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                 cond_resched();
 314         }
 315 }
 316
 317 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 318                                       const struct macvlan_dev *src,
 319                                       struct sk_buff *skb)
 320 {
 321         struct sk_buff *nskb;
 322         int err = -ENOMEM;
 323
 324         nskb = skb_clone(skb, GFP_ATOMIC);
 325         if (!nskb)
 326                 goto err;
 327
 328         MACVLAN_SKB_CB(nskb)->src = src;
 329
 330         spin_lock(&port->bc_queue.lock);
 331         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 332                 if (src)
 333                         dev_hold(src->dev);
 334                 __skb_queue_tail(&port->bc_queue, nskb);
 335                 err = 0;
 336         }
 337         spin_unlock(&port->bc_queue.lock);
 338
 339         schedule_work(&port->bc_work);
 340
 341         if (err)
 342                 goto free_nskb;
 343
 344         return;
 345
 346 free_nskb:
 347         kfree_skb(nskb);
 348 err:
 349         atomic_long_inc(&skb->dev->rx_dropped);
 350 }
 351
 352 static void macvlan_flush_sources(struct macvlan_port *port,
 353                                   struct macvlan_dev *vlan)
 354 {
 355         int i;
 356
 357         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 358                 struct hlist_node *h, *n;
 359
 360                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 361                         struct macvlan_source_entry *entry;
 362
 363                         entry = hlist_entry(h, struct macvlan_source_entry,
 364                                             hlist);
 365                         if (entry->vlan == vlan)
 366                                 macvlan_hash_del_source(entry);
 367                 }
 368         }
 369         vlan->macaddr_count = 0;
 370 }
 371
 372 static void macvlan_forward_source_one(struct sk_buff *skb,
 373                                        struct macvlan_dev *vlan)
 374 {
 375         struct sk_buff *nskb;
 376         struct net_device *dev;
 377         int len;
 378         int ret;
 379
 380         dev = vlan->dev;
 381         if (unlikely(!(dev->flags & IFF_UP)))
 382                 return;
 383
 384         nskb = skb_clone(skb, GFP_ATOMIC);
 385         if (!nskb)
 386                 return;
 387
 388         len = nskb->len + ETH_HLEN;
 389         nskb->dev = dev;
 390         nskb->pkt_type = PACKET_HOST;
 391
 392         ret = netif_rx(nskb);
 393         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 394 }
 395
 396 static void macvlan_forward_source(struct sk_buff *skb,
 397                                    struct macvlan_port *port,
 398                                    const unsigned char *addr)
 399 {
 400         struct macvlan_source_entry *entry;
 401         u32 idx = macvlan_eth_hash(addr);
 402         struct hlist_head *h = &port->vlan_source_hash[idx];
 403
 404         hlist_for_each_entry_rcu(entry, h, hlist) {
 405                 if (ether_addr_equal_64bits(entry->addr, addr))
 406                         if (entry->vlan->dev->flags & IFF_UP)
 407                                 macvlan_forward_source_one(skb, entry->vlan);
 408         }
 409 }
 410
 411 /* called under rcu_read_lock() from netif_receive_skb */
 412 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 413 {
 414         struct macvlan_port *port;
 415         struct sk_buff *skb = *pskb;
 416         const struct ethhdr *eth = eth_hdr(skb);
 417         const struct macvlan_dev *vlan;
 418         const struct macvlan_dev *src;
 419         struct net_device *dev;
 420         unsigned int len = 0;
 421         int ret;
 422         rx_handler_result_t handle_res;
 423
 424         /* Packets from dev_loopback_xmit() do not have L2 header, bail out */
 425         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 426                 return RX_HANDLER_PASS;
 427
 428         port = macvlan_port_get_rcu(skb->dev);
 429         if (is_multicast_ether_addr(eth->h_dest)) {
 430                 unsigned int hash;
 431
 432                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 433                 if (!skb)
 434                         return RX_HANDLER_CONSUMED;
 435                 *pskb = skb;
 436                 eth = eth_hdr(skb);
 437                 macvlan_forward_source(skb, port, eth->h_source);
 438                 src = macvlan_hash_lookup(port, eth->h_source);
 439                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 440                     src->mode != MACVLAN_MODE_BRIDGE) {
 441                         /* forward to original port. */
 442                         vlan = src;
 443                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 444                               netif_rx(skb);
 445                         handle_res = RX_HANDLER_CONSUMED;
 446                         goto out;
 447                 }
 448
 449                 hash = mc_hash(NULL, eth->h_dest);
 450                 if (test_bit(hash, port->mc_filter))
 451                         macvlan_broadcast_enqueue(port, src, skb);
 452
 453                 return RX_HANDLER_PASS;
 454         }
 455
 456         macvlan_forward_source(skb, port, eth->h_source);
 457         if (port->passthru)
 458                 vlan = list_first_or_null_rcu(&port->vlans,
 459                                               struct macvlan_dev, list);
 460         else
 461                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 462         if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
 463                 return RX_HANDLER_PASS;
 464
 465         dev = vlan->dev;
 466         if (unlikely(!(dev->flags & IFF_UP))) {
 467                 kfree_skb(skb);
 468                 return RX_HANDLER_CONSUMED;
 469         }
 470         len = skb->len + ETH_HLEN;
 471         skb = skb_share_check(skb, GFP_ATOMIC);
