| 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 | |
| 28 | struct pcpu_vstats { |
| 29 | u64 packets; |
| 30 | u64 bytes; |
| 31 | struct u64_stats_sync syncp; |
| 32 | }; |
| 33 | |
| 34 | struct veth_priv { |
| 35 | struct net_device __rcu *peer; |
| 36 | atomic64_t dropped; |
| 37 | }; |
| 38 | |
| 39 | /* |
| 40 | * ethtool interface |
| 41 | */ |
| 42 | |
| 43 | static struct { |
| 44 | const char string[ETH_GSTRING_LEN]; |
| 45 | } ethtool_stats_keys[] = { |
| 46 | { "peer_ifindex" }, |
| 47 | }; |
| 48 | |
| 49 | static 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 | |
| 64 | static 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 | |
| 70 | static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf) |
| 71 | { |
| 72 | switch(stringset) { |
| 73 | case ETH_SS_STATS: |
| 74 | memcpy(buf, ðtool_stats_keys, sizeof(ethtool_stats_keys)); |
| 75 | break; |
| 76 | } |
| 77 | } |
| 78 | |
| 79 | static 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 | |
| 89 | static 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 | |
| 98 | static 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 | |
| 107 | static 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 | |
| 120 | if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) { |
| 121 | struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats); |
| 122 | |
| 123 | u64_stats_update_begin(&stats->syncp); |
| 124 | stats->bytes += length; |
| 125 | stats->packets++; |
| 126 | u64_stats_update_end(&stats->syncp); |
| 127 | } else { |
| 128 | drop: |
| 129 | atomic64_inc(&priv->dropped); |
| 130 | } |
| 131 | rcu_read_unlock(); |
| 132 | return NETDEV_TX_OK; |
| 133 | } |
| 134 | |
| 135 | /* |
| 136 | * general routines |
| 137 | */ |
| 138 | |
| 139 | static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev) |
| 140 | { |
| 141 | struct veth_priv *priv = netdev_priv(dev); |
| 142 | int cpu; |
| 143 | |
| 144 | result->packets = 0; |
| 145 | result->bytes = 0; |
| 146 | for_each_possible_cpu(cpu) { |
| 147 | struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu); |
| 148 | u64 packets, bytes; |
| 149 | unsigned int start; |
| 150 | |
| 151 | do { |
| 152 | start = u64_stats_fetch_begin_bh(&stats->syncp); |
| 153 | packets = stats->packets; |
| 154 | bytes = stats->bytes; |
| 155 | } while (u64_stats_fetch_retry_bh(&stats->syncp, start)); |
| 156 | result->packets += packets; |
| 157 | result->bytes += bytes; |
| 158 | } |
| 159 | return atomic64_read(&priv->dropped); |
| 160 | } |
| 161 | |
| 162 | static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev, |
| 163 | struct rtnl_link_stats64 *tot) |
| 164 | { |
| 165 | struct veth_priv *priv = netdev_priv(dev); |
| 166 | struct net_device *peer; |
| 167 | struct pcpu_vstats one; |
| 168 | |
| 169 | tot->tx_dropped = veth_stats_one(&one, dev); |
| 170 | tot->tx_bytes = one.bytes; |
| 171 | tot->tx_packets = one.packets; |
| 172 | |
| 173 | rcu_read_lock(); |
| 174 | peer = rcu_dereference(priv->peer); |
| 175 | if (peer) { |
| 176 | tot->rx_dropped = veth_stats_one(&one, peer); |
| 177 | tot->rx_bytes = one.bytes; |
| 178 | tot->rx_packets = one.packets; |
| 179 | } |
| 180 | rcu_read_unlock(); |
| 181 | |
| 182 | return tot; |
| 183 | } |
| 184 | |
| 185 | static int veth_open(struct net_device *dev) |
| 186 | { |
| 187 | struct veth_priv *priv = netdev_priv(dev); |
| 188 | struct net_device *peer = rtnl_dereference(priv->peer); |
| 189 | |
| 190 | if (!peer) |
| 191 | return -ENOTCONN; |
| 192 | |
| 193 | if (peer->flags & IFF_UP) { |
| 194 | netif_carrier_on(dev); |
| 195 | netif_carrier_on(peer); |
| 196 | } |
| 197 | return 0; |
| 198 | } |
| 199 | |
| 200 | static int veth_close(struct net_device *dev) |
| 201 | { |
| 202 | struct veth_priv *priv = netdev_priv(dev); |
| 203 | struct net_device *peer = rtnl_dereference(priv->peer); |
| 204 | |
| 205 | netif_carrier_off(dev); |
| 206 | if (peer) |
| 207 | netif_carrier_off(peer); |
