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