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 #if IS_ENABLED(CONFIG_IPV6)
  26         {
  27                 .hook     = ipvlan_nf_input,
  28                 .pf       = NFPROTO_IPV6,
  29                 .hooknum  = NF_INET_LOCAL_IN,
  30                 .priority = INT_MAX,
  31         },
  32 #endif
  33 };
  34
  35 static const struct l3mdev_ops ipvl_l3mdev_ops = {
  36         .l3mdev_l3_rcv = ipvlan_l3_rcv,
  37 };
  38
  39 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
  40 {
  41         ipvlan->dev->mtu = dev->mtu;
  42 }
  43
  44 static int ipvlan_register_nf_hook(struct net *net)
  45 {
  46         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
  47         int err = 0;
  48
  49         if (!vnet->ipvl_nf_hook_refcnt) {
  50                 err = nf_register_net_hooks(net, ipvl_nfops,
  51                                             ARRAY_SIZE(ipvl_nfops));
  52                 if (!err)
  53                         vnet->ipvl_nf_hook_refcnt = 1;
  54         } else {
  55                 vnet->ipvl_nf_hook_refcnt++;
  56         }
  57
  58         return err;
  59 }
  60
  61 static void ipvlan_unregister_nf_hook(struct net *net)
  62 {
  63         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
  64
  65         if (WARN_ON(!vnet->ipvl_nf_hook_refcnt))
  66                 return;
  67
  68         vnet->ipvl_nf_hook_refcnt--;
  69         if (!vnet->ipvl_nf_hook_refcnt)
  70                 nf_unregister_net_hooks(net, ipvl_nfops,
  71                                         ARRAY_SIZE(ipvl_nfops));
  72 }
  73
  74 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
  75                                 struct netlink_ext_ack *extack)
  76 {
  77         struct ipvl_dev *ipvlan;
  78         struct net_device *mdev = port->dev;
  79         unsigned int flags;
  80         int err;
  81
  82         ASSERT_RTNL();
  83         if (port->mode != nval) {
  84                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
  85                         flags = ipvlan->dev->flags;
  86                         if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
  87                                 err = dev_change_flags(ipvlan->dev,
  88                                                        flags | IFF_NOARP,
  89                                                        extack);
  90                         } else {
  91                                 err = dev_change_flags(ipvlan->dev,
  92                                                        flags & ~IFF_NOARP,
  93                                                        extack);
  94                         }
  95                         if (unlikely(err))
  96                                 goto fail;
  97                 }
  98                 if (nval == IPVLAN_MODE_L3S) {
  99                         /* New mode is L3S */
 100                         err = ipvlan_register_nf_hook(read_pnet(&port->pnet));
 101                         if (!err) {
 102                                 mdev->l3mdev_ops = &ipvl_l3mdev_ops;
 103                                 mdev->priv_flags |= IFF_L3MDEV_RX_HANDLER;
 104                         } else
 105                                 goto fail;
 106                 } else if (port->mode == IPVLAN_MODE_L3S) {
 107                         /* Old mode was L3S */
 108                         mdev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
 109                         ipvlan_unregister_nf_hook(read_pnet(&port->pnet));
 110                         mdev->l3mdev_ops = NULL;
 111                 }
 112                 port->mode = nval;
 113         }
 114         return 0;
 115
 116 fail:
 117         /* Undo the flags changes that have been done so far. */
 118         list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
 119                 flags = ipvlan->dev->flags;
 120                 if (port->mode == IPVLAN_MODE_L3 ||
 121                     port->mode == IPVLAN_MODE_L3S)
 122                         dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
 123                                          NULL);
 124                 else
 125                         dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
 126                                          NULL);
 127         }
 128
 129         return err;
 130 }
 131
 132 static int ipvlan_port_create(struct net_device *dev)
 133 {
 134         struct ipvl_port *port;
 135         int err, idx;
 136
 137         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
 138         if (!port)
 139                 return -ENOMEM;
 140
 141         write_pnet(&port->pnet, dev_net(dev));
 142         port->dev = dev;
 143         port->mode = IPVLAN_MODE_L3;
 144         INIT_LIST_HEAD(&port->ipvlans);
 145         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
 146                 INIT_HLIST_HEAD(&port->hlhead[idx]);
 147
 148         skb_queue_head_init(&port->backlog);
 149         INIT_WORK(&port->wq, ipvlan_process_multicast);
 150         ida_init(&port->ida);
 151         port->dev_id_start = 1;
 152
 153         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
 154         if (err)
 155                 goto err;
 156
 157         return 0;
 158
 159 err:
 160         kfree(port);
 161         return err;
 162 }
 163
 164 static void ipvlan_port_destroy(struct net_device *dev)
