datapath/flow.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include "flow.h"
  10 #include "datapath.h"
  11 #include <asm/uaccess.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/if_ether.h>
  15 #include <linux/if_vlan.h>
  16 #include <net/llc_pdu.h>
  17 #include <linux/kernel.h>
  18 #include <linux/jhash.h>
  19 #include <linux/jiffies.h>
  20 #include <linux/llc.h>
  21 #include <linux/module.h>
  22 #include <linux/in.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/if_arp.h>
  25 #include <linux/if_ether.h>
  26 #include <linux/ip.h>
  27 #include <linux/ipv6.h>
  28 #include <linux/tcp.h>
  29 #include <linux/udp.h>
  30 #include <linux/icmp.h>
  31 #include <linux/icmpv6.h>
  32 #include <linux/rculist.h>
  33 #include <net/inet_ecn.h>
  34 #include <net/ip.h>
  35 #include <net/ipv6.h>
  36 #include <net/ndisc.h>
  37
  38 #include "vlan.h"
  39
  40 static struct kmem_cache *flow_cache;
  41 static unsigned int hash_seed __read_mostly;
  42
  43 static int check_header(struct sk_buff *skb, int len)
  44 {
  45         if (unlikely(skb->len < len))
  46                 return -EINVAL;
  47         if (unlikely(!pskb_may_pull(skb, len)))
  48                 return -ENOMEM;
  49         return 0;
  50 }
  51
  52 static inline bool arphdr_ok(struct sk_buff *skb)
  53 {
  54         return pskb_may_pull(skb, skb_network_offset(skb) +
  55                                   sizeof(struct arp_eth_header));
  56 }
  57
  58 static inline int check_iphdr(struct sk_buff *skb)
  59 {
  60         unsigned int nh_ofs = skb_network_offset(skb);
  61         unsigned int ip_len;
  62         int err;
  63
  64         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  65         if (unlikely(err))
  66                 return err;
  67
  68         ip_len = ip_hdrlen(skb);
  69         if (unlikely(ip_len < sizeof(struct iphdr) ||
  70                      skb->len < nh_ofs + ip_len))
  71                 return -EINVAL;
  72
  73         skb_set_transport_header(skb, nh_ofs + ip_len);
  74         return 0;
  75 }
  76
  77 static inline bool tcphdr_ok(struct sk_buff *skb)
  78 {
  79         int th_ofs = skb_transport_offset(skb);
  80         int tcp_len;
  81
  82         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  83                 return false;
  84
  85         tcp_len = tcp_hdrlen(skb);
  86         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  87                      skb->len < th_ofs + tcp_len))
  88                 return false;
  89
  90         return true;
  91 }
  92
  93 static inline bool udphdr_ok(struct sk_buff *skb)
  94 {
  95         return pskb_may_pull(skb, skb_transport_offset(skb) +
  96                                   sizeof(struct udphdr));
  97 }
  98
  99 static inline bool icmphdr_ok(struct sk_buff *skb)
 100 {
 101         return pskb_may_pull(skb, skb_transport_offset(skb) +
 102                                   sizeof(struct icmphdr));
 103 }
 104
 105 u64 flow_used_time(unsigned long flow_jiffies)
 106 {
 107         struct timespec cur_ts;
 108         u64 cur_ms, idle_ms;
 109
 110         ktime_get_ts(&cur_ts);
 111         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 112         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 113                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 114
 115         return cur_ms - idle_ms;
 116 }
 117
 118 #define SW_FLOW_KEY_OFFSET(field)               \
 119         offsetof(struct sw_flow_key, field) +   \
 120         FIELD_SIZEOF(struct sw_flow_key, field)
 121
 122 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 123                          int *key_lenp)
 124 {
 125         unsigned int nh_ofs = skb_network_offset(skb);
 126         unsigned int nh_len;
 127         int payload_ofs;
 128         struct ipv6hdr *nh;
 129         uint8_t nexthdr;
 130         int err;
 131
 132         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.addr);
 133
 134         err = check_header(skb, nh_ofs + sizeof(*nh));
 135         if (unlikely(err))
 136                 return err;
 137
 138         nh = ipv6_hdr(skb);
 139         nexthdr = nh->nexthdr;
 140         payload_ofs = (u8 *)(nh + 1) - skb->data;
 141
 142         key->ip.proto = NEXTHDR_NONE;
 143         key->ip.tos = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
 144         ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
 145         ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
 146
 147         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr);
 148         if (unlikely(payload_ofs < 0))
 149                 return -EINVAL;
 150
 151         nh_len = payload_ofs - nh_ofs;
 152         skb_set_transport_header(skb, nh_ofs + nh_len);
 153         key->ip.proto = nexthdr;
 154         return nh_len;
 155 }
 156
 157 static bool icmp6hdr_ok(struct sk_buff *skb)
 158 {
 159         return pskb_may_pull(skb, skb_transport_offset(skb) +
 160                                   sizeof(struct icmp6hdr));
 161 }
 162
 163 #define TCP_FLAGS_OFFSET 13
 164 #define TCP_FLAG_MASK 0x3f
 165
 166 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 167 {
 168         u8 tcp_flags = 0;
 169
 170         if (flow->key.eth.type == htons(ETH_P_IP) &&
 171             flow->key.ip.proto == IPPROTO_TCP) {
 172                 u8 *tcp = (u8 *)tcp_hdr(skb);
 173                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 174         }
 175
 176         spin_lock(&flow->lock);
 177         flow->used = jiffies;
 178         flow->packet_count++;
 179         flow->byte_count += skb->len;
 180         flow->tcp_flags |= tcp_flags;
 181         spin_unlock(&flow->lock);
 182 }
 183
 184 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 185 {
 186         int actions_len = nla_len(actions);
 187         struct sw_flow_actions *sfa;
 188
