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       *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;
  93
  94         priv = netdev_priv(dev);
  95         data[0] = priv->peer->ifindex;
  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 = priv->peer;
 111         int length = skb->len;
 112
 113         /* don't change ip_summed == CHECKSUM_PARTIAL, as that
 114          * will cause bad checksum on forwarded packets
 115          */
 116         if (skb->ip_summed == CHECKSUM_NONE &&
 117             rcv->features & NETIF_F_RXCSUM)
 118                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 119
 120         if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
 121                 struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 122
 123                 u64_stats_update_begin(&stats->syncp);
 124                 stats->bytes += length;
 125                 stats->packets++;
 126                 u64_stats_update_end(&stats->syncp);
 127         } else {
 128                 atomic64_inc(&priv->dropped);
 129         }
 130
 131         return NETDEV_TX_OK;
 132 }
 133
 134 /*
 135  * general routines
 136  */
 137
 138 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 139 {
 140         struct veth_priv *priv = netdev_priv(dev);
 141         int cpu;
 142
 143         result->packets = 0;
 144         result->bytes = 0;
 145         for_each_possible_cpu(cpu) {
 146                 struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
 147                 u64 packets, bytes;
 148                 unsigned int start;
 149
 150                 do {
 151                         start = u64_stats_fetch_begin_bh(&stats->syncp);
 152                         packets = stats->packets;
 153                         bytes = stats->bytes;
 154                 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
 155                 result->packets += packets;
 156                 result->bytes += bytes;
 157         }
 158         return atomic64_read(&priv->dropped);
 159 }
 160
 161 static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
 162                                                   struct rtnl_link_stats64 *tot)
 163 {
 164         struct veth_priv *priv = netdev_priv(dev);
 165         struct pcpu_vstats one;
 166
 167         tot->tx_dropped = veth_stats_one(&one, dev);
 168         tot->tx_bytes = one.bytes;
 169         tot->tx_packets = one.packets;
 170
 171         tot->rx_dropped = veth_stats_one(&one, priv->peer);
 172         tot->rx_bytes = one.bytes;
 173         tot->rx_packets = one.packets;
 174
 175         return tot;
 176 }
 177
 178 static int veth_open(struct net_device *dev)
 179 {
 180         struct veth_priv *priv;
 181
 182         priv = netdev_priv(dev);
 183         if (priv->peer == NULL)
 184                 return -ENOTCONN;
 185
 186         if (priv->peer->flags & IFF_UP) {
 187                 netif_carrier_on(dev);
 188                 netif_carrier_on(priv->peer);
 189         }
 190         return 0;
 191 }
 192
 193 static int veth_close(struct net_device *dev)
 194 {
 195         struct veth_priv *priv = netdev_priv(dev);
 196
 197         netif_carrier_off(dev);
 198         netif_carrier_off(priv->peer);
 199
 200         return 0;
 201 }
 202
 203 static int is_valid_veth_mtu(int new_mtu)
 204 {
 205         return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
 206 }
 207
 208 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 209 {
 210         if (!is_valid_veth_mtu(new_mtu))
 211                 return -EINVAL;
 212         dev->mtu = new_mtu;
 213         return 0;
 214 }
 215
 216 static int veth_dev_init(struct net_device *dev)
 217 {
 218         dev->vstats = alloc_percpu(struct pcpu_vstats);
 219         if (!dev->vstats)
 220                 return -ENOMEM;
 221
 222         return 0;
 223 }
 224
 225 static void veth_dev_free(struct net_device *dev)
 226 {
 227         free_percpu(dev->vstats);
 228         free_netdev(dev);
 229 }
 230
 231 static const struct net_device_ops veth_netdev_ops = {
 232         .ndo_init            = veth_dev_init,
 233         .ndo_open            = veth_open,
 234         .ndo_stop            = veth_close,
 235         .ndo_start_xmit      = veth_xmit,
 236         .ndo_change_mtu      = veth_change_mtu,
 237         .ndo_get_stats64     = veth_get_stats64,
 238         .ndo_set_mac_address = eth_mac_addr,
 239 };
 240
 241 static void veth_setup(struct net_device *dev)
 242 {
 243         ether_setup(dev);
 244
 245         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 246         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 247
 248         dev->netdev_ops = &veth_netdev_ops;
 249         dev->ethtool_ops = &veth_ethtool_ops;
 250         dev->features |= NETIF_F_LLTX;
 251         dev->destructor = veth_dev_free;
 252
 253         dev->hw_features = NETIF_F_HW_CSUM | NETIF_F_SG | NETIF_F_RXCSUM;
 254 }
 255
 256 /*
 257  * netlink interface
 258  */
 259
 260 static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
 261 {
 262         if (tb[IFLA_ADDRESS]) {
 263                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 264                         return -EINVAL;
 265                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 266                         return -EADDRNOTAVAIL;
 267         }
 268         if (tb[IFLA_MTU]) {
