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
  37 #define MACVLAN_HASH_SIZE       (1 << BITS_PER_BYTE)
  38
  39 struct macvlan_port {
  40         struct net_device       *dev;
  41         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  42         struct list_head        vlans;
  43         struct rcu_head         rcu;
  44         struct sk_buff_head     bc_queue;
  45         struct work_struct      bc_work;
  46         bool                    passthru;
  47 };
  48
  49 #define MACVLAN_PORT_IS_EMPTY(port)    list_empty(&port->vlans)
  50
  51 struct macvlan_skb_cb {
  52         const struct macvlan_dev *src;
  53 };
  54
  55 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
  56
  57 static void macvlan_port_destroy(struct net_device *dev);
  58
  59 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
  60 {
  61         return rcu_dereference(dev->rx_handler_data);
  62 }
  63
  64 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
  65 {
  66         return rtnl_dereference(dev->rx_handler_data);
  67 }
  68
  69 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
  70
  71 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
  72                                                const unsigned char *addr)
  73 {
  74         struct macvlan_dev *vlan;
  75
  76         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) {
  77                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
  78                         return vlan;
  79         }
  80         return NULL;
  81 }
  82
  83 static void macvlan_hash_add(struct macvlan_dev *vlan)
  84 {
  85         struct macvlan_port *port = vlan->port;
  86         const unsigned char *addr = vlan->dev->dev_addr;
  87
  88         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
  89 }
  90
  91 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
  92 {
  93         hlist_del_rcu(&vlan->hlist);
  94         if (sync)
  95                 synchronize_rcu();
  96 }
  97
  98 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
  99                                         const unsigned char *addr)
 100 {
 101         macvlan_hash_del(vlan, true);
 102         /* Now that we are unhashed it is safe to change the device
 103          * address without confusing packet delivery.
 104          */
 105         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
 106         macvlan_hash_add(vlan);
 107 }
 108
 109 static int macvlan_addr_busy(const struct macvlan_port *port,
 110                                 const unsigned char *addr)
 111 {
 112         /* Test to see if the specified multicast address is
 113          * currently in use by the underlying device or
 114          * another macvlan.
 115          */
 116         if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
 117                 return 1;
 118
 119         if (macvlan_hash_lookup(port, addr))
 120                 return 1;
 121
 122         return 0;
 123 }
 124
 125
 126 static int macvlan_broadcast_one(struct sk_buff *skb,
 127                                  const struct macvlan_dev *vlan,
 128                                  const struct ethhdr *eth, bool local)
 129 {
 130         struct net_device *dev = vlan->dev;
 131
 132         if (local)
 133                 return __dev_forward_skb(dev, skb);
 134
 135         skb->dev = dev;
 136         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 137                 skb->pkt_type = PACKET_BROADCAST;
 138         else
 139                 skb->pkt_type = PACKET_MULTICAST;
 140
 141         return 0;
 142 }
 143
 144 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 145 {
 146         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 147 }
 148
 149
 150 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 151                             const unsigned char *addr)
 152 {
 153         u32 val = __get_unaligned_cpu32(addr + 2);
 154
 155         val ^= macvlan_hash_mix(vlan);
 156         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 157 }
 158
 159 static void macvlan_broadcast(struct sk_buff *skb,
 160                               const struct macvlan_port *port,
 161                               struct net_device *src,
 162                               enum macvlan_mode mode)
 163 {
 164         const struct ethhdr *eth = eth_hdr(skb);
 165         const struct macvlan_dev *vlan;
 166         struct sk_buff *nskb;
 167         unsigned int i;
 168         int err;
 169         unsigned int hash;
 170
 171         if (skb->protocol == htons(ETH_P_PAUSE))
 172                 return;
 173
 174         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 175                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 176                         if (vlan->dev == src || !(vlan->mode & mode))
 177                                 continue;
 178
 179                         hash = mc_hash(vlan, eth->h_dest);
 180                         if (!test_bit(hash, vlan->mc_filter))
 181                                 continue;
 182
 183                         err = NET_RX_DROP;
 184                         nskb = skb_clone(skb, GFP_ATOMIC);
 185                         if (likely(nskb))
 186                                 err = macvlan_broadcast_one(
 187                                         nskb, vlan, eth,
 188                                         mode == MACVLAN_MODE_BRIDGE) ?:
 189                                       netif_rx_ni(nskb);
 190                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 191                                          err == NET_RX_SUCCESS, 1);
 192                 }
 193         }
 194 }
 195
 196 static void macvlan_process_broadcast(struct work_struct *w)
 197 {
 198         struct macvlan_port *port = container_of(w, struct macvlan_port,
 199                                                  bc_work);
 200         struct sk_buff *skb;
 201         struct sk_buff_head list;
 202
 203         skb_queue_head_init(&list);
 204
 205         spin_lock_bh(&port->bc_queue.lock);
 206         skb_queue_splice_tail_init(&port->bc_queue, &list);
 207         spin_unlock_bh(&port->bc_queue.lock);
 208
 209         while ((skb = __skb_dequeue(&list))) {
 210                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
 211
 212                 rcu_read_lock();
 213
 214                 if (!src)
 215                         /* frame comes from an external address */
 216                         macvlan_broadcast(skb, port, NULL,
 217                                           MACVLAN_MODE_PRIVATE |
 218                                           MACVLAN_MODE_VEPA    |
 219                                           MACVLAN_MODE_PASSTHRU|
 220                                           MACVLAN_MODE_BRIDGE);
 221                 else if (src->mode == MACVLAN_MODE_VEPA)
 222                         /* flood to everyone except source */
 223                         macvlan_broadcast(skb, port, src->dev,
 224                                           MACVLAN_MODE_VEPA |
 225                                           MACVLAN_MODE_BRIDGE);
 226                 else
 227                         /*
 228                          * flood only to VEPA ports, bridge ports
 229                          * already saw the frame on the way out.
