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 || vlan->mode == MACVLAN_MODE_SOURCE)
 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         if (dev->flags & IFF_ALLMULTI)
 627                 dev_set_allmulti(lowerdev, -1);
 628 del_unicast:
 629         dev_uc_del(lowerdev, dev->dev_addr);
 630 out:
 631         if (vlan->fwd_priv) {
 632                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 633                                                            vlan->fwd_priv);
 634                 vlan->fwd_priv = NULL;
 635         }
 636         return err;
 637 }
 638
 639 static int macvlan_stop(struct net_device *dev)
 640 {
 641         struct macvlan_dev *vlan = netdev_priv(dev);
 642         struct net_device *lowerdev = vlan->lowerdev;
 643
 644         if (vlan->fwd_priv) {
 645                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 646                                                            vlan->fwd_priv);
 647                 vlan->fwd_priv = NULL;
 648                 return 0;
 649         }
 650
 651         dev_uc_unsync(lowerdev, dev);
 652         dev_mc_unsync(lowerdev, dev);
 653
 654         if (vlan->port->passthru) {
 655                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 656                         dev_set_promiscuity(lowerdev, -1);
 657                 goto hash_del;
 658         }
 659
 660         if (dev->flags & IFF_ALLMULTI)
 661                 dev_set_allmulti(lowerdev, -1);
 662
 663         if (dev->flags & IFF_PROMISC)
 664                 dev_set_promiscuity(lowerdev, -1);
 665
 666         dev_uc_del(lowerdev, dev->dev_addr);
 667
 668 hash_del:
 669         macvlan_hash_del(vlan, !dev->dismantle);
 670         return 0;
 671 }
 672
 673 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 674 {
 675         struct macvlan_dev *vlan = netdev_priv(dev);
 676         struct net_device *lowerdev = vlan->lowerdev;
 677         int err;
 678
 679         if (!(dev->flags & IFF_UP)) {
 680                 /* Just copy in the new address */
 681                 ether_addr_copy(dev->dev_addr, addr);
 682         } else {
 683                 /* Rehash and update the device filters */
 684                 if (macvlan_addr_busy(vlan->port, addr))
 685                         return -EBUSY;
 686
 687                 if (!vlan->port->passthru) {
 688                         err = dev_uc_add(lowerdev, addr);
 689                         if (err)
 690                                 return err;
 691
 692                         dev_uc_del(lowerdev, dev->dev_addr);
 693                 }
 694
 695                 macvlan_hash_change_addr(vlan, addr);
 696         }
 697         return 0;
 698 }
 699
 700 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 701 {
 702         struct macvlan_dev *vlan = netdev_priv(dev);
 703         struct sockaddr *addr = p;
 704
 705         if (!is_valid_ether_addr(addr->sa_data))
 706                 return -EADDRNOTAVAIL;
 707
 708         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 709                 dev_set_mac_address(vlan->lowerdev, addr);
 710                 return 0;
 711         }
 712
 713         return macvlan_sync_address(dev, addr->sa_data);
 714 }
 715
 716 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 717 {
 718         struct macvlan_dev *vlan = netdev_priv(dev);
 719         struct net_device *lowerdev = vlan->lowerdev;
 720
 721         if (dev->flags & IFF_UP) {
 722                 if (change & IFF_ALLMULTI)
 723                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 724                 if (change & IFF_PROMISC)
 725                         dev_set_promiscuity(lowerdev,
 726                                             dev->flags & IFF_PROMISC ? 1 : -1);
 727
 728         }
 729 }
 730
 731 static void macvlan_compute_filter(unsigned long *mc_filter,
 732                                    struct net_device *dev,
 733                                    struct macvlan_dev *vlan)
 734 {
 735         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 736                 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
 737         } else {
 738                 struct netdev_hw_addr *ha;
 739                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 740
 741                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 742                 netdev_for_each_mc_addr(ha, dev) {
 743                         __set_bit(mc_hash(vlan, ha->addr), filter);
 744                 }
 745
 746                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 747
 748                 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 749         }
 750 }
 751
 752 static void macvlan_set_mac_lists(struct net_device *dev)
 753 {
 754         struct macvlan_dev *vlan = netdev_priv(dev);
 755
 756         macvlan_compute_filter(vlan->mc_filter, dev, vlan);
 757
 758         dev_uc_sync(vlan->lowerdev, dev);
 759         dev_mc_sync(vlan->lowerdev, dev);
 760
 761         /* This is slightly inaccurate as we're including the subscription
 762          * list of vlan->lowerdev too.
