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_broadcast(struct sk_buff *skb,
  64                               const struct macvlan_port *port)
  65 {
  66         const struct ethhdr *eth = eth_hdr(skb);
  67         const struct macvlan_dev *vlan;
  68         struct hlist_node *n;
  69         struct net_device *dev;
  70         struct sk_buff *nskb;
  71         unsigned int i;
  72
  73         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
  74                 hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
  75                         dev = vlan->dev;
  76
  77                         nskb = skb_clone(skb, GFP_ATOMIC);
  78                         if (nskb == NULL) {
  79                                 dev->stats.rx_errors++;
  80                                 dev->stats.rx_dropped++;
  81                                 continue;
  82                         }
  83
  84                         dev->stats.rx_bytes += skb->len + ETH_HLEN;
  85                         dev->stats.rx_packets++;
  86                         dev->stats.multicast++;
  87                         dev->last_rx = jiffies;
  88
  89                         nskb->dev = dev;
  90                         if (!compare_ether_addr(eth->h_dest, dev->broadcast))
  91                                 nskb->pkt_type = PACKET_BROADCAST;
  92                         else
  93                                 nskb->pkt_type = PACKET_MULTICAST;
  94
  95                         netif_rx(nskb);
  96                 }
  97         }
  98 }
  99
 100 /* called under rcu_read_lock() from netif_receive_skb */
 101 static struct sk_buff *macvlan_handle_frame(struct sk_buff *skb)
 102 {
 103         const struct ethhdr *eth = eth_hdr(skb);
 104         const struct macvlan_port *port;
 105         const struct macvlan_dev *vlan;
 106         struct net_device *dev;
 107
 108         port = rcu_dereference(skb->dev->macvlan_port);
 109         if (port == NULL)
 110                 return skb;
 111
 112         if (is_multicast_ether_addr(eth->h_dest)) {
 113                 macvlan_broadcast(skb, port);
 114                 return skb;
 115         }
 116
 117         vlan = macvlan_hash_lookup(port, eth->h_dest);
 118         if (vlan == NULL)
 119                 return skb;
 120
 121         dev = vlan->dev;
 122         if (unlikely(!(dev->flags & IFF_UP))) {
 123                 kfree_skb(skb);
 124                 return NULL;
 125         }
 126
 127         skb = skb_share_check(skb, GFP_ATOMIC);
 128         if (skb == NULL) {
 129                 dev->stats.rx_errors++;
 130                 dev->stats.rx_dropped++;
 131                 return NULL;
 132         }
 133
 134         dev->stats.rx_bytes += skb->len + ETH_HLEN;
 135         dev->stats.rx_packets++;
 136         dev->last_rx = jiffies;
 137
 138         skb->dev = dev;
 139         skb->pkt_type = PACKET_HOST;
 140
 141         netif_rx(skb);
 142         return NULL;
 143 }
 144
 145 static int macvlan_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
 146 {
 147         const struct macvlan_dev *vlan = netdev_priv(dev);
 148         unsigned int len = skb->len;
 149         int ret;
 150
 151         skb->dev = vlan->lowerdev;
 152         ret = dev_queue_xmit(skb);
 153
 154         if (likely(ret == NET_XMIT_SUCCESS)) {
 155                 dev->stats.tx_packets++;
 156                 dev->stats.tx_bytes += len;
 157         } else {
 158                 dev->stats.tx_errors++;
 159                 dev->stats.tx_aborted_errors++;
 160         }
 161         return NETDEV_TX_OK;
 162 }
 163
 164 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 165                                unsigned short type, const void *daddr,