 230                          */
 231                         macvlan_broadcast(skb, port, src->dev,
 232                                           MACVLAN_MODE_VEPA);
 233
 234                 rcu_read_unlock();
 235
 236                 kfree_skb(skb);
 237         }
 238 }
 239
 240 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 241                                       struct sk_buff *skb)
 242 {
 243         struct sk_buff *nskb;
 244         int err = -ENOMEM;
 245
 246         nskb = skb_clone(skb, GFP_ATOMIC);
 247         if (!nskb)
 248                 goto err;
 249
 250         spin_lock(&port->bc_queue.lock);
 251         if (skb_queue_len(&port->bc_queue) < skb->dev->tx_queue_len) {
 252                 __skb_queue_tail(&port->bc_queue, nskb);
 253                 err = 0;
 254         }
 255         spin_unlock(&port->bc_queue.lock);
 256
 257         if (err)
 258                 goto free_nskb;
 259
 260         schedule_work(&port->bc_work);
 261         return;
 262
 263 free_nskb:
 264         kfree_skb(nskb);
 265 err:
 266         atomic_long_inc(&skb->dev->rx_dropped);
 267 }
 268
 269 /* called under rcu_read_lock() from netif_receive_skb */
 270 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 271 {
 272         struct macvlan_port *port;
 273         struct sk_buff *skb = *pskb;
 274         const struct ethhdr *eth = eth_hdr(skb);
 275         const struct macvlan_dev *vlan;
 276         const struct macvlan_dev *src;
 277         struct net_device *dev;
 278         unsigned int len = 0;
 279         int ret = NET_RX_DROP;
 280
 281         port = macvlan_port_get_rcu(skb->dev);
 282         if (is_multicast_ether_addr(eth->h_dest)) {
 283                 skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
 284                 if (!skb)
 285                         return RX_HANDLER_CONSUMED;
 286                 eth = eth_hdr(skb);
 287                 src = macvlan_hash_lookup(port, eth->h_source);
 288                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 289                     src->mode != MACVLAN_MODE_BRIDGE) {
 290                         /* forward to original port. */
 291                         vlan = src;
 292                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 293                               netif_rx(skb);
 294                         goto out;
 295                 }
 296
 297                 MACVLAN_SKB_CB(skb)->src = src;
 298                 macvlan_broadcast_enqueue(port, skb);
 299
 300                 return RX_HANDLER_PASS;
 301         }
 302
 303         if (port->passthru)
 304                 vlan = list_first_or_null_rcu(&port->vlans,
 305                                               struct macvlan_dev, list);
 306         else
 307                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 308         if (vlan == NULL)
 309                 return RX_HANDLER_PASS;
 310
 311         dev = vlan->dev;
 312         if (unlikely(!(dev->flags & IFF_UP))) {
 313                 kfree_skb(skb);
 314                 return RX_HANDLER_CONSUMED;
 315         }
 316         len = skb->len + ETH_HLEN;
 317         skb = skb_share_check(skb, GFP_ATOMIC);
 318         if (!skb)
 319                 goto out;
 320
 321         skb->dev = dev;
 322         skb->pkt_type = PACKET_HOST;
 323
 324         ret = netif_rx(skb);
 325
 326 out:
 327         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
 328         return RX_HANDLER_CONSUMED;
 329 }
 330
 331 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 332 {
 333         const struct macvlan_dev *vlan = netdev_priv(dev);
 334         const struct macvlan_port *port = vlan->port;
 335         const struct macvlan_dev *dest;
 336
 337         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 338                 const struct ethhdr *eth = (void *)skb->data;
 339
 340                 /* send to other bridge ports directly */
 341                 if (is_multicast_ether_addr(eth->h_dest)) {
 342                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 343                         goto xmit_world;
 344                 }
 345
 346                 dest = macvlan_hash_lookup(port, eth->h_dest);
 347                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 348                         /* send to lowerdev first for its network taps */
