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
 121         if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
 122                 struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 123
 124                 u64_stats_update_begin(&stats->syncp);
 125                 stats->bytes += length;
 126                 stats->packets++;
 127                 u64_stats_update_end(&stats->syncp);
 128         } else {
 129 drop:
 130                 atomic64_inc(&priv->dropped);
 131         }
 132         rcu_read_unlock();
 133         return NETDEV_TX_OK;
 134 }
 135
 136 /*
 137  * general routines
 138  */
 139
 140 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 141 {
 142         struct veth_priv *priv = netdev_priv(dev);
 143         int cpu;
 144
 145         result->packets = 0;
 146         result->bytes = 0;
 147         for_each_possible_cpu(cpu) {
 148                 struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
 149                 u64 packets, bytes;
 150                 unsigned int start;
 151
 152                 do {
 153                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 154                         packets = stats->packets;
 155                         bytes = stats->bytes;
 156                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 157                 result->packets += packets;
 158                 result->bytes += bytes;
 159         }
 160         return atomic64_read(&priv->dropped);
 161 }
 162
 163 static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
 164                                                   struct rtnl_link_stats64 *tot)
 165 {
 166         struct veth_priv *priv = netdev_priv(dev);
 167         struct net_device *peer;
 168         struct pcpu_vstats one;
 169
 170         tot->tx_dropped = veth_stats_one(&one, dev);
 171         tot->tx_bytes = one.bytes;
 172         tot->tx_packets = one.packets;
 173
 174         rcu_read_lock();
 175         peer = rcu_dereference(priv->peer);
 176         if (peer) {
 177                 tot->rx_dropped = veth_stats_one(&one, peer);
 178                 tot->rx_bytes = one.bytes;
 179                 tot->rx_packets = one.packets;
 180         }
 181         rcu_read_unlock();
 182
 183         return tot;
 184 }
 185
 186 /* fake multicast ability */
 187 static void veth_set_multicast_list(struct net_device *dev)
 188 {
 189 }
 190
 191 static int veth_open(struct net_device *dev)
 192 {
 193         struct veth_priv *priv = netdev_priv(dev);
 194         struct net_device *peer = rtnl_dereference(priv->peer);
 195
 196         if (!peer)
 197                 return -ENOTCONN;
 198
 199         if (peer->flags & IFF_UP) {
 200                 netif_carrier_on(dev);
 201                 netif_carrier_on(peer);
 202         }
 203         return 0;
 204 }
 205
 206 static int veth_close(struct net_device *dev)
 207 {
 208         struct veth_priv *priv = netdev_priv(dev);
 209         struct net_device *peer = rtnl_dereference(priv->peer);
 210
 211         netif_carrier_off(dev);
 212         if (peer)
 213                 netif_carrier_off(peer);
 214
 215         return 0;
 216 }
 217
 218 static int is_valid_veth_mtu(int new_mtu)
 219 {
 220         return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
 221 }
 222
 223 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 224 {
 225         if (!is_valid_veth_mtu(new_mtu))
 226                 return -EINVAL;
 227         dev->mtu = new_mtu;
 228         return 0;
 229 }
 230
 231 static int veth_dev_init(struct net_device *dev)
 232 {
 233         dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
 234         if (!dev->vstats)
 235                 return -ENOMEM;
 236         return 0;
 237 }
 238
 239 static void veth_dev_free(struct net_device *dev)
 240 {
 241         free_percpu(dev->vstats);
 242         free_netdev(dev);
 243 }
 244
 245 static const struct net_device_ops veth_netdev_ops = {
 246         .ndo_init            = veth_dev_init,
 247         .ndo_open            = veth_open,
 248         .ndo_stop            = veth_close,
 249         .ndo_start_xmit      = veth_xmit,
 250         .ndo_change_mtu      = veth_change_mtu,
 251         .ndo_get_stats64     = veth_get_stats64,
 252         .ndo_set_rx_mode     = veth_set_multicast_list,
 253         .ndo_set_mac_address = eth_mac_addr,
 254 };
 255
 256 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO |    \
 257                        NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \
 258                        NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL |           \
 259                        NETIF_F_GSO_IPIP | NETIF_F_GSO_SIT | NETIF_F_UFO |   \
 260                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
 261                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
 262
 263 static void veth_setup(struct net_device *dev)
 264 {
 265         ether_setup(dev);
 266
 267         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 268         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 269
 270         dev->netdev_ops = &veth_netdev_ops;
 271         dev->ethtool_ops = &veth_ethtool_ops;
 272         dev->features |= NETIF_F_LLTX;
 273         dev->features |= VETH_FEATURES;
 274         dev->vlan_features = dev->features &
 275                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 276                                NETIF_F_HW_VLAN_STAG_TX |
 277                                NETIF_F_HW_VLAN_CTAG_RX |
 278                                NETIF_F_HW_VLAN_STAG_RX);
 279         dev->destructor = veth_dev_free;
 280
 281         dev->hw_features = VETH_FEATURES;
 282         dev->hw_enc_features = VETH_FEATURES;
 283 }
 284
 285 /*
 286  * netlink interface
 287  */
 288
 289 static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
 290 {
 291         if (tb[IFLA_ADDRESS]) {
