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 <linux/workqueue.h>
  34 #include <net/rtnetlink.h>
  35 #include <net/xfrm.h>
  36 #include <linux/netpoll.h>
  37
  38 #define MACVLAN_HASH_BITS       8
  39 #define MACVLAN_HASH_SIZE       (1<<MACVLAN_HASH_BITS)
  40 #define MACVLAN_BC_QUEUE_LEN    1000
  41
  42 struct macvlan_port {
  43         struct net_device       *dev;
  44         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  45         struct list_head        vlans;
  46         struct rcu_head         rcu;
  47         struct sk_buff_head     bc_queue;
  48         struct work_struct      bc_work;
  49         bool                    passthru;
  50         int                     count;
  51         struct hlist_head       vlan_source_hash[MACVLAN_HASH_SIZE];
  52 };
  53
  54 struct macvlan_source_entry {
  55         struct hlist_node       hlist;
  56         struct macvlan_dev      *vlan;
  57         unsigned char           addr[6+2] __aligned(sizeof(u16));
  58         struct rcu_head         rcu;
  59 };
  60
  61 struct macvlan_skb_cb {
  62         const struct macvlan_dev *src;
  63 };
  64
  65 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
  66
  67 static void macvlan_port_destroy(struct net_device *dev);
  68
  69 /* Hash Ethernet address */
  70 static u32 macvlan_eth_hash(const unsigned char *addr)
  71 {
  72         u64 value = get_unaligned((u64 *)addr);
  73
  74         /* only want 6 bytes */
  75 #ifdef __BIG_ENDIAN
  76         value >>= 16;
  77 #else
  78         value <<= 16;
  79 #endif
  80         return hash_64(value, MACVLAN_HASH_BITS);
  81 }
  82
  83 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
  84 {
  85         return rcu_dereference(dev->rx_handler_data);
  86 }
  87
  88 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
  89 {
  90         return rtnl_dereference(dev->rx_handler_data);
  91 }
  92
  93 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
  94
  95 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
  96                                                const unsigned char *addr)
  97 {
  98         struct macvlan_dev *vlan;
  99         u32 idx = macvlan_eth_hash(addr);
 100
 101         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
 102                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
 103                         return vlan;
 104         }
 105         return NULL;
 106 }
 107
 108 static struct macvlan_source_entry *macvlan_hash_lookup_source(
 109         const struct macvlan_dev *vlan,
 110         const unsigned char *addr)
 111 {
 112         struct macvlan_source_entry *entry;
 113         u32 idx = macvlan_eth_hash(addr);
 114         struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
 115
 116         hlist_for_each_entry_rcu(entry, h, hlist) {
 117                 if (ether_addr_equal_64bits(entry->addr, addr) &&
 118                     entry->vlan == vlan)
 119                         return entry;
 120         }
 121         return NULL;
 122 }
 123
 124 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
 125                                    const unsigned char *addr)
 126 {
 127         struct macvlan_port *port = vlan->port;
 128         struct macvlan_source_entry *entry;
 129         struct hlist_head *h;
 130
 131         entry = macvlan_hash_lookup_source(vlan, addr);
 132         if (entry)
 133                 return 0;
 134
 135         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 136         if (!entry)
 137                 return -ENOMEM;
 138
 139         ether_addr_copy(entry->addr, addr);
 140         entry->vlan = vlan;
 141         h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
 142         hlist_add_head_rcu(&entry->hlist, h);
 143         vlan->macaddr_count++;
 144
 145         return 0;
 146 }
 147
 148 static void macvlan_hash_add(struct macvlan_dev *vlan)
 149 {
 150         struct macvlan_port *port = vlan->port;
 151         const unsigned char *addr = vlan->dev->dev_addr;
 152         u32 idx = macvlan_eth_hash(addr);
 153
 154         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
 155 }
 156
 157 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
 158 {
 159         hlist_del_rcu(&entry->hlist);
 160         kfree_rcu(entry, rcu);
 161 }
 162
 163 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
 164 {
 165         hlist_del_rcu(&vlan->hlist);
 166         if (sync)
 167                 synchronize_rcu();
 168 }
 169
 170 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
 171                                         const unsigned char *addr)
 172 {
 173         macvlan_hash_del(vlan, true);
 174         /* Now that we are unhashed it is safe to change the device
 175          * address without confusing packet delivery.
 176          */
 177         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
 178         macvlan_hash_add(vlan);
 179 }
 180
 181 static int macvlan_addr_busy(const struct macvlan_port *port,
 182                                 const unsigned char *addr)
 183 {
 184         /* Test to see if the specified multicast address is
 185          * currently in use by the underlying device or
 186          * another macvlan.
