Pileus Git - ~andy/linux/blob - 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
 122         if (local)
 123                 return vlan->forward(dev, skb);
 124
 125         skb->dev = dev;
 126         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 127                 skb->pkt_type = PACKET_BROADCAST;
 128         else
 129                 skb->pkt_type = PACKET_MULTICAST;
 130
 131         return vlan->receive(skb);
 132 }
 133
 134 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 135 {
 136         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 137 }
 138
 139
 140 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 141                             const unsigned char *addr)
 142 {
 143         u32 val = __get_unaligned_cpu32(addr + 2);
 144
 145         val ^= macvlan_hash_mix(vlan);
 146         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 147 }
 148
 149 static void macvlan_broadcast(struct sk_buff *skb,
 150                               const struct macvlan_port *port,
 151                               struct net_device *src,
 152                               enum macvlan_mode mode)
 153 {
 154         const struct ethhdr *eth = eth_hdr(skb);
 155         const struct macvlan_dev *vlan;
 156         struct sk_buff *nskb;
 157         unsigned int i;
 158         int err;
 159         unsigned int hash;
 160
 161         if (skb->protocol == htons(ETH_P_PAUSE))
 162                 return;
 163
 164         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 165                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 166                         if (vlan->dev == src || !(vlan->mode & mode))
 167                                 continue;
 168
 169                         hash = mc_hash(vlan, eth->h_dest);
 170                         if (!test_bit(hash, vlan->mc_filter))
 171                                 continue;
 172
 173                         err = NET_RX_DROP;
 174                         nskb = skb_clone(skb, GFP_ATOMIC);
 175                         if (likely(nskb))
 176                                 err = macvlan_broadcast_one(
 177                                         nskb, vlan, eth,
 178                                         mode == MACVLAN_MODE_BRIDGE);
 179                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 180                                          err == NET_RX_SUCCESS, 1);
 181                 }
 182         }
 183 }
 184
 185 /* called under rcu_read_lock() from netif_receive_skb */
 186 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 187 {
 188         struct macvlan_port *port;
 189         struct sk_buff *skb = *pskb;
 190         const struct ethhdr *eth = eth_hdr(skb);
 191         const struct macvlan_dev *vlan;
 192         const struct macvlan_dev *src;
 193         struct net_device *dev;
 194         unsigned int len = 0;
 195         int ret = NET_RX_DROP;
 196
 197         port = macvlan_port_get_rcu(skb->dev);
 198         if (is_multicast_ether_addr(eth->h_dest)) {
 199                 skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
 200                 if (!skb)
 201                         return RX_HANDLER_CONSUMED;
 202                 eth = eth_hdr(skb);
 203                 src = macvlan_hash_lookup(port, eth->h_source);
 204                 if (!src)
 205                         /* frame comes from an external address */
 206                         macvlan_broadcast(skb, port, NULL,
 207                                           MACVLAN_MODE_PRIVATE |
 208                                           MACVLAN_MODE_VEPA    |
 209                                           MACVLAN_MODE_PASSTHRU|
 210                                           MACVLAN_MODE_BRIDGE);
 211                 else if (src->mode == MACVLAN_MODE_VEPA)
 212                         /* flood to everyone except source */
 213                         macvlan_broadcast(skb, port, src->dev,
 214                                           MACVLAN_MODE_VEPA |
 215                                           MACVLAN_MODE_BRIDGE);
 216                 else if (src->mode == MACVLAN_MODE_BRIDGE)
 217                         /*
 218                          * flood only to VEPA ports, bridge ports
 219                          * already saw the frame on the way out.