| 208 | |
| 209 | return 0; |
| 210 | } |
| 211 | |
| 212 | static int is_valid_veth_mtu(int new_mtu) |
| 213 | { |
| 214 | return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU; |
| 215 | } |
| 216 | |
| 217 | static int veth_change_mtu(struct net_device *dev, int new_mtu) |
| 218 | { |
| 219 | if (!is_valid_veth_mtu(new_mtu)) |
| 220 | return -EINVAL; |
| 221 | dev->mtu = new_mtu; |
| 222 | return 0; |
| 223 | } |
| 224 | |
| 225 | static int veth_dev_init(struct net_device *dev) |
| 226 | { |
| 227 | dev->vstats = alloc_percpu(struct pcpu_vstats); |
| 228 | if (!dev->vstats) |
| 229 | return -ENOMEM; |
| 230 | |
| 231 | return 0; |
| 232 | } |
| 233 | |
| 234 | static void veth_dev_free(struct net_device *dev) |
| 235 | { |
| 236 | free_percpu(dev->vstats); |
| 237 | free_netdev(dev); |
| 238 | } |
| 239 | |
| 240 | static const struct net_device_ops veth_netdev_ops = { |
| 241 | .ndo_init = veth_dev_init, |
| 242 | .ndo_open = veth_open, |
| 243 | .ndo_stop = veth_close, |
| 244 | .ndo_start_xmit = veth_xmit, |
| 245 | .ndo_change_mtu = veth_change_mtu, |
| 246 | .ndo_get_stats64 = veth_get_stats64, |
| 247 | .ndo_set_mac_address = eth_mac_addr, |
| 248 | }; |
| 249 | |
| 250 | #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ |
| 251 | NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \ |
| 252 | NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \ |
| 253 | NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX ) |
| 254 | |
| 255 | static void veth_setup(struct net_device *dev) |
| 256 | { |
| 257 | ether_setup(dev); |
| 258 | |
| 259 | dev->priv_flags &= ~IFF_TX_SKB_SHARING; |
| 260 | dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; |
| 261 | |
| 262 | dev->netdev_ops = &veth_netdev_ops; |
| 263 | dev->ethtool_ops = &veth_ethtool_ops; |
| 264 | dev->features |= NETIF_F_LLTX; |
| 265 | dev->features |= VETH_FEATURES; |
| 266 | dev->destructor = veth_dev_free; |
| 267 | |
| 268 | dev->hw_features = VETH_FEATURES; |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * netlink interface |
| 273 | */ |
| 274 | |
| 275 | static int veth_validate(struct nlattr *tb[], struct nlattr *data[]) |
| 276 | { |
| 277 | if (tb[IFLA_ADDRESS]) { |
| 278 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) |
| 279 | return -EINVAL; |
| 280 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) |
| 281 | return -EADDRNOTAVAIL; |
| 282 | } |
| 283 | if (tb[IFLA_MTU]) { |
| 284 | if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU]))) |
| 285 | return -EINVAL; |
| 286 | } |
| 287 | return 0; |
| 288 | } |
| 289 | |
| 290 | static struct rtnl_link_ops veth_link_ops; |
| 291 | |
| 292 | static int veth_newlink(struct net *src_net, struct net_device *dev, |
| 293 | struct nlattr *tb[], struct nlattr *data[]) |
| 294 | { |
| 295 | int err; |
| 296 | struct net_device *peer; |
| 297 | struct veth_priv *priv; |
| 298 | char ifname[IFNAMSIZ]; |
| 299 | struct nlattr *peer_tb[IFLA_MAX + 1], **tbp; |
| 300 | struct ifinfomsg *ifmp; |
| 301 | struct net *net; |
| 302 | |
| 303 | /* |
| 304 | * create and register peer first |
| 305 | */ |
| 306 | if (data != NULL && data[VETH_INFO_PEER] != NULL) { |
| 307 | struct nlattr *nla_peer; |
| 308 | |
| 309 | nla_peer = data[VETH_INFO_PEER]; |
| 310 | ifmp = nla_data(nla_peer); |
| 311 | err = nla_parse(peer_tb, IFLA_MAX, |
| 312 | nla_data(nla_peer) + sizeof(struct ifinfomsg), |
| 313 | nla_len(nla_peer) - sizeof(struct ifinfomsg), |
| 314 | ifla_policy); |
| 315 | if (err < 0) |
| 316 | return err; |
| 317 | |
| 318 | err = veth_validate(peer_tb, NULL); |
| 319 | if (err < 0) |
| 320 | return err; |
| 321 | |
| 322 | tbp = peer_tb; |
| 323 | } else { |
| 324 | ifmp = NULL; |
| 325 | tbp = tb; |
| 326 | } |
| 327 | |
| 328 | if (tbp[IFLA_IFNAME]) |
| 329 | nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ); |
| 330 | else |
| 331 | snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d"); |
| 332 | |
