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_or_null_rcu(&port->vlans,
 208                                               struct macvlan_dev, list);
 209         else
 210                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 211         if (vlan == NULL)
 212                 return RX_HANDLER_PASS;
 213
 214         dev = vlan->dev;
 215         if (unlikely(!(dev->flags & IFF_UP))) {
 216                 kfree_skb(skb);
 217                 return RX_HANDLER_CONSUMED;
 218         }
 219         len = skb->len + ETH_HLEN;
 220         skb = skb_share_check(skb, GFP_ATOMIC);
 221         if (!skb)
 222                 goto out;
 223
 224         skb->dev = dev;
 225         skb->pkt_type = PACKET_HOST;
 226
 227         ret = vlan->receive(skb);
 228
 229 out:
 230         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
 231         return RX_HANDLER_CONSUMED;
 232 }
 233
 234 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 235 {
 236         const struct macvlan_dev *vlan = netdev_priv(dev);
 237         const struct macvlan_port *port = vlan->port;
 238         const struct macvlan_dev *dest;
 239         __u8 ip_summed = skb->ip_summed;
 240
 241         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 242                 const struct ethhdr *eth = (void *)skb->data;
 243                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 244
 245                 /* send to other bridge ports directly */
 246                 if (is_multicast_ether_addr(eth->h_dest)) {
 247                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 248                         goto xmit_world;
 249                 }
 250
 251                 dest = macvlan_hash_lookup(port, eth->h_dest);
 252                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 253                         /* send to lowerdev first for its network taps */
 254                         dev_forward_skb(vlan->lowerdev, skb);
 255
 256                         return NET_XMIT_SUCCESS;
 257                 }
 258         }
 259
 260 xmit_world:
 261         skb->ip_summed = ip_summed;
 262         skb->dev = vlan->lowerdev;
 263         return dev_queue_xmit(skb);
 264 }
 265
 266 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 267                                struct net_device *dev)
 268 {
 269         unsigned int len = skb->len;
 270         int ret;
 271         const struct macvlan_dev *vlan = netdev_priv(dev);
 272
 273         ret = macvlan_queue_xmit(skb, dev);
 274         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 275                 struct macvlan_pcpu_stats *pcpu_stats;
 276
 277                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 278                 u64_stats_update_begin(&pcpu_stats->syncp);
 279                 pcpu_stats->tx_packets++;
 280                 pcpu_stats->tx_bytes += len;
 281                 u64_stats_update_end(&pcpu_stats->syncp);
 282         } else {
 283                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 284         }
 285         return ret;
 286 }
 287 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
 288
 289 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 290                                unsigned short type, const void *daddr,
 291                                const void *saddr, unsigned len)
 292 {
 293         const struct macvlan_dev *vlan = netdev_priv(dev);
 294         struct net_device *lowerdev = vlan->lowerdev;
 295
 296         return dev_hard_header(skb, lowerdev, type, daddr,
 297                                saddr ? : dev->dev_addr, len);
 298 }
 299
 300 static const struct header_ops macvlan_hard_header_ops = {
 301         .create         = macvlan_hard_header,
 302         .rebuild        = eth_rebuild_header,
 303         .parse          = eth_header_parse,
 304         .cache          = eth_header_cache,
 305         .cache_update   = eth_header_cache_update,
 306 };
 307
 308 static int macvlan_open(struct net_device *dev)
 309 {
 310         struct macvlan_dev *vlan = netdev_priv(dev);
 311         struct net_device *lowerdev = vlan->lowerdev;
 312         int err;
 313
 314         if (vlan->port->passthru) {
 315                 dev_set_promiscuity(lowerdev, 1);
 316                 goto hash_add;
 317         }
 318
 319         err = -EBUSY;
 320         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 321                 goto out;
 322
 323         err = dev_uc_add(lowerdev, dev->dev_addr);
 324         if (err < 0)
 325                 goto out;
 326         if (dev->flags & IFF_ALLMULTI) {
 327                 err = dev_set_allmulti(lowerdev, 1);
 328                 if (err < 0)
 329                         goto del_unicast;
 330         }
 331
 332 hash_add:
 333         macvlan_hash_add(vlan);
 334         return 0;
 335
 336 del_unicast:
 337         dev_uc_del(lowerdev, dev->dev_addr);
 338 out:
 339         return err;
 340 }
 341
 342 static int macvlan_stop(struct net_device *dev)
 343 {
 344         struct macvlan_dev *vlan = netdev_priv(dev);
 345         struct net_device *lowerdev = vlan->lowerdev;
 346
 347         if (vlan->port->passthru) {
 348                 dev_set_promiscuity(lowerdev, -1);
 349                 goto hash_del;
 350         }
 351
 352         dev_mc_unsync(lowerdev, dev);
 353         if (dev->flags & IFF_ALLMULTI)
 354                 dev_set_allmulti(lowerdev, -1);
 355
 356         dev_uc_del(lowerdev, dev->dev_addr);
 357
 358 hash_del:
 359         macvlan_hash_del(vlan, !dev->dismantle);
 360         return 0;
 361 }
 362
 363 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 364 {
 365         struct macvlan_dev *vlan = netdev_priv(dev);
 366         struct net_device *lowerdev = vlan->lowerdev;
 367         struct sockaddr *addr = p;
