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 /* 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 { 134 drop: 135 atomic64_inc(&priv->dropped); 136 } 137 rcu_read_unlock(); 138 return NETDEV_TX_OK; 139 } 140 141 /* 142 * general routines 143 */ 144 145 static 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 } 167 168 static 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 } 190 191 /* fake multicast ability */ 192 static void veth_set_multicast_list(struct net_device *dev) 193 { 194 } 195 196 static int veth_open(struct net_device *dev) 197 { 198 struct veth_priv *priv = netdev_priv(dev); 199 struct net_device *peer = rtnl_dereference(priv->peer); 200 201 if (!peer) 202 return -ENOTCONN; 203 204 if (peer->flags & IFF_UP) { 205 netif_carrier_on(dev); 206 netif_carrier_on(peer); 207 } 208 return 0; 209 } 210 211 static int veth_close(struct net_device *dev) 212 { 213 struct veth_priv *priv = netdev_priv(dev); 214 struct net_device *peer = rtnl_dereference(priv->peer); 215 216 netif_carrier_off(dev); 217 if (peer) 218 netif_carrier_off(peer); 219 220 return 0; 221 } 222 223 static int is_valid_veth_mtu(int new_mtu) 224 { 225 return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU; 226 } 227 228 static int veth_change_mtu(struct net_device *dev, int new_mtu) 229 { 230 if (!is_valid_veth_mtu(new_mtu)) 231 return -EINVAL; 232 dev->mtu = new_mtu; 233 return 0; 234 } 235 236 static int veth_dev_init(struct net_device *dev) 237 { 238 int i; 239 240 dev->vstats = alloc_percpu(struct pcpu_vstats); 241 if (!dev->vstats) 242 return -ENOMEM; 243 244 for_each_possible_cpu(i) { 245 struct pcpu_vstats *veth_stats; 246 veth_stats = per_cpu_ptr(dev->vstats, i); 247 u64_stats_init(&veth_stats->syncp); 248 } 249 250 return 0; 251 } 252 253 static void veth_dev_free(struct net_device *dev) 254 { 255 free_percpu(dev->vstats); 256 free_netdev(dev); 257 } 258 259 static const struct net_device_ops veth_netdev_ops = { 260 .ndo_init = veth_dev_init, 261 .ndo_open = veth_open, 262 .ndo_stop = veth_close, 263 .ndo_start_xmit = veth_xmit, 264 .ndo_change_mtu = veth_change_mtu, 265 .ndo_get_stats64 = veth_get_stats64, 266 .ndo_set_rx_mode = veth_set_multicast_list, 267 .ndo_set_mac_address = eth_mac_addr, 268 }; 269 270 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 271 NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \ 272 NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL | \ 273 NETIF_F_GSO_IPIP | NETIF_F_GSO_SIT | NETIF_F_UFO | \ 274 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \ 275 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX ) 276 277 static void veth_setup(struct net_device *dev) 278 { 279 ether_setup(dev); 280 281 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 282 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 283 284 dev->netdev_ops = &veth_netdev_ops; 285 dev->ethtool_ops = &veth_ethtool_ops; 286 dev->features |= NETIF_F_LLTX; 287 dev->features |= VETH_FEATURES; 288 dev->vlan_features = dev->features & 289 ~(NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX); 290 dev->destructor = veth_dev_free; 291 292 dev->hw_features = VETH_FEATURES; 293 dev->hw_enc_features = VETH_FEATURES; 294 } 295 296 /* 297 * netlink interface 298 */ 299 300 static int veth_validate(struct nlattr *tb[], struct nlattr *data[]) 301 { 302 if (tb[IFLA_ADDRESS]) { 303 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 304 return -EINVAL; 305 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 306 return -EADDRNOTAVAIL; 307 } 308 if (tb[IFLA_MTU]) { 309 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU]))) 310 return -EINVAL; 311 } 312 return 0; 313 } 314 315 static struct rtnl_link_ops veth_link_ops; 316 317 static int veth_newlink(struct net *src_net, struct net_device *dev, 318 struct nlattr *tb[], struct nlattr *data[]) 319 { 320 int err; 321 struct net_device *peer; 322 struct