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/cpumask.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/ip.h>
  35 #include <linux/ipv6.h>
  36 #include <linux/mpls.h>
  37 #include <linux/sctp.h>
  38 #include <linux/smp.h>
  39 #include <linux/tcp.h>
  40 #include <linux/udp.h>
  41 #include <linux/icmp.h>
  42 #include <linux/icmpv6.h>
  43 #include <linux/rculist.h>
  44 #include <net/ip.h>
  45 #include <net/ip_tunnels.h>
  46 #include <net/ipv6.h>
  47 #include <net/mpls.h>
  48 #include <net/ndisc.h>
  49 #include <net/nsh.h>
  50
  51 #include "conntrack.h"
  52 #include "datapath.h"
  53 #include "flow.h"
  54 #include "flow_netlink.h"
  55 #include "vport.h"
  56
  57 u64 ovs_flow_used_time(unsigned long flow_jiffies)
  58 {
  59         struct timespec cur_ts;
  60         u64 cur_ms, idle_ms;
  61
  62         ktime_get_ts(&cur_ts);
  63         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  64         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
  65                  cur_ts.tv_nsec / NSEC_PER_MSEC;
  66
  67         return cur_ms - idle_ms;
  68 }
  69
  70 #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF))
  71
  72 void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags,
  73                            const struct sk_buff *skb)
  74 {
  75         struct flow_stats *stats;
  76         unsigned int cpu = smp_processor_id();
  77         int len = skb->len + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
  78
  79         stats = rcu_dereference(flow->stats[cpu]);
  80
  81         /* Check if already have CPU-specific stats. */
  82         if (likely(stats)) {
  83                 spin_lock(&stats->lock);
  84                 /* Mark if we write on the pre-allocated stats. */
  85                 if (cpu == 0 && unlikely(flow->stats_last_writer != cpu))
  86                         flow->stats_last_writer = cpu;
  87         } else {
  88                 stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
  89                 spin_lock(&stats->lock);
  90
  91                 /* If the current CPU is the only writer on the
  92                  * pre-allocated stats keep using them.
  93                  */
  94                 if (unlikely(flow->stats_last_writer != cpu)) {
  95                         /* A previous locker may have already allocated the
  96                          * stats, so we need to check again.  If CPU-specific
  97                          * stats were already allocated, we update the pre-
  98                          * allocated stats as we have already locked them.
  99                          */
 100                         if (likely(flow->stats_last_writer != -1) &&
 101                             likely(!rcu_access_pointer(flow->stats[cpu]))) {
 102                                 /* Try to allocate CPU-specific stats. */
 103                                 struct flow_stats *new_stats;
 104
 105                                 new_stats =
 106                                         kmem_cache_alloc_node(flow_stats_cache,
 107                                                               GFP_NOWAIT |
 108                                                               __GFP_THISNODE |
 109                                                               __GFP_NOWARN |
 110                                                               __GFP_NOMEMALLOC,
 111                                                               numa_node_id());
 112                                 if (likely(new_stats)) {
 113                                         new_stats->used = jiffies;
 114                                         new_stats->packet_count = 1;
 115                                         new_stats->byte_count = len;
 116                                         new_stats->tcp_flags = tcp_flags;
 117                                         spin_lock_init(&new_stats->lock);
 118
 119                                         rcu_assign_pointer(flow->stats[cpu],
 120                                                            new_stats);
 121                                         cpumask_set_cpu(cpu, &flow->cpu_used_mask);
 122                                         goto unlock;
 123                                 }
 124                         }
 125                         flow->stats_last_writer = cpu;
 126                 }
 127         }
 128
 129         stats->used = jiffies;
 130         stats->packet_count++;
 131         stats->byte_count += len;
 132         stats->tcp_flags |= tcp_flags;
 133 unlock:
 134         spin_unlock(&stats->lock);
 135 }
 136
 137 /* Must be called with rcu_read_lock or ovs_mutex. */
 138 void ovs_flow_stats_get(const struct sw_flow *flow,
 139                         struct ovs_flow_stats *ovs_stats,
 140                         unsigned long *used, __be16 *tcp_flags)
 141 {
 142         int cpu;
 143
 144         *used = 0;
 145         *tcp_flags = 0;
 146         memset(ovs_stats, 0, sizeof(*ovs_stats));
 147
 148         /* We open code this to make sure cpu 0 is always considered */
 149         for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
 150                 struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[cpu]);
 151
 152                 if (stats) {
 153                         /* Local CPU may write on non-local stats, so we must
 154                          * block bottom-halves here.
