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 "checksum.h"
  33 #include "datapath.h"
  34 #include "vlan.h"
  35 #include "vport-internal_dev.h"
  36 #include "vport-netdev.h"
  37
  38 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) && \
  39         !defined(HAVE_VLAN_BUG_WORKAROUND)
  40 #include <linux/module.h>
  41
  42 static int vlan_tso __read_mostly;
  43 module_param(vlan_tso, int, 0644);
  44 MODULE_PARM_DESC(vlan_tso, "Enable TSO for VLAN packets");
  45 #else
  46 #define vlan_tso true
  47 #endif
  48
  49 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  50
  51 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
  52 /* Called with rcu_read_lock and bottom-halves disabled. */
  53 static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
  54 {
  55         struct sk_buff *skb = *pskb;
  56         struct vport *vport;
  57
  58         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  59                 return RX_HANDLER_PASS;
  60
  61         vport = ovs_netdev_get_vport(skb->dev);
  62
  63         netdev_port_receive(vport, skb);
  64
  65         return RX_HANDLER_CONSUMED;
  66 }
  67 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
  68       defined HAVE_RHEL_OVS_HOOK
  69 /* Called with rcu_read_lock and bottom-halves disabled. */
  70 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  71 {
  72         struct vport *vport;
  73
  74         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  75                 return skb;
  76
  77         vport = ovs_netdev_get_vport(skb->dev);
  78
  79         netdev_port_receive(vport, skb);
  80
  81         return NULL;
  82 }
  83 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,32)
  84 /*
  85  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  86  * different set of devices!)
  87  */
  88 /* Called with rcu_read_lock and bottom-halves disabled. */
  89 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  90                                          struct sk_buff *skb)
  91 {
  92         netdev_port_receive((struct vport *)p, skb);
  93         return NULL;
  94 }
  95 #else
  96 #error
  97 #endif
  98
  99 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
 100     defined HAVE_RHEL_OVS_HOOK
 101 static int netdev_init(void) { return 0; }
 102 static void netdev_exit(void) { }
 103 #else
 104 static int port_count;
 105
 106 static void netdev_init(void)
 107 {
 108         port_count++;
 109         if (port_count > 1)
 110                 return;
 111
 112         /* Hook into callback used by the bridge to intercept packets.
 113          * Parasites we are. */
 114         br_handle_frame_hook = netdev_frame_hook;
 115
 116         return;
 117 }
 118
 119 static void netdev_exit(void)
 120 {
 121         port_count--;
 122         if (port_count > 0)
 123                 return;
 124
 125         br_handle_frame_hook = NULL;
 126 }
 127 #endif
 128
 129 static struct net_device *get_dpdev(struct datapath *dp)
 130 {
 131         struct vport *local;
 132
 133         local = ovs_vport_ovsl(dp, OVSP_LOCAL);
 134         BUG_ON(!local);
 135         return netdev_vport_priv(local)->dev;
 136 }
 137
 138 static struct vport *netdev_create(const struct vport_parms *parms)
 139 {
 140         struct vport *vport;
 141         struct netdev_vport *netdev_vport;
 142         int err;
 143
 144         vport = ovs_vport_alloc(sizeof(struct netdev_vport),
 145                                 &ovs_netdev_vport_ops, parms);
 146         if (IS_ERR(vport)) {
 147                 err = PTR_ERR(vport);
 148                 goto error;
 149         }
 150
 151         netdev_vport = netdev_vport_priv(vport);
 152
 153         netdev_vport->dev = dev_get_by_name(ovs_dp_get_net(vport->dp), parms->name);
 154         if (!netdev_vport->dev) {
 155                 err = -ENODEV;
 156                 goto error_free_vport;
 157         }
 158
 159         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 160             netdev_vport->dev->type != ARPHRD_ETHER ||
 161             ovs_is_internal_dev(netdev_vport->dev)) {
 162                 err = -EINVAL;
 163                 goto error_put;
 164         }
 165
 166         rtnl_lock();
 167         err = netdev_master_upper_dev_link(netdev_vport->dev,
 168                                            get_dpdev(vport->dp));
 169         if (err)
 170                 goto error_unlock;
 171
 172         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 173                                          vport);
 174         if (err)
 175                 goto error_master_upper_dev_unlink;
 176
 177         dev_set_promiscuity(netdev_vport->dev, 1);
 178         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 179         rtnl_unlock();
 180
 181         netdev_init();
 182         return vport;
 183
 184 error_master_upper_dev_unlink:
 185         netdev_upper_dev_unlink(netdev_vport->dev, get_dpdev(vport->dp));
 186 error_unlock:
 187         rtnl_unlock();
 188 error_put:
 189         dev_put(netdev_vport->dev);
 190 error_free_vport:
 191         ovs_vport_free(vport);
 192 error:
 193         return ERR_PTR(err);
 194 }
 195
 196 static void free_port_rcu(struct rcu_head *rcu)
 197 {
 198         struct netdev_vport *netdev_vport = container_of(rcu,
 199                                         struct netdev_vport, rcu);
 200
 201         dev_put(netdev_vport->dev);
 202         ovs_vport_free(vport_from_priv(netdev_vport));
 203 }
 204
 205 static void netdev_destroy(struct vport *vport)
 206 {
 207         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 208
 209         netdev_exit();
 210         rtnl_lock();
 211         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 212         netdev_rx_handler_unregister(netdev_vport->dev);
 213         netdev_upper_dev_unlink(netdev_vport->dev, get_dpdev(vport->dp));
 214         dev_set_promiscuity(netdev_vport->dev, -1);
 215         rtnl_unlock();
 216
 217         call_rcu(&netdev_vport->rcu, free_port_rcu);
 218 }
 219
 220 const char *ovs_netdev_get_name(const struct vport *vport)
 221 {
 222         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 223         return netdev_vport->dev->name;
 224 }
 225
 226 /* Must be called with rcu_read_lock. */
 227 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 228 {
 229         if (unlikely(!vport))
 230                 goto error;
