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