drivers/net/macvlan.c

   1 /*
   2  * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public License as
   6  * published by the Free Software Foundation; either version 2 of
   7  * the License, or (at your option) any later version.
   8  *
   9  * The code this is based on carried the following copyright notice:
  10  * ---
  11  * (C) Copyright 2001-2006
  12  * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
  13  * Re-worked by Ben Greear <greearb@candelatech.com>
  14  * ---
  15  */
  16 #include <linux/kernel.h>
  17 #include <linux/types.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/errno.h>
  21 #include <linux/slab.h>
  22 #include <linux/string.h>
  23 #include <linux/rculist.h>
  24 #include <linux/notifier.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/etherdevice.h>
  27 #include <linux/ethtool.h>
  28 #include <linux/if_arp.h>
  29 #include <linux/if_vlan.h>
  30 #include <linux/if_link.h>
  31 #include <linux/if_macvlan.h>
  32 #include <linux/hash.h>
  33 #include <linux/workqueue.h>
  34 #include <net/rtnetlink.h>
  35 #include <net/xfrm.h>
  36 #include <linux/netpoll.h>
  37
  38 #define MACVLAN_HASH_BITS       8
  39 #define MACVLAN_HASH_SIZE       (1<<MACVLAN_HASH_BITS)
  40 #define MACVLAN_BC_QUEUE_LEN    1000
  41
  42 struct macvlan_port {
  43         struct net_device       *dev;
  44         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  45         struct list_head        vlans;
  46         struct rcu_head         rcu;
  47         struct sk_buff_head     bc_queue;
  48         struct work_struct      bc_work;
  49         bool                    passthru;
  50         int                     count;
  51         struct hlist_head       vlan_source_hash[MACVLAN_HASH_SIZE];
  52 };
  53
  54 struct macvlan_source_entry {
  55         struct hlist_node       hlist;
  56         struct macvlan_dev      *vlan;
  57         unsigned char           addr[6+2] __aligned(sizeof(u16));
  58         struct rcu_head         rcu;
  59 };
  60
  61 struct macvlan_skb_cb {
  62         const struct macvlan_dev *src;
  63 };
  64
  65 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
  66
  67 static void macvlan_port_destroy(struct net_device *dev);
  68
  69 /* Hash Ethernet address */
  70 static u32 macvlan_eth_hash(const unsigned char *addr)
  71 {
  72         u64 value = get_unaligned((u64 *)addr);
  73
  74         /* only want 6 bytes */
  75 #ifdef __BIG_ENDIAN
  76         value >>= 16;
  77 #else
  78         value <<= 16;
  79 #endif
  80         return hash_64(value, MACVLAN_HASH_BITS);
  81 }
  82
  83 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
  84 {
  85         return rcu_dereference(dev->rx_handler_data);
  86 }
  87
  88 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
  89 {
  90         return rtnl_dereference(dev->rx_handler_data);
  91 }
  92
  93 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
  94
  95 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
  96                                                const unsigned char *addr)
  97 {
  98         struct macvlan_dev *vlan;
  99         u32 idx = macvlan_eth_hash(addr);
 100
 101         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
 102                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
 103                         return vlan;
 104         }
 105         return NULL;
 106 }
 107
 108 static struct macvlan_source_entry *macvlan_hash_lookup_source(
 109         const struct macvlan_dev *vlan,
 110         const unsigned char *addr)
 111 {
 112         struct macvlan_source_entry *entry;
 113         u32 idx = macvlan_eth_hash(addr);
 114         struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
 115
 116         hlist_for_each_entry_rcu(entry, h, hlist) {
 117                 if (ether_addr_equal_64bits(entry->addr, addr) &&
 118                     entry->vlan == vlan)
 119                         return entry;
 120         }
 121         return NULL;
 122 }
 123
 124 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
 125                                    const unsigned char *addr)
 126 {
 127         struct macvlan_port *port = vlan->port;
 128         struct macvlan_source_entry *entry;
 129         struct hlist_head *h;
 130
 131         entry = macvlan_hash_lookup_source(vlan, addr);
 132         if (entry)
 133                 return 0;
 134
 135         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 136         if (!entry)
 137                 return -ENOMEM;
 138
 139         ether_addr_copy(entry->addr, addr);
 140         entry->vlan = vlan;
 141         h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
 142         hlist_add_head_rcu(&entry->hlist, h);
 143         vlan->macaddr_count++;
 144
 145         return 0;
 146 }
 147
 148 static void macvlan_hash_add(struct macvlan_dev *vlan)
 149 {
 150         struct macvlan_port *port = vlan->port;
 151         const unsigned char *addr = vlan->dev->dev_addr;
 152         u32 idx = macvlan_eth_hash(addr);
 153
 154         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
 155 }
 156
 157 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
 158 {
 159         hlist_del_rcu(&entry->hlist);
 160         kfree_rcu(entry, rcu);
 161 }
 162
 163 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
 164 {
 165         hlist_del_rcu(&vlan->hlist);
 166         if (sync)
 167                 synchronize_rcu();
 168 }
 169
 170 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
 171                                         const unsigned char *addr)
 172 {
 173         macvlan_hash_del(vlan, true);
 174         /* Now that we are unhashed it is safe to change the device
 175          * address without confusing packet delivery.
 176          */
 177         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
 178         macvlan_hash_add(vlan);
 179 }
 180
 181 static int macvlan_addr_busy(const struct macvlan_port *port,
 182                                 const unsigned char *addr)
 183 {
 184         /* Test to see if the specified multicast address is
 185          * currently in use by the underlying device or
 186          * another macvlan.
