datapath/vport-netdev.c

   1 /*
   2  * Copyright (c) 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/if_arp.h>
  10 #include <linux/if_bridge.h>
  11 #include <linux/if_vlan.h>
  12 #include <linux/kernel.h>
  13 #include <linux/llc.h>
  14 #include <linux/rtnetlink.h>
  15 #include <linux/skbuff.h>
  16
  17 #include <net/llc.h>
  18
  19 #include "datapath.h"
  20 #include "vport-internal_dev.h"
  21 #include "vport-netdev.h"
  22
  23 #include "compat.h"
  24
  25 /* If the native device stats aren't 64 bit use the vport stats tracking instead. */
  26 #define USE_VPORT_STATS (sizeof(((struct net_device_stats *)0)->rx_bytes) < sizeof(u64))
  27
  28 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  29
  30 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  31 /* Called with rcu_read_lock and bottom-halves disabled. */
  32 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  33 {
  34         struct vport *vport;
  35
  36         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  37                 return skb;
  38
  39         vport = (struct vport *)rcu_dereference(skb->dev->br_port);
  40
  41         netdev_port_receive(vport, skb);
  42
  43         return NULL;
  44 }
  45 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
  46 /*
  47  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  48  * different set of devices!)
  49  */
  50 /* Called with rcu_read_lock and bottom-halves disabled. */
  51 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  52                                          struct sk_buff *skb)
  53 {
  54         netdev_port_receive((struct vport *)p, skb);
  55         return NULL;
  56 }
  57 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  58 /*
  59  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  60  * different set of devices!)
  61  */
  62 /* Called with rcu_read_lock and bottom-halves disabled. */
  63 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
  64 {
  65         netdev_port_receive((struct vport *)p, *pskb);
  66         return 1;
  67 }
  68 #else
  69 #error
  70 #endif
  71
  72 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  73 static int netdev_init(void) { return 0; }
  74 static void netdev_exit(void) { }
  75 #else
  76 static int netdev_init(void)
  77 {
  78         /* Hook into callback used by the bridge to intercept packets.
  79          * Parasites we are. */
  80         br_handle_frame_hook = netdev_frame_hook;
  81
  82         return 0;
  83 }
  84
  85 static void netdev_exit(void)
  86 {
  87         br_handle_frame_hook = NULL;
  88 }
  89 #endif
  90
  91 static struct vport *netdev_create(const char *name, const void __user *config)
  92 {
  93         struct vport *vport;
  94         struct netdev_vport *netdev_vport;
  95         int err;
  96
  97         vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops);
  98         if (IS_ERR(vport)) {
  99                 err = PTR_ERR(vport);
 100                 goto error;
 101         }
 102
 103         netdev_vport = netdev_vport_priv(vport);
 104
 105         netdev_vport->dev = dev_get_by_name(&init_net, name);
 106         if (!netdev_vport->dev) {
 107                 err = -ENODEV;
 108                 goto error_free_vport;
 109         }
 110
 111         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 112             netdev_vport->dev->type != ARPHRD_ETHER ||
 113             is_internal_dev(netdev_vport->dev)) {
 114                 err = -EINVAL;
 115                 goto error_put;
 116         }
 117
 118         /* If we are using the vport stats layer initialize it to the current
 119          * values so we are roughly consistent with the device stats. */
 120         if (USE_VPORT_STATS) {
 121                 struct odp_vport_stats stats;
 122
 123                 err = netdev_get_stats(vport, &stats);
 124                 if (!err)
 125                         vport_set_stats(vport, &stats);
 126         }
 127
 128         return vport;
 129
 130 error_put:
 131         dev_put(netdev_vport->dev);
 132 error_free_vport:
 133         vport_free(vport);
 134 error:
 135         return ERR_PTR(err);
 136 }
 137
 138 static int netdev_destroy(struct vport *vport)
 139 {
 140         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 141
 142         dev_put(netdev_vport->dev);
 143         vport_free(vport);
 144
 145         return 0;
 146 }
 147
 148 static int netdev_attach(struct vport *vport)
 149 {
 150         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 151         int err;
 152
 153         rcu_assign_pointer(netdev_vport->dev->br_port,
 154                            (struct net_bridge_port *)vport);
 155         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 156                                          NULL);
 157         if (err)
 158                 return err;
 159
 160         dev_set_promiscuity(netdev_vport->dev, 1);
 161         dev_disable_lro(netdev_vport->dev);
 162
 163         return 0;
 164 }
 165
 166 static int netdev_detach(struct vport *vport)
 167 {
 168         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 169
 170         netdev_rx_handler_unregister(netdev_vport->dev);
 171         rcu_assign_pointer(netdev_vport->dev->br_port, NULL);
 172         dev_set_promiscuity(netdev_vport->dev, -1);
 173
 174         return 0;
 175 }
 176
 177 int netdev_set_mtu(struct vport *vport, int mtu)
 178 {
 179         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 180         return dev_set_mtu(netdev_vport->dev, mtu);
 181 }
 182
 183 int netdev_set_addr(struct vport *vport, const unsigned char *addr)
 184 {
 185         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 186         struct sockaddr sa;
 187
 188         sa.sa_family = ARPHRD_ETHER;
 189         memcpy(sa.sa_data, addr, ETH_ALEN);
 190
 191         return dev_set_mac_address(netdev_vport->dev, &sa);
 192 }
 193
 194 const char *netdev_get_name(const struct vport *vport)
 195 {