 763          *
 764          * Bug alert: This only works if everyone has the same broadcast
 765          * address as lowerdev.  As soon as someone changes theirs this
 766          * will break.
 767          *
 768          * However, this is already broken as when you change your broadcast
 769          * address we don't get called.
 770          *
 771          * The solution is to maintain a list of broadcast addresses like
 772          * we do for uc/mc, if you care.
 773          */
 774         macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 775 }
 776
 777 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 778 {
 779         struct macvlan_dev *vlan = netdev_priv(dev);
 780
 781         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 782                 return -EINVAL;
 783         dev->mtu = new_mtu;
 784         return 0;
 785 }
 786
 787 /*
 788  * macvlan network devices have devices nesting below it and are a special
 789  * "super class" of normal network devices; split their locks off into a
 790  * separate class since they always nest.
 791  */
 792 static struct lock_class_key macvlan_netdev_addr_lock_key;
 793
 794 #define ALWAYS_ON_FEATURES \
 795         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX | \
 796          NETIF_F_GSO_ROBUST)
 797
 798 #define MACVLAN_FEATURES \
 799         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 800          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
 801          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 802          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 803
 804 #define MACVLAN_STATE_MASK \
 805         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 806
 807 static int macvlan_get_nest_level(struct net_device *dev)
 808 {
 809         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 810 }
 811
 812 static void macvlan_set_lockdep_class(struct net_device *dev)
 813 {
 814         netdev_lockdep_set_classes(dev);
 815         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 816                                        &macvlan_netdev_addr_lock_key,
 817                                        macvlan_get_nest_level(dev));
 818 }
 819
 820 static int macvlan_init(struct net_device *dev)
 821 {
 822         struct macvlan_dev *vlan = netdev_priv(dev);
 823         const struct net_device *lowerdev = vlan->lowerdev;
 824         struct macvlan_port *port = vlan->port;
 825
 826         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 827                                   (lowerdev->state & MACVLAN_STATE_MASK);
 828         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 829         dev->features           |= ALWAYS_ON_FEATURES;
 830         dev->hw_features        |= NETIF_F_LRO;
 831         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 832         dev->gso_max_size       = lowerdev->gso_max_size;
 833         dev->gso_max_segs       = lowerdev->gso_max_segs;
 834         dev->hard_header_len    = lowerdev->hard_header_len;
 835
 836         macvlan_set_lockdep_class(dev);
 837
 838         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 839         if (!vlan->pcpu_stats)
 840                 return -ENOMEM;
 841
 842         port->count += 1;
 843
 844         return 0;
 845 }
 846
 847 static void macvlan_uninit(struct net_device *dev)
 848 {
 849         struct macvlan_dev *vlan = netdev_priv(dev);
 850         struct macvlan_port *port = vlan->port;
 851
 852         free_percpu(vlan->pcpu_stats);
 853
 854         macvlan_flush_sources(port, vlan);
 855         port->count -= 1;
 856         if (!port->count)
 857                 macvlan_port_destroy(port->dev);
 858 }
 859
 860 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 861                                                          struct rtnl_link_stats64 *stats)
 862 {
 863         struct macvlan_dev *vlan = netdev_priv(dev);
 864
 865         if (vlan->pcpu_stats) {
 866                 struct vlan_pcpu_stats *p;
 867                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 868                 u32 rx_errors = 0, tx_dropped = 0;
 869                 unsigned int start;
 870                 int i;
 871
 872                 for_each_possible_cpu(i) {
 873                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 874                         do {
 875                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 876                                 rx_packets      = p->rx_packets;
 877                                 rx_bytes        = p->rx_bytes;
 878                                 rx_multicast    = p->rx_multicast;
 879                                 tx_packets      = p->tx_packets;
 880                                 tx_bytes        = p->tx_bytes;
 881                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 882
 883                         stats->rx_packets       += rx_packets;
 884                         stats->rx_bytes         += rx_bytes;
 885                         stats->multicast        += rx_multicast;
 886                         stats->tx_packets       += tx_packets;
 887                         stats->tx_bytes         += tx_bytes;
 888                         /* rx_errors & tx_dropped are u32, updated
 889                          * without syncp protection.
