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                 *pskb = skb;
 176                 eth = eth_hdr(skb);
 177                 src = macvlan_hash_lookup(port, eth->h_source);
 178                 if (!src)
 179                         /* frame comes from an external address */
 180                         macvlan_broadcast(skb, port, NULL,
 181                                           MACVLAN_MODE_PRIVATE |
 182                                           MACVLAN_MODE_VEPA    |
 183                                           MACVLAN_MODE_PASSTHRU|
 184                                           MACVLAN_MODE_BRIDGE);
 185                 else if (src->mode == MACVLAN_MODE_VEPA)
 186                         /* flood to everyone except source */
 187                         macvlan_broadcast(skb, port, src->dev,
 188                                           MACVLAN_MODE_VEPA |
 189                                           MACVLAN_MODE_BRIDGE);
 190                 else if (src->mode == MACVLAN_MODE_BRIDGE)
 191                         /*
 192                          * flood only to VEPA ports, bridge ports
 193                          * already saw the frame on the way out.
 194                          */
 195                         macvlan_broadcast(skb, port, src->dev,
 196                                           MACVLAN_MODE_VEPA);
 197                 else {
 198                         /* forward to original port. */
 199                         vlan = src;
 200                         ret = macvlan_broadcast_one(skb, vlan, eth, 0);
 201                         goto out;
 202                 }
 203
 204                 return RX_HANDLER_PASS;
 205         }
 206
 207         if (port->passthru)
 208                 vlan = list_first_or_null_rcu(&port->vlans,
 209                                               struct macvlan_dev, list);
 210         else
 211                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 212         if (vlan == NULL)
 213                 return RX_HANDLER_PASS;
 214
 215         dev = vlan->dev;
 216         if (unlikely(!(dev->flags & IFF_UP))) {
 217                 kfree_skb(skb);
 218                 return RX_HANDLER_CONSUMED;
 219         }
 220         len = skb->len + ETH_HLEN;
 221         skb = skb_share_check(skb, GFP_ATOMIC);
 222         if (!skb)
 223                 goto out;
 224
 225         *pskb = skb;
 226         skb->dev = dev;
 227         skb->pkt_type = PACKET_HOST;
 228
 229         ret = vlan->receive(skb);
 230
 231 out:
 232         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
 233         return RX_HANDLER_CONSUMED;
 234 }
 235
 236 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 237 {
 238         const struct macvlan_dev *vlan = netdev_priv(dev);
 239         const struct macvlan_port *port = vlan->port;
 240         const struct macvlan_dev *dest;
 241
 242         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 243                 const struct ethhdr *eth = (void *)skb->data;
 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->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 (dev->flags & IFF_UP) {
 397                 if (change & IFF_ALLMULTI)
 398                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 399         }
 400 }
 401
 402 static void macvlan_set_multicast_list(struct net_device *dev)
 403 {
 404         struct macvlan_dev *vlan = netdev_priv(dev);
 405
 406         dev_mc_sync(vlan->lowerdev, dev);
 407 }
 408
 409 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 410 {
 411         struct macvlan_dev *vlan = netdev_priv(dev);
 412
 413         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 414                 return -EINVAL;
 415         dev->mtu = new_mtu;
 416         return 0;
 417 }
 418
 419 /*
 420  * macvlan network devices have devices nesting below it and are a special
 421  * "super class" of normal network devices; split their locks off into a
 422  * separate class since they always nest.