 472         if (!skb) {
 473                 ret = NET_RX_DROP;
 474                 handle_res = RX_HANDLER_CONSUMED;
 475                 goto out;
 476         }
 477
 478         *pskb = skb;
 479         skb->dev = dev;
 480         skb->pkt_type = PACKET_HOST;
 481
 482         ret = NET_RX_SUCCESS;
 483         handle_res = RX_HANDLER_ANOTHER;
 484 out:
 485         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 486         return handle_res;
 487 }
 488
 489 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 490 {
 491         const struct macvlan_dev *vlan = netdev_priv(dev);
 492         const struct macvlan_port *port = vlan->port;
 493         const struct macvlan_dev *dest;
 494
 495         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 496                 const struct ethhdr *eth = skb_eth_hdr(skb);
 497
 498                 /* send to other bridge ports directly */
 499                 if (is_multicast_ether_addr(eth->h_dest)) {
 500                         skb_reset_mac_header(skb);
 501                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 502                         goto xmit_world;
 503                 }
 504
 505                 dest = macvlan_hash_lookup(port, eth->h_dest);
 506                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 507                         /* send to lowerdev first for its network taps */
 508                         dev_forward_skb(vlan->lowerdev, skb);
 509
 510                         return NET_XMIT_SUCCESS;
 511                 }
 512         }
 513
 514 xmit_world:
 515         skb->dev = vlan->lowerdev;
 516         return dev_queue_xmit(skb);
 517 }
 518
 519 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 520 {
 521 #ifdef CONFIG_NET_POLL_CONTROLLER
 522         if (vlan->netpoll)
 523                 netpoll_send_skb(vlan->netpoll, skb);
 524 #else
 525         BUG();
 526 #endif
 527         return NETDEV_TX_OK;
 528 }
 529
 530 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 531                                       struct net_device *dev)
 532 {
 533         unsigned int len = skb->len;
 534         int ret;
 535         struct macvlan_dev *vlan = netdev_priv(dev);
 536
 537         if (unlikely(netpoll_tx_running(dev)))
 538                 return macvlan_netpoll_send_skb(vlan, skb);
 539
 540         if (vlan->fwd_priv) {
 541                 skb->dev = vlan->lowerdev;
 542                 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
 543         } else {
 544                 ret = macvlan_queue_xmit(skb, dev);
 545         }
 546
 547         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 548                 struct vlan_pcpu_stats *pcpu_stats;
 549
 550                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 551                 u64_stats_update_begin(&pcpu_stats->syncp);
 552                 pcpu_stats->tx_packets++;
 553                 pcpu_stats->tx_bytes += len;
 554                 u64_stats_update_end(&pcpu_stats->syncp);
 555         } else {
 556                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 557         }
 558         return ret;
 559 }
 560
 561 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 562                                unsigned short type, const void *daddr,
 563                                const void *saddr, unsigned len)
 564 {
 565         const struct macvlan_dev *vlan = netdev_priv(dev);
 566         struct net_device *lowerdev = vlan->lowerdev;
 567
 568         return dev_hard_header(skb, lowerdev, type, daddr,
 569                                saddr ? : dev->dev_addr, len);
 570 }
 571
 572 static const struct header_ops macvlan_hard_header_ops = {
 573         .create         = macvlan_hard_header,
 574         .parse          = eth_header_parse,
 575         .cache          = eth_header_cache,
 576         .cache_update   = eth_header_cache_update,
 577 };
 578
 579 static struct rtnl_link_ops macvlan_link_ops;
 580
 581 static int macvlan_open(struct net_device *dev)
 582 {
 583         struct macvlan_dev *vlan = netdev_priv(dev);
 584         struct net_device *lowerdev = vlan->lowerdev;
 585         int err;
 586
 587         if (vlan->port->passthru) {
 588                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 589                         err = dev_set_promiscuity(lowerdev, 1);
 590                         if (err < 0)
 591                                 goto out;
 592                 }
 593                 goto hash_add;
 594         }
 595
 596         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
 597             dev->rtnl_link_ops == &macvlan_link_ops) {
 598                 vlan->fwd_priv =
 599                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 600
 601                 /* If we get a NULL pointer back, or if we get an error
 602                  * then we should just fall through to the non accelerated path
 603                  */
 604                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
 605                         vlan->fwd_priv = NULL;
 606                 } else
 607                         return 0;
 608         }
 609
 610         err = -EBUSY;
 611         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 612                 goto out;
 613
 614         err = dev_uc_add(lowerdev, dev->dev_addr);
 615         if (err < 0)
 616                 goto out;
 617         if (dev->flags & IFF_ALLMULTI) {
 618                 err = dev_set_allmulti(lowerdev, 1);
 619                 if (err < 0)
 620                         goto del_unicast;
 621         }
 622
 623         if (dev->flags & IFF_PROMISC) {
 624                 err = dev_set_promiscuity(lowerdev, 1);
 625                 if (err < 0)
 626                         goto clear_multi;
 627         }
 628
 629 hash_add:
 630         macvlan_hash_add(vlan);
 631         return 0;
 632
 633 clear_multi:
 634         if (dev->flags & IFF_ALLMULTI)
 635                 dev_set_allmulti(lowerdev, -1);
