datapath/vport-netdev.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira Networks.
   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 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  49
  50 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
  51 /* Called with rcu_read_lock and bottom-halves disabled. */
  52 static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
  53 {
  54         struct sk_buff *skb = *pskb;
  55         struct vport *vport;
  56
  57         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  58                 return RX_HANDLER_PASS;
  59
  60         vport = ovs_netdev_get_vport(skb->dev);
  61
  62         netdev_port_receive(vport, skb);
  63
  64         return RX_HANDLER_CONSUMED;
  65 }
  66 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  67 /* Called with rcu_read_lock and bottom-halves disabled. */
  68 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  69 {
  70         struct vport *vport;
  71
  72         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  73                 return skb;
  74
  75         vport = netdev_get_vport(skb->dev);
  76
  77         netdev_port_receive(vport, skb);
  78
  79         return NULL;
  80 }
  81 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
  82 /*
  83  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  84  * different set of devices!)
  85  */
  86 /* Called with rcu_read_lock and bottom-halves disabled. */
  87 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  88                                          struct sk_buff *skb)
  89 {
  90         netdev_port_receive((struct vport *)p, skb);
  91         return NULL;
  92 }
  93 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  94 /*
  95  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  96  * different set of devices!)
  97  */
  98 /* Called with rcu_read_lock and bottom-halves disabled. */
  99 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 100 {
 101         netdev_port_receive((struct vport *)p, *pskb);
 102         return 1;
 103 }
 104 #else
 105 #error
 106 #endif
 107
 108 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
 109 static int netdev_init(void) { return 0; }
 110 static void netdev_exit(void) { }
 111 #else
 112 static int netdev_init(void)
 113 {
 114         /* Hook into callback used by the bridge to intercept packets.
 115          * Parasites we are. */
 116         br_handle_frame_hook = netdev_frame_hook;
 117
 118         return 0;
 119 }
 120
 121 static void netdev_exit(void)
 122 {
 123         br_handle_frame_hook = NULL;
 124 }
 125 #endif
 126
 127 static struct vport *netdev_create(const struct vport_parms *parms)
 128 {
 129         struct vport *vport;
 130         struct netdev_vport *netdev_vport;
 131         int err;
 132
 133         vport = ovs_vport_alloc(sizeof(struct netdev_vport),
 134                                 &ovs_netdev_vport_ops, parms);
 135         if (IS_ERR(vport)) {
 136                 err = PTR_ERR(vport);
 137                 goto error;
 138         }
 139
 140         netdev_vport = netdev_vport_priv(vport);
 141
 142         netdev_vport->dev = dev_get_by_name(&init_net, parms->name);
 143         if (!netdev_vport->dev) {
 144                 err = -ENODEV;
 145                 goto error_free_vport;
 146         }
 147
 148         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 149             netdev_vport->dev->type != ARPHRD_ETHER ||
 150             ovs_is_internal_dev(netdev_vport->dev)) {
 151                 err = -EINVAL;
 152                 goto error_put;
 153         }
 154
 155         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 156                                          vport);
 157         if (err)
 158                 goto error_put;
 159
 160         dev_set_promiscuity(netdev_vport->dev, 1);
 161 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
 162         dev_disable_lro(netdev_vport->dev);
 163 #endif
 164         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 165
 166         return vport;
 167
 168 error_put:
 169         dev_put(netdev_vport->dev);
 170 error_free_vport:
 171         ovs_vport_free(vport);
 172 error:
 173         return ERR_PTR(err);
 174 }
 175
 176 static void netdev_destroy(struct vport *vport)
 177 {
 178         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 179
 180         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 181         netdev_rx_handler_unregister(netdev_vport->dev);
 182         dev_set_promiscuity(netdev_vport->dev, -1);
 183
 184         synchronize_rcu();
 185
 186         dev_put(netdev_vport->dev);
