1e314dbdcSPavel Emelyanov /* 2e314dbdcSPavel Emelyanov * drivers/net/veth.c 3e314dbdcSPavel Emelyanov * 4e314dbdcSPavel Emelyanov * Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc 5e314dbdcSPavel Emelyanov * 6e314dbdcSPavel Emelyanov * Author: Pavel Emelianov <xemul@openvz.org> 7e314dbdcSPavel Emelyanov * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com> 8e314dbdcSPavel Emelyanov * 9e314dbdcSPavel Emelyanov */ 10e314dbdcSPavel Emelyanov 11e314dbdcSPavel Emelyanov #include <linux/netdevice.h> 125a0e3ad6STejun Heo #include <linux/slab.h> 13e314dbdcSPavel Emelyanov #include <linux/ethtool.h> 14e314dbdcSPavel Emelyanov #include <linux/etherdevice.h> 15cf05c700SEric Dumazet #include <linux/u64_stats_sync.h> 16e314dbdcSPavel Emelyanov 17f7b12606SJiri Pirko #include <net/rtnetlink.h> 18e314dbdcSPavel Emelyanov #include <net/dst.h> 19e314dbdcSPavel Emelyanov #include <net/xfrm.h> 20ecef969eSStephen Hemminger #include <linux/veth.h> 219d9779e7SPaul Gortmaker #include <linux/module.h> 22e314dbdcSPavel Emelyanov 23e314dbdcSPavel Emelyanov #define DRV_NAME "veth" 24e314dbdcSPavel Emelyanov #define DRV_VERSION "1.0" 25e314dbdcSPavel Emelyanov 2638d40815SEric Biederman #define MIN_MTU 68 /* Min L3 MTU */ 2738d40815SEric Biederman #define MAX_MTU 65535 /* Max L3 MTU (arbitrary) */ 2838d40815SEric Biederman 292681128fSEric Dumazet struct pcpu_vstats { 302681128fSEric Dumazet u64 packets; 312681128fSEric Dumazet u64 bytes; 32cf05c700SEric Dumazet struct u64_stats_sync syncp; 33e314dbdcSPavel Emelyanov }; 34e314dbdcSPavel Emelyanov 35e314dbdcSPavel Emelyanov struct veth_priv { 36d0e2c55eSEric Dumazet struct net_device __rcu *peer; 372681128fSEric Dumazet atomic64_t dropped; 38163e5292SPaolo Abeni unsigned requested_headroom; 39e314dbdcSPavel Emelyanov }; 40e314dbdcSPavel Emelyanov 41e314dbdcSPavel Emelyanov /* 42e314dbdcSPavel Emelyanov * ethtool interface 43e314dbdcSPavel Emelyanov */ 44e314dbdcSPavel Emelyanov 45e314dbdcSPavel Emelyanov static struct { 46e314dbdcSPavel Emelyanov const char string[ETH_GSTRING_LEN]; 47e314dbdcSPavel Emelyanov } ethtool_stats_keys[] = { 48e314dbdcSPavel Emelyanov { "peer_ifindex" }, 49e314dbdcSPavel Emelyanov }; 50e314dbdcSPavel Emelyanov 51e314dbdcSPavel Emelyanov static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) 52e314dbdcSPavel Emelyanov { 53e314dbdcSPavel Emelyanov cmd->supported = 0; 54e314dbdcSPavel Emelyanov cmd->advertising = 0; 5570739497SDavid Decotigny ethtool_cmd_speed_set(cmd, SPEED_10000); 56e314dbdcSPavel Emelyanov cmd->duplex = DUPLEX_FULL; 57e314dbdcSPavel Emelyanov cmd->port = PORT_TP; 58e314dbdcSPavel Emelyanov cmd->phy_address = 0; 59e314dbdcSPavel Emelyanov cmd->transceiver = XCVR_INTERNAL; 60e314dbdcSPavel Emelyanov cmd->autoneg = AUTONEG_DISABLE; 61e314dbdcSPavel Emelyanov cmd->maxtxpkt = 0; 62e314dbdcSPavel Emelyanov cmd->maxrxpkt = 0; 63e314dbdcSPavel Emelyanov return 0; 64e314dbdcSPavel Emelyanov } 65e314dbdcSPavel Emelyanov 66e314dbdcSPavel Emelyanov static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) 67e314dbdcSPavel Emelyanov { 6833a5ba14SRick Jones strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 6933a5ba14SRick Jones strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 70e314dbdcSPavel Emelyanov } 71e314dbdcSPavel Emelyanov 72e314dbdcSPavel Emelyanov