datapath/vport-internal_dev.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include <linux/hardirq.h>
  10 #include <linux/if_vlan.h>
  11 #include <linux/kernel.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/ethtool.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/version.h>
  17
  18 #include "checksum.h"
  19 #include "datapath.h"
  20 #include "vlan.h"
  21 #include "vport-generic.h"
  22 #include "vport-internal_dev.h"
  23 #include "vport-netdev.h"
  24
  25 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
  26 #define HAVE_NET_DEVICE_OPS
  27 #endif
  28
  29 struct internal_dev {
  30         struct vport *vport;
  31 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
  32         struct net_device_stats stats;
  33 #endif
  34 };
  35
  36 static struct internal_dev *internal_dev_priv(struct net_device *netdev)
  37 {
  38         return netdev_priv(netdev);
  39 }
  40
  41 /* This function is only called by the kernel network layer.*/
  42 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  43 static struct rtnl_link_stats64 *internal_dev_get_stats(struct net_device *netdev,
  44                                                         struct rtnl_link_stats64 *stats)
  45 {
  46 #else
  47 static struct net_device_stats *internal_dev_sys_stats(struct net_device *netdev)
  48 {
  49 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
  50         struct net_device_stats *stats = &internal_dev_priv(netdev)->stats;
  51 #else
  52         struct net_device_stats *stats = &netdev->stats;
  53 #endif
  54 #endif
  55         struct vport *vport = internal_dev_get_vport(netdev);
  56         struct ovs_vport_stats vport_stats;
  57
  58         vport_get_stats(vport, &vport_stats);
  59
  60         /* The tx and rx stats need to be swapped because the
  61          * switch and host OS have opposite perspectives. */
  62         stats->rx_packets       = vport_stats.tx_packets;
  63         stats->tx_packets       = vport_stats.rx_packets;
  64         stats->rx_bytes         = vport_stats.tx_bytes;
  65         stats->tx_bytes         = vport_stats.rx_bytes;
  66         stats->rx_errors        = vport_stats.tx_errors;
  67         stats->tx_errors        = vport_stats.rx_errors;
  68         stats->rx_dropped       = vport_stats.tx_dropped;
  69         stats->tx_dropped       = vport_stats.rx_dropped;
  70
  71         return stats;
  72 }
  73
  74 static int internal_dev_mac_addr(struct net_device *dev, void *p)
  75 {
  76         struct sockaddr *addr = p;
  77
  78         if (!is_valid_ether_addr(addr->sa_data))
  79                 return -EADDRNOTAVAIL;
  80         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
  81         return 0;
  82 }
  83
  84 /* Called with rcu_read_lock_bh. */
  85 static int internal_dev_xmit(struct sk_buff *skb, struct net_device *netdev)
  86 {
  87         if (unlikely(compute_ip_summed(skb, true))) {
  88                 kfree_skb(skb);
  89                 return 0;
  90         }
  91
  92         vlan_copy_skb_tci(skb);
  93         OVS_CB(skb)->flow = NULL;
  94
  95         rcu_read_lock();
  96         vport_receive(internal_dev_priv(netdev)->vport, skb);
  97         rcu_read_unlock();
  98         return 0;
  99 }
 100
 101 static int internal_dev_open(struct net_device *netdev)
 102 {
 103         netif_start_queue(netdev);
 104         return 0;
 105 }
 106
 107 static int internal_dev_stop(struct net_device *netdev)
 108 {
 109         netif_stop_queue(netdev);
 110         return 0;
 111 }
 112
 113 static void internal_dev_getinfo(struct net_device *netdev,
 114                                  struct ethtool_drvinfo *info)
 115 {
 116         strcpy(info->driver, "openvswitch");
 117 }
 118
 119 static const struct ethtool_ops internal_dev_ethtool_ops = {
 120         .get_drvinfo    = internal_dev_getinfo,
 121         .get_link       = ethtool_op_get_link,
 122 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 123         .get_sg         = ethtool_op_get_sg,
 124         .set_sg         = ethtool_op_set_sg,
 125         .get_tx_csum    = ethtool_op_get_tx_csum,
 126         .set_tx_csum    = ethtool_op_set_tx_hw_csum,
 127         .get_tso        = ethtool_op_get_tso,
 128         .set_tso        = ethtool_op_set_tso,
 129 #endif
 130 };
 131
 132 static int internal_dev_change_mtu(struct net_device *netdev, int new_mtu)
 133 {
 134         if (new_mtu < 68)
 135                 return -EINVAL;
 136
 137         netdev->mtu = new_mtu;
 138         return 0;
 139 }
 140
 141 static int internal_dev_do_ioctl(struct net_device *dev,
 142                                  struct ifreq *ifr, int cmd)
 143 {
 144         if (dp_ioctl_hook)
 145                 return dp_ioctl_hook(dev, ifr, cmd);
 146
 147         return -EOPNOTSUPP;
 148 }
 149
 150 static void internal_dev_destructor(struct net_device *dev)
 151 {
 152         struct vport *vport = internal_dev_get_vport(dev);
 153
 154         vport_free(vport);
 155         free_netdev(dev);
 156 }
 157
 158 #ifdef HAVE_NET_DEVICE_OPS
 159 static const struct net_device_ops internal_dev_netdev_ops = {
 160         .ndo_open = internal_dev_open,
 161         .ndo_stop = internal_dev_stop,
 162         .ndo_start_xmit = internal_dev_xmit,
 163         .ndo_set_mac_address = internal_dev_mac_addr,
 164         .ndo_do_ioctl = internal_dev_do_ioctl,
 165         .ndo_change_mtu = internal_dev_change_mtu,
 166 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
