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 {
  76         struct ipvl_dev *ipvlan;
  77         struct net_device *mdev = port->dev;
  78         unsigned int flags;
  79         int err;
  80
  81         ASSERT_RTNL();
  82         if (port->mode != nval) {
  83                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
  84                         flags = ipvlan->dev->flags;
  85                         if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
  86                                 err = dev_change_flags(ipvlan->dev,
  87                                                        flags | IFF_NOARP);
  88                         } else {
  89                                 err = dev_change_flags(ipvlan->dev,
  90                                                        flags & ~IFF_NOARP);
  91                         }
  92                         if (unlikely(err))
  93                                 goto fail;
  94                 }
  95                 if (nval == IPVLAN_MODE_L3S) {
  96                         /* New mode is L3S */
  97                         err = ipvlan_register_nf_hook(read_pnet(&port->pnet));
  98                         if (!err) {
  99                                 mdev->l3mdev_ops = &ipvl_l3mdev_ops;
 100                                 mdev->priv_flags |= IFF_L3MDEV_RX_HANDLER;
 101                         } else
 102                                 goto fail;
 103                 } else if (port->mode == IPVLAN_MODE_L3S) {
 104                         /* Old mode was L3S */
 105                         mdev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
 106                         ipvlan_unregister_nf_hook(read_pnet(&port->pnet));
 107                         mdev->l3mdev_ops = NULL;
 108                 }
 109                 port->mode = nval;
 110         }
 111         return 0;
 112
 113 fail:
 114         /* Undo the flags changes that have been done so far. */
 115         list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
 116                 flags = ipvlan->dev->flags;
 117                 if (port->mode == IPVLAN_MODE_L3 ||
 118                     port->mode == IPVLAN_MODE_L3S)
 119                         dev_change_flags(ipvlan->dev, flags | IFF_NOARP);
 120                 else
 121                         dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP);
 122         }
 123
 124         return err;
 125 }
 126
 127 static int ipvlan_port_create(struct net_device *dev)
 128 {
 129         struct ipvl_port *port;
 130         int err, idx;
 131
 132         port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
 133         if (!port)
 134                 return -ENOMEM;
 135
 136         write_pnet(&port->pnet, dev_net(dev));
 137         port->dev = dev;
 138         port->mode = IPVLAN_MODE_L3;
 139         INIT_LIST_HEAD(&port->ipvlans);
 140         for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
 141                 INIT_HLIST_HEAD(&port->hlhead[idx]);
 142
 143         skb_queue_head_init(&port->backlog);
 144         INIT_WORK(&port->wq, ipvlan_process_multicast);
 145         ida_init(&port->ida);
 146         port->dev_id_start = 1;
 147
 148         err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
 149         if (err)
 150                 goto err;
 151
 152         return 0;
 153
 154 err:
 155         kfree(port);
 156         return err;
 157 }
 158
 159 static void ipvlan_port_destroy(struct net_device *dev)
 160 {
 161         struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
 162         struct sk_buff *skb;
 163
 164         if (port->mode == IPVLAN_MODE_L3S) {
 165                 dev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
 166                 ipvlan_unregister_nf_hook(dev_net(dev));
 167                 dev->l3mdev_ops = NULL;
 168         }
 169         netdev_rx_handler_unregister(dev);
 170         cancel_work_sync(&port->wq);
 171         while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
 172                 if (skb->dev)
 173                         dev_put(skb->dev);
 174                 kfree_skb(skb);
 175         }
 176         ida_destroy(&port->ida);
 177         kfree(port);
 178 }
 179
 180 #define IPVLAN_ALWAYS_ON_OFLOADS \
 181         (NETIF_F_SG | NETIF_F_HW_CSUM | \
 182          NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
 183
 184 #define IPVLAN_ALWAYS_ON \
 185         (IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED)
 186
 187 #define IPVLAN_FEATURES \
 188         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 189          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
 190          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 191          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 192
 193         /* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
 194
 195 #define IPVLAN_STATE_MASK \
 196         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 197
 198 static int ipvlan_init(struct net_device *dev)
 199 {
 200         struct ipvl_dev *ipvlan = netdev_priv(dev);
