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_set_mac_address(struct net_device *dev, void *p)
 498 {
 499         struct macvlan_dev *vlan = netdev_priv(dev);
 500         struct net_device *lowerdev = vlan->lowerdev;
 501         struct sockaddr *addr = p;
 502         int err;
 503
 504         if (!is_valid_ether_addr(addr->sa_data))
 505                 return -EADDRNOTAVAIL;
 506
 507         if (!(dev->flags & IFF_UP)) {
 508                 /* Just copy in the new address */
 509                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 510         } else {
 511                 /* Rehash and update the device filters */
 512                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
 513                         return -EBUSY;
 514
 515                 err = dev_uc_add(lowerdev, addr->sa_data);
 516                 if (err)
 517                         return err;
 518
 519                 dev_uc_del(lowerdev, dev->dev_addr);
 520
 521                 macvlan_hash_change_addr(vlan, addr->sa_data);
 522         }
 523         return 0;
 524 }
 525
 526 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 527 {
 528         struct macvlan_dev *vlan = netdev_priv(dev);
 529         struct net_device *lowerdev = vlan->lowerdev;
 530
 531         if (dev->flags & IFF_UP) {
 532                 if (change & IFF_ALLMULTI)
 533                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 534         }
 535 }
 536
 537 static void macvlan_set_mac_lists(struct net_device *dev)
 538 {
 539         struct macvlan_dev *vlan = netdev_priv(dev);
 540
 541         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 542                 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
 543         } else {
 544                 struct netdev_hw_addr *ha;
 545                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 546
 547                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 548                 netdev_for_each_mc_addr(ha, dev) {
 549                         __set_bit(mc_hash(vlan, ha->addr), filter);
 550                 }
 551
 552                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 553
 554                 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 555         }
 556         dev_uc_sync(vlan->lowerdev, dev);
 557         dev_mc_sync(vlan->lowerdev, dev);
 558 }
 559
 560 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 561 {
 562         struct macvlan_dev *vlan = netdev_priv(dev);
 563
 564         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 565                 return -EINVAL;
 566         dev->mtu = new_mtu;
 567         return 0;
 568 }
 569
 570 /*
 571  * macvlan network devices have devices nesting below it and are a special
 572  * "super class" of normal network devices; split their locks off into a
 573  * separate class since they always nest.
 574  */
 575 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 576 static struct lock_class_key macvlan_netdev_addr_lock_key;
 577
 578 #define ALWAYS_ON_FEATURES \
 579         (NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX)
 580
 581 #define MACVLAN_FEATURES \
 582         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 583          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 584          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 585          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 586
 587 #define MACVLAN_STATE_MASK \
 588         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 589
 590 static int macvlan_get_nest_level(struct net_device *dev)
 591 {
 592         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 593 }
 594
 595 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 596                                           struct netdev_queue *txq,
 597                                           void *_unused)
 598 {
 599         lockdep_set_class(&txq->_xmit_lock,
 600                           &macvlan_netdev_xmit_lock_key);
 601 }
 602
 603 static void macvlan_set_lockdep_class(struct net_device *dev)
 604 {
 605         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 606                                        &macvlan_netdev_addr_lock_key,
 607                                        macvlan_get_nest_level(dev));
 608         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 609 }
 610
 611 static int macvlan_init(struct net_device *dev)
 612 {
 613         struct macvlan_dev *vlan = netdev_priv(dev);
 614         const struct net_device *lowerdev = vlan->lowerdev;
 615
 616         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 617                                   (lowerdev->state & MACVLAN_STATE_MASK);
 618         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 619         dev->features           |= ALWAYS_ON_FEATURES;
 620         dev->gso_max_size       = lowerdev->gso_max_size;
 621         dev->iflink             = lowerdev->ifindex;
 622         dev->hard_header_len    = lowerdev->hard_header_len;
 623
 624         macvlan_set_lockdep_class(dev);
 625
 626         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 627         if (!vlan->pcpu_stats)
 628                 return -ENOMEM;
 629
 630         return 0;
 631 }
 632
 633 static void macvlan_uninit(struct net_device *dev)
 634 {
 635         struct macvlan_dev *vlan = netdev_priv(dev);
 636         struct macvlan_port *port = vlan->port;
 637
 638         free_percpu(vlan->pcpu_stats);
 639
 640         if (MACVLAN_PORT_IS_EMPTY(port))
 641                 macvlan_port_destroy(port->dev);
 642 }
 643
 644 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
 645                                                          struct rtnl_link_stats64 *stats)
 646 {
 647         struct macvlan_dev *vlan = netdev_priv(dev);
 648
 649         if (vlan->pcpu_stats) {
 650                 struct vlan_pcpu_stats *p;
 651                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 652                 u32 rx_errors = 0, tx_dropped = 0;
 653                 unsigned int start;
 654                 int i;
 655
 656                 for_each_possible_cpu(i) {
 657                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 658                         do {
 659                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 660                                 rx_packets      = p->rx_packets;
 661                                 rx_bytes        = p->rx_bytes;
 662                                 rx_multicast    = p->rx_multicast;
 663                                 tx_packets      = p->tx_packets;
 664                                 tx_bytes        = p->tx_bytes;
 665                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 666
 667                         stats->rx_packets       += rx_packets;
 668                         stats->rx_bytes         += rx_bytes;
 669                         stats->multicast        += rx_multicast;
 670                         stats->tx_packets       += tx_packets;
 671                         stats->tx_bytes         += tx_bytes;
 672                         /* rx_errors & tx_dropped are u32, updated
 673                          * without syncp protection.