 189         /* At least DP_MAX_PORTS actions are required to be able to flood a
 190          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 191          * etc. */
 192         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 193                 return ERR_PTR(-EINVAL);
 194
 195         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 196         if (!sfa)
 197                 return ERR_PTR(-ENOMEM);
 198
 199         sfa->actions_len = actions_len;
 200         memcpy(sfa->actions, nla_data(actions), actions_len);
 201         return sfa;
 202 }
 203
 204 struct sw_flow *flow_alloc(void)
 205 {
 206         struct sw_flow *flow;
 207
 208         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 209         if (!flow)
 210                 return ERR_PTR(-ENOMEM);
 211
 212         spin_lock_init(&flow->lock);
 213         atomic_set(&flow->refcnt, 1);
 214         flow->sf_acts = NULL;
 215         flow->dead = false;
 216
 217         return flow;
 218 }
 219
 220 static struct hlist_head __rcu *find_bucket(struct flow_table * table, u32 hash)
 221 {
 222         return flex_array_get(table->buckets,
 223                                 (hash & (table->n_buckets - 1)));
 224 }
 225
 226 static struct flex_array  __rcu *alloc_buckets(unsigned int n_buckets)
 227 {
 228         struct flex_array  __rcu * buckets;
 229         int i, err;
 230
 231         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 232                                    n_buckets, GFP_KERNEL);
 233         if (!buckets)
 234                 return NULL;
 235
 236         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 237         if (err) {
 238                 flex_array_free(buckets);
 239                 return NULL;
 240         }
 241
 242         for (i = 0; i < n_buckets; i++)
 243                 INIT_HLIST_HEAD((struct hlist_head *)
 244                                         flex_array_get(buckets, i));
 245
 246         return buckets;
 247 }
 248
 249 static void free_buckets(struct flex_array * buckets)
 250 {
 251         flex_array_free(buckets);
 252 }
 253
 254 struct flow_table *flow_tbl_alloc(int new_size)
 255 {
 256         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 257
 258         if (!table)
 259                 return NULL;
 260
 261         table->buckets = alloc_buckets(new_size);
 262
 263         if (!table->buckets) {
 264                 kfree(table);
 265                 return NULL;
 266         }
 267         table->n_buckets = new_size;
 268         table->count = 0;
 269
 270         return table;
 271 }
 272
 273 static void flow_free(struct sw_flow *flow)
 274 {
 275         flow->dead = true;
 276         flow_put(flow);
 277 }
 278
 279 void flow_tbl_destroy(struct flow_table *table)
 280 {
 281         int i;
 282
 283         if (!table)
 284                 return;
 285
 286         for (i = 0; i < table->n_buckets; i++) {
 287                 struct sw_flow *flow;
 288                 struct hlist_head *head = flex_array_get(table->buckets, i);
 289                 struct hlist_node *node, *n;
 290
 291                 hlist_for_each_entry_safe(flow, node, n, head, hash_node) {
 292                         hlist_del_init_rcu(&flow->hash_node);
 293                         flow_free(flow);
 294                 }
 295         }
 296
 297         free_buckets(table->buckets);
 298         kfree(table);
 299 }
 300
 301 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 302 {
 303         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 304
 305         flow_tbl_destroy(table);
 306 }
 307
 308 void flow_tbl_deferred_destroy(struct flow_table *table)
 309 {
 310         if (!table)
 311                 return;
 312
 313         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 314 }
 315
 316 struct sw_flow *flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 317 {
 318         struct sw_flow *flow;
 319         struct hlist_head *head;
 320         struct hlist_node *n;
 321         int i;
 322
 323         while (*bucket < table->n_buckets) {
 324                 i = 0;
 325                 head = flex_array_get(table->buckets, *bucket);
 326                 hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 327                         if (i < *last) {
 328                                 i++;
 329                                 continue;
 330                         }
 331                         *last = i + 1;
 332                         return flow;
 333                 }
 334                 (*bucket)++;
 335                 *last = 0;
 336         }
 337
 338         return NULL;
 339 }
 340
 341 struct flow_table *flow_tbl_expand(struct flow_table *table)
 342 {
 343         struct flow_table *new_table;
 344         int n_buckets = table->n_buckets * 2;
 345         int i;
 346
 347         new_table = flow_tbl_alloc(n_buckets);
 348         if (!new_table)
 349                 return ERR_PTR(-ENOMEM);
 350
 351         for (i = 0; i < table->n_buckets; i++) {
 352                 struct sw_flow *flow;
 353                 struct hlist_head *head;
 354                 struct hlist_node *n, *pos;
 355
 356                 head = flex_array_get(table->buckets, i);
 357
 358                 hlist_for_each_entry_safe(flow, n, pos, head, hash_node) {
 359                         hlist_del_init_rcu(&flow->hash_node);
 360                         flow_tbl_insert(new_table, flow);
 361                 }
 362         }
 363
 364         return new_table;
 365 }
 366
 367 /* RCU callback used by flow_deferred_free. */
 368 static void rcu_free_flow_callback(struct rcu_head *rcu)
 369 {
 370         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 371
 372         flow->dead = true;
 373         flow_put(flow);
 374 }
 375
 376 /* Schedules 'flow' to be freed after the next RCU grace period.