 201         struct net_device *phy_dev = ipvlan->phy_dev;
 202         struct ipvl_port *port;
 203         int err;
 204
 205         dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
 206                      (phy_dev->state & IPVLAN_STATE_MASK);
 207         dev->features = phy_dev->features & IPVLAN_FEATURES;
 208         dev->features |= IPVLAN_ALWAYS_ON;
 209         dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
 210         dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
 211         dev->gso_max_size = phy_dev->gso_max_size;
 212         dev->gso_max_segs = phy_dev->gso_max_segs;
 213         dev->hard_header_len = phy_dev->hard_header_len;
 214
 215         netdev_lockdep_set_classes(dev);
 216
 217         ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
 218         if (!ipvlan->pcpu_stats)
 219                 return -ENOMEM;
 220
 221         if (!netif_is_ipvlan_port(phy_dev)) {
 222                 err = ipvlan_port_create(phy_dev);
 223                 if (err < 0) {
 224                         free_percpu(ipvlan->pcpu_stats);
 225                         return err;
 226                 }
 227         }
 228         port = ipvlan_port_get_rtnl(phy_dev);
 229         port->count += 1;
 230         return 0;
 231 }
 232
 233 static void ipvlan_uninit(struct net_device *dev)
 234 {
 235         struct ipvl_dev *ipvlan = netdev_priv(dev);
 236         struct net_device *phy_dev = ipvlan->phy_dev;
 237         struct ipvl_port *port;
 238
 239         free_percpu(ipvlan->pcpu_stats);
 240
 241         port = ipvlan_port_get_rtnl(phy_dev);
 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         rcu_read_lock();
 259         list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
 260                 ipvlan_ht_addr_add(ipvlan, addr);
 261         rcu_read_unlock();
 262
 263         return 0;
 264 }
 265
 266 static int ipvlan_stop(struct net_device *dev)
 267 {
 268         struct ipvl_dev *ipvlan = netdev_priv(dev);
 269         struct net_device *phy_dev = ipvlan->phy_dev;
 270         struct ipvl_addr *addr;
 271
 272         dev_uc_unsync(phy_dev, dev);
 273         dev_mc_unsync(phy_dev, dev);
 274
 275         rcu_read_lock();
 276         list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
 277                 ipvlan_ht_addr_del(addr);
 278         rcu_read_unlock();
 279
 280         return 0;
 281 }
 282
 283 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
 284                                      struct net_device *dev)
 285 {
 286         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 287         int skblen = skb->len;
 288         int ret;
 289
 290         ret = ipvlan_queue_xmit(skb, dev);
 291         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 292                 struct ipvl_pcpu_stats *pcptr;
 293
 294                 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
 295
 296                 u64_stats_update_begin(&pcptr->syncp);
 297                 pcptr->tx_pkts++;
 298                 pcptr->tx_bytes += skblen;
 299                 u64_stats_update_end(&pcptr->syncp);
 300         } else {
 301                 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
 302         }
 303         return ret;
 304 }
 305
 306 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
 307                                              netdev_features_t features)
 308 {
 309         struct ipvl_dev *ipvlan = netdev_priv(dev);
 310
 311         features |= NETIF_F_ALL_FOR_ALL;
 312         features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
 313         features = netdev_increment_features(ipvlan->phy_dev->features,
 314                                              features, features);
 315         features |= IPVLAN_ALWAYS_ON;
 316         features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
 317
 318         return features;
 319 }
 320
 321 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
 322 {
 323         struct ipvl_dev *ipvlan = netdev_priv(dev);
 324         struct net_device *phy_dev = ipvlan->phy_dev;
 325
 326         if (change & IFF_ALLMULTI)
 327                 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
 328 }
 329
 330 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
 331 {
 332         struct ipvl_dev *ipvlan = netdev_priv(dev);
 333
 334         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 335                 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
 336         } else {
 337                 struct netdev_hw_addr *ha;
 338                 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 339
 340                 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
 341                 netdev_for_each_mc_addr(ha, dev)
 342                         __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
 343
 344                 /* Turn-on broadcast bit irrespective of address family,
 345                  * since broadcast is deferred to a work-queue, hence no
 346                  * impact on fast-path processing.
