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
  29 #include <net/llc.h>
  30
  31 #include "checksum.h"
  32 #include "datapath.h"
  33 #include "vlan.h"
  34 #include "vport-internal_dev.h"
  35 #include "vport-netdev.h"
  36
  37 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) && \
  38         !defined(HAVE_VLAN_BUG_WORKAROUND)
  39 #include <linux/module.h>
  40
  41 static int vlan_tso __read_mostly;
  42 module_param(vlan_tso, int, 0644);
  43 MODULE_PARM_DESC(vlan_tso, "Enable TSO for VLAN packets");
  44 #else
  45 #define vlan_tso true
  46 #endif
  47
  48 #ifdef HAVE_RHEL_OVS_HOOK
  49 static atomic_t nr_bridges = ATOMIC_INIT(0);
  50
  51 extern struct sk_buff *(*openvswitch_handle_frame_hook)(struct sk_buff *skb);
  52 #endif
  53
  54 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  55
  56 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
  57 /* Called with rcu_read_lock and bottom-halves disabled. */
  58 static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
  59 {
  60         struct sk_buff *skb = *pskb;
  61         struct vport *vport;
  62
  63         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  64                 return RX_HANDLER_PASS;
  65
  66         vport = ovs_netdev_get_vport(skb->dev);
  67
  68         netdev_port_receive(vport, skb);
  69
  70         return RX_HANDLER_CONSUMED;
  71 }
  72 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
  73       defined HAVE_RHEL_OVS_HOOK
  74 /* Called with rcu_read_lock and bottom-halves disabled. */
  75 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  76 {
  77         struct vport *vport;
  78
  79         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  80                 return skb;
  81
  82         vport = ovs_netdev_get_vport(skb->dev);
  83
  84         netdev_port_receive(vport, skb);
  85
  86         return NULL;
  87 }
  88 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
  89 /*
  90  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  91  * different set of devices!)
  92  */
  93 /* Called with rcu_read_lock and bottom-halves disabled. */
  94 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  95                                          struct sk_buff *skb)
  96 {
  97         netdev_port_receive((struct vport *)p, skb);
  98         return NULL;
  99 }
 100 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 101 /*
 102  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 103  * different set of devices!)
 104  */
 105 /* Called with rcu_read_lock and bottom-halves disabled. */
 106 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 107 {
 108         netdev_port_receive((struct vport *)p, *pskb);
 109         return 1;
 110 }
 111 #else
 112 #error
 113 #endif
 114
 115 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
 116     defined HAVE_RHEL_OVS_HOOK
 117 static int netdev_init(void) { return 0; }
 118 static void netdev_exit(void) { }
 119 #else
 120 static int port_count;
 121
 122 static void netdev_init(void)
 123 {
 124         port_count++;
 125         if (port_count > 1)
 126                 return;
 127
 128         /* Hook into callback used by the bridge to intercept packets.