 220                          */
 221                         macvlan_broadcast(skb, port, src->dev,
 222                                           MACVLAN_MODE_VEPA);
 223                 else {
 224                         /* forward to original port. */
 225                         vlan = src;
 226                         ret = macvlan_broadcast_one(skb, vlan, eth, 0);
 227                         goto out;
 228                 }
 229
 230                 return RX_HANDLER_PASS;
 231         }
 232
 233         if (port->passthru)
 234                 vlan = list_first_or_null_rcu(&port->vlans,
 235                                               struct macvlan_dev, list);
 236         else
 237                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 238         if (vlan == NULL)
 239                 return RX_HANDLER_PASS;
 240
 241         dev = vlan->dev;
 242         if (unlikely(!(dev->flags & IFF_UP))) {
 243                 kfree_skb(skb);
 244                 return RX_HANDLER_CONSUMED;
 245         }
 246         len = skb->len + ETH_HLEN;
 247         skb = skb_share_check(skb, GFP_ATOMIC);
 248         if (!skb)
 249                 goto out;
 250
 251         skb->dev = dev;
 252         skb->pkt_type = PACKET_HOST;
 253
 254         ret = vlan->receive(skb);
 255
 256 out:
 257         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
 258         return RX_HANDLER_CONSUMED;
 259 }
 260
 261 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 262 {
 263         const struct macvlan_dev *vlan = netdev_priv(dev);
 264         const struct macvlan_port *port = vlan->port;
 265         const struct macvlan_dev *dest;
 266         __u8 ip_summed = skb->ip_summed;
 267
 268         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 269                 const struct ethhdr *eth = (void *)skb->data;
 270                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 271
 272                 /* send to other bridge ports directly */
 273                 if (is_multicast_ether_addr(eth->h_dest)) {
 274                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 275                         goto xmit_world;
 276                 }
 277
 278                 dest = macvlan_hash_lookup(port, eth->h_dest);
 279                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 280                         /* send to lowerdev first for its network taps */
 281                         dev_forward_skb(vlan->lowerdev, skb);
 282
 283                         return NET_XMIT_SUCCESS;
 284                 }
 285         }
 286
 287 xmit_world:
 288         skb->ip_summed = ip_summed;
 289         skb->dev = vlan->lowerdev;
 290         return dev_queue_xmit(skb);
 291 }
 292
 293 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 294                                struct net_device *dev)
 295 {
 296         unsigned int len = skb->len;
 297         int ret;
 298         const struct macvlan_dev *vlan = netdev_priv(dev);
 299
 300         if (vlan->fwd_priv) {
 301                 skb->dev = vlan->lowerdev;
 302                 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
 303         } else {
 304                 ret = macvlan_queue_xmit(skb, dev);
 305         }
 306
 307         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 308                 struct macvlan_pcpu_stats *pcpu_stats;
 309
 310                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 311                 u64_stats_update_begin(&pcpu_stats->syncp);
 312                 pcpu_stats->tx_packets++;
 313                 pcpu_stats->tx_bytes += len;
 314                 u64_stats_update_end(&pcpu_stats->syncp);
 315         } else {
 316                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 317         }
 318         return ret;
 319 }
 320 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
 321
 322 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 323                                unsigned short type, const void *daddr,
 324                                const void *saddr, unsigned len)
 325 {
 326         const struct macvlan_dev *vlan = netdev_priv(dev);
 327         struct net_device *lowerdev = vlan->lowerdev;
 328
 329         return dev_hard_header(skb, lowerdev, type, daddr,
 330                                saddr ? : dev->dev_addr, len);
 331 }
 332
 333 static const struct header_ops macvlan_hard_header_ops = {
 334         .create         = macvlan_hard_header,
 335         .rebuild        = eth_rebuild_header,
 336         .parse          = eth_header_parse,
 337         .cache          = eth_header_cache,
 338         .cache_update   = eth_header_cache_update,
 339 };
 340
 341 static struct rtnl_link_ops macvlan_link_ops;
 342
 343 static int macvlan_open(struct net_device *dev)
 344 {
 345         struct macvlan_dev *vlan = netdev_priv(dev);
 346         struct net_device *lowerdev = vlan->lowerdev;
 347         int err;
 348
 349         if (vlan->port->passthru) {
 350                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 351                         err = dev_set_promiscuity(lowerdev, 1);
 352                         if (err < 0)
 353                                 goto out;
 354                 }
 355                 goto hash_add;
 356         }
 357
