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 short frag_off;
 258         unsigned int payload_ofs = 0;
 259         unsigned int nh_ofs = skb_network_offset(skb);
 260         unsigned int nh_len;
 261         struct ipv6hdr *nh;
 262         int err, nexthdr, flags = 0;
 263
 264         err = check_header(skb, nh_ofs + sizeof(*nh));
 265         if (unlikely(err))
 266                 return err;
 267
 268         nh = ipv6_hdr(skb);
 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         nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
 278         if (flags & IP6_FH_F_FRAG) {
 279                 if (frag_off)
 280                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 281                 else
 282                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 283         } else {
 284                 key->ip.frag = OVS_FRAG_TYPE_NONE;
 285         }
 286
 287         /* Delayed handling of error in ipv6_find_hdr() as it
 288          * always sets flags and frag_off to a valid value which may be
 289          * used to set key->ip.frag above.
 290          */
 291         if (unlikely(nexthdr < 0))
 292                 return -EPROTO;
 293
 294         nh_len = payload_ofs - nh_ofs;
 295         skb_set_transport_header(skb, nh_ofs + nh_len);
 296         key->ip.proto = nexthdr;
 297         return nh_len;
 298 }
 299
 300 static bool icmp6hdr_ok(struct sk_buff *skb)
 301 {
 302         return pskb_may_pull(skb, skb_transport_offset(skb) +
 303                                   sizeof(struct icmp6hdr));
 304 }
 305
 306 /**
 307  * Parse vlan tag from vlan header.
 308  * Returns ERROR on memory error.
 309  * Returns 0 if it encounters a non-vlan or incomplete packet.
 310  * Returns 1 after successfully parsing vlan tag.
 311  */
 312 static int parse_vlan_tag(struct sk_buff *skb, struct vlan_head *key_vh,
 313                           bool untag_vlan)
 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         if (unlikely(untag_vlan)) {
 332                 int offset = skb->data - skb_mac_header(skb);
 333                 u16 tci;
 334                 int err;
 335
 336                 __skb_push(skb, offset);
 337                 err = __skb_vlan_pop(skb, &tci);
 338                 __skb_pull(skb, offset);
 339                 if (err)
 340                         return err;
 341                 __vlan_hwaccel_put_tag(skb, key_vh->tpid, tci);
 342         } else {
 343                 __skb_pull(skb, sizeof(struct vlan_head));
 344         }
 345         return 1;
 346 }
 347
 348 static void clear_vlan(struct sw_flow_key *key)
 349 {
 350         key->eth.vlan.tci = 0;
 351         key->eth.vlan.tpid = 0;
 352         key->eth.cvlan.tci = 0;
 353         key->eth.cvlan.tpid = 0;
 354 }
 355
 356 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 357 {
 358         int res;
 359
 360         key->eth.vlan.tci = 0;
 361         key->eth.vlan.tpid = 0;
 362         key->eth.cvlan.tci = 0;
 363         key->eth.cvlan.tpid = 0;
 364
 365         if (skb_vlan_tag_present(skb)) {
 366                 key->eth.vlan.tci = htons(skb->vlan_tci);
 367                 key->eth.vlan.tpid = skb->vlan_proto;
 368         } else {
 369                 /* Parse outer vlan tag in the non-accelerated case. */
 370                 res = parse_vlan_tag(skb, &key->eth.vlan, true);
 371                 if (res <= 0)
 372                         return res;
 373         }
 374
 375         /* Parse inner vlan tag. */
 376         res = parse_vlan_tag(skb, &key->eth.cvlan, false);
 377         if (res <= 0)
 378                 return res;
 379
 380         return 0;
 381 }
 382
 383 static __be16 parse_ethertype(struct sk_buff *skb)
 384 {
 385         struct llc_snap_hdr {
 386                 u8  dsap;  /* Always 0xAA */
 387                 u8  ssap;  /* Always 0xAA */
 388                 u8  ctrl;
 389                 u8  oui[3];
 390                 __be16 ethertype;
 391         };
 392         struct llc_snap_hdr *llc;
 393         __be16 proto;
 394
 395         proto = *(__be16 *) skb->data;
 396         __skb_pull(skb, sizeof(__be16));
 397
 398         if (eth_proto_is_802_3(proto))
 399                 return proto;
 400
 401         if (skb->len < sizeof(struct llc_snap_hdr))
 402                 return htons(ETH_P_802_2);
 403
 404         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 405                 return htons(0);
 406
 407         llc = (struct llc_snap_hdr *) skb->data;
 408         if (llc->dsap != LLC_SAP_SNAP ||
 409             llc->ssap != LLC_SAP_SNAP ||
 410             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 411                 return htons(ETH_P_802_2);
 412
 413         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 414
 415         if (eth_proto_is_802_3(llc->ethertype))
 416                 return llc->ethertype;
 417
 418         return htons(ETH_P_802_2);
 419 }
 420
 421 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 422                         int nh_len)
 423 {
 424         struct icmp6hdr *icmp = icmp6_hdr(skb);
 425
 426         /* The ICMPv6 type and code fields use the 16-bit transport port
 427          * fields, so we need to store them in 16-bit network byte order.
