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/dst.h>
  18#include <net/xfrm.h>
  19#include <linux/veth.h>
  20#include <linux/module.h>
  21
  22#define DRV_NAME        "veth"
  23#define DRV_VERSION     "1.0"
  24
  25#define MIN_MTU 68              /* Min L3 MTU */
  26#define MAX_MTU 65535           /* Max L3 MTU (arbitrary) */
  27
  28struct pcpu_vstats {
  29        u64                     packets;
  30        u64                     bytes;
  31        struct u64_stats_sync   syncp;
  32};
  33
  34struct veth_priv {
  35        struct net_device __rcu *peer;
  36        atomic64_t              dropped;
  37};
  38
  39/*
  40 * ethtool interface
  41 */
  42
  43static struct {
  44        const char string[ETH_GSTRING_LEN];
  45} ethtool_stats_keys[] = {
  46        { "peer_ifindex" },
  47};
  48
  49static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
  50{
  51        cmd->supported          = 0;
  52        cmd->advertising        = 0;
  53        ethtool_cmd_speed_set(cmd, SPEED_10000);
  54        cmd->duplex             = DUPLEX_FULL;
  55        cmd->port               = PORT_TP;
  56        cmd->phy_address        = 0;
  57        cmd->transceiver        = XCVR_INTERNAL;
  58        cmd->autoneg            = AUTONEG_DISABLE;
  59        cmd->maxtxpkt           = 0;
  60        cmd->maxrxpkt           = 0;
  61        return 0;
  62}
  63
  64static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
  65{
  66        strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
  67        strlcpy(info->version, DRV_VERSION, sizeof(info->version));
  68}
  69
  70static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
  71{
  72        switch(stringset) {
  73        case ETH_SS_STATS:
  74                memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
  75                break;
  76        }
  77}
  78
  79static int veth_get_sset_count(struct net_device *dev, int sset)
  80{
  81        switch (sset) {
  82        case ETH_SS_STATS:
  83                return ARRAY_SIZE(ethtool_stats_keys);
  84        default:
  85                return -EOPNOTSUPP;
  86        }
  87}
  88
  89static void veth_get_ethtool_stats(struct net_device *dev,
  90                struct ethtool_stats *stats, u64 *data)
  91{
  92        struct veth_priv *priv = netdev_priv(dev);
  93        struct net_device *peer = rtnl_dereference(priv->peer);
  94
  95        data[0] = peer ? peer->ifindex : 0;
  96}
  97
  98static const struct ethtool_ops veth_ethtool_ops = {
  99        .get_settings           = veth_get_settings,
 100        .get_drvinfo            = veth_get_drvinfo,
 101        .get_link               = ethtool_op_get_link,
 102        .get_strings            = veth_get_strings,
 103        .get_sset_count         = veth_get_sset_count,
 104        .get_ethtool_stats      = veth_get_ethtool_stats,
 105};
 106
 107static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
 108{
 109        struct veth_priv *priv = netdev_priv(dev);
 110        struct net_device *rcv;
 111        int length = skb->len;
 112
 113        rcu_read_lock();
 114        rcv = rcu_dereference(priv->peer);
 115        if (unlikely(!rcv)) {
 116                kfree_skb(skb);
 117                goto drop;
 118        }
 119        /* don't change ip_summed == CHECKSUM_PARTIAL, as that
 120         * will cause bad checksum on forwarded packets
 121         */
 122        if (skb->ip_summed == CHECKSUM_NONE &&
 123            rcv->features & NETIF_F_RXCSUM)
 124                skb->ip_summed = CHECKSUM_UNNECESSARY;
 125
 126        if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
 127                struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
 128
 129                u64_stats_update_begin(&stats->syncp);
 130                stats->bytes += length;
 131                stats->packets++;
 132                u64_stats_update_end(&stats->syncp);
 133        } else {
 134drop:
 135                atomic64_inc(&priv->dropped);
 136        }
 137        rcu_read_unlock();
 138        return NETDEV_TX_OK;
 139}
 140
 141/*
 142 * general routines
 143 */
 144
 145static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
 146{
 147        struct veth_priv *priv = netdev_priv(dev);
 148        int cpu;
 149
 150        result->packets = 0;
 151        result->bytes = 0;
 152        for_each_possible_cpu(cpu) {
 153                struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
 154                u64 packets, bytes;
 155                unsigned int start;
 156
 157                do {
