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_node_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         /* We open code this to make sure node 0 is always considered */
 146         for (node = 0; node < MAX_NUMNODES; node = next_node(node, node_possible_map)) {
 147                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]);
 148
 149                 if (stats) {
 150                         /* Local CPU may write on non-local stats, so we must
 151                          * block bottom-halves here.
 152                          */
 153                         spin_lock_bh(&stats->lock);
 154                         if (!*used || time_after(stats->used, *used))
 155                                 *used = stats->used;
 156                         *tcp_flags |= stats->tcp_flags;
 157                         ovs_stats->n_packets += stats->packet_count;
 158                         ovs_stats->n_bytes += stats->byte_count;
 159                         spin_unlock_bh(&stats->lock);
 160                 }
 161         }
 162 }
 163
 164 /* Called with ovs_mutex. */
 165 void ovs_flow_stats_clear(struct sw_flow *flow)
 166 {
 167         int node;
 168
 169         /* We open code this to make sure node 0 is always considered */
 170         for (node = 0; node < MAX_NUMNODES; node = next_node(node, node_possible_map)) {
 171                 struct flow_stats *stats = ovsl_dereference(flow->stats[node]);
 172
 173                 if (stats) {
 174                         spin_lock_bh(&stats->lock);
 175                         stats->used = 0;
 176                         stats->packet_count = 0;
 177                         stats->byte_count = 0;
 178                         stats->tcp_flags = 0;
 179                         spin_unlock_bh(&stats->lock);
 180                 }
 181         }
 182 }
 183
 184 static int check_header(struct sk_buff *skb, int len)
 185 {
 186         if (unlikely(skb->len < len))
 187                 return -EINVAL;
 188         if (unlikely(!pskb_may_pull(skb, len)))
 189                 return -ENOMEM;
 190         return 0;
 191 }
 192
 193 static bool arphdr_ok(struct sk_buff *skb)
 194 {
 195         return pskb_may_pull(skb, skb_network_offset(skb) +
 196                                   sizeof(struct arp_eth_header));
 197 }
 198
 199 static int check_iphdr(struct sk_buff *skb)
 200 {
 201         unsigned int nh_ofs = skb_network_offset(skb);
 202         unsigned int ip_len;
 203         int err;
 204
 205         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 206         if (unlikely(err))
 207                 return err;
 208
 209         ip_len = ip_hdrlen(skb);
 210         if (unlikely(ip_len < sizeof(struct iphdr) ||
 211                      skb->len < nh_ofs + ip_len))
 212                 return -EINVAL;
 213
 214         skb_set_transport_header(skb, nh_ofs + ip_len);
 215         return 0;
 216 }
 217
 218 static bool tcphdr_ok(struct sk_buff *skb)
 219 {
 220         int th_ofs = skb_transport_offset(skb);
 221         int tcp_len;
 222
 223         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 224                 return false;
 225
 226         tcp_len = tcp_hdrlen(skb);
 227         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 228                      skb->len < th_ofs + tcp_len))
 229                 return false;
 230
 231         return true;
 232 }
 233
 234 static bool udphdr_ok(struct sk_buff *skb)
 235 {
 236         return pskb_may_pull(skb, skb_transport_offset(skb) +
 237                                   sizeof(struct udphdr));
 238 }
 239
 240 static bool sctphdr_ok(struct sk_buff *skb)
 241 {
 242         return pskb_may_pull(skb, skb_transport_offset(skb) +
 243                                   sizeof(struct sctphdr));
 244 }
 245
 246 static bool icmphdr_ok(struct sk_buff *skb)
 247 {
 248         return pskb_may_pull(skb, skb_transport_offset(skb) +
 249                                   sizeof(struct icmphdr));
 250 }
 251
 252 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 253 {
 254         unsigned int nh_ofs = skb_network_offset(skb);
 255         unsigned int nh_len;
 256         int payload_ofs;
 257         struct ipv6hdr *nh;
 258         uint8_t nexthdr;
 259         __be16 frag_off;
 260         int err;
 261
 262         err = check_header(skb, nh_ofs + sizeof(*nh));
 263         if (unlikely(err))
 264                 return err;
 265
 266         nh = ipv6_hdr(skb);
 267         nexthdr = nh->nexthdr;
 268         payload_ofs = (u8 *)(nh + 1) - skb->data;
 269
 270         key->ip.proto = NEXTHDR_NONE;
 271         key->ip.tos = ipv6_get_dsfield(nh);
 272         key->ip.ttl = nh->hop_limit;
 273         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 274         key->ipv6.addr.src = nh->saddr;
 275         key->ipv6.addr.dst = nh->daddr;
 276
 277         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 278
 279         if (frag_off) {
 280                 if (frag_off & htons(~0x7))
 281                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 282                 else
 283                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 284         } else {
 285                 key->ip.frag = OVS_FRAG_TYPE_NONE;
 286         }
 287
 288         /* Delayed handling of error in ipv6_skip_exthdr() as it
 289          * always sets frag_off to a valid value which may be
 290          * used to set key->ip.frag above.