static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf) 73e314dbdcSPavel Emelyanov { 74e314dbdcSPavel Emelyanov switch(stringset) { 75e314dbdcSPavel Emelyanov case ETH_SS_STATS: 76e314dbdcSPavel Emelyanov memcpy(buf, ðtool_stats_keys, sizeof(ethtool_stats_keys)); 77e314dbdcSPavel Emelyanov break; 78e314dbdcSPavel Emelyanov } 79e314dbdcSPavel Emelyanov } 80e314dbdcSPavel Emelyanov 81b9f2c044SJeff Garzik static int veth_get_sset_count(struct net_device *dev, int sset) 82e314dbdcSPavel Emelyanov { 83b9f2c044SJeff Garzik switch (sset) { 84b9f2c044SJeff Garzik case ETH_SS_STATS: 85e314dbdcSPavel Emelyanov return ARRAY_SIZE(ethtool_stats_keys); 86b9f2c044SJeff Garzik default: 87b9f2c044SJeff Garzik return -EOPNOTSUPP; 88b9f2c044SJeff Garzik } 89e314dbdcSPavel Emelyanov } 90e314dbdcSPavel Emelyanov 91e314dbdcSPavel Emelyanov static void veth_get_ethtool_stats(struct net_device *dev, 92e314dbdcSPavel Emelyanov struct ethtool_stats *stats, u64 *data) 93e314dbdcSPavel Emelyanov { 94d0e2c55eSEric Dumazet struct veth_priv *priv = netdev_priv(dev); 95d0e2c55eSEric Dumazet struct net_device *peer = rtnl_dereference(priv->peer); 96e314dbdcSPavel Emelyanov 97d0e2c55eSEric Dumazet data[0] = peer ? peer->ifindex : 0; 98e314dbdcSPavel Emelyanov } 99e314dbdcSPavel Emelyanov 1000fc0b732SStephen Hemminger static const struct ethtool_ops veth_ethtool_ops = { 101e314dbdcSPavel Emelyanov .get_settings = veth_get_settings, 102e314dbdcSPavel Emelyanov .get_drvinfo = veth_get_drvinfo, 103e314dbdcSPavel Emelyanov .get_link = ethtool_op_get_link, 104e314dbdcSPavel Emelyanov .get_strings = veth_get_strings, 105b9f2c044SJeff Garzik .get_sset_count = veth_get_sset_count, 106e314dbdcSPavel Emelyanov .get_ethtool_stats = veth_get_ethtool_stats, 107e314dbdcSPavel Emelyanov }; 108e314dbdcSPavel Emelyanov 109424efe9cSStephen Hemminger static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev) 110e314dbdcSPavel Emelyanov { 1112681128fSEric Dumazet struct veth_priv *priv = netdev_priv(dev); 112d0e2c55eSEric Dumazet struct net_device *rcv; 1132681128fSEric Dumazet int length = skb->len; 114e314dbdcSPavel Emelyanov 115d0e2c55eSEric Dumazet rcu_read_lock(); 116d0e2c55eSEric Dumazet rcv = rcu_dereference(priv->peer); 117d0e2c55eSEric Dumazet if (unlikely(!rcv)) { 118d0e2c55eSEric Dumazet kfree_skb(skb); 119d0e2c55eSEric Dumazet goto drop; 120d0e2c55eSEric Dumazet } 121e314dbdcSPavel Emelyanov 1222681128fSEric Dumazet if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) { 1232681128fSEric Dumazet struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats); 124e314dbdcSPavel Emelyanov 125cf05c700SEric Dumazet u64_stats_update_begin(&stats->syncp); 1262681128fSEric Dumazet stats->bytes += length; 1272681128fSEric Dumazet stats->packets++; 128cf05c700SEric Dumazet u64_stats_update_end(&stats->syncp); 1292681128fSEric Dumazet } else { 130d0e2c55eSEric Dumazet drop: 1312681128fSEric Dumazet atomic64_inc(&priv->dropped); 1322681128fSEric Dumazet } 133d0e2c55eSEric Dumazet rcu_read_unlock(); 1346ed10654SPatrick McHardy return NETDEV_TX_OK; 135e314dbdcSPavel Emelyanov } 136e314dbdcSPavel Emelyanov 137e314dbdcSPavel Emelyanov /* 138e314dbdcSPavel Emelyanov * general routines 139e314dbdcSPavel Emelyanov */ 140e314dbdcSPavel Emelyanov 1412681128fSEric Dumazet static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev) 