 155                          */
 156                         spin_lock_bh(&stats->lock);
 157                         if (!*used || time_after(stats->used, *used))
 158                                 *used = stats->used;
 159                         *tcp_flags |= stats->tcp_flags;
 160                         ovs_stats->n_packets += stats->packet_count;
 161                         ovs_stats->n_bytes += stats->byte_count;
 162                         spin_unlock_bh(&stats->lock);
 163                 }
 164         }
 165 }
 166
 167 /* Called with ovs_mutex. */
 168 void ovs_flow_stats_clear(struct sw_flow *flow)
 169 {
 170         int cpu;
 171
 172         /* We open code this to make sure cpu 0 is always considered */
 173         for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) {
 174                 struct flow_stats *stats = ovsl_dereference(flow->stats[cpu]);
 175
 176                 if (stats) {
 177                         spin_lock_bh(&stats->lock);
 178                         stats->used = 0;
 179                         stats->packet_count = 0;
 180                         stats->byte_count = 0;
 181                         stats->tcp_flags = 0;
 182                         spin_unlock_bh(&stats->lock);
 183                 }
 184         }
 185 }
 186
 187 static int check_header(struct sk_buff *skb, int len)
 188 {
 189         if (unlikely(skb->len < len))
 190                 return -EINVAL;
 191         if (unlikely(!pskb_may_pull(skb, len)))
 192                 return -ENOMEM;
 193         return 0;
 194 }
 195
 196 static bool arphdr_ok(struct sk_buff *skb)
 197 {
 198         return pskb_may_pull(skb, skb_network_offset(skb) +
 199                                   sizeof(struct arp_eth_header));
 200 }
 201
 202 static int check_iphdr(struct sk_buff *skb)
 203 {
 204         unsigned int nh_ofs = skb_network_offset(skb);
 205         unsigned int ip_len;
 206         int err;
 207
 208         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 209         if (unlikely(err))
 210                 return err;
 211
 212         ip_len = ip_hdrlen(skb);
 213         if (unlikely(ip_len < sizeof(struct iphdr) ||
 214                      skb->len < nh_ofs + ip_len))
 215                 return -EINVAL;
 216
 217         skb_set_transport_header(skb, nh_ofs + ip_len);
 218         return 0;
 219 }
 220
 221 static bool tcphdr_ok(struct sk_buff *skb)
 222 {
 223         int th_ofs = skb_transport_offset(skb);
 224         int tcp_len;
 225
 226         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 227                 return false;
 228
 229         tcp_len = tcp_hdrlen(skb);
 230         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 231                      skb->len < th_ofs + tcp_len))
 232                 return false;
 233
 234         return true;
 235 }
 236
 237 static bool udphdr_ok(struct sk_buff *skb)
 238 {
 239         return pskb_may_pull(skb, skb_transport_offset(skb) +
 240                                   sizeof(struct udphdr));
 241 }
 242
 243 static bool sctphdr_ok(struct sk_buff *skb)
 244 {
 245         return pskb_may_pull(skb, skb_transport_offset(skb) +
 246                                   sizeof(struct sctphdr));
 247 }
 248
 249 static bool icmphdr_ok(struct sk_buff *skb)
 250 {
 251         return pskb_may_pull(skb, skb_transport_offset(skb) +
 252                                   sizeof(struct icmphdr));
 253 }
 254
 255 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 256 {
 257         unsigned int nh_ofs = skb_network_offset(skb);
 258         unsigned int nh_len;
 259         int payload_ofs;
 260         struct ipv6hdr *nh;
 261         uint8_t nexthdr;
 262         __be16 frag_off;
 263         int err;
 264
 265         err = check_header(skb, nh_ofs + sizeof(*nh));
 266         if (unlikely(err))
 267                 return err;
 268
 269         nh = ipv6_hdr(skb);
 270         nexthdr = nh->nexthdr;
 271         payload_ofs = (u8 *)(nh + 1) - skb->data;
 272
 273         key->ip.proto = NEXTHDR_NONE;
 274         key->ip.tos = ipv6_get_dsfield(nh);
 275         key->ip.ttl = nh->hop_limit;
 276         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 277         key->ipv6.addr.src = nh->saddr;
 278         key->ipv6.addr.dst = nh->daddr;
 279
 280         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 281
 282         if (frag_off) {
 283                 if (frag_off & htons(~0x7))
 284                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 285                 else
 286                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 287         } else {
 288                 key->ip.frag = OVS_FRAG_TYPE_NONE;
 289         }
 290
 291         /* Delayed handling of error in ipv6_skip_exthdr() as it
 292          * always sets frag_off to a valid value which may be
 293          * used to set key->ip.frag above.
