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 struct internal_dev {
  26         struct vport *vport;
  27         struct net_device_stats stats;
  28 };
  29
  30 static inline struct internal_dev *internal_dev_priv(struct net_device *netdev)
  31 {
  32         return netdev_priv(netdev);
  33 }
  34
  35 /* This function is only called by the kernel network layer.  It is not a vport
  36  * get_stats() function.  If a vport get_stats() function is defined that
  37  * results in this being called it will cause infinite recursion. */
  38 static struct net_device_stats *internal_dev_sys_stats(struct net_device *netdev)
  39 {
  40         struct vport *vport = internal_dev_get_vport(netdev);
  41         struct net_device_stats *stats = &internal_dev_priv(netdev)->stats;
  42
  43         if (vport) {
  44                 struct rtnl_link_stats64 vport_stats;
  45
  46                 vport_get_stats(vport, &vport_stats);
  47
  48                 /* The tx and rx stats need to be swapped because the switch
  49                  * and host OS have opposite perspectives. */
  50                 stats->rx_packets       = vport_stats.tx_packets;
  51                 stats->tx_packets       = vport_stats.rx_packets;
  52                 stats->rx_bytes         = vport_stats.tx_bytes;
  53                 stats->tx_bytes         = vport_stats.rx_bytes;
  54                 stats->rx_errors        = vport_stats.tx_errors;
  55                 stats->tx_errors        = vport_stats.rx_errors;
  56                 stats->rx_dropped       = vport_stats.tx_dropped;
  57                 stats->tx_dropped       = vport_stats.rx_dropped;
  58                 stats->collisions       = vport_stats.collisions;
  59         }
  60
  61         return stats;
  62 }
  63
  64 static int internal_dev_mac_addr(struct net_device *dev, void *p)
  65 {
  66         struct sockaddr *addr = p;
  67
  68         if (!is_valid_ether_addr(addr->sa_data))
  69                 return -EADDRNOTAVAIL;
  70         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
  71         return 0;
  72 }
  73
  74 /* Called with rcu_read_lock and bottom-halves disabled. */
  75 static int internal_dev_xmit(struct sk_buff *skb, struct net_device *netdev)
  76 {
  77         if (unlikely(compute_ip_summed(skb, true))) {
  78                 kfree_skb(skb);
  79                 return 0;
  80         }
  81
  82         vlan_copy_skb_tci(skb);
  83         OVS_CB(skb)->flow = NULL;
  84
  85         vport_receive(internal_dev_priv(netdev)->vport, skb);
  86         return 0;
  87 }
  88
  89 static int internal_dev_open(struct net_device *netdev)
  90 {
  91         netif_start_queue(netdev);
  92         return 0;
  93 }
  94
  95 static int internal_dev_stop(struct net_device *netdev)
  96 {
  97         netif_stop_queue(netdev);
  98         return 0;
  99 }
 100
 101 static void internal_dev_getinfo(struct net_device *netdev,
 102                                  struct ethtool_drvinfo *info)
 103 {
 104         strcpy(info->driver, "openvswitch");
 105 }
 106
 107 static const struct ethtool_ops internal_dev_ethtool_ops = {
 108         .get_drvinfo    = internal_dev_getinfo,
 109         .get_link       = ethtool_op_get_link,
 110         .get_sg         = ethtool_op_get_sg,
 111         .set_sg         = ethtool_op_set_sg,
 112         .get_tx_csum    = ethtool_op_get_tx_csum,
 113         .set_tx_csum    = ethtool_op_set_tx_hw_csum,
 114         .get_tso        = ethtool_op_get_tso,
 115         .set_tso        = ethtool_op_set_tso,
 116 };
 117
 118 static int internal_dev_change_mtu(struct net_device *netdev, int new_mtu)
 119 {
 120         if (new_mtu < 68)
 121                 return -EINVAL;
 122
 123         netdev->mtu = new_mtu;
 124         return 0;
 125 }
 126
 127 static int internal_dev_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 128 {
 129         if (dp_ioctl_hook)
 130                 return dp_ioctl_hook(dev, ifr, cmd);
 131
 132         return -EOPNOTSUPP;
 133 }
 134
 135 static void internal_dev_destructor(struct net_device *dev)
 136 {
 137         struct vport *vport = internal_dev_get_vport(dev);
 138
 139         vport_free(vport);
 140         free_netdev(dev);
 141 }
 142
 143 #ifdef HAVE_NET_DEVICE_OPS
 144 static const struct net_device_ops internal_dev_netdev_ops = {
 145         .ndo_open = internal_dev_open,
 146         .ndo_stop = internal_dev_stop,
 147         .ndo_start_xmit = internal_dev_xmit,
 148         .ndo_set_mac_address = internal_dev_mac_addr,
 149         .ndo_do_ioctl = internal_dev_do_ioctl,
 150         .ndo_change_mtu = internal_dev_change_mtu,
 151         .ndo_get_stats = internal_dev_sys_stats,
 152 };
 153 #endif
 154
 155 static void do_setup(struct net_device *netdev)
 156 {
 157         ether_setup(netdev);
 158
 159 #ifdef HAVE_NET_DEVICE_OPS
 160         netdev->netdev_ops = &internal_dev_netdev_ops;
 161 #else
 162         netdev->do_ioctl = internal_dev_do_ioctl;