 428          */
 429         key->tp.src = htons(icmp->icmp6_type);
 430         key->tp.dst = htons(icmp->icmp6_code);
 431         memset(&key->ipv6.nd, 0, sizeof(key->ipv6.nd));
 432
 433         if (icmp->icmp6_code == 0 &&
 434             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 435              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 436                 int icmp_len = skb->len - skb_transport_offset(skb);
 437                 struct nd_msg *nd;
 438                 int offset;
 439
 440                 /* In order to process neighbor discovery options, we need the
 441                  * entire packet.
 442                  */
 443                 if (unlikely(icmp_len < sizeof(*nd)))
 444                         return 0;
 445
 446                 if (unlikely(skb_linearize(skb)))
 447                         return -ENOMEM;
 448
 449                 nd = (struct nd_msg *)skb_transport_header(skb);
 450                 key->ipv6.nd.target = nd->target;
 451
 452                 icmp_len -= sizeof(*nd);
 453                 offset = 0;
 454                 while (icmp_len >= 8) {
 455                         struct nd_opt_hdr *nd_opt =
 456                                  (struct nd_opt_hdr *)(nd->opt + offset);
 457                         int opt_len = nd_opt->nd_opt_len * 8;
 458
 459                         if (unlikely(!opt_len || opt_len > icmp_len))
 460                                 return 0;
 461
 462                         /* Store the link layer address if the appropriate
 463                          * option is provided.  It is considered an error if
 464                          * the same link layer option is specified twice.
 465                          */
 466                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 467                             && opt_len == 8) {
 468                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 469                                         goto invalid;
 470                                 ether_addr_copy(key->ipv6.nd.sll,
 471                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 472                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 473                                    && opt_len == 8) {
 474                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 475                                         goto invalid;
 476                                 ether_addr_copy(key->ipv6.nd.tll,
 477                                                 &nd->opt[offset+sizeof(*nd_opt)]);
 478                         }
 479
 480                         icmp_len -= opt_len;
 481                         offset += opt_len;
 482                 }
 483         }
 484
 485         return 0;
 486
 487 invalid:
 488         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 489         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 490         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 491
 492         return 0;
 493 }
 494
 495 static int parse_nsh(struct sk_buff *skb, struct sw_flow_key *key)
 496 {
 497         struct nshhdr *nh;
 498         unsigned int nh_ofs = skb_network_offset(skb);
 499         u8 version, length;
 500         int err;
 501
 502         err = check_header(skb, nh_ofs + NSH_BASE_HDR_LEN);
 503         if (unlikely(err))
 504                 return err;
 505
 506         nh = nsh_hdr(skb);
 507         version = nsh_get_ver(nh);
 508         length = nsh_hdr_len(nh);
 509
 510         if (version != 0)
 511                 return -EINVAL;
 512
 513         err = check_header(skb, nh_ofs + length);
 514         if (unlikely(err))
 515                 return err;
 516
 517         nh = nsh_hdr(skb);
 518         key->nsh.base.flags = nsh_get_flags(nh);
 519         key->nsh.base.ttl = nsh_get_ttl(nh);
 520         key->nsh.base.mdtype = nh->mdtype;
 521         key->nsh.base.np = nh->np;
 522         key->nsh.base.path_hdr = nh->path_hdr;
 523         switch (key->nsh.base.mdtype) {
 524         case NSH_M_TYPE1:
 525                 if (length != NSH_M_TYPE1_LEN)
 526                         return -EINVAL;
 527                 memcpy(key->nsh.context, nh->md1.context,
 528                        sizeof(nh->md1));
 529                 break;
 530         case NSH_M_TYPE2:
 531                 memset(key->nsh.context, 0,
 532                        sizeof(nh->md1));
 533                 break;
 534         default:
 535                 return -EINVAL;
 536         }
 537
 538         return 0;
 539 }
 540
 541 /**
 542  * key_extract - extracts a flow key from an Ethernet frame.