 423  */
 424 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 425 static struct lock_class_key macvlan_netdev_addr_lock_key;
 426
 427 #define MACVLAN_FEATURES \
 428         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 429          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 430          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 431          NETIF_F_HW_VLAN_FILTER)
 432
 433 #define MACVLAN_STATE_MASK \
 434         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 435
 436 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 437                                           struct netdev_queue *txq,
 438                                           void *_unused)
 439 {
 440         lockdep_set_class(&txq->_xmit_lock,
 441                           &macvlan_netdev_xmit_lock_key);
 442 }
 443
 444 static void macvlan_set_lockdep_class(struct net_device *dev)
 445 {
 446         lockdep_set_class(&dev->addr_list_lock,
 447                           &macvlan_netdev_addr_lock_key);
 448         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 449 }
 450
 451 static int macvlan_init(struct net_device *dev)
 452 {
 453         struct macvlan_dev *vlan = netdev_priv(dev);
 454         const struct net_device *lowerdev = vlan->lowerdev;
 455
 456         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 457                                   (lowerdev->state & MACVLAN_STATE_MASK);
 458         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 459         dev->features           |= NETIF_F_LLTX;
 460         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 461         dev->gso_max_size       = lowerdev->gso_max_size;
 462         dev->iflink             = lowerdev->ifindex;
 463         dev->hard_header_len    = lowerdev->hard_header_len;
 464
 465         macvlan_set_lockdep_class(dev);
 466
 467         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
 468         if (!vlan->pcpu_stats)
 469                 return -ENOMEM;
 470
 471         return 0;
 472 }
 473
 474 static void macvlan_uninit(struct net_device *dev)
 475 {
 476         struct macvlan_dev *vlan = netdev_priv(dev);
 477         struct macvlan_port *port = vlan->port;
 478
 479         free_percpu(vlan->pcpu_stats);
 480
 481         port->count -= 1;
 482         if (!port->count)
 483                 macvlan_port_destroy(port->dev);
 484 }
 485
 486 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 487                                                          struct rtnl_link_stats64 *stats)
 488 {
 489         struct macvlan_dev *vlan = netdev_priv(dev);
 490
 491         if (vlan->pcpu_stats) {
 492                 struct macvlan_pcpu_stats *p;
 493                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 494                 u32 rx_errors = 0, tx_dropped = 0;
 495                 unsigned int start;
 496                 int i;
 497
 498                 for_each_possible_cpu(i) {
 499                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 500                         do {
 501                                 start = u64_stats_fetch_begin_bh(&p->syncp);
 502                                 rx_packets      = p->rx_packets;
 503                                 rx_bytes        = p->rx_bytes;
 504                                 rx_multicast    = p->rx_multicast;
 505                                 tx_packets      = p->tx_packets;
 506                                 tx_bytes        = p->tx_bytes;
 507                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
 508
 509                         stats->rx_packets       += rx_packets;
 510                         stats->rx_bytes         += rx_bytes;
 511                         stats->multicast        += rx_multicast;
 512                         stats->tx_packets       += tx_packets;
 513                         stats->tx_bytes         += tx_bytes;
 514                         /* rx_errors & tx_dropped are u32, updated
 515                          * without syncp protection.
 516                          */
 517                         rx_errors       += p->rx_errors;
 518                         tx_dropped      += p->tx_dropped;
 519                 }
 520                 stats->rx_errors        = rx_errors;
 521                 stats->rx_dropped       = rx_errors;
 522                 stats->tx_dropped       = tx_dropped;
 523         }
 524         return stats;
 525 }
 526
 527 static void macvlan_vlan_rx_add_vid(struct net_device *dev,
 528                                     unsigned short vid)
 529 {
 530         struct macvlan_dev *vlan = netdev_priv(dev);
 531         struct net_device *lowerdev = vlan->lowerdev;
 532         const struct net_device_ops *ops = lowerdev->netdev_ops;
 533
 534         if (ops->ndo_vlan_rx_add_vid)
 535                 ops->ndo_vlan_rx_add_vid(lowerdev, vid);
 536 }
 537
 538 static void macvlan_vlan_rx_kill_vid(struct net_device *dev,
 539                                      unsigned short vid)
 540 {
 541         struct macvlan_dev *vlan = netdev_priv(dev);
 542         struct net_device *lowerdev = vlan->lowerdev;
 543         const struct net_device_ops *ops = lowerdev->netdev_ops;
 544
 545         if (ops->ndo_vlan_rx_kill_vid)
