datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #include "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 #include "vlan.h"
  47
  48 static struct kmem_cache *flow_cache;
  49
  50 static int check_header(struct sk_buff *skb, int len)
  51 {
  52         if (unlikely(skb->len < len))
  53                 return -EINVAL;
  54         if (unlikely(!pskb_may_pull(skb, len)))
  55                 return -ENOMEM;
  56         return 0;
  57 }
  58
  59 static bool arphdr_ok(struct sk_buff *skb)
  60 {
  61         return pskb_may_pull(skb, skb_network_offset(skb) +
  62                                   sizeof(struct arp_eth_header));
  63 }
  64
  65 static int check_iphdr(struct sk_buff *skb)
  66 {
  67         unsigned int nh_ofs = skb_network_offset(skb);
  68         unsigned int ip_len;
  69         int err;
  70
  71         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  72         if (unlikely(err))
  73                 return err;
  74
  75         ip_len = ip_hdrlen(skb);
  76         if (unlikely(ip_len < sizeof(struct iphdr) ||
  77                      skb->len < nh_ofs + ip_len))
  78                 return -EINVAL;
  79
  80         skb_set_transport_header(skb, nh_ofs + ip_len);
  81         return 0;
  82 }
  83
  84 static bool tcphdr_ok(struct sk_buff *skb)
  85 {
  86         int th_ofs = skb_transport_offset(skb);
  87         int tcp_len;
  88
  89         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  90                 return false;
  91
  92         tcp_len = tcp_hdrlen(skb);
  93         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  94                      skb->len < th_ofs + tcp_len))
  95                 return false;
  96
  97         return true;
  98 }
  99
 100 static bool udphdr_ok(struct sk_buff *skb)
 101 {
 102         return pskb_may_pull(skb, skb_transport_offset(skb) +
 103                                   sizeof(struct udphdr));
 104 }
 105
 106 static bool icmphdr_ok(struct sk_buff *skb)
 107 {
 108         return pskb_may_pull(skb, skb_transport_offset(skb) +
 109                                   sizeof(struct icmphdr));
 110 }
 111
 112 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 113 {
 114         struct timespec cur_ts;
 115         u64 cur_ms, idle_ms;
 116
 117         ktime_get_ts(&cur_ts);
 118         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 119         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 120                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 121
 122         return cur_ms - idle_ms;
 123 }
 124
 125 #define SW_FLOW_KEY_OFFSET(field)               \
 126         (offsetof(struct sw_flow_key, field) +  \
 127          FIELD_SIZEOF(struct sw_flow_key, field))
 128
 129 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 130                          int *key_lenp)
 131 {
 132         unsigned int nh_ofs = skb_network_offset(skb);
 133         unsigned int nh_len;
 134         int payload_ofs;
 135         struct ipv6hdr *nh;
 136         uint8_t nexthdr;
 137         __be16 frag_off;
 138         int err;
 139
 140         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 141
 142         err = check_header(skb, nh_ofs + sizeof(*nh));
 143         if (unlikely(err))
 144                 return err;
 145
 146         nh = ipv6_hdr(skb);
 147         nexthdr = nh->nexthdr;
 148         payload_ofs = (u8 *)(nh + 1) - skb->data;
 149
 150         key->ip.proto = NEXTHDR_NONE;
 151         key->ip.tos = ipv6_get_dsfield(nh);
 152         key->ip.ttl = nh->hop_limit;
 153         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 154         key->ipv6.addr.src = nh->saddr;
 155         key->ipv6.addr.dst = nh->daddr;
 156
 157         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 158         if (unlikely(payload_ofs < 0))
 159                 return -EINVAL;
 160
 161         if (frag_off) {
 162                 if (frag_off & htons(~0x7))
 163                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 164                 else
 165                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 166         }
 167
 168         nh_len = payload_ofs - nh_ofs;
 169         skb_set_transport_header(skb, nh_ofs + nh_len);
 170         key->ip.proto = nexthdr;
 171         return nh_len;
 172 }
 173
 174 static bool icmp6hdr_ok(struct sk_buff *skb)
 175 {
 176         return pskb_may_pull(skb, skb_transport_offset(skb) +
 177                                   sizeof(struct icmp6hdr));
 178 }
 179
 180 #define TCP_FLAGS_OFFSET 13
 181 #define TCP_FLAG_MASK 0x3f
 182
 183 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 184 {
 185         u8 tcp_flags = 0;
 186
 187         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 188              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 189             flow->key.ip.proto == IPPROTO_TCP &&
 190             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 191                 u8 *tcp = (u8 *)tcp_hdr(skb);
 192                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 193         }
 194
 195         spin_lock(&flow->lock);
 196         flow->used = jiffies;
 197         flow->packet_count++;
 198         flow->byte_count += skb->len;
 199         flow->tcp_flags |= tcp_flags;
 200         spin_unlock(&flow->lock);
 201 }
 202
 203 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 204 {
 205         int actions_len = nla_len(actions);
 206         struct sw_flow_actions *sfa;
 207
 208         if (actions_len > MAX_ACTIONS_BUFSIZE)
 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         atomic_set(&flow->refcnt, 1);
 230         flow->sf_acts = NULL;
 231         flow->dead = false;
 232
 233         return flow;
 234 }
 235
 236 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 237 {