 543  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 544  * Ethernet header
 545  * @key: output flow key
 546  *
 547  * The caller must ensure that skb->len >= ETH_HLEN.
 548  *
 549  * Returns 0 if successful, otherwise a negative errno value.
 550  *
 551  * Initializes @skb header fields as follows:
 552  *
 553  *    - skb->mac_header: the L2 header.
 554  *
 555  *    - skb->network_header: just past the L2 header, or just past the
 556  *      VLAN header, to the first byte of the L2 payload.
 557  *
 558  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 559  *      on output, then just past the IP header, if one is present and
 560  *      of a correct length, otherwise the same as skb->network_header.
 561  *      For other key->eth.type values it is left untouched.
 562  *
 563  *    - skb->protocol: the type of the data starting at skb->network_header.
 564  *      Equals to key->eth.type.
 565  */
 566 static int key_extract(struct sk_buff *skb, struct sw_flow_key *key)
 567 {
 568         int error;
 569         struct ethhdr *eth;
 570
 571         /* Flags are always used as part of stats */
 572         key->tp.flags = 0;
 573
 574         skb_reset_mac_header(skb);
 575
 576         /* Link layer. */
 577         clear_vlan(key);
 578         if (ovs_key_mac_proto(key) == MAC_PROTO_NONE) {
 579                 if (unlikely(eth_type_vlan(skb->protocol)))
 580                         return -EINVAL;
 581
 582                 skb_reset_network_header(skb);
 583                 key->eth.type = skb->protocol;
 584         } else {
 585                 eth = eth_hdr(skb);
 586                 ether_addr_copy(key->eth.src, eth->h_source);
 587                 ether_addr_copy(key->eth.dst, eth->h_dest);
 588
 589                 __skb_pull(skb, 2 * ETH_ALEN);
 590                 /* We are going to push all headers that we pull, so no need to
 591                  * update skb->csum here.
 592                  */
 593
 594                 if (unlikely(parse_vlan(skb, key)))
 595                         return -ENOMEM;
 596
 597                 key->eth.type = parse_ethertype(skb);
 598                 if (unlikely(key->eth.type == htons(0)))
 599                         return -ENOMEM;
 600
 601                 /* Multiple tagged packets need to retain TPID to satisfy
 602                  * skb_vlan_pop(), which will later shift the ethertype into
 603                  * skb->protocol.