 166                                const void *saddr, unsigned len)
 167 {
 168         const struct macvlan_dev *vlan = netdev_priv(dev);
 169         struct net_device *lowerdev = vlan->lowerdev;
 170
 171         return dev_hard_header(skb, lowerdev, type, daddr,
 172                                saddr ? : dev->dev_addr, len);
 173 }
 174
 175 static const struct header_ops macvlan_hard_header_ops = {
 176         .create         = macvlan_hard_header,
 177         .rebuild        = eth_rebuild_header,
 178         .parse          = eth_header_parse,
 179         .cache          = eth_header_cache,
 180         .cache_update   = eth_header_cache_update,
 181 };
 182
 183 static int macvlan_open(struct net_device *dev)
 184 {
 185         struct macvlan_dev *vlan = netdev_priv(dev);
 186         struct macvlan_port *port = vlan->port;
 187         struct net_device *lowerdev = vlan->lowerdev;
 188         int err;
 189
 190         err = dev_unicast_add(lowerdev, dev->dev_addr, ETH_ALEN);
 191         if (err < 0)
 192                 goto out;
 193         if (dev->flags & IFF_ALLMULTI) {
 194                 err = dev_set_allmulti(lowerdev, 1);
 195                 if (err < 0)
 196                         goto del_unicast;
 197         }
 198
 199         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[dev->dev_addr[5]]);
 200         return 0;
 201
 202 del_unicast:
 203         dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
 204 out:
 205         return err;
 206 }
 207
 208 static int macvlan_stop(struct net_device *dev)
 209 {
 210         struct macvlan_dev *vlan = netdev_priv(dev);
 211         struct net_device *lowerdev = vlan->lowerdev;
 212
 213         dev_mc_unsync(lowerdev, dev);
 214         if (dev->flags & IFF_ALLMULTI)
 215                 dev_set_allmulti(lowerdev, -1);
 216
 217         dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
 218
 219         hlist_del_rcu(&vlan->hlist);
 220         synchronize_rcu();
 221         return 0;
 222 }
 223
 224 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 225 {
 226         struct macvlan_dev *vlan = netdev_priv(dev);
 227         struct net_device *lowerdev = vlan->lowerdev;
 228         struct sockaddr *addr = p;
 229         int err;
 230
 231         if (!is_valid_ether_addr(addr->sa_data))
 232                 return -EADDRNOTAVAIL;
 233
 234         if (!(dev->flags & IFF_UP))
 235                 goto out;
 236
 237         err = dev_unicast_add(lowerdev, addr->sa_data, ETH_ALEN);
 238         if (err < 0)
 239                 return err;
 240         dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
 241
 242 out:
 243         memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 244         return 0;
 245 }
 246
 247 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 248 {
 249         struct macvlan_dev *vlan = netdev_priv(dev);
 250         struct net_device *lowerdev = vlan->lowerdev;
 251
 252         if (change & IFF_ALLMULTI)
 253                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 254 }
 255
 256 static void macvlan_set_multicast_list(struct net_device *dev)
 257 {
 258         struct macvlan_dev *vlan = netdev_priv(dev);
 259
 260         dev_mc_sync(vlan->lowerdev, dev);
 261 }
 262
 263 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 264 {
 265         struct macvlan_dev *vlan = netdev_priv(dev);
 266
 267         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 268                 return -EINVAL;
 269         dev->mtu = new_mtu;
 270         return 0;
 271 }
 272
 273 /*
 274  * macvlan network devices have devices nesting below it and are a special
 275  * "super class" of normal network devices; split their locks off into a
 276  * separate class since they always nest.
