipvlan = netdev_priv(dev);
 205         struct net_device *phy_dev = ipvlan->phy_dev;
 206         struct ipvl_port *port;
 207
 208         free_percpu(ipvlan->pcpu_stats);
 209
 210         port = ipvlan_port_get_rtnl(phy_dev);
 211         port->count -= 1;
 212         if (!port->count)
 213                 ipvlan_port_destroy(port->dev);
 214 }
 215
 216 static int ipvlan_open(struct net_device *dev)
 217 {
 218         struct ipvl_dev *ipvlan = netdev_priv(dev);
 219         struct net_device *phy_dev = ipvlan->phy_dev;
 220         struct ipvl_addr *addr;
 221
 222         if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
 223             ipvlan->port->mode == IPVLAN_MODE_L3S)
 224                 dev->flags |= IFF_NOARP;
 225         else
 226                 dev->flags &= ~IFF_NOARP;
 227
 228         list_for_each_entry(addr, &ipvlan->addrs, anode)
 229                 ipvlan_ht_addr_add(ipvlan, addr);
 230
 231         return dev_uc_add(phy_dev, phy_dev->dev_addr);
 232 }
 233
 234 static int ipvlan_stop(struct net_device *dev)
 235 {
 236         struct ipvl_dev *ipvlan = netdev_priv(dev);
 237         struct net_device *phy_dev = ipvlan->phy_dev;
 238         struct ipvl_addr *addr;
 239
 240         dev_uc_unsync(phy_dev, dev);
 241         dev_mc_unsync(phy_dev, dev);
 242
 243         dev_uc_del(phy_dev, phy_dev->dev_addr);
 244
 245         list_for_each_entry(addr, &ipvlan->addrs, anode)
 246                 ipvlan_ht_addr_del(addr);
 247
 248         return 0;
 249 }
 250
 251 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 252                                      struct net_device *dev)
 253 {
 254         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 255         int skblen = skb->len;
 256         int ret;
 257
 258         ret = ipvlan_queue_xmit(skb, dev);
 259         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 260                 struct ipvl_pcpu_stats *pcptr;
 261
 262                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 263
 264                 u64_stats_update_begin(&pcptr->syncp);
 265                 pcptr->tx_pkts++;
 266                 pcptr->tx_bytes += skblen;
 267                 u64_stats_update_end(&pcptr->syncp);
 268         } else {
 269                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 270         }
 271         return ret;
 272 }
 273
 274 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 275                                              netdev_features_t features)
 276 {
 277         struct ipvl_dev *ipvlan = netdev_priv(dev);
 278
 279         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 280 }
 281
 282 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 283 {
 284         struct ipvl_dev *ipvlan = netdev_priv(dev);
 285         struct net_device *phy_dev = ipvlan->phy_dev;
 286
 287         if (change & IFF_ALLMULTI)
 288                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 289 }
 290
 291 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 292 {
 293         struct ipvl_dev *ipvlan = netdev_priv(dev);
 294
 295         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 296                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 297         } else {
 298                 struct netdev_hw_addr *ha;
 299                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 300
 301                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 302                 netdev_for_each_mc_addr(ha, dev)
 303                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 304
 305                 /* Turn-on broadcast bit irrespective of address family,
 306                  * since broadcast is deferred to a work-queue, hence no
 307                  * impact on fast-path processing.
