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 #define MACVLAN_F_PASSTHRU      1
  43 #define MACVLAN_F_ADDRCHANGE    2
  44
  45 struct macvlan_port {
  46         struct net_device       *dev;
  47         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  48         struct list_head        vlans;
  49         struct sk_buff_head     bc_queue;
  50         struct work_struct      bc_work;
  51         u32                     flags;
  52         int                     count;
  53         struct hlist_head       vlan_source_hash[MACVLAN_HASH_SIZE];
  54         DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
  55         unsigned char           perm_addr[ETH_ALEN];
  56 };
  57
  58 struct macvlan_source_entry {
  59         struct hlist_node       hlist;
  60         struct macvlan_dev      *vlan;
  61         unsigned char           addr[6+2] __aligned(sizeof(u16));
  62         struct rcu_head         rcu;
  63 };
  64
  65 struct macvlan_skb_cb {
  66         const struct macvlan_dev *src;
  67 };
  68
  69 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
  70
  71 static void macvlan_port_destroy(struct net_device *dev);
  72
  73 static inline bool macvlan_passthru(const struct macvlan_port *port)
  74 {
  75         return port->flags & MACVLAN_F_PASSTHRU;
  76 }
  77
  78 static inline void macvlan_set_passthru(struct macvlan_port *port)
  79 {
  80         port->flags |= MACVLAN_F_PASSTHRU;
  81 }
  82
  83 static inline bool macvlan_addr_change(const struct macvlan_port *port)
  84 {
  85         return port->flags & MACVLAN_F_ADDRCHANGE;
  86 }
  87
  88 static inline void macvlan_set_addr_change(struct macvlan_port *port)
  89 {
  90         port->flags |= MACVLAN_F_ADDRCHANGE;
  91 }
  92
  93 static inline void macvlan_clear_addr_change(struct macvlan_port *port)
  94 {
  95         port->flags &= ~MACVLAN_F_ADDRCHANGE;
  96 }
  97
  98 /* Hash Ethernet address */
  99 static u32 macvlan_eth_hash(const unsigned char *addr)
 100 {
 101         u64 value = get_unaligned((u64 *)addr);
 102
 103         /* only want 6 bytes */
 104 #ifdef __BIG_ENDIAN
 105         value >>= 16;
 106 #else
 107         value <<= 16;
 108 #endif
 109         return hash_64(value, MACVLAN_HASH_BITS);
 110 }
 111
 112 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
 113 {
 114         return rcu_dereference(dev->rx_handler_data);
 115 }
 116
 117 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
 118 {
 119         return rtnl_dereference(dev->rx_handler_data);
 120 }
 121
 122 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
 123
 124 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
 125                                                const unsigned char *addr)
 126 {
 127         struct macvlan_dev *vlan;
 128         u32 idx = macvlan_eth_hash(addr);
 129
 130         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
 131                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
 132                         return vlan;
 133         }
 134         return NULL;
 135 }
 136
 137 static struct macvlan_source_entry *macvlan_hash_lookup_source(
 138         const struct macvlan_dev *vlan,
 139         const unsigned char *addr)
 140 {
 141         struct macvlan_source_entry *entry;
 142         u32 idx = macvlan_eth_hash(addr);
 143         struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
 144
 145         hlist_for_each_entry_rcu(entry, h, hlist) {
 146                 if (ether_addr_equal_64bits(entry->addr, addr) &&
 147                     entry->vlan == vlan)
 148                         return entry;
 149         }
 150         return NULL;
 151 }
 152
 153 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
 154                                    const unsigned char *addr)
 155 {
 156         struct macvlan_port *port = vlan->port;
 157         struct macvlan_source_entry *entry;
 158         struct hlist_head *h;
 159
 160         entry = macvlan_hash_lookup_source(vlan, addr);
 161         if (entry)
 162                 return 0;
 163
 164         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 165         if (!entry)
 166                 return -ENOMEM;
 167
 168         ether_addr_copy(entry->addr, addr);
 169         entry->vlan = vlan;
 170         h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
 171         hlist_add_head_rcu(&entry->hlist, h);
 172         vlan->macaddr_count++;
 173
 174         return 0;
 175 }
 176
 177 static void macvlan_hash_add(struct macvlan_dev *vlan)
 178 {
 179         struct macvlan_port *port = vlan->port;
 180         const unsigned char *addr = vlan->dev->dev_addr;
 181         u32 idx = macvlan_eth_hash(addr);
 182
 183         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
 184 }
 185
 186 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
 187 {
 188         hlist_del_rcu(&entry->hlist);
 189         kfree_rcu(entry, rcu);
 190 }
 191
 192 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
 193 {
 194         hlist_del_rcu(&vlan->hlist);
 195         if (sync)
 196                 synchronize_rcu();
 197 }
 198
 199 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
 200                                         const unsigned char *addr)
 201 {
 202         macvlan_hash_del(vlan, true);
 203         /* Now that we are unhashed it is safe to change the device
 204          * address without confusing packet delivery.
 205          */
 206         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
 207         macvlan_hash_add(vlan);
 208 }
 209
 210 static bool macvlan_addr_busy(const struct macvlan_port *port,
 211                               const unsigned char *addr)
 212 {
 213         /* Test to see if the specified address is
 214          * currently in use by the underlying device or
 215          * another macvlan.