 187         ovs_vport_free(vport);
 188 }
 189
 190 int ovs_netdev_set_addr(struct vport *vport, const unsigned char *addr)
 191 {
 192         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 193         struct sockaddr sa;
 194
 195         sa.sa_family = ARPHRD_ETHER;
 196         memcpy(sa.sa_data, addr, ETH_ALEN);
 197
 198         return dev_set_mac_address(netdev_vport->dev, &sa);
 199 }
 200
 201 const char *ovs_netdev_get_name(const struct vport *vport)
 202 {
 203         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 204         return netdev_vport->dev->name;
 205 }
 206
 207 const unsigned char *ovs_netdev_get_addr(const struct vport *vport)
 208 {
 209         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 210         return netdev_vport->dev->dev_addr;
 211 }
 212
 213 struct kobject *ovs_netdev_get_kobj(const struct vport *vport)
 214 {
 215         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 216         return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 217 }
 218
 219 unsigned ovs_netdev_get_dev_flags(const struct vport *vport)
 220 {
 221         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 222         return dev_get_flags(netdev_vport->dev);
 223 }
 224
 225 int ovs_netdev_is_running(const struct vport *vport)
 226 {
 227         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 228         return netif_running(netdev_vport->dev);
 229 }
 230
 231 unsigned char ovs_netdev_get_operstate(const struct vport *vport)
 232 {
 233         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 234         return netdev_vport->dev->operstate;
 235 }
 236
 237 int ovs_netdev_get_ifindex(const struct vport *vport)
 238 {
 239         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 240         return netdev_vport->dev->ifindex;
 241 }
 242
 243 int ovs_netdev_get_mtu(const struct vport *vport)
 244 {
 245         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 246         return netdev_vport->dev->mtu;
 247 }
 248
 249 /* Must be called with rcu_read_lock. */
 250 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 251 {
 252         if (unlikely(!vport)) {
 253                 kfree_skb(skb);
 254                 return;
 255         }
 256
 257         /* Make our own copy of the packet.  Otherwise we will mangle the
 258          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 259          * (No one comes after us, since we tell handle_bridge() that we took
 260          * the packet.) */
 261         skb = skb_share_check(skb, GFP_ATOMIC);
 262         if (unlikely(!skb))
 263                 return;
 264
 265         skb_push(skb, ETH_HLEN);
 266
 267         if (unlikely(compute_ip_summed(skb, false))) {
 268                 kfree_skb(skb);
 269                 return;
 270         }
 271         vlan_copy_skb_tci(skb);
 272
 273         ovs_vport_receive(vport, skb);
 274 }
 275
 276 static unsigned packet_length(const struct sk_buff *skb)
 277 {
 278         unsigned length = skb->len - ETH_HLEN;
 279
 280         if (skb->protocol == htons(ETH_P_8021Q))
 281                 length -= VLAN_HLEN;
 282
 283         return length;
 284 }
 285
 286 static bool dev_supports_vlan_tx(struct net_device *dev)
 287 {
 288 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
 289         /* Software fallback means every device supports vlan_tci on TX. */
 290         return true;
 291 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
 292         return dev->features & NETIF_F_HW_VLAN_TX;
 293 #else
 294         /* Assume that the driver is buggy. */
 295         return false;
 296 #endif
 297 }
 298
 299 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 300 {
 301         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 302         int mtu = netdev_vport->dev->mtu;
 303         int len;
 304
 305         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
 306                 if (net_ratelimit())
 307                         pr_warn("%s: dropped over-mtu packet: %d > %d\n",
 308                                 ovs_dp_name(vport->dp), packet_length(skb), mtu);
 309                 goto error;
 310         }
 311
 312         if (unlikely(skb_warn_if_lro(skb)))
 313                 goto error;
 314
 315         skb->dev = netdev_vport->dev;
 316         forward_ip_summed(skb, true);
 317
 318         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
 319                 int features;
 320
 321                 features = netif_skb_features(skb);
 322
 323                 if (!vlan_tso)
 324                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