 238         hash = jhash_1word(hash, table->hash_seed);
 239         return flex_array_get(table->buckets,
 240                                 (hash & (table->n_buckets - 1)));
 241 }
 242
 243 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 244 {
 245         struct flex_array *buckets;
 246         int i, err;
 247
 248         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 249                                    n_buckets, GFP_KERNEL);
 250         if (!buckets)
 251                 return NULL;
 252
 253         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 254         if (err) {
 255                 flex_array_free(buckets);
 256                 return NULL;
 257         }
 258
 259         for (i = 0; i < n_buckets; i++)
 260                 INIT_HLIST_HEAD((struct hlist_head *)
 261                                         flex_array_get(buckets, i));
 262
 263         return buckets;
 264 }
 265
 266 static void free_buckets(struct flex_array *buckets)
 267 {
 268         flex_array_free(buckets);
 269 }
 270
 271 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 272 {
 273         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 274
 275         if (!table)
 276                 return NULL;
 277
 278         table->buckets = alloc_buckets(new_size);
 279
 280         if (!table->buckets) {
 281                 kfree(table);
 282                 return NULL;
 283         }
 284         table->n_buckets = new_size;
 285         table->count = 0;
 286         table->node_ver = 0;
 287         table->keep_flows = false;
 288         get_random_bytes(&table->hash_seed, sizeof(u32));
 289
 290         return table;
 291 }
 292
 293 static void flow_free(struct sw_flow *flow)
 294 {
 295         flow->dead = true;
 296         ovs_flow_put(flow);
 297 }
 298
 299 void ovs_flow_tbl_destroy(struct flow_table *table)
 300 {
 301         int i;
 302
 303         if (!table)
 304                 return;
 305
 306         if (table->keep_flows)
 307                 goto skip_flows;
 308
 309         for (i = 0; i < table->n_buckets; i++) {
 310                 struct sw_flow *flow;
 311                 struct hlist_head *head = flex_array_get(table->buckets, i);
 312                 struct hlist_node *node, *n;
 313                 int ver = table->node_ver;
 314
 315                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 316                         hlist_del_rcu(&flow->hash_node[ver]);
 317                         flow_free(flow);
 318                 }
 319         }
 320
 321 skip_flows:
 322         free_buckets(table->buckets);
 323         kfree(table);
 324 }
 325
 326 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 327 {
 328         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 329
 330         ovs_flow_tbl_destroy(table);
 331 }
 332
 333 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 334 {
 335         if (!table)
 336                 return;
 337
 338         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 339 }
 340
 341 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 342 {
 343         struct sw_flow *flow;
 344         struct hlist_head *head;
 345         struct hlist_node *n;
 346         int ver;
 347         int i;
 348
 349         ver = table->node_ver;
 350         while (*bucket < table->n_buckets) {
 351                 i = 0;
 352                 head = flex_array_get(table->buckets, *bucket);
 353                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 354                         if (i < *last) {
 355                                 i++;
 356                                 continue;
 357                         }
 358                         *last = i + 1;
 359                         return flow;
 360                 }
 361                 (*bucket)++;
 362                 *last = 0;
 363         }
 364
 365         return NULL;
 366 }
 367
 368 static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 369 {
 370         struct hlist_head *head;
 371         head = find_bucket(table, flow->hash);
 372         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 373         table->count++;
 374 }
 375
 376 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 377 {
 378         int old_ver;
 379         int i;
 380
 381         old_ver = old->node_ver;
 382         new->node_ver = !old_ver;
 383
 384         /* Insert in new table. */
 385         for (i = 0; i < old->n_buckets; i++) {
 386                 struct sw_flow *flow;
 387                 struct hlist_head *head;
 388                 struct hlist_node *n;
 389
 390                 head = flex_array_get(old->buckets, i);
 391
 392                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 393                         __flow_tbl_insert(new, flow);
 394         }
 395         old->keep_flows = true;
 396 }
 397
 398 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 399 {
 400         struct flow_table *new_table;
 401
 402         new_table = ovs_flow_tbl_alloc(n_buckets);
 403         if (!new_table)
 404                 return ERR_PTR(-ENOMEM);
 405
 406         flow_table_copy_flows(table, new_table);
 407
 408         return new_table;
 409 }
 410
 411 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 412 {
 413         return __flow_tbl_rehash(table, table->n_buckets);
 414 }
 415
 416 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 417 {
 418         return __flow_tbl_rehash(table, table->n_buckets * 2);
 419 }
 420
 421 /* RCU callback used by ovs_flow_deferred_free. */
 422 static void rcu_free_flow_callback(struct rcu_head *rcu)
 423 {
 424         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 425
 426         flow->dead = true;
 427         ovs_flow_put(flow);
 428 }
 429
 430 /* Schedules 'flow' to be freed after the next RCU grace period.
