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                 cond_resched();
 343         }
 344 }
 345
 346 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 347                                       const struct macvlan_dev *src,
 348                                       struct sk_buff *skb)
 349 {
 350         struct sk_buff *nskb;
 351         int err = -ENOMEM;
 352
 353         nskb = skb_clone(skb, GFP_ATOMIC);
 354         if (!nskb)
 355                 goto err;
 356
 357         MACVLAN_SKB_CB(nskb)->src = src;
 358
 359         spin_lock(&port->bc_queue.lock);
 360         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 361                 if (src)
 362                         dev_hold(src->dev);
 363                 __skb_queue_tail(&port->bc_queue, nskb);
 364                 err = 0;
 365         }
 366         spin_unlock(&port->bc_queue.lock);
 367
 368         schedule_work(&port->bc_work);
 369
 370         if (err)
 371                 goto free_nskb;
 372
 373         return;
 374
 375 free_nskb:
 376         kfree_skb(nskb);
 377 err:
 378         atomic_long_inc(&skb->dev->rx_dropped);
 379 }
 380
 381 static void macvlan_flush_sources(struct macvlan_port *port,
 382                                   struct macvlan_dev *vlan)
 383 {
 384         int i;
 385
 386         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 387                 struct hlist_node *h, *n;
 388
 389                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 390                         struct macvlan_source_entry *entry;
 391
 392                         entry = hlist_entry(h, struct macvlan_source_entry,
 393                                             hlist);
 394                         if (entry->vlan == vlan)
 395                                 macvlan_hash_del_source(entry);
 396                 }
 397         }
 398         vlan->macaddr_count = 0;
 399 }
 400
 401 static void macvlan_forward_source_one(struct sk_buff *skb,
 402                                        struct macvlan_dev *vlan)
 403 {
 404         struct sk_buff *nskb;
 405         struct net_device *dev;
 406         int len;
 407         int ret;
 408
 409         dev = vlan->dev;
 410         if (unlikely(!(dev->flags & IFF_UP)))
 411                 return;
 412
 413         nskb = skb_clone(skb, GFP_ATOMIC);
 414         if (!nskb)
 415                 return;
 416
 417         len = nskb->len + ETH_HLEN;
 418         nskb->dev = dev;
 419
 420         if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
 421                 nskb->pkt_type = PACKET_HOST;
 422
 423         ret = netif_rx(nskb);
 424         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 425 }
 426
 427 static void macvlan_forward_source(struct sk_buff *skb,
 428                                    struct macvlan_port *port,
 429                                    const unsigned char *addr)
 430 {
 431         struct macvlan_source_entry *entry;
 432         u32 idx = macvlan_eth_hash(addr);
 433         struct hlist_head *h = &port->vlan_source_hash[idx];
 434
 435         hlist_for_each_entry_rcu(entry, h, hlist) {
 436                 if (ether_addr_equal_64bits(entry->addr, addr))
 437                         macvlan_forward_source_one(skb, entry->vlan);
 438         }
 439 }
 440
 441 /* called under rcu_read_lock() from netif_receive_skb */
 442 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 443 {
 444         struct macvlan_port *port;
 445         struct sk_buff *skb = *pskb;
 446         const struct ethhdr *eth = eth_hdr(skb);
 447         const struct macvlan_dev *vlan;
 448         const struct macvlan_dev *src;
 449         struct net_device *dev;
 450         unsigned int len = 0;
 451         int ret;
 452         rx_handler_result_t handle_res;
 453
 454         /* Packets from dev_loopback_xmit() do not have L2 header, bail out */
 455         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 456                 return RX_HANDLER_PASS;
 457
 458         port = macvlan_port_get_rcu(skb->dev);
 459         if (is_multicast_ether_addr(eth->h_dest)) {
 460                 unsigned int hash;
 461
 462                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 463                 if (!skb)
 464                         return RX_HANDLER_CONSUMED;
 465                 *pskb = skb;
 466                 eth = eth_hdr(skb);
 467                 macvlan_forward_source(skb, port, eth->h_source);
 468                 src = macvlan_hash_lookup(port, eth->h_source);
 469                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 470                     src->mode != MACVLAN_MODE_BRIDGE) {
 471                         /* forward to original port. */
 472                         vlan = src;
 473                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 474                               netif_rx(skb);
 475                         handle_res = RX_HANDLER_CONSUMED;
 476                         goto out;
 477                 }
 478
 479                 hash = mc_hash(NULL, eth->h_dest);
 480                 if (test_bit(hash, port->mc_filter))
 481                         macvlan_broadcast_enqueue(port, src, skb);
 482
 483                 return RX_HANDLER_PASS;
 484         }
 485
 486         macvlan_forward_source(skb, port, eth->h_source);
 487         if (macvlan_passthru(port))
 488                 vlan = list_first_or_null_rcu(&port->vlans,
 489                                               struct macvlan_dev, list);
 490         else
 491                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 492         if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
 493                 return RX_HANDLER_PASS;
 494
 495         dev = vlan->dev;
 496         if (unlikely(!(dev->flags & IFF_UP))) {
 497                 kfree_skb(skb);
 498                 return RX_HANDLER_CONSUMED;
 499         }
 500         len = skb->len + ETH_HLEN;
 501         skb = skb_share_check(skb, GFP_ATOMIC);
 502         if (!skb) {
 503                 ret = NET_RX_DROP;
 504                 handle_res = RX_HANDLER_CONSUMED;
 505                 goto out;
 506         }
 507
 508         *pskb = skb;
 509         skb->dev = dev;
 510         skb->pkt_type = PACKET_HOST;
 511
 512         ret = NET_RX_SUCCESS;
 513         handle_res = RX_HANDLER_ANOTHER;
 514 out:
 515         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 516         return handle_res;
 517 }
 518
 519 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 520 {
 521         const struct macvlan_dev *vlan = netdev_priv(dev);
 522         const struct macvlan_port *port = vlan->port;
 523         const struct macvlan_dev *dest;
 524
 525         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 526                 const struct ethhdr *eth = skb_eth_hdr(skb);
 527
 528                 /* send to other bridge ports directly */
 529                 if (is_multicast_ether_addr(eth->h_dest)) {
 530                         skb_reset_mac_header(skb);
 531                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 532                         goto xmit_world;
 533                 }
 534
 535                 dest = macvlan_hash_lookup(port, eth->h_dest);
 536                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 537                         /* send to lowerdev first for its network taps */
