net/openvswitch/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/ip_tunnels.h>
  45 #include <net/ipv6.h>
  46 #include <net/mpls.h>
  47 #include <net/ndisc.h>
  48
  49 #include "datapath.h"
  50 #include "flow.h"
  51 #include "flow_netlink.h"
  52
  53 u64 ovs_flow_used_time(unsigned long flow_jiffies)
  54 {
  55         struct timespec cur_ts;
  56         u64 cur_ms, idle_ms;
  57
  58         ktime_get_ts(&cur_ts);
  59         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  60         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
  61                  cur_ts.tv_nsec / NSEC_PER_MSEC;
  62
  63         return cur_ms - idle_ms;
  64 }
  65
  66 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
  67
  68 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
  69                            const struct sk_buff *skb)
  70 {
  71         struct flow_stats *stats;
  72         int node = numa_node_id();
  73
  74         stats = rcu_dereference(flow->stats[node]);
  75
  76         /* Check if already have node-specific stats. */
  77         if (likely(stats)) {
  78                 spin_lock(&stats->lock);
  79                 /* Mark if we write on the pre-allocated stats. */
  80                 if (node == 0 && unlikely(flow->stats_last_writer != node))
  81                         flow->stats_last_writer = node;
  82         } else {
  83                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
  84                 spin_lock(&stats->lock);
  85
  86                 /* If the current NUMA-node is the only writer on the
  87                  * pre-allocated stats keep using them.
  88                  */
  89                 if (unlikely(flow->stats_last_writer != node)) {
  90                         /* A previous locker may have already allocated the
  91                          * stats, so we need to check again.  If node-specific
  92                          * stats were already allocated, we update the pre-
  93                          * allocated stats as we have already locked them.
  94                          */
  95                         if (likely(flow->stats_last_writer != NUMA_NO_NODE)
  96                             && likely(!rcu_access_pointer(flow->stats[node]))) {
  97                                 /* Try to allocate node-specific stats. */
  98                                 struct flow_stats *new_stats;
  99
 100                                 new_stats =
 101                                         kmem_cache_alloc_node(flow_stats_cache,
 102                                                               GFP_THISNODE |
 103                                                               __GFP_NOMEMALLOC,
 104                                                               node);
 105                                 if (likely(new_stats)) {
 106                                         new_stats->used = jiffies;
 107                                         new_stats->packet_count = 1;
 108                                         new_stats->byte_count = skb->len;
 109                                         new_stats->tcp_flags = tcp_flags;
 110                                         spin_lock_init(&new_stats->lock);
 111
 112                                         rcu_assign_pointer(flow->stats[node],
 113                                                            new_stats);
 114                                         goto unlock;
 115                                 }
 116                         }
 117                         flow->stats_last_writer = node;
 118                 }
 119         }
 120
 121         stats->used = jiffies;
 122         stats->packet_count++;
 123         stats->byte_count += skb->len;
 124         stats->tcp_flags |= tcp_flags;
 125 unlock:
 126         spin_unlock(&stats->lock);
 127 }
 128
 129 /* Must be called with rcu_read_lock or ovs_mutex. */
 130 void ovs_flow_stats_get(const struct sw_flow *flow,
 131                         struct ovs_flow_stats *ovs_stats,
 132                         unsigned long *used, __be16 *tcp_flags)
 133 {
 134         int node;
 135
 136         *used = 0;
 137         *tcp_flags = 0;
 138         memset(ovs_stats, 0, sizeof(*ovs_stats));
 139
 140         for_each_node(node) {
 141                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
 142
 143                 if (stats) {
 144                         /* Local CPU may write on non-local stats, so we must
 145                          * block bottom-halves here.