 292                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 293                         return -EINVAL;
 294                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 295                         return -EADDRNOTAVAIL;
 296         }
 297         if (tb[IFLA_MTU]) {
 298                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 299                         return -EINVAL;
 300         }
 301         return 0;
 302 }
 303
 304 static struct rtnl_link_ops veth_link_ops;
 305
 306 static int veth_newlink(struct net *src_net, struct net_device *dev,
 307                          struct nlattr *tb[], struct nlattr *data[])
 308 {
 309         int err;
 310         struct net_device *peer;
 311         struct veth_priv *priv;
 312         char ifname[IFNAMSIZ];
 313         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 314         struct ifinfomsg *ifmp;
 315         struct net *net;
 316
 317         /*
 318          * create and register peer first
 319          */
 320         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 321                 struct nlattr *nla_peer;
 322
 323                 nla_peer = data[VETH_INFO_PEER];
 324                 ifmp = nla_data(nla_peer);
 325                 err = rtnl_nla_parse_ifla(peer_tb,
 326                                           nla_data(nla_peer) + sizeof(struct ifinfomsg),
 327                                           nla_len(nla_peer) - sizeof(struct ifinfomsg));
 328                 if (err < 0)
 329                         return err;
 330
 331                 err = veth_validate(peer_tb, NULL);
 332                 if (err < 0)
 333                         return err;
 334
 335                 tbp = peer_tb;
 336         } else {
 337                 ifmp = NULL;
 338                 tbp = tb;
 339         }
 340
 341         if (tbp[IFLA_IFNAME])
 342                 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 343         else
 344                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 345
 346         net = rtnl_link_get_net(src_net, tbp);
 347         if (IS_ERR(net))
 348                 return PTR_ERR(net);
 349
 350         peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp);
 351         if (IS_ERR(peer)) {
 352                 put_net(net);
 353                 return PTR_ERR(peer);
 354         }
 355
 356         if (tbp[IFLA_ADDRESS] == NULL)
 357                 eth_hw_addr_random(peer);
 358
 359         if (ifmp && (dev->ifindex != 0))
 360                 peer->ifindex = ifmp->ifi_index;
 361
 362         err = register_netdevice(peer);
 363         put_net(net);
 364         net = NULL;
 365         if (err < 0)
 366                 goto err_register_peer;
 367
 368         netif_carrier_off(peer);
 369
 370         err = rtnl_configure_link(peer, ifmp);
 371         if (err < 0)
 372                 goto err_configure_peer;
 373
 374         /*
 375          * register dev last
 376          *
 377          * note, that since we've registered new device the dev's name
 378          * should be re-allocated
 379          */
 380
 381         if (tb[IFLA_ADDRESS] == NULL)
 382                 eth_hw_addr_random(dev);
 383
 384         if (tb[IFLA_IFNAME])
 385                 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 386         else
 387                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 388
 389         err = register_netdevice(dev);
 390         if (err < 0)
 391                 goto err_register_dev;
 392
 393         netif_carrier_off(dev);
 394
 395         /*
 396          * tie the deviced together
 397          */
 398
 399         priv = netdev_priv(dev);
 400         rcu_assign_pointer(priv->peer, peer);
 401
 402         priv = netdev_priv(peer);
 403         rcu_assign_pointer(priv->peer, dev);
 404         return 0;
 405
 406 err_register_dev:
 407         /* nothing to do */
 408 err_configure_peer:
 409         unregister_netdevice(peer);
 410         return err;
 411
 412 err_register_peer:
 413         free_netdev(peer);
 414         return err;
 415 }
 416
 417 static void veth_dellink(struct net_device *dev, struct list_head *head)
 418 {
 419         struct veth_priv *priv;
 420         struct net_device *peer;
 421
 422         priv = netdev_priv(dev);
 423         peer = rtnl_dereference(priv->peer);
 424
 425         /* Note : dellink() is called from default_device_exit_batch(),
 426          * before a rcu_synchronize() point. The devices are guaranteed
 427          * not being freed before one RCU grace period.
 428          */
 429         RCU_INIT_POINTER(priv->peer, NULL);
 430         unregister_netdevice_queue(dev, head);
 431
 432         if (peer) {
 433                 priv = netdev_priv(peer);
 434                 RCU_INIT_POINTER(priv->peer, NULL);
 435                 unregister_netdevice_queue(peer, head);
 436         }
 437 }
 438
 439 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 440         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 441 };
 442
 443 static struct rtnl_link_ops veth_link_ops = {
 444         .kind           = DRV_NAME,
 445         .priv_size      = sizeof(struct veth_priv),
 446         .setup          = veth_setup,
 447         .validate       = veth_validate,
 448         .newlink        = veth_newlink,
 449         .dellink        = veth_dellink,
 450         .policy         = veth_policy,
 451         .maxtype        = VETH_INFO_MAX,
 452 };
 453
 454 /*
 455  * init/fini
 456  */
 457
 458 static __init int veth_init(void)
 459 {
 460         return rtnl_link_register(&veth_link_ops);
 461 }
 462
 463 static __exit void veth_exit(void)
 464 {
 465         rtnl_link_unregister(&veth_link_ops);
 466 }
 467
 468 module_init(veth_init);
 469 module_exit(veth_exit);
 470
 471 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 472 MODULE_LICENSE("GPL v2");
 473 MODULE_ALIAS_RTNL_LINK(DRV_NAME);