 187          */
 188         if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
 189                 return 1;
 190
 191         if (macvlan_hash_lookup(port, addr))
 192                 return 1;
 193
 194         return 0;
 195 }
 196
 197
 198 static int macvlan_broadcast_one(struct sk_buff *skb,
 199                                  const struct macvlan_dev *vlan,
 200                                  const struct ethhdr *eth, bool local)
 201 {
 202         struct net_device *dev = vlan->dev;
 203
 204         if (local)
 205                 return __dev_forward_skb(dev, skb);
 206
 207         skb->dev = dev;
 208         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 209                 skb->pkt_type = PACKET_BROADCAST;
 210         else
 211                 skb->pkt_type = PACKET_MULTICAST;
 212
 213         return 0;
 214 }
 215
 216 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 217 {
 218         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 219 }
 220
 221
 222 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 223                             const unsigned char *addr)
 224 {
 225         u32 val = __get_unaligned_cpu32(addr + 2);
 226
 227         val ^= macvlan_hash_mix(vlan);
 228         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 229 }
 230
 231 static void macvlan_broadcast(struct sk_buff *skb,
 232                               const struct macvlan_port *port,
 233                               struct net_device *src,
 234                               enum macvlan_mode mode)
 235 {
 236         const struct ethhdr *eth = eth_hdr(skb);
 237         const struct macvlan_dev *vlan;
 238         struct sk_buff *nskb;
 239         unsigned int i;
 240         int err;
 241         unsigned int hash;
 242
 243         if (skb->protocol == htons(ETH_P_PAUSE))
 244                 return;
 245
 246         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 247                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 248                         if (vlan->dev == src || !(vlan->mode & mode))
 249                                 continue;
 250
 251                         hash = mc_hash(vlan, eth->h_dest);
 252                         if (!test_bit(hash, vlan->mc_filter))
 253                                 continue;
 254
 255                         err = NET_RX_DROP;
 256                         nskb = skb_clone(skb, GFP_ATOMIC);
 257                         if (likely(nskb))
 258                                 err = macvlan_broadcast_one(
 259                                         nskb, vlan, eth,
 260                                         mode == MACVLAN_MODE_BRIDGE) ?:
 261                                       netif_rx_ni(nskb);
 262                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 263                                          err == NET_RX_SUCCESS, true);
 264                 }
 265         }
 266 }
 267
 268 static void macvlan_process_broadcast(struct work_struct *w)
 269 {
 270         struct macvlan_port *port = container_of(w, struct macvlan_port,
 271                                                  bc_work);
 272         struct sk_buff *skb;
 273         struct sk_buff_head list;
 274
 275         __skb_queue_head_init(&list);
 276
 277         spin_lock_bh(&port->bc_queue.lock);
 278         skb_queue_splice_tail_init(&port->bc_queue, &list);
 279         spin_unlock_bh(&port->bc_queue.lock);
 280
 281         while ((skb = __skb_dequeue(&list))) {
 282                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
 283
 284                 rcu_read_lock();
 285
 286                 if (!src)
 287                         /* frame comes from an external address */
 288                         macvlan_broadcast(skb, port, NULL,
 289                                           MACVLAN_MODE_PRIVATE |
 290                                           MACVLAN_MODE_VEPA    |
 291                                           MACVLAN_MODE_PASSTHRU|
 292                                           MACVLAN_MODE_BRIDGE);
 293                 else if (src->mode == MACVLAN_MODE_VEPA)
 294                         /* flood to everyone except source */
 295                         macvlan_broadcast(skb, port, src->dev,
 296                                           MACVLAN_MODE_VEPA |
 297                                           MACVLAN_MODE_BRIDGE);
 298                 else
 299                         /*
 300                          * flood only to VEPA ports, bridge ports
 301                          * already saw the frame on the way out.
 302                          */
 303                         macvlan_broadcast(skb, port, src->dev,
 304                                           MACVLAN_MODE_VEPA);
 305
 306                 rcu_read_unlock();
 307
 308                 if (src)
 309                         dev_put(src->dev);
 310                 kfree_skb(skb);
 311         }
 312 }
 313
 314 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 315                                       const struct macvlan_dev *src,
 316                                       struct sk_buff *skb)
 317 {
 318         struct sk_buff *nskb;
 319         int err = -ENOMEM;
 320
 321         nskb = skb_clone(skb, GFP_ATOMIC);
 322         if (!nskb)
 323                 goto err;
 324
 325         MACVLAN_SKB_CB(nskb)->src = src;
 326
 327         spin_lock(&port->bc_queue.lock);
 328         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 329                 if (src)
 330                         dev_hold(src->dev);
 331                 __skb_queue_tail(&port->bc_queue, nskb);
 332                 err = 0;
 333         }
 334         spin_unlock(&port->bc_queue.lock);
 335
 336         if (err)
 337                 goto free_nskb;
 338
 339         schedule_work(&port->bc_work);
 340         return;
 341
 342 free_nskb:
 343         kfree_skb(nskb);
 344 err:
 345         atomic_long_inc(&skb->dev->rx_dropped);
 346 }
 347
 348 static void macvlan_flush_sources(struct macvlan_port *port,
 349                                   struct macvlan_dev *vlan)
 350 {
 351         int i;
 352
 353         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 354                 struct hlist_node *h, *n;
 355
 356                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 357                         struct macvlan_source_entry *entry;
 358
 359                         entry = hlist_entry(h, struct macvlan_source_entry,
 360                                             hlist);
 361                         if (entry->vlan == vlan)
 362                                 macvlan_hash_del_source(entry);
 363                 }
 364         }
 365         vlan->macaddr_count = 0;
 366 }
 367
 368 static void macvlan_forward_source_one(struct sk_buff *skb,
 369                                        struct macvlan_dev *vlan)
 370 {
 371         struct sk_buff *nskb;
 372         struct net_device *dev;
 373         int len;
 374         int ret;
 375
 376         dev = vlan->dev;
 377         if (unlikely(!(dev->flags & IFF_UP)))