 431  * The caller must hold rcu_read_lock for this to be sensible. */
 432 void ovs_flow_deferred_free(struct sw_flow *flow)
 433 {
 434         call_rcu(&flow->rcu, rcu_free_flow_callback);
 435 }
 436
 437 void ovs_flow_hold(struct sw_flow *flow)
 438 {
 439         atomic_inc(&flow->refcnt);
 440 }
 441
 442 void ovs_flow_put(struct sw_flow *flow)
 443 {
 444         if (unlikely(!flow))
 445                 return;
 446
 447         if (atomic_dec_and_test(&flow->refcnt)) {
 448                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 449                 kmem_cache_free(flow_cache, flow);
 450         }
 451 }
 452
 453 /* RCU callback used by ovs_flow_deferred_free_acts. */
 454 static void rcu_free_acts_callback(struct rcu_head *rcu)
 455 {
 456         struct sw_flow_actions *sf_acts = container_of(rcu,
 457                         struct sw_flow_actions, rcu);
 458         kfree(sf_acts);
 459 }
 460
 461 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 462  * The caller must hold rcu_read_lock for this to be sensible. */
 463 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 464 {
 465         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 466 }
 467
 468 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 469 {
 470         struct qtag_prefix {
 471                 __be16 eth_type; /* ETH_P_8021Q */
 472                 __be16 tci;
 473         };
 474         struct qtag_prefix *qp;
 475
 476         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 477                 return 0;
 478
 479         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 480                                          sizeof(__be16))))
 481                 return -ENOMEM;
 482
 483         qp = (struct qtag_prefix *) skb->data;
 484         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 485         __skb_pull(skb, sizeof(struct qtag_prefix));
 486
 487         return 0;
 488 }
 489
 490 static __be16 parse_ethertype(struct sk_buff *skb)
 491 {
 492         struct llc_snap_hdr {
 493                 u8  dsap;  /* Always 0xAA */
 494                 u8  ssap;  /* Always 0xAA */
 495                 u8  ctrl;
 496                 u8  oui[3];
 497                 __be16 ethertype;
 498         };
 499         struct llc_snap_hdr *llc;
 500         __be16 proto;
 501
 502         proto = *(__be16 *) skb->data;
 503         __skb_pull(skb, sizeof(__be16));
 504
 505         if (ntohs(proto) >= 1536)
 506                 return proto;
 507
 508         if (skb->len < sizeof(struct llc_snap_hdr))
 509                 return htons(ETH_P_802_2);
 510
 511         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 512                 return htons(0);
 513
 514         llc = (struct llc_snap_hdr *) skb->data;
 515         if (llc->dsap != LLC_SAP_SNAP ||
 516             llc->ssap != LLC_SAP_SNAP ||
 517             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 518                 return htons(ETH_P_802_2);
 519
 520         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 521         return llc->ethertype;
 522 }
 523
 524 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 525                         int *key_lenp, int nh_len)
 526 {
 527         struct icmp6hdr *icmp = icmp6_hdr(skb);
 528         int error = 0;
 529         int key_len;
 530
 531         /* The ICMPv6 type and code fields use the 16-bit transport port
 532          * fields, so we need to store them in 16-bit network byte order.
 533          */
 534         key->ipv6.tp.src = htons(icmp->icmp6_type);
 535         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 536         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 537
 538         if (icmp->icmp6_code == 0 &&
 539             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 540              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 541                 int icmp_len = skb->len - skb_transport_offset(skb);
 542                 struct nd_msg *nd;
 543                 int offset;
 544
 545                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 546
 547                 /* In order to process neighbor discovery options, we need the
 548                  * entire packet.
 549                  */
 550                 if (unlikely(icmp_len < sizeof(*nd)))
 551                         goto out;
 552                 if (unlikely(skb_linearize(skb))) {
 553                         error = -ENOMEM;
 554                         goto out;
 555                 }
 556
 557                 nd = (struct nd_msg *)skb_transport_header(skb);
 558                 key->ipv6.nd.target = nd->target;
 559                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 560
 561                 icmp_len -= sizeof(*nd);
 562                 offset = 0;
 563                 while (icmp_len >= 8) {
 564                         struct nd_opt_hdr *nd_opt =
 565                                  (struct nd_opt_hdr *)(nd->opt + offset);
 566                         int opt_len = nd_opt->nd_opt_len * 8;
 567
 568                         if (unlikely(!opt_len || opt_len > icmp_len))
 569                                 goto invalid;
 570
 571                         /* Store the link layer address if the appropriate
 572                          * option is provided.  It is considered an error if
 573                          * the same link layer option is specified twice.
 574                          */
 575                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 576                             && opt_len == 8) {
 577                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 578                                         goto invalid;
 579                                 memcpy(key->ipv6.nd.sll,
 580                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 581                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 582                                    && opt_len == 8) {
 583                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 584                                         goto invalid;
 585                                 memcpy(key->ipv6.nd.tll,
 586                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 587                         }
 588
 589                         icmp_len -= opt_len;
 590                         offset += opt_len;
 591                 }
 592         }
 593
 594         goto out;
 595
 596 invalid:
 597         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 598         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 599         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 600
 601 out:
 602         *key_lenp = key_len;
 603         return error;
 604 }
 605
 606 /**
 607  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 608  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 609  * Ethernet header
 610  * @in_port: port number on which @skb was received.
