drivers/net/veth.c

   1 /*
   2  *  drivers/net/veth.c
   3  *
   4  *  Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
   5  *
   6  * Author: Pavel Emelianov <xemul@openvz.org>
   7  * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
   8  *
   9  */
  10
  11 #include <linux/netdevice.h>
  12 #include <linux/slab.h>
  13 #include <linux/ethtool.h>
  14 #include <linux/etherdevice.h>
  15 #include <linux/u64_stats_sync.h>
  16
  17 #include <net/rtnetlink.h>
  18 #include <net/dst.h>
  19 #include <net/xfrm.h>
  20 #include <linux/veth.h>
  21 #include <linux/module.h>
  22
  23 #define DRV_NAME        "veth"
  24 #define DRV_VERSION     "1.0"
  25
  26 #define MIN_MTU 68              /* Min L3 MTU */
  27 #define MAX_MTU 65535           /* Max L3 MTU (arbitrary) */
  28
  29 struct pcpu_vstats {
  30         u64                     packets;
  31         u64                     bytes;
  32         struct u64_stats_sync   syncp;
  33 };
  34
  35 struct veth_priv {
  36         struct net_device __rcu *peer;
  37         atomic64_t              dropped;
  38 };
  39
  40 /*
  41  * ethtool interface
  42  */
  43
  44 static struct {
  45         const char string[ETH_GSTRING_LEN];
  46 } ethtool_stats_keys[] = {
  47         { "peer_ifindex" },
  48 };
  49
  50 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
  51 {
  52         cmd->supported          = 0;
  53         cmd->advertising        = 0;
  54         ethtool_cmd_speed_set(cmd, SPEED_10000);
  55         cmd->duplex             = DUPLEX_FULL;
  56         cmd->port               = PORT_TP;
  57         cmd->phy_address        = 0;
  58         cmd->transceiver        = XCVR_INTERNAL;
  59         cmd->autoneg            = AUTONEG_DISABLE;
  60         cmd->maxtxpkt           = 0;
  61         cmd->maxrxpkt           = 0;
  62         return 0;
  63 }
  64
  65 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
  66 {
  67         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
  68         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
  69 }
  70
  71 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
  72 {
  73         switch(stringset) {
  74         case ETH_SS_STATS:
  75                 memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
  76                 break;
  77         }
  78 }
  79
  80 static int veth_get_sset_count(struct net_device *dev, int sset)
  81 {
  82         switch (sset) {
  83         case ETH_SS_STATS:
  84                 return ARRAY_SIZE(ethtool_stats_keys);
  85         default:
  86                 return -EOPNOTSUPP;
  87         }
  88 }
  89
  90 static void veth_get_ethtool_stats(struct net_device *dev,
  91                 struct ethtool_stats *stats, u64 *data)
  92 {
  93         struct veth_priv *priv = netdev_priv(dev);
  94         struct net_device *peer = rtnl_dereference(priv->peer);
  95
  96         data[0] = peer ? peer->ifindex : 0;
  97 }
  98
  99 static const struct ethtool_ops veth_ethtool_ops = {
 100         .get_settings           = veth_get_settings,
 101         .get_drvinfo            = veth_get_drvinfo,
 102         .get_link               = ethtool_op_get_link,
 103         .get_strings            = veth_get_strings,
 104         .get_sset_count         = veth_get_sset_count,
 105         .get_ethtool_stats      = veth_get_ethtool_stats,
 106 };
 107
 108 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 109 {
 110         struct veth_priv *priv = netdev_priv(dev);
 111         struct net_device *rcv;
 112         int length = skb->len;
 113
 114         rcu_read_lock();
 115         rcv = rcu_dereference(priv->peer);
 116         if (unlikely(!rcv)) {
 117                 kfree_skb(skb);
 118                 goto drop;
 119         }
 120         /* don't change ip_summed == CHECKSUM_PARTIAL, as that
 121          * will cause bad checksum on forwarded packets
 122          */
 123         if (skb->ip_summed == CHECKSUM_NONE &&
 124             rcv->features & NETIF_F_RXCSUM)
 125                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 126
 127         if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