 187          */
 188         if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
 189                 return 1;
 190
 191         if (macvlan_hash_lookup(port, addr))
 192                 return 1;
 193
 194         return 0;
 195 }
 196
 197
 198 static int macvlan_broadcast_one(struct sk_buff *skb,
 199                                  const struct macvlan_dev *vlan,
 200                                  const struct ethhdr *eth, bool local)
 201 {
 202         struct net_device *dev = vlan->dev;
 203
 204         if (local)
 205                 return __dev_forward_skb(dev, skb);
 206
 207         skb->dev = dev;
 208         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 209                 skb->pkt_type = PACKET_BROADCAST;
 210         else
 211                 skb->pkt_type = PACKET_MULTICAST;
 212
 213         return 0;
 214 }
 215
 216 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 217 {
 218         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 219 }
 220
 221
 222 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 223                             const unsigned char *addr)
 224 {
 225         u32 val = __get_unaligned_cpu32(addr + 2);
 226
 227         val ^= macvlan_hash_mix(vlan);
 228         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 229 }
 230
 231 static void macvlan_broadcast(struct sk_buff *skb,
 232                               const struct macvlan_port *port,
 233                               struct net_device *src,
 234                               enum macvlan_mode mode)
 235 {
 236         const struct ethhdr *eth = eth_hdr(skb);
 237         const struct macvlan_dev *vlan;
 238         struct sk_buff *nskb;
 239         unsigned int i;
 240         int err;
 241         unsigned int hash;
 242
 243         if (skb->protocol == htons(ETH_P_PAUSE))
 244                 return;
 245
 246         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 247                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 248                         if (vlan->dev == src || !(vlan->mode & mode))
 249                                 continue;
 250
 251                         hash = mc_hash(vlan, eth->h_dest);
 252                         if (!test_bit(hash, vlan->mc_filter))
 253                                 continue;
 254
 255                         err = NET_RX_DROP;
 256                         nskb = skb_clone(skb, GFP_ATOMIC);
 257                         if (likely(nskb))
 258                                 err = macvlan_broadcast_one(
 259                                         nskb, vlan, eth,
 260                                         mode == MACVLAN_MODE_BRIDGE) ?:
 261                                       netif_rx_ni(nskb);
 262                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 263                                          err == NET_RX_SUCCESS, true);
 264                 }
 265         }
 266 }
 267
 268 static void macvlan_process_broadcast(struct work_struct *w)
 269 {
 270         struct macvlan_port *port = container_of(w, struct macvlan_port,
 271                                                  bc_work);
 272         struct sk_buff *skb;
 273         struct sk_buff_head list;
 274
 275         __skb_queue_head_init(&list);
 276
 277         spin_lock_bh(&port->bc_queue.lock);
 278         skb_queue_splice_tail_init(&port->bc_queue, &list);
 279         spin_unlock_bh(&port->bc_queue.lock);
 280
 281         while ((skb = __skb_dequeue(&list))) {
 282                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
 283
 284                 rcu_read_lock();
 285
 286                 if (!src)
 287                         /* frame comes from an external address */
 288                         macvlan_broadcast(skb, port, NULL,
 289                                           MACVLAN_MODE_PRIVATE |
 290                                           MACVLAN_MODE_VEPA    |
 291                                           MACVLAN_MODE_PASSTHRU|
 292                                           MACVLAN_MODE_BRIDGE);
 293                 else if (src->mode == MACVLAN_MODE_VEPA)
 294                         /* flood to everyone except source */
 295                         macvlan_broadcast(skb, port, src->dev,
 296                                           MACVLAN_MODE_VEPA |
 297                                           MACVLAN_MODE_BRIDGE);
 298                 else
 299                         /*
 300                          * flood only to VEPA ports, bridge ports
 301                          * already saw the frame on the way out.
 302                          */
 303                         macvlan_broadcast(skb, port, src->dev,
 304                                           MACVLAN_MODE_VEPA);
 305
 306                 rcu_read_unlock();
 307
 308                 if (src)
 309                         dev_put(src->dev);
 310                 kfree_skb(skb);
 311
 312                 cond_resched();
 313         }
 314 }
 315
 316 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 317                                       const struct macvlan_dev *src,
 318                                       struct sk_buff *skb)
 319 {
 320         struct sk_buff *nskb;
 321         int err = -ENOMEM;
 322
 323         nskb = skb_clone(skb, GFP_ATOMIC);
 324         if (!nskb)
 325                 goto err;
 326
 327         MACVLAN_SKB_CB(nskb)->src = src;
 328
 329         spin_lock(&port->bc_queue.lock);
 330         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 331                 if (src)
 332                         dev_hold(src->dev);
 333                 __skb_queue_tail(&port->bc_queue, nskb);
 334                 err = 0;
 335         }
 336         spin_unlock(&port->bc_queue.lock);
 337
 338         schedule_work(&port->bc_work);
 339
 340         if (err)
 341                 goto free_nskb;
 342
 343         return;
 344
 345 free_nskb:
 346         kfree_skb(nskb);
 347 err:
 348         atomic_long_inc(&skb->dev->rx_dropped);
 349 }
 350
 351 static void macvlan_flush_sources(struct macvlan_port *port,
 352                                   struct macvlan_dev *vlan)
 353 {
 354         int i;
 355
 356         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 357                 struct hlist_node *h, *n;
 358
 359                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 360                         struct macvlan_source_entry *entry;
 361
 362                         entry = hlist_entry(h, struct macvlan_source_entry,
 363                                             hlist);
 364                         if (entry->vlan == vlan)
 365                                 macvlan_hash_del_source(entry);
 366                 }
 367         }
 368         vlan->macaddr_count = 0;
 369 }
 370
 371 static void macvlan_forward_source_one(struct sk_buff *skb,
 372                                        struct macvlan_dev *vlan)
 373 {
 374         struct sk_buff *nskb;
 375         struct net_device *dev;
 376         int len;
 377         int ret;
 378
 379         dev = vlan->dev;
 380         if (unlikely(!(dev->flags & IFF_UP)))
 381                 return;
 382
 383         nskb = skb_clone(skb, GFP_ATOMIC);
 384         if (!nskb)
 385                 return;
 386
 387         len = nskb->len + ETH_HLEN;
 388         nskb->dev = dev;
 389         nskb->pkt_type = PACKET_HOST;
 390
 391         ret = netif_rx(nskb);
 392         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 393 }
 394
 395 static void macvlan_forward_source(struct sk_buff *skb,
 396                                    struct macvlan_port *port,
 397                                    const unsigned char *addr)
 398 {
 399         struct macvlan_source_entry *entry;
 400         u32 idx = macvlan_eth_hash(addr);
 401         struct hlist_head *h = &port->vlan_source_hash[idx];
 402
 403         hlist_for_each_entry_rcu(entry, h, hlist) {
 404                 if (ether_addr_equal_64bits(entry->addr, addr))
 405                         if (entry->vlan->dev->flags & IFF_UP)
 406                                 macvlan_forward_source_one(skb, entry->vlan);
 407         }
 408 }
 409
 410 /* called under rcu_read_lock() from netif_receive_skb */
 411 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 412 {
 413         struct macvlan_port *port;
 414         struct sk_buff *skb = *pskb;
 415         const struct ethhdr *eth = eth_hdr(skb);
 416         const struct macvlan_dev *vlan;
 417         const struct macvlan_dev *src;
 418         struct net_device *dev;
 419         unsigned int len = 0;
 420         int ret;
 421         rx_handler_result_t handle_res;
 422
