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_vlan.h>
  30 #include <linux/if_link.h>
  31 #include <linux/if_macvlan.h>
  32 #include <linux/hash.h>
  33 #include <net/rtnetlink.h>
  34 #include <net/xfrm.h>
  35
  36 #define MACVLAN_HASH_SIZE       (1 << BITS_PER_BYTE)
  37
  38 struct macvlan_port {
  39         struct net_device       *dev;
  40         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  41         struct list_head        vlans;
  42         struct rcu_head         rcu;
  43         bool                    passthru;
  44         int                     count;
  45 };
  46
  47 static void macvlan_port_destroy(struct net_device *dev);
  48
  49 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
  50 {
  51         return rcu_dereference(dev->rx_handler_data);
  52 }
  53
  54 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
  55 {
  56         return rtnl_dereference(dev->rx_handler_data);
  57 }
  58
  59 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
  60
  61 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
  62                                                const unsigned char *addr)
  63 {
  64         struct macvlan_dev *vlan;
  65
  66         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) {
  67                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
  68                         return vlan;
  69         }
  70         return NULL;
  71 }
  72
  73 static void macvlan_hash_add(struct macvlan_dev *vlan)
  74 {
  75         struct macvlan_port *port = vlan->port;
  76         const unsigned char *addr = vlan->dev->dev_addr;
  77
  78         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
  79 }
  80
  81 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
  82 {
  83         hlist_del_rcu(&vlan->hlist);
  84         if (sync)
  85                 synchronize_rcu();
  86 }
  87
  88 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
  89                                         const unsigned char *addr)
  90 {
  91         macvlan_hash_del(vlan, true);
  92         /* Now that we are unhashed it is safe to change the device
  93          * address without confusing packet delivery.
  94          */
  95         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
  96         macvlan_hash_add(vlan);
  97 }
  98
  99 static int macvlan_addr_busy(const struct macvlan_port *port,
 100                                 const unsigned char *addr)
 101 {
 102         /* Test to see if the specified multicast address is
 103          * currently in use by the underlying device or
 104          * another macvlan.
 105          */
 106         if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
 107                 return 1;
 108
 109         if (macvlan_hash_lookup(port, addr))
 110                 return 1;
 111
 112         return 0;
 113 }
 114
 115
 116 static int macvlan_broadcast_one(struct sk_buff *skb,
 117                                  const struct macvlan_dev *vlan,
 118                                  const struct ethhdr *eth, bool local)
 119 {
 120         struct net_device *dev = vlan->dev;
 121         if (!skb)
 122                 return NET_RX_DROP;
 123
 124         if (local)
 125                 return vlan->forward(dev, skb);
 126
 127         skb->dev = dev;
 128         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 129                 skb->pkt_type = PACKET_BROADCAST;
 130         else
 131                 skb->pkt_type = PACKET_MULTICAST;
 132
 133         return vlan->receive(skb);
 134 }
 135
 136 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 137 {
 138         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 139 }
 140
 141
 142 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 143                             const unsigned char *addr)
 144 {
 145         u32 val = __get_unaligned_cpu32(addr + 2);
 146
 147         val ^= macvlan_hash_mix(vlan);
 148         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 149 }
 150
 151 static void macvlan_broadcast(struct sk_buff *skb,
 152                               const struct macvlan_port *port,
 153                               struct net_device *src,
 154                               enum macvlan_mode mode)
 155 {
 156         const struct ethhdr *eth = eth_hdr(skb);
 157         const struct macvlan_dev *vlan;
 158         struct sk_buff *nskb;
 159         unsigned int i;
 160         int err;
 161         unsigned int hash;
 162
 163         if (skb->protocol == htons(ETH_P_PAUSE))
 164                 return;
 165
 166         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 167                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 168                         if (vlan->dev == src || !(vlan->mode & mode))
 169                                 continue;
 170
 171                         hash = mc_hash(vlan, eth->h_dest);
 172                         if (!test_bit(hash, vlan->mc_filter))
 173                                 continue;
 174                         nskb = skb_clone(skb, GFP_ATOMIC);