 349                         dev_forward_skb(vlan->lowerdev, skb);
 350
 351                         return NET_XMIT_SUCCESS;
 352                 }
 353         }
 354
 355 xmit_world:
 356         skb->dev = vlan->lowerdev;
 357         return dev_queue_xmit(skb);
 358 }
 359
 360 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 361                                       struct net_device *dev)
 362 {
 363         unsigned int len = skb->len;
 364         int ret;
 365         const struct macvlan_dev *vlan = netdev_priv(dev);
 366
 367         if (vlan->fwd_priv) {
 368                 skb->dev = vlan->lowerdev;
 369                 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
 370         } else {
 371                 ret = macvlan_queue_xmit(skb, dev);
 372         }
 373
 374         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 375                 struct vlan_pcpu_stats *pcpu_stats;
 376
 377                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 378                 u64_stats_update_begin(&pcpu_stats->syncp);
 379                 pcpu_stats->tx_packets++;
 380                 pcpu_stats->tx_bytes += len;
 381                 u64_stats_update_end(&pcpu_stats->syncp);
 382         } else {
 383                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 384         }
 385         return ret;
 386 }
 387
 388 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 389                                unsigned short type, const void *daddr,
 390                                const void *saddr, unsigned len)
 391 {
 392         const struct macvlan_dev *vlan = netdev_priv(dev);
 393         struct net_device *lowerdev = vlan->lowerdev;
 394
 395         return dev_hard_header(skb, lowerdev, type, daddr,
 396                                saddr ? : dev->dev_addr, len);
 397 }
 398
 399 static const struct header_ops macvlan_hard_header_ops = {
 400         .create         = macvlan_hard_header,
 401         .rebuild        = eth_rebuild_header,
 402         .parse          = eth_header_parse,
 403         .cache          = eth_header_cache,
 404         .cache_update   = eth_header_cache_update,
 405 };
 406
 407 static struct rtnl_link_ops macvlan_link_ops;
 408
 409 static int macvlan_open(struct net_device *dev)
 410 {
 411         struct macvlan_dev *vlan = netdev_priv(dev);
 412         struct net_device *lowerdev = vlan->lowerdev;
 413         int err;
 414
 415         if (vlan->port->passthru) {
 416                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 417                         err = dev_set_promiscuity(lowerdev, 1);
 418                         if (err < 0)
 419                                 goto out;
 420                 }
 421                 goto hash_add;
 422         }
 423
 424         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
 425             dev->rtnl_link_ops == &macvlan_link_ops) {
 426                 vlan->fwd_priv =
 427                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 428
 429                 /* If we get a NULL pointer back, or if we get an error
 430                  * then we should just fall through to the non accelerated path
 431                  */
 432                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
 433                         vlan->fwd_priv = NULL;
 434                 } else
 435                         return 0;
 436         }
 437
 438         err = -EBUSY;
 439         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 440                 goto out;
 441
 442         err = dev_uc_add(lowerdev, dev->dev_addr);
 443         if (err < 0)
 444                 goto out;
 445         if (dev->flags & IFF_ALLMULTI) {
 446                 err = dev_set_allmulti(lowerdev, 1);
 447                 if (err < 0)
 448                         goto del_unicast;
 449         }
 450
 451 hash_add:
 452         macvlan_hash_add(vlan);
 453         return 0;
 454
 455 del_unicast:
 456         dev_uc_del(lowerdev, dev->dev_addr);
 457 out:
 458         if (vlan->fwd_priv) {
 459                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 460                                                            vlan->fwd_priv);
 461                 vlan->fwd_priv = NULL;
 462         }
 463         return err;
 464 }
 465