 129          * Parasites we are. */
 130         br_handle_frame_hook = netdev_frame_hook;
 131
 132         return;
 133 }
 134
 135 static void netdev_exit(void)
 136 {
 137         port_count--;
 138         if (port_count > 0)
 139                 return;
 140
 141         br_handle_frame_hook = NULL;
 142 }
 143 #endif
 144
 145 static struct vport *netdev_create(const struct vport_parms *parms)
 146 {
 147         struct vport *vport;
 148         struct netdev_vport *netdev_vport;
 149         int err;
 150
 151         vport = ovs_vport_alloc(sizeof(struct netdev_vport),
 152                                 &ovs_netdev_vport_ops, parms);
 153         if (IS_ERR(vport)) {
 154                 err = PTR_ERR(vport);
 155                 goto error;
 156         }
 157
 158         netdev_vport = netdev_vport_priv(vport);
 159
 160         netdev_vport->dev = dev_get_by_name(ovs_dp_get_net(vport->dp), parms->name);
 161         if (!netdev_vport->dev) {
 162                 err = -ENODEV;
 163                 goto error_free_vport;
 164         }
 165
 166         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 167             netdev_vport->dev->type != ARPHRD_ETHER ||
 168             ovs_is_internal_dev(netdev_vport->dev)) {
 169                 err = -EINVAL;
 170                 goto error_put;
 171         }
 172
 173         rtnl_lock();
 174 #ifdef HAVE_RHEL_OVS_HOOK
 175         rcu_assign_pointer(netdev_vport->dev->ax25_ptr, vport);
 176         atomic_inc(&nr_bridges);
 177         rcu_assign_pointer(openvswitch_handle_frame_hook, netdev_frame_hook);
 178 #else
 179         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 180                                          vport);
 181         if (err)
 182                 goto error_unlock;
 183 #endif
 184
 185         dev_set_promiscuity(netdev_vport->dev, 1);
 186 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
 187         dev_disable_lro(netdev_vport->dev);
 188 #endif
 189         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 190         rtnl_unlock();
 191
 192         netdev_init();
 193         return vport;
 194
 195 #ifndef HAVE_RHEL_OVS_HOOK
 196 error_unlock:
 197 #endif
 198         rtnl_unlock();
 199 error_put:
 200         dev_put(netdev_vport->dev);
 201 error_free_vport:
 202         ovs_vport_free(vport);
 203 error:
 204         return ERR_PTR(err);
 205 }
 206
 207 static void free_port_rcu(struct rcu_head *rcu)
 208 {
 209         struct netdev_vport *netdev_vport = container_of(rcu,
 210                                         struct netdev_vport, rcu);
 211
 212 #ifdef HAVE_RHEL_OVS_HOOK
 213         rcu_assign_pointer(netdev_vport->dev->ax25_ptr, NULL);
 214
 215         if (atomic_dec_and_test(&nr_bridges))
 216                 rcu_assign_pointer(openvswitch_handle_frame_hook, NULL);
 217 #endif
 218         dev_put(netdev_vport->dev);
 219         ovs_vport_free(vport_from_priv(netdev_vport));
 220 }
 221
 222 static void netdev_destroy(struct vport *vport)
 223 {
 224         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 225
 226         netdev_exit();
 227         rtnl_lock();
 228         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 229         netdev_rx_handler_unregister(netdev_vport->dev);
 230         dev_set_promiscuity(netdev_vport->dev, -1);
 231         rtnl_unlock();
 232
 233         call_rcu(&netdev_vport->rcu, free_port_rcu);
 234 }
 235
 236 const char *ovs_netdev_get_name(const struct vport *vport)
 237 {
 238         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 239         return netdev_vport->dev->name;
 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         if (unlikely(!vport))
 246                 goto error;
 247
 248         if (unlikely(skb_warn_if_lro(skb)))
 249                 goto error;
 250
 251         /* Make our own copy of the packet.  Otherwise we will mangle the
 252          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 253          * (No one comes after us, since we tell handle_bridge() that we took
 254          * the packet.) */
 255         skb = skb_share_check(skb, GFP_ATOMIC);
 256         if (unlikely(!skb))
 257                 return;
 258
 259         if (unlikely(compute_ip_summed(skb, false)))
 260                 goto error;
 261
 262         skb_push(skb, ETH_HLEN);
 263         ovs_skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
 264
 265         vlan_copy_skb_tci(skb);
 266
 267         ovs_vport_receive(vport, skb, NULL);
 268         return;
 269
 270 error:
 271         kfree_skb(skb);
 272 }
 273
 274 static unsigned int packet_length(const struct sk_buff *skb)
 275 {
 276         unsigned int length = skb->len - ETH_HLEN;
 277
 278         if (skb->protocol == htons(ETH_P_8021Q))
 279                 length -= VLAN_HLEN;
 280
 281         return length;
 282 }
 283
 284 static bool dev_supports_vlan_tx(struct net_device *dev)
 285 {
 286 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
 287         /* Software fallback means every device supports vlan_tci on TX. */
 288         return true;
 289 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
