datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2015 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 <linux/timekeeping.h>
  45 #include <net/ip.h>
  46 #include <net/ipv6.h>
  47 #include <net/mpls.h>
  48 #include <net/ndisc.h>
  49 #include <net/nsh.h>
  50
  51 #include "datapath.h"
  52 #include "conntrack.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 timespec64 cur_ts;
  60         u64 cur_ms, idle_ms;
  61
  62         ktime_get_ts64(&cur_ts);
  63         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
  64         cur_ms = (u64)(u32)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         key->eth.vlan.tci = 0;
 365         key->eth.vlan.tpid = 0;
 366         key->eth.cvlan.tci = 0;
 367         key->eth.cvlan.tpid = 0;
 368
 369         if (skb_vlan_tag_present(skb)) {
 370                 key->eth.vlan.tci = htons(skb->vlan_tci);
 371                 key->eth.vlan.tpid = skb->vlan_proto;
 372         } else {
 373                 /* Parse outer vlan tag in the non-accelerated case. */
 374                 res = parse_vlan_tag(skb, &key->eth.vlan, true);
 375                 if (res <= 0)
 376                         return res;
 377         }
 378
 379         /* Parse inner vlan tag. */
 380         res = parse_vlan_tag(skb, &key->eth.cvlan, false);
 381         if (res <= 0)
 382                 return res;
 383
 384         return 0;
 385 }
 386
 387 static __be16 parse_ethertype(struct sk_buff *skb)
 388 {
 389         struct llc_snap_hdr {
 390                 u8  dsap;  /* Always 0xAA */
 391                 u8  ssap;  /* Always 0xAA */
 392                 u8  ctrl;
 393                 u8  oui[3];
 394                 __be16 ethertype;
 395         };
 396         struct llc_snap_hdr *llc;
 397         __be16 proto;
 398
 399         proto = *(__be16 *) skb->data;
 400         __skb_pull(skb, sizeof(__be16));
 401
 402         if (eth_proto_is_802_3(proto))
 403                 return proto;
 404
 405         if (skb->len < sizeof(struct llc_snap_hdr))
 406                 return htons(ETH_P_802_2);
 407
 408         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 409                 return htons(0);
 410
 411         llc = (struct llc_snap_hdr *) skb->data;
 412         if (llc->dsap != LLC_SAP_SNAP ||
 413             llc->ssap != LLC_SAP_SNAP ||
 414             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 415                 return htons(ETH_P_802_2);
 416
 417         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 418
 419         if (eth_proto_is_802_3(llc->ethertype))
 420                 return llc->ethertype;
 421
 422         return htons(ETH_P_802_2);
 423 }
 424
 425 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 426                         int nh_len)
 427 {
 428         struct icmp6hdr *icmp = icmp6_hdr(skb);
 429
 430         /* The ICMPv6 type and code fields use the 16-bit transport port
 431          * fields, so we need to store them in 16-bit network byte order.
 432          */
 433         key->tp.src = htons(icmp->icmp6_type);
 434         key->tp.dst = htons(icmp->icmp6_code);
 435         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 436
 437         if (icmp->icmp6_code == 0 &&
 438             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 439              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 440                 int icmp_len = skb->len - skb_transport_offset(skb);
 441                 struct nd_msg *nd;
 442                 int offset;
 443
 444                 /* In order to process neighbor discovery options, we need the
 445                  * entire packet.
 446                  */
 447                 if (unlikely(icmp_len < sizeof(*nd)))
 448                         return 0;
 449
 450                 if (unlikely(skb_linearize(skb)))
 451                         return -ENOMEM;
 452
 453                 nd = (struct nd_msg *)skb_transport_header(skb);
 454                 key->ipv6.nd.target = nd->target;
 455
 456                 icmp_len -= sizeof(*nd);
 457                 offset = 0;
 458                 while (icmp_len >= 8) {
 459                         struct nd_opt_hdr *nd_opt =
 460                                  (struct nd_opt_hdr *)(nd->opt + offset);
 461                         int opt_len = nd_opt->nd_opt_len * 8;
 462
 463                         if (unlikely(!opt_len || opt_len > icmp_len))
 464                                 return 0;
 465
 466                         /* Store the link layer address if the appropriate
 467                          * option is provided.  It is considered an error if
 468                          * the same link layer option is specified twice.
