Pileus Git - ~andy/linux/blob - net/openvswitch/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira Networks.
   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 "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 static struct kmem_cache *flow_cache;
  47
  48 static int check_header(struct sk_buff *skb, int len)
  49 {
  50         if (unlikely(skb->len < len))
  51                 return -EINVAL;
  52         if (unlikely(!pskb_may_pull(skb, len)))
  53                 return -ENOMEM;
  54         return 0;
  55 }
  56
  57 static bool arphdr_ok(struct sk_buff *skb)
  58 {
  59         return pskb_may_pull(skb, skb_network_offset(skb) +
  60                                   sizeof(struct arp_eth_header));
  61 }
  62
  63 static int check_iphdr(struct sk_buff *skb)
  64 {
  65         unsigned int nh_ofs = skb_network_offset(skb);
  66         unsigned int ip_len;
  67         int err;
  68
  69         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  70         if (unlikely(err))
  71                 return err;
  72
  73         ip_len = ip_hdrlen(skb);
  74         if (unlikely(ip_len < sizeof(struct iphdr) ||
  75                      skb->len < nh_ofs + ip_len))
  76                 return -EINVAL;
  77
  78         skb_set_transport_header(skb, nh_ofs + ip_len);
  79         return 0;
  80 }
  81
  82 static bool tcphdr_ok(struct sk_buff *skb)
  83 {
  84         int th_ofs = skb_transport_offset(skb);
  85         int tcp_len;
  86
  87         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  88                 return false;
  89
  90         tcp_len = tcp_hdrlen(skb);
  91         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  92                      skb->len < th_ofs + tcp_len))
  93                 return false;
  94
  95         return true;
  96 }
  97
  98 static bool udphdr_ok(struct sk_buff *skb)
  99 {
 100         return pskb_may_pull(skb, skb_transport_offset(skb) +
 101                                   sizeof(struct udphdr));
 102 }
 103
 104 static bool icmphdr_ok(struct sk_buff *skb)
 105 {
 106         return pskb_may_pull(skb, skb_transport_offset(skb) +
 107                                   sizeof(struct icmphdr));
 108 }
 109
 110 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 111 {
 112         struct timespec cur_ts;
 113         u64 cur_ms, idle_ms;
 114
 115         ktime_get_ts(&cur_ts);
 116         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 117         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 118                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 119
 120         return cur_ms - idle_ms;
 121 }
 122
 123 #define SW_FLOW_KEY_OFFSET(field)               \
 124         (offsetof(struct sw_flow_key, field) +  \
 125          FIELD_SIZEOF(struct sw_flow_key, field))
 126
 127 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 128                          int *key_lenp)
 129 {
 130         unsigned int nh_ofs = skb_network_offset(skb);
 131         unsigned int nh_len;
 132         int payload_ofs;
 133         struct ipv6hdr *nh;
 134         uint8_t nexthdr;
 135         __be16 frag_off;
 136         int err;
 137
 138         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 139
 140         err = check_header(skb, nh_ofs + sizeof(*nh));
 141         if (unlikely(err))
 142                 return err;
 143
 144         nh = ipv6_hdr(skb);
 145         nexthdr = nh->nexthdr;
 146         payload_ofs = (u8 *)(nh + 1) - skb->data;
 147
 148         key->ip.proto = NEXTHDR_NONE;
 149         key->ip.tos = ipv6_get_dsfield(nh);
 150         key->ip.ttl = nh->hop_limit;
 151         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 152         key->ipv6.addr.src = nh->saddr;
 153         key->ipv6.addr.dst = nh->daddr;
 154
 155         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 156         if (unlikely(payload_ofs < 0))
 157                 return -EINVAL;
 158
 159         if (frag_off) {
 160                 if (frag_off & htons(~0x7))
 161                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 162                 else
 163                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 164         }
 165
 166         nh_len = payload_ofs - nh_ofs;
 167         skb_set_transport_header(skb, nh_ofs + nh_len);
 168         key->ip.proto = nexthdr;
 169         return nh_len;
 170 }
 171
 172 static bool icmp6hdr_ok(struct sk_buff *skb)
 173 {
 174         return pskb_may_pull(skb, skb_transport_offset(skb) +
 175                                   sizeof(struct icmp6hdr));
 176 }
 177
 178 #define TCP_FLAGS_OFFSET 13
 179 #define TCP_FLAG_MASK 0x3f
 180
 181 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 182 {
 183         u8 tcp_flags = 0;
 184
 185         if (flow->key.eth.type == htons(ETH_P_IP) &&
 186             flow->key.ip.proto == IPPROTO_TCP) {
 187                 u8 *tcp = (u8 *)tcp_hdr(skb);
 188                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 189         }
 190
 191         spin_lock(&flow->lock);
 192         flow->used = jiffies;
 193         flow->packet_count++;
 194         flow->byte_count += skb->len;
 195         flow->tcp_flags |= tcp_flags;
 196         spin_unlock(&flow->lock);
 197 }
 198
