datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2014 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 #include <linux/uaccess.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/etherdevice.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/if_vlan.h>
  24 #include <net/llc_pdu.h>
  25 #include <linux/kernel.h>
  26 #include <linux/jhash.h>
  27 #include <linux/jiffies.h>
  28 #include <linux/llc.h>
  29 #include <linux/module.h>
  30 #include <linux/in.h>
  31 #include <linux/rcupdate.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/ip.h>
  34 #include <linux/ipv6.h>
  35 #include <linux/mpls.h>
  36 #include <linux/sctp.h>
  37 #include <linux/smp.h>
  38 #include <linux/tcp.h>
  39 #include <linux/udp.h>
  40 #include <linux/icmp.h>
  41 #include <linux/icmpv6.h>
  42 #include <linux/rculist.h>
  43 #include <net/ip.h>
  44 #include <net/ipv6.h>
  45 #include <net/mpls.h>
  46 #include <net/ndisc.h>
  47
  48 #include "datapath.h"
  49 #include "flow.h"
  50 #include "flow_netlink.h"
  51
  52 #include "vlan.h"
  53
  54 u64 ovs_flow_used_time(unsigned long flow_jiffies)
  55 {
  56         struct timespec cur_ts;
  57         u64 cur_ms, idle_ms;
  58
  59         ktime_get_ts(&cur_ts);
  60         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  61         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
  62                  cur_ts.tv_nsec / NSEC_PER_MSEC;
  63
  64         return cur_ms - idle_ms;
  65 }
  66
  67 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
  68
  69 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
  70                            const struct sk_buff *skb)
  71 {
  72         struct flow_stats *stats;
  73         int node = numa_node_id();
  74         int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
  75
  76         stats = rcu_dereference(flow->stats[node]);
  77
  78         /* Check if already have node-specific stats. */
  79         if (likely(stats)) {
  80                 spin_lock(&stats->lock);
  81                 /* Mark if we write on the pre-allocated stats. */
  82                 if (node == 0 && unlikely(flow->stats_last_writer != node))
  83                         flow->stats_last_writer = node;
  84         } else {
  85                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
  86                 spin_lock(&stats->lock);
  87
  88                 /* If the current NUMA-node is the only writer on the
  89                  * pre-allocated stats keep using them.
  90                  */
  91                 if (unlikely(flow->stats_last_writer != node)) {
  92                         /* A previous locker may have already allocated the
  93                          * stats, so we need to check again.  If node-specific
  94                          * stats were already allocated, we update the pre-
  95                          * allocated stats as we have already locked them.
  96                          */
  97                         if (likely(flow->stats_last_writer != NUMA_NO_NODE)
  98                             && likely(!rcu_access_pointer(flow->stats[node]))) {
  99                                 /* Try to allocate node-specific stats. */
 100                                 struct flow_stats *new_stats;
 101
 102                                 new_stats =
 103                                         kmem_cache_alloc_node(flow_stats_cache,
 104                                                               GFP_THISNODE |
 105                                                               __GFP_NOMEMALLOC,
 106                                                               node);
 107                                 if (likely(new_stats)) {
 108                                         new_stats->used = jiffies;
 109                                         new_stats->packet_count = 1;
 110                                         new_stats->byte_count = len;
 111                                         new_stats->tcp_flags = tcp_flags;
 112                                         spin_lock_init(&new_stats->lock);
 113
 114                                         rcu_assign_pointer(flow->stats[node],
 115                                                            new_stats);
 116                                         goto unlock;
 117                                 }
 118                         }
 119                         flow->stats_last_writer = node;
 120                 }
 121         }
 122
 123         stats->used = jiffies;
 124         stats->packet_count++;
 125         stats->byte_count += len;
 126         stats->tcp_flags |= tcp_flags;
 127 unlock:
 128         spin_unlock(&stats->lock);
 129 }
 130
 131 /* Must be called with rcu_read_lock or ovs_mutex. */
 132 void ovs_flow_stats_get(const struct sw_flow *flow,
 133                         struct ovs_flow_stats *ovs_stats,
 134                         unsigned long *used, __be16 *tcp_flags)
 135 {
 136         int node;
 137
 138         *used = 0;
 139         *tcp_flags = 0;
 140         memset(ovs_stats, 0, sizeof(*ovs_stats));
 141
 142         for_each_node(node) {
 143                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
 144
 145                 if (stats) {
 146                         /* Local CPU may write on non-local stats, so we must
 147                          * block bottom-halves here.