 469                          */
 470                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 471                             && opt_len == 8) {
 472                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 473                                         goto invalid;
 474                                 ether_addr_copy(key->ipv6.nd.sll,
 475                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 476                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 477                                    && opt_len == 8) {
 478                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 479                                         goto invalid;
 480                                 ether_addr_copy(key->ipv6.nd.tll,
 481                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 482                         }
 483
 484                         icmp_len -= opt_len;
 485                         offset += opt_len;
 486                 }
 487         }
 488
 489         return 0;
 490
 491 invalid:
 492         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 493         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 494         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 495
 496         return 0;
 497 }
 498
 499 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 500 {
 501         struct nshhdr *nh;
 502         unsigned int nh_ofs = skb_network_offset(skb);
 503         u8 version, length;
 504         int err;
 505
 506         err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
 507         if (unlikely(err))
 508                 return err;
 509
 510         nh = nsh_hdr(skb);
 511         version = nsh_get_ver(nh);
 512         length = nsh_hdr_len(nh);
 513
 514         if (version != 0)
 515                 return -EINVAL;
 516
 517         err = check_header(skb, nh_ofs + length);
 518         if (unlikely(err))
 519                 return err;
 520
 521         nh = nsh_hdr(skb);
 522         key->nsh.base.flags = nsh_get_flags(nh);
 523         key->nsh.base.ttl = nsh_get_ttl(nh);
 524         key->nsh.base.mdtype = nh->mdtype;
 525         key->nsh.base.np = nh->np;
 526         key->nsh.base.path_hdr = nh->path_hdr;
 527         switch (key->nsh.base.mdtype) {
 528         case NSH_M_TYPE1:
 529                 if (length != NSH_M_TYPE1_LEN)
 530                         return -EINVAL;
 531                 memcpy(key->nsh.context, nh->md1.context,
 532                        sizeof(nh->md1));
 533                 break;
 534         case NSH_M_TYPE2:
 535                 memset(key->nsh.context, 0,
 536                        sizeof(nh->md1));
 537                 break;
 538         default:
 539                 return -EINVAL;
 540         }
 541
 542         return 0;
 543 }
 544
 545 /**
 546  * key_extract - extracts a flow key from an Ethernet frame.
 547  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 548  * Ethernet header
 549  * @key: output flow key
 550  *
 551  * The caller must ensure that skb->len >= ETH_HLEN.
 552  *
 553  * Returns 0 if successful, otherwise a negative errno value.
 554  *
 555  * Initializes @skb header fields as follows:
 556  *
 557  *    - skb->mac_header: the L2 header.
 558  *
 559  *    - skb->network_header: just past the L2 header, or just past the
 560  *      VLAN header, to the first byte of the L2 payload.
 561  *
 562  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 563  *      on output, then just past the IP header, if one is present and
 564  *      of a correct length, otherwise the same as skb->network_header.
 565  *      For other key->eth.type values it is left untouched.
 566  *
 567  *    - skb->protocol: the type of the data starting at skb->network_header.
 568  *      Equals to key->eth.type.
 569  */
 570 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 571 {
 572         int error;
 573         struct ethhdr *eth;
 574
 575         /* Flags are always used as part of stats */
 576         key->tp.flags = 0;
 577
 578         skb_reset_mac_header(skb);
 579
 580         /* Link layer. */
 581         clear_vlan(key);
 582         if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
 583                 if (unlikely(eth_type_vlan(skb->protocol)))
 584                         return -EINVAL;
 585
 586                 skb_reset_network_header(skb);
 587                 key->eth.type = skb->protocol;
 588         } else {
 589                 eth = eth_hdr(skb);
 590                 ether_addr_copy(key->eth.src, eth->h_source);
 591                 ether_addr_copy(key->eth.dst, eth->h_dest);
 592
 593                 __skb_pull(skb, 2 * ETH_ALEN);
 594                 /* We are going to push all headers that we pull, so no need to
 595                  * update skb->csum here.
