linux/drivers/net/veth.c
<<
>>
Prefs
   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
  29struct pcpu_vstats {
  30        u64                     packets;
  31        u64                     bytes;
  32        struct u64_stats_sync   syncp;
  33};
  34
  35struct veth_priv {
  36        struct net_device __rcu *peer;
  37        atomic64_t              dropped;
  38        unsigned                requested_headroom;
  39};
  40
  41/*
  42 * ethtool interface
  43 */
  44
  45static struct {
  46        const char string[ETH_GSTRING_LEN];
  47} ethtool_stats_keys[] = {
  48        { "peer_ifindex" },
  49};
  50
  51static 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
  66static 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
  72static 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
  81static 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
  91static 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
 100static 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
 109static 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 {
 130drop:
 131                atomic64_inc(&priv->dropped);
 132        }
 133        rcu_read_unlock();
 134        return NETDEV_TX_OK;
 135}
 136
 137/*
 138 * general routines
 139 */
 140
 141static 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
 164static void 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
 185/* fake multicast ability */
 186static void veth_set_multicast_list(struct net_device *dev)
 187{
 188}
 189
 190static int veth_open(struct net_device *dev)
 191{
 192        struct veth_priv *priv = netdev_priv(dev);
 193        struct net_device *peer = rtnl_dereference(priv->peer);
 194
 195        if (!peer)
 196                return -ENOTCONN;
 197
 198        if (peer->flags & IFF_UP) {
 199                netif_carrier_on(dev);
 200                netif_carrier_on(peer);
 201        }
 202        return 0;
 203}
 204
 205static int veth_close(struct net_device *dev)
 206{
 207        struct veth_priv *priv = netdev_priv(dev);
 208        struct net_device *peer = rtnl_dereference(priv->peer);
 209
 210        netif_carrier_off(dev);
 211        if (peer)
 212                netif_carrier_off(peer);
 213
 214        return 0;
 215}
 216
 217static int is_valid_veth_mtu(int new_mtu)
 218{
 219        return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
 220}
 221
 222static int veth_change_mtu(struct net_device *dev, int new_mtu)
 223{
 224        if (!is_valid_veth_mtu(new_mtu))
 225                return -EINVAL;
 226        dev->mtu = new_mtu;
 227        return 0;
 228}
 229
 230static int veth_dev_init(struct net_device *dev)
 231{
 232        dev->vstats = alloc_percpu(struct pcpu_vstats);
 233        if (!dev->vstats)
 234                return -ENOMEM;
 235
 236        return 0;
 237}
 238
 239static void veth_dev_free(struct net_device *dev)
 240{
 241        free_percpu(dev->vstats);
 242}
 243
 244static int veth_get_iflink(const struct net_device *dev)
 245{
 246        struct veth_priv *priv = netdev_priv(dev);
 247        struct net_device *peer;
 248        int iflink;
 249
 250        rcu_read_lock();
 251        peer = rcu_dereference(priv->peer);
 252        iflink = peer ? peer->ifindex : 0;
 253        rcu_read_unlock();
 254
 255        return iflink;
 256}
 257
 258static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
 259{
 260        struct veth_priv *peer_priv, *priv = netdev_priv(dev);
 261        struct net_device *peer;
 262
 263        if (new_hr < 0)
 264                new_hr = 0;
 265
 266        rcu_read_lock();
 267        peer = rcu_dereference(priv->peer);
 268        if (unlikely(!peer))
 269                goto out;
 270
 271        peer_priv = netdev_priv(peer);
 272        priv->requested_headroom = new_hr;
 273        new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
 274        dev->needed_headroom = new_hr;
 275        peer->needed_headroom = new_hr;
 276
 277out:
 278        rcu_read_unlock();
 279}
 280
 281static const struct net_device_ops veth_netdev_ops = {
 282        .ndo_init            = veth_dev_init,
 283        .ndo_open            = veth_open,
 284        .ndo_stop            = veth_close,
 285        .ndo_start_xmit      = veth_xmit,
 286        .ndo_change_mtu_rh74 = veth_change_mtu,
 287        .ndo_get_stats64     = veth_get_stats64,
 288        .ndo_set_rx_mode     = veth_set_multicast_list,
 289        .ndo_set_mac_address = eth_mac_addr,
 290        .ndo_get_iflink         = veth_get_iflink,
 291        .ndo_size               = sizeof(struct net_device_ops),
 292        .extended.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
 301static 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 | NETIF_F_HW_VLAN_STAG_TX);
 316        dev->extended->needs_free_netdev = true;
 317        dev->extended->priv_destructor = veth_dev_free;
 318
 319        dev->hw_features = VETH_FEATURES;
 320        dev->hw_enc_features = VETH_FEATURES;
 321        dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
 322}
 323
 324/*
 325 * netlink interface
 326 */
 327
 328static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
 329{
 330        if (tb[IFLA_ADDRESS]) {
 331                if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 332                        return -EINVAL;
 333                if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 334                        return -EADDRNOTAVAIL;
 335        }
 336        if (tb[IFLA_MTU]) {
 337                if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 338                        return -EINVAL;
 339        }
 340        return 0;
 341}
 342
 343static struct rtnl_link_ops veth_link_ops;
 344
 345static int veth_newlink(struct net *src_net, struct net_device *dev,
