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 struct pcpu_vstats {
  27         u64                     packets;
  28         u64                     bytes;
  29         struct u64_stats_sync   syncp;
  30 };
  31
  32 struct veth_priv {
  33         struct net_device __rcu *peer;
  34         atomic64_t              dropped;
  35         unsigned                requested_headroom;
  36 };
  37
  38 /*
  39  * ethtool interface
  40  */
  41
  42 static struct {
  43         const char string[ETH_GSTRING_LEN];
  44 } ethtool_stats_keys[] = {
  45         { "peer_ifindex" },
  46 };
  47
  48 static int veth_get_link_ksettings(struct net_device *dev,
  49                                    struct ethtool_link_ksettings *cmd)
  50 {
  51         cmd->base.speed         = SPEED_10000;
  52         cmd->base.duplex        = DUPLEX_FULL;
  53         cmd->base.port          = PORT_TP;
  54         cmd->base.autoneg       = AUTONEG_DISABLE;
  55         return 0;
  56 }
  57
  58 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
  59 {
  60         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
  61         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
  62 }
  63
  64 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
  65 {
  66         switch(stringset) {
  67         case ETH_SS_STATS:
  68                 memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
  69                 break;
  70         }
  71 }
  72
  73 static int veth_get_sset_count(struct net_device *dev, int sset)
  74 {
  75         switch (sset) {
  76         case ETH_SS_STATS:
  77                 return ARRAY_SIZE(ethtool_stats_keys);
  78         default:
  79                 return -EOPNOTSUPP;
  80         }
  81 }
  82
  83 static void veth_get_ethtool_stats(struct net_device *dev,
  84                 struct ethtool_stats *stats, u64 *data)
  85 {
  86         struct veth_priv *priv = netdev_priv(dev);
  87         struct net_device *peer = rtnl_dereference(priv->peer);
  88
  89         data[0] = peer ? peer->ifindex : 0;
  90 }
  91
  92 static const struct ethtool_ops veth_ethtool_ops = {
  93         .get_drvinfo            = veth_get_drvinfo,
  94         .get_link               = ethtool_op_get_link,
  95         .get_strings            = veth_get_strings,
  96         .get_sset_count         = veth_get_sset_count,
  97         .get_ethtool_stats      = veth_get_ethtool_stats,
  98         .get_link_ksettings     = veth_get_link_ksettings,
  99 };
 100
 101 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 102 {
 103         struct veth_priv *priv = netdev_priv(dev);
 104         struct net_device *rcv;
 105         int length = skb->len;
 106
 107         rcu_read_lock();
 108         rcv = rcu_dereference(priv->peer);
 109         if (unlikely(!rcv)) {
 110                 kfree_skb(skb);
 111                 goto drop;
 112         }
 113
 114         if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
 115                 struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 116
 117                 u64_stats_update_begin(&stats->syncp);
 118                 stats->bytes += length;
 119                 stats->packets++;
 120                 u64_stats_update_end(&stats->syncp);
 121         } else {
 122 drop:
 123                 atomic64_inc(&priv->dropped);
 124         }
 125         rcu_read_unlock();
 126         return NETDEV_TX_OK;
 127 }
 128
 129 /*
 130  * general routines
 131  */
 132
 133 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 134 {
 135         struct veth_priv *priv = netdev_priv(dev);
 136         int cpu;
 137
 138         result->packets = 0;
 139         result->bytes = 0;
 140         for_each_possible_cpu(cpu) {
 141                 struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
 142                 u64 packets, bytes;
 143                 unsigned int start;
 144
 145                 do {
 146                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 147                         packets = stats->packets;
 148                         bytes = stats->bytes;
 149                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 150                 result->packets += packets;
 151                 result->bytes += bytes;
 152         }
 153         return atomic64_read(&priv->dropped);
 154 }
 155