| 333 | net = rtnl_link_get_net(src_net, tbp); |
| 334 | if (IS_ERR(net)) |
| 335 | return PTR_ERR(net); |
| 336 | |
| 337 | peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp); |
| 338 | if (IS_ERR(peer)) { |
| 339 | put_net(net); |
| 340 | return PTR_ERR(peer); |
| 341 | } |
| 342 | |
| 343 | if (tbp[IFLA_ADDRESS] == NULL) |
| 344 | eth_hw_addr_random(peer); |
| 345 | |
| 346 | if (ifmp && (dev->ifindex != 0)) |
| 347 | peer->ifindex = ifmp->ifi_index; |
| 348 | |
| 349 | err = register_netdevice(peer); |
| 350 | put_net(net); |
| 351 | net = NULL; |
| 352 | if (err < 0) |
| 353 | goto err_register_peer; |
| 354 | |
| 355 | netif_carrier_off(peer); |
| 356 | |
| 357 | err = rtnl_configure_link(peer, ifmp); |
| 358 | if (err < 0) |
| 359 | goto err_configure_peer; |
| 360 | |
| 361 | /* |
| 362 | * register dev last |
| 363 | * |
| 364 | * note, that since we've registered new device the dev's name |
| 365 | * should be re-allocated |
| 366 | */ |
| 367 | |
| 368 | if (tb[IFLA_ADDRESS] == NULL) |
| 369 | eth_hw_addr_random(dev); |
| 370 | |
| 371 | if (tb[IFLA_IFNAME]) |
| 372 | nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ); |
| 373 | else |
| 374 | snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d"); |
| 375 | |
| 376 | if (strchr(dev->name, '%')) { |
| 377 | err = dev_alloc_name(dev, dev->name); |
| 378 | if (err < 0) |
| 379 | goto err_alloc_name; |
| 380 | } |
| 381 | |
| 382 | err = register_netdevice(dev); |
| 383 | if (err < 0) |
| 384 | goto err_register_dev; |
| 385 | |
| 386 | netif_carrier_off(dev); |
| 387 | |
| 388 | /* |
| 389 | * tie the deviced together |
| 390 | */ |
| 391 | |
| 392 | priv = netdev_priv(dev); |
| 393 | rcu_assign_pointer(priv->peer, peer); |
| 394 | |
| 395 | priv = netdev_priv(peer); |
| 396 | rcu_assign_pointer(priv->peer, dev); |
| 397 | return 0; |
| 398 | |
| 399 | err_register_dev: |
| 400 | /* nothing to do */ |
| 401 | err_alloc_name: |
| 402 | err_configure_peer: |
| 403 | unregister_netdevice(peer); |
| 404 | return err; |
| 405 | |
| 406 | err_register_peer: |
| 407 | free_netdev(peer); |
| 408 | return err; |
| 409 | } |
| 410 | |
| 411 | static void veth_dellink(struct net_device *dev, struct list_head *head) |
| 412 | { |
| 413 | struct veth_priv *priv; |
| 414 | struct net_device *peer; |
| 415 | |
| 416 | priv = netdev_priv(dev); |
| 417 | peer = rtnl_dereference(priv->peer); |
| 418 | |
| 419 | /* Note : dellink() is called from default_device_exit_batch(), |
| 420 | * before a rcu_synchronize() point. The devices are guaranteed |
| 421 | * not being freed before one RCU grace period. |
| 422 | */ |
| 423 | RCU_INIT_POINTER(priv->peer, NULL); |
| 424 | unregister_netdevice_queue(dev, head); |
| 425 | |
| 426 | if (peer) { |
| 427 | priv = netdev_priv(peer); |
| 428 | RCU_INIT_POINTER(priv->peer, NULL); |
| 429 | unregister_netdevice_queue(peer, head); |
| 430 | } |
| 431 | } |
| 432 | |
| 433 | static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = { |
| 434 | [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) }, |
| 435 | }; |
| 436 | |
| 437 | static struct rtnl_link_ops veth_link_ops = { |
| 438 | .kind = DRV_NAME, |
| 439 | .priv_size = sizeof(struct veth_priv), |
| 440 | .setup = veth_setup, |
| 441 | .validate = veth_validate, |
| 442 | .newlink = veth_newlink, |
| 443 | .dellink = veth_dellink, |
| 444 | .policy = veth_policy, |
| 445 | .maxtype = VETH_INFO_MAX, |
| 446 | }; |
| 447 | |
| 448 | /* |
| 449 | * init/fini |
| 450 | */ |
| 451 | |
| 452 | static __init int veth_init(void) |
| 453 | { |
| 454 | return rtnl_link_register(&veth_link_ops); |
| 455 | } |
| 456 | |
| 457 | static __exit void veth_exit(void) |
| 458 | { |
| 459 | rtnl_link_unregister(&veth_link_ops); |
| 460 | } |
| 461 | |
| 462 | module_init(veth_init); |
| 463 | module_exit(veth_exit); |
| 464 | |
| 465 | MODULE_DESCRIPTION("Virtual Ethernet Tunnel"); |
| 466 | MODULE_LICENSE("GPL v2"); |
| 467 | MODULE_ALIAS_RTNL_LINK(DRV_NAME); |