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_link.h>
  30 #include <linux/if_macvlan.h>
  31 #include <net/rtnetlink.h>
  32
  33 #define MACVLAN_HASH_SIZE       (1 << BITS_PER_BYTE)
  34
  35 struct macvlan_port {
  36         struct net_device       *dev;
  37         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  38         struct list_head        vlans;
  39 };
  40
  41 struct macvlan_dev {
  42         struct net_device       *dev;
  43         struct list_head        list;
  44         struct hlist_node       hlist;
  45         struct macvlan_port     *port;
  46         struct net_device       *lowerdev;
  47 };
  48
  49
  50 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
  51                                                const unsigned char *addr)
  52 {
  53         struct macvlan_dev *vlan;
  54         struct hlist_node *n;
  55
  56         hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
  57                 if (!compare_ether_addr(vlan->dev->dev_addr, addr))
  58                         return vlan;
  59         }
  60         return NULL;
  61 }
  62
  63 static void macvlan_hash_add(struct macvlan_dev *vlan)
  64 {
  65         struct macvlan_port *port = vlan->port;
  66         const unsigned char *addr = vlan->dev->dev_addr;
  67
  68         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
  69 }
  70
  71 static void macvlan_hash_del(struct macvlan_dev *vlan)
  72 {
  73         hlist_del_rcu(&vlan->hlist);
  74         synchronize_rcu();
  75 }
  76
  77 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
  78                                         const unsigned char *addr)
  79 {
  80         macvlan_hash_del(vlan);
  81         /* Now that we are unhashed it is safe to change the device
  82          * address without confusing packet delivery.
  83          */
  84         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
  85         macvlan_hash_add(vlan);
  86 }
  87
  88 static int macvlan_addr_busy(const struct macvlan_port *port,
  89                                 const unsigned char *addr)
  90 {
  91         /* Test to see if the specified multicast address is
  92          * currently in use by the underlying device or
  93          * another macvlan.
  94          */
  95         if (memcmp(port->dev->dev_addr, addr, ETH_ALEN) == 0)
  96                 return 1;
  97
  98         if (macvlan_hash_lookup(port, addr))
  99                 return 1;
 100
 101         return 0;
 102 }
 103
 104 static void macvlan_broadcast(struct sk_buff *skb,
 105                               const struct macvlan_port *port)
 106 {
 107         const struct ethhdr *eth = eth_hdr(skb);
 108         const struct macvlan_dev *vlan;
 109         struct hlist_node *n;
 110         struct net_device *dev;
 111         struct sk_buff *nskb;
 112         unsigned int i;
 113
 114         if (skb->protocol == htons(ETH_P_PAUSE))
 115                 return;
 116
 117         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 118                 hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
 119                         dev = vlan->dev;
 120
 121                         nskb = skb_clone(skb, GFP_ATOMIC);
 122                         if (nskb == NULL) {
 123                                 dev->stats.rx_errors++;
 124                                 dev->stats.rx_dropped++;
 125                                 continue;
 126                         }
 127
 128                         dev->stats.rx_bytes += skb->len + ETH_HLEN;
 129                         dev->stats.rx_packets++;
 130                         dev->stats.multicast++;
 131
 132                         nskb->dev = dev;
 133                         if (!compare_ether_addr(eth->h_dest, dev->broadcast))
 134                                 nskb->pkt_type = PACKET_BROADCAST;
 135                         else
 136                                 nskb->pkt_type = PACKET_MULTICAST;
 137
 138                         netif_rx(nskb);
 139                 }
 140         }
 141 }
 142
 143 /* called under rcu_read_lock() from netif_receive_skb */
 144 static struct sk_buff *macvlan_handle_frame(struct sk_buff *skb)
 145 {
 146         const struct ethhdr *eth = eth_hdr(skb);