 175                         err = macvlan_broadcast_one(nskb, vlan, eth,
 176                                          mode == MACVLAN_MODE_BRIDGE);
 177                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 178                                          err == NET_RX_SUCCESS, 1);
 179                 }
 180         }
 181 }
 182
 183 /* called under rcu_read_lock() from netif_receive_skb */
 184 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 185 {
 186         struct macvlan_port *port;
 187         struct sk_buff *skb = *pskb;
 188         const struct ethhdr *eth = eth_hdr(skb);
 189         const struct macvlan_dev *vlan;
 190         const struct macvlan_dev *src;
 191         struct net_device *dev;
 192         unsigned int len = 0;
 193         int ret = NET_RX_DROP;
 194
 195         port = macvlan_port_get_rcu(skb->dev);
 196         if (is_multicast_ether_addr(eth->h_dest)) {
 197                 skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
 198                 if (!skb)
 199                         return RX_HANDLER_CONSUMED;
 200                 eth = eth_hdr(skb);
 201                 src = macvlan_hash_lookup(port, eth->h_source);
 202                 if (!src)
 203                         /* frame comes from an external address */
 204                         macvlan_broadcast(skb, port, NULL,
 205                                           MACVLAN_MODE_PRIVATE |
 206                                           MACVLAN_MODE_VEPA    |
 207                                           MACVLAN_MODE_PASSTHRU|
 208                                           MACVLAN_MODE_BRIDGE);
 209                 else if (src->mode == MACVLAN_MODE_VEPA)
 210                         /* flood to everyone except source */
 211                         macvlan_broadcast(skb, port, src->dev,
 212                                           MACVLAN_MODE_VEPA |
 213                                           MACVLAN_MODE_BRIDGE);
 214                 else if (src->mode == MACVLAN_MODE_BRIDGE)
 215                         /*
 216                          * flood only to VEPA ports, bridge ports
 217                          * already saw the frame on the way out.
 218                          */
 219                         macvlan_broadcast(skb, port, src->dev,
 220                                           MACVLAN_MODE_VEPA);
 221                 else {
 222                         /* forward to original port. */
 223                         vlan = src;
 224                         ret = macvlan_broadcast_one(skb, vlan, eth, 0);
 225                         goto out;
 226                 }
 227
 228                 return RX_HANDLER_PASS;
 229         }
 230
 231         if (port->passthru)
 232                 vlan = list_first_or_null_rcu(&port->vlans,
 233                                               struct macvlan_dev, list);
 234         else
 235                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 236         if (vlan == NULL)
 237                 return RX_HANDLER_PASS;
 238
 239         dev = vlan->dev;
 240         if (unlikely(!(dev->flags & IFF_UP))) {
 241                 kfree_skb(skb);
 242                 return RX_HANDLER_CONSUMED;
 243         }
 244         len = skb->len + ETH_HLEN;
 245         skb = skb_share_check(skb, GFP_ATOMIC);
 246         if (!skb)
 247                 goto out;
 248
 249         skb->dev = dev;
 250         skb->pkt_type = PACKET_HOST;
 251
 252         ret = vlan->receive(skb);
 253
 254 out:
 255         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
 256         return RX_HANDLER_CONSUMED;
 257 }
 258
 259 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 260 {
 261         const struct macvlan_dev *vlan = netdev_priv(dev);
 262         const struct macvlan_port *port = vlan->port;
 263         const struct macvlan_dev *dest;
 264
 265         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 266                 const struct ethhdr *eth = (void *)skb->data;
 267
 268                 /* send to other bridge ports directly */
 269                 if (is_multicast_ether_addr(eth->h_dest)) {
 270                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 271                         goto xmit_world;
 272                 }
 273
 274                 dest = macvlan_hash_lookup(port, eth->h_dest);
 275                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 276                         /* send to lowerdev first for its network taps */
 277                         dev_forward_skb(vlan->lowerdev, skb);
 278
 279                         return NET_XMIT_SUCCESS;
 280                 }
 281         }
 282
 283 xmit_world:
 284         skb->dev = vlan->lowerdev;
 285         return dev_queue_xmit(skb);
 286 }
 287
 288 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 289                                struct net_device *dev)
 290 {
 291         unsigned int len = skb->len;
 292         int ret;
 293         const struct macvlan_dev *vlan = netdev_priv(dev);
 294
 295         ret = macvlan_queue_xmit(skb, dev);
