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);
 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);
 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);
 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                 err = dev_unicast_add(lowerdev, addr->sa_data);
 286                 if (err)
 287                         return err;
 288
 289                 dev_unicast_delete(lowerdev, dev->dev_addr);
 290
 291                 macvlan_hash_change_addr(vlan, addr->sa_data);
 292         }
 293         return 0;
 294 }
 295
 296 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 297 {
 298         struct macvlan_dev *vlan = netdev_priv(dev);
 299         struct net_device *lowerdev = vlan->lowerdev;
 300
 301         if (change & IFF_ALLMULTI)
 302                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 303 }
 304
 305 static void macvlan_set_multicast_list(struct net_device *dev)
 306 {
 307         struct macvlan_dev *vlan = netdev_priv(dev);
 308
 309         dev_mc_sync(vlan->lowerdev, dev);
 310 }
 311
 312 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 313 {
 314         struct macvlan_dev *vlan = netdev_priv(dev);
 315
 316         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
 317                 return -EINVAL;
 318         dev->mtu = new_mtu;
 319         return 0;
 320 }
 321
 322 /*
 323  * macvlan network devices have devices nesting below it and are a special
 324  * "super class" of normal network devices; split their locks off into a
 325  * separate class since they always nest.
 326  */
 327 static struct lock_class_key macvlan_netdev_xmit_lock_key;
 328 static struct lock_class_key macvlan_netdev_addr_lock_key;
 329
 330 #define MACVLAN_FEATURES \
 331         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 332          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
 333          NETIF_F_TSO_ECN | NETIF_F_TSO6)
 334
 335 #define MACVLAN_STATE_MASK \
 336         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 337
 338 static void macvlan_set_lockdep_class_one(struct net_device *dev,
 339                                           struct netdev_queue *txq,
 340                                           void *_unused)
 341 {
 342         lockdep_set_class(&txq->_xmit_lock,
 343                           &macvlan_netdev_xmit_lock_key);
 344 }
 345
 346 static void macvlan_set_lockdep_class(struct net_device *dev)
 347 {
 348         lockdep_set_class(&dev->addr_list_lock,
 349                           &macvlan_netdev_addr_lock_key);
 350         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
 351 }
 352
 353 static int macvlan_init(struct net_device *dev)
 354 {
 355         struct macvlan_dev *vlan = netdev_priv(dev);
 356         const struct net_device *lowerdev = vlan->lowerdev;
 357
 358         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 359                                   (lowerdev->state & MACVLAN_STATE_MASK);
 360         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 361         dev->iflink             = lowerdev->ifindex;
 362         dev->hard_header_len    = lowerdev->hard_header_len;
 363
 364         macvlan_set_lockdep_class(dev);
 365
 366         return 0;
 367 }
 368
 369 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
 370                                         struct ethtool_drvinfo *drvinfo)
 371 {
 372         snprintf(drvinfo->driver, 32, "macvlan");
 373         snprintf(drvinfo->version, 32, "0.1");
 374 }
 375
 376 static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev)
 377 {
 378         const struct macvlan_dev *vlan = netdev_priv(dev);
 379         return dev_ethtool_get_rx_csum(vlan->lowerdev);
 380 }
 381
 382 static int macvlan_ethtool_get_settings(struct net_device *dev,
 383                                         struct ethtool_cmd *cmd)
 384 {
 385         const struct macvlan_dev *vlan = netdev_priv(dev);
 386         return dev_ethtool_get_settings(vlan->lowerdev, cmd);
 387 }
 388
 389 static u32 macvlan_ethtool_get_flags(struct net_device *dev)
 390 {
 391         const struct macvlan_dev *vlan = netdev_priv(dev);