 358         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
 359             dev->rtnl_link_ops == &macvlan_link_ops) {
 360                 vlan->fwd_priv =
 361                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 362
 363                 /* If we get a NULL pointer back, or if we get an error
 364                  * then we should just fall through to the non accelerated path
 365                  */
 366                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
 367                         vlan->fwd_priv = NULL;
 368                 } else
 369                         return 0;
 370         }
 371
 372         err = -EBUSY;
 373         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 374                 goto out;
 375
 376         err = dev_uc_add(lowerdev, dev->dev_addr);
 377         if (err < 0)
 378                 goto out;
 379         if (dev->flags & IFF_ALLMULTI) {
 380                 err = dev_set_allmulti(lowerdev, 1);
 381                 if (err < 0)
 382                         goto del_unicast;
 383         }
 384
 385 hash_add:
 386         macvlan_hash_add(vlan);
 387         return 0;
 388
 389 del_unicast:
 390         dev_uc_del(lowerdev, dev->dev_addr);
 391 out:
 392         if (vlan->fwd_priv) {
 393                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 394                                                            vlan->fwd_priv);
 395                 vlan->fwd_priv = NULL;
 396         }
 397         return err;
 398 }
 399
 400 static int macvlan_stop(struct net_device *dev)
 401 {
 402         struct macvlan_dev *vlan = netdev_priv(dev);
 403         struct net_device *lowerdev = vlan->lowerdev;
 404
 405         if (vlan->fwd_priv) {
 406                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 407                                                            vlan->fwd_priv);
 408                 vlan->fwd_priv = NULL;
 409                 return 0;
 410         }
 411
 412         dev_uc_unsync(lowerdev, dev);
 413         dev_mc_unsync(lowerdev, dev);
 414
 415         if (vlan->port->passthru) {
 416                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 417                         dev_set_promiscuity(lowerdev, -1);
 418                 goto hash_del;
 419         }
 420
 421         if (dev->flags & IFF_ALLMULTI)
 422                 dev_set_allmulti(lowerdev, -1);
 423
 424         dev_uc_del(lowerdev, dev->dev_addr);
 425
 426 hash_del:
 427         macvlan_hash_del(vlan, !dev->dismantle);
 428         return 0;
 429 }
 430
 431 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 432 {
 433         struct macvlan_dev *vlan = netdev_priv(dev);
 434         struct net_device *lowerdev = vlan->lowerdev;
 435         struct sockaddr *addr = p;
 436         int err;
 437
 438         if (!is_valid_ether_addr(addr->sa_data))
 439                 return -EADDRNOTAVAIL;
 440
 441         if (!(dev->flags & IFF_UP)) {
 442                 /* Just copy in the new address */
 443                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 444         } else {
 445                 /* Rehash and update the device filters */
 446                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
 447                         return -EBUSY;
 448
 449                 err = dev_uc_add(lowerdev, addr->sa_data);
 450                 if (err)
 451                         return err;
 452
 453                 dev_uc_del(lowerdev, dev->dev_addr);
 454
 455                 macvlan_hash_change_addr(vlan, addr->sa_data);
 456         }
 457         return 0;
 458 }
 459
 460 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 461 {
 462         struct macvlan_dev *vlan = netdev_priv(dev);
 463         struct net_device *lowerdev = vlan->lowerdev;
 464
 465         if (change & IFF_ALLMULTI)
 466                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 467 }
 468
 469 static void macvlan_set_mac_lists(struct net_device *dev)
 470 {
 471         struct macvlan_dev *vlan = netdev_priv(dev);
 472
 473         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 474                 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
 475         } else {
 476                 struct netdev_hw_addr *ha;
 477                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 478
 479                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 480                 netdev_for_each_mc_addr(ha, dev) {
 481                         __set_bit(mc_hash(vlan, ha->addr), filter);
 482                 }
 483
 484                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 485
 486                 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 487         }
 488         dev_uc_sync(vlan->lowerdev, dev);
 489         dev_mc_sync(vlan->lowerdev, dev);
 490 }
 491
 492 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 493 {
 494         struct macvlan_dev *vlan = netdev_priv(dev);
 495
 496         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 497                 return -EINVAL;
 498         dev->mtu = new_mtu;
 499         return 0;
 500 }
 501
 502 /*
 503  * macvlan network devices have devices nesting below it and are a special
 504  * "super class" of normal network devices; split their locks off into a
 505  * separate class since they always nest.