 368         int err;
 369
 370         if (!is_valid_ether_addr(addr->sa_data))
 371                 return -EADDRNOTAVAIL;
 372
 373         if (!(dev->flags & IFF_UP)) {
 374                 /* Just copy in the new address */
 375                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 376         } else {
 377                 /* Rehash and update the device filters */
 378                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
 379                         return -EBUSY;
 380
 381                 err = dev_uc_add(lowerdev, addr->sa_data);
 382                 if (err)
 383                         return err;
 384
 385                 dev_uc_del(lowerdev, dev->dev_addr);
 386
 387                 macvlan_hash_change_addr(vlan, addr->sa_data);
 388         }
 389         return 0;
 390 }
 391
 392 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 393 {
 394         struct macvlan_dev *vlan = netdev_priv(dev);
 395         struct net_device *lowerdev = vlan->lowerdev;
 396
 397         if (change & IFF_ALLMULTI)
 398                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 399 }
 400
 401 static void macvlan_set_multicast_list(struct net_device *dev)
 402 {
 403         struct macvlan_dev *vlan = netdev_priv(dev);
 404
 405         dev_mc_sync(vlan->lowerdev, dev);
 406 }
 407
 408 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 409 {
 410         struct macvlan_dev *vlan = netdev_priv(dev);
 411
 412         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 413                 return -EINVAL;
 414         dev->mtu = new_mtu;
 415         return 0;
 416 }
 417
 418 /*
 419  * macvlan network devices have devices nesting below it and are a special
 420  * "super class" of normal network devices; split their locks off into a
 421  * separate class since they always nest.
 422  */
 423 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 424 static struct lock_class_key macvlan_netdev_addr_lock_key;
 425
 426 #define MACVLAN_FEATURES \
 427         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 428          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 429          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 430          NETIF_F_HW_VLAN_FILTER)
 431
 432 #define MACVLAN_STATE_MASK \
 433         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 434
 435 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 436                                           struct netdev_queue *txq,
 437                                           void *_unused)
 438 {
 439         lockdep_set_class(&txq->_xmit_lock,
 440                           &macvlan_netdev_xmit_lock_key);
 441 }
 442
 443 static void macvlan_set_lockdep_class(struct net_device *dev)
 444 {
 445         lockdep_set_class(&dev->addr_list_lock,
 446                           &macvlan_netdev_addr_lock_key);
 447         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 448 }
 449
 450 static int macvlan_init(struct net_device *dev)
 451 {
 452         struct macvlan_dev *vlan = netdev_priv(dev);
 453         const struct net_device *lowerdev = vlan->lowerdev;
 454
 455         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 456                                   (lowerdev->state & MACVLAN_STATE_MASK);
 457         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 458         dev->features           |= NETIF_F_LLTX;
 459         dev->gso_max_size       = lowerdev->gso_max_size;
 460         dev->iflink             = lowerdev->ifindex;
 461         dev->hard_header_len    = lowerdev->hard_header_len;
 462
 463         macvlan_set_lockdep_class(dev);
 464
 465         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
 466         if (!vlan->pcpu_stats)
 467                 return -ENOMEM;
 468
 469         return 0;
 470 }
 471
 472 static void macvlan_uninit(struct net_device *dev)
 473 {
 474         struct macvlan_dev *vlan = netdev_priv(dev);
 475         struct macvlan_port *port = vlan->port;
 476
 477         free_percpu(vlan->pcpu_stats);
 478
 479         port->count -= 1;
 480         if (!port->count)
 481                 macvlan_port_destroy(port->dev);
 482 }
 483
 484 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 485                                                          struct rtnl_link_stats64 *stats)
 486 {
 487         struct macvlan_dev *vlan = netdev_priv(dev);
 488
 489         if (vlan->pcpu_stats) {
 490                 struct macvlan_pcpu_stats *p;
 491                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 492                 u32 rx_errors = 0, tx_dropped = 0;
 493                 unsigned int start;
 494                 int i;
 495
 496                 for_each_possible_cpu(i) {
 497                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 498                         do {
 499                                 start = u64_stats_fetch_begin_bh(&p->syncp);
 500                                 rx_packets      = p->rx_packets;
 501                                 rx_bytes        = p->rx_bytes;
 502                                 rx_multicast    = p->rx_multicast;
 503                                 tx_packets      = p->tx_packets;
 504                                 tx_bytes        = p->tx_bytes;
 505                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
 506
 507                         stats->rx_packets       += rx_packets;
 508                         stats->rx_bytes         += rx_bytes;
 509                         stats->multicast        += rx_multicast;
 510                         stats->tx_packets       += tx_packets;
 511                         stats->tx_bytes         += tx_bytes;
 512                         /* rx_errors & tx_dropped are u32, updated
 513                          * without syncp protection.
