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