 165 {
 166         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
 167         struct sk_buff *skb;
 168
 169         if (port->mode == IPVLAN_MODE_L3S) {
 170                 dev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
 171                 ipvlan_unregister_nf_hook(dev_net(dev));
 172                 dev->l3mdev_ops = NULL;
 173         }
 174         netdev_rx_handler_unregister(dev);
 175         cancel_work_sync(&port->wq);
 176         while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
 177                 if (skb->dev)
 178                         dev_put(skb->dev);
 179                 kfree_skb(skb);
 180         }
 181         ida_destroy(&port->ida);
 182         kfree(port);
 183 }
 184
 185 #define IPVLAN_FEATURES \
 186         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 187          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
 188          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 189          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 190
 191 #define IPVLAN_STATE_MASK \
 192         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 193
 194 static int ipvlan_init(struct net_device *dev)
 195 {
 196         struct ipvl_dev *ipvlan = netdev_priv(dev);
 197         struct net_device *phy_dev = ipvlan->phy_dev;
 198         struct ipvl_port *port;
 199         int err;
 200
 201         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 202                      (phy_dev->state & IPVLAN_STATE_MASK);
 203         dev->features = phy_dev->features & IPVLAN_FEATURES;
 204         dev->features |= NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED;
 205         dev->gso_max_size = phy_dev->gso_max_size;
 206         dev->gso_max_segs = phy_dev->gso_max_segs;
 207         dev->hard_header_len = phy_dev->hard_header_len;
 208
 209         netdev_lockdep_set_classes(dev);
 210
 211         ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
 212         if (!ipvlan->pcpu_stats)
 213                 return -ENOMEM;
 214
 215         if (!netif_is_ipvlan_port(phy_dev)) {
 216                 err = ipvlan_port_create(phy_dev);
 217                 if (err < 0) {
 218                         free_percpu(ipvlan->pcpu_stats);
 219                         return err;
 220                 }
 221         }
 222         port = ipvlan_port_get_rtnl(phy_dev);
 223         port->count += 1;
 224         return 0;
 225 }
 226
 227 static void ipvlan_uninit(struct net_device *dev)
 228 {
 229         struct ipvl_dev *ipvlan = netdev_priv(dev);
 230         struct net_device *phy_dev = ipvlan->phy_dev;
 231         struct ipvl_port *port;
 232
 233         free_percpu(ipvlan->pcpu_stats);
 234
 235         port = ipvlan_port_get_rtnl(phy_dev);
 236         port->count -= 1;
 237         if (!port->count)
 238                 ipvlan_port_destroy(port->dev);
 239 }
 240
 241 static int ipvlan_open(struct net_device *dev)
 242 {
 243         struct ipvl_dev *ipvlan = netdev_priv(dev);
 244         struct net_device *phy_dev = ipvlan->phy_dev;
 245         struct ipvl_addr *addr;
 246
 247         if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
 248             ipvlan->port->mode == IPVLAN_MODE_L3S)
 249                 dev->flags |= IFF_NOARP;
 250         else
 251                 dev->flags &= ~IFF_NOARP;
 252
 253         rcu_read_lock();
 254         list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
 255                 ipvlan_ht_addr_add(ipvlan, addr);
 256         rcu_read_unlock();
 257
 258         return dev_uc_add(phy_dev, phy_dev->dev_addr);
 259 }
 260
 261 static int ipvlan_stop(struct net_device *dev)
 262 {
 263         struct ipvl_dev *ipvlan = netdev_priv(dev);
 264         struct net_device *phy_dev = ipvlan->phy_dev;
 265         struct ipvl_addr *addr;
 266
 267         dev_uc_unsync(phy_dev, dev);
 268         dev_mc_unsync(phy_dev, dev);
 269
 270         dev_uc_del(phy_dev, phy_dev->dev_addr);
 271
 272         rcu_read_lock();
 273         list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
 274                 ipvlan_ht_addr_del(addr);
 275         rcu_read_unlock();
 276
 277         return 0;
 278 }
 279
 280 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 281                                      struct net_device *dev)
 282 {
 283         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 284         int skblen = skb->len;
 285         int ret;
 286
 287         ret = ipvlan_queue_xmit(skb, dev);
 288         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 289                 struct ipvl_pcpu_stats *pcptr;
 290
 291                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 292
 293                 u64_stats_update_begin(&pcptr->syncp);
 294                 pcptr->tx_pkts++;
 295                 pcptr->tx_bytes += skblen;
 296                 u64_stats_update_end(&pcptr->syncp);
 297         } else {
 298                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 299         }
 300         return ret;
 301 }
 302
 303 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 304                                              netdev_features_t features)
 305 {
 306         struct ipvl_dev *ipvlan = netdev_priv(dev);
 307
 308         return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 309 }
 310
 311 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 312 {