 392         return dev_ethtool_get_flags(vlan->lowerdev);
 393 }
 394
 395 static const struct ethtool_ops macvlan_ethtool_ops = {
 396         .get_link               = ethtool_op_get_link,
 397         .get_settings           = macvlan_ethtool_get_settings,
 398         .get_rx_csum            = macvlan_ethtool_get_rx_csum,
 399         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
 400         .get_flags              = macvlan_ethtool_get_flags,
 401 };
 402
 403 static const struct net_device_ops macvlan_netdev_ops = {
 404         .ndo_init               = macvlan_init,
 405         .ndo_open               = macvlan_open,
 406         .ndo_stop               = macvlan_stop,
 407         .ndo_start_xmit         = macvlan_start_xmit,
 408         .ndo_change_mtu         = macvlan_change_mtu,
 409         .ndo_change_rx_flags    = macvlan_change_rx_flags,
 410         .ndo_set_mac_address    = macvlan_set_mac_address,
 411         .ndo_set_multicast_list = macvlan_set_multicast_list,
 412         .ndo_validate_addr      = eth_validate_addr,
 413 };
 414
 415 static void macvlan_setup(struct net_device *dev)
 416 {
 417         ether_setup(dev);
 418
 419         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
 420         dev->netdev_ops         = &macvlan_netdev_ops;
 421         dev->destructor         = free_netdev;
 422         dev->header_ops         = &macvlan_hard_header_ops,
 423         dev->ethtool_ops        = &macvlan_ethtool_ops;
 424         dev->tx_queue_len       = 0;
 425 }
 426
 427 static int macvlan_port_create(struct net_device *dev)
 428 {
 429         struct macvlan_port *port;
 430         unsigned int i;
 431
 432         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
 433                 return -EINVAL;
 434
 435         port = kzalloc(sizeof(*port), GFP_KERNEL);
 436         if (port == NULL)
 437                 return -ENOMEM;
 438
 439         port->dev = dev;
 440         INIT_LIST_HEAD(&port->vlans);
 441         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
 442                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
 443         rcu_assign_pointer(dev->macvlan_port, port);
 444         return 0;
 445 }
 446
 447 static void macvlan_port_destroy(struct net_device *dev)
 448 {
 449         struct macvlan_port *port = dev->macvlan_port;
 450
 451         rcu_assign_pointer(dev->macvlan_port, NULL);
 452         synchronize_rcu();
 453         kfree(port);
 454 }
 455
 456 static void macvlan_transfer_operstate(struct net_device *dev)
 457 {
 458         struct macvlan_dev *vlan = netdev_priv(dev);
 459         const struct net_device *lowerdev = vlan->lowerdev;
 460
 461         if (lowerdev->operstate == IF_OPER_DORMANT)
 462                 netif_dormant_on(dev);
 463         else
 464                 netif_dormant_off(dev);
 465
 466         if (netif_carrier_ok(lowerdev)) {
 467                 if (!netif_carrier_ok(dev))
 468                         netif_carrier_on(dev);
 469         } else {
 470                 if (netif_carrier_ok(dev))
 471                         netif_carrier_off(dev);
 472         }
 473 }
 474
 475 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
 476 {
 477         if (tb[IFLA_ADDRESS]) {
 478                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 479                         return -EINVAL;
 480                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 481                         return -EADDRNOTAVAIL;
 482         }
 483         return 0;
 484 }
 485
 486 static int macvlan_newlink(struct net_device *dev,
 487                            struct nlattr *tb[], struct nlattr *data[])
 488 {
 489         struct macvlan_dev *vlan = netdev_priv(dev);
 490         struct macvlan_port *port;
 491         struct net_device *lowerdev;
 492         int err;
 493
 494         if (!tb[IFLA_LINK])
 495                 return -EINVAL;
 496
 497         lowerdev = __dev_get_by_index(dev_net(dev), nla_get_u32(tb[IFLA_LINK]));
 498         if (lowerdev == NULL)
 499                 return -ENODEV;
 500
 501         /* When creating macvlans on top of other macvlans - use
 502          * the real device as the lowerdev.