 506  */
 507 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 508 static struct lock_class_key macvlan_netdev_addr_lock_key;
 509
 510 #define MACVLAN_FEATURES \
 511         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 512          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 513          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 514          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 515
 516 #define MACVLAN_STATE_MASK \
 517         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 518
 519 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 520                                           struct netdev_queue *txq,
 521                                           void *_unused)
 522 {
 523         lockdep_set_class(&txq->_xmit_lock,
 524                           &macvlan_netdev_xmit_lock_key);
 525 }
 526
 527 static void macvlan_set_lockdep_class(struct net_device *dev)
 528 {
 529         lockdep_set_class(&dev->addr_list_lock,
 530                           &macvlan_netdev_addr_lock_key);
 531         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 532 }
 533
 534 static int macvlan_init(struct net_device *dev)
 535 {
 536         struct macvlan_dev *vlan = netdev_priv(dev);
 537         const struct net_device *lowerdev = vlan->lowerdev;
 538         int i;
 539
 540         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 541                                   (lowerdev->state & MACVLAN_STATE_MASK);
 542         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 543         dev->features           |= NETIF_F_LLTX;
 544         dev->gso_max_size       = lowerdev->gso_max_size;
 545         dev->iflink             = lowerdev->ifindex;
 546         dev->hard_header_len    = lowerdev->hard_header_len;
 547
 548         macvlan_set_lockdep_class(dev);
 549
 550         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
 551         if (!vlan->pcpu_stats)
 552                 return -ENOMEM;
 553
 554         for_each_possible_cpu(i) {
 555                 struct macvlan_pcpu_stats *mvlstats;
 556                 mvlstats = per_cpu_ptr(vlan->pcpu_stats, i);
 557                 u64_stats_init(&mvlstats->syncp);
 558         }
 559
 560         return 0;
 561 }
 562
 563 static void macvlan_uninit(struct net_device *dev)
 564 {
 565         struct macvlan_dev *vlan = netdev_priv(dev);
 566         struct macvlan_port *port = vlan->port;
 567
 568         free_percpu(vlan->pcpu_stats);
 569
 570         port->count -= 1;
 571         if (!port->count)
 572                 macvlan_port_destroy(port->dev);
 573 }
 574
 575 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 576                                                          struct rtnl_link_stats64 *stats)
 577 {
 578         struct macvlan_dev *vlan = netdev_priv(dev);
 579
 580         if (vlan->pcpu_stats) {
 581                 struct macvlan_pcpu_stats *p;
 582                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 583                 u32 rx_errors = 0, tx_dropped = 0;
 584                 unsigned int start;
 585                 int i;
 586
 587                 for_each_possible_cpu(i) {
 588                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 589                         do {
 590                                 start = u64_stats_fetch_begin_bh(&p->syncp);
 591                                 rx_packets      = p->rx_packets;
 592                                 rx_bytes        = p->rx_bytes;
 593                                 rx_multicast    = p->rx_multicast;
 594                                 tx_packets      = p->tx_packets;
 595                                 tx_bytes        = p->tx_bytes;
 596                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
 597
 598                         stats->rx_packets       += rx_packets;
 599                         stats->rx_bytes         += rx_bytes;
 600                         stats->multicast        += rx_multicast;
 601                         stats->tx_packets       += tx_packets;
 602                         stats->tx_bytes         += tx_bytes;
 603                         /* rx_errors & tx_dropped are u32, updated
 604                          * without syncp protection.