 423         /* Packets from dev_loopback_xmit() do not have L2 header, bail out */
 424         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 425                 return RX_HANDLER_PASS;
 426
 427         port = macvlan_port_get_rcu(skb->dev);
 428         if (is_multicast_ether_addr(eth->h_dest)) {
 429                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 430                 if (!skb)
 431                         return RX_HANDLER_CONSUMED;
 432                 *pskb = skb;
 433                 eth = eth_hdr(skb);
 434                 macvlan_forward_source(skb, port, eth->h_source);
 435                 src = macvlan_hash_lookup(port, eth->h_source);
 436                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 437                     src->mode != MACVLAN_MODE_BRIDGE) {
 438                         /* forward to original port. */
 439                         vlan = src;
 440                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 441                               netif_rx(skb);
 442                         handle_res = RX_HANDLER_CONSUMED;
 443                         goto out;
 444                 }
 445
 446                 macvlan_broadcast_enqueue(port, src, skb);
 447
 448                 return RX_HANDLER_PASS;
 449         }
 450
 451         macvlan_forward_source(skb, port, eth->h_source);
 452         if (port->passthru)
 453                 vlan = list_first_or_null_rcu(&port->vlans,
 454                                               struct macvlan_dev, list);
 455         else
 456                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 457         if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
 458                 return RX_HANDLER_PASS;
 459
 460         dev = vlan->dev;
 461         if (unlikely(!(dev->flags & IFF_UP))) {
 462                 kfree_skb(skb);
 463                 return RX_HANDLER_CONSUMED;
 464         }
 465         len = skb->len + ETH_HLEN;
 466         skb = skb_share_check(skb, GFP_ATOMIC);
 467         if (!skb) {
 468                 ret = NET_RX_DROP;
 469                 handle_res = RX_HANDLER_CONSUMED;
 470                 goto out;
 471         }
 472
 473         *pskb = skb;
 474         skb->dev = dev;
 475         skb->pkt_type = PACKET_HOST;
 476
 477         ret = NET_RX_SUCCESS;
 478         handle_res = RX_HANDLER_ANOTHER;
 479 out:
 480         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 481         return handle_res;
 482 }
 483
 484 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 485 {
 486         const struct macvlan_dev *vlan = netdev_priv(dev);
 487         const struct macvlan_port *port = vlan->port;
 488         const struct macvlan_dev *dest;
 489
 490         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 491                 const struct ethhdr *eth = skb_eth_hdr(skb);
 492
 493                 /* send to other bridge ports directly */
 494                 if (is_multicast_ether_addr(eth->h_dest)) {
 495                         skb_reset_mac_header(skb);
 496                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 497                         goto xmit_world;
 498                 }
 499
 500                 dest = macvlan_hash_lookup(port, eth->h_dest);
 501                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 502                         /* send to lowerdev first for its network taps */
 503                         dev_forward_skb(vlan->lowerdev, skb);
 504
 505                         return NET_XMIT_SUCCESS;
 506                 }
 507         }
 508
 509 xmit_world:
 510         skb->dev = vlan->lowerdev;
 511         return dev_queue_xmit(skb);
 512 }
 513
 514 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 515 {
 516 #ifdef CONFIG_NET_POLL_CONTROLLER
 517         if (vlan->netpoll)
 518                 netpoll_send_skb(vlan->netpoll, skb);
 519 #else
 520         BUG();
 521 #endif
 522         return NETDEV_TX_OK;
 523 }
 524
 525 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 526                                       struct net_device *dev)
 527 {
 528         unsigned int len = skb->len;
 529         int ret;
 530         struct macvlan_dev *vlan = netdev_priv(dev);
 531
 532         if (unlikely(netpoll_tx_running(dev)))
 533                 return macvlan_netpoll_send_skb(vlan, skb);
 534
 535         if (vlan->fwd_priv) {
 536                 skb->dev = vlan->lowerdev;
 537                 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
 538         } else {
 539                 ret = macvlan_queue_xmit(skb, dev);
 540         }
 541
 542         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 543                 struct vlan_pcpu_stats *pcpu_stats;
 544
 545                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 546                 u64_stats_update_begin(&pcpu_stats->syncp);
 547                 pcpu_stats->tx_packets++;
 548                 pcpu_stats->tx_bytes += len;
 549                 u64_stats_update_end(&pcpu_stats->syncp);
 550         } else {
 551                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 552         }
 553         return ret;
 554 }
 555
 556 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 557                                unsigned short type, const void *daddr,
 558                                const void *saddr, unsigned len)
 559 {
 560         const struct macvlan_dev *vlan = netdev_priv(dev);
 561         struct net_device *lowerdev = vlan->lowerdev;
 562
 563         return dev_hard_header(skb, lowerdev, type, daddr,
 564                                saddr ? : dev->dev_addr, len);
 565 }
 566
 567 static const struct header_ops macvlan_hard_header_ops = {
 568         .create         = macvlan_hard_header,
 569         .parse          = eth_header_parse,
 570         .cache          = eth_header_cache,
 571         .cache_update   = eth_header_cache_update,
 572 };
 573
 574 static struct rtnl_link_ops macvlan_link_ops;
 575
 576 static int macvlan_open(struct net_device *dev)
 577 {
 578         struct macvlan_dev *vlan = netdev_priv(dev);
 579         struct net_device *lowerdev = vlan->lowerdev;
 580         int err;
 581
 582         if (vlan->port->passthru) {
 583                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 584                         err = dev_set_promiscuity(lowerdev, 1);
 585                         if (err < 0)
 586                                 goto out;
 587                 }
 588                 goto hash_add;
 589         }
 590
 591         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
 592             dev->rtnl_link_ops == &macvlan_link_ops) {
 593                 vlan->fwd_priv =
 594                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 595
 596                 /* If we get a NULL pointer back, or if we get an error
 597                  * then we should just fall through to the non accelerated path
 598                  */
 599                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
 600                         vlan->fwd_priv = NULL;
 601                 } else
 602                         return 0;
 603         }
 604
 605         err = -EBUSY;
 606         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 607                 goto out;
 608
 609         err = dev_uc_add(lowerdev, dev->dev_addr);
 610         if (err < 0)
 611                 goto out;
 612         if (dev->flags & IFF_ALLMULTI) {
 613                 err = dev_set_allmulti(lowerdev, 1);
 614                 if (err < 0)
 615                         goto del_unicast;
 616         }
 617
 618         if (dev->flags & IFF_PROMISC) {
 619                 err = dev_set_promiscuity(lowerdev, 1);
 620                 if (err < 0)
 621                         goto clear_multi;
 622         }
 623
 624 hash_add:
 625         macvlan_hash_add(vlan);
 626         return 0;
 627
 628 clear_multi:
 629         dev_set_allmulti(lowerdev, -1);
 630 del_unicast:
 631         dev_uc_del(lowerdev, dev->dev_addr);
 632 out:
 633         if (vlan->fwd_priv) {
 634                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 635                                                            vlan->fwd_priv);
 636                 vlan->fwd_priv = NULL;
 637         }
 638         return err;
 639 }
 640
 641 static int macvlan_stop(struct net_device *dev)
 642 {
 643         struct macvlan_dev *vlan = netdev_priv(dev);
 644         struct net_device *lowerdev = vlan->lowerdev;
 645
 646         if (vlan->fwd_priv) {
 647                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 648                                                            vlan->fwd_priv);
 649                 vlan->fwd_priv = NULL;