 148                          */
 149                         spin_lock_bh(&stats->lock);
 150                         if (!*used || time_after(stats->used, *used))
 151                                 *used = stats->used;
 152                         *tcp_flags |= stats->tcp_flags;
 153                         ovs_stats->n_packets += stats->packet_count;
 154                         ovs_stats->n_bytes += stats->byte_count;
 155                         spin_unlock_bh(&stats->lock);
 156                 }
 157         }
 158 }
 159
 160 /* Called with ovs_mutex. */
 161 void ovs_flow_stats_clear(struct sw_flow *flow)
 162 {
 163         int node;
 164
 165         for_each_node(node) {
 166                 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
 167
 168                 if (stats) {
 169                         spin_lock_bh(&stats->lock);
 170                         stats->used = 0;
 171                         stats->packet_count = 0;
 172                         stats->byte_count = 0;
 173                         stats->tcp_flags = 0;
 174                         spin_unlock_bh(&stats->lock);
 175                 }
 176         }
 177 }
 178
 179 static int check_header(struct sk_buff *skb, int len)
 180 {
 181         if (unlikely(skb->len < len))
 182                 return -EINVAL;
 183         if (unlikely(!pskb_may_pull(skb, len)))
 184                 return -ENOMEM;
 185         return 0;
 186 }
 187
 188 static bool arphdr_ok(struct sk_buff *skb)
 189 {
 190         return pskb_may_pull(skb, skb_network_offset(skb) +
 191                                   sizeof(struct arp_eth_header));
 192 }
 193
 194 static int check_iphdr(struct sk_buff *skb)
 195 {
 196         unsigned int nh_ofs = skb_network_offset(skb);
 197         unsigned int ip_len;
 198         int err;
 199
 200         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 201         if (unlikely(err))
 202                 return err;
 203
 204         ip_len = ip_hdrlen(skb);
 205         if (unlikely(ip_len < sizeof(struct iphdr) ||
 206                      skb->len < nh_ofs + ip_len))
 207                 return -EINVAL;
 208
 209         skb_set_transport_header(skb, nh_ofs + ip_len);
 210         return 0;
 211 }
 212
 213 static bool tcphdr_ok(struct sk_buff *skb)
 214 {
 215         int th_ofs = skb_transport_offset(skb);
 216         int tcp_len;
 217
 218         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 219                 return false;
 220
 221         tcp_len = tcp_hdrlen(skb);
 222         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 223                      skb->len < th_ofs + tcp_len))
 224                 return false;
 225
 226         return true;
 227 }
 228
 229 static bool udphdr_ok(struct sk_buff *skb)
 230 {
 231         return pskb_may_pull(skb, skb_transport_offset(skb) +
 232                                   sizeof(struct udphdr));
 233 }
 234
 235 static bool sctphdr_ok(struct sk_buff *skb)
 236 {
 237         return pskb_may_pull(skb, skb_transport_offset(skb) +
 238                                   sizeof(struct sctphdr));
 239 }
 240
 241 static bool icmphdr_ok(struct sk_buff *skb)
 242 {
 243         return pskb_may_pull(skb, skb_transport_offset(skb) +
 244                                   sizeof(struct icmphdr));
 245 }
 246
 247 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 248 {
 249         unsigned int nh_ofs = skb_network_offset(skb);
 250         unsigned int nh_len;
 251         int payload_ofs;
 252         struct ipv6hdr *nh;
 253         uint8_t nexthdr;
 254         __be16 frag_off;
 255         int err;
 256
 257         err = check_header(skb, nh_ofs + sizeof(*nh));
 258         if (unlikely(err))
 259                 return err;
 260
 261         nh = ipv6_hdr(skb);
 262         nexthdr = nh->nexthdr;
 263         payload_ofs = (u8 *)(nh + 1) - skb->data;
 264
 265         key->ip.proto = NEXTHDR_NONE;
 266         key->ip.tos = ipv6_get_dsfield(nh);
 267         key->ip.ttl = nh->hop_limit;