 146                          */
 147                         spin_lock_bh(&stats->lock);
 148                         if (!*used || time_after(stats->used, *used))
 149                                 *used = stats->used;
 150                         *tcp_flags |= stats->tcp_flags;
 151                         ovs_stats->n_packets += stats->packet_count;
 152                         ovs_stats->n_bytes += stats->byte_count;
 153                         spin_unlock_bh(&stats->lock);
 154                 }
 155         }
 156 }
 157
 158 /* Called with ovs_mutex. */
 159 void ovs_flow_stats_clear(struct sw_flow *flow)
 160 {
 161         int node;
 162
 163         for_each_node(node) {
 164                 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
 165
 166                 if (stats) {
 167                         spin_lock_bh(&stats->lock);
 168                         stats->used = 0;
 169                         stats->packet_count = 0;
 170                         stats->byte_count = 0;
 171                         stats->tcp_flags = 0;
 172                         spin_unlock_bh(&stats->lock);
 173                 }
 174         }
 175 }
 176
 177 static int check_header(struct sk_buff *skb, int len)
 178 {
 179         if (unlikely(skb->len < len))
 180                 return -EINVAL;
 181         if (unlikely(!pskb_may_pull(skb, len)))
 182                 return -ENOMEM;
 183         return 0;
 184 }
 185
 186 static bool arphdr_ok(struct sk_buff *skb)
 187 {
 188         return pskb_may_pull(skb, skb_network_offset(skb) +
 189                                   sizeof(struct arp_eth_header));
 190 }
 191
 192 static int check_iphdr(struct sk_buff *skb)
 193 {
 194         unsigned int nh_ofs = skb_network_offset(skb);
 195         unsigned int ip_len;
 196         int err;
 197
 198         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 199         if (unlikely(err))
 200                 return err;
 201
 202         ip_len = ip_hdrlen(skb);
 203         if (unlikely(ip_len < sizeof(struct iphdr) ||
 204                      skb->len < nh_ofs + ip_len))
 205                 return -EINVAL;
 206
 207         skb_set_transport_header(skb, nh_ofs + ip_len);
 208         return 0;
 209 }
 210
 211 static bool tcphdr_ok(struct sk_buff *skb)
 212 {
 213         int th_ofs = skb_transport_offset(skb);
 214         int tcp_len;
 215
 216         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 217                 return false;
 218
 219         tcp_len = tcp_hdrlen(skb);
 220         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 221                      skb->len < th_ofs + tcp_len))
 222                 return false;
 223
 224         return true;
 225 }
 226
 227 static bool udphdr_ok(struct sk_buff *skb)
 228 {
 229         return pskb_may_pull(skb, skb_transport_offset(skb) +
 230                                   sizeof(struct udphdr));
 231 }
 232
 233 static bool sctphdr_ok(struct sk_buff *skb)
 234 {
 235         return pskb_may_pull(skb, skb_transport_offset(skb) +
 236                                   sizeof(struct sctphdr));
 237 }
 238
 239 static bool icmphdr_ok(struct sk_buff *skb)
 240 {
 241         return pskb_may_pull(skb, skb_transport_offset(skb) +
 242                                   sizeof(struct icmphdr));
 243 }
 244
 245 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 246 {
 247         unsigned int nh_ofs = skb_network_offset(skb);
 248         unsigned int nh_len;
 249         int payload_ofs;
 250         struct ipv6hdr *nh;
 251         uint8_t nexthdr;
 252         __be16 frag_off;
 253         int err;
 254
 255         err = check_header(skb, nh_ofs + sizeof(*nh));
 256         if (unlikely(err))
 257                 return err;
 258
 259         nh = ipv6_hdr(skb);
 260         nexthdr = nh->nexthdr;
 261         payload_ofs = (u8 *)(nh + 1) - skb->data;
 262
 263         key->ip.proto = NEXTHDR_NONE;
 264         key->ip.tos = ipv6_get_dsfield(nh);
 265         key->ip.ttl = nh->hop_limit;
 266         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 267         key->ipv6.addr.src = nh->saddr;
 268         key->ipv6.addr.dst = nh->daddr;
 269
 270         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 271         if (unlikely(payload_ofs < 0))
 272                 return -EINVAL;
 273
 274         if (frag_off) {
 275                 if (frag_off & htons(~0x7))
 276                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 277                 else
 278                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 279         } else {
