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/if_ether.h>
  36 #include <linux/ip.h>
  37 #include <linux/ipv6.h>
  38 #include <linux/tcp.h>
  39 #include <linux/udp.h>
  40 #include <linux/icmp.h>
  41 #include <linux/icmpv6.h>
  42 #include <linux/rculist.h>
  43 #include <net/ip.h>
  44 #include <net/ipv6.h>
  45 #include <net/ndisc.h>
  46
  47 static struct kmem_cache *flow_cache;
  48
  49 static int check_header(struct sk_buff *skb, int len)
  50 {
  51         if (unlikely(skb->len < len))
  52                 return -EINVAL;
  53         if (unlikely(!pskb_may_pull(skb, len)))
  54                 return -ENOMEM;
  55         return 0;
  56 }
  57
  58 static bool arphdr_ok(struct sk_buff *skb)
  59 {
  60         return pskb_may_pull(skb, skb_network_offset(skb) +
  61                                   sizeof(struct arp_eth_header));
  62 }
  63
  64 static int check_iphdr(struct sk_buff *skb)
  65 {
  66         unsigned int nh_ofs = skb_network_offset(skb);
  67         unsigned int ip_len;
  68         int err;
  69
  70         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  71         if (unlikely(err))
  72                 return err;
  73
  74         ip_len = ip_hdrlen(skb);
  75         if (unlikely(ip_len < sizeof(struct iphdr) ||
  76                      skb->len < nh_ofs + ip_len))
  77                 return -EINVAL;
  78
  79         skb_set_transport_header(skb, nh_ofs + ip_len);
  80         return 0;
  81 }
  82
  83 static bool tcphdr_ok(struct sk_buff *skb)
  84 {
  85         int th_ofs = skb_transport_offset(skb);
  86         int tcp_len;
  87
  88         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  89                 return false;
  90
  91         tcp_len = tcp_hdrlen(skb);
  92         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  93                      skb->len < th_ofs + tcp_len))
  94                 return false;
  95
  96         return true;
  97 }
  98
  99 static bool udphdr_ok(struct sk_buff *skb)
 100 {
 101         return pskb_may_pull(skb, skb_transport_offset(skb) +
 102                                   sizeof(struct udphdr));
 103 }
 104
 105 static bool icmphdr_ok(struct sk_buff *skb)
 106 {
 107         return pskb_may_pull(skb, skb_transport_offset(skb) +
 108                                   sizeof(struct icmphdr));
 109 }
 110
 111 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 112 {
 113         struct timespec cur_ts;
 114         u64 cur_ms, idle_ms;
 115
 116         ktime_get_ts(&cur_ts);
 117         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 118         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 119                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 120
 121         return cur_ms - idle_ms;
 122 }
 123
 124 #define SW_FLOW_KEY_OFFSET(field)               \
 125         (offsetof(struct sw_flow_key, field) +  \
 126          FIELD_SIZEOF(struct sw_flow_key, field))
 127
 128 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 129                          int *key_lenp)
 130 {
 131         unsigned int nh_ofs = skb_network_offset(skb);
 132         unsigned int nh_len;
 133         int payload_ofs;
 134         struct ipv6hdr *nh;
 135         uint8_t nexthdr;
 136         __be16 frag_off;
 137         int err;
 138
 139         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 140
 141         err = check_header(skb, nh_ofs + sizeof(*nh));
 142         if (unlikely(err))
 143                 return err;
 144
 145         nh = ipv6_hdr(skb);
 146         nexthdr = nh->nexthdr;
 147         payload_ofs = (u8 *)(nh + 1) - skb->data;
 148
 149         key->ip.proto = NEXTHDR_NONE;
 150         key->ip.tos = ipv6_get_dsfield(nh);
 151         key->ip.ttl = nh->hop_limit;
 152         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 153         key->ipv6.addr.src = nh->saddr;
 154         key->ipv6.addr.dst = nh->daddr;
 155
 156         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 157         if (unlikely(payload_ofs < 0))
 158                 return -EINVAL;
 159
 160         if (frag_off) {
 161                 if (frag_off & htons(~0x7))
 162                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 163                 else
 164                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 165         }
 166
 167         nh_len = payload_ofs - nh_ofs;
 168         skb_set_transport_header(skb, nh_ofs + nh_len);
 169         key->ip.proto = nexthdr;
 170         return nh_len;
 171 }
 172
 173 static bool icmp6hdr_ok(struct sk_buff *skb)
 174 {
 175         return pskb_may_pull(skb, skb_transport_offset(skb) +
 176                                   sizeof(struct icmp6hdr));
 177 }
 178
 179 #define TCP_FLAGS_OFFSET 13
 180 #define TCP_FLAG_MASK 0x3f
 181
 182 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 183 {
 184         u8 tcp_flags = 0;
 185
 186         if (flow->key.eth.type == htons(ETH_P_IP) &&
 187             flow->key.ip.proto == IPPROTO_TCP) {
 188                 u8 *tcp = (u8 *)tcp_hdr(skb);