 674                          */
 675                         rx_errors       += p->rx_errors;
 676                         tx_dropped      += p->tx_dropped;
 677                 }
 678                 stats->rx_errors        = rx_errors;
 679                 stats->rx_dropped       = rx_errors;
 680                 stats->tx_dropped       = tx_dropped;
 681         }
 682         return stats;
 683 }
 684
 685 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 686                                    __be16 proto, u16 vid)
 687 {
 688         struct macvlan_dev *vlan = netdev_priv(dev);
 689         struct net_device *lowerdev = vlan->lowerdev;
 690
 691         return vlan_vid_add(lowerdev, proto, vid);
 692 }
 693
 694 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 695                                     __be16 proto, u16 vid)
 696 {
 697         struct macvlan_dev *vlan = netdev_priv(dev);
 698         struct net_device *lowerdev = vlan->lowerdev;
 699
 700         vlan_vid_del(lowerdev, proto, vid);
 701         return 0;
 702 }
 703
 704 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 705                            struct net_device *dev,
 706                            const unsigned char *addr,
 707                            u16 flags)
 708 {
 709         struct macvlan_dev *vlan = netdev_priv(dev);
 710         int err = -EINVAL;
 711
 712         if (!vlan->port->passthru)
 713                 return -EOPNOTSUPP;
 714
 715         if (flags & NLM_F_REPLACE)
 716                 return -EOPNOTSUPP;
 717
 718         if (is_unicast_ether_addr(addr))
 719                 err = dev_uc_add_excl(dev, addr);
 720         else if (is_multicast_ether_addr(addr))
 721                 err = dev_mc_add_excl(dev, addr);
 722
 723         return err;
 724 }
 725
 726 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
 727                            struct net_device *dev,
 728                            const unsigned char *addr)
 729 {
 730         struct macvlan_dev *vlan = netdev_priv(dev);
 731         int err = -EINVAL;
 732
 733         if (!vlan->port->passthru)
 734                 return -EOPNOTSUPP;
 735
 736         if (is_unicast_ether_addr(addr))
 737                 err = dev_uc_del(dev, addr);
 738         else if (is_multicast_ether_addr(addr))
 739                 err = dev_mc_del(dev, addr);
 740
 741         return err;
 742 }
 743
 744 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 745                                         struct ethtool_drvinfo *drvinfo)
 746 {
 747         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
 748         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
 749 }
 750
 751 static int macvlan_ethtool_get_settings(struct net_device *dev,
 752                                         struct ethtool_cmd *cmd)
 753 {
 754         const struct macvlan_dev *vlan = netdev_priv(dev);
 755
 756         return __ethtool_get_settings(vlan->lowerdev, cmd);
 757 }
 758
 759 static netdev_features_t macvlan_fix_features(struct net_device *dev,
 760                                               netdev_features_t features)
 761 {
 762         struct macvlan_dev *vlan = netdev_priv(dev);
 763         netdev_features_t mask;
 764
 765         features |= NETIF_F_ALL_FOR_ALL;
 766         features &= (vlan->set_features | ~MACVLAN_FEATURES);
 767         mask = features;
 768
 769         features = netdev_increment_features(vlan->lowerdev->features,
 770                                              features,
 771                                              mask);
 772         features |= ALWAYS_ON_FEATURES;
 773
 774         return features;
 775 }
 776
 777 static const struct ethtool_ops macvlan_ethtool_ops = {
 778         .get_link               = ethtool_op_get_link,
 779         .get_settings           = macvlan_ethtool_get_settings,
 780         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 781 };
 782
 783 static const struct net_device_ops macvlan_netdev_ops = {
 784         .ndo_init               = macvlan_init,
 785         .ndo_uninit             = macvlan_uninit,
 786         .ndo_open               = macvlan_open,
 787         .ndo_stop               = macvlan_stop,
 788         .ndo_start_xmit         = macvlan_start_xmit,
 789         .ndo_change_mtu         = macvlan_change_mtu,
 790         .ndo_fix_features       = macvlan_fix_features,