 163         netdev->get_stats = internal_dev_sys_stats;
 164         netdev->hard_start_xmit = internal_dev_xmit;
 165         netdev->open = internal_dev_open;
 166         netdev->stop = internal_dev_stop;
 167         netdev->set_mac_address = internal_dev_mac_addr;
 168         netdev->change_mtu = internal_dev_change_mtu;
 169 #endif
 170
 171         netdev->destructor = internal_dev_destructor;
 172         SET_ETHTOOL_OPS(netdev, &internal_dev_ethtool_ops);
 173         netdev->tx_queue_len = 0;
 174
 175         netdev->flags = IFF_BROADCAST | IFF_MULTICAST;
 176         netdev->features = NETIF_F_LLTX | NETIF_F_SG | NETIF_F_FRAGLIST |
 177                                 NETIF_F_HIGHDMA | NETIF_F_HW_CSUM | NETIF_F_TSO;
 178
 179 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)
 180         netdev->vlan_features = netdev->features;
 181         netdev->features |= NETIF_F_HW_VLAN_TX;
 182 #endif
 183
 184         vport_gen_rand_ether_addr(netdev->dev_addr);
 185 }
 186
 187 static struct vport *internal_dev_create(const struct vport_parms *parms)
 188 {
 189         struct vport *vport;
 190         struct netdev_vport *netdev_vport;
 191         struct internal_dev *internal_dev;
 192         int err;
 193
 194         vport = vport_alloc(sizeof(struct netdev_vport), &internal_vport_ops, parms);
 195         if (IS_ERR(vport)) {
 196                 err = PTR_ERR(vport);
 197                 goto error;
 198         }
 199
 200         netdev_vport = netdev_vport_priv(vport);
 201
 202         netdev_vport->dev = alloc_netdev(sizeof(struct internal_dev), parms->name, do_setup);
 203         if (!netdev_vport->dev) {
 204                 err = -ENOMEM;
 205                 goto error_free_vport;
 206         }
 207
 208         internal_dev = internal_dev_priv(netdev_vport->dev);
 209         internal_dev->vport = vport;
 210
 211         err = register_netdevice(netdev_vport->dev);
 212         if (err)
 213                 goto error_free_netdev;
 214
 215         dev_set_promiscuity(netdev_vport->dev, 1);
 216         netif_start_queue(netdev_vport->dev);
 217
 218         return vport;
 219
 220 error_free_netdev:
 221         free_netdev(netdev_vport->dev);
 222 error_free_vport:
 223         vport_free(vport);
 224 error:
 225         return ERR_PTR(err);
 226 }
 227
 228 static void internal_dev_destroy(struct vport *vport)
 229 {
 230         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 231
 232         netif_stop_queue(netdev_vport->dev);
 233         dev_set_promiscuity(netdev_vport->dev, -1);
 234
 235         /* unregister_netdevice() waits for an RCU grace period. */
 236         unregister_netdevice(netdev_vport->dev);
 237 }
 238
 239 static int internal_dev_recv(struct vport *vport, struct sk_buff *skb)
 240 {
 241         struct net_device *netdev = netdev_vport_priv(vport)->dev;
 242         int len;
 243
 244 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37)
 245         if (unlikely(vlan_deaccel_tag(skb)))
 246                 return 0;
 247 #endif
 248
 249         len = skb->len;
 250         skb->dev = netdev;
 251         skb->pkt_type = PACKET_HOST;
 252         skb->protocol = eth_type_trans(skb, netdev);
 253         forward_ip_summed(skb, false);
 254
 255         if (in_interrupt())
 256                 netif_rx(skb);
 257         else
 258                 netif_rx_ni(skb);
 259
 260 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,29)
 261         netdev->last_rx = jiffies;
 262 #endif
 263
 264         return len;
 265 }
 266
 267 const struct vport_ops internal_vport_ops = {
 268         .type           = OVS_VPORT_TYPE_INTERNAL,
 269         .flags          = VPORT_F_REQUIRED | VPORT_F_GEN_STATS | VPORT_F_FLOW,
 270         .create         = internal_dev_create,
 271         .destroy        = internal_dev_destroy,
 272         .set_addr       = netdev_set_addr,
 273         .get_name       = netdev_get_name,
 274         .get_addr       = netdev_get_addr,
 275         .get_kobj       = netdev_get_kobj,
 276         .get_dev_flags  = netdev_get_dev_flags,
 277         .is_running     = netdev_is_running,
 278         .get_operstate  = netdev_get_operstate,
 279         .get_ifindex    = netdev_get_ifindex,
 280         .get_mtu        = netdev_get_mtu,
 281         .send           = internal_dev_recv,
 282 };
 283
 284 int is_internal_dev(const struct net_device *netdev)
 285 {
 286 #ifdef HAVE_NET_DEVICE_OPS
 287         return netdev->netdev_ops == &internal_dev_netdev_ops;
 288 #else
 289         return netdev->open == internal_dev_open;
 290 #endif
 291 }
 292
 293 int is_internal_vport(const struct vport *vport)
 294 {
 295         return vport->ops == &internal_vport_ops;
 296 }
 297
 298 struct vport *internal_dev_get_vport(struct net_device *netdev)
 299 {
 300         if (!is_internal_dev(netdev))
 301                 return NULL;
 302
 303         return internal_dev_priv(netdev)->vport;
 304 }