 296         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 297                 struct macvlan_pcpu_stats *pcpu_stats;
 298
 299                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 300                 u64_stats_update_begin(&pcpu_stats->syncp);
 301                 pcpu_stats->tx_packets++;
 302                 pcpu_stats->tx_bytes += len;
 303                 u64_stats_update_end(&pcpu_stats->syncp);
 304         } else {
 305                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 306         }
 307         return ret;
 308 }
 309 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
 310
 311 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 312                                unsigned short type, const void *daddr,
 313                                const void *saddr, unsigned len)
 314 {
 315         const struct macvlan_dev *vlan = netdev_priv(dev);
 316         struct net_device *lowerdev = vlan->lowerdev;
 317
 318         return dev_hard_header(skb, lowerdev, type, daddr,
 319                                saddr ? : dev->dev_addr, len);
 320 }
 321
 322 static const struct header_ops macvlan_hard_header_ops = {
 323         .create         = macvlan_hard_header,
 324         .rebuild        = eth_rebuild_header,
 325         .parse          = eth_header_parse,
 326         .cache          = eth_header_cache,
 327         .cache_update   = eth_header_cache_update,
 328 };
 329
 330 static int macvlan_open(struct net_device *dev)
 331 {
 332         struct macvlan_dev *vlan = netdev_priv(dev);
 333         struct net_device *lowerdev = vlan->lowerdev;
 334         int err;
 335
 336         if (vlan->port->passthru) {
 337                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 338                         dev_set_promiscuity(lowerdev, 1);
 339                 goto hash_add;
 340         }
 341
 342         err = -EBUSY;
 343         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 344                 goto out;
 345
 346         err = dev_uc_add(lowerdev, dev->dev_addr);
 347         if (err < 0)
 348                 goto out;
 349         if (dev->flags & IFF_ALLMULTI) {
 350                 err = dev_set_allmulti(lowerdev, 1);
 351                 if (err < 0)
 352                         goto del_unicast;
 353         }
 354
 355 hash_add:
 356         macvlan_hash_add(vlan);
 357         return 0;
 358
 359 del_unicast:
 360         dev_uc_del(lowerdev, dev->dev_addr);
 361 out:
 362         return err;
 363 }
 364
 365 static int macvlan_stop(struct net_device *dev)
 366 {
 367         struct macvlan_dev *vlan = netdev_priv(dev);
 368         struct net_device *lowerdev = vlan->lowerdev;
 369
 370         dev_uc_unsync(lowerdev, dev);
 371         dev_mc_unsync(lowerdev, dev);
 372
 373         if (vlan->port->passthru) {
 374                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 375                         dev_set_promiscuity(lowerdev, -1);
 376                 goto hash_del;
 377         }
 378
 379         if (dev->flags & IFF_ALLMULTI)
 380                 dev_set_allmulti(lowerdev, -1);
 381
 382         dev_uc_del(lowerdev, dev->dev_addr);
 383
 384 hash_del:
 385         macvlan_hash_del(vlan, !dev->dismantle);
 386         return 0;
 387 }
 388
 389 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 390 {
 391         struct macvlan_dev *vlan = netdev_priv(dev);
 392         struct net_device *lowerdev = vlan->lowerdev;
 393         struct sockaddr *addr = p;
 394         int err;
 395
 396         if (!is_valid_ether_addr(addr->sa_data))
 397                 return -EADDRNOTAVAIL;
 398
 399         if (!(dev->flags & IFF_UP)) {
 400                 /* Just copy in the new address */
 401                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 402         } else {
 403                 /* Rehash and update the device filters */
 404                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
 405                         return -EBUSY;
 406
 407                 err = dev_uc_add(lowerdev, addr->sa_data);
 408                 if (err)
 409                         return err;
 410
 411                 dev_uc_del(lowerdev, dev->dev_addr);
 412
 413                 macvlan_hash_change_addr(vlan, addr->sa_data);
 414         }
 415         return 0;
 416 }
 417
 418 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 419 {
 420         struct macvlan_dev *vlan = netdev_priv(dev);
 421         struct net_device *lowerdev = vlan->lowerdev;
 422
 423         if (dev->flags & IFF_UP) {
 424                 if (change & IFF_ALLMULTI)
 425                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 426         }
 427 }
 428
 429 static void macvlan_set_mac_lists(struct net_device *dev)
 430 {
 431         struct macvlan_dev *vlan = netdev_priv(dev);
 432
 433         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 434                 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
 435         } else {
 436                 struct netdev_hw_addr *ha;
 437                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 438