 189                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 190         }
 191
 192         spin_lock(&flow->lock);
 193         flow->used = jiffies;
 194         flow->packet_count++;
 195         flow->byte_count += skb->len;
 196         flow->tcp_flags |= tcp_flags;
 197         spin_unlock(&flow->lock);
 198 }
 199
 200 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 201 {
 202         int actions_len = nla_len(actions);
 203         struct sw_flow_actions *sfa;
 204
 205         /* At least DP_MAX_PORTS actions are required to be able to flood a
 206          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 207          * etc. */
 208         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 209                 return ERR_PTR(-EINVAL);
 210
 211         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 212         if (!sfa)
 213                 return ERR_PTR(-ENOMEM);
 214
 215         sfa->actions_len = actions_len;
 216         memcpy(sfa->actions, nla_data(actions), actions_len);
 217         return sfa;
 218 }
 219
 220 struct sw_flow *ovs_flow_alloc(void)
 221 {
 222         struct sw_flow *flow;
 223
 224         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 225         if (!flow)
 226                 return ERR_PTR(-ENOMEM);
 227
 228         spin_lock_init(&flow->lock);
 229         flow->sf_acts = NULL;
 230
 231         return flow;
 232 }
 233
 234 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 235 {
 236         hash = jhash_1word(hash, table->hash_seed);
 237         return flex_array_get(table->buckets,
 238                                 (hash & (table->n_buckets - 1)));
 239 }
 240
 241 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 242 {
 243         struct flex_array *buckets;
 244         int i, err;
 245
 246         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 247                                    n_buckets, GFP_KERNEL);
 248         if (!buckets)
 249                 return NULL;
 250
 251         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 252         if (err) {
 253                 flex_array_free(buckets);
 254                 return NULL;
 255         }
 256
 257         for (i = 0; i < n_buckets; i++)
 258                 INIT_HLIST_HEAD((struct hlist_head *)
 259                                         flex_array_get(buckets, i));
 260
 261         return buckets;
 262 }
 263
 264 static void free_buckets(struct flex_array *buckets)
 265 {
 266         flex_array_free(buckets);
 267 }
 268
 269 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 270 {
 271         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 272
 273         if (!table)
 274                 return NULL;
 275
 276         table->buckets = alloc_buckets(new_size);
 277
 278         if (!table->buckets) {
 279                 kfree(table);
 280                 return NULL;
 281         }
 282         table->n_buckets = new_size;
 283         table->count = 0;
 284         table->node_ver = 0;
 285         table->keep_flows = false;
 286         get_random_bytes(&table->hash_seed, sizeof(u32));
 287
 288         return table;
 289 }
 290
 291 void ovs_flow_tbl_destroy(struct flow_table *table)
 292 {
 293         int i;
 294
 295         if (!table)
 296                 return;
 297
 298         if (table->keep_flows)
 299                 goto skip_flows;
 300
 301         for (i = 0; i < table->n_buckets; i++) {
 302                 struct sw_flow *flow;
 303                 struct hlist_head *head = flex_array_get(table->buckets, i);
 304                 struct hlist_node *node, *n;
 305                 int ver = table->node_ver;
 306
 307                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 308                         hlist_del_rcu(&flow->hash_node[ver]);
 309                         ovs_flow_free(flow);
 310                 }
 311         }
 312
 313 skip_flows:
 314         free_buckets(table->buckets);
 315         kfree(table);
 316 }
 317
 318 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 319 {
 320         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 321
 322         ovs_flow_tbl_destroy(table);
 323 }
 324
 325 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 326 {
 327         if (!table)
 328                 return;
 329
 330         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 331 }
 332
 333 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 334 {
 335         struct sw_flow *flow;
 336         struct hlist_head *head;
 337         struct hlist_node *n;
 338         int ver;
 339         int i;
 340
 341         ver = table->node_ver;
 342         while (*bucket < table->n_buckets) {
 343                 i = 0;
 344                 head = flex_array_get(table->buckets, *bucket);
 345                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 346                         if (i < *last) {
 347                                 i++;
 348                                 continue;
 349                         }
 350                         *last = i + 1;
 351                         return flow;