 216          */
 217         if (!macvlan_passthru(port) && !macvlan_addr_change(port) &&
 218             ether_addr_equal_64bits(port->dev->dev_addr, addr))
 219                 return true;
 220
 221         if (macvlan_hash_lookup(port, addr))
 222                 return true;
 223
 224         return false;
 225 }
 226
 227
 228 static int macvlan_broadcast_one(struct sk_buff *skb,
 229                                  const struct macvlan_dev *vlan,
 230                                  const struct ethhdr *eth, bool local)
 231 {
 232         struct net_device *dev = vlan->dev;
 233
 234         if (local)
 235                 return __dev_forward_skb(dev, skb);
 236
 237         skb->dev = dev;
 238         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 239                 skb->pkt_type = PACKET_BROADCAST;
 240         else
 241                 skb->pkt_type = PACKET_MULTICAST;
 242
 243         return 0;
 244 }
 245
 246 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 247 {
 248         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 249 }
 250
 251
 252 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 253                             const unsigned char *addr)
 254 {
 255         u32 val = __get_unaligned_cpu32(addr + 2);
 256
 257         val ^= macvlan_hash_mix(vlan);
 258         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 259 }
 260
 261 static void macvlan_broadcast(struct sk_buff *skb,
 262                               const struct macvlan_port *port,
 263                               struct net_device *src,
 264                               enum macvlan_mode mode)
 265 {
 266         const struct ethhdr *eth = eth_hdr(skb);
 267         const struct macvlan_dev *vlan;
 268         struct sk_buff *nskb;
 269         unsigned int i;
 270         int err;
 271         unsigned int hash;
 272
 273         if (skb->protocol == htons(ETH_P_PAUSE))
 274                 return;
 275
 276         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 277                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 278                         if (vlan->dev == src || !(vlan->mode & mode))
 279                                 continue;
 280
 281                         hash = mc_hash(vlan, eth->h_dest);
 282                         if (!test_bit(hash, vlan->mc_filter))
 283                                 continue;
 284
 285                         err = NET_RX_DROP;
 286                         nskb = skb_clone(skb, GFP_ATOMIC);
 287                         if (likely(nskb))
 288                                 err = macvlan_broadcast_one(
 289                                         nskb, vlan, eth,
 290                                         mode == MACVLAN_MODE_BRIDGE) ?:
 291                                       netif_rx_ni(nskb);
 292                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 293                                          err == NET_RX_SUCCESS, true);
 294                 }
 295         }
 296 }
 297
 298 static void macvlan_process_broadcast(struct work_struct *w)
 299 {
 300         struct macvlan_port *port = container_of(w, struct macvlan_port,
 301                                                  bc_work);
 302         struct sk_buff *skb;
 303         struct sk_buff_head list;
 304
 305         __skb_queue_head_init(&list);
 306
 307         spin_lock_bh(&port->bc_queue.lock);
 308         skb_queue_splice_tail_init(&port->bc_queue, &list);
 309         spin_unlock_bh(&port->bc_queue.lock);
 310
 311         while ((skb = __skb_dequeue(&list))) {
 312                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
 313
 314                 rcu_read_lock();
 315
 316                 if (!src)
 317                         /* frame comes from an external address */
 318                         macvlan_broadcast(skb, port, NULL,
 319                                           MACVLAN_MODE_PRIVATE |
 320                                           MACVLAN_MODE_VEPA    |
 321                                           MACVLAN_MODE_PASSTHRU|
 322                                           MACVLAN_MODE_BRIDGE);
 323                 else if (src->mode == MACVLAN_MODE_VEPA)
 324                         /* flood to everyone except source */
 325                         macvlan_broadcast(skb, port, src->dev,
 326                                           MACVLAN_MODE_VEPA |
 327                                           MACVLAN_MODE_BRIDGE);
 328                 else
 329                         /*
 330                          * flood only to VEPA ports, bridge ports
 331                          * already saw the frame on the way out.
 332                          */
 333                         macvlan_broadcast(skb, port, src->dev,
 334                                           MACVLAN_MODE_VEPA);
 335
 336                 rcu_read_unlock();
 337
 338                 if (src)
 339                         dev_put(src->dev);
 340                 kfree_skb(skb);
 341         }
 342 }
 343
 344 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 345                                       const struct macvlan_dev *src,
 346                                       struct sk_buff *skb)
 347 {
 348         struct sk_buff *nskb;
 349         int err = -ENOMEM;
 350
 351         nskb = skb_clone(skb, GFP_ATOMIC);
 352         if (!nskb)
 353                 goto err;
 354
 355         MACVLAN_SKB_CB(nskb)->src = src;
 356
 357         spin_lock(&port->bc_queue.lock);
 358         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 359                 if (src)
 360                         dev_hold(src->dev);
 361                 __skb_queue_tail(&port->bc_queue, nskb);
 362                 err = 0;
 363         }
 364         spin_unlock(&port->bc_queue.lock);
 365
 366         if (err)
 367                 goto free_nskb;
 368
 369         schedule_work(&port->bc_work);
 370         return;
 371
 372 free_nskb:
 373         kfree_skb(nskb);
 374 err:
 375         atomic_long_inc(&skb->dev->rx_dropped);
 376 }
 377
 378 static void macvlan_flush_sources(struct macvlan_port *port,
 379                                   struct macvlan_dev *vlan)
 380 {
 381         int i;
 382
 383         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 384                 struct hlist_node *h, *n;
 385
 386                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 387                         struct macvlan_source_entry *entry;
 388
 389                         entry = hlist_entry(h, struct macvlan_source_entry,
 390                                             hlist);
 391                         if (entry->vlan == vlan)
 392                                 macvlan_hash_del_source(entry);
 393                 }
 394         }
 395         vlan->macaddr_count = 0;
 396 }
 397
 398 static void macvlan_forward_source_one(struct sk_buff *skb,
 399                                        struct macvlan_dev *vlan)
 400 {
 401         struct sk_buff *nskb;
 402         struct net_device *dev;
 403         int len;
 404         int ret;
 405
 406         dev = vlan->dev;
 407         if (unlikely(!(dev->flags & IFF_UP)))
 408                 return;
 409
 410         nskb = skb_clone(skb, GFP_ATOMIC);
 411         if (!nskb)
 412                 return;
 413
 414         len = nskb->len + ETH_HLEN;
 415         nskb->dev = dev;
 416         nskb->pkt_type = PACKET_HOST;
 417
 418         ret = netif_rx(nskb);
 419         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 420 }
 421
 422 static void macvlan_forward_source(struct sk_buff *skb,
 423                                    struct macvlan_port *port,
 424                                    const unsigned char *addr)
 425 {