 605                          */
 606                         rx_errors       += p->rx_errors;
 607                         tx_dropped      += p->tx_dropped;
 608                 }
 609                 stats->rx_errors        = rx_errors;
 610                 stats->rx_dropped       = rx_errors;
 611                 stats->tx_dropped       = tx_dropped;
 612         }
 613         return stats;
 614 }
 615
 616 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 617                                    __be16 proto, u16 vid)
 618 {
 619         struct macvlan_dev *vlan = netdev_priv(dev);
 620         struct net_device *lowerdev = vlan->lowerdev;
 621
 622         return vlan_vid_add(lowerdev, proto, vid);
 623 }
 624
 625 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 626                                     __be16 proto, u16 vid)
 627 {
 628         struct macvlan_dev *vlan = netdev_priv(dev);
 629         struct net_device *lowerdev = vlan->lowerdev;
 630
 631         vlan_vid_del(lowerdev, proto, vid);
 632         return 0;
 633 }
 634
 635 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 636                            struct net_device *dev,
 637                            const unsigned char *addr,
 638                            u16 flags)
 639 {
 640         struct macvlan_dev *vlan = netdev_priv(dev);
 641         int err = -EINVAL;
 642
 643         if (!vlan->port->passthru)
 644                 return -EOPNOTSUPP;
 645
 646         if (flags & NLM_F_REPLACE)
 647                 return -EOPNOTSUPP;
 648
 649         if (is_unicast_ether_addr(addr))
 650                 err = dev_uc_add_excl(dev, addr);
 651         else if (is_multicast_ether_addr(addr))
 652                 err = dev_mc_add_excl(dev, addr);
 653
 654         return err;
 655 }
 656
 657 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 658                            struct net_device *dev,
 659                            const unsigned char *addr)
 660 {
 661         struct macvlan_dev *vlan = netdev_priv(dev);
 662         int err = -EINVAL;
 663
 664         if (!vlan->port->passthru)
 665                 return -EOPNOTSUPP;
 666
 667         if (is_unicast_ether_addr(addr))
 668                 err = dev_uc_del(dev, addr);
 669         else if (is_multicast_ether_addr(addr))
 670                 err = dev_mc_del(dev, addr);
 671
 672         return err;
 673 }
 674
 675 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 676                                         struct ethtool_drvinfo *drvinfo)
 677 {
 678         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 679         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 680 }
 681
 682 static int macvlan_ethtool_get_settings(struct net_device *dev,
 683                                         struct ethtool_cmd *cmd)
 684 {
 685         const struct macvlan_dev *vlan = netdev_priv(dev);
 686
 687         return __ethtool_get_settings(vlan->lowerdev, cmd);
 688 }
 689
 690 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 691                                               netdev_features_t features)
 692 {
 693         struct macvlan_dev *vlan = netdev_priv(dev);
 694         netdev_features_t mask;
 695
 696         features |= NETIF_F_ALL_FOR_ALL;
 697         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 698         mask = features;
 699
 700         features = netdev_increment_features(vlan->lowerdev->features,
 701                                              features,
 702                                              mask);
 703         features |= NETIF_F_LLTX;
 704
 705         return features;
 706 }
 707
 708 static const struct ethtool_ops macvlan_ethtool_ops = {
 709         .get_link               = ethtool_op_get_link,
 710         .get_settings           = macvlan_ethtool_get_settings,
 711         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 712 };
 713
 714 static const struct net_device_ops macvlan_netdev_ops = {
 715         .ndo_init               = macvlan_init,
 716         .ndo_uninit             = macvlan_uninit,
 717         .ndo_open               = macvlan_open,
 718         .ndo_stop               = macvlan_stop,
 719         .ndo_start_xmit         = macvlan_start_xmit,
 720         .ndo_change_mtu         = macvlan_change_mtu,
 721         .ndo_fix_features       = macvlan_fix_features,