 636 del_unicast:
 637         dev_uc_del(lowerdev, dev->dev_addr);
 638 out:
 639         if (vlan->fwd_priv) {
 640                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 641                                                            vlan->fwd_priv);
 642                 vlan->fwd_priv = NULL;
 643         }
 644         return err;
 645 }
 646
 647 static int macvlan_stop(struct net_device *dev)
 648 {
 649         struct macvlan_dev *vlan = netdev_priv(dev);
 650         struct net_device *lowerdev = vlan->lowerdev;
 651
 652         if (vlan->fwd_priv) {
 653                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 654                                                            vlan->fwd_priv);
 655                 vlan->fwd_priv = NULL;
 656                 return 0;
 657         }
 658
 659         dev_uc_unsync(lowerdev, dev);
 660         dev_mc_unsync(lowerdev, dev);
 661
 662         if (vlan->port->passthru) {
 663                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 664                         dev_set_promiscuity(lowerdev, -1);
 665                 goto hash_del;
 666         }
 667
 668         if (dev->flags & IFF_ALLMULTI)
 669                 dev_set_allmulti(lowerdev, -1);
 670
 671         if (dev->flags & IFF_PROMISC)
 672                 dev_set_promiscuity(lowerdev, -1);
 673
 674         dev_uc_del(lowerdev, dev->dev_addr);
 675
 676 hash_del:
 677         macvlan_hash_del(vlan, !dev->dismantle);
 678         return 0;
 679 }
 680
 681 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 682 {
 683         struct macvlan_dev *vlan = netdev_priv(dev);
 684         struct net_device *lowerdev = vlan->lowerdev;
 685         int err;
 686
 687         if (!(dev->flags & IFF_UP)) {
 688                 /* Just copy in the new address */
 689                 ether_addr_copy(dev->dev_addr, addr);
 690         } else {
 691                 /* Rehash and update the device filters */
 692                 if (macvlan_addr_busy(vlan->port, addr))
 693                         return -EBUSY;
 694
 695                 if (!vlan->port->passthru) {
 696                         err = dev_uc_add(lowerdev, addr);
 697                         if (err)
 698                                 return err;
 699
 700                         dev_uc_del(lowerdev, dev->dev_addr);
 701                 }
 702
 703                 macvlan_hash_change_addr(vlan, addr);
 704         }
 705         return 0;
 706 }
 707
 708 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 709 {
 710         struct macvlan_dev *vlan = netdev_priv(dev);
 711         struct sockaddr *addr = p;
 712
 713         if (!is_valid_ether_addr(addr->sa_data))
 714                 return -EADDRNOTAVAIL;
 715
 716         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 717                 dev_set_mac_address(vlan->lowerdev, addr);
 718                 return 0;
 719         }
 720
 721         return macvlan_sync_address(dev, addr->sa_data);
 722 }
 723
 724 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 725 {
 726         struct macvlan_dev *vlan = netdev_priv(dev);
 727         struct net_device *lowerdev = vlan->lowerdev;
 728
 729         if (dev->flags & IFF_UP) {
 730                 if (change & IFF_ALLMULTI)
 731                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 732                 if (change & IFF_PROMISC)
 733                         dev_set_promiscuity(lowerdev,
 734                                             dev->flags & IFF_PROMISC ? 1 : -1);
 735
 736         }
 737 }
 738
 739 static void macvlan_compute_filter(unsigned long *mc_filter,
 740                                    struct net_device *dev,
 741                                    struct macvlan_dev *vlan)
 742 {
 743         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 744                 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
 745         } else {
 746                 struct netdev_hw_addr *ha;
 747                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 748
 749                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 750                 netdev_for_each_mc_addr(ha, dev) {
 751                         __set_bit(mc_hash(vlan, ha->addr), filter);
 752                 }
 753
 754                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 755
 756                 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 757         }
 758 }
 759
 760 static void macvlan_set_mac_lists(struct net_device *dev)
 761 {
 762         struct macvlan_dev *vlan = netdev_priv(dev);
 763
 764         macvlan_compute_filter(vlan->mc_filter, dev, vlan);
 765
 766         dev_uc_sync(vlan->lowerdev, dev);
 767         dev_mc_sync(vlan->lowerdev, dev);
 768
 769         /* This is slightly inaccurate as we're including the subscription
 770          * list of vlan->lowerdev too.
 771          *
 772          * Bug alert: This only works if everyone has the same broadcast
 773          * address as lowerdev.  As soon as someone changes theirs this
 774          * will break.
 775          *
 776          * However, this is already broken as when you change your broadcast
 777          * address we don't get called.
 778          *
 779          * The solution is to maintain a list of broadcast addresses like
 780          * we do for uc/mc, if you care.
 781          */
 782         macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 783 }
 784
 785 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 786 {
 787         struct macvlan_dev *vlan = netdev_priv(dev);
 788
 789         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 790                 return -EINVAL;
 791         dev->mtu = new_mtu;
 792         return 0;
 793 }
 794
 795 /*
 796  * macvlan network devices have devices nesting below it and are a special
 797  * "super class" of normal network devices; split their locks off into a
 798  * separate class since they always nest.
