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 "conntrack.h"
  50 #include "datapath.h"
  51 #include "flow.h"
  52 #include "flow_netlink.h"
  53 #include "vport.h"
  54
  55 u64 ovs_flow_used_time(unsigned long flow_jiffies)
  56 {
  57         struct timespec cur_ts;
  58         u64 cur_ms, idle_ms;
  59
  60         ktime_get_ts(&cur_ts);
  61         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  62         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
  63                  cur_ts.tv_nsec / NSEC_PER_MSEC;
  64
  65         return cur_ms - idle_ms;
  66 }
  67
  68 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
  69
  70 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
  71                            const struct sk_buff *skb)
  72 {
  73         struct flow_stats *stats;
  74         int node = numa_mem_id();
  75         int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
  76
  77         stats = rcu_dereference(flow->stats[node]);
  78
  79         /* Check if already have node-specific stats. */
  80         if (likely(stats)) {
  81                 spin_lock(&stats->lock);
  82                 /* Mark if we write on the pre-allocated stats. */
  83                 if (node == 0 && unlikely(flow->stats_last_writer != node))
  84                         flow->stats_last_writer = node;
  85         } else {
  86                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
  87                 spin_lock(&stats->lock);
  88
  89                 /* If the current NUMA-node is the only writer on the
  90                  * pre-allocated stats keep using them.
  91                  */
  92                 if (unlikely(flow->stats_last_writer != node)) {
  93                         /* A previous locker may have already allocated the
  94                          * stats, so we need to check again.  If node-specific
  95                          * stats were already allocated, we update the pre-
  96                          * allocated stats as we have already locked them.
  97                          */
  98                         if (likely(flow->stats_last_writer != NUMA_NO_NODE)
  99                             && likely(!rcu_access_pointer(flow->stats[node]))) {
 100                                 /* Try to allocate node-specific stats. */
 101                                 struct flow_stats *new_stats;
 102
 103                                 new_stats =
 104                                         kmem_cache_alloc_node(flow_stats_cache,
 105                                                               GFP_NOWAIT |
 106                                                               __GFP_THISNODE |
 107                                                               __GFP_NOWARN |
 108                                                               __GFP_NOMEMALLOC,
 109                                                               node);
 110                                 if (likely(new_stats)) {
 111                                         new_stats->used = jiffies;
 112                                         new_stats->packet_count = 1;
 113                                         new_stats->byte_count = len;
 114                                         new_stats->tcp_flags = tcp_flags;
 115                                         spin_lock_init(&new_stats->lock);
 116
 117                                         rcu_assign_pointer(flow->stats[node],
 118                                                            new_stats);
 119                                         goto unlock;
 120                                 }
 121                         }
 122                         flow->stats_last_writer = node;
 123                 }
 124         }
 125
 126         stats->used = jiffies;
 127         stats->packet_count++;
 128         stats->byte_count += len;
 129         stats->tcp_flags |= tcp_flags;
 130 unlock:
 131         spin_unlock(&stats->lock);
 132 }
 133
 134 /* Must be called with rcu_read_lock or ovs_mutex. */
 135 void ovs_flow_stats_get(const struct sw_flow *flow,
 136                         struct ovs_flow_stats *ovs_stats,
 137                         unsigned long *used, __be16 *tcp_flags)
 138 {
 139         int node;
 140
 141         *used = 0;
 142         *tcp_flags = 0;
 143         memset(ovs_stats, 0, sizeof(*ovs_stats));
 144
 145         for_each_node(node) {
 146                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
 147
 148                 if (stats) {
 149                         /* Local CPU may write on non-local stats, so we must
 150                          * block bottom-halves here.