 722         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 723         .ndo_set_mac_address    = macvlan_set_mac_address,
 724         .ndo_set_rx_mode        = macvlan_set_mac_lists,
 725         .ndo_get_stats64        = macvlan_dev_get_stats64,
 726         .ndo_validate_addr      = eth_validate_addr,
 727         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
 728         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
 729         .ndo_fdb_add            = macvlan_fdb_add,
 730         .ndo_fdb_del            = macvlan_fdb_del,
 731         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
 732 };
 733
 734 void macvlan_common_setup(struct net_device *dev)
 735 {
 736         ether_setup(dev);
 737
 738         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 739         dev->priv_flags        |= IFF_UNICAST_FLT;
 740         dev->netdev_ops         = &macvlan_netdev_ops;
 741         dev->destructor         = free_netdev;
 742         dev->header_ops         = &macvlan_hard_header_ops;
 743         dev->ethtool_ops        = &macvlan_ethtool_ops;
 744 }
 745 EXPORT_SYMBOL_GPL(macvlan_common_setup);
 746
 747 static void macvlan_setup(struct net_device *dev)
 748 {
 749         macvlan_common_setup(dev);
 750         dev->tx_queue_len       = 0;
 751 }
 752
 753 static int macvlan_port_create(struct net_device *dev)
 754 {
 755         struct macvlan_port *port;
 756         unsigned int i;
 757         int err;
 758
 759         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 760                 return -EINVAL;
 761
 762         port = kzalloc(sizeof(*port), GFP_KERNEL);
 763         if (port == NULL)
 764                 return -ENOMEM;
 765
 766         port->passthru = false;
 767         port->dev = dev;
 768         INIT_LIST_HEAD(&port->vlans);
 769         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 770                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 771
 772         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
 773         if (err)
 774                 kfree(port);
 775         else
 776                 dev->priv_flags |= IFF_MACVLAN_PORT;
 777         return err;
 778 }
 779
 780 static void macvlan_port_destroy(struct net_device *dev)
 781 {
 782         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
 783
 784         dev->priv_flags &= ~IFF_MACVLAN_PORT;
 785         netdev_rx_handler_unregister(dev);
 786         kfree_rcu(port, rcu);
 787 }
 788
 789 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 790 {
 791         if (tb[IFLA_ADDRESS]) {
 792                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 793                         return -EINVAL;
 794                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 795                         return -EADDRNOTAVAIL;
 796         }
 797
 798         if (data && data[IFLA_MACVLAN_FLAGS] &&
 799             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
 800                 return -EINVAL;
 801
 802         if (data && data[IFLA_MACVLAN_MODE]) {
 803                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
 804                 case MACVLAN_MODE_PRIVATE:
 805                 case MACVLAN_MODE_VEPA:
 806                 case MACVLAN_MODE_BRIDGE:
 807                 case MACVLAN_MODE_PASSTHRU:
 808                         break;
 809                 default:
 810                         return -EINVAL;
 811                 }
 812         }
 813         return 0;
 814 }
 815
 816 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
 817                            struct nlattr *tb[], struct nlattr *data[],
 818                            int (*receive)(struct sk_buff *skb),
 819                            int (*forward)(struct net_device *dev,
 820                                           struct sk_buff *skb))
 821 {
 822         struct macvlan_dev *vlan = netdev_priv(dev);
 823         struct macvlan_port *port;
 824         struct net_device *lowerdev;
 825         int err;
 826
 827         if (!tb[IFLA_LINK])
 828                 return -EINVAL;
 829
 830         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 831         if (lowerdev == NULL)
 832                 return -ENODEV;
 833
 834         /* When creating macvlans on top of other macvlans - use
 835          * the real device as the lowerdev.