 156 static void veth_get_stats64(struct net_device *dev,
 157                              struct rtnl_link_stats64 *tot)
 158 {
 159         struct veth_priv *priv = netdev_priv(dev);
 160         struct net_device *peer;
 161         struct pcpu_vstats one;
 162
 163         tot->tx_dropped = veth_stats_one(&one, dev);
 164         tot->tx_bytes = one.bytes;
 165         tot->tx_packets = one.packets;
 166
 167         rcu_read_lock();
 168         peer = rcu_dereference(priv->peer);
 169         if (peer) {
 170                 tot->rx_dropped = veth_stats_one(&one, peer);
 171                 tot->rx_bytes = one.bytes;
 172                 tot->rx_packets = one.packets;
 173         }
 174         rcu_read_unlock();
 175 }
 176
 177 /* fake multicast ability */
 178 static void veth_set_multicast_list(struct net_device *dev)
 179 {
 180 }
 181
 182 static int veth_open(struct net_device *dev)
 183 {
 184         struct veth_priv *priv = netdev_priv(dev);
 185         struct net_device *peer = rtnl_dereference(priv->peer);
 186
 187         if (!peer)
 188                 return -ENOTCONN;
 189
 190         if (peer->flags & IFF_UP) {
 191                 netif_carrier_on(dev);
 192                 netif_carrier_on(peer);
 193         }
 194         return 0;
 195 }
 196
 197 static int veth_close(struct net_device *dev)
 198 {
 199         struct veth_priv *priv = netdev_priv(dev);
 200         struct net_device *peer = rtnl_dereference(priv->peer);
 201
 202         netif_carrier_off(dev);
 203         if (peer)
 204                 netif_carrier_off(peer);
 205
 206         return 0;
 207 }
 208
 209 static int is_valid_veth_mtu(int mtu)
 210 {
 211         return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
 212 }
 213
 214 static int veth_dev_init(struct net_device *dev)
 215 {
 216         dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
 217         if (!dev->vstats)
 218                 return -ENOMEM;
 219         return 0;
 220 }
 221
 222 static void veth_dev_free(struct net_device *dev)
 223 {
 224         free_percpu(dev->vstats);
 225 }
 226
 227 #ifdef CONFIG_NET_POLL_CONTROLLER
 228 static void veth_poll_controller(struct net_device *dev)
 229 {
 230         /* veth only receives frames when its peer sends one
 231          * Since it's a synchronous operation, we are guaranteed
 232          * never to have pending data when we poll for it so
 233          * there is nothing to do here.
 234          *
 235          * We need this though so netpoll recognizes us as an interface that
 236          * supports polling, which enables bridge devices in virt setups to
 237          * still use netconsole
 238          */
 239 }
 240 #endif  /* CONFIG_NET_POLL_CONTROLLER */
 241
 242 static int veth_get_iflink(const struct net_device *dev)
 243 {
 244         struct veth_priv *priv = netdev_priv(dev);
 245         struct net_device *peer;
 246         int iflink;
 247
 248         rcu_read_lock();
 249         peer = rcu_dereference(priv->peer);
 250         iflink = peer ? peer->ifindex : 0;
 251         rcu_read_unlock();
 252
 253         return iflink;
 254 }
 255
 256 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
 257 {
 258         struct veth_priv *peer_priv, *priv = netdev_priv(dev);
 259         struct net_device *peer;
 260
 261         if (new_hr < 0)
 262                 new_hr = 0;
 263
 264         rcu_read_lock();
 265         peer = rcu_dereference(priv->peer);
 266         if (unlikely(!peer))
 267                 goto out;
 268
 269         peer_priv = netdev_priv(peer);
 270         priv->requested_headroom = new_hr;
 271         new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
 272         dev->needed_headroom = new_hr;
 273         peer->needed_headroom = new_hr;
 274
 275 out:
 276         rcu_read_unlock();
 277 }
 278
 279 static const struct net_device_ops veth_netdev_ops = {
 280         .ndo_init            = veth_dev_init,
 281         .ndo_open            = veth_open,
 282         .ndo_stop            = veth_close,
 283         .ndo_start_xmit      = veth_xmit,
 284         .ndo_get_stats64     = veth_get_stats64,
 285         .ndo_set_rx_mode     = veth_set_multicast_list,
 286         .ndo_set_mac_address = eth_mac_addr,
 287 #ifdef CONFIG_NET_POLL_CONTROLLER
 288         .ndo_poll_controller    = veth_poll_controller,