 308                  */
 309                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 310
 311                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 312                             IPVLAN_MAC_FILTER_SIZE);
 313         }
 314         dev_uc_sync(ipvlan->phy_dev, dev);
 315         dev_mc_sync(ipvlan->phy_dev, dev);
 316 }
 317
 318 static void ipvlan_get_stats64(struct net_device *dev,
 319                                struct rtnl_link_stats64 *s)
 320 {
 321         struct ipvl_dev *ipvlan = netdev_priv(dev);
 322
 323         if (ipvlan->pcpu_stats) {
 324                 struct ipvl_pcpu_stats *pcptr;
 325                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 326                 u32 rx_errs = 0, tx_drps = 0;
 327                 u32 strt;
 328                 int idx;
 329
 330                 for_each_possible_cpu(idx) {
 331                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 332                         do {
 333                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 334                                 rx_pkts = pcptr->rx_pkts;
 335                                 rx_bytes = pcptr->rx_bytes;
 336                                 rx_mcast = pcptr->rx_mcast;
 337                                 tx_pkts = pcptr->tx_pkts;
 338                                 tx_bytes = pcptr->tx_bytes;
 339                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 340                                                            strt));
 341
 342                         s->rx_packets += rx_pkts;
 343                         s->rx_bytes += rx_bytes;
 344                         s->multicast += rx_mcast;
 345                         s->tx_packets += tx_pkts;
 346                         s->tx_bytes += tx_bytes;
 347
 348                         /* u32 values are updated without syncp protection. */
 349                         rx_errs += pcptr->rx_errs;
 350                         tx_drps += pcptr->tx_drps;
 351                 }
 352                 s->rx_errors = rx_errs;
 353                 s->rx_dropped = rx_errs;
 354                 s->tx_dropped = tx_drps;
 355         }
 356 }
 357
 358 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 359 {
 360         struct ipvl_dev *ipvlan = netdev_priv(dev);
 361         struct net_device *phy_dev = ipvlan->phy_dev;
 362
 363         return vlan_vid_add(phy_dev, proto, vid);
 364 }
 365
 366 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 367                                    u16 vid)
 368 {
 369         struct ipvl_dev *ipvlan = netdev_priv(dev);
 370         struct net_device *phy_dev = ipvlan->phy_dev;
 371
 372         vlan_vid_del(phy_dev, proto, vid);
 373         return 0;
 374 }
 375
 376 static int ipvlan_get_iflink(const struct net_device *dev)
 377 {
 378         struct ipvl_dev *ipvlan = netdev_priv(dev);
 379
 380         return ipvlan->phy_dev->ifindex;
 381 }
 382
 383 static const struct net_device_ops ipvlan_netdev_ops = {
 384         .ndo_init               = ipvlan_init,
 385         .ndo_uninit             = ipvlan_uninit,
 386         .ndo_open               = ipvlan_open,
 387         .ndo_stop               = ipvlan_stop,
 388         .ndo_start_xmit         = ipvlan_start_xmit,
 389         .ndo_fix_features       = ipvlan_fix_features,
 390         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 391         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 392         .ndo_get_stats64        = ipvlan_get_stats64,
 393         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 394         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 395         .ndo_get_iflink         = ipvlan_get_iflink,
 396 };
 397
 398 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 399                               unsigned short type, const void *daddr,
 400                               const void *saddr, unsigned len)
 401 {
 402         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 403         struct net_device *phy_dev = ipvlan->phy_dev;
 404
 405         /* TODO Probably use a different field than dev_addr so that the
 406          * mac-address on the virtual device is portable and can be carried
 407          * while the packets use the mac-addr on the physical device.