 147         const struct macvlan_port *port;
 148         const struct macvlan_dev *vlan;
 149         struct net_device *dev;
 150
 151         port = rcu_dereference(skb->dev->macvlan_port);
 152         if (port == NULL)
 153                 return skb;
 154
 155         if (is_multicast_ether_addr(eth->h_dest)) {
 156                 macvlan_broadcast(skb, port);
 157                 return skb;
 158         }
 159
 160         vlan = macvlan_hash_lookup(port, eth->h_dest);
 161         if (vlan == NULL)
 162                 return skb;
 163
 164         dev = vlan->dev;
 165         if (unlikely(!(dev->flags & IFF_UP))) {
 166                 kfree_skb(skb);
 167                 return NULL;
 168         }
 169
 170         skb = skb_share_check(skb, GFP_ATOMIC);
 171         if (skb == NULL) {
 172                 dev->stats.rx_errors++;
 173                 dev->stats.rx_dropped++;
 174                 return NULL;
 175         }
 176
 177         dev->stats.rx_bytes += skb->len + ETH_HLEN;
 178         dev->stats.rx_packets++;
 179
 180         skb->dev = dev;
 181         skb->pkt_type = PACKET_HOST;
 182
 183         netif_rx(skb);
 184         return NULL;
 185 }
 186
 187 static int macvlan_start_xmit(struct sk_buff *skb, struct net_device *dev)
 188 {
 189         const struct macvlan_dev *vlan = netdev_priv(dev);
 190         unsigned int len = skb->len;
 191         int ret;
 192
 193         skb->dev = vlan->lowerdev;
 194         ret = dev_queue_xmit(skb);
 195
 196         if (likely(ret == NET_XMIT_SUCCESS)) {
 197                 dev->stats.tx_packets++;
 198                 dev->stats.tx_bytes += len;
 199         } else {
 200                 dev->stats.tx_errors++;
 201                 dev->stats.tx_aborted_errors++;
 202         }
 203         return NETDEV_TX_OK;
 204 }
 205
 206 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 207                                unsigned short type, const void *daddr,
 208                                const void *saddr, unsigned len)
 209 {
 210         const struct macvlan_dev *vlan = netdev_priv(dev);
 211         struct net_device *lowerdev = vlan->lowerdev;
 212
 213         return dev_hard_header(skb, lowerdev, type, daddr,
 214                                saddr ? : dev->dev_addr, len);
 215 }
 216
 217 static const struct header_ops macvlan_hard_header_ops = {
 218         .create         = macvlan_hard_header,
 219         .rebuild        = eth_rebuild_header,
 220         .parse          = eth_header_parse,
 221         .cache          = eth_header_cache,
 222         .cache_update   = eth_header_cache_update,
 223 };
 224
 225 static int macvlan_open(struct net_device *dev)
 226 {
 227         struct macvlan_dev *vlan = netdev_priv(dev);
 228         struct net_device *lowerdev = vlan->lowerdev;
 229         int err;
 230
 231         err = -EBUSY;
 232         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 233                 goto out;
 234
 235         err = dev_unicast_add(lowerdev, dev->dev_addr, ETH_ALEN);
 236         if (err < 0)
 237                 goto out;
 238         if (dev->flags & IFF_ALLMULTI) {
 239                 err = dev_set_allmulti(lowerdev, 1);
 240                 if (err < 0)
 241                         goto del_unicast;
 242         }
 243         macvlan_hash_add(vlan);
 244         return 0;
 245
 246 del_unicast:
 247         dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
 248 out:
 249         return err;
 250 }
 251
 252 static int macvlan_stop(struct net_device *dev)
 253 {
 254         struct macvlan_dev *vlan = netdev_priv(dev);
 255         struct net_device *lowerdev = vlan->lowerdev;
 256
 257         dev_mc_unsync(lowerdev, dev);
 258         if (dev->flags & IFF_ALLMULTI)
 259                 dev_set_allmulti(lowerdev, -1);
 260
 261         dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
 262
 263         macvlan_hash_del(vlan);
 264         return 0;
 265 }
 266
 267 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 268 {
 269         struct macvlan_dev *vlan = netdev_priv(dev);
 270         struct net_device *lowerdev = vlan->lowerdev;
 271         struct sockaddr *addr = p;
 272         int err;
 273
