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         DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
  53 };
  54
  55 struct macvlan_source_entry {
  56         struct hlist_node       hlist;
  57         struct macvlan_dev      *vlan;
  58         unsigned char           addr[6+2] __aligned(sizeof(u16));
  59         struct rcu_head         rcu;
  60 };
  61
  62 struct macvlan_skb_cb {
  63         const struct macvlan_dev *src;
  64 };
  65
  66 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
  67
  68 static void macvlan_port_destroy(struct net_device *dev);
  69
  70 /* Hash Ethernet address */
  71 static u32 macvlan_eth_hash(const unsigned char *addr)
  72 {
  73         u64 value = get_unaligned((u64 *)addr);
  74
  75         /* only want 6 bytes */
  76 #ifdef __BIG_ENDIAN
  77         value >>= 16;
  78 #else
  79         value <<= 16;
  80 #endif
  81         return hash_64(value, MACVLAN_HASH_BITS);
  82 }
  83
  84 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
  85 {
  86         return rcu_dereference(dev->rx_handler_data);
  87 }
  88
  89 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
  90 {
  91         return rtnl_dereference(dev->rx_handler_data);
  92 }
  93
  94 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
  95
  96 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
  97                                                const unsigned char *addr)
  98 {
  99         struct macvlan_dev *vlan;
 100         u32 idx = macvlan_eth_hash(addr);
 101
 102         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
 103                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
 104                         return vlan;
 105         }
 106         return NULL;
 107 }
 108
 109 static struct macvlan_source_entry *macvlan_hash_lookup_source(
 110         const struct macvlan_dev *vlan,
 111         const unsigned char *addr)
 112 {
 113         struct macvlan_source_entry *entry;
 114         u32 idx = macvlan_eth_hash(addr);
 115         struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
 116
 117         hlist_for_each_entry_rcu(entry, h, hlist) {
 118                 if (ether_addr_equal_64bits(entry->addr, addr) &&
 119                     entry->vlan == vlan)
 120                         return entry;
 121         }
 122         return NULL;
 123 }
 124
 125 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
 126                                    const unsigned char *addr)
 127 {
 128         struct macvlan_port *port = vlan->port;
 129         struct macvlan_source_entry *entry;
 130         struct hlist_head *h;
 131
 132         entry = macvlan_hash_lookup_source(vlan, addr);
 133         if (entry)
 134                 return 0;
 135
 136         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 137         if (!entry)
 138                 return -ENOMEM;
 139
 140         ether_addr_copy(entry->addr, addr);
 141         entry->vlan = vlan;
 142         h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
 143         hlist_add_head_rcu(&entry->hlist, h);
 144         vlan->macaddr_count++;
 145
 146         return 0;
 147 }
 148
 149 static void macvlan_hash_add(struct macvlan_dev *vlan)
 150 {
 151         struct macvlan_port *port = vlan->port;
 152         const unsigned char *addr = vlan->dev->dev_addr;
 153         u32 idx = macvlan_eth_hash(addr);
 154
 155         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
 156 }
 157
 158 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
 159 {
 160         hlist_del_rcu(&entry->hlist);
 161         kfree_rcu(entry, rcu);
 162 }
 163
 164 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
 165 {
 166         hlist_del_rcu(&vlan->hlist);
 167         if (sync)
 168                 synchronize_rcu();
 169 }
 170
 171 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
 172                                         const unsigned char *addr)
 173 {
 174         macvlan_hash_del(vlan, true);
 175         /* Now that we are unhashed it is safe to change the device
 176          * address without confusing packet delivery.
 177          */
 178         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
 179         macvlan_hash_add(vlan);
 180 }
 181
 182 static int macvlan_addr_busy(const struct macvlan_port *port,
 183                                 const unsigned char *addr)
 184 {
 185         /* Test to see if the specified multicast address is
 186          * currently in use by the underlying device or
 187          * another macvlan.
 188          */
 189         if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
 190                 return 1;
 191
 192         if (macvlan_hash_lookup(port, addr))
 193                 return 1;
 194
 195         return 0;
 196 }
 197
 198
 199 static int macvlan_broadcast_one(struct sk_buff *skb,
 200                                  const struct macvlan_dev *vlan,
 201                                  const struct ethhdr *eth, bool local)
 202 {
 203         struct net_device *dev = vlan->dev;
 204
 205         if (local)
 206                 return __dev_forward_skb(dev, skb);
 207
 208         skb->dev = dev;
 209         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 210                 skb->pkt_type = PACKET_BROADCAST;
 211         else
 212                 skb->pkt_type = PACKET_MULTICAST;
 213
 214         return 0;
 215 }
 216
 217 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 218 {
 219         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 220 }
 221
 222
 223 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 224                             const unsigned char *addr)
 225 {
 226         u32 val = __get_unaligned_cpu32(addr + 2);
 227
 228         val ^= macvlan_hash_mix(vlan);
 229         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 230 }
 231
 232 static void macvlan_broadcast(struct sk_buff *skb,
 233                               const struct macvlan_port *port,
 234                               struct net_device *src,
 235                               enum macvlan_mode mode)
 236 {
 237         const struct ethhdr *eth = eth_hdr(skb);
 238         const struct macvlan_dev *vlan;
 239         struct sk_buff *nskb;
 240         unsigned int i;
 241         int err;
 242         unsigned int hash;
 243
 244         if (skb->protocol == htons(ETH_P_PAUSE))
 245                 return;
 246
 247         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 248                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 249                         if (vlan->dev == src || !(vlan->mode & mode))
 250                                 continue;
 251
 252                         hash = mc_hash(vlan, eth->h_dest);
 253                         if (!test_bit(hash, vlan->mc_filter))
 254                                 continue;
 255
 256                         err = NET_RX_DROP;
 257                         nskb = skb_clone(skb, GFP_ATOMIC);
 258                         if (likely(nskb))
 259                                 err = macvlan_broadcast_one(
 260                                         nskb, vlan, eth,
 261                                         mode == MACVLAN_MODE_BRIDGE) ?:
 262                                       netif_rx_ni(nskb);
 263                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 264                                          err == NET_RX_SUCCESS, true);
 265                 }
 266         }
 267 }
 268
 269 static void macvlan_process_broadcast(struct work_struct *w)
 270 {
 271         struct macvlan_port *port = container_of(w, struct macvlan_port,
 272                                                  bc_work);
 273         struct sk_buff *skb;
 274         struct sk_buff_head list;
 275
 276         __skb_queue_head_init(&list);
 277
 278         spin_lock_bh(&port->bc_queue.lock);
 279         skb_queue_splice_tail_init(&port->bc_queue, &list);
 280         spin_unlock_bh(&port->bc_queue.lock);
 281
 282         while ((skb = __skb_dequeue(&list))) {
 283                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
 284
 285                 rcu_read_lock();
 286
 287                 if (!src)
 288                         /* frame comes from an external address */
 289                         macvlan_broadcast(skb, port, NULL,
 290                                           MACVLAN_MODE_PRIVATE |
 291                                           MACVLAN_MODE_VEPA    |
 292                                           MACVLAN_MODE_PASSTHRU|
 293                                           MACVLAN_MODE_BRIDGE);
 294                 else if (src->mode == MACVLAN_MODE_VEPA)
 295                         /* flood to everyone except source */
 296                         macvlan_broadcast(skb, port, src->dev,
 297                                           MACVLAN_MODE_VEPA |
 298                                           MACVLAN_MODE_BRIDGE);
 299                 else
 300                         /*
 301                          * flood only to VEPA ports, bridge ports
 302                          * already saw the frame on the way out.