 791         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 792         .ndo_set_mac_address    = macvlan_set_mac_address,
 793         .ndo_set_rx_mode        = macvlan_set_mac_lists,
 794         .ndo_get_stats64        = macvlan_dev_get_stats64,
 795         .ndo_validate_addr      = eth_validate_addr,
 796         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
 797         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
 798         .ndo_fdb_add            = macvlan_fdb_add,
 799         .ndo_fdb_del            = macvlan_fdb_del,
 800         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
 801         .ndo_get_lock_subclass  = macvlan_get_nest_level,
 802 };
 803
 804 void macvlan_common_setup(struct net_device *dev)
 805 {
 806         ether_setup(dev);
 807
 808         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 809         dev->priv_flags        |= IFF_UNICAST_FLT;
 810         dev->netdev_ops         = &macvlan_netdev_ops;
 811         dev->destructor         = free_netdev;
 812         dev->header_ops         = &macvlan_hard_header_ops;
 813         dev->ethtool_ops        = &macvlan_ethtool_ops;
 814 }
 815 EXPORT_SYMBOL_GPL(macvlan_common_setup);
 816
 817 static void macvlan_setup(struct net_device *dev)
 818 {
 819         macvlan_common_setup(dev);
 820         dev->tx_queue_len       = 0;
 821 }
 822
 823 static int macvlan_port_create(struct net_device *dev)
 824 {
 825         struct macvlan_port *port;
 826         unsigned int i;
 827         int err;
 828
 829         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 830                 return -EINVAL;
 831
 832         port = kzalloc(sizeof(*port), GFP_KERNEL);
 833         if (port == NULL)
 834                 return -ENOMEM;
 835
 836         port->passthru = false;
 837         port->dev = dev;
 838         INIT_LIST_HEAD(&port->vlans);
 839         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 840                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 841
 842         skb_queue_head_init(&port->bc_queue);
 843         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
 844
 845         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
 846         if (err)
 847                 kfree(port);
 848         else
 849                 dev->priv_flags |= IFF_MACVLAN_PORT;
 850         return err;
 851 }
 852
 853 static void macvlan_port_destroy(struct net_device *dev)
 854 {
 855         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
 856
 857         cancel_work_sync(&port->bc_work);
 858         dev->priv_flags &= ~IFF_MACVLAN_PORT;
 859         netdev_rx_handler_unregister(dev);
 860         kfree_rcu(port, rcu);
 861 }
 862
 863 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 864 {
 865         if (tb[IFLA_ADDRESS]) {
 866                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 867                         return -EINVAL;
 868                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 869                         return -EADDRNOTAVAIL;
 870         }
 871
 872         if (data && data[IFLA_MACVLAN_FLAGS] &&
 873             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
 874                 return -EINVAL;
 875
 876         if (data && data[IFLA_MACVLAN_MODE]) {
 877                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
 878                 case MACVLAN_MODE_PRIVATE:
 879                 case MACVLAN_MODE_VEPA:
 880                 case MACVLAN_MODE_BRIDGE:
 881                 case MACVLAN_MODE_PASSTHRU:
 882                         break;
 883                 default:
 884                         return -EINVAL;
 885                 }
 886         }
 887         return 0;
 888 }
 889
 890 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
 891                            struct nlattr *tb[], struct nlattr *data[])
 892 {
 893         struct macvlan_dev *vlan = netdev_priv(dev);
 894         struct macvlan_port *port;
 895         struct net_device *lowerdev;
 896         int err;
 897
 898         if (!tb[IFLA_LINK])
 899                 return -EINVAL;
 900
 901         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
 902         if (lowerdev == NULL)
 903                 return -ENODEV;
 904
 905         /* When creating macvlans or macvtaps on top of other macvlans - use
 906          * the real device as the lowerdev.