 426         struct macvlan_source_entry *entry;
 427         u32 idx = macvlan_eth_hash(addr);
 428         struct hlist_head *h = &port->vlan_source_hash[idx];
 429
 430         hlist_for_each_entry_rcu(entry, h, hlist) {
 431                 if (ether_addr_equal_64bits(entry->addr, addr))
 432                         macvlan_forward_source_one(skb, entry->vlan);
 433         }
 434 }
 435
 436 /* called under rcu_read_lock() from netif_receive_skb */
 437 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 438 {
 439         struct macvlan_port *port;
 440         struct sk_buff *skb = *pskb;
 441         const struct ethhdr *eth = eth_hdr(skb);
 442         const struct macvlan_dev *vlan;
 443         const struct macvlan_dev *src;
 444         struct net_device *dev;
 445         unsigned int len = 0;
 446         int ret;
 447         rx_handler_result_t handle_res;
 448
 449         port = macvlan_port_get_rcu(skb->dev);
 450         if (is_multicast_ether_addr(eth->h_dest)) {
 451                 unsigned int hash;
 452
 453                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 454                 if (!skb)
 455                         return RX_HANDLER_CONSUMED;
 456                 *pskb = skb;
 457                 eth = eth_hdr(skb);
 458                 macvlan_forward_source(skb, port, eth->h_source);
 459                 src = macvlan_hash_lookup(port, eth->h_source);
 460                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 461                     src->mode != MACVLAN_MODE_BRIDGE) {
 462                         /* forward to original port. */
 463                         vlan = src;
 464                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 465                               netif_rx(skb);
 466                         handle_res = RX_HANDLER_CONSUMED;
 467                         goto out;
 468                 }
 469
 470                 hash = mc_hash(NULL, eth->h_dest);
 471                 if (test_bit(hash, port->mc_filter))
 472                         macvlan_broadcast_enqueue(port, src, skb);
 473
 474                 return RX_HANDLER_PASS;
 475         }
 476
 477         macvlan_forward_source(skb, port, eth->h_source);
 478         if (macvlan_passthru(port))
 479                 vlan = list_first_or_null_rcu(&port->vlans,
 480                                               struct macvlan_dev, list);
 481         else
 482                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 483         if (vlan == NULL)
 484                 return RX_HANDLER_PASS;
 485
 486         dev = vlan->dev;
 487         if (unlikely(!(dev->flags & IFF_UP))) {
 488                 kfree_skb(skb);
 489                 return RX_HANDLER_CONSUMED;
 490         }
 491         len = skb->len + ETH_HLEN;
 492         skb = skb_share_check(skb, GFP_ATOMIC);
 493         if (!skb) {
 494                 ret = NET_RX_DROP;
 495                 handle_res = RX_HANDLER_CONSUMED;
 496                 goto out;
 497         }
 498
 499         *pskb = skb;
 500         skb->dev = dev;
 501         skb->pkt_type = PACKET_HOST;
 502
 503         ret = NET_RX_SUCCESS;
 504         handle_res = RX_HANDLER_ANOTHER;
 505 out:
 506         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 507         return handle_res;
 508 }
 509
 510 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 511 {
 512         const struct macvlan_dev *vlan = netdev_priv(dev);
 513         const struct macvlan_port *port = vlan->port;
 514         const struct macvlan_dev *dest;
 515
 516         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 517                 const struct ethhdr *eth = (void *)skb->data;
 518
 519                 /* send to other bridge ports directly */
 520                 if (is_multicast_ether_addr(eth->h_dest)) {
 521                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 522                         goto xmit_world;
 523                 }
 524
 525                 dest = macvlan_hash_lookup(port, eth->h_dest);
 526                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 527                         /* send to lowerdev first for its network taps */
 528                         dev_forward_skb(vlan->lowerdev, skb);
 529
 530                         return NET_XMIT_SUCCESS;
 531                 }
 532         }
 533
 534 xmit_world:
 535         skb->dev = vlan->lowerdev;
 536         return dev_queue_xmit(skb);
 537 }
 538
 539 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 540 {
 541 #ifdef CONFIG_NET_POLL_CONTROLLER
 542         if (vlan->netpoll)
 543                 netpoll_send_skb(vlan->netpoll, skb);
 544 #else
 545         BUG();
 546 #endif
 547         return NETDEV_TX_OK;
 548 }
 549
 550 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 551                                       struct net_device *dev)
 552 {
 553         unsigned int len = skb->len;
 554         int ret;
 555         struct macvlan_dev *vlan = netdev_priv(dev);
 556
 557         if (unlikely(netpoll_tx_running(dev)))
 558                 return macvlan_netpoll_send_skb(vlan, skb);
 559
 560         if (vlan->fwd_priv) {
 561                 skb->dev = vlan->lowerdev;
 562                 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
 563         } else {
 564                 ret = macvlan_queue_xmit(skb, dev);
 565         }
 566
 567         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 568                 struct vlan_pcpu_stats *pcpu_stats;
 569
 570                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 571                 u64_stats_update_begin(&pcpu_stats->syncp);
 572                 pcpu_stats->tx_packets++;
 573                 pcpu_stats->tx_bytes += len;
 574                 u64_stats_update_end(&pcpu_stats->syncp);
 575         } else {
 576                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 577         }
 578         return ret;
 579 }
 580
 581 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 582                                unsigned short type, const void *daddr,
 583                                const void *saddr, unsigned len)
 584 {
 585         const struct macvlan_dev *vlan = netdev_priv(dev);
 586         struct net_device *lowerdev = vlan->lowerdev;
 587
 588         return dev_hard_header(skb, lowerdev, type, daddr,
 589                                saddr ? : dev->dev_addr, len);
 590 }
 591
 592 static const struct header_ops macvlan_hard_header_ops = {
 593         .create         = macvlan_hard_header,
 594         .parse          = eth_header_parse,
 595         .cache          = eth_header_cache,
 596         .cache_update   = eth_header_cache_update,
 597 };
 598
 599 static struct rtnl_link_ops macvlan_link_ops;
 600
 601 static int macvlan_open(struct net_device *dev)
 602 {
 603         struct macvlan_dev *vlan = netdev_priv(dev);
 604         struct net_device *lowerdev = vlan->lowerdev;
 605         int err;
 606
 607         if (macvlan_passthru(vlan->port)) {
 608                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 609                         err = dev_set_promiscuity(lowerdev, 1);
 610                         if (err < 0)
 611                                 goto out;
 612                 }
 613                 goto hash_add;
 614         }
 615
 616         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
 617             dev->rtnl_link_ops == &macvlan_link_ops) {
 618                 vlan->fwd_priv =
 619                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 620
 621                 /* If we get a NULL pointer back, or if we get an error
 622                  * then we should just fall through to the non accelerated path
 623                  */