 280                 key->ip.frag = OVS_FRAG_TYPE_NONE;
 281         }
 282
 283         nh_len = payload_ofs - nh_ofs;
 284         skb_set_transport_header(skb, nh_ofs + nh_len);
 285         key->ip.proto = nexthdr;
 286         return nh_len;
 287 }
 288
 289 static bool icmp6hdr_ok(struct sk_buff *skb)
 290 {
 291         return pskb_may_pull(skb, skb_transport_offset(skb) +
 292                                   sizeof(struct icmp6hdr));
 293 }
 294
 295 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 296 {
 297         struct qtag_prefix {
 298                 __be16 eth_type; /* ETH_P_8021Q */
 299                 __be16 tci;
 300         };
 301         struct qtag_prefix *qp;
 302
 303         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 304                 return 0;
 305
 306         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 307                                          sizeof(__be16))))
 308                 return -ENOMEM;
 309
 310         qp = (struct qtag_prefix *) skb->data;
 311         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 312         __skb_pull(skb, sizeof(struct qtag_prefix));
 313
 314         return 0;
 315 }
 316
 317 static __be16 parse_ethertype(struct sk_buff *skb)
 318 {
 319         struct llc_snap_hdr {
 320                 u8  dsap;  /* Always 0xAA */
 321                 u8  ssap;  /* Always 0xAA */
 322                 u8  ctrl;
 323                 u8  oui[3];
 324                 __be16 ethertype;
 325         };
 326         struct llc_snap_hdr *llc;
 327         __be16 proto;
 328
 329         proto = *(__be16 *) skb->data;
 330         __skb_pull(skb, sizeof(__be16));
 331
 332         if (ntohs(proto) >= ETH_P_802_3_MIN)
 333                 return proto;
 334
 335         if (skb->len < sizeof(struct llc_snap_hdr))
 336                 return htons(ETH_P_802_2);
 337
 338         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 339                 return htons(0);
 340
 341         llc = (struct llc_snap_hdr *) skb->data;
 342         if (llc->dsap != LLC_SAP_SNAP ||
 343             llc->ssap != LLC_SAP_SNAP ||
 344             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 345                 return htons(ETH_P_802_2);
 346
 347         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 348
 349         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 350                 return llc->ethertype;
 351
 352         return htons(ETH_P_802_2);
 353 }
 354
 355 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 356                         int nh_len)
 357 {
 358         struct icmp6hdr *icmp = icmp6_hdr(skb);
 359
 360         /* The ICMPv6 type and code fields use the 16-bit transport port
 361          * fields, so we need to store them in 16-bit network byte order.
 362          */
 363         key->tp.src = htons(icmp->icmp6_type);
 364         key->tp.dst = htons(icmp->icmp6_code);
 365         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 366
 367         if (icmp->icmp6_code == 0 &&
 368             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 369              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 370                 int icmp_len = skb->len - skb_transport_offset(skb);
 371                 struct nd_msg *nd;
 372                 int offset;
 373
 374                 /* In order to process neighbor discovery options, we need the
 375                  * entire packet.
 376                  */
 377                 if (unlikely(icmp_len < sizeof(*nd)))
 378                         return 0;
 379
 380                 if (unlikely(skb_linearize(skb)))
 381                         return -ENOMEM;
 382
 383                 nd = (struct nd_msg *)skb_transport_header(skb);
 384                 key->ipv6.nd.target = nd->target;
 385
 386                 icmp_len -= sizeof(*nd);
 387                 offset = 0;
 388                 while (icmp_len >= 8) {
 389                         struct nd_opt_hdr *nd_opt =
 390                                  (struct nd_opt_hdr *)(nd->opt + offset);
 391                         int opt_len = nd_opt->nd_opt_len * 8;
 392
 393                         if (unlikely(!opt_len || opt_len > icmp_len))
 394                                 return 0;
 395
 396                         /* Store the link layer address if the appropriate
 397                          * option is provided.  It is considered an error if
 398                          * the same link layer option is specified twice.
