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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/if_arp.h>
  12 #include <linux/if_bridge.h>
  13 #include <linux/if_vlan.h>
  14 #include <linux/kernel.h>
  15 #include <linux/llc.h>
  16 #include <linux/rtnetlink.h>
  17 #include <linux/skbuff.h>
  18
  19 #include <net/llc.h>
  20
  21 #include "checksum.h"
  22 #include "datapath.h"
  23 #include "vlan.h"
  24 #include "vport-internal_dev.h"
  25 #include "vport-netdev.h"
  26
  27 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) && \
  28     !defined(HAVE_VLAN_BUG_WORKAROUND)
  29 #include <linux/module.h>
  30
  31 static int vlan_tso __read_mostly = 0;
  32 module_param(vlan_tso, int, 0644);
  33 MODULE_PARM_DESC(vlan_tso, "Enable TSO for VLAN packets");
  34 #else
  35 #define vlan_tso true
  36 #endif
  37
  38 /* If the native device stats aren't 64 bit use the vport stats tracking instead. */
  39 #define USE_VPORT_STATS (sizeof(((struct net_device_stats *)0)->rx_bytes) < sizeof(u64))
  40
  41 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
  42
  43 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
  44 /* Called with rcu_read_lock and bottom-halves disabled. */
  45 static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
  46 {
  47         struct sk_buff *skb = *pskb;
  48         struct vport *vport;
  49
  50         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  51                 return RX_HANDLER_PASS;
  52
  53         vport = netdev_get_vport(skb->dev);
  54
  55         netdev_port_receive(vport, skb);
  56
  57         return RX_HANDLER_CONSUMED;
  58 }
  59 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  60 /* Called with rcu_read_lock and bottom-halves disabled. */
  61 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
  62 {
  63         struct vport *vport;
  64
  65         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
  66                 return skb;
  67
  68         vport = netdev_get_vport(skb->dev);
  69
  70         netdev_port_receive(vport, skb);
  71
  72         return NULL;
  73 }
  74 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
  75 /*
  76  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  77  * different set of devices!)
  78  */
  79 /* Called with rcu_read_lock and bottom-halves disabled. */
  80 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
  81                                          struct sk_buff *skb)
  82 {
  83         netdev_port_receive((struct vport *)p, skb);
  84         return NULL;
  85 }
  86 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  87 /*
  88  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
  89  * different set of devices!)
  90  */
  91 /* Called with rcu_read_lock and bottom-halves disabled. */
  92 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
  93 {
  94         netdev_port_receive((struct vport *)p, *pskb);
  95         return 1;
  96 }
  97 #else
  98 #error
  99 #endif
 100
 101 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
 102 static int netdev_init(void) { return 0; }
 103 static void netdev_exit(void) { }
 104 #else
 105 static int netdev_init(void)
 106 {
 107         /* Hook into callback used by the bridge to intercept packets.
 108          * Parasites we are. */
 109         br_handle_frame_hook = netdev_frame_hook;
 110
 111         return 0;
 112 }
 113
 114 static void netdev_exit(void)
 115 {
 116         br_handle_frame_hook = NULL;
 117 }
 118 #endif
 119
 120 static struct vport *netdev_create(const struct vport_parms *parms)
 121 {
 122         struct vport *vport;
 123         struct netdev_vport *netdev_vport;
 124         int err;
 125
 126         vport = vport_alloc(sizeof(struct netdev_vport), &netdev_vport_ops, parms);
 127         if (IS_ERR(vport)) {
 128                 err = PTR_ERR(vport);
 129                 goto error;
 130         }
 131
 132         netdev_vport = netdev_vport_priv(vport);
 133
 134         netdev_vport->dev = dev_get_by_name(&init_net, parms->name);
 135         if (!netdev_vport->dev) {
 136                 err = -ENODEV;
 137                 goto error_free_vport;
 138         }
 139
 140         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
 141             netdev_vport->dev->type != ARPHRD_ETHER ||
 142             is_internal_dev(netdev_vport->dev)) {
 143                 err = -EINVAL;
 144                 goto error_put;
 145         }
 146
 147         /* If we are using the vport stats layer initialize it to the current
 148          * values so we are roughly consistent with the device stats. */
 149         if (USE_VPORT_STATS) {
 150                 struct rtnl_link_stats64 stats;
 151
 152                 err = netdev_get_stats(vport, &stats);
 153                 if (!err)
 154                         vport_set_stats(vport, &stats);
 155         }
 156
 157         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
 158                                          vport);
 159         if (err)
 160                 goto error_put;
 161
 162         dev_set_promiscuity(netdev_vport->dev, 1);
 163 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
 164         dev_disable_lro(netdev_vport->dev);
 165 #endif
 166         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