 538                         dev_forward_skb(vlan->lowerdev, skb);
 539
 540                         return NET_XMIT_SUCCESS;
 541                 }
 542         }
 543 xmit_world:
 544         skb->dev = vlan->lowerdev;
 545         return dev_queue_xmit_accel(skb,
 546                                     netdev_get_sb_channel(dev) ? dev : NULL);
 547 }
 548
 549 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 550 {
 551 #ifdef CONFIG_NET_POLL_CONTROLLER
 552         if (vlan->netpoll)
 553                 netpoll_send_skb(vlan->netpoll, skb);
 554 #else
 555         BUG();
 556 #endif
 557         return NETDEV_TX_OK;
 558 }
 559
 560 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 561                                       struct net_device *dev)
 562 {
 563         struct macvlan_dev *vlan = netdev_priv(dev);
 564         unsigned int len = skb->len;
 565         int ret;
 566
 567         if (unlikely(netpoll_tx_running(dev)))
 568                 return macvlan_netpoll_send_skb(vlan, skb);
 569
 570         ret = macvlan_queue_xmit(skb, dev);
 571
 572         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 573                 struct vlan_pcpu_stats *pcpu_stats;
 574
 575                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 576                 u64_stats_update_begin(&pcpu_stats->syncp);
 577                 pcpu_stats->tx_packets++;
 578                 pcpu_stats->tx_bytes += len;
 579                 u64_stats_update_end(&pcpu_stats->syncp);
 580         } else {
 581                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 582         }
 583         return ret;
 584 }
 585
 586 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 587                                unsigned short type, const void *daddr,
 588                                const void *saddr, unsigned len)
 589 {
 590         const struct macvlan_dev *vlan = netdev_priv(dev);
 591         struct net_device *lowerdev = vlan->lowerdev;
 592
 593         return dev_hard_header(skb, lowerdev, type, daddr,
 594                                saddr ? : dev->dev_addr, len);
 595 }
 596
 597 static const struct header_ops macvlan_hard_header_ops = {
 598         .create         = macvlan_hard_header,
 599         .parse          = eth_header_parse,
 600         .cache          = eth_header_cache,
 601         .cache_update   = eth_header_cache_update,
 602 };
 603
 604 static int macvlan_open(struct net_device *dev)
 605 {
 606         struct macvlan_dev *vlan = netdev_priv(dev);
 607         struct net_device *lowerdev = vlan->lowerdev;
 608         int err;
 609
 610         if (macvlan_passthru(vlan->port)) {
 611                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 612                         err = dev_set_promiscuity(lowerdev, 1);
 613                         if (err < 0)
 614                                 goto out;
 615                 }
 616                 goto hash_add;
 617         }
 618
 619         err = -EADDRINUSE;
 620         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 621                 goto out;
 622
 623         /* Attempt to populate accel_priv which is used to offload the L2
 624          * forwarding requests for unicast packets.
 625          */
 626         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD)
 627                 vlan->accel_priv =
 628                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 629
 630         /* If earlier attempt to offload failed, or accel_priv is not
 631          * populated we must add the unicast address to the lower device.
 632          */
 633         if (IS_ERR_OR_NULL(vlan->accel_priv)) {
 634                 vlan->accel_priv = NULL;
 635                 err = dev_uc_add(lowerdev, dev->dev_addr);
 636                 if (err < 0)
 637                         goto out;
 638         }
 639
 640         if (dev->flags & IFF_ALLMULTI) {
 641                 err = dev_set_allmulti(lowerdev, 1);
 642                 if (err < 0)
 643                         goto del_unicast;
 644         }
 645
 646         if (dev->flags & IFF_PROMISC) {
 647                 err = dev_set_promiscuity(lowerdev, 1);
 648                 if (err < 0)
 649                         goto clear_multi;
 650         }
 651
 652 hash_add:
 653         macvlan_hash_add(vlan);
 654         return 0;
 655
 656 clear_multi:
 657         if (dev->flags & IFF_ALLMULTI)
 658                 dev_set_allmulti(lowerdev, -1);
 659 del_unicast:
 660         if (vlan->accel_priv) {
 661                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 662                                                            vlan->accel_priv);
 663                 vlan->accel_priv = NULL;
 664         } else {
 665                 dev_uc_del(lowerdev, dev->dev_addr);
 666         }
 667 out:
 668         return err;
 669 }
 670
 671 static int macvlan_stop(struct net_device *dev)
 672 {
 673         struct macvlan_dev *vlan = netdev_priv(dev);
 674         struct net_device *lowerdev = vlan->lowerdev;
 675
 676         if (vlan->accel_priv) {
 677                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 678                                                            vlan->accel_priv);
 679                 vlan->accel_priv = NULL;
 680         }
 681
 682         dev_uc_unsync(lowerdev, dev);
 683         dev_mc_unsync(lowerdev, dev);
 684
 685         if (macvlan_passthru(vlan->port)) {
 686                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 687                         dev_set_promiscuity(lowerdev, -1);
 688                 goto hash_del;
 689         }
 690
 691         if (dev->flags & IFF_ALLMULTI)
 692                 dev_set_allmulti(lowerdev, -1);
 693
 694         if (dev->flags & IFF_PROMISC)
 695                 dev_set_promiscuity(lowerdev, -1);
 696
 697         dev_uc_del(lowerdev, dev->dev_addr);
 698
 699 hash_del:
 700         macvlan_hash_del(vlan, !dev->dismantle);
 701         return 0;
 702 }
 703
 704 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 705 {
 706         struct macvlan_dev *vlan = netdev_priv(dev);
 707         struct net_device *lowerdev = vlan->lowerdev;
 708         struct macvlan_port *port = vlan->port;
 709         int err;
 710
 711         if (!(dev->flags & IFF_UP)) {
 712                 /* Just copy in the new address */
 713                 ether_addr_copy(dev->dev_addr, addr);
 714         } else {
 715                 /* Rehash and update the device filters */
 716                 if (macvlan_addr_busy(vlan->port, addr))
 717                         return -EADDRINUSE;
 718
 719                 if (!macvlan_passthru(port)) {
 720                         err = dev_uc_add(lowerdev, addr);
 721                         if (err)
 722                                 return err;
 723
 724                         dev_uc_del(lowerdev, dev->dev_addr);
 725                 }
 726
 727                 macvlan_hash_change_addr(vlan, addr);
 728         }
 729         if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
 730                 /* Since addr_change isn't set, we are here due to lower
 731                  * device change.  Save the lower-dev address so we can
 732                  * restore it later.