 277  */
 278 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 279
 280 #define MACVLAN_FEATURES \
 281         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 282          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 283          NETIF_F_TSO_ECN | NETIF_F_TSO6)
 284
 285 #define MACVLAN_STATE_MASK \
 286         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 287
 288 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 289                                           struct netdev_queue *txq,
 290                                           void *_unused)
 291 {
 292         lockdep_set_class(&txq->_xmit_lock,
 293                           &macvlan_netdev_xmit_lock_key);
 294 }
 295
 296 static void macvlan_set_lockdep_class(struct net_device *dev)
 297 {
 298         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 299 }
 300
 301 static int macvlan_init(struct net_device *dev)
 302 {
 303         struct macvlan_dev *vlan = netdev_priv(dev);
 304         const struct net_device *lowerdev = vlan->lowerdev;
 305
 306         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 307                                   (lowerdev->state & MACVLAN_STATE_MASK);
 308         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 309         dev->iflink             = lowerdev->ifindex;
 310
 311         macvlan_set_lockdep_class(dev);
 312
 313         return 0;
 314 }
 315
 316 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 317                                         struct ethtool_drvinfo *drvinfo)
 318 {
 319         snprintf(drvinfo->driver, 32, "macvlan");
 320         snprintf(drvinfo->version, 32, "0.1");
 321 }
 322
 323 static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev)
 324 {
 325         const struct macvlan_dev *vlan = netdev_priv(dev);
 326         struct net_device *lowerdev = vlan->lowerdev;
 327
 328         if (lowerdev->ethtool_ops->get_rx_csum == NULL)
 329                 return 0;
 330         return lowerdev->ethtool_ops->get_rx_csum(lowerdev);
 331 }
 332
 333 static const struct ethtool_ops macvlan_ethtool_ops = {
 334         .get_link               = ethtool_op_get_link,
 335         .get_rx_csum            = macvlan_ethtool_get_rx_csum,
 336         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 337 };
 338
 339 static void macvlan_setup(struct net_device *dev)
 340 {
 341         ether_setup(dev);
 342
 343         dev->init               = macvlan_init;
 344         dev->open               = macvlan_open;
 345         dev->stop               = macvlan_stop;
 346         dev->change_mtu         = macvlan_change_mtu;
 347         dev->change_rx_flags    = macvlan_change_rx_flags;
 348         dev->set_mac_address    = macvlan_set_mac_address;
 349         dev->set_multicast_list = macvlan_set_multicast_list;
 350         dev->hard_start_xmit    = macvlan_hard_start_xmit;
 351         dev->destructor         = free_netdev;
 352         dev->header_ops         = &macvlan_hard_header_ops,
 353         dev->ethtool_ops        = &macvlan_ethtool_ops;
 354         dev->tx_queue_len       = 0;
 355 }
 356
 357 static int macvlan_port_create(struct net_device *dev)
 358 {
 359         struct macvlan_port *port;
 360         unsigned int i;
 361
 362         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 363                 return -EINVAL;
 364
 365         port = kzalloc(sizeof(*port), GFP_KERNEL);
 366         if (port == NULL)
 367                 return -ENOMEM;
 368
 369         port->dev = dev;
 370         INIT_LIST_HEAD(&port->vlans);
 371         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 372                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 373         rcu_assign_pointer(dev->macvlan_port, port);
 374         return 0;
 375 }
 376
 377 static void macvlan_port_destroy(struct net_device *dev)
 378 {
 379         struct macvlan_port *port = dev->macvlan_port;
 380
 381         rcu_assign_pointer(dev->macvlan_port, NULL);
 382         synchronize_rcu();
 383         kfree(port);
 384 }
 385
 386 static void macvlan_transfer_operstate(struct net_device *dev)
 387 {
 388         struct macvlan_dev *vlan = netdev_priv(dev);
 389         const struct net_device *lowerdev = vlan->lowerdev;
 390
 391         if (lowerdev->operstate == IF_OPER_DORMANT)
 392                 netif_dormant_on(dev);
 393         else
 394                 netif_dormant_off(dev);
 395
 396         if (netif_carrier_ok(lowerdev)) {
 397                 if (!netif_carrier_ok(dev))
 398                         netif_carrier_on(dev);
 399         } else {
 400                 if (netif_carrier_ok(dev))
 401                         netif_carrier_off(dev);
 402         }
 403 }
 404
 405 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 406 {
 407         if (tb[IFLA_ADDRESS]) {
 408                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 409                         return -EINVAL;
 410                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 411                         return -EADDRNOTAVAIL;
 412         }
 413         return 0;
 414 }
 415
 416 static int macvlan_newlink(struct net_device *dev,
 417                            struct nlattr *tb[], struct nlattr *data[])
 418 {
 419         struct macvlan_dev *vlan = netdev_priv(dev);
 420         struct macvlan_port *port;
 421         struct net_device *lowerdev;
 422         int err;
 423
 424         if (!tb[IFLA_LINK])
 425                 return -EINVAL;
 426
 427         lowerdev = __dev_get_by_index(dev_net(dev), nla_get_u32(tb[IFLA_LINK]));
 428         if (lowerdev == NULL)
 429                 return -ENODEV;
 430
 431         /* Don't allow macvlans on top of other macvlans - its not really
 432          * wrong, but lockdep can't handle it and its not useful for anything
 433          * you couldn't do directly on top of the real device.