 890                          */
 891                         rx_errors       += p->rx_errors;
 892                         tx_dropped      += p->tx_dropped;
 893                 }
 894                 stats->rx_errors        = rx_errors;
 895                 stats->rx_dropped       = rx_errors;
 896                 stats->tx_dropped       = tx_dropped;
 897         }
 898         return stats;
 899 }
 900
 901 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 902                                    __be16 proto, u16 vid)
 903 {
 904         struct macvlan_dev *vlan = netdev_priv(dev);
 905         struct net_device *lowerdev = vlan->lowerdev;
 906
 907         return vlan_vid_add(lowerdev, proto, vid);
 908 }
 909
 910 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 911                                     __be16 proto, u16 vid)
 912 {
 913         struct macvlan_dev *vlan = netdev_priv(dev);
 914         struct net_device *lowerdev = vlan->lowerdev;
 915
 916         vlan_vid_del(lowerdev, proto, vid);
 917         return 0;
 918 }
 919
 920 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 921                            struct net_device *dev,
 922                            const unsigned char *addr, u16 vid,
 923                            u16 flags)
 924 {
 925         struct macvlan_dev *vlan = netdev_priv(dev);
 926         int err = -EINVAL;
 927
 928         /* Support unicast filter only on passthru devices.
 929          * Multicast filter should be allowed on all devices.
 930          */
 931         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 932                 return -EOPNOTSUPP;
 933
 934         if (flags & NLM_F_REPLACE)
 935                 return -EOPNOTSUPP;
 936
 937         if (is_unicast_ether_addr(addr))
 938                 err = dev_uc_add_excl(dev, addr);
 939         else if (is_multicast_ether_addr(addr))
 940                 err = dev_mc_add_excl(dev, addr);
 941
 942         return err;
 943 }
 944
 945 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 946                            struct net_device *dev,
 947                            const unsigned char *addr, u16 vid)
 948 {
 949         struct macvlan_dev *vlan = netdev_priv(dev);
 950         int err = -EINVAL;
 951
 952         /* Support unicast filter only on passthru devices.
 953          * Multicast filter should be allowed on all devices.
 954          */
 955         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 956                 return -EOPNOTSUPP;
 957
 958         if (is_unicast_ether_addr(addr))
 959                 err = dev_uc_del(dev, addr);
 960         else if (is_multicast_ether_addr(addr))
 961                 err = dev_mc_del(dev, addr);
 962
 963         return err;
 964 }
 965
 966 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 967                                         struct ethtool_drvinfo *drvinfo)
 968 {
 969         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 970         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 971 }
 972
 973 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
 974                                               struct ethtool_link_ksettings *cmd)
 975 {
 976         const struct macvlan_dev *vlan = netdev_priv(dev);
 977
 978         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
 979 }
 980
 981 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 982                                               netdev_features_t features)
 983 {
 984         struct macvlan_dev *vlan = netdev_priv(dev);
 985         netdev_features_t lowerdev_features = vlan->lowerdev->features;
 986         netdev_features_t mask;
 987
 988         features |= NETIF_F_ALL_FOR_ALL;
 989         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 990         mask = features;
 991
 992         lowerdev_features &= (features | ~NETIF_F_LRO);
 993         features = netdev_increment_features(lowerdev_features, features, mask);
 994         features |= ALWAYS_ON_FEATURES;
 995         features &= ~NETIF_F_NETNS_LOCAL;
 996
 997         return features;
 998 }
 999
1000 #ifdef CONFIG_NET_POLL_CONTROLLER
1001 static void macvlan_dev_poll_controller(struct net_device *dev)
1002 {
1003         return;
1004 }
1005
1006 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1007 {
1008         struct macvlan_dev *vlan = netdev_priv(dev);
1009         struct net_device *real_dev = vlan->lowerdev;
1010         struct netpoll *netpoll;
1011         int err = 0;
1012
1013         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1014         err = -ENOMEM;
1015         if (!netpoll)
1016                 goto out;
1017
1018         err = __netpoll_setup(netpoll, real_dev);
1019         if (err) {
1020                 kfree(netpoll);
1021                 goto out;
1022         }
1023
1024         vlan->netpoll = netpoll;
1025
1026 out:
1027         return err;
1028 }
1029
1030 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1031 {
1032         struct macvlan_dev *vlan = netdev_priv(dev);