 158                        start = u64_stats_fetch_begin_bh(&stats->syncp);
 159                        packets = stats->packets;
 160                        bytes = stats->bytes;
 161                } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
 162                result->packets += packets;
 163                result->bytes += bytes;
 164        }
 165        return atomic64_read(&priv->dropped);
 166}
 167
 168static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
 169                                                  struct rtnl_link_stats64 *tot)
 170{
 171        struct veth_priv *priv = netdev_priv(dev);
 172        struct net_device *peer;
 173        struct pcpu_vstats one;
 174
 175        tot->tx_dropped = veth_stats_one(&one, dev);
 176        tot->tx_bytes = one.bytes;
 177        tot->tx_packets = one.packets;
 178
 179        rcu_read_lock();
 180        peer = rcu_dereference(priv->peer);
 181        if (peer) {
 182                tot->rx_dropped = veth_stats_one(&one, peer);
 183                tot->rx_bytes = one.bytes;
 184                tot->rx_packets = one.packets;
 185        }
 186        rcu_read_unlock();
 187
 188        return tot;
 189}
 190
 191static 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
 206static 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
 218static int is_valid_veth_mtu(int new_mtu)
 219{
 220        return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
 221}
 222
 223static 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
 231static int veth_dev_init(struct net_device *dev)
 232{
 233        dev->vstats = alloc_percpu(struct pcpu_vstats);
 234        if (!dev->vstats)
 235                return -ENOMEM;
 236
 237        return 0;
 238}
 239
 240static void veth_dev_free(struct net_device *dev)
 241{
 242        free_percpu(dev->vstats);
 243        free_netdev(dev);
 244}
 245
 246static const struct net_device_ops veth_netdev_ops = {
 247        .ndo_init            = veth_dev_init,
 248        .ndo_open            = veth_open,
 249        .ndo_stop            = veth_close,
 250        .ndo_start_xmit      = veth_xmit,
 251        .ndo_change_mtu      = veth_change_mtu,
 252        .ndo_get_stats64     = veth_get_stats64,
 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_HW_VLAN_TX | NETIF_F_HW_VLAN_RX)
 259
 260static void veth_setup(struct net_device *dev)
 261{
 262        ether_setup(dev);
 263
 264        dev->priv_flags &= ~IFF_TX_SKB_SHARING;
 265        dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 266
 267        dev->netdev_ops = &veth_netdev_ops;
 268        dev->ethtool_ops = &veth_ethtool_ops;
 269        dev->features |= NETIF_F_LLTX;
 270        dev->features |= VETH_FEATURES;
 271        dev->destructor = veth_dev_free;
 272
 273        dev->hw_features = VETH_FEATURES;
 274}
 275
 276/*
 277 * netlink interface
 278 */
 279
 280static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
 281{
 282        if (tb[IFLA_ADDRESS]) {
 283                if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
 284                        return -EINVAL;
 285                if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
 286                        return -EADDRNOTAVAIL;
 287        }
 288        if (tb[IFLA_MTU]) {
 289                if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
 290                        return -EINVAL;
 291        }
 292        return 0;
 293}
 294
 295static struct rtnl_link_ops veth_link_ops;
 296
 297static int veth_newlink(struct net *src_net, struct net_device *dev,
 298                         struct nlattr *tb[], struct nlattr *data[])
 299{
 300        int err;
 301        struct net_device *peer;
 302        struct veth_priv *priv;
 303        char ifname[IFNAMSIZ];
 304        struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
 305        struct ifinfomsg *ifmp;
 306        struct net *net;
 307
 308        /*
 309         * create and register peer first
 310         */
 311        if (data != NULL && data[VETH_INFO_PEER] != NULL) {
 312                struct nlattr *nla_peer;
 313
 314                nla_peer = data[VETH_INFO_PEER];
 315                ifmp = nla_data(nla_peer);
 316                err = nla_parse(peer_tb, IFLA_MAX,
 317                                nla_data(nla_peer) + sizeof(struct ifinfomsg),
 318                                nla_len(nla_peer) - sizeof(struct ifinfomsg),
 319                                ifla_policy);
 320                if (err < 0)
 321                        return err;
 322