 439                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 440                 netdev_for_each_mc_addr(ha, dev) {
 441                         __set_bit(mc_hash(vlan, ha->addr), filter);
 442                 }
 443
 444                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 445
 446                 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 447         }
 448         dev_uc_sync(vlan->lowerdev, dev);
 449         dev_mc_sync(vlan->lowerdev, dev);
 450 }
 451
 452 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 453 {
 454         struct macvlan_dev *vlan = netdev_priv(dev);
 455
 456         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 457                 return -EINVAL;
 458         dev->mtu = new_mtu;
 459         return 0;
 460 }
 461
 462 /*
 463  * macvlan network devices have devices nesting below it and are a special
 464  * "super class" of normal network devices; split their locks off into a
 465  * separate class since they always nest.
 466  */
 467 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 468 static struct lock_class_key macvlan_netdev_addr_lock_key;
 469
 470 #define MACVLAN_FEATURES \
 471         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 472          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 473          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 474          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 475
 476 #define MACVLAN_STATE_MASK \
 477         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 478
 479 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 480                                           struct netdev_queue *txq,
 481                                           void *_unused)
 482 {
 483         lockdep_set_class(&txq->_xmit_lock,
 484                           &macvlan_netdev_xmit_lock_key);
 485 }
 486
 487 static void macvlan_set_lockdep_class(struct net_device *dev)
 488 {
 489         lockdep_set_class(&dev->addr_list_lock,
 490                           &macvlan_netdev_addr_lock_key);
 491         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 492 }
 493
 494 static int macvlan_init(struct net_device *dev)
 495 {
 496         struct macvlan_dev *vlan = netdev_priv(dev);
 497         const struct net_device *lowerdev = vlan->lowerdev;
 498
 499         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 500                                   (lowerdev->state & MACVLAN_STATE_MASK);
 501         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 502         dev->features           |= NETIF_F_LLTX;
 503         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 504         dev->gso_max_size       = lowerdev->gso_max_size;
 505         dev->iflink             = lowerdev->ifindex;
 506         dev->hard_header_len    = lowerdev->hard_header_len;
 507
 508         macvlan_set_lockdep_class(dev);
 509
 510         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
 511         if (!vlan->pcpu_stats)
 512                 return -ENOMEM;
 513
 514         return 0;
 515 }
 516
 517 static void macvlan_uninit(struct net_device *dev)
 518 {
 519         struct macvlan_dev *vlan = netdev_priv(dev);
 520         struct macvlan_port *port = vlan->port;
 521
 522         free_percpu(vlan->pcpu_stats);
 523
 524         port->count -= 1;
 525         if (!port->count)
 526                 macvlan_port_destroy(port->dev);
 527 }
 528
 529 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 530                                                          struct rtnl_link_stats64 *stats)
 531 {
 532         struct macvlan_dev *vlan = netdev_priv(dev);
 533
 534         if (vlan->pcpu_stats) {
 535                 struct macvlan_pcpu_stats *p;
 536                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 537                 u32 rx_errors = 0, tx_dropped = 0;
 538                 unsigned int start;
 539                 int i;
 540
 541                 for_each_possible_cpu(i) {
 542                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 543                         do {
 544                                 start = u64_stats_fetch_begin_bh(&p->syncp);
 545                                 rx_packets      = p->rx_packets;
 546                                 rx_bytes        = p->rx_bytes;
 547                                 rx_multicast    = p->rx_multicast;
 548                                 tx_packets      = p->tx_packets;
 549                                 tx_bytes        = p->tx_bytes;
 550                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
 551
 552                         stats->rx_packets       += rx_packets;
 553                         stats->rx_bytes         += rx_bytes;
 554                         stats->multicast        += rx_multicast;
 555                         stats->tx_packets       += tx_packets;
 556                         stats->tx_bytes         += tx_bytes;
 557                         /* rx_errors & tx_dropped are u32, updated
 558                          * without syncp protection.