 294          */
 295         if (unlikely(payload_ofs < 0))
 296                 return -EPROTO;
 297
 298         nh_len = payload_ofs - nh_ofs;
 299         skb_set_transport_header(skb, nh_ofs + nh_len);
 300         key->ip.proto = nexthdr;
 301         return nh_len;
 302 }
 303
 304 static bool icmp6hdr_ok(struct sk_buff *skb)
 305 {
 306         return pskb_may_pull(skb, skb_transport_offset(skb) +
 307                                   sizeof(struct icmp6hdr));
 308 }
 309
 310 /**
 311  * Parse vlan tag from vlan header.
 312  * Returns ERROR on memory error.
 313  * Returns 0 if it encounters a non-vlan or incomplete packet.
 314  * Returns 1 after successfully parsing vlan tag.
 315  */
 316 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
 317                           bool untag_vlan)
 318 {
 319         struct vlan_head *vh = (struct vlan_head *)skb->data;
 320
 321         if (likely(!eth_type_vlan(vh->tpid)))
 322                 return 0;
 323
 324         if (unlikely(skb->len < sizeof(struct vlan_head) + sizeof(__be16)))
 325                 return 0;
 326
 327         if (unlikely(!pskb_may_pull(skb, sizeof(struct vlan_head) +
 328                                  sizeof(__be16))))
 329                 return -ENOMEM;
 330
 331         vh = (struct vlan_head *)skb->data;
 332         key_vh->tci = vh->tci | htons(VLAN_TAG_PRESENT);
 333         key_vh->tpid = vh->tpid;
 334
 335         if (unlikely(untag_vlan)) {
 336                 int offset = skb->data - skb_mac_header(skb);
 337                 u16 tci;
 338                 int err;
 339
 340                 __skb_push(skb, offset);
 341                 err = __skb_vlan_pop(skb, &tci);
 342                 __skb_pull(skb, offset);
 343                 if (err)
 344                         return err;
 345                 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
 346         } else {
 347                 __skb_pull(skb, sizeof(struct vlan_head));
 348         }
 349         return 1;
 350 }
 351
 352 static void clear_vlan(struct sw_flow_key *key)
 353 {
 354         key->eth.vlan.tci = 0;
 355         key->eth.vlan.tpid = 0;
 356         key->eth.cvlan.tci = 0;
 357         key->eth.cvlan.tpid = 0;
 358 }
 359
 360 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 361 {
 362         int res;
 363
 364         if (skb_vlan_tag_present(skb)) {
 365                 key->eth.vlan.tci = htons(skb->vlan_tci);
 366                 key->eth.vlan.tpid = skb->vlan_proto;
 367         } else {
 368                 /* Parse outer vlan tag in the non-accelerated case. */
 369                 res = parse_vlan_tag(skb, &key->eth.vlan, true);
 370                 if (res <= 0)
 371                         return res;
 372         }
 373
 374         /* Parse inner vlan tag. */
 375         res = parse_vlan_tag(skb, &key->eth.cvlan, false);
 376         if (res <= 0)
 377                 return res;
 378
 379         return 0;
 380 }
 381
 382 static __be16 parse_ethertype(struct sk_buff *skb)
 383 {
 384         struct llc_snap_hdr {
 385                 u8  dsap;  /* Always 0xAA */
 386                 u8  ssap;  /* Always 0xAA */
 387                 u8  ctrl;
 388                 u8  oui[3];
 389                 __be16 ethertype;
 390         };
 391         struct llc_snap_hdr *llc;
 392         __be16 proto;
 393
 394         proto = *(__be16 *) skb->data;
 395         __skb_pull(skb, sizeof(__be16));
 396
 397         if (eth_proto_is_802_3(proto))
 398                 return proto;
 399
 400         if (skb->len < sizeof(struct llc_snap_hdr))
 401                 return htons(ETH_P_802_2);
 402
 403         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 404                 return htons(0);
 405
 406         llc = (struct llc_snap_hdr *) skb->data;
 407         if (llc->dsap != LLC_SAP_SNAP ||
 408             llc->ssap != LLC_SAP_SNAP ||
 409             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 410                 return htons(ETH_P_802_2);
 411
 412         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 413
 414         if (eth_proto_is_802_3(llc->ethertype))
 415                 return llc->ethertype;
 416
 417         return htons(ETH_P_802_2);
 418 }
 419
 420 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 421                         int nh_len)
 422 {
 423         struct icmp6hdr *icmp = icmp6_hdr(skb);
 424
 425         /* The ICMPv6 type and code fields use the 16-bit transport port
 426          * fields, so we need to store them in 16-bit network byte order.