 268         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 269         key->ipv6.addr.src = nh->saddr;
 270         key->ipv6.addr.dst = nh->daddr;
 271
 272         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 273         if (unlikely(payload_ofs < 0))
 274                 return -EINVAL;
 275
 276         if (frag_off) {
 277                 if (frag_off & htons(~0x7))
 278                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 279                 else
 280                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 281         } else {
 282                 key->ip.frag = OVS_FRAG_TYPE_NONE;
 283         }
 284
 285         nh_len = payload_ofs - nh_ofs;
 286         skb_set_transport_header(skb, nh_ofs + nh_len);
 287         key->ip.proto = nexthdr;
 288         return nh_len;
 289 }
 290
 291 static bool icmp6hdr_ok(struct sk_buff *skb)
 292 {
 293         return pskb_may_pull(skb, skb_transport_offset(skb) +
 294                                   sizeof(struct icmp6hdr));
 295 }
 296
 297 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 298 {
 299         struct qtag_prefix {
 300                 __be16 eth_type; /* ETH_P_8021Q */
 301                 __be16 tci;
 302         };
 303         struct qtag_prefix *qp;
 304
 305         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 306                 return 0;
 307
 308         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 309                                          sizeof(__be16))))
 310                 return -ENOMEM;
 311
 312         qp = (struct qtag_prefix *) skb->data;
 313         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 314         __skb_pull(skb, sizeof(struct qtag_prefix));
 315
 316         return 0;
 317 }
 318
 319 static __be16 parse_ethertype(struct sk_buff *skb)
 320 {
 321         struct llc_snap_hdr {
 322                 u8  dsap;  /* Always 0xAA */
 323                 u8  ssap;  /* Always 0xAA */
 324                 u8  ctrl;
 325                 u8  oui[3];
 326                 __be16 ethertype;
 327         };
 328         struct llc_snap_hdr *llc;
 329         __be16 proto;
 330
 331         proto = *(__be16 *) skb->data;
 332         __skb_pull(skb, sizeof(__be16));
 333
 334         if (eth_proto_is_802_3(proto))
 335                 return proto;
 336
 337         if (skb->len < sizeof(struct llc_snap_hdr))
 338                 return htons(ETH_P_802_2);
 339
 340         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 341                 return htons(0);
 342
 343         llc = (struct llc_snap_hdr *) skb->data;
 344         if (llc->dsap != LLC_SAP_SNAP ||
 345             llc->ssap != LLC_SAP_SNAP ||
 346             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 347                 return htons(ETH_P_802_2);
 348
 349         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 350
 351         if (eth_proto_is_802_3(llc->ethertype))
 352                 return llc->ethertype;
 353
 354         return htons(ETH_P_802_2);
 355 }
 356
 357 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 358                         int nh_len)
 359 {
 360         struct icmp6hdr *icmp = icmp6_hdr(skb);
 361
 362         /* The ICMPv6 type and code fields use the 16-bit transport port
 363          * fields, so we need to store them in 16-bit network byte order.
 364          */
 365         key->tp.src = htons(icmp->icmp6_type);
 366         key->tp.dst = htons(icmp->icmp6_code);
 367         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 368
 369         if (icmp->icmp6_code == 0 &&
 370             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 371              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 372                 int icmp_len = skb->len - skb_transport_offset(skb);
 373                 struct nd_msg *nd;
 374                 int offset;
 375
 376                 /* In order to process neighbor discovery options, we need the
 377                  * entire packet.