 303                          */
 304                         macvlan_broadcast(skb, port, src->dev,
 305                                           MACVLAN_MODE_VEPA);
 306
 307                 rcu_read_unlock();
 308
 309                 if (src)
 310                         dev_put(src->dev);
 311                 kfree_skb(skb);
 312         }
 313 }
 314
 315 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 316                                       const struct macvlan_dev *src,
 317                                       struct sk_buff *skb)
 318 {
 319         struct sk_buff *nskb;
 320         int err = -ENOMEM;
 321
 322         nskb = skb_clone(skb, GFP_ATOMIC);
 323         if (!nskb)
 324                 goto err;
 325
 326         MACVLAN_SKB_CB(nskb)->src = src;
 327
 328         spin_lock(&port->bc_queue.lock);
 329         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 330                 if (src)
 331                         dev_hold(src->dev);
 332                 __skb_queue_tail(&port->bc_queue, nskb);
 333                 err = 0;
 334         }
 335         spin_unlock(&port->bc_queue.lock);
 336
 337         if (err)
 338                 goto free_nskb;
 339
 340         schedule_work(&port->bc_work);
 341         return;
 342
 343 free_nskb:
 344         kfree_skb(nskb);
 345 err:
 346         atomic_long_inc(&skb->dev->rx_dropped);
 347 }
 348
 349 static void macvlan_flush_sources(struct macvlan_port *port,
 350                                   struct macvlan_dev *vlan)
 351 {
 352         int i;
 353
 354         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 355                 struct hlist_node *h, *n;
 356
 357                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 358                         struct macvlan_source_entry *entry;
 359
 360                         entry = hlist_entry(h, struct macvlan_source_entry,
 361                                             hlist);
 362                         if (entry->vlan == vlan)
 363                                 macvlan_hash_del_source(entry);
 364                 }
 365         }
 366         vlan->macaddr_count = 0;
 367 }
 368
 369 static void macvlan_forward_source_one(struct sk_buff *skb,
 370                                        struct macvlan_dev *vlan)
 371 {
 372         struct sk_buff *nskb;
 373         struct net_device *dev;
 374         int len;
 375         int ret;
 376
 377         dev = vlan->dev;
 378         if (unlikely(!(dev->flags & IFF_UP)))
 379                 return;
 380
 381         nskb = skb_clone(skb, GFP_ATOMIC);
 382         if (!nskb)
 383                 return;
 384
 385         len = nskb->len + ETH_HLEN;
 386         nskb->dev = dev;
 387         nskb->pkt_type = PACKET_HOST;
 388
 389         ret = netif_rx(nskb);
 390         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 391 }
 392
 393 static void macvlan_forward_source(struct sk_buff *skb,
 394                                    struct macvlan_port *port,
 395                                    const unsigned char *addr)
 396 {
 397         struct macvlan_source_entry *entry;
 398         u32 idx = macvlan_eth_hash(addr);
 399         struct hlist_head *h = &port->vlan_source_hash[idx];
 400
 401         hlist_for_each_entry_rcu(entry, h, hlist) {
 402                 if (ether_addr_equal_64bits(entry->addr, addr))
 403                         if (entry->vlan->dev->flags & IFF_UP)
 404                                 macvlan_forward_source_one(skb, entry->vlan);
 405         }
 406 }
 407
 408 /* called under rcu_read_lock() from netif_receive_skb */
 409 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 410 {
 411         struct macvlan_port *port;
 412         struct sk_buff *skb = *pskb;
 413         const struct ethhdr *eth = eth_hdr(skb);
 414         const struct macvlan_dev *vlan;
 415         const struct macvlan_dev *src;
 416         struct net_device *dev;
 417         unsigned int len = 0;
 418         int ret;
 419         rx_handler_result_t handle_res;
 420
 421         port = macvlan_port_get_rcu(skb->dev);
 422         if (is_multicast_ether_addr(eth->h_dest)) {
 423                 unsigned int hash;
 424
 425                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 426                 if (!skb)
 427                         return RX_HANDLER_CONSUMED;
 428                 *pskb = skb;
 429                 eth = eth_hdr(skb);
 430                 macvlan_forward_source(skb, port, eth->h_source);
 431                 src = macvlan_hash_lookup(port, eth->h_source);
 432                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 433                     src->mode != MACVLAN_MODE_BRIDGE) {
 434                         /* forward to original port. */
 435                         vlan = src;
 436                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 437                               netif_rx(skb);
 438                         handle_res = RX_HANDLER_CONSUMED;
 439                         goto out;
 440                 }
 441
 442                 hash = mc_hash(NULL, eth->h_dest);
 443                 if (test_bit(hash, port->mc_filter))
 444                         macvlan_broadcast_enqueue(port, src, skb);
 445
 446                 return RX_HANDLER_PASS;
 447         }
 448
 449         macvlan_forward_source(skb, port, eth->h_source);
 450         if (port->passthru)
 451                 vlan = list_first_or_null_rcu(&port->vlans,
 452                                               struct macvlan_dev, list);
 453         else
 454                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 455         if (vlan == NULL)
 456                 return RX_HANDLER_PASS;
 457
 458         dev = vlan->dev;
 459         if (unlikely(!(dev->flags & IFF_UP))) {
 460                 kfree_skb(skb);
 461                 return RX_HANDLER_CONSUMED;
 462         }
 463         len = skb->len + ETH_HLEN;
 464         skb = skb_share_check(skb, GFP_ATOMIC);
 465         if (!skb) {
 466                 ret = NET_RX_DROP;
 467                 handle_res = RX_HANDLER_CONSUMED;
 468                 goto out;
 469         }
 470
 471         *pskb = skb;
 472         skb->dev = dev;
 473         skb->pkt_type = PACKET_HOST;
 474
 475         ret = NET_RX_SUCCESS;
 476         handle_res = RX_HANDLER_ANOTHER;
 477 out:
 478         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 479         return handle_res;
 480 }
 481
 482 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 483 {
 484         const struct macvlan_dev *vlan = netdev_priv(dev);
 485         const struct macvlan_port *port = vlan->port;
 486         const struct macvlan_dev *dest;
 487
 488         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 489                 const struct ethhdr *eth = (void *)skb->data;
 490
 491                 /* send to other bridge ports directly */
 492                 if (is_multicast_ether_addr(eth->h_dest)) {
 493                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 494                         goto xmit_world;
 495                 }
 496
 497                 dest = macvlan_hash_lookup(port, eth->h_dest);
 498                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 499                         /* send to lowerdev first for its network taps */
 500                         dev_forward_skb(vlan->lowerdev, skb);