 167
 168         return vport;
 169
 170 error_put:
 171         dev_put(netdev_vport->dev);
 172 error_free_vport:
 173         vport_free(vport);
 174 error:
 175         return ERR_PTR(err);
 176 }
 177
 178 static int netdev_destroy(struct vport *vport)
 179 {
 180         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 181
 182         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
 183         netdev_rx_handler_unregister(netdev_vport->dev);
 184         dev_set_promiscuity(netdev_vport->dev, -1);
 185
 186         synchronize_rcu();
 187
 188         dev_put(netdev_vport->dev);
 189         vport_free(vport);
 190
 191         return 0;
 192 }
 193
 194 int netdev_set_mtu(struct vport *vport, int mtu)
 195 {
 196         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 197         return dev_set_mtu(netdev_vport->dev, mtu);
 198 }
 199
 200 int netdev_set_addr(struct vport *vport, const unsigned char *addr)
 201 {
 202         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 203         struct sockaddr sa;
 204
 205         sa.sa_family = ARPHRD_ETHER;
 206         memcpy(sa.sa_data, addr, ETH_ALEN);
 207
 208         return dev_set_mac_address(netdev_vport->dev, &sa);
 209 }
 210
 211 const char *netdev_get_name(const struct vport *vport)
 212 {
 213         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 214         return netdev_vport->dev->name;
 215 }
 216
 217 const unsigned char *netdev_get_addr(const struct vport *vport)
 218 {
 219         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 220         return netdev_vport->dev->dev_addr;
 221 }
 222
 223 struct kobject *netdev_get_kobj(const struct vport *vport)
 224 {
 225         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 226         return &netdev_vport->dev->NETDEV_DEV_MEMBER.kobj;
 227 }
 228
 229 int netdev_get_stats(const struct vport *vport, struct rtnl_link_stats64 *stats)
 230 {
 231         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 232         dev_get_stats(netdev_vport->dev, stats);
 233         return 0;
 234 }
 235
 236 unsigned netdev_get_dev_flags(const struct vport *vport)
 237 {
 238         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 239         return dev_get_flags(netdev_vport->dev);
 240 }
 241
 242 int netdev_is_running(const struct vport *vport)
 243 {
 244         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 245         return netif_running(netdev_vport->dev);
 246 }
 247
 248 unsigned char netdev_get_operstate(const struct vport *vport)
 249 {
 250         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 251         return netdev_vport->dev->operstate;
 252 }
 253
 254 int netdev_get_ifindex(const struct vport *vport)
 255 {
 256         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 257         return netdev_vport->dev->ifindex;
 258 }
 259
 260 int netdev_get_mtu(const struct vport *vport)
 261 {
 262         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 263         return netdev_vport->dev->mtu;
 264 }
 265
 266 /* Must be called with rcu_read_lock. */
 267 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
 268 {
 269         if (unlikely(!vport)) {
 270                 kfree_skb(skb);
 271                 return;
 272         }
 273
 274         /* Make our own copy of the packet.  Otherwise we will mangle the
 275          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 276          * (No one comes after us, since we tell handle_bridge() that we took
 277          * the packet.) */
 278         skb = skb_share_check(skb, GFP_ATOMIC);
 279         if (unlikely(!skb))
 280                 return;
 281
 282         skb_push(skb, ETH_HLEN);
 283
 284         if (unlikely(compute_ip_summed(skb, false))) {
 285                 kfree_skb(skb);
 286                 return;
 287         }
 288         vlan_copy_skb_tci(skb);
 289
 290         vport_receive(vport, skb);
 291 }
 292
 293 static inline unsigned packet_length(const struct sk_buff *skb)
 294 {
 295         unsigned length = skb->len - ETH_HLEN;
 296
 297         if (skb->protocol == htons(ETH_P_8021Q))
 298                 length -= VLAN_HLEN;
 299
 300         return length;
 301 }
 302
 303 static bool dev_supports_vlan_tx(struct net_device *dev)
 304 {
 305 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
 306         /* Software fallback means every device supports vlan_tci on TX. */
 307         return true;
 308 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
 309         return dev->features & NETIF_F_HW_VLAN_TX;
 310 #else
 311         /* Assume that the driver is buggy. */
 312         return false;
 313 #endif
 314 }
 315
 316 static int netdev_send(struct vport *vport, struct sk_buff *skb)
 317 {
 318         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
 319         int mtu = netdev_vport->dev->mtu;
 320         int len;
 321
 322         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
 323                 if (net_ratelimit())
 324                         pr_warn("%s: dropped over-mtu packet: %d > %d\n",
 325                                 dp_name(vport->dp), packet_length(skb), mtu);
 326                 goto error;
 327         }
 328
 329         if (unlikely(skb_warn_if_lro(skb)))
 330                 goto error;
 331