 799  */
 800 static struct lock_class_key macvlan_netdev_addr_lock_key;
 801
 802 #define ALWAYS_ON_FEATURES \
 803         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX | \
 804          NETIF_F_GSO_ROBUST)
 805
 806 #define MACVLAN_FEATURES \
 807         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 808          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
 809          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 810          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 811
 812 #define MACVLAN_STATE_MASK \
 813         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 814
 815 static int macvlan_get_nest_level(struct net_device *dev)
 816 {
 817         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 818 }
 819
 820 static void macvlan_set_lockdep_class(struct net_device *dev)
 821 {
 822         netdev_lockdep_set_classes(dev);
 823         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 824                                        &macvlan_netdev_addr_lock_key,
 825                                        macvlan_get_nest_level(dev));
 826 }
 827
 828 static int macvlan_init(struct net_device *dev)
 829 {
 830         struct macvlan_dev *vlan = netdev_priv(dev);
 831         const struct net_device *lowerdev = vlan->lowerdev;
 832         struct macvlan_port *port = vlan->port;
 833
 834         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 835                                   (lowerdev->state & MACVLAN_STATE_MASK);
 836         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 837         dev->features           |= ALWAYS_ON_FEATURES;
 838         dev->hw_features        |= NETIF_F_LRO;
 839         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 840         dev->gso_max_size       = lowerdev->gso_max_size;
 841         dev->gso_max_segs       = lowerdev->gso_max_segs;
 842         dev->hard_header_len    = lowerdev->hard_header_len;
 843
 844         macvlan_set_lockdep_class(dev);
 845
 846         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 847         if (!vlan->pcpu_stats)
 848                 return -ENOMEM;
 849
 850         port->count += 1;
 851
 852         return 0;
 853 }
 854
 855 static void macvlan_uninit(struct net_device *dev)
 856 {
 857         struct macvlan_dev *vlan = netdev_priv(dev);
 858         struct macvlan_port *port = vlan->port;
 859
 860         free_percpu(vlan->pcpu_stats);
 861
 862         macvlan_flush_sources(port, vlan);
 863         port->count -= 1;
 864         if (!port->count)
 865                 macvlan_port_destroy(port->dev);
 866 }
 867
 868 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 869                                                          struct rtnl_link_stats64 *stats)
 870 {
 871         struct macvlan_dev *vlan = netdev_priv(dev);
 872
 873         if (vlan->pcpu_stats) {
 874                 struct vlan_pcpu_stats *p;
 875                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 876                 u32 rx_errors = 0, tx_dropped = 0;
 877                 unsigned int start;
 878                 int i;
 879
 880                 for_each_possible_cpu(i) {
 881                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 882                         do {
 883                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 884                                 rx_packets      = p->rx_packets;
 885                                 rx_bytes        = p->rx_bytes;
 886                                 rx_multicast    = p->rx_multicast;
 887                                 tx_packets      = p->tx_packets;
 888                                 tx_bytes        = p->tx_bytes;
 889                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 890
 891                         stats->rx_packets       += rx_packets;
 892                         stats->rx_bytes         += rx_bytes;
 893                         stats->multicast        += rx_multicast;
 894                         stats->tx_packets       += tx_packets;
 895                         stats->tx_bytes         += tx_bytes;
 896                         /* rx_errors & tx_dropped are u32, updated
 897                          * without syncp protection.
 898                          */
 899                         rx_errors       += p->rx_errors;
 900                         tx_dropped      += p->tx_dropped;
 901                 }
 902                 stats->rx_errors        = rx_errors;
 903                 stats->rx_dropped       = rx_errors;
 904                 stats->tx_dropped       = tx_dropped;
 905         }
 906         return stats;
 907 }
 908
 909 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 910                                    __be16 proto, u16 vid)
 911 {
 912         struct macvlan_dev *vlan = netdev_priv(dev);
 913         struct net_device *lowerdev = vlan->lowerdev;
 914
 915         return vlan_vid_add(lowerdev, proto, vid);
 916 }
 917
 918 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 919                                     __be16 proto, u16 vid)
 920 {
 921         struct macvlan_dev *vlan = netdev_priv(dev);
 922         struct net_device *lowerdev = vlan->lowerdev;
 923
 924         vlan_vid_del(lowerdev, proto, vid);
 925         return 0;
 926 }
 927
 928 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 929                            struct net_device *dev,
 930                            const unsigned char *addr, u16 vid,
 931                            u16 flags)
 932 {
 933         struct macvlan_dev *vlan = netdev_priv(dev);
 934         int err = -EINVAL;
 935
 936         /* Support unicast filter only on passthru devices.
 937          * Multicast filter should be allowed on all devices.
 938          */
 939         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 940                 return -EOPNOTSUPP;
 941
 942         if (flags & NLM_F_REPLACE)
 943                 return -EOPNOTSUPP;
 944
 945         if (is_unicast_ether_addr(addr))
 946                 err = dev_uc_add_excl(dev, addr);
 947         else if (is_multicast_ether_addr(addr))
 948                 err = dev_mc_add_excl(dev, addr);
 949
 950         return err;
 951 }
 952
 953 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 954                            struct net_device *dev,
 955                            const unsigned char *addr, u16 vid)
 956 {
 957         struct macvlan_dev *vlan = netdev_priv(dev);
 958         int err = -EINVAL;
 959
 960         /* Support unicast filter only on passthru devices.
 961          * Multicast filter should be allowed on all devices.