 291          */
 292         if (unlikely(payload_ofs < 0))
 293                 return -EPROTO;
 294
 295         nh_len = payload_ofs - nh_ofs;
 296         skb_set_transport_header(skb, nh_ofs + nh_len);
 297         key->ip.proto = nexthdr;
 298         return nh_len;
 299 }
 300
 301 static bool icmp6hdr_ok(struct sk_buff *skb)
 302 {
 303         return pskb_may_pull(skb, skb_transport_offset(skb) +
 304                                   sizeof(struct icmp6hdr));
 305 }
 306
 307 /**
 308  * Parse vlan tag from vlan header.
 309  * Returns ERROR on memory error.
 310  * Returns 0 if it encounters a non-vlan or incomplete packet.
 311  * Returns 1 after successfully parsing vlan tag.
 312  */
 313 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh)
 314 {
 315         struct vlan_head *vh = (struct vlan_head *)skb->data;
 316
 317         if (likely(!eth_type_vlan(vh->tpid)))
 318                 return 0;
 319
 320         if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
 321                 return 0;
 322
 323         if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
 324                                  sizeof(__be16))))
 325                 return -ENOMEM;
 326
 327         vh = (struct vlan_head *)skb->data;
 328         key_vh->tci = vh->tci | htons(VLAN_TAG_PRESENT);
 329         key_vh->tpid = vh->tpid;
 330
 331         __skb_pull(skb, sizeof(struct vlan_head));
 332         return 1;
 333 }
 334
 335 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 336 {
 337         int res;
 338
 339         key->eth.vlan.tci = 0;
 340         key->eth.vlan.tpid = 0;
 341         key->eth.cvlan.tci = 0;
 342         key->eth.cvlan.tpid = 0;
 343
 344         if (likely(skb_vlan_tag_present(skb))) {
 345                 key->eth.vlan.tci = htons(skb->vlan_tci);
 346                 key->eth.vlan.tpid = skb->vlan_proto;
 347         } else {
 348                 /* Parse outer vlan tag in the non-accelerated case. */
 349                 res = parse_vlan_tag(skb, &key->eth.vlan);
 350                 if (res <= 0)
 351                         return res;
 352         }
 353
 354         /* Parse inner vlan tag. */
 355         res = parse_vlan_tag(skb, &key->eth.cvlan);
 356         if (res <= 0)
 357                 return res;
 358
 359         return 0;
 360 }
 361
 362 static __be16 parse_ethertype(struct sk_buff *skb)
 363 {
 364         struct llc_snap_hdr {
 365                 u8  dsap;  /* Always 0xAA */
 366                 u8  ssap;  /* Always 0xAA */
 367                 u8  ctrl;
 368                 u8  oui[3];
 369                 __be16 ethertype;
 370         };
 371         struct llc_snap_hdr *llc;
 372         __be16 proto;
 373
 374         proto = *(__be16 *) skb->data;
 375         __skb_pull(skb, sizeof(__be16));
 376
 377         if (eth_proto_is_802_3(proto))
 378                 return proto;
 379
 380         if (skb->len < sizeof(struct llc_snap_hdr))
 381                 return htons(ETH_P_802_2);
 382
 383         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 384                 return htons(0);
 385
 386         llc = (struct llc_snap_hdr *) skb->data;
 387         if (llc->dsap != LLC_SAP_SNAP ||
 388             llc->ssap != LLC_SAP_SNAP ||
 389             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 390                 return htons(ETH_P_802_2);
 391
 392         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 393
 394         if (eth_proto_is_802_3(llc->ethertype))
 395                 return llc->ethertype;
 396
 397         return htons(ETH_P_802_2);
 398 }
 399
 400 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 401                         int nh_len)
 402 {
 403         struct icmp6hdr *icmp = icmp6_hdr(skb);
 404
 405         /* The ICMPv6 type and code fields use the 16-bit transport port
 406          * fields, so we need to store them in 16-bit network byte order.