 611  * @key: output flow key
 612  * @key_lenp: length of output flow key
 613  *
 614  * The caller must ensure that skb->len >= ETH_HLEN.
 615  *
 616  * Returns 0 if successful, otherwise a negative errno value.
 617  *
 618  * Initializes @skb header pointers as follows:
 619  *
 620  *    - skb->mac_header: the Ethernet header.
 621  *
 622  *    - skb->network_header: just past the Ethernet header, or just past the
 623  *      VLAN header, to the first byte of the Ethernet payload.
 624  *
 625  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 626  *      on output, then just past the IP header, if one is present and
 627  *      of a correct length, otherwise the same as skb->network_header.
 628  *      For other key->dl_type values it is left untouched.
 629  */
 630 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 631                  int *key_lenp)
 632 {
 633         int error = 0;
 634         int key_len = SW_FLOW_KEY_OFFSET(eth);
 635         struct ethhdr *eth;
 636
 637         memset(key, 0, sizeof(*key));
 638
 639         key->phy.priority = skb->priority;
 640         if (OVS_CB(skb)->tun_key)
 641                 memcpy(&key->phy.tun.tun_key, OVS_CB(skb)->tun_key, sizeof(key->phy.tun.tun_key));
 642         key->phy.in_port = in_port;
 643
 644         skb_reset_mac_header(skb);
 645
 646         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 647          * header in the linear data area.
 648          */
 649         eth = eth_hdr(skb);
 650         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 651         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 652
 653         __skb_pull(skb, 2 * ETH_ALEN);
 654
 655         if (vlan_tx_tag_present(skb))
 656                 key->eth.tci = htons(vlan_get_tci(skb));
 657         else if (eth->h_proto == htons(ETH_P_8021Q))
 658                 if (unlikely(parse_vlan(skb, key)))
 659                         return -ENOMEM;
 660
 661         key->eth.type = parse_ethertype(skb);
 662         if (unlikely(key->eth.type == htons(0)))
 663                 return -ENOMEM;
 664
 665         skb_reset_network_header(skb);
 666         __skb_push(skb, skb->data - skb_mac_header(skb));
 667
 668         /* Network layer. */
 669         if (key->eth.type == htons(ETH_P_IP)) {
 670                 struct iphdr *nh;
 671                 __be16 offset;
 672
 673                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 674
 675                 error = check_iphdr(skb);
 676                 if (unlikely(error)) {
 677                         if (error == -EINVAL) {
 678                                 skb->transport_header = skb->network_header;
 679                                 error = 0;
 680                         }
 681                         goto out;
 682                 }
 683
 684                 nh = ip_hdr(skb);
 685                 key->ipv4.addr.src = nh->saddr;
 686                 key->ipv4.addr.dst = nh->daddr;
 687
 688                 key->ip.proto = nh->protocol;
 689                 key->ip.tos = nh->tos;
 690                 key->ip.ttl = nh->ttl;
 691
 692                 offset = nh->frag_off & htons(IP_OFFSET);
 693                 if (offset) {
 694                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 695                         goto out;
 696                 }
 697                 if (nh->frag_off & htons(IP_MF) ||
 698                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 699                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 700
 701                 /* Transport layer. */
 702                 if (key->ip.proto == IPPROTO_TCP) {
 703                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 704                         if (tcphdr_ok(skb)) {
 705                                 struct tcphdr *tcp = tcp_hdr(skb);
 706                                 key->ipv4.tp.src = tcp->source;
 707                                 key->ipv4.tp.dst = tcp->dest;
 708                         }
 709                 } else if (key->ip.proto == IPPROTO_UDP) {
 710                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 711                         if (udphdr_ok(skb)) {
 712                                 struct udphdr *udp = udp_hdr(skb);
 713                                 key->ipv4.tp.src = udp->source;
 714                                 key->ipv4.tp.dst = udp->dest;
 715                         }
 716                 } else if (key->ip.proto == IPPROTO_ICMP) {
 717                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 718                         if (icmphdr_ok(skb)) {
 719                                 struct icmphdr *icmp = icmp_hdr(skb);
 720                                 /* The ICMP type and code fields use the 16-bit
 721                                  * transport port fields, so we need to store
 722                                  * them in 16-bit network byte order. */
 723                                 key->ipv4.tp.src = htons(icmp->type);
 724                                 key->ipv4.tp.dst = htons(icmp->code);
 725                         }
 726                 }
 727
 728         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 729                 struct arp_eth_header *arp;
 730
 731                 arp = (struct arp_eth_header *)skb_network_header(skb);
 732
 733                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 734                                 && arp->ar_pro == htons(ETH_P_IP)
 735                                 && arp->ar_hln == ETH_ALEN
 736                                 && arp->ar_pln == 4) {
 737
 738                         /* We only match on the lower 8 bits of the opcode. */
 739                         if (ntohs(arp->ar_op) <= 0xff)
 740                                 key->ip.proto = ntohs(arp->ar_op);
 741
 742                         if (key->ip.proto == ARPOP_REQUEST
 743                                         || key->ip.proto == ARPOP_REPLY) {
 744                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 745                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 746                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 747                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 748                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 749                         }
 750                 }
 751         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 752                 int nh_len;             /* IPv6 Header + Extensions */
 753
 754                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 755                 if (unlikely(nh_len < 0)) {
 756                         if (nh_len == -EINVAL)
 757                                 skb->transport_header = skb->network_header;
 758                         else
 759                                 error = nh_len;
 760                         goto out;
 761                 }
 762
 763                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 764                         goto out;
 765                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 766                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 767
 768                 /* Transport layer. */
 769                 if (key->ip.proto == NEXTHDR_TCP) {
 770                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 771                         if (tcphdr_ok(skb)) {
 772                                 struct tcphdr *tcp = tcp_hdr(skb);
 773                                 key->ipv6.tp.src = tcp->source;
 774                                 key->ipv6.tp.dst = tcp->dest;
 775                         }
 776                 } else if (key->ip.proto == NEXTHDR_UDP) {
 777                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 778                         if (udphdr_ok(skb)) {
 779                                 struct udphdr *udp = udp_hdr(skb);
 780                                 key->ipv6.tp.src = udp->source;
 781                                 key->ipv6.tp.dst = udp->dest;
 782                         }
 783                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 784                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 785                         if (icmp6hdr_ok(skb)) {
 786                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 787                                 if (error < 0)
 788                                         goto out;
 789                         }
 790                 }
 791         }
 792
 793 out:
 794         *key_lenp = key_len;
 795         return error;
 796 }
 797
 798 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 799 {
 800         return jhash2((u32 *)((u8 *)key + key_start),
 801                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 802 }