 274         if (!is_valid_ether_addr(addr->sa_data))
 275                 return -EADDRNOTAVAIL;
 276
 277         if (!(dev->flags & IFF_UP)) {
 278                 /* Just copy in the new address */
 279                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 280         } else {
 281                 /* Rehash and update the device filters */
 282                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
 283                         return -EBUSY;
 284
 285                 if ((err = dev_unicast_add(lowerdev, addr->sa_data, ETH_ALEN)))
 286                         return err;
 287
 288                 dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
 289
 290                 macvlan_hash_change_addr(vlan, addr->sa_data);
 291         }
 292         return 0;
 293 }
 294
 295 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 296 {
 297         struct macvlan_dev *vlan = netdev_priv(dev);
 298         struct net_device *lowerdev = vlan->lowerdev;
 299
 300         if (change & IFF_ALLMULTI)
 301                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 302 }
 303
 304 static void macvlan_set_multicast_list(struct net_device *dev)
 305 {
 306         struct macvlan_dev *vlan = netdev_priv(dev);
 307
 308         dev_mc_sync(vlan->lowerdev, dev);
 309 }
 310
 311 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 312 {
 313         struct macvlan_dev *vlan = netdev_priv(dev);
 314
 315         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 316                 return -EINVAL;
 317         dev->mtu = new_mtu;
 318         return 0;
 319 }
 320
 321 /*
 322  * macvlan network devices have devices nesting below it and are a special
 323  * "super class" of normal network devices; split their locks off into a
 324  * separate class since they always nest.
 325  */
 326 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 327 static struct lock_class_key macvlan_netdev_addr_lock_key;
 328
 329 #define MACVLAN_FEATURES \
 330         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 331          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 332          NETIF_F_TSO_ECN | NETIF_F_TSO6)
 333
 334 #define MACVLAN_STATE_MASK \
 335         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 336
 337 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 338                                           struct netdev_queue *txq,
 339                                           void *_unused)
 340 {
 341         lockdep_set_class(&txq->_xmit_lock,
 342                           &macvlan_netdev_xmit_lock_key);
 343 }
 344
 345 static void macvlan_set_lockdep_class(struct net_device *dev)
 346 {
 347         lockdep_set_class(&dev->addr_list_lock,
 348                           &macvlan_netdev_addr_lock_key);
 349         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 350 }
 351
 352 static int macvlan_init(struct net_device *dev)
 353 {
 354         struct macvlan_dev *vlan = netdev_priv(dev);
 355         const struct net_device *lowerdev = vlan->lowerdev;
 356
 357         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 358                                   (lowerdev->state & MACVLAN_STATE_MASK);
 359         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 360         dev->iflink             = lowerdev->ifindex;
 361
 362         macvlan_set_lockdep_class(dev);
 363
 364         return 0;
 365 }
 366
 367 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 368                                         struct ethtool_drvinfo *drvinfo)
 369 {
 370         snprintf(drvinfo->driver, 32, "macvlan");
 371         snprintf(drvinfo->version, 32, "0.1");
 372 }
 373
 374 static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev)
 375 {
 376         const struct macvlan_dev *vlan = netdev_priv(dev);
 377         return dev_ethtool_get_rx_csum(vlan->lowerdev);
 378 }
 379
 380 static int macvlan_ethtool_get_settings(struct net_device *dev,
 381                                         struct ethtool_cmd *cmd)
 382 {
 383         const struct macvlan_dev *vlan = netdev_priv(dev);
 384         return dev_ethtool_get_settings(vlan->lowerdev, cmd);
 385 }
 386
 387 static u32 macvlan_ethtool_get_flags(struct net_device *dev)
 388 {
