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(int size)
 204 {
 205         struct sw_flow_actions *sfa;
 206
 207         if (size > MAX_ACTIONS_BUFSIZE)
 208                 return ERR_PTR(-EINVAL);
 209
 210         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 211         if (!sfa)
 212                 return ERR_PTR(-ENOMEM);
 213
 214         sfa->actions_len = 0;
 215         return sfa;
 216 }
 217
 218 struct sw_flow *ovs_flow_alloc(void)
 219 {
 220         struct sw_flow *flow;
 221
 222         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 223         if (!flow)
 224                 return ERR_PTR(-ENOMEM);
 225
 226         spin_lock_init(&flow->lock);
 227         flow->sf_acts = NULL;
 228
 229         return flow;
 230 }
 231
 232 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 233 {
 234         hash = jhash_1word(hash, table->hash_seed);
 235         return flex_array_get(table->buckets,
 236                                 (hash & (table->n_buckets - 1)));
 237 }
 238
 239 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 240 {
 241         struct flex_array *buckets;
 242         int i, err;
 243
 244         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 245                                    n_buckets, GFP_KERNEL);
 246         if (!buckets)
 247                 return NULL;
 248
 249         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 250         if (err) {
 251                 flex_array_free(buckets);
 252                 return NULL;
 253         }
 254
 255         for (i = 0; i < n_buckets; i++)
 256                 INIT_HLIST_HEAD((struct hlist_head *)
 257                                         flex_array_get(buckets, i));
 258
 259         return buckets;
 260 }
 261
 262 static void free_buckets(struct flex_array *buckets)
 263 {
 264         flex_array_free(buckets);
 265 }
 266
 267 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 268 {
 269         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 270
 271         if (!table)
 272                 return NULL;
 273
 274         table->buckets = alloc_buckets(new_size);
 275
 276         if (!table->buckets) {
 277                 kfree(table);
 278                 return NULL;
 279         }
 280         table->n_buckets = new_size;
 281         table->count = 0;
 282         table->node_ver = 0;
 283         table->keep_flows = false;
 284         get_random_bytes(&table->hash_seed, sizeof(u32));
 285
 286         return table;
 287 }
 288
 289 void ovs_flow_tbl_destroy(struct flow_table *table)
 290 {
 291         int i;
 292
 293         if (!table)
 294                 return;
 295
 296         if (table->keep_flows)
 297                 goto skip_flows;
 298
 299         for (i = 0; i < table->n_buckets; i++) {
 300                 struct sw_flow *flow;
 301                 struct hlist_head *head = flex_array_get(table->buckets, i);
 302                 struct hlist_node *node, *n;
 303                 int ver = table->node_ver;
 304
 305                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 306                         hlist_del_rcu(&flow->hash_node[ver]);
 307                         ovs_flow_free(flow);
 308                 }
 309         }
 310
 311 skip_flows:
 312         free_buckets(table->buckets);
 313         kfree(table);
 314 }
 315
 316 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 317 {
 318         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 319
 320         ovs_flow_tbl_destroy(table);
 321 }
 322
 323 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 324 {
 325         if (!table)
 326                 return;
 327
 328         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 329 }
 330
 331 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 332 {
 333         struct sw_flow *flow;
 334         struct hlist_head *head;
 335         struct hlist_node *n;
 336         int ver;
 337         int i;
 338
 339         ver = table->node_ver;
 340         while (*bucket < table->n_buckets) {
 341                 i = 0;
 342                 head = flex_array_get(table->buckets, *bucket);
 343                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 344                         if (i < *last) {
 345                                 i++;
 346                                 continue;
 347                         }
 348                         *last = i + 1;
 349                         return flow;
 350                 }
 351                 (*bucket)++;
 352                 *last = 0;
 353         }
 354
 355         return NULL;
 356 }
 357
 358 static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 359 {
 360         struct hlist_head *head;
 361         head = find_bucket(table, flow->hash);
 362         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 363         table->count++;
 364 }
 365
 366 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 367 {
 368         int old_ver;
 369         int i;
 370
 371         old_ver = old->node_ver;
 372         new->node_ver = !old_ver;
 373
 374         /* Insert in new table. */
 375         for (i = 0; i < old->n_buckets; i++) {
 376                 struct sw_flow *flow;
 377                 struct hlist_head *head;
 378                 struct hlist_node *n;
 379
 380                 head = flex_array_get(old->buckets, i);
 381
 382                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 383                         __flow_tbl_insert(new, flow);
 384         }
 385         old->keep_flows = true;
 386 }
 387
 388 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 389 {
 390         struct flow_table *new_table;
 391
 392         new_table = ovs_flow_tbl_alloc(n_buckets);
 393         if (!new_table)
 394                 return ERR_PTR(-ENOMEM);
 395
 396         flow_table_copy_flows(table, new_table);
 397
 398         return new_table;
 399 }
 400
 401 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 402 {
 403         return __flow_tbl_rehash(table, table->n_buckets);
 404 }
 405
 406 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 407 {
 408         return __flow_tbl_rehash(table, table->n_buckets * 2);
 409 }
 410
 411 void ovs_flow_free(struct sw_flow *flow)
 412 {
 413         if (unlikely(!flow))
 414                 return;
 415
 416         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 417         kmem_cache_free(flow_cache, flow);
 418 }
 419
 420 /* RCU callback used by ovs_flow_deferred_free. */
 421 static void rcu_free_flow_callback(struct rcu_head *rcu)
 422 {
 423         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 424
 425         ovs_flow_free(flow);
 426 }
 427
 428 /* Schedules 'flow' to be freed after the next RCU grace period.