 378                  */
 379                 if (unlikely(icmp_len < sizeof(*nd)))
 380                         return 0;
 381
 382                 if (unlikely(skb_linearize(skb)))
 383                         return -ENOMEM;
 384
 385                 nd = (struct nd_msg *)skb_transport_header(skb);
 386                 key->ipv6.nd.target = nd->target;
 387
 388                 icmp_len -= sizeof(*nd);
 389                 offset = 0;
 390                 while (icmp_len >= 8) {
 391                         struct nd_opt_hdr *nd_opt =
 392                                  (struct nd_opt_hdr *)(nd->opt + offset);
 393                         int opt_len = nd_opt->nd_opt_len * 8;
 394
 395                         if (unlikely(!opt_len || opt_len > icmp_len))
 396                                 return 0;
 397
 398                         /* Store the link layer address if the appropriate
 399                          * option is provided.  It is considered an error if
 400                          * the same link layer option is specified twice.
 401                          */
 402                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 403                             && opt_len == 8) {
 404                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 405                                         goto invalid;
 406                                 ether_addr_copy(key->ipv6.nd.sll,
 407                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 408                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 409                                    && opt_len == 8) {
 410                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 411                                         goto invalid;
 412                                 ether_addr_copy(key->ipv6.nd.tll,
 413                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 414                         }
 415
 416                         icmp_len -= opt_len;
 417                         offset += opt_len;
 418                 }
 419         }
 420
 421         return 0;
 422
 423 invalid:
 424         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 425         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 426         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 427
 428         return 0;
 429 }
 430
 431 /**
 432  * key_extract - extracts a flow key from an Ethernet frame.
 433  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 434  * Ethernet header
 435  * @key: output flow key
 436  *
 437  * The caller must ensure that skb->len >= ETH_HLEN.
 438  *
 439  * Returns 0 if successful, otherwise a negative errno value.
 440  *
 441  * Initializes @skb header pointers as follows:
 442  *
 443  *    - skb->mac_header: the Ethernet header.
 444  *
 445  *    - skb->network_header: just past the Ethernet header, or just past the
 446  *      VLAN header, to the first byte of the Ethernet payload.
 447  *
 448  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 449  *      on output, then just past the IP header, if one is present and
 450  *      of a correct length, otherwise the same as skb->network_header.
 451  *      For other key->eth.type values it is left untouched.
 452  */
 453 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 454 {
 455         int error;
 456         struct ethhdr *eth;
 457
 458         /* Flags are always used as part of stats */
 459         key->tp.flags = 0;
 460
 461         skb_reset_mac_header(skb);
 462
 463         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 464          * header in the linear data area.
 465          */
 466         eth = eth_hdr(skb);
 467         ether_addr_copy(key->eth.src, eth->h_source);
 468         ether_addr_copy(key->eth.dst, eth->h_dest);
 469
 470         __skb_pull(skb, 2 * ETH_ALEN);
 471         /* We are going to push all headers that we pull, so no need to
 472          * update skb->csum here.
 473          */
 474
 475         key->eth.tci = 0;
 476         if (skb_vlan_tag_present(skb))
 477                 key->eth.tci = htons(vlan_get_tci(skb));
 478         else if (eth->h_proto == htons(ETH_P_8021Q))
 479                 if (unlikely(parse_vlan(skb, key)))
 480                         return -ENOMEM;
 481
 482         key->eth.type = parse_ethertype(skb);
 483         if (unlikely(key->eth.type == htons(0)))
 484                 return -ENOMEM;
 485
 486         skb_reset_network_header(skb);
 487         skb_reset_mac_len(skb);
 488         __skb_push(skb, skb->data - skb_mac_header(skb));
 489
 490         /* Network layer. */
 491         if (key->eth.type == htons(ETH_P_IP)) {
 492                 struct iphdr *nh;
 493                 __be16 offset;
 494
 495                 error = check_iphdr(skb);
 496                 if (unlikely(error)) {
 497                         memset(&key->ip, 0, sizeof(key->ip));
 498                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 499                         if (error == -EINVAL) {
 500                                 skb->transport_header = skb->network_header;