 151                          */
 152                         spin_lock_bh(&stats->lock);
 153                         if (!*used || time_after(stats->used, *used))
 154                                 *used = stats->used;
 155                         *tcp_flags |= stats->tcp_flags;
 156                         ovs_stats->n_packets += stats->packet_count;
 157                         ovs_stats->n_bytes += stats->byte_count;
 158                         spin_unlock_bh(&stats->lock);
 159                 }
 160         }
 161 }
 162
 163 /* Called with ovs_mutex. */
 164 void ovs_flow_stats_clear(struct sw_flow *flow)
 165 {
 166         int node;
 167
 168         for_each_node(node) {
 169                 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
 170
 171                 if (stats) {
 172                         spin_lock_bh(&stats->lock);
 173                         stats->used = 0;
 174                         stats->packet_count = 0;
 175                         stats->byte_count = 0;
 176                         stats->tcp_flags = 0;
 177                         spin_unlock_bh(&stats->lock);
 178                 }
 179         }
 180 }
 181
 182 static int check_header(struct sk_buff *skb, int len)
 183 {
 184         if (unlikely(skb->len < len))
 185                 return -EINVAL;
 186         if (unlikely(!pskb_may_pull(skb, len)))
 187                 return -ENOMEM;
 188         return 0;
 189 }
 190
 191 static bool arphdr_ok(struct sk_buff *skb)
 192 {
 193         return pskb_may_pull(skb, skb_network_offset(skb) +
 194                                   sizeof(struct arp_eth_header));
 195 }
 196
 197 static int check_iphdr(struct sk_buff *skb)
 198 {
 199         unsigned int nh_ofs = skb_network_offset(skb);
 200         unsigned int ip_len;
 201         int err;
 202
 203         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 204         if (unlikely(err))
 205                 return err;
 206
 207         ip_len = ip_hdrlen(skb);
 208         if (unlikely(ip_len < sizeof(struct iphdr) ||
 209                      skb->len < nh_ofs + ip_len))
 210                 return -EINVAL;
 211
 212         skb_set_transport_header(skb, nh_ofs + ip_len);
 213         return 0;
 214 }
 215
 216 static bool tcphdr_ok(struct sk_buff *skb)
 217 {
 218         int th_ofs = skb_transport_offset(skb);
 219         int tcp_len;
 220
 221         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 222                 return false;
 223
 224         tcp_len = tcp_hdrlen(skb);
 225         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 226                      skb->len < th_ofs + tcp_len))
 227                 return false;
 228
 229         return true;
 230 }
 231
 232 static bool udphdr_ok(struct sk_buff *skb)
 233 {
 234         return pskb_may_pull(skb, skb_transport_offset(skb) +
 235                                   sizeof(struct udphdr));
 236 }
 237
 238 static bool sctphdr_ok(struct sk_buff *skb)
 239 {
 240         return pskb_may_pull(skb, skb_transport_offset(skb) +
 241                                   sizeof(struct sctphdr));
 242 }
 243
 244 static bool icmphdr_ok(struct sk_buff *skb)
 245 {
 246         return pskb_may_pull(skb, skb_transport_offset(skb) +
 247                                   sizeof(struct icmphdr));
 248 }
 249
 250 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 251 {
 252         unsigned int nh_ofs = skb_network_offset(skb);
 253         unsigned int nh_len;
 254         int payload_ofs;
 255         struct ipv6hdr *nh;
 256         uint8_t nexthdr;
 257         __be16 frag_off;
 258         int err;
 259
 260         err = check_header(skb, nh_ofs + sizeof(*nh));
 261         if (unlikely(err))
 262                 return err;
 263
 264         nh = ipv6_hdr(skb);
 265         nexthdr = nh->nexthdr;
 266         payload_ofs = (u8 *)(nh + 1) - skb->data;
 267
 268         key->ip.proto = NEXTHDR_NONE;
 269         key->ip.tos = ipv6_get_dsfield(nh);
 270         key->ip.ttl = nh->hop_limit;
 271         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 272         key->ipv6.addr.src = nh->saddr;
 273         key->ipv6.addr.dst = nh->daddr;
 274
 275         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 276
 277         if (frag_off) {
 278                 if (frag_off & htons(~0x7))
 279                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 280                 else
 281                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 282         } else {
 283                 key->ip.frag = OVS_FRAG_TYPE_NONE;
 284         }
 285
 286         /* Delayed handling of error in ipv6_skip_exthdr() as it
 287          * always sets frag_off to a valid value which may be
 288          * used to set key->ip.frag above.