 313         struct ipvl_dev *ipvlan = netdev_priv(dev);
 314         struct net_device *phy_dev = ipvlan->phy_dev;
 315
 316         if (change & IFF_ALLMULTI)
 317                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 318 }
 319
 320 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 321 {
 322         struct ipvl_dev *ipvlan = netdev_priv(dev);
 323
 324         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 325                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 326         } else {
 327                 struct netdev_hw_addr *ha;
 328                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 329
 330                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 331                 netdev_for_each_mc_addr(ha, dev)
 332                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 333
 334                 /* Turn-on broadcast bit irrespective of address family,
 335                  * since broadcast is deferred to a work-queue, hence no
 336                  * impact on fast-path processing.
 337                  */
 338                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 339
 340                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 341                             IPVLAN_MAC_FILTER_SIZE);
 342         }
 343         dev_uc_sync(ipvlan->phy_dev, dev);
 344         dev_mc_sync(ipvlan->phy_dev, dev);
 345 }
 346
 347 static void ipvlan_get_stats64(struct net_device *dev,
 348                                struct rtnl_link_stats64 *s)
 349 {
 350         struct ipvl_dev *ipvlan = netdev_priv(dev);
 351
 352         if (ipvlan->pcpu_stats) {
 353                 struct ipvl_pcpu_stats *pcptr;
 354                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 355                 u32 rx_errs = 0, tx_drps = 0;
 356                 u32 strt;
 357                 int idx;
 358
 359                 for_each_possible_cpu(idx) {
 360                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 361                         do {
 362                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 363                                 rx_pkts = pcptr->rx_pkts;
 364                                 rx_bytes = pcptr->rx_bytes;
 365                                 rx_mcast = pcptr->rx_mcast;
 366                                 tx_pkts = pcptr->tx_pkts;
 367                                 tx_bytes = pcptr->tx_bytes;
 368                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 369                                                            strt));
 370
 371                         s->rx_packets += rx_pkts;
 372                         s->rx_bytes += rx_bytes;
 373                         s->multicast += rx_mcast;
 374                         s->tx_packets += tx_pkts;
 375                         s->tx_bytes += tx_bytes;
 376
 377                         /* u32 values are updated without syncp protection. */
 378                         rx_errs += pcptr->rx_errs;
 379                         tx_drps += pcptr->tx_drps;
 380                 }
 381                 s->rx_errors = rx_errs;
 382                 s->rx_dropped = rx_errs;
 383                 s->tx_dropped = tx_drps;
 384         }
 385 }
 386
 387 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 388 {
 389         struct ipvl_dev *ipvlan = netdev_priv(dev);
 390         struct net_device *phy_dev = ipvlan->phy_dev;
 391
 392         return vlan_vid_add(phy_dev, proto, vid);
 393 }
 394
 395 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 396                                    u16 vid)
 397 {
 398         struct ipvl_dev *ipvlan = netdev_priv(dev);
 399         struct net_device *phy_dev = ipvlan->phy_dev;
 400
 401         vlan_vid_del(phy_dev, proto, vid);
 402         return 0;
 403 }
 404
 405 static int ipvlan_get_iflink(const struct net_device *dev)
 406 {
 407         struct ipvl_dev *ipvlan = netdev_priv(dev);
 408
 409         return ipvlan->phy_dev->ifindex;
 410 }
 411
 412 static const struct net_device_ops ipvlan_netdev_ops = {
 413         .ndo_init               = ipvlan_init,
 414         .ndo_uninit             = ipvlan_uninit,
 415         .ndo_open               = ipvlan_open,
 416         .ndo_stop               = ipvlan_stop,
 417         .ndo_start_xmit         = ipvlan_start_xmit,
 418         .ndo_fix_features       = ipvlan_fix_features,
 419         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 420         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 421         .ndo_get_stats64        = ipvlan_get_stats64,
 422         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 423         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 424         .ndo_get_iflink         = ipvlan_get_iflink,
 425 };
 426
 427 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 428                               unsigned short type, const void *daddr,
 429                               const void *saddr, unsigned len)
 430 {
 431         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 432         struct net_device *phy_dev = ipvlan->phy_dev;
 433
 434         /* TODO Probably use a different field than dev_addr so that the
 435          * mac-address on the virtual device is portable and can be carried
 436          * while the packets use the mac-addr on the physical device.