 596                  */
 597
 598                 if (unlikely(parse_vlan(skb, key)))
 599                         return -ENOMEM;
 600
 601                 key->eth.type = parse_ethertype(skb);
 602                 if (unlikely(key->eth.type == htons(0)))
 603                         return -ENOMEM;
 604
 605                 /* Multiple tagged packets need to retain TPID to satisfy
 606                  * skb_vlan_pop(), which will later shift the ethertype into
 607                  * skb->protocol.
 608                  */
 609                 if (key->eth.cvlan.tci & htons(VLAN_TAG_PRESENT))
 610                         skb->protocol = key->eth.cvlan.tpid;
 611                 else
 612                         skb->protocol = key->eth.type;
 613
 614                 skb_reset_network_header(skb);
 615                 __skb_push(skb, skb->data - skb_mac_header(skb));
 616         }
 617
 618         skb_reset_mac_len(skb);
 619
 620         /* Network layer. */
 621         if (key->eth.type == htons(ETH_P_IP)) {
 622                 struct iphdr *nh;
 623                 __be16 offset;
 624
 625                 error = check_iphdr(skb);
 626                 if (unlikely(error)) {
 627                         memset(&key->ip, 0, sizeof(key->ip));
 628                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 629                         if (error == -EINVAL) {
 630                                 skb->transport_header = skb->network_header;
 631                                 error = 0;
 632                         }
 633                         return error;
 634                 }
 635
 636                 nh = ip_hdr(skb);
 637                 key->ipv4.addr.src = nh->saddr;
 638                 key->ipv4.addr.dst = nh->daddr;
 639
 640                 key->ip.proto = nh->protocol;
 641                 key->ip.tos = nh->tos;
 642                 key->ip.ttl = nh->ttl;
 643
 644                 offset = nh->frag_off & htons(IP_OFFSET);
 645                 if (offset) {
 646                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 647                         return 0;
 648                 }
 649 #ifdef HAVE_SKB_GSO_UDP
 650                 if (nh->frag_off & htons(IP_MF) ||
 651                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 652 #else
 653                 if (nh->frag_off & htons(IP_MF))
 654 #endif
 655                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 656                 else
 657                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 658
 659                 /* Transport layer. */
 660                 if (key->ip.proto == IPPROTO_TCP) {
 661                         if (tcphdr_ok(skb)) {
 662                                 struct tcphdr *tcp = tcp_hdr(skb);
 663                                 key->tp.src = tcp->source;
 664                                 key->tp.dst = tcp->dest;
 665                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 666                         } else {
 667                                 memset(&key->tp, 0, sizeof(key->tp));
 668                         }
 669
 670                 } else if (key->ip.proto == IPPROTO_UDP) {
 671                         if (udphdr_ok(skb)) {
 672                                 struct udphdr *udp = udp_hdr(skb);
 673                                 key->tp.src = udp->source;
 674                                 key->tp.dst = udp->dest;
 675                         } else {
 676                                 memset(&key->tp, 0, sizeof(key->tp));
 677                         }
 678                 } else if (key->ip.proto == IPPROTO_SCTP) {
 679                         if (sctphdr_ok(skb)) {
 680                                 struct sctphdr *sctp = sctp_hdr(skb);
 681                                 key->tp.src = sctp->source;
 682                                 key->tp.dst = sctp->dest;
 683                         } else {
 684                                 memset(&key->tp, 0, sizeof(key->tp));
 685                         }
 686                 } else if (key->ip.proto == IPPROTO_ICMP) {
 687                         if (icmphdr_ok(skb)) {
 688                                 struct icmphdr *icmp = icmp_hdr(skb);
 689                                 /* The ICMP type and code fields use the 16-bit
 690                                  * transport port fields, so we need to store
 691                                  * them in 16-bit network byte order.