 559                          */
 560                         rx_errors       += p->rx_errors;
 561                         tx_dropped      += p->tx_dropped;
 562                 }
 563                 stats->rx_errors        = rx_errors;
 564                 stats->rx_dropped       = rx_errors;
 565                 stats->tx_dropped       = tx_dropped;
 566         }
 567         return stats;
 568 }
 569
 570 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 571                                    __be16 proto, u16 vid)
 572 {
 573         struct macvlan_dev *vlan = netdev_priv(dev);
 574         struct net_device *lowerdev = vlan->lowerdev;
 575
 576         return vlan_vid_add(lowerdev, proto, vid);
 577 }
 578
 579 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 580                                     __be16 proto, u16 vid)
 581 {
 582         struct macvlan_dev *vlan = netdev_priv(dev);
 583         struct net_device *lowerdev = vlan->lowerdev;
 584
 585         vlan_vid_del(lowerdev, proto, vid);
 586         return 0;
 587 }
 588
 589 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 590                            struct net_device *dev,
 591                            const unsigned char *addr,
 592                            u16 flags)
 593 {
 594         struct macvlan_dev *vlan = netdev_priv(dev);
 595         int err = -EINVAL;
 596
 597         if (!vlan->port->passthru)
 598                 return -EOPNOTSUPP;
 599
 600         if (is_unicast_ether_addr(addr))
 601                 err = dev_uc_add_excl(dev, addr);
 602         else if (is_multicast_ether_addr(addr))
 603                 err = dev_mc_add_excl(dev, addr);
 604
 605         return err;
 606 }
 607
 608 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 609                            struct net_device *dev,
 610                            const unsigned char *addr)
 611 {
 612         struct macvlan_dev *vlan = netdev_priv(dev);
 613         int err = -EINVAL;
 614
 615         if (!vlan->port->passthru)
 616                 return -EOPNOTSUPP;
 617
 618         if (is_unicast_ether_addr(addr))
 619                 err = dev_uc_del(dev, addr);
 620         else if (is_multicast_ether_addr(addr))
 621                 err = dev_mc_del(dev, addr);
 622
 623         return err;
 624 }
 625
 626 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 627                                         struct ethtool_drvinfo *drvinfo)
 628 {
 629         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 630         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 631 }
 632
 633 static int macvlan_ethtool_get_settings(struct net_device *dev,
 634                                         struct ethtool_cmd *cmd)
 635 {
 636         const struct macvlan_dev *vlan = netdev_priv(dev);
 637
 638         return __ethtool_get_settings(vlan->lowerdev, cmd);
 639 }
 640
 641 static const struct ethtool_ops macvlan_ethtool_ops = {
 642         .get_link               = ethtool_op_get_link,
 643         .get_settings           = macvlan_ethtool_get_settings,
 644         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 645 };
 646
 647 static const struct net_device_ops macvlan_netdev_ops = {
 648         .ndo_init               = macvlan_init,
 649         .ndo_uninit             = macvlan_uninit,
 650         .ndo_open               = macvlan_open,
 651         .ndo_stop               = macvlan_stop,
 652         .ndo_start_xmit         = macvlan_start_xmit,
 653         .ndo_change_mtu         = macvlan_change_mtu,
 654         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 655         .ndo_set_mac_address    = macvlan_set_mac_address,
 656         .ndo_set_rx_mode        = macvlan_set_mac_lists,
 657         .ndo_get_stats64        = macvlan_dev_get_stats64,
 658         .ndo_validate_addr      = eth_validate_addr,
 659         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