 546                 ops->ndo_vlan_rx_kill_vid(lowerdev, vid);
 547 }
 548
 549 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 550                                         struct ethtool_drvinfo *drvinfo)
 551 {
 552         snprintf(drvinfo->driver, 32, "macvlan");
 553         snprintf(drvinfo->version, 32, "0.1");
 554 }
 555
 556 static int macvlan_ethtool_get_settings(struct net_device *dev,
 557                                         struct ethtool_cmd *cmd)
 558 {
 559         const struct macvlan_dev *vlan = netdev_priv(dev);
 560
 561         return __ethtool_get_settings(vlan->lowerdev, cmd);
 562 }
 563
 564 static const struct ethtool_ops macvlan_ethtool_ops = {
 565         .get_link               = ethtool_op_get_link,
 566         .get_settings           = macvlan_ethtool_get_settings,
 567         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 568 };
 569
 570 static const struct net_device_ops macvlan_netdev_ops = {
 571         .ndo_init               = macvlan_init,
 572         .ndo_uninit             = macvlan_uninit,
 573         .ndo_open               = macvlan_open,
 574         .ndo_stop               = macvlan_stop,
 575         .ndo_start_xmit         = macvlan_start_xmit,
 576         .ndo_change_mtu         = macvlan_change_mtu,
 577         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 578         .ndo_set_mac_address    = macvlan_set_mac_address,
 579         .ndo_set_rx_mode        = macvlan_set_multicast_list,
 580         .ndo_get_stats64        = macvlan_dev_get_stats64,
 581         .ndo_validate_addr      = eth_validate_addr,
 582         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
 583         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
 584 };
 585
 586 void macvlan_common_setup(struct net_device *dev)
 587 {
 588         ether_setup(dev);
 589
 590         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 591         dev->priv_flags        |= IFF_UNICAST_FLT;
 592         dev->netdev_ops         = &macvlan_netdev_ops;
 593         dev->destructor         = free_netdev;
 594         dev->header_ops         = &macvlan_hard_header_ops,
 595         dev->ethtool_ops        = &macvlan_ethtool_ops;
 596 }
 597 EXPORT_SYMBOL_GPL(macvlan_common_setup);
 598
 599 static void macvlan_setup(struct net_device *dev)
 600 {
 601         macvlan_common_setup(dev);
 602         dev->tx_queue_len       = 0;
 603 }
 604
 605 static int macvlan_port_create(struct net_device *dev)
 606 {
 607         struct macvlan_port *port;
 608         unsigned int i;
 609         int err;
 610
 611         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 612                 return -EINVAL;
 613
 614         port = kzalloc(sizeof(*port), GFP_KERNEL);
 615         if (port == NULL)
 616                 return -ENOMEM;
 617
 618         port->passthru = false;
 619         port->dev = dev;
 620         INIT_LIST_HEAD(&port->vlans);
 621         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 622                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 623
 624         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
 625         if (err)
 626                 kfree(port);
 627         else
 628                 dev->priv_flags |= IFF_MACVLAN_PORT;
 629         return err;
 630 }
 631
 632 static void macvlan_port_destroy(struct net_device *dev)
 633 {
 634         struct macvlan_port *port = macvlan_port_get(dev);
 635
 636         dev->priv_flags &= ~IFF_MACVLAN_PORT;
 637         netdev_rx_handler_unregister(dev);
 638         kfree_rcu(port, rcu);
 639 }
 640
 641 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 642 {
 643         if (tb[IFLA_ADDRESS]) {
 644                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 645                         return -EINVAL;
 646                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 647                         return -EADDRNOTAVAIL;
 648         }
 649
 650         if (data && data[IFLA_MACVLAN_MODE]) {
 651                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
 652                 case MACVLAN_MODE_PRIVATE:
 653                 case MACVLAN_MODE_VEPA:
 654                 case MACVLAN_MODE_BRIDGE:
 655                 case MACVLAN_MODE_PASSTHRU:
 656                         break;
 657                 default:
 658                         return -EINVAL;
 659                 }
 660         }
 661         return 0;
 662 }
 663
 664 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
 665                            struct nlattr *tb[], struct nlattr *data[],
 666                            int (*receive)(struct sk_buff *skb),
 667                            int (*forward)(struct net_device *dev,
 668                                           struct sk_buff *skb))
 669 {
 670         struct macvlan_dev *vlan = netdev_priv(dev);
 671         struct macvlan_port *port;
 672         struct net_device *lowerdev;
 673         int err;
 674
 675         if (!tb[IFLA_LINK])
 676                 return -EINVAL;
 677
 678         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 679         if (lowerdev == NULL)
 680                 return -ENODEV;
 681
 682         /* When creating macvlans on top of other macvlans - use
 683          * the real device as the lowerdev.