 962          */
 963         if (!vlan->port->passthru && is_unicast_ether_addr(addr))
 964                 return -EOPNOTSUPP;
 965
 966         if (is_unicast_ether_addr(addr))
 967                 err = dev_uc_del(dev, addr);
 968         else if (is_multicast_ether_addr(addr))
 969                 err = dev_mc_del(dev, addr);
 970
 971         return err;
 972 }
 973
 974 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 975                                         struct ethtool_drvinfo *drvinfo)
 976 {
 977         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 978         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 979 }
 980
 981 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
 982                                               struct ethtool_link_ksettings *cmd)
 983 {
 984         const struct macvlan_dev *vlan = netdev_priv(dev);
 985
 986         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
 987 }
 988
 989 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 990                                               netdev_features_t features)
 991 {
 992         struct macvlan_dev *vlan = netdev_priv(dev);
 993         netdev_features_t lowerdev_features = vlan->lowerdev->features;
 994         netdev_features_t mask;
 995
 996         features |= NETIF_F_ALL_FOR_ALL;
 997         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 998         mask = features;
 999
1000         lowerdev_features &= (features | ~NETIF_F_LRO);
1001         features = netdev_increment_features(lowerdev_features, features, mask);
1002         features |= ALWAYS_ON_FEATURES;
1003         features &= ~NETIF_F_NETNS_LOCAL;
1004
1005         return features;
1006 }
1007
1008 #ifdef CONFIG_NET_POLL_CONTROLLER
1009 static void macvlan_dev_poll_controller(struct net_device *dev)
1010 {
1011         return;
1012 }
1013
1014 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1015 {
1016         struct macvlan_dev *vlan = netdev_priv(dev);
1017         struct net_device *real_dev = vlan->lowerdev;
1018         struct netpoll *netpoll;
1019         int err = 0;
1020
1021         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1022         err = -ENOMEM;
1023         if (!netpoll)
1024                 goto out;
1025
1026         err = __netpoll_setup(netpoll, real_dev);
1027         if (err) {
1028                 kfree(netpoll);
1029                 goto out;
1030         }
1031
1032         vlan->netpoll = netpoll;
1033
1034 out:
1035         return err;
1036 }
1037
1038 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1039 {
1040         struct macvlan_dev *vlan = netdev_priv(dev);
1041         struct netpoll *netpoll = vlan->netpoll;
1042
1043         if (!netpoll)
1044                 return;
1045
1046         vlan->netpoll = NULL;
1047
1048         __netpoll_free_async(netpoll);
1049 }
1050 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1051
1052 static int macvlan_dev_get_iflink(const struct net_device *dev)
1053 {
1054         struct macvlan_dev *vlan = netdev_priv(dev);
1055
1056         return vlan->lowerdev->ifindex;
1057 }
1058
1059 static const struct ethtool_ops macvlan_ethtool_ops = {
1060         .get_link               = ethtool_op_get_link,
1061         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1062         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1063 };
1064
1065 static const struct net_device_ops macvlan_netdev_ops = {
1066         .ndo_init               = macvlan_init,
1067         .ndo_uninit             = macvlan_uninit,
1068         .ndo_open               = macvlan_open,
1069         .ndo_stop               = macvlan_stop,
1070         .ndo_start_xmit         = macvlan_start_xmit,
1071         .ndo_change_mtu         = macvlan_change_mtu,
1072         .ndo_fix_features       = macvlan_fix_features,
1073         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1074         .ndo_set_mac_address    = macvlan_set_mac_address,
1075         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1076         .ndo_get_stats64        = macvlan_dev_get_stats64,
1077         .ndo_validate_addr      = eth_validate_addr,
1078         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1079         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1080         .ndo_fdb_add            = macvlan_fdb_add,
1081         .ndo_fdb_del            = macvlan_fdb_del,
1082         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1083         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1084 #ifdef CONFIG_NET_POLL_CONTROLLER
1085         .ndo_poll_controller    = macvlan_dev_poll_controller,
1086         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1087         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1088 #endif
1089         .ndo_get_iflink         = macvlan_dev_get_iflink,
1090         .ndo_features_check     = passthru_features_check,
1091 };
1092
1093 void macvlan_common_setup(struct net_device *dev)
1094 {
1095         ether_setup(dev);
1096
1097         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1098         netif_keep_dst(dev);
1099         dev->priv_flags        |= IFF_UNICAST_FLT;
1100         dev->netdev_ops         = &macvlan_netdev_ops;
1101         dev->destructor         = free_netdev;
1102         dev->header_ops         = &macvlan_hard_header_ops;
1103         dev->ethtool_ops        = &macvlan_ethtool_ops;
1104 }
1105 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1106
1107 static void macvlan_setup(struct net_device *dev)
1108 {
1109         macvlan_common_setup(dev);
1110         dev->priv_flags |= IFF_NO_QUEUE;
1111 }
1112
1113 static int macvlan_port_create(struct net_device *dev)
1114 {
1115         struct macvlan_port *port;
1116         unsigned int i;
1117         int err;
1118
1119         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1120                 return -EINVAL;
1121
1122         if (netif_is_ipvlan_port(dev))
1123                 return -EBUSY;
1124
1125         port = kzalloc(sizeof(*port), GFP_KERNEL);
1126         if (port == NULL)
1127                 return -ENOMEM;
1128
1129         port->passthru = false;
1130         port->dev = dev;
1131         INIT_LIST_HEAD(&port->vlans);
1132         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1133                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1134         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1135                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1136
1137         skb_queue_head_init(&port->bc_queue);
1138         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1139
1140         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1141         if (err)
1142                 kfree(port);
1143         else
1144                 dev->priv_flags |= IFF_MACVLAN_PORT;
1145         return err;
1146 }
1147
1148 static void macvlan_port_destroy(struct net_device *dev)
1149 {
1150         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1151         struct sk_buff *skb;
1152
1153         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1154         netdev_rx_handler_unregister(dev);
1155
1156         /* After this point, no packet can schedule bc_work anymore,
1157          * but we need to cancel it and purge left skbs if any.