1033         struct netpoll *netpoll = vlan->netpoll;
1034
1035         if (!netpoll)
1036                 return;
1037
1038         vlan->netpoll = NULL;
1039
1040         __netpoll_free_async(netpoll);
1041 }
1042 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1043
1044 static int macvlan_dev_get_iflink(const struct net_device *dev)
1045 {
1046         struct macvlan_dev *vlan = netdev_priv(dev);
1047
1048         return vlan->lowerdev->ifindex;
1049 }
1050
1051 static const struct ethtool_ops macvlan_ethtool_ops = {
1052         .get_link               = ethtool_op_get_link,
1053         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1054         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1055 };
1056
1057 static const struct net_device_ops macvlan_netdev_ops = {
1058         .ndo_init               = macvlan_init,
1059         .ndo_uninit             = macvlan_uninit,
1060         .ndo_open               = macvlan_open,
1061         .ndo_stop               = macvlan_stop,
1062         .ndo_start_xmit         = macvlan_start_xmit,
1063         .ndo_change_mtu         = macvlan_change_mtu,
1064         .ndo_fix_features       = macvlan_fix_features,
1065         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1066         .ndo_set_mac_address    = macvlan_set_mac_address,
1067         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1068         .ndo_get_stats64        = macvlan_dev_get_stats64,
1069         .ndo_validate_addr      = eth_validate_addr,
1070         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1071         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1072         .ndo_fdb_add            = macvlan_fdb_add,
1073         .ndo_fdb_del            = macvlan_fdb_del,
1074         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1075         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1076 #ifdef CONFIG_NET_POLL_CONTROLLER
1077         .ndo_poll_controller    = macvlan_dev_poll_controller,
1078         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1079         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1080 #endif
1081         .ndo_get_iflink         = macvlan_dev_get_iflink,
1082         .ndo_features_check     = passthru_features_check,
1083 };
1084
1085 void macvlan_common_setup(struct net_device *dev)
1086 {
1087         ether_setup(dev);
1088
1089         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1090         netif_keep_dst(dev);
1091         dev->priv_flags        |= IFF_UNICAST_FLT;
1092         dev->netdev_ops         = &macvlan_netdev_ops;
1093         dev->destructor         = free_netdev;
1094         dev->header_ops         = &macvlan_hard_header_ops;
1095         dev->ethtool_ops        = &macvlan_ethtool_ops;
1096 }
1097 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1098
1099 static void macvlan_setup(struct net_device *dev)
1100 {
1101         macvlan_common_setup(dev);
1102         dev->priv_flags |= IFF_NO_QUEUE;
1103 }
1104
1105 static int macvlan_port_create(struct net_device *dev)
1106 {
1107         struct macvlan_port *port;
1108         unsigned int i;
1109         int err;
1110
1111         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1112                 return -EINVAL;
1113
1114         if (netif_is_ipvlan_port(dev))
1115                 return -EBUSY;
1116
1117         port = kzalloc(sizeof(*port), GFP_KERNEL);
1118         if (port == NULL)
1119                 return -ENOMEM;
1120
1121         port->passthru = false;
1122         port->dev = dev;
1123         INIT_LIST_HEAD(&port->vlans);
1124         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1125                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1126         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1127                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1128
1129         skb_queue_head_init(&port->bc_queue);
1130         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1131
1132         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1133         if (err)
1134                 kfree(port);
1135         else
1136                 dev->priv_flags |= IFF_MACVLAN_PORT;
1137         return err;
1138 }
1139
1140 static void macvlan_port_destroy(struct net_device *dev)
1141 {
1142         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1143         struct sk_buff *skb;
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
1153         while ((skb = __skb_dequeue(&port->bc_queue))) {
1154                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1155
1156                 if (src)
1157                         dev_put(src->dev);
1158
1159                 kfree_skb(skb);
1160         }
1161
1162         kfree_rcu(port, rcu);
1163 }
1164
1165 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
1166 {
1167         if (tb[IFLA_ADDRESS]) {
1168                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1169                         return -EINVAL;
