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         unsigned                requested_headroom;
  39 };
  40
  41 /*
  42  * ethtool interface
  43  */
  44
  45 static struct {
  46         const char string[ETH_GSTRING_LEN];
  47 } ethtool_stats_keys[] = {
  48         { "peer_ifindex" },
  49 };
  50
  51 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
  52 {
  53         cmd->supported          = 0;
  54         cmd->advertising        = 0;
  55         ethtool_cmd_speed_set(cmd, SPEED_10000);
  56         cmd->duplex             = DUPLEX_FULL;
  57         cmd->port               = PORT_TP;
  58         cmd->phy_address        = 0;
  59         cmd->transceiver        = XCVR_INTERNAL;
  60         cmd->autoneg            = AUTONEG_DISABLE;
  61         cmd->maxtxpkt           = 0;
  62         cmd->maxrxpkt           = 0;
  63         return 0;
  64 }
  65
  66 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
  67 {
  68         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
  69         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
  70 }
  71
  72 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
  73 {
  74         switch(stringset) {
  75         case ETH_SS_STATS:
  76                 memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
  77                 break;
  78         }
  79 }
  80
  81 static int veth_get_sset_count(struct net_device *dev, int sset)
  82 {
  83         switch (sset) {
  84         case ETH_SS_STATS:
  85                 return ARRAY_SIZE(ethtool_stats_keys);
  86         default:
  87                 return -EOPNOTSUPP;
  88         }
  89 }
  90
  91 static void veth_get_ethtool_stats(struct net_device *dev,
  92                 struct ethtool_stats *stats, u64 *data)
  93 {
  94         struct veth_priv *priv = netdev_priv(dev);
  95         struct net_device *peer = rtnl_dereference(priv->peer);
  96
  97         data[0] = peer ? peer->ifindex : 0;
  98 }
  99
 100 static const struct ethtool_ops veth_ethtool_ops = {
 101         .get_settings           = veth_get_settings,
 102         .get_drvinfo            = veth_get_drvinfo,
 103         .get_link               = ethtool_op_get_link,
 104         .get_strings            = veth_get_strings,
 105         .get_sset_count         = veth_get_sset_count,
 106         .get_ethtool_stats      = veth_get_ethtool_stats,
 107 };
 108
 109 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 110 {
 111         struct veth_priv *priv = netdev_priv(dev);
 112         struct net_device *rcv;
 113         int length = skb->len;
 114
 115         rcu_read_lock();
 116         rcv = rcu_dereference(priv->peer);
 117         if (unlikely(!rcv)) {
 118                 kfree_skb(skb);
 119                 goto drop;
 120         }
 121
 122         if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
 123                 struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 124
 125                 u64_stats_update_begin(&stats->syncp);
 126                 stats->bytes += length;
 127                 stats->packets++;
 128                 u64_stats_update_end(&stats->syncp);
 129         } else {
 130 drop:
 131                 atomic64_inc(&priv->dropped);
 132         }
 133         rcu_read_unlock();
 134         return NETDEV_TX_OK;
 135 }
 136
 137 /*
 138  * general routines
 139  */
 140
 141 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 142 {
 143         struct veth_priv *priv = netdev_priv(dev);
 144         int cpu;
 145
 146         result->packets = 0;
 147         result->bytes = 0;
 148         for_each_possible_cpu(cpu) {
 149                 struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
 150                 u64 packets, bytes;
 151                 unsigned int start;
 152
 153                 do {
 154                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 155                         packets = stats->packets;
 156                         bytes = stats->bytes;
 157                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 158                 result->packets += packets;
 159                 result->bytes += bytes;
 160         }
 161         return atomic64_read(&priv->dropped);
 162 }
 163
 164 static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