 407          */
 408         key->tp.src = htons(icmp->icmp6_type);
 409         key->tp.dst = htons(icmp->icmp6_code);
 410         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 411
 412         if (icmp->icmp6_code == 0 &&
 413             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 414              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 415                 int icmp_len = skb->len - skb_transport_offset(skb);
 416                 struct nd_msg *nd;
 417                 int offset;
 418
 419                 /* In order to process neighbor discovery options, we need the
 420                  * entire packet.
 421                  */
 422                 if (unlikely(icmp_len < sizeof(*nd)))
 423                         return 0;
 424
 425                 if (unlikely(skb_linearize(skb)))
 426                         return -ENOMEM;
 427
 428                 nd = (struct nd_msg *)skb_transport_header(skb);
 429                 key->ipv6.nd.target = nd->target;
 430
 431                 icmp_len -= sizeof(*nd);
 432                 offset = 0;
 433                 while (icmp_len >= 8) {
 434                         struct nd_opt_hdr *nd_opt =
 435                                  (struct nd_opt_hdr *)(nd->opt + offset);
 436                         int opt_len = nd_opt->nd_opt_len * 8;
 437
 438                         if (unlikely(!opt_len || opt_len > icmp_len))
 439                                 return 0;
 440
 441                         /* Store the link layer address if the appropriate
 442                          * option is provided.  It is considered an error if
 443                          * the same link layer option is specified twice.
 444                          */
 445                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 446                             && opt_len == 8) {
 447                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 448                                         goto invalid;
 449                                 ether_addr_copy(key->ipv6.nd.sll,
 450                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 451                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 452                                    && opt_len == 8) {
 453                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 454                                         goto invalid;
 455                                 ether_addr_copy(key->ipv6.nd.tll,
 456                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 457                         }
 458
 459                         icmp_len -= opt_len;
 460                         offset += opt_len;
 461                 }
 462         }
 463
 464         return 0;
 465
 466 invalid:
 467         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 468         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 469         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 470
 471         return 0;
 472 }
 473
 474 /**
 475  * key_extract - extracts a flow key from an Ethernet frame.
 476  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 477  * Ethernet header
 478  * @key: output flow key
 479  *
 480  * The caller must ensure that skb->len >= ETH_HLEN.
 481  *
 482  * Returns 0 if successful, otherwise a negative errno value.
 483  *
 484  * Initializes @skb header pointers as follows:
 485  *
 486  *    - skb->mac_header: the Ethernet header.
 487  *
 488  *    - skb->network_header: just past the Ethernet header, or just past the
 489  *      VLAN header, to the first byte of the Ethernet payload.
 490  *
 491  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 492  *      on output, then just past the IP header, if one is present and
 493  *      of a correct length, otherwise the same as skb->network_header.
 494  *      For other key->eth.type values it is left untouched.
 495  */
 496 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 497 {
 498         int error;
 499         struct ethhdr *eth;
 500
 501         /* Flags are always used as part of stats */
 502         key->tp.flags = 0;
 503
 504         skb_reset_mac_header(skb);
 505
 506         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 507          * header in the linear data area.
 508          */
 509         eth = eth_hdr(skb);
 510         ether_addr_copy(key->eth.src, eth->h_source);
 511         ether_addr_copy(key->eth.dst, eth->h_dest);
 512
 513         __skb_pull(skb, 2 * ETH_ALEN);
 514         /* We are going to push all headers that we pull, so no need to
 515          * update skb->csum here.