 429  * The caller must hold rcu_read_lock for this to be sensible. */
 430 void ovs_flow_deferred_free(struct sw_flow *flow)
 431 {
 432         call_rcu(&flow->rcu, rcu_free_flow_callback);
 433 }
 434
 435 /* RCU callback used by ovs_flow_deferred_free_acts. */
 436 static void rcu_free_acts_callback(struct rcu_head *rcu)
 437 {
 438         struct sw_flow_actions *sf_acts = container_of(rcu,
 439                         struct sw_flow_actions, rcu);
 440         kfree(sf_acts);
 441 }
 442
 443 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 444  * The caller must hold rcu_read_lock for this to be sensible. */
 445 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 446 {
 447         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 448 }
 449
 450 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 451 {
 452         struct qtag_prefix {
 453                 __be16 eth_type; /* ETH_P_8021Q */
 454                 __be16 tci;
 455         };
 456         struct qtag_prefix *qp;
 457
 458         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 459                 return 0;
 460
 461         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 462                                          sizeof(__be16))))
 463                 return -ENOMEM;
 464
 465         qp = (struct qtag_prefix *) skb->data;
 466         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 467         __skb_pull(skb, sizeof(struct qtag_prefix));
 468
 469         return 0;
 470 }
 471
 472 static __be16 parse_ethertype(struct sk_buff *skb)
 473 {
 474         struct llc_snap_hdr {
 475                 u8  dsap;  /* Always 0xAA */
 476                 u8  ssap;  /* Always 0xAA */
 477                 u8  ctrl;
 478                 u8  oui[3];
 479                 __be16 ethertype;
 480         };
 481         struct llc_snap_hdr *llc;
 482         __be16 proto;
 483
 484         proto = *(__be16 *) skb->data;
 485         __skb_pull(skb, sizeof(__be16));
 486
 487         if (ntohs(proto) >= 1536)
 488                 return proto;
 489
 490         if (skb->len < sizeof(struct llc_snap_hdr))
 491                 return htons(ETH_P_802_2);
 492
 493         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 494                 return htons(0);
 495
 496         llc = (struct llc_snap_hdr *) skb->data;
 497         if (llc->dsap != LLC_SAP_SNAP ||
 498             llc->ssap != LLC_SAP_SNAP ||
 499             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 500                 return htons(ETH_P_802_2);
 501
 502         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 503         return llc->ethertype;
 504 }
 505
 506 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 507                         int *key_lenp, int nh_len)
 508 {
 509         struct icmp6hdr *icmp = icmp6_hdr(skb);
 510         int error = 0;
 511         int key_len;
 512
 513         /* The ICMPv6 type and code fields use the 16-bit transport port
 514          * fields, so we need to store them in 16-bit network byte order.
 515          */
 516         key->ipv6.tp.src = htons(icmp->icmp6_type);
 517         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 518         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 519
 520         if (icmp->icmp6_code == 0 &&
 521             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 522              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 523                 int icmp_len = skb->len - skb_transport_offset(skb);
 524                 struct nd_msg *nd;
 525                 int offset;
 526
 527                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 528
 529                 /* In order to process neighbor discovery options, we need the
 530                  * entire packet.
 531                  */
 532                 if (unlikely(icmp_len < sizeof(*nd)))
 533                         goto out;
 534                 if (unlikely(skb_linearize(skb))) {
 535                         error = -ENOMEM;
 536                         goto out;
 537                 }
 538
 539                 nd = (struct nd_msg *)skb_transport_header(skb);
 540                 key->ipv6.nd.target = nd->target;
 541                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 542
 543                 icmp_len -= sizeof(*nd);
 544                 offset = 0;
 545                 while (icmp_len >= 8) {
 546                         struct nd_opt_hdr *nd_opt =
 547                                  (struct nd_opt_hdr *)(nd->opt + offset);
 548                         int opt_len = nd_opt->nd_opt_len * 8;
 549
 550                         if (unlikely(!opt_len || opt_len > icmp_len))
 551                                 goto invalid;
 552
 553                         /* Store the link layer address if the appropriate
 554                          * option is provided.  It is considered an error if
 555                          * the same link layer option is specified twice.
 556                          */
 557                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 558                             && opt_len == 8) {
 559                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 560                                         goto invalid;
 561                                 memcpy(key->ipv6.nd.sll,
 562                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 563                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 564                                    && opt_len == 8) {
 565                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 566                                         goto invalid;
 567                                 memcpy(key->ipv6.nd.tll,
 568                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 569                         }
 570
 571                         icmp_len -= opt_len;
 572                         offset += opt_len;
 573                 }
 574         }
 575
 576         goto out;
 577
 578 invalid:
 579         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 580         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 581         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 582
 583 out:
 584         *key_lenp = key_len;
 585         return error;
 586 }
 587
 588 /**
 589  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 590  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 591  * Ethernet header
 592  * @in_port: port number on which @skb was received.
 593  * @key: output flow key
 594  * @key_lenp: length of output flow key
 595  *
 596  * The caller must ensure that skb->len >= ETH_HLEN.
 597  *
 598  * Returns 0 if successful, otherwise a negative errno value.
 599  *
 600  * Initializes @skb header pointers as follows:
 601  *
 602  *    - skb->mac_header: the Ethernet header.
 603  *
 604  *    - skb->network_header: just past the Ethernet header, or just past the
 605  *      VLAN header, to the first byte of the Ethernet payload.
 606  *
 607  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 608  *      on output, then just past the IP header, if one is present and
 609  *      of a correct length, otherwise the same as skb->network_header.
 610  *      For other key->dl_type values it is left untouched.
 611  */
 612 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 613                  int *key_lenp)
 614 {
 615         int error = 0;
 616         int key_len = SW_FLOW_KEY_OFFSET(eth);
 617         struct ethhdr *eth;
 618
 619         memset(key, 0, sizeof(*key