 733                  */
 734                 ether_addr_copy(vlan->port->perm_addr,
 735                                 lowerdev->dev_addr);
 736         }
 737         macvlan_clear_addr_change(port);
 738         return 0;
 739 }
 740
 741 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 742 {
 743         struct macvlan_dev *vlan = netdev_priv(dev);
 744         struct sockaddr *addr = p;
 745
 746         if (!is_valid_ether_addr(addr->sa_data))
 747                 return -EADDRNOTAVAIL;
 748
 749         /* If the addresses are the same, this is a no-op */
 750         if (ether_addr_equal(dev->dev_addr, addr->sa_data))
 751                 return 0;
 752
 753         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 754                 macvlan_set_addr_change(vlan->port);
 755                 return dev_set_mac_address(vlan->lowerdev, addr);
 756         }
 757
 758         if (macvlan_addr_busy(vlan->port, addr->sa_data))
 759                 return -EADDRINUSE;
 760
 761         return macvlan_sync_address(dev, addr->sa_data);
 762 }
 763
 764 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 765 {
 766         struct macvlan_dev *vlan = netdev_priv(dev);
 767         struct net_device *lowerdev = vlan->lowerdev;
 768
 769         if (dev->flags & IFF_UP) {
 770                 if (change & IFF_ALLMULTI)
 771                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 772                 if (change & IFF_PROMISC)
 773                         dev_set_promiscuity(lowerdev,
 774                                             dev->flags & IFF_PROMISC ? 1 : -1);
 775
 776         }
 777 }
 778
 779 static void macvlan_compute_filter(unsigned long *mc_filter,
 780                                    struct net_device *dev,
 781                                    struct macvlan_dev *vlan)
 782 {
 783         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 784                 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
 785         } else {
 786                 struct netdev_hw_addr *ha;
 787                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 788
 789                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 790                 netdev_for_each_mc_addr(ha, dev) {
 791                         __set_bit(mc_hash(vlan, ha->addr), filter);
 792                 }
 793
 794                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 795
 796                 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 797         }
 798 }
 799
 800 static void macvlan_set_mac_lists(struct net_device *dev)
 801 {
 802         struct macvlan_dev *vlan = netdev_priv(dev);
 803
 804         macvlan_compute_filter(vlan->mc_filter, dev, vlan);
 805
 806         dev_uc_sync(vlan->lowerdev, dev);
 807         dev_mc_sync(vlan->lowerdev, dev);
 808
 809         /* This is slightly inaccurate as we're including the subscription
 810          * list of vlan->lowerdev too.
 811          *
 812          * Bug alert: This only works if everyone has the same broadcast
 813          * address as lowerdev.  As soon as someone changes theirs this
 814          * will break.
 815          *
 816          * However, this is already broken as when you change your broadcast
 817          * address we don't get called.
 818          *
 819          * The solution is to maintain a list of broadcast addresses like
 820          * we do for uc/mc, if you care.
 821          */
 822         macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 823 }
 824
 825 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 826 {
 827         struct macvlan_dev *vlan = netdev_priv(dev);
 828
 829         if (vlan->lowerdev->mtu < new_mtu)
 830                 return -EINVAL;
 831         dev->mtu = new_mtu;
 832         return 0;
 833 }
 834
 835 /*
 836  * macvlan network devices have devices nesting below it and are a special
 837  * "super class" of normal network devices; split their locks off into a
 838  * separate class since they always nest.