 347                  */
 348                 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
 349
 350                 bitmap_copy(ipvlan->mac_filters, mc_filters,
 351                             IPVLAN_MAC_FILTER_SIZE);
 352         }
 353         dev_uc_sync(ipvlan->phy_dev, dev);
 354         dev_mc_sync(ipvlan->phy_dev, dev);
 355 }
 356
 357 static void ipvlan_get_stats64(struct net_device *dev,
 358                                struct rtnl_link_stats64 *s)
 359 {
 360         struct ipvl_dev *ipvlan = netdev_priv(dev);
 361
 362         if (ipvlan->pcpu_stats) {
 363                 struct ipvl_pcpu_stats *pcptr;
 364                 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
 365                 u32 rx_errs = 0, tx_drps = 0;
 366                 u32 strt;
 367                 int idx;
 368
 369                 for_each_possible_cpu(idx) {
 370                         pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
 371                         do {
 372                                 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
 373                                 rx_pkts = pcptr->rx_pkts;
 374                                 rx_bytes = pcptr->rx_bytes;
 375                                 rx_mcast = pcptr->rx_mcast;
 376                                 tx_pkts = pcptr->tx_pkts;
 377                                 tx_bytes = pcptr->tx_bytes;
 378                         } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
 379                                                            strt));
 380
 381                         s->rx_packets += rx_pkts;
 382                         s->rx_bytes += rx_bytes;
 383                         s->multicast += rx_mcast;
 384                         s->tx_packets += tx_pkts;
 385                         s->tx_bytes += tx_bytes;
 386
 387                         /* u32 values are updated without syncp protection. */
 388                         rx_errs += pcptr->rx_errs;
 389                         tx_drps += pcptr->tx_drps;
 390                 }
 391                 s->rx_errors = rx_errs;
 392                 s->rx_dropped = rx_errs;
 393                 s->tx_dropped = tx_drps;
 394         }
 395 }
 396
 397 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
 398 {
 399         struct ipvl_dev *ipvlan = netdev_priv(dev);
 400         struct net_device *phy_dev = ipvlan->phy_dev;
 401
 402         return vlan_vid_add(phy_dev, proto, vid);
 403 }
 404
 405 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
 406                                    u16 vid)
 407 {
 408         struct ipvl_dev *ipvlan = netdev_priv(dev);
 409         struct net_device *phy_dev = ipvlan->phy_dev;
 410
 411         vlan_vid_del(phy_dev, proto, vid);
 412         return 0;
 413 }
 414
 415 static int ipvlan_get_iflink(const struct net_device *dev)
 416 {
 417         struct ipvl_dev *ipvlan = netdev_priv(dev);
 418
 419         return ipvlan->phy_dev->ifindex;
 420 }
 421
 422 static const struct net_device_ops ipvlan_netdev_ops = {
 423         .ndo_init               = ipvlan_init,
 424         .ndo_uninit             = ipvlan_uninit,
 425         .ndo_open               = ipvlan_open,
 426         .ndo_stop               = ipvlan_stop,
 427         .ndo_start_xmit         = ipvlan_start_xmit,
 428         .ndo_fix_features       = ipvlan_fix_features,
 429         .ndo_change_rx_flags    = ipvlan_change_rx_flags,
 430         .ndo_set_rx_mode        = ipvlan_set_multicast_mac_filter,
 431         .ndo_get_stats64        = ipvlan_get_stats64,
 432         .ndo_vlan_rx_add_vid    = ipvlan_vlan_rx_add_vid,
 433         .ndo_vlan_rx_kill_vid   = ipvlan_vlan_rx_kill_vid,
 434         .ndo_get_iflink         = ipvlan_get_iflink,
 435 };
 436
 437 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 438                               unsigned short type, const void *daddr,
 439                               const void *saddr, unsigned len)
 440 {
 441         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 442         struct net_device *phy_dev = ipvlan->phy_dev;
 443
 444         /* TODO Probably use a different field than dev_addr so that the
 445          * mac-address on the virtual device is portable and can be carried
 446          * while the packets use the mac-addr on the physical device.