 604                  */
 605                 if (key->eth.cvlan.tci & htons(VLAN_TAG_PRESENT))
 606                         skb->protocol = key->eth.cvlan.tpid;
 607                 else
 608                         skb->protocol = key->eth.type;
 609
 610                 skb_reset_network_header(skb);
 611                 __skb_push(skb, skb->data - skb_mac_header(skb));
 612         }
 613
 614         skb_reset_mac_len(skb);
 615
 616         /* Network layer. */
 617         if (key->eth.type == htons(ETH_P_IP)) {
 618                 struct iphdr *nh;
 619                 __be16 offset;
 620
 621                 error = check_iphdr(skb);
 622                 if (unlikely(error)) {
 623                         memset(&key->ip, 0, sizeof(key->ip));
 624                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 625                         if (error == -EINVAL) {
 626                                 skb->transport_header = skb->network_header;
 627                                 error = 0;
 628                         }
 629                         return error;
 630                 }
 631
 632                 nh = ip_hdr(skb);
 633                 key->ipv4.addr.src = nh->saddr;
 634                 key->ipv4.addr.dst = nh->daddr;
 635
 636                 key->ip.proto = nh->protocol;
 637                 key->ip.tos = nh->tos;
 638                 key->ip.ttl = nh->ttl;
 639
 640                 offset = nh->frag_off & htons(IP_OFFSET);
 641                 if (offset) {
 642                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 643                         return 0;
 644                 }
 645 #ifdef HAVE_SKB_GSO_UDP
 646                 if (nh->frag_off & htons(IP_MF) ||
 647                         skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 648 #else
 649                 if (nh->frag_off & htons(IP_MF))
 650 #endif
 651                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 652                 else
 653                         key->ip.frag = OVS_FRAG_TYPE_NONE;
 654
 655                 /* Transport layer. */
 656                 if (key->ip.proto == IPPROTO_TCP) {
 657                         if (tcphdr_ok(skb)) {
 658                                 struct tcphdr *tcp = tcp_hdr(skb);
 659                                 key->tp.src = tcp->source;
 660                                 key->tp.dst = tcp->dest;
 661                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 662                         } else {
 663                                 memset(&key->tp, 0, sizeof(key->tp));
 664                         }
 665
 666                 } else if (key->ip.proto == IPPROTO_UDP) {
 667                         if (udphdr_ok(skb)) {
 668                                 struct udphdr *udp = udp_hdr(skb);
 669                                 key->tp.src = udp->source;
 670                                 key->tp.dst = udp->dest;
 671                         } else {
 672                                 memset(&key->tp, 0, sizeof(key->tp));
 673                         }
 674                 } else if (key->ip.proto == IPPROTO_SCTP) {
 675                         if (sctphdr_ok(skb)) {
 676                                 struct sctphdr *sctp = sctp_hdr(skb);
 677                                 key->tp.src = sctp->source;
 678                                 key->tp.dst = sctp->dest;
 679                         } else {
 680                                 memset(&key->tp, 0, sizeof(key->tp));
 681                         }
 682                 } else if (key->ip.proto == IPPROTO_ICMP) {
 683                         if (icmphdr_ok(skb)) {
 684                                 struct icmphdr *icmp = icmp_hdr(skb);
 685                                 /* The ICMP type and code fields use the 16-bit
 686                                  * transport port fields, so we need to store
 687                                  * them in 16-bit network byte order.