 427          */
 428         key->tp.src = htons(icmp->icmp6_type);
 429         key->tp.dst = htons(icmp->icmp6_code);
 430         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 431
 432         if (icmp->icmp6_code == 0 &&
 433             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 434              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 435                 int icmp_len = skb->len - skb_transport_offset(skb);
 436                 struct nd_msg *nd;
 437                 int offset;
 438
 439                 /* In order to process neighbor discovery options, we need the
 440                  * entire packet.
 441                  */
 442                 if (unlikely(icmp_len < sizeof(*nd)))
 443                         return 0;
 444
 445                 if (unlikely(skb_linearize(skb)))
 446                         return -ENOMEM;
 447
 448                 nd = (struct nd_msg *)skb_transport_header(skb);
 449                 key->ipv6.nd.target = nd->target;
 450
 451                 icmp_len -= sizeof(*nd);
 452                 offset = 0;
 453                 while (icmp_len >= 8) {
 454                         struct nd_opt_hdr *nd_opt =
 455                                  (struct nd_opt_hdr *)(nd->opt + offset);
 456                         int opt_len = nd_opt->nd_opt_len * 8;
 457
 458                         if (unlikely(!opt_len || opt_len > icmp_len))
 459                                 return 0;
 460
 461                         /* Store the link layer address if the appropriate
 462                          * option is provided.  It is considered an error if
 463                          * the same link layer option is specified twice.
 464                          */
 465                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 466                             && opt_len == 8) {
 467                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 468                                         goto invalid;
 469                                 ether_addr_copy(key->ipv6.nd.sll,
 470                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 471                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 472                                    && opt_len == 8) {
 473                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 474                                         goto invalid;
 475                                 ether_addr_copy(key->ipv6.nd.tll,
 476                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 477                         }
 478
 479                         icmp_len -= opt_len;
 480                         offset += opt_len;
 481                 }
 482         }
 483
 484         return 0;
 485
 486 invalid:
 487         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 488         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 489         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 490
 491         return 0;
 492 }
 493
 494 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 495 {
 496         struct nshhdr *nh;
 497         unsigned int nh_ofs = skb_network_offset(skb);
 498         u8 version, length;
 499         int err;
 500
 501         err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
 502         if (unlikely(err))
 503                 return err;
 504
 505         nh = nsh_hdr(skb);
 506         version = nsh_get_ver(nh);
 507         length = nsh_hdr_len(nh);
 508
 509         if (version != 0)
 510                 return -EINVAL;
 511
 512         err = check_header(skb, nh_ofs + length);
 513         if (unlikely(err))
 514                 return err;
 515
 516         nh = nsh_hdr(skb);
 517         key->nsh.base.flags = nsh_get_flags(nh);
 518         key->nsh.base.ttl = nsh_get_ttl(nh);
 519         key->nsh.base.mdtype = nh->mdtype;
 520         key->nsh.base.np = nh->np;
 521         key->nsh.base.path_hdr = nh->path_hdr;
 522         switch (key->nsh.base.mdtype) {
 523         case NSH_M_TYPE1:
 524                 if (length != NSH_M_TYPE1_LEN)
 525                         return -EINVAL;
 526                 memcpy(key->nsh.context, nh->md1.context,
 527                        sizeof(nh->md1));
 528                 break;
 529         case NSH_M_TYPE2:
 530                 memset(key->nsh.context, 0,
 531                        sizeof(nh->md1));
 532                 break;
 533         default:
 534                 return -EINVAL;
 535         }
 536
 537         return 0;
 538 }
 539
 540 /**
 541  * key_extract - extracts a flow key from an Ethernet frame.