142e314dbdcSPavel Emelyanov { 143cf05c700SEric Dumazet struct veth_priv *priv = netdev_priv(dev); 14411687a10SDavid S. Miller int cpu; 14511687a10SDavid S. Miller 1462681128fSEric Dumazet result->packets = 0; 1472681128fSEric Dumazet result->bytes = 0; 1482b1c8b0fSEric Dumazet for_each_possible_cpu(cpu) { 1492681128fSEric Dumazet struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu); 1502681128fSEric Dumazet u64 packets, bytes; 151cf05c700SEric Dumazet unsigned int start; 152e314dbdcSPavel Emelyanov 153cf05c700SEric Dumazet do { 15457a7744eSEric W. Biederman start = u64_stats_fetch_begin_irq(&stats->syncp); 1552681128fSEric Dumazet packets = stats->packets; 1562681128fSEric Dumazet bytes = stats->bytes; 15757a7744eSEric W. Biederman } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1582681128fSEric Dumazet result->packets += packets; 1592681128fSEric Dumazet result->bytes += bytes; 160e314dbdcSPavel Emelyanov } 1612681128fSEric Dumazet return atomic64_read(&priv->dropped); 1622681128fSEric Dumazet } 1632681128fSEric Dumazet 1642681128fSEric Dumazet static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev, 1652681128fSEric Dumazet struct rtnl_link_stats64 *tot) 1662681128fSEric Dumazet { 1672681128fSEric Dumazet struct veth_priv *priv = netdev_priv(dev); 168d0e2c55eSEric Dumazet struct net_device *peer; 1692681128fSEric Dumazet struct pcpu_vstats one; 1702681128fSEric Dumazet 1712681128fSEric Dumazet tot->tx_dropped = veth_stats_one(&one, dev); 1722681128fSEric Dumazet tot->tx_bytes = one.bytes; 1732681128fSEric Dumazet tot->tx_packets = one.packets; 1742681128fSEric Dumazet 175d0e2c55eSEric Dumazet rcu_read_lock(); 176d0e2c55eSEric Dumazet peer = rcu_dereference(priv->peer); 177d0e2c55eSEric Dumazet if (peer) { 178d0e2c55eSEric Dumazet tot->rx_dropped = veth_stats_one(&one, peer); 1792681128fSEric Dumazet tot->rx_bytes = one.bytes; 1802681128fSEric Dumazet tot->rx_packets = one.packets; 181d0e2c55eSEric Dumazet } 182d0e2c55eSEric Dumazet rcu_read_unlock(); 183e314dbdcSPavel Emelyanov 1846311cc44Sstephen hemminger return tot; 185e314dbdcSPavel Emelyanov } 186e314dbdcSPavel Emelyanov 1875c70ef85SGao feng /* fake multicast ability */ 1885c70ef85SGao feng static void veth_set_multicast_list(struct net_device *dev) 1895c70ef85SGao feng { 1905c70ef85SGao feng } 1915c70ef85SGao feng 192e314dbdcSPavel Emelyanov static int veth_open(struct net_device *dev) 193e314dbdcSPavel Emelyanov { 194d0e2c55eSEric Dumazet struct veth_priv *priv = netdev_priv(dev); 195d0e2c55eSEric Dumazet struct net_device *peer = rtnl_dereference(priv->peer); 196e314dbdcSPavel Emelyanov 197d0e2c55eSEric Dumazet if (!peer) 198e314dbdcSPavel Emelyanov return -ENOTCONN; 199e314dbdcSPavel Emelyanov 200d0e2c55eSEric Dumazet if (peer->flags & IFF_UP) { 201e314dbdcSPavel Emelyanov netif_carrier_on(dev); 202d0e2c55eSEric Dumazet netif_carrier_on(peer); 203e314dbdcSPavel Emelyanov } 204e314dbdcSPavel Emelyanov return 0; 205e314dbdcSPavel Emelyanov } 206e314dbdcSPavel Emelyanov 2072cf48a10SEric W. Biederman static int veth_close(struct net_device *dev) 2082cf48a10SEric W. Biederman { 2092cf48a10SEric W. Biederman struct veth_priv *priv = netdev_priv(dev); 2102efd32eeSEric Dumazet struct net_device *peer = rtnl_dereference(priv->peer); 2112cf48a10SEric W. Biederman 