 269                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 270                         return -EINVAL;
 271         }
 272         return 0;
 273 }
 274
 275 static struct rtnl_link_ops veth_link_ops;
 276
 277 static int veth_newlink(struct net *src_net, struct net_device *dev,
 278                          struct nlattr *tb[], struct nlattr *data[])
 279 {
 280         int err;
 281         struct net_device *peer;
 282         struct veth_priv *priv;
 283         char ifname[IFNAMSIZ];
 284         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 285         struct ifinfomsg *ifmp;
 286         struct net *net;
 287
 288         /*
 289          * create and register peer first
 290          */
 291         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 292                 struct nlattr *nla_peer;
 293
 294                 nla_peer = data[VETH_INFO_PEER];
 295                 ifmp = nla_data(nla_peer);
 296                 err = nla_parse(peer_tb, IFLA_MAX,
 297                                 nla_data(nla_peer) + sizeof(struct ifinfomsg),
 298                                 nla_len(nla_peer) - sizeof(struct ifinfomsg),
 299                                 ifla_policy);
 300                 if (err < 0)
 301                         return err;
 302
 303                 err = veth_validate(peer_tb, NULL);
 304                 if (err < 0)
 305                         return err;
 306
 307                 tbp = peer_tb;
 308         } else {
 309                 ifmp = NULL;
 310                 tbp = tb;
 311         }
 312
 313         if (tbp[IFLA_IFNAME])
 314                 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 315         else
 316                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 317
 318         net = rtnl_link_get_net(src_net, tbp);
 319         if (IS_ERR(net))
 320                 return PTR_ERR(net);
 321
 322         peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp);
 323         if (IS_ERR(peer)) {
 324                 put_net(net);
 325                 return PTR_ERR(peer);
 326         }
 327
 328         if (tbp[IFLA_ADDRESS] == NULL)
 329                 eth_hw_addr_random(peer);
 330
 331         if (ifmp && (dev->ifindex != 0))
 332                 peer->ifindex = ifmp->ifi_index;
 333
 334         err = register_netdevice(peer);
 335         put_net(net);
 336         net = NULL;
 337         if (err < 0)
 338                 goto err_register_peer;
 339
 340         netif_carrier_off(peer);
 341
 342         err = rtnl_configure_link(peer, ifmp);
 343         if (err < 0)
 344                 goto err_configure_peer;
 345
 346         /*
 347          * register dev last
 348          *
 349          * note, that since we've registered new device the dev's name
 350          * should be re-allocated
 351          */
 352
 353         if (tb[IFLA_ADDRESS] == NULL)
 354                 eth_hw_addr_random(dev);
 355
 356         if (tb[IFLA_IFNAME])
 357                 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 358         else
 359                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 360
 361         if (strchr(dev->name, '%')) {
 362                 err = dev_alloc_name(dev, dev->name);
 363                 if (err < 0)
 364                         goto err_alloc_name;
 365         }
 366
 367         err = register_netdevice(dev);
 368         if (err < 0)
 369                 goto err_register_dev;
 370
 371         netif_carrier_off(dev);
 372
 373         /*
 374          * tie the deviced together
 375          */
 376
 377         priv = netdev_priv(dev);
 378         priv->peer = peer;
 379
 380         priv = netdev_priv(peer);
 381         priv->peer = dev;
 382         return 0;
 383
 384 err_register_dev:
 385         /* nothing to do */
 386 err_alloc_name:
 387 err_configure_peer:
 388         unregister_netdevice(peer);
 389         return err;
 390
 391 err_register_peer:
 392         free_netdev(peer);
 393         return err;
 394 }
 395
 396 static void veth_dellink(struct net_device *dev, struct list_head *head)
 397 {
 398         struct veth_priv *priv;
 399         struct net_device *peer;
 400
 401         priv = netdev_priv(dev);
 402         peer = priv->peer;
 403
 404         unregister_netdevice_queue(dev, head);
 405         unregister_netdevice_queue(peer, head);
 406 }
 407
 408 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 409         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 410 };
 411
 412 static struct rtnl_link_ops veth_link_ops = {
 413         .kind           = DRV_NAME,
 414         .priv_size      = sizeof(struct veth_priv),
 415         .setup          = veth_setup,
 416         .validate       = veth_validate,
 417         .newlink        = veth_newlink,
 418         .dellink        = veth_dellink,
 419         .policy         = veth_policy,
 420         .maxtype        = VETH_INFO_MAX,
 421 };
 422
 423 /*
 424  * init/fini
 425  */
 426
 427 static __init int veth_init(void)
 428 {
 429         return rtnl_link_register(&veth_link_ops);
 430 }
 431
 432 static __exit void veth_exit(void)
 433 {
 434         rtnl_link_unregister(&veth_link_ops);
 435 }
 436
 437 module_init(veth_init);
 438 module_exit(veth_exit);
 439
 440 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 441 MODULE_LICENSE("GPL v2");
 442 MODULE_ALIAS_RTNL_LINK(DRV_NAME);