 839  */
 840 static struct lock_class_key macvlan_netdev_addr_lock_key;
 841
 842 #define ALWAYS_ON_OFFLOADS \
 843         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
 844          NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
 845
 846 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
 847
 848 #define MACVLAN_FEATURES \
 849         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 850          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
 851          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 852          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 853
 854 #define MACVLAN_STATE_MASK \
 855         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 856
 857 static int macvlan_get_nest_level(struct net_device *dev)
 858 {
 859         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 860 }
 861
 862 static void macvlan_set_lockdep_class(struct net_device *dev)
 863 {
 864         netdev_lockdep_set_classes(dev);
 865         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 866                                        &macvlan_netdev_addr_lock_key,
 867                                        macvlan_get_nest_level(dev));
 868 }
 869
 870 static int macvlan_init(struct net_device *dev)
 871 {
 872         struct macvlan_dev *vlan = netdev_priv(dev);
 873         const struct net_device *lowerdev = vlan->lowerdev;
 874         struct macvlan_port *port = vlan->port;
 875
 876         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 877                                   (lowerdev->state & MACVLAN_STATE_MASK);
 878         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 879         dev->features           |= ALWAYS_ON_FEATURES;
 880         dev->hw_features        |= NETIF_F_LRO;
 881         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 882         dev->vlan_features      |= ALWAYS_ON_OFFLOADS;
 883         dev->hw_enc_features    |= dev->features;
 884         dev->gso_max_size       = lowerdev->gso_max_size;
 885         dev->gso_max_segs       = lowerdev->gso_max_segs;
 886         dev->hard_header_len    = lowerdev->hard_header_len;
 887
 888         macvlan_set_lockdep_class(dev);
 889
 890         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 891         if (!vlan->pcpu_stats)
 892                 return -ENOMEM;
 893
 894         port->count += 1;
 895
 896         return 0;
 897 }
 898
 899 static void macvlan_uninit(struct net_device *dev)
 900 {
 901         struct macvlan_dev *vlan = netdev_priv(dev);
 902         struct macvlan_port *port = vlan->port;
 903
 904         free_percpu(vlan->pcpu_stats);
 905
 906         macvlan_flush_sources(port, vlan);
 907         port->count -= 1;
 908         if (!port->count)
 909                 macvlan_port_destroy(port->dev);
 910 }
 911
 912 static void macvlan_dev_get_stats64(struct net_device *dev,
 913                                     struct rtnl_link_stats64 *stats)
 914 {
 915         struct macvlan_dev *vlan = netdev_priv(dev);
 916
 917         if (vlan->pcpu_stats) {
 918                 struct vlan_pcpu_stats *p;
 919                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 920                 u32 rx_errors = 0, tx_dropped = 0;
 921                 unsigned int start;
 922                 int i;
 923
 924                 for_each_possible_cpu(i) {
 925                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 926                         do {
 927                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 928                                 rx_packets      = p->rx_packets;
 929                                 rx_bytes        = p->rx_bytes;
 930                                 rx_multicast    = p->rx_multicast;
 931                                 tx_packets      = p->tx_packets;
 932                                 tx_bytes        = p->tx_bytes;
 933                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 934
 935                         stats->rx_packets       += rx_packets;
 936                         stats->rx_bytes         += rx_bytes;
 937                         stats->multicast        += rx_multicast;
 938                         stats->tx_packets       += tx_packets;
 939                         stats->tx_bytes         += tx_bytes;
 940                         /* rx_errors & tx_dropped are u32, updated
 941                          * without syncp protection.
 942                          */
 943                         rx_errors       += p->rx_errors;
 944                         tx_dropped      += p->tx_dropped;
 945                 }
 946                 stats->rx_errors        = rx_errors;
 947                 stats->rx_dropped       = rx_errors;
 948                 stats->tx_dropped       = tx_dropped;
 949         }
 950 }
 951
 952 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 953                                    __be16 proto, u16 vid)
 954 {
 955         struct macvlan_dev *vlan = netdev_priv(dev);
 956         struct net_device *lowerdev = vlan->lowerdev;
 957
 958         return vlan_vid_add(lowerdev, proto, vid);
 959 }
 960
 961 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 962                                     __be16 proto, u16 vid)
 963 {
 964         struct macvlan_dev *vlan = netdev_priv(dev);
 965         struct net_device *lowerdev = vlan->lowerdev;
 966
 967         vlan_vid_del(lowerdev, proto, vid);
 968         return 0;
 969 }
 970
 971 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 972                            struct net_device *dev,
 973                            const unsigned char *addr, u16 vid,
 974                            u16 flags)
 975 {
 976         struct macvlan_dev *vlan = netdev_priv(dev);
 977         int err = -EINVAL;
 978
 979         /* Support unicast filter only on passthru devices.
 980          * Multicast filter should be allowed on all devices.
 981          */
 982         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
 983                 return -EOPNOTSUPP;
 984
 985         if (flags & NLM_F_REPLACE)
 986                 return -EOPNOTSUPP;
 987
 988         if (is_unicast_ether_addr(addr))
 989                 err = dev_uc_add_excl(dev, addr);
 990         else if (is_multicast_ether_addr(addr))
 991                 err = dev_mc_add_excl(dev, addr);
 992
 993         return err;
 994 }
 995
 996 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 997                            struct net_device *dev,
 998                            const unsigned char *addr, u16 vid)
 999 {
1000         struct macvlan_dev *vlan = netdev_priv(dev);
1001         int err = -EINVAL;
1002
1003         /* Support unicast filter only on passthru devices.
1004          * Multicast filter should be allowed on all devices.