2122cf48a10SEric W. Biederman netif_carrier_off(dev); 2132efd32eeSEric Dumazet if (peer) 2142efd32eeSEric Dumazet netif_carrier_off(peer); 2152cf48a10SEric W. Biederman 2162cf48a10SEric W. Biederman return 0; 2172cf48a10SEric W. Biederman } 2182cf48a10SEric W. Biederman 21938d40815SEric Biederman static int is_valid_veth_mtu(int new_mtu) 22038d40815SEric Biederman { 221807540baSEric Dumazet return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU; 22238d40815SEric Biederman } 22338d40815SEric Biederman 22438d40815SEric Biederman static int veth_change_mtu(struct net_device *dev, int new_mtu) 22538d40815SEric Biederman { 22638d40815SEric Biederman if (!is_valid_veth_mtu(new_mtu)) 22738d40815SEric Biederman return -EINVAL; 22838d40815SEric Biederman dev->mtu = new_mtu; 22938d40815SEric Biederman return 0; 23038d40815SEric Biederman } 23138d40815SEric Biederman 232e314dbdcSPavel Emelyanov static int veth_dev_init(struct net_device *dev) 233e314dbdcSPavel Emelyanov { 2341c213bd2SWANG Cong dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats); 2352681128fSEric Dumazet if (!dev->vstats) 236e314dbdcSPavel Emelyanov return -ENOMEM; 237e314dbdcSPavel Emelyanov return 0; 238e314dbdcSPavel Emelyanov } 239e314dbdcSPavel Emelyanov 24011687a10SDavid S. Miller static void veth_dev_free(struct net_device *dev) 24111687a10SDavid S. Miller { 2422681128fSEric Dumazet free_percpu(dev->vstats); 24311687a10SDavid S. Miller free_netdev(dev); 24411687a10SDavid S. Miller } 24511687a10SDavid S. Miller 246bb446c19SWANG Cong #ifdef CONFIG_NET_POLL_CONTROLLER 247bb446c19SWANG Cong static void veth_poll_controller(struct net_device *dev) 248bb446c19SWANG Cong { 249bb446c19SWANG Cong /* veth only receives frames when its peer sends one 250bb446c19SWANG Cong * Since it's a synchronous operation, we are guaranteed 251bb446c19SWANG Cong * never to have pending data when we poll for it so 252bb446c19SWANG Cong * there is nothing to do here. 253bb446c19SWANG Cong * 254bb446c19SWANG Cong * We need this though so netpoll recognizes us as an interface that 255bb446c19SWANG Cong * supports polling, which enables bridge devices in virt setups to 256bb446c19SWANG Cong * still use netconsole 257bb446c19SWANG Cong */ 258bb446c19SWANG Cong } 259bb446c19SWANG Cong #endif /* CONFIG_NET_POLL_CONTROLLER */ 260bb446c19SWANG Cong 261a45253bfSNicolas Dichtel static int veth_get_iflink(const struct net_device *dev) 262a45253bfSNicolas Dichtel { 263a45253bfSNicolas Dichtel struct veth_priv *priv = netdev_priv(dev); 264a45253bfSNicolas Dichtel struct net_device *peer; 265a45253bfSNicolas Dichtel int iflink; 266a45253bfSNicolas Dichtel 267a45253bfSNicolas Dichtel rcu_read_lock(); 268a45253bfSNicolas Dichtel peer = rcu_dereference(priv->peer); 269a45253bfSNicolas Dichtel iflink = peer ? peer->ifindex : 0; 270a45253bfSNicolas Dichtel rcu_read_unlock(); 271a45253bfSNicolas Dichtel 272a45253bfSNicolas Dichtel return iflink; 273a45253bfSNicolas Dichtel } 274a45253bfSNicolas Dichtel 275163e5292SPaolo Abeni static void veth_set_rx_headroom(struct net_device *dev, int new_hr) 276163e5292SPaolo Abeni { 277163e5292SPaolo Abeni struct veth_priv *peer_priv, *priv = netdev_priv(dev); 278163e5292SPaolo Abeni struct net_device *peer; 279163e5292SPaolo Abeni 280163e5292SPaolo Abeni if (new_hr < 0) 281163e5292SPaolo