 803
 804 static int flow_key_start(struct sw_flow_key *key)
 805 {
 806         if (key->phy.tun.tun_key.ipv4_dst)
 807                 return 0;
 808         else
 809                 return offsetof(struct sw_flow_key, phy.priority);
 810 }
 811
 812 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 813                                 struct sw_flow_key *key, int key_len)
 814 {
 815         struct sw_flow *flow;
 816         struct hlist_node *n;
 817         struct hlist_head *head;
 818         u8 *_key;
 819         int key_start;
 820         u32 hash;
 821
 822         key_start = flow_key_start(key);
 823         hash = ovs_flow_hash(key, key_start, key_len);
 824
 825         _key = (u8 *) key + key_start;
 826         head = find_bucket(table, hash);
 827         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 828
 829                 if (flow->hash == hash &&
 830                     !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
 831                         return flow;
 832                 }
 833         }
 834         return NULL;
 835 }
 836
 837 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 838                          struct sw_flow_key *key, int key_len)
 839 {
 840         flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
 841         memcpy(&flow->key, key, sizeof(flow->key));
 842         __flow_tbl_insert(table, flow);
 843 }
 844
 845 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 846 {
 847         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 848         table->count--;
 849         BUG_ON(table->count < 0);
 850 }
 851
 852 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 853 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 854         [OVS_KEY_ATTR_ENCAP] = -1,
 855         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 856         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 857         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 858         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 859         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 860         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 861         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 862         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 863         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 864         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 865         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 866         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 867         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 868
 869         /* Not upstream. */
 870         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 871         [OVS_KEY_ATTR_IPV4_TUNNEL] = sizeof(struct ovs_key_ipv4_tunnel),
 872 };
 873
 874 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 875                                   const struct nlattr *a[], u64 *attrs)
 876 {
 877         const struct ovs_key_icmp *icmp_key;
 878         const struct ovs_key_tcp *tcp_key;
 879         const struct ovs_key_udp *udp_key;
 880
 881         switch (swkey->ip.proto) {
 882         case IPPROTO_TCP:
 883                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 884                         return -EINVAL;
 885                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 886
 887                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 888                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 889                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 890                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 891                 break;
 892
 893         case IPPROTO_UDP:
 894                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 895                         return -EINVAL;
 896                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 897
 898                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 899                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 900                 swkey->ipv4.tp.src = udp_key->udp_src;
 901                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 902                 break;
 903
 904         case IPPROTO_ICMP:
 905                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 906                         return -EINVAL;
 907                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 908
 909                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 910                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 911                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 912                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 913                 break;
 914         }
 915
 916         return 0;
 917 }
 918
 919 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 920                                   const struct nlattr *a[], u64 *attrs)
 921 {
 922         const struct ovs_key_icmpv6 *icmpv6_key;
 923         const struct ovs_key_tcp *tcp_key;
 924         const struct ovs_key_udp *udp_key;
 925
 926         switch (swkey->ip.proto) {
 927         case IPPROTO_TCP:
 928                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 929                         return -EINVAL;
 930                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 931
 932                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 933                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 934                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 935                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 936                 break;
 937
 938         case IPPROTO_UDP:
 939                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 940                         return -EINVAL;
 941                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 942
 943                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 944                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 945                 swkey->ipv6.tp.src = udp_key->udp_src;
 946                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 947                 break;
 948
 949         case IPPROTO_ICMPV6:
 950                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 951                         return -EINVAL;
 952                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 953
 954                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 955                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 956                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 957                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 958
 959                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 960                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 961                         const struct ovs_key_nd *nd_key;
 962
 963                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 964                                 return -EINVAL;
 965                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 966
 967                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 968                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 969                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 970                                sizeof(swkey->ipv6.nd.target));
 971                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 972                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 973                 }
 974                 break;
 975         }
 976
 977         return 0;
 978 }
 979
 980 static int parse_flow_nlattrs(const struct nlattr *attr,
 981                               const struct nlattr *a[], u64 *attrsp)
 982 {
 983         const struct nlattr *nla;
 984         u64 attrs;
 985         int rem;
 986
 987         attrs = 0;
 988         nla_for_each_nested(nla, attr, rem) {
 989                 u16 type = nla_type(nla);
 990                 int expected_len;
 991
 992                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 993                         return -EINVAL;
 994
 995                 expected_len = ovs_key_lens[type];
 996                 if (nla_len(nla) != expected_len && expected_len != -1)
 997                         return -EINVAL;
 998
 999                 attrs |= 1ULL << type;
1000                 a[type] = nla;
1001         }
1002         if (rem)
1003                 return -EINVAL;
1004
1005         *attrsp = attrs;
1006         return 0;
1007 }
1008
1009 /**
1010  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
1011  * @swkey: receives the extracted flow key.