 399                          */
 400                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 401                             && opt_len == 8) {
 402                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 403                                         goto invalid;
 404                                 ether_addr_copy(key->ipv6.nd.sll,
 405                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 406                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 407                                    && opt_len == 8) {
 408                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 409                                         goto invalid;
 410                                 ether_addr_copy(key->ipv6.nd.tll,
 411                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 412                         }
 413
 414                         icmp_len -= opt_len;
 415                         offset += opt_len;
 416                 }
 417         }
 418
 419         return 0;
 420
 421 invalid:
 422         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 423         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 424         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 425
 426         return 0;
 427 }
 428
 429 /**
 430  * key_extract - extracts a flow key from an Ethernet frame.
 431  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 432  * Ethernet header
 433  * @key: output flow key
 434  *
 435  * The caller must ensure that skb->len >= ETH_HLEN.
 436  *
 437  * Returns 0 if successful, otherwise a negative errno value.
 438  *
 439  * Initializes @skb header pointers as follows:
 440  *
 441  *    - skb->mac_header: the Ethernet header.
 442  *
 443  *    - skb->network_header: just past the Ethernet header, or just past the
 444  *      VLAN header, to the first byte of the Ethernet payload.
 445  *
 446  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 447  *      on output, then just past the IP header, if one is present and
 448  *      of a correct length, otherwise the same as skb->network_header.
 449  *      For other key->eth.type values it is left untouched.
 450  */
 451 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 452 {
 453         int error;
 454         struct ethhdr *eth;
 455
 456         /* Flags are always used as part of stats */
 457         key->tp.flags = 0;
 458
 459         skb_reset_mac_header(skb);
 460
 461         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 462          * header in the linear data area.
 463          */
 464         eth = eth_hdr(skb);
 465         ether_addr_copy(key->eth.src, eth->h_source);
 466         ether_addr_copy(key->eth.dst, eth->h_dest);
 467
 468         __skb_pull(skb, 2 * ETH_ALEN);
 469         /* We are going to push all headers that we pull, so no need to
 470          * update skb->csum here.
 471          */
 472
 473         key->eth.tci = 0;
 474         if (vlan_tx_tag_present(skb))
 475                 key->eth.tci = htons(skb->vlan_tci);
 476         else if (eth->h_proto == htons(ETH_P_8021Q))
 477                 if (unlikely(parse_vlan(skb, key)))
 478                         return -ENOMEM;
 479
 480         key->eth.type = parse_ethertype(skb);
 481         if (unlikely(key->eth.type == htons(0)))
 482                 return -ENOMEM;
 483
 484         skb_reset_network_header(skb);
 485         skb_reset_mac_len(skb);
 486         __skb_push(skb, skb->data - skb_mac_header(skb));
 487
 488         /* Network layer. */
 489         if (key->eth.type == htons(ETH_P_IP)) {
 490                 struct iphdr *nh;
 491                 __be16 offset;
 492
 493                 error = check_iphdr(skb);
 494                 if (unlikely(error)) {
 495                         memset(&key->ip, 0, sizeof(key->ip));
 496                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 497                         if (error == -EINVAL) {
 498                                 skb->transport_header = skb->network_header;
 499                                 error = 0;
 500                         }
 501                         return error;
 502                 }
 503
 504                 nh = ip_hdr(skb);
 505                 key->ipv4.addr.src = nh->saddr;
 506                 key->ipv4.addr.dst = nh->daddr;
 507
 508                 key->ip.proto = nh->protocol;
 509                 key->ip.tos = nh->tos;
 510                 key->ip.ttl = nh->ttl;
 511
 512                 offset = nh->frag_off & htons(IP_OFFSET);