 516          */
 517
 518         if (unlikely(parse_vlan(skb, key)))
 519                 return -ENOMEM;
 520
 521         key->eth.type = parse_ethertype(skb);
 522         if (unlikely(key->eth.type == htons(0)))
 523                 return -ENOMEM;
 524
 525         skb_reset_network_header(skb);
 526         skb_reset_mac_len(skb);
 527         __skb_push(skb, skb->data - skb_mac_header(skb));
 528
 529         /* Network layer. */
 530         if (key->eth.type == htons(ETH_P_IP)) {
 531                 struct iphdr *nh;
 532                 __be16 offset;
 533
 534                 error = check_iphdr(skb);
 535                 if (unlikely(error)) {
 536                         memset(&key->ip, 0, sizeof(key->ip));
 537                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 538                         if (error == -EINVAL) {
 539                                 skb->transport_header = skb->network_header;
 540                                 error = 0;
 541                         }
 542                         return error;
 543                 }
 544
 545                 nh = ip_hdr(skb);
 546                 key->ipv4.addr.src = nh->saddr;
 547                 key->ipv4.addr.dst = nh->daddr;
 548
 549                 key->ip.proto = nh->protocol;
 550                 key->ip.tos = nh->tos;
 551                 key->ip.ttl = nh->ttl;
 552
 553                 offset = nh->frag_off & htons(IP_OFFSET);
 554                 if (offset) {
 555                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 556                         return 0;
 557                 }
 558                 if (nh->frag_off & htons(IP_MF) ||
 559                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 560                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 561                 else
 562                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 563
 564                 /* Transport layer. */
 565                 if (key->ip.proto == IPPROTO_TCP) {
 566                         if (tcphdr_ok(skb)) {
 567                                 struct tcphdr *tcp = tcp_hdr(skb);
 568                                 key->tp.src = tcp->source;
 569                                 key->tp.dst = tcp->dest;
 570                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 571                         } else {
 572                                 memset(&key->tp, 0, sizeof(key->tp));
 573                         }
 574
 575                 } else if (key->ip.proto == IPPROTO_UDP) {
 576                         if (udphdr_ok(skb)) {
 577                                 struct udphdr *udp = udp_hdr(skb);
 578                                 key->tp.src = udp->source;
 579                                 key->tp.dst = udp->dest;
 580                         } else {
 581                                 memset(&key->tp, 0, sizeof(key->tp));
 582                         }
 583                 } else if (key->ip.proto == IPPROTO_SCTP) {
 584                         if (sctphdr_ok(skb)) {
 585                                 struct sctphdr *sctp = sctp_hdr(skb);
 586                                 key->tp.src = sctp->source;
 587                                 key->tp.dst = sctp->dest;
 588                         } else {
 589                                 memset(&key->tp, 0, sizeof(key->tp));
 590                         }
 591                 } else if (key->ip.proto == IPPROTO_ICMP) {
 592                         if (icmphdr_ok(skb)) {
 593                                 struct icmphdr *icmp = icmp_hdr(skb);
 594                                 /* The ICMP type and code fields use the 16-bit
 595                                  * transport port fields, so we need to store
 596                                  * them in 16-bit network byte order. */
 597                                 key->tp.src = htons(icmp->type);
 598                                 key->tp.dst = htons(icmp->code);
 599                         } else {
 600                                 memset(&key->tp, 0, sizeof(key->tp));
 601                         }
 602                 }
 603
 604         } else if (key->eth.type == htons(ETH_P_ARP) ||
 605                    key->eth.type == htons(ETH_P_RARP)) {
 606                 struct arp_eth_header *arp;
 607                 bool arp_available = arphdr_ok(skb);
 608
 609                 arp = (struct arp_eth_header *)skb_network_header(skb);
 610
 611                 if (arp_available &&
 612                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 613                     arp->ar_pro == htons(ETH_P_IP) &&
 614                     arp->ar_hln == ETH_ALEN &&
 615                     arp->ar_pln == 4) {
 616
 617                         /* We only match on the lower 8 bits of the opcode. */
 618                         if (ntohs(arp->ar_op) <= 0xff)
 619                                 key->ip.proto = ntohs(arp->ar_op);
 620                         else
 621                                 key->ip.proto = 0;
 622
 623                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 624                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 625                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 626                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 627                 } else {
 628                         memset(&key->ip, 0, sizeof(key->ip));
 629                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 630                 }
 631         } else if (eth_p_mpls(key->eth.type)) {
 632                 size_t stack_len = MPLS_HLEN;
 633
 634                 /* In the presence of an MPLS label stack the end of the L2
 635                  * header and the beginning of the L3 header differ.
 636                  *
 637                  * Advance network_header to the beginning of the L3
 638                  * header. mac_len corresponds to the end of the L2 header.