 378                 return;
 379
 380         nskb = skb_clone(skb, GFP_ATOMIC);
 381         if (!nskb)
 382                 return;
 383
 384         len = nskb->len + ETH_HLEN;
 385         nskb->dev = dev;
 386         nskb->pkt_type = PACKET_HOST;
 387
 388         ret = netif_rx(nskb);
 389         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 390 }
 391
 392 static void macvlan_forward_source(struct sk_buff *skb,
 393                                    struct macvlan_port *port,
 394                                    const unsigned char *addr)
 395 {
 396         struct macvlan_source_entry *entry;
 397         u32 idx = macvlan_eth_hash(addr);
 398         struct hlist_head *h = &port->vlan_source_hash[idx];
 399
 400         hlist_for_each_entry_rcu(entry, h, hlist) {
 401                 if (ether_addr_equal_64bits(entry->addr, addr))
 402                         if (entry->vlan->dev->flags & IFF_UP)
 403                                 macvlan_forward_source_one(skb, entry->vlan);
 404         }
 405 }
 406
 407 /* called under rcu_read_lock() from netif_receive_skb */
 408 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 409 {
 410         struct macvlan_port *port;
 411         struct sk_buff *skb = *pskb;
 412         const struct ethhdr *eth = eth_hdr(skb);
 413         const struct macvlan_dev *vlan;
 414         const struct macvlan_dev *src;
 415         struct net_device *dev;
 416         unsigned int len = 0;
 417         int ret;
 418         rx_handler_result_t handle_res;
 419
 420         port = macvlan_port_get_rcu(skb->dev);
 421         if (is_multicast_ether_addr(eth->h_dest)) {
 422                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 423                 if (!skb)
 424                         return RX_HANDLER_CONSUMED;
 425                 *pskb = skb;
 426                 eth = eth_hdr(skb);
 427                 macvlan_forward_source(skb, port, eth->h_source);
 428                 src = macvlan_hash_lookup(port, eth->h_source);
 429                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 430                     src->mode != MACVLAN_MODE_BRIDGE) {
 431                         /* forward to original port. */
 432                         vlan = src;
 433                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 434                               netif_rx(skb);
 435                         handle_res = RX_HANDLER_CONSUMED;
 436                         goto out;
 437                 }
 438
 439                 macvlan_broadcast_enqueue(port, src, skb);
 440
 441                 return RX_HANDLER_PASS;
 442         }
 443
 444         macvlan_forward_source(skb, port, eth->h_source);
 445         if (port->passthru)
 446                 vlan = list_first_or_null_rcu(&port->vlans,
 447                                               struct macvlan_dev, list);
 448         else
 449                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 450         if (vlan == NULL)
 451                 return RX_HANDLER_PASS;
 452
 453         dev = vlan->dev;
 454         if (unlikely(!(dev->flags & IFF_UP))) {
 455                 kfree_skb(skb);
 456                 return RX_HANDLER_CONSUMED;
 457         }
 458         len = skb->len + ETH_HLEN;
 459         skb = skb_share_check(skb, GFP_ATOMIC);
 460         if (!skb) {
 461                 ret = NET_RX_DROP;
 462                 handle_res = RX_HANDLER_CONSUMED;
 463                 goto out;
 464         }
 465
 466         *pskb = skb;
 467         skb->dev = dev;
 468         skb->pkt_type = PACKET_HOST;
 469
 470         ret = NET_RX_SUCCESS;
 471         handle_res = RX_HANDLER_ANOTHER;
 472 out:
 473         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 474         return handle_res;
 475 }
 476
 477 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 478 {
 479         const struct macvlan_dev *vlan = netdev_priv(dev);
 480         const struct macvlan_port *port = vlan->port;
 481         const struct macvlan_dev *dest;
 482
 483         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 484                 const struct ethhdr *eth = (void *)skb->data;
 485
 486                 /* send to other bridge ports directly */
 487                 if (is_multicast_ether_addr(eth->h_dest)) {
 488                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 489                         goto xmit_world;
 490                 }
 491
 492                 dest = macvlan_hash_lookup(port, eth->h_dest);
 493                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 494                         /* send to lowerdev first for its network taps */
 495                         dev_forward_skb(vlan->lowerdev, skb);
 496
 497                         return NET_XMIT_SUCCESS;
 498                 }
 499         }
 500
 501 xmit_world:
 502         skb->dev = vlan->lowerdev;
 503         return dev_queue_xmit(skb);
 504 }
 505
 506 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 507 {
 508 #ifdef CONFIG_NET_POLL_CONTROLLER
 509         if (vlan->netpoll)
 510                 netpoll_send_skb(vlan->netpoll, skb);
 511 #else
 512         BUG();
 513 #endif
 514         return NETDEV_TX_OK;
 515 }
 516
 517 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 518                                       struct net_device *dev)
 519 {
 520         unsigned int len = skb->len;
 521         int ret;
 522         struct macvlan_dev *vlan = netdev_priv(dev);
 523
 524         if (unlikely(netpoll_tx_running(dev)))
 525                 return macvlan_netpoll_send_skb(vlan, skb);
 526
 527         if (vlan->fwd_priv) {
 528                 skb->dev = vlan->lowerdev;
 529                 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
 530         } else {
 531                 ret = macvlan_queue_xmit(skb, dev);
 532         }
 533
 534         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 535                 struct vlan_pcpu_stats *pcpu_stats;
 536
 537                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 538                 u64_stats_update_begin(&pcpu_stats->syncp);
 539                 pcpu_stats->tx_packets++;
 540                 pcpu_stats->tx_bytes += len;
 541                 u64_stats_update_end(&pcpu_stats->syncp);
 542         } else {
 543                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 544         }
 545         return ret;
 546 }
 547
 548 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 549                                unsigned short type, const void *daddr,
 550                                const void *saddr, unsigned len)
 551 {
 552         const struct macvlan_dev *vlan = netdev_priv(dev);
 553         struct net_device *lowerdev = vlan->lowerdev;
 554
 555         return dev_hard_header(skb, lowerdev, type, daddr,
 556                                saddr ? : dev->dev_addr, len);
 557 }
 558
 559 static const struct header_ops macvlan_hard_header_ops = {
 560         .create         = macvlan_hard_header,
 561         .parse          = eth_header_parse,
 562         .cache          = eth_header_cache,