 836          */
 837         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
 838                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
 839                 lowerdev = lowervlan->lowerdev;
 840         }
 841
 842         if (!tb[IFLA_MTU])
 843                 dev->mtu = lowerdev->mtu;
 844         else if (dev->mtu > lowerdev->mtu)
 845                 return -EINVAL;
 846
 847         if (!tb[IFLA_ADDRESS])
 848                 eth_hw_addr_random(dev);
 849
 850         if (!macvlan_port_exists(lowerdev)) {
 851                 err = macvlan_port_create(lowerdev);
 852                 if (err < 0)
 853                         return err;
 854         }
 855         port = macvlan_port_get_rtnl(lowerdev);
 856
 857         /* Only 1 macvlan device can be created in passthru mode */
 858         if (port->passthru)
 859                 return -EINVAL;
 860
 861         vlan->lowerdev = lowerdev;
 862         vlan->dev      = dev;
 863         vlan->port     = port;
 864         vlan->receive  = receive;
 865         vlan->forward  = forward;
 866         vlan->set_features = MACVLAN_FEATURES;
 867
 868         vlan->mode     = MACVLAN_MODE_VEPA;
 869         if (data && data[IFLA_MACVLAN_MODE])
 870                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 871
 872         if (data && data[IFLA_MACVLAN_FLAGS])
 873                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
 874
 875         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 876                 if (port->count)
 877                         return -EINVAL;
 878                 port->passthru = true;
 879                 eth_hw_addr_inherit(dev, lowerdev);
 880         }
 881
 882         port->count += 1;
 883         err = register_netdevice(dev);
 884         if (err < 0)
 885                 goto destroy_port;
 886
 887         dev->priv_flags |= IFF_MACVLAN;
 888         err = netdev_upper_dev_link(lowerdev, dev);
 889         if (err)
 890                 goto destroy_port;
 891
 892
 893         list_add_tail_rcu(&vlan->list, &port->vlans);
 894         netif_stacked_transfer_operstate(lowerdev, dev);
 895
 896         return 0;
 897
 898 destroy_port:
 899         port->count -= 1;
 900         if (!port->count)
 901                 macvlan_port_destroy(lowerdev);
 902
 903         return err;
 904 }
 905 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
 906
 907 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
 908                            struct nlattr *tb[], struct nlattr *data[])
 909 {
 910         return macvlan_common_newlink(src_net, dev, tb, data,
 911                                       netif_rx,
 912                                       dev_forward_skb);
 913 }
 914
 915 void macvlan_dellink(struct net_device *dev, struct list_head *head)
 916 {
 917         struct macvlan_dev *vlan = netdev_priv(dev);
 918
 919         list_del_rcu(&vlan->list);
 920         unregister_netdevice_queue(dev, head);
 921         netdev_upper_dev_unlink(vlan->lowerdev, dev);
 922 }
 923 EXPORT_SYMBOL_GPL(macvlan_dellink);
 924
 925 static int macvlan_changelink(struct net_device *dev,
 926                 struct nlattr *tb[], struct nlattr *data[])
 927 {
 928         struct macvlan_dev *vlan = netdev_priv(dev);
 929         enum macvlan_mode mode;
 930         bool set_mode = false;
 931
 932         /* Validate mode, but don't set yet: setting flags may fail. */
 933         if (data && data[IFLA_MACVLAN_MODE]) {
 934                 set_mode = true;
 935                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 936                 /* Passthrough mode can't be set or cleared dynamically */
 937                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
 938                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
 939                         return -EINVAL;
 940         }
 941
 942         if (data && data[IFLA_MACVLAN_FLAGS]) {
 943                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
 944                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
 945                 if (vlan->port->passthru && promisc) {
 946                         int err;
 947
 948                         if (flags & MACVLAN_FLAG_NOPROMISC)
 949                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
 950                         else
 951                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
 952                         if (err < 0)
 953                                 return err;
 954                 }
 955                 vlan->flags = flags;
 956         }
 957         if (set_mode)
 958                 vlan->mode = mode;