 650                 return 0;
 651         }
 652
 653         dev_uc_unsync(lowerdev, dev);
 654         dev_mc_unsync(lowerdev, dev);
 655
 656         if (vlan->port->passthru) {
 657                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 658                         dev_set_promiscuity(lowerdev, -1);
 659                 goto hash_del;
 660         }
 661
 662         if (dev->flags & IFF_ALLMULTI)
 663                 dev_set_allmulti(lowerdev, -1);
 664
 665         if (dev->flags & IFF_PROMISC)
 666                 dev_set_promiscuity(lowerdev, -1);
 667
 668         dev_uc_del(lowerdev, dev->dev_addr);
 669
 670 hash_del:
 671         macvlan_hash_del(vlan, !dev->dismantle);
 672         return 0;
 673 }
 674
 675 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 676 {
 677         struct macvlan_dev *vlan = netdev_priv(dev);
 678         struct net_device *lowerdev = vlan->lowerdev;
 679         int err;
 680
 681         if (!(dev->flags & IFF_UP)) {
 682                 /* Just copy in the new address */
 683                 ether_addr_copy(dev->dev_addr, addr);
 684         } else {
 685                 /* Rehash and update the device filters */
 686                 if (macvlan_addr_busy(vlan->port, addr))
 687                         return -EBUSY;
 688
 689                 if (!vlan->port->passthru) {
 690                         err = dev_uc_add(lowerdev, addr);
 691                         if (err)
 692                                 return err;
 693
 694                         dev_uc_del(lowerdev, dev->dev_addr);
 695                 }
 696
 697                 macvlan_hash_change_addr(vlan, addr);
 698         }
 699         return 0;
 700 }
 701
 702 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 703 {
 704         struct macvlan_dev *vlan = netdev_priv(dev);
 705         struct sockaddr *addr = p;
 706
 707         if (!is_valid_ether_addr(addr->sa_data))
 708                 return -EADDRNOTAVAIL;
 709
 710         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 711                 dev_set_mac_address(vlan->lowerdev, addr);
 712                 return 0;
 713         }
 714
 715         return macvlan_sync_address(dev, addr->sa_data);
 716 }
 717
 718 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 719 {
 720         struct macvlan_dev *vlan = netdev_priv(dev);
 721         struct net_device *lowerdev = vlan->lowerdev;
 722
 723         if (dev->flags & IFF_UP) {
 724                 if (change & IFF_ALLMULTI)
 725                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 726                 if (change & IFF_PROMISC)
 727                         dev_set_promiscuity(lowerdev,
 728                                             dev->flags & IFF_PROMISC ? 1 : -1);
 729
 730         }
 731 }
 732
 733 static void macvlan_set_mac_lists(struct net_device *dev)
 734 {
 735         struct macvlan_dev *vlan = netdev_priv(dev);
 736
 737         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 738                 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
 739         } else {
 740                 struct netdev_hw_addr *ha;
 741                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 742
 743                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 744                 netdev_for_each_mc_addr(ha, dev) {
 745                         __set_bit(mc_hash(vlan, ha->addr), filter);
 746                 }
 747
 748                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 749
 750                 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 751         }
 752         dev_uc_sync(vlan->lowerdev, dev);
 753         dev_mc_sync(vlan->lowerdev, dev);
 754 }
 755
 756 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 757 {
 758         struct macvlan_dev *vlan = netdev_priv(dev);
 759
 760         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 761                 return -EINVAL;
 762         dev->mtu = new_mtu;
 763         return 0;
 764 }
 765
 766 /*
 767  * macvlan network devices have devices nesting below it and are a special
 768  * "super class" of normal network devices; split their locks off into a
 769  * separate class since they always nest.
 770  */
 771 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 772 static struct lock_class_key macvlan_netdev_addr_lock_key;
 773
 774 #define ALWAYS_ON_FEATURES \
 775         (NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX | \
 776          NETIF_F_GSO_ROBUST)
 777
 778 #define MACVLAN_FEATURES \
 779         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 780          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
 781          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 782          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 783
 784 #define MACVLAN_STATE_MASK \
 785         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 786
 787 static int macvlan_get_nest_level(struct net_device *dev)
 788 {
 789         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 790 }
 791
 792 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 793                                           struct netdev_queue *txq,
 794                                           void *_unused)
 795 {
 796         lockdep_set_class(&txq->_xmit_lock,
 797                           &macvlan_netdev_xmit_lock_key);
 798 }
 799
 800 static void macvlan_set_lockdep_class(struct net_device *dev)
 801 {
 802         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 803                                        &macvlan_netdev_addr_lock_key,
 804                                        macvlan_get_nest_level(dev));
 805         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 806 }
 807
 808 static int macvlan_init(struct net_device *dev)
 809 {
 810         struct macvlan_dev *vlan = netdev_priv(dev);
 811         const struct net_device *lowerdev = vlan->lowerdev;
 812
 813         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 814                                   (lowerdev->state & MACVLAN_STATE_MASK);
 815         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 816         dev->features           |= ALWAYS_ON_FEATURES;
 817         dev->hw_features        |= NETIF_F_LRO;
 818         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 819         dev->gso_max_size       = lowerdev->gso_max_size;
 820         dev->hard_header_len    = lowerdev->hard_header_len;
 821
 822         macvlan_set_lockdep_class(dev);
 823
 824         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 825         if (!vlan->pcpu_stats)
 826                 return -ENOMEM;
 827
 828         return 0;
 829 }
 830
 831 static void macvlan_uninit(struct net_device *dev)
 832 {
 833         struct macvlan_dev *vlan = netdev_priv(dev);
 834         struct macvlan_port *port = vlan->port;
 835
 836         free_percpu(vlan->pcpu_stats);
 837
 838         macvlan_flush_sources(port, vlan);
 839         port->count -= 1;
 840         if (!port->count)
 841                 macvlan_port_destroy(port->dev);
 842 }
 843
 844 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 845                                                          struct rtnl_link_stats64 *stats)
 846 {
 847         struct macvlan_dev *vlan = netdev_priv(dev);
 848
 849         if (vlan->pcpu_stats) {
 850                 struct vlan_pcpu_stats *p;
 851                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 852                 u32 rx_errors = 0, tx_dropped = 0;
 853                 unsigned int start;
 854                 int i;
 855
 856                 for_each_possible_cpu(i) {
 857                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 858                         do {
 859                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 860                                 rx_packets      = p->rx_packets;
 861                                 rx_bytes        = p->rx_bytes;
 862                                 rx_multicast    = p->rx_multicast;
 863                                 tx_packets      = p->tx_packets;
 864                                 tx_bytes        = p->tx_bytes;
 865                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 866
 867                         stats->rx_packets       += rx_packets;
 868                         stats->rx_bytes         += rx_bytes;
 869                         stats->multicast        += rx_multicast;
 870                         stats->tx_packets       += tx_packets;
 871                         stats->tx_bytes         += tx_bytes;
 872                         /* rx_errors & tx_dropped are u32, updated
 873                          * without syncp protection.