 542  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 543  * Ethernet header
 544  * @key: output flow key
 545  *
 546  * The caller must ensure that skb->len >= ETH_HLEN.
 547  *
 548  * Returns 0 if successful, otherwise a negative errno value.
 549  *
 550  * Initializes @skb header fields as follows:
 551  *
 552  *    - skb->mac_header: the L2 header.
 553  *
 554  *    - skb->network_header: just past the L2 header, or just past the
 555  *      VLAN header, to the first byte of the L2 payload.
 556  *
 557  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 558  *      on output, then just past the IP header, if one is present and
 559  *      of a correct length, otherwise the same as skb->network_header.
 560  *      For other key->eth.type values it is left untouched.
 561  *
 562  *    - skb->protocol: the type of the data starting at skb->network_header.
 563  *      Equals to key->eth.type.
 564  */
 565 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 566 {
 567         int error;
 568         struct ethhdr *eth;
 569
 570         /* Flags are always used as part of stats */
 571         key->tp.flags = 0;
 572
 573         skb_reset_mac_header(skb);
 574
 575         /* Link layer. */
 576         clear_vlan(key);
 577         if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
 578                 if (unlikely(eth_type_vlan(skb->protocol)))
 579                         return -EINVAL;
 580
 581                 skb_reset_network_header(skb);
 582         } else {
 583                 eth = eth_hdr(skb);
 584                 ether_addr_copy(key->eth.src, eth->h_source);
 585                 ether_addr_copy(key->eth.dst, eth->h_dest);
 586
 587                 __skb_pull(skb, 2 * ETH_ALEN);
 588                 /* We are going to push all headers that we pull, so no need to
 589                 * update skb->csum here.
 590                 */
 591
 592                 if (unlikely(parse_vlan(skb, key)))
 593                         return -ENOMEM;
 594
 595                 skb->protocol = parse_ethertype(skb);
 596                 if (unlikely(skb->protocol == htons(0)))
 597                         return -ENOMEM;
 598
 599                 skb_reset_network_header(skb);
 600                 __skb_push(skb, skb->data - skb_mac_header(skb));
 601         }
 602         skb_reset_mac_len(skb);
 603         key->eth.type = skb->protocol;
 604
 605         /* Network layer. */
 606         if (key->eth.type == htons(ETH_P_IP)) {
 607                 struct iphdr *nh;
 608                 __be16 offset;
 609
 610                 error = check_iphdr(skb);
 611                 if (unlikely(error)) {
 612                         memset(&key->ip, 0, sizeof(key->ip));
 613                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 614                         if (error == -EINVAL) {
 615                                 skb->transport_header = skb->network_header;
 616                                 error = 0;
 617                         }
 618                         return error;
 619                 }
 620
 621                 nh = ip_hdr(skb);
 622                 key->ipv4.addr.src = nh->saddr;
 623                 key->ipv4.addr.dst = nh->daddr;
 624
 625                 key->ip.proto = nh->protocol;
 626                 key->ip.tos = nh->tos;
 627                 key->ip.ttl = nh->ttl;
 628
 629                 offset = nh->frag_off & htons(IP_OFFSET);
 630                 if (offset) {
 631                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 632                         return 0;
 633                 }
 634                 if (nh->frag_off & htons(IP_MF) ||
 635                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 636                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 637                 else
 638                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 639
 640                 /* Transport layer. */
 641                 if (key->ip.proto == IPPROTO_TCP) {
 642                         if (tcphdr_ok(skb)) {
 643                                 struct tcphdr *tcp = tcp_hdr(skb);
 644                                 key->tp.src = tcp->source;
 645                                 key->tp.dst = tcp->dest;
 646                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 647                         } else {
 648                                 memset(&key->tp, 0, sizeof(key->tp));
 649                         }
 650
 651                 } else if (key->ip.proto == IPPROTO_UDP) {
 652                         if (udphdr_ok(skb)) {
 653                                 struct udphdr *udp = udp_hdr(skb);
 654                                 key->tp.src = udp->source;
 655                                 key->tp.dst = udp->dest;
 656                         } else {
 657                                 memset(&key->tp, 0, sizeof(key->tp));
 658                         }
 659                 } else if (key->ip.proto == IPPROTO_SCTP) {
 660                         if (sctphdr_ok(skb)) {
 661                                 struct sctphdr *sctp = sctp_hdr(skb);
 662                                 key->tp.src = sctp->source;
 663                                 key->tp.dst = sctp->dest;
 664                         } else {
 665                                 memset(&key->tp, 0, sizeof(key->tp));
 666                         }
