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/dst.h>
  18 #include <net/xfrm.h>
  19 #include <linux/veth.h>
  20 #include <linux/module.h>
  21
  22 #define DRV_NAME        "veth"
  23 #define DRV_VERSION     "1.0"
  24
  25 #define MIN_MTU 68              /* Min L3 MTU */
  26 #define MAX_MTU 65535           /* Max L3 MTU (arbitrary) */
  27
  28 struct pcpu_vstats {
  29         u64                     packets;
  30         u64                     bytes;
  31         struct u64_stats_sync   syncp;
  32 };
  33
  34 struct veth_priv {
  35         struct net_device __rcu *peer;
  36         atomic64_t              dropped;
  37 };
  38
  39 /*
  40  * ethtool interface
  41  */
  42
  43 static struct {
  44         const char string[ETH_GSTRING_LEN];
  45 } ethtool_stats_keys[] = {
  46         { "peer_ifindex" },
  47 };
  48
  49 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
  50 {
  51         cmd->supported          = 0;
  52         cmd->advertising        = 0;
  53         ethtool_cmd_speed_set(cmd, SPEED_10000);
  54         cmd->duplex             = DUPLEX_FULL;
  55         cmd->port               = PORT_TP;
  56         cmd->phy_address        = 0;
  57         cmd->transceiver        = XCVR_INTERNAL;
  58         cmd->autoneg            = AUTONEG_DISABLE;
  59         cmd->maxtxpkt           = 0;
  60         cmd->maxrxpkt           = 0;
  61         return 0;
  62 }
  63
  64 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
  65 {
  66         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
  67         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
  68 }
  69
  70 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
  71 {
  72         switch(stringset) {
  73         case ETH_SS_STATS:
  74                 memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
  75                 break;
  76         }
  77 }
  78
  79 static int veth_get_sset_count(struct net_device *dev, int sset)
  80 {
  81         switch (sset) {
  82         case ETH_SS_STATS:
  83                 return ARRAY_SIZE(ethtool_stats_keys);
  84         default:
  85                 return -EOPNOTSUPP;
  86         }
  87 }
  88
  89 static void veth_get_ethtool_stats(struct net_device *dev,
  90                 struct ethtool_stats *stats, u64 *data)
  91 {
  92         struct veth_priv *priv = netdev_priv(dev);
  93         struct net_device *peer = rtnl_dereference(priv->peer);
  94
  95         data[0] = peer ? peer->ifindex : 0;
  96 }
  97
  98 static const struct ethtool_ops veth_ethtool_ops = {
  99         .get_settings           = veth_get_settings,
 100         .get_drvinfo            = veth_get_drvinfo,
 101         .get_link               = ethtool_op_get_link,
 102         .get_strings            = veth_get_strings,
 103         .get_sset_count         = veth_get_sset_count,
 104         .get_ethtool_stats      = veth_get_ethtool_stats,
 105 };
 106
 107 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 108 {
 109         struct veth_priv *priv = netdev_priv(dev);
 110         struct net_device *rcv;
 111         int length = skb->len;
 112
 113         rcu_read_lock();
 114         rcv = rcu_dereference(priv->peer);
 115         if (unlikely(!rcv)) {
 116                 kfree_skb(skb);
 117                 goto drop;
 118         }
 119         /* don't change ip_summed == CHECKSUM_PARTIAL, as that
 120          * will cause bad checksum on forwarded packets
 121          */
 122         if (skb->ip_summed == CHECKSUM_NONE &&
 123             rcv->features & NETIF_F_RXCSUM)
 124                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 125
 126         if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
 127                 struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 128
 129                 u64_stats_update_begin(&stats->syncp);
 130                 stats->bytes += length;
 131                 stats->packets++;
 132                 u64_stats_update_end(&stats->syncp);
 133         } else {
 134 drop:
 135                 atomic64_inc(&priv->dropped);
 136         }
 137         rcu_read_unlock();
 138         return NETDEV_TX_OK;
 139 }
 140
 141 /*
 142  * general routines
 143  */
 144
 145 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 146 {
 147         struct veth_priv *priv = netdev_priv(dev);
 148         int cpu;
 149
 150         result->packets = 0;
 151         result->bytes = 0;
 152         for_each_possible_cpu(cpu) {
 153                 struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
 154                 u64 packets, bytes;
 155                 unsigned int start;
 156
 157                 do {
 158                         start = u64_stats_fetch_begin_bh(&stats->syncp);
 159                         packets = stats->packets;
 160                         bytes = stats->bytes;
 161                 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
 162                 result->packets += packets;
 163                 result->bytes += bytes;
 164         }
 165         return atomic64_read(&priv->dropped);
