drivers/net/macvlan.c

   1 /*
   2  * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public License as
   6  * published by the Free Software Foundation; either version 2 of
   7  * the License, or (at your option) any later version.
   8  *
   9  * The code this is based on carried the following copyright notice:
  10  * ---
  11  * (C) Copyright 2001-2006
  12  * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
  13  * Re-worked by Ben Greear <greearb@candelatech.com>
  14  * ---
  15  */
  16 #include <linux/kernel.h>
  17 #include <linux/types.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/errno.h>
  21 #include <linux/slab.h>
  22 #include <linux/string.h>
  23 #include <linux/rculist.h>
  24 #include <linux/notifier.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/etherdevice.h>
  27 #include <linux/ethtool.h>
  28 #include <linux/if_arp.h>
  29 #include <linux/if_link.h>
  30 #include <linux/if_macvlan.h>
  31 #include <net/rtnetlink.h>
  32 #include <net/xfrm.h>
  33
  34 #define MACVLAN_HASH_SIZE       (1 << BITS_PER_BYTE)
  35
  36 struct macvlan_port {
  37         struct net_device       *dev;
  38         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  39         struct list_head        vlans;
  40         struct rcu_head         rcu;
  41         bool                    passthru;
  42         int                     count;
  43 };
  44
  45 static void macvlan_port_destroy(struct net_device *dev);
  46
  47 #define macvlan_port_get_rcu(dev) \
  48         ((struct macvlan_port *) rcu_dereference(dev->rx_handler_data))
  49 #define macvlan_port_get(dev) ((struct macvlan_port *) dev->rx_handler_data)
  50 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
  51
  52 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
  53                                                const unsigned char *addr)
  54 {
  55         struct macvlan_dev *vlan;
  56         struct hlist_node *n;
  57
  58         hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
  59                 if (!compare_ether_addr_64bits(vlan->dev->dev_addr, addr))
  60                         return vlan;
  61         }
  62         return NULL;
  63 }
  64
  65 static void macvlan_hash_add(struct macvlan_dev *vlan)
  66 {
  67         struct macvlan_port *port = vlan->port;
  68         const unsigned char *addr = vlan->dev->dev_addr;
  69
  70         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
  71 }
  72
  73 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
  74 {
  75         hlist_del_rcu(&vlan->hlist);
  76         if (sync)
  77                 synchronize_rcu();
  78 }
  79
  80 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
  81                                         const unsigned char *addr)
  82 {
  83         macvlan_hash_del(vlan, true);
  84         /* Now that we are unhashed it is safe to change the device
  85          * address without confusing packet delivery.
  86          */
  87         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
  88         macvlan_hash_add(vlan);
  89 }
  90
  91 static int macvlan_addr_busy(const struct macvlan_port *port,
  92                                 const unsigned char *addr)
  93 {
  94         /* Test to see if the specified multicast address is
  95          * currently in use by the underlying device or
  96          * another macvlan.
  97          */
  98         if (!compare_ether_addr_64bits(port->dev->dev_addr, addr))
  99                 return 1;
 100
 101         if (macvlan_hash_lookup(port, addr))
 102                 return 1;
 103
 104         return 0;
 105 }
 106
 107
 108 static int macvlan_broadcast_one(struct sk_buff *skb,
 109                                  const struct macvlan_dev *vlan,
 110                                  const struct ethhdr *eth, bool local)
 111 {
 112         struct net_device *dev = vlan->dev;
 113         if (!skb)
 114                 return NET_RX_DROP;
 115
 116         if (local)
 117                 return vlan->forward(dev, skb);
 118
 119         skb->dev = dev;
 120         if (!compare_ether_addr_64bits(eth->h_dest,
 121                                        dev->broadcast))
 122                 skb->pkt_type = PACKET_BROADCAST;
 123         else
 124                 skb->pkt_type = PACKET_MULTICAST;
 125
 126         return vlan->receive(skb);
 127 }
 128
 129 static void macvlan_broadcast(struct sk_buff *skb,
 130                               const struct macvlan_port *port,
 131                               struct net_device *src,
 132                               enum macvlan_mode mode)
 133 {
 134         const struct ethhdr *eth = eth_hdr(skb);
 135         const struct macvlan_dev *vlan;
 136         struct hlist_node *n;
 137         struct sk_buff *nskb;
 138         unsigned int i;
 139         int err;
 140
 141         if (skb->protocol == htons(ETH_P_PAUSE))
 142                 return;
 143
 144         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 145                 hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
 146                         if (vlan->dev == src || !(vlan->mode & mode))
 147                                 continue;