 513                 if (offset) {
 514                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 515                         return 0;
 516                 }
 517                 if (nh->frag_off & htons(IP_MF) ||
 518                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 519                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 520                 else
 521                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 522
 523                 /* Transport layer. */
 524                 if (key->ip.proto == IPPROTO_TCP) {
 525                         if (tcphdr_ok(skb)) {
 526                                 struct tcphdr *tcp = tcp_hdr(skb);
 527                                 key->tp.src = tcp->source;
 528                                 key->tp.dst = tcp->dest;
 529                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 530                         } else {
 531                                 memset(&key->tp, 0, sizeof(key->tp));
 532                         }
 533
 534                 } else if (key->ip.proto == IPPROTO_UDP) {
 535                         if (udphdr_ok(skb)) {
 536                                 struct udphdr *udp = udp_hdr(skb);
 537                                 key->tp.src = udp->source;
 538                                 key->tp.dst = udp->dest;
 539                         } else {
 540                                 memset(&key->tp, 0, sizeof(key->tp));
 541                         }
 542                 } else if (key->ip.proto == IPPROTO_SCTP) {
 543                         if (sctphdr_ok(skb)) {
 544                                 struct sctphdr *sctp = sctp_hdr(skb);
 545                                 key->tp.src = sctp->source;
 546                                 key->tp.dst = sctp->dest;
 547                         } else {
 548                                 memset(&key->tp, 0, sizeof(key->tp));
 549                         }
 550                 } else if (key->ip.proto == IPPROTO_ICMP) {
 551                         if (icmphdr_ok(skb)) {
 552                                 struct icmphdr *icmp = icmp_hdr(skb);
 553                                 /* The ICMP type and code fields use the 16-bit
 554                                  * transport port fields, so we need to store
 555                                  * them in 16-bit network byte order. */
 556                                 key->tp.src = htons(icmp->type);
 557                                 key->tp.dst = htons(icmp->code);
 558                         } else {
 559                                 memset(&key->tp, 0, sizeof(key->tp));
 560                         }
 561                 }
 562
 563         } else if (key->eth.type == htons(ETH_P_ARP) ||
 564                    key->eth.type == htons(ETH_P_RARP)) {
 565                 struct arp_eth_header *arp;
 566                 bool arp_available = arphdr_ok(skb);
 567
 568                 arp = (struct arp_eth_header *)skb_network_header(skb);
 569
 570                 if (arp_available &&
 571                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 572                     arp->ar_pro == htons(ETH_P_IP) &&
 573                     arp->ar_hln == ETH_ALEN &&
 574                     arp->ar_pln == 4) {
 575
 576                         /* We only match on the lower 8 bits of the opcode. */
 577                         if (ntohs(arp->ar_op) <= 0xff)
 578                                 key->ip.proto = ntohs(arp->ar_op);
 579                         else
 580                                 key->ip.proto = 0;
 581
 582                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 583                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 584                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 585                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 586                 } else {
 587                         memset(&key->ip, 0, sizeof(key->ip));
 588                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 589                 }
 590         } else if (eth_p_mpls(key->eth.type)) {
 591                 size_t stack_len = MPLS_HLEN;
 592
 593                 /* In the presence of an MPLS label stack the end of the L2
 594                  * header and the beginning of the L3 header differ.
 595                  *
 596                  * Advance network_header to the beginning of the L3
 597                  * header. mac_len corresponds to the end of the L2 header.