 692                                  */
 693                                 key->tp.src = htons(icmp->type);
 694                                 key->tp.dst = htons(icmp->code);
 695                         } else {
 696                                 memset(&key->tp, 0, sizeof(key->tp));
 697                         }
 698                 }
 699
 700         } else if (key->eth.type == htons(ETH_P_ARP) ||
 701                    key->eth.type == htons(ETH_P_RARP)) {
 702                 struct arp_eth_header *arp;
 703                 bool arp_available = arphdr_ok(skb);
 704
 705                 arp = (struct arp_eth_header *)skb_network_header(skb);
 706
 707                 if (arp_available &&
 708                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 709                     arp->ar_pro == htons(ETH_P_IP) &&
 710                     arp->ar_hln == ETH_ALEN &&
 711                     arp->ar_pln == 4) {
 712
 713                         /* We only match on the lower 8 bits of the opcode. */
 714                         if (ntohs(arp->ar_op) <= 0xff)
 715                                 key->ip.proto = ntohs(arp->ar_op);
 716                         else
 717                                 key->ip.proto = 0;
 718
 719                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 720                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 721                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 722                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 723                 } else {
 724                         memset(&key->ip, 0, sizeof(key->ip));
 725                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 726                 }
 727         } else if (eth_p_mpls(key->eth.type)) {
 728                 size_t stack_len = MPLS_HLEN;
 729
 730                 skb_set_inner_network_header(skb, skb->mac_len);
 731                 while (1) {
 732                         __be32 lse;
 733
 734                         error = check_header(skb, skb->mac_len + stack_len);
 735                         if (unlikely(error))
 736                                 return 0;
 737
 738                         memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
 739
 740                         if (stack_len == MPLS_HLEN)
 741                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
 742
 743                         skb_set_inner_network_header(skb, skb->mac_len + stack_len);
 744                         if (lse & htonl(MPLS_LS_S_MASK))
 745                                 break;
 746
 747                         stack_len += MPLS_HLEN;
 748                 }
 749         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 750                 int nh_len;             /* IPv6 Header + Extensions */
 751
 752                 nh_len = parse_ipv6hdr(skb, key);
 753                 if (unlikely(nh_len < 0)) {
 754                         switch (nh_len) {
 755                         case -EINVAL:
 756                                 memset(&key->ip, 0, sizeof(key->ip));
 757                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 758                                 /* fall-through */
 759                         case -EPROTO:
 760                                 skb->transport_header = skb->network_header;
 761                                 error = 0;
 762                                 break;
 763                         default:
 764                                 error = nh_len;
 765                         }
 766                         return error;
 767                 }
 768
 769                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 770                         return 0;
 771 #ifdef HAVE_SKB_GSO_UDP
 772                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 773                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 774
 775 #endif
 776                 /* Transport layer. */
 777                 if (key->ip.proto == NEXTHDR_TCP) {
 778                         if (tcphdr_ok(skb)) {
 779                                 struct tcphdr *tcp = tcp_hdr(skb);
 780                                 key->tp.src = tcp->source;
 781                                 key->tp.dst = tcp->dest;
 782                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 783                         } else {
 784                                 memset(&key->tp, 0, sizeof(key->tp));
 785                         }
 786                 } else if (key->ip.proto == NEXTHDR_UDP) {
 787                         if (udphdr_ok(skb)) {
 788                                 struct udphdr *udp = udp_hdr(skb);
 789                                 key->tp.src = udp->source;
 790                                 key->tp.dst = udp->dest;
 791                         } else {
 792                                 memset(&key->tp, 0, sizeof(key->tp));
 793                         }
 794                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 795                         if (sctphdr_ok(skb)) {
 796                                 struct sctphdr *sctp = sctp_hdr(skb);
 797                                 key->tp.src = sctp->source;