1012  * @key_lenp: number of bytes used in @swkey.
1013  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1014  * sequence.
1015  */
1016 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1017                       const struct nlattr *attr)
1018 {
1019         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1020         const struct ovs_key_ethernet *eth_key;
1021         int key_len;
1022         u64 attrs;
1023         int err;
1024
1025         memset(swkey, 0, sizeof(struct sw_flow_key));
1026         key_len = SW_FLOW_KEY_OFFSET(eth);
1027
1028         err = parse_flow_nlattrs(attr, a, &attrs);
1029         if (err)
1030                 return err;
1031
1032         /* Metadata attributes. */
1033         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1034                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1035                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1036         }
1037         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1038                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1039                 if (in_port >= DP_MAX_PORTS)
1040                         return -EINVAL;
1041                 swkey->phy.in_port = in_port;
1042                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1043         } else {
1044                 swkey->phy.in_port = DP_MAX_PORTS;
1045         }
1046
1047         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID) &&
1048             attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1049                 struct ovs_key_ipv4_tunnel *tun_key;
1050                 __be64 tun_id;
1051
1052                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1053
1054                 if (!tun_key->ipv4_dst)
1055                         return -EINVAL;
1056                 if (!(tun_key->tun_flags & OVS_FLOW_TNL_F_KEY))
1057                         return -EINVAL;
1058
1059                 tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1060                 if (tun_id != tun_key->tun_id)
1061                         return -EINVAL;
1062
1063                 memcpy(&swkey->phy.tun.tun_key, tun_key, sizeof(swkey->phy.tun.tun_key));
1064                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1065                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1066         } else if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
1067                 swkey->phy.tun.tun_key.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1068                 swkey->phy.tun.tun_key.tun_flags |= OVS_FLOW_TNL_F_KEY;
1069
1070                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1071         } else if (attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1072                 struct ovs_key_ipv4_tunnel *tun_key;
1073                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1074
1075                 if (!tun_key->ipv4_dst)
1076                         return -EINVAL;
1077
1078                 memcpy(&swkey->phy.tun.tun_key, tun_key, sizeof(swkey->phy.tun.tun_key));
1079                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1080         }
1081
1082         /* Data attributes. */
1083         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1084                 return -EINVAL;
1085         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1086
1087         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1088         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1089         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1090
1091         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1092             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1093                 const struct nlattr *encap;
1094                 __be16 tci;
1095
1096                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1097                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1098                               (1 << OVS_KEY_ATTR_ENCAP)))
1099                         return -EINVAL;
1100
1101                 encap = a[OVS_KEY_ATTR_ENCAP];
1102                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1103                 if (tci & htons(VLAN_TAG_PRESENT)) {
1104                         swkey->eth.tci = tci;
1105
1106                         err = parse_flow_nlattrs(encap, a, &attrs);
1107                         if (err)
1108                                 return err;
1109                 } else if (!tci) {
1110                         /* Corner case for truncated 802.1Q header. */
1111                         if (nla_len(encap))
1112                                 return -EINVAL;
1113
1114                         swkey->eth.type = htons(ETH_P_8021Q);
1115                         *key_lenp = key_len;
1116                         return 0;
1117                 } else {
1118                         return -EINVAL;
1119                 }
1120         }
1121
1122         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1123                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1124                 if (ntohs(swkey->eth.type) < 1536)
1125                         return -EINVAL;
1126                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1127         } else {
1128                 swkey->eth.type = htons(ETH_P_802_2);
1129         }
1130
1131         if (swkey->eth.type == htons(ETH_P_IP)) {
1132                 const struct ovs_key_ipv4 *ipv4_key;
1133
1134                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1135                         return -EINVAL;
1136                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1137
1138                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1139                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1140                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1141                         return -EINVAL;
1142                 swkey->ip.proto = ipv4_key->ipv4_proto;
1143                 swkey->ip.tos = ipv4_key->ipv4_tos;
1144                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1145                 swkey->ip.frag = ipv4_key->ipv4_frag;
1146                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1147                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1148
1149                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1150                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1151                         if (err)
1152                                 return err;
1153                 }
1154         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1155                 const struct ovs_key_ipv6 *ipv6_key;
1156
1157                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1158                         return -EINVAL;
1159                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1160
1161                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1162                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1163                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1164                         return -EINVAL;
1165                 swkey->ipv6.label = ipv6_key->ipv6_label;
1166                 swkey->ip.proto = ipv6_key->ipv6_proto;