 199 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 200 {
 201         int actions_len = nla_len(actions);
 202         struct sw_flow_actions *sfa;
 203
 204         /* At least DP_MAX_PORTS actions are required to be able to flood a
 205          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 206          * etc. */
 207         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 208                 return ERR_PTR(-EINVAL);
 209
 210         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 211         if (!sfa)
 212                 return ERR_PTR(-ENOMEM);
 213
 214         sfa->actions_len = actions_len;
 215         memcpy(sfa->actions, nla_data(actions), actions_len);
 216         return sfa;
 217 }
 218
 219 struct sw_flow *ovs_flow_alloc(void)
 220 {
 221         struct sw_flow *flow;
 222
 223         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 224         if (!flow)
 225                 return ERR_PTR(-ENOMEM);
 226
 227         spin_lock_init(&flow->lock);
 228         flow->sf_acts = NULL;
 229
 230         return flow;
 231 }
 232
 233 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 234 {
 235         hash = jhash_1word(hash, table->hash_seed);
 236         return flex_array_get(table->buckets,
 237                                 (hash & (table->n_buckets - 1)));
 238 }
 239
 240 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 241 {
 242         struct flex_array *buckets;
 243         int i, err;
 244
 245         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 246                                    n_buckets, GFP_KERNEL);
 247         if (!buckets)
 248                 return NULL;
 249
 250         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 251         if (err) {
 252                 flex_array_free(buckets);
 253                 return NULL;
 254         }
 255
 256         for (i = 0; i < n_buckets; i++)
 257                 INIT_HLIST_HEAD((struct hlist_head *)
 258                                         flex_array_get(buckets, i));
 259
 260         return buckets;
 261 }
 262
 263 static void free_buckets(struct flex_array *buckets)
 264 {
 265         flex_array_free(buckets);
 266 }
 267
 268 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 269 {
 270         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 271
 272         if (!table)
 273                 return NULL;
 274
 275         table->buckets = alloc_buckets(new_size);
 276
 277         if (!table->buckets) {
 278                 kfree(table);
 279                 return NULL;
 280         }
 281         table->n_buckets = new_size;
 282         table->count = 0;
 283         table->node_ver = 0;
 284         table->keep_flows = false;
 285         get_random_bytes(&table->hash_seed, sizeof(u32));
 286
 287         return table;
 288 }
 289
 290 void ovs_flow_tbl_destroy(struct flow_table *table)
 291 {
 292         int i;
 293
 294         if (!table)
 295                 return;
 296
 297         if (table->keep_flows)
 298                 goto skip_flows;
 299
 300         for (i = 0; i < table->n_buckets; i++) {
 301                 struct sw_flow *flow;
 302                 struct hlist_head *head = flex_array_get(table->buckets, i);
 303                 struct hlist_node *node, *n;
 304                 int ver = table->node_ver;
 305
 306                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 307                         hlist_del_rcu(&flow->hash_node[ver]);
 308                         ovs_flow_free(flow);
 309                 }
 310         }
 311
 312 skip_flows:
 313         free_buckets(table->buckets);
 314         kfree(table);
 315 }
 316
 317 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 318 {
 319         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 320
 321         ovs_flow_tbl_destroy(table);
 322 }
 323
 324 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 325 {
 326         if (!table)
 327                 return;
 328
 329         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 330 }
 331
 332 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 333 {
 334         struct sw_flow *flow;
 335         struct hlist_head *head;
 336         struct hlist_node *n;
 337         int ver;
 338         int i;
 339
 340         ver = table->node_ver;
 341         while (*bucket < table->n_buckets) {
 342                 i = 0;
 343                 head = flex_array_get(table->buckets, *bucket);
 344                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 345                         if (i < *last) {
 346                                 i++;
 347                                 continue;
 348                         }
 349                         *last = i + 1;
 350                         return flow;
 351                 }
 352                 (*bucket)++;
 353                 *last = 0;
 354         }
 355
 356         return NULL;
 357 }
 358
 359 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 360 {
 361         int old_ver;
 362         int i;
 363
 364         old_ver = old->node_ver;
 365         new->node_ver = !old_ver;
 366
 367         /* Insert in new table. */
 368         for (i = 0; i < old->n_buckets; i++) {
 369                 struct sw_flow *flow;
 370                 struct hlist_head *head;