 166 }
 167
 168 static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
 169                                                   struct rtnl_link_stats64 *tot)
 170 {
 171         struct veth_priv *priv = netdev_priv(dev);
 172         struct net_device *peer;
 173         struct pcpu_vstats one;
 174
 175         tot->tx_dropped = veth_stats_one(&one, dev);
 176         tot->tx_bytes = one.bytes;
 177         tot->tx_packets = one.packets;
 178
 179         rcu_read_lock();
 180         peer = rcu_dereference(priv->peer);
 181         if (peer) {
 182                 tot->rx_dropped = veth_stats_one(&one, peer);
 183                 tot->rx_bytes = one.bytes;
 184                 tot->rx_packets = one.packets;
 185         }
 186         rcu_read_unlock();
 187
 188         return tot;
 189 }
 190
 191 /* fake multicast ability */
 192 static void veth_set_multicast_list(struct net_device *dev)
 193 {
 194 }
 195
 196 static int veth_open(struct net_device *dev)
 197 {
 198         struct veth_priv *priv = netdev_priv(dev);
 199         struct net_device *peer = rtnl_dereference(priv->peer);
 200
 201         if (!peer)
 202                 return -ENOTCONN;
 203
 204         if (peer->flags & IFF_UP) {
 205                 netif_carrier_on(dev);
 206                 netif_carrier_on(peer);
 207         }
 208         return 0;
 209 }
 210
 211 static int veth_close(struct net_device *dev)
 212 {
 213         struct veth_priv *priv = netdev_priv(dev);
 214         struct net_device *peer = rtnl_dereference(priv->peer);
 215
 216         netif_carrier_off(dev);
 217         if (peer)
 218                 netif_carrier_off(peer);
 219
 220         return 0;
 221 }
 222
 223 static int is_valid_veth_mtu(int new_mtu)
 224 {
 225         return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
 226 }
 227
 228 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 229 {
 230         if (!is_valid_veth_mtu(new_mtu))
 231                 return -EINVAL;
 232         dev->mtu = new_mtu;
 233         return 0;
 234 }
 235
 236 static int veth_dev_init(struct net_device *dev)
 237 {
 238         dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
 239         if (!dev->vstats)
 240                 return -ENOMEM;
 241         return 0;
 242 }
 243
 244 static void veth_dev_free(struct net_device *dev)
 245 {
 246         free_percpu(dev->vstats);
 247         free_netdev(dev);
 248 }
 249
 250 static const struct net_device_ops veth_netdev_ops = {
 251         .ndo_init            = veth_dev_init,
 252         .ndo_open            = veth_open,
 253         .ndo_stop            = veth_close,
 254         .ndo_start_xmit      = veth_xmit,
 255         .ndo_change_mtu      = veth_change_mtu,
 256         .ndo_get_stats64     = veth_get_stats64,
 257         .ndo_set_rx_mode     = veth_set_multicast_list,
 258         .ndo_set_mac_address = eth_mac_addr,
 259 };
 260
 261 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO |    \
 262                        NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \
 263                        NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL |           \
 264                        NETIF_F_GSO_IPIP | NETIF_F_GSO_SIT | NETIF_F_UFO |   \
 265                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
 266                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
 267
 268 static void veth_setup(struct net_device *dev)
 269 {
 270         ether_setup(dev);
 271
 272         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 273         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 274
 275         dev->netdev_ops = &veth_netdev_ops;
 276         dev->ethtool_ops = &veth_ethtool_ops;
 277         dev->features |= NETIF_F_LLTX;
 278         dev->features |= VETH_FEATURES;
 279         dev->vlan_features = dev->features;
 280         dev->destructor = veth_dev_free;
 281
 282         dev->hw_features = VETH_FEATURES;
 283         dev->hw_enc_features = VETH_FEATURES;
 284 }
 285
 286 /*
 287  * netlink interface
 288  */
 289
 290 static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
 291 {
 292         if (tb[IFLA_ADDRESS]) {
 293                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 294                         return -EINVAL;
 295                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 296                         return -EADDRNOTAVAIL;
 297         }
 298         if (tb[IFLA_MTU]) {
 299                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 300                         return -EINVAL;
 301         }
 302         return 0;
 303 }
 304
 305 static struct rtnl_link_ops veth_link_ops;
 306
 307 static int veth_newlink(struct net *src_net, struct net_device *dev,
 308                          struct nlattr *tb[], struct nlattr *data[])
 309 {
 310         int err;
 311         struct net_device *peer;
 312         struct veth_priv *priv;
 313         char ifname[IFNAMSIZ];
 314         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 315         struct ifinfomsg *ifmp;