 660         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
 661         .ndo_fdb_add            = macvlan_fdb_add,
 662         .ndo_fdb_del            = macvlan_fdb_del,
 663         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
 664 };
 665
 666 void macvlan_common_setup(struct net_device *dev)
 667 {
 668         ether_setup(dev);
 669
 670         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 671         dev->priv_flags        |= IFF_UNICAST_FLT;
 672         dev->netdev_ops         = &macvlan_netdev_ops;
 673         dev->destructor         = free_netdev;
 674         dev->header_ops         = &macvlan_hard_header_ops,
 675         dev->ethtool_ops        = &macvlan_ethtool_ops;
 676 }
 677 EXPORT_SYMBOL_GPL(macvlan_common_setup);
 678
 679 static void macvlan_setup(struct net_device *dev)
 680 {
 681         macvlan_common_setup(dev);
 682         dev->tx_queue_len       = 0;
 683 }
 684
 685 static int macvlan_port_create(struct net_device *dev)
 686 {
 687         struct macvlan_port *port;
 688         unsigned int i;
 689         int err;
 690
 691         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 692                 return -EINVAL;
 693
 694         port = kzalloc(sizeof(*port), GFP_KERNEL);
 695         if (port == NULL)
 696                 return -ENOMEM;
 697
 698         port->passthru = false;
 699         port->dev = dev;
 700         INIT_LIST_HEAD(&port->vlans);
 701         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 702                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 703
 704         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
 705         if (err)
 706                 kfree(port);
 707         else
 708                 dev->priv_flags |= IFF_MACVLAN_PORT;
 709         return err;
 710 }
 711
 712 static void macvlan_port_destroy(struct net_device *dev)
 713 {
 714         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
 715
 716         dev->priv_flags &= ~IFF_MACVLAN_PORT;
 717         netdev_rx_handler_unregister(dev);
 718         kfree_rcu(port, rcu);
 719 }
 720
 721 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 722 {
 723         if (tb[IFLA_ADDRESS]) {
 724                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 725                         return -EINVAL;
 726                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 727                         return -EADDRNOTAVAIL;
 728         }
 729
 730         if (data && data[IFLA_MACVLAN_FLAGS] &&
 731             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
 732                 return -EINVAL;
 733
 734         if (data && data[IFLA_MACVLAN_MODE]) {
 735                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
 736                 case MACVLAN_MODE_PRIVATE:
 737                 case MACVLAN_MODE_VEPA:
 738                 case MACVLAN_MODE_BRIDGE:
 739                 case MACVLAN_MODE_PASSTHRU:
 740                         break;
 741                 default:
 742                         return -EINVAL;
 743                 }
 744         }
 745         return 0;
 746 }
 747
 748 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
 749                            struct nlattr *tb[], struct nlattr *data[],
 750                            int (*receive)(struct sk_buff *skb),
 751                            int (*forward)(struct net_device *dev,
 752                                           struct sk_buff *skb))
 753 {
 754         struct macvlan_dev *vlan = netdev_priv(dev);
 755         struct macvlan_port *port;
 756         struct net_device *lowerdev;
 757         int err;
 758
 759         if (!tb[IFLA_LINK])
 760                 return -EINVAL;
 761
 762         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 763         if (lowerdev == NULL)
 764                 return -ENODEV;
 765
 766         /* When creating macvlans on top of other macvlans - use
 767          * the real device as the lowerdev.