 371                 struct hlist_node *n;
 372
 373                 head = flex_array_get(old->buckets, i);
 374
 375                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 376                         ovs_flow_tbl_insert(new, flow);
 377         }
 378         old->keep_flows = true;
 379 }
 380
 381 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 382 {
 383         struct flow_table *new_table;
 384
 385         new_table = ovs_flow_tbl_alloc(n_buckets);
 386         if (!new_table)
 387                 return ERR_PTR(-ENOMEM);
 388
 389         flow_table_copy_flows(table, new_table);
 390
 391         return new_table;
 392 }
 393
 394 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 395 {
 396         return __flow_tbl_rehash(table, table->n_buckets);
 397 }
 398
 399 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 400 {
 401         return __flow_tbl_rehash(table, table->n_buckets * 2);
 402 }
 403
 404 void ovs_flow_free(struct sw_flow *flow)
 405 {
 406         if (unlikely(!flow))
 407                 return;
 408
 409         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 410         kmem_cache_free(flow_cache, flow);
 411 }
 412
 413 /* RCU callback used by ovs_flow_deferred_free. */
 414 static void rcu_free_flow_callback(struct rcu_head *rcu)
 415 {
 416         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 417
 418         ovs_flow_free(flow);
 419 }
 420
 421 /* Schedules 'flow' to be freed after the next RCU grace period.
 422  * The caller must hold rcu_read_lock for this to be sensible. */
 423 void ovs_flow_deferred_free(struct sw_flow *flow)
 424 {
 425         call_rcu(&flow->rcu, rcu_free_flow_callback);
 426 }
 427
 428 /* RCU callback used by ovs_flow_deferred_free_acts. */
 429 static void rcu_free_acts_callback(struct rcu_head *rcu)
 430 {
 431         struct sw_flow_actions *sf_acts = container_of(rcu,
 432                         struct sw_flow_actions, rcu);
 433         kfree(sf_acts);
 434 }
 435
 436 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 437  * The caller must hold rcu_read_lock for this to be sensible. */
 438 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 439 {
 440         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 441 }
 442
 443 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 444 {
 445         struct qtag_prefix {
 446                 __be16 eth_type; /* ETH_P_8021Q */
 447                 __be16 tci;
 448         };
 449         struct qtag_prefix *qp;
 450
 451         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 452                 return 0;
 453
 454         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 455                                          sizeof(__be16))))
 456                 return -ENOMEM;
 457
 458         qp = (struct qtag_prefix *) skb->data;
 459         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 460         __skb_pull(skb, sizeof(struct qtag_prefix));
 461
 462         return 0;
 463 }
 464
 465 static __be16 parse_ethertype(struct sk_buff *skb)
 466 {
 467         struct llc_snap_hdr {
 468                 u8  dsap;  /* Always 0xAA */
 469                 u8  ssap;  /* Always 0xAA */
 470                 u8  ctrl;
 471                 u8  oui[3];
 472                 __be16 ethertype;
 473         };
 474         struct llc_snap_hdr *llc;
 475         __be16 proto;
 476
 477         proto = *(__be16 *) skb->data;
 478         __skb_pull(skb, sizeof(__be16));
 479
 480         if (ntohs(proto) >= 1536)
 481                 return proto;
 482
 483         if (skb->len < sizeof(struct llc_snap_hdr))
 484                 return htons(ETH_P_802_2);
 485
 486         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 487                 return htons(0);
 488
 489         llc = (struct llc_snap_hdr *) skb->data;
 490         if (llc->dsap != LLC_SAP_SNAP ||
 491             llc->ssap != LLC_SAP_SNAP ||
 492             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 493                 return htons(ETH_P_802_2);
 494
 495         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 496         return llc->ethertype;
 497 }
 498
 499 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 500                         int *key_lenp, int nh_len)
 501 {
 502         struct icmp6hdr *icmp = icmp6_hdr(skb);
 503         int error = 0;
 504         int key_len;
 505
 506         /* The ICMPv6 type and code fields use the 16-bit transport port
 507          * fields, so we need to store them in 16-bit network byte order.
 508          */
 509         key->ipv6.tp.src = htons(icmp->icmp6_type);
 510         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 511         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 512
 513         if (icmp->icmp6_code == 0 &&
 514             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 515              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 516                 int icmp_len = skb->len - skb_transport_offset(skb);
 517                 struct nd_msg *nd;
 518                 int offset;
 519
 520                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 521
 522                 /* In order to process neighbor discovery options, we need the
 523                  * entire packet.