 332         skb->dev = netdev_vport->dev;
 333         forward_ip_summed(skb, true);
 334
 335         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
 336                 int features = 0;
 337
 338 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,26)
 339                 features = skb->dev->features & skb->dev->vlan_features;
 340 #endif
 341
 342                 if (!vlan_tso)
 343                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
 344                                       NETIF_F_UFO | NETIF_F_FSO);
 345
 346                 if (skb_is_gso(skb) &&
 347                     (!skb_gso_ok(skb, features) ||
 348                      unlikely(skb->ip_summed != CHECKSUM_PARTIAL))) {
 349                         struct sk_buff *nskb;
 350
 351                         nskb = skb_gso_segment(skb, features);
 352                         if (!nskb) {
 353                                 if (unlikely(skb_cloned(skb) &&
 354                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) {
 355                                         kfree_skb(skb);
 356                                         return 0;
 357                                 }
 358
 359                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
 360                                 goto tag;
 361                         }
 362
 363                         if (IS_ERR(nskb)) {
 364                                 kfree_skb(skb);
 365                                 return 0;
 366                         }
 367                         consume_skb(skb);
 368                         skb = nskb;
 369
 370                         len = 0;
 371                         do {
 372                                 nskb = skb->next;
 373                                 skb->next = NULL;
 374
 375                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 376                                 if (likely(skb)) {
 377                                         len += skb->len;
 378                                         vlan_set_tci(skb, 0);
 379                                         dev_queue_xmit(skb);
 380                                 }
 381
 382                                 skb = nskb;
 383                         } while (skb);
 384
 385                         return len;
 386                 }
 387
 388 tag:
 389                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 390                 if (unlikely(!skb))
 391                         return 0;
 392                 vlan_set_tci(skb, 0);
 393         }
 394
 395         len = skb->len;
 396         dev_queue_xmit(skb);
 397
 398         return len;
 399
 400 error:
 401         kfree_skb(skb);
 402         vport_record_error(vport, VPORT_E_TX_DROPPED);
 403         return 0;
 404 }
 405
 406 /* Returns null if this device is not attached to a datapath. */
 407 struct vport *netdev_get_vport(struct net_device *dev)
 408 {
 409 #ifdef IFF_BRIDGE_PORT
 410 #if IFF_BRIDGE_PORT != IFF_OVS_DATAPATH
 411         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
 412 #else
 413         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
 414 #endif
 415                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
 416         else
 417                 return NULL;
 418 #else
 419         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
 420 #endif
 421 }
 422
 423 const struct vport_ops netdev_vport_ops = {
 424         .type           = OVS_VPORT_TYPE_NETDEV,
 425         .flags          = (VPORT_F_REQUIRED |
 426                           (USE_VPORT_STATS ? VPORT_F_GEN_STATS : 0)),
 427         .init           = netdev_init,
 428         .exit           = netdev_exit,
 429         .create         = netdev_create,
 430         .destroy        = netdev_destroy,
 431         .set_mtu        = netdev_set_mtu,
 432         .set_addr       = netdev_set_addr,
 433         .get_name       = netdev_get_name,
 434         .get_addr       = netdev_get_addr,
 435         .get_kobj       = netdev_get_kobj,
 436         .get_stats      = netdev_get_stats,
 437         .get_dev_flags  = netdev_get_dev_flags,
 438         .is_running     = netdev_is_running,
 439         .get_operstate  = netdev_get_operstate,
 440         .get_ifindex    = netdev_get_ifindex,
 441         .get_mtu        = netdev_get_mtu,
 442         .send           = netdev_send,
 443 };
 444
 445 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36)
 446 /*
 447  * In kernels earlier than 2.6.36, Open vSwitch cannot safely coexist with
 448  * the Linux bridge module on any released version of Linux, because there
 449  * is only a single bridge hook function and only a single br_port member
 450  * in struct net_device.
 451  *
 452  * Declaring and exporting this symbol enforces mutual exclusion.  The bridge
 453  * module also exports the same symbol, so the module loader will refuse to
 454  * load both modules at the same time (e.g. "bridge: exports duplicate symbol
 455  * br_should_route_hook (owned by openvswitch_mod)").
 456  *
 457  * The use of "typeof" here avoids the need to track changes in the type of
 458  * br_should_route_hook over various kernel versions.
 459  */
 460 typeof(br_should_route_hook) br_should_route_hook;
 461 EXPORT_SYMBOL(br_should_route_hook);
 462 #endif