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 "checksum.h"
  20 #include "datapath.h"
  21 #include "vport-internal_dev.h"
  22 #include "vport-netdev.h"
  23
  24 #include "compat.h"
  25
  26 /* If the native device stats aren't 64 bit use the vport stats tracking instead. */
  27 #define USE_VPORT_STATS (sizeof(((struct net_device_stats *)0)->rx_bytes) < sizeof(u64))
  28
  29 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  30
  31 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  32 /* Called with rcu_read_lock and bottom-halves disabled. */
  33 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  34 {
  35         struct vport *vport;
  36
  37         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  38                 return skb;
  39
  40         vport = netdev_get_vport(skb->dev);
  41
  42         netdev_port_receive(vport, skb);
  43
  44         return NULL;
  45 }
  46 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
  47 /*
  48  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  49  * different set of devices!)
  50  */
  51 /* Called with rcu_read_lock and bottom-halves disabled. */
  52 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  53                                          struct sk_buff *skb)
  54 {
  55         netdev_port_receive((struct vport *)p, skb);
  56         return NULL;
  57 }
  58 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  59 /*
  60  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  61  * different set of devices!)
  62  */
  63 /* Called with rcu_read_lock and bottom-halves disabled. */
  64 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
  65 {
  66         netdev_port_receive((struct vport *)p, *pskb);
  67         return 1;
  68 }
  69 #else
  70 #error
  71 #endif
  72
  73 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  74 static int netdev_init(void) { return 0; }
  75 static void netdev_exit(void) { }
  76 #else
  77 static int netdev_init(void)
  78 {
  79         /* Hook into callback used by the bridge to intercept packets.
  80          * Parasites we are. */
  81         br_handle_frame_hook = netdev_frame_hook;
  82
  83         return 0;
  84 }
  85
  86 static void netdev_exit(void)
  87 {
  88         br_handle_frame_hook = NULL;
  89 }
  90 #endif
  91
  92 static struct vport *netdev_create(const struct vport_parms *parms)
  93 {
  94         struct vport *vport;
  95         struct netdev_vport *netdev_vport;
  96         int err;
  97
  98         vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops, parms);
  99         if (IS_ERR(vport)) {
 100                 err = PTR_ERR(vport);
 101                 goto error;
 102         }
 103
 104         netdev_vport = netdev_vport_priv(vport);
 105
 106         netdev_vport->dev = dev_get_by_name(&init_net, parms->name);
 107         if (!netdev_vport->dev) {
 108                 err = -ENODEV;
 109                 goto error_free_vport;
 110         }
 111
 112         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 113             netdev_vport->dev->type != ARPHRD_ETHER ||
 114             is_internal_dev(netdev_vport->dev)) {
 115                 err = -EINVAL;
 116                 goto error_put;
 117         }
 118
 119         /* If we are using the vport stats layer initialize it to the current
 120          * values so we are roughly consistent with the device stats. */
 121         if (USE_VPORT_STATS) {
 122                 struct rtnl_link_stats64 stats;
 123
 124                 err = netdev_get_stats(vport, &stats);
 125                 if (!err)
 126                         vport_set_stats(vport, &stats);
 127         }
 128
 129         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 130                                          vport);
 131         if (err)
 132                 goto error_put;
 133
 134         dev_set_promiscuity(netdev_vport->dev, 1);
 135         dev_disable_lro(netdev_vport->dev);
 136         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 137
 138         return vport;
 139
 140 error_put:
 141         dev_put(netdev_vport->dev);
 142 error_free_vport:
 143         vport_free(vport);
 144 error:
 145         return ERR_PTR(err);
 146 }
 147
 148 static int netdev_destroy(struct vport *vport)
 149 {
 150         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 151
 152         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 153         netdev_rx_handler_unregister(netdev_vport->dev);
 154         dev_set_promiscuity(netdev_vport->dev, -1);
 155
 156         synchronize_rcu();
 157
 158         dev_put(netdev_vport->dev);
 159         vport_free(vport);
 160
 161         return 0;
 162 }
 163
 164 int netdev_set_mtu(struct vport *vport, int mtu)
 165 {
 166         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 167         return dev_set_mtu(netdev_vport->dev, mtu);
 168 }
 169
 170 int netdev_set_addr(struct vport *vport, const unsigned char *addr)
 171 {
 172         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 173         struct sockaddr sa;
 174
 175         sa.sa_family = ARPHRD_ETHER;
 176         memcpy(sa.sa_data, addr, ETH_ALEN);
 177
 178         return dev_set_mac_address(netdev_vport->dev, &sa);
 179 }
 180
 181 const char *netdev_get_name(const struct vport *vport)
 182 {
 183         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 184         return netdev_vport->dev->name;