 323                err = veth_validate(peer_tb, NULL);
 324                if (err < 0)
 325                        return err;
 326
 327                tbp = peer_tb;
 328        } else {
 329                ifmp = NULL;
 330                tbp = tb;
 331        }
 332
 333        if (tbp[IFLA_IFNAME])
 334                nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
 335        else
 336                snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
 337
 338        net = rtnl_link_get_net(src_net, tbp);
 339        if (IS_ERR(net))
 340                return PTR_ERR(net);
 341
 342        peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp);
 343        if (IS_ERR(peer)) {
 344                put_net(net);
 345                return PTR_ERR(peer);
 346        }
 347
 348        if (tbp[IFLA_ADDRESS] == NULL)
 349                eth_hw_addr_random(peer);
 350
 351        if (ifmp && (dev->ifindex != 0))
 352                peer->ifindex = ifmp->ifi_index;
 353
 354        err = register_netdevice(peer);
 355        put_net(net);
 356        net = NULL;
 357        if (err < 0)
 358                goto err_register_peer;
 359
 360        netif_carrier_off(peer);
 361
 362        err = rtnl_configure_link(peer, ifmp);
 363        if (err < 0)
 364                goto err_configure_peer;
 365
 366        /*
 367         * register dev last
 368         *
 369         * note, that since we've registered new device the dev's name
 370         * should be re-allocated
 371         */
 372
 373        if (tb[IFLA_ADDRESS] == NULL)
 374                eth_hw_addr_random(dev);
 375
 376        if (tb[IFLA_IFNAME])
 377                nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
 378        else
 379                snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
 380
 381        if (strchr(dev->name, '%')) {
 382                err = dev_alloc_name(dev, dev->name);
 383                if (err < 0)
 384                        goto err_alloc_name;
 385        }
 386
 387        err = register_netdevice(dev);
 388        if (err < 0)
 389                goto err_register_dev;
 390
 391        netif_carrier_off(dev);
 392
 393        /*
 394         * tie the deviced together
 395         */
 396
 397        priv = netdev_priv(dev);
 398        rcu_assign_pointer(priv->peer, peer);
 399
 400        priv = netdev_priv(peer);
 401        rcu_assign_pointer(priv->peer, dev);
 402        return 0;
 403
 404err_register_dev:
 405        /* nothing to do */
 406err_alloc_name:
 407err_configure_peer:
 408        unregister_netdevice(peer);
 409        return err;
 410
 411err_register_peer:
 412        free_netdev(peer);
 413        return err;
 414}
 415
 416static void veth_dellink(struct net_device *dev, struct list_head *head)
 417{
 418        struct veth_priv *priv;
 419        struct net_device *peer;
 420
 421        priv = netdev_priv(dev);
 422        peer = rtnl_dereference(priv->peer);
 423
 424        /* Note : dellink() is called from default_device_exit_batch(),
 425         * before a rcu_synchronize() point. The devices are guaranteed
 426         * not being freed before one RCU grace period.
 427         */
 428        RCU_INIT_POINTER(priv->peer, NULL);
 429        unregister_netdevice_queue(dev, head);
 430
 431        if (peer) {
 432                priv = netdev_priv(peer);
 433                RCU_INIT_POINTER(priv->peer, NULL);
 434                unregister_netdevice_queue(peer, head);
 435        }
 436}
 437
 438static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
 439        [VETH_INFO_PEER]        = { .len = sizeof(struct ifinfomsg) },
 440};
 441
 442static struct rtnl_link_ops veth_link_ops = {
 443        .kind           = DRV_NAME,
 444        .priv_size      = sizeof(struct veth_priv),
 445        .setup          = veth_setup,
 446        .validate       = veth_validate,
 447        .newlink        = veth_newlink,
 448        .dellink        = veth_dellink,
 449        .policy         = veth_policy,
 450        .maxtype        = VETH_INFO_MAX,
 451};
 452
 453/*
 454 * init/fini
 455 */
 456
 457static __init int veth_init(void)
 458{
 459        return rtnl_link_register(&veth_link_ops);
 460}
 461
 462static __exit void veth_exit(void)
 463{
 464        rtnl_link_unregister(&veth_link_ops);
 465}
 466
 467module_init(veth_init);
 468module_exit(veth_exit);
 469
 470MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
 471MODULE_LICENSE("GPL v2");
 472MODULE_ALIAS_RTNL_LINK(DRV_NAME);
 473
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.