 447          */
 448         return dev_hard_header(skb, phy_dev, type, daddr,
 449                                saddr ? : phy_dev->dev_addr, len);
 450 }
 451
 452 static const struct header_ops ipvlan_header_ops = {
 453         .create         = ipvlan_hard_header,
 454         .parse          = eth_header_parse,
 455         .cache          = eth_header_cache,
 456         .cache_update   = eth_header_cache_update,
 457 };
 458
 459 static bool netif_is_ipvlan(const struct net_device *dev)
 460 {
 461         /* both ipvlan and ipvtap devices use the same netdev_ops */
 462         return dev->netdev_ops == &ipvlan_netdev_ops;
 463 }
 464
 465 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
 466                                              struct ethtool_link_ksettings *cmd)
 467 {
 468         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 469
 470         return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
 471 }
 472
 473 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
 474                                        struct ethtool_drvinfo *drvinfo)
 475 {
 476         strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
 477         strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
 478 }
 479
 480 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
 481 {
 482         const struct ipvl_dev *ipvlan = netdev_priv(dev);
 483
 484         return ipvlan->msg_enable;
 485 }
 486
 487 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
 488 {
 489         struct ipvl_dev *ipvlan = netdev_priv(dev);
 490
 491         ipvlan->msg_enable = value;
 492 }
 493
 494 static const struct ethtool_ops ipvlan_ethtool_ops = {
 495         .get_link       = ethtool_op_get_link,
 496         .get_link_ksettings     = ipvlan_ethtool_get_link_ksettings,
 497         .get_drvinfo    = ipvlan_ethtool_get_drvinfo,
 498         .get_msglevel   = ipvlan_ethtool_get_msglevel,
 499         .set_msglevel   = ipvlan_ethtool_set_msglevel,
 500 };
 501
 502 static int ipvlan_nl_changelink(struct net_device *dev,
 503                                 struct nlattr *tb[], struct nlattr *data[],
 504                                 struct netlink_ext_ack *extack)
 505 {
 506         struct ipvl_dev *ipvlan = netdev_priv(dev);
 507         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 508         int err = 0;
 509
 510         if (!data)
 511                 return 0;
 512         if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
 513                 return -EPERM;
 514
 515         if (data[IFLA_IPVLAN_MODE]) {
 516                 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 517
 518                 err = ipvlan_set_port_mode(port, nmode);
 519         }
 520
 521         if (!err && data[IFLA_IPVLAN_FLAGS]) {
 522                 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 523
 524                 if (flags & IPVLAN_F_PRIVATE)
 525                         ipvlan_mark_private(port);
 526                 else
 527                         ipvlan_clear_private(port);
 528
 529                 if (flags & IPVLAN_F_VEPA)
 530                         ipvlan_mark_vepa(port);
 531                 else
 532                         ipvlan_clear_vepa(port);
 533         }
 534
 535         return err;
 536 }
 537
 538 static size_t ipvlan_nl_getsize(const struct net_device *dev)
 539 {
 540         return (0
 541                 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
 542                 + nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
 543                 );
 544 }
 545
 546 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
 547                               struct netlink_ext_ack *extack)
 548 {
 549         if (!data)
 550                 return 0;
 551
 552         if (data[IFLA_IPVLAN_MODE]) {
 553                 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 554
 555                 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
 556                         return -EINVAL;
 557         }
 558         if (data[IFLA_IPVLAN_FLAGS]) {
 559                 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 560
 561                 /* Only two bits are used at this moment. */
 562                 if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
 563                         return -EINVAL;
 564                 /* Also both flags can't be active at the same time. */
 565                 if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
 566                     (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
 567                         return -EINVAL;
 568         }
 569
 570         return 0;
 571 }
 572
 573 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
 574                               const struct net_device *dev)
 575 {
 576         struct ipvl_dev *ipvlan = netdev_priv(dev);
 577         struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
 578         int ret = -EINVAL;
 579
 580         if (!port)
 581                 goto err;
 582
 583         ret = -EMSGSIZE;
 584         if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
 585                 goto err;
 586         if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
 587                 goto err;
 588
 589         return 0;
 590
 591 err:
 592         return ret;
 593 }
 594
 595 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
 596                     struct nlattr *tb[], struct nlattr *data[],
 597                     struct netlink_ext_ack *extack)
 598 {
 599         struct ipvl_dev *ipvlan = netdev_priv(dev);
 600         struct ipvl_port *port;
 601         struct net_device *phy_dev;
 602         int err;
 603         u16 mode = IPVLAN_MODE_L3;
 604
 605         if (!tb[IFLA_LINK])
 606                 return -EINVAL;
 607
 608         phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 609         if (!phy_dev)
 610                 return -ENODEV;
 611
 612         if (netif_is_ipvlan(phy_dev)) {
 613                 struct ipvl_dev *tmp = netdev_priv(phy_dev);
 614
 615                 phy_dev = tmp->phy_dev;
 616                 if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
 617                         return -EPERM;
 618         } else if (!netif_is_ipvlan_port(phy_dev)) {
 619                 /* Exit early if the underlying link is invalid or busy */
 620                 if (phy_dev->type != ARPHRD_ETHER ||
 621                     phy_dev->flags & IFF_LOOPBACK) {
 622                         netdev_err(phy_dev,
 623                                    "Master is either lo or non-ether device\n");
 624                         return -EINVAL;
 625                 }
 626
 627                 if (netdev_is_rx_handler_busy(phy_dev)) {
 628                         netdev_err(phy_dev, "Device is already in use.\n");
 629                         return -EBUSY;
 630                 }
 631         }
 632
 633         ipvlan->phy_dev = phy_dev;
 634         ipvlan->dev = dev;
 635         ipvlan->sfeatures = IPVLAN_FEATURES;
 636         if (!tb[IFLA_MTU])
 637                 ipvlan_adjust_mtu(ipvlan, phy_dev);
 638         INIT_LIST_HEAD(&ipvlan->addrs);
 639         spin_lock_init(&ipvlan->addrs_lock);
 640
 641         /* TODO Probably put random address here to be presented to the
 642          * world but keep using the physical-dev address for the outgoing
 643          * packets.
