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