 874                          */
 875                         rx_errors       += p->rx_errors;
 876                         tx_dropped      += p->tx_dropped;
 877                 }
 878                 stats->rx_errors        = rx_errors;
 879                 stats->rx_dropped       = rx_errors;
 880                 stats->tx_dropped       = tx_dropped;
 881         }
 882         return stats;
 883 }
 884
 885 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 886                                    __be16 proto, u16 vid)
 887 {
 888         struct macvlan_dev *vlan = netdev_priv(dev);
 889         struct net_device *lowerdev = vlan->lowerdev;
 890
 891         return vlan_vid_add(lowerdev, proto, vid);
 892 }
 893
 894 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 895                                     __be16 proto, u16 vid)
 896 {
 897         struct macvlan_dev *vlan = netdev_priv(dev);
 898         struct net_device *lowerdev = vlan->lowerdev;
 899
 900         vlan_vid_del(lowerdev, proto, vid);
 901         return 0;
 902 }
 903
 904 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 905                            struct net_device *dev,
 906                            const unsigned char *addr, u16 vid,
 907                            u16 flags)
 908 {
 909         struct macvlan_dev *vlan = netdev_priv(dev);
 910         int err = -EINVAL;
 911
 912         /* Support unicast filter only on passthru devices.
 913          * Multicast filter should be allowed on all devices.
 914          */
 915         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 916                 return -EOPNOTSUPP;
 917
 918         if (flags & NLM_F_REPLACE)
 919                 return -EOPNOTSUPP;
 920
 921         if (is_unicast_ether_addr(addr))
 922                 err = dev_uc_add_excl(dev, addr);
 923         else if (is_multicast_ether_addr(addr))
 924                 err = dev_mc_add_excl(dev, addr);
 925
 926         return err;
 927 }
 928
 929 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 930                            struct net_device *dev,
 931                            const unsigned char *addr, u16 vid)
 932 {
 933         struct macvlan_dev *vlan = netdev_priv(dev);
 934         int err = -EINVAL;
 935
 936         /* Support unicast filter only on passthru devices.
 937          * Multicast filter should be allowed on all devices.
 938          */
 939         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 940                 return -EOPNOTSUPP;
 941
 942         if (is_unicast_ether_addr(addr))
 943                 err = dev_uc_del(dev, addr);
 944         else if (is_multicast_ether_addr(addr))
 945                 err = dev_mc_del(dev, addr);
 946
 947         return err;
 948 }
 949
 950 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 951                                         struct ethtool_drvinfo *drvinfo)
 952 {
 953         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 954         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 955 }
 956
 957 static int macvlan_ethtool_get_settings(struct net_device *dev,
 958                                         struct ethtool_cmd *cmd)
 959 {
 960         const struct macvlan_dev *vlan = netdev_priv(dev);
 961
 962         return __ethtool_get_settings(vlan->lowerdev, cmd);
 963 }
 964
 965 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 966                                               netdev_features_t features)
 967 {
 968         struct macvlan_dev *vlan = netdev_priv(dev);
 969         netdev_features_t lowerdev_features = vlan->lowerdev->features;
 970         netdev_features_t mask;
 971
 972         features |= NETIF_F_ALL_FOR_ALL;
 973         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 974         mask = features;
 975
 976         lowerdev_features &= (features | ~NETIF_F_LRO);
 977         features = netdev_increment_features(lowerdev_features, features, mask);
 978         features |= ALWAYS_ON_FEATURES;
 979         features &= ~NETIF_F_NETNS_LOCAL;
 980
 981         return features;
 982 }
 983
 984 #ifdef CONFIG_NET_POLL_CONTROLLER
 985 static void macvlan_dev_poll_controller(struct net_device *dev)
 986 {
 987         return;
 988 }
 989
 990 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
 991 {
 992         struct macvlan_dev *vlan = netdev_priv(dev);
 993         struct net_device *real_dev = vlan->lowerdev;
 994         struct netpoll *netpoll;
 995         int err = 0;
 996
 997         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
 998         err = -ENOMEM;
 999         if (!netpoll)
1000                 goto out;
1001
1002         err = __netpoll_setup(netpoll, real_dev);
1003         if (err) {
1004                 kfree(netpoll);
1005                 goto out;
1006         }
1007
1008         vlan->netpoll = netpoll;
1009
1010 out:
1011         return err;
1012 }
1013
1014 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1015 {
1016         struct macvlan_dev *vlan = netdev_priv(dev);
1017         struct netpoll *netpoll = vlan->netpoll;
1018
1019         if (!netpoll)
1020                 return;
1021
1022         vlan->netpoll = NULL;
1023
1024         __netpoll_free_async(netpoll);
1025 }
1026 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1027
1028 static int macvlan_dev_get_iflink(const struct net_device *dev)
1029 {
1030         struct macvlan_dev *vlan = netdev_priv(dev);
1031
1032         return vlan->lowerdev->ifindex;
1033 }
1034
1035 static const struct ethtool_ops macvlan_ethtool_ops = {
1036         .get_link               = ethtool_op_get_link,
1037         .get_settings           = macvlan_ethtool_get_settings,
1038         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1039 };
1040
1041 static const struct net_device_ops macvlan_netdev_ops = {
1042         .ndo_init               = macvlan_init,
1043         .ndo_uninit             = macvlan_uninit,
1044         .ndo_open               = macvlan_open,
1045         .ndo_stop               = macvlan_stop,
1046         .ndo_start_xmit         = macvlan_start_xmit,
1047         .ndo_change_mtu         = macvlan_change_mtu,
1048         .ndo_fix_features       = macvlan_fix_features,
1049         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1050         .ndo_set_mac_address    = macvlan_set_mac_address,
1051         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1052         .ndo_get_stats64        = macvlan_dev_get_stats64,
1053         .ndo_validate_addr      = eth_validate_addr,
1054         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1055         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1056         .ndo_fdb_add            = macvlan_fdb_add,
1057         .ndo_fdb_del            = macvlan_fdb_del,
1058         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1059         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1060 #ifdef CONFIG_NET_POLL_CONTROLLER
1061         .ndo_poll_controller    = macvlan_dev_poll_controller,
1062         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1063         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1064 #endif
1065         .ndo_get_iflink         = macvlan_dev_get_iflink,
1066         .ndo_features_check     = passthru_features_check,
1067 };
1068
1069 void macvlan_common_setup(struct net_device *dev)
1070 {
1071         ether_setup(dev);
1072
1073         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1074         netif_keep_dst(dev);
1075         dev->priv_flags        |= IFF_UNICAST_FLT;
1076         dev->netdev_ops         = &macvlan_netdev_ops;
1077         dev->destructor         = free_netdev;