 598                  */
 599                 while (1) {
 600                         __be32 lse;
 601
 602                         error = check_header(skb, skb->mac_len + stack_len);
 603                         if (unlikely(error))
 604                                 return 0;
 605
 606                         memcpy(&lse, skb_network_header(skb), MPLS_HLEN);
 607
 608                         if (stack_len == MPLS_HLEN)
 609                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
 610
 611                         skb_set_network_header(skb, skb->mac_len + stack_len);
 612                         if (lse & htonl(MPLS_LS_S_MASK))
 613                                 break;
 614
 615                         stack_len += MPLS_HLEN;
 616                 }
 617         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 618                 int nh_len;             /* IPv6 Header + Extensions */
 619
 620                 nh_len = parse_ipv6hdr(skb, key);
 621                 if (unlikely(nh_len < 0)) {
 622                         memset(&key->ip, 0, sizeof(key->ip));
 623                         memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 624                         if (nh_len == -EINVAL) {
 625                                 skb->transport_header = skb->network_header;
 626                                 error = 0;
 627                         } else {
 628                                 error = nh_len;
 629                         }
 630                         return error;
 631                 }
 632
 633                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 634                         return 0;
 635                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 636                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 637
 638                 /* Transport layer. */
 639                 if (key->ip.proto == NEXTHDR_TCP) {
 640                         if (tcphdr_ok(skb)) {
 641                                 struct tcphdr *tcp = tcp_hdr(skb);
 642                                 key->tp.src = tcp->source;
 643                                 key->tp.dst = tcp->dest;
 644                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 645                         } else {
 646                                 memset(&key->tp, 0, sizeof(key->tp));
 647                         }
 648                 } else if (key->ip.proto == NEXTHDR_UDP) {
 649                         if (udphdr_ok(skb)) {
 650                                 struct udphdr *udp = udp_hdr(skb);
 651                                 key->tp.src = udp->source;
 652                                 key->tp.dst = udp->dest;
 653                         } else {
 654                                 memset(&key->tp, 0, sizeof(key->tp));
 655                         }
 656                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 657                         if (sctphdr_ok(skb)) {
 658                                 struct sctphdr *sctp = sctp_hdr(skb);
 659                                 key->tp.src = sctp->source;
 660                                 key->tp.dst = sctp->dest;
 661                         } else {
 662                                 memset(&key->tp, 0, sizeof(key->tp));
 663                         }
 664                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 665                         if (icmp6hdr_ok(skb)) {
 666                                 error = parse_icmpv6(skb, key, nh_len);
 667                                 if (error)
 668                                         return error;
 669                         } else {
 670                                 memset(&key->tp, 0, sizeof(key->tp));
 671                         }
 672                 }
 673         }
 674         return 0;
 675 }
 676
 677 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 678 {
 679         return key_extract(skb, key);
 680 }
 681
 682 int ovs_flow_key_extract(const struct ovs_tunnel_info *tun_info,
 683                          struct sk_buff *skb, struct sw_flow_key *key)
 684 {
 685         /* Extract metadata from packet. */
 686         if (tun_info) {
 687                 memcpy(&key->tun_key, &tun_info->tunnel, sizeof(key->tun_key));
 688
 689                 if (tun_info->options) {
 690                         BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
 691                                                    8)) - 1
 692                                         > sizeof(key->tun_opts));
 693                         memcpy(GENEVE_OPTS(key, tun_info->options_len),
 694                                tun_info->options, tun_info->options_len);
 695                         key->tun_opts_len = tun_info->options_len;
 696                 } else {
 697                         key->tun_opts_len = 0;
 698                 }
 699         } else  {
 700                 key->tun_opts_len = 0;
 701                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 702         }
 703
 704         key->phy.priority = skb->priority;
 705         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 706         key->phy.skb_mark = skb->mark;
 707         key->ovs_flow_hash = 0;
 708         key->recirc_id = 0;
 709
 710         return key_extract(skb, key);
 711 }
 712
 713 int ovs_flow_key_extract_userspace(const struct nlattr *attr,
 714                                    struct sk_buff *skb,
 715                                    struct sw_flow_key *key, bool log)
 716 {
 717         int err;
 718
 719         /* Extract metadata from netlink attributes. */
 720         err = ovs_nla_get_flow_metadata(attr, key, log);
 721         if (err)
 722                 return err;
 723
 724         return key_extract(skb, key);
 725 }