 503          */
 504         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
 505                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
 506                 lowerdev = lowervlan->lowerdev;
 507         }
 508
 509         if (!tb[IFLA_MTU])
 510                 dev->mtu = lowerdev->mtu;
 511         else if (dev->mtu > lowerdev->mtu)
 512                 return -EINVAL;
 513
 514         if (!tb[IFLA_ADDRESS])
 515                 random_ether_addr(dev->dev_addr);
 516
 517         if (lowerdev->macvlan_port == NULL) {
 518                 err = macvlan_port_create(lowerdev);
 519                 if (err < 0)
 520                         return err;
 521         }
 522         port = lowerdev->macvlan_port;
 523
 524         vlan->lowerdev = lowerdev;
 525         vlan->dev      = dev;
 526         vlan->port     = port;
 527
 528         err = register_netdevice(dev);
 529         if (err < 0)
 530                 return err;
 531
 532         list_add_tail(&vlan->list, &port->vlans);
 533         macvlan_transfer_operstate(dev);
 534         return 0;
 535 }
 536
 537 static void macvlan_dellink(struct net_device *dev)
 538 {
 539         struct macvlan_dev *vlan = netdev_priv(dev);
 540         struct macvlan_port *port = vlan->port;
 541
 542         list_del(&vlan->list);
 543         unregister_netdevice(dev);
 544
 545         if (list_empty(&port->vlans))
 546                 macvlan_port_destroy(port->dev);
 547 }
 548
 549 static struct rtnl_link_ops macvlan_link_ops __read_mostly = {
 550         .kind           = "macvlan",
 551         .priv_size      = sizeof(struct macvlan_dev),
 552         .setup          = macvlan_setup,
 553         .validate       = macvlan_validate,
 554         .newlink        = macvlan_newlink,
 555         .dellink        = macvlan_dellink,
 556 };
 557
 558 static int macvlan_device_event(struct notifier_block *unused,
 559                                 unsigned long event, void *ptr)
 560 {
 561         struct net_device *dev = ptr;
 562         struct macvlan_dev *vlan, *next;
 563         struct macvlan_port *port;
 564
 565         port = dev->macvlan_port;
 566         if (port == NULL)
 567                 return NOTIFY_DONE;
 568
 569         switch (event) {
 570         case NETDEV_CHANGE:
 571                 list_for_each_entry(vlan, &port->vlans, list)
 572                         macvlan_transfer_operstate(vlan->dev);
 573                 break;
 574         case NETDEV_FEAT_CHANGE:
 575                 list_for_each_entry(vlan, &port->vlans, list) {
 576                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
 577                         netdev_features_change(vlan->dev);
 578                 }
 579                 break;
 580         case NETDEV_UNREGISTER:
 581                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
 582                         macvlan_dellink(vlan->dev);
 583                 break;
 584         }
 585         return NOTIFY_DONE;
 586 }
 587
 588 static struct notifier_block macvlan_notifier_block __read_mostly = {
 589         .notifier_call  = macvlan_device_event,
 590 };
 591
 592 static int __init macvlan_init_module(void)
 593 {
 594         int err;
 595
 596         register_netdevice_notifier(&macvlan_notifier_block);
 597         macvlan_handle_frame_hook = macvlan_handle_frame;
 598
 599         err = rtnl_link_register(&macvlan_link_ops);
 600         if (err < 0)
 601                 goto err1;
 602         return 0;
 603 err1:
 604         macvlan_handle_frame_hook = NULL;
 605         unregister_netdevice_notifier(&macvlan_notifier_block);
 606         return err;
 607 }
 608
 609 static void __exit macvlan_cleanup_module(void)
 610 {
 611         rtnl_link_unregister(&macvlan_link_ops);
 612         macvlan_handle_frame_hook = NULL;
 613         unregister_netdevice_notifier(&macvlan_notifier_block);
 614 }
 615
 616 module_init(macvlan_init_module);
 617 module_exit(macvlan_cleanup_module);
 618
 619 MODULE_LICENSE("GPL");
 620 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 621 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
 622 MODULE_ALIAS_RTNL_LINK("macvlan");