datapath/vport-netdev.c

   1 /*
   2  * Copyright (c) 2010, 2011 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 "vlan.h"
  22 #include "vport-internal_dev.h"
  23 #include "vport-netdev.h"
  24
  25 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) && \
  26     !defined(HAVE_VLAN_BUG_WORKAROUND)
  27 #include <linux/module.h>
  28
  29 static int vlan_tso __read_mostly = 0;
  30 module_param(vlan_tso, int, 0644);
  31 MODULE_PARM_DESC(vlan_tso, "Enable TSO for VLAN packets");
  32 #else
  33 #define vlan_tso true
  34 #endif
  35
  36 /* If the native device stats aren't 64 bit use the vport stats tracking instead. */
  37 #define USE_VPORT_STATS (sizeof(((struct net_device_stats *)0)->rx_bytes) < sizeof(u64))
  38
  39 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  40
  41 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  42 /* Called with rcu_read_lock and bottom-halves disabled. */
  43 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  44 {
  45         struct vport *vport;
  46
  47         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  48                 return skb;
  49
  50         vport = netdev_get_vport(skb->dev);
  51
  52         netdev_port_receive(vport, skb);
  53
  54         return NULL;
  55 }
  56 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
  57 /*
  58  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  59  * different set of devices!)
  60  */
  61 /* Called with rcu_read_lock and bottom-halves disabled. */
  62 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  63                                          struct sk_buff *skb)
  64 {
  65         netdev_port_receive((struct vport *)p, skb);
  66         return NULL;
  67 }
  68 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  69 /*
  70  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  71  * different set of devices!)
  72  */
  73 /* Called with rcu_read_lock and bottom-halves disabled. */
  74 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
  75 {
  76         netdev_port_receive((struct vport *)p, *pskb);
  77         return 1;
  78 }
  79 #else
  80 #error
  81 #endif
  82
  83 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  84 static int netdev_init(void) { return 0; }
  85 static void netdev_exit(void) { }
  86 #else
  87 static int netdev_init(void)
  88 {
  89         /* Hook into callback used by the bridge to intercept packets.
  90          * Parasites we are. */
  91         br_handle_frame_hook = netdev_frame_hook;
  92
  93         return 0;
  94 }
  95
  96 static void netdev_exit(void)
  97 {
  98         br_handle_frame_hook = NULL;
  99 }
 100 #endif
 101
 102 static struct vport *netdev_create(const struct vport_parms *parms)
 103 {
 104         struct vport *vport;
 105         struct netdev_vport *netdev_vport;
 106         int err;
 107
 108         vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops, parms);
 109         if (IS_ERR(vport)) {
 110                 err = PTR_ERR(vport);
 111                 goto error;
 112         }
 113
 114         netdev_vport = netdev_vport_priv(vport);
 115
 116         netdev_vport->dev = dev_get_by_name(&init_net, parms->name);
 117         if (!netdev_vport->dev) {
 118                 err = -ENODEV;
 119                 goto error_free_vport;
 120         }
 121
 122         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 123             netdev_vport->dev->type != ARPHRD_ETHER ||
 124             is_internal_dev(netdev_vport->dev)) {
 125                 err = -EINVAL;
 126                 goto error_put;
 127         }
 128
 129         /* If we are using the vport stats layer initialize it to the current
 130          * values so we are roughly consistent with the device stats. */
 131         if (USE_VPORT_STATS) {
 132                 struct rtnl_link_stats64 stats;
 133
 134                 err = netdev_get_stats(vport, &stats);
 135                 if (!err)
 136                         vport_set_stats(vport, &stats);
 137         }
 138
 139         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 140                                          vport);
 141         if (err)
 142                 goto error_put;
 143
 144         dev_set_promiscuity(netdev_vport->dev, 1);
 145         dev_disable_lro(netdev_vport->dev);
 146         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 147
 148         return vport;
 149
 150 error_put:
 151         dev_put(netdev_vport->dev);
 152 error_free_vport:
 153         vport_free(vport);
 154 error:
 155         return ERR_PTR(err);
 156 }
 157
 158 static int netdev_destroy(struct vport *vport)
 159 {
 160         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 161