 325                                       NETIF_F_UFO | NETIF_F_FSO);
 326
 327                 if (netif_needs_gso(skb, features)) {
 328                         struct sk_buff *nskb;
 329
 330                         nskb = skb_gso_segment(skb, features);
 331                         if (!nskb) {
 332                                 if (unlikely(skb_cloned(skb) &&
 333                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) {
 334                                         kfree_skb(skb);
 335                                         return 0;
 336                                 }
 337
 338                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
 339                                 goto tag;
 340                         }
 341
 342                         if (IS_ERR(nskb)) {
 343                                 kfree_skb(skb);
 344                                 return 0;
 345                         }
 346                         consume_skb(skb);
 347                         skb = nskb;
 348
 349                         len = 0;
 350                         do {
 351                                 nskb = skb->next;
 352                                 skb->next = NULL;
 353
 354                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 355                                 if (likely(skb)) {
 356                                         len += skb->len;
 357                                         vlan_set_tci(skb, 0);
 358                                         dev_queue_xmit(skb);
 359                                 }
 360
 361                                 skb = nskb;
 362                         } while (skb);
 363
 364                         return len;
 365                 }
 366
 367 tag:
 368                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 369                 if (unlikely(!skb))
 370                         return 0;
 371                 vlan_set_tci(skb, 0);
 372         }
 373
 374         len = skb->len;
 375         dev_queue_xmit(skb);
 376
 377         return len;
 378
 379 error:
 380         kfree_skb(skb);
 381         ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
 382         return 0;
 383 }
 384
 385 /* Returns null if this device is not attached to a datapath. */
 386 struct vport *ovs_netdev_get_vport(struct net_device *dev)
 387 {
 388 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
 389 #if IFF_BRIDGE_PORT != IFF_OVS_DATAPATH
 390         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 391 #else
 392         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 393 #endif
 394                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 395         else
 396                 return NULL;
 397 #else
 398         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
 399 #endif
 400 }
 401
 402 const struct vport_ops ovs_netdev_vport_ops = {
 403         .type           = OVS_VPORT_TYPE_NETDEV,
 404         .flags          = VPORT_F_REQUIRED,
 405         .init           = netdev_init,
 406         .exit           = netdev_exit,
 407         .create         = netdev_create,
 408         .destroy        = netdev_destroy,
 409         .set_addr       = ovs_netdev_set_addr,
 410         .get_name       = ovs_netdev_get_name,
 411         .get_addr       = ovs_netdev_get_addr,
 412         .get_kobj       = ovs_netdev_get_kobj,
 413         .get_dev_flags  = ovs_netdev_get_dev_flags,
 414         .is_running     = ovs_netdev_is_running,
 415         .get_operstate  = ovs_netdev_get_operstate,
 416         .get_ifindex    = ovs_netdev_get_ifindex,
 417         .get_mtu        = ovs_netdev_get_mtu,
 418         .send           = netdev_send,
 419 };
 420
 421 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36)
 422 /*
 423  * In kernels earlier than 2.6.36, Open vSwitch cannot safely coexist with
 424  * the Linux bridge module on any released version of Linux, because there
 425  * is only a single bridge hook function and only a single br_port member
 426  * in struct net_device.
 427  *
 428  * Declaring and exporting this symbol enforces mutual exclusion.  The bridge
 429  * module also exports the same symbol, so the module loader will refuse to
 430  * load both modules at the same time (e.g. "bridge: exports duplicate symbol
 431  * br_should_route_hook (owned by openvswitch_mod)").
 432  *
 433  * The use of "typeof" here avoids the need to track changes in the type of
 434  * br_should_route_hook over various kernel versions.
 435  */
 436 typeof(br_should_route_hook) br_should_route_hook;
 437 EXPORT_SYMBOL(br_should_route_hook);
 438 #endif