1158          */
1159         cancel_work_sync(&port->bc_work);
1160
1161         while ((skb = __skb_dequeue(&port->bc_queue))) {
1162                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1163
1164                 if (src)
1165                         dev_put(src->dev);
1166
1167                 kfree_skb(skb);
1168         }
1169
1170         kfree_rcu(port, rcu);
1171 }
1172
1173 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
1174 {
1175         if (tb[IFLA_ADDRESS]) {
1176                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1177                         return -EINVAL;
1178                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1179                         return -EADDRNOTAVAIL;
1180         }
1181
1182         if (data && data[IFLA_MACVLAN_FLAGS] &&
1183             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1184                 return -EINVAL;
1185
1186         if (data && data[IFLA_MACVLAN_MODE]) {
1187                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1188                 case MACVLAN_MODE_PRIVATE:
1189                 case MACVLAN_MODE_VEPA:
1190                 case MACVLAN_MODE_BRIDGE:
1191                 case MACVLAN_MODE_PASSTHRU:
1192                 case MACVLAN_MODE_SOURCE:
1193                         break;
1194                 default:
1195                         return -EINVAL;
1196                 }
1197         }
1198
1199         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1200                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1201                 case MACVLAN_MACADDR_ADD:
1202                 case MACVLAN_MACADDR_DEL:
1203                 case MACVLAN_MACADDR_FLUSH:
1204                 case MACVLAN_MACADDR_SET:
1205                         break;
1206                 default:
1207                         return -EINVAL;
1208                 }
1209         }
1210
1211         if (data && data[IFLA_MACVLAN_MACADDR]) {
1212                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1213                         return -EINVAL;
1214
1215                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1216                         return -EADDRNOTAVAIL;
1217         }
1218
1219         if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1220                 return -EINVAL;
1221
1222         return 0;
1223 }
1224
1225 /**
1226  * reconfigure list of remote source mac address
1227  * (only for macvlan devices in source mode)
1228  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1229  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1230  */
1231 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1232                                       struct nlattr *data[])
1233 {
1234         char *addr = NULL;
1235         int ret, rem, len;
1236         struct nlattr *nla, *head;
1237         struct macvlan_source_entry *entry;
1238
1239         if (data[IFLA_MACVLAN_MACADDR])
1240                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1241
1242         if (mode == MACVLAN_MACADDR_ADD) {
1243                 if (!addr)
1244                         return -EINVAL;
1245
1246                 return macvlan_hash_add_source(vlan, addr);
1247
1248         } else if (mode == MACVLAN_MACADDR_DEL) {
1249                 if (!addr)
1250                         return -EINVAL;
1251
1252                 entry = macvlan_hash_lookup_source(vlan, addr);
1253                 if (entry) {
1254                         macvlan_hash_del_source(entry);
1255                         vlan->macaddr_count--;
1256                 }
1257         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1258                 macvlan_flush_sources(vlan->port, vlan);
1259         } else if (mode == MACVLAN_MACADDR_SET) {
1260                 macvlan_flush_sources(vlan->port, vlan);
1261
1262                 if (addr) {
1263                         ret = macvlan_hash_add_source(vlan, addr);
1264                         if (ret)
1265                                 return ret;
1266                 }
1267
1268                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1269                         return 0;
1270
1271                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1272                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1273
1274                 nla_for_each_attr(nla, head, len, rem) {
1275                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1276                             nla_len(nla) != ETH_ALEN)
1277                                 continue;
1278
1279                         addr = nla_data(nla);
1280                         ret = macvlan_hash_add_source(vlan, addr);
1281                         if (ret)
1282                                 return ret;
1283                 }
1284         } else {
1285                 return -EINVAL;
1286         }
1287
1288         return 0;
1289 }
1290
1291 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1292                            struct nlattr *tb[], struct nlattr *data[])
1293 {
1294         struct macvlan_dev *vlan = netdev_priv(dev);
1295         struct macvlan_port *port;
1296         struct net_device *lowerdev;
1297         int err;
1298         int macmode;
1299         bool create = false;
1300
1301         if (!tb[IFLA_LINK])
1302                 return -EINVAL;
1303
1304         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1305         if (lowerdev == NULL)
1306                 return -ENODEV;
1307
1308         /* When creating macvlans or macvtaps on top of other macvlans - use
1309          * the real device as the lowerdev.
1310          */
1311         if (netif_is_macvlan(lowerdev))
1312                 lowerdev = macvlan_dev_real_dev(lowerdev);
1313
1314         if (!tb[IFLA_MTU])
1315                 dev->mtu = lowerdev->mtu;
1316         else if (dev->mtu > lowerdev->mtu)
1317                 return -EINVAL;
1318
1319         if (!tb[IFLA_ADDRESS])
1320                 eth_hw_addr_random(dev);
1321
1322         if (!macvlan_port_exists(lowerdev)) {
1323                 err = macvlan_port_create(lowerdev);
1324                 if (err < 0)
1325                         return err;
1326                 create = true;
1327         }
1328         port = macvlan_port_get_rtnl(lowerdev);
1329
1330         /* Only 1 macvlan device can be created in passthru mode */
1331         if (port->passthru) {
1332                 /* The macvlan port must be not created this time,
1333                  * still goto destroy_macvlan_port for readability.