 377  * The caller must hold rcu_read_lock for this to be sensible. */
 378 void flow_deferred_free(struct sw_flow *flow)
 379 {
 380         call_rcu(&flow->rcu, rcu_free_flow_callback);
 381 }
 382
 383 void flow_hold(struct sw_flow *flow)
 384 {
 385         atomic_inc(&flow->refcnt);
 386 }
 387
 388 void flow_put(struct sw_flow *flow)
 389 {
 390         if (unlikely(!flow))
 391                 return;
 392
 393         if (atomic_dec_and_test(&flow->refcnt)) {
 394                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 395                 kmem_cache_free(flow_cache, flow);
 396         }
 397 }
 398
 399 /* RCU callback used by flow_deferred_free_acts. */
 400 static void rcu_free_acts_callback(struct rcu_head *rcu)
 401 {
 402         struct sw_flow_actions *sf_acts = container_of(rcu,
 403                         struct sw_flow_actions, rcu);
 404         kfree(sf_acts);
 405 }
 406
 407 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 408  * The caller must hold rcu_read_lock for this to be sensible. */
 409 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 410 {
 411         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 412 }
 413
 414 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 415 {
 416         struct qtag_prefix {
 417                 __be16 eth_type; /* ETH_P_8021Q */
 418                 __be16 tci;
 419         };
 420         struct qtag_prefix *qp;
 421
 422         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 423                                          sizeof(__be16))))
 424                 return -ENOMEM;
 425
 426         qp = (struct qtag_prefix *) skb->data;
 427         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 428         __skb_pull(skb, sizeof(struct qtag_prefix));
 429
 430         return 0;
 431 }
 432
 433 static __be16 parse_ethertype(struct sk_buff *skb)
 434 {
 435         struct llc_snap_hdr {
 436                 u8  dsap;  /* Always 0xAA */
 437                 u8  ssap;  /* Always 0xAA */
 438                 u8  ctrl;
 439                 u8  oui[3];
 440                 __be16 ethertype;
 441         };
 442         struct llc_snap_hdr *llc;
 443         __be16 proto;
 444
 445         proto = *(__be16 *) skb->data;
 446         __skb_pull(skb, sizeof(__be16));
 447
 448         if (ntohs(proto) >= 1536)
 449                 return proto;
 450
 451         if (skb->len < sizeof(struct llc_snap_hdr))
 452                 return htons(ETH_P_802_2);
 453
 454         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 455                 return htons(0);
 456
 457         llc = (struct llc_snap_hdr *) skb->data;
 458         if (llc->dsap != LLC_SAP_SNAP ||
 459             llc->ssap != LLC_SAP_SNAP ||
 460             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 461                 return htons(ETH_P_802_2);
 462
 463         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 464         return llc->ethertype;
 465 }
 466
 467 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 468                         int *key_lenp, int nh_len)
 469 {
 470         struct icmp6hdr *icmp = icmp6_hdr(skb);
 471         int error = 0;
 472         int key_len;
 473
 474         /* The ICMPv6 type and code fields use the 16-bit transport port
 475          * fields, so we need to store them in 16-bit network byte order.
 476          */
 477         key->ipv6.tp.src = htons(icmp->icmp6_type);
 478         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 479         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 480
 481         if (icmp->icmp6_code == 0 &&
 482             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 483              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 484                 int icmp_len = skb->len - skb_transport_offset(skb);
 485                 struct nd_msg *nd;
 486                 int offset;
 487
 488                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 489
 490                 /* In order to process neighbor discovery options, we need the
 491                  * entire packet.
 492                  */
 493                 if (unlikely(icmp_len < sizeof(*nd)))
 494                         goto out;
 495                 if (unlikely(skb_linearize(skb))) {
 496                         error = -ENOMEM;
 497                         goto out;
 498                 }
 499
 500                 nd = (struct nd_msg *)skb_transport_header(skb);
 501                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
 502                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 503
 504                 icmp_len -= sizeof(*nd);
 505                 offset = 0;
 506                 while (icmp_len >= 8) {
 507                         struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd->opt + offset);
 508                         int opt_len = nd_opt->nd_opt_len * 8;
 509
 510                         if (unlikely(!opt_len || opt_len > icmp_len))
 511                                 goto invalid;
 512
 513                         /* Store the link layer address if the appropriate
 514                          * option is provided.  It is considered an error if
 515                          * the same link layer option is specified twice.