 644          */
 645         memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
 646
 647         dev->priv_flags |= IFF_NO_RX_HANDLER;
 648
 649         err = register_netdevice(dev);
 650         if (err < 0)
 651                 return err;
 652
 653         /* ipvlan_init() would have created the port, if required */
 654         port = ipvlan_port_get_rtnl(phy_dev);
 655         ipvlan->port = port;
 656
 657         /* If the port-id base is at the MAX value, then wrap it around and
 658          * begin from 0x1 again. This may be due to a busy system where lots
 659          * of slaves are getting created and deleted.
 660          */
 661         if (port->dev_id_start == 0xFFFE)
 662                 port->dev_id_start = 0x1;
 663
 664         /* Since L2 address is shared among all IPvlan slaves including
 665          * master, use unique 16 bit dev-ids to diffentiate among them.
 666          * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
 667          * slave link [see addrconf_ifid_eui48()].
 668          */
 669         err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
 670                              GFP_KERNEL);
 671         if (err < 0)
 672                 err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
 673                                      GFP_KERNEL);
 674         if (err < 0)
 675                 goto unregister_netdev;
 676         dev->dev_id = err;
 677
 678         /* Increment id-base to the next slot for the future assignment */
 679         port->dev_id_start = err + 1;
 680
 681         err = netdev_upper_dev_link(phy_dev, dev, extack);
 682         if (err)
 683                 goto remove_ida;
 684
 685         /* Flags are per port and latest update overrides. User has
 686          * to be consistent in setting it just like the mode attribute.
 687          */
 688         if (data && data[IFLA_IPVLAN_FLAGS])
 689                 port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
 690
 691         if (data && data[IFLA_IPVLAN_MODE])
 692                 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
 693
 694         err = ipvlan_set_port_mode(port, mode);
 695         if (err)
 696                 goto unlink_netdev;
 697
 698         list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
 699         netif_stacked_transfer_operstate(phy_dev, dev);
 700         return 0;
 701
 702 unlink_netdev:
 703         netdev_upper_dev_unlink(phy_dev, dev);
 704 remove_ida:
 705         ida_simple_remove(&port->ida, dev->dev_id);
 706 unregister_netdev:
 707         unregister_netdevice(dev);
 708         return err;
 709 }
 710 EXPORT_SYMBOL_GPL(ipvlan_link_new);
 711
 712 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
 713 {
 714         struct ipvl_dev *ipvlan = netdev_priv(dev);
 715         struct ipvl_addr *addr, *next;
 716
 717         spin_lock_bh(&ipvlan->addrs_lock);
 718         list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
 719                 ipvlan_ht_addr_del(addr);
 720                 list_del_rcu(&addr->anode);
 721                 kfree_rcu(addr, rcu);
 722         }
 723         spin_unlock_bh(&ipvlan->addrs_lock);
 724
 725         ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
 726         list_del_rcu(&ipvlan->pnode);
 727         unregister_netdevice_queue(dev, head);
 728         netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
 729 }
 730 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
 731
 732 void ipvlan_link_setup(struct net_device *dev)
 733 {
 734         ether_setup(dev);
 735
 736         dev->max_mtu = ETH_MAX_MTU;
 737         dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 738         dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
 739         dev->netdev_ops = &ipvlan_netdev_ops;
 740         dev->needs_free_netdev = true;
 741         dev->header_ops = &ipvlan_header_ops;
 742         dev->ethtool_ops = &ipvlan_ethtool_ops;
 743 }
 744 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
 745
 746 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
 747 {
 748         [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
 749         [IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
 750 };
 751
 752 static struct rtnl_link_ops ipvlan_link_ops = {
 753         .kind           = "ipvlan",
 754         .priv_size      = sizeof(struct ipvl_dev),
 755
 756         .setup          = ipvlan_link_setup,
 757         .newlink        = ipvlan_link_new,