 148
 149                         nskb = skb_clone(skb, GFP_ATOMIC);
 150                         err = macvlan_broadcast_one(nskb, vlan, eth,
 151                                          mode == MACVLAN_MODE_BRIDGE);
 152                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 153                                          err == NET_RX_SUCCESS, 1);
 154                 }
 155         }
 156 }
 157
 158 /* called under rcu_read_lock() from netif_receive_skb */
 159 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 160 {
 161         struct macvlan_port *port;
 162         struct sk_buff *skb = *pskb;
 163         const struct ethhdr *eth = eth_hdr(skb);
 164         const struct macvlan_dev *vlan;
 165         const struct macvlan_dev *src;
 166         struct net_device *dev;
 167         unsigned int len = 0;
 168         int ret = NET_RX_DROP;
 169
 170         port = macvlan_port_get_rcu(skb->dev);
 171         if (is_multicast_ether_addr(eth->h_dest)) {
 172                 skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
 173                 if (!skb)
 174                         return RX_HANDLER_CONSUMED;
 175                 eth = eth_hdr(skb);
 176                 src = macvlan_hash_lookup(port, eth->h_source);
 177                 if (!src)
 178                         /* frame comes from an external address */
 179                         macvlan_broadcast(skb, port, NULL,
 180                                           MACVLAN_MODE_PRIVATE |
 181                                           MACVLAN_MODE_VEPA    |
 182                                           MACVLAN_MODE_PASSTHRU|
 183                                           MACVLAN_MODE_BRIDGE);
 184                 else if (src->mode == MACVLAN_MODE_VEPA)
 185                         /* flood to everyone except source */
 186                         macvlan_broadcast(skb, port, src->dev,
 187                                           MACVLAN_MODE_VEPA |
 188                                           MACVLAN_MODE_BRIDGE);
 189                 else if (src->mode == MACVLAN_MODE_BRIDGE)
 190                         /*
 191                          * flood only to VEPA ports, bridge ports
 192                          * already saw the frame on the way out.
 193                          */
 194                         macvlan_broadcast(skb, port, src->dev,
 195                                           MACVLAN_MODE_VEPA);
 196                 else {
 197                         /* forward to original port. */
 198                         vlan = src;
 199                         ret = macvlan_broadcast_one(skb, vlan, eth, 0);
 200                         goto out;
 201                 }
 202
 203                 return RX_HANDLER_PASS;
 204         }
 205
 206         if (port->passthru)
 207                 vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
 208         else
 209                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 210         if (vlan == NULL)
 211                 return RX_HANDLER_PASS;
 212
 213         dev = vlan->dev;
 214         if (unlikely(!(dev->flags & IFF_UP))) {
 215                 kfree_skb(skb);
 216                 return RX_HANDLER_CONSUMED;
 217         }
 218         len = skb->len + ETH_HLEN;
 219         skb = skb_share_check(skb, GFP_ATOMIC);
 220         if (!skb)
 221                 goto out;
 222
 223         skb->dev = dev;
 224         skb->pkt_type = PACKET_HOST;
 225
 226         ret = vlan->receive(skb);
 227
 228 out:
 229         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
 230         return RX_HANDLER_CONSUMED;
 231 }
 232
 233 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 234 {
 235         const struct macvlan_dev *vlan = netdev_priv(dev);
 236         const struct macvlan_port *port = vlan->port;
 237         const struct macvlan_dev *dest;
 238         __u8 ip_summed = skb->ip_summed;
 239
 240         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 241                 const struct ethhdr *eth = (void *)skb->data;
 242                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 243
 244                 /* send to other bridge ports directly */
 245                 if (is_multicast_ether_addr(eth->h_dest)) {
 246                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 247                         goto xmit_world;
 248                 }
 249
 250                 dest = macvlan_hash_lookup(port, eth->h_dest);
 251                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 252                         /* send to lowerdev first for its network taps */
 253                         dev_forward_skb(vlan->lowerdev, skb);
 254
 255                         return NET_XMIT_SUCCESS;
 256                 }
 257         }
 258
 259 xmit_world:
 260         skb->ip_summed = ip_summed;
 261         skb->dev = vlan->lowerdev;
 262         return dev_queue_xmit(skb);
 263 }
 264
 265 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 266                                struct net_device *dev)
 267 {
 268         unsigned int len = skb->len;
 269         int ret;
 270         const struct macvlan_dev *vlan = netdev_priv(dev);