 466 static int macvlan_stop(struct net_device *dev)
 467 {
 468         struct macvlan_dev *vlan = netdev_priv(dev);
 469         struct net_device *lowerdev = vlan->lowerdev;
 470
 471         if (vlan->fwd_priv) {
 472                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 473                                                            vlan->fwd_priv);
 474                 vlan->fwd_priv = NULL;
 475                 return 0;
 476         }
 477
 478         dev_uc_unsync(lowerdev, dev);
 479         dev_mc_unsync(lowerdev, dev);
 480
 481         if (vlan->port->passthru) {
 482                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 483                         dev_set_promiscuity(lowerdev, -1);
 484                 goto hash_del;
 485         }
 486
 487         if (dev->flags & IFF_ALLMULTI)
 488                 dev_set_allmulti(lowerdev, -1);
 489
 490         dev_uc_del(lowerdev, dev->dev_addr);
 491
 492 hash_del:
 493         macvlan_hash_del(vlan, !dev->dismantle);
 494         return 0;
 495 }
 496
 497 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 498 {
 499         struct macvlan_dev *vlan = netdev_priv(dev);
 500         struct net_device *lowerdev = vlan->lowerdev;
 501         int err;
 502
 503         if (!(dev->flags & IFF_UP)) {
 504                 /* Just copy in the new address */
 505                 ether_addr_copy(dev->dev_addr, addr);
 506         } else {
 507                 /* Rehash and update the device filters */
 508                 if (macvlan_addr_busy(vlan->port, addr))
 509                         return -EBUSY;
 510
 511                 if (!vlan->port->passthru) {
 512                         err = dev_uc_add(lowerdev, addr);
 513                         if (err)
 514                                 return err;
 515
 516                         dev_uc_del(lowerdev, dev->dev_addr);
 517                 }
 518
 519                 macvlan_hash_change_addr(vlan, addr);
 520         }
 521         return 0;
 522 }
 523
 524 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 525 {
 526         struct macvlan_dev *vlan = netdev_priv(dev);
 527         struct sockaddr *addr = p;
 528
 529         if (!is_valid_ether_addr(addr->sa_data))
 530                 return -EADDRNOTAVAIL;
 531
 532         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 533                 dev_set_mac_address(vlan->lowerdev, addr);
 534                 return 0;
 535         }
 536
 537         return macvlan_sync_address(dev, addr->sa_data);
 538 }
 539
 540 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 541 {
 542         struct macvlan_dev *vlan = netdev_priv(dev);
 543         struct net_device *lowerdev = vlan->lowerdev;
 544
 545         if (dev->flags & IFF_UP) {
 546                 if (change & IFF_ALLMULTI)
 547                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 548         }
 549 }
 550
 551 static void macvlan_set_mac_lists(struct net_device *dev)
 552 {
 553         struct macvlan_dev *vlan = netdev_priv(dev);
 554
 555         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 556                 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
 557         } else {
 558                 struct netdev_hw_addr *ha;
 559                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 560
 561                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 562                 netdev_for_each_mc_addr(ha, dev) {
 563                         __set_bit(mc_hash(vlan, ha->addr), filter);
 564                 }
 565
 566                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 567
 568                 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 569         }
 570         dev_uc_sync(vlan->lowerdev, dev);
 571         dev_mc_sync(vlan->lowerdev, dev);
 572 }
 573
 574 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 575 {
 576         struct macvlan_dev *vlan = netdev_priv(dev);
 577
 578         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 579                 return -EINVAL;
 580         dev->mtu = new_mtu;
 581         return 0;
 582 }
 583
 584 /*
 585  * macvlan network devices have devices nesting below it and are a special
 586  * "super class" of normal network devices; split their locks off into a
 587  * separate class since they always nest.