 128                 struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 129
 130                 u64_stats_update_begin(&stats->syncp);
 131                 stats->bytes += length;
 132                 stats->packets++;
 133                 u64_stats_update_end(&stats->syncp);
 134         } else {
 135 drop:
 136                 atomic64_inc(&priv->dropped);
 137         }
 138         rcu_read_unlock();
 139         return NETDEV_TX_OK;
 140 }
 141
 142 /*
 143  * general routines
 144  */
 145
 146 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 147 {
 148         struct veth_priv *priv = netdev_priv(dev);
 149         int cpu;
 150
 151         result->packets = 0;
 152         result->bytes = 0;
 153         for_each_possible_cpu(cpu) {
 154                 struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
 155                 u64 packets, bytes;
 156                 unsigned int start;
 157
 158                 do {
 159                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 160                         packets = stats->packets;
 161                         bytes = stats->bytes;
 162                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 163                 result->packets += packets;
 164                 result->bytes += bytes;
 165         }
 166         return atomic64_read(&priv->dropped);
 167 }
 168
 169 static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
 170                                                   struct rtnl_link_stats64 *tot)
 171 {
 172         struct veth_priv *priv = netdev_priv(dev);
 173         struct net_device *peer;
 174         struct pcpu_vstats one;
 175
 176         tot->tx_dropped = veth_stats_one(&one, dev);
 177         tot->tx_bytes = one.bytes;
 178         tot->tx_packets = one.packets;
 179
 180         rcu_read_lock();
 181         peer = rcu_dereference(priv->peer);
 182         if (peer) {
 183                 tot->rx_dropped = veth_stats_one(&one, peer);
 184                 tot->rx_bytes = one.bytes;
 185                 tot->rx_packets = one.packets;
 186         }
 187         rcu_read_unlock();
 188
 189         return tot;
 190 }
 191
 192 /* fake multicast ability */
 193 static void veth_set_multicast_list(struct net_device *dev)
 194 {
 195 }
 196
 197 static int veth_open(struct net_device *dev)
 198 {
 199         struct veth_priv *priv = netdev_priv(dev);
 200         struct net_device *peer = rtnl_dereference(priv->peer);
 201
 202         if (!peer)
 203                 return -ENOTCONN;
 204
 205         if (peer->flags & IFF_UP) {
 206                 netif_carrier_on(dev);
 207                 netif_carrier_on(peer);
 208         }
 209         return 0;
 210 }
 211
 212 static int veth_close(struct net_device *dev)
 213 {
 214         struct veth_priv *priv = netdev_priv(dev);
 215         struct net_device *peer = rtnl_dereference(priv->peer);
 216
 217         netif_carrier_off(dev);
 218         if (peer)
 219                 netif_carrier_off(peer);
 220
 221         return 0;
 222 }
 223
 224 static int is_valid_veth_mtu(int new_mtu)
 225 {
 226         return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
 227 }
 228
 229 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 230 {
 231         if (!is_valid_veth_mtu(new_mtu))
 232                 return -EINVAL;
 233         dev->mtu = new_mtu;
 234         return 0;
 235 }
 236
 237 static int veth_dev_init(struct net_device *dev)
 238 {
 239         dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
 240         if (!dev->vstats)
 241                 return -ENOMEM;
 242         return 0;
 243 }
 244
 245 static void veth_dev_free(struct net_device *dev)
 246 {
 247         free_percpu(dev->vstats);
 248         free_netdev(dev);
 249 }
 250
 251 #ifdef CONFIG_NET_POLL_CONTROLLER
 252 static void veth_poll_controller(struct net_device *dev)
 253 {
 254         /* veth only receives frames when its peer sends one
 255          * Since it's a synchronous operation, we are guaranteed
 256          * never to have pending data when we poll for it so
 257          * there is nothing to do here.