 437          */
 438         return dev_hard_header(skb, phy_dev, type, daddr,
 439                                saddr ? : phy_dev->dev_addr, len);
 440 }
 441
 442 static const struct header_ops ipvlan_header_ops = {
 443         .create         = ipvlan_hard_header,
 444         .parse          = eth_header_parse,
 445         .cache          = eth_header_cache,
 446         .cache_update   = eth_header_cache_update,
 447 };
 448
 449 static bool netif_is_ipvlan(const struct net_device *dev)
 450 {
 451         /* both ipvlan and ipvtap devices use the same netdev_ops */
 452         return dev->netdev_ops == &ipvlan_netdev_ops;
 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
 503         if (data[IFLA_IPVLAN_MODE]) {
 504                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 505
 506                 err = ipvlan_set_port_mode(port, nmode, extack);
 507         }
 508
 509         if (!err && data[IFLA_IPVLAN_FLAGS]) {
 510                 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 511
 512                 if (flags & IPVLAN_F_PRIVATE)
 513                         ipvlan_mark_private(port);
 514                 else
 515                         ipvlan_clear_private(port);
 516
 517                 if (flags & IPVLAN_F_VEPA)
 518                         ipvlan_mark_vepa(port);
 519                 else
 520                         ipvlan_clear_vepa(port);
 521         }
 522
 523         return err;
 524 }
 525
 526 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 527 {
 528         return (0
 529                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 530                 + nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
 531                 );
 532 }
 533
 534 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
 535                               struct netlink_ext_ack *extack)
 536 {
 537         if (!data)
 538                 return 0;
 539
 540         if (data[IFLA_IPVLAN_MODE]) {
 541                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 542
 543                 if (mode >= IPVLAN_MODE_MAX)
 544                         return -EINVAL;
 545         }
 546         if (data[IFLA_IPVLAN_FLAGS]) {
 547                 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 548
 549                 /* Only two bits are used at this moment. */
 550                 if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
 551                         return -EINVAL;
 552                 /* Also both flags can't be active at the same time. */
 553                 if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
 554                     (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
 555                         return -EINVAL;
 556         }
 557
 558         return 0;
 559 }
 560
 561 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 562                               const struct net_device *dev)
 563 {
 564         struct ipvl_dev *ipvlan = netdev_priv(dev);
 565         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 566         int ret = -EINVAL;
 567
 568         if (!port)
 569                 goto err;
 570
 571         ret = -EMSGSIZE;
 572         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 573                 goto err;
 574         if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
 575                 goto err;
 576
 577         return 0;
 578
 579 err:
 580         return ret;
 581 }
 582
 583 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 584                     struct nlattr *tb[], struct nlattr *data[],
 585                     struct netlink_ext_ack *extack)
 586 {
 587         struct ipvl_dev *ipvlan = netdev_priv(dev);
 588         struct ipvl_port *port;
 589         struct net_device *phy_dev;
 590         int err;
 591         u16 mode = IPVLAN_MODE_L3;
 592
 593         if (!tb[IFLA_LINK])
 594                 return -EINVAL;
 595
 596         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 597         if (!phy_dev)
 598                 return -ENODEV;
 599
 600         if (netif_is_ipvlan(phy_dev)) {
 601                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 602
 603                 phy_dev = tmp->phy_dev;
 604         } else if (!netif_is_ipvlan_port(phy_dev)) {
 605                 /* Exit early if the underlying link is invalid or busy */
 606                 if (phy_dev->type != ARPHRD_ETHER ||
 607                     phy_dev->flags & IFF_LOOPBACK) {
 608                         netdev_err(phy_dev,
 609                                    "Master is either lo or non-ether device\n");
 610                         return -EINVAL;
 611                 }
 612
 613                 if (netdev_is_rx_handler_busy(phy_dev)) {
 614                         netdev_err(phy_dev, "Device is already in use.\n");
 615                         return -EBUSY;
 616                 }
 617         }
 618
 619         ipvlan->phy_dev = phy_dev;
 620         ipvlan->dev = dev;
 621         ipvlan->sfeatures = IPVLAN_FEATURES;
 622         if (!tb[IFLA_MTU])
 623                 ipvlan_adjust_mtu(ipvlan, phy_dev);
 624         INIT_LIST_HEAD(&ipvlan->addrs);
 625         spin_lock_init(&ipvlan->addrs_lock);
 626
 627         /* TODO Probably put random address here to be presented to the
 628          * world but keep using the physical-dev address for the outgoing
 629          * packets.