 352                 }
 353                 (*bucket)++;
 354                 *last = 0;
 355         }
 356
 357         return NULL;
 358 }
 359
 360 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 361 {
 362         int old_ver;
 363         int i;
 364
 365         old_ver = old->node_ver;
 366         new->node_ver = !old_ver;
 367
 368         /* Insert in new table. */
 369         for (i = 0; i < old->n_buckets; i++) {
 370                 struct sw_flow *flow;
 371                 struct hlist_head *head;
 372                 struct hlist_node *n;
 373
 374                 head = flex_array_get(old->buckets, i);
 375
 376                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 377                         ovs_flow_tbl_insert(new, flow);
 378         }
 379         old->keep_flows = true;
 380 }
 381
 382 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 383 {
 384         struct flow_table *new_table;
 385
 386         new_table = ovs_flow_tbl_alloc(n_buckets);
 387         if (!new_table)
 388                 return ERR_PTR(-ENOMEM);
 389
 390         flow_table_copy_flows(table, new_table);
 391
 392         return new_table;
 393 }
 394
 395 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 396 {
 397         return __flow_tbl_rehash(table, table->n_buckets);
 398 }
 399
 400 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 401 {
 402         return __flow_tbl_rehash(table, table->n_buckets * 2);
 403 }
 404
 405 void ovs_flow_free(struct sw_flow *flow)
 406 {
 407         if (unlikely(!flow))
 408                 return;
 409
 410         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 411         kmem_cache_free(flow_cache, flow);
 412 }
 413
 414 /* RCU callback used by ovs_flow_deferred_free. */
 415 static void rcu_free_flow_callback(struct rcu_head *rcu)
 416 {
 417         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 418
 419         ovs_flow_free(flow);
 420 }
 421
 422 /* Schedules 'flow' to be freed after the next RCU grace period.
 423  * The caller must hold rcu_read_lock for this to be sensible. */
 424 void ovs_flow_deferred_free(struct sw_flow *flow)
 425 {
 426         call_rcu(&flow->rcu, rcu_free_flow_callback);
 427 }
 428
 429 /* RCU callback used by ovs_flow_deferred_free_acts. */
 430 static void rcu_free_acts_callback(struct rcu_head *rcu)
 431 {
 432         struct sw_flow_actions *sf_acts = container_of(rcu,
 433                         struct sw_flow_actions, rcu);
 434         kfree(sf_acts);
 435 }
 436
 437 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 438  * The caller must hold rcu_read_lock for this to be sensible. */
 439 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 440 {
 441         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 442 }
 443
 444 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 445 {
 446         struct qtag_prefix {
 447                 __be16 eth_type; /* ETH_P_8021Q */
 448                 __be16 tci;
 449         };
 450         struct qtag_prefix *qp;
 451
 452         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 453                 return 0;
 454
 455         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 456                                          sizeof(__be16))))
 457                 return -ENOMEM;
 458
 459         qp = (struct qtag_prefix *) skb->data;
 460         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 461         __skb_pull(skb, sizeof(struct qtag_prefix));
 462
 463         return 0;
 464 }
 465
 466 static __be16 parse_ethertype(struct sk_buff *skb)
 467 {
 468         struct llc_snap_hdr {
 469                 u8  dsap;  /* Always 0xAA */
 470                 u8  ssap;  /* Always 0xAA */
 471                 u8  ctrl;
 472                 u8  oui[3];
 473                 __be16 ethertype;
 474         };
 475         struct llc_snap_hdr *llc;
 476         __be16 proto;
 477
 478         proto = *(__be16 *) skb->data;
 479         __skb_pull(skb, sizeof(__be16));
 480
 481         if (ntohs(proto) >= 1536)
 482                 return proto;
 483
 484         if (skb->len < sizeof(struct llc_snap_hdr))
 485                 return htons(ETH_P_802_2);
 486
 487         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 488                 return htons(0);
 489
 490         llc = (struct llc_snap_hdr *) skb->data;
 491         if (llc->dsap != LLC_SAP_SNAP ||
 492             llc->ssap != LLC_SAP_SNAP ||
 493             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 494                 return htons(ETH_P_802_2);
 495
 496         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 497         return llc->ethertype;
 498 }
 499
 500 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 501                         int *key_lenp, int nh_len)
 502 {
 503         struct icmp6hdr *icmp = icmp6_hdr(skb);
 504         int error = 0;
 505         int key_len;
 506
 507         /* The ICMPv6 type and code fields use the 16-bit transport port
 508          * fields, so we need to store them in 16-bit network byte order.