1170                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1171                         return -EADDRNOTAVAIL;
1172         }
1173
1174         if (data && data[IFLA_MACVLAN_FLAGS] &&
1175             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1176                 return -EINVAL;
1177
1178         if (data && data[IFLA_MACVLAN_MODE]) {
1179                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1180                 case MACVLAN_MODE_PRIVATE:
1181                 case MACVLAN_MODE_VEPA:
1182                 case MACVLAN_MODE_BRIDGE:
1183                 case MACVLAN_MODE_PASSTHRU:
1184                 case MACVLAN_MODE_SOURCE:
1185                         break;
1186                 default:
1187                         return -EINVAL;
1188                 }
1189         }
1190
1191         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1192                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1193                 case MACVLAN_MACADDR_ADD:
1194                 case MACVLAN_MACADDR_DEL:
1195                 case MACVLAN_MACADDR_FLUSH:
1196                 case MACVLAN_MACADDR_SET:
1197                         break;
1198                 default:
1199                         return -EINVAL;
1200                 }
1201         }
1202
1203         if (data && data[IFLA_MACVLAN_MACADDR]) {
1204                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1205                         return -EINVAL;
1206
1207                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1208                         return -EADDRNOTAVAIL;
1209         }
1210
1211         if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1212                 return -EINVAL;
1213
1214         return 0;
1215 }
1216
1217 /**
1218  * reconfigure list of remote source mac address
1219  * (only for macvlan devices in source mode)
1220  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1221  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1222  */
1223 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1224                                       struct nlattr *data[])
1225 {
1226         char *addr = NULL;
1227         int ret, rem, len;
1228         struct nlattr *nla, *head;
1229         struct macvlan_source_entry *entry;
1230
1231         if (data[IFLA_MACVLAN_MACADDR])
1232                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1233
1234         if (mode == MACVLAN_MACADDR_ADD) {
1235                 if (!addr)
1236                         return -EINVAL;
1237
1238                 return macvlan_hash_add_source(vlan, addr);
1239
1240         } else if (mode == MACVLAN_MACADDR_DEL) {
1241                 if (!addr)
1242                         return -EINVAL;
1243
1244                 entry = macvlan_hash_lookup_source(vlan, addr);
1245                 if (entry) {
1246                         macvlan_hash_del_source(entry);
1247                         vlan->macaddr_count--;
1248                 }
1249         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1250                 macvlan_flush_sources(vlan->port, vlan);
1251         } else if (mode == MACVLAN_MACADDR_SET) {
1252                 macvlan_flush_sources(vlan->port, vlan);
1253
1254                 if (addr) {
1255                         ret = macvlan_hash_add_source(vlan, addr);
1256                         if (ret)
1257                                 return ret;
1258                 }
1259
1260                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1261                         return 0;
1262
1263                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1264                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1265
1266                 nla_for_each_attr(nla, head, len, rem) {
1267                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1268                             nla_len(nla) != ETH_ALEN)
1269                                 continue;
1270
1271                         addr = nla_data(nla);
1272                         ret = macvlan_hash_add_source(vlan, addr);
1273                         if (ret)
1274                                 return ret;
1275                 }
1276         } else {
1277                 return -EINVAL;
1278         }
1279
1280         return 0;
1281 }
1282
1283 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1284                            struct nlattr *tb[], struct nlattr *data[])
1285 {
1286         struct macvlan_dev *vlan = netdev_priv(dev);
1287         struct macvlan_port *port;
1288         struct net_device *lowerdev;
1289         int err;
1290         int macmode;
1291         bool create = false;
1292
1293         if (!tb[IFLA_LINK])
1294                 return -EINVAL;
1295
1296         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1297         if (lowerdev == NULL)
1298                 return -ENODEV;
1299
1300         /* When creating macvlans or macvtaps on top of other macvlans - use
1301          * the real device as the lowerdev.