 389         const struct macvlan_dev *vlan = netdev_priv(dev);
 390         return dev_ethtool_get_flags(vlan->lowerdev);
 391 }
 392
 393 static const struct ethtool_ops macvlan_ethtool_ops = {
 394         .get_link               = ethtool_op_get_link,
 395         .get_settings           = macvlan_ethtool_get_settings,
 396         .get_rx_csum            = macvlan_ethtool_get_rx_csum,
 397         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 398         .get_flags              = macvlan_ethtool_get_flags,
 399 };
 400
 401 static const struct net_device_ops macvlan_netdev_ops = {
 402         .ndo_init               = macvlan_init,
 403         .ndo_open               = macvlan_open,
 404         .ndo_stop               = macvlan_stop,
 405         .ndo_start_xmit         = macvlan_start_xmit,
 406         .ndo_change_mtu         = macvlan_change_mtu,
 407         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 408         .ndo_set_mac_address    = macvlan_set_mac_address,
 409         .ndo_set_multicast_list = macvlan_set_multicast_list,
 410         .ndo_validate_addr      = eth_validate_addr,
 411 };
 412
 413 static void macvlan_setup(struct net_device *dev)
 414 {
 415         ether_setup(dev);
 416
 417         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
 418         dev->netdev_ops         = &macvlan_netdev_ops;
 419         dev->destructor         = free_netdev;
 420         dev->header_ops         = &macvlan_hard_header_ops,
 421         dev->ethtool_ops        = &macvlan_ethtool_ops;
 422         dev->tx_queue_len       = 0;
 423 }
 424
 425 static int macvlan_port_create(struct net_device *dev)
 426 {
 427         struct macvlan_port *port;
 428         unsigned int i;
 429
 430         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 431                 return -EINVAL;
 432
 433         port = kzalloc(sizeof(*port), GFP_KERNEL);
 434         if (port == NULL)
 435                 return -ENOMEM;
 436
 437         port->dev = dev;
 438         INIT_LIST_HEAD(&port->vlans);
 439         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 440                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 441         rcu_assign_pointer(dev->macvlan_port, port);
 442         return 0;
 443 }
 444
 445 static void macvlan_port_destroy(struct net_device *dev)
 446 {
 447         struct macvlan_port *port = dev->macvlan_port;
 448
 449         rcu_assign_pointer(dev->macvlan_port, NULL);
 450         synchronize_rcu();
 451         kfree(port);
 452 }
 453
 454 static void macvlan_transfer_operstate(struct net_device *dev)
 455 {
 456         struct macvlan_dev *vlan = netdev_priv(dev);
 457         const struct net_device *lowerdev = vlan->lowerdev;
 458
 459         if (lowerdev->operstate == IF_OPER_DORMANT)
 460                 netif_dormant_on(dev);
 461         else
 462                 netif_dormant_off(dev);
 463
 464         if (netif_carrier_ok(lowerdev)) {
 465                 if (!netif_carrier_ok(dev))
 466                         netif_carrier_on(dev);
 467         } else {
 468                 if (netif_carrier_ok(dev))
 469                         netif_carrier_off(dev);
 470         }
 471 }
 472
 473 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 474 {
 475         if (tb[IFLA_ADDRESS]) {
 476                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 477                         return -EINVAL;
 478                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 479                         return -EADDRNOTAVAIL;
 480         }
 481         return 0;
 482 }
 483
 484 static int macvlan_newlink(struct net_device *dev,
 485                            struct nlattr *tb[], struct nlattr *data[])
 486 {
 487         struct macvlan_dev *vlan = netdev_priv(dev);
 488         struct macvlan_port *port;
 489         struct net_device *lowerdev;
 490         int err;
 491
 492         if (!tb[IFLA_LINK])
 493                 return -EINVAL;
 494
 495         lowerdev = __dev_get_by_index(dev_net(dev), nla_get_u32(tb[IFLA_LINK]));
 496         if (lowerdev == NULL)
 497                 return -ENODEV;
 498
 499         /* When creating macvlans on top of other macvlans - use
 500          * the real device as the lowerdev.
