datapath/vport-internal_dev.c

   1 /*
   2  * Copyright (c) 2009, 2010 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/kernel.h>
  10 #include <linux/netdevice.h>
  11 #include <linux/etherdevice.h>
  12 #include <linux/ethtool.h>
  13 #include <linux/rcupdate.h>
  14 #include <linux/skbuff.h>
  15
  16 #include "datapath.h"
  17 #include "vport-generic.h"
  18 #include "vport-internal_dev.h"
  19 #include "vport-netdev.h"
  20
  21 struct internal_dev {
  22         struct vport *attached_vport, *vport;
  23         struct net_device_stats stats;
  24 };
  25
  26 static inline struct internal_dev *internal_dev_priv(struct net_device *netdev)
  27 {
  28         return netdev_priv(netdev);
  29 }
  30
  31 /* This function is only called by the kernel network layer.  It is not a vport
  32  * get_stats() function.  If a vport get_stats() function is defined that
  33  * results in this being called it will cause infinite recursion. */
  34 static struct net_device_stats *internal_dev_sys_stats(struct net_device *netdev)
  35 {
  36         struct vport *vport = internal_dev_get_vport(netdev);
  37         struct net_device_stats *stats = &internal_dev_priv(netdev)->stats;
  38
  39         if (vport) {
  40                 struct odp_vport_stats vport_stats;
  41
  42                 vport_get_stats(vport, &vport_stats);
  43
  44                 /* The tx and rx stats need to be swapped because the switch
  45                  * and host OS have opposite perspectives. */
  46                 stats->rx_packets       = vport_stats.tx_packets;
  47                 stats->tx_packets       = vport_stats.rx_packets;
  48                 stats->rx_bytes         = vport_stats.tx_bytes;
  49                 stats->tx_bytes         = vport_stats.rx_bytes;
  50                 stats->rx_errors        = vport_stats.tx_errors;
  51                 stats->tx_errors        = vport_stats.rx_errors;
  52                 stats->rx_dropped       = vport_stats.tx_dropped;
  53                 stats->tx_dropped       = vport_stats.rx_dropped;
  54                 stats->collisions       = vport_stats.collisions;
  55         }
  56
  57         return stats;
  58 }
  59
  60 static int internal_dev_mac_addr(struct net_device *dev, void *p)
  61 {
  62         struct sockaddr *addr = p;
  63
  64         if (!is_valid_ether_addr(addr->sa_data))
  65                 return -EADDRNOTAVAIL;
  66         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
  67         return 0;
  68 }
  69
  70 /* Called with rcu_read_lock and bottom-halves disabled. */
  71 static int internal_dev_xmit(struct sk_buff *skb, struct net_device *netdev)
  72 {
  73         struct internal_dev *internal_dev = internal_dev_priv(netdev);
  74         struct vport *vport = rcu_dereference(internal_dev->vport);
  75
  76         /* We need our own clone. */
  77         skb = skb_share_check(skb, GFP_ATOMIC);
  78         if (unlikely(!skb)) {
  79                 vport_record_error(vport, VPORT_E_RX_DROPPED);
  80                 return 0;
  81         }
  82
  83         skb_reset_mac_header(skb);
  84         compute_ip_summed(skb, true);
  85         OVS_CB(skb)->flow = NULL;
  86
  87         vport_receive(vport, skb);
  88
  89         return 0;
  90 }
  91
  92 static int internal_dev_open(struct net_device *netdev)
  93 {
  94         netif_start_queue(netdev);
  95         return 0;
  96 }
  97
  98 static int internal_dev_stop(struct net_device *netdev)
  99 {
 100         netif_stop_queue(netdev);
 101         return 0;
 102 }
 103
 104 static void internal_dev_getinfo(struct net_device *netdev,
 105                                  struct ethtool_drvinfo *info)
 106 {
 107         struct vport *vport = internal_dev_get_vport(netdev);
 108         struct dp_port *dp_port;
 109