veth_priv *priv; 323 char ifname[IFNAMSIZ]; 324 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp; 325 struct ifinfomsg *ifmp; 326 struct net *net; 327 328 /* 329 * create and register peer first 330 */ 331 if (data != NULL && data[VETH_INFO_PEER] != NULL) { 332 struct nlattr *nla_peer; 333 334 nla_peer = data[VETH_INFO_PEER]; 335 ifmp = nla_data(nla_peer); 336 err = nla_parse(peer_tb, IFLA_MAX, 337 nla_data(nla_peer) + sizeof(struct ifinfomsg), 338 nla_len(nla_peer) - sizeof(struct ifinfomsg), 339 ifla_policy); 340 if (err < 0) 341 return err; 342 343 err = veth_validate(peer_tb, NULL); 344 if (err < 0) 345 return err; 346 347 tbp = peer_tb; 348 } else { 349 ifmp = NULL; 350 tbp = tb; 351 } 352 353 if (tbp[IFLA_IFNAME]) 354 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ); 355 else 356 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d"); 357 358 net = rtnl_link_get_net(src_net, tbp); 359 if (IS_ERR(net)) 360 return PTR_ERR(net); 361 362 peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp); 363 if (IS_ERR(peer)) { 364 put_net(net); 365 return PTR_ERR(peer); 366 } 367 368 if (tbp[IFLA_ADDRESS] == NULL) 369 eth_hw_addr_random(peer); 370 371 if (ifmp && (dev->ifindex != 0)) 372 peer->ifindex = ifmp->ifi_index; 373 374 err = register_netdevice(peer); 375 put_net(net); 376 net = NULL; 377 if (err < 0) 378 goto err_register_peer; 379 380 netif_carrier_off(peer); 381 382 err = rtnl_configure_link(peer, ifmp); 383 if (err < 0) 384 goto err_configure_peer; 385 386 /* 387 * register dev last 388 * 389 * note, that since we've registered new device the dev's name 390 * should be re-allocated 391 */ 392 393 if (tb[IFLA_ADDRESS] == NULL) 394 eth_hw_addr_random(dev); 395 396 if (tb[IFLA_IFNAME]) 397 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ); 398 else 399 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d"); 400 401 err = register_netdevice(dev); 402 if (err < 0) 403 goto err_register_dev; 404 405 netif_carrier_off(dev); 406 407 /* 408 * tie the deviced together 409 */ 410 411 priv = netdev_priv(dev); 412 rcu_assign_pointer(priv->peer, peer); 413 414 priv = netdev_priv(peer); 415 rcu_assign_pointer(priv->peer, dev); 416 return 0; 417 418 err_register_dev: 419 /* nothing to do */ 420 err_configure_peer: 421 unregister_netdevice(peer); 422 return err; 423 424 err_register_peer: 425 free_netdev(peer); 426 return err; 427 } 428 429 static void veth_dellink(struct net_device *dev, struct list_head *head) 430 { 431 struct veth_priv *priv; 432 struct net_device *peer; 433 434 priv = netdev_priv(dev); 435 peer = rtnl_dereference(priv->peer); 436 437 /* Note : dellink() is called from default_device_exit_batch(), 438 * before a rcu_synchronize() point. The devices are guaranteed 439 * not being freed before one RCU grace period. 440 */ 441 RCU_INIT_POINTER(priv->peer, NULL); 442 unregister_netdevice_queue(dev, head); 443 444 if (peer) { 445 priv = netdev_priv(peer); 446 RCU_INIT_POINTER(priv->peer, NULL); 447 unregister_netdevice_queue(peer, head); 448 } 449 } 450 451 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = { 452 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) }, 453 }; 454 455 static struct rtnl_link_ops veth_link_ops = { 456 .kind = DRV_NAME, 457 .priv_size = sizeof(struct veth_priv), 458 .setup = veth_setup, 459 .validate = veth_validate, 460 .newlink = veth_newlink, 461 .dellink = veth_dellink, 462 .policy = veth_policy, 463 .maxtype = VETH_INFO_MAX, 464 }; 465 466 /* 467 * init/fini 468 */ 469 470 static __init int veth_init(void) 471 { 472 return rtnl_link_register(&veth_link_ops); 473 } 474 475 static __exit void veth_exit(void) 476 { 477 rtnl_link_unregister(&veth_link_ops); 478 } 479 480 module_init(veth_init); 481 module_exit(veth_exit); 482 483 MODULE_DESCRIPTION("Virtual Ethernet Tunnel"); 484 MODULE_LICENSE("GPL v2"); 485 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 486