 408          */
 409         return dev_hard_header(skb, phy_dev, type, daddr,
 410                                saddr ? : phy_dev->dev_addr, len);
 411 }
 412
 413 static const struct header_ops ipvlan_header_ops = {
 414         .create         = ipvlan_hard_header,
 415         .parse          = eth_header_parse,
 416         .cache          = eth_header_cache,
 417         .cache_update   = eth_header_cache_update,
 418 };
 419
 420 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 421                                              struct ethtool_link_ksettings *cmd)
 422 {
 423         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 424
 425         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 426 }
 427
 428 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 429                                        struct ethtool_drvinfo *drvinfo)
 430 {
 431         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 432         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 433 }
 434
 435 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 436 {
 437         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 438
 439         return ipvlan->msg_enable;
 440 }
 441
 442 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 443 {
 444         struct ipvl_dev *ipvlan = netdev_priv(dev);
 445
 446         ipvlan->msg_enable = value;
 447 }
 448
 449 static const struct ethtool_ops ipvlan_ethtool_ops = {
 450         .get_link       = ethtool_op_get_link,
 451         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 452         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 453         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 454         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 455 };
 456
 457 static int ipvlan_nl_changelink(struct net_device *dev,
 458                                 struct nlattr *tb[], struct nlattr *data[],
 459                                 struct netlink_ext_ack *extack)
 460 {
 461         struct ipvl_dev *ipvlan = netdev_priv(dev);
 462         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 463         int err = 0;
 464
 465         if (!data)
 466                 return 0;
 467
 468         if (data[IFLA_IPVLAN_MODE]) {
 469                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 470
 471                 err = ipvlan_set_port_mode(port, nmode);
 472         }
 473
 474         if (!err && data[IFLA_IPVLAN_FLAGS]) {
 475                 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 476
 477                 if (flags & IPVLAN_F_PRIVATE)
 478                         ipvlan_mark_private(port);
 479                 else
 480                         ipvlan_clear_private(port);
 481
 482                 if (flags & IPVLAN_F_VEPA)
 483                         ipvlan_mark_vepa(port);
 484                 else
 485                         ipvlan_clear_vepa(port);
 486         }
 487
 488         return err;
 489 }
 490
 491 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 492 {
 493         return (0
 494                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 495                 + nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
 496                 );
 497 }
 498
 499 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
 500                               struct netlink_ext_ack *extack)
 501 {
 502         if (!data)
 503                 return 0;
 504
 505         if (data[IFLA_IPVLAN_MODE]) {
 506                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 507
 508                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 509                         return -EINVAL;
 510         }
 511         if (data[IFLA_IPVLAN_FLAGS]) {
 512                 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 513
 514                 /* Only two bits are used at this moment. */
 515                 if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
 516                         return -EINVAL;
 517                 /* Also both flags can't be active at the same time. */
 518                 if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
 519                     (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
 520                         return -EINVAL;
 521         }
 522
 523         return 0;
 524 }
 525
 526 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 527                               const struct net_device *dev)
 528 {
 529         struct ipvl_dev *ipvlan = netdev_priv(dev);
 530         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 531         int ret = -EINVAL;
 532
 533         if (!port)
 534                 goto err;
 535
 536         ret = -EMSGSIZE;
 537         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 538                 goto err;
 539         if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
 540                 goto err;
 541
 542         return 0;
 543
 544 err:
 545         return ret;
 546 }
 547
 548 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 549                     struct nlattr *tb[], struct nlattr *data[],
 550                     struct netlink_ext_ack *extack)
 551 {
 552         struct ipvl_dev *ipvlan = netdev_priv(dev);
 553         struct ipvl_port *port;
 554         struct net_device *phy_dev;
 555         int err;
 556         u16 mode = IPVLAN_MODE_L3;
 557
 558         if (!tb[IFLA_LINK])
 559                 return -EINVAL;
 560
 561         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 562         if (!phy_dev)
 563                 return -ENODEV;
 564
 565         if (netif_is_ipvlan(phy_dev)) {
 566                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 567
 568                 phy_dev = tmp->phy_dev;
 569         } else if (!netif_is_ipvlan_port(phy_dev)) {
 570                 /* Exit early if the underlying link is invalid or busy */
 571                 if (phy_dev->type != ARPHRD_ETHER ||
 572                     phy_dev->flags & IFF_LOOPBACK) {
 573                         netdev_err(phy_dev,
 574                                    "Master is either lo or non-ether device\n");
 575                         return -EINVAL;
 576                 }
 577
 578                 if (netdev_is_rx_handler_busy(phy_dev)) {
 579                         netdev_err(phy_dev, "Device is already in use.\n");
 580                         return -EBUSY;
 581                 }
 582         }
 583
 584         ipvlan->phy_dev = phy_dev;
 585         ipvlan->dev = dev;
 586         ipvlan->sfeatures = IPVLAN_FEATURES;
 587         ipvlan_adjust_mtu(ipvlan, phy_dev);
 588         INIT_LIST_HEAD(&ipvlan->addrs);
 589
 590         /* TODO Probably put random address here to be presented to the
 591          * world but keep using the physical-dev address for the outgoing
 592          * packets.