 624                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
 625                         vlan->fwd_priv = NULL;
 626                 } else
 627                         return 0;
 628         }
 629
 630         err = -EBUSY;
 631         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 632                 goto out;
 633
 634         err = dev_uc_add(lowerdev, dev->dev_addr);
 635         if (err < 0)
 636                 goto out;
 637         if (dev->flags & IFF_ALLMULTI) {
 638                 err = dev_set_allmulti(lowerdev, 1);
 639                 if (err < 0)
 640                         goto del_unicast;
 641         }
 642
 643         if (dev->flags & IFF_PROMISC) {
 644                 err = dev_set_promiscuity(lowerdev, 1);
 645                 if (err < 0)
 646                         goto clear_multi;
 647         }
 648
 649 hash_add:
 650         macvlan_hash_add(vlan);
 651         return 0;
 652
 653 clear_multi:
 654         if (dev->flags & IFF_ALLMULTI)
 655                 dev_set_allmulti(lowerdev, -1);
 656 del_unicast:
 657         dev_uc_del(lowerdev, dev->dev_addr);
 658 out:
 659         if (vlan->fwd_priv) {
 660                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 661                                                            vlan->fwd_priv);
 662                 vlan->fwd_priv = NULL;
 663         }
 664         return err;
 665 }
 666
 667 static int macvlan_stop(struct net_device *dev)
 668 {
 669         struct macvlan_dev *vlan = netdev_priv(dev);
 670         struct net_device *lowerdev = vlan->lowerdev;
 671
 672         if (vlan->fwd_priv) {
 673                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 674                                                            vlan->fwd_priv);
 675                 vlan->fwd_priv = NULL;
 676                 return 0;
 677         }
 678
 679         dev_uc_unsync(lowerdev, dev);
 680         dev_mc_unsync(lowerdev, dev);
 681
 682         if (macvlan_passthru(vlan->port)) {
 683                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 684                         dev_set_promiscuity(lowerdev, -1);
 685                 goto hash_del;
 686         }
 687
 688         if (dev->flags & IFF_ALLMULTI)
 689                 dev_set_allmulti(lowerdev, -1);
 690
 691         if (dev->flags & IFF_PROMISC)
 692                 dev_set_promiscuity(lowerdev, -1);
 693
 694         dev_uc_del(lowerdev, dev->dev_addr);
 695
 696 hash_del:
 697         macvlan_hash_del(vlan, !dev->dismantle);
 698         return 0;
 699 }
 700
 701 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 702 {
 703         struct macvlan_dev *vlan = netdev_priv(dev);
 704         struct net_device *lowerdev = vlan->lowerdev;
 705         struct macvlan_port *port = vlan->port;
 706         int err;
 707
 708         if (!(dev->flags & IFF_UP)) {
 709                 /* Just copy in the new address */
 710                 ether_addr_copy(dev->dev_addr, addr);
 711         } else {
 712                 /* Rehash and update the device filters */
 713                 if (macvlan_addr_busy(vlan->port, addr))
 714                         return -EBUSY;
 715
 716                 if (!macvlan_passthru(port)) {
 717                         err = dev_uc_add(lowerdev, addr);
 718                         if (err)
 719                                 return err;
 720
 721                         dev_uc_del(lowerdev, dev->dev_addr);
 722                 }
 723
 724                 macvlan_hash_change_addr(vlan, addr);
 725         }
 726         if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
 727                 /* Since addr_change isn't set, we are here due to lower
 728                  * device change.  Save the lower-dev address so we can
 729                  * restore it later.
 730                  */
 731                 ether_addr_copy(vlan->port->perm_addr,
 732                                 lowerdev->dev_addr);
 733         }
 734         macvlan_clear_addr_change(port);
 735         return 0;
 736 }
 737
 738 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 739 {
 740         struct macvlan_dev *vlan = netdev_priv(dev);
 741         struct sockaddr *addr = p;
 742
 743         if (!is_valid_ether_addr(addr->sa_data))
 744                 return -EADDRNOTAVAIL;
 745
 746         /* If the addresses are the same, this is a no-op */
 747         if (ether_addr_equal(dev->dev_addr, addr->sa_data))
 748                 return 0;
 749
 750         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 751                 macvlan_set_addr_change(vlan->port);
 752                 dev_set_mac_address(vlan->lowerdev, addr);
 753                 return 0;
 754         }
 755
 756         return macvlan_sync_address(dev, addr->sa_data);
 757 }
 758
 759 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 760 {
 761         struct macvlan_dev *vlan = netdev_priv(dev);
 762         struct net_device *lowerdev = vlan->lowerdev;
 763
 764         if (dev->flags & IFF_UP) {
 765                 if (change & IFF_ALLMULTI)
 766                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 767                 if (change & IFF_PROMISC)
 768                         dev_set_promiscuity(lowerdev,
 769                                             dev->flags & IFF_PROMISC ? 1 : -1);
 770
 771         }
 772 }
 773
 774 static void macvlan_compute_filter(unsigned long *mc_filter,
 775                                    struct net_device *dev,
 776                                    struct macvlan_dev *vlan)
 777 {
 778         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 779                 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
 780         } else {
 781                 struct netdev_hw_addr *ha;
 782                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 783
 784                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 785                 netdev_for_each_mc_addr(ha, dev) {
 786                         __set_bit(mc_hash(vlan, ha->addr), filter);
 787                 }
 788
 789                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 790
 791                 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 792         }
 793 }
 794
 795 static void macvlan_set_mac_lists(struct net_device *dev)
 796 {
 797         struct macvlan_dev *vlan = netdev_priv(dev);
 798
 799         macvlan_compute_filter(vlan->mc_filter, dev, vlan);
 800
 801         dev_uc_sync(vlan->lowerdev, dev);
 802         dev_mc_sync(vlan->lowerdev, dev);
 803
 804         /* This is slightly inaccurate as we're including the subscription
 805          * list of vlan->lowerdev too.
 806          *
 807          * Bug alert: This only works if everyone has the same broadcast
 808          * address as lowerdev.  As soon as someone changes theirs this
 809          * will break.
 810          *
 811          * However, this is already broken as when you change your broadcast
 812          * address we don't get called.
 813          *
 814          * The solution is to maintain a list of broadcast addresses like
 815          * we do for uc/mc, if you care.
 816          */
 817         macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 818 }
 819
 820 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 821 {
 822         struct macvlan_dev *vlan = netdev_priv(dev);
 823
 824         if (vlan->lowerdev->mtu < new_mtu)
 825                 return -EINVAL;
 826         dev->mtu = new_mtu;
 827         return 0;
 828 }
 829
 830 /*
 831  * macvlan network devices have devices nesting below it and are a special
 832  * "super class" of normal network devices; split their locks off into a
 833  * separate class since they always nest.