 165                                                   struct rtnl_link_stats64 *tot)
 166 {
 167         struct veth_priv *priv = netdev_priv(dev);
 168         struct net_device *peer;
 169         struct pcpu_vstats one;
 170
 171         tot->tx_dropped = veth_stats_one(&one, dev);
 172         tot->tx_bytes = one.bytes;
 173         tot->tx_packets = one.packets;
 174
 175         rcu_read_lock();
 176         peer = rcu_dereference(priv->peer);
 177         if (peer) {
 178                 tot->rx_dropped = veth_stats_one(&one, peer);
 179                 tot->rx_bytes = one.bytes;
 180                 tot->rx_packets = one.packets;
 181         }
 182         rcu_read_unlock();
 183
 184         return tot;
 185 }
 186
 187 /* fake multicast ability */
 188 static void veth_set_multicast_list(struct net_device *dev)
 189 {
 190 }
 191
 192 static int veth_open(struct net_device *dev)
 193 {
 194         struct veth_priv *priv = netdev_priv(dev);
 195         struct net_device *peer = rtnl_dereference(priv->peer);
 196
 197         if (!peer)
 198                 return -ENOTCONN;
 199
 200         if (peer->flags & IFF_UP) {
 201                 netif_carrier_on(dev);
 202                 netif_carrier_on(peer);
 203         }
 204         return 0;
 205 }
 206
 207 static int veth_close(struct net_device *dev)
 208 {
 209         struct veth_priv *priv = netdev_priv(dev);
 210         struct net_device *peer = rtnl_dereference(priv->peer);
 211
 212         netif_carrier_off(dev);
 213         if (peer)
 214                 netif_carrier_off(peer);
 215
 216         return 0;
 217 }
 218
 219 static int is_valid_veth_mtu(int new_mtu)
 220 {
 221         return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
 222 }
 223
 224 static int veth_change_mtu(struct net_device *dev, int new_mtu)
 225 {
 226         if (!is_valid_veth_mtu(new_mtu))
 227                 return -EINVAL;
 228         dev->mtu = new_mtu;
 229         return 0;
 230 }
 231
 232 static int veth_dev_init(struct net_device *dev)
 233 {
 234         dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
 235         if (!dev->vstats)
 236                 return -ENOMEM;
 237         return 0;
 238 }
 239
 240 static void veth_dev_free(struct net_device *dev)
 241 {
 242         free_percpu(dev->vstats);
 243         free_netdev(dev);
 244 }
 245
 246 #ifdef CONFIG_NET_POLL_CONTROLLER
 247 static void veth_poll_controller(struct net_device *dev)
 248 {
 249         /* veth only receives frames when its peer sends one
 250          * Since it's a synchronous operation, we are guaranteed
 251          * never to have pending data when we poll for it so
 252          * there is nothing to do here.
 253          *
 254          * We need this though so netpoll recognizes us as an interface that
 255          * supports polling, which enables bridge devices in virt setups to
 256          * still use netconsole
 257          */
 258 }
 259 #endif  /* CONFIG_NET_POLL_CONTROLLER */
 260
 261 static int veth_get_iflink(const struct net_device *dev)
 262 {
 263         struct veth_priv *priv = netdev_priv(dev);
 264         struct net_device *peer;
 265         int iflink;
 266
 267         rcu_read_lock();
 268         peer = rcu_dereference(priv->peer);
 269         iflink = peer ? peer->ifindex : 0;
 270         rcu_read_unlock();
 271
 272         return iflink;
 273 }
 274
 275 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
 276 {
 277         struct veth_priv *peer_priv, *priv = netdev_priv(dev);
 278         struct net_device *peer;
 279
 280         if (new_hr < 0)
 281                 new_hr = 0;
 282
 283         rcu_read_lock();
 284         peer = rcu_dereference(priv->peer);
 285         if (unlikely(!peer))
 286                 goto out;
 287
 288         peer_priv = netdev_priv(peer);
 289         priv->requested_headroom = new_hr;
 290         new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
 291         dev->needed_headroom = new_hr;
 292         peer->needed_headroom = new_hr;
 293
 294 out:
 295         rcu_read_unlock();
 296 }
 297
 298 static const struct net_device_ops veth_netdev_ops = {
 299         .ndo_init            = veth_dev_init,
 300         .ndo_open            = veth_open,
 301         .ndo_stop            = veth_close,