 167         .ndo_get_stats64 = internal_dev_get_stats,
 168 #else
 169         .ndo_get_stats = internal_dev_sys_stats,
 170 #endif
 171 };
 172 #endif
 173
 174 static void do_setup(struct net_device *netdev)
 175 {
 176         ether_setup(netdev);
 177
 178 #ifdef HAVE_NET_DEVICE_OPS
 179         netdev->netdev_ops = &internal_dev_netdev_ops;
 180 #else
 181         netdev->do_ioctl = internal_dev_do_ioctl;
 182         netdev->get_stats = internal_dev_sys_stats;
 183         netdev->hard_start_xmit = internal_dev_xmit;
 184         netdev->open = internal_dev_open;
 185         netdev->stop = internal_dev_stop;
 186         netdev->set_mac_address = internal_dev_mac_addr;
 187         netdev->change_mtu = internal_dev_change_mtu;
 188 #endif
 189
 190         netdev->priv_flags &= ~IFF_TX_SKB_SHARING;
 191         netdev->destructor = internal_dev_destructor;
 192         SET_ETHTOOL_OPS(netdev, &internal_dev_ethtool_ops);
 193         netdev->tx_queue_len = 0;
 194
 195         netdev->features = NETIF_F_LLTX | NETIF_F_SG | NETIF_F_FRAGLIST |
 196                                 NETIF_F_HIGHDMA | NETIF_F_HW_CSUM | NETIF_F_TSO;
 197
 198 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)
 199         netdev->vlan_features = netdev->features;
 200         netdev->features |= NETIF_F_HW_VLAN_TX;
 201 #endif
 202
 203 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
 204         netdev->hw_features = netdev->features & ~NETIF_F_LLTX;
 205 #endif
 206         vport_gen_rand_ether_addr(netdev->dev_addr);
 207 }
 208
 209 static struct vport *internal_dev_create(const struct vport_parms *parms)
 210 {
 211         struct vport *vport;
 212         struct netdev_vport *netdev_vport;
 213         struct internal_dev *internal_dev;
 214         int err;
 215
 216         vport = vport_alloc(sizeof(struct netdev_vport),
 217                             &internal_vport_ops, parms);
 218         if (IS_ERR(vport)) {
 219                 err = PTR_ERR(vport);
 220                 goto error;
 221         }
 222
 223         netdev_vport = netdev_vport_priv(vport);
 224
 225         netdev_vport->dev = alloc_netdev(sizeof(struct internal_dev),
 226                                          parms->name, do_setup);
 227         if (!netdev_vport->dev) {
 228                 err = -ENOMEM;
 229                 goto error_free_vport;
 230         }
 231
 232         internal_dev = internal_dev_priv(netdev_vport->dev);
 233         internal_dev->vport = vport;
 234
 235         err = register_netdevice(netdev_vport->dev);
 236         if (err)
 237                 goto error_free_netdev;
 238
 239         dev_set_promiscuity(netdev_vport->dev, 1);
 240         netif_start_queue(netdev_vport->dev);
 241
 242         return vport;
 243
 244 error_free_netdev:
 245         free_netdev(netdev_vport->dev);
 246 error_free_vport:
 247         vport_free(vport);
 248 error:
 249         return ERR_PTR(err);
 250 }
 251
 252 static void internal_dev_destroy(struct vport *vport)
 253 {
 254         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 255
 256         netif_stop_queue(netdev_vport->dev);
 257         dev_set_promiscuity(netdev_vport->dev, -1);
 258
 259         /* unregister_netdevice() waits for an RCU grace period. */
 260         unregister_netdevice(netdev_vport->dev);
 261 }
 262
 263 static int internal_dev_recv(struct vport *vport, struct sk_buff *skb)
 264 {
 265         struct net_device *netdev = netdev_vport_priv(vport)->dev;
 266         int len;
 267
 268 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37)
 269         if (unlikely(vlan_deaccel_tag(skb)))
 270                 return 0;
 271 #endif
 272
 273         len = skb->len;
 274         skb->dev = netdev;
 275         skb->pkt_type = PACKET_HOST;
 276         skb->protocol = eth_type_trans(skb, netdev);
 277         forward_ip_summed(skb, false);
 278
 279         netif_rx(skb);
 280
 281 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,29)
 282         netdev->last_rx = jiffies;
 283 #endif
 284
 285         return len;
 286 }
 287
 288 const struct vport_ops internal_vport_ops = {
 289         .type           = OVS_VPORT_TYPE_INTERNAL,
 290         .flags          = VPORT_F_REQUIRED | VPORT_F_FLOW,
 291         .create         = internal_dev_create,
 292         .destroy        = internal_dev_destroy,
 293         .set_addr       = netdev_set_addr,
 294         .get_name       = netdev_get_name,
 295         .get_addr       = netdev_get_addr,
 296         .get_kobj       = netdev_get_kobj,
 297         .get_dev_flags  = netdev_get_dev_flags,
 298         .is_running     = netdev_is_running,
 299         .get_operstate  = netdev_get_operstate,
 300         .get_ifindex    = netdev_get_ifindex,
 301         .get_mtu        = netdev_get_mtu,
 302         .send           = internal_dev_recv,
 303 };
 304
 305 int is_internal_dev(const struct net_device *netdev)
 306 {
 307 #ifdef HAVE_NET_DEVICE_OPS
 308         return netdev->netdev_ops == &internal_dev_netdev_ops;
 309 #else
 310         return netdev->open == internal_dev_open;
 311 #endif
 312 }
 313
 314 int is_internal_vport(const struct vport *vport)
 315 {
 316         return vport->ops == &internal_vport_ops;
 317 }
 318
 319 struct vport *internal_dev_get_vport(struct net_device *netdev)
 320 {
 321         if (!is_internal_dev(netdev))
 322                 return NULL;
 323
 324         return internal_dev_priv(netdev)->vport;
 325 }