 289          */
 290         if (unlikely(payload_ofs < 0))
 291                 return -EPROTO;
 292
 293         nh_len = payload_ofs - nh_ofs;
 294         skb_set_transport_header(skb, nh_ofs + nh_len);
 295         key->ip.proto = nexthdr;
 296         return nh_len;
 297 }
 298
 299 static bool icmp6hdr_ok(struct sk_buff *skb)
 300 {
 301         return pskb_may_pull(skb, skb_transport_offset(skb) +
 302                                   sizeof(struct icmp6hdr));
 303 }
 304
 305 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 306 {
 307         struct qtag_prefix {
 308                 __be16 eth_type; /* ETH_P_8021Q */
 309                 __be16 tci;
 310         };
 311         struct qtag_prefix *qp;
 312
 313         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 314                 return 0;
 315
 316         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 317                                          sizeof(__be16))))
 318                 return -ENOMEM;
 319
 320         qp = (struct qtag_prefix *) skb->data;
 321         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 322         __skb_pull(skb, sizeof(struct qtag_prefix));
 323
 324         return 0;
 325 }
 326
 327 static __be16 parse_ethertype(struct sk_buff *skb)
 328 {
 329         struct llc_snap_hdr {
 330                 u8  dsap;  /* Always 0xAA */
 331                 u8  ssap;  /* Always 0xAA */
 332                 u8  ctrl;
 333                 u8  oui[3];
 334                 __be16 ethertype;
 335         };
 336         struct llc_snap_hdr *llc;
 337         __be16 proto;
 338
 339         proto = *(__be16 *) skb->data;
 340         __skb_pull(skb, sizeof(__be16));
 341
 342         if (eth_proto_is_802_3(proto))
 343                 return proto;
 344
 345         if (skb->len < sizeof(struct llc_snap_hdr))
 346                 return htons(ETH_P_802_2);
 347
 348         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 349                 return htons(0);
 350
 351         llc = (struct llc_snap_hdr *) skb->data;
 352         if (llc->dsap != LLC_SAP_SNAP ||
 353             llc->ssap != LLC_SAP_SNAP ||
 354             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 355                 return htons(ETH_P_802_2);
 356
 357         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 358
 359         if (eth_proto_is_802_3(llc->ethertype))
 360                 return llc->ethertype;
 361
 362         return htons(ETH_P_802_2);
 363 }
 364
 365 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 366                         int nh_len)
 367 {
 368         struct icmp6hdr *icmp = icmp6_hdr(skb);
 369
 370         /* The ICMPv6 type and code fields use the 16-bit transport port
 371          * fields, so we need to store them in 16-bit network byte order.
 372          */
 373         key->tp.src = htons(icmp->icmp6_type);
 374         key->tp.dst = htons(icmp->icmp6_code);
 375         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 376
 377         if (icmp->icmp6_code == 0 &&
 378             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 379              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 380                 int icmp_len = skb->len - skb_transport_offset(skb);
 381                 struct nd_msg *nd;
 382                 int offset;
 383
 384                 /* In order to process neighbor discovery options, we need the
 385                  * entire packet.
 386                  */
 387                 if (unlikely(icmp_len < sizeof(*nd)))
 388                         return 0;
 389
 390                 if (unlikely(skb_linearize(skb)))
 391                         return -ENOMEM;
 392
 393                 nd = (struct nd_msg *)skb_transport_header(skb);
 394                 key->ipv6.nd.target = nd->target;
 395
 396                 icmp_len -= sizeof(*nd);
 397                 offset = 0;
 398                 while (icmp_len >= 8) {
 399                         struct nd_opt_hdr *nd_opt =
 400                                  (struct nd_opt_hdr *)(nd->opt + offset);
 401                         int opt_len = nd_opt->nd_opt_len * 8;
 402
 403                         if (unlikely(!opt_len || opt_len > icmp_len))
 404                                 return 0;
 405
 406                         /* Store the link layer address if the appropriate
 407                          * option is provided.  It is considered an error if
 408                          * the same link layer option is specified twice.