 758         .dellink        = ipvlan_link_delete,
 759 };
 760
 761 int ipvlan_link_register(struct rtnl_link_ops *ops)
 762 {
 763         ops->get_size   = ipvlan_nl_getsize;
 764         ops->policy     = ipvlan_nl_policy;
 765         ops->validate   = ipvlan_nl_validate;
 766         ops->fill_info  = ipvlan_nl_fillinfo;
 767         ops->changelink = ipvlan_nl_changelink;
 768         ops->maxtype    = IFLA_IPVLAN_MAX;
 769         return rtnl_link_register(ops);
 770 }
 771 EXPORT_SYMBOL_GPL(ipvlan_link_register);
 772
 773 static int ipvlan_device_event(struct notifier_block *unused,
 774                                unsigned long event, void *ptr)
 775 {
 776         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 777         struct ipvl_dev *ipvlan, *next;
 778         struct ipvl_port *port;
 779         LIST_HEAD(lst_kill);
 780
 781         if (!netif_is_ipvlan_port(dev))
 782                 return NOTIFY_DONE;
 783
 784         port = ipvlan_port_get_rtnl(dev);
 785
 786         switch (event) {
 787         case NETDEV_CHANGE:
 788                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 789                         netif_stacked_transfer_operstate(ipvlan->phy_dev,
 790                                                          ipvlan->dev);
 791                 break;
 792
 793         case NETDEV_REGISTER: {
 794                 struct net *oldnet, *newnet = dev_net(dev);
 795                 struct ipvlan_netns *old_vnet;
 796
 797                 oldnet = read_pnet(&port->pnet);
 798                 if (net_eq(newnet, oldnet))
 799                         break;
 800
 801                 write_pnet(&port->pnet, newnet);
 802
 803                 old_vnet = net_generic(oldnet, ipvlan_netid);
 804                 if (!old_vnet->ipvl_nf_hook_refcnt)
 805                         break;
 806
 807                 ipvlan_register_nf_hook(newnet);
 808                 ipvlan_unregister_nf_hook(oldnet);
 809                 break;
 810         }
 811         case NETDEV_UNREGISTER:
 812                 if (dev->reg_state != NETREG_UNREGISTERING)
 813                         break;
 814
 815                 list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
 816                         ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
 817                                                             &lst_kill);
 818                 unregister_netdevice_many(&lst_kill);
 819                 break;
 820
 821         case NETDEV_FEAT_CHANGE:
 822                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 823                         ipvlan->dev->gso_max_size = dev->gso_max_size;
 824                         ipvlan->dev->gso_max_segs = dev->gso_max_segs;
 825                         netdev_update_features(ipvlan->dev);
 826                 }
 827                 break;
 828
 829         case NETDEV_CHANGEMTU:
 830                 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
 831                         ipvlan_adjust_mtu(ipvlan, dev);
 832                 break;
 833
 834         case NETDEV_CHANGEADDR:
 835                 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
 836                         ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
 837                         call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
 838                 }
 839                 break;
 840
 841         case NETDEV_PRE_TYPE_CHANGE:
 842                 /* Forbid underlying device to change its type. */
 843                 return NOTIFY_BAD;
 844         }
 845         return NOTIFY_DONE;
 846 }
 847
 848 /* the caller must held the addrs lock */
 849 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 850 {
 851         struct ipvl_addr *addr;
 852
 853         addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
 854         if (!addr)
 855                 return -ENOMEM;
 856
 857         addr->master = ipvlan;
 858         if (!is_v6) {
 859                 memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
 860                 addr->atype = IPVL_IPV4;
 861 #if IS_ENABLED(CONFIG_IPV6)
 862         } else {
 863                 memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
 864                 addr->atype = IPVL_IPV6;
 865 #endif
 866         }
 867
 868         list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
 869
 870         /* If the interface is not up, the address will be added to the hash
 871          * list by ipvlan_open.