 501                                 error = 0;
 502                         }
 503                         return error;
 504                 }
 505
 506                 nh = ip_hdr(skb);
 507                 key->ipv4.addr.src = nh->saddr;
 508                 key->ipv4.addr.dst = nh->daddr;
 509
 510                 key->ip.proto = nh->protocol;
 511                 key->ip.tos = nh->tos;
 512                 key->ip.ttl = nh->ttl;
 513
 514                 offset = nh->frag_off & htons(IP_OFFSET);
 515                 if (offset) {
 516                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 517                         return 0;
 518                 }
 519                 if (nh->frag_off & htons(IP_MF) ||
 520                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 521                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 522                 else
 523                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 524
 525                 /* Transport layer. */
 526                 if (key->ip.proto == IPPROTO_TCP) {
 527                         if (tcphdr_ok(skb)) {
 528                                 struct tcphdr *tcp = tcp_hdr(skb);
 529                                 key->tp.src = tcp->source;
 530                                 key->tp.dst = tcp->dest;
 531                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 532                         } else {
 533                                 memset(&key->tp, 0, sizeof(key->tp));
 534                         }
 535
 536                 } else if (key->ip.proto == IPPROTO_UDP) {
 537                         if (udphdr_ok(skb)) {
 538                                 struct udphdr *udp = udp_hdr(skb);
 539                                 key->tp.src = udp->source;
 540                                 key->tp.dst = udp->dest;
 541                         } else {
 542                                 memset(&key->tp, 0, sizeof(key->tp));
 543                         }
 544                 } else if (key->ip.proto == IPPROTO_SCTP) {
 545                         if (sctphdr_ok(skb)) {
 546                                 struct sctphdr *sctp = sctp_hdr(skb);
 547                                 key->tp.src = sctp->source;
 548                                 key->tp.dst = sctp->dest;
 549                         } else {
 550                                 memset(&key->tp, 0, sizeof(key->tp));
 551                         }
 552                 } else if (key->ip.proto == IPPROTO_ICMP) {
 553                         if (icmphdr_ok(skb)) {
 554                                 struct icmphdr *icmp = icmp_hdr(skb);
 555                                 /* The ICMP type and code fields use the 16-bit
 556                                  * transport port fields, so we need to store
 557                                  * them in 16-bit network byte order.
 558                                  */
 559                                 key->tp.src = htons(icmp->type);
 560                                 key->tp.dst = htons(icmp->code);
 561                         } else {
 562                                 memset(&key->tp, 0, sizeof(key->tp));
 563                         }
 564                 }
 565
 566         } else if (key->eth.type == htons(ETH_P_ARP) ||
 567                    key->eth.type == htons(ETH_P_RARP)) {
 568                 struct arp_eth_header *arp;
 569                 bool arp_available = arphdr_ok(skb);
 570
 571                 arp = (struct arp_eth_header *)skb_network_header(skb);
 572
 573                 if (arp_available &&
 574                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 575                     arp->ar_pro == htons(ETH_P_IP) &&
 576                     arp->ar_hln == ETH_ALEN &&
 577                     arp->ar_pln == 4) {
 578
 579                         /* We only match on the lower 8 bits of the opcode. */
 580                         if (ntohs(arp->ar_op) <= 0xff)
 581                                 key->ip.proto = ntohs(arp->ar_op);
 582                         else
 583                                 key->ip.proto = 0;
 584
 585                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 586                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 587                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 588                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 589                 } else {
 590                         memset(&key->ip, 0, sizeof(key->ip));
 591                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 592                 }
 593         } else if (eth_p_mpls(key->eth.type)) {
 594                 size_t stack_len = MPLS_HLEN;
 595
 596                 /* In the presence of an MPLS label stack the end of the L2
 597                  * header and the beginning of the L3 header differ.
 598                  *
 599                  * Advance network_header to the beginning of the L3
 600                  * header. mac_len corresponds to the end of the L2 header.