 834  */
 835 static struct lock_class_key macvlan_netdev_addr_lock_key;
 836
 837 #define ALWAYS_ON_OFFLOADS \
 838         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
 839          NETIF_F_GSO_ROBUST)
 840
 841 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
 842
 843 #define MACVLAN_FEATURES \
 844         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 845          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
 846          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 847          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 848
 849 #define MACVLAN_STATE_MASK \
 850         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 851
 852 static int macvlan_get_nest_level(struct net_device *dev)
 853 {
 854         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 855 }
 856
 857 static void macvlan_set_lockdep_class(struct net_device *dev)
 858 {
 859         netdev_lockdep_set_classes(dev);
 860         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 861                                        &macvlan_netdev_addr_lock_key,
 862                                        macvlan_get_nest_level(dev));
 863 }
 864
 865 static int macvlan_init(struct net_device *dev)
 866 {
 867         struct macvlan_dev *vlan = netdev_priv(dev);
 868         const struct net_device *lowerdev = vlan->lowerdev;
 869         struct macvlan_port *port = vlan->port;
 870
 871         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 872                                   (lowerdev->state & MACVLAN_STATE_MASK);
 873         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 874         dev->features           |= ALWAYS_ON_FEATURES;
 875         dev->hw_features        |= NETIF_F_LRO;
 876         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 877         dev->vlan_features      |= ALWAYS_ON_OFFLOADS;
 878         dev->gso_max_size       = lowerdev->gso_max_size;
 879         dev->gso_max_segs       = lowerdev->gso_max_segs;
 880         dev->hard_header_len    = lowerdev->hard_header_len;
 881
 882         macvlan_set_lockdep_class(dev);
 883
 884         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 885         if (!vlan->pcpu_stats)
 886                 return -ENOMEM;
 887
 888         port->count += 1;
 889
 890         return 0;
 891 }
 892
 893 static void macvlan_uninit(struct net_device *dev)
 894 {
 895         struct macvlan_dev *vlan = netdev_priv(dev);
 896         struct macvlan_port *port = vlan->port;
 897
 898         free_percpu(vlan->pcpu_stats);
 899
 900         macvlan_flush_sources(port, vlan);
 901         port->count -= 1;
 902         if (!port->count)
 903                 macvlan_port_destroy(port->dev);
 904 }
 905
 906 static void macvlan_dev_get_stats64(struct net_device *dev,
 907                                     struct rtnl_link_stats64 *stats)
 908 {
 909         struct macvlan_dev *vlan = netdev_priv(dev);
 910
 911         if (vlan->pcpu_stats) {
 912                 struct vlan_pcpu_stats *p;
 913                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 914                 u32 rx_errors = 0, tx_dropped = 0;
 915                 unsigned int start;
 916                 int i;
 917
 918                 for_each_possible_cpu(i) {
 919                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 920                         do {
 921                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 922                                 rx_packets      = p->rx_packets;
 923                                 rx_bytes        = p->rx_bytes;
 924                                 rx_multicast    = p->rx_multicast;
 925                                 tx_packets      = p->tx_packets;
 926                                 tx_bytes        = p->tx_bytes;
 927                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 928
 929                         stats->rx_packets       += rx_packets;
 930                         stats->rx_bytes         += rx_bytes;
 931                         stats->multicast        += rx_multicast;
 932                         stats->tx_packets       += tx_packets;
 933                         stats->tx_bytes         += tx_bytes;
 934                         /* rx_errors & tx_dropped are u32, updated
 935                          * without syncp protection.
 936                          */
 937                         rx_errors       += p->rx_errors;
 938                         tx_dropped      += p->tx_dropped;
 939                 }
 940                 stats->rx_errors        = rx_errors;
 941                 stats->rx_dropped       = rx_errors;
 942                 stats->tx_dropped       = tx_dropped;
 943         }
 944 }
 945
 946 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 947                                    __be16 proto, u16 vid)
 948 {
 949         struct macvlan_dev *vlan = netdev_priv(dev);
 950         struct net_device *lowerdev = vlan->lowerdev;
 951
 952         return vlan_vid_add(lowerdev, proto, vid);
 953 }
 954
 955 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 956                                     __be16 proto, u16 vid)
 957 {
 958         struct macvlan_dev *vlan = netdev_priv(dev);
 959         struct net_device *lowerdev = vlan->lowerdev;
 960
 961         vlan_vid_del(lowerdev, proto, vid);
 962         return 0;
 963 }
 964
 965 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 966                            struct net_device *dev,
 967                            const unsigned char *addr, u16 vid,
 968                            u16 flags)
 969 {
 970         struct macvlan_dev *vlan = netdev_priv(dev);
 971         int err = -EINVAL;
 972
 973         /* Support unicast filter only on passthru devices.
 974          * Multicast filter should be allowed on all devices.
 975          */
 976         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
 977                 return -EOPNOTSUPP;
 978
 979         if (flags & NLM_F_REPLACE)
 980                 return -EOPNOTSUPP;
 981
 982         if (is_unicast_ether_addr(addr))
 983                 err = dev_uc_add_excl(dev, addr);
 984         else if (is_multicast_ether_addr(addr))
 985                 err = dev_mc_add_excl(dev, addr);
 986
 987         return err;
 988 }
 989
 990 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 991                            struct net_device *dev,
 992                            const unsigned char *addr, u16 vid)
 993 {
 994         struct macvlan_dev *vlan = netdev_priv(dev);
 995         int err = -EINVAL;
 996
 997         /* Support unicast filter only on passthru devices.
 998          * Multicast filter should be allowed on all devices.
 999          */
1000         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1001                 return -EOPNOTSUPP;
1002
1003         if (is_unicast_ether_addr(addr))
1004                 err = dev_uc_del(dev, addr);
1005         else if (is_multicast_ether_addr(addr))
1006                 err = dev_mc_del(dev, addr);
1007
1008         return err;
1009 }
1010
1011 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1012                                         struct ethtool_drvinfo *drvinfo)
1013 {
1014         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1015         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1016 }
1017
1018 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1019                                               struct ethtool_link_ksettings *cmd)
1020 {
1021         const struct macvlan_dev *vlan = netdev_priv(dev);
1022
1023         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1024 }
1025
1026 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1027                                               netdev_features_t features)
1028 {
1029         struct macvlan_dev *vlan = netdev_priv(dev);
1030         netdev_features_t lowerdev_features = vlan->lowerdev->features;
1031         netdev_features_t mask;
1032
1033         features |= NETIF_F_ALL_FOR_ALL;
1034         features &= (vlan->set_features | ~MACVLAN_FEATURES);
1035         mask = features;
1036
1037         lowerdev_features &= (features | ~NETIF_F_LRO);
1038         features = netdev_increment_features(lowerdev_features, features, mask);
1039         features |= ALWAYS_ON_FEATURES;
1040         features &= ~NETIF_F_NETNS_LOCAL;
1041
1042         return features;
1043 }
1044
1045 #ifdef CONFIG_NET_POLL_CONTROLLER
1046 static void macvlan_dev_poll_controller(struct net_device *dev)
1047 {
1048         return;
1049 }
1050
1051 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1052 {
1053         struct macvlan_dev *vlan = netdev_priv(dev);
1054         struct net_device *real_dev = vlan->lowerdev;
1055         struct netpoll *netpoll;
1056         int err = 0;
1057
1058         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1059         err = -ENOMEM;
1060         if (!netpoll)
1061                 goto out;
1062
1063         err = __netpoll_setup(netpoll, real_dev);
1064         if (err) {
1065                 kfree(netpoll);
1066                 goto out;
1067         }
1068
1069         vlan->netpoll = netpoll;
1070
1071 out:
1072         return err;
1073 }
1074
1075 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1076 {
1077         struct macvlan_dev *vlan = netdev_priv(dev);
1078         struct netpoll *netpoll = vlan->netpoll;
1079
1080         if (!netpoll)
1081                 return;
1082
1083         vlan->netpoll = NULL;
1084
1085         __netpoll_free_async(netpoll);
1086 }
1087 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1088
1089 static int macvlan_dev_get_iflink(const struct net_device *dev)
1090 {
1091         struct macvlan_dev *vlan = netdev_priv(dev);
1092
1093         return vlan->lowerdev->ifindex;
1094 }
1095
1096 static const struct ethtool_ops macvlan_ethtool_ops = {
1097         .get_link               = ethtool_op_get_link,
1098         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1099         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1100 };
1101
1102 static const struct net_device_ops macvlan_netdev_ops = {