1078         dev->header_ops         = &macvlan_hard_header_ops;
1079         dev->ethtool_ops        = &macvlan_ethtool_ops;
1080 }
1081 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1082
1083 static void macvlan_setup(struct net_device *dev)
1084 {
1085         macvlan_common_setup(dev);
1086         dev->tx_queue_len       = 0;
1087 }
1088
1089 static int macvlan_port_create(struct net_device *dev)
1090 {
1091         struct macvlan_port *port;
1092         unsigned int i;
1093         int err;
1094
1095         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1096                 return -EINVAL;
1097
1098         if (netif_is_ipvlan_port(dev))
1099                 return -EBUSY;
1100
1101         port = kzalloc(sizeof(*port), GFP_KERNEL);
1102         if (port == NULL)
1103                 return -ENOMEM;
1104
1105         port->passthru = false;
1106         port->dev = dev;
1107         INIT_LIST_HEAD(&port->vlans);
1108         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1109                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1110         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1111                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1112
1113         skb_queue_head_init(&port->bc_queue);
1114         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1115
1116         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1117         if (err)
1118                 kfree(port);
1119         else
1120                 dev->priv_flags |= IFF_MACVLAN_PORT;
1121         return err;
1122 }
1123
1124 static void macvlan_port_destroy(struct net_device *dev)
1125 {
1126         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1127         struct sk_buff *skb;
1128
1129         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1130         netdev_rx_handler_unregister(dev);
1131
1132         /* After this point, no packet can schedule bc_work anymore,
1133          * but we need to cancel it and purge left skbs if any.
1134          */
1135         cancel_work_sync(&port->bc_work);
1136
1137         while ((skb = __skb_dequeue(&port->bc_queue))) {
1138                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1139
1140                 if (src)
1141                         dev_put(src->dev);
1142
1143                 kfree_skb(skb);
1144         }
1145
1146         kfree_rcu(port, rcu);
1147 }
1148
1149 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
1150 {
1151         if (tb[IFLA_ADDRESS]) {
1152                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1153                         return -EINVAL;
1154                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1155                         return -EADDRNOTAVAIL;
1156         }
1157
1158         if (data && data[IFLA_MACVLAN_FLAGS] &&
1159             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1160                 return -EINVAL;
1161
1162         if (data && data[IFLA_MACVLAN_MODE]) {
1163                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1164                 case MACVLAN_MODE_PRIVATE:
1165                 case MACVLAN_MODE_VEPA:
1166                 case MACVLAN_MODE_BRIDGE:
1167                 case MACVLAN_MODE_PASSTHRU:
1168                 case MACVLAN_MODE_SOURCE:
1169                         break;
1170                 default:
1171                         return -EINVAL;
1172                 }
1173         }
1174
1175         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1176                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1177                 case MACVLAN_MACADDR_ADD:
1178                 case MACVLAN_MACADDR_DEL:
1179                 case MACVLAN_MACADDR_FLUSH:
1180                 case MACVLAN_MACADDR_SET:
1181                         break;
1182                 default:
1183                         return -EINVAL;
1184                 }
1185         }
1186
1187         if (data && data[IFLA_MACVLAN_MACADDR]) {
1188                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1189                         return -EINVAL;
1190
1191                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1192                         return -EADDRNOTAVAIL;
1193         }
1194
1195         if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1196                 return -EINVAL;
1197
1198         return 0;
1199 }
1200
1201 /**
1202  * reconfigure list of remote source mac address
1203  * (only for macvlan devices in source mode)
1204  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1205  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1206  */
1207 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1208                                       struct nlattr *data[])
1209 {
1210         char *addr = NULL;
1211         int ret, rem, len;
1212         struct nlattr *nla, *head;
1213         struct macvlan_source_entry *entry;
1214
1215         if (data[IFLA_MACVLAN_MACADDR])
1216                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1217
1218         if (mode == MACVLAN_MACADDR_ADD) {
1219                 if (!addr)
1220                         return -EINVAL;
1221
1222                 return macvlan_hash_add_source(vlan, addr);
1223
1224         } else if (mode == MACVLAN_MACADDR_DEL) {
1225                 if (!addr)
1226                         return -EINVAL;
1227
1228                 entry = macvlan_hash_lookup_source(vlan, addr);
1229                 if (entry) {
1230                         macvlan_hash_del_source(entry);
1231                         vlan->macaddr_count--;
1232                 }
1233         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1234                 macvlan_flush_sources(vlan->port, vlan);
1235         } else if (mode == MACVLAN_MACADDR_SET) {
1236                 macvlan_flush_sources(vlan->port, vlan);
1237
1238                 if (addr) {
1239                         ret = macvlan_hash_add_source(vlan, addr);
1240                         if (ret)
1241                                 return ret;
1242                 }
1243
1244                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1245                         return 0;
1246
1247                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1248                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1249
1250                 nla_for_each_attr(nla, head, len, rem) {
1251                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1252                             nla_len(nla) != ETH_ALEN)
1253                                 continue;
1254
1255                         addr = nla_data(nla);
1256                         ret = macvlan_hash_add_source(vlan, addr);
1257                         if (ret)
1258                                 return ret;
1259                 }
1260         } else {
1261                 return -EINVAL;
1262         }
1263
1264         return 0;
1265 }
1266
1267 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1268                            struct nlattr *tb[], struct nlattr *data[])
1269 {
1270         struct macvlan_dev *vlan = netdev_priv(dev);
1271         struct macvlan_port *port;
1272         struct net_device *lowerdev;
1273         int err;
1274         int macmode;
1275
1276         if (!tb[IFLA_LINK])
1277                 return -EINVAL;
1278
1279         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1280         if (lowerdev == NULL)
1281                 return -ENODEV;
1282
1283         /* When creating macvlans or macvtaps on top of other macvlans - use
1284          * the real device as the lowerdev.