 514                          */
 515                         rx_errors       += p->rx_errors;
 516                         tx_dropped      += p->tx_dropped;
 517                 }
 518                 stats->rx_errors        = rx_errors;
 519                 stats->rx_dropped       = rx_errors;
 520                 stats->tx_dropped       = tx_dropped;
 521         }
 522         return stats;
 523 }
 524
 525 static void macvlan_vlan_rx_add_vid(struct net_device *dev,
 526                                     unsigned short vid)
 527 {
 528         struct macvlan_dev *vlan = netdev_priv(dev);
 529         struct net_device *lowerdev = vlan->lowerdev;
 530         const struct net_device_ops *ops = lowerdev->netdev_ops;
 531
 532         if (ops->ndo_vlan_rx_add_vid)
 533                 ops->ndo_vlan_rx_add_vid(lowerdev, vid);
 534 }
 535
 536 static void macvlan_vlan_rx_kill_vid(struct net_device *dev,
 537                                      unsigned short vid)
 538 {
 539         struct macvlan_dev *vlan = netdev_priv(dev);
 540         struct net_device *lowerdev = vlan->lowerdev;
 541         const struct net_device_ops *ops = lowerdev->netdev_ops;
 542
 543         if (ops->ndo_vlan_rx_kill_vid)
 544                 ops->ndo_vlan_rx_kill_vid(lowerdev, vid);
 545 }
 546
 547 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 548                                         struct ethtool_drvinfo *drvinfo)
 549 {
 550         snprintf(drvinfo->driver, 32, "macvlan");
 551         snprintf(drvinfo->version, 32, "0.1");
 552 }
 553
 554 static int macvlan_ethtool_get_settings(struct net_device *dev,
 555                                         struct ethtool_cmd *cmd)
 556 {
 557         const struct macvlan_dev *vlan = netdev_priv(dev);
 558
 559         return __ethtool_get_settings(vlan->lowerdev, cmd);
 560 }
 561
 562 static const struct ethtool_ops macvlan_ethtool_ops = {
 563         .get_link               = ethtool_op_get_link,
 564         .get_settings           = macvlan_ethtool_get_settings,
 565         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 566 };
 567
 568 static const struct net_device_ops macvlan_netdev_ops = {
 569         .ndo_init               = macvlan_init,
 570         .ndo_uninit             = macvlan_uninit,
 571         .ndo_open               = macvlan_open,
 572         .ndo_stop               = macvlan_stop,
 573         .ndo_start_xmit         = macvlan_start_xmit,
 574         .ndo_change_mtu         = macvlan_change_mtu,
 575         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 576         .ndo_set_mac_address    = macvlan_set_mac_address,
 577         .ndo_set_rx_mode        = macvlan_set_multicast_list,
 578         .ndo_get_stats64        = macvlan_dev_get_stats64,
 579         .ndo_validate_addr      = eth_validate_addr,
 580         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
 581         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
 582 };
 583
 584 void macvlan_common_setup(struct net_device *dev)
 585 {
 586         ether_setup(dev);
 587
 588         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 589         dev->priv_flags        |= IFF_UNICAST_FLT;
 590         dev->netdev_ops         = &macvlan_netdev_ops;
 591         dev->destructor         = free_netdev;
 592         dev->header_ops         = &macvlan_hard_header_ops,
 593         dev->ethtool_ops        = &macvlan_ethtool_ops;
 594 }
 595 EXPORT_SYMBOL_GPL(macvlan_common_setup);
 596
 597 static void macvlan_setup(struct net_device *dev)
 598 {
 599         macvlan_common_setup(dev);
 600         dev->tx_queue_len       = 0;
 601 }
 602
 603 static int macvlan_port_create(struct net_device *dev)
 604 {
 605         struct macvlan_port *port;
 606         unsigned int i;
 607         int err;
 608
 609         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 610                 return -EINVAL;
 611
 612         port = kzalloc(sizeof(*port), GFP_KERNEL);
 613         if (port == NULL)
 614                 return -ENOMEM;
 615
 616         port->passthru = false;
 617         port->dev = dev;
 618         INIT_LIST_HEAD(&port->vlans);
 619         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 620                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 621
 622         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
 623         if (err)
 624                 kfree(port);
 625         else
 626                 dev->priv_flags |= IFF_MACVLAN_PORT;
 627         return err;
 628 }
 629
 630 static void macvlan_port_destroy(struct net_device *dev)
 631 {
 632         struct macvlan_port *port = macvlan_port_get(dev);