 501
 502                         return NET_XMIT_SUCCESS;
 503                 }
 504         }
 505
 506 xmit_world:
 507         skb->dev = vlan->lowerdev;
 508         return dev_queue_xmit(skb);
 509 }
 510
 511 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 512 {
 513 #ifdef CONFIG_NET_POLL_CONTROLLER
 514         if (vlan->netpoll)
 515                 netpoll_send_skb(vlan->netpoll, skb);
 516 #else
 517         BUG();
 518 #endif
 519         return NETDEV_TX_OK;
 520 }
 521
 522 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 523                                       struct net_device *dev)
 524 {
 525         unsigned int len = skb->len;
 526         int ret;
 527         struct macvlan_dev *vlan = netdev_priv(dev);
 528
 529         if (unlikely(netpoll_tx_running(dev)))
 530                 return macvlan_netpoll_send_skb(vlan, skb);
 531
 532         if (vlan->fwd_priv) {
 533                 skb->dev = vlan->lowerdev;
 534                 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
 535         } else {
 536                 ret = macvlan_queue_xmit(skb, dev);
 537         }
 538
 539         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 540                 struct vlan_pcpu_stats *pcpu_stats;
 541
 542                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 543                 u64_stats_update_begin(&pcpu_stats->syncp);
 544                 pcpu_stats->tx_packets++;
 545                 pcpu_stats->tx_bytes += len;
 546                 u64_stats_update_end(&pcpu_stats->syncp);
 547         } else {
 548                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 549         }
 550         return ret;
 551 }
 552
 553 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 554                                unsigned short type, const void *daddr,
 555                                const void *saddr, unsigned len)
 556 {
 557         const struct macvlan_dev *vlan = netdev_priv(dev);
 558         struct net_device *lowerdev = vlan->lowerdev;
 559
 560         return dev_hard_header(skb, lowerdev, type, daddr,
 561                                saddr ? : dev->dev_addr, len);
 562 }
 563
 564 static const struct header_ops macvlan_hard_header_ops = {
 565         .create         = macvlan_hard_header,
 566         .parse          = eth_header_parse,
 567         .cache          = eth_header_cache,
 568         .cache_update   = eth_header_cache_update,
 569 };
 570
 571 static struct rtnl_link_ops macvlan_link_ops;
 572
 573 static int macvlan_open(struct net_device *dev)
 574 {
 575         struct macvlan_dev *vlan = netdev_priv(dev);
 576         struct net_device *lowerdev = vlan->lowerdev;
 577         int err;
 578
 579         if (vlan->port->passthru) {
 580                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 581                         err = dev_set_promiscuity(lowerdev, 1);
 582                         if (err < 0)
 583                                 goto out;
 584                 }
 585                 goto hash_add;
 586         }
 587
 588         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
 589             dev->rtnl_link_ops == &macvlan_link_ops) {
 590                 vlan->fwd_priv =
 591                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 592
 593                 /* If we get a NULL pointer back, or if we get an error
 594                  * then we should just fall through to the non accelerated path
 595                  */
 596                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
 597                         vlan->fwd_priv = NULL;
 598                 } else
 599                         return 0;
 600         }
 601
 602         err = -EBUSY;
 603         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 604                 goto out;
 605
 606         err = dev_uc_add(lowerdev, dev->dev_addr);
 607         if (err < 0)
 608                 goto out;
 609         if (dev->flags & IFF_ALLMULTI) {
 610                 err = dev_set_allmulti(lowerdev, 1);
 611                 if (err < 0)
 612                         goto del_unicast;
 613         }
 614
 615         if (dev->flags & IFF_PROMISC) {
 616                 err = dev_set_promiscuity(lowerdev, 1);
 617                 if (err < 0)
 618                         goto clear_multi;
 619         }
 620
 621 hash_add:
 622         macvlan_hash_add(vlan);
 623         return 0;
 624
 625 clear_multi:
 626         dev_set_allmulti(lowerdev, -1);
 627 del_unicast:
 628         dev_uc_del(lowerdev, dev->dev_addr);
 629 out:
 630         if (vlan->fwd_priv) {
 631                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 632                                                            vlan->fwd_priv);
 633                 vlan->fwd_priv = NULL;
 634         }
 635         return err;
 636 }
 637
 638 static int macvlan_stop(struct net_device *dev)
 639 {
 640         struct macvlan_dev *vlan = netdev_priv(dev);
 641         struct net_device *lowerdev = vlan->lowerdev;
 642
 643         if (vlan->fwd_priv) {
 644                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 645                                                            vlan->fwd_priv);
 646                 vlan->fwd_priv = NULL;
 647                 return 0;
 648         }
 649
 650         dev_uc_unsync(lowerdev, dev);
 651         dev_mc_unsync(lowerdev, dev);
 652
 653         if (vlan->port->passthru) {
 654                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 655                         dev_set_promiscuity(lowerdev, -1);
 656                 goto hash_del;
 657         }
 658
 659         if (dev->flags & IFF_ALLMULTI)
 660                 dev_set_allmulti(lowerdev, -1);
 661
 662         if (dev->flags & IFF_PROMISC)
 663                 dev_set_promiscuity(lowerdev, -1);
 664
 665         dev_uc_del(lowerdev, dev->dev_addr);
 666
 667 hash_del:
 668         macvlan_hash_del(vlan, !dev->dismantle);
 669         return 0;
 670 }
 671
 672 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 673 {
 674         struct macvlan_dev *vlan = netdev_priv(dev);
 675         struct net_device *lowerdev = vlan->lowerdev;
 676         int err;
 677
 678         if (!(dev->flags & IFF_UP)) {
 679                 /* Just copy in the new address */
 680                 ether_addr_copy(dev->dev_addr, addr);
 681         } else {
 682                 /* Rehash and update the device filters */
 683                 if (macvlan_addr_busy(vlan->port, addr))
 684                         return -EBUSY;
 685
 686                 if (!vlan->port->passthru) {
 687                         err = dev_uc_add(lowerdev, addr);
 688                         if (err)
 689                                 return err;
 690
 691                         dev_uc_del(lowerdev, dev->dev_addr);
 692                 }
 693
 694                 macvlan_hash_change_addr(vlan, addr);
 695         }
 696         return 0;
 697 }
 698
 699 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 700 {
 701         struct macvlan_dev *vlan = netdev_priv(dev);
 702         struct sockaddr *addr = p;
 703
 704         if (!is_valid_ether_addr(addr->sa_data))
 705                 return -EADDRNOTAVAIL;
 706
 707         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 708                 dev_set_mac_address(vlan->lowerdev, addr);