 588  */
 589 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 590 static struct lock_class_key macvlan_netdev_addr_lock_key;
 591
 592 #define ALWAYS_ON_FEATURES \
 593         (NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX)
 594
 595 #define MACVLAN_FEATURES \
 596         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 597          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 598          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 599          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 600
 601 #define MACVLAN_STATE_MASK \
 602         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 603
 604 static int macvlan_get_nest_level(struct net_device *dev)
 605 {
 606         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 607 }
 608
 609 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 610                                           struct netdev_queue *txq,
 611                                           void *_unused)
 612 {
 613         lockdep_set_class(&txq->_xmit_lock,
 614                           &macvlan_netdev_xmit_lock_key);
 615 }
 616
 617 static void macvlan_set_lockdep_class(struct net_device *dev)
 618 {
 619         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 620                                        &macvlan_netdev_addr_lock_key,
 621                                        macvlan_get_nest_level(dev));
 622         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 623 }
 624
 625 static int macvlan_init(struct net_device *dev)
 626 {
 627         struct macvlan_dev *vlan = netdev_priv(dev);
 628         const struct net_device *lowerdev = vlan->lowerdev;
 629
 630         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 631                                   (lowerdev->state & MACVLAN_STATE_MASK);
 632         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 633         dev->features           |= ALWAYS_ON_FEATURES;
 634         dev->gso_max_size       = lowerdev->gso_max_size;
 635         dev->iflink             = lowerdev->ifindex;
 636         dev->hard_header_len    = lowerdev->hard_header_len;
 637
 638         macvlan_set_lockdep_class(dev);
 639
 640         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 641         if (!vlan->pcpu_stats)
 642                 return -ENOMEM;
 643
 644         return 0;
 645 }
 646
 647 static void macvlan_uninit(struct net_device *dev)
 648 {
 649         struct macvlan_dev *vlan = netdev_priv(dev);
 650         struct macvlan_port *port = vlan->port;
 651
 652         free_percpu(vlan->pcpu_stats);
 653
 654         if (MACVLAN_PORT_IS_EMPTY(port))
 655                 macvlan_port_destroy(port->dev);
 656 }
 657
 658 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 659                                                          struct rtnl_link_stats64 *stats)
 660 {
 661         struct macvlan_dev *vlan = netdev_priv(dev);
 662
 663         if (vlan->pcpu_stats) {
 664                 struct vlan_pcpu_stats *p;
 665                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 666                 u32 rx_errors = 0, tx_dropped = 0;
 667                 unsigned int start;
 668                 int i;
 669
 670                 for_each_possible_cpu(i) {
 671                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 672                         do {
 673                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 674                                 rx_packets      = p->rx_packets;
 675                                 rx_bytes        = p->rx_bytes;
 676                                 rx_multicast    = p->rx_multicast;
 677                                 tx_packets      = p->tx_packets;
 678                                 tx_bytes        = p->tx_bytes;
 679                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 680
 681                         stats->rx_packets       += rx_packets;
 682                         stats->rx_bytes         += rx_bytes;
 683                         stats->multicast        += rx_multicast;
 684                         stats->tx_packets       += tx_packets;
 685                         stats->tx_bytes         += tx_bytes;
 686                         /* rx_errors & tx_dropped are u32, updated
 687                          * without syncp protection.
 688                          */
 689                         rx_errors       += p->rx_errors;
 690                         tx_dropped      += p->tx_dropped;
 691                 }
 692                 stats->rx_errors        = rx_errors;
 693                 stats->rx_dropped       = rx_errors;
 694                 stats->tx_dropped       = tx_dropped;
 695         }
 696         return stats;
 697 }
 698
 699 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 700                                    __be16 proto, u16 vid)
 701 {
 702         struct macvlan_dev *vlan = netdev_priv(dev);
 703         struct net_device *lowerdev = vlan->lowerdev;
 704
 705         return vlan_vid_add(lowerdev, proto, vid);
 706 }
 707
 708 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 709                                     __be16 proto, u16 vid)
 710 {
 711         struct macvlan_dev *vlan = netdev_priv(dev);
 712         struct net_device *lowerdev = vlan->lowerdev;
 713
 714         vlan_vid_del(lowerdev, proto, vid);
 715         return 0;
 716 }
 717
 718 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 719                            struct net_device *dev,
 720                            const unsigned char *addr,
 721                            u16 flags)
 722 {
 723         struct macvlan_dev *vlan = netdev_priv(dev);
 724         int err = -EINVAL;
 725
 726         if (!vlan->port->passthru)
 727                 return -EOPNOTSUPP;
 728
 729         if (flags & NLM_F_REPLACE)
 730                 return -EOPNOTSUPP;
 731
 732         if (is_unicast_ether_addr(addr))
 733                 err = dev_uc_add_excl(dev, addr);
 734         else if (is_multicast_ether_addr(addr))
 735                 err = dev_mc_add_excl(dev, addr);
 736
 737         return err;
 738 }
 739
 740 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 741                            struct net_device *dev,
 742                            const unsigned char *addr)
 743 {
 744         struct macvlan_dev *vlan = netdev_priv(dev);
 745         int err = -EINVAL;
 746
 747         if (!vlan->port->passthru)
 748                 return -EOPNOTSUPP;
 749
 750         if (is_unicast_ether_addr(addr))