Abeni new_hr = 0; 282163e5292SPaolo Abeni 283163e5292SPaolo Abeni rcu_read_lock(); 284163e5292SPaolo Abeni peer = rcu_dereference(priv->peer); 285163e5292SPaolo Abeni if (unlikely(!peer)) 286163e5292SPaolo Abeni goto out; 287163e5292SPaolo Abeni 288163e5292SPaolo Abeni peer_priv = netdev_priv(peer); 289163e5292SPaolo Abeni priv->requested_headroom = new_hr; 290163e5292SPaolo Abeni new_hr = max(priv->requested_headroom, peer_priv->requested_headroom); 291163e5292SPaolo Abeni dev->needed_headroom = new_hr; 292163e5292SPaolo Abeni peer->needed_headroom = new_hr; 293163e5292SPaolo Abeni 294163e5292SPaolo Abeni out: 295163e5292SPaolo Abeni rcu_read_unlock(); 296163e5292SPaolo Abeni } 297163e5292SPaolo Abeni 2984456e7bdSStephen Hemminger static const struct net_device_ops veth_netdev_ops = { 2994456e7bdSStephen Hemminger .ndo_init = veth_dev_init, 3004456e7bdSStephen Hemminger .ndo_open = veth_open, 3012cf48a10SEric W. Biederman .ndo_stop = veth_close, 30200829823SStephen Hemminger .ndo_start_xmit = veth_xmit, 30338d40815SEric Biederman .ndo_change_mtu = veth_change_mtu, 3046311cc44Sstephen hemminger .ndo_get_stats64 = veth_get_stats64, 3055c70ef85SGao feng .ndo_set_rx_mode = veth_set_multicast_list, 306ee923623SDaniel Lezcano .ndo_set_mac_address = eth_mac_addr, 307bb446c19SWANG Cong #ifdef CONFIG_NET_POLL_CONTROLLER 308bb446c19SWANG Cong .ndo_poll_controller = veth_poll_controller, 309bb446c19SWANG Cong #endif 310a45253bfSNicolas Dichtel .ndo_get_iflink = veth_get_iflink, 3111a04a821SToshiaki Makita .ndo_features_check = passthru_features_check, 312163e5292SPaolo Abeni .ndo_set_rx_headroom = veth_set_rx_headroom, 3134456e7bdSStephen Hemminger }; 3144456e7bdSStephen Hemminger 3158093315aSEric Dumazet #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 3168093315aSEric Dumazet NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \ 31782d81898SEric Dumazet NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL | \ 31882d81898SEric Dumazet NETIF_F_GSO_IPIP | NETIF_F_GSO_SIT | NETIF_F_UFO | \ 31928d2b136SPatrick McHardy NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \ 32028d2b136SPatrick McHardy NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX ) 3218093315aSEric Dumazet 322e314dbdcSPavel Emelyanov static void veth_setup(struct net_device *dev) 323e314dbdcSPavel Emelyanov { 324e314dbdcSPavel Emelyanov ether_setup(dev); 325e314dbdcSPavel Emelyanov 326550fd08cSNeil Horman dev->priv_flags &= ~IFF_TX_SKB_SHARING; 32723ea5a96SHannes Frederic Sowa dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 32802f01ec1SPhil Sutter dev->priv_flags |= IFF_NO_QUEUE; 329163e5292SPaolo Abeni dev->priv_flags |= IFF_PHONY_HEADROOM; 330550fd08cSNeil Horman 3314456e7bdSStephen Hemminger dev->netdev_ops = &veth_netdev_ops; 332e314dbdcSPavel Emelyanov dev->ethtool_ops = &veth_ethtool_ops; 333e314dbdcSPavel Emelyanov dev->features |= NETIF_F_LLTX; 3348093315aSEric Dumazet dev->features |= VETH_FEATURES; 3358d0d21f4SToshiaki Makita dev->vlan_features = dev->features & 3363f8c707bSVlad Yasevich ~(NETIF_F_HW_VLAN_CTAG_TX | 3373f8c707bSVlad Yasevich NETIF_F_HW_VLAN_STAG_TX | 3383f8c707bSVlad Yasevich NETIF_F_HW_VLAN_CTAG_RX | 3393f8c707bSVlad Yasevich NETIF_F_HW_VLAN_STAG_RX); 34011687a10SDavid S. Miller dev->destructor = veth_dev_free; 341a2c725faSMichał Mirosław 