 258          *
 259          * We need this though so netpoll recognizes us as an interface that
 260          * supports polling, which enables bridge devices in virt setups to
 261          * still use netconsole
 262          */
 263 }
 264 #endif  /* CONFIG_NET_POLL_CONTROLLER */
 265
 266 static int veth_get_iflink(const struct net_device *dev)
 267 {
 268         struct veth_priv *priv = netdev_priv(dev);
 269         struct net_device *peer;
 270         int iflink;
 271
 272         rcu_read_lock();
 273         peer = rcu_dereference(priv->peer);
 274         iflink = peer ? peer->ifindex : 0;
 275         rcu_read_unlock();
 276
 277         return iflink;
 278 }
 279
 280 static const struct net_device_ops veth_netdev_ops = {
 281         .ndo_init            = veth_dev_init,
 282         .ndo_open            = veth_open,
 283         .ndo_stop            = veth_close,
 284         .ndo_start_xmit      = veth_xmit,
 285         .ndo_change_mtu      = veth_change_mtu,
 286         .ndo_get_stats64     = veth_get_stats64,
 287         .ndo_set_rx_mode     = veth_set_multicast_list,
 288         .ndo_set_mac_address = eth_mac_addr,
 289 #ifdef CONFIG_NET_POLL_CONTROLLER
 290         .ndo_poll_controller    = veth_poll_controller,
 291 #endif
 292         .ndo_get_iflink         = veth_get_iflink,
 293         .ndo_features_check     = passthru_features_check,
 294 };
 295
 296 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO |    \
 297                        NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \
 298                        NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL |           \
 299                        NETIF_F_GSO_IPIP | NETIF_F_GSO_SIT | NETIF_F_UFO |   \
 300                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
 301                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
 302
 303 static void veth_setup(struct net_device *dev)
 304 {
 305         ether_setup(dev);
 306
 307         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 308         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 309         dev->priv_flags |= IFF_NO_QUEUE;
 310
 311         dev->netdev_ops = &veth_netdev_ops;
 312         dev->ethtool_ops = &veth_ethtool_ops;
 313         dev->features |= NETIF_F_LLTX;
 314         dev->features |= VETH_FEATURES;
 315         dev->vlan_features = dev->features &
 316                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 317                                NETIF_F_HW_VLAN_STAG_TX |
 318                                NETIF_F_HW_VLAN_CTAG_RX |
 319                                NETIF_F_HW_VLAN_STAG_RX);
 320         dev->destructor = veth_dev_free;
 321
 322         dev->hw_features = VETH_FEATURES;
 323         dev->hw_enc_features = VETH_FEATURES;
 324 }
 325
 326 /*
 327  * netlink interface
 328  */
 329
 330 static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
 331 {
 332         if (tb[IFLA_ADDRESS]) {
 333                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 334                         return -EINVAL;
 335                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 336                         return -EADDRNOTAVAIL;
 337         }
 338         if (tb[IFLA_MTU]) {
 339                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 340                         return -EINVAL;
 341         }
 342         return 0;
 343 }
 344
 345 static struct rtnl_link_ops veth_link_ops;
 346
 347 static int veth_newlink(struct net *src_net, struct net_device *dev,
 348                          struct nlattr *tb[], struct nlattr *data[])
 349 {
 350         int err;
 351         struct net_device *peer;
 352         struct veth_priv *priv;
 353         char ifname[IFNAMSIZ];
 354         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 355         unsigned char name_assign_type;
 356         struct ifinfomsg *ifmp;
 357         struct net *net;
 358
 359         /*
 360          * create and register peer first
 361          */
 362         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 363                 struct nlattr *nla_peer;
 364
 365                 nla_peer = data[VETH_INFO_PEER];
 366                 ifmp = nla_data(nla_peer);
 367                 err = rtnl_nla_parse_ifla(peer_tb,
 368                                           nla_data(nla_peer) + sizeof(struct ifinfomsg),
 369                                           nla_len(nla_peer) - sizeof(struct ifinfomsg));
 370                 if (err < 0)
 371                         return err;
 372
 373                 err = veth_validate(peer_tb, NULL);
 374                 if (err < 0)
 375                         return err;
 376
 377                 tbp = peer_tb;
