datapath/vport-netdev.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/if_arp.h>
  22 #include <linux/if_bridge.h>
  23 #include <linux/if_vlan.h>
  24 #include <linux/kernel.h>
  25 #include <linux/llc.h>
  26 #include <linux/rtnetlink.h>
  27 #include <linux/skbuff.h>
  28 #include <linux/openvswitch.h>
  29
  30 #include <net/llc.h>
  31
  32 #include "datapath.h"
  33 #include "vlan.h"
  34 #include "vport-internal_dev.h"
  35 #include "vport-netdev.h"
  36
  37 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  38
  39 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
  40 /* Called with rcu_read_lock and bottom-halves disabled. */
  41 static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
  42 {
  43         struct sk_buff *skb = *pskb;
  44         struct vport *vport;
  45
  46         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  47                 return RX_HANDLER_PASS;
  48
  49         vport = ovs_netdev_get_vport(skb->dev);
  50
  51         netdev_port_receive(vport, skb);
  52
  53         return RX_HANDLER_CONSUMED;
  54 }
  55 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
  56       defined HAVE_RHEL_OVS_HOOK
  57 /* Called with rcu_read_lock and bottom-halves disabled. */
  58 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  59 {
  60         struct vport *vport;
  61
  62         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  63                 return skb;
  64
  65         vport = ovs_netdev_get_vport(skb->dev);
  66
  67         netdev_port_receive(vport, skb);
  68
  69         return NULL;
  70 }
  71 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,32)
  72 /*
  73  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  74  * different set of devices!)
  75  */
  76 /* Called with rcu_read_lock and bottom-halves disabled. */
  77 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  78                                          struct sk_buff *skb)
  79 {
  80         netdev_port_receive((struct vport *)p, skb);
  81         return NULL;
  82 }
  83 #else
  84 #error
  85 #endif
  86
  87 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
  88     defined HAVE_RHEL_OVS_HOOK
  89 static int netdev_init(void) { return 0; }
  90 static void netdev_exit(void) { }
  91 #else
  92 static int port_count;
  93
  94 static void netdev_init(void)
  95 {
  96         port_count++;
  97         if (port_count > 1)
  98                 return;
  99
 100         /* Hook into callback used by the bridge to intercept packets.
 101          * Parasites we are. */
 102         br_handle_frame_hook = netdev_frame_hook;
 103
 104         return;
 105 }
 106
 107 static void netdev_exit(void)
 108 {
 109         port_count--;
 110         if (port_count > 0)
 111                 return;
 112
 113         br_handle_frame_hook = NULL;
 114 }
 115 #endif
 116
 117 static struct net_device *get_dpdev(struct datapath *dp)
 118 {
 119         struct vport *local;
 120
 121         local = ovs_vport_ovsl(dp, OVSP_LOCAL);
 122         BUG_ON(!local);
 123         return netdev_vport_priv(local)->dev;
 124 }
 125
 126 static struct vport *netdev_create(const struct vport_parms *parms)
 127 {
 128         struct vport *vport;
 129         struct netdev_vport *netdev_vport;
 130         int err;
 131
 132         vport = ovs_vport_alloc(sizeof(struct netdev_vport),
 133                                 &ovs_netdev_vport_ops, parms);
 134         if (IS_ERR(vport)) {
 135                 err = PTR_ERR(vport);
 136                 goto error;
 137         }
 138
 139         netdev_vport = netdev_vport_priv(vport);
 140
 141         netdev_vport->dev = dev_get_by_name(ovs_dp_get_net(vport->dp), parms->name);
 142         if (!netdev_vport->dev) {
 143                 err = -ENODEV;
 144                 goto error_free_vport;
 145         }
 146
 147         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 148             netdev_vport->dev->type != ARPHRD_ETHER ||
 149             ovs_is_internal_dev(netdev_vport->dev)) {
 150                 err = -EINVAL;
 151                 goto error_put;
 152         }
 153
 154         rtnl_lock();
 155         err = netdev_master_upper_dev_link(netdev_vport->dev,
 156                                            get_dpdev(vport->dp));
 157         if (err)
 158                 goto error_unlock;
 159
 160         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 161                                          vport);
 162         if (err)
 163                 goto error_master_upper_dev_unlink;
 164
 165         dev_set_promiscuity(netdev_vport->dev, 1);
 166         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 167         rtnl_unlock();
 168
 169         netdev_init();
 170         return vport;
 171
 172 error_master_upper_dev_unlink:
 173         netdev_upper_dev_unlink(netdev_vport->dev, get_dpdev(vport->dp));
 174 error_unlock:
 175         rtnl_unlock();
 176 error_put:
 177         dev_put(netdev_vport->dev);
 178 error_free_vport:
 179         ovs_vport_free(vport);
 180 error:
 181         return ERR_PTR(err);
 182 }
 183
 184 static void free_port_rcu(struct rcu_head *rcu)
 185 {
 186         struct netdev_vport *netdev_vport = container_of(rcu,
 187                                         struct netdev_vport, rcu);
 188
 189         dev_put(netdev_vport->dev);
 190         ovs_vport_free(vport_from_priv(netdev_vport));
 191 }
 192
 193 static void netdev_destroy(struct vport *vport)
 194 {
 195         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 196
 197         netdev_exit();
 198         rtnl_lock();
 199         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 200         netdev_rx_handler_unregister(netdev_vport->dev);
 201         netdev_upper_dev_unlink(netdev_vport->dev, get_dpdev(vport->dp));
 202         dev_set_promiscuity(netdev_vport->dev, -1);
 203         rtnl_unlock();
 204
 205         call_rcu(&netdev_vport->rcu, free_port_rcu);
 206 }
 207
 208 const char *ovs_netdev_get_name(const struct vport *vport)
 209 {
 210         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 211         return netdev_vport->dev->name;
 212 }
 213
 214 /* Must be called with rcu_read_lock. */
 215 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 216 {
 217         if (unlikely(!vport))
 218                 goto error;
 219
 220         if (unlikely(skb_warn_if_lro(skb)))
 221                 goto error;
 222
 223         /* Make our own copy of the packet.  Otherwise we will mangle the
 224          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 225          * (No one comes after us, since we tell handle_bridge() that we took
 226          * the packet.) */
 227         skb = skb_share_check(skb, GFP_ATOMIC);
 228         if (unlikely(!skb))
 229                 return;
 230
 231         skb_push(skb, ETH_HLEN);
 232         ovs_skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
 233
 234         ovs_vport_receive(vport, skb, NULL);
 235         return;
 236
 237 error:
 238         kfree_skb(skb);
 239 }
 240
 241 static unsigned int packet_length(const struct sk_buff *skb)
 242 {
 243         unsigned int length = skb->len - ETH_HLEN;
 244
 245         if (skb->protocol == htons(ETH_P_8021Q))
 246                 length -= VLAN_HLEN;
 247
 248         return length;
 249 }
 250
 251 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 252 {
 253         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 254         int mtu = netdev_vport->dev->mtu;
 255         int len;
 256
 257         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
 258                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
 259                                      netdev_vport->dev->name,
 260                                      packet_length(skb), mtu);
 261                 goto drop;
 262         }
 263
 264         skb->dev = netdev_vport->dev;
 265         len = skb->len;
 266         dev_queue_xmit(skb);
 267
 268         return len;
 269
 270 drop:
 271         kfree_skb(skb);
 272         return 0;
 273 }
 274
 275 /* Returns null if this device is not attached to a datapath. */
 276 struct vport *ovs_netdev_get_vport(struct net_device *dev)
 277 {
 278 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
 279     defined HAVE_RHEL_OVS_HOOK
 280 #if IFF_OVS_DATAPATH != 0
 281         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 282 #else
 283         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 284 #endif
 285 #ifdef HAVE_RHEL_OVS_HOOK
 286                 return (struct vport *)rcu_dereference_rtnl(dev->ax25_ptr);
 287 #else
 288                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 289 #endif
 290         else
 291                 return NULL;
 292 #else
 293         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
 294 #endif
 295 }
 296
 297 const struct vport_ops ovs_netdev_vport_ops = {
 298         .type           = OVS_VPORT_TYPE_NETDEV,
 299         .create         = netdev_create,
 300         .destroy        = netdev_destroy,
 301         .get_name       = ovs_netdev_get_name,
 302         .send           = netdev_send,
 303 };
 304
 305 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36) && \
 306     !defined HAVE_RHEL_OVS_HOOK
 307 /*
 308  * Enforces, mutual exclusion with the Linux bridge module, by declaring and
 309  * exporting br_should_route_hook.  Because the bridge module also exports the
 310  * same symbol, the module loader will refuse to load both modules at the same
 311  * time (e.g. "bridge: exports duplicate symbol br_should_route_hook (owned by
 312  * openvswitch)").
 313  *
 314  * Before Linux 2.6.36, Open vSwitch cannot safely coexist with the Linux
 315  * bridge module, so openvswitch uses this macro in those versions.  In
 316  * Linux 2.6.36 and later, Open vSwitch can coexist with the bridge module.
 317  *
 318  * The use of "typeof" here avoids the need to track changes in the type of
 319  * br_should_route_hook over various kernel versions.
 320  */
 321 typeof(br_should_route_hook) br_should_route_hook;
 322 EXPORT_SYMBOL(br_should_route_hook);
 323 #endif