 593          */
 594         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 595
 596         dev->priv_flags |= IFF_IPVLAN_SLAVE;
 597
 598         err = register_netdevice(dev);
 599         if (err < 0)
 600                 return err;
 601
 602         /* ipvlan_init() would have created the port, if required */
 603         port = ipvlan_port_get_rtnl(phy_dev);
 604         ipvlan->port = port;
 605
 606         /* If the port-id base is at the MAX value, then wrap it around and
 607          * begin from 0x1 again. This may be due to a busy system where lots
 608          * of slaves are getting created and deleted.
 609          */
 610         if (port->dev_id_start == 0xFFFE)
 611                 port->dev_id_start = 0x1;
 612
 613         /* Since L2 address is shared among all IPvlan slaves including
 614          * master, use unique 16 bit dev-ids to diffentiate among them.
 615          * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
 616          * slave link [see addrconf_ifid_eui48()].
 617          */
 618         err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
 619                              GFP_KERNEL);
 620         if (err < 0)
 621                 err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
 622                                      GFP_KERNEL);
 623         if (err < 0)
 624                 goto unregister_netdev;
 625         dev->dev_id = err;
 626
 627         /* Increment id-base to the next slot for the future assignment */
 628         port->dev_id_start = err + 1;
 629
 630         err = netdev_upper_dev_link(phy_dev, dev, extack);
 631         if (err)
 632                 goto remove_ida;
 633
 634         /* Flags are per port and latest update overrides. User has
 635          * to be consistent in setting it just like the mode attribute.
 636          */
 637         if (data && data[IFLA_IPVLAN_FLAGS])
 638                 port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 639
 640         if (data && data[IFLA_IPVLAN_MODE])
 641                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 642
 643         err = ipvlan_set_port_mode(port, mode);
 644         if (err)
 645                 goto unlink_netdev;
 646
 647         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 648         netif_stacked_transfer_operstate(phy_dev, dev);
 649         return 0;
 650
 651 unlink_netdev:
 652         netdev_upper_dev_unlink(phy_dev, dev);
 653 remove_ida:
 654         ida_simple_remove(&port->ida, dev->dev_id);
 655 unregister_netdev:
 656         unregister_netdevice(dev);
 657         return err;
 658 }
 659 EXPORT_SYMBOL_GPL(ipvlan_link_new);
 660
 661 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 662 {
 663         struct ipvl_dev *ipvlan = netdev_priv(dev);
 664         struct ipvl_addr *addr, *next;
 665
 666         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 667                 ipvlan_ht_addr_del(addr);
 668                 list_del(&addr->anode);
 669                 kfree_rcu(addr, rcu);
 670         }
 671
 672         ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
 673         list_del_rcu(&ipvlan->pnode);
 674         unregister_netdevice_queue(dev, head);
 675         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 676 }
 677 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
 678
 679 void ipvlan_link_setup(struct net_device *dev)
 680 {
 681         ether_setup(dev);
 682
 683         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 684         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 685         dev->netdev_ops = &ipvlan_netdev_ops;
 686         dev->needs_free_netdev = true;
 687         dev->header_ops = &ipvlan_header_ops;
 688         dev->ethtool_ops = &ipvlan_ethtool_ops;
 689 }
 690 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
 691
 692 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 693 {
 694         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 695         [IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
 696 };
 697
 698 static struct rtnl_link_ops ipvlan_link_ops = {
 699         .kind           = "ipvlan",
 700         .priv_size      = sizeof(struct ipvl_dev),
 701
 702         .setup          = ipvlan_link_setup,
 703         .newlink        = ipvlan_link_new,
 704         .dellink        = ipvlan_link_delete,