 563         .cache_update   = eth_header_cache_update,
 564 };
 565
 566 static struct rtnl_link_ops macvlan_link_ops;
 567
 568 static int macvlan_open(struct net_device *dev)
 569 {
 570         struct macvlan_dev *vlan = netdev_priv(dev);
 571         struct net_device *lowerdev = vlan->lowerdev;
 572         int err;
 573
 574         if (vlan->port->passthru) {
 575                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 576                         err = dev_set_promiscuity(lowerdev, 1);
 577                         if (err < 0)
 578                                 goto out;
 579                 }
 580                 goto hash_add;
 581         }
 582
 583         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
 584             dev->rtnl_link_ops == &macvlan_link_ops) {
 585                 vlan->fwd_priv =
 586                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 587
 588                 /* If we get a NULL pointer back, or if we get an error
 589                  * then we should just fall through to the non accelerated path
 590                  */
 591                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
 592                         vlan->fwd_priv = NULL;
 593                 } else
 594                         return 0;
 595         }
 596
 597         err = -EBUSY;
 598         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 599                 goto out;
 600
 601         err = dev_uc_add(lowerdev, dev->dev_addr);
 602         if (err < 0)
 603                 goto out;
 604         if (dev->flags & IFF_ALLMULTI) {
 605                 err = dev_set_allmulti(lowerdev, 1);
 606                 if (err < 0)
 607                         goto del_unicast;
 608         }
 609
 610         if (dev->flags & IFF_PROMISC) {
 611                 err = dev_set_promiscuity(lowerdev, 1);
 612                 if (err < 0)
 613                         goto clear_multi;
 614         }
 615
 616 hash_add:
 617         macvlan_hash_add(vlan);
 618         return 0;
 619
 620 clear_multi:
 621         dev_set_allmulti(lowerdev, -1);
 622 del_unicast:
 623         dev_uc_del(lowerdev, dev->dev_addr);
 624 out:
 625         if (vlan->fwd_priv) {
 626                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 627                                                            vlan->fwd_priv);
 628                 vlan->fwd_priv = NULL;
 629         }
 630         return err;
 631 }
 632
 633 static int macvlan_stop(struct net_device *dev)
 634 {
 635         struct macvlan_dev *vlan = netdev_priv(dev);
 636         struct net_device *lowerdev = vlan->lowerdev;
 637
 638         if (vlan->fwd_priv) {
 639                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 640                                                            vlan->fwd_priv);
 641                 vlan->fwd_priv = NULL;
 642                 return 0;
 643         }
 644
 645         dev_uc_unsync(lowerdev, dev);
 646         dev_mc_unsync(lowerdev, dev);
 647
 648         if (vlan->port->passthru) {
 649                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 650                         dev_set_promiscuity(lowerdev, -1);
 651                 goto hash_del;
 652         }
 653
 654         if (dev->flags & IFF_ALLMULTI)
 655                 dev_set_allmulti(lowerdev, -1);
 656
 657         if (dev->flags & IFF_PROMISC)
 658                 dev_set_promiscuity(lowerdev, -1);
 659
 660         dev_uc_del(lowerdev, dev->dev_addr);
 661
 662 hash_del:
 663         macvlan_hash_del(vlan, !dev->dismantle);
 664         return 0;
 665 }
 666
 667 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 668 {
 669         struct macvlan_dev *vlan = netdev_priv(dev);
 670         struct net_device *lowerdev = vlan->lowerdev;
 671         int err;
 672
 673         if (!(dev->flags & IFF_UP)) {
 674                 /* Just copy in the new address */
 675                 ether_addr_copy(dev->dev_addr, addr);
 676         } else {
 677                 /* Rehash and update the device filters */
 678                 if (macvlan_addr_busy(vlan->port, addr))
 679                         return -EBUSY;
 680
 681                 if (!vlan->port->passthru) {
 682                         err = dev_uc_add(lowerdev, addr);
 683                         if (err)
 684                                 return err;
 685
 686                         dev_uc_del(lowerdev, dev->dev_addr);
 687                 }
 688
 689                 macvlan_hash_change_addr(vlan, addr);
 690         }
 691         return 0;
 692 }
 693
 694 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 695 {
 696         struct macvlan_dev *vlan = netdev_priv(dev);
 697         struct sockaddr *addr = p;
 698
 699         if (!is_valid_ether_addr(addr->sa_data))
 700                 return -EADDRNOTAVAIL;
 701
 702         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 703                 dev_set_mac_address(vlan->lowerdev, addr);
 704                 return 0;
 705         }
 706
 707         return macvlan_sync_address(dev, addr->sa_data);
 708 }
 709
 710 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 711 {
 712         struct macvlan_dev *vlan = netdev_priv(dev);
 713         struct net_device *lowerdev = vlan->lowerdev;
 714
 715         if (dev->flags & IFF_UP) {
 716                 if (change & IFF_ALLMULTI)
 717                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 718                 if (change & IFF_PROMISC)
 719                         dev_set_promiscuity(lowerdev,
 720                                             dev->flags & IFF_PROMISC ? 1 : -1);
 721
 722         }
 723 }
 724
 725 static void macvlan_set_mac_lists(struct net_device *dev)
 726 {
 727         struct macvlan_dev *vlan = netdev_priv(dev);
 728
 729         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 730                 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
 731         } else {
 732                 struct netdev_hw_addr *ha;
 733                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 734
 735                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 736                 netdev_for_each_mc_addr(ha, dev) {
 737                         __set_bit(mc_hash(vlan, ha->addr), filter);
 738                 }
 739
 740                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 741
 742                 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 743         }
 744         dev_uc_sync(vlan->lowerdev, dev);
 745         dev_mc_sync(vlan->lowerdev, dev);
 746 }
 747
 748 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 749 {
 750         struct macvlan_dev *vlan = netdev_priv(dev);
 751
 752         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 753                 return -EINVAL;
 754         dev->mtu = new_mtu;
 755         return 0;
 756 }
 757
 758 /*
 759  * macvlan network devices have devices nesting below it and are a special
 760  * "super class" of normal network devices; split their locks off into a
 761  * separate class since they always nest.