 509          */
 510         key->ipv6.tp.src = htons(icmp->icmp6_type);
 511         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 512         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 513
 514         if (icmp->icmp6_code == 0 &&
 515             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 516              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 517                 int icmp_len = skb->len - skb_transport_offset(skb);
 518                 struct nd_msg *nd;
 519                 int offset;
 520
 521                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 522
 523                 /* In order to process neighbor discovery options, we need the
 524                  * entire packet.
 525                  */
 526                 if (unlikely(icmp_len < sizeof(*nd)))
 527                         goto out;
 528                 if (unlikely(skb_linearize(skb))) {
 529                         error = -ENOMEM;
 530                         goto out;
 531                 }
 532
 533                 nd = (struct nd_msg *)skb_transport_header(skb);
 534                 key->ipv6.nd.target = nd->target;
 535                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 536
 537                 icmp_len -= sizeof(*nd);
 538                 offset = 0;
 539                 while (icmp_len >= 8) {
 540                         struct nd_opt_hdr *nd_opt =
 541                                  (struct nd_opt_hdr *)(nd->opt + offset);
 542                         int opt_len = nd_opt->nd_opt_len * 8;
 543
 544                         if (unlikely(!opt_len || opt_len > icmp_len))
 545                                 goto invalid;
 546
 547                         /* Store the link layer address if the appropriate
 548                          * option is provided.  It is considered an error if
 549                          * the same link layer option is specified twice.
 550                          */
 551                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 552                             && opt_len == 8) {
 553                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 554                                         goto invalid;
 555                                 memcpy(key->ipv6.nd.sll,
 556                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 557                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 558                                    && opt_len == 8) {
 559                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 560                                         goto invalid;
 561                                 memcpy(key->ipv6.nd.tll,
 562                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 563                         }
 564
 565                         icmp_len -= opt_len;
 566                         offset += opt_len;
 567                 }
 568         }
 569
 570         goto out;
 571
 572 invalid:
 573         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 574         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 575         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 576
 577 out:
 578         *key_lenp = key_len;
 579         return error;
 580 }
 581
 582 /**
 583  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 584  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 585  * Ethernet header
 586  * @in_port: port number on which @skb was received.
 587  * @key: output flow key
 588  * @key_lenp: length of output flow key
 589  *
 590  * The caller must ensure that skb->len >= ETH_HLEN.
 591  *
 592  * Returns 0 if successful, otherwise a negative errno value.
 593  *
 594  * Initializes @skb header pointers as follows:
 595  *
 596  *    - skb->mac_header: the Ethernet header.
 597  *
 598  *    - skb->network_header: just past the Ethernet header, or just past the
 599  *      VLAN header, to the first byte of the Ethernet payload.
 600  *
 601  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 602  *      on output, then just past the IP header, if one is present and
 603  *      of a correct length, otherwise the same as skb->network_header.
 604  *      For other key->dl_type values it is left untouched.
 605  */
 606 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 607                  int *key_lenp)
 608 {
 609         int error = 0;
 610         int key_len = SW_FLOW_KEY_OFFSET(eth);
 611         struct ethhdr *eth;
 612
 613         memset(key, 0, sizeof(*key));
 614
 615         key->phy.priority = skb->priority;
 616         key->phy.in_port = in_port;
 617
 618         skb_reset_mac_header(skb);
 619
 620         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 621          * header in the linear data area.