 601                  */
 602                 while (1) {
 603                         __be32 lse;
 604
 605                         error = check_header(skb, skb->mac_len + stack_len);
 606                         if (unlikely(error))
 607                                 return 0;
 608
 609                         memcpy(&lse, skb_network_header(skb), MPLS_HLEN);
 610
 611                         if (stack_len == MPLS_HLEN)
 612                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
 613
 614                         skb_set_network_header(skb, skb->mac_len + stack_len);
 615                         if (lse & htonl(MPLS_LS_S_MASK))
 616                                 break;
 617
 618                         stack_len += MPLS_HLEN;
 619                 }
 620         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 621                 int nh_len;             /* IPv6 Header + Extensions */
 622
 623                 nh_len = parse_ipv6hdr(skb, key);
 624                 if (unlikely(nh_len < 0)) {
 625                         memset(&key->ip, 0, sizeof(key->ip));
 626                         memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 627                         if (nh_len == -EINVAL) {
 628                                 skb->transport_header = skb->network_header;
 629                                 error = 0;
 630                         } else {
 631                                 error = nh_len;
 632                         }
 633                         return error;
 634                 }
 635
 636                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 637                         return 0;
 638                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 639                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 640
 641                 /* Transport layer. */
 642                 if (key->ip.proto == NEXTHDR_TCP) {
 643                         if (tcphdr_ok(skb)) {
 644                                 struct tcphdr *tcp = tcp_hdr(skb);
 645                                 key->tp.src = tcp->source;
 646                                 key->tp.dst = tcp->dest;
 647                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 648                         } else {
 649                                 memset(&key->tp, 0, sizeof(key->tp));
 650                         }
 651                 } else if (key->ip.proto == NEXTHDR_UDP) {
 652                         if (udphdr_ok(skb)) {
 653                                 struct udphdr *udp = udp_hdr(skb);
 654                                 key->tp.src = udp->source;
 655                                 key->tp.dst = udp->dest;
 656                         } else {
 657                                 memset(&key->tp, 0, sizeof(key->tp));
 658                         }
 659                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 660                         if (sctphdr_ok(skb)) {
 661                                 struct sctphdr *sctp = sctp_hdr(skb);
 662                                 key->tp.src = sctp->source;
 663                                 key->tp.dst = sctp->dest;
 664                         } else {
 665                                 memset(&key->tp, 0, sizeof(key->tp));
 666                         }
 667                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 668                         if (icmp6hdr_ok(skb)) {
 669                                 error = parse_icmpv6(skb, key, nh_len);
 670                                 if (error)
 671                                         return error;
 672                         } else {
 673                                 memset(&key->tp, 0, sizeof(key->tp));
 674                         }
 675                 }
 676         }
 677         return 0;
 678 }
 679
 680 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 681 {
 682         return key_extract(skb, key);
 683 }
 684
 685 int ovs_flow_key_extract(const struct ovs_tunnel_info *tun_info,
 686                          struct sk_buff *skb, struct sw_flow_key *key)
 687 {
 688         /* Extract metadata from packet. */
 689         if (tun_info) {
 690                 memcpy(&key->tun_key, &tun_info->tunnel, sizeof(key->tun_key));
 691
 692                 BUILD_BUG_ON(((1 << (sizeof(tun_info->options_len) * 8)) - 1) >
 693                              sizeof(key->tun_opts));
 694
 695                 if (tun_info->options) {
 696                         memcpy(TUN_METADATA_OPTS(key, tun_info->options_len),
 697                                tun_info->options, tun_info->options_len);
 698                         key->tun_opts_len = tun_info->options_len;
 699                 } else {
 700                         key->tun_opts_len = 0;
 701                 }
 702         } else {
 703                 key->tun_opts_len = 0;
 704                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 705         }
 706
 707         key->phy.priority = skb->priority;
 708         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 709         key->phy.skb_mark = skb->mark;
 710         key->ovs_flow_hash = 0;
 711         key->recirc_id = 0;
 712
 713         return key_extract(skb, key);
 714 }
 715
 716 int ovs_flow_key_extract_userspace(const struct nlattr *attr,
 717                                    struct sk_buff *skb,
 718                                    struct sw_flow_key *key, bool log)
 719 {
 720         int err;
 721
 722         /* Extract metadata from netlink attributes. */
 723         err = ovs_nla_get_flow_metadata(attr, key, log);
 724         if (err)
 725                 return err;
 726
 727         return key_extract(skb, key);
 728 }