1103         .ndo_init               = macvlan_init,
1104         .ndo_uninit             = macvlan_uninit,
1105         .ndo_open               = macvlan_open,
1106         .ndo_stop               = macvlan_stop,
1107         .ndo_start_xmit         = macvlan_start_xmit,
1108         .ndo_change_mtu         = macvlan_change_mtu,
1109         .ndo_fix_features       = macvlan_fix_features,
1110         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1111         .ndo_set_mac_address    = macvlan_set_mac_address,
1112         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1113         .ndo_get_stats64        = macvlan_dev_get_stats64,
1114         .ndo_validate_addr      = eth_validate_addr,
1115         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1116         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1117         .ndo_fdb_add            = macvlan_fdb_add,
1118         .ndo_fdb_del            = macvlan_fdb_del,
1119         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1120         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1121 #ifdef CONFIG_NET_POLL_CONTROLLER
1122         .ndo_poll_controller    = macvlan_dev_poll_controller,
1123         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1124         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1125 #endif
1126         .ndo_get_iflink         = macvlan_dev_get_iflink,
1127         .ndo_features_check     = passthru_features_check,
1128 };
1129
1130 void macvlan_common_setup(struct net_device *dev)
1131 {
1132         ether_setup(dev);
1133
1134         dev->min_mtu            = 0;
1135         dev->max_mtu            = ETH_MAX_MTU;
1136         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1137         netif_keep_dst(dev);
1138         dev->priv_flags        |= IFF_UNICAST_FLT;
1139         dev->netdev_ops         = &macvlan_netdev_ops;
1140         dev->needs_free_netdev  = true;
1141         dev->header_ops         = &macvlan_hard_header_ops;
1142         dev->ethtool_ops        = &macvlan_ethtool_ops;
1143 }
1144 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1145
1146 static void macvlan_setup(struct net_device *dev)
1147 {
1148         macvlan_common_setup(dev);
1149         dev->priv_flags |= IFF_NO_QUEUE;
1150 }
1151
1152 static int macvlan_port_create(struct net_device *dev)
1153 {
1154         struct macvlan_port *port;
1155         unsigned int i;
1156         int err;
1157
1158         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1159                 return -EINVAL;
1160
1161         if (netdev_is_rx_handler_busy(dev))
1162                 return -EBUSY;
1163
1164         port = kzalloc(sizeof(*port), GFP_KERNEL);
1165         if (port == NULL)
1166                 return -ENOMEM;
1167
1168         port->dev = dev;
1169         ether_addr_copy(port->perm_addr, dev->dev_addr);
1170         INIT_LIST_HEAD(&port->vlans);
1171         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1172                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1173         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1174                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1175
1176         skb_queue_head_init(&port->bc_queue);
1177         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1178
1179         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1180         if (err)
1181                 kfree(port);
1182         else
1183                 dev->priv_flags |= IFF_MACVLAN_PORT;
1184         return err;
1185 }
1186
1187 static void macvlan_port_destroy(struct net_device *dev)
1188 {
1189         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1190         struct sk_buff *skb;
1191
1192         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1193         netdev_rx_handler_unregister(dev);
1194
1195         /* After this point, no packet can schedule bc_work anymore,
1196          * but we need to cancel it and purge left skbs if any.
1197          */
1198         cancel_work_sync(&port->bc_work);
1199
1200         while ((skb = __skb_dequeue(&port->bc_queue))) {
1201                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1202
1203                 if (src)
1204                         dev_put(src->dev);
1205
1206                 kfree_skb(skb);
1207         }
1208
1209         /* If the lower device address has been changed by passthru
1210          * macvlan, put it back.
1211          */
1212         if (macvlan_passthru(port) &&
1213             !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1214                 struct sockaddr sa;
1215
1216                 sa.sa_family = port->dev->type;
1217                 memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1218                 dev_set_mac_address(port->dev, &sa);
1219         }
1220
1221         kfree(port);
1222 }
1223
1224 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
1225 {
1226         if (tb[IFLA_ADDRESS]) {
1227                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1228                         return -EINVAL;
1229                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1230                         return -EADDRNOTAVAIL;
1231         }
1232
1233         if (data && data[IFLA_MACVLAN_FLAGS] &&
1234             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1235                 return -EINVAL;
1236
1237         if (data && data[IFLA_MACVLAN_MODE]) {
1238                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1239                 case MACVLAN_MODE_PRIVATE:
1240                 case MACVLAN_MODE_VEPA:
1241                 case MACVLAN_MODE_BRIDGE:
1242                 case MACVLAN_MODE_PASSTHRU:
1243                 case MACVLAN_MODE_SOURCE:
1244                         break;
1245                 default:
1246                         return -EINVAL;
1247                 }
1248         }
1249
1250         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1251                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1252                 case MACVLAN_MACADDR_ADD:
1253                 case MACVLAN_MACADDR_DEL:
1254                 case MACVLAN_MACADDR_FLUSH:
1255                 case MACVLAN_MACADDR_SET:
1256                         break;
1257                 default:
1258                         return -EINVAL;
1259                 }
1260         }
1261
1262         if (data && data[IFLA_MACVLAN_MACADDR]) {
1263                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1264                         return -EINVAL;
1265
1266                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1267                         return -EADDRNOTAVAIL;
1268         }
1269
1270         if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1271                 return -EINVAL;
1272
1273         return 0;
1274 }
1275
1276 /**
1277  * reconfigure list of remote source mac address
1278  * (only for macvlan devices in source mode)
1279  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1280  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1281  */
1282 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1283                                       struct nlattr *data[])
1284 {
1285         char *addr = NULL;
1286         int ret, rem, len;
1287         struct nlattr *nla, *head;
1288         struct macvlan_source_entry *entry;
1289
1290         if (data[IFLA_MACVLAN_MACADDR])
1291                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1292
1293         if (mode == MACVLAN_MACADDR_ADD) {
1294                 if (!addr)
1295                         return -EINVAL;
1296
1297                 return macvlan_hash_add_source(vlan, addr);
1298
1299         } else if (mode == MACVLAN_MACADDR_DEL) {
1300                 if (!addr)
1301                         return -EINVAL;
1302
1303                 entry = macvlan_hash_lookup_source(vlan, addr);
1304                 if (entry) {
1305                         macvlan_hash_del_source(entry);
1306                         vlan->macaddr_count--;
1307                 }
1308         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1309                 macvlan_flush_sources(vlan->port, vlan);
1310         } else if (mode == MACVLAN_MACADDR_SET) {
1311                 macvlan_flush_sources(vlan->port, vlan);
1312
1313                 if (addr) {
1314                         ret = macvlan_hash_add_source(vlan, addr);
1315                         if (ret)
1316                                 return ret;
1317                 }
1318
1319                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1320                         return 0;
1321
1322                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1323                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1324
1325                 nla_for_each_attr(nla, head, len, rem) {
1326                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1327                             nla_len(nla) != ETH_ALEN)
1328                                 continue;
1329
1330                         addr = nla_data(nla);
1331                         ret = macvlan_hash_add_source(vlan, addr);
1332                         if (ret)
1333                                 return ret;
1334                 }
1335         } else {
1336                 return -EINVAL;
1337         }
1338
1339         return 0;
1340 }
1341
1342 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1343                            struct nlattr *tb[], struct nlattr *data[])
1344 {
1345         struct macvlan_dev *vlan = netdev_priv(dev);
1346         struct macvlan_port *port;
1347         struct net_device *lowerdev;
1348         int err;
1349         int macmode;
1350         bool create = false;
1351
1352         if (!tb[IFLA_LINK])
1353                 return -EINVAL;
1354
1355         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1356         if (lowerdev == NULL)
1357                 return -ENODEV;
1358
1359         /* When creating macvlans or macvtaps on top of other macvlans - use
1360          * the real device as the lowerdev.