3428093315aSEric Dumazet dev->hw_features = VETH_FEATURES; 34382d81898SEric Dumazet dev->hw_enc_features = VETH_FEATURES; 344e314dbdcSPavel Emelyanov } 345e314dbdcSPavel Emelyanov 346e314dbdcSPavel Emelyanov /* 347e314dbdcSPavel Emelyanov * netlink interface 348e314dbdcSPavel Emelyanov */ 349e314dbdcSPavel Emelyanov 350e314dbdcSPavel Emelyanov static int veth_validate(struct nlattr *tb[], struct nlattr *data[]) 351e314dbdcSPavel Emelyanov { 352e314dbdcSPavel Emelyanov if (tb[IFLA_ADDRESS]) { 353e314dbdcSPavel Emelyanov if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 354e314dbdcSPavel Emelyanov return -EINVAL; 355e314dbdcSPavel Emelyanov if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 356e314dbdcSPavel Emelyanov return -EADDRNOTAVAIL; 357e314dbdcSPavel Emelyanov } 35838d40815SEric Biederman if (tb[IFLA_MTU]) { 35938d40815SEric Biederman if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU]))) 36038d40815SEric Biederman return -EINVAL; 36138d40815SEric Biederman } 362e314dbdcSPavel Emelyanov return 0; 363e314dbdcSPavel Emelyanov } 364e314dbdcSPavel Emelyanov 365e314dbdcSPavel Emelyanov static struct rtnl_link_ops veth_link_ops; 366e314dbdcSPavel Emelyanov 36781adee47SEric W. Biederman static int veth_newlink(struct net *src_net, struct net_device *dev, 368e314dbdcSPavel Emelyanov struct nlattr *tb[], struct nlattr *data[]) 369e314dbdcSPavel Emelyanov { 370e314dbdcSPavel Emelyanov int err; 371e314dbdcSPavel Emelyanov struct net_device *peer; 372e314dbdcSPavel Emelyanov struct veth_priv *priv; 373e314dbdcSPavel Emelyanov char ifname[IFNAMSIZ]; 374e314dbdcSPavel Emelyanov struct nlattr *peer_tb[IFLA_MAX + 1], **tbp; 3755517750fSTom Gundersen unsigned char name_assign_type; 3763729d502SPatrick McHardy struct ifinfomsg *ifmp; 37781adee47SEric W. Biederman struct net *net; 378e314dbdcSPavel Emelyanov 379e314dbdcSPavel Emelyanov /* 380e314dbdcSPavel Emelyanov * create and register peer first 381e314dbdcSPavel Emelyanov */ 382e314dbdcSPavel Emelyanov if (data != NULL && data[VETH_INFO_PEER] != NULL) { 383e314dbdcSPavel Emelyanov struct nlattr *nla_peer; 384e314dbdcSPavel Emelyanov 385e314dbdcSPavel Emelyanov nla_peer = data[VETH_INFO_PEER]; 3863729d502SPatrick McHardy ifmp = nla_data(nla_peer); 387f7b12606SJiri Pirko err = rtnl_nla_parse_ifla(peer_tb, 388e314dbdcSPavel Emelyanov nla_data(nla_peer) + sizeof(struct ifinfomsg), 389f7b12606SJiri Pirko nla_len(nla_peer) - sizeof(struct ifinfomsg)); 390e314dbdcSPavel Emelyanov if (err < 0) 391e314dbdcSPavel Emelyanov return err; 392e314dbdcSPavel Emelyanov 393e314dbdcSPavel Emelyanov err = veth_validate(peer_tb, NULL); 394e314dbdcSPavel Emelyanov if (err < 0) 395e314dbdcSPavel Emelyanov return err; 396e314dbdcSPavel Emelyanov 397e314dbdcSPavel Emelyanov tbp = peer_tb; 3983729d502SPatrick McHardy } else { 3993729d502SPatrick McHardy ifmp = NULL; 400e314dbdcSPavel Emelyanov tbp = tb; 4013729d502SPatrick McHardy } 402e314dbdcSPavel Emelyanov 4035517750fSTom Gundersen if (tbp[IFLA_IFNAME]) { 404e314dbdcSPavel Emelyanov nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ); 4055517750fSTom Gundersen name_assign_type = NET_NAME_USER; 4065517750fSTom Gundersen } else { 407e314dbdcSPavel Emelyanov snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d"); 4085517750fSTom Gundersen name_assign_type = NET_NAME_ENUM; 4095517750fSTom Gundersen } 410e314dbdcSPavel Emelyanov 41181adee47SEric W. Biederman net = rtnl_link_get_net(src_net, tbp); 41281adee47SEric W. Biederman if (IS_ERR(net)) 41381adee47SEric W. Biederman return PTR_ERR(net); 41481adee47SEric W. Biederman 4155517750fSTom Gundersen peer = rtnl_create_link(net, ifname, name_assign_type, 4165517750fSTom Gundersen &veth_link_ops, tbp); 41781adee47SEric W. Biederman if (IS_ERR(peer)) { 41881adee47SEric W. Biederman put_net(net); 419e314dbdcSPavel Emelyanov return PTR_ERR(peer); 42081adee47SEric W. Biederman } 421e314dbdcSPavel Emelyanov 422e314dbdcSPavel Emelyanov if (tbp[IFLA_ADDRESS] == NULL) 423f2cedb63SDanny Kukawka eth_hw_addr_random(peer); 424e314dbdcSPavel Emelyanov 425e6f8f1a7SPavel Emelyanov if (ifmp && (dev->ifindex != 0)) 426e6f8f1a7SPavel Emelyanov peer->ifindex = ifmp->ifi_index; 427e6f8f1a7SPavel Emelyanov 428e314dbdcSPavel Emelyanov err = register_netdevice(peer); 42981adee47SEric W. Biederman put_net(net); 43081adee47SEric W. Biederman net = NULL; 431e314dbdcSPavel Emelyanov if (err < 0) 432e314dbdcSPavel Emelyanov goto err_register_peer; 433e314dbdcSPavel Emelyanov 434e314dbdcSPavel Emelyanov netif_carrier_off(peer); 435e314dbdcSPavel Emelyanov 4363729d502SPatrick McHardy err = rtnl_configure_link(peer, ifmp); 4373729d502SPatrick McHardy if (err < 0) 4383729d502SPatrick McHardy goto err_configure_peer; 4393729d502SPatrick McHardy 440e314dbdcSPavel Emelyanov /* 441e314dbdcSPavel Emelyanov * register dev last 442e314dbdcSPavel Emelyanov * 443e314dbdcSPavel Emelyanov * note, that since we've registered new device the dev's name 444e314dbdcSPavel Emelyanov * should be re-allocated 445e314dbdcSPavel Emelyanov */ 446e314dbdcSPavel Emelyanov 447e314dbdcSPavel Emelyanov if (tb[IFLA_ADDRESS] == NULL) 448f2cedb63SDanny Kukawka eth_hw_addr_random(dev); 449e314dbdcSPavel Emelyanov 4506c8c4446SJiri Pirko if (tb[IFLA_IFNAME]) 4516c8c4446SJiri Pirko nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ); 4526c8c4446SJiri Pirko else 4536c8c4446SJiri Pirko snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d"); 4546c8c4446SJiri Pirko 455e314dbdcSPavel Emelyanov err = register_netdevice(dev); 456e314dbdcSPavel Emelyanov if (err < 0) 457e314dbdcSPavel Emelyanov goto err_register_dev; 458e314dbdcSPavel Emelyanov 459e314dbdcSPavel Emelyanov netif_carrier_off(dev); 460e314dbdcSPavel Emelyanov 461e314dbdcSPavel Emelyanov /* 462e314dbdcSPavel Emelyanov * tie the deviced together 463e314dbdcSPavel Emelyanov */ 464e314dbdcSPavel Emelyanov 465e314dbdcSPavel Emelyanov priv = netdev_priv(dev); 466d0e2c55eSEric Dumazet rcu_assign_pointer(priv->peer, peer); 467e314dbdcSPavel Emelyanov 468e314dbdcSPavel Emelyanov priv = netdev_priv(peer); 469d0e2c55eSEric Dumazet rcu_assign_pointer(priv->peer, dev); 470e314dbdcSPavel Emelyanov return 0; 471e314dbdcSPavel Emelyanov 472e314dbdcSPavel Emelyanov err_register_dev: 473e314dbdcSPavel Emelyanov /* nothing to do */ 4743729d502SPatrick McHardy err_configure_peer: 475e314dbdcSPavel Emelyanov unregister_netdevice(peer); 476e314dbdcSPavel Emelyanov return err; 477e314dbdcSPavel Emelyanov 478e314dbdcSPavel Emelyanov err_register_peer: 479e314dbdcSPavel Emelyanov free_netdev(peer); 480e314dbdcSPavel Emelyanov return err; 481e314dbdcSPavel Emelyanov } 482e314dbdcSPavel Emelyanov 