 667                 } else if (key->ip.proto == IPPROTO_ICMP) {
 668                         if (icmphdr_ok(skb)) {
 669                                 struct icmphdr *icmp = icmp_hdr(skb);
 670                                 /* The ICMP type and code fields use the 16-bit
 671                                  * transport port fields, so we need to store
 672                                  * them in 16-bit network byte order. */
 673                                 key->tp.src = htons(icmp->type);
 674                                 key->tp.dst = htons(icmp->code);
 675                         } else {
 676                                 memset(&key->tp, 0, sizeof(key->tp));
 677                         }
 678                 }
 679
 680         } else if (key->eth.type == htons(ETH_P_ARP) ||
 681                    key->eth.type == htons(ETH_P_RARP)) {
 682                 struct arp_eth_header *arp;
 683                 bool arp_available = arphdr_ok(skb);
 684
 685                 arp = (struct arp_eth_header *)skb_network_header(skb);
 686
 687                 if (arp_available &&
 688                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 689                     arp->ar_pro == htons(ETH_P_IP) &&
 690                     arp->ar_hln == ETH_ALEN &&
 691                     arp->ar_pln == 4) {
 692
 693                         /* We only match on the lower 8 bits of the opcode. */
 694                         if (ntohs(arp->ar_op) <= 0xff)
 695                                 key->ip.proto = ntohs(arp->ar_op);
 696                         else
 697                                 key->ip.proto = 0;
 698
 699                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 700                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 701                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 702                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 703                 } else {
 704                         memset(&key->ip, 0, sizeof(key->ip));
 705                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 706                 }
 707         } else if (eth_p_mpls(key->eth.type)) {
 708                 size_t stack_len = MPLS_HLEN;
 709
 710                 skb_set_inner_network_header(skb, skb->mac_len);
 711                 while (1) {
 712                         __be32 lse;
 713
 714                         error = check_header(skb, skb->mac_len + stack_len);
 715                         if (unlikely(error))
 716                                 return 0;
 717
 718                         memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
 719
 720                         if (stack_len == MPLS_HLEN)
 721                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
 722
 723                         skb_set_inner_network_header(skb, skb->mac_len + stack_len);
 724                         if (lse & htonl(MPLS_LS_S_MASK))
 725                                 break;
 726
 727                         stack_len += MPLS_HLEN;
 728                 }
 729         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 730                 int nh_len;             /* IPv6 Header + Extensions */
 731
 732                 nh_len = parse_ipv6hdr(skb, key);
 733                 if (unlikely(nh_len < 0)) {
 734                         switch (nh_len) {
 735                         case -EINVAL:
 736                                 memset(&key->ip, 0, sizeof(key->ip));
 737                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 738                                 /* fall-through */
 739                         case -EPROTO:
 740                                 skb->transport_header = skb->network_header;
 741                                 error = 0;
 742                                 break;
 743                         default:
 744                                 error = nh_len;
 745                         }
 746                         return error;
 747                 }
 748
 749                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 750                         return 0;
 751                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 752                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 753
 754                 /* Transport layer. */
 755                 if (key->ip.proto == NEXTHDR_TCP) {
 756                         if (tcphdr_ok(skb)) {
 757                                 struct tcphdr *tcp = tcp_hdr(skb);
 758                                 key->tp.src = tcp->source;
 759                                 key->tp.dst = tcp->dest;
 760                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 761                         } else {
 762                                 memset(&key->tp, 0, sizeof(key->tp));
 763                         }
 764                 } else if (key->ip.proto == NEXTHDR_UDP) {
 765                         if (udphdr_ok(skb)) {
 766                                 struct udphdr *udp = udp_hdr(skb);
 767                                 key->tp.src = udp->source;
 768                                 key->tp.dst = udp->dest;
 769                         } else {
 770                                 memset(&key->tp, 0, sizeof(key->tp));
 771                         }
 772                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 773                         if (sctphdr_ok(skb)) {