 524                  */
 525                 if (unlikely(icmp_len < sizeof(*nd)))
 526                         goto out;
 527                 if (unlikely(skb_linearize(skb))) {
 528                         error = -ENOMEM;
 529                         goto out;
 530                 }
 531
 532                 nd = (struct nd_msg *)skb_transport_header(skb);
 533                 key->ipv6.nd.target = nd->target;
 534                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 535
 536                 icmp_len -= sizeof(*nd);
 537                 offset = 0;
 538                 while (icmp_len >= 8) {
 539                         struct nd_opt_hdr *nd_opt =
 540                                  (struct nd_opt_hdr *)(nd->opt + offset);
 541                         int opt_len = nd_opt->nd_opt_len * 8;
 542
 543                         if (unlikely(!opt_len || opt_len > icmp_len))
 544                                 goto invalid;
 545
 546                         /* Store the link layer address if the appropriate
 547                          * option is provided.  It is considered an error if
 548                          * the same link layer option is specified twice.
 549                          */
 550                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 551                             && opt_len == 8) {
 552                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 553                                         goto invalid;
 554                                 memcpy(key->ipv6.nd.sll,
 555                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 556                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 557                                    && opt_len == 8) {
 558                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 559                                         goto invalid;
 560                                 memcpy(key->ipv6.nd.tll,
 561                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 562                         }
 563
 564                         icmp_len -= opt_len;
 565                         offset += opt_len;
 566                 }
 567         }
 568
 569         goto out;
 570
 571 invalid:
 572         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 573         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 574         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 575
 576 out:
 577         *key_lenp = key_len;
 578         return error;
 579 }
 580
 581 /**
 582  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 583  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 584  * Ethernet header
 585  * @in_port: port number on which @skb was received.
 586  * @key: output flow key
 587  * @key_lenp: length of output flow key
 588  *
 589  * The caller must ensure that skb->len >= ETH_HLEN.
 590  *
 591  * Returns 0 if successful, otherwise a negative errno value.
 592  *
 593  * Initializes @skb header pointers as follows:
 594  *
 595  *    - skb->mac_header: the Ethernet header.
 596  *
 597  *    - skb->network_header: just past the Ethernet header, or just past the
 598  *      VLAN header, to the first byte of the Ethernet payload.
 599  *
 600  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 601  *      on output, then just past the IP header, if one is present and
 602  *      of a correct length, otherwise the same as skb->network_header.
 603  *      For other key->dl_type values it is left untouched.
 604  */
 605 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 606                  int *key_lenp)
 607 {
 608         int error = 0;
 609         int key_len = SW_FLOW_KEY_OFFSET(eth);
 610         struct ethhdr *eth;
 611
 612         memset(key, 0, sizeof(*key));
 613
 614         key->phy.priority = skb->priority;
 615         key->phy.in_port = in_port;
 616
 617         skb_reset_mac_header(skb);
 618
 619         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 620          * header in the linear data area.
 621          */
 622         eth = eth_hdr(skb);
 623         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 624         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 625
 626         __skb_pull(skb, 2 * ETH_ALEN);
 627
 628         if (vlan_tx_tag_present(skb))
 629                 key->eth.tci = htons(skb->vlan_tci);
 630         else if (eth->h_proto == htons(ETH_P_8021Q))
 631                 if (unlikely(parse_vlan(skb, key)))
 632                         return -ENOMEM;
 633
 634         key->eth.type = parse_ethertype(skb);
 635         if (unlikely(key->eth.type == htons(0)))
 636                 return -ENOMEM;
 637
 638         skb_reset_network_header(skb);
 639         __skb_push(skb, skb->data - skb_mac_header(skb));
 640
 641         /* Network layer. */
 642         if (key->eth.type == htons(ETH_P_IP)) {
 643                 struct iphdr *nh;
 644                 __be16 offset;
 645
 646                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 647
 648                 error = check_iphdr(skb);
 649                 if (unlikely(error)) {
 650                         if (error == -EINVAL) {
 651                                 skb->transport_header = skb->network_header;
 652                                 error = 0;
 653                         }
 654                         goto out;
 655                 }
 656
 657                 nh = ip_hdr(skb);
 658                 key->ipv4.addr.src = nh->saddr;
 659                 key->ipv4.addr.dst = nh->daddr;
 660
 661                 key->ip.proto = nh->protocol;
 662                 key->ip.tos = nh->tos;
 663                 key->ip.ttl = nh->ttl;
 664
 665                 offset = nh->frag_off & htons(IP_OFFSET);
 666                 if (offset) {
 667                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 668                         goto out;
 669                 }
 670                 if (nh->frag_off & htons(IP_MF) ||
 671                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 672                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 673
 674                 /* Transport layer. */
 675                 if (key->ip.proto == IPPROTO_TCP) {
 676                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 677                         if (tcphdr_ok(skb)) {
 678                                 struct tcphdr *tcp = tcp_hdr(skb);
 679                                 key->ipv4.tp.src = tcp->source;
 680                                 key->ipv4.tp.dst = tcp->dest;