 271
 272         ret = macvlan_queue_xmit(skb, dev);
 273         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 274                 struct macvlan_pcpu_stats *pcpu_stats;
 275
 276                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 277                 u64_stats_update_begin(&pcpu_stats->syncp);
 278                 pcpu_stats->tx_packets++;
 279                 pcpu_stats->tx_bytes += len;
 280                 u64_stats_update_end(&pcpu_stats->syncp);
 281         } else {
 282                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 283         }
 284         return ret;
 285 }
 286 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
 287
 288 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 289                                unsigned short type, const void *daddr,
 290                                const void *saddr, unsigned len)
 291 {
 292         const struct macvlan_dev *vlan = netdev_priv(dev);
 293         struct net_device *lowerdev = vlan->lowerdev;
 294
 295         return dev_hard_header(skb, lowerdev, type, daddr,
 296                                saddr ? : dev->dev_addr, len);
 297 }
 298
 299 static const struct header_ops macvlan_hard_header_ops = {
 300         .create         = macvlan_hard_header,
 301         .rebuild        = eth_rebuild_header,
 302         .parse          = eth_header_parse,
 303         .cache          = eth_header_cache,
 304         .cache_update   = eth_header_cache_update,
 305 };
 306
 307 static int macvlan_open(struct net_device *dev)
 308 {
 309         struct macvlan_dev *vlan = netdev_priv(dev);
 310         struct net_device *lowerdev = vlan->lowerdev;
 311         int err;
 312
 313         if (vlan->port->passthru) {
 314                 dev_set_promiscuity(lowerdev, 1);
 315                 goto hash_add;
 316         }
 317
 318         err = -EBUSY;
 319         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 320                 goto out;
 321
 322         err = dev_uc_add(lowerdev, dev->dev_addr);
 323         if (err < 0)
 324                 goto out;
 325         if (dev->flags & IFF_ALLMULTI) {
 326                 err = dev_set_allmulti(lowerdev, 1);
 327                 if (err < 0)
 328                         goto del_unicast;
 329         }
 330
 331 hash_add:
 332         macvlan_hash_add(vlan);
 333         return 0;
 334
 335 del_unicast:
 336         dev_uc_del(lowerdev, dev->dev_addr);
 337 out:
 338         return err;
 339 }
 340
 341 static int macvlan_stop(struct net_device *dev)
 342 {
 343         struct macvlan_dev *vlan = netdev_priv(dev);
 344         struct net_device *lowerdev = vlan->lowerdev;
 345
 346         if (vlan->port->passthru) {
 347                 dev_set_promiscuity(lowerdev, -1);
 348                 goto hash_del;
 349         }
 350
 351         dev_mc_unsync(lowerdev, dev);
 352         if (dev->flags & IFF_ALLMULTI)
 353                 dev_set_allmulti(lowerdev, -1);
 354
 355         dev_uc_del(lowerdev, dev->dev_addr);
 356
 357 hash_del:
 358         macvlan_hash_del(vlan, !dev->dismantle);
 359         return 0;
 360 }
 361
 362 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 363 {
 364         struct macvlan_dev *vlan = netdev_priv(dev);
 365         struct net_device *lowerdev = vlan->lowerdev;
 366         struct sockaddr *addr = p;
 367         int err;
 368
 369         if (!is_valid_ether_addr(addr->sa_data))
 370                 return -EADDRNOTAVAIL;
 371
 372         if (!(dev->flags & IFF_UP)) {
 373                 /* Just copy in the new address */
 374                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 375         } else {
 376                 /* Rehash and update the device filters */
 377                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
 378                         return -EBUSY;
 379
 380                 err = dev_uc_add(lowerdev, addr->sa_data);
 381                 if (err)
 382                         return err;
 383
 384                 dev_uc_del(lowerdev, dev->dev_addr);
 385
 386                 macvlan_hash_change_addr(vlan, addr->sa_data);
 387         }
 388         return 0;
 389 }
 390
 391 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 392 {
 393         struct macvlan_dev *vlan = netdev_priv(dev);
 394         struct net_device *lowerdev = vlan->lowerdev;
 395
 396         if (change & IFF_ALLMULTI)
 397                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 398 }
 399
 400 static void macvlan_set_multicast_list(struct net_device *dev)
 401 {
 402         struct macvlan_dev *vlan = netdev_priv(dev);
 403
 404         dev_mc_sync(vlan->lowerdev, dev);
 405 }
 406
 407 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 408 {
 409         struct macvlan_dev *vlan = netdev_priv(dev);
 410
 411         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 412                 return -EINVAL;
 413         dev->mtu = new_mtu;
 414         return 0;
 415 }
 416
 417 /*
 418  * macvlan network devices have devices nesting below it and are a special
 419  * "super class" of normal network devices; split their locks off into a
 420  * separate class since they always nest.
