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(2,6,22)
  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 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  96 /*
  97  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  98  * different set of devices!)
  99  */
 100 /* Called with rcu_read_lock and bottom-halves disabled. */
 101 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 102 {
 103         netdev_port_receive((struct vport *)p, *pskb);
 104         return 1;
 105 }
 106 #else
 107 #error
 108 #endif
 109
 110 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
 111     defined HAVE_RHEL_OVS_HOOK
 112 static int netdev_init(void) { return 0; }
 113 static void netdev_exit(void) { }
 114 #else
 115 static int port_count;
 116
 117 static void netdev_init(void)
 118 {
 119         port_count++;
 120         if (port_count > 1)
 121                 return;
 122
 123         /* Hook into callback used by the bridge to intercept packets.
 124          * Parasites we are. */
 125         br_handle_frame_hook = netdev_frame_hook;
 126
 127         return;
 128 }
 129
 130 static void netdev_exit(void)
 131 {
 132         port_count--;
 133         if (port_count > 0)
 134                 return;
 135
 136         br_handle_frame_hook = NULL;
 137 }
 138 #endif
 139
 140 static struct net_device *get_dpdev(struct datapath *dp)
 141 {
 142         struct vport *local;
 143
 144         local = ovs_vport_ovsl(dp, OVSP_LOCAL);
 145         BUG_ON(!local);
 146         return netdev_vport_priv(local)->dev;
 147 }
 148
 149 static struct vport *netdev_create(const struct vport_parms *parms)
 150 {
 151         struct vport *vport;
 152         struct netdev_vport *netdev_vport;
 153         int err;
 154
 155         vport = ovs_vport_alloc(sizeof(struct netdev_vport),
 156                                 &ovs_netdev_vport_ops, parms);
 157         if (IS_ERR(vport)) {
 158                 err = PTR_ERR(vport);
 159                 goto error;
 160         }
 161
 162         netdev_vport = netdev_vport_priv(vport);
 163
 164         netdev_vport->dev = dev_get_by_name(ovs_dp_get_net(vport->dp), parms->name);
 165         if (!netdev_vport->dev) {
 166                 err = -ENODEV;
 167                 goto error_free_vport;
 168         }
 169
 170         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 171             netdev_vport->dev->type != ARPHRD_ETHER ||
 172             ovs_is_internal_dev(netdev_vport->dev)) {
 173                 err = -EINVAL;
 174                 goto error_put;
 175         }
 176
 177         rtnl_lock();
 178         err = netdev_master_upper_dev_link(netdev_vport->dev,
 179                                            get_dpdev(vport->dp));
 180         if (err)
 181                 goto error_unlock;
 182
 183         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 184                                          vport);
 185         if (err)
 186                 goto error_master_upper_dev_unlink;
 187
 188         dev_set_promiscuity(netdev_vport->dev, 1);
 189 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
 190         dev_disable_lro(netdev_vport->dev);
 191 #endif
 192         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 193         rtnl_unlock();
 194
 195         netdev_init();
 196         return vport;
 197
 198 error_master_upper_dev_unlink:
 199         netdev_upper_dev_unlink(netdev_vport->dev, get_dpdev(vport->dp));
 200 error_unlock:
 201         rtnl_unlock();
 202 error_put:
 203         dev_put(netdev_vport->dev);
 204 error_free_vport:
 205         ovs_vport_free(vport);
 206 error:
 207         return ERR_PTR(err);
 208 }
 209
 210 static void free_port_rcu(struct rcu_head *rcu)
 211 {
 212         struct netdev_vport *netdev_vport = container_of(rcu,
 213                                         struct netdev_vport, rcu);
 214
 215         dev_put(netdev_vport->dev);
 216         ovs_vport_free(vport_from_priv(netdev_vport));
 217 }
 218
 219 static void netdev_destroy(struct vport *vport)
 220 {
 221         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 222
 223         netdev_exit();
 224         rtnl_lock();
 225         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 226         netdev_rx_handler_unregister(netdev_vport->dev);
 227         netdev_upper_dev_unlink(netdev_vport->dev, get_dpdev(vport->dp));
 228         dev_set_promiscuity(netdev_vport->dev, -1);
 229         rtnl_unlock();
 230
 231         call_rcu(&netdev_vport->rcu, free_port_rcu);
 232 }
 233
 234 const char *ovs_netdev_get_name(const struct vport *vport)
 235 {
 236         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 237         return netdev_vport->dev->name;
 238 }
 239
 240 /* Must be called with rcu_read_lock. */
 241 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 242 {
 243         if (unlikely(!vport))
 244                 goto error;
 245
 246         if (unlikely(skb_warn_if_lro(skb)))
 247                 goto error;
 248
 249         /* Make our own copy of the packet.  Otherwise we will mangle the
 250          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 251          * (No one comes after us, since we tell handle_bridge() that we took
 252          * the packet.) */
 253         skb = skb_share_check(skb, GFP_ATOMIC);
 254         if (unlikely(!skb))
 255                 return;
 256
 257         if (unlikely(compute_ip_summed(skb, false)))
 258                 goto error;
 259
 260         skb_push(skb, ETH_HLEN);
 261         ovs_skb_postpush_rcsum(skb, skb->data, ETH_HLEN);
 262
 263         vlan_copy_skb_tci(skb);