 290         return dev->features & NETIF_F_HW_VLAN_TX;
 291 #else
 292         /* Assume that the driver is buggy. */
 293         return false;
 294 #endif
 295 }
 296
 297 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 298 {
 299         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 300         int mtu = netdev_vport->dev->mtu;
 301         int len;
 302
 303         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
 304                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
 305                                      netdev_vport->dev->name,
 306                                      packet_length(skb), mtu);
 307                 goto drop;
 308         }
 309
 310         skb->dev = netdev_vport->dev;
 311         forward_ip_summed(skb, true);
 312
 313         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
 314                 int features;
 315
 316                 features = netif_skb_features(skb);
 317
 318                 if (!vlan_tso)
 319                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
 320                                       NETIF_F_UFO | NETIF_F_FSO);
 321
 322                 if (netif_needs_gso(skb, features)) {
 323                         struct sk_buff *nskb;
 324
 325                         nskb = skb_gso_segment(skb, features);
 326                         if (!nskb) {
 327                                 if (unlikely(skb_cloned(skb) &&
 328                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
 329                                         goto drop;
 330
 331                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
 332                                 goto tag;
 333                         }
 334
 335                         if (IS_ERR(nskb))
 336                                 goto drop;
 337                         consume_skb(skb);
 338                         skb = nskb;
 339
 340                         len = 0;
 341                         do {
 342                                 nskb = skb->next;
 343                                 skb->next = NULL;
 344
 345                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 346                                 if (likely(skb)) {
 347                                         len += skb->len;
 348                                         vlan_set_tci(skb, 0);
 349                                         dev_queue_xmit(skb);
 350                                 }
 351
 352                                 skb = nskb;
 353                         } while (skb);
 354
 355                         return len;
 356                 }
 357
 358 tag:
 359                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 360                 if (unlikely(!skb))
 361                         return 0;
 362                 vlan_set_tci(skb, 0);
 363         }
 364
 365         len = skb->len;
 366         dev_queue_xmit(skb);
 367
 368         return len;
 369
 370 drop:
 371         kfree_skb(skb);
 372         return 0;
 373 }
 374
 375 /* Returns null if this device is not attached to a datapath. */
 376 struct vport *ovs_netdev_get_vport(struct net_device *dev)
 377 {
 378 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
 379     defined HAVE_RHEL_OVS_HOOK
 380 #if IFF_OVS_DATAPATH != 0
 381         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 382 #else
 383         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 384 #endif
 385 #ifdef HAVE_RHEL_OVS_HOOK
 386                 return (struct vport *)rcu_dereference_rtnl(dev->ax25_ptr);
 387 #else
 388                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 389 #endif
 390         else
 391                 return NULL;
 392 #else
 393         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
 394 #endif
 395 }
 396
 397 const struct vport_ops ovs_netdev_vport_ops = {
 398         .type           = OVS_VPORT_TYPE_NETDEV,
 399         .create         = netdev_create,
 400         .destroy        = netdev_destroy,
 401         .get_name       = ovs_netdev_get_name,
 402         .send           = netdev_send,
 403 };
 404
 405 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36) && \
 406     !defined HAVE_RHEL_OVS_HOOK
 407 /*
 408  * Enforces, mutual exclusion with the Linux bridge module, by declaring and
 409  * exporting br_should_route_hook.  Because the bridge module also exports the
 410  * same symbol, the module loader will refuse to load both modules at the same
 411  * time (e.g. "bridge: exports duplicate symbol br_should_route_hook (owned by
 412  * openvswitch)").
 413  *
 414  * Before Linux 2.6.36, Open vSwitch cannot safely coexist with the Linux
 415  * bridge module, so openvswitch uses this macro in those versions.  In
 416  * Linux 2.6.36 and later, Open vSwitch can coexist with the bridge module.
 417  *
 418  * The use of "typeof" here avoids the need to track changes in the type of
 419  * br_should_route_hook over various kernel versions.
 420  */
 421 typeof(br_should_route_hook) br_should_route_hook;
 422 EXPORT_SYMBOL(br_should_route_hook);
 423 #endif