 768          */
 769         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
 770                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
 771                 lowerdev = lowervlan->lowerdev;
 772         }
 773
 774         if (!tb[IFLA_MTU])
 775                 dev->mtu = lowerdev->mtu;
 776         else if (dev->mtu > lowerdev->mtu)
 777                 return -EINVAL;
 778
 779         if (!tb[IFLA_ADDRESS])
 780                 eth_hw_addr_random(dev);
 781
 782         if (!macvlan_port_exists(lowerdev)) {
 783                 err = macvlan_port_create(lowerdev);
 784                 if (err < 0)
 785                         return err;
 786         }
 787         port = macvlan_port_get_rtnl(lowerdev);
 788
 789         /* Only 1 macvlan device can be created in passthru mode */
 790         if (port->passthru)
 791                 return -EINVAL;
 792
 793         vlan->lowerdev = lowerdev;
 794         vlan->dev      = dev;
 795         vlan->port     = port;
 796         vlan->receive  = receive;
 797         vlan->forward  = forward;
 798
 799         vlan->mode     = MACVLAN_MODE_VEPA;
 800         if (data && data[IFLA_MACVLAN_MODE])
 801                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 802
 803         if (data && data[IFLA_MACVLAN_FLAGS])
 804                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
 805
 806         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 807                 if (port->count)
 808                         return -EINVAL;
 809                 port->passthru = true;
 810                 memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
 811         }
 812
 813         err = netdev_upper_dev_link(lowerdev, dev);
 814         if (err)
 815                 goto destroy_port;
 816
 817         port->count += 1;
 818         err = register_netdevice(dev);
 819         if (err < 0)
 820                 goto upper_dev_unlink;
 821
 822         list_add_tail_rcu(&vlan->list, &port->vlans);
 823         netif_stacked_transfer_operstate(lowerdev, dev);
 824
 825         return 0;
 826
 827 upper_dev_unlink:
 828         netdev_upper_dev_unlink(lowerdev, dev);
 829 destroy_port:
 830         port->count -= 1;
 831         if (!port->count)
 832                 macvlan_port_destroy(lowerdev);
 833
 834         return err;
 835 }
 836 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
 837
 838 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
 839                            struct nlattr *tb[], struct nlattr *data[])
 840 {
 841         return macvlan_common_newlink(src_net, dev, tb, data,
 842                                       netif_rx,
 843                                       dev_forward_skb);
 844 }
 845
 846 void macvlan_dellink(struct net_device *dev, struct list_head *head)
 847 {
 848         struct macvlan_dev *vlan = netdev_priv(dev);
 849
 850         list_del_rcu(&vlan->list);
 851         unregister_netdevice_queue(dev, head);
 852         netdev_upper_dev_unlink(vlan->lowerdev, dev);
 853 }
 854 EXPORT_SYMBOL_GPL(macvlan_dellink);
 855
 856 static int macvlan_changelink(struct net_device *dev,
 857                 struct nlattr *tb[], struct nlattr *data[])
 858 {
 859         struct macvlan_dev *vlan = netdev_priv(dev);
 860
 861         if (data && data[IFLA_MACVLAN_FLAGS]) {
 862                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
 863                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
 864                 if (vlan->port->passthru && promisc) {
 865                         int err;
 866
 867                         if (flags & MACVLAN_FLAG_NOPROMISC)
 868                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
 869                         else
 870                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
 871                         if (err < 0)
 872                                 return err;
 873                 }
 874                 vlan->flags = flags;
 875         }
 876         if (data && data[IFLA_MACVLAN_MODE])
 877                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 878         return 0;
 879 }
 880
 881 static size_t macvlan_get_size(const struct net_device *dev)
 882 {
 883         return (0
 884                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
 885                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