 622          */
 623         eth = eth_hdr(skb);
 624         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 625         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 626
 627         __skb_pull(skb, 2 * ETH_ALEN);
 628
 629         if (vlan_tx_tag_present(skb))
 630                 key->eth.tci = htons(skb->vlan_tci);
 631         else if (eth->h_proto == htons(ETH_P_8021Q))
 632                 if (unlikely(parse_vlan(skb, key)))
 633                         return -ENOMEM;
 634
 635         key->eth.type = parse_ethertype(skb);
 636         if (unlikely(key->eth.type == htons(0)))
 637                 return -ENOMEM;
 638
 639         skb_reset_network_header(skb);
 640         __skb_push(skb, skb->data - skb_mac_header(skb));
 641
 642         /* Network layer. */
 643         if (key->eth.type == htons(ETH_P_IP)) {
 644                 struct iphdr *nh;
 645                 __be16 offset;
 646
 647                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 648
 649                 error = check_iphdr(skb);
 650                 if (unlikely(error)) {
 651                         if (error == -EINVAL) {
 652                                 skb->transport_header = skb->network_header;
 653                                 error = 0;
 654                         }
 655                         goto out;
 656                 }
 657
 658                 nh = ip_hdr(skb);
 659                 key->ipv4.addr.src = nh->saddr;
 660                 key->ipv4.addr.dst = nh->daddr;
 661
 662                 key->ip.proto = nh->protocol;
 663                 key->ip.tos = nh->tos;
 664                 key->ip.ttl = nh->ttl;
 665
 666                 offset = nh->frag_off & htons(IP_OFFSET);
 667                 if (offset) {
 668                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 669                         goto out;
 670                 }
 671                 if (nh->frag_off & htons(IP_MF) ||
 672                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 673                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 674
 675                 /* Transport layer. */
 676                 if (key->ip.proto == IPPROTO_TCP) {
 677                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 678                         if (tcphdr_ok(skb)) {
 679                                 struct tcphdr *tcp = tcp_hdr(skb);
 680                                 key->ipv4.tp.src = tcp->source;
 681                                 key->ipv4.tp.dst = tcp->dest;
 682                         }
 683                 } else if (key->ip.proto == IPPROTO_UDP) {
 684                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 685                         if (udphdr_ok(skb)) {
 686                                 struct udphdr *udp = udp_hdr(skb);
 687                                 key->ipv4.tp.src = udp->source;
 688                                 key->ipv4.tp.dst = udp->dest;
 689                         }
 690                 } else if (key->ip.proto == IPPROTO_ICMP) {
 691                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 692                         if (icmphdr_ok(skb)) {
 693                                 struct icmphdr *icmp = icmp_hdr(skb);
 694                                 /* The ICMP type and code fields use the 16-bit
 695                                  * transport port fields, so we need to store
 696                                  * them in 16-bit network byte order. */
 697                                 key->ipv4.tp.src = htons(icmp->type);
 698                                 key->ipv4.tp.dst = htons(icmp->code);
 699                         }
 700                 }
 701
 702         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 703                 struct arp_eth_header *arp;
 704
 705                 arp = (struct arp_eth_header *)skb_network_header(skb);
 706
 707                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 708                                 && arp->ar_pro == htons(ETH_P_IP)
 709                                 && arp->ar_hln == ETH_ALEN
 710                                 && arp->ar_pln == 4) {
 711
 712                         /* We only match on the lower 8 bits of the opcode. */
 713                         if (ntohs(arp->ar_op) <= 0xff)
 714                                 key->ip.proto = ntohs(arp->ar_op);
 715
 716                         if (key->ip.proto == ARPOP_REQUEST
 717                                         || key->ip.proto == ARPOP_REPLY) {
 718                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 719                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 720                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 721                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 722                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 723                         }
 724                 }
 725         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 726                 int nh_len;             /* IPv6 Header + Extensions */
 727
 728                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 729                 if (unlikely(nh_len < 0)) {
 730                         if (nh_len == -EINVAL)
 731                                 skb->transport_header = skb->network_header;
 732                         else
 733                                 error = nh_len;
 734                         goto out;
 735                 }
 736
 737                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 738                         goto out;
 739                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 740                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 741
 742                 /* Transport layer. */
 743                 if (key->ip.proto == NEXTHDR_TCP) {
 744                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 745                         if (tcphdr_ok(skb)) {
 746                                 struct tcphdr *tcp = tcp_hdr(skb);
 747                                 key->ipv6.tp.src = tcp->source;
 748                                 key->ipv6.tp.dst = tcp->dest;
 749                         }
 750                 } else if (key->ip.proto == NEXTHDR_UDP) {
 751                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 752                         if (udphdr_ok(skb)) {
 753                                 struct udphdr *udp = udp_hdr(skb);
 754                                 key->ipv6.tp.src = udp->source;
 755                                 key->ipv6.tp.dst = udp->dest;
 756                         }
 757                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 758                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 759                         if (icmp6hdr_ok(skb)) {
 760                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 761                                 if (error < 0)
 762                                         goto out;
 763                         }
 764                 }
 765         }
 766
 767 out:
 768         *key_lenp = key_len;
 769         return error;
 770 }
 771
 772 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 773 {
 774         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 775 }
 776
 777 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 778                                 struct sw_flow_key *key, int key_len)
 779 {
 780         struct sw_flow *flow;
 781         struct hlist_node *n;
 782         struct hlist_head *head;
 783         u32 hash;
 784
 785         hash = ovs_flow_hash(key, key_len);
 786
 787         head = find_bucket(table, hash);
 788         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 789