 684          */
 685         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
 686                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
 687                 lowerdev = lowervlan->lowerdev;
 688         }
 689
 690         if (!tb[IFLA_MTU])
 691                 dev->mtu = lowerdev->mtu;
 692         else if (dev->mtu > lowerdev->mtu)
 693                 return -EINVAL;
 694
 695         if (!tb[IFLA_ADDRESS])
 696                 random_ether_addr(dev->dev_addr);
 697
 698         if (!macvlan_port_exists(lowerdev)) {
 699                 err = macvlan_port_create(lowerdev);
 700                 if (err < 0)
 701                         return err;
 702         }
 703         port = macvlan_port_get(lowerdev);
 704
 705         /* Only 1 macvlan device can be created in passthru mode */
 706         if (port->passthru)
 707                 return -EINVAL;
 708
 709         vlan->lowerdev = lowerdev;
 710         vlan->dev      = dev;
 711         vlan->port     = port;
 712         vlan->receive  = receive;
 713         vlan->forward  = forward;
 714
 715         vlan->mode     = MACVLAN_MODE_VEPA;
 716         if (data && data[IFLA_MACVLAN_MODE])
 717                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 718
 719         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 720                 if (port->count)
 721                         return -EINVAL;
 722                 port->passthru = true;
 723                 memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
 724         }
 725
 726         port->count += 1;
 727         err = register_netdevice(dev);
 728         if (err < 0)
 729                 goto destroy_port;
 730
 731         list_add_tail_rcu(&vlan->list, &port->vlans);
 732         netif_stacked_transfer_operstate(lowerdev, dev);
 733
 734         return 0;
 735
 736 destroy_port:
 737         port->count -= 1;
 738         if (!port->count)
 739                 macvlan_port_destroy(lowerdev);
 740
 741         return err;
 742 }
 743 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
 744
 745 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
 746                            struct nlattr *tb[], struct nlattr *data[])
 747 {
 748         return macvlan_common_newlink(src_net, dev, tb, data,
 749                                       netif_rx,
 750                                       dev_forward_skb);
 751 }
 752
 753 void macvlan_dellink(struct net_device *dev, struct list_head *head)
 754 {
 755         struct macvlan_dev *vlan = netdev_priv(dev);
 756
 757         list_del_rcu(&vlan->list);
 758         unregister_netdevice_queue(dev, head);
 759 }
 760 EXPORT_SYMBOL_GPL(macvlan_dellink);
 761
 762 static int macvlan_changelink(struct net_device *dev,
 763                 struct nlattr *tb[], struct nlattr *data[])
 764 {
 765         struct macvlan_dev *vlan = netdev_priv(dev);
 766         if (data && data[IFLA_MACVLAN_MODE])
 767                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 768         return 0;
 769 }
 770
 771 static size_t macvlan_get_size(const struct net_device *dev)
 772 {
 773         return nla_total_size(4);
 774 }
 775
 776 static int macvlan_fill_info(struct sk_buff *skb,
 777                                 const struct net_device *dev)
 778 {
 779         struct macvlan_dev *vlan = netdev_priv(dev);
 780
 781         NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
 782         return 0;
 783
 784 nla_put_failure:
 785         return -EMSGSIZE;
 786 }
 787
 788 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
 789         [IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
 790 };
 791