 762  */
 763 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 764 static struct lock_class_key macvlan_netdev_addr_lock_key;
 765
 766 #define ALWAYS_ON_FEATURES \
 767         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX | \
 768          NETIF_F_GSO_ROBUST)
 769
 770 #define MACVLAN_FEATURES \
 771         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 772          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
 773          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 774          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 775
 776 #define MACVLAN_STATE_MASK \
 777         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 778
 779 static int macvlan_get_nest_level(struct net_device *dev)
 780 {
 781         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 782 }
 783
 784 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 785                                           struct netdev_queue *txq,
 786                                           void *_unused)
 787 {
 788         lockdep_set_class(&txq->_xmit_lock,
 789                           &macvlan_netdev_xmit_lock_key);
 790 }
 791
 792 static void macvlan_set_lockdep_class(struct net_device *dev)
 793 {
 794         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 795                                        &macvlan_netdev_addr_lock_key,
 796                                        macvlan_get_nest_level(dev));
 797         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 798 }
 799
 800 static int macvlan_init(struct net_device *dev)
 801 {
 802         struct macvlan_dev *vlan = netdev_priv(dev);
 803         const struct net_device *lowerdev = vlan->lowerdev;
 804         struct macvlan_port *port = vlan->port;
 805
 806         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 807                                   (lowerdev->state & MACVLAN_STATE_MASK);
 808         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 809         dev->features           |= ALWAYS_ON_FEATURES;
 810         dev->hw_features        |= NETIF_F_LRO;
 811         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 812         dev->gso_max_size       = lowerdev->gso_max_size;
 813         dev->gso_max_segs       = lowerdev->gso_max_segs;
 814         dev->hard_header_len    = lowerdev->hard_header_len;
 815
 816         macvlan_set_lockdep_class(dev);
 817
 818         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 819         if (!vlan->pcpu_stats)
 820                 return -ENOMEM;
 821
 822         port->count += 1;
 823
 824         return 0;
 825 }
 826
 827 static void macvlan_uninit(struct net_device *dev)
 828 {
 829         struct macvlan_dev *vlan = netdev_priv(dev);
 830         struct macvlan_port *port = vlan->port;
 831
 832         free_percpu(vlan->pcpu_stats);
 833
 834         macvlan_flush_sources(port, vlan);
 835         port->count -= 1;
 836         if (!port->count)
 837                 macvlan_port_destroy(port->dev);
 838 }
 839
 840 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 841                                                          struct rtnl_link_stats64 *stats)
 842 {
 843         struct macvlan_dev *vlan = netdev_priv(dev);
 844
 845         if (vlan->pcpu_stats) {
 846                 struct vlan_pcpu_stats *p;
 847                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 848                 u32 rx_errors = 0, tx_dropped = 0;
 849                 unsigned int start;
 850                 int i;
 851
 852                 for_each_possible_cpu(i) {
 853                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 854                         do {
 855                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 856                                 rx_packets      = p->rx_packets;
 857                                 rx_bytes        = p->rx_bytes;
 858                                 rx_multicast    = p->rx_multicast;
 859                                 tx_packets      = p->tx_packets;
 860                                 tx_bytes        = p->tx_bytes;
 861                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 862
 863                         stats->rx_packets       += rx_packets;
 864                         stats->rx_bytes         += rx_bytes;
 865                         stats->multicast        += rx_multicast;
 866                         stats->tx_packets       += tx_packets;
 867                         stats->tx_bytes         += tx_bytes;
 868                         /* rx_errors & tx_dropped are u32, updated
 869                          * without syncp protection.
 870                          */
 871                         rx_errors       += p->rx_errors;
 872                         tx_dropped      += p->tx_dropped;
 873                 }
 874                 stats->rx_errors        = rx_errors;
 875                 stats->rx_dropped       = rx_errors;
 876                 stats->tx_dropped       = tx_dropped;
 877         }
 878         return stats;
 879 }
 880
 881 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 882                                    __be16 proto, u16 vid)
 883 {
 884         struct macvlan_dev *vlan = netdev_priv(dev);
 885         struct net_device *lowerdev = vlan->lowerdev;
 886
 887         return vlan_vid_add(lowerdev, proto, vid);
 888 }
 889
 890 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 891                                     __be16 proto, u16 vid)
 892 {
 893         struct macvlan_dev *vlan = netdev_priv(dev);
 894         struct net_device *lowerdev = vlan->lowerdev;
 895
 896         vlan_vid_del(lowerdev, proto, vid);
 897         return 0;
 898 }
 899
 900 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 901                            struct net_device *dev,
 902                            const unsigned char *addr, u16 vid,
 903                            u16 flags)
 904 {
 905         struct macvlan_dev *vlan = netdev_priv(dev);
 906         int err = -EINVAL;
 907
 908         /* Support unicast filter only on passthru devices.
 909          * Multicast filter should be allowed on all devices.
 910          */
 911         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 912                 return -EOPNOTSUPP;
 913
 914         if (flags & NLM_F_REPLACE)
 915                 return -EOPNOTSUPP;
 916
 917         if (is_unicast_ether_addr(addr))
 918                 err = dev_uc_add_excl(dev, addr);
 919         else if (is_multicast_ether_addr(addr))
 920                 err = dev_mc_add_excl(dev, addr);
 921
 922         return err;
 923 }
 924
 925 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 926                            struct net_device *dev,
 927                            const unsigned char *addr, u16 vid)
 928 {
 929         struct macvlan_dev *vlan = netdev_priv(dev);
 930         int err = -EINVAL;
 931
 932         /* Support unicast filter only on passthru devices.
 933          * Multicast filter should be allowed on all devices.
 934          */
 935         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 936                 return -EOPNOTSUPP;
 937
 938         if (is_unicast_ether_addr(addr))
 939                 err = dev_uc_del(dev, addr);
 940         else if (is_multicast_ether_addr(addr))
 941                 err = dev_mc_del(dev, addr);
 942
 943         return err;
 944 }
 945
 946 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 947                                         struct ethtool_drvinfo *drvinfo)
 948 {
 949         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 950         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 951 }
 952
 953 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
 954                                               struct ethtool_link_ksettings *cmd)
 955 {
 956         const struct macvlan_dev *vlan = netdev_priv(dev);
 957
 958         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
 959 }
 960
 961 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 962                                               netdev_features_t features)
 963 {
 964         struct macvlan_dev *vlan = netdev_priv(dev);
 965         netdev_features_t lowerdev_features = vlan->lowerdev->features;
 966         netdev_features_t mask;
 967
 968         features |= NETIF_F_ALL_FOR_ALL;
 969         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 970         mask = features;
 971
 972         lowerdev_features &= (features | ~NETIF_F_LRO);
 973         features = netdev_increment_features(lowerdev_features, features, mask);
 974         features |= ALWAYS_ON_FEATURES;
 975         features &= ~NETIF_F_NETNS_LOCAL;
 976
 977         return features;
 978 }
 979
 980 #ifdef CONFIG_NET_POLL_CONTROLLER
 981 static void macvlan_dev_poll_controller(struct net_device *dev)
 982 {
 983         return;
 984 }
 985
 986 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
 987 {
 988         struct macvlan_dev *vlan = netdev_priv(dev);
 989         struct net_device *real_dev = vlan->lowerdev;
 990         struct netpoll *netpoll;
 991         int err = 0;
 992
 993         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
 994         err = -ENOMEM;
 995         if (!netpoll)
 996                 goto out;
 997
 998         err = __netpoll_setup(netpoll, real_dev);
 999         if (err) {
1000                 kfree(netpoll);
1001                 goto out;
1002         }
1003
1004         vlan->netpoll = netpoll;
1005
1006 out:
1007         return err;
1008 }
1009
1010 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1011 {
1012         struct macvlan_dev *vlan = netdev_priv(dev);
1013         struct netpoll *netpoll = vlan->netpoll;
1014
1015         if (!netpoll)
1016                 return;
1017
1018         vlan->netpoll = NULL;
1019