 681                         }
 682                 } else if (key->ip.proto == IPPROTO_UDP) {
 683                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 684                         if (udphdr_ok(skb)) {
 685                                 struct udphdr *udp = udp_hdr(skb);
 686                                 key->ipv4.tp.src = udp->source;
 687                                 key->ipv4.tp.dst = udp->dest;
 688                         }
 689                 } else if (key->ip.proto == IPPROTO_ICMP) {
 690                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 691                         if (icmphdr_ok(skb)) {
 692                                 struct icmphdr *icmp = icmp_hdr(skb);
 693                                 /* The ICMP type and code fields use the 16-bit
 694                                  * transport port fields, so we need to store
 695                                  * them in 16-bit network byte order. */
 696                                 key->ipv4.tp.src = htons(icmp->type);
 697                                 key->ipv4.tp.dst = htons(icmp->code);
 698                         }
 699                 }
 700
 701         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 702                 struct arp_eth_header *arp;
 703
 704                 arp = (struct arp_eth_header *)skb_network_header(skb);
 705
 706                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 707                                 && arp->ar_pro == htons(ETH_P_IP)
 708                                 && arp->ar_hln == ETH_ALEN
 709                                 && arp->ar_pln == 4) {
 710
 711                         /* We only match on the lower 8 bits of the opcode. */
 712                         if (ntohs(arp->ar_op) <= 0xff)
 713                                 key->ip.proto = ntohs(arp->ar_op);
 714
 715                         if (key->ip.proto == ARPOP_REQUEST
 716                                         || key->ip.proto == ARPOP_REPLY) {
 717                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 718                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 719                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 720                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 721                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 722                         }
 723                 }
 724         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 725                 int nh_len;             /* IPv6 Header + Extensions */
 726
 727                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 728                 if (unlikely(nh_len < 0)) {
 729                         if (nh_len == -EINVAL)
 730                                 skb->transport_header = skb->network_header;
 731                         else
 732                                 error = nh_len;
 733                         goto out;
 734                 }
 735
 736                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 737                         goto out;
 738                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 739                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 740
 741                 /* Transport layer. */
 742                 if (key->ip.proto == NEXTHDR_TCP) {
 743                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 744                         if (tcphdr_ok(skb)) {
 745                                 struct tcphdr *tcp = tcp_hdr(skb);
 746                                 key->ipv6.tp.src = tcp->source;
 747                                 key->ipv6.tp.dst = tcp->dest;
 748                         }
 749                 } else if (key->ip.proto == NEXTHDR_UDP) {
 750                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 751                         if (udphdr_ok(skb)) {
 752                                 struct udphdr *udp = udp_hdr(skb);
 753                                 key->ipv6.tp.src = udp->source;
 754                                 key->ipv6.tp.dst = udp->dest;
 755                         }
 756                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 757                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 758                         if (icmp6hdr_ok(skb)) {
 759                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 760                                 if (error < 0)
 761                                         goto out;
 762                         }
 763                 }
 764         }
 765
 766 out:
 767         *key_lenp = key_len;
 768         return error;
 769 }
 770
 771 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 772 {
 773         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 774 }
 775
 776 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 777                                 struct sw_flow_key *key, int key_len)
 778 {
 779         struct sw_flow *flow;
 780         struct hlist_node *n;
 781         struct hlist_head *head;
 782         u32 hash;
 783
 784         hash = ovs_flow_hash(key, key_len);
 785
 786         head = find_bucket(table, hash);
 787         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 788
 789                 if (flow->hash == hash &&
 790                     !memcmp(&flow->key, key, key_len)) {
 791                         return flow;
 792                 }
 793         }
 794         return NULL;
 795 }
 796
 797 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 798 {
 799         struct hlist_head *head;
 800
 801         head = find_bucket(table, flow->hash);
 802         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 803         table->count++;
 804 }
 805
 806 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 807 {
 808         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 809         table->count--;
 810         BUG_ON(table->count < 0);
 811 }
 812
 813 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 814 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 815         [OVS_KEY_ATTR_ENCAP] = -1,