 516                          */
 517                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 518                             && opt_len == 8) {
 519                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 520                                         goto invalid;
 521                                 memcpy(key->ipv6.nd.sll,
 522                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 523                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 524                                    && opt_len == 8) {
 525                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 526                                         goto invalid;
 527                                 memcpy(key->ipv6.nd.tll,
 528                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 529                         }
 530
 531                         icmp_len -= opt_len;
 532                         offset += opt_len;
 533                 }
 534         }
 535
 536         goto out;
 537
 538 invalid:
 539         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 540         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 541         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 542
 543 out:
 544         *key_lenp = key_len;
 545         return error;
 546 }
 547
 548 /**
 549  * flow_extract - extracts a flow key from an Ethernet frame.
 550  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 551  * Ethernet header
 552  * @in_port: port number on which @skb was received.
 553  * @key: output flow key
 554  * @key_lenp: length of output flow key
 555  * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
 556  * not contain an IPv4 packet or if it is not a fragment.
 557  *
 558  * The caller must ensure that skb->len >= ETH_HLEN.
 559  *
 560  * Returns 0 if successful, otherwise a negative errno value.
 561  *
 562  * Initializes @skb header pointers as follows:
 563  *
 564  *    - skb->mac_header: the Ethernet header.
 565  *
 566  *    - skb->network_header: just past the Ethernet header, or just past the
 567  *      VLAN header, to the first byte of the Ethernet payload.
 568  *
 569  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 570  *      on output, then just past the IP header, if one is present and
 571  *      of a correct length, otherwise the same as skb->network_header.
 572  *      For other key->dl_type values it is left untouched.
 573  */
 574 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 575                  int *key_lenp, bool *is_frag)
 576 {
 577         int error = 0;
 578         int key_len = SW_FLOW_KEY_OFFSET(eth);
 579         struct ethhdr *eth;
 580
 581         memset(key, 0, sizeof(*key));
 582         key->eth.tun_id = OVS_CB(skb)->tun_id;
 583         key->eth.in_port = in_port;
 584         *is_frag = false;
 585
 586         skb_reset_mac_header(skb);
 587
 588         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 589          * header in the linear data area.
 590          */
 591         eth = eth_hdr(skb);
 592         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 593         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 594
 595         __skb_pull(skb, 2 * ETH_ALEN);
 596
 597         if (vlan_tx_tag_present(skb))
 598                 key->eth.tci = htons(vlan_get_tci(skb));
 599         else if (eth->h_proto == htons(ETH_P_8021Q))
 600                 if (unlikely(parse_vlan(skb, key)))
 601                         return -ENOMEM;
 602
 603         key->eth.type = parse_ethertype(skb);
 604         if (unlikely(key->eth.type == htons(0)))
 605                 return -ENOMEM;
 606
 607         skb_reset_network_header(skb);
 608         __skb_push(skb, skb->data - skb_mac_header(skb));
 609
 610         /* Network layer. */
 611         if (key->eth.type == htons(ETH_P_IP)) {
 612                 struct iphdr *nh;
 613
 614                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 615
 616                 error = check_iphdr(skb);
 617                 if (unlikely(error)) {
 618                         if (error == -EINVAL) {
 619                                 skb->transport_header = skb->network_header;
 620                                 error = 0;
 621                         }
 622                         goto out;
 623                 }
 624
 625                 nh = ip_hdr(skb);
 626                 key->ipv4.addr.src = nh->saddr;
 627                 key->ipv4.addr.dst = nh->daddr;
 628                 key->ip.tos = nh->tos & ~INET_ECN_MASK;
 629                 key->ip.proto = nh->protocol;
 630
 631                 /* Transport layer. */
 632                 if ((nh->frag_off & htons(IP_MF | IP_OFFSET)) ||
 633                     (skb_shinfo(skb)->gso_type & SKB_GSO_UDP))
 634                         *is_frag = true;
 635
 636                 if (key->ip.proto == IPPROTO_TCP) {
 637                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 638                         if (!*is_frag && tcphdr_ok(skb)) {
 639                                 struct tcphdr *tcp = tcp_hdr(skb);
 640                                 key->ipv4.tp.src = tcp->source;
 641                                 key->ipv4.tp.dst = tcp->dest;
 642                         }
 643                 } else if (key->ip.proto == IPPROTO_UDP) {
 644                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 645                         if (!*is_frag && udphdr_ok(skb)) {
 646                                 struct udphdr *udp = udp_hdr(skb);
 647                                 key->ipv4.tp.src = udp->source;
 648                                 key->ipv4.tp.dst = udp->dest;
 649                         }
 650                 } else if (key->ip.proto == IPPROTO_ICMP) {
 651                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 652                         if (!*is_frag && icmphdr_ok(skb)) {
 653                                 struct icmphdr *icmp = icmp_hdr(skb);
 654                                 /* The ICMP type and code fields use the 16-bit