 302         .ndo_start_xmit      = veth_xmit,
 303         .ndo_change_mtu      = veth_change_mtu,
 304         .ndo_get_stats64     = veth_get_stats64,
 305         .ndo_set_rx_mode     = veth_set_multicast_list,
 306         .ndo_set_mac_address = eth_mac_addr,
 307 #ifdef CONFIG_NET_POLL_CONTROLLER
 308         .ndo_poll_controller    = veth_poll_controller,
 309 #endif
 310         .ndo_get_iflink         = veth_get_iflink,
 311         .ndo_features_check     = passthru_features_check,
 312         .ndo_set_rx_headroom    = veth_set_rx_headroom,
 313 };
 314
 315 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
 316                        NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \
 317                        NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
 318                        NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
 319                        NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
 320
 321 static void veth_setup(struct net_device *dev)
 322 {
 323         ether_setup(dev);
 324
 325         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 326         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 327         dev->priv_flags |= IFF_NO_QUEUE;
 328         dev->priv_flags |= IFF_PHONY_HEADROOM;
 329
 330         dev->netdev_ops = &veth_netdev_ops;
 331         dev->ethtool_ops = &veth_ethtool_ops;
 332         dev->features |= NETIF_F_LLTX;
 333         dev->features |= VETH_FEATURES;
 334         dev->vlan_features = dev->features &
 335                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 336                                NETIF_F_HW_VLAN_STAG_TX |
 337                                NETIF_F_HW_VLAN_CTAG_RX |
 338                                NETIF_F_HW_VLAN_STAG_RX);
 339         dev->destructor = veth_dev_free;
 340
 341         dev->hw_features = VETH_FEATURES;
 342         dev->hw_enc_features = VETH_FEATURES;
 343 }
 344
 345 /*
 346  * netlink interface
 347  */
 348
 349 static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
 350 {
 351         if (tb[IFLA_ADDRESS]) {
 352                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 353                         return -EINVAL;
 354                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 355                         return -EADDRNOTAVAIL;
 356         }
 357         if (tb[IFLA_MTU]) {
 358                 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 359                         return -EINVAL;
 360         }
 361         return 0;
 362 }
 363
 364 static struct rtnl_link_ops veth_link_ops;
 365
 366 static int veth_newlink(struct net *src_net, struct net_device *dev,
 367                          struct nlattr *tb[], struct nlattr *data[])
 368 {
 369         int err;
 370         struct net_device *peer;
 371         struct veth_priv *priv;
 372         char ifname[IFNAMSIZ];
 373         struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 374         unsigned char name_assign_type;
 375         struct ifinfomsg *ifmp;
 376         struct net *net;
 377
 378         /*
 379          * create and register peer first
 380          */
 381         if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 382                 struct nlattr *nla_peer;
 383
 384                 nla_peer = data[VETH_INFO_PEER];
 385                 ifmp = nla_data(nla_peer);
 386                 err = rtnl_nla_parse_ifla(peer_tb,
 387                                           nla_data(nla_peer) + sizeof(struct ifinfomsg),
 388                                           nla_len(nla_peer) - sizeof(struct ifinfomsg));
 389                 if (err < 0)
 390                         return err;
 391
 392                 err = veth_validate(peer_tb, NULL);
 393                 if (err < 0)
 394                         return err;
 395
 396                 tbp = peer_tb;
 397         } else {
 398                 ifmp = NULL;
 399                 tbp = tb;
 400         }
 401
 402         if (tbp[IFLA_IFNAME]) {
 403                 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 404                 name_assign_type = NET_NAME_USER;
 405         } else {
 406                 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 407                 name_assign_type = NET_NAME_ENUM;