 421  */
 422 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 423 static struct lock_class_key macvlan_netdev_addr_lock_key;
 424
 425 #define MACVLAN_FEATURES \
 426         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 427          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 428          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 429          NETIF_F_HW_VLAN_FILTER)
 430
 431 #define MACVLAN_STATE_MASK \
 432         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 433
 434 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 435                                           struct netdev_queue *txq,
 436                                           void *_unused)
 437 {
 438         lockdep_set_class(&txq->_xmit_lock,
 439                           &macvlan_netdev_xmit_lock_key);
 440 }
 441
 442 static void macvlan_set_lockdep_class(struct net_device *dev)
 443 {
 444         lockdep_set_class(&dev->addr_list_lock,
 445                           &macvlan_netdev_addr_lock_key);
 446         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 447 }
 448
 449 static int macvlan_init(struct net_device *dev)
 450 {
 451         struct macvlan_dev *vlan = netdev_priv(dev);
 452         const struct net_device *lowerdev = vlan->lowerdev;
 453
 454         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 455                                   (lowerdev->state & MACVLAN_STATE_MASK);
 456         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 457         dev->features           |= NETIF_F_LLTX;
 458         dev->gso_max_size       = lowerdev->gso_max_size;
 459         dev->iflink             = lowerdev->ifindex;
 460         dev->hard_header_len    = lowerdev->hard_header_len;
 461
 462         macvlan_set_lockdep_class(dev);
 463
 464         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
 465         if (!vlan->pcpu_stats)
 466                 return -ENOMEM;
 467
 468         return 0;
 469 }
 470
 471 static void macvlan_uninit(struct net_device *dev)
 472 {
 473         struct macvlan_dev *vlan = netdev_priv(dev);
 474         struct macvlan_port *port = vlan->port;
 475
 476         free_percpu(vlan->pcpu_stats);
 477
 478         port->count -= 1;
 479         if (!port->count)
 480                 macvlan_port_destroy(port->dev);
 481 }
 482
 483 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 484                                                          struct rtnl_link_stats64 *stats)
 485 {
 486         struct macvlan_dev *vlan = netdev_priv(dev);
 487
 488         if (vlan->pcpu_stats) {
 489                 struct macvlan_pcpu_stats *p;
 490                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 491                 u32 rx_errors = 0, tx_dropped = 0;
 492                 unsigned int start;
 493                 int i;
 494
 495                 for_each_possible_cpu(i) {
 496                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 497                         do {
 498                                 start = u64_stats_fetch_begin_bh(&p->syncp);
 499                                 rx_packets      = p->rx_packets;
 500                                 rx_bytes        = p->rx_bytes;
 501                                 rx_multicast    = p->rx_multicast;
 502                                 tx_packets      = p->tx_packets;
 503                                 tx_bytes        = p->tx_bytes;
 504                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
 505
 506                         stats->rx_packets       += rx_packets;
 507                         stats->rx_bytes         += rx_bytes;
 508                         stats->multicast        += rx_multicast;
 509                         stats->tx_packets       += tx_packets;
 510                         stats->tx_bytes         += tx_bytes;
 511                         /* rx_errors & tx_dropped are u32, updated
 512                          * without syncp protection.