1302          */
1303         if (netif_is_macvlan(lowerdev))
1304                 lowerdev = macvlan_dev_real_dev(lowerdev);
1305
1306         if (!tb[IFLA_MTU])
1307                 dev->mtu = lowerdev->mtu;
1308         else if (dev->mtu > lowerdev->mtu)
1309                 return -EINVAL;
1310
1311         if (!tb[IFLA_ADDRESS])
1312                 eth_hw_addr_random(dev);
1313
1314         if (!macvlan_port_exists(lowerdev)) {
1315                 err = macvlan_port_create(lowerdev);
1316                 if (err < 0)
1317                         return err;
1318                 create = true;
1319         }
1320         port = macvlan_port_get_rtnl(lowerdev);
1321
1322         /* Only 1 macvlan device can be created in passthru mode */
1323         if (port->passthru) {
1324                 /* The macvlan port must be not created this time,
1325                  * still goto destroy_macvlan_port for readability.
1326                  */
1327                 err = -EINVAL;
1328                 goto destroy_macvlan_port;
1329         }
1330
1331         vlan->lowerdev = lowerdev;
1332         vlan->dev      = dev;
1333         vlan->port     = port;
1334         vlan->set_features = MACVLAN_FEATURES;
1335         vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1336
1337         vlan->mode     = MACVLAN_MODE_VEPA;
1338         if (data && data[IFLA_MACVLAN_MODE])
1339                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1340
1341         if (data && data[IFLA_MACVLAN_FLAGS])
1342                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1343
1344         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1345                 if (port->count) {
1346                         err = -EINVAL;
1347                         goto destroy_macvlan_port;
1348                 }
1349                 port->passthru = true;
1350                 eth_hw_addr_inherit(dev, lowerdev);
1351         }
1352
1353         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1354                 if (vlan->mode != MACVLAN_MODE_SOURCE) {
1355                         err = -EINVAL;
1356                         goto destroy_macvlan_port;
1357                 }
1358                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1359                 err = macvlan_changelink_sources(vlan, macmode, data);
1360                 if (err)
1361                         goto destroy_macvlan_port;
1362         }
1363
1364         err = register_netdevice(dev);
1365         if (err < 0)
1366                 goto destroy_macvlan_port;
1367
1368         dev->priv_flags |= IFF_MACVLAN;
1369         err = netdev_upper_dev_link(lowerdev, dev);
1370         if (err)
1371                 goto unregister_netdev;
1372
1373         list_add_tail_rcu(&vlan->list, &port->vlans);
1374         netif_stacked_transfer_operstate(lowerdev, dev);
1375         linkwatch_fire_event(dev);
1376
1377         return 0;
1378
1379 unregister_netdev:
1380         /* macvlan_uninit would free the macvlan port */
1381         unregister_netdevice(dev);
1382         return err;
1383 destroy_macvlan_port:
1384         /* the macvlan port may be freed by macvlan_uninit when fail to register.
1385          * so we destroy the macvlan port only when it's valid.