 751                 err = dev_uc_del(dev, addr);
 752         else if (is_multicast_ether_addr(addr))
 753                 err = dev_mc_del(dev, addr);
 754
 755         return err;
 756 }
 757
 758 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 759                                         struct ethtool_drvinfo *drvinfo)
 760 {
 761         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 762         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 763 }
 764
 765 static int macvlan_ethtool_get_settings(struct net_device *dev,
 766                                         struct ethtool_cmd *cmd)
 767 {
 768         const struct macvlan_dev *vlan = netdev_priv(dev);
 769
 770         return __ethtool_get_settings(vlan->lowerdev, cmd);
 771 }
 772
 773 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 774                                               netdev_features_t features)
 775 {
 776         struct macvlan_dev *vlan = netdev_priv(dev);
 777         netdev_features_t mask;
 778
 779         features |= NETIF_F_ALL_FOR_ALL;
 780         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 781         mask = features;
 782
 783         features = netdev_increment_features(vlan->lowerdev->features,
 784                                              features,
 785                                              mask);
 786         features |= ALWAYS_ON_FEATURES;
 787
 788         return features;
 789 }
 790
 791 static const struct ethtool_ops macvlan_ethtool_ops = {
 792         .get_link               = ethtool_op_get_link,
 793         .get_settings           = macvlan_ethtool_get_settings,
 794         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 795 };
 796
 797 static const struct net_device_ops macvlan_netdev_ops = {
 798         .ndo_init               = macvlan_init,
 799         .ndo_uninit             = macvlan_uninit,
 800         .ndo_open               = macvlan_open,
 801         .ndo_stop               = macvlan_stop,
 802         .ndo_start_xmit         = macvlan_start_xmit,
 803         .ndo_change_mtu         = macvlan_change_mtu,
 804         .ndo_fix_features       = macvlan_fix_features,
 805         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 806         .ndo_set_mac_address    = macvlan_set_mac_address,
 807         .ndo_set_rx_mode        = macvlan_set_mac_lists,
 808         .ndo_get_stats64        = macvlan_dev_get_stats64,
 809         .ndo_validate_addr      = eth_validate_addr,
 810         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
 811         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
 812         .ndo_fdb_add            = macvlan_fdb_add,
 813         .ndo_fdb_del            = macvlan_fdb_del,
 814         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
 815         .ndo_get_lock_subclass  = macvlan_get_nest_level,
 816 };
 817
 818 void macvlan_common_setup(struct net_device *dev)
 819 {
 820         ether_setup(dev);
 821
 822         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 823         dev->priv_flags        |= IFF_UNICAST_FLT;
 824         dev->netdev_ops         = &macvlan_netdev_ops;
 825         dev->destructor         = free_netdev;
 826         dev->header_ops         = &macvlan_hard_header_ops;
 827         dev->ethtool_ops        = &macvlan_ethtool_ops;
 828 }
 829 EXPORT_SYMBOL_GPL(macvlan_common_setup);
 830
 831 static void macvlan_setup(struct net_device *dev)
 832 {
 833         macvlan_common_setup(dev);
 834         dev->tx_queue_len       = 0;
 835 }
 836
 837 static int macvlan_port_create(struct net_device *dev)
 838 {
 839         struct macvlan_port *port;
 840         unsigned int i;
 841         int err;
 842
 843         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 844                 return -EINVAL;
 845
 846         port = kzalloc(sizeof(*port), GFP_KERNEL);
 847         if (port == NULL)
 848                 return -ENOMEM;
 849
 850         port->passthru = false;
 851         port->dev = dev;
 852         INIT_LIST_HEAD(&port->vlans);
 853         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 854                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 855
 856         skb_queue_head_init(&port->bc_queue);
 857         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
 858
 859         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
 860         if (err)
 861                 kfree(port);
 862         else
 863                 dev->priv_flags |= IFF_MACVLAN_PORT;
 864         return err;
 865 }
 866
 867 static void macvlan_port_destroy(struct net_device *dev)
 868 {
 869         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
 870
 871         cancel_work_sync(&port->bc_work);
 872         dev->priv_flags &= ~IFF_MACVLAN_PORT;
 873         netdev_rx_handler_unregister(dev);
 874         kfree_rcu(port, rcu);
 875 }
 876
 877 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 878 {
 879         if (tb[IFLA_ADDRESS]) {
 880                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 881                         return -EINVAL;
 882                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 883                         return -EADDRNOTAVAIL;
 884         }
 885
 886         if (data && data[IFLA_MACVLAN_FLAGS] &&
 887             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
 888                 return -EINVAL;
 889
 890         if (data && data[IFLA_MACVLAN_MODE]) {
 891                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
 892                 case MACVLAN_MODE_PRIVATE:
 893                 case MACVLAN_MODE_VEPA:
 894                 case MACVLAN_MODE_BRIDGE:
 895                 case MACVLAN_MODE_PASSTHRU:
 896                         break;
 897                 default:
 898                         return -EINVAL;
 899                 }
 900         }
 901         return 0;
 902 }
 903
 904 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
 905                            struct nlattr *tb[], struct nlattr *data[])
 906 {
 907         struct macvlan_dev *vlan = netdev_priv(dev);
 908         struct macvlan_port *port;
 909         struct net_device *lowerdev;
 910         int err;
 911
 912         if (!tb[IFLA_LINK])
 913                 return -EINVAL;
 914
 915         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 916         if (lowerdev == NULL)
 917                 return -ENODEV;
 918
 919         /* When creating macvlans or macvtaps on top of other macvlans - use
 920          * the real device as the lowerdev.