 907          */
 908         if (netif_is_macvlan(lowerdev))
 909                 lowerdev = macvlan_dev_real_dev(lowerdev);
 910
 911         if (!tb[IFLA_MTU])
 912                 dev->mtu = lowerdev->mtu;
 913         else if (dev->mtu > lowerdev->mtu)
 914                 return -EINVAL;
 915
 916         if (!tb[IFLA_ADDRESS])
 917                 eth_hw_addr_random(dev);
 918
 919         if (!macvlan_port_exists(lowerdev)) {
 920                 err = macvlan_port_create(lowerdev);
 921                 if (err < 0)
 922                         return err;
 923         }
 924         port = macvlan_port_get_rtnl(lowerdev);
 925
 926         /* Only 1 macvlan device can be created in passthru mode */
 927         if (port->passthru)
 928                 return -EINVAL;
 929
 930         vlan->lowerdev = lowerdev;
 931         vlan->dev      = dev;
 932         vlan->port     = port;
 933         vlan->set_features = MACVLAN_FEATURES;
 934         vlan->nest_level = dev_get_nest_level(lowerdev, netif_is_macvlan) + 1;
 935
 936         vlan->mode     = MACVLAN_MODE_VEPA;
 937         if (data && data[IFLA_MACVLAN_MODE])
 938                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
 939
 940         if (data && data[IFLA_MACVLAN_FLAGS])
 941                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
 942
 943         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 944                 if (!MACVLAN_PORT_IS_EMPTY(port))
 945                         return -EINVAL;
 946                 port->passthru = true;
 947                 eth_hw_addr_inherit(dev, lowerdev);
 948         }
 949
 950         err = register_netdevice(dev);
 951         if (err < 0)
 952                 goto destroy_port;
 953
 954         dev->priv_flags |= IFF_MACVLAN;
 955         err = netdev_upper_dev_link(lowerdev, dev);
 956         if (err)
 957                 goto unregister_netdev;
 958
 959         list_add_tail_rcu(&vlan->list, &port->vlans);
 960         netif_stacked_transfer_operstate(lowerdev, dev);
 961
 962         return 0;
 963
 964 unregister_netdev:
 965         unregister_netdevice(dev);
 966 destroy_port:
 967         if (MACVLAN_PORT_IS_EMPTY(port))
 968                 macvlan_port_destroy(lowerdev);
 969
 970         return err;
 971 }
 972 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
 973
 974 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
 975                            struct nlattr *tb[], struct nlattr *data[])
 976 {
 977         return macvlan_common_newlink(src_net, dev, tb, data);
 978 }
 979
 980 void macvlan_dellink(struct net_device *dev, struct list_head *head)
 981 {
 982         struct macvlan_dev *vlan = netdev_priv(dev);
 983
 984         list_del_rcu(&vlan->list);
 985         unregister_netdevice_queue(dev, head);
 986         netdev_upper_dev_unlink(vlan->lowerdev, dev);
 987 }
 988 EXPORT_SYMBOL_GPL(macvlan_dellink);
 989
 990 static int macvlan_changelink(struct net_device *dev,
 991                 struct nlattr *tb[], struct nlattr *data[])
 992 {
 993         struct macvlan_dev *vlan = netdev_priv(dev);
 994         enum macvlan_mode mode;
 995         bool set_mode = false;
 996
 997         /* Validate mode, but don't set yet: setting flags may fail. */
 998         if (data && data[IFLA_MACVLAN_MODE]) {
 999                 set_mode = true;
1000                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1001                 /* Passthrough mode can't be set or cleared dynamically */
1002                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1003                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1004                         return -EINVAL;
1005         }
1006
1007         if (data && data[IFLA_MACVLAN_FLAGS]) {
1008                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1009                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1010                 if (vlan->port->passthru && promisc) {
1011                         int err;
1012
1013                         if (flags & MACVLAN_FLAG_NOPROMISC)
1014                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1015                         else
1016                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1017                         if (err < 0)
1018                                 return err;
1019                 }
1020                 vlan->flags = flags;
1021         }
1022         if (set_mode)
1023                 vlan->mode = mode;
1024         return 0;
1025 }
1026
1027 static size_t macvlan_get_size(const struct net_device *dev)
1028 {
1029         return (0
1030                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1031                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1032                 );