 709                 return 0;
 710         }
 711
 712         return macvlan_sync_address(dev, addr->sa_data);
 713 }
 714
 715 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 716 {
 717         struct macvlan_dev *vlan = netdev_priv(dev);
 718         struct net_device *lowerdev = vlan->lowerdev;
 719
 720         if (dev->flags & IFF_UP) {
 721                 if (change & IFF_ALLMULTI)
 722                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 723                 if (change & IFF_PROMISC)
 724                         dev_set_promiscuity(lowerdev,
 725                                             dev->flags & IFF_PROMISC ? 1 : -1);
 726
 727         }
 728 }
 729
 730 static void macvlan_compute_filter(unsigned long *mc_filter,
 731                                    struct net_device *dev,
 732                                    struct macvlan_dev *vlan)
 733 {
 734         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 735                 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
 736         } else {
 737                 struct netdev_hw_addr *ha;
 738                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 739
 740                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 741                 netdev_for_each_mc_addr(ha, dev) {
 742                         __set_bit(mc_hash(vlan, ha->addr), filter);
 743                 }
 744
 745                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 746
 747                 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 748         }
 749 }
 750
 751 static void macvlan_set_mac_lists(struct net_device *dev)
 752 {
 753         struct macvlan_dev *vlan = netdev_priv(dev);
 754
 755         macvlan_compute_filter(vlan->mc_filter, dev, vlan);
 756
 757         dev_uc_sync(vlan->lowerdev, dev);
 758         dev_mc_sync(vlan->lowerdev, dev);
 759
 760         /* This is slightly inaccurate as we're including the subscription
 761          * list of vlan->lowerdev too.
 762          *
 763          * Bug alert: This only works if everyone has the same broadcast
 764          * address as lowerdev.  As soon as someone changes theirs this
 765          * will break.
 766          *
 767          * However, this is already broken as when you change your broadcast
 768          * address we don't get called.
 769          *
 770          * The solution is to maintain a list of broadcast addresses like
 771          * we do for uc/mc, if you care.
 772          */
 773         macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 774 }
 775
 776 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 777 {
 778         struct macvlan_dev *vlan = netdev_priv(dev);
 779
 780         if (vlan->lowerdev->mtu < new_mtu)
 781                 return -EINVAL;
 782         dev->mtu = new_mtu;
 783         return 0;
 784 }
 785
 786 /*
 787  * macvlan network devices have devices nesting below it and are a special
 788  * "super class" of normal network devices; split their locks off into a
 789  * separate class since they always nest.
 790  */
 791 static struct lock_class_key macvlan_netdev_addr_lock_key;
 792
 793 #define ALWAYS_ON_FEATURES \
 794         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX | \
 795          NETIF_F_GSO_ROBUST)
 796
 797 #define MACVLAN_FEATURES \
 798         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 799          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
 800          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 801          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 802
 803 #define MACVLAN_STATE_MASK \
 804         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 805
 806 static int macvlan_get_nest_level(struct net_device *dev)
 807 {
 808         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 809 }
 810
 811 static void macvlan_set_lockdep_class(struct net_device *dev)
 812 {
 813         netdev_lockdep_set_classes(dev);
 814         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 815                                        &macvlan_netdev_addr_lock_key,
 816                                        macvlan_get_nest_level(dev));
 817 }
 818
 819 static int macvlan_init(struct net_device *dev)
 820 {
 821         struct macvlan_dev *vlan = netdev_priv(dev);
 822         const struct net_device *lowerdev = vlan->lowerdev;
 823         struct macvlan_port *port = vlan->port;
 824
 825         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 826                                   (lowerdev->state & MACVLAN_STATE_MASK);
 827         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 828         dev->features           |= ALWAYS_ON_FEATURES;
 829         dev->hw_features        |= NETIF_F_LRO;
 830         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 831         dev->gso_max_size       = lowerdev->gso_max_size;
 832         dev->gso_max_segs       = lowerdev->gso_max_segs;
 833         dev->hard_header_len    = lowerdev->hard_header_len;
 834
 835         macvlan_set_lockdep_class(dev);
 836
 837         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 838         if (!vlan->pcpu_stats)
 839                 return -ENOMEM;
 840
 841         port->count += 1;
 842
 843         return 0;
 844 }
 845
 846 static void macvlan_uninit(struct net_device *dev)
 847 {
 848         struct macvlan_dev *vlan = netdev_priv(dev);
 849         struct macvlan_port *port = vlan->port;
 850
 851         free_percpu(vlan->pcpu_stats);
 852
 853         macvlan_flush_sources(port, vlan);
 854         port->count -= 1;
 855         if (!port->count)
 856                 macvlan_port_destroy(port->dev);
 857 }
 858
 859 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 860                                                          struct rtnl_link_stats64 *stats)
 861 {
 862         struct macvlan_dev *vlan = netdev_priv(dev);
 863
 864         if (vlan->pcpu_stats) {
 865                 struct vlan_pcpu_stats *p;
 866                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 867                 u32 rx_errors = 0, tx_dropped = 0;
 868                 unsigned int start;
 869                 int i;
 870
 871                 for_each_possible_cpu(i) {
 872                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 873                         do {
 874                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 875                                 rx_packets      = p->rx_packets;
 876                                 rx_bytes        = p->rx_bytes;
 877                                 rx_multicast    = p->rx_multicast;
 878                                 tx_packets      = p->tx_packets;
 879                                 tx_bytes        = p->tx_bytes;
 880                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 881
 882                         stats->rx_packets       += rx_packets;
 883                         stats->rx_bytes         += rx_bytes;
 884                         stats->multicast        += rx_multicast;
 885                         stats->tx_packets       += tx_packets;
 886                         stats->tx_bytes         += tx_bytes;
 887                         /* rx_errors & tx_dropped are u32, updated
 888                          * without syncp protection.