1167                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1168                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1169                 swkey->ip.frag = ipv6_key->ipv6_frag;
1170                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1171                        sizeof(swkey->ipv6.addr.src));
1172                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1173                        sizeof(swkey->ipv6.addr.dst));
1174
1175                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1176                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1177                         if (err)
1178                                 return err;
1179                 }
1180         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1181                 const struct ovs_key_arp *arp_key;
1182
1183                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1184                         return -EINVAL;
1185                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1186
1187                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1188                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1189                 swkey->ipv4.addr.src = arp_key->arp_sip;
1190                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1191                 if (arp_key->arp_op & htons(0xff00))
1192                         return -EINVAL;
1193                 swkey->ip.proto = ntohs(arp_key->arp_op);
1194                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1195                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1196         }
1197
1198         if (attrs)
1199                 return -EINVAL;
1200         *key_lenp = key_len;
1201
1202         return 0;
1203 }
1204
1205 /**
1206  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1207  * @in_port: receives the extracted input port.
1208  * @tun_id: receives the extracted tunnel ID.
1209  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1210  * sequence.
1211  *
1212  * This parses a series of Netlink attributes that form a flow key, which must
1213  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1214  * get the metadata, that is, the parts of the flow key that cannot be
1215  * extracted from the packet itself.
1216  */
1217
1218 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
1219 {
1220         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.phy.tun.tun_key;
1221         const struct nlattr *nla;
1222         int rem;
1223         __be64 tun_id = 0;
1224
1225         flow->key.phy.in_port = DP_MAX_PORTS;
1226         flow->key.phy.priority = 0;
1227         memset(tun_key, 0, sizeof(flow->key.phy.tun.tun_key));
1228
1229         nla_for_each_nested(nla, attr, rem) {
1230                 int type = nla_type(nla);
1231
1232                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1233                         if (nla_len(nla) != ovs_key_lens[type])
1234                                 return -EINVAL;
1235
1236                         switch (type) {
1237                         case OVS_KEY_ATTR_PRIORITY:
1238                                 flow->key.phy.priority = nla_get_u32(nla);
1239                                 break;
1240
1241                         case OVS_KEY_ATTR_TUN_ID:
1242                                 tun_id = nla_get_be64(nla);
1243
1244                                 if (tun_key->ipv4_dst) {
1245                                         if (!(tun_key->tun_flags & OVS_FLOW_TNL_F_KEY))
1246                                                 return -EINVAL;
1247                                         if (tun_key->tun_id != tun_id)
1248                                                 return -EINVAL;
1249                                         break;
1250                                 }
1251                                 tun_key->tun_id = tun_id;
1252                                 tun_key->tun_flags |= OVS_FLOW_TNL_F_KEY;
1253
1254                                 break;
1255
1256                         case OVS_KEY_ATTR_IPV4_TUNNEL:
1257                                 if (tun_key->tun_flags & OVS_FLOW_TNL_F_KEY) {
1258                                         tun_id = tun_key->tun_id;
1259
1260                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1261                                         if (!(tun_key->tun_flags & OVS_FLOW_TNL_F_KEY))
1262                                                 return -EINVAL;
1263
1264                                         if (tun_key->tun_id != tun_id)
1265                                                 return -EINVAL;
1266                                 } else
1267                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1268
1269                                 if (!tun_key->ipv4_dst)
1270                                         return -EINVAL;
1271                                 break;
1272
1273                         case OVS_KEY_ATTR_IN_PORT:
1274                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1275                                         return -EINVAL;
1276                                 flow->key.phy.in_port = nla_get_u32(nla);
1277                                 break;
1278                         }
1279                 }
1280         }
1281         if (rem)
1282                 return -EINVAL;
1283
1284         flow->hash = ovs_flow_hash(&flow->key,
1285                                    flow_key_start(&flow->key), key_len);
1286
1287         return 0;
1288 }
1289
1290 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1291 {
1292         struct ovs_key_ethernet *eth_key;
1293         struct nlattr *nla, *encap;
1294
1295         if (swkey->phy.priority &&
1296             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1297                 goto nla_put_failure;
1298
1299         if (swkey->phy.tun.tun_key.ipv4_dst) {
1300                 struct ovs_key_ipv4_tunnel *tun_key;
1301                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4_TUNNEL, sizeof(*tun_key));
1302                 if (!nla)
1303                         goto nla_put_failure;
1304                 tun_key = nla_data(nla);
1305                 memcpy(tun_key, &swkey->phy.tun.tun_key, sizeof(*tun_key));
1306         }
1307         if ((swkey->phy.tun.tun_key.tun_flags & OVS_FLOW_TNL_F_KEY) &&
1308             nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun.tun_key.tun_id))
1309                 goto nla_put_failure;
1310
1311         if (swkey->phy.in_port != DP_MAX_PORTS &&
1312             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1313                 goto nla_put_failure;
1314
1315         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1316         if (!nla)
1317                 goto nla_put_failure;
1318         eth_key = nla_data(nla);
1319         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1320         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1321
1322         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1323                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1324                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1325                         goto nla_put_failure;
1326                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1327                 if (!swkey->eth.tci)