 655                                  * transport port fields, so we need to store them
 656                                  * in 16-bit network byte order. */
 657                                 key->ipv4.tp.src = htons(icmp->type);
 658                                 key->ipv4.tp.dst = htons(icmp->code);
 659                         }
 660                 }
 661
 662         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 663                 struct arp_eth_header *arp;
 664
 665                 arp = (struct arp_eth_header *)skb_network_header(skb);
 666
 667                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 668                                 && arp->ar_pro == htons(ETH_P_IP)
 669                                 && arp->ar_hln == ETH_ALEN
 670                                 && arp->ar_pln == 4) {
 671
 672                         /* We only match on the lower 8 bits of the opcode. */
 673                         if (ntohs(arp->ar_op) <= 0xff)
 674                                 key->ip.proto = ntohs(arp->ar_op);
 675
 676                         if (key->ip.proto == ARPOP_REQUEST
 677                                         || key->ip.proto == ARPOP_REPLY) {
 678                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 679                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 680                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 681                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 682                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 683                         }
 684                 }
 685         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 686                 int nh_len;             /* IPv6 Header + Extensions */
 687
 688                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 689                 if (unlikely(nh_len < 0)) {
 690                         if (nh_len == -EINVAL)
 691                                 skb->transport_header = skb->network_header;
 692                         else
 693                                 error = nh_len;
 694                         goto out;
 695                 }
 696
 697                 /* Transport layer. */
 698                 if (key->ip.proto == NEXTHDR_TCP) {
 699                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 700                         if (tcphdr_ok(skb)) {
 701                                 struct tcphdr *tcp = tcp_hdr(skb);
 702                                 key->ipv6.tp.src = tcp->source;
 703                                 key->ipv6.tp.dst = tcp->dest;
 704                         }
 705                 } else if (key->ip.proto == NEXTHDR_UDP) {
 706                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 707                         if (udphdr_ok(skb)) {
 708                                 struct udphdr *udp = udp_hdr(skb);
 709                                 key->ipv6.tp.src = udp->source;
 710                                 key->ipv6.tp.dst = udp->dest;
 711                         }
 712                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 713                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 714                         if (icmp6hdr_ok(skb)) {
 715                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 716                                 if (error < 0)
 717                                         goto out;
 718                         }
 719                 }
 720         }
 721
 722 out:
 723         *key_lenp = key_len;
 724         return error;
 725 }
 726
 727 u32 flow_hash(const struct sw_flow_key *key, int key_len)
 728 {
 729         return jhash2((u32*)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
 730 }
 731
 732 struct sw_flow * flow_tbl_lookup(struct flow_table *table,
 733                                 struct sw_flow_key *key, int key_len)
 734 {
 735         struct sw_flow *flow;
 736         struct hlist_node *n;
 737         struct hlist_head *head;
 738         u32 hash;
 739
 740         hash = flow_hash(key, key_len);
 741
 742         head = find_bucket(table, hash);
 743         hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 744
 745                 if (flow->hash == hash &&
 746                     !memcmp(&flow->key, key, key_len)) {
 747                         return flow;
 748                 }
 749         }
 750         return NULL;
 751 }
 752
 753 void flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 754 {
 755         struct hlist_head *head;
 756
 757         head = find_bucket(table, flow->hash);
 758         hlist_add_head_rcu(&flow->hash_node, head);
 759         table->count++;
 760 }
 761
 762 void flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 763 {
 764         if (!hlist_unhashed(&flow->hash_node)) {
 765                 hlist_del_init_rcu(&flow->hash_node);
 766                 table->count--;
 767                 BUG_ON(table->count < 0);
 768         }
 769 }
 770
 771 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 772 const u32 ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 773         [OVS_KEY_ATTR_TUN_ID] = 8,
 774         [OVS_KEY_ATTR_IN_PORT] = 4,
 775         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 776         [OVS_KEY_ATTR_8021Q] = sizeof(struct ovs_key_8021q),
 777         [OVS_KEY_ATTR_ETHERTYPE] = 2,
 778         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 779         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 780         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 781         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 782         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 783         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 784         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 785         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 786 };
 787
 788 /**
 789  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 790  * @swkey: receives the extracted flow key.
 791  * @key_lenp: number of bytes used in @swkey.
 792  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 793  * sequence.