 196         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 197         return netdev_vport->dev->name;
 198 }
 199
 200 const unsigned char *netdev_get_addr(const struct vport *vport)
 201 {
 202         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 203         return netdev_vport->dev->dev_addr;
 204 }
 205
 206 struct kobject *netdev_get_kobj(const struct vport *vport)
 207 {
 208         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 209         return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 210 }
 211
 212 int netdev_get_stats(const struct vport *vport, struct odp_vport_stats *stats)
 213 {
 214         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 215         const struct net_device_stats *netdev_stats;
 216
 217         netdev_stats = dev_get_stats(netdev_vport->dev);
 218
 219         stats->rx_bytes         = netdev_stats->rx_bytes;
 220         stats->rx_packets       = netdev_stats->rx_packets;
 221         stats->tx_bytes         = netdev_stats->tx_bytes;
 222         stats->tx_packets       = netdev_stats->tx_packets;
 223         stats->rx_dropped       = netdev_stats->rx_dropped;
 224         stats->rx_errors        = netdev_stats->rx_errors;
 225         stats->rx_frame_err     = netdev_stats->rx_frame_errors;
 226         stats->rx_over_err      = netdev_stats->rx_over_errors;
 227         stats->rx_crc_err       = netdev_stats->rx_crc_errors;
 228         stats->tx_dropped       = netdev_stats->tx_dropped;
 229         stats->tx_errors        = netdev_stats->tx_errors;
 230         stats->collisions       = netdev_stats->collisions;
 231
 232         return 0;
 233 }
 234
 235 unsigned netdev_get_dev_flags(const struct vport *vport)
 236 {
 237         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 238         return dev_get_flags(netdev_vport->dev);
 239 }
 240
 241 int netdev_is_running(const struct vport *vport)
 242 {
 243         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 244         return netif_running(netdev_vport->dev);
 245 }
 246
 247 unsigned char netdev_get_operstate(const struct vport *vport)
 248 {
 249         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 250         return netdev_vport->dev->operstate;
 251 }
 252
 253 int netdev_get_ifindex(const struct vport *vport)
 254 {
 255         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 256         return netdev_vport->dev->ifindex;
 257 }
 258
 259 int netdev_get_iflink(const struct vport *vport)
 260 {
 261         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 262         return netdev_vport->dev->iflink;
 263 }
 264
 265 int netdev_get_mtu(const struct vport *vport)
 266 {
 267         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 268         return netdev_vport->dev->mtu;
 269 }
 270
 271 /* Must be called with rcu_read_lock. */
 272 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 273 {
 274         /* Make our own copy of the packet.  Otherwise we will mangle the
 275          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 276          * (No one comes after us, since we tell handle_bridge() that we took
 277          * the packet.) */
 278         skb = skb_share_check(skb, GFP_ATOMIC);
 279         if (unlikely(!skb))
 280                 return;
 281
 282         skb_warn_if_lro(skb);
 283
 284         /* Push the Ethernet header back on. */
 285         skb_push(skb, ETH_HLEN);
 286         skb_reset_mac_header(skb);
 287         compute_ip_summed(skb, false);
 288
 289         vport_receive(vport, skb);
 290 }
 291
 292 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 293 {
 294         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 295         int len = skb->len;
 296
 297         skb->dev = netdev_vport->dev;
 298         forward_ip_summed(skb);
 299         dev_queue_xmit(skb);
 300
 301         return len;
 302 }
 303
 304 /* Returns null if this device is not attached to a datapath. */
 305 struct vport *netdev_get_vport(struct net_device *dev)
 306 {
 307         return (struct vport *)dev->br_port;
 308 }
 309
 310 struct vport_ops netdev_vport_ops = {
 311         .type           = "netdev",
 312         .flags          = (VPORT_F_REQUIRED |
 313                           (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)),
 314         .init           = netdev_init,
 315         .exit           = netdev_exit,
 316         .create         = netdev_create,
 317         .destroy        = netdev_destroy,
 318         .attach         = netdev_attach,
 319         .detach         = netdev_detach,
 320         .set_mtu        = netdev_set_mtu,
 321         .set_addr       = netdev_set_addr,
 322         .get_name       = netdev_get_name,
 323         .get_addr       = netdev_get_addr,
 324         .get_kobj       = netdev_get_kobj,
 325         .get_stats      = netdev_get_stats,
 326         .get_dev_flags  = netdev_get_dev_flags,
 327         .is_running     = netdev_is_running,
 328         .get_operstate  = netdev_get_operstate,
 329         .get_ifindex    = netdev_get_ifindex,
 330         .get_iflink     = netdev_get_iflink,
 331         .get_mtu        = netdev_get_mtu,
 332         .send           = netdev_send,
 333 };
 334
 335 /*
 336  * Open vSwitch cannot safely coexist with the Linux bridge module on any
 337  * released version of Linux, because there is only a single bridge hook
 338  * function and only a single br_port member in struct net_device.
 339  *
 340  * Declaring and exporting this symbol enforces mutual exclusion.  The bridge
 341  * module also exports the same symbol, so the module loader will refuse to
 342  * load both modules at the same time (e.g. "bridge: exports duplicate symbol
 343  * br_should_route_hook (owned by openvswitch_mod)").
 344  *
 345  * The use of "typeof" here avoids the need to track changes in the type of
 346  * br_should_route_hook over various kernel versions.
 347  */
 348 typeof(br_should_route_hook) br_should_route_hook;
 349 EXPORT_SYMBOL(br_should_route_hook);