1005          */
1006         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1007                 return -EOPNOTSUPP;
1008
1009         if (is_unicast_ether_addr(addr))
1010                 err = dev_uc_del(dev, addr);
1011         else if (is_multicast_ether_addr(addr))
1012                 err = dev_mc_del(dev, addr);
1013
1014         return err;
1015 }
1016
1017 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1018                                         struct ethtool_drvinfo *drvinfo)
1019 {
1020         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1021         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1022 }
1023
1024 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1025                                               struct ethtool_link_ksettings *cmd)
1026 {
1027         const struct macvlan_dev *vlan = netdev_priv(dev);
1028
1029         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1030 }
1031
1032 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1033                                               netdev_features_t features)
1034 {
1035         struct macvlan_dev *vlan = netdev_priv(dev);
1036         netdev_features_t lowerdev_features = vlan->lowerdev->features;
1037         netdev_features_t mask;
1038
1039         features |= NETIF_F_ALL_FOR_ALL;
1040         features &= (vlan->set_features | ~MACVLAN_FEATURES);
1041         mask = features;
1042
1043         lowerdev_features &= (features | ~NETIF_F_LRO);
1044         features = netdev_increment_features(lowerdev_features, features, mask);
1045         features |= ALWAYS_ON_FEATURES;
1046         features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1047
1048         return features;
1049 }
1050
1051 #ifdef CONFIG_NET_POLL_CONTROLLER
1052 static void macvlan_dev_poll_controller(struct net_device *dev)
1053 {
1054         return;
1055 }
1056
1057 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1058 {
1059         struct macvlan_dev *vlan = netdev_priv(dev);
1060         struct net_device *real_dev = vlan->lowerdev;
1061         struct netpoll *netpoll;
1062         int err = 0;
1063
1064         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1065         err = -ENOMEM;
1066         if (!netpoll)
1067                 goto out;
1068
1069         err = __netpoll_setup(netpoll, real_dev);
1070         if (err) {
1071                 kfree(netpoll);
1072                 goto out;
1073         }
1074
1075         vlan->netpoll = netpoll;
1076
1077 out:
1078         return err;
1079 }
1080
1081 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1082 {
1083         struct macvlan_dev *vlan = netdev_priv(dev);
1084         struct netpoll *netpoll = vlan->netpoll;
1085
1086         if (!netpoll)
1087                 return;
1088
1089         vlan->netpoll = NULL;
1090
1091         __netpoll_free_async(netpoll);
1092 }
1093 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1094
1095 static int macvlan_dev_get_iflink(const struct net_device *dev)
1096 {
1097         struct macvlan_dev *vlan = netdev_priv(dev);
1098
1099         return vlan->lowerdev->ifindex;
1100 }
1101
1102 static const struct ethtool_ops macvlan_ethtool_ops = {
1103         .get_link               = ethtool_op_get_link,
1104         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1105         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1106 };
1107
1108 static const struct net_device_ops macvlan_netdev_ops = {
1109         .ndo_init               = macvlan_init,
1110         .ndo_uninit             = macvlan_uninit,
1111         .ndo_open               = macvlan_open,
1112         .ndo_stop               = macvlan_stop,
1113         .ndo_start_xmit         = macvlan_start_xmit,
1114         .ndo_change_mtu         = macvlan_change_mtu,
1115         .ndo_fix_features       = macvlan_fix_features,
1116         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1117         .ndo_set_mac_address    = macvlan_set_mac_address,
1118         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1119         .ndo_get_stats64        = macvlan_dev_get_stats64,
1120         .ndo_validate_addr      = eth_validate_addr,
1121         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1122         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1123         .ndo_fdb_add            = macvlan_fdb_add,
1124         .ndo_fdb_del            = macvlan_fdb_del,
1125         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1126         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1127 #ifdef CONFIG_NET_POLL_CONTROLLER
1128         .ndo_poll_controller    = macvlan_dev_poll_controller,
1129         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1130         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1131 #endif
1132         .ndo_get_iflink         = macvlan_dev_get_iflink,
1133         .ndo_features_check     = passthru_features_check,
1134 };
1135
1136 void macvlan_common_setup(struct net_device *dev)
1137 {
1138         ether_setup(dev);
1139
1140         /* ether_setup() has set dev->min_mtu to ETH_MIN_MTU. */
1141         dev->max_mtu            = ETH_MAX_MTU;
1142         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1143         netif_keep_dst(dev);
1144         dev->priv_flags        |= IFF_UNICAST_FLT;
1145         dev->netdev_ops         = &macvlan_netdev_ops;
1146         dev->needs_free_netdev  = true;
1147         dev->header_ops         = &macvlan_hard_header_ops;
1148         dev->ethtool_ops        = &macvlan_ethtool_ops;
1149 }
1150 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1151
1152 static void macvlan_setup(struct net_device *dev)
1153 {
1154         macvlan_common_setup(dev);
1155         dev->priv_flags |= IFF_NO_QUEUE;
1156 }
1157
1158 static int macvlan_port_create(struct net_device *dev)
1159 {
1160         struct macvlan_port *port;
1161         unsigned int i;
1162         int err;
1163
1164         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1165                 return -EINVAL;
1166
1167         if (netdev_is_rx_handler_busy(dev))
1168                 return -EBUSY;
1169
1170         port = kzalloc(sizeof(*port), GFP_KERNEL);
1171         if (port == NULL)
1172                 return -ENOMEM;
1173
1174         port->dev = dev;
1175         ether_addr_copy(port->perm_addr, dev->dev_addr);
1176         INIT_LIST_HEAD(&port->vlans);
1177         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1178                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1179         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1180                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1181
1182         skb_queue_head_init(&port->bc_queue);
1183         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1184
1185         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1186         if (err)
1187                 kfree(port);
1188         else
1189                 dev->priv_flags |= IFF_MACVLAN_PORT;
1190         return err;
1191 }
1192
1193 static void macvlan_port_destroy(struct net_device *dev)
1194 {
1195         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1196         struct sk_buff *skb;
1197
1198         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1199         netdev_rx_handler_unregister(dev);
1200
1201         /* After this point, no packet can schedule bc_work anymore,
1202          * but we need to cancel it and purge left skbs if any.
1203          */
1204         cancel_work_sync(&port->bc_work);
1205
1206         while ((skb = __skb_dequeue(&port->bc_queue))) {
1207                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1208
1209                 if (src)
1210                         dev_put(src->dev);
1211
1212                 kfree_skb(skb);
1213         }
1214
1215         /* If the lower device address has been changed by passthru
1216          * macvlan, put it back.