1334                  */
1335                 err = -EINVAL;
1336                 goto destroy_macvlan_port;
1337         }
1338
1339         vlan->lowerdev = lowerdev;
1340         vlan->dev      = dev;
1341         vlan->port     = port;
1342         vlan->set_features = MACVLAN_FEATURES;
1343         vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1344
1345         vlan->mode     = MACVLAN_MODE_VEPA;
1346         if (data && data[IFLA_MACVLAN_MODE])
1347                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1348
1349         if (data && data[IFLA_MACVLAN_FLAGS])
1350                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1351
1352         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1353                 if (port->count) {
1354                         err = -EINVAL;
1355                         goto destroy_macvlan_port;
1356                 }
1357                 port->passthru = true;
1358                 eth_hw_addr_inherit(dev, lowerdev);
1359         }
1360
1361         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1362                 if (vlan->mode != MACVLAN_MODE_SOURCE) {
1363                         err = -EINVAL;
1364                         goto destroy_macvlan_port;
1365                 }
1366                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1367                 err = macvlan_changelink_sources(vlan, macmode, data);
1368                 if (err)
1369                         goto destroy_macvlan_port;
1370         }
1371
1372         err = register_netdevice(dev);
1373         if (err < 0)
1374                 goto destroy_macvlan_port;
1375
1376         dev->priv_flags |= IFF_MACVLAN;
1377         err = netdev_upper_dev_link(lowerdev, dev);
1378         if (err)
1379                 goto unregister_netdev;
1380
1381         list_add_tail_rcu(&vlan->list, &port->vlans);
1382         netif_stacked_transfer_operstate(lowerdev, dev);
1383         linkwatch_fire_event(dev);
1384
1385         return 0;
1386
1387 unregister_netdev:
1388         /* macvlan_uninit would free the macvlan port */
1389         unregister_netdevice(dev);
1390         return err;
1391 destroy_macvlan_port:
1392         /* the macvlan port may be freed by macvlan_uninit when fail to register.
1393          * so we destroy the macvlan port only when it's valid.
1394          */
1395         if (create && macvlan_port_get_rtnl(lowerdev)) {
1396                 macvlan_flush_sources(port, vlan);
1397                 macvlan_port_destroy(port->dev);
1398         }
1399         return err;
1400 }
1401 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1402
1403 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1404                            struct nlattr *tb[], struct nlattr *data[])
1405 {
1406         return macvlan_common_newlink(src_net, dev, tb, data);
1407 }
1408
1409 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1410 {
1411         struct macvlan_dev *vlan = netdev_priv(dev);
1412
1413         if (vlan->mode == MACVLAN_MODE_SOURCE)
1414                 macvlan_flush_sources(vlan->port, vlan);
1415         list_del_rcu(&vlan->list);
1416         unregister_netdevice_queue(dev, head);
1417         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1418 }
1419 EXPORT_SYMBOL_GPL(macvlan_dellink);
1420
1421 static int macvlan_changelink(struct net_device *dev,
1422                 struct nlattr *tb[], struct nlattr *data[])
1423 {
1424         struct macvlan_dev *vlan = netdev_priv(dev);
1425         enum macvlan_mode mode;
1426         bool set_mode = false;
1427         enum macvlan_macaddr_mode macmode;
1428         int ret;
1429
1430         /* Validate mode, but don't set yet: setting flags may fail. */
1431         if (data && data[IFLA_MACVLAN_MODE]) {
1432                 set_mode = true;
1433                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1434                 /* Passthrough mode can't be set or cleared dynamically */
1435                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1436                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1437                         return -EINVAL;
1438                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1439                     vlan->mode != mode)
1440                         macvlan_flush_sources(vlan->port, vlan);
1441         }
1442
1443         if (data && data[IFLA_MACVLAN_FLAGS]) {
1444                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1445                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1446                 if (vlan->port->passthru && promisc) {
1447                         int err;
1448
1449                         if (flags & MACVLAN_FLAG_NOPROMISC)
1450                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1451                         else
1452                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1453                         if (err < 0)
1454                                 return err;
1455                 }
1456                 vlan->flags = flags;
1457         }
1458         if (set_mode)
1459                 vlan->mode = mode;
1460         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1461                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1462                         return -EINVAL;
1463                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1464                 ret = macvlan_changelink_sources(vlan, macmode, data);
1465                 if (ret)
1466                         return ret;
1467         }
1468         return 0;
1469 }
1470
1471 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1472 {
1473         if (vlan->macaddr_count == 0)
1474                 return 0;
1475         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1476                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1477 }
1478
1479 static size_t macvlan_get_size(const struct net_device *dev)
1480 {
1481         struct macvlan_dev *vlan = netdev_priv(dev);
1482
1483         return (0
1484                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1485                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1486                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1487                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1488                 );
1489 }
1490
1491 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1492                                      const struct macvlan_dev *vlan,
1493                                      const int i)
1494 {
1495         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1496         struct macvlan_source_entry *entry;
1497
1498         hlist_for_each_entry_rcu(entry, h, hlist) {
1499                 if (entry->vlan != vlan)
1500                         continue;