1020         __netpoll_free_async(netpoll);
1021 }
1022 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1023
1024 static int macvlan_dev_get_iflink(const struct net_device *dev)
1025 {
1026         struct macvlan_dev *vlan = netdev_priv(dev);
1027
1028         return vlan->lowerdev->ifindex;
1029 }
1030
1031 static const struct ethtool_ops macvlan_ethtool_ops = {
1032         .get_link               = ethtool_op_get_link,
1033         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1034         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1035 };
1036
1037 static const struct net_device_ops macvlan_netdev_ops = {
1038         .ndo_init               = macvlan_init,
1039         .ndo_uninit             = macvlan_uninit,
1040         .ndo_open               = macvlan_open,
1041         .ndo_stop               = macvlan_stop,
1042         .ndo_start_xmit         = macvlan_start_xmit,
1043         .ndo_change_mtu         = macvlan_change_mtu,
1044         .ndo_fix_features       = macvlan_fix_features,
1045         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1046         .ndo_set_mac_address    = macvlan_set_mac_address,
1047         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1048         .ndo_get_stats64        = macvlan_dev_get_stats64,
1049         .ndo_validate_addr      = eth_validate_addr,
1050         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1051         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1052         .ndo_fdb_add            = macvlan_fdb_add,
1053         .ndo_fdb_del            = macvlan_fdb_del,
1054         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1055         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1056 #ifdef CONFIG_NET_POLL_CONTROLLER
1057         .ndo_poll_controller    = macvlan_dev_poll_controller,
1058         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1059         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1060 #endif
1061         .ndo_get_iflink         = macvlan_dev_get_iflink,
1062         .ndo_features_check     = passthru_features_check,
1063 };
1064
1065 void macvlan_common_setup(struct net_device *dev)
1066 {
1067         ether_setup(dev);
1068
1069         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1070         netif_keep_dst(dev);
1071         dev->priv_flags        |= IFF_UNICAST_FLT;
1072         dev->netdev_ops         = &macvlan_netdev_ops;
1073         dev->destructor         = free_netdev;
1074         dev->header_ops         = &macvlan_hard_header_ops;
1075         dev->ethtool_ops        = &macvlan_ethtool_ops;
1076 }
1077 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1078
1079 static void macvlan_setup(struct net_device *dev)
1080 {
1081         macvlan_common_setup(dev);
1082         dev->priv_flags |= IFF_NO_QUEUE;
1083 }
1084
1085 static int macvlan_port_create(struct net_device *dev)
1086 {
1087         struct macvlan_port *port;
1088         unsigned int i;
1089         int err;
1090
1091         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1092                 return -EINVAL;
1093
1094         if (netif_is_ipvlan_port(dev))
1095                 return -EBUSY;
1096
1097         port = kzalloc(sizeof(*port), GFP_KERNEL);
1098         if (port == NULL)
1099                 return -ENOMEM;
1100
1101         port->passthru = false;
1102         port->dev = dev;
1103         INIT_LIST_HEAD(&port->vlans);
1104         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1105                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1106         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1107                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1108
1109         skb_queue_head_init(&port->bc_queue);
1110         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1111
1112         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1113         if (err)
1114                 kfree(port);
1115         else
1116                 dev->priv_flags |= IFF_MACVLAN_PORT;
1117         return err;
1118 }
1119
1120 static void macvlan_port_destroy(struct net_device *dev)
1121 {
1122         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1123
1124         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1125         netdev_rx_handler_unregister(dev);
1126
1127         /* After this point, no packet can schedule bc_work anymore,
1128          * but we need to cancel it and purge left skbs if any.
1129          */
1130         cancel_work_sync(&port->bc_work);
1131         __skb_queue_purge(&port->bc_queue);
1132
1133         kfree_rcu(port, rcu);
1134 }
1135
1136 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
1137 {
1138         if (tb[IFLA_ADDRESS]) {
1139                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1140                         return -EINVAL;
1141                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1142                         return -EADDRNOTAVAIL;
1143         }
1144
1145         if (data && data[IFLA_MACVLAN_FLAGS] &&
1146             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1147                 return -EINVAL;
1148
1149         if (data && data[IFLA_MACVLAN_MODE]) {
1150                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1151                 case MACVLAN_MODE_PRIVATE:
1152                 case MACVLAN_MODE_VEPA:
1153                 case MACVLAN_MODE_BRIDGE:
1154                 case MACVLAN_MODE_PASSTHRU:
1155                 case MACVLAN_MODE_SOURCE:
1156                         break;
1157                 default:
1158                         return -EINVAL;
1159                 }
1160         }
1161
1162         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1163                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1164                 case MACVLAN_MACADDR_ADD:
1165                 case MACVLAN_MACADDR_DEL:
1166                 case MACVLAN_MACADDR_FLUSH:
1167                 case MACVLAN_MACADDR_SET:
1168                         break;
1169                 default:
1170                         return -EINVAL;
1171                 }
1172         }
1173
1174         if (data && data[IFLA_MACVLAN_MACADDR]) {
1175                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1176                         return -EINVAL;
1177
1178                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1179                         return -EADDRNOTAVAIL;
1180         }
1181
1182         if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1183                 return -EINVAL;
1184
1185         return 0;
1186 }
1187
1188 /**
1189  * reconfigure list of remote source mac address
1190  * (only for macvlan devices in source mode)
1191  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1192  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1193  */
1194 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1195                                       struct nlattr *data[])
1196 {
1197         char *addr = NULL;
1198         int ret, rem, len;
1199         struct nlattr *nla, *head;
1200         struct macvlan_source_entry *entry;
1201
1202         if (data[IFLA_MACVLAN_MACADDR])
1203                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1204
1205         if (mode == MACVLAN_MACADDR_ADD) {
1206                 if (!addr)
1207                         return -EINVAL;
1208
1209                 return macvlan_hash_add_source(vlan, addr);
1210
1211         } else if (mode == MACVLAN_MACADDR_DEL) {
1212                 if (!addr)
1213                         return -EINVAL;
1214
1215                 entry = macvlan_hash_lookup_source(vlan, addr);
1216                 if (entry) {
1217                         macvlan_hash_del_source(entry);
1218                         vlan->macaddr_count--;
1219                 }
1220         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1221                 macvlan_flush_sources(vlan->port, vlan);
1222         } else if (mode == MACVLAN_MACADDR_SET) {
1223                 macvlan_flush_sources(vlan->port, vlan);
1224
1225                 if (addr) {
1226                         ret = macvlan_hash_add_source(vlan, addr);
1227                         if (ret)
1228                                 return ret;
1229                 }
1230
1231                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1232                         return 0;
1233
1234                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1235                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1236
1237                 nla_for_each_attr(nla, head, len, rem) {
1238                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1239                             nla_len(nla) != ETH_ALEN)
1240                                 continue;
1241
1242                         addr = nla_data(nla);
1243                         ret = macvlan_hash_add_source(vlan, addr);
1244                         if (ret)
1245                                 return ret;
1246                 }
1247         } else {
1248                 return -EINVAL;
1249         }
1250
1251         return 0;
1252 }
1253
1254 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1255                            struct nlattr *tb[], struct nlattr *data[])
1256 {
1257         struct macvlan_dev *vlan = netdev_priv(dev);
1258         struct macvlan_port *port;
1259         struct net_device *lowerdev;
1260         int err;
1261         int macmode;
1262
1263         if (!tb[IFLA_LINK])
1264                 return -EINVAL;
1265
1266         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1267         if (lowerdev == NULL)
1268                 return -ENODEV;
1269
1270         /* When creating macvlans or macvtaps on top of other macvlans - use
1271          * the real device as the lowerdev.