 921          */
 922         if (netif_is_macvlan(lowerdev))
 923                 lowerdev = macvlan_dev_real_dev(lowerdev);
 924
 925         if (!tb[IFLA_MTU])
 926                 dev->mtu = lowerdev->mtu;
 927         else if (dev->mtu > lowerdev->mtu)
 928                 return -EINVAL;
 929
 930         if (!tb[IFLA_ADDRESS])
 931                 eth_hw_addr_random(dev);
 932
 933         if (!macvlan_port_exists(lowerdev)) {
 934                 err = macvlan_port_create(lowerdev);
 935                 if (err < 0)
 936                         return err;
 937         }
 938         port = macvlan_port_get_rtnl(lowerdev);
 939
 940         /* Only 1 macvlan device can be created in passthru mode */
 941         if (port->passthru)
 942                 return -EINVAL;
 943
 944         vlan->lowerdev = lowerdev;
 945         vlan->dev      = dev;
 946         vlan->port     = port;
 947         vlan->set_features = MACVLAN_FEATURES;
 948         vlan->nest_level = dev_get_nest_level(lowerdev, netif_is_macvlan) + 1;
 949
 950         vlan->mode     = MACVLAN_MODE_VEPA;
 951         if (data && data[IFLA_MACVLAN_MODE])
 952                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 953
 954         if (data && data[IFLA_MACVLAN_FLAGS])
 955                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
 956
 957         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 958                 if (!MACVLAN_PORT_IS_EMPTY(port))
 959                         return -EINVAL;
 960                 port->passthru = true;
 961                 eth_hw_addr_inherit(dev, lowerdev);
 962         }
 963
 964         err = register_netdevice(dev);
 965         if (err < 0)
 966                 goto destroy_port;
 967
 968         dev->priv_flags |= IFF_MACVLAN;
 969         err = netdev_upper_dev_link(lowerdev, dev);
 970         if (err)
 971                 goto unregister_netdev;
 972
 973         list_add_tail_rcu(&vlan->list, &port->vlans);
 974         netif_stacked_transfer_operstate(lowerdev, dev);
 975
 976         return 0;
 977
 978 unregister_netdev:
 979         unregister_netdevice(dev);
 980 destroy_port:
 981         if (MACVLAN_PORT_IS_EMPTY(port))
 982                 macvlan_port_destroy(lowerdev);
 983
 984         return err;
 985 }
 986 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
 987
 988 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
 989                            struct nlattr *tb[], struct nlattr *data[])
 990 {
 991         return macvlan_common_newlink(src_net, dev, tb, data);
 992 }
 993
 994 void macvlan_dellink(struct net_device *dev, struct list_head *head)
 995 {
 996         struct macvlan_dev *vlan = netdev_priv(dev);
 997
 998         list_del_rcu(&vlan->list);
 999         unregister_netdevice_queue(dev, head);
1000         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1001 }
1002 EXPORT_SYMBOL_GPL(macvlan_dellink);
1003
1004 static int macvlan_changelink(struct net_device *dev,
1005                 struct nlattr *tb[], struct nlattr *data[])
1006 {
1007         struct macvlan_dev *vlan = netdev_priv(dev);
1008         enum macvlan_mode mode;
1009         bool set_mode = false;
1010
1011         /* Validate mode, but don't set yet: setting flags may fail. */
1012         if (data && data[IFLA_MACVLAN_MODE]) {
1013                 set_mode = true;
1014                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1015                 /* Passthrough mode can't be set or cleared dynamically */
1016                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1017                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1018                         return -EINVAL;
1019         }
1020
1021         if (data && data[IFLA_MACVLAN_FLAGS]) {
1022                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1023                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1024                 if (vlan->port->passthru && promisc) {
1025                         int err;
1026
1027                         if (flags & MACVLAN_FLAG_NOPROMISC)
1028                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1029                         else
1030                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1031                         if (err < 0)
1032                                 return err;
1033                 }
1034                 vlan->flags = flags;
1035         }
1036         if (set_mode)
1037                 vlan->mode = mode;
1038         return 0;
1039 }
1040
1041 static size_t macvlan_get_size(const struct net_device *dev)
1042 {
1043         return (0
1044                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1045                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1046                 );
1047 }
1048
1049 static int macvlan_fill_info(struct sk_buff *skb,
1050                                 const struct net_device *dev)
1051 {
1052         struct macvlan_dev *vlan = netdev_priv(dev);