 790                 if (flow->hash == hash &&
 791                     !memcmp(&flow->key, key, key_len)) {
 792                         return flow;
 793                 }
 794         }
 795         return NULL;
 796 }
 797
 798 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 799 {
 800         struct hlist_head *head;
 801
 802         head = find_bucket(table, flow->hash);
 803         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 804         table->count++;
 805 }
 806
 807 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 808 {
 809         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 810         table->count--;
 811         BUG_ON(table->count < 0);
 812 }
 813
 814 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 815 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 816         [OVS_KEY_ATTR_ENCAP] = -1,
 817         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 818         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 819         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 820         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 821         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 822         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 823         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 824         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 825         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 826         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 827         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 828         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 829         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 830 };
 831
 832 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 833                                   const struct nlattr *a[], u32 *attrs)
 834 {
 835         const struct ovs_key_icmp *icmp_key;
 836         const struct ovs_key_tcp *tcp_key;
 837         const struct ovs_key_udp *udp_key;
 838
 839         switch (swkey->ip.proto) {
 840         case IPPROTO_TCP:
 841                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 842                         return -EINVAL;
 843                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 844
 845                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 846                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 847                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 848                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 849                 break;
 850
 851         case IPPROTO_UDP:
 852                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 853                         return -EINVAL;
 854                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 855
 856                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 857                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 858                 swkey->ipv4.tp.src = udp_key->udp_src;
 859                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 860                 break;
 861
 862         case IPPROTO_ICMP:
 863                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 864                         return -EINVAL;
 865                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 866
 867                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 868                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 869                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 870                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 871                 break;
 872         }
 873
 874         return 0;
 875 }
 876
 877 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 878                                   const struct nlattr *a[], u32 *attrs)
 879 {
 880         const struct ovs_key_icmpv6 *icmpv6_key;
 881         const struct ovs_key_tcp *tcp_key;
 882         const struct ovs_key_udp *udp_key;
 883
 884         switch (swkey->ip.proto) {
 885         case IPPROTO_TCP:
 886                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 887                         return -EINVAL;
 888                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 889
 890                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 891                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 892                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 893                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 894                 break;
 895
 896         case IPPROTO_UDP:
 897                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 898                         return -EINVAL;
 899                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 900
 901                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 902                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 903                 swkey->ipv6.tp.src = udp_key->udp_src;
 904                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 905                 break;
 906
 907         case IPPROTO_ICMPV6:
 908                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 909                         return -EINVAL;
 910                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 911
 912                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 913                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 914                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 915                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 916
 917                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 918                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 919                         const struct ovs_key_nd *nd_key;
 920
 921                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 922                                 return -EINVAL;
 923                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 924
 925                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 926                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 927                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 928                                sizeof(swkey->ipv6.nd.target));
 929                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 930                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 931                 }
 932                 break;
 933         }
 934
 935         return 0;
 936 }
 937
 938 static int parse_flow_nlattrs(const struct nlattr *attr,
 939                               const struct nlattr *a[], u32 *attrsp)
 940 {
 941         const struct nlattr *nla;
 942         u32 attrs;
 943         int rem;
 944
 945         attrs = 0;
 946         nla_for_each_nested(nla, attr, rem) {
 947                 u16 type = nla_type(nla);
 948                 int expected_len;
 949
 950                 if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
 951                         return -EINVAL;
 952
 953                 expected_len = ovs_key_lens[type];
 954                 if (nla_len(nla) != expected_len && expected_len != -1)
 955                         return -EINVAL;
 956
 957                 attrs |= 1 << type;
 958                 a[type] = nla;
 959         }
 960         if (rem)
 961                 return -EINVAL;
 962
 963         *attrsp = attrs;
 964         return 0;
 965 }
 966
 967 /**
 968  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 969  * @swkey: receives the extracted flow key.
 970  * @key_lenp: number of bytes used in @swkey.
 971  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 972  * sequence.