 231
 232         if (unlikely(skb_warn_if_lro(skb)))
 233                 goto error;
 234
 235         /* Make our own copy of the packet.  Otherwise we will mangle the
 236          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 237          * (No one comes after us, since we tell handle_bridge() that we took
 238          * the packet.) */
 239         skb = skb_share_check(skb, GFP_ATOMIC);
 240         if (unlikely(!skb))
 241                 return;
 242
 243         if (unlikely(compute_ip_summed(skb, false)))
 244                 goto error;
 245
 246         skb_push(skb, ETH_HLEN);
 247         ovs_skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
 248
 249         vlan_copy_skb_tci(skb);
 250
 251         ovs_vport_receive(vport, skb, NULL);
 252         return;
 253
 254 error:
 255         kfree_skb(skb);
 256 }
 257
 258 static unsigned int packet_length(const struct sk_buff *skb)
 259 {
 260         unsigned int length = skb->len - ETH_HLEN;
 261
 262         if (skb->protocol == htons(ETH_P_8021Q))
 263                 length -= VLAN_HLEN;
 264
 265         return length;
 266 }
 267
 268 static bool dev_supports_vlan_tx(struct net_device *dev)
 269 {
 270 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
 271         /* Software fallback means every device supports vlan_tci on TX. */
 272         return true;
 273 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
 274         return dev->features & NETIF_F_HW_VLAN_TX;
 275 #else
 276         /* Assume that the driver is buggy. */
 277         return false;
 278 #endif
 279 }
 280
 281 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 282 {
 283         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 284         int mtu = netdev_vport->dev->mtu;
 285         int len;
 286
 287         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
 288                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
 289                                      netdev_vport->dev->name,
 290                                      packet_length(skb), mtu);
 291                 goto drop;
 292         }
 293
 294         skb->dev = netdev_vport->dev;
 295         forward_ip_summed(skb, true);
 296
 297         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
 298                 int features;
 299
 300                 features = netif_skb_features(skb);
 301
 302                 if (!vlan_tso)
 303                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
 304                                       NETIF_F_UFO | NETIF_F_FSO);
 305
 306                 if (netif_needs_gso(skb, features)) {
 307                         struct sk_buff *nskb;
 308
 309                         nskb = skb_gso_segment(skb, features);
 310                         if (!nskb) {
 311                                 if (unlikely(skb_cloned(skb) &&
 312                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
 313                                         goto drop;
 314
 315                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
 316                                 goto tag;
 317                         }
 318
 319                         if (IS_ERR(nskb))
 320                                 goto drop;
 321                         consume_skb(skb);
 322                         skb = nskb;
 323
 324                         len = 0;
 325                         do {
 326                                 nskb = skb->next;
 327                                 skb->next = NULL;
 328
 329                                 skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
 330                                 if (likely(skb)) {
 331                                         len += skb->len;
 332                                         vlan_set_tci(skb, 0);
 333                                         dev_queue_xmit(skb);
 334                                 }
 335
 336                                 skb = nskb;
 337                         } while (skb);
 338
 339                         return len;
 340                 }
 341
 342 tag:
 343                 skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
 344                 if (unlikely(!skb))
 345                         return 0;
 346                 vlan_set_tci(skb, 0);
 347         }
 348
 349         len = skb->len;
 350         dev_queue_xmit(skb);
 351
 352         return len;
 353
 354 drop:
 355         kfree_skb(skb);
 356         return 0;
 357 }
 358
 359 /* Returns null if this device is not attached to a datapath. */
 360 struct vport *ovs_netdev_get_vport(struct net_device *dev)
 361 {
 362 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
 363     defined HAVE_RHEL_OVS_HOOK
 364 #if IFF_OVS_DATAPATH != 0
 365         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 366 #else
 367         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 368 #endif
 369 #ifdef HAVE_RHEL_OVS_HOOK
 370                 return (struct vport *)rcu_dereference_rtnl(dev->ax25_ptr);
 371 #else
 372                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 373 #endif
 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 ovs_netdev_vport_ops = {
 382         .type           = OVS_VPORT_TYPE_NETDEV,
 383         .create         = netdev_create,
 384         .destroy        = netdev_destroy,
 385         .get_name       = ovs_netdev_get_name,
 386         .send           = netdev_send,
 387 };
 388
 389 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36) && \
 390     !defined HAVE_RHEL_OVS_HOOK
 391 /*
 392  * Enforces, mutual exclusion with the Linux bridge module, by declaring and
 393  * exporting br_should_route_hook.  Because the bridge module also exports the
 394  * same symbol, the module loader will refuse to load both modules at the same
 395  * time (e.g. "bridge: exports duplicate symbol br_should_route_hook (owned by
 396  * openvswitch)").
 397  *
 398  * Before Linux 2.6.36, Open vSwitch cannot safely coexist with the Linux
 399  * bridge module, so openvswitch uses this macro in those versions.  In
 400  * Linux 2.6.36 and later, Open vSwitch can coexist with the bridge module.
 401  *
 402  * The use of "typeof" here avoids the need to track changes in the type of
 403  * br_should_route_hook over various kernel versions.
 404  */
 405 typeof(br_should_route_hook) br_should_route_hook;
 406 EXPORT_SYMBOL(br_should_route_hook);
 407 #endif