1285          */
1286         if (netif_is_macvlan(lowerdev))
1287                 lowerdev = macvlan_dev_real_dev(lowerdev);
1288
1289         if (!tb[IFLA_MTU])
1290                 dev->mtu = lowerdev->mtu;
1291         else if (dev->mtu > lowerdev->mtu)
1292                 return -EINVAL;
1293
1294         if (!tb[IFLA_ADDRESS])
1295                 eth_hw_addr_random(dev);
1296
1297         if (!macvlan_port_exists(lowerdev)) {
1298                 err = macvlan_port_create(lowerdev);
1299                 if (err < 0)
1300                         return err;
1301         }
1302         port = macvlan_port_get_rtnl(lowerdev);
1303
1304         /* Only 1 macvlan device can be created in passthru mode */
1305         if (port->passthru)
1306                 return -EINVAL;
1307
1308         vlan->lowerdev = lowerdev;
1309         vlan->dev      = dev;
1310         vlan->port     = port;
1311         vlan->set_features = MACVLAN_FEATURES;
1312         vlan->nest_level = dev_get_nest_level(lowerdev, netif_is_macvlan) + 1;
1313
1314         vlan->mode     = MACVLAN_MODE_VEPA;
1315         if (data && data[IFLA_MACVLAN_MODE])
1316                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1317
1318         if (data && data[IFLA_MACVLAN_FLAGS])
1319                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1320
1321         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1322                 if (port->count)
1323                         return -EINVAL;
1324                 port->passthru = true;
1325                 eth_hw_addr_inherit(dev, lowerdev);
1326         }
1327
1328         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1329                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1330                         return -EINVAL;
1331                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1332                 err = macvlan_changelink_sources(vlan, macmode, data);
1333                 if (err)
1334                         return err;
1335         }
1336
1337         port->count += 1;
1338         err = register_netdevice(dev);
1339         if (err < 0)
1340                 goto destroy_port;
1341
1342         dev->priv_flags |= IFF_MACVLAN;
1343         err = netdev_upper_dev_link(lowerdev, dev);
1344         if (err)
1345                 goto unregister_netdev;
1346
1347         list_add_tail_rcu(&vlan->list, &port->vlans);
1348         netif_stacked_transfer_operstate(lowerdev, dev);
1349
1350         return 0;
1351
1352 unregister_netdev:
1353         unregister_netdevice(dev);
1354 destroy_port:
1355         port->count -= 1;
1356         if (!port->count)
1357                 macvlan_port_destroy(lowerdev);
1358
1359         return err;
1360 }
1361 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1362
1363 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1364                            struct nlattr *tb[], struct nlattr *data[])
1365 {
1366         return macvlan_common_newlink(src_net, dev, tb, data);
1367 }
1368
1369 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1370 {
1371         struct macvlan_dev *vlan = netdev_priv(dev);
1372
1373         if (vlan->mode == MACVLAN_MODE_SOURCE)
1374                 macvlan_flush_sources(vlan->port, vlan);
1375         list_del_rcu(&vlan->list);
1376         unregister_netdevice_queue(dev, head);
1377         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1378 }
1379 EXPORT_SYMBOL_GPL(macvlan_dellink);
1380
1381 static int macvlan_changelink(struct net_device *dev,
1382                 struct nlattr *tb[], struct nlattr *data[])
1383 {
1384         struct macvlan_dev *vlan = netdev_priv(dev);
1385         enum macvlan_mode mode;
1386         bool set_mode = false;
1387         enum macvlan_macaddr_mode macmode;
1388         int ret;
1389
1390         /* Validate mode, but don't set yet: setting flags may fail. */
1391         if (data && data[IFLA_MACVLAN_MODE]) {
1392                 set_mode = true;
1393                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1394                 /* Passthrough mode can't be set or cleared dynamically */
1395                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1396                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1397                         return -EINVAL;
1398                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1399                     vlan->mode != mode)
1400                         macvlan_flush_sources(vlan->port, vlan);
1401         }
1402
1403         if (data && data[IFLA_MACVLAN_FLAGS]) {
1404                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1405                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1406                 if (vlan->port->passthru && promisc) {
1407                         int err;
1408
1409                         if (flags & MACVLAN_FLAG_NOPROMISC)
1410                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1411                         else
1412                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1413                         if (err < 0)
1414                                 return err;
1415                 }
1416                 vlan->flags = flags;
1417         }
1418         if (set_mode)
1419                 vlan->mode = mode;
1420         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1421                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1422                         return -EINVAL;
1423                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1424                 ret = macvlan_changelink_sources(vlan, macmode, data);
1425                 if (ret)
1426                         return ret;
1427         }
1428         return 0;
1429 }
1430
1431 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1432 {
1433         if (vlan->macaddr_count == 0)
1434                 return 0;
1435         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1436                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1437 }
1438
1439 static size_t macvlan_get_size(const struct net_device *dev)
1440 {
1441         struct macvlan_dev *vlan = netdev_priv(dev);
1442
1443         return (0
1444                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1445                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1446                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1447                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1448                 );
1449 }
1450
1451 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1452                                      const struct macvlan_dev *vlan,
1453                                      const int i)
1454 {
1455         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1456         struct macvlan_source_entry *entry;
1457