 110         strcpy(info->driver, "openvswitch");
 111
 112         if (!vport)
 113                 return;
 114
 115         dp_port = vport_get_dp_port(vport);
 116         if (dp_port)
 117                 sprintf(info->bus_info, "%d.%d", dp_port->dp->dp_idx, dp_port->port_no);
 118 }
 119
 120 static struct ethtool_ops internal_dev_ethtool_ops = {
 121         .get_drvinfo    = internal_dev_getinfo,
 122         .get_link       = ethtool_op_get_link,
 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 };
 130
 131 static int internal_dev_change_mtu(struct net_device *netdev, int new_mtu)
 132 {
 133         struct vport *vport = internal_dev_get_vport(netdev);
 134
 135         if (new_mtu < 68)
 136                 return -EINVAL;
 137
 138         if (vport) {
 139                 struct dp_port *dp_port = vport_get_dp_port(vport);
 140
 141                 if (dp_port) {
 142                         if (new_mtu > dp_min_mtu(dp_port->dp))
 143                                 return -EINVAL;
 144                 }
 145         }
 146
 147         netdev->mtu = new_mtu;
 148         return 0;
 149 }
 150
 151 static void internal_dev_free(struct net_device *netdev)
 152 {
 153         free_netdev(netdev);
 154 }
 155
 156 static int internal_dev_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 157 {
 158         if (dp_ioctl_hook)
 159                 return dp_ioctl_hook(dev, ifr, cmd);
 160
 161         return -EOPNOTSUPP;
 162 }
 163
 164 #ifdef HAVE_NET_DEVICE_OPS
 165 static const struct net_device_ops internal_dev_netdev_ops = {
 166         .ndo_open = internal_dev_open,
 167         .ndo_stop = internal_dev_stop,
 168         .ndo_start_xmit = internal_dev_xmit,
 169         .ndo_set_mac_address = internal_dev_mac_addr,
 170         .ndo_do_ioctl = internal_dev_do_ioctl,
 171         .ndo_change_mtu = internal_dev_change_mtu,
 172         .ndo_get_stats = internal_dev_sys_stats,
 173 };
 174 #endif
 175
 176 static void do_setup(struct net_device *netdev)
 177 {
 178         ether_setup(netdev);
 179
 180 #ifdef HAVE_NET_DEVICE_OPS
 181         netdev->netdev_ops = &internal_dev_netdev_ops;
 182 #else
 183         netdev->do_ioctl = internal_dev_do_ioctl;
 184         netdev->get_stats = internal_dev_sys_stats;
 185         netdev->hard_start_xmit = internal_dev_xmit;
 186         netdev->open = internal_dev_open;
 187         netdev->stop = internal_dev_stop;
 188         netdev->set_mac_address = internal_dev_mac_addr;
 189         netdev->change_mtu = internal_dev_change_mtu;
 190 #endif
 191
 192         netdev->destructor = internal_dev_free;
 193         SET_ETHTOOL_OPS(netdev, &internal_dev_ethtool_ops);
 194         netdev->tx_queue_len = 0;
 195
 196         netdev->flags = IFF_BROADCAST | IFF_MULTICAST;
 197         netdev->features = NETIF_F_LLTX | NETIF_F_SG | NETIF_F_HIGHDMA
 198                                 | NETIF_F_HW_CSUM | NETIF_F_TSO;
 199
 200         vport_gen_rand_ether_addr(netdev->dev_addr);
 201 }
 202
 203 static struct vport *internal_dev_create(const char *name,
 204                                          const void __user *config)
 205 {
 206         struct vport *vport;
 207         struct netdev_vport *netdev_vport;
 208         struct internal_dev *internal_dev;
 209         int err;
 210
 211         vport = vport_alloc(sizeof(struct netdev_vport), &internal_vport_ops);
 212         if (IS_ERR(vport)) {
 213                 err = PTR_ERR(vport);
 214                 goto error;
 215         }
 216
 217         netdev_vport = netdev_vport_priv(vport);
 218
 219         netdev_vport->dev = alloc_netdev(sizeof(struct internal_dev), name, do_setup);
 220         if (!netdev_vport->dev) {
 221                 err = -ENOMEM;
 222                 goto error_free_vport;
 223         }
 224
 225         internal_dev = internal_dev_priv(netdev_vport->dev);