1272          */
1273         if (netif_is_macvlan(lowerdev))
1274                 lowerdev = macvlan_dev_real_dev(lowerdev);
1275
1276         if (!tb[IFLA_MTU])
1277                 dev->mtu = lowerdev->mtu;
1278         else if (dev->mtu > lowerdev->mtu)
1279                 return -EINVAL;
1280
1281         if (!tb[IFLA_ADDRESS])
1282                 eth_hw_addr_random(dev);
1283
1284         if (!macvlan_port_exists(lowerdev)) {
1285                 err = macvlan_port_create(lowerdev);
1286                 if (err < 0)
1287                         return err;
1288         }
1289         port = macvlan_port_get_rtnl(lowerdev);
1290
1291         /* Only 1 macvlan device can be created in passthru mode */
1292         if (port->passthru)
1293                 return -EINVAL;
1294
1295         vlan->lowerdev = lowerdev;
1296         vlan->dev      = dev;
1297         vlan->port     = port;
1298         vlan->set_features = MACVLAN_FEATURES;
1299         vlan->nest_level = dev_get_nest_level(lowerdev, netif_is_macvlan) + 1;
1300
1301         vlan->mode     = MACVLAN_MODE_VEPA;
1302         if (data && data[IFLA_MACVLAN_MODE])
1303                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1304
1305         if (data && data[IFLA_MACVLAN_FLAGS])
1306                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1307
1308         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1309                 if (port->count)
1310                         return -EINVAL;
1311                 port->passthru = true;
1312                 eth_hw_addr_inherit(dev, lowerdev);
1313         }
1314
1315         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1316                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1317                         return -EINVAL;
1318                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1319                 err = macvlan_changelink_sources(vlan, macmode, data);
1320                 if (err)
1321                         return err;
1322         }
1323
1324         err = register_netdevice(dev);
1325         if (err < 0)
1326                 return err;
1327
1328         dev->priv_flags |= IFF_MACVLAN;
1329         err = netdev_upper_dev_link(lowerdev, dev);
1330         if (err)
1331                 goto unregister_netdev;
1332
1333         list_add_tail_rcu(&vlan->list, &port->vlans);
1334         netif_stacked_transfer_operstate(lowerdev, dev);
1335         linkwatch_fire_event(dev);
1336
1337         return 0;
1338
1339 unregister_netdev:
1340         unregister_netdevice(dev);
1341
1342         return err;
1343 }
1344 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1345
1346 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1347                            struct nlattr *tb[], struct nlattr *data[])
1348 {
1349         return macvlan_common_newlink(src_net, dev, tb, data);
1350 }
1351
1352 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1353 {
1354         struct macvlan_dev *vlan = netdev_priv(dev);
1355
1356         if (vlan->mode == MACVLAN_MODE_SOURCE)
1357                 macvlan_flush_sources(vlan->port, vlan);
1358         list_del_rcu(&vlan->list);
1359         unregister_netdevice_queue(dev, head);
1360         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1361 }
1362 EXPORT_SYMBOL_GPL(macvlan_dellink);
1363
1364 static int macvlan_changelink(struct net_device *dev,
1365                 struct nlattr *tb[], struct nlattr *data[])
1366 {
1367         struct macvlan_dev *vlan = netdev_priv(dev);
1368         enum macvlan_mode mode;
1369         bool set_mode = false;
1370         enum macvlan_macaddr_mode macmode;
1371         int ret;
1372
1373         /* Validate mode, but don't set yet: setting flags may fail. */
1374         if (data && data[IFLA_MACVLAN_MODE]) {
1375                 set_mode = true;
1376                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1377                 /* Passthrough mode can't be set or cleared dynamically */
1378                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1379                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1380                         return -EINVAL;
1381                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1382                     vlan->mode != mode)
1383                         macvlan_flush_sources(vlan->port, vlan);
1384         }
1385
1386         if (data && data[IFLA_MACVLAN_FLAGS]) {
1387                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1388                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1389                 if (vlan->port->passthru && promisc) {
1390                         int err;
1391
1392                         if (flags & MACVLAN_FLAG_NOPROMISC)
1393                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1394                         else
1395                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1396                         if (err < 0)
1397                                 return err;
1398                 }
1399                 vlan->flags = flags;
1400         }
1401         if (set_mode)
1402                 vlan->mode = mode;
1403         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1404                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1405                         return -EINVAL;
1406                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1407                 ret = macvlan_changelink_sources(vlan, macmode, data);
1408                 if (ret)
1409                         return ret;
1410         }
1411         return 0;
1412 }
1413
1414 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1415 {
1416         if (vlan->macaddr_count == 0)
1417                 return 0;
1418         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1419                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1420 }
1421
1422 static size_t macvlan_get_size(const struct net_device *dev)
1423 {
1424         struct macvlan_dev *vlan = netdev_priv(dev);
1425
1426         return (0
1427                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1428                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1429                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1430                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1431                 );
1432 }
1433
1434 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1435                                      const struct macvlan_dev *vlan,
1436                                      const int i)
1437 {
1438         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1439         struct macvlan_source_entry *entry;
1440
1441         hlist_for_each_entry_rcu(entry, h, hlist) {
1442                 if (entry->vlan != vlan)
1443                         continue;