 973  */
 974 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 975                       const struct nlattr *attr)
 976 {
 977         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 978         const struct ovs_key_ethernet *eth_key;
 979         int key_len;
 980         u32 attrs;
 981         int err;
 982
 983         memset(swkey, 0, sizeof(struct sw_flow_key));
 984         key_len = SW_FLOW_KEY_OFFSET(eth);
 985
 986         err = parse_flow_nlattrs(attr, a, &attrs);
 987         if (err)
 988                 return err;
 989
 990         /* Metadata attributes. */
 991         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
 992                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
 993                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
 994         }
 995         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
 996                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
 997                 if (in_port >= DP_MAX_PORTS)
 998                         return -EINVAL;
 999                 swkey->phy.in_port = in_port;
1000                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1001         } else {
1002                 swkey->phy.in_port = USHRT_MAX;
1003         }
1004
1005         /* Data attributes. */
1006         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1007                 return -EINVAL;
1008         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1009
1010         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1011         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1012         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1013
1014         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1015             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1016                 const struct nlattr *encap;
1017                 __be16 tci;
1018
1019                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1020                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1021                               (1 << OVS_KEY_ATTR_ENCAP)))
1022                         return -EINVAL;
1023
1024                 encap = a[OVS_KEY_ATTR_ENCAP];
1025                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1026                 if (tci & htons(VLAN_TAG_PRESENT)) {
1027                         swkey->eth.tci = tci;
1028
1029                         err = parse_flow_nlattrs(encap, a, &attrs);
1030                         if (err)
1031                                 return err;
1032                 } else if (!tci) {
1033                         /* Corner case for truncated 802.1Q header. */
1034                         if (nla_len(encap))
1035                                 return -EINVAL;
1036
1037                         swkey->eth.type = htons(ETH_P_8021Q);
1038                         *key_lenp = key_len;
1039                         return 0;
1040                 } else {
1041                         return -EINVAL;
1042                 }
1043         }
1044
1045         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1046                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1047                 if (ntohs(swkey->eth.type) < 1536)
1048                         return -EINVAL;
1049                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1050         } else {
1051                 swkey->eth.type = htons(ETH_P_802_2);
1052         }
1053
1054         if (swkey->eth.type == htons(ETH_P_IP)) {
1055                 const struct ovs_key_ipv4 *ipv4_key;
1056
1057                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1058                         return -EINVAL;
1059                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1060
1061                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1062                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1063                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1064                         return -EINVAL;
1065                 swkey->ip.proto = ipv4_key->ipv4_proto;
1066                 swkey->ip.tos = ipv4_key->ipv4_tos;
1067                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1068                 swkey->ip.frag = ipv4_key->ipv4_frag;
1069                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1070                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1071
1072                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1073                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1074                         if (err)
1075                                 return err;
1076                 }
1077         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1078                 const struct ovs_key_ipv6 *ipv6_key;
1079
1080                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1081                         return -EINVAL;
1082                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1083
1084                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1085                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1086                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1087                         return -EINVAL;
1088                 swkey->ipv6.label = ipv6_key->ipv6_label;
1089                 swkey->ip.proto = ipv6_key->ipv6_proto;
1090                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1091                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1092                 swkey->ip.frag = ipv6_key->ipv6_frag;
1093                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1094                        sizeof(swkey->ipv6.addr.src));
1095                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1096                        sizeof(swkey->ipv6.addr.dst));
1097
1098                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1099                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1100                         if (err)
1101                                 return err;
1102                 }
1103         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1104                 const struct ovs_key_arp *arp_key;
1105
1106                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1107                         return -EINVAL;
1108                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1109
1110                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1111                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1112                 swkey->ipv4.addr.src = arp_key->arp_sip;
1113                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1114                 if (arp_key->arp_op & htons(0xff00))
1115                         return -EINVAL;
1116                 swkey->ip.proto = ntohs(arp_key->arp_op);
1117                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1118                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1119         }
1120
1121         if (attrs)
1122                 return -EINVAL;
1123         *key_lenp = key_len;
1124
1125         return 0;
1126 }
1127
1128 /**
1129  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1130  * @in_port: receives the extracted input port.
1131  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1132  * sequence.
1133  *
1134  * This parses a series of Netlink attributes that form a flow key, which must
1135  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1136  * get the metadata, that is, the parts of the flow key that cannot be
1137  * extracted from the packet itself.