 639                  */
 640                 while (1) {
 641                         __be32 lse;
 642
 643                         error = check_header(skb, skb->mac_len + stack_len);
 644                         if (unlikely(error))
 645                                 return 0;
 646
 647                         memcpy(&lse, skb_network_header(skb), MPLS_HLEN);
 648
 649                         if (stack_len == MPLS_HLEN)
 650                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
 651
 652                         skb_set_network_header(skb, skb->mac_len + stack_len);
 653                         if (lse & htonl(MPLS_LS_S_MASK))
 654                                 break;
 655
 656                         stack_len += MPLS_HLEN;
 657                 }
 658         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 659                 int nh_len;             /* IPv6 Header + Extensions */
 660
 661                 nh_len = parse_ipv6hdr(skb, key);
 662                 if (unlikely(nh_len < 0)) {
 663                         switch (nh_len) {
 664                         case -EINVAL:
 665                                 memset(&key->ip, 0, sizeof(key->ip));
 666                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 667                                 /* fall-through */
 668                         case -EPROTO:
 669                                 skb->transport_header = skb->network_header;
 670                                 error = 0;
 671                                 break;
 672                         default:
 673                                 error = nh_len;
 674                         }
 675                         return error;
 676                 }
 677
 678                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 679                         return 0;
 680                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 681                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 682
 683                 /* Transport layer. */
 684                 if (key->ip.proto == NEXTHDR_TCP) {
 685                         if (tcphdr_ok(skb)) {
 686                                 struct tcphdr *tcp = tcp_hdr(skb);
 687                                 key->tp.src = tcp->source;
 688                                 key->tp.dst = tcp->dest;
 689                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 690                         } else {
 691                                 memset(&key->tp, 0, sizeof(key->tp));
 692                         }
 693                 } else if (key->ip.proto == NEXTHDR_UDP) {
 694                         if (udphdr_ok(skb)) {
 695                                 struct udphdr *udp = udp_hdr(skb);
 696                                 key->tp.src = udp->source;
 697                                 key->tp.dst = udp->dest;
 698                         } else {
 699                                 memset(&key->tp, 0, sizeof(key->tp));
 700                         }
 701                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 702                         if (sctphdr_ok(skb)) {
 703                                 struct sctphdr *sctp = sctp_hdr(skb);
 704                                 key->tp.src = sctp->source;
 705                                 key->tp.dst = sctp->dest;
 706                         } else {
 707                                 memset(&key->tp, 0, sizeof(key->tp));
 708                         }
 709                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 710                         if (icmp6hdr_ok(skb)) {
 711                                 error = parse_icmpv6(skb, key, nh_len);
 712                                 if (error)
 713                                         return error;
 714                         } else {
 715                                 memset(&key->tp, 0, sizeof(key->tp));
 716                         }
 717                 }
 718         }
 719         return 0;
 720 }
 721
 722 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 723 {
 724         return key_extract(skb, key);
 725 }
 726
 727 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
 728                          struct sk_buff *skb, struct sw_flow_key *key)
 729 {
 730         /* Extract metadata from packet. */
 731         if (tun_info) {
 732                 key->tun_proto = ip_tunnel_info_af(tun_info);
 733                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
 734
 735                 if (tun_info->options_len) {
 736                         BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
 737                                                    8)) - 1
 738                                         > sizeof(key->tun_opts));
 739
 740                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
 741                                                 tun_info);
 742                         key->tun_opts_len = tun_info->options_len;
 743                 } else {
 744                         key->tun_opts_len = 0;
 745                 }
 746         } else  {
 747                 key->tun_proto = 0;
 748                 key->tun_opts_len = 0;
 749                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 750         }
 751
 752         key->phy.priority = skb->priority;
 753         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 754         key->phy.skb_mark = skb->mark;
 755         ovs_ct_fill_key(skb, key);
 756         key->ovs_flow_hash = 0;
 757         key->recirc_id = 0;
 758
 759         return key_extract(skb, key);
 760 }
 761
 762 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
 763                                    struct sk_buff *skb,
 764                                    struct sw_flow_key *key, bool log)
 765 {
 766         int err;
 767
 768         memset(key, 0, OVS_SW_FLOW_KEY_METADATA_SIZE);
 769
 770         /* Extract metadata from netlink attributes. */
 771         err = ovs_nla_get_flow_metadata(net, attr, key, log);
 772         if (err)
 773                 return err;
 774
 775         return key_extract(skb, key);
 776 }