1361          */
1362         if (netif_is_macvlan(lowerdev))
1363                 lowerdev = macvlan_dev_real_dev(lowerdev);
1364
1365         if (!tb[IFLA_MTU])
1366                 dev->mtu = lowerdev->mtu;
1367         else if (dev->mtu > lowerdev->mtu)
1368                 return -EINVAL;
1369
1370         /* MTU range: 68 - lowerdev->max_mtu */
1371         dev->min_mtu = ETH_MIN_MTU;
1372         dev->max_mtu = lowerdev->max_mtu;
1373
1374         if (!tb[IFLA_ADDRESS])
1375                 eth_hw_addr_random(dev);
1376
1377         if (!macvlan_port_exists(lowerdev)) {
1378                 err = macvlan_port_create(lowerdev);
1379                 if (err < 0)
1380                         return err;
1381                 create = true;
1382         }
1383         port = macvlan_port_get_rtnl(lowerdev);
1384
1385         /* Only 1 macvlan device can be created in passthru mode */
1386         if (macvlan_passthru(port)) {
1387                 /* The macvlan port must be not created this time,
1388                  * still goto destroy_macvlan_port for readability.
1389                  */
1390                 err = -EINVAL;
1391                 goto destroy_macvlan_port;
1392         }
1393
1394         vlan->lowerdev = lowerdev;
1395         vlan->dev      = dev;
1396         vlan->port     = port;
1397         vlan->set_features = MACVLAN_FEATURES;
1398         vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1399
1400         vlan->mode     = MACVLAN_MODE_VEPA;
1401         if (data && data[IFLA_MACVLAN_MODE])
1402                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1403
1404         if (data && data[IFLA_MACVLAN_FLAGS])
1405                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1406
1407         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1408                 if (port->count) {
1409                         err = -EINVAL;
1410                         goto destroy_macvlan_port;
1411                 }
1412                 macvlan_set_passthru(port);
1413                 eth_hw_addr_inherit(dev, lowerdev);
1414         }
1415
1416         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1417                 if (vlan->mode != MACVLAN_MODE_SOURCE) {
1418                         err = -EINVAL;
1419                         goto destroy_macvlan_port;
1420                 }
1421                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1422                 err = macvlan_changelink_sources(vlan, macmode, data);
1423                 if (err)
1424                         goto destroy_macvlan_port;
1425         }
1426
1427         err = register_netdevice(dev);
1428         if (err < 0)
1429                 goto destroy_macvlan_port;
1430
1431         dev->priv_flags |= IFF_MACVLAN;
1432         err = netdev_upper_dev_link(lowerdev, dev);
1433         if (err)
1434                 goto unregister_netdev;
1435
1436         list_add_tail_rcu(&vlan->list, &port->vlans);
1437         netif_stacked_transfer_operstate(lowerdev, dev);
1438         linkwatch_fire_event(dev);
1439
1440         return 0;
1441
1442 unregister_netdev:
1443         unregister_netdevice(dev);
1444 destroy_macvlan_port:
1445         if (create)
1446                 macvlan_port_destroy(port->dev);
1447         return err;
1448 }
1449 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1450
1451 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1452                            struct nlattr *tb[], struct nlattr *data[])
1453 {
1454         return macvlan_common_newlink(src_net, dev, tb, data);
1455 }
1456
1457 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1458 {
1459         struct macvlan_dev *vlan = netdev_priv(dev);
1460
1461         if (vlan->mode == MACVLAN_MODE_SOURCE)
1462                 macvlan_flush_sources(vlan->port, vlan);
1463         list_del_rcu(&vlan->list);
1464         unregister_netdevice_queue(dev, head);
1465         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1466 }
1467 EXPORT_SYMBOL_GPL(macvlan_dellink);
1468
1469 static int macvlan_changelink(struct net_device *dev,
1470                 struct nlattr *tb[], struct nlattr *data[])
1471 {
1472         struct macvlan_dev *vlan = netdev_priv(dev);
1473         enum macvlan_mode mode;
1474         bool set_mode = false;
1475         enum macvlan_macaddr_mode macmode;
1476         int ret;
1477
1478         /* Validate mode, but don't set yet: setting flags may fail. */
1479         if (data && data[IFLA_MACVLAN_MODE]) {
1480                 set_mode = true;
1481                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1482                 /* Passthrough mode can't be set or cleared dynamically */
1483                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1484                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1485                         return -EINVAL;
1486                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1487                     vlan->mode != mode)
1488                         macvlan_flush_sources(vlan->port, vlan);
1489         }
1490
1491         if (data && data[IFLA_MACVLAN_FLAGS]) {
1492                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1493                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1494                 if (macvlan_passthru(vlan->port) && promisc) {
1495                         int err;
1496
1497                         if (flags & MACVLAN_FLAG_NOPROMISC)
1498                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1499                         else
1500                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1501                         if (err < 0)
1502                                 return err;
1503                 }
1504                 vlan->flags = flags;
1505         }
1506         if (set_mode)
1507                 vlan->mode = mode;
1508         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1509                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1510                         return -EINVAL;
1511                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1512                 ret = macvlan_changelink_sources(vlan, macmode, data);
1513                 if (ret)
1514                         return ret;
1515         }
1516         return 0;
1517 }
1518
1519 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1520 {
1521         if (vlan->macaddr_count == 0)
1522                 return 0;
1523         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1524                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1525 }
1526
1527 static size_t macvlan_get_size(const struct net_device *dev)
1528 {
1529         struct macvlan_dev *vlan = netdev_priv(dev);
1530
1531         return (0
1532                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1533                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1534                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1535                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1536                 );
1537 }
1538
1539 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1540                                      const struct macvlan_dev *vlan,