 513                          */
 514                         rx_errors       += p->rx_errors;
 515                         tx_dropped      += p->tx_dropped;
 516                 }
 517                 stats->rx_errors        = rx_errors;
 518                 stats->rx_dropped       = rx_errors;
 519                 stats->tx_dropped       = tx_dropped;
 520         }
 521         return stats;
 522 }
 523
 524 static void macvlan_vlan_rx_add_vid(struct net_device *dev,
 525                                     unsigned short vid)
 526 {
 527         struct macvlan_dev *vlan = netdev_priv(dev);
 528         struct net_device *lowerdev = vlan->lowerdev;
 529         const struct net_device_ops *ops = lowerdev->netdev_ops;
 530
 531         if (ops->ndo_vlan_rx_add_vid)
 532                 ops->ndo_vlan_rx_add_vid(lowerdev, vid);
 533 }
 534
 535 static void macvlan_vlan_rx_kill_vid(struct net_device *dev,
 536                                      unsigned short vid)
 537 {
 538         struct macvlan_dev *vlan = netdev_priv(dev);
 539         struct net_device *lowerdev = vlan->lowerdev;
 540         const struct net_device_ops *ops = lowerdev->netdev_ops;
 541
 542         if (ops->ndo_vlan_rx_kill_vid)
 543                 ops->ndo_vlan_rx_kill_vid(lowerdev, vid);
 544 }
 545
 546 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 547                                         struct ethtool_drvinfo *drvinfo)
 548 {
 549         snprintf(drvinfo->driver, 32, "macvlan");
 550         snprintf(drvinfo->version, 32, "0.1");
 551 }
 552
 553 static int macvlan_ethtool_get_settings(struct net_device *dev,
 554                                         struct ethtool_cmd *cmd)
 555 {
 556         const struct macvlan_dev *vlan = netdev_priv(dev);
 557
 558         return __ethtool_get_settings(vlan->lowerdev, cmd);
 559 }
 560
 561 static const struct ethtool_ops macvlan_ethtool_ops = {
 562         .get_link               = ethtool_op_get_link,
 563         .get_settings           = macvlan_ethtool_get_settings,
 564         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 565 };
 566
 567 static const struct net_device_ops macvlan_netdev_ops = {
 568         .ndo_init               = macvlan_init,
 569         .ndo_uninit             = macvlan_uninit,
 570         .ndo_open               = macvlan_open,
 571         .ndo_stop               = macvlan_stop,
 572         .ndo_start_xmit         = macvlan_start_xmit,
 573         .ndo_change_mtu         = macvlan_change_mtu,
 574         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 575         .ndo_set_mac_address    = macvlan_set_mac_address,
 576         .ndo_set_rx_mode        = macvlan_set_multicast_list,
 577         .ndo_get_stats64        = macvlan_dev_get_stats64,
 578         .ndo_validate_addr      = eth_validate_addr,
 579         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
 580         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
 581 };
 582
 583 void macvlan_common_setup(struct net_device *dev)
 584 {
 585         ether_setup(dev);
 586
 587         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 588         dev->netdev_ops         = &macvlan_netdev_ops;
 589         dev->destructor         = free_netdev;
 590         dev->header_ops         = &macvlan_hard_header_ops,
 591         dev->ethtool_ops        = &macvlan_ethtool_ops;
 592 }
 593 EXPORT_SYMBOL_GPL(macvlan_common_setup);
 594
 595 static void macvlan_setup(struct net_device *dev)
 596 {
 597         macvlan_common_setup(dev);
 598         dev->tx_queue_len       = 0;
 599 }
 600
 601 static int macvlan_port_create(struct net_device *dev)
 602 {
 603         struct macvlan_port *port;
 604         unsigned int i;
 605         int err;
 606
 607         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 608                 return -EINVAL;
 609
 610         port = kzalloc(sizeof(*port), GFP_KERNEL);
 611         if (port == NULL)
 612                 return -ENOMEM;
 613
 614         port->passthru = false;
 615         port->dev = dev;
 616         INIT_LIST_HEAD(&port->vlans);
 617         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 618                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 619
 620         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
 621         if (err)
 622                 kfree(port);
 623         else
 624                 dev->priv_flags |= IFF_MACVLAN_PORT;
 625         return err;
 626 }
 627
 628 static void macvlan_port_destroy(struct net_device *dev)
 629 {
 630         struct macvlan_port *port = macvlan_port_get(dev);
 631