 705 };
 706
 707 int ipvlan_link_register(struct rtnl_link_ops *ops)
 708 {
 709         ops->get_size   = ipvlan_nl_getsize;
 710         ops->policy     = ipvlan_nl_policy;
 711         ops->validate   = ipvlan_nl_validate;
 712         ops->fill_info  = ipvlan_nl_fillinfo;
 713         ops->changelink = ipvlan_nl_changelink;
 714         ops->maxtype    = IFLA_IPVLAN_MAX;
 715         return rtnl_link_register(ops);
 716 }
 717 EXPORT_SYMBOL_GPL(ipvlan_link_register);
 718
 719 static int ipvlan_device_event(struct notifier_block *unused,
 720                                unsigned long event, void *ptr)
 721 {
 722         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 723         struct ipvl_dev *ipvlan, *next;
 724         struct ipvl_port *port;
 725         LIST_HEAD(lst_kill);
 726
 727         if (!netif_is_ipvlan_port(dev))
 728                 return NOTIFY_DONE;
 729
 730         port = ipvlan_port_get_rtnl(dev);
 731
 732         switch (event) {
 733         case NETDEV_CHANGE:
 734                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 735                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 736                                                          ipvlan->dev);
 737                 break;
 738
 739         case NETDEV_REGISTER: {
 740                 struct net *oldnet, *newnet = dev_net(dev);
 741                 struct ipvlan_netns *old_vnet;
 742
 743                 oldnet = read_pnet(&port->pnet);
 744                 if (net_eq(newnet, oldnet))
 745                         break;
 746
 747                 write_pnet(&port->pnet, newnet);
 748
 749                 old_vnet = net_generic(oldnet, ipvlan_netid);
 750                 if (!old_vnet->ipvl_nf_hook_refcnt)
 751                         break;
 752
 753                 ipvlan_register_nf_hook(newnet);
 754                 ipvlan_unregister_nf_hook(oldnet);
 755                 break;
 756         }
 757         case NETDEV_UNREGISTER:
 758                 if (dev->reg_state != NETREG_UNREGISTERING)
 759                         break;
 760
 761                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
 762                                          pnode)
 763                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 764                                                             &lst_kill);
 765                 unregister_netdevice_many(&lst_kill);
 766                 break;
 767
 768         case NETDEV_FEAT_CHANGE:
 769                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 770                         ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
 771                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 772                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 773                         netdev_features_change(ipvlan->dev);
 774                 }
 775                 break;
 776
 777         case NETDEV_CHANGEMTU:
 778                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 779                         ipvlan_adjust_mtu(ipvlan, dev);
 780                 break;
 781
 782         case NETDEV_CHANGEADDR:
 783                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 784                         ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
 785                 break;
 786
 787         case NETDEV_PRE_TYPE_CHANGE:
 788                 /* Forbid underlying device to change its type. */
 789                 return NOTIFY_BAD;
 790         }
 791         return NOTIFY_DONE;
 792 }
 793
 794 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 795 {
 796         struct ipvl_addr *addr;
 797
 798         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 799         if (!addr)
 800                 return -ENOMEM;
 801
 802         addr->master = ipvlan;
 803         if (is_v6) {
 804                 memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
 805                 addr->atype = IPVL_IPV6;
 806         } else {
 807                 memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
 808                 addr->atype = IPVL_IPV4;
 809         }
 810         list_add_tail(&addr->anode, &ipvlan->addrs);
 811
 812         /* If the interface is not up, the address will be added to the hash
 813          * list by ipvlan_open.