 630          */
 631         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 632
 633         dev->priv_flags |= IFF_NO_RX_HANDLER;
 634
 635         err = register_netdevice(dev);
 636         if (err < 0)
 637                 return err;
 638
 639         /* ipvlan_init() would have created the port, if required */
 640         port = ipvlan_port_get_rtnl(phy_dev);
 641         ipvlan->port = port;
 642
 643         /* If the port-id base is at the MAX value, then wrap it around and
 644          * begin from 0x1 again. This may be due to a busy system where lots
 645          * of slaves are getting created and deleted.
 646          */
 647         if (port->dev_id_start == 0xFFFE)
 648                 port->dev_id_start = 0x1;
 649
 650         /* Since L2 address is shared among all IPvlan slaves including
 651          * master, use unique 16 bit dev-ids to diffentiate among them.
 652          * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
 653          * slave link [see addrconf_ifid_eui48()].
 654          */
 655         err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
 656                              GFP_KERNEL);
 657         if (err < 0)
 658                 err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
 659                                      GFP_KERNEL);
 660         if (err < 0)
 661                 goto unregister_netdev;
 662         dev->dev_id = err;
 663
 664         /* Increment id-base to the next slot for the future assignment */
 665         port->dev_id_start = err + 1;
 666
 667         err = netdev_upper_dev_link(phy_dev, dev, extack);
 668         if (err)
 669                 goto remove_ida;
 670
 671         /* Flags are per port and latest update overrides. User has
 672          * to be consistent in setting it just like the mode attribute.
 673          */
 674         if (data && data[IFLA_IPVLAN_FLAGS])
 675                 port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 676
 677         if (data && data[IFLA_IPVLAN_MODE])
 678                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 679
 680         err = ipvlan_set_port_mode(port, mode, extack);
 681         if (err)
 682                 goto unlink_netdev;
 683
 684         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 685         netif_stacked_transfer_operstate(phy_dev, dev);
 686         return 0;
 687
 688 unlink_netdev:
 689         netdev_upper_dev_unlink(phy_dev, dev);
 690 remove_ida:
 691         ida_simple_remove(&port->ida, dev->dev_id);
 692 unregister_netdev:
 693         unregister_netdevice(dev);
 694         return err;
 695 }
 696 EXPORT_SYMBOL_GPL(ipvlan_link_new);
 697
 698 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 699 {
 700         struct ipvl_dev *ipvlan = netdev_priv(dev);
 701         struct ipvl_addr *addr, *next;
 702
 703         spin_lock_bh(&ipvlan->addrs_lock);
 704         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 705                 ipvlan_ht_addr_del(addr);
 706                 list_del_rcu(&addr->anode);
 707                 kfree_rcu(addr, rcu);
 708         }
 709         spin_unlock_bh(&ipvlan->addrs_lock);
 710
 711         ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
 712         list_del_rcu(&ipvlan->pnode);
 713         unregister_netdevice_queue(dev, head);
 714         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 715 }
 716 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
 717
 718 void ipvlan_link_setup(struct net_device *dev)
 719 {
 720         ether_setup(dev);
 721
 722         dev->max_mtu = ETH_MAX_MTU;
 723         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 724         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 725         dev->netdev_ops = &ipvlan_netdev_ops;
 726         dev->needs_free_netdev = true;
 727         dev->header_ops = &ipvlan_header_ops;
 728         dev->ethtool_ops = &ipvlan_ethtool_ops;
 729 }
 730 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
 731
 732 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 733 {
 734         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 735         [IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
 736 };
 737
 738 static struct rtnl_link_ops ipvlan_link_ops = {
 739         .kind           = "ipvlan",
 740         .priv_size      = sizeof(struct ipvl_dev),
 741
 742         .setup          = ipvlan_link_setup,
 743         .newlink        = ipvlan_link_new,
 744         .dellink        = ipvlan_link_delete,
 745 };
 746
 747 int ipvlan_link_register(struct rtnl_link_ops *ops)
 748 {
 749         ops->get_size   = ipvlan_nl_getsize;
 750         ops->policy     = ipvlan_nl_policy;
 751         ops->validate   = ipvlan_nl_validate;
 752         ops->fill_info  = ipvlan_nl_fillinfo;
 753         ops->changelink = ipvlan_nl_changelink;