 289 #endif
 290         .ndo_get_iflink         = veth_get_iflink,
 291         .ndo_features_check     = passthru_features_check,
 292         .ndo_set_rx_headroom    = veth_set_rx_headroom,
 293 };
 294
 295 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
 296                        NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
 297                        NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
 298                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
 299                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
 300
 301 static void veth_setup(struct net_device *dev)
 302 {
 303         ether_setup(dev);
 304
 305         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 306         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 307         dev->priv_flags |= IFF_NO_QUEUE;
 308         dev->priv_flags |= IFF_PHONY_HEADROOM;
 309
 310         dev->netdev_ops = &veth_netdev_ops;
 311         dev->ethtool_ops = &veth_ethtool_ops;
 312         dev->features |= NETIF_F_LLTX;
 313         dev->features |= VETH_FEATURES;
 314         dev->vlan_features = dev->features &
 315                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 316                                NETIF_F_HW_VLAN_STAG_TX |
 317                                NETIF_F_HW_VLAN_CTAG_RX |
 318                                NETIF_F_HW_VLAN_STAG_RX);
 319         dev->needs_free_netdev = true;
 320         dev->priv_destructor = veth_dev_free;
 321         dev->max_mtu = ETH_MAX_MTU;
 322
 323         dev->hw_features = VETH_FEATURES;
 324         dev->hw_enc_features = VETH_FEATURES;
 325         dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
 326 }
 327
 328 /*
 329  * netlink interface
 330  */
 331
 332 static int veth_validate(struct nlattr *tb[], struct nlattr *data[],
 333                          struct netlink_ext_ack *extack)
 334 {
 335         if (tb[IFLA_ADDRESS]) {
 336                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 337                         return -EINVAL;
 338                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 339                         return -EADDRNOTAVAIL;
 340         }
 341         if (tb[IFLA_MTU]) {
 342                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 343                         return -EINVAL;
 344         }
 345         return 0;
 346 }
 347
 348 static struct rtnl_link_ops veth_link_ops;
 349
 350 static int veth_newlink(struct net *src_net, struct net_device *dev,
 351                         struct nlattr *tb[], struct nlattr *data[],
 352                         struct netlink_ext_ack *extack)
 353 {
 354         int err;
 355         struct net_device *peer;
 356         struct veth_priv *priv;
 357         char ifname[IFNAMSIZ];
 358         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 359         unsigned char name_assign_type;
 360         struct ifinfomsg *ifmp;
 361         struct net *net;
 362
 363         /*
 364          * create and register peer first
 365          */
 366         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 367                 struct nlattr *nla_peer;
 368
 369                 nla_peer = data[VETH_INFO_PEER];
 370                 ifmp = nla_data(nla_peer);
 371                 err = rtnl_nla_parse_ifla(peer_tb,
 372                                           nla_data(nla_peer) + sizeof(struct ifinfomsg),
 373                                           nla_len(nla_peer) - sizeof(struct ifinfomsg),
 374                                           NULL);
 375                 if (err < 0)
 376                         return err;
 377
 378                 err = veth_validate(peer_tb, NULL, extack);
 379                 if (err < 0)
 380                         return err;
 381
 382                 tbp = peer_tb;
 383         } else {
 384                 ifmp = NULL;
 385                 tbp = tb;
 386         }
 387
 388         if (ifmp && tbp[IFLA_IFNAME]) {
 389                 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 390                 name_assign_type = NET_NAME_USER;
 391         } else {
 392                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 393                 name_assign_type = NET_NAME_ENUM;
 394         }
 395
 396         net = rtnl_link_get_net(src_net, tbp);
