veth.c revision d0e2c55e7c940a3ee91e9e23a2683b593690f1e9
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, ðtool_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); 209 210 netif_carrier_off(dev); 211 netif_carrier_off(rtnl_dereference(priv->peer)); 212 213 return 0; 214} 215 216static int is_valid_veth_mtu(int new_mtu) 217{ 218 return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU; 219} 220 221static int veth_change_mtu(struct net_device *dev, int new_mtu) 222{ 223 if (!is_valid_veth_mtu(new_mtu)) 224 return -EINVAL; 225 dev->mtu = new_mtu; 226 return 0; 227} 228 229static int veth_dev_init(struct net_device *dev) 230{ 231 dev->vstats = alloc_percpu(struct pcpu_vstats); 232 if (!dev->vstats) 233 return -ENOMEM; 234 235 return 0; 236} 237 238static void veth_dev_free(struct net_device *dev) 239{ 240 free_percpu(dev->vstats); 241 free_netdev(dev); 242} 243 244static const struct net_device_ops veth_netdev_ops = { 245 .ndo_init = veth_dev_init, 246 .ndo_open = veth_open, 247 .ndo_stop = veth_close, 248 .ndo_start_xmit = veth_xmit, 249 .ndo_change_mtu = veth_change_mtu, 250 .ndo_get_stats64 = veth_get_stats64, 251 .ndo_set_mac_address = eth_mac_addr, 252}; 253 254#define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 255 NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \ 256 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX) 257 258static void veth_setup(struct net_device *dev) 259{ 260 ether_setup(dev); 261 262 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 263 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 264 265 dev->netdev_ops = &veth_netdev_ops; 266 dev->ethtool_ops = &veth_ethtool_ops; 267 dev->features |= NETIF_F_LLTX; 268 dev->features |= VETH_FEATURES; 269 dev->destructor = veth_dev_free; 270 271 dev->hw_features = VETH_FEATURES; 272} 273 274/* 275 * netlink interface 276 */ 277 278static int veth_validate(struct nlattr *tb[], struct nlattr *data[]) 279{ 280 if (tb[IFLA_ADDRESS]) { 281 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 282 return -EINVAL; 283 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 284 return -EADDRNOTAVAIL; 285 } 286 if (tb[IFLA_MTU]) { 287 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU]))) 288 return -EINVAL; 289 } 290 return 0; 291} 292 293static struct rtnl_link_ops veth_link_ops; 294 295static int veth_newlink(struct net *src_net, struct net_device *dev, 296 struct nlattr *tb[], struct nlattr *data[]) 297{ 298 int err; 299 struct net_device *peer; 300 struct veth_priv *priv; 301 char ifname[IFNAMSIZ]; 302 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp; 303 struct ifinfomsg *ifmp; 304 struct net *net; 305 306 /* 307 * create and register peer first 308 */ 309 if (data != NULL && data[VETH_INFO_PEER] != NULL) { 310 struct nlattr *nla_peer; 311 312 nla_peer = data[VETH_INFO_PEER]; 313 ifmp = nla_data(nla_peer); 314 err = nla_parse(peer_tb, IFLA_MAX, 315 nla_data(nla_peer) + sizeof(struct ifinfomsg), 316 nla_len(nla_peer) - sizeof(struct ifinfomsg), 317 ifla_policy); 318 if (err < 0) 319 return err; 320 321 err = veth_validate(peer_tb, NULL); 322 if (err < 0) 323 return err; 324 325 tbp = peer_tb; 326 } else { 327 ifmp = NULL; 328 tbp = tb; 329 } 330 331 if (tbp[IFLA_IFNAME]) 332 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ); 333 else 334 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d"); 335 336 net = rtnl_link_get_net(src_net, tbp); 337 if (IS_ERR(net)) 338 return PTR_ERR(net); 339 340 peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp); 341 if (IS_ERR(peer)) { 342 put_net(net); 343 return PTR_ERR(peer); 344 } 345 346 if (tbp[IFLA_ADDRESS] == NULL) 347 eth_hw_addr_random(peer); 348 349 if (ifmp && (dev->ifindex != 0)) 350 peer->ifindex = ifmp->ifi_index; 351 352 err = register_netdevice(peer); 353 put_net(net); 354 net = NULL; 355 if (err < 0) 356 goto err_register_peer; 357 358 netif_carrier_off(peer); 359 360 err = rtnl_configure_link(peer, ifmp); 361 if (err < 0) 362 goto err_configure_peer; 363 364 /* 365 * register dev last 366 * 367 * note, that since we've registered new device the dev's name 368 * should be re-allocated 369 */ 370 371 if (tb[IFLA_ADDRESS] == NULL) 372 eth_hw_addr_random(dev); 373 374 if (tb[IFLA_IFNAME]) 375 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ); 376 else 377 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d"); 378 379 if (strchr(dev->name, '%')) { 380 err = dev_alloc_name(dev, dev->name); 381 if (err < 0) 382 goto err_alloc_name; 383 } 384 385 err = register_netdevice(dev); 386 if (err < 0) 387 goto err_register_dev; 388 389 netif_carrier_off(dev); 390 391 /* 392 * tie the deviced together 393 */ 394 395 priv = netdev_priv(dev); 396 rcu_assign_pointer(priv->peer, peer); 397 398 priv = netdev_priv(peer); 399 rcu_assign_pointer(priv->peer, dev); 400 return 0; 401 402err_register_dev: 403 /* nothing to do */ 404err_alloc_name: 405err_configure_peer: 406 unregister_netdevice(peer); 407 return err; 408 409err_register_peer: 410 free_netdev(peer); 411 return err; 412} 413 414static void veth_dellink(struct net_device *dev, struct list_head *head) 415{ 416 struct veth_priv *priv; 417 struct net_device *peer; 418 419 priv = netdev_priv(dev); 420 peer = rtnl_dereference(priv->peer); 421 422 /* Note : dellink() is called from default_device_exit_batch(), 423 * before a rcu_synchronize() point. The devices are guaranteed 424 * not being freed before one RCU grace period. 425 */ 426 RCU_INIT_POINTER(priv->peer, NULL); 427 428 priv = netdev_priv(peer); 429 RCU_INIT_POINTER(priv->peer, NULL); 430 431 unregister_netdevice_queue(dev, head); 432 unregister_netdevice_queue(peer, head); 433} 434 435static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = { 436 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) }, 437}; 438 439static struct rtnl_link_ops veth_link_ops = { 440 .kind = DRV_NAME, 441 .priv_size = sizeof(struct veth_priv), 442 .setup = veth_setup, 443 .validate = veth_validate, 444 .newlink = veth_newlink, 445 .dellink = veth_dellink, 446 .policy = veth_policy, 447 .maxtype = VETH_INFO_MAX, 448}; 449 450/* 451 * init/fini 452 */ 453 454static __init int veth_init(void) 455{ 456 return rtnl_link_register(&veth_link_ops); 457} 458 459static __exit void veth_exit(void) 460{ 461 rtnl_link_unregister(&veth_link_ops); 462} 463 464module_init(veth_init); 465module_exit(veth_exit); 466 467MODULE_DESCRIPTION("Virtual Ethernet Tunnel"); 468MODULE_LICENSE("GPL v2"); 469MODULE_ALIAS_RTNL_LINK(DRV_NAME); 470