 754         ops->maxtype    = IFLA_IPVLAN_MAX;
 755         return rtnl_link_register(ops);
 756 }
 757 EXPORT_SYMBOL_GPL(ipvlan_link_register);
 758
 759 static int ipvlan_device_event(struct notifier_block *unused,
 760                                unsigned long event, void *ptr)
 761 {
 762         struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
 763         struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
 764         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 765         struct ipvl_dev *ipvlan, *next;
 766         struct ipvl_port *port;
 767         LIST_HEAD(lst_kill);
 768         int err;
 769
 770         if (!netif_is_ipvlan_port(dev))
 771                 return NOTIFY_DONE;
 772
 773         port = ipvlan_port_get_rtnl(dev);
 774
 775         switch (event) {
 776         case NETDEV_CHANGE:
 777                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 778                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 779                                                          ipvlan->dev);
 780                 break;
 781
 782         case NETDEV_REGISTER: {
 783                 struct net *oldnet, *newnet = dev_net(dev);
 784                 struct ipvlan_netns *old_vnet;
 785
 786                 oldnet = read_pnet(&port->pnet);
 787                 if (net_eq(newnet, oldnet))
 788                         break;
 789
 790                 write_pnet(&port->pnet, newnet);
 791
 792                 old_vnet = net_generic(oldnet, ipvlan_netid);
 793                 if (!old_vnet->ipvl_nf_hook_refcnt)
 794                         break;
 795
 796                 ipvlan_register_nf_hook(newnet);
 797                 ipvlan_unregister_nf_hook(oldnet);
 798                 break;
 799         }
 800         case NETDEV_UNREGISTER:
 801                 if (dev->reg_state != NETREG_UNREGISTERING)
 802                         break;
 803
 804                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
 805                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 806                                                             &lst_kill);
 807                 unregister_netdevice_many(&lst_kill);
 808                 break;
 809
 810         case NETDEV_FEAT_CHANGE:
 811                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 812                         ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
 813                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 814                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 815                         netdev_features_change(ipvlan->dev);
 816                 }
 817                 break;
 818
 819         case NETDEV_CHANGEMTU:
 820                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 821                         ipvlan_adjust_mtu(ipvlan, dev);
 822                 break;
 823
 824         case NETDEV_PRE_CHANGEADDR:
 825                 prechaddr_info = ptr;
 826                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 827                         err = dev_pre_changeaddr_notify(ipvlan->dev,
 828                                                     prechaddr_info->dev_addr,
 829                                                     extack);
 830                         if (err)
 831                                 return notifier_from_errno(err);
 832                 }
 833                 break;
 834
 835         case NETDEV_CHANGEADDR:
 836                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 837                         ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
 838                         call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
 839                 }
 840                 break;
 841
 842         case NETDEV_PRE_TYPE_CHANGE:
 843                 /* Forbid underlying device to change its type. */
 844                 return NOTIFY_BAD;
 845         }
 846         return NOTIFY_DONE;
 847 }
 848
 849 /* the caller must held the addrs lock */
 850 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 851 {
 852         struct ipvl_addr *addr;
 853
 854         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 855         if (!addr)
 856                 return -ENOMEM;
 857
 858         addr->master = ipvlan;
 859         if (!is_v6) {
 860                 memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
 861                 addr->atype = IPVL_IPV4;
 862 #if IS_ENABLED(CONFIG_IPV6)
 863         } else {
 864                 memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
 865                 addr->atype = IPVL_IPV6;
 866 #endif
 867         }
 868
 869         list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
 870
 871         /* If the interface is not up, the address will be added to the hash
 872          * list by ipvlan_open.