 397         if (IS_ERR(net))
 398                 return PTR_ERR(net);
 399
 400         peer = rtnl_create_link(net, ifname, name_assign_type,
 401                                 &veth_link_ops, tbp);
 402         if (IS_ERR(peer)) {
 403                 put_net(net);
 404                 return PTR_ERR(peer);
 405         }
 406
 407         if (!ifmp || !tbp[IFLA_ADDRESS])
 408                 eth_hw_addr_random(peer);
 409
 410         if (ifmp && (dev->ifindex != 0))
 411                 peer->ifindex = ifmp->ifi_index;
 412
 413         peer->gso_max_size = dev->gso_max_size;
 414         peer->gso_max_segs = dev->gso_max_segs;
 415
 416         err = register_netdevice(peer);
 417         put_net(net);
 418         net = NULL;
 419         if (err < 0)
 420                 goto err_register_peer;
 421
 422         netif_carrier_off(peer);
 423
 424         err = rtnl_configure_link(peer, ifmp);
 425         if (err < 0)
 426                 goto err_configure_peer;
 427
 428         /*
 429          * register dev last
 430          *
 431          * note, that since we've registered new device the dev's name
 432          * should be re-allocated
 433          */
 434
 435         if (tb[IFLA_ADDRESS] == NULL)
 436                 eth_hw_addr_random(dev);
 437
 438         if (tb[IFLA_IFNAME])
 439                 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 440         else
 441                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 442
 443         err = register_netdevice(dev);
 444         if (err < 0)
 445                 goto err_register_dev;
 446
 447         netif_carrier_off(dev);
 448
 449         /*
 450          * tie the deviced together
 451          */
 452
 453         priv = netdev_priv(dev);
 454         rcu_assign_pointer(priv->peer, peer);
 455
 456         priv = netdev_priv(peer);
 457         rcu_assign_pointer(priv->peer, dev);
 458         return 0;
 459
 460 err_register_dev:
 461         /* nothing to do */
 462 err_configure_peer:
 463         unregister_netdevice(peer);
 464         return err;
 465
 466 err_register_peer:
 467         free_netdev(peer);
 468         return err;
 469 }
 470
 471 static void veth_dellink(struct net_device *dev, struct list_head *head)
 472 {
 473         struct veth_priv *priv;
 474         struct net_device *peer;
 475
 476         priv = netdev_priv(dev);
 477         peer = rtnl_dereference(priv->peer);
 478
 479         /* Note : dellink() is called from default_device_exit_batch(),
 480          * before a rcu_synchronize() point. The devices are guaranteed
 481          * not being freed before one RCU grace period.
 482          */
 483         RCU_INIT_POINTER(priv->peer, NULL);
 484         unregister_netdevice_queue(dev, head);
 485
 486         if (peer) {
 487                 priv = netdev_priv(peer);
 488                 RCU_INIT_POINTER(priv->peer, NULL);
 489                 unregister_netdevice_queue(peer, head);
 490         }
 491 }
 492
 493 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 494         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 495 };
 496
 497 static struct net *veth_get_link_net(const struct net_device *dev)
 498 {
 499         struct veth_priv *priv = netdev_priv(dev);
 500         struct net_device *peer = rtnl_dereference(priv->peer);
 501
 502         return peer ? dev_net(peer) : dev_net(dev);
 503 }
 504
 505 static struct rtnl_link_ops veth_link_ops = {
 506         .kind           = DRV_NAME,
 507         .priv_size      = sizeof(struct veth_priv),
 508         .setup          = veth_setup,
 509         .validate       = veth_validate,
 510         .newlink        = veth_newlink,
 511         .dellink        = veth_dellink,
 512         .policy         = veth_policy,
 513         .maxtype        = VETH_INFO_MAX,
 514         .get_link_net   = veth_get_link_net,
 515 };
 516
 517 /*
 518  * init/fini
 519  */
 520
 521 static __init int veth_init(void)
 522 {
 523         return rtnl_link_register(&veth_link_ops);
 524 }
 525
 526 static __exit void veth_exit(void)
 527 {
 528         rtnl_link_unregister(&veth_link_ops);
 529 }
 530
 531 module_init(veth_init);
 532 module_exit(veth_exit);
 533
 534 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 535 MODULE_LICENSE("GPL v2");
 536 MODULE_ALIAS_RTNL_LINK(DRV_NAME);