1541                                      const int i)
1542 {
1543         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1544         struct macvlan_source_entry *entry;
1545
1546         hlist_for_each_entry_rcu(entry, h, hlist) {
1547                 if (entry->vlan != vlan)
1548                         continue;
1549                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1550                         return 1;
1551         }
1552         return 0;
1553 }
1554
1555 static int macvlan_fill_info(struct sk_buff *skb,
1556                                 const struct net_device *dev)
1557 {
1558         struct macvlan_dev *vlan = netdev_priv(dev);
1559         int i;
1560         struct nlattr *nest;
1561
1562         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1563                 goto nla_put_failure;
1564         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1565                 goto nla_put_failure;
1566         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1567                 goto nla_put_failure;
1568         if (vlan->macaddr_count > 0) {
1569                 nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1570                 if (nest == NULL)
1571                         goto nla_put_failure;
1572
1573                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1574                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1575                                 goto nla_put_failure;
1576                 }
1577                 nla_nest_end(skb, nest);
1578         }
1579         return 0;
1580
1581 nla_put_failure:
1582         return -EMSGSIZE;
1583 }
1584
1585 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1586         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1587         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1588         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1589         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1590         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1591         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1592 };
1593
1594 int macvlan_link_register(struct rtnl_link_ops *ops)
1595 {
1596         /* common fields */
1597         ops->validate           = macvlan_validate;
1598         ops->maxtype            = IFLA_MACVLAN_MAX;
1599         ops->policy             = macvlan_policy;
1600         ops->changelink         = macvlan_changelink;
1601         ops->get_size           = macvlan_get_size;
1602         ops->fill_info          = macvlan_fill_info;
1603
1604         return rtnl_link_register(ops);
1605 };
1606 EXPORT_SYMBOL_GPL(macvlan_link_register);
1607
1608 static struct net *macvlan_get_link_net(const struct net_device *dev)
1609 {
1610         return dev_net(macvlan_dev_real_dev(dev));
1611 }
1612
1613 static struct rtnl_link_ops macvlan_link_ops = {
1614         .kind           = "macvlan",
1615         .setup          = macvlan_setup,
1616         .newlink        = macvlan_newlink,
1617         .dellink        = macvlan_dellink,
1618         .get_link_net   = macvlan_get_link_net,
1619         .priv_size      = sizeof(struct macvlan_dev),
1620 };
1621
1622 static int macvlan_device_event(struct notifier_block *unused,
1623                                 unsigned long event, void *ptr)
1624 {
1625         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1626         struct macvlan_dev *vlan, *next;
1627         struct macvlan_port *port;
1628         LIST_HEAD(list_kill);
1629
1630         if (!macvlan_port_exists(dev))
1631                 return NOTIFY_DONE;
1632
1633         port = macvlan_port_get_rtnl(dev);
1634
1635         switch (event) {
1636         case NETDEV_UP:
1637         case NETDEV_CHANGE:
1638                 list_for_each_entry(vlan, &port->vlans, list)
1639                         netif_stacked_transfer_operstate(vlan->lowerdev,
1640                                                          vlan->dev);
1641                 break;
1642         case NETDEV_FEAT_CHANGE:
1643                 list_for_each_entry(vlan, &port->vlans, list) {
1644                         vlan->dev->gso_max_size = dev->gso_max_size;
1645                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1646                         netdev_update_features(vlan->dev);
1647                 }
1648                 break;
1649         case NETDEV_CHANGEMTU:
1650                 list_for_each_entry(vlan, &port->vlans, list) {
1651                         if (vlan->dev->mtu <= dev->mtu)
1652                                 continue;
1653                         dev_set_mtu(vlan->dev, dev->mtu);
1654                 }
1655                 break;
1656         case NETDEV_CHANGEADDR:
1657                 if (!macvlan_passthru(port))
1658                         return NOTIFY_DONE;
1659
1660                 vlan = list_first_entry_or_null(&port->vlans,
1661                                                 struct macvlan_dev,
1662                                                 list);
1663
1664                 if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1665                         return NOTIFY_BAD;
1666
1667                 break;
1668         case NETDEV_UNREGISTER:
1669                 /* twiddle thumbs on netns device moves */
1670                 if (dev->reg_state != NETREG_UNREGISTERING)
1671                         break;
1672
1673                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1674                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1675                 unregister_netdevice_many(&list_kill);
1676                 break;
1677         case NETDEV_PRE_TYPE_CHANGE:
1678                 /* Forbid underlaying device to change its type. */
1679                 return NOTIFY_BAD;
1680
1681         case NETDEV_NOTIFY_PEERS:
1682         case NETDEV_BONDING_FAILOVER:
1683         case NETDEV_RESEND_IGMP:
1684                 /* Propagate to all vlans */
1685                 list_for_each_entry(vlan, &port->vlans, list)
1686                         call_netdevice_notifiers(event, vlan->dev);
1687         }
1688         return NOTIFY_DONE;
1689 }
1690
1691 static struct notifier_block macvlan_notifier_block __read_mostly = {
1692         .notifier_call  = macvlan_device_event,
1693 };
1694
1695 static int __init macvlan_init_module(void)
1696 {
1697         int err;
1698
1699         register_netdevice_notifier(&macvlan_notifier_block);
1700
1701         err = macvlan_link_register(&macvlan_link_ops);
1702         if (err < 0)
1703                 goto err1;
1704         return 0;
1705 err1:
1706         unregister_netdevice_notifier(&macvlan_notifier_block);
1707         return err;
1708 }
1709
1710 static void __exit macvlan_cleanup_module(void)
1711 {
1712         rtnl_link_unregister(&macvlan_link_ops);
1713         unregister_netdevice_notifier(&macvlan_notifier_block);
1714 }
1715
1716 module_init(macvlan_init_module);
1717 module_exit(macvlan_cleanup_module);
1718
1719 MODULE_LICENSE("GPL");
1720 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1721 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1722 MODULE_ALIAS_RTNL_LINK("macvlan");