1033 }
1034
1035 static int macvlan_fill_info(struct sk_buff *skb,
1036                                 const struct net_device *dev)
1037 {
1038         struct macvlan_dev *vlan = netdev_priv(dev);
1039
1040         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1041                 goto nla_put_failure;
1042         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1043                 goto nla_put_failure;
1044         return 0;
1045
1046 nla_put_failure:
1047         return -EMSGSIZE;
1048 }
1049
1050 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1051         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1052         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1053 };
1054
1055 int macvlan_link_register(struct rtnl_link_ops *ops)
1056 {
1057         /* common fields */
1058         ops->priv_size          = sizeof(struct macvlan_dev);
1059         ops->validate           = macvlan_validate;
1060         ops->maxtype            = IFLA_MACVLAN_MAX;
1061         ops->policy             = macvlan_policy;
1062         ops->changelink         = macvlan_changelink;
1063         ops->get_size           = macvlan_get_size;
1064         ops->fill_info          = macvlan_fill_info;
1065
1066         return rtnl_link_register(ops);
1067 };
1068 EXPORT_SYMBOL_GPL(macvlan_link_register);
1069
1070 static struct rtnl_link_ops macvlan_link_ops = {
1071         .kind           = "macvlan",
1072         .setup          = macvlan_setup,
1073         .newlink        = macvlan_newlink,
1074         .dellink        = macvlan_dellink,
1075 };
1076
1077 static int macvlan_device_event(struct notifier_block *unused,
1078                                 unsigned long event, void *ptr)
1079 {
1080         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1081         struct macvlan_dev *vlan, *next;
1082         struct macvlan_port *port;
1083         LIST_HEAD(list_kill);
1084
1085         if (!macvlan_port_exists(dev))
1086                 return NOTIFY_DONE;
1087
1088         port = macvlan_port_get_rtnl(dev);
1089
1090         switch (event) {
1091         case NETDEV_CHANGE:
1092                 list_for_each_entry(vlan, &port->vlans, list)
1093                         netif_stacked_transfer_operstate(vlan->lowerdev,
1094                                                          vlan->dev);
1095                 break;
1096         case NETDEV_FEAT_CHANGE:
1097                 list_for_each_entry(vlan, &port->vlans, list) {
1098                         vlan->dev->gso_max_size = dev->gso_max_size;
1099                         netdev_update_features(vlan->dev);
1100                 }
1101                 break;
1102         case NETDEV_CHANGEMTU:
1103                 list_for_each_entry(vlan, &port->vlans, list) {
1104                         if (vlan->dev->mtu <= dev->mtu)
1105                                 continue;
1106                         dev_set_mtu(vlan->dev, dev->mtu);
1107                 }
1108                 break;
1109         case NETDEV_UNREGISTER:
1110                 /* twiddle thumbs on netns device moves */
1111                 if (dev->reg_state != NETREG_UNREGISTERING)
1112                         break;
1113
1114                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1115                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1116                 unregister_netdevice_many(&list_kill);
1117                 list_del(&list_kill);
1118                 break;
1119         case NETDEV_PRE_TYPE_CHANGE:
1120                 /* Forbid underlaying device to change its type. */
1121                 return NOTIFY_BAD;
1122         }
1123         return NOTIFY_DONE;
1124 }
1125
1126 static struct notifier_block macvlan_notifier_block __read_mostly = {
1127         .notifier_call  = macvlan_device_event,
1128 };
1129
1130 static int __init macvlan_init_module(void)
1131 {
1132         int err;
1133
1134         register_netdevice_notifier(&macvlan_notifier_block);
1135
1136         err = macvlan_link_register(&macvlan_link_ops);
1137         if (err < 0)
1138                 goto err1;
1139         return 0;
1140 err1:
1141         unregister_netdevice_notifier(&macvlan_notifier_block);
1142         return err;
1143 }
1144
1145 static void __exit macvlan_cleanup_module(void)
1146 {
1147         rtnl_link_unregister(&macvlan_link_ops);
1148         unregister_netdevice_notifier(&macvlan_notifier_block);
1149 }
1150
1151 module_init(macvlan_init_module);
1152 module_exit(macvlan_cleanup_module);
1153
1154 MODULE_LICENSE("GPL");
1155 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1156 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1157 MODULE_ALIAS_RTNL_LINK("macvlan");