 226         rcu_assign_pointer(internal_dev->vport, vport);
 227
 228         err = register_netdevice(netdev_vport->dev);
 229         if (err)
 230                 goto error_free_netdev;
 231
 232         return vport;
 233
 234 error_free_netdev:
 235         free_netdev(netdev_vport->dev);
 236 error_free_vport:
 237         vport_free(vport);
 238 error:
 239         return ERR_PTR(err);
 240 }
 241
 242 static int internal_dev_destroy(struct vport *vport)
 243 {
 244         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 245
 246         unregister_netdevice(netdev_vport->dev);
 247         vport_free(vport);
 248
 249         return 0;
 250 }
 251
 252 static int internal_dev_attach(struct vport *vport)
 253 {
 254         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 255         struct internal_dev *internal_dev = internal_dev_priv(netdev_vport->dev);
 256
 257         rcu_assign_pointer(internal_dev->attached_vport, internal_dev->vport);
 258         dev_set_promiscuity(netdev_vport->dev, 1);
 259         netif_start_queue(netdev_vport->dev);
 260
 261         return 0;
 262 }
 263
 264 static int internal_dev_detach(struct vport *vport)
 265 {
 266         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 267         struct internal_dev *internal_dev = internal_dev_priv(netdev_vport->dev);
 268
 269         netif_stop_queue(netdev_vport->dev);
 270         dev_set_promiscuity(netdev_vport->dev, -1);
 271         rcu_assign_pointer(internal_dev->attached_vport, NULL);
 272
 273         return 0;
 274 }
 275
 276 static int internal_dev_recv(struct vport *vport, struct sk_buff *skb)
 277 {
 278         struct net_device *netdev = netdev_vport_priv(vport)->dev;
 279         int len;
 280
 281         skb->dev = netdev;
 282         len = skb->len;
 283         skb->pkt_type = PACKET_HOST;
 284         skb->protocol = eth_type_trans(skb, netdev);
 285
 286         if (in_interrupt())
 287                 netif_rx(skb);
 288         else
 289                 netif_rx_ni(skb);
 290         netdev->last_rx = jiffies;
 291
 292         return len;
 293 }
 294
 295 struct vport_ops internal_vport_ops = {
 296         .type           = "internal",
 297         .flags          = VPORT_F_REQUIRED | VPORT_F_GEN_STATS | VPORT_F_FLOW,
 298         .create         = internal_dev_create,
 299         .destroy        = internal_dev_destroy,
 300         .attach         = internal_dev_attach,
 301         .detach         = internal_dev_detach,
 302         .set_mtu        = netdev_set_mtu,
 303         .set_addr       = netdev_set_addr,
 304         .get_name       = netdev_get_name,
 305         .get_addr       = netdev_get_addr,
 306         .get_kobj       = netdev_get_kobj,
 307         .get_dev_flags  = netdev_get_dev_flags,
 308         .is_running     = netdev_is_running,
 309         .get_operstate  = netdev_get_operstate,
 310         .get_ifindex    = netdev_get_ifindex,
 311         .get_iflink     = netdev_get_iflink,
 312         .get_mtu        = netdev_get_mtu,
 313         .send           = internal_dev_recv,
 314 };
 315
 316 int is_internal_dev(const struct net_device *netdev)
 317 {
 318 #ifdef HAVE_NET_DEVICE_OPS
 319         return netdev->netdev_ops == &internal_dev_netdev_ops;
 320 #else
 321         return netdev->open == internal_dev_open;
 322 #endif
 323 }
 324
 325 int is_internal_vport(const struct vport *vport)
 326 {
 327         return vport->ops == &internal_vport_ops;
 328 }
 329
 330 struct vport *internal_dev_get_vport(struct net_device *netdev)
 331 {
 332         struct internal_dev *internal_dev;
 333
 334         if (!is_internal_dev(netdev))
 335                 return NULL;
 336
 337         internal_dev = internal_dev_priv(netdev);
 338         return rcu_dereference(internal_dev->attached_vport);
 339 }