 409                          */
 410                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 411                             && opt_len == 8) {
 412                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 413                                         goto invalid;
 414                                 ether_addr_copy(key->ipv6.nd.sll,
 415                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 416                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 417                                    && opt_len == 8) {
 418                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 419                                         goto invalid;
 420                                 ether_addr_copy(key->ipv6.nd.tll,
 421                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 422                         }
 423
 424                         icmp_len -= opt_len;
 425                         offset += opt_len;
 426                 }
 427         }
 428
 429         return 0;
 430
 431 invalid:
 432         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 433         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 434         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 435
 436         return 0;
 437 }
 438
 439 /**
 440  * key_extract - extracts a flow key from an Ethernet frame.
 441  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 442  * Ethernet header
 443  * @key: output flow key
 444  *
 445  * The caller must ensure that skb->len >= ETH_HLEN.
 446  *
 447  * Returns 0 if successful, otherwise a negative errno value.
 448  *
 449  * Initializes @skb header pointers as follows:
 450  *
 451  *    - skb->mac_header: the Ethernet header.
 452  *
 453  *    - skb->network_header: just past the Ethernet header, or just past the
 454  *      VLAN header, to the first byte of the Ethernet payload.
 455  *
 456  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 457  *      on output, then just past the IP header, if one is present and
 458  *      of a correct length, otherwise the same as skb->network_header.
 459  *      For other key->eth.type values it is left untouched.
 460  */
 461 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 462 {
 463         int error;
 464         struct ethhdr *eth;
 465
 466         /* Flags are always used as part of stats */
 467         key->tp.flags = 0;
 468
 469         skb_reset_mac_header(skb);
 470
 471         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 472          * header in the linear data area.
 473          */
 474         eth = eth_hdr(skb);
 475         ether_addr_copy(key->eth.src, eth->h_source);
 476         ether_addr_copy(key->eth.dst, eth->h_dest);
 477
 478         __skb_pull(skb, 2 * ETH_ALEN);
 479         /* We are going to push all headers that we pull, so no need to
 480          * update skb->csum here.
 481          */
 482
 483         key->eth.tci = 0;
 484         if (skb_vlan_tag_present(skb))
 485                 key->eth.tci = htons(skb->vlan_tci);
 486         else if (eth->h_proto == htons(ETH_P_8021Q))
 487                 if (unlikely(parse_vlan(skb, key)))
 488                         return -ENOMEM;
 489
 490         key->eth.type = parse_ethertype(skb);
 491         if (unlikely(key->eth.type == htons(0)))
 492                 return -ENOMEM;
 493
 494         skb_reset_network_header(skb);
 495         skb_reset_mac_len(skb);
 496         __skb_push(skb, skb->data - skb_mac_header(skb));
 497
 498         /* Network layer. */
 499         if (key->eth.type == htons(ETH_P_IP)) {
 500                 struct iphdr *nh;
 501                 __be16 offset;
 502
 503                 error = check_iphdr(skb);
 504                 if (unlikely(error)) {
 505                         memset(&key->ip, 0, sizeof(key->ip));
 506                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 507                         if (error == -EINVAL) {
 508                                 skb->transport_header = skb->network_header;
 509                                 error = 0;
 510                         }
 511                         return error;
 512                 }
 513
 514                 nh = ip_hdr(skb);
 515                 key->ipv4.addr.src = nh->saddr;
 516                 key->ipv4.addr.dst = nh->daddr;
 517
 518                 key->ip.proto = nh->protocol;
 519                 key->ip.tos = nh->tos;
 520                 key->ip.ttl = nh->ttl;
 521
 522                 offset = nh->frag_off & htons(IP_OFFSET);