1458         hlist_for_each_entry_rcu(entry, h, hlist) {
1459                 if (entry->vlan != vlan)
1460                         continue;
1461                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1462                         return 1;
1463         }
1464         return 0;
1465 }
1466
1467 static int macvlan_fill_info(struct sk_buff *skb,
1468                                 const struct net_device *dev)
1469 {
1470         struct macvlan_dev *vlan = netdev_priv(dev);
1471         int i;
1472         struct nlattr *nest;
1473
1474         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1475                 goto nla_put_failure;
1476         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1477                 goto nla_put_failure;
1478         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1479                 goto nla_put_failure;
1480         if (vlan->macaddr_count > 0) {
1481                 nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1482                 if (nest == NULL)
1483                         goto nla_put_failure;
1484
1485                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1486                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1487                                 goto nla_put_failure;
1488                 }
1489                 nla_nest_end(skb, nest);
1490         }
1491         return 0;
1492
1493 nla_put_failure:
1494         return -EMSGSIZE;
1495 }
1496
1497 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1498         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1499         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1500         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1501         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1502         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1503         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1504 };
1505
1506 int macvlan_link_register(struct rtnl_link_ops *ops)
1507 {
1508         /* common fields */
1509         ops->priv_size          = sizeof(struct macvlan_dev);
1510         ops->validate           = macvlan_validate;
1511         ops->maxtype            = IFLA_MACVLAN_MAX;
1512         ops->policy             = macvlan_policy;
1513         ops->changelink         = macvlan_changelink;
1514         ops->get_size           = macvlan_get_size;
1515         ops->fill_info          = macvlan_fill_info;
1516
1517         return rtnl_link_register(ops);
1518 };
1519 EXPORT_SYMBOL_GPL(macvlan_link_register);
1520
1521 static struct net *macvlan_get_link_net(const struct net_device *dev)
1522 {
1523         return dev_net(macvlan_dev_real_dev(dev));
1524 }
1525
1526 static struct rtnl_link_ops macvlan_link_ops = {
1527         .kind           = "macvlan",
1528         .setup          = macvlan_setup,
1529         .newlink        = macvlan_newlink,
1530         .dellink        = macvlan_dellink,
1531         .get_link_net   = macvlan_get_link_net,
1532 };
1533
1534 static int macvlan_device_event(struct notifier_block *unused,
1535                                 unsigned long event, void *ptr)
1536 {
1537         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1538         struct macvlan_dev *vlan, *next;
1539         struct macvlan_port *port;
1540         LIST_HEAD(list_kill);
1541
1542         if (!macvlan_port_exists(dev))
1543                 return NOTIFY_DONE;
1544
1545         port = macvlan_port_get_rtnl(dev);
1546
1547         switch (event) {
1548         case NETDEV_CHANGE:
1549                 list_for_each_entry(vlan, &port->vlans, list)
1550                         netif_stacked_transfer_operstate(vlan->lowerdev,
1551                                                          vlan->dev);
1552                 break;
1553         case NETDEV_FEAT_CHANGE:
1554                 list_for_each_entry(vlan, &port->vlans, list) {
1555                         vlan->dev->gso_max_size = dev->gso_max_size;
1556                         netdev_update_features(vlan->dev);
1557                 }
1558                 break;
1559         case NETDEV_CHANGEMTU:
1560                 list_for_each_entry(vlan, &port->vlans, list) {
1561                         if (vlan->dev->mtu <= dev->mtu)
1562                                 continue;
1563                         dev_set_mtu(vlan->dev, dev->mtu);
1564                 }
1565                 break;
1566         case NETDEV_CHANGEADDR:
1567                 if (!port->passthru)
1568                         return NOTIFY_DONE;
1569
1570                 vlan = list_first_entry_or_null(&port->vlans,
1571                                                 struct macvlan_dev,
1572                                                 list);
1573
1574                 if (vlan && macvlan_sync_address(vlan->dev, dev->dev_addr))
1575                         return NOTIFY_BAD;
1576
1577                 break;
1578         case NETDEV_UNREGISTER:
1579                 /* twiddle thumbs on netns device moves */
1580                 if (dev->reg_state != NETREG_UNREGISTERING)
1581                         break;
1582
1583                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1584                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1585                 unregister_netdevice_many(&list_kill);
1586                 break;
1587         case NETDEV_PRE_TYPE_CHANGE:
1588                 /* Forbid underlaying device to change its type. */
1589                 return NOTIFY_BAD;
1590
1591         case NETDEV_NOTIFY_PEERS:
1592         case NETDEV_BONDING_FAILOVER:
1593         case NETDEV_RESEND_IGMP:
1594                 /* Propagate to all vlans */
1595                 list_for_each_entry(vlan, &port->vlans, list)
1596                         call_netdevice_notifiers(event, vlan->dev);
1597         }
1598         return NOTIFY_DONE;
1599 }
1600
1601 static struct notifier_block macvlan_notifier_block __read_mostly = {
1602         .notifier_call  = macvlan_device_event,
1603 };
1604
1605 static int __init macvlan_init_module(void)
1606 {
1607         int err;
1608
1609         register_netdevice_notifier(&macvlan_notifier_block);
1610
1611         err = macvlan_link_register(&macvlan_link_ops);
1612         if (err < 0)
1613                 goto err1;
1614         return 0;
1615 err1:
1616         unregister_netdevice_notifier(&macvlan_notifier_block);
1617         return err;
1618 }
1619
1620 static void __exit macvlan_cleanup_module(void)
1621 {
1622         rtnl_link_unregister(&macvlan_link_ops);
1623         unregister_netdevice_notifier(&macvlan_notifier_block);
1624 }
1625
1626 module_init(macvlan_init_module);
1627 module_exit(macvlan_cleanup_module);
1628
1629 MODULE_LICENSE("GPL");
1630 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1631 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1632 MODULE_ALIAS_RTNL_LINK("macvlan");