1053
1054         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1055                 goto nla_put_failure;
1056         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1057                 goto nla_put_failure;
1058         return 0;
1059
1060 nla_put_failure:
1061         return -EMSGSIZE;
1062 }
1063
1064 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1065         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1066         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1067 };
1068
1069 int macvlan_link_register(struct rtnl_link_ops *ops)
1070 {
1071         /* common fields */
1072         ops->priv_size          = sizeof(struct macvlan_dev);
1073         ops->validate           = macvlan_validate;
1074         ops->maxtype            = IFLA_MACVLAN_MAX;
1075         ops->policy             = macvlan_policy;
1076         ops->changelink         = macvlan_changelink;
1077         ops->get_size           = macvlan_get_size;
1078         ops->fill_info          = macvlan_fill_info;
1079
1080         return rtnl_link_register(ops);
1081 };
1082 EXPORT_SYMBOL_GPL(macvlan_link_register);
1083
1084 static struct rtnl_link_ops macvlan_link_ops = {
1085         .kind           = "macvlan",
1086         .setup          = macvlan_setup,
1087         .newlink        = macvlan_newlink,
1088         .dellink        = macvlan_dellink,
1089 };
1090
1091 static int macvlan_device_event(struct notifier_block *unused,
1092                                 unsigned long event, void *ptr)
1093 {
1094         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1095         struct macvlan_dev *vlan, *next;
1096         struct macvlan_port *port;
1097         LIST_HEAD(list_kill);
1098
1099         if (!macvlan_port_exists(dev))
1100                 return NOTIFY_DONE;
1101
1102         port = macvlan_port_get_rtnl(dev);
1103
1104         switch (event) {
1105         case NETDEV_CHANGE:
1106                 list_for_each_entry(vlan, &port->vlans, list)
1107                         netif_stacked_transfer_operstate(vlan->lowerdev,
1108                                                          vlan->dev);
1109                 break;
1110         case NETDEV_FEAT_CHANGE:
1111                 list_for_each_entry(vlan, &port->vlans, list) {
1112                         vlan->dev->gso_max_size = dev->gso_max_size;
1113                         netdev_update_features(vlan->dev);
1114                 }
1115                 break;
1116         case NETDEV_CHANGEMTU:
1117                 list_for_each_entry(vlan, &port->vlans, list) {
1118                         if (vlan->dev->mtu <= dev->mtu)
1119                                 continue;
1120                         dev_set_mtu(vlan->dev, dev->mtu);
1121                 }
1122                 break;
1123         case NETDEV_CHANGEADDR:
1124                 if (!port->passthru)
1125                         return NOTIFY_DONE;
1126
1127                 vlan = list_first_entry_or_null(&port->vlans,
1128                                                 struct macvlan_dev,
1129                                                 list);
1130
1131                 if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1132                         return NOTIFY_BAD;
1133
1134                 break;
1135         case NETDEV_UNREGISTER:
1136                 /* twiddle thumbs on netns device moves */
1137                 if (dev->reg_state != NETREG_UNREGISTERING)
1138                         break;
1139
1140                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1141                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1142                 unregister_netdevice_many(&list_kill);
1143                 list_del(&list_kill);
1144                 break;
1145         case NETDEV_PRE_TYPE_CHANGE:
1146                 /* Forbid underlaying device to change its type. */
1147                 return NOTIFY_BAD;
1148         }
1149         return NOTIFY_DONE;
1150 }
1151
1152 static struct notifier_block macvlan_notifier_block __read_mostly = {
1153         .notifier_call  = macvlan_device_event,
1154 };
1155
1156 static int __init macvlan_init_module(void)
1157 {
1158         int err;
1159
1160         register_netdevice_notifier(&macvlan_notifier_block);
1161
1162         err = macvlan_link_register(&macvlan_link_ops);
1163         if (err < 0)
1164                 goto err1;
1165         return 0;
1166 err1:
1167         unregister_netdevice_notifier(&macvlan_notifier_block);
1168         return err;
1169 }
1170
1171 static void __exit macvlan_cleanup_module(void)
1172 {
1173         rtnl_link_unregister(&macvlan_link_ops);
1174         unregister_netdevice_notifier(&macvlan_notifier_block);
1175 }
1176
1177 module_init(macvlan_init_module);
1178 module_exit(macvlan_cleanup_module);
1179
1180 MODULE_LICENSE("GPL");
1181 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1182 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1183 MODULE_ALIAS_RTNL_LINK("macvlan");