1217          */
1218         if (macvlan_passthru(port) &&
1219             !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1220                 struct sockaddr sa;
1221
1222                 sa.sa_family = port->dev->type;
1223                 memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1224                 dev_set_mac_address(port->dev, &sa);
1225         }
1226
1227         kfree(port);
1228 }
1229
1230 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1231                             struct netlink_ext_ack *extack)
1232 {
1233         struct nlattr *nla, *head;
1234         int rem, len;
1235
1236         if (tb[IFLA_ADDRESS]) {
1237                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1238                         return -EINVAL;
1239                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1240                         return -EADDRNOTAVAIL;
1241         }
1242
1243         if (!data)
1244                 return 0;
1245
1246         if (data[IFLA_MACVLAN_FLAGS] &&
1247             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1248                 return -EINVAL;
1249
1250         if (data[IFLA_MACVLAN_MODE]) {
1251                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1252                 case MACVLAN_MODE_PRIVATE:
1253                 case MACVLAN_MODE_VEPA:
1254                 case MACVLAN_MODE_BRIDGE:
1255                 case MACVLAN_MODE_PASSTHRU:
1256                 case MACVLAN_MODE_SOURCE:
1257                         break;
1258                 default:
1259                         return -EINVAL;
1260                 }
1261         }
1262
1263         if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1264                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1265                 case MACVLAN_MACADDR_ADD:
1266                 case MACVLAN_MACADDR_DEL:
1267                 case MACVLAN_MACADDR_FLUSH:
1268                 case MACVLAN_MACADDR_SET:
1269                         break;
1270                 default:
1271                         return -EINVAL;
1272                 }
1273         }
1274
1275         if (data[IFLA_MACVLAN_MACADDR]) {
1276                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1277                         return -EINVAL;
1278
1279                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1280                         return -EADDRNOTAVAIL;
1281         }
1282
1283         if (data[IFLA_MACVLAN_MACADDR_DATA]) {
1284                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1285                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1286
1287                 nla_for_each_attr(nla, head, len, rem) {
1288                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1289                             nla_len(nla) != ETH_ALEN)
1290                                 return -EINVAL;
1291
1292                         if (!is_valid_ether_addr(nla_data(nla)))
1293                                 return -EADDRNOTAVAIL;
1294                 }
1295         }
1296
1297         if (data[IFLA_MACVLAN_MACADDR_COUNT])
1298                 return -EINVAL;
1299
1300         return 0;
1301 }
1302
1303 /**
1304  * reconfigure list of remote source mac address
1305  * (only for macvlan devices in source mode)
1306  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1307  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1308  */
1309 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1310                                       struct nlattr *data[])
1311 {
1312         char *addr = NULL;
1313         int ret, rem, len;
1314         struct nlattr *nla, *head;
1315         struct macvlan_source_entry *entry;
1316
1317         if (data[IFLA_MACVLAN_MACADDR])
1318                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1319
1320         if (mode == MACVLAN_MACADDR_ADD) {
1321                 if (!addr)
1322                         return -EINVAL;
1323
1324                 return macvlan_hash_add_source(vlan, addr);
1325
1326         } else if (mode == MACVLAN_MACADDR_DEL) {
1327                 if (!addr)
1328                         return -EINVAL;
1329
1330                 entry = macvlan_hash_lookup_source(vlan, addr);
1331                 if (entry) {
1332                         macvlan_hash_del_source(entry);
1333                         vlan->macaddr_count--;
1334                 }
1335         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1336                 macvlan_flush_sources(vlan->port, vlan);
1337         } else if (mode == MACVLAN_MACADDR_SET) {
1338                 macvlan_flush_sources(vlan->port, vlan);
1339
1340                 if (addr) {
1341                         ret = macvlan_hash_add_source(vlan, addr);
1342                         if (ret)
1343                                 return ret;
1344                 }
1345
1346                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1347                         return 0;
1348
1349                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1350                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1351
1352                 nla_for_each_attr(nla, head, len, rem) {
1353                         addr = nla_data(nla);
1354                         ret = macvlan_hash_add_source(vlan, addr);
1355                         if (ret)
1356                                 return ret;
1357                 }
1358         } else {
1359                 return -EINVAL;
1360         }
1361
1362         return 0;
1363 }
1364
1365 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1366                            struct nlattr *tb[], struct nlattr *data[],
1367                            struct netlink_ext_ack *extack)
1368 {
1369         struct macvlan_dev *vlan = netdev_priv(dev);
1370         struct macvlan_port *port;
1371         struct net_device *lowerdev;
1372         int err;
1373         int macmode;
1374         bool create = false;
1375
1376         if (!tb[IFLA_LINK])
1377                 return -EINVAL;
1378
1379         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1380         if (lowerdev == NULL)
1381                 return -ENODEV;
1382
1383         /* When creating macvlans or macvtaps on top of other macvlans - use
1384          * the real device as the lowerdev.
1385          */
1386         if (netif_is_macvlan(lowerdev))
1387                 lowerdev = macvlan_dev_real_dev(lowerdev);
1388
1389         if (!tb[IFLA_MTU])
1390                 dev->mtu = lowerdev->mtu;
1391         else if (dev->mtu > lowerdev->mtu)
1392                 return -EINVAL;
1393
1394         /* MTU range: 68 - lowerdev->max_mtu */
1395         dev->min_mtu = ETH_MIN_MTU;
1396         dev->max_mtu = lowerdev->max_mtu;
1397
1398         if (!tb[IFLA_ADDRESS])
1399                 eth_hw_addr_random(dev);
1400
1401         if (!macvlan_port_exists(lowerdev)) {
1402                 err = macvlan_port_create(lowerdev);
1403                 if (err < 0)
1404                         return err;
1405                 create = true;
1406         }
1407         port = macvlan_port_get_rtnl(lowerdev);
1408
1409         /* Only 1 macvlan device can be created in passthru mode */
1410         if (macvlan_passthru(port)) {
1411                 /* The macvlan port must be not created this time,
1412                  * still goto destroy_macvlan_port for readability.