 816         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 817         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 818         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 819         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 820         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 821         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 822         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 823         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 824         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 825         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 826         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 827         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 828         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 829 };
 830
 831 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 832                                   const struct nlattr *a[], u32 *attrs)
 833 {
 834         const struct ovs_key_icmp *icmp_key;
 835         const struct ovs_key_tcp *tcp_key;
 836         const struct ovs_key_udp *udp_key;
 837
 838         switch (swkey->ip.proto) {
 839         case IPPROTO_TCP:
 840                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 841                         return -EINVAL;
 842                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 843
 844                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 845                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 846                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 847                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 848                 break;
 849
 850         case IPPROTO_UDP:
 851                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 852                         return -EINVAL;
 853                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 854
 855                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 856                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 857                 swkey->ipv4.tp.src = udp_key->udp_src;
 858                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 859                 break;
 860
 861         case IPPROTO_ICMP:
 862                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 863                         return -EINVAL;
 864                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 865
 866                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 867                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 868                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 869                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 870                 break;
 871         }
 872
 873         return 0;
 874 }
 875
 876 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 877                                   const struct nlattr *a[], u32 *attrs)
 878 {
 879         const struct ovs_key_icmpv6 *icmpv6_key;
 880         const struct ovs_key_tcp *tcp_key;
 881         const struct ovs_key_udp *udp_key;
 882
 883         switch (swkey->ip.proto) {
 884         case IPPROTO_TCP:
 885                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 886                         return -EINVAL;
 887                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 888
 889                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 890                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 891                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 892                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 893                 break;
 894
 895         case IPPROTO_UDP:
 896                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 897                         return -EINVAL;
 898                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 899
 900                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 901                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 902                 swkey->ipv6.tp.src = udp_key->udp_src;
 903                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 904                 break;
 905
 906         case IPPROTO_ICMPV6:
 907                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 908                         return -EINVAL;
 909                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 910
 911                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 912                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 913                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 914                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 915
 916                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 917                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 918                         const struct ovs_key_nd *nd_key;
 919
 920                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 921                                 return -EINVAL;
 922                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 923
 924                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 925                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 926                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 927                                sizeof(swkey->ipv6.nd.target));
 928                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 929                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 930                 }
 931                 break;
 932         }
 933
 934         return 0;
 935 }
 936
 937 static int parse_flow_nlattrs(const struct nlattr *attr,
 938                               const struct nlattr *a[], u32 *attrsp)
 939 {
 940         const struct nlattr *nla;
 941         u32 attrs;
 942         int rem;
 943
 944         attrs = 0;
 945         nla_for_each_nested(nla, attr, rem) {
 946                 u16 type = nla_type(nla);
 947                 int expected_len;
 948
 949                 if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
 950                         return -EINVAL;
 951
 952                 expected_len = ovs_key_lens[type];
 953                 if (nla_len(nla) != expected_len && expected_len != -1)
 954                         return -EINVAL;
 955
 956                 attrs |= 1 << type;
 957                 a[type] = nla;
 958         }
 959         if (rem)
 960                 return -EINVAL;
 961
 962         *attrsp = attrs;
 963         return 0;
 964 }
 965
 966 /**
 967  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 968  * @swkey: receives the extracted flow key.
 969  * @key_lenp: number of bytes used in @swkey.
 970  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 971  * sequence.
