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