 316         struct net *net;
 317
 318         /*
 319          * create and register peer first
 320          */
 321         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 322                 struct nlattr *nla_peer;
 323
 324                 nla_peer = data[VETH_INFO_PEER];
 325                 ifmp = nla_data(nla_peer);
 326                 err = nla_parse(peer_tb, IFLA_MAX,
 327                                 nla_data(nla_peer) + sizeof(struct ifinfomsg),
 328                                 nla_len(nla_peer) - sizeof(struct ifinfomsg),
 329                                 ifla_policy);
 330                 if (err < 0)
 331                         return err;
 332
 333                 err = veth_validate(peer_tb, NULL);
 334                 if (err < 0)
 335                         return err;
 336
 337                 tbp = peer_tb;
 338         } else {
 339                 ifmp = NULL;
 340                 tbp = tb;
 341         }
 342
 343         if (tbp[IFLA_IFNAME])
 344                 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 345         else
 346                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 347
 348         net = rtnl_link_get_net(src_net, tbp);
 349         if (IS_ERR(net))
 350                 return PTR_ERR(net);
 351
 352         peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp);
 353         if (IS_ERR(peer)) {
 354                 put_net(net);
 355                 return PTR_ERR(peer);
 356         }
 357
 358         if (tbp[IFLA_ADDRESS] == NULL)
 359                 eth_hw_addr_random(peer);
 360
 361         if (ifmp && (dev->ifindex != 0))
 362                 peer->ifindex = ifmp->ifi_index;
 363
 364         err = register_netdevice(peer);
 365         put_net(net);
 366         net = NULL;
 367         if (err < 0)
 368                 goto err_register_peer;
 369
 370         netif_carrier_off(peer);
 371
 372         err = rtnl_configure_link(peer, ifmp);
 373         if (err < 0)
 374                 goto err_configure_peer;
 375
 376         /*
 377          * register dev last
 378          *
 379          * note, that since we've registered new device the dev's name
 380          * should be re-allocated
 381          */
 382
 383         if (tb[IFLA_ADDRESS] == NULL)
 384                 eth_hw_addr_random(dev);
 385
 386         if (tb[IFLA_IFNAME])
 387                 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 388         else
 389                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 390
 391         err = register_netdevice(dev);
 392         if (err < 0)
 393                 goto err_register_dev;
 394
 395         netif_carrier_off(dev);
 396
 397         /*
 398          * tie the deviced together
 399          */
 400
 401         priv = netdev_priv(dev);
 402         rcu_assign_pointer(priv->peer, peer);
 403
 404         priv = netdev_priv(peer);
 405         rcu_assign_pointer(priv->peer, dev);
 406         return 0;
 407
 408 err_register_dev:
 409         /* nothing to do */
 410 err_configure_peer:
 411         unregister_netdevice(peer);
 412         return err;
 413
 414 err_register_peer:
 415         free_netdev(peer);
 416         return err;
 417 }
 418
 419 static void veth_dellink(struct net_device *dev, struct list_head *head)
 420 {
 421         struct veth_priv *priv;
 422         struct net_device *peer;
 423
 424         priv = netdev_priv(dev);
 425         peer = rtnl_dereference(priv->peer);
 426
 427         /* Note : dellink() is called from default_device_exit_batch(),
 428          * before a rcu_synchronize() point. The devices are guaranteed
 429          * not being freed before one RCU grace period.
 430          */
 431         RCU_INIT_POINTER(priv->peer, NULL);
 432         unregister_netdevice_queue(dev, head);
 433
 434         if (peer) {
 435                 priv = netdev_priv(peer);
 436                 RCU_INIT_POINTER(priv->peer, NULL);
 437                 unregister_netdevice_queue(peer, head);
 438         }
 439 }
 440
 441 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 442         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 443 };
 444
 445 static struct rtnl_link_ops veth_link_ops = {
 446         .kind           = DRV_NAME,
 447         .priv_size      = sizeof(struct veth_priv),
 448         .setup          = veth_setup,
 449         .validate       = veth_validate,
 450         .newlink        = veth_newlink,
 451         .dellink        = veth_dellink,
 452         .policy         = veth_policy,
 453         .maxtype        = VETH_INFO_MAX,
 454 };
 455
 456 /*
 457  * init/fini
 458  */
 459
 460 static __init int veth_init(void)
 461 {
 462         return rtnl_link_register(&veth_link_ops);
 463 }
 464
 465 static __exit void veth_exit(void)
 466 {
 467         rtnl_link_unregister(&veth_link_ops);
 468 }
 469
 470 module_init(veth_init);
 471 module_exit(veth_exit);
 472
 473 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 474 MODULE_LICENSE("GPL v2");
 475 MODULE_ALIAS_RTNL_LINK(DRV_NAME);