 872          */
 873         if (netif_running(ipvlan->dev))
 874                 ipvlan_ht_addr_add(ipvlan, addr);
 875
 876         return 0;
 877 }
 878
 879 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
 880 {
 881         struct ipvl_addr *addr;
 882
 883         spin_lock_bh(&ipvlan->addrs_lock);
 884         addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
 885         if (!addr) {
 886                 spin_unlock_bh(&ipvlan->addrs_lock);
 887                 return;
 888         }
 889
 890         ipvlan_ht_addr_del(addr);
 891         list_del_rcu(&addr->anode);
 892         spin_unlock_bh(&ipvlan->addrs_lock);
 893         kfree_rcu(addr, rcu);
 894 }
 895
 896 static bool ipvlan_is_valid_dev(const struct net_device *dev)
 897 {
 898         struct ipvl_dev *ipvlan = netdev_priv(dev);
 899
 900         if (!netif_is_ipvlan(dev))
 901                 return false;
 902
 903         if (!ipvlan || !ipvlan->port)
 904                 return false;
 905
 906         return true;
 907 }
 908
 909 #if IS_ENABLED(CONFIG_IPV6)
 910 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 911 {
 912         int ret = -EINVAL;
 913
 914         spin_lock_bh(&ipvlan->addrs_lock);
 915         if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
 916                 netif_err(ipvlan, ifup, ipvlan->dev,
 917                           "Failed to add IPv6=%pI6c addr for %s intf\n",
 918                           ip6_addr, ipvlan->dev->name);
 919         else
 920                 ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
 921         spin_unlock_bh(&ipvlan->addrs_lock);
 922         return ret;
 923 }
 924
 925 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
 926 {
 927         return ipvlan_del_addr(ipvlan, ip6_addr, true);
 928 }
 929
 930 static int ipvlan_addr6_event(struct notifier_block *unused,
 931                               unsigned long event, void *ptr)
 932 {
 933         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
 934         struct net_device *dev = (struct net_device *)if6->idev->dev;
 935         struct ipvl_dev *ipvlan = netdev_priv(dev);
 936
 937         if (!ipvlan_is_valid_dev(dev))
 938                 return NOTIFY_DONE;
 939
 940         switch (event) {
 941         case NETDEV_UP:
 942                 if (ipvlan_add_addr6(ipvlan, &if6->addr))
 943                         return NOTIFY_BAD;
 944                 break;
 945
 946         case NETDEV_DOWN:
 947                 ipvlan_del_addr6(ipvlan, &if6->addr);
 948                 break;
 949         }
 950
 951         return NOTIFY_OK;
 952 }
 953
 954 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
 955                                         unsigned long event, void *ptr)
 956 {
 957         struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
 958         struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
 959         struct ipvl_dev *ipvlan = netdev_priv(dev);
 960
 961         if (!ipvlan_is_valid_dev(dev))
 962                 return NOTIFY_DONE;
 963
 964         switch (event) {
 965         case NETDEV_UP:
 966                 if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
 967                         NL_SET_ERR_MSG(i6vi->extack,
 968                                        "Address already assigned to an ipvlan device");
 969                         return notifier_from_errno(-EADDRINUSE);
 970                 }
 971                 break;
 972         }
 973
 974         return NOTIFY_OK;
 975 }
 976 #endif
 977
 978 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 979 {
 980         int ret = -EINVAL;
 981
 982         spin_lock_bh(&ipvlan->addrs_lock);
 983         if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
 984                 netif_err(ipvlan, ifup, ipvlan->dev,
 985                           "Failed to add IPv4=%pI4 on %s intf.\n",
 986                           ip4_addr, ipvlan->dev->name);
 987         else
 988                 ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
 989         spin_unlock_bh(&ipvlan->addrs_lock);
 990         return ret;
 991 }
 992
 993 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
 994 {
 995         return ipvlan_del_addr(ipvlan, ip4_addr, false);
 996 }
 997
 998 static int ipvlan_addr4_event(struct notifier_block *unused,
 999                               unsigned long event, void *ptr)
1000 {
1001         struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
1002         struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
1003         struct ipvl_dev *ipvlan = netdev_priv(dev);
1004         struct in_addr ip4_addr;
1005
1006         if (!ipvlan_is_valid_dev(dev))
1007                 return NOTIFY_DONE;
1008
1009         switch (event) {
1010         case NETDEV_UP:
1011                 ip4_addr.s_addr = if4->ifa_address;