 972  */
 973 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 974                       const struct nlattr *attr)
 975 {
 976         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 977         const struct ovs_key_ethernet *eth_key;
 978         int key_len;
 979         u32 attrs;
 980         int err;
 981
 982         memset(swkey, 0, sizeof(struct sw_flow_key));
 983         key_len = SW_FLOW_KEY_OFFSET(eth);
 984
 985         err = parse_flow_nlattrs(attr, a, &attrs);
 986         if (err)
 987                 return err;
 988
 989         /* Metadata attributes. */
 990         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 991                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
 992                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 993         }
 994         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 995                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 996                 if (in_port >= DP_MAX_PORTS)
 997                         return -EINVAL;
 998                 swkey->phy.in_port = in_port;
 999                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1000         } else {
1001                 swkey->phy.in_port = USHRT_MAX;
1002         }
1003
1004         /* Data attributes. */
1005         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1006                 return -EINVAL;
1007         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1008
1009         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1010         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1011         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1012
1013         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1014             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1015                 const struct nlattr *encap;
1016                 __be16 tci;
1017
1018                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1019                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1020                               (1 << OVS_KEY_ATTR_ENCAP)))
1021                         return -EINVAL;
1022
1023                 encap = a[OVS_KEY_ATTR_ENCAP];
1024                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1025                 if (tci & htons(VLAN_TAG_PRESENT)) {
1026                         swkey->eth.tci = tci;
1027
1028                         err = parse_flow_nlattrs(encap, a, &attrs);
1029                         if (err)
1030                                 return err;
1031                 } else if (!tci) {
1032                         /* Corner case for truncated 802.1Q header. */
1033                         if (nla_len(encap))
1034                                 return -EINVAL;
1035
1036                         swkey->eth.type = htons(ETH_P_8021Q);
1037                         *key_lenp = key_len;
1038                         return 0;
1039                 } else {
1040                         return -EINVAL;
1041                 }
1042         }
1043
1044         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1045                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1046                 if (ntohs(swkey->eth.type) < 1536)
1047                         return -EINVAL;
1048                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1049         } else {
1050                 swkey->eth.type = htons(ETH_P_802_2);
1051         }
1052
1053         if (swkey->eth.type == htons(ETH_P_IP)) {
1054                 const struct ovs_key_ipv4 *ipv4_key;
1055
1056                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1057                         return -EINVAL;
1058                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1059
1060                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1061                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1062                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1063                         return -EINVAL;
1064                 swkey->ip.proto = ipv4_key->ipv4_proto;
1065                 swkey->ip.tos = ipv4_key->ipv4_tos;
1066                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1067                 swkey->ip.frag = ipv4_key->ipv4_frag;
1068                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1069                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1070
1071                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1072                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1073                         if (err)
1074                                 return err;
1075                 }
1076         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1077                 const struct ovs_key_ipv6 *ipv6_key;
1078
1079                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1080                         return -EINVAL;
1081                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1082
1083                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1084                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1085                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1086                         return -EINVAL;
1087                 swkey->ipv6.label = ipv6_key->ipv6_label;
1088                 swkey->ip.proto = ipv6_key->ipv6_proto;
1089                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1090                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1091                 swkey->ip.frag = ipv6_key->ipv6_frag;
1092                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1093                        sizeof(swkey->ipv6.addr.src));
1094                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1095                        sizeof(swkey->ipv6.addr.dst));
1096
1097                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1098                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1099                         if (err)
1100                                 return err;
1101                 }
1102         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1103                 const struct ovs_key_arp *arp_key;
1104
1105                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1106                         return -EINVAL;
1107                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1108
1109                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1110                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1111                 swkey->ipv4.addr.src = arp_key->arp_sip;
1112                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1113                 if (arp_key->arp_op & htons(0xff00))
1114                         return -EINVAL;
1115                 swkey->ip.proto = ntohs(arp_key->arp_op);
1116                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1117                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1118         }
1119
1120         if (attrs)
1121                 return -EINVAL;
1122         *key_lenp = key_len;
1123
1124         return 0;
1125 }
1126
1127 /**
1128  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1129  * @in_port: receives the extracted input port.
1130  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1131  * sequence.
1132  *
1133  * This parses a series of Netlink attributes that form a flow key, which must
1134  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1135  * get the metadata, that is, the parts of the flow key that cannot be
1136  * extracted from the packet itself.
1137  */
1138 int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
1139                                const struct nlattr *attr)
1140 {
1141         const struct nlattr *nla;
1142         int rem;
1143
1144         *in_port = USHRT_MAX;
1145         *priority = 0;
1146
1147         nla_for_each_nested(nla, attr, rem) {
1148                 int type = nla_type(nla);
1149
1150                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1151                         if (nla_len(nla) != ovs_key_lens[type])
1152                                 return -EINVAL;
1153
1154                         switch (type) {
1155                         case OVS_KEY_ATTR_PRIORITY:
1156                                 *priority = nla_get_u32(nla);
1157                                 break;
1158
1159                         case OVS_KEY_ATTR_IN_PORT:
1160                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1161                                         return -EINVAL;