 346                         struct nlattr *tb[], struct nlattr *data[])
 347{
 348        int err;
 349        struct net_device *peer;
 350        struct veth_priv *priv;
 351        char ifname[IFNAMSIZ];
 352        struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 353        struct ifinfomsg *ifmp;
 354        struct net *net;
 355
 356        /*
 357         * create and register peer first
 358         */
 359        if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 360                struct nlattr *nla_peer;
 361
 362                nla_peer = data[VETH_INFO_PEER];
 363                ifmp = nla_data(nla_peer);
 364                err = rtnl_nla_parse_ifla(peer_tb,
 365                                          nla_data(nla_peer) + sizeof(struct ifinfomsg),
 366                                          nla_len(nla_peer) - sizeof(struct ifinfomsg));
 367                if (err < 0)
 368                        return err;
 369
 370                err = veth_validate(peer_tb, NULL);
 371                if (err < 0)
 372                        return err;
 373
 374                tbp = peer_tb;
 375        } else {
 376                ifmp = NULL;
 377                tbp = tb;
 378        }
 379
 380        if (ifmp && tbp[IFLA_IFNAME])
 381                nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 382        else
 383                snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 384
 385        net = rtnl_link_get_net(src_net, tbp);
 386        if (IS_ERR(net))
 387                return PTR_ERR(net);
 388
 389        peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp);
 390        if (IS_ERR(peer)) {
 391                put_net(net);
 392                return PTR_ERR(peer);
 393        }
 394
 395        if (!ifmp || !tbp[IFLA_ADDRESS])
 396                eth_hw_addr_random(peer);
 397
 398        if (ifmp && (dev->ifindex != 0))
 399                peer->ifindex = ifmp->ifi_index;
 400
 401        err = register_netdevice(peer);
 402        put_net(net);
 403        net = NULL;
 404        if (err < 0)
 405                goto err_register_peer;
 406
 407        netif_carrier_off(peer);
 408
 409        err = rtnl_configure_link(peer, ifmp);
 410        if (err < 0)
 411                goto err_configure_peer;
 412
 413        /*
 414         * register dev last
 415         *
 416         * note, that since we've registered new device the dev's name
 417         * should be re-allocated
 418         */
 419
 420        if (tb[IFLA_ADDRESS] == NULL)
 421                eth_hw_addr_random(dev);
 422
 423        if (tb[IFLA_IFNAME])
 424                nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 425        else
 426                snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 427
 428        if (strchr(dev->name, '%')) {
 429                err = dev_alloc_name(dev, dev->name);
 430                if (err < 0)
 431                        goto err_alloc_name;
 432        }
 433
 434        err = register_netdevice(dev);
 435        if (err < 0)
 436                goto err_register_dev;
 437
 438        netif_carrier_off(dev);
 439
 440        /*
 441         * tie the deviced together
 442         */
 443
 444        priv = netdev_priv(dev);
 445        rcu_assign_pointer(priv->peer, peer);
 446
 447        priv = netdev_priv(peer);
 448        rcu_assign_pointer(priv->peer, dev);
 449        return 0;
 450
 451err_register_dev:
 452        /* nothing to do */
 453err_alloc_name:
 454err_configure_peer:
 455        unregister_netdevice(peer);
 456        return err;
 457
 458err_register_peer:
 459        free_netdev(peer);
 460        return err;
 461}
 462
 463static void veth_dellink(struct net_device *dev, struct list_head *head)
 464{
 465        struct veth_priv *priv;
 466        struct net_device *peer;
 467
 468        priv = netdev_priv(dev);
 469        peer = rtnl_dereference(priv->peer);
 470
 471        /* Note : dellink() is called from default_device_exit_batch(),
 472         * before a rcu_synchronize() point. The devices are guaranteed
 473         * not being freed before one RCU grace period.
 474         */
 475        RCU_INIT_POINTER(priv->peer, NULL);
 476        unregister_netdevice_queue(dev, head);
 477
 478        if (peer) {
 479                priv = netdev_priv(peer);
 480                RCU_INIT_POINTER(priv->peer, NULL);
 481                unregister_netdevice_queue(peer, head);
 482        }
 483}
 484
 485static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 486        [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 487};
 488
 489static struct net *veth_get_link_net(const struct net_device *dev)
 490{
 491        struct veth_priv *priv = netdev_priv(dev);
 492        struct net_device *peer = rtnl_dereference(priv->peer);
 493
 494        return peer ? dev_net(peer) : dev_net(dev);
 495}
 496
 497static struct rtnl_link_ops veth_link_ops = {
 498        .kind           = DRV_NAME,
 499        .priv_size      = sizeof(struct veth_priv),
 500        .setup          = veth_setup,
 501        .validate       = veth_validate,
 502        .newlink        = veth_newlink,
 503        .dellink        = veth_dellink,
 504        .policy         = veth_policy,
 505        .maxtype        = VETH_INFO_MAX,
 506        .get_link_net   = veth_get_link_net,
 507};
 508
 509/*
 510 * init/fini
 511 */
 512
 513static __init int veth_init(void)
 514{
 515        return rtnl_link_register(&veth_link_ops);
 516}
 517
 518static __exit void veth_exit(void)
 519{
 520        rtnl_link_unregister(&veth_link_ops);
 521}
 522
 523module_init(veth_init);
 524module_exit(veth_exit);
 525
 526MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 527MODULE_LICENSE("GPL v2");
 528MODULE_ALIAS_RTNL_LINK(DRV_NAME);
 529