48323289a37SEric Dumazet static void veth_dellink(struct net_device *dev, struct list_head *head) 484e314dbdcSPavel Emelyanov { 485e314dbdcSPavel Emelyanov struct veth_priv *priv; 486e314dbdcSPavel Emelyanov struct net_device *peer; 487e314dbdcSPavel Emelyanov 488e314dbdcSPavel Emelyanov priv = netdev_priv(dev); 489d0e2c55eSEric Dumazet peer = rtnl_dereference(priv->peer); 490d0e2c55eSEric Dumazet 491d0e2c55eSEric Dumazet /* Note : dellink() is called from default_device_exit_batch(), 492d0e2c55eSEric Dumazet * before a rcu_synchronize() point. The devices are guaranteed 493d0e2c55eSEric Dumazet * not being freed before one RCU grace period. 494d0e2c55eSEric Dumazet */ 495d0e2c55eSEric Dumazet RCU_INIT_POINTER(priv->peer, NULL); 496f45a5c26SEric Dumazet unregister_netdevice_queue(dev, head); 497d0e2c55eSEric Dumazet 498f45a5c26SEric Dumazet if (peer) { 499d0e2c55eSEric Dumazet priv = netdev_priv(peer); 500d0e2c55eSEric Dumazet RCU_INIT_POINTER(priv->peer, NULL); 50124540535SEric Dumazet unregister_netdevice_queue(peer, head); 502e314dbdcSPavel Emelyanov } 503f45a5c26SEric Dumazet } 504e314dbdcSPavel Emelyanov 50523711438SThomas Graf static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = { 50623711438SThomas Graf [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) }, 50723711438SThomas Graf }; 508e314dbdcSPavel Emelyanov 509e5f4e7b9SNicolas Dichtel static struct net *veth_get_link_net(const struct net_device *dev) 510e5f4e7b9SNicolas Dichtel { 511e5f4e7b9SNicolas Dichtel struct veth_priv *priv = netdev_priv(dev); 512e5f4e7b9SNicolas Dichtel struct net_device *peer = rtnl_dereference(priv->peer); 513e5f4e7b9SNicolas Dichtel 514e5f4e7b9SNicolas Dichtel return peer ? dev_net(peer) : dev_net(dev); 515e5f4e7b9SNicolas Dichtel } 516e5f4e7b9SNicolas Dichtel 517e314dbdcSPavel Emelyanov static struct rtnl_link_ops veth_link_ops = { 518e314dbdcSPavel Emelyanov .kind = DRV_NAME, 519e314dbdcSPavel Emelyanov .priv_size = sizeof(struct veth_priv), 520e314dbdcSPavel Emelyanov .setup = veth_setup, 521e314dbdcSPavel Emelyanov .validate = veth_validate, 522e314dbdcSPavel Emelyanov .newlink = veth_newlink, 523e314dbdcSPavel Emelyanov .dellink = veth_dellink, 524e314dbdcSPavel Emelyanov .policy = veth_policy, 525e314dbdcSPavel Emelyanov .maxtype = VETH_INFO_MAX, 526e5f4e7b9SNicolas Dichtel .get_link_net = veth_get_link_net, 527e314dbdcSPavel Emelyanov }; 528e314dbdcSPavel Emelyanov 529e314dbdcSPavel Emelyanov /* 530e314dbdcSPavel Emelyanov * init/fini 531e314dbdcSPavel Emelyanov */ 532e314dbdcSPavel Emelyanov 533e314dbdcSPavel Emelyanov static __init int veth_init(void) 534e314dbdcSPavel Emelyanov { 535e314dbdcSPavel Emelyanov return rtnl_link_register(&veth_link_ops); 536e314dbdcSPavel Emelyanov } 537e314dbdcSPavel Emelyanov 538e314dbdcSPavel Emelyanov static __exit void veth_exit(void) 539e314dbdcSPavel Emelyanov { 54068365458SPatrick McHardy rtnl_link_unregister(&veth_link_ops); 541e314dbdcSPavel Emelyanov } 542e314dbdcSPavel Emelyanov 543e314dbdcSPavel Emelyanov module_init(veth_init); 544e314dbdcSPavel Emelyanov module_exit(veth_exit); 545e314dbdcSPavel Emelyanov 546e314dbdcSPavel Emelyanov MODULE_DESCRIPTION("Virtual Ethernet Tunnel"); 547e314dbdcSPavel Emelyanov MODULE_LICENSE("GPL v2"); 548e314dbdcSPavel Emelyanov MODULE_ALIAS_RTNL_LINK(DRV_NAME); 549