 378         } else {
 379                 ifmp = NULL;
 380                 tbp = tb;
 381         }
 382
 383         if (tbp[IFLA_IFNAME]) {
 384                 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 385                 name_assign_type = NET_NAME_USER;
 386         } else {
 387                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 388                 name_assign_type = NET_NAME_ENUM;
 389         }
 390
 391         net = rtnl_link_get_net(src_net, tbp);
 392         if (IS_ERR(net))
 393                 return PTR_ERR(net);
 394
 395         peer = rtnl_create_link(net, ifname, name_assign_type,
 396                                 &veth_link_ops, tbp);
 397         if (IS_ERR(peer)) {
 398                 put_net(net);
 399                 return PTR_ERR(peer);
 400         }
 401
 402         if (tbp[IFLA_ADDRESS] == NULL)
 403                 eth_hw_addr_random(peer);
 404
 405         if (ifmp && (dev->ifindex != 0))
 406                 peer->ifindex = ifmp->ifi_index;
 407
 408         err = register_netdevice(peer);
 409         put_net(net);
 410         net = NULL;
 411         if (err < 0)
 412                 goto err_register_peer;
 413
 414         netif_carrier_off(peer);
 415
 416         err = rtnl_configure_link(peer, ifmp);
 417         if (err < 0)
 418                 goto err_configure_peer;
 419
 420         /*
 421          * register dev last
 422          *
 423          * note, that since we've registered new device the dev's name
 424          * should be re-allocated
 425          */
 426
 427         if (tb[IFLA_ADDRESS] == NULL)
 428                 eth_hw_addr_random(dev);
 429
 430         if (tb[IFLA_IFNAME])
 431                 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 432         else
 433                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 434
 435         err = register_netdevice(dev);
 436         if (err < 0)
 437                 goto err_register_dev;
 438
 439         netif_carrier_off(dev);
 440
 441         /*
 442          * tie the deviced together
 443          */
 444
 445         priv = netdev_priv(dev);
 446         rcu_assign_pointer(priv->peer, peer);
 447
 448         priv = netdev_priv(peer);
 449         rcu_assign_pointer(priv->peer, dev);
 450         return 0;
 451
 452 err_register_dev:
 453         /* nothing to do */
 454 err_configure_peer:
 455         unregister_netdevice(peer);
 456         return err;
 457
 458 err_register_peer:
 459         free_netdev(peer);
 460         return err;
 461 }
 462
 463 static void veth_dellink(struct net_device *dev, struct list_head *head)
 464 {
 465         struct veth_priv *priv;
 466         struct net_device *peer;
 467
 468         priv = netdev_priv(dev);
 469         peer = rtnl_dereference(priv->peer);
 470
 471         /* Note : dellink() is called from default_device_exit_batch(),
 472          * before a rcu_synchronize() point. The devices are guaranteed
 473          * not being freed before one RCU grace period.
 474          */
 475         RCU_INIT_POINTER(priv->peer, NULL);
 476         unregister_netdevice_queue(dev, head);
 477
 478         if (peer) {
 479                 priv = netdev_priv(peer);
 480                 RCU_INIT_POINTER(priv->peer, NULL);
 481                 unregister_netdevice_queue(peer, head);
 482         }
 483 }
 484
 485 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 486         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 487 };
 488
 489 static struct net *veth_get_link_net(const struct net_device *dev)
 490 {
 491         struct veth_priv *priv = netdev_priv(dev);
 492         struct net_device *peer = rtnl_dereference(priv->peer);
 493
 494         return peer ? dev_net(peer) : dev_net(dev);
 495 }
 496
 497 static struct rtnl_link_ops veth_link_ops = {
 498         .kind           = DRV_NAME,
 499         .priv_size      = sizeof(struct veth_priv),
 500         .setup          = veth_setup,
 501         .validate       = veth_validate,
 502         .newlink        = veth_newlink,
 503         .dellink        = veth_dellink,
 504         .policy         = veth_policy,
 505         .maxtype        = VETH_INFO_MAX,
 506         .get_link_net   = veth_get_link_net,
 507 };
 508
 509 /*
 510  * init/fini
 511  */
 512
 513 static __init int veth_init(void)
 514 {
 515         return rtnl_link_register(&veth_link_ops);
 516 }
 517
 518 static __exit void veth_exit(void)
 519 {
 520         rtnl_link_unregister(&veth_link_ops);
 521 }
 522
 523 module_init(veth_init);
 524 module_exit(veth_exit);
 525
 526 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 527 MODULE_LICENSE("GPL v2");
 528 MODULE_ALIAS_RTNL_LINK(DRV_NAME);