1501                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1502                         return 1;
1503         }
1504         return 0;
1505 }
1506
1507 static int macvlan_fill_info(struct sk_buff *skb,
1508                                 const struct net_device *dev)
1509 {
1510         struct macvlan_dev *vlan = netdev_priv(dev);
1511         int i;
1512         struct nlattr *nest;
1513
1514         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1515                 goto nla_put_failure;
1516         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1517                 goto nla_put_failure;
1518         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1519                 goto nla_put_failure;
1520         if (vlan->macaddr_count > 0) {
1521                 nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1522                 if (nest == NULL)
1523                         goto nla_put_failure;
1524
1525                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1526                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1527                                 goto nla_put_failure;
1528                 }
1529                 nla_nest_end(skb, nest);
1530         }
1531         return 0;
1532
1533 nla_put_failure:
1534         return -EMSGSIZE;
1535 }
1536
1537 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1538         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1539         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1540         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1541         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1542         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1543         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1544 };
1545
1546 int macvlan_link_register(struct rtnl_link_ops *ops)
1547 {
1548         /* common fields */
1549         ops->priv_size          = sizeof(struct macvlan_dev);
1550         ops->validate           = macvlan_validate;
1551         ops->maxtype            = IFLA_MACVLAN_MAX;
1552         ops->policy             = macvlan_policy;
1553         ops->changelink         = macvlan_changelink;
1554         ops->get_size           = macvlan_get_size;
1555         ops->fill_info          = macvlan_fill_info;
1556
1557         return rtnl_link_register(ops);
1558 };
1559 EXPORT_SYMBOL_GPL(macvlan_link_register);
1560
1561 static struct net *macvlan_get_link_net(const struct net_device *dev)
1562 {
1563         return dev_net(macvlan_dev_real_dev(dev));
1564 }
1565
1566 static struct rtnl_link_ops macvlan_link_ops = {
1567         .kind           = "macvlan",
1568         .setup          = macvlan_setup,
1569         .newlink        = macvlan_newlink,
1570         .dellink        = macvlan_dellink,
1571         .get_link_net   = macvlan_get_link_net,
1572 };
1573
1574 static int macvlan_device_event(struct notifier_block *unused,
1575                                 unsigned long event, void *ptr)
1576 {
1577         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1578         struct macvlan_dev *vlan, *next;
1579         struct macvlan_port *port;
1580         LIST_HEAD(list_kill);
1581
1582         if (!macvlan_port_exists(dev))
1583                 return NOTIFY_DONE;
1584
1585         port = macvlan_port_get_rtnl(dev);
1586
1587         switch (event) {
1588         case NETDEV_UP:
1589         case NETDEV_CHANGE:
1590                 list_for_each_entry(vlan, &port->vlans, list)
1591                         netif_stacked_transfer_operstate(vlan->lowerdev,
1592                                                          vlan->dev);
1593                 break;
1594         case NETDEV_FEAT_CHANGE:
1595                 list_for_each_entry(vlan, &port->vlans, list) {
1596                         vlan->dev->gso_max_size = dev->gso_max_size;
1597                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1598                         netdev_update_features(vlan->dev);
1599                 }
1600                 break;
1601         case NETDEV_CHANGEMTU:
1602                 list_for_each_entry(vlan, &port->vlans, list) {
1603                         if (vlan->dev->mtu <= dev->mtu)
1604                                 continue;
1605                         dev_set_mtu(vlan->dev, dev->mtu);
1606                 }
1607                 break;
1608         case NETDEV_CHANGEADDR:
1609                 if (!port->passthru)
1610                         return NOTIFY_DONE;
1611
1612                 vlan = list_first_entry_or_null(&port->vlans,
1613                                                 struct macvlan_dev,
1614                                                 list);
1615
1616                 if (vlan && macvlan_sync_address(vlan->dev, dev->dev_addr))
1617                         return NOTIFY_BAD;
1618
1619                 break;
1620         case NETDEV_UNREGISTER:
1621                 /* twiddle thumbs on netns device moves */
1622                 if (dev->reg_state != NETREG_UNREGISTERING)
1623                         break;
1624
1625                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1626                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1627                 unregister_netdevice_many(&list_kill);
1628                 break;
1629         case NETDEV_PRE_TYPE_CHANGE:
1630                 /* Forbid underlaying device to change its type. */
1631                 return NOTIFY_BAD;
1632
1633         case NETDEV_NOTIFY_PEERS:
1634         case NETDEV_BONDING_FAILOVER:
1635         case NETDEV_RESEND_IGMP:
1636                 /* Propagate to all vlans */
1637                 list_for_each_entry(vlan, &port->vlans, list)
1638                         call_netdevice_notifiers(event, vlan->dev);
1639         }
1640         return NOTIFY_DONE;
1641 }
1642
1643 static struct notifier_block macvlan_notifier_block __read_mostly = {
1644         .notifier_call  = macvlan_device_event,
1645 };
1646
1647 static int __init macvlan_init_module(void)
1648 {
1649         int err;
1650
1651         register_netdevice_notifier(&macvlan_notifier_block);
1652
1653         err = macvlan_link_register(&macvlan_link_ops);
1654         if (err < 0)
1655                 goto err1;
1656         return 0;
1657 err1:
1658         unregister_netdevice_notifier(&macvlan_notifier_block);
1659         return err;
1660 }
1661
1662 static void __exit macvlan_cleanup_module(void)
1663 {
1664         rtnl_link_unregister(&macvlan_link_ops);
1665         unregister_netdevice_notifier(&macvlan_notifier_block);
1666 }
1667
1668 module_init(macvlan_init_module);
1669 module_exit(macvlan_cleanup_module);
1670
1671 MODULE_LICENSE("GPL");
1672 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1673 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1674 MODULE_ALIAS_RTNL_LINK("macvlan");