1162                                 *in_port = nla_get_u32(nla);
1163                                 break;
1164                         }
1165                 }
1166         }
1167         if (rem)
1168                 return -EINVAL;
1169         return 0;
1170 }
1171
1172 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1173 {
1174         struct ovs_key_ethernet *eth_key;
1175         struct nlattr *nla, *encap;
1176
1177         if (swkey->phy.priority)
1178                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1179
1180         if (swkey->phy.in_port != USHRT_MAX)
1181                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1182
1183         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1184         if (!nla)
1185                 goto nla_put_failure;
1186         eth_key = nla_data(nla);
1187         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1188         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1189
1190         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1191                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q));
1192                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci);
1193                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1194                 if (!swkey->eth.tci)
1195                         goto unencap;
1196         } else {
1197                 encap = NULL;
1198         }
1199
1200         if (swkey->eth.type == htons(ETH_P_802_2))
1201                 goto unencap;
1202
1203         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1204
1205         if (swkey->eth.type == htons(ETH_P_IP)) {
1206                 struct ovs_key_ipv4 *ipv4_key;
1207
1208                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1209                 if (!nla)
1210                         goto nla_put_failure;
1211                 ipv4_key = nla_data(nla);
1212                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1213                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1214                 ipv4_key->ipv4_proto = swkey->ip.proto;
1215                 ipv4_key->ipv4_tos = swkey->ip.tos;
1216                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1217                 ipv4_key->ipv4_frag = swkey->ip.frag;
1218         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1219                 struct ovs_key_ipv6 *ipv6_key;
1220
1221                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1222                 if (!nla)
1223                         goto nla_put_failure;
1224                 ipv6_key = nla_data(nla);
1225                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1226                                 sizeof(ipv6_key->ipv6_src));
1227                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1228                                 sizeof(ipv6_key->ipv6_dst));
1229                 ipv6_key->ipv6_label = swkey->ipv6.label;
1230                 ipv6_key->ipv6_proto = swkey->ip.proto;
1231                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1232                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1233                 ipv6_key->ipv6_frag = swkey->ip.frag;
1234         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1235                 struct ovs_key_arp *arp_key;
1236
1237                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1238                 if (!nla)
1239                         goto nla_put_failure;
1240                 arp_key = nla_data(nla);
1241                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1242                 arp_key->arp_sip = swkey->ipv4.addr.src;
1243                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1244                 arp_key->arp_op = htons(swkey->ip.proto);
1245                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1246                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1247         }
1248
1249         if ((swkey->eth.type == htons(ETH_P_IP) ||
1250              swkey->eth.type == htons(ETH_P_IPV6)) &&
1251              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1252
1253                 if (swkey->ip.proto == IPPROTO_TCP) {
1254                         struct ovs_key_tcp *tcp_key;
1255
1256                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1257                         if (!nla)
1258                                 goto nla_put_failure;
1259                         tcp_key = nla_data(nla);
1260                         if (swkey->eth.type == htons(ETH_P_IP)) {
1261                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1262                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1263                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1264                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1265                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1266                         }
1267                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1268                         struct ovs_key_udp *udp_key;
1269
1270                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1271                         if (!nla)
1272                                 goto nla_put_failure;
1273                         udp_key = nla_data(nla);
1274                         if (swkey->eth.type == htons(ETH_P_IP)) {
1275                                 udp_key->udp_src = swkey->ipv4.tp.src;
1276                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1277                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1278                                 udp_key->udp_src = swkey->ipv6.tp.src;
1279                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1280                         }
1281                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1282                            swkey->ip.proto == IPPROTO_ICMP) {
1283                         struct ovs_key_icmp *icmp_key;
1284
1285                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1286                         if (!nla)
1287                                 goto nla_put_failure;
1288                         icmp_key = nla_data(nla);
1289                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1290                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1291                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1292                            swkey->ip.proto == IPPROTO_ICMPV6) {
1293                         struct ovs_key_icmpv6 *icmpv6_key;
1294
1295                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1296                                                 sizeof(*icmpv6_key));
1297                         if (!nla)
1298                                 goto nla_put_failure;
1299                         icmpv6_key = nla_data(nla);
1300                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1301                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1302
1303                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1304                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1305                                 struct ovs_key_nd *nd_key;
1306
1307                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1308                                 if (!nla)
1309                                         goto nla_put_failure;
1310                                 nd_key = nla_data(nla);
1311                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1312                                                         sizeof(nd_key->nd_target));
1313                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1314                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1315                         }
1316                 }
1317         }
1318
1319 unencap:
1320         if (encap)
1321                 nla_nest_end(skb, encap);
1322
1323         return 0;
1324
1325 nla_put_failure:
1326         return -EMSGSIZE;
1327 }
1328
1329 /* Initializes the flow module.
1330  * Returns zero if successful or a negative error code. */
1331 int ovs_flow_init(void)
1332 {
1333         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1334                                         0, NULL);
1335         if (flow_cache == NULL)
1336                 return -ENOMEM;
1337
1338         return 0;
1339 }
1340
1341 /* Uninitializes the flow module. */
1342 void ovs_flow_exit(void)
1343 {
1344         kmem_cache_destroy(flow_cache);
1345 }