1444                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1445                         return 1;
1446         }
1447         return 0;
1448 }
1449
1450 static int macvlan_fill_info(struct sk_buff *skb,
1451                                 const struct net_device *dev)
1452 {
1453         struct macvlan_dev *vlan = netdev_priv(dev);
1454         int i;
1455         struct nlattr *nest;
1456
1457         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1458                 goto nla_put_failure;
1459         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1460                 goto nla_put_failure;
1461         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1462                 goto nla_put_failure;
1463         if (vlan->macaddr_count > 0) {
1464                 nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1465                 if (nest == NULL)
1466                         goto nla_put_failure;
1467
1468                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1469                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1470                                 goto nla_put_failure;
1471                 }
1472                 nla_nest_end(skb, nest);
1473         }
1474         return 0;
1475
1476 nla_put_failure:
1477         return -EMSGSIZE;
1478 }
1479
1480 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1481         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1482         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1483         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1484         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1485         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1486         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1487 };
1488
1489 int macvlan_link_register(struct rtnl_link_ops *ops)
1490 {
1491         /* common fields */
1492         ops->priv_size          = sizeof(struct macvlan_dev);
1493         ops->validate           = macvlan_validate;
1494         ops->maxtype            = IFLA_MACVLAN_MAX;
1495         ops->policy             = macvlan_policy;
1496         ops->changelink         = macvlan_changelink;
1497         ops->get_size           = macvlan_get_size;
1498         ops->fill_info          = macvlan_fill_info;
1499
1500         return rtnl_link_register(ops);
1501 };
1502 EXPORT_SYMBOL_GPL(macvlan_link_register);
1503
1504 static struct net *macvlan_get_link_net(const struct net_device *dev)
1505 {
1506         return dev_net(macvlan_dev_real_dev(dev));
1507 }
1508
1509 static struct rtnl_link_ops macvlan_link_ops = {
1510         .kind           = "macvlan",
1511         .setup          = macvlan_setup,
1512         .newlink        = macvlan_newlink,
1513         .dellink        = macvlan_dellink,
1514         .get_link_net   = macvlan_get_link_net,
1515 };
1516
1517 static int macvlan_device_event(struct notifier_block *unused,
1518                                 unsigned long event, void *ptr)
1519 {
1520         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1521         struct macvlan_dev *vlan, *next;
1522         struct macvlan_port *port;
1523         LIST_HEAD(list_kill);
1524
1525         if (!macvlan_port_exists(dev))
1526                 return NOTIFY_DONE;
1527
1528         port = macvlan_port_get_rtnl(dev);
1529
1530         switch (event) {
1531         case NETDEV_UP:
1532         case NETDEV_CHANGE:
1533                 list_for_each_entry(vlan, &port->vlans, list)
1534                         netif_stacked_transfer_operstate(vlan->lowerdev,
1535                                                          vlan->dev);
1536                 break;
1537         case NETDEV_FEAT_CHANGE:
1538                 list_for_each_entry(vlan, &port->vlans, list) {
1539                         vlan->dev->gso_max_size = dev->gso_max_size;
1540                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1541                         netdev_update_features(vlan->dev);
1542                 }
1543                 break;
1544         case NETDEV_CHANGEMTU:
1545                 list_for_each_entry(vlan, &port->vlans, list) {
1546                         if (vlan->dev->mtu <= dev->mtu)
1547                                 continue;
1548                         dev_set_mtu(vlan->dev, dev->mtu);
1549                 }
1550                 break;
1551         case NETDEV_CHANGEADDR:
1552                 if (!port->passthru)
1553                         return NOTIFY_DONE;
1554
1555                 vlan = list_first_entry_or_null(&port->vlans,
1556                                                 struct macvlan_dev,
1557                                                 list);
1558
1559                 if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1560                         return NOTIFY_BAD;
1561
1562                 break;
1563         case NETDEV_UNREGISTER:
1564                 /* twiddle thumbs on netns device moves */
1565                 if (dev->reg_state != NETREG_UNREGISTERING)
1566                         break;
1567
1568                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1569                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1570                 unregister_netdevice_many(&list_kill);
1571                 break;
1572         case NETDEV_PRE_TYPE_CHANGE:
1573                 /* Forbid underlaying device to change its type. */
1574                 return NOTIFY_BAD;
1575
1576         case NETDEV_NOTIFY_PEERS:
1577         case NETDEV_BONDING_FAILOVER:
1578         case NETDEV_RESEND_IGMP:
1579                 /* Propagate to all vlans */
1580                 list_for_each_entry(vlan, &port->vlans, list)
1581                         call_netdevice_notifiers(event, vlan->dev);
1582         }
1583         return NOTIFY_DONE;
1584 }
1585
1586 static struct notifier_block macvlan_notifier_block __read_mostly = {
1587         .notifier_call  = macvlan_device_event,
1588 };
1589
1590 static int __init macvlan_init_module(void)
1591 {
1592         int err;
1593
1594         register_netdevice_notifier(&macvlan_notifier_block);
1595
1596         err = macvlan_link_register(&macvlan_link_ops);
1597         if (err < 0)
1598                 goto err1;
1599         return 0;
1600 err1:
1601         unregister_netdevice_notifier(&macvlan_notifier_block);
1602         return err;
1603 }
1604
1605 static void __exit macvlan_cleanup_module(void)
1606 {
1607         rtnl_link_unregister(&macvlan_link_ops);
1608         unregister_netdevice_notifier(&macvlan_notifier_block);
1609 }
1610
1611 module_init(macvlan_init_module);
1612 module_exit(macvlan_cleanup_module);
1613
1614 MODULE_LICENSE("GPL");
1615 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1616 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1617 MODULE_ALIAS_RTNL_LINK("macvlan");