 873          */
 874         if (netif_running(ipvlan->dev))
 875                 ipvlan_ht_addr_add(ipvlan, addr);
 876
 877         return 0;
 878 }
 879
 880 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 881 {
 882         struct ipvl_addr *addr;
 883
 884         spin_lock_bh(&ipvlan->addrs_lock);
 885         addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
 886         if (!addr) {
 887                 spin_unlock_bh(&ipvlan->addrs_lock);
 888                 return;
 889         }
 890
 891         ipvlan_ht_addr_del(addr);
 892         list_del_rcu(&addr->anode);
 893         spin_unlock_bh(&ipvlan->addrs_lock);
 894         kfree_rcu(addr, rcu);
 895 }
 896
 897 static bool ipvlan_is_valid_dev(const struct net_device *dev)
 898 {
 899         struct ipvl_dev *ipvlan = netdev_priv(dev);
 900
 901         if (!netif_is_ipvlan(dev))
 902                 return false;
 903
 904         if (!ipvlan || !ipvlan->port)
 905                 return false;
 906
 907         return true;
 908 }
 909
 910 #if IS_ENABLED(CONFIG_IPV6)
 911 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 912 {
 913         int ret = -EINVAL;
 914
 915         spin_lock_bh(&ipvlan->addrs_lock);
 916         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
 917                 netif_err(ipvlan, ifup, ipvlan->dev,
 918                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 919                           ip6_addr, ipvlan->dev->name);
 920         else
 921                 ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
 922         spin_unlock_bh(&ipvlan->addrs_lock);
 923         return ret;
 924 }
 925
 926 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 927 {
 928         return ipvlan_del_addr(ipvlan, ip6_addr, true);
 929 }
 930
 931 static int ipvlan_addr6_event(struct notifier_block *unused,
 932                               unsigned long event, void *ptr)
 933 {
 934         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 935         struct net_device *dev = (struct net_device *)if6->idev->dev;
 936         struct ipvl_dev *ipvlan = netdev_priv(dev);
 937
 938         if (!ipvlan_is_valid_dev(dev))
 939                 return NOTIFY_DONE;
 940
 941         switch (event) {
 942         case NETDEV_UP:
 943                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 944                         return NOTIFY_BAD;
 945                 break;
 946
 947         case NETDEV_DOWN:
 948                 ipvlan_del_addr6(ipvlan, &if6->addr);
 949                 break;
 950         }
 951
 952         return NOTIFY_OK;
 953 }
 954
 955 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
 956                                         unsigned long event, void *ptr)
 957 {
 958         struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
 959         struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
 960         struct ipvl_dev *ipvlan = netdev_priv(dev);
 961
 962         if (!ipvlan_is_valid_dev(dev))
 963                 return NOTIFY_DONE;
 964
 965         switch (event) {
 966         case NETDEV_UP:
 967                 if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
 968                         NL_SET_ERR_MSG(i6vi->extack,
 969                                        "Address already assigned to an ipvlan device");
 970                         return notifier_from_errno(-EADDRINUSE);
 971                 }
 972                 break;
 973         }
 974
 975         return NOTIFY_OK;
 976 }
 977 #endif
 978
 979 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 980 {
 981         int ret = -EINVAL;
 982
 983         spin_lock_bh(&ipvlan->addrs_lock);
 984         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
 985                 netif_err(ipvlan, ifup, ipvlan->dev,
 986                           "Failed to add IPv4=%pI4 on %s intf.\n",
 987                           ip4_addr, ipvlan->dev->name);
 988         else
 989                 ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
 990         spin_unlock_bh(&ipvlan->addrs_lock);
 991         return ret;
 992 }
 993
 994 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 995 {
 996         return ipvlan_del_addr(ipvlan, ip4_addr, false);
 997 }
 998
 999 static int ipvlan_addr4_event(struct notifier_block *unused,
1000                               unsigned long event, void *ptr)
1001 {
1002         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
1003         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
1004         struct ipvl_dev *ipvlan = netdev_priv(dev);
1005         struct in_addr ip4_addr;
1006
1007         if (!ipvlan_is_valid_dev(dev))
1008                 return NOTIFY_DONE;
1009
1010         switch (event) {
1011         case NETDEV_UP:
1012                 ip4_addr.s_addr = if4->ifa_address;