1413                  */
1414                 err = -EINVAL;
1415                 goto destroy_macvlan_port;
1416         }
1417
1418         vlan->lowerdev = lowerdev;
1419         vlan->dev      = dev;
1420         vlan->port     = port;
1421         vlan->set_features = MACVLAN_FEATURES;
1422         vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1423
1424         vlan->mode     = MACVLAN_MODE_VEPA;
1425         if (data && data[IFLA_MACVLAN_MODE])
1426                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1427
1428         if (data && data[IFLA_MACVLAN_FLAGS])
1429                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1430
1431         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1432                 if (port->count) {
1433                         err = -EINVAL;
1434                         goto destroy_macvlan_port;
1435                 }
1436                 macvlan_set_passthru(port);
1437                 eth_hw_addr_inherit(dev, lowerdev);
1438         }
1439
1440         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1441                 if (vlan->mode != MACVLAN_MODE_SOURCE) {
1442                         err = -EINVAL;
1443                         goto destroy_macvlan_port;
1444                 }
1445                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1446                 err = macvlan_changelink_sources(vlan, macmode, data);
1447                 if (err)
1448                         goto destroy_macvlan_port;
1449         }
1450
1451         err = register_netdevice(dev);
1452         if (err < 0)
1453                 goto destroy_macvlan_port;
1454
1455         dev->priv_flags |= IFF_MACVLAN;
1456         err = netdev_upper_dev_link(lowerdev, dev, extack);
1457         if (err)
1458                 goto unregister_netdev;
1459
1460         list_add_tail_rcu(&vlan->list, &port->vlans);
1461         netif_stacked_transfer_operstate(lowerdev, dev);
1462         linkwatch_fire_event(dev);
1463
1464         return 0;
1465
1466 unregister_netdev:
1467         /* macvlan_uninit would free the macvlan port */
1468         unregister_netdevice(dev);
1469         return err;
1470 destroy_macvlan_port:
1471         /* the macvlan port may be freed by macvlan_uninit when fail to register.
1472          * so we destroy the macvlan port only when it's valid.
1473          */
1474         if (create && macvlan_port_get_rtnl(lowerdev)) {
1475                 macvlan_flush_sources(port, vlan);
1476                 macvlan_port_destroy(port->dev);
1477         }
1478         return err;
1479 }
1480 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1481
1482 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1483                            struct nlattr *tb[], struct nlattr *data[],
1484                            struct netlink_ext_ack *extack)
1485 {
1486         return macvlan_common_newlink(src_net, dev, tb, data, extack);
1487 }
1488
1489 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1490 {
1491         struct macvlan_dev *vlan = netdev_priv(dev);
1492
1493         if (vlan->mode == MACVLAN_MODE_SOURCE)
1494                 macvlan_flush_sources(vlan->port, vlan);
1495         list_del_rcu(&vlan->list);
1496         unregister_netdevice_queue(dev, head);
1497         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1498 }
1499 EXPORT_SYMBOL_GPL(macvlan_dellink);
1500
1501 static int macvlan_changelink(struct net_device *dev,
1502                               struct nlattr *tb[], struct nlattr *data[],
1503                               struct netlink_ext_ack *extack)
1504 {
1505         struct macvlan_dev *vlan = netdev_priv(dev);
1506         enum macvlan_mode mode;
1507         bool set_mode = false;
1508         enum macvlan_macaddr_mode macmode;
1509         int ret;
1510
1511         /* Validate mode, but don't set yet: setting flags may fail. */
1512         if (data && data[IFLA_MACVLAN_MODE]) {
1513                 set_mode = true;
1514                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1515                 /* Passthrough mode can't be set or cleared dynamically */
1516                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1517                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1518                         return -EINVAL;
1519                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1520                     vlan->mode != mode)
1521                         macvlan_flush_sources(vlan->port, vlan);
1522         }
1523
1524         if (data && data[IFLA_MACVLAN_FLAGS]) {
1525                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1526                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1527                 if (macvlan_passthru(vlan->port) && promisc) {
1528                         int err;
1529
1530                         if (flags & MACVLAN_FLAG_NOPROMISC)
1531                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1532                         else
1533                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1534                         if (err < 0)
1535                                 return err;
1536                 }
1537                 vlan->flags = flags;
1538         }
1539         if (set_mode)
1540                 vlan->mode = mode;
1541         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1542                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1543                         return -EINVAL;
1544                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1545                 ret = macvlan_changelink_sources(vlan, macmode, data);
1546                 if (ret)
1547                         return ret;
1548         }
1549         return 0;
1550 }
1551
1552 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1553 {
1554         if (vlan->macaddr_count == 0)
1555                 return 0;
1556         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1557                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1558 }
1559
1560 static size_t macvlan_get_size(const struct net_device *dev)
1561 {
1562         struct macvlan_dev *vlan = netdev_priv(dev);
1563
1564         return (0
1565                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1566                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1567                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1568                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1569                 );
1570 }
1571
1572 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1573                                      const struct macvlan_dev *vlan,
1574                                      const int i)
1575 {
1576         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1577         struct macvlan_source_entry *entry;
1578
1579         hlist_for_each_entry_rcu(entry, h, hlist) {
1580                 if (entry->vlan != vlan)