 633
 634         dev->priv_flags &= ~IFF_MACVLAN_PORT;
 635         netdev_rx_handler_unregister(dev);
 636         kfree_rcu(port, rcu);
 637 }
 638
 639 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 640 {
 641         if (tb[IFLA_ADDRESS]) {
 642                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 643                         return -EINVAL;
 644                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 645                         return -EADDRNOTAVAIL;
 646         }
 647
 648         if (data && data[IFLA_MACVLAN_MODE]) {
 649                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
 650                 case MACVLAN_MODE_PRIVATE:
 651                 case MACVLAN_MODE_VEPA:
 652                 case MACVLAN_MODE_BRIDGE:
 653                 case MACVLAN_MODE_PASSTHRU:
 654                         break;
 655                 default:
 656                         return -EINVAL;
 657                 }
 658         }
 659         return 0;
 660 }
 661
 662 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
 663                            struct nlattr *tb[], struct nlattr *data[],
 664                            int (*receive)(struct sk_buff *skb),
 665                            int (*forward)(struct net_device *dev,
 666                                           struct sk_buff *skb))
 667 {
 668         struct macvlan_dev *vlan = netdev_priv(dev);
 669         struct macvlan_port *port;
 670         struct net_device *lowerdev;
 671         int err;
 672
 673         if (!tb[IFLA_LINK])
 674                 return -EINVAL;
 675
 676         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 677         if (lowerdev == NULL)
 678                 return -ENODEV;
 679
 680         /* When creating macvlans on top of other macvlans - use
 681          * the real device as the lowerdev.
 682          */
 683         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
 684                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
 685                 lowerdev = lowervlan->lowerdev;
 686         }
 687
 688         if (!tb[IFLA_MTU])
 689                 dev->mtu = lowerdev->mtu;
 690         else if (dev->mtu > lowerdev->mtu)
 691                 return -EINVAL;
 692
 693         if (!tb[IFLA_ADDRESS])
 694                 random_ether_addr(dev->dev_addr);
 695
 696         if (!macvlan_port_exists(lowerdev)) {
 697                 err = macvlan_port_create(lowerdev);
 698                 if (err < 0)
 699                         return err;
 700         }
 701         port = macvlan_port_get(lowerdev);
 702
 703         /* Only 1 macvlan device can be created in passthru mode */
 704         if (port->passthru)
 705                 return -EINVAL;
 706
 707         vlan->lowerdev = lowerdev;
 708         vlan->dev      = dev;
 709         vlan->port     = port;
 710         vlan->receive  = receive;
 711         vlan->forward  = forward;
 712
 713         vlan->mode     = MACVLAN_MODE_VEPA;
 714         if (data && data[IFLA_MACVLAN_MODE])
 715                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 716
 717         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 718                 if (port->count)
 719                         return -EINVAL;
 720                 port->passthru = true;
 721                 memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
 722         }
 723
 724         port->count += 1;
 725         err = register_netdevice(dev);
 726         if (err < 0)
 727                 goto destroy_port;
 728
 729         list_add_tail_rcu(&vlan->list, &port->vlans);
 730         netif_stacked_transfer_operstate(lowerdev, dev);
 731
 732         return 0;
 733
 734 destroy_port:
 735         port->count -= 1;
 736         if (!port->count)
 737                 macvlan_port_destroy(lowerdev);
 738
 739         return err;
 740 }
 741 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
 742
 743 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
 744                            struct nlattr *tb[], struct nlattr *data[])
 745 {
 746         return macvlan_common_newlink(src_net, dev, tb, data,
 747                                       netif_rx,
 748                                       dev_forward_skb);
 749 }
 750
 751 void macvlan_dellink(struct net_device *dev, struct list_head *head)
 752 {
 753         struct macvlan_dev *vlan = netdev_priv(dev);
 754
 755         list_del_rcu(&vlan->list);
 756         unregister_netdevice_queue(dev, head);
 757 }
 758 EXPORT_SYMBOL_GPL(macvlan_dellink);
 759
 760 static int macvlan_changelink(struct net_device *dev,