1012                 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
1013                         return NOTIFY_BAD;
1014                 break;
1015
1016         case NETDEV_DOWN:
1017                 ip4_addr.s_addr = if4->ifa_address;
1018                 ipvlan_del_addr4(ipvlan, &ip4_addr);
1019                 break;
1020         }
1021
1022         return NOTIFY_OK;
1023 }
1024
1025 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
1026                                         unsigned long event, void *ptr)
1027 {
1028         struct in_validator_info *ivi = (struct in_validator_info *)ptr;
1029         struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
1030         struct ipvl_dev *ipvlan = netdev_priv(dev);
1031
1032         if (!ipvlan_is_valid_dev(dev))
1033                 return NOTIFY_DONE;
1034
1035         switch (event) {
1036         case NETDEV_UP:
1037                 if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
1038                         NL_SET_ERR_MSG(ivi->extack,
1039                                        "Address already assigned to an ipvlan device");
1040                         return notifier_from_errno(-EADDRINUSE);
1041                 }
1042                 break;
1043         }
1044
1045         return NOTIFY_OK;
1046 }
1047
1048 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1049         .notifier_call = ipvlan_addr4_event,
1050 };
1051
1052 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1053         .notifier_call = ipvlan_addr4_validator_event,
1054 };
1055
1056 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1057         .notifier_call = ipvlan_device_event,
1058 };
1059
1060 #if IS_ENABLED(CONFIG_IPV6)
1061 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1062         .notifier_call = ipvlan_addr6_event,
1063 };
1064
1065 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1066         .notifier_call = ipvlan_addr6_validator_event,
1067 };
1068 #endif
1069
1070 static void ipvlan_ns_exit(struct net *net)
1071 {
1072         struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
1073
1074         if (WARN_ON_ONCE(vnet->ipvl_nf_hook_refcnt)) {
1075                 vnet->ipvl_nf_hook_refcnt = 0;
1076                 nf_unregister_net_hooks(net, ipvl_nfops,
1077                                         ARRAY_SIZE(ipvl_nfops));
1078         }
1079 }
1080
1081 static struct pernet_operations ipvlan_net_ops = {
1082         .id = &ipvlan_netid,
1083         .size = sizeof(struct ipvlan_netns),
1084         .exit = ipvlan_ns_exit,
1085 };
1086
1087 static int __init ipvlan_init_module(void)
1088 {
1089         int err;
1090
1091         ipvlan_init_secret();
1092         register_netdevice_notifier(&ipvlan_notifier_block);
1093 #if IS_ENABLED(CONFIG_IPV6)
1094         register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1095         register_inet6addr_validator_notifier(
1096             &ipvlan_addr6_vtor_notifier_block);
1097 #endif
1098         register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1099         register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1100
1101         err = register_pernet_subsys(&ipvlan_net_ops);
1102         if (err < 0)
1103                 goto error;
1104
1105         err = ipvlan_link_register(&ipvlan_link_ops);
1106         if (err < 0) {
1107                 unregister_pernet_subsys(&ipvlan_net_ops);
1108                 goto error;
1109         }
1110
1111         return 0;
1112 error:
1113         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1114         unregister_inetaddr_validator_notifier(
1115             &ipvlan_addr4_vtor_notifier_block);
1116 #if IS_ENABLED(CONFIG_IPV6)
1117         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1118         unregister_inet6addr_validator_notifier(
1119             &ipvlan_addr6_vtor_notifier_block);
1120 #endif
1121         unregister_netdevice_notifier(&ipvlan_notifier_block);
1122         return err;
1123 }
1124
1125 static void __exit ipvlan_cleanup_module(void)
1126 {
1127         rtnl_link_unregister(&ipvlan_link_ops);
1128         unregister_pernet_subsys(&ipvlan_net_ops);
1129         unregister_netdevice_notifier(&ipvlan_notifier_block);
1130         unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1131         unregister_inetaddr_validator_notifier(
1132             &ipvlan_addr4_vtor_notifier_block);
1133 #if IS_ENABLED(CONFIG_IPV6)
1134         unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1135         unregister_inet6addr_validator_notifier(
1136             &ipvlan_addr6_vtor_notifier_block);
1137 #endif
1138 }
1139
1140 module_init(ipvlan_init_module);
1141 module_exit(ipvlan_cleanup_module);
1142
1143 MODULE_LICENSE("GPL");
1144 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1145 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1146 MODULE_ALIAS_RTNL_LINK("ipvlan");