 688                                  */
 689                                 key->tp.src = htons(icmp->type);
 690                                 key->tp.dst = htons(icmp->code);
 691                         } else {
 692                                 memset(&key->tp, 0, sizeof(key->tp));
 693                         }
 694                 }
 695
 696         } else if (key->eth.type == htons(ETH_P_ARP) ||
 697                    key->eth.type == htons(ETH_P_RARP)) {
 698                 struct arp_eth_header *arp;
 699                 bool arp_available = arphdr_ok(skb);
 700
 701                 arp = (struct arp_eth_header *)skb_network_header(skb);
 702
 703                 if (arp_available &&
 704                     arp->ar_hrd == htons(ARPHRD_ETHER) &&
 705                     arp->ar_pro == htons(ETH_P_IP) &&
 706                     arp->ar_hln == ETH_ALEN &&
 707                     arp->ar_pln == 4) {
 708
 709                         /* We only match on the lower 8 bits of the opcode. */
 710                         if (ntohs(arp->ar_op) <= 0xff)
 711                                 key->ip.proto = ntohs(arp->ar_op);
 712                         else
 713                                 key->ip.proto = 0;
 714
 715                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 716                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 717                         ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha);
 718                         ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha);
 719                 } else {
 720                         memset(&key->ip, 0, sizeof(key->ip));
 721                         memset(&key->ipv4, 0, sizeof(key->ipv4));
 722                 }
 723         } else if (eth_p_mpls(key->eth.type)) {
 724                 size_t stack_len = MPLS_HLEN;
 725
 726                 skb_set_inner_network_header(skb, skb->mac_len);
 727                 while (1) {
 728                         __be32 lse;
 729
 730                         error = check_header(skb, skb->mac_len + stack_len);
 731                         if (unlikely(error))
 732                                 return 0;
 733
 734                         memcpy(&lse, skb_inner_network_header(skb), MPLS_HLEN);
 735
 736                         if (stack_len == MPLS_HLEN)
 737                                 memcpy(&key->mpls.top_lse, &lse, MPLS_HLEN);
 738
 739                         skb_set_inner_network_header(skb, skb->mac_len + stack_len);
 740                         if (lse & htonl(MPLS_LS_S_MASK))
 741                                 break;
 742
 743                         stack_len += MPLS_HLEN;
 744                 }
 745         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 746                 int nh_len;             /* IPv6 Header + Extensions */
 747
 748                 nh_len = parse_ipv6hdr(skb, key);
 749                 if (unlikely(nh_len < 0)) {
 750                         switch (nh_len) {
 751                         case -EINVAL:
 752                                 memset(&key->ip, 0, sizeof(key->ip));
 753                                 memset(&key->ipv6.addr, 0, sizeof(key->ipv6.addr));
 754                                 /* fall-through */
 755                         case -EPROTO:
 756                                 skb->transport_header = skb->network_header;
 757                                 error = 0;
 758                                 break;
 759                         default:
 760                                 error = nh_len;
 761                         }
 762                         return error;
 763                 }
 764
 765                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 766                         return 0;
 767 #ifdef HAVE_SKB_GSO_UDP
 768                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 769                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 770
 771 #endif
 772                 /* Transport layer. */
 773                 if (key->ip.proto == NEXTHDR_TCP) {
 774                         if (tcphdr_ok(skb)) {
 775                                 struct tcphdr *tcp = tcp_hdr(skb);
 776                                 key->tp.src = tcp->source;
 777                                 key->tp.dst = tcp->dest;
 778                                 key->tp.flags = TCP_FLAGS_BE16(tcp);
 779                         } else {
 780                                 memset(&key->tp, 0, sizeof(key->tp));
 781                         }
 782                 } else if (key->ip.proto == NEXTHDR_UDP) {
 783                         if (udphdr_ok(skb)) {
 784                                 struct udphdr *udp = udp_hdr(skb);
 785                                 key->tp.src = udp->source;
 786                                 key->tp.dst = udp->dest;
 787                         } else {
 788                                 memset(&key->tp, 0, sizeof(key->tp));
 789                         }
 790                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 791                         if (sctphdr_ok(skb)) {
 792                                 struct sctphdr *sctp = sctp_hdr(skb);
 793                                 key->tp.src = sctp->source;