 632         dev->priv_flags &= ~IFF_MACVLAN_PORT;
 633         netdev_rx_handler_unregister(dev);
 634         kfree_rcu(port, rcu);
 635 }
 636
 637 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 638 {
 639         if (tb[IFLA_ADDRESS]) {
 640                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 641                         return -EINVAL;
 642                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 643                         return -EADDRNOTAVAIL;
 644         }
 645
 646         if (data && data[IFLA_MACVLAN_MODE]) {
 647                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
 648                 case MACVLAN_MODE_PRIVATE:
 649                 case MACVLAN_MODE_VEPA:
 650                 case MACVLAN_MODE_BRIDGE:
 651                 case MACVLAN_MODE_PASSTHRU:
 652                         break;
 653                 default:
 654                         return -EINVAL;
 655                 }
 656         }
 657         return 0;
 658 }
 659
 660 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
 661                            struct nlattr *tb[], struct nlattr *data[],
 662                            int (*receive)(struct sk_buff *skb),
 663                            int (*forward)(struct net_device *dev,
 664                                           struct sk_buff *skb))
 665 {
 666         struct macvlan_dev *vlan = netdev_priv(dev);
 667         struct macvlan_port *port;
 668         struct net_device *lowerdev;
 669         int err;
 670
 671         if (!tb[IFLA_LINK])
 672                 return -EINVAL;
 673
 674         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 675         if (lowerdev == NULL)
 676                 return -ENODEV;
 677
 678         /* When creating macvlans on top of other macvlans - use
 679          * the real device as the lowerdev.
 680          */
 681         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
 682                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
 683                 lowerdev = lowervlan->lowerdev;
 684         }
 685
 686         if (!tb[IFLA_MTU])
 687                 dev->mtu = lowerdev->mtu;
 688         else if (dev->mtu > lowerdev->mtu)
 689                 return -EINVAL;
 690
 691         if (!tb[IFLA_ADDRESS])
 692                 random_ether_addr(dev->dev_addr);
 693
 694         if (!macvlan_port_exists(lowerdev)) {
 695                 err = macvlan_port_create(lowerdev);
 696                 if (err < 0)
 697                         return err;
 698         }
 699         port = macvlan_port_get(lowerdev);
 700
 701         /* Only 1 macvlan device can be created in passthru mode */
 702         if (port->passthru)
 703                 return -EINVAL;
 704
 705         vlan->lowerdev = lowerdev;
 706         vlan->dev      = dev;
 707         vlan->port     = port;
 708         vlan->receive  = receive;
 709         vlan->forward  = forward;
 710
 711         vlan->mode     = MACVLAN_MODE_VEPA;
 712         if (data && data[IFLA_MACVLAN_MODE])
 713                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 714
 715         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 716                 if (port->count)
 717                         return -EINVAL;
 718                 port->passthru = true;
 719                 memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
 720         }
 721
 722         port->count += 1;
 723         err = register_netdevice(dev);
 724         if (err < 0)
 725                 goto destroy_port;
 726
 727         list_add_tail(&vlan->list, &port->vlans);
 728         netif_stacked_transfer_operstate(lowerdev, dev);
 729
 730         return 0;
 731
 732 destroy_port:
 733         port->count -= 1;
 734         if (!port->count)
 735                 macvlan_port_destroy(lowerdev);
 736
 737         return err;
 738 }
 739 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
 740
 741 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
 742                            struct nlattr *tb[], struct nlattr *data[])
 743 {
 744         return macvlan_common_newlink(src_net, dev, tb, data,
 745                                       netif_rx,
 746                                       dev_forward_skb);
 747 }
 748
 749 void macvlan_dellink(struct net_device *dev, struct list_head *head)
 750 {
 751         struct macvlan_dev *vlan = netdev_priv(dev);
 752
 753         list_del(&vlan->list);
 754         unregister_netdevice_queue(dev, head);
 755 }
 756 EXPORT_SYMBOL_GPL(macvlan_dellink);
 757
 758 static int macvlan_changelink(struct net_device *dev,