 264
 265         ovs_vport_receive(vport, skb, NULL);
 266         return;
 267
 268 error:
 269         kfree_skb(skb);
 270 }
 271
 272 static unsigned int packet_length(const struct sk_buff *skb)
 273 {
 274         unsigned int length = skb->len - ETH_HLEN;
 275
 276         if (skb->protocol == htons(ETH_P_8021Q))
 277                 length -= VLAN_HLEN;
 278
 279         return length;
 280 }
 281
 282 static bool dev_supports_vlan_tx(struct net_device *dev)
 283 {
 284 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
 285         /* Software fallback means every device supports vlan_tci on TX. */
 286         return true;
 287 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
 288         return dev->features & NETIF_F_HW_VLAN_TX;
 289 #else
 290         /* Assume that the driver is buggy. */
 291         return false;
 292 #endif
 293 }
 294
 295 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 296 {
 297         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 298         int mtu = netdev_vport->dev->mtu;
 299         int len;
 300
 301         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
 302                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
 303                                      netdev_vport->dev->name,
 304                                      packet_length(skb), mtu);
 305                 goto drop;
 306         }
 307
 308         skb->dev = netdev_vport->dev;
 309         forward_ip_summed(skb, true);
 310
 311         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
 312                 int features;
 313
 314                 features = netif_skb_features(skb);
 315
 316                 if (!vlan_tso)
 317                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
 318                                       NETIF_F_UFO | NETIF_F_FSO);
 319
 320                 if (netif_needs_gso(skb, features)) {
 321                         struct sk_buff *nskb;
 322
 323                         nskb = skb_gso_segment(skb, features);
 324                         if (!nskb) {
 325                                 if (unlikely(skb_cloned(skb) &&
 326                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
 327                                         goto drop;
 328
 329                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
 330                                 goto tag;
 331                         }
 332
 333                         if (IS_ERR(nskb))
 334                                 goto drop;
 335                         consume_skb(skb);
 336                         skb = nskb;
 337
 338                         len = 0;
 339                         do {
 340                                 nskb = skb->next;
 341                                 skb->next = NULL;
 342
 343                                 skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
 344                                 if (likely(skb)) {
 345                                         len += skb->len;
 346                                         vlan_set_tci(skb, 0);
 347                                         dev_queue_xmit(skb);
 348                                 }
 349
 350                                 skb = nskb;
 351                         } while (skb);
 352
 353                         return len;
 354                 }
 355
 356 tag:
 357                 skb = __vlan_put_tag(skb, skb->vlan_proto, vlan_tx_tag_get(skb));
 358                 if (unlikely(!skb))
 359                         return 0;
 360                 vlan_set_tci(skb, 0);
 361         }
 362
 363         len = skb->len;
 364         dev_queue_xmit(skb);
 365
 366         return len;
 367
 368 drop:
 369         kfree_skb(skb);
 370         return 0;
 371 }
 372
 373 /* Returns null if this device is not attached to a datapath. */
 374 struct vport *ovs_netdev_get_vport(struct net_device *dev)
 375 {
 376 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
 377     defined HAVE_RHEL_OVS_HOOK
 378 #if IFF_OVS_DATAPATH != 0
 379         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 380 #else
 381         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 382 #endif
 383 #ifdef HAVE_RHEL_OVS_HOOK
 384                 return (struct vport *)rcu_dereference_rtnl(dev->ax25_ptr);
 385 #else
 386                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 387 #endif
 388         else
 389                 return NULL;
 390 #else
 391         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
 392 #endif
 393 }
 394
 395 const struct vport_ops ovs_netdev_vport_ops = {
 396         .type           = OVS_VPORT_TYPE_NETDEV,
 397         .create         = netdev_create,
 398         .destroy        = netdev_destroy,
 399         .get_name       = ovs_netdev_get_name,
 400         .send           = netdev_send,
 401 };
 402
 403 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36) && \
 404     !defined HAVE_RHEL_OVS_HOOK
 405 /*
 406  * Enforces, mutual exclusion with the Linux bridge module, by declaring and
 407  * exporting br_should_route_hook.  Because the bridge module also exports the
 408  * same symbol, the module loader will refuse to load both modules at the same
 409  * time (e.g. "bridge: exports duplicate symbol br_should_route_hook (owned by
 410  * openvswitch)").
 411  *
 412  * Before Linux 2.6.36, Open vSwitch cannot safely coexist with the Linux
 413  * bridge module, so openvswitch uses this macro in those versions.  In
 414  * Linux 2.6.36 and later, Open vSwitch can coexist with the bridge module.
 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