 889                          */
 890                         rx_errors       += p->rx_errors;
 891                         tx_dropped      += p->tx_dropped;
 892                 }
 893                 stats->rx_errors        = rx_errors;
 894                 stats->rx_dropped       = rx_errors;
 895                 stats->tx_dropped       = tx_dropped;
 896         }
 897         return stats;
 898 }
 899
 900 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 901                                    __be16 proto, u16 vid)
 902 {
 903         struct macvlan_dev *vlan = netdev_priv(dev);
 904         struct net_device *lowerdev = vlan->lowerdev;
 905
 906         return vlan_vid_add(lowerdev, proto, vid);
 907 }
 908
 909 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 910                                     __be16 proto, u16 vid)
 911 {
 912         struct macvlan_dev *vlan = netdev_priv(dev);
 913         struct net_device *lowerdev = vlan->lowerdev;
 914
 915         vlan_vid_del(lowerdev, proto, vid);
 916         return 0;
 917 }
 918
 919 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 920                            struct net_device *dev,
 921                            const unsigned char *addr, u16 vid,
 922                            u16 flags)
 923 {
 924         struct macvlan_dev *vlan = netdev_priv(dev);
 925         int err = -EINVAL;
 926
 927         /* Support unicast filter only on passthru devices.
 928          * Multicast filter should be allowed on all devices.
 929          */
 930         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 931                 return -EOPNOTSUPP;
 932
 933         if (flags & NLM_F_REPLACE)
 934                 return -EOPNOTSUPP;
 935
 936         if (is_unicast_ether_addr(addr))
 937                 err = dev_uc_add_excl(dev, addr);
 938         else if (is_multicast_ether_addr(addr))
 939                 err = dev_mc_add_excl(dev, addr);
 940
 941         return err;
 942 }
 943
 944 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 945                            struct net_device *dev,
 946                            const unsigned char *addr, u16 vid)
 947 {
 948         struct macvlan_dev *vlan = netdev_priv(dev);
 949         int err = -EINVAL;
 950
 951         /* Support unicast filter only on passthru devices.
 952          * Multicast filter should be allowed on all devices.
 953          */
 954         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 955                 return -EOPNOTSUPP;
 956
 957         if (is_unicast_ether_addr(addr))
 958                 err = dev_uc_del(dev, addr);
 959         else if (is_multicast_ether_addr(addr))
 960                 err = dev_mc_del(dev, addr);
 961
 962         return err;
 963 }
 964
 965 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 966                                         struct ethtool_drvinfo *drvinfo)
 967 {
 968         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 969         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 970 }
 971
 972 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
 973                                               struct ethtool_link_ksettings *cmd)
 974 {
 975         const struct macvlan_dev *vlan = netdev_priv(dev);
 976
 977         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
 978 }
 979
 980 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 981                                               netdev_features_t features)
 982 {
 983         struct macvlan_dev *vlan = netdev_priv(dev);
 984         netdev_features_t lowerdev_features = vlan->lowerdev->features;
 985         netdev_features_t mask;
 986
 987         features |= NETIF_F_ALL_FOR_ALL;
 988         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 989         mask = features;
 990
 991         lowerdev_features &= (features | ~NETIF_F_LRO);
 992         features = netdev_increment_features(lowerdev_features, features, mask);
 993         features |= ALWAYS_ON_FEATURES;
 994         features &= ~NETIF_F_NETNS_LOCAL;
 995
 996         return features;
 997 }
 998
 999 #ifdef CONFIG_NET_POLL_CONTROLLER
1000 static void macvlan_dev_poll_controller(struct net_device *dev)
1001 {
1002         return;
1003 }
1004
1005 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1006 {
1007         struct macvlan_dev *vlan = netdev_priv(dev);
1008         struct net_device *real_dev = vlan->lowerdev;
1009         struct netpoll *netpoll;
1010         int err = 0;
1011
1012         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1013         err = -ENOMEM;
1014         if (!netpoll)
1015                 goto out;
1016
1017         err = __netpoll_setup(netpoll, real_dev);
1018         if (err) {
1019                 kfree(netpoll);
1020                 goto out;
1021         }
1022
1023         vlan->netpoll = netpoll;
1024
1025 out:
1026         return err;
1027 }
1028
1029 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1030 {
1031         struct macvlan_dev *vlan = netdev_priv(dev);
1032         struct netpoll *netpoll = vlan->netpoll;
1033
1034         if (!netpoll)
1035                 return;
1036
1037         vlan->netpoll = NULL;
1038
1039         __netpoll_free_async(netpoll);
1040 }
1041 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1042
1043 static int macvlan_dev_get_iflink(const struct net_device *dev)
1044 {
1045         struct macvlan_dev *vlan = netdev_priv(dev);
1046
1047         return vlan->lowerdev->ifindex;
1048 }
1049
1050 static const struct ethtool_ops macvlan_ethtool_ops = {
1051         .get_link               = ethtool_op_get_link,
1052         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1053         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1054 };
1055
1056 static const struct net_device_ops macvlan_netdev_ops = {
1057         .ndo_init               = macvlan_init,
1058         .ndo_uninit             = macvlan_uninit,
1059         .ndo_open               = macvlan_open,
1060         .ndo_stop               = macvlan_stop,
1061         .ndo_start_xmit         = macvlan_start_xmit,
1062         .ndo_change_mtu         = macvlan_change_mtu,
1063         .ndo_fix_features       = macvlan_fix_features,
1064         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1065         .ndo_set_mac_address    = macvlan_set_mac_address,
1066         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1067         .ndo_get_stats64        = macvlan_dev_get_stats64,
1068         .ndo_validate_addr      = eth_validate_addr,
1069         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1070         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1071         .ndo_fdb_add            = macvlan_fdb_add,
1072         .ndo_fdb_del            = macvlan_fdb_del,
1073         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1074         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1075 #ifdef CONFIG_NET_POLL_CONTROLLER
1076         .ndo_poll_controller    = macvlan_dev_poll_controller,
1077         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1078         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1079 #endif
1080         .ndo_get_iflink         = macvlan_dev_get_iflink,
1081         .ndo_features_check     = passthru_features_check,
1082 };
1083
1084 void macvlan_common_setup(struct net_device *dev)
1085 {
1086         ether_setup(dev);
1087
1088         dev->min_mtu            = 0;
1089         dev->max_mtu            = ETH_MAX_MTU;
1090         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1091         netif_keep_dst(dev);