 792 int macvlan_link_register(struct rtnl_link_ops *ops)
 793 {
 794         /* common fields */
 795         ops->priv_size          = sizeof(struct macvlan_dev);
 796         ops->validate           = macvlan_validate;
 797         ops->maxtype            = IFLA_MACVLAN_MAX;
 798         ops->policy             = macvlan_policy;
 799         ops->changelink         = macvlan_changelink;
 800         ops->get_size           = macvlan_get_size;
 801         ops->fill_info          = macvlan_fill_info;
 802
 803         return rtnl_link_register(ops);
 804 };
 805 EXPORT_SYMBOL_GPL(macvlan_link_register);
 806
 807 static struct rtnl_link_ops macvlan_link_ops = {
 808         .kind           = "macvlan",
 809         .setup          = macvlan_setup,
 810         .newlink        = macvlan_newlink,
 811         .dellink        = macvlan_dellink,
 812 };
 813
 814 static int macvlan_device_event(struct notifier_block *unused,
 815                                 unsigned long event, void *ptr)
 816 {
 817         struct net_device *dev = ptr;
 818         struct macvlan_dev *vlan, *next;
 819         struct macvlan_port *port;
 820         LIST_HEAD(list_kill);
 821
 822         if (!macvlan_port_exists(dev))
 823                 return NOTIFY_DONE;
 824
 825         port = macvlan_port_get(dev);
 826
 827         switch (event) {
 828         case NETDEV_CHANGE:
 829                 list_for_each_entry(vlan, &port->vlans, list)
 830                         netif_stacked_transfer_operstate(vlan->lowerdev,
 831                                                          vlan->dev);
 832                 break;
 833         case NETDEV_FEAT_CHANGE:
 834                 list_for_each_entry(vlan, &port->vlans, list) {
 835                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
 836                         vlan->dev->gso_max_size = dev->gso_max_size;
 837                         netdev_features_change(vlan->dev);
 838                 }
 839                 break;
 840         case NETDEV_UNREGISTER:
 841                 /* twiddle thumbs on netns device moves */
 842                 if (dev->reg_state != NETREG_UNREGISTERING)
 843                         break;
 844
 845                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
 846                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
 847                 unregister_netdevice_many(&list_kill);
 848                 list_del(&list_kill);
 849                 break;
 850         case NETDEV_PRE_TYPE_CHANGE:
 851                 /* Forbid underlaying device to change its type. */
 852                 return NOTIFY_BAD;
 853         }
 854         return NOTIFY_DONE;
 855 }
 856
 857 static struct notifier_block macvlan_notifier_block __read_mostly = {
 858         .notifier_call  = macvlan_device_event,
 859 };
 860
 861 static int __init macvlan_init_module(void)
 862 {
 863         int err;
 864
 865         register_netdevice_notifier(&macvlan_notifier_block);
 866
 867         err = macvlan_link_register(&macvlan_link_ops);
 868         if (err < 0)
 869                 goto err1;
 870         return 0;
 871 err1:
 872         unregister_netdevice_notifier(&macvlan_notifier_block);
 873         return err;
 874 }
 875
 876 static void __exit macvlan_cleanup_module(void)
 877 {
 878         rtnl_link_unregister(&macvlan_link_ops);
 879         unregister_netdevice_notifier(&macvlan_notifier_block);
 880 }
 881
 882 module_init(macvlan_init_module);
 883 module_exit(macvlan_cleanup_module);
 884
 885 MODULE_LICENSE("GPL");
 886 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 887 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
 888 MODULE_ALIAS_RTNL_LINK("macvlan");