 434          */
 435         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops)
 436                 return -ENODEV;
 437
 438         if (!tb[IFLA_MTU])
 439                 dev->mtu = lowerdev->mtu;
 440         else if (dev->mtu > lowerdev->mtu)
 441                 return -EINVAL;
 442
 443         if (!tb[IFLA_ADDRESS])
 444                 random_ether_addr(dev->dev_addr);
 445
 446         if (lowerdev->macvlan_port == NULL) {
 447                 err = macvlan_port_create(lowerdev);
 448                 if (err < 0)
 449                         return err;
 450         }
 451         port = lowerdev->macvlan_port;
 452
 453         vlan->lowerdev = lowerdev;
 454         vlan->dev      = dev;
 455         vlan->port     = port;
 456
 457         err = register_netdevice(dev);
 458         if (err < 0)
 459                 return err;
 460
 461         list_add_tail(&vlan->list, &port->vlans);
 462         macvlan_transfer_operstate(dev);
 463         return 0;
 464 }
 465
 466 static void macvlan_dellink(struct net_device *dev)
 467 {
 468         struct macvlan_dev *vlan = netdev_priv(dev);
 469         struct macvlan_port *port = vlan->port;
 470
 471         list_del(&vlan->list);
 472         unregister_netdevice(dev);
 473
 474         if (list_empty(&port->vlans))
 475                 macvlan_port_destroy(port->dev);
 476 }
 477
 478 static struct rtnl_link_ops macvlan_link_ops __read_mostly = {
 479         .kind           = "macvlan",
 480         .priv_size      = sizeof(struct macvlan_dev),
 481         .setup          = macvlan_setup,
 482         .validate       = macvlan_validate,
 483         .newlink        = macvlan_newlink,
 484         .dellink        = macvlan_dellink,
 485 };
 486
 487 static int macvlan_device_event(struct notifier_block *unused,
 488                                 unsigned long event, void *ptr)
 489 {
 490         struct net_device *dev = ptr;
 491         struct macvlan_dev *vlan, *next;
 492         struct macvlan_port *port;
 493
 494         port = dev->macvlan_port;
 495         if (port == NULL)
 496                 return NOTIFY_DONE;
 497
 498         switch (event) {
 499         case NETDEV_CHANGE:
 500                 list_for_each_entry(vlan, &port->vlans, list)
 501                         macvlan_transfer_operstate(vlan->dev);
 502                 break;
 503         case NETDEV_FEAT_CHANGE:
 504                 list_for_each_entry(vlan, &port->vlans, list) {
 505                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
 506                         netdev_features_change(vlan->dev);
 507                 }
 508                 break;
 509         case NETDEV_UNREGISTER:
 510                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
 511                         macvlan_dellink(vlan->dev);
 512                 break;
 513         }
 514         return NOTIFY_DONE;
 515 }
 516
 517 static struct notifier_block macvlan_notifier_block __read_mostly = {
 518         .notifier_call  = macvlan_device_event,
 519 };
 520
 521 static int __init macvlan_init_module(void)
 522 {
 523         int err;
 524
 525         register_netdevice_notifier(&macvlan_notifier_block);
 526         macvlan_handle_frame_hook = macvlan_handle_frame;
 527
 528         err = rtnl_link_register(&macvlan_link_ops);
 529         if (err < 0)
 530                 goto err1;
 531         return 0;
 532 err1:
 533         macvlan_handle_frame_hook = NULL;
 534         unregister_netdevice_notifier(&macvlan_notifier_block);
 535         return err;
 536 }
 537
 538 static void __exit macvlan_cleanup_module(void)
 539 {
 540         rtnl_link_unregister(&macvlan_link_ops);
 541         macvlan_handle_frame_hook = NULL;
 542         unregister_netdevice_notifier(&macvlan_notifier_block);
 543 }
 544
 545 module_init(macvlan_init_module);
 546 module_exit(macvlan_cleanup_module);
 547
 548 MODULE_LICENSE("GPL");
 549 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 550 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
 551 MODULE_ALIAS_RTNL_LINK("macvlan");