1092         dev->priv_flags        |= IFF_UNICAST_FLT;
1093         dev->netdev_ops         = &macvlan_netdev_ops;
1094         dev->destructor         = free_netdev;
1095         dev->header_ops         = &macvlan_hard_header_ops;
1096         dev->ethtool_ops        = &macvlan_ethtool_ops;
1097 }
1098 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1099
1100 static void macvlan_setup(struct net_device *dev)
1101 {
1102         macvlan_common_setup(dev);
1103         dev->priv_flags |= IFF_NO_QUEUE;
1104 }
1105
1106 static int macvlan_port_create(struct net_device *dev)
1107 {
1108         struct macvlan_port *port;
1109         unsigned int i;
1110         int err;
1111
1112         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1113                 return -EINVAL;
1114
1115         if (netif_is_ipvlan_port(dev))
1116                 return -EBUSY;
1117
1118         port = kzalloc(sizeof(*port), GFP_KERNEL);
1119         if (port == NULL)
1120                 return -ENOMEM;
1121
1122         port->passthru = false;
1123         port->dev = dev;
1124         INIT_LIST_HEAD(&port->vlans);
1125         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1126                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1127         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1128                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1129
1130         skb_queue_head_init(&port->bc_queue);
1131         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1132
1133         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1134         if (err)
1135                 kfree(port);
1136         else
1137                 dev->priv_flags |= IFF_MACVLAN_PORT;
1138         return err;
1139 }
1140
1141 static void macvlan_port_destroy(struct net_device *dev)
1142 {
1143         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1144
1145         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1146         netdev_rx_handler_unregister(dev);
1147
1148         /* After this point, no packet can schedule bc_work anymore,
1149          * but we need to cancel it and purge left skbs if any.
1150          */
1151         cancel_work_sync(&port->bc_work);
1152         __skb_queue_purge(&port->bc_queue);
1153
1154         kfree_rcu(port, rcu);
1155 }
1156
1157 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
1158 {
1159         if (tb[IFLA_ADDRESS]) {
1160                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1161                         return -EINVAL;
1162                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1163                         return -EADDRNOTAVAIL;
1164         }
1165
1166         if (data && data[IFLA_MACVLAN_FLAGS] &&
1167             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1168                 return -EINVAL;
1169
1170         if (data && data[IFLA_MACVLAN_MODE]) {
1171                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1172                 case MACVLAN_MODE_PRIVATE:
1173                 case MACVLAN_MODE_VEPA:
1174                 case MACVLAN_MODE_BRIDGE:
1175                 case MACVLAN_MODE_PASSTHRU:
1176                 case MACVLAN_MODE_SOURCE:
1177                         break;
1178                 default:
1179                         return -EINVAL;
1180                 }
1181         }
1182
1183         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1184                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1185                 case MACVLAN_MACADDR_ADD:
1186                 case MACVLAN_MACADDR_DEL:
1187                 case MACVLAN_MACADDR_FLUSH:
1188                 case MACVLAN_MACADDR_SET:
1189                         break;
1190                 default:
1191                         return -EINVAL;
1192                 }
1193         }
1194
1195         if (data && data[IFLA_MACVLAN_MACADDR]) {
1196                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1197                         return -EINVAL;
1198
1199                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1200                         return -EADDRNOTAVAIL;
1201         }
1202
1203         if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1204                 return -EINVAL;
1205
1206         return 0;
1207 }
1208
1209 /**
1210  * reconfigure list of remote source mac address
1211  * (only for macvlan devices in source mode)
1212  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1213  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1214  */
1215 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1216                                       struct nlattr *data[])
1217 {
1218         char *addr = NULL;
1219         int ret, rem, len;
1220         struct nlattr *nla, *head;
1221         struct macvlan_source_entry *entry;
1222
1223         if (data[IFLA_MACVLAN_MACADDR])
1224                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1225
1226         if (mode == MACVLAN_MACADDR_ADD) {
1227                 if (!addr)
1228                         return -EINVAL;
1229
1230                 return macvlan_hash_add_source(vlan, addr);
1231
1232         } else if (mode == MACVLAN_MACADDR_DEL) {
1233                 if (!addr)
1234                         return -EINVAL;
1235
1236                 entry = macvlan_hash_lookup_source(vlan, addr);
1237                 if (entry) {
1238                         macvlan_hash_del_source(entry);
1239                         vlan->macaddr_count--;
1240                 }
1241         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1242                 macvlan_flush_sources(vlan->port, vlan);
1243         } else if (mode == MACVLAN_MACADDR_SET) {
1244                 macvlan_flush_sources(vlan->port, vlan);
1245
1246                 if (addr) {
1247                         ret = macvlan_hash_add_source(vlan, addr);
1248                         if (ret)
1249                                 return ret;
1250                 }
1251
1252                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1253                         return 0;
1254
1255                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1256                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1257
1258                 nla_for_each_attr(nla, head, len, rem) {
1259                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1260                             nla_len(nla) != ETH_ALEN)
1261                                 continue;
1262
1263                         addr = nla_data(nla);
1264                         ret = macvlan_hash_add_source(vlan, addr);
1265                         if (ret)
1266                                 return ret;
1267                 }
1268         } else {
1269                 return -EINVAL;
1270         }
1271
1272         return 0;
1273 }
1274
1275 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1276                            struct nlattr *tb[], struct nlattr *data[])
1277 {
1278         struct macvlan_dev *vlan = netdev_priv(dev);
1279         struct macvlan_port *port;
1280         struct net_device *lowerdev;
1281         int err;
1282         int macmode;
1283
1284         if (!tb[IFLA_LINK])
1285                 return -EINVAL;
1286
1287         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1288         if (lowerdev == NULL)
1289                 return -ENODEV;
1290
1291         /* When creating macvlans or macvtaps on top of other macvlans - use
1292          * the real device as the lowerdev.