1138  */
1139 int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
1140                                const struct nlattr *attr)
1141 {
1142         const struct nlattr *nla;
1143         int rem;
1144
1145         *in_port = USHRT_MAX;
1146         *priority = 0;
1147
1148         nla_for_each_nested(nla, attr, rem) {
1149                 int type = nla_type(nla);
1150
1151                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1152                         if (nla_len(nla) != ovs_key_lens[type])
1153                                 return -EINVAL;
1154
1155                         switch (type) {
1156                         case OVS_KEY_ATTR_PRIORITY:
1157                                 *priority = nla_get_u32(nla);
1158                                 break;
1159
1160                         case OVS_KEY_ATTR_IN_PORT:
1161                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1162                                         return -EINVAL;
1163                                 *in_port = nla_get_u32(nla);
1164                                 break;
1165                         }
1166                 }
1167         }
1168         if (rem)
1169                 return -EINVAL;
1170         return 0;
1171 }
1172
1173 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1174 {
1175         struct ovs_key_ethernet *eth_key;
1176         struct nlattr *nla, *encap;
1177
1178         if (swkey->phy.priority)
1179                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1180
1181         if (swkey->phy.in_port != USHRT_MAX)
1182                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1183
1184         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1185         if (!nla)
1186                 goto nla_put_failure;
1187         eth_key = nla_data(nla);
1188         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1189         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1190
1191         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1192                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q));
1193                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci);
1194                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1195                 if (!swkey->eth.tci)
1196                         goto unencap;
1197         } else {
1198                 encap = NULL;
1199         }
1200
1201         if (swkey->eth.type == htons(ETH_P_802_2))
1202                 goto unencap;
1203
1204         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1205
1206         if (swkey->eth.type == htons(ETH_P_IP)) {
1207                 struct ovs_key_ipv4 *ipv4_key;
1208
1209                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1210                 if (!nla)
1211                         goto nla_put_failure;
1212                 ipv4_key = nla_data(nla);
1213                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1214                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1215                 ipv4_key->ipv4_proto = swkey->ip.proto;
1216                 ipv4_key->ipv4_tos = swkey->ip.tos;
1217                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1218                 ipv4_key->ipv4_frag = swkey->ip.frag;
1219         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1220                 struct ovs_key_ipv6 *ipv6_key;
1221
1222                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1223                 if (!nla)
1224                         goto nla_put_failure;
1225                 ipv6_key = nla_data(nla);
1226                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1227                                 sizeof(ipv6_key->ipv6_src));
1228                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1229                                 sizeof(ipv6_key->ipv6_dst));
1230                 ipv6_key->ipv6_label = swkey->ipv6.label;
1231                 ipv6_key->ipv6_proto = swkey->ip.proto;
1232                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1233                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1234                 ipv6_key->ipv6_frag = swkey->ip.frag;
1235         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1236                 struct ovs_key_arp *arp_key;
1237
1238                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1239                 if (!nla)
1240                         goto nla_put_failure;
1241                 arp_key = nla_data(nla);
1242                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1243                 arp_key->arp_sip = swkey->ipv4.addr.src;
1244                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1245                 arp_key->arp_op = htons(swkey->ip.proto);
1246                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1247                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1248         }
1249
1250         if ((swkey->eth.type == htons(ETH_P_IP) ||
1251              swkey->eth.type == htons(ETH_P_IPV6)) &&
1252              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1253
1254                 if (swkey->ip.proto == IPPROTO_TCP) {
1255                         struct ovs_key_tcp *tcp_key;
1256
1257                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1258                         if (!nla)
1259                                 goto nla_put_failure;
1260                         tcp_key = nla_data(nla);
1261                         if (swkey->eth.type == htons(ETH_P_IP)) {
1262                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1263                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1264                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1265                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1266                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1267                         }
1268                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1269                         struct ovs_key_udp *udp_key;
1270
1271                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1272                         if (!nla)
1273                                 goto nla_put_failure;
1274                         udp_key = nla_data(nla);
1275                         if (swkey->eth.type == htons(ETH_P_IP)) {
1276                                 udp_key->udp_src = swkey->ipv4.tp.src;
1277                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1278                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1279                                 udp_key->udp_src = swkey->ipv6.tp.src;
1280                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1281                         }
1282                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1283                            swkey->ip.proto == IPPROTO_ICMP) {
1284                         struct ovs_key_icmp *icmp_key;
1285
1286                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1287                         if (!nla)
1288                                 goto nla_put_failure;
1289                         icmp_key = nla_data(nla);
1290                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1291                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1292                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1293                            swkey->ip.proto == IPPROTO_ICMPV6) {
1294                         struct ovs_key_icmpv6 *icmpv6_key;
1295
1296                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1297                                                 sizeof(*icmpv6_key));
1298                         if (!nla)
1299                                 goto nla_put_failure;
1300                         icmpv6_key = nla_data(nla);
1301                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1302                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1303
1304                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1305                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1306                                 struct ovs_key_nd *nd_key;
1307
1308                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1309                                 if (!nla)
1310                                         goto nla_put_failure;
1311                                 nd_key = nla_data(nla);
1312                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1313                                                         sizeof(nd_key->nd_target));
1314                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1315                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1316                         }
1317                 }
1318         }
1319
1320 unencap:
1321         if (encap)
1322                 nla_nest_end(skb, encap);
1323
1324         return 0;
1325
1326 nla_put_failure:
1327         return -EMSGSIZE;
1328 }
1329
1330 /* Initializes the flow module.
1331  * Returns zero if successful or a negative error code. */
1332 int ovs_flow_init(void)
1333 {
1334         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1335                                         0, NULL);
1336         if (flow_cache == NULL)
1337                 return -ENOMEM;
1338
1339         return 0;
1340 }
1341
1342 /* Uninitializes the flow module. */
1343 void ovs_flow_exit(void)
1344 {
1345         kmem_cache_destroy(flow_cache);
1346 }