 501          */
 502         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
 503                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
 504                 lowerdev = lowervlan->lowerdev;
 505         }
 506
 507         if (!tb[IFLA_MTU])
 508                 dev->mtu = lowerdev->mtu;
 509         else if (dev->mtu > lowerdev->mtu)
 510                 return -EINVAL;
 511
 512         if (!tb[IFLA_ADDRESS])
 513                 random_ether_addr(dev->dev_addr);
 514
 515         if (lowerdev->macvlan_port == NULL) {
 516                 err = macvlan_port_create(lowerdev);
 517                 if (err < 0)
 518                         return err;
 519         }
 520         port = lowerdev->macvlan_port;
 521
 522         vlan->lowerdev = lowerdev;
 523         vlan->dev      = dev;
 524         vlan->port     = port;
 525
 526         err = register_netdevice(dev);
 527         if (err < 0)
 528                 return err;
 529
 530         list_add_tail(&vlan->list, &port->vlans);
 531         macvlan_transfer_operstate(dev);
 532         return 0;
 533 }
 534
 535 static void macvlan_dellink(struct net_device *dev)
 536 {
 537         struct macvlan_dev *vlan = netdev_priv(dev);
 538         struct macvlan_port *port = vlan->port;
 539
 540         list_del(&vlan->list);
 541         unregister_netdevice(dev);
 542
 543         if (list_empty(&port->vlans))
 544                 macvlan_port_destroy(port->dev);
 545 }
 546
 547 static struct rtnl_link_ops macvlan_link_ops __read_mostly = {
 548         .kind           = "macvlan",
 549         .priv_size      = sizeof(struct macvlan_dev),
 550         .setup          = macvlan_setup,
 551         .validate       = macvlan_validate,
 552         .newlink        = macvlan_newlink,
 553         .dellink        = macvlan_dellink,
 554 };
 555
 556 static int macvlan_device_event(struct notifier_block *unused,
 557                                 unsigned long event, void *ptr)
 558 {
 559         struct net_device *dev = ptr;
 560         struct macvlan_dev *vlan, *next;
 561         struct macvlan_port *port;
 562
 563         port = dev->macvlan_port;
 564         if (port == NULL)
 565                 return NOTIFY_DONE;
 566
 567         switch (event) {
 568         case NETDEV_CHANGE:
 569                 list_for_each_entry(vlan, &port->vlans, list)
 570                         macvlan_transfer_operstate(vlan->dev);
 571                 break;
 572         case NETDEV_FEAT_CHANGE:
 573                 list_for_each_entry(vlan, &port->vlans, list) {
 574                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
 575                         netdev_features_change(vlan->dev);
 576                 }
 577                 break;
 578         case NETDEV_UNREGISTER:
 579                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
 580                         macvlan_dellink(vlan->dev);
 581                 break;
 582         }
 583         return NOTIFY_DONE;
 584 }
 585
 586 static struct notifier_block macvlan_notifier_block __read_mostly = {
 587         .notifier_call  = macvlan_device_event,
 588 };
 589
 590 static int __init macvlan_init_module(void)
 591 {
 592         int err;
 593
 594         register_netdevice_notifier(&macvlan_notifier_block);
 595         macvlan_handle_frame_hook = macvlan_handle_frame;
 596
 597         err = rtnl_link_register(&macvlan_link_ops);
 598         if (err < 0)
 599                 goto err1;
 600         return 0;
 601 err1:
 602         macvlan_handle_frame_hook = NULL;
 603         unregister_netdevice_notifier(&macvlan_notifier_block);
 604         return err;
 605 }
 606
 607 static void __exit macvlan_cleanup_module(void)
 608 {
 609         rtnl_link_unregister(&macvlan_link_ops);
 610         macvlan_handle_frame_hook = NULL;
 611         unregister_netdevice_notifier(&macvlan_notifier_block);
 612 }
 613
 614 module_init(macvlan_init_module);
 615 module_exit(macvlan_cleanup_module);
 616
 617 MODULE_LICENSE("GPL");
 618 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 619 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
 620 MODULE_ALIAS_RTNL_LINK("macvlan");