 523                 if (offset) {
 524                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 525                         return 0;
 526                 }
 527                 if (nh->frag_off & htons(IP_MF) ||
 528                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 529                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 530                 else
 531                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 532
 533                 /* Transport layer. */
 534                 if (key->ip.proto == IPPROTO_TCP) {
 535                         if (tcphdr_ok(skb)) {
 536                                 struct tcphdr *tcp = tcp_hdr(skb);
 537                                 key->tp.src = tcp->source;
 538                                 key->tp.dst = tcp->dest;
 539                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 540                         } else {
 541                                 memset(&key->tp, 0, sizeof(key->tp));
 542                         }
 543
 544                 } else if (key->ip.proto == IPPROTO_UDP) {
 545                         if (udphdr_ok(skb)) {
 546                                 struct udphdr *udp = udp_hdr(skb);
 547                                 key->tp.src = udp->source;
 548                                 key->tp.dst = udp->dest;
 549                         } else {
 550                                 memset(&key->tp, 0, sizeof(key->tp));
 551                         }
 552                 } else if (key->ip.proto == IPPROTO_SCTP) {
 553                         if (sctphdr_ok(skb)) {
 554                                 struct sctphdr *sctp = sctp_hdr(skb);
 555                                 key->tp.src = sctp->source;
 556                                 key->tp.dst = sctp->dest;
 557                         } else {
 558                                 memset(&key->tp, 0, sizeof(key->tp));
 559                         }
 560                 } else if (key->ip.proto == IPPROTO_ICMP) {
 561                         if (icmphdr_ok(skb)) {
 562                                 struct icmphdr *icmp = icmp_hdr(skb);
 563                                 /* The ICMP type and code fields use the 16-bit
 564                                  * transport port fields, so we need to store
 565                                  * them in 16-bit network byte order. */
 566                                 key->tp.src = htons(icmp->type);
 567                                 key->tp.dst = htons(icmp->code);
 568                         } else {
 569                                 memset(&key->tp, 0, sizeof(key->tp));
 570                         }
 571                 }
 572
 573         } else if (key->eth.type == htons(ETH_P_ARP) ||
 574                    key->eth.type == htons(ETH_P_RARP)) {
 575                 struct arp_eth_header *arp;
 576                 bool arp_available = arphdr_ok(skb);
 577
 578                 arp = (struct arp_eth_header *)skb_network_header(skb);
 579
 580                 if (arp_available &&
 581                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 582                     arp->ar_pro == htons(ETH_P_IP) &&
 583                     arp->ar_hln == ETH_ALEN &&
 584                     arp->ar_pln == 4) {
 585
 586                         /* We only match on the lower 8 bits of the opcode. */
 587                         if (ntohs(arp->ar_op) <= 0xff)
 588                                 key->ip.proto = ntohs(arp->ar_op);
 589                         else
 590                                 key->ip.proto = 0;
 591
 592                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 593                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 594                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 595                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 596                 } else {
 597                         memset(&key->ip, 0, sizeof(key->ip));
 598                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 599                 }
 600         } else if (eth_p_mpls(key->eth.type)) {
 601                 size_t stack_len = MPLS_HLEN;
 602
 603                 /* In the presence of an MPLS label stack the end of the L2
 604                  * header and the beginning of the L3 header differ.
 605                  *
 606                  * Advance network_header to the beginning of the L3
 607                  * header. mac_len corresponds to the end of the L2 header.