 794  *
 795  * This state machine accepts the following forms, with [] for optional
 796  * elements and | for alternatives:
 797  *
 798  * [tun_id] [in_port] ethernet [8021q] [ethertype \
 799  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 800  */
 801 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 802                       const struct nlattr *attr)
 803 {
 804         int error = 0;
 805         const struct nlattr *nla;
 806         u16 prev_type;
 807         int rem;
 808         int key_len;
 809
 810         memset(swkey, 0, sizeof(*swkey));
 811         swkey->eth.in_port = USHRT_MAX;
 812         swkey->eth.type = htons(ETH_P_802_2);
 813         key_len = SW_FLOW_KEY_OFFSET(eth);
 814
 815         prev_type = OVS_KEY_ATTR_UNSPEC;
 816         nla_for_each_nested(nla, attr, rem) {
 817                 const struct ovs_key_ethernet *eth_key;
 818                 const struct ovs_key_8021q *q_key;
 819                 const struct ovs_key_ipv4 *ipv4_key;
 820                 const struct ovs_key_ipv6 *ipv6_key;
 821                 const struct ovs_key_tcp *tcp_key;
 822                 const struct ovs_key_udp *udp_key;
 823                 const struct ovs_key_icmp *icmp_key;
 824                 const struct ovs_key_icmpv6 *icmpv6_key;
 825                 const struct ovs_key_arp *arp_key;
 826                 const struct ovs_key_nd *nd_key;
 827
 828                 int type = nla_type(nla);
 829
 830                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
 831                         goto invalid;
 832
 833 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 834                 switch (TRANSITION(prev_type, type)) {
 835                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
 836                         swkey->eth.tun_id = nla_get_be64(nla);
 837                         break;
 838
 839                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
 840                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
 841                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 842                                 goto invalid;
 843                         swkey->eth.in_port = nla_get_u32(nla);
 844                         break;
 845
 846                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_ETHERNET):
 847                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_ETHERNET):
 848                 case TRANSITION(OVS_KEY_ATTR_IN_PORT, OVS_KEY_ATTR_ETHERNET):
 849                         eth_key = nla_data(nla);
 850                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
 851                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
 852                         break;
 853
 854                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_8021Q):
 855                         q_key = nla_data(nla);
 856                         /* Only standard 0x8100 VLANs currently supported. */
 857                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 858                                 goto invalid;
 859                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 860                                 goto invalid;
 861                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 862                         break;
 863
 864                 case TRANSITION(OVS_KEY_ATTR_8021Q, OVS_KEY_ATTR_ETHERTYPE):
 865                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_ETHERTYPE):
 866                         swkey->eth.type = nla_get_be16(nla);
 867                         if (ntohs(swkey->eth.type) < 1536)
 868                                 goto invalid;
 869                         break;
 870
 871                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV4):
 872                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 873                         if (swkey->eth.type != htons(ETH_P_IP))
 874                                 goto invalid;
 875                         ipv4_key = nla_data(nla);
 876                         swkey->ip.proto = ipv4_key->ipv4_proto;
 877                         swkey->ip.tos = ipv4_key->ipv4_tos;
 878                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
 879                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
 880                         if (swkey->ip.tos & INET_ECN_MASK)
 881                                 goto invalid;
 882                         break;
 883
 884                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV6):
 885                         key_len = SW_FLOW_KEY_OFFSET(ipv6.addr);
 886                         if (swkey->eth.type != htons(ETH_P_IPV6))
 887                                 goto invalid;
 888                         ipv6_key = nla_data(nla);
 889                         swkey->ip.proto = ipv6_key->ipv6_proto;
 890                         swkey->ip.tos = ipv6_key->ipv6_tos;
 891                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
 892                                         sizeof(swkey->ipv6.addr.src));
 893                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
 894                                         sizeof(swkey->ipv6.addr.dst));
 895                         if (swkey->ip.tos & INET_ECN_MASK)
 896                                 goto invalid;
 897                         break;
 898
 899                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_TCP):
 900                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 901                         if (swkey->ip.proto != IPPROTO_TCP)
 902                                 goto invalid;
 903                         tcp_key = nla_data(nla);
 904                         swkey->ipv4.tp.src = tcp_key->tcp_src;
 905                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 906                         break;
 907
 908                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_TCP):
 909                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 910                         if (swkey->ip.proto != IPPROTO_TCP)
 911                                 goto invalid;
 912                         tcp_key = nla_data(nla);
 913                         swkey->ipv6.tp.src = tcp_key->tcp_src;