1293          */
1294         if (netif_is_macvlan(lowerdev))
1295                 lowerdev = macvlan_dev_real_dev(lowerdev);
1296
1297         if (!tb[IFLA_MTU])
1298                 dev->mtu = lowerdev->mtu;
1299         else if (dev->mtu > lowerdev->mtu)
1300                 return -EINVAL;
1301
1302         /* MTU range: 68 - lowerdev->max_mtu */
1303         dev->min_mtu = ETH_MIN_MTU;
1304         dev->max_mtu = lowerdev->max_mtu;
1305
1306         if (!tb[IFLA_ADDRESS])
1307                 eth_hw_addr_random(dev);
1308
1309         if (!macvlan_port_exists(lowerdev)) {
1310                 err = macvlan_port_create(lowerdev);
1311                 if (err < 0)
1312                         return err;
1313         }
1314         port = macvlan_port_get_rtnl(lowerdev);
1315
1316         /* Only 1 macvlan device can be created in passthru mode */
1317         if (port->passthru)
1318                 return -EINVAL;
1319
1320         vlan->lowerdev = lowerdev;
1321         vlan->dev      = dev;
1322         vlan->port     = port;
1323         vlan->set_features = MACVLAN_FEATURES;
1324         vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1325
1326         vlan->mode     = MACVLAN_MODE_VEPA;
1327         if (data && data[IFLA_MACVLAN_MODE])
1328                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1329
1330         if (data && data[IFLA_MACVLAN_FLAGS])
1331                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1332
1333         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1334                 if (port->count)
1335                         return -EINVAL;
1336                 port->passthru = true;
1337                 eth_hw_addr_inherit(dev, lowerdev);
1338         }
1339
1340         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1341                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1342                         return -EINVAL;
1343                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1344                 err = macvlan_changelink_sources(vlan, macmode, data);
1345                 if (err)
1346                         return err;
1347         }
1348
1349         err = register_netdevice(dev);
1350         if (err < 0)
1351                 return err;
1352
1353         dev->priv_flags |= IFF_MACVLAN;
1354         err = netdev_upper_dev_link(lowerdev, dev);
1355         if (err)
1356                 goto unregister_netdev;
1357
1358         list_add_tail_rcu(&vlan->list, &port->vlans);
1359         netif_stacked_transfer_operstate(lowerdev, dev);
1360         linkwatch_fire_event(dev);
1361
1362         return 0;
1363
1364 unregister_netdev:
1365         unregister_netdevice(dev);
1366
1367         return err;
1368 }
1369 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1370
1371 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1372                            struct nlattr *tb[], struct nlattr *data[])
1373 {
1374         return macvlan_common_newlink(src_net, dev, tb, data);
1375 }
1376
1377 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1378 {
1379         struct macvlan_dev *vlan = netdev_priv(dev);
1380
1381         if (vlan->mode == MACVLAN_MODE_SOURCE)
1382                 macvlan_flush_sources(vlan->port, vlan);
1383         list_del_rcu(&vlan->list);
1384         unregister_netdevice_queue(dev, head);
1385         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1386 }
1387 EXPORT_SYMBOL_GPL(macvlan_dellink);
1388
1389 static int macvlan_changelink(struct net_device *dev,
1390                 struct nlattr *tb[], struct nlattr *data[])
1391 {
1392         struct macvlan_dev *vlan = netdev_priv(dev);
1393         enum macvlan_mode mode;
1394         bool set_mode = false;
1395         enum macvlan_macaddr_mode macmode;
1396         int ret;
1397
1398         /* Validate mode, but don't set yet: setting flags may fail. */
1399         if (data && data[IFLA_MACVLAN_MODE]) {
1400                 set_mode = true;
1401                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1402                 /* Passthrough mode can't be set or cleared dynamically */
1403                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1404                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1405                         return -EINVAL;
1406                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1407                     vlan->mode != mode)
1408                         macvlan_flush_sources(vlan->port, vlan);
1409         }
1410
1411         if (data && data[IFLA_MACVLAN_FLAGS]) {
1412                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1413                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1414                 if (vlan->port->passthru && promisc) {
1415                         int err;
1416
1417                         if (flags & MACVLAN_FLAG_NOPROMISC)
1418                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1419                         else
1420                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1421                         if (err < 0)
1422                                 return err;
1423                 }
1424                 vlan->flags = flags;
1425         }
1426         if (set_mode)
1427                 vlan->mode = mode;
1428         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1429                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1430                         return -EINVAL;
1431                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1432                 ret = macvlan_changelink_sources(vlan, macmode, data);
1433                 if (ret)
1434                         return ret;
1435         }
1436         return 0;
1437 }
1438
1439 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1440 {
1441         if (vlan->macaddr_count == 0)
1442                 return 0;
1443         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1444                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1445 }
1446
1447 static size_t macvlan_get_size(const struct net_device *dev)
1448 {
1449         struct macvlan_dev *vlan = netdev_priv(dev);
1450
1451         return (0
1452                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1453                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1454                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1455                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1456                 );
1457 }
1458
1459 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1460                                      const struct macvlan_dev *vlan,
1461                                      const int i)
1462 {
1463         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1464         struct macvlan_source_entry *entry;
1465
1466         hlist_for_each_entry_rcu(entry, h, hlist) {