1581                         continue;
1582                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1583                         return 1;
1584         }
1585         return 0;
1586 }
1587
1588 static int macvlan_fill_info(struct sk_buff *skb,
1589                                 const struct net_device *dev)
1590 {
1591         struct macvlan_dev *vlan = netdev_priv(dev);
1592         int i;
1593         struct nlattr *nest;
1594
1595         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1596                 goto nla_put_failure;
1597         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1598                 goto nla_put_failure;
1599         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1600                 goto nla_put_failure;
1601         if (vlan->macaddr_count > 0) {
1602                 nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1603                 if (nest == NULL)
1604                         goto nla_put_failure;
1605
1606                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1607                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1608                                 goto nla_put_failure;
1609                 }
1610                 nla_nest_end(skb, nest);
1611         }
1612         return 0;
1613
1614 nla_put_failure:
1615         return -EMSGSIZE;
1616 }
1617
1618 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1619         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1620         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1621         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1622         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1623         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1624         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1625 };
1626
1627 int macvlan_link_register(struct rtnl_link_ops *ops)
1628 {
1629         /* common fields */
1630         ops->validate           = macvlan_validate;
1631         ops->maxtype            = IFLA_MACVLAN_MAX;
1632         ops->policy             = macvlan_policy;
1633         ops->changelink         = macvlan_changelink;
1634         ops->get_size           = macvlan_get_size;
1635         ops->fill_info          = macvlan_fill_info;
1636
1637         return rtnl_link_register(ops);
1638 };
1639 EXPORT_SYMBOL_GPL(macvlan_link_register);
1640
1641 static struct net *macvlan_get_link_net(const struct net_device *dev)
1642 {
1643         return dev_net(macvlan_dev_real_dev(dev));
1644 }
1645
1646 static struct rtnl_link_ops macvlan_link_ops = {
1647         .kind           = "macvlan",
1648         .setup          = macvlan_setup,
1649         .newlink        = macvlan_newlink,
1650         .dellink        = macvlan_dellink,
1651         .get_link_net   = macvlan_get_link_net,
1652         .priv_size      = sizeof(struct macvlan_dev),
1653 };
1654
1655 static int macvlan_device_event(struct notifier_block *unused,
1656                                 unsigned long event, void *ptr)
1657 {
1658         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1659         struct macvlan_dev *vlan, *next;
1660         struct macvlan_port *port;
1661         LIST_HEAD(list_kill);
1662
1663         if (!macvlan_port_exists(dev))
1664                 return NOTIFY_DONE;
1665
1666         port = macvlan_port_get_rtnl(dev);
1667
1668         switch (event) {
1669         case NETDEV_UP:
1670         case NETDEV_DOWN:
1671         case NETDEV_CHANGE:
1672                 list_for_each_entry(vlan, &port->vlans, list)
1673                         netif_stacked_transfer_operstate(vlan->lowerdev,
1674                                                          vlan->dev);
1675                 break;
1676         case NETDEV_FEAT_CHANGE:
1677                 list_for_each_entry(vlan, &port->vlans, list) {
1678                         vlan->dev->gso_max_size = dev->gso_max_size;
1679                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1680                         netdev_update_features(vlan->dev);
1681                 }
1682                 break;
1683         case NETDEV_CHANGEMTU:
1684                 list_for_each_entry(vlan, &port->vlans, list) {
1685                         if (vlan->dev->mtu <= dev->mtu)
1686                                 continue;
1687                         dev_set_mtu(vlan->dev, dev->mtu);
1688                 }
1689                 break;
1690         case NETDEV_CHANGEADDR:
1691                 if (!macvlan_passthru(port))
1692                         return NOTIFY_DONE;
1693
1694                 vlan = list_first_entry_or_null(&port->vlans,
1695                                                 struct macvlan_dev,
1696                                                 list);
1697
1698                 if (vlan && macvlan_sync_address(vlan->dev, dev->dev_addr))
1699                         return NOTIFY_BAD;
1700
1701                 break;
1702         case NETDEV_UNREGISTER:
1703                 /* twiddle thumbs on netns device moves */
1704                 if (dev->reg_state != NETREG_UNREGISTERING)
1705                         break;
1706
1707                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1708                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1709                 unregister_netdevice_many(&list_kill);
1710                 break;
1711         case NETDEV_PRE_TYPE_CHANGE:
1712                 /* Forbid underlaying device to change its type. */
1713                 return NOTIFY_BAD;
1714
1715         case NETDEV_NOTIFY_PEERS:
1716         case NETDEV_BONDING_FAILOVER:
1717         case NETDEV_RESEND_IGMP:
1718                 /* Propagate to all vlans */
1719                 list_for_each_entry(vlan, &port->vlans, list)
1720                         call_netdevice_notifiers(event, vlan->dev);
1721         }
1722         return NOTIFY_DONE;
1723 }
1724
1725 static struct notifier_block macvlan_notifier_block __read_mostly = {
1726         .notifier_call  = macvlan_device_event,
1727 };
1728
1729 static int __init macvlan_init_module(void)
1730 {
1731         int err;
1732
1733         register_netdevice_notifier(&macvlan_notifier_block);
1734
1735         err = macvlan_link_register(&macvlan_link_ops);
1736         if (err < 0)
1737                 goto err1;
1738         return 0;
1739 err1:
1740         unregister_netdevice_notifier(&macvlan_notifier_block);
1741         return err;
1742 }
1743
1744 static void __exit macvlan_cleanup_module(void)
1745 {
1746         rtnl_link_unregister(&macvlan_link_ops);
1747         unregister_netdevice_notifier(&macvlan_notifier_block);
1748 }
1749
1750 module_init(macvlan_init_module);
1751 module_exit(macvlan_cleanup_module);
1752
1753 MODULE_LICENSE("GPL");
1754 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1755 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1756 MODULE_ALIAS_RTNL_LINK("macvlan");