 914                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 915                         break;
 916
 917                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_UDP):
 918                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 919                         if (swkey->ip.proto != IPPROTO_UDP)
 920                                 goto invalid;
 921                         udp_key = nla_data(nla);
 922                         swkey->ipv4.tp.src = udp_key->udp_src;
 923                         swkey->ipv4.tp.dst = udp_key->udp_dst;
 924                         break;
 925
 926                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_UDP):
 927                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 928                         if (swkey->ip.proto != IPPROTO_UDP)
 929                                 goto invalid;
 930                         udp_key = nla_data(nla);
 931                         swkey->ipv6.tp.src = udp_key->udp_src;
 932                         swkey->ipv6.tp.dst = udp_key->udp_dst;
 933                         break;
 934
 935                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_ICMP):
 936                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 937                         if (swkey->ip.proto != IPPROTO_ICMP)
 938                                 goto invalid;
 939                         icmp_key = nla_data(nla);
 940                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 941                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 942                         break;
 943
 944                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_ICMPV6):
 945                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 946                         if (swkey->ip.proto != IPPROTO_ICMPV6)
 947                                 goto invalid;
 948                         icmpv6_key = nla_data(nla);
 949                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 950                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 951                         break;
 952
 953                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_ARP):
 954                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 955                         if (swkey->eth.type != htons(ETH_P_ARP))
 956                                 goto invalid;
 957                         arp_key = nla_data(nla);
 958                         swkey->ipv4.addr.src = arp_key->arp_sip;
 959                         swkey->ipv4.addr.dst = arp_key->arp_tip;
 960                         if (arp_key->arp_op & htons(0xff00))
 961                                 goto invalid;
 962                         swkey->ip.proto = ntohs(arp_key->arp_op);
 963                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
 964                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
 965                         break;
 966
 967                 case TRANSITION(OVS_KEY_ATTR_ICMPV6, OVS_KEY_ATTR_ND):
 968                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 969                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
 970                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
 971                                 goto invalid;
 972                         nd_key = nla_data(nla);
 973                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 974                                         sizeof(swkey->ipv6.nd.target));
 975                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 976                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 977                         break;
 978
 979                 default:
 980                         goto invalid;
 981                 }
 982
 983                 prev_type = type;
 984         }
 985         if (rem)
 986                 goto invalid;
 987
 988         switch (prev_type) {
 989         case OVS_KEY_ATTR_UNSPEC:
 990                 goto invalid;
 991
 992         case OVS_KEY_ATTR_TUN_ID:
 993         case OVS_KEY_ATTR_IN_PORT:
 994                 goto invalid;
 995
 996         case OVS_KEY_ATTR_ETHERNET:
 997         case OVS_KEY_ATTR_8021Q:
 998                 goto ok;
 999
1000         case OVS_KEY_ATTR_ETHERTYPE:
1001                 if (swkey->eth.type == htons(ETH_P_IP) ||
1002                     swkey->eth.type == htons(ETH_P_ARP))
1003                         goto invalid;
1004                 goto ok;
1005
1006         case OVS_KEY_ATTR_IPV4:
1007                 if (swkey->ip.proto == IPPROTO_TCP ||
1008                     swkey->ip.proto == IPPROTO_UDP ||
1009                     swkey->ip.proto == IPPROTO_ICMP)
1010                         goto invalid;
1011                 goto ok;
1012
1013         case OVS_KEY_ATTR_IPV6:
1014                 if (swkey->ip.proto == IPPROTO_TCP ||
1015                     swkey->ip.proto == IPPROTO_UDP ||
1016                     swkey->ip.proto == IPPROTO_ICMPV6)
1017                         goto invalid;
1018                 goto ok;
1019
1020         case OVS_KEY_ATTR_ICMPV6:
1021                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1022                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
1023                         goto invalid;
1024                 goto ok;
1025
1026         case OVS_KEY_ATTR_TCP:
1027         case OVS_KEY_ATTR_UDP:
1028         case OVS_KEY_ATTR_ICMP:
1029         case OVS_KEY_ATTR_ARP:
1030         case OVS_KEY_ATTR_ND:
1031                 goto ok;
1032
1033         default:
1034                 WARN_ON_ONCE(1);
1035         }
1036
1037 invalid:
1038         error = -EINVAL;
1039
1040 ok:
1041         WARN_ON_ONCE(!key_len && !error);
1042         *key_lenp = key_len;
1043         return error;
1044 }
1045
1046 /**
1047  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1048  * @in_port: receives the extracted input port.
1049  * @tun_id: receives the extracted tunnel ID.
1050  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1051  * sequence.
1052  *
1053  * This parses a series of Netlink attributes that form a flow key, which must
1054  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1055  * get the metadata, that is, the parts of the flow key that cannot be
1056  * extracted from the packet itself.