 886                 );
 887 }
 888
 889 static int macvlan_fill_info(struct sk_buff *skb,
 890                                 const struct net_device *dev)
 891 {
 892         struct macvlan_dev *vlan = netdev_priv(dev);
 893
 894         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
 895                 goto nla_put_failure;
 896         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
 897                 goto nla_put_failure;
 898         return 0;
 899
 900 nla_put_failure:
 901         return -EMSGSIZE;
 902 }
 903
 904 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
 905         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
 906         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
 907 };
 908
 909 int macvlan_link_register(struct rtnl_link_ops *ops)
 910 {
 911         /* common fields */
 912         ops->priv_size          = sizeof(struct macvlan_dev);
 913         ops->validate           = macvlan_validate;
 914         ops->maxtype            = IFLA_MACVLAN_MAX;
 915         ops->policy             = macvlan_policy;
 916         ops->changelink         = macvlan_changelink;
 917         ops->get_size           = macvlan_get_size;
 918         ops->fill_info          = macvlan_fill_info;
 919
 920         return rtnl_link_register(ops);
 921 };
 922 EXPORT_SYMBOL_GPL(macvlan_link_register);
 923
 924 static struct rtnl_link_ops macvlan_link_ops = {
 925         .kind           = "macvlan",
 926         .setup          = macvlan_setup,
 927         .newlink        = macvlan_newlink,
 928         .dellink        = macvlan_dellink,
 929 };
 930
 931 static int macvlan_device_event(struct notifier_block *unused,
 932                                 unsigned long event, void *ptr)
 933 {
 934         struct net_device *dev = ptr;
 935         struct macvlan_dev *vlan, *next;
 936         struct macvlan_port *port;
 937         LIST_HEAD(list_kill);
 938
 939         if (!macvlan_port_exists(dev))
 940                 return NOTIFY_DONE;
 941
 942         port = macvlan_port_get_rtnl(dev);
 943
 944         switch (event) {
 945         case NETDEV_CHANGE:
 946                 list_for_each_entry(vlan, &port->vlans, list)
 947                         netif_stacked_transfer_operstate(vlan->lowerdev,
 948                                                          vlan->dev);
 949                 break;
 950         case NETDEV_FEAT_CHANGE:
 951                 list_for_each_entry(vlan, &port->vlans, list) {
 952                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
 953                         vlan->dev->gso_max_size = dev->gso_max_size;
 954                         netdev_features_change(vlan->dev);
 955                 }
 956                 break;
 957         case NETDEV_UNREGISTER:
 958                 /* twiddle thumbs on netns device moves */
 959                 if (dev->reg_state != NETREG_UNREGISTERING)
 960                         break;
 961
 962                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
 963                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
 964                 unregister_netdevice_many(&list_kill);
 965                 break;
 966         case NETDEV_PRE_TYPE_CHANGE:
 967                 /* Forbid underlaying device to change its type. */
 968                 return NOTIFY_BAD;
 969         }
 970         return NOTIFY_DONE;
 971 }
 972
 973 static struct notifier_block macvlan_notifier_block __read_mostly = {
 974         .notifier_call  = macvlan_device_event,
 975 };
 976
 977 static int __init macvlan_init_module(void)
 978 {
 979         int err;
 980
 981         register_netdevice_notifier(&macvlan_notifier_block);
 982
 983         err = macvlan_link_register(&macvlan_link_ops);
 984         if (err < 0)
 985                 goto err1;
 986         return 0;
 987 err1:
 988         unregister_netdevice_notifier(&macvlan_notifier_block);
 989         return err;
 990 }
 991
 992 static void __exit macvlan_cleanup_module(void)
 993 {
 994         rtnl_link_unregister(&macvlan_link_ops);
 995         unregister_netdevice_notifier(&macvlan_notifier_block);
 996 }
 997
 998 module_init(macvlan_init_module);
 999 module_exit(macvlan_cleanup_module);
1000
1001 MODULE_LICENSE("GPL");
1002 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1003 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1004 MODULE_ALIAS_RTNL_LINK("macvlan");