 608                  */
 609                 while (1) {
 610                         __be32 lse;
 611
 612                         error = check_header(skb, skb->mac_len + stack_len);
 613                         if (unlikely(error))
 614                                 return 0;
 615
 616                         memcpy(&lse, skb_network_header(skb), MPLS_HLEN);
 617
 618                         if (stack_len == MPLS_HLEN)
 619                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
 620
 621                         skb_set_network_header(skb, skb->mac_len + stack_len);
 622                         if (lse & htonl(MPLS_LS_S_MASK))
 623                                 break;
 624
 625                         stack_len += MPLS_HLEN;
 626                 }
 627         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 628                 int nh_len;             /* IPv6 Header + Extensions */
 629
 630                 nh_len = parse_ipv6hdr(skb, key);
 631                 if (unlikely(nh_len < 0)) {
 632                         switch (nh_len) {
 633                         case -EINVAL:
 634                                 memset(&key->ip, 0, sizeof(key->ip));
 635                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 636                                 /* fall-through */
 637                         case -EPROTO:
 638                                 skb->transport_header = skb->network_header;
 639                                 error = 0;
 640                                 break;
 641                         default:
 642                                 error = nh_len;
 643                         }
 644                         return error;
 645                 }
 646
 647                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 648                         return 0;
 649                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 650                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 651
 652                 /* Transport layer. */
 653                 if (key->ip.proto == NEXTHDR_TCP) {
 654                         if (tcphdr_ok(skb)) {
 655                                 struct tcphdr *tcp = tcp_hdr(skb);
 656                                 key->tp.src = tcp->source;
 657                                 key->tp.dst = tcp->dest;
 658                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 659                         } else {
 660                                 memset(&key->tp, 0, sizeof(key->tp));
 661                         }
 662                 } else if (key->ip.proto == NEXTHDR_UDP) {
 663                         if (udphdr_ok(skb)) {
 664                                 struct udphdr *udp = udp_hdr(skb);
 665                                 key->tp.src = udp->source;
 666                                 key->tp.dst = udp->dest;
 667                         } else {
 668                                 memset(&key->tp, 0, sizeof(key->tp));
 669                         }
 670                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 671                         if (sctphdr_ok(skb)) {
 672                                 struct sctphdr *sctp = sctp_hdr(skb);
 673                                 key->tp.src = sctp->source;
 674                                 key->tp.dst = sctp->dest;
 675                         } else {
 676                                 memset(&key->tp, 0, sizeof(key->tp));
 677                         }
 678                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 679                         if (icmp6hdr_ok(skb)) {
 680                                 error = parse_icmpv6(skb, key, nh_len);
 681                                 if (error)
 682                                         return error;
 683                         } else {
 684                                 memset(&key->tp, 0, sizeof(key->tp));
 685                         }
 686                 }
 687         }
 688         return 0;
 689 }
 690
 691 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 692 {
 693         return key_extract(skb, key);
 694 }
 695
 696 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
 697                          struct sk_buff *skb, struct sw_flow_key *key)
 698 {
 699         /* Extract metadata from packet. */
 700         if (tun_info) {
 701                 key->tun_proto = ip_tunnel_info_af(tun_info);
 702                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
 703
 704                 if (tun_info->options_len) {
 705                         BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
 706                                                    8)) - 1
 707                                         > sizeof(key->tun_opts));
 708
 709                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
 710                                                 tun_info);
 711                         key->tun_opts_len = tun_info->options_len;
 712                 } else {
 713                         key->tun_opts_len = 0;
 714                 }
 715         } else  {
 716                 key->tun_proto = 0;
 717                 key->tun_opts_len = 0;
 718                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 719         }
 720
 721         key->phy.priority = skb->priority;
 722         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 723         key->phy.skb_mark = skb->mark;
 724         ovs_ct_fill_key(skb, key);
 725         key->ovs_flow_hash = 0;
 726         key->recirc_id = 0;
 727
 728         return key_extract(skb, key);
 729 }
 730
 731 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
 732                                    struct sk_buff *skb,
 733                                    struct sw_flow_key *key, bool log)
 734 {
 735         int err;
 736
 737         memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
 738
 739         /* Extract metadata from netlink attributes. */
 740         err = ovs_nla_get_flow_metadata(net, attr, key, log);
 741         if (err)
 742                 return err;
 743
 744         return key_extract(skb, key);
 745 }