 798                                 key->tp.dst = sctp->dest;
 799                         } else {
 800                                 memset(&key->tp, 0, sizeof(key->tp));
 801                         }
 802                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 803                         if (icmp6hdr_ok(skb)) {
 804                                 error = parse_icmpv6(skb, key, nh_len);
 805                                 if (error)
 806                                         return error;
 807                         } else {
 808                                 memset(&key->tp, 0, sizeof(key->tp));
 809                         }
 810                 }
 811         } else if (key->eth.type == htons(ETH_P_NSH)) {
 812                 error = parse_nsh(skb, key);
 813                 if (error)
 814                         return error;
 815         }
 816         return 0;
 817 }
 818
 819 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 820 {
 821         int res;
 822
 823         res = key_extract(skb, key);
 824         if (!res)
 825                 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
 826
 827         return res;
 828 }
 829
 830 static int key_extract_mac_proto(struct sk_buff *skb)
 831 {
 832         switch (skb->dev->type) {
 833         case ARPHRD_ETHER:
 834                 return MAC_PROTO_ETHERNET;
 835         case ARPHRD_NONE:
 836                 if (skb->protocol == htons(ETH_P_TEB))
 837                         return MAC_PROTO_ETHERNET;
 838                 return MAC_PROTO_NONE;
 839         }
 840         WARN_ON_ONCE(1);
 841         return -EINVAL;
 842 }
 843
 844 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
 845                          struct sk_buff *skb, struct sw_flow_key *key)
 846 {
 847         int res, err;
 848
 849         /* Extract metadata from packet. */
 850         if (tun_info) {
 851                 key->tun_proto = ip_tunnel_info_af(tun_info);
 852                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
 853                 BUILD_BUG_ON(((1 << (sizeof(tun_info->options_len) * 8)) - 1) >
 854                              sizeof(key->tun_opts));
 855
 856                 if (tun_info->options_len) {
 857                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
 858                                                 tun_info);
 859                         key->tun_opts_len = tun_info->options_len;
 860                 } else {
 861                         key->tun_opts_len = 0;
 862                 }
 863         } else  {
 864                 key->tun_proto = 0;
 865                 key->tun_opts_len = 0;
 866                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 867         }
 868
 869         key->phy.priority = skb->priority;
 870         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 871         key->phy.skb_mark = skb->mark;
 872         key->ovs_flow_hash = 0;
 873         res = key_extract_mac_proto(skb);
 874         if (res < 0)
 875                 return res;
 876         key->mac_proto = res;
 877         key->recirc_id = 0;
 878
 879         err = key_extract(skb, key);
 880         if (!err)
 881                 ovs_ct_fill_key(skb, key);   /* Must be after key_extract(). */
 882         return err;
 883 }
 884
 885 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
 886                                    struct sk_buff *skb,
 887                                    struct sw_flow_key *key, bool log)
 888 {
 889         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 890         u64 attrs = 0;
 891         int err;
 892
 893         err = parse_flow_nlattrs(attr, a, &attrs, log);
 894         if (err)
 895                 return -EINVAL;
 896
 897         /* Extract metadata from netlink attributes. */
 898         err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
 899         if (err)
 900                 return err;
 901
 902         /* key_extract assumes that skb->protocol is set-up for
 903          * layer 3 packets which is the case for other callers,
 904          * in particular packets received from the network stack.
 905          * Here the correct value can be set from the metadata
 906          * extracted above.
 907          * For L2 packet key eth type would be zero. skb protocol
 908          * would be set to correct value later during key-extact.
 909          */
 910
 911         skb->protocol = key->eth.type;
 912         err = key_extract(skb, key);
 913         if (err)
 914                 return err;
 915
 916         /* Check that we have conntrack original direction tuple metadata only
 917          * for packets for which it makes sense.  Otherwise the key may be
 918          * corrupted due to overlapping key fields.
 919          */
 920         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
 921             key->eth.type != htons(ETH_P_IP))
 922                 return -EINVAL;
 923         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
 924             (key->eth.type != htons(ETH_P_IPV6) ||
 925              sw_flow_key_is_nd(key)))
 926                 return -EINVAL;
 927
 928         return 0;
 929 }