 408         }
 409
 410         net = rtnl_link_get_net(src_net, tbp);
 411         if (IS_ERR(net))
 412                 return PTR_ERR(net);
 413
 414         peer = rtnl_create_link(net, ifname, name_assign_type,
 415                                 &veth_link_ops, tbp);
 416         if (IS_ERR(peer)) {
 417                 put_net(net);
 418                 return PTR_ERR(peer);
 419         }
 420
 421         if (tbp[IFLA_ADDRESS] == NULL)
 422                 eth_hw_addr_random(peer);
 423
 424         if (ifmp && (dev->ifindex != 0))
 425                 peer->ifindex = ifmp->ifi_index;
 426
 427         err = register_netdevice(peer);
 428         put_net(net);
 429         net = NULL;
 430         if (err < 0)
 431                 goto err_register_peer;
 432
 433         netif_carrier_off(peer);
 434
 435         err = rtnl_configure_link(peer, ifmp);
 436         if (err < 0)
 437                 goto err_configure_peer;
 438
 439         /*
 440          * register dev last
 441          *
 442          * note, that since we've registered new device the dev's name
 443          * should be re-allocated
 444          */
 445
 446         if (tb[IFLA_ADDRESS] == NULL)
 447                 eth_hw_addr_random(dev);
 448
 449         if (tb[IFLA_IFNAME])
 450                 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 451         else
 452                 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 453
 454         err = register_netdevice(dev);
 455         if (err < 0)
 456                 goto err_register_dev;
 457
 458         netif_carrier_off(dev);
 459
 460         /*
 461          * tie the deviced together
 462          */
 463
 464         priv = netdev_priv(dev);
 465         rcu_assign_pointer(priv->peer, peer);
 466
 467         priv = netdev_priv(peer);
 468         rcu_assign_pointer(priv->peer, dev);
 469         return 0;
 470
 471 err_register_dev:
 472         /* nothing to do */
 473 err_configure_peer:
 474         unregister_netdevice(peer);
 475         return err;
 476
 477 err_register_peer:
 478         free_netdev(peer);
 479         return err;
 480 }
 481
 482 static void veth_dellink(struct net_device *dev, struct list_head *head)
 483 {
 484         struct veth_priv *priv;
 485         struct net_device *peer;
 486
 487         priv = netdev_priv(dev);
 488         peer = rtnl_dereference(priv->peer);
 489
 490         /* Note : dellink() is called from default_device_exit_batch(),
 491          * before a rcu_synchronize() point. The devices are guaranteed
 492          * not being freed before one RCU grace period.
 493          */
 494         RCU_INIT_POINTER(priv->peer, NULL);
 495         unregister_netdevice_queue(dev, head);
 496
 497         if (peer) {
 498                 priv = netdev_priv(peer);
 499                 RCU_INIT_POINTER(priv->peer, NULL);
 500                 unregister_netdevice_queue(peer, head);
 501         }
 502 }
 503
 504 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 505         [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 506 };
 507
 508 static struct net *veth_get_link_net(const struct net_device *dev)
 509 {
 510         struct veth_priv *priv = netdev_priv(dev);
 511         struct net_device *peer = rtnl_dereference(priv->peer);
 512
 513         return peer ? dev_net(peer) : dev_net(dev);
 514 }
 515
 516 static struct rtnl_link_ops veth_link_ops = {
 517         .kind           = DRV_NAME,
 518         .priv_size      = sizeof(struct veth_priv),
 519         .setup          = veth_setup,
 520         .validate       = veth_validate,
 521         .newlink        = veth_newlink,
 522         .dellink        = veth_dellink,
 523         .policy         = veth_policy,
 524         .maxtype        = VETH_INFO_MAX,
 525         .get_link_net   = veth_get_link_net,
 526 };
 527
 528 /*
 529  * init/fini
 530  */
 531
 532 static __init int veth_init(void)
 533 {
 534         return rtnl_link_register(&veth_link_ops);
 535 }
 536
 537 static __exit void veth_exit(void)
 538 {
 539         rtnl_link_unregister(&veth_link_ops);
 540 }
 541
 542 module_init(veth_init);
 543 module_exit(veth_exit);
 544
 545 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 546 MODULE_LICENSE("GPL v2");
 547 MODULE_ALIAS_RTNL_LINK(DRV_NAME);