 162         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 163         netdev_rx_handler_unregister(netdev_vport->dev);
 164         dev_set_promiscuity(netdev_vport->dev, -1);
 165
 166         synchronize_rcu();
 167
 168         dev_put(netdev_vport->dev);
 169         vport_free(vport);
 170
 171         return 0;
 172 }
 173
 174 int netdev_set_mtu(struct vport *vport, int mtu)
 175 {
 176         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 177         return dev_set_mtu(netdev_vport->dev, mtu);
 178 }
 179
 180 int netdev_set_addr(struct vport *vport, const unsigned char *addr)
 181 {
 182         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 183         struct sockaddr sa;
 184
 185         sa.sa_family = ARPHRD_ETHER;
 186         memcpy(sa.sa_data, addr, ETH_ALEN);
 187
 188         return dev_set_mac_address(netdev_vport->dev, &sa);
 189 }
 190
 191 const char *netdev_get_name(const struct vport *vport)
 192 {
 193         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 194         return netdev_vport->dev->name;
 195 }
 196
 197 const unsigned char *netdev_get_addr(const struct vport *vport)
 198 {
 199         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 200         return netdev_vport->dev->dev_addr;
 201 }
 202
 203 struct kobject *netdev_get_kobj(const struct vport *vport)
 204 {
 205         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 206         return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 207 }
 208
 209 int netdev_get_stats(const struct vport *vport, struct rtnl_link_stats64 *stats)
 210 {
 211         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 212         dev_get_stats(netdev_vport->dev, stats);
 213         return 0;
 214 }
 215
 216 unsigned netdev_get_dev_flags(const struct vport *vport)
 217 {
 218         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 219         return dev_get_flags(netdev_vport->dev);
 220 }
 221
 222 int netdev_is_running(const struct vport *vport)
 223 {
 224         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 225         return netif_running(netdev_vport->dev);
 226 }
 227
 228 unsigned char netdev_get_operstate(const struct vport *vport)
 229 {
 230         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 231         return netdev_vport->dev->operstate;
 232 }
 233
 234 int netdev_get_ifindex(const struct vport *vport)
 235 {
 236         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 237         return netdev_vport->dev->ifindex;
 238 }
 239
 240 int netdev_get_iflink(const struct vport *vport)
 241 {
 242         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 243         return netdev_vport->dev->iflink;
 244 }
 245
 246 int netdev_get_mtu(const struct vport *vport)
 247 {
 248         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 249         return netdev_vport->dev->mtu;
 250 }
 251
 252 /* Must be called with rcu_read_lock. */
 253 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 254 {
 255         /* Make our own copy of the packet.  Otherwise we will mangle the
 256          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 257          * (No one comes after us, since we tell handle_bridge() that we took
 258          * the packet.) */
 259         skb = skb_share_check(skb, GFP_ATOMIC);
 260         if (unlikely(!skb))
 261                 return;
 262
 263         skb_warn_if_lro(skb);
 264
 265         skb_push(skb, ETH_HLEN);
 266         compute_ip_summed(skb, false);
 267         vlan_copy_skb_tci(skb);
 268
 269         vport_receive(vport, skb);
 270 }
 271
 272 static bool dev_supports_vlan_tx(struct net_device *dev)
 273 {
 274 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
 275         /* Software fallback means every device supports vlan_tci on TX. */
 276         return true;
 277 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
 278         return dev->features & NETIF_F_HW_VLAN_TX;
 279 #else
 280         /* Assume that the driver is buggy. */
 281         return false;
 282 #endif
 283 }
 284
 285 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 286 {
 287         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 288         int len;
 289
 290         skb->dev = netdev_vport->dev;
 291         forward_ip_summed(skb);
 292
 293         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
 294                 int err;
 295                 int features = 0;
 296
 297 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,26)
 298                 features = skb->dev->features & skb->dev->vlan_features;
 299 #endif
 300
 301                 err = vswitch_skb_checksum_setup(skb);
 302                 if (unlikely(err)) {
 303                         kfree_skb(skb);
 304                         return 0;
 305                 }
 306