 759                 struct nlattr *tb[], struct nlattr *data[])
 760 {
 761         struct macvlan_dev *vlan = netdev_priv(dev);
 762         if (data && data[IFLA_MACVLAN_MODE])
 763                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 764         return 0;
 765 }
 766
 767 static size_t macvlan_get_size(const struct net_device *dev)
 768 {
 769         return nla_total_size(4);
 770 }
 771
 772 static int macvlan_fill_info(struct sk_buff *skb,
 773                                 const struct net_device *dev)
 774 {
 775         struct macvlan_dev *vlan = netdev_priv(dev);
 776
 777         NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
 778         return 0;
 779
 780 nla_put_failure:
 781         return -EMSGSIZE;
 782 }
 783
 784 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
 785         [IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
 786 };
 787
 788 int macvlan_link_register(struct rtnl_link_ops *ops)
 789 {
 790         /* common fields */
 791         ops->priv_size          = sizeof(struct macvlan_dev);
 792         ops->validate           = macvlan_validate;
 793         ops->maxtype            = IFLA_MACVLAN_MAX;
 794         ops->policy             = macvlan_policy;
 795         ops->changelink         = macvlan_changelink;
 796         ops->get_size           = macvlan_get_size;
 797         ops->fill_info          = macvlan_fill_info;
 798
 799         return rtnl_link_register(ops);
 800 };
 801 EXPORT_SYMBOL_GPL(macvlan_link_register);
 802
 803 static struct rtnl_link_ops macvlan_link_ops = {
 804         .kind           = "macvlan",
 805         .setup          = macvlan_setup,
 806         .newlink        = macvlan_newlink,
 807         .dellink        = macvlan_dellink,
 808 };
 809
 810 static int macvlan_device_event(struct notifier_block *unused,
 811                                 unsigned long event, void *ptr)
 812 {
 813         struct net_device *dev = ptr;
 814         struct macvlan_dev *vlan, *next;
 815         struct macvlan_port *port;
 816         LIST_HEAD(list_kill);
 817
 818         if (!macvlan_port_exists(dev))
 819                 return NOTIFY_DONE;
 820
 821         port = macvlan_port_get(dev);
 822
 823         switch (event) {
 824         case NETDEV_CHANGE:
 825                 list_for_each_entry(vlan, &port->vlans, list)
 826                         netif_stacked_transfer_operstate(vlan->lowerdev,
 827                                                          vlan->dev);
 828                 break;
 829         case NETDEV_FEAT_CHANGE:
 830                 list_for_each_entry(vlan, &port->vlans, list) {
 831                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
 832                         vlan->dev->gso_max_size = dev->gso_max_size;
 833                         netdev_features_change(vlan->dev);
 834                 }
 835                 break;
 836         case NETDEV_UNREGISTER:
 837                 /* twiddle thumbs on netns device moves */
 838                 if (dev->reg_state != NETREG_UNREGISTERING)
 839                         break;
 840
 841                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
 842                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
 843                 unregister_netdevice_many(&list_kill);
 844                 list_del(&list_kill);
 845                 break;
 846         case NETDEV_PRE_TYPE_CHANGE:
 847                 /* Forbid underlaying device to change its type. */
 848                 return NOTIFY_BAD;
 849         }
 850         return NOTIFY_DONE;
 851 }
 852
 853 static struct notifier_block macvlan_notifier_block __read_mostly = {
 854         .notifier_call  = macvlan_device_event,
 855 };
 856
 857 static int __init macvlan_init_module(void)
 858 {
 859         int err;
 860
 861         register_netdevice_notifier(&macvlan_notifier_block);
 862
 863         err = macvlan_link_register(&macvlan_link_ops);
 864         if (err < 0)
 865                 goto err1;
 866         return 0;
 867 err1:
 868         unregister_netdevice_notifier(&macvlan_notifier_block);
 869         return err;
 870 }
 871
 872 static void __exit macvlan_cleanup_module(void)
 873 {
 874         rtnl_link_unregister(&macvlan_link_ops);
 875         unregister_netdevice_notifier(&macvlan_notifier_block);
 876 }
 877
 878 module_init(macvlan_init_module);
 879 module_exit(macvlan_cleanup_module);
 880
 881 MODULE_LICENSE("GPL");
 882 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 883 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
 884 MODULE_ALIAS_RTNL_LINK("macvlan");