 814          */
 815         if (netif_running(ipvlan->dev))
 816                 ipvlan_ht_addr_add(ipvlan, addr);
 817
 818         return 0;
 819 }
 820
 821 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 822 {
 823         struct ipvl_addr *addr;
 824
 825         addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
 826         if (!addr)
 827                 return;
 828
 829         ipvlan_ht_addr_del(addr);
 830         list_del(&addr->anode);
 831         kfree_rcu(addr, rcu);
 832
 833         return;
 834 }
 835
 836 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 837 {
 838         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
 839                 netif_err(ipvlan, ifup, ipvlan->dev,
 840                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 841                           ip6_addr, ipvlan->dev->name);
 842                 return -EINVAL;
 843         }
 844
 845         return ipvlan_add_addr(ipvlan, ip6_addr, true);
 846 }
 847
 848 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 849 {
 850         return ipvlan_del_addr(ipvlan, ip6_addr, true);
 851 }
 852
 853 static int ipvlan_addr6_event(struct notifier_block *unused,
 854                               unsigned long event, void *ptr)
 855 {
 856         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 857         struct net_device *dev = (struct net_device *)if6->idev->dev;
 858         struct ipvl_dev *ipvlan = netdev_priv(dev);
 859
 860         if (!netif_is_ipvlan(dev))
 861                 return NOTIFY_DONE;
 862
 863         if (!ipvlan || !ipvlan->port)
 864                 return NOTIFY_DONE;
 865
 866         switch (event) {
 867         case NETDEV_UP:
 868                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 869                         return NOTIFY_BAD;
 870                 break;
 871
 872         case NETDEV_DOWN:
 873                 ipvlan_del_addr6(ipvlan, &if6->addr);
 874                 break;
 875         }
 876
 877         return NOTIFY_OK;
 878 }
 879
 880 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
 881                                         unsigned long event, void *ptr)
 882 {
 883         struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
 884         struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
 885         struct ipvl_dev *ipvlan = netdev_priv(dev);
 886
 887         /* FIXME IPv6 autoconf calls us from bh without RTNL */
 888         if (in_softirq())
 889                 return NOTIFY_DONE;
 890
 891         if (!netif_is_ipvlan(dev))
 892                 return NOTIFY_DONE;
 893
 894         if (!ipvlan || !ipvlan->port)
 895                 return NOTIFY_DONE;
 896
 897         switch (event) {
 898         case NETDEV_UP:
 899                 if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
 900                         NL_SET_ERR_MSG(i6vi->extack,
 901                                        "Address already assigned to an ipvlan device");
 902                         return notifier_from_errno(-EADDRINUSE);
 903                 }
 904                 break;
 905         }
 906
 907         return NOTIFY_OK;
 908 }
 909
 910 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 911 {
 912         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
 913                 netif_err(ipvlan, ifup, ipvlan->dev,
 914                           "Failed to add IPv4=%pI4 on %s intf.\n",
 915                           ip4_addr, ipvlan->dev->name);
 916                 return -EINVAL;
 917         }
 918
 919         return ipvlan_add_addr(ipvlan, ip4_addr, false);
 920 }
 921
 922 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 923 {
 924         return ipvlan_del_addr(ipvlan, ip4_addr, false);
 925 }
 926
 927 static int ipvlan_addr4_event(struct notifier_block *unused,
 928                               unsigned long event, void *ptr)
 929 {
 930         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
 931         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
 932         struct ipvl_dev *ipvlan = netdev_priv(dev);
 933         struct in_addr ip4_addr;
 934
 935         if (!netif_is_ipvlan(dev))
 936                 return NOTIFY_DONE;
 937
 938         if (!ipvlan || !ipvlan->port)
 939                 return NOTIFY_DONE;
 940
 941         switch (event) {
 942         case NETDEV_UP:
 943                 ip4_addr.s_addr = if4->ifa_address;
 944                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
 945                         return NOTIFY_BAD;
 946                 break;
 947
 948         case NETDEV_DOWN:
 949                 ip4_addr.s_addr = if4->ifa_address;
 950                 ipvlan_del_addr4(ipvlan, &ip4_addr);
 951                 break;
 952         }
 953
 954         return NOTIFY_OK;
 955 }
 956
 957 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
 958                                         unsigned long event, void *ptr)
 959 {
 960         struct in_validator_info *ivi = (struct in_validator_info *)ptr;
 961         struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
 962         struct ipvl_dev *ipvlan = netdev_priv(dev);
 963
 964         if (!netif_is_ipvlan(dev))
 965                 return NOTIFY_DONE;
 966
 967         if (!ipvlan || !ipvlan->port)
 968                 return NOTIFY_DONE;
 969
 970         switch (event) {
 971         case NETDEV_UP:
 972                 if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
 973                         NL_SET_ERR_MSG(ivi->extack,
 974                                        "Address already assigned to an ipvlan device");
 975                         return notifier_from_errno(-EADDRINUSE);
 976                 }
 977                 break;
 978         }
 979
 980         return NOTIFY_OK;
 981 }
 982
 983 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
 984         .notifier_call = ipvlan_addr4_event,
 985 };
 986
 987 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
 988         .notifier_call = ipvlan_addr4_validator_event,
 989 };
 990
 991 static struct notifier_block ipvlan_notifier_block __read_mostly = {
 992         .notifier_call = ipvlan_device_event,
 993 };
 994
 995 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
 996         .notifier_call = ipvlan_addr6_event,
 997 };
 998
 999 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1000         .notifier_call = ipvlan_addr6_validator_event,
1001 };
1002
1003 static void ipvlan_ns_exit(struct net *net)
1004 {
1005         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
1006
1007         if (WARN_ON_ONCE(vnet->ipvl_nf_hook_refcnt)) {
1008                 vnet->ipvl_nf_hook_refcnt = 0;
1009                 nf_unregister_net_hooks(net, ipvl_nfops,
1010                                         ARRAY_SIZE(ipvl_nfops));
1011         }
1012 }
1013
1014 static struct pernet_operations ipvlan_net_ops = {
1015         .id = &ipvlan_netid,
1016         .size = sizeof(struct ipvlan_netns),
1017         .exit = ipvlan_ns_exit,
1018 };
1019
1020 static int __init ipvlan_init_module(void)
1021 {
1022         int err;
1023
1024         ipvlan_init_secret();
1025         register_netdevice_notifier(&ipvlan_notifier_block);
1026         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1027         register_inet6addr_validator_notifier(
1028             &ipvlan_addr6_vtor_notifier_block);
1029         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1030         register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1031
1032         err = register_pernet_subsys(&ipvlan_net_ops);
1033         if (err < 0)
1034                 goto error;
1035
1036         err = ipvlan_link_register(&ipvlan_link_ops);
1037         if (err < 0) {
1038                 unregister_pernet_subsys(&ipvlan_net_ops);
1039                 goto error;
1040         }
1041
1042         return 0;
1043 error:
1044         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1045         unregister_inetaddr_validator_notifier(
1046             &ipvlan_addr4_vtor_notifier_block);
1047         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1048         unregister_inet6addr_validator_notifier(
1049             &ipvlan_addr6_vtor_notifier_block);
1050         unregister_netdevice_notifier(&ipvlan_notifier_block);
1051         return err;
1052 }
1053
1054 static void __exit ipvlan_cleanup_module(void)
1055 {
1056         rtnl_link_unregister(&ipvlan_link_ops);
1057         unregister_pernet_subsys(&ipvlan_net_ops);
1058         unregister_netdevice_notifier(&ipvlan_notifier_block);
1059         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1060         unregister_inetaddr_validator_notifier(
1061             &ipvlan_addr4_vtor_notifier_block);
1062         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1063         unregister_inet6addr_validator_notifier(
1064             &ipvlan_addr6_vtor_notifier_block);
1065 }
1066
1067 module_init(ipvlan_init_module);
1068 module_exit(ipvlan_cleanup_module);
1069
1070 MODULE_LICENSE("GPL");
1071 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1072 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1073 MODULE_ALIAS_RTNL_LINK("ipvlan");