 794                                 key->tp.dst = sctp->dest;
 795                         } else {
 796                                 memset(&key->tp, 0, sizeof(key->tp));
 797                         }
 798                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 799                         if (icmp6hdr_ok(skb)) {
 800                                 error = parse_icmpv6(skb, key, nh_len);
 801                                 if (error)
 802                                         return error;
 803                         } else {
 804                                 memset(&key->tp, 0, sizeof(key->tp));
 805                         }
 806                 }
 807         } else if (key->eth.type == htons(ETH_P_NSH)) {
 808                 error = parse_nsh(skb, key);
 809                 if (error)
 810                         return error;
 811         }
 812         return 0;
 813 }
 814
 815 int ovs_flow_key_update(struct sk_buff *skb, struct sw_flow_key *key)
 816 {
 817         int res;
 818
 819         res = key_extract(skb, key);
 820         if (!res)
 821                 key->mac_proto &= ~SW_FLOW_KEY_INVALID;
 822
 823         return res;
 824 }
 825
 826 static int key_extract_mac_proto(struct sk_buff *skb)
 827 {
 828         switch (skb->dev->type) {
 829         case ARPHRD_ETHER:
 830                 return MAC_PROTO_ETHERNET;
 831         case ARPHRD_NONE:
 832                 if (skb->protocol == htons(ETH_P_TEB))
 833                         return MAC_PROTO_ETHERNET;
 834                 return MAC_PROTO_NONE;
 835         }
 836         WARN_ON_ONCE(1);
 837         return -EINVAL;
 838 }
 839
 840 int ovs_flow_key_extract(const struct ip_tunnel_info *tun_info,
 841                          struct sk_buff *skb, struct sw_flow_key *key)
 842 {
 843         int res, err;
 844
 845         /* Extract metadata from packet. */
 846         if (tun_info) {
 847                 key->tun_proto = ip_tunnel_info_af(tun_info);
 848                 memcpy(&key->tun_key, &tun_info->key, sizeof(key->tun_key));
 849                 BUILD_BUG_ON(((1 << (sizeof(tun_info->options_len) * 8)) - 1) >
 850                              sizeof(key->tun_opts));
 851
 852                 if (tun_info->options_len) {
 853                         ip_tunnel_info_opts_get(TUN_METADATA_OPTS(key, tun_info->options_len),
 854                                                 tun_info);
 855                         key->tun_opts_len = tun_info->options_len;
 856                 } else {
 857                         key->tun_opts_len = 0;
 858                 }
 859         } else  {
 860                 key->tun_proto = 0;
 861                 key->tun_opts_len = 0;
 862                 memset(&key->tun_key, 0, sizeof(key->tun_key));
 863         }
 864
 865         key->phy.priority = skb->priority;
 866         key->phy.in_port = OVS_CB(skb)->input_vport->port_no;
 867         key->phy.skb_mark = skb->mark;
 868         key->ovs_flow_hash = 0;
 869         res = key_extract_mac_proto(skb);
 870         if (res < 0)
 871                 return res;
 872         key->mac_proto = res;
 873         key->recirc_id = 0;
 874
 875         err = key_extract(skb, key);
 876         if (!err)
 877                 ovs_ct_fill_key(skb, key);   /* Must be after key_extract(). */
 878         return err;
 879 }
 880
 881 int ovs_flow_key_extract_userspace(struct net *net, const struct nlattr *attr,
 882                                    struct sk_buff *skb,
 883                                    struct sw_flow_key *key, bool log)
 884 {
 885         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 886         u64 attrs = 0;
 887         int err;
 888
 889         err = parse_flow_nlattrs(attr, a, &attrs, log);
 890         if (err)
 891                 return -EINVAL;
 892
 893         /* Extract metadata from netlink attributes. */
 894         err = ovs_nla_get_flow_metadata(net, a, attrs, key, log);
 895         if (err)
 896                 return err;
 897
 898         /* key_extract assumes that skb->protocol is set-up for
 899          * layer 3 packets which is the case for other callers,
 900          * in particular packets received from the network stack.
 901          * Here the correct value can be set from the metadata
 902          * extracted above.
 903          * For L2 packet key eth type would be zero. skb protocol
 904          * would be set to correct value later during key-extact.
 905          */
 906
 907         skb->protocol = key->eth.type;
 908         err = key_extract(skb, key);
 909         if (err)
 910                 return err;
 911
 912         /* Check that we have conntrack original direction tuple metadata only
 913          * for packets for which it makes sense.  Otherwise the key may be
 914          * corrupted due to overlapping key fields.
 915          */
 916         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4) &&
 917             key->eth.type != htons(ETH_P_IP))
 918                 return -EINVAL;
 919         if (attrs & (1 << OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6) &&
 920             (key->eth.type != htons(ETH_P_IPV6) ||
 921              sw_flow_key_is_nd(key)))
 922                 return -EINVAL;
 923
 924         return 0;
 925 }