 774                                 struct sctphdr *sctp = sctp_hdr(skb);
 775                                 key->tp.src = sctp->source;
 776                                 key->tp.dst = sctp->dest;
 777                         } else {
 778                                 memset(&key->tp, 0, sizeof(key->tp));
 779                         }
 780                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 781                         if (icmp6hdr_ok(skb)) {
 782                                 error = parse_icmpv6(skb, key, nh_len);
 783                                 if (error)
 784                                         return error;
 785                         } else {
 786                                 memset(&key->tp, 0, sizeof(key->tp));
 787                         }
 788                 }
 789         } else if (key->eth.type == htons(ETH_P_NSH)) {
 790                 error = parse_nsh(skb, key);
 791                 if (error)
 792                         return error;
 793         }
 794         return 0;
 795 }
 796
 797 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 798 {
 799         int res;
 800
 801         res = key_extract(skb, key);
 802         if (!res)
 803                 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
 804
 805         return res;
 806 }
 807
 808 static int key_extract_mac_proto(struct sk_buff *skb)
 809 {
 810         switch (skb->dev->type) {
 811         case ARPHRD_ETHER:
 812                 return MAC_PROTO_ETHERNET;
 813         case ARPHRD_NONE:
 814                 if (skb->protocol == htons(ETH_P_TEB))
 815                         return MAC_PROTO_ETHERNET;
 816                 return MAC_PROTO_NONE;
 817         }
 818         WARN_ON_ONCE(1);
 819         return -EINVAL;
 820 }
 821
 822 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
 823                          struct sk_buff *skb, struct sw_flow_key *key)
 824 {
 825         int res, err;
 826
 827         /* Extract metadata from packet. */
 828         if (tun_info) {
 829                 key->tun_proto = ip_tunnel_info_af(tun_info);
 830                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
 831
 832                 if (tun_info->options_len) {
 833                         BUILD_BUG_ON((1 << (sizeof(tun_info->options_len) *
 834                                                    8)) - 1
 835                                         > sizeof(key->tun_opts));
 836
 837                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
 838                                                 tun_info);
 839                         key->tun_opts_len = tun_info->options_len;
 840                 } else {
 841                         key->tun_opts_len = 0;
 842                 }
 843         } else  {
 844                 key->tun_proto = 0;
 845                 key->tun_opts_len = 0;
 846                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 847         }
 848
 849         key->phy.priority = skb->priority;
 850         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 851         key->phy.skb_mark = skb->mark;
 852         key->ovs_flow_hash = 0;
 853         res = key_extract_mac_proto(skb);
 854         if (res < 0)
 855                 return res;
 856         key->mac_proto = res;
 857         key->recirc_id = 0;
 858
 859         err = key_extract(skb, key);
 860         if (!err)
 861                 ovs_ct_fill_key(skb, key);   /* Must be after key_extract(). */
 862         return err;
 863 }
 864
 865 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
 866                                    struct sk_buff *skb,
 867                                    struct sw_flow_key *key, bool log)
 868 {
 869         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 870         u64 attrs = 0;
 871         int err;
 872
 873         err = parse_flow_nlattrs(attr, a, &attrs, log);
 874         if (err)
 875                 return -EINVAL;
 876
 877         /* Extract metadata from netlink attributes. */
 878         err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
 879         if (err)
 880                 return err;
 881
 882         /* key_extract assumes that skb->protocol is set-up for
 883          * layer 3 packets which is the case for other callers,
 884          * in particular packets received from the network stack.
 885          * Here the correct value can be set from the metadata
 886          * extracted above.
 887          * For L2 packet key eth type would be zero. skb protocol
 888          * would be set to correct value later during key-extact.
 889          */
 890
 891         skb->protocol = key->eth.type;
 892         err = key_extract(skb, key);
 893         if (err)
 894                 return err;
 895
 896         /* Check that we have conntrack original direction tuple metadata only
 897          * for packets for which it makes sense.  Otherwise the key may be
 898          * corrupted due to overlapping key fields.
 899          */
 900         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
 901             key->eth.type != htons(ETH_P_IP))
 902                 return -EINVAL;
 903         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
 904             (key->eth.type != htons(ETH_P_IPV6) ||
 905              sw_flow_key_is_nd(key)))
 906                 return -EINVAL;
 907
 908         return 0;
 909 }