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 = netdev_get_vport(skb->dev);
  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 rtnl_link_stats64 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         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 154                                          vport);
 155         if (err)
 156                 return err;
 157
 158         dev_set_promiscuity(netdev_vport->dev, 1);
 159         dev_disable_lro(netdev_vport->dev);
 160         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 161
 162         return 0;
 163 }
 164
 165 static int netdev_detach(struct vport *vport)
 166 {
 167         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 168
 169         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 170         netdev_rx_handler_unregister(netdev_vport->dev);
 171         dev_set_promiscuity(netdev_vport->dev, -1);
 172
 173         return 0;
 174 }
 175
 176 int netdev_set_mtu(struct vport *vport, int mtu)
 177 {
 178         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 179         return dev_set_mtu(netdev_vport->dev, mtu);
 180 }
 181
 182 int netdev_set_addr(struct vport *vport, const unsigned char *addr)
 183 {
 184         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 185         struct sockaddr sa;
 186
 187         sa.sa_family = ARPHRD_ETHER;
 188         memcpy(sa.sa_data, addr, ETH_ALEN);
 189
 190         return dev_set_mac_address(netdev_vport->dev, &sa);
 191 }
 192
 193 const char *netdev_get_name(const struct vport *vport)
 194 {
 195         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 196         return netdev_vport->dev->name;
 197 }
 198
 199 const unsigned char *netdev_get_addr(const struct vport *vport)
 200 {
 201         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 202         return netdev_vport->dev->dev_addr;
 203 }
 204
 205 struct kobject *netdev_get_kobj(const struct vport *vport)
 206 {
 207         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 208         return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 209 }
 210
 211 int netdev_get_stats(const struct vport *vport, struct rtnl_link_stats64 *stats)
 212 {
 213         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 214         dev_get_stats(netdev_vport->dev, stats);
 215         return 0;
 216 }
 217
 218 unsigned netdev_get_dev_flags(const struct vport *vport)
 219 {
 220         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 221         return dev_get_flags(netdev_vport->dev);
 222 }
 223
 224 int netdev_is_running(const struct vport *vport)
 225 {
 226         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 227         return netif_running(netdev_vport->dev);
 228 }
 229
 230 unsigned char netdev_get_operstate(const struct vport *vport)
 231 {
 232         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 233         return netdev_vport->dev->operstate;
 234 }
 235
 236 int netdev_get_ifindex(const struct vport *vport)
 237 {
 238         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 239         return netdev_vport->dev->ifindex;
 240 }
 241
 242 int netdev_get_iflink(const struct vport *vport)
 243 {
 244         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 245         return netdev_vport->dev->iflink;
 246 }
 247
 248 int netdev_get_mtu(const struct vport *vport)
 249 {
 250         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 251         return netdev_vport->dev->mtu;
 252 }
 253
 254 /* Must be called with rcu_read_lock. */
 255 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 256 {
 257         /* Make our own copy of the packet.  Otherwise we will mangle the
 258          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 259          * (No one comes after us, since we tell handle_bridge() that we took
 260          * the packet.) */
 261         skb = skb_share_check(skb, GFP_ATOMIC);
 262         if (unlikely(!skb))
 263                 return;
 264
 265         skb_warn_if_lro(skb);
 266
 267         skb_push(skb, ETH_HLEN);
 268         compute_ip_summed(skb, false);
 269
 270         vport_receive(vport, skb);
 271 }
 272
 273 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 274 {
 275         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 276         int len = skb->len;
 277
 278         skb->dev = netdev_vport->dev;
 279         forward_ip_summed(skb);
 280         dev_queue_xmit(skb);
 281
 282         return len;
 283 }
 284
 285 /* Returns null if this device is not attached to a datapath. */
 286 struct vport *netdev_get_vport(struct net_device *dev)
 287 {
 288 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
 289         /* XXX: The bridge code may have registered the data.
 290          * So check that the handler pointer is the datapath's.
 291          * Once the merge is done and IFF_OVS_DATAPATH stops
 292          * being the same value as IFF_BRIDGE_PORT the check can
 293          * simply be netdev_vport->dev->priv_flags & IFF_OVS_DATAPATH. */
 294         if (rcu_dereference(dev->rx_handler) != netdev_frame_hook)
 295                 return NULL;
 296         return (struct vport *)rcu_dereference(dev->rx_handler_data);
 297 #else
 298         return (struct vport *)rcu_dereference(dev->br_port);
 299 #endif
 300 }
 301
 302 struct vport_ops netdev_vport_ops = {
 303         .type           = "netdev",
 304         .flags          = (VPORT_F_REQUIRED |
 305                           (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)),
 306         .init           = netdev_init,
 307         .exit           = netdev_exit,
 308         .create         = netdev_create,
 309         .destroy        = netdev_destroy,
 310         .attach         = netdev_attach,
 311         .detach         = netdev_detach,
 312         .set_mtu        = netdev_set_mtu,
 313         .set_addr       = netdev_set_addr,
 314         .get_name       = netdev_get_name,
 315         .get_addr       = netdev_get_addr,
 316         .get_kobj       = netdev_get_kobj,
 317         .get_stats      = netdev_get_stats,
 318         .get_dev_flags  = netdev_get_dev_flags,
 319         .is_running     = netdev_is_running,
 320         .get_operstate  = netdev_get_operstate,
 321         .get_ifindex    = netdev_get_ifindex,
 322         .get_iflink     = netdev_get_iflink,
 323         .get_mtu        = netdev_get_mtu,
 324         .send           = netdev_send,
 325 };
 326
 327 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36)
 328 /*
 329  * In kernels earlier than 2.6.36, Open vSwitch cannot safely coexist with
 330  * the Linux bridge module on any released version of Linux, because there
 331  * is only a single bridge hook function and only a single br_port member
 332  * in struct net_device.
 333  *
 334  * Declaring and exporting this symbol enforces mutual exclusion.  The bridge
 335  * module also exports the same symbol, so the module loader will refuse to
 336  * load both modules at the same time (e.g. "bridge: exports duplicate symbol
 337  * br_should_route_hook (owned by openvswitch_mod)").
 338  *
 339  * The use of "typeof" here avoids the need to track changes in the type of
 340  * br_should_route_hook over various kernel versions.
 341  */
 342 typeof(br_should_route_hook) br_should_route_hook;
 343 EXPORT_SYMBOL(br_should_route_hook);
 344 #endif