1328                         goto unencap;
1329         } else {
1330                 encap = NULL;
1331         }
1332
1333         if (swkey->eth.type == htons(ETH_P_802_2))
1334                 goto unencap;
1335
1336         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1337                 goto nla_put_failure;
1338
1339         if (swkey->eth.type == htons(ETH_P_IP)) {
1340                 struct ovs_key_ipv4 *ipv4_key;
1341
1342                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1343                 if (!nla)
1344                         goto nla_put_failure;
1345                 ipv4_key = nla_data(nla);
1346                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1347                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1348                 ipv4_key->ipv4_proto = swkey->ip.proto;
1349                 ipv4_key->ipv4_tos = swkey->ip.tos;
1350                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1351                 ipv4_key->ipv4_frag = swkey->ip.frag;
1352         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1353                 struct ovs_key_ipv6 *ipv6_key;
1354
1355                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1356                 if (!nla)
1357                         goto nla_put_failure;
1358                 ipv6_key = nla_data(nla);
1359                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1360                                 sizeof(ipv6_key->ipv6_src));
1361                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1362                                 sizeof(ipv6_key->ipv6_dst));
1363                 ipv6_key->ipv6_label = swkey->ipv6.label;
1364                 ipv6_key->ipv6_proto = swkey->ip.proto;
1365                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1366                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1367                 ipv6_key->ipv6_frag = swkey->ip.frag;
1368         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1369                 struct ovs_key_arp *arp_key;
1370
1371                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1372                 if (!nla)
1373                         goto nla_put_failure;
1374                 arp_key = nla_data(nla);
1375                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1376                 arp_key->arp_sip = swkey->ipv4.addr.src;
1377                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1378                 arp_key->arp_op = htons(swkey->ip.proto);
1379                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1380                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1381         }
1382
1383         if ((swkey->eth.type == htons(ETH_P_IP) ||
1384              swkey->eth.type == htons(ETH_P_IPV6)) &&
1385              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1386
1387                 if (swkey->ip.proto == IPPROTO_TCP) {
1388                         struct ovs_key_tcp *tcp_key;
1389
1390                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1391                         if (!nla)
1392                                 goto nla_put_failure;
1393                         tcp_key = nla_data(nla);
1394                         if (swkey->eth.type == htons(ETH_P_IP)) {
1395                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1396                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1397                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1398                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1399                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1400                         }
1401                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1402                         struct ovs_key_udp *udp_key;
1403
1404                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1405                         if (!nla)
1406                                 goto nla_put_failure;
1407                         udp_key = nla_data(nla);
1408                         if (swkey->eth.type == htons(ETH_P_IP)) {
1409                                 udp_key->udp_src = swkey->ipv4.tp.src;
1410                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1411                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1412                                 udp_key->udp_src = swkey->ipv6.tp.src;
1413                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1414                         }
1415                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1416                            swkey->ip.proto == IPPROTO_ICMP) {
1417                         struct ovs_key_icmp *icmp_key;
1418
1419                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1420                         if (!nla)
1421                                 goto nla_put_failure;
1422                         icmp_key = nla_data(nla);
1423                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1424                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1425                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1426                            swkey->ip.proto == IPPROTO_ICMPV6) {
1427                         struct ovs_key_icmpv6 *icmpv6_key;
1428
1429                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1430                                                 sizeof(*icmpv6_key));
1431                         if (!nla)
1432                                 goto nla_put_failure;
1433                         icmpv6_key = nla_data(nla);
1434                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1435                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1436
1437                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1438                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1439                                 struct ovs_key_nd *nd_key;
1440
1441                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1442                                 if (!nla)
1443                                         goto nla_put_failure;
1444                                 nd_key = nla_data(nla);
1445                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1446                                                         sizeof(nd_key->nd_target));
1447                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1448                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1449                         }
1450                 }
1451         }
1452
1453 unencap:
1454         if (encap)
1455                 nla_nest_end(skb, encap);
1456
1457         return 0;
1458
1459 nla_put_failure:
1460         return -EMSGSIZE;
1461 }
1462
1463 /* Initializes the flow module.
1464  * Returns zero if successful or a negative error code. */
1465 int ovs_flow_init(void)
1466 {
1467         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1468                                         0, NULL);
1469         if (flow_cache == NULL)
1470                 return -ENOMEM;
1471
1472         return 0;
1473 }
1474
1475 /* Uninitializes the flow module. */
1476 void ovs_flow_exit(void)
1477 {
1478         kmem_cache_destroy(flow_cache);
1479 }