 185 }
 186
 187 const unsigned char *netdev_get_addr(const struct vport *vport)
 188 {
 189         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 190         return netdev_vport->dev->dev_addr;
 191 }
 192
 193 struct kobject *netdev_get_kobj(const struct vport *vport)
 194 {
 195         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 196         return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 197 }
 198
 199 int netdev_get_stats(const struct vport *vport, struct rtnl_link_stats64 *stats)
 200 {
 201         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 202         dev_get_stats(netdev_vport->dev, stats);
 203         return 0;
 204 }
 205
 206 unsigned netdev_get_dev_flags(const struct vport *vport)
 207 {
 208         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 209         return dev_get_flags(netdev_vport->dev);
 210 }
 211
 212 int netdev_is_running(const struct vport *vport)
 213 {
 214         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 215         return netif_running(netdev_vport->dev);
 216 }
 217
 218 unsigned char netdev_get_operstate(const struct vport *vport)
 219 {
 220         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 221         return netdev_vport->dev->operstate;
 222 }
 223
 224 int netdev_get_ifindex(const struct vport *vport)
 225 {
 226         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 227         return netdev_vport->dev->ifindex;
 228 }
 229
 230 int netdev_get_iflink(const struct vport *vport)
 231 {
 232         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 233         return netdev_vport->dev->iflink;
 234 }
 235
 236 int netdev_get_mtu(const struct vport *vport)
 237 {
 238         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 239         return netdev_vport->dev->mtu;
 240 }
 241
 242 /* Must be called with rcu_read_lock. */
 243 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 244 {
 245         /* Make our own copy of the packet.  Otherwise we will mangle the
 246          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 247          * (No one comes after us, since we tell handle_bridge() that we took
 248          * the packet.) */
 249         skb = skb_share_check(skb, GFP_ATOMIC);
 250         if (unlikely(!skb))
 251                 return;
 252
 253         skb_warn_if_lro(skb);
 254
 255         skb_push(skb, ETH_HLEN);
 256         compute_ip_summed(skb, false);
 257
 258         vport_receive(vport, skb);
 259 }
 260
 261 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 262 {
 263         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 264         int len = skb->len;
 265
 266         skb->dev = netdev_vport->dev;
 267         forward_ip_summed(skb);
 268         dev_queue_xmit(skb);
 269
 270         return len;
 271 }
 272
 273 /* Returns null if this device is not attached to a datapath. */
 274 struct vport *netdev_get_vport(struct net_device *dev)
 275 {
 276 #ifdef IFF_BRIDGE_PORT
 277 #if IFF_BRIDGE_PORT != IFF_OVS_DATAPATH
 278         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 279 #else
 280         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 281 #endif
 282                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 283         else
 284                 return NULL;
 285 #else
 286         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
 287 #endif
 288 }
 289
 290 const struct vport_ops netdev_vport_ops = {
 291         .type           = "netdev",
 292         .flags          = (VPORT_F_REQUIRED |
 293                           (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)),
 294         .init           = netdev_init,
 295         .exit           = netdev_exit,
 296         .create         = netdev_create,
 297         .destroy        = netdev_destroy,
 298         .set_mtu        = netdev_set_mtu,
 299         .set_addr       = netdev_set_addr,
 300         .get_name       = netdev_get_name,
 301         .get_addr       = netdev_get_addr,
 302         .get_kobj       = netdev_get_kobj,
 303         .get_stats      = netdev_get_stats,
 304         .get_dev_flags  = netdev_get_dev_flags,
 305         .is_running     = netdev_is_running,
 306         .get_operstate  = netdev_get_operstate,
 307         .get_ifindex    = netdev_get_ifindex,
 308         .get_iflink     = netdev_get_iflink,
 309         .get_mtu        = netdev_get_mtu,
 310         .send           = netdev_send,
 311 };
 312
 313 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36)
 314 /*
 315  * In kernels earlier than 2.6.36, Open vSwitch cannot safely coexist with
 316  * the Linux bridge module on any released version of Linux, because there
 317  * is only a single bridge hook function and only a single br_port member
 318  * in struct net_device.
 319  *
 320  * Declaring and exporting this symbol enforces mutual exclusion.  The bridge
 321  * module also exports the same symbol, so the module loader will refuse to
 322  * load both modules at the same time (e.g. "bridge: exports duplicate symbol
 323  * br_should_route_hook (owned by openvswitch_mod)").
 324  *
 325  * The use of "typeof" here avoids the need to track changes in the type of
 326  * br_should_route_hook over various kernel versions.
 327  */
 328 typeof(br_should_route_hook) br_should_route_hook;
 329 EXPORT_SYMBOL(br_should_route_hook);
 330 #endif