 761                 struct nlattr *tb[], struct nlattr *data[])
 762 {
 763         struct macvlan_dev *vlan = netdev_priv(dev);
 764         if (data && data[IFLA_MACVLAN_MODE])
 765                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 766         return 0;
 767 }
 768
 769 static size_t macvlan_get_size(const struct net_device *dev)
 770 {
 771         return nla_total_size(4);
 772 }
 773
 774 static int macvlan_fill_info(struct sk_buff *skb,
 775                                 const struct net_device *dev)
 776 {
 777         struct macvlan_dev *vlan = netdev_priv(dev);
 778
 779         NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
 780         return 0;
 781
 782 nla_put_failure:
 783         return -EMSGSIZE;
 784 }
 785
 786 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
 787         [IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
 788 };
 789
 790 int macvlan_link_register(struct rtnl_link_ops *ops)
 791 {
 792         /* common fields */
 793         ops->priv_size          = sizeof(struct macvlan_dev);
 794         ops->validate           = macvlan_validate;
 795         ops->maxtype            = IFLA_MACVLAN_MAX;
 796         ops->policy             = macvlan_policy;
 797         ops->changelink         = macvlan_changelink;
 798         ops->get_size           = macvlan_get_size;
 799         ops->fill_info          = macvlan_fill_info;
 800
 801         return rtnl_link_register(ops);
 802 };
 803 EXPORT_SYMBOL_GPL(macvlan_link_register);
 804
 805 static struct rtnl_link_ops macvlan_link_ops = {
 806         .kind           = "macvlan",
 807         .setup          = macvlan_setup,
 808         .newlink        = macvlan_newlink,
 809         .dellink        = macvlan_dellink,
 810 };
 811
 812 static int macvlan_device_event(struct notifier_block *unused,
 813                                 unsigned long event, void *ptr)
 814 {
 815         struct net_device *dev = ptr;
 816         struct macvlan_dev *vlan, *next;
 817         struct macvlan_port *port;
 818         LIST_HEAD(list_kill);
 819
 820         if (!macvlan_port_exists(dev))
 821                 return NOTIFY_DONE;
 822
 823         port = macvlan_port_get(dev);
 824
 825         switch (event) {
 826         case NETDEV_CHANGE:
 827                 list_for_each_entry(vlan, &port->vlans, list)
 828                         netif_stacked_transfer_operstate(vlan->lowerdev,
 829                                                          vlan->dev);
 830                 break;
 831         case NETDEV_FEAT_CHANGE:
 832                 list_for_each_entry(vlan, &port->vlans, list) {
 833                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
 834                         vlan->dev->gso_max_size = dev->gso_max_size;
 835                         netdev_features_change(vlan->dev);
 836                 }
 837                 break;
 838         case NETDEV_UNREGISTER:
 839                 /* twiddle thumbs on netns device moves */
 840                 if (dev->reg_state != NETREG_UNREGISTERING)
 841                         break;
 842
 843                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
 844                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
 845                 unregister_netdevice_many(&list_kill);
 846                 list_del(&list_kill);
 847                 break;
 848         case NETDEV_PRE_TYPE_CHANGE:
 849                 /* Forbid underlaying device to change its type. */
 850                 return NOTIFY_BAD;
 851         }
 852         return NOTIFY_DONE;
 853 }
 854
 855 static struct notifier_block macvlan_notifier_block __read_mostly = {
 856         .notifier_call  = macvlan_device_event,
 857 };
 858
 859 static int __init macvlan_init_module(void)
 860 {
 861         int err;
 862
 863         register_netdevice_notifier(&macvlan_notifier_block);
 864
 865         err = macvlan_link_register(&macvlan_link_ops);
 866         if (err < 0)
 867                 goto err1;
 868         return 0;
 869 err1:
 870         unregister_netdevice_notifier(&macvlan_notifier_block);
 871         return err;
 872 }
 873
 874 static void __exit macvlan_cleanup_module(void)
 875 {
 876         rtnl_link_unregister(&macvlan_link_ops);
 877         unregister_netdevice_notifier(&macvlan_notifier_block);
 878 }
 879
 880 module_init(macvlan_init_module);
 881 module_exit(macvlan_cleanup_module);
 882
 883 MODULE_LICENSE("GPL");
 884 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 885 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
 886 MODULE_ALIAS_RTNL_LINK("macvlan");