 307                 if (!vlan_tso)
 308                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
 309                                       NETIF_F_UFO | NETIF_F_FSO);
 310
 311                 if (skb_is_gso(skb) &&
 312                     (!skb_gso_ok(skb, features) ||
 313                      unlikely(skb->ip_summed != CHECKSUM_PARTIAL))) {
 314                         struct sk_buff *nskb;
 315
 316                         nskb = skb_gso_segment(skb, features);
 317                         if (!nskb) {
 318                                 if (unlikely(skb_cloned(skb) &&
 319                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) {
 320                                         kfree_skb(skb);
 321                                         return 0;
 322                                 }
 323
 324                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
 325                                 goto tag;
 326                         }
 327
 328                         kfree_skb(skb);
 329                         skb = nskb;
 330                         if (IS_ERR(skb))
 331                                 return 0;
 332
 333                         len = 0;
 334                         do {
 335                                 nskb = skb->next;
 336                                 skb->next = NULL;
 337
 338                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 339                                 if (likely(skb)) {
 340                                         len += skb->len;
 341                                         vlan_set_tci(skb, 0);
 342                                         dev_queue_xmit(skb);
 343                                 }
 344
 345                                 skb = nskb;
 346                         } while (skb);
 347
 348                         return len;
 349                 }
 350
 351 tag:
 352                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 353                 if (unlikely(!skb))
 354                         return 0;
 355                 vlan_set_tci(skb, 0);
 356         }
 357
 358         len = skb->len;
 359         dev_queue_xmit(skb);
 360
 361         return len;
 362 }
 363
 364 /* Returns null if this device is not attached to a datapath. */
 365 struct vport *netdev_get_vport(struct net_device *dev)
 366 {
 367 #ifdef IFF_BRIDGE_PORT
 368 #if IFF_BRIDGE_PORT != IFF_OVS_DATAPATH
 369         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 370 #else
 371         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 372 #endif
 373                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 374         else
 375                 return NULL;
 376 #else
 377         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
 378 #endif
 379 }
 380
 381 const struct vport_ops netdev_vport_ops = {
 382         .type           = ODP_VPORT_TYPE_NETDEV,
 383         .flags          = (VPORT_F_REQUIRED |
 384                           (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)),
 385         .init           = netdev_init,
 386         .exit           = netdev_exit,
 387         .create         = netdev_create,
 388         .destroy        = netdev_destroy,
 389         .set_mtu        = netdev_set_mtu,
 390         .set_addr       = netdev_set_addr,
 391         .get_name       = netdev_get_name,
 392         .get_addr       = netdev_get_addr,
 393         .get_kobj       = netdev_get_kobj,
 394         .get_stats      = netdev_get_stats,
 395         .get_dev_flags  = netdev_get_dev_flags,
 396         .is_running     = netdev_is_running,
 397         .get_operstate  = netdev_get_operstate,
 398         .get_ifindex    = netdev_get_ifindex,
 399         .get_iflink     = netdev_get_iflink,
 400         .get_mtu        = netdev_get_mtu,
 401         .send           = netdev_send,
 402 };
 403
 404 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36)
 405 /*
 406  * In kernels earlier than 2.6.36, Open vSwitch cannot safely coexist with
 407  * the Linux bridge module on any released version of Linux, because there
 408  * is only a single bridge hook function and only a single br_port member
 409  * in struct net_device.
 410  *
 411  * Declaring and exporting this symbol enforces mutual exclusion.  The bridge
 412  * module also exports the same symbol, so the module loader will refuse to
 413  * load both modules at the same time (e.g. "bridge: exports duplicate symbol
 414  * br_should_route_hook (owned by openvswitch_mod)").
 415  *
 416  * The use of "typeof" here avoids the need to track changes in the type of
 417  * br_should_route_hook over various kernel versions.
 418  */
 419 typeof(br_should_route_hook) br_should_route_hook;
 420 EXPORT_SYMBOL(br_should_route_hook);
 421 #endif