 959         return 0;
 960 }
 961
 962 static size_t macvlan_get_size(const struct net_device *dev)
 963 {
 964         return (0
 965                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
 966                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
 967                 );
 968 }
 969
 970 static int macvlan_fill_info(struct sk_buff *skb,
 971                                 const struct net_device *dev)
 972 {
 973         struct macvlan_dev *vlan = netdev_priv(dev);
 974
 975         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
 976                 goto nla_put_failure;
 977         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
 978                 goto nla_put_failure;
 979         return 0;
 980
 981 nla_put_failure:
 982         return -EMSGSIZE;
 983 }
 984
 985 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
 986         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
 987         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
 988 };
 989
 990 int macvlan_link_register(struct rtnl_link_ops *ops)
 991 {
 992         /* common fields */
 993         ops->priv_size          = sizeof(struct macvlan_dev);
 994         ops->validate           = macvlan_validate;
 995         ops->maxtype            = IFLA_MACVLAN_MAX;
 996         ops->policy             = macvlan_policy;
 997         ops->changelink         = macvlan_changelink;
 998         ops->get_size           = macvlan_get_size;
 999         ops->fill_info          = macvlan_fill_info;
1000
1001         return rtnl_link_register(ops);
1002 };
1003 EXPORT_SYMBOL_GPL(macvlan_link_register);
1004
1005 static struct rtnl_link_ops macvlan_link_ops = {
1006         .kind           = "macvlan",
1007         .setup          = macvlan_setup,
1008         .newlink        = macvlan_newlink,
1009         .dellink        = macvlan_dellink,
1010 };
1011
1012 static int macvlan_device_event(struct notifier_block *unused,
1013                                 unsigned long event, void *ptr)
1014 {
1015         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1016         struct macvlan_dev *vlan, *next;
1017         struct macvlan_port *port;
1018         LIST_HEAD(list_kill);
1019
1020         if (!macvlan_port_exists(dev))
1021                 return NOTIFY_DONE;
1022
1023         port = macvlan_port_get_rtnl(dev);
1024
1025         switch (event) {
1026         case NETDEV_CHANGE:
1027                 list_for_each_entry(vlan, &port->vlans, list)
1028                         netif_stacked_transfer_operstate(vlan->lowerdev,
1029                                                          vlan->dev);
1030                 break;
1031         case NETDEV_FEAT_CHANGE:
1032                 list_for_each_entry(vlan, &port->vlans, list) {
1033                         vlan->dev->gso_max_size = dev->gso_max_size;
1034                         netdev_update_features(vlan->dev);
1035                 }
1036                 break;
1037         case NETDEV_UNREGISTER:
1038                 /* twiddle thumbs on netns device moves */
1039                 if (dev->reg_state != NETREG_UNREGISTERING)
1040                         break;
1041
1042                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1043                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1044                 unregister_netdevice_many(&list_kill);
1045                 list_del(&list_kill);
1046                 break;
1047         case NETDEV_PRE_TYPE_CHANGE:
1048                 /* Forbid underlaying device to change its type. */
1049                 return NOTIFY_BAD;
1050         }
1051         return NOTIFY_DONE;
1052 }
1053
1054 static struct notifier_block macvlan_notifier_block __read_mostly = {
1055         .notifier_call  = macvlan_device_event,
1056 };
1057
1058 static int __init macvlan_init_module(void)
1059 {
1060         int err;
1061
1062         register_netdevice_notifier(&macvlan_notifier_block);
1063
1064         err = macvlan_link_register(&macvlan_link_ops);
1065         if (err < 0)
1066                 goto err1;
1067         return 0;
1068 err1:
1069         unregister_netdevice_notifier(&macvlan_notifier_block);
1070         return err;
1071 }
1072
1073 static void __exit macvlan_cleanup_module(void)
1074 {
1075         rtnl_link_unregister(&macvlan_link_ops);
1076         unregister_netdevice_notifier(&macvlan_notifier_block);
1077 }
1078
1079 module_init(macvlan_init_module);
1080 module_exit(macvlan_cleanup_module);
1081
1082 MODULE_LICENSE("GPL");
1083 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1084 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1085 MODULE_ALIAS_RTNL_LINK("macvlan");