1467                 if (entry->vlan != vlan)
1468                         continue;
1469                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1470                         return 1;
1471         }
1472         return 0;
1473 }
1474
1475 static int macvlan_fill_info(struct sk_buff *skb,
1476                                 const struct net_device *dev)
1477 {
1478         struct macvlan_dev *vlan = netdev_priv(dev);
1479         int i;
1480         struct nlattr *nest;
1481
1482         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1483                 goto nla_put_failure;
1484         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1485                 goto nla_put_failure;
1486         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1487                 goto nla_put_failure;
1488         if (vlan->macaddr_count > 0) {
1489                 nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1490                 if (nest == NULL)
1491                         goto nla_put_failure;
1492
1493                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1494                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1495                                 goto nla_put_failure;
1496                 }
1497                 nla_nest_end(skb, nest);
1498         }
1499         return 0;
1500
1501 nla_put_failure:
1502         return -EMSGSIZE;
1503 }
1504
1505 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1506         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1507         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1508         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1509         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1510         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1511         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1512 };
1513
1514 int macvlan_link_register(struct rtnl_link_ops *ops)
1515 {
1516         /* common fields */
1517         ops->priv_size          = sizeof(struct macvlan_dev);
1518         ops->validate           = macvlan_validate;
1519         ops->maxtype            = IFLA_MACVLAN_MAX;
1520         ops->policy             = macvlan_policy;
1521         ops->changelink         = macvlan_changelink;
1522         ops->get_size           = macvlan_get_size;
1523         ops->fill_info          = macvlan_fill_info;
1524
1525         return rtnl_link_register(ops);
1526 };
1527 EXPORT_SYMBOL_GPL(macvlan_link_register);
1528
1529 static struct net *macvlan_get_link_net(const struct net_device *dev)
1530 {
1531         return dev_net(macvlan_dev_real_dev(dev));
1532 }
1533
1534 static struct rtnl_link_ops macvlan_link_ops = {
1535         .kind           = "macvlan",
1536         .setup          = macvlan_setup,
1537         .newlink        = macvlan_newlink,
1538         .dellink        = macvlan_dellink,
1539         .get_link_net   = macvlan_get_link_net,
1540 };
1541
1542 static int macvlan_device_event(struct notifier_block *unused,
1543                                 unsigned long event, void *ptr)
1544 {
1545         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1546         struct macvlan_dev *vlan, *next;
1547         struct macvlan_port *port;
1548         LIST_HEAD(list_kill);
1549
1550         if (!macvlan_port_exists(dev))
1551                 return NOTIFY_DONE;
1552
1553         port = macvlan_port_get_rtnl(dev);
1554
1555         switch (event) {
1556         case NETDEV_UP:
1557         case NETDEV_CHANGE:
1558                 list_for_each_entry(vlan, &port->vlans, list)
1559                         netif_stacked_transfer_operstate(vlan->lowerdev,
1560                                                          vlan->dev);
1561                 break;
1562         case NETDEV_FEAT_CHANGE:
1563                 list_for_each_entry(vlan, &port->vlans, list) {
1564                         vlan->dev->gso_max_size = dev->gso_max_size;
1565                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1566                         netdev_update_features(vlan->dev);
1567                 }
1568                 break;
1569         case NETDEV_CHANGEMTU:
1570                 list_for_each_entry(vlan, &port->vlans, list) {
1571                         if (vlan->dev->mtu <= dev->mtu)
1572                                 continue;
1573                         dev_set_mtu(vlan->dev, dev->mtu);
1574                 }
1575                 break;
1576         case NETDEV_CHANGEADDR:
1577                 if (!port->passthru)
1578                         return NOTIFY_DONE;
1579
1580                 vlan = list_first_entry_or_null(&port->vlans,
1581                                                 struct macvlan_dev,
1582                                                 list);
1583
1584                 if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1585                         return NOTIFY_BAD;
1586
1587                 break;
1588         case NETDEV_UNREGISTER:
1589                 /* twiddle thumbs on netns device moves */
1590                 if (dev->reg_state != NETREG_UNREGISTERING)
1591                         break;
1592
1593                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1594                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1595                 unregister_netdevice_many(&list_kill);
1596                 break;
1597         case NETDEV_PRE_TYPE_CHANGE:
1598                 /* Forbid underlaying device to change its type. */
1599                 return NOTIFY_BAD;
1600
1601         case NETDEV_NOTIFY_PEERS:
1602         case NETDEV_BONDING_FAILOVER:
1603         case NETDEV_RESEND_IGMP:
1604                 /* Propagate to all vlans */
1605                 list_for_each_entry(vlan, &port->vlans, list)
1606                         call_netdevice_notifiers(event, vlan->dev);
1607         }
1608         return NOTIFY_DONE;
1609 }
1610
1611 static struct notifier_block macvlan_notifier_block __read_mostly = {
1612         .notifier_call  = macvlan_device_event,
1613 };
1614
1615 static int __init macvlan_init_module(void)
1616 {
1617         int err;
1618
1619         register_netdevice_notifier(&macvlan_notifier_block);
1620
1621         err = macvlan_link_register(&macvlan_link_ops);
1622         if (err < 0)
1623                 goto err1;
1624         return 0;
1625 err1:
1626         unregister_netdevice_notifier(&macvlan_notifier_block);
1627         return err;
1628 }
1629
1630 static void __exit macvlan_cleanup_module(void)
1631 {
1632         rtnl_link_unregister(&macvlan_link_ops);
1633         unregister_netdevice_notifier(&macvlan_notifier_block);
1634 }
1635
1636 module_init(macvlan_init_module);
1637 module_exit(macvlan_cleanup_module);
1638
1639 MODULE_LICENSE("GPL");
1640 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1641 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1642 MODULE_ALIAS_RTNL_LINK("macvlan");