1013                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
1014                         return NOTIFY_BAD;
1015                 break;
1016
1017         case NETDEV_DOWN:
1018                 ip4_addr.s_addr = if4->ifa_address;
1019                 ipvlan_del_addr4(ipvlan, &ip4_addr);
1020                 break;
1021         }
1022
1023         return NOTIFY_OK;
1024 }
1025
1026 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
1027                                         unsigned long event, void *ptr)
1028 {
1029         struct in_validator_info *ivi = (struct in_validator_info *)ptr;
1030         struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
1031         struct ipvl_dev *ipvlan = netdev_priv(dev);
1032
1033         if (!ipvlan_is_valid_dev(dev))
1034                 return NOTIFY_DONE;
1035
1036         switch (event) {
1037         case NETDEV_UP:
1038                 if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
1039                         NL_SET_ERR_MSG(ivi->extack,
1040                                        "Address already assigned to an ipvlan device");
1041                         return notifier_from_errno(-EADDRINUSE);
1042                 }
1043                 break;
1044         }
1045
1046         return NOTIFY_OK;
1047 }
1048
1049 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1050         .notifier_call = ipvlan_addr4_event,
1051 };
1052
1053 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1054         .notifier_call = ipvlan_addr4_validator_event,
1055 };
1056
1057 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1058         .notifier_call = ipvlan_device_event,
1059 };
1060
1061 #if IS_ENABLED(CONFIG_IPV6)
1062 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1063         .notifier_call = ipvlan_addr6_event,
1064 };
1065
1066 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1067         .notifier_call = ipvlan_addr6_validator_event,
1068 };
1069 #endif
1070
1071 static void ipvlan_ns_exit(struct net *net)
1072 {
1073         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
1074
1075         if (WARN_ON_ONCE(vnet->ipvl_nf_hook_refcnt)) {
1076                 vnet->ipvl_nf_hook_refcnt = 0;
1077                 nf_unregister_net_hooks(net, ipvl_nfops,
1078                                         ARRAY_SIZE(ipvl_nfops));
1079         }
1080 }
1081
1082 static struct pernet_operations ipvlan_net_ops = {
1083         .id = &ipvlan_netid,
1084         .size = sizeof(struct ipvlan_netns),
1085         .exit = ipvlan_ns_exit,
1086 };
1087
1088 static int __init ipvlan_init_module(void)
1089 {
1090         int err;
1091
1092         ipvlan_init_secret();
1093         register_netdevice_notifier(&ipvlan_notifier_block);
1094 #if IS_ENABLED(CONFIG_IPV6)
1095         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1096         register_inet6addr_validator_notifier(
1097             &ipvlan_addr6_vtor_notifier_block);
1098 #endif
1099         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1100         register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1101
1102         err = register_pernet_subsys(&ipvlan_net_ops);
1103         if (err < 0)
1104                 goto error;
1105
1106         err = ipvlan_link_register(&ipvlan_link_ops);
1107         if (err < 0) {
1108                 unregister_pernet_subsys(&ipvlan_net_ops);
1109                 goto error;
1110         }
1111
1112         return 0;
1113 error:
1114         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1115         unregister_inetaddr_validator_notifier(
1116             &ipvlan_addr4_vtor_notifier_block);
1117 #if IS_ENABLED(CONFIG_IPV6)
1118         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1119         unregister_inet6addr_validator_notifier(
1120             &ipvlan_addr6_vtor_notifier_block);
1121 #endif
1122         unregister_netdevice_notifier(&ipvlan_notifier_block);
1123         return err;
1124 }
1125
1126 static void __exit ipvlan_cleanup_module(void)
1127 {
1128         rtnl_link_unregister(&ipvlan_link_ops);
1129         unregister_pernet_subsys(&ipvlan_net_ops);
1130         unregister_netdevice_notifier(&ipvlan_notifier_block);
1131         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1132         unregister_inetaddr_validator_notifier(
1133             &ipvlan_addr4_vtor_notifier_block);
1134 #if IS_ENABLED(CONFIG_IPV6)
1135         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1136         unregister_inet6addr_validator_notifier(
1137             &ipvlan_addr6_vtor_notifier_block);
1138 #endif
1139 }
1140
1141 module_init(ipvlan_init_module);
1142 module_exit(ipvlan_cleanup_module);
1143
1144 MODULE_LICENSE("GPL");
1145 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1146 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1147 MODULE_ALIAS_RTNL_LINK("ipvlan");