1057  */
1058 int flow_metadata_from_nlattrs(u16 *in_port, __be64 *tun_id,
1059                                const struct nlattr *attr)
1060 {
1061         const struct nlattr *nla;
1062         u16 prev_type;
1063         int rem;
1064
1065         *in_port = USHRT_MAX;
1066         *tun_id = 0;
1067
1068         prev_type = OVS_KEY_ATTR_UNSPEC;
1069         nla_for_each_nested(nla, attr, rem) {
1070                 int type = nla_type(nla);
1071
1072                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
1073                         return -EINVAL;
1074
1075                 switch (TRANSITION(prev_type, type)) {
1076                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
1077                         *tun_id = nla_get_be64(nla);
1078                         break;
1079
1080                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
1081                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
1082                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
1083                                 return -EINVAL;
1084                         *in_port = nla_get_u32(nla);
1085                         break;
1086
1087                 default:
1088                         return 0;
1089                 }
1090
1091                 prev_type = type;
1092         }
1093         if (rem)
1094                 return -EINVAL;
1095         return 0;
1096 }
1097
1098 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1099 {
1100         struct ovs_key_ethernet *eth_key;
1101         struct nlattr *nla;
1102
1103         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
1104          * to be updated, but will at least raise awareness when new
1105          * datapath key types are added. */
1106         BUILD_BUG_ON(__OVS_KEY_ATTR_MAX != 14);
1107
1108         if (swkey->eth.tun_id != cpu_to_be64(0))
1109                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->eth.tun_id);
1110
1111         if (swkey->eth.in_port != USHRT_MAX)
1112                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->eth.in_port);
1113
1114         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1115         if (!nla)
1116                 goto nla_put_failure;
1117         eth_key = nla_data(nla);
1118         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1119         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1120
1121         if (swkey->eth.tci != htons(0)) {
1122                 struct ovs_key_8021q q_key;
1123
1124                 q_key.q_tpid = htons(ETH_P_8021Q);
1125                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
1126                 NLA_PUT(skb, OVS_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
1127         }
1128
1129         if (swkey->eth.type == htons(ETH_P_802_2))
1130                 return 0;
1131
1132         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1133
1134         if (swkey->eth.type == htons(ETH_P_IP)) {
1135                 struct ovs_key_ipv4 *ipv4_key;
1136
1137                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1138                 if (!nla)
1139                         goto nla_put_failure;
1140                 ipv4_key = nla_data(nla);
1141                 memset(ipv4_key, 0, sizeof(struct ovs_key_ipv4));
1142                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1143                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1144                 ipv4_key->ipv4_proto = swkey->ip.proto;
1145                 ipv4_key->ipv4_tos = swkey->ip.tos;
1146         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1147                 struct ovs_key_ipv6 *ipv6_key;
1148
1149                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1150                 if (!nla)
1151                         goto nla_put_failure;
1152                 ipv6_key = nla_data(nla);
1153                 memset(ipv6_key, 0, sizeof(struct ovs_key_ipv6));
1154                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1155                                 sizeof(ipv6_key->ipv6_src));
1156                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1157                                 sizeof(ipv6_key->ipv6_dst));
1158                 ipv6_key->ipv6_proto = swkey->ip.proto;
1159                 ipv6_key->ipv6_tos = swkey->ip.tos;
1160         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1161                 struct ovs_key_arp *arp_key;
1162
1163                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1164                 if (!nla)
1165                         goto nla_put_failure;
1166                 arp_key = nla_data(nla);
1167                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1168                 arp_key->arp_sip = swkey->ipv4.addr.src;
1169                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1170                 arp_key->arp_op = htons(swkey->ip.proto);
1171                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1172                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1173         }
1174
1175         if (swkey->eth.type == htons(ETH_P_IP) ||
1176             swkey->eth.type == htons(ETH_P_IPV6)) {
1177
1178                 if (swkey->ip.proto == IPPROTO_TCP) {
1179                         struct ovs_key_tcp *tcp_key;
1180
1181                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1182                         if (!nla)
1183                                 goto nla_put_failure;
1184                         tcp_key = nla_data(nla);
1185                         if (swkey->eth.type == htons(ETH_P_IP)) {
1186                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1187                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1188                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1189                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1190                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1191                         }
1192                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1193                         struct ovs_key_udp *udp_key;
1194
1195                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1196                         if (!nla)
1197                                 goto nla_put_failure;
1198                         udp_key = nla_data(nla);
1199                         if (swkey->eth.type == htons(ETH_P_IP)) {
1200                                 udp_key->udp_src = swkey->ipv4.tp.src;
1201                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1202                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1203                                 udp_key->udp_src = swkey->ipv6.tp.src;
1204                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1205                         }
1206                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1207                            swkey->ip.proto == IPPROTO_ICMP) {
1208                         struct ovs_key_icmp *icmp_key;
1209
1210                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1211                         if (!nla)
1212                                 goto nla_put_failure;
1213                         icmp_key = nla_data(nla);
1214                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1215                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1216                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1217                            swkey->ip.proto == IPPROTO_ICMPV6) {
1218                         struct ovs_key_icmpv6 *icmpv6_key;
1219
1220                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1221                                                 sizeof(*icmpv6_key));
1222                         if (!nla)
1223                                 goto nla_put_failure;
1224                         icmpv6_key = nla_data(nla);
1225                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1226                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1227
1228                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1229                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1230                                 struct ovs_key_nd *nd_key;
1231
1232                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1233                                 if (!nla)
1234                                         goto nla_put_failure;
1235                                 nd_key = nla_data(nla);
1236                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1237                                                         sizeof(nd_key->nd_target));
1238                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1239                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1240                         }
1241                 }
1242         }
1243
1244         return 0;
1245
1246 nla_put_failure:
1247         return -EMSGSIZE;
1248 }
1249
1250 /* Initializes the flow module.
1251  * Returns zero if successful or a negative error code. */
1252 int flow_init(void)
1253 {
1254         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1255                                         0, NULL);
1256         if (flow_cache == NULL)
1257                 return -ENOMEM;
1258
1259         get_random_bytes(&hash_seed, sizeof(hash_seed));
1260
1261         return 0;
1262 }
1263
1264 /* Uninitializes the flow module. */
1265 void flow_exit(void)
1266 {
1267         kmem_cache_destroy(flow_cache);
1268 }