1386          */
1387         if (create && macvlan_port_get_rtnl(lowerdev))
1388                 macvlan_port_destroy(port->dev);
1389         return err;
1390 }
1391 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1392
1393 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1394                            struct nlattr *tb[], struct nlattr *data[])
1395 {
1396         return macvlan_common_newlink(src_net, dev, tb, data);
1397 }
1398
1399 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1400 {
1401         struct macvlan_dev *vlan = netdev_priv(dev);
1402
1403         if (vlan->mode == MACVLAN_MODE_SOURCE)
1404                 macvlan_flush_sources(vlan->port, vlan);
1405         list_del_rcu(&vlan->list);
1406         unregister_netdevice_queue(dev, head);
1407         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1408 }
1409 EXPORT_SYMBOL_GPL(macvlan_dellink);
1410
1411 static int macvlan_changelink(struct net_device *dev,
1412                 struct nlattr *tb[], struct nlattr *data[])
1413 {
1414         struct macvlan_dev *vlan = netdev_priv(dev);
1415         enum macvlan_mode mode;
1416         bool set_mode = false;
1417         enum macvlan_macaddr_mode macmode;
1418         int ret;
1419
1420         /* Validate mode, but don't set yet: setting flags may fail. */
1421         if (data && data[IFLA_MACVLAN_MODE]) {
1422                 set_mode = true;
1423                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1424                 /* Passthrough mode can't be set or cleared dynamically */
1425                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1426                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1427                         return -EINVAL;
1428                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1429                     vlan->mode != mode)
1430                         macvlan_flush_sources(vlan->port, vlan);
1431         }
1432
1433         if (data && data[IFLA_MACVLAN_FLAGS]) {
1434                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1435                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1436                 if (vlan->port->passthru && promisc) {
1437                         int err;
1438
1439                         if (flags & MACVLAN_FLAG_NOPROMISC)
1440                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1441                         else
1442                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1443                         if (err < 0)
1444                                 return err;
1445                 }
1446                 vlan->flags = flags;
1447         }
1448         if (set_mode)
1449                 vlan->mode = mode;
1450         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1451                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1452                         return -EINVAL;
1453                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1454                 ret = macvlan_changelink_sources(vlan, macmode, data);
1455                 if (ret)
1456                         return ret;
1457         }
1458         return 0;
1459 }
1460
1461 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1462 {
1463         if (vlan->macaddr_count == 0)
1464                 return 0;
1465         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1466                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1467 }
1468
1469 static size_t macvlan_get_size(const struct net_device *dev)
1470 {
1471         struct macvlan_dev *vlan = netdev_priv(dev);
1472
1473         return (0
1474                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1475                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1476                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1477                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1478                 );
1479 }
1480
1481 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1482                                      const struct macvlan_dev *vlan,
1483                                      const int i)
1484 {
1485         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1486         struct macvlan_source_entry *entry;
1487
1488         hlist_for_each_entry_rcu(entry, h, hlist) {
1489                 if (entry->vlan != vlan)
1490                         continue;
1491                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1492                         return 1;
1493         }
1494         return 0;
1495 }
1496
1497 static int macvlan_fill_info(struct sk_buff *skb,
1498                                 const struct net_device *dev)
1499 {
1500         struct macvlan_dev *vlan = netdev_priv(dev);
1501         int i;
1502         struct nlattr *nest;
1503
1504         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1505                 goto nla_put_failure;
1506         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1507                 goto nla_put_failure;
1508         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1509                 goto nla_put_failure;
1510         if (vlan->macaddr_count > 0) {
1511                 nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1512                 if (nest == NULL)
1513                         goto nla_put_failure;
1514
1515                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1516                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1517                                 goto nla_put_failure;
1518                 }
1519                 nla_nest_end(skb, nest);
1520         }
1521         return 0;
1522
1523 nla_put_failure:
1524         return -EMSGSIZE;
1525 }
1526
1527 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1528         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1529         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1530         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1531         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1532         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1533         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1534 };
1535
1536 int macvlan_link_register(struct rtnl_link_ops *ops)
1537 {
1538         /* common fields */
1539         ops->priv_size          = sizeof(struct macvlan_dev);
1540         ops->validate           = macvlan_validate;
1541         ops->maxtype            = IFLA_MACVLAN_MAX;
1542         ops->policy             = macvlan_policy;
1543         ops->changelink         = macvlan_changelink;
1544         ops->get_size           = macvlan_get_size;
1545         ops->fill_info          = macvlan_fill_info;
1546
1547         return rtnl_link_register(ops);
1548 };
1549 EXPORT_SYMBOL_GPL(macvlan_link_register);
1550
1551 static struct net *macvlan_get_link_net(const struct net_device *dev)
1552 {
1553         return dev_net(macvlan_dev_real_dev(dev));
1554 }
1555
1556 static struct rtnl_link_ops macvlan_link_ops = {
1557         .kind           = "macvlan",
1558         .setup          = macvlan_setup,
1559         .newlink        = macvlan_newlink,
1560         .dellink        = macvlan_dellink,
1561         .get_link_net   = macvlan_get_link_net,
1562 };
1563
1564 static int macvlan_device_event(struct notifier_block *unused,
1565                                 unsigned long event, void *ptr)
1566 {
1567         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1568         struct macvlan_dev *vlan, *next;
1569         struct macvlan_port *port;
1570         LIST_HEAD(list_kill);
1571
1572         if (!macvlan_port_exists(dev))
1573                 return NOTIFY_DONE;
1574
1575         port = macvlan_port_get_rtnl(dev);
1576
1577         switch (event) {
1578         case NETDEV_UP:
1579         case NETDEV_CHANGE:
1580                 list_for_each_entry(vlan, &port->vlans, list)
1581                         netif_stacked_transfer_operstate(vlan->lowerdev,
1582                                                          vlan->dev);
1583                 break;
1584         case NETDEV_FEAT_CHANGE:
1585                 list_for_each_entry(vlan, &port->vlans, list) {
1586                         vlan->dev->gso_max_size = dev->gso_max_size;
1587                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1588                         netdev_update_features(vlan->dev);
1589                 }
1590                 break;
1591         case NETDEV_CHANGEMTU:
1592                 list_for_each_entry(vlan, &port->vlans, list) {
1593                         if (vlan->dev->mtu <= dev->mtu)
1594                                 continue;
1595                         dev_set_mtu(vlan->dev, dev->mtu);
1596                 }
1597                 break;
1598         case NETDEV_CHANGEADDR:
1599                 if (!port->passthru)
1600                         return NOTIFY_DONE;
1601
1602                 vlan = list_first_entry_or_null(&port->vlans,
1603                                                 struct macvlan_dev,
1604                                                 list);
1605
1606                 if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1607                         return NOTIFY_BAD;
1608
1609                 break;
1610         case NETDEV_UNREGISTER:
1611                 /* twiddle thumbs on netns device moves */
1612                 if (dev->reg_state != NETREG_UNREGISTERING)
1613                         break;
1614
1615                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1616                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1617                 unregister_netdevice_many(&list_kill);
1618                 break;
1619         case NETDEV_PRE_TYPE_CHANGE:
1620                 /* Forbid underlaying device to change its type. */
1621                 return NOTIFY_BAD;
1622
1623         case NETDEV_NOTIFY_PEERS:
1624         case NETDEV_BONDING_FAILOVER:
1625         case NETDEV_RESEND_IGMP:
1626                 /* Propagate to all vlans */
1627                 list_for_each_entry(vlan, &port->vlans, list)
1628                         call_netdevice_notifiers(event, vlan->dev);
1629         }
1630         return NOTIFY_DONE;
1631 }
1632
1633 static struct notifier_block macvlan_notifier_block __read_mostly = {
1634         .notifier_call  = macvlan_device_event,
1635 };
1636
1637 static int __init macvlan_init_module(void)
1638 {
1639         int err;
1640
1641         register_netdevice_notifier(&macvlan_notifier_block);
1642
1643         err = macvlan_link_register(&macvlan_link_ops);
1644         if (err < 0)
1645                 goto err1;
1646         return 0;
1647 err1:
1648         unregister_netdevice_notifier(&macvlan_notifier_block);
1649         return err;
1650 }
1651
1652 static void __exit macvlan_cleanup_module(void)
1653 {
1654         rtnl_link_unregister(&macvlan_link_ops);
1655         unregister_netdevice_notifier(&macvlan_notifier_block);
1656 }
1657
1658 module_init(macvlan_init_module);
1659 module_exit(macvlan_cleanup_module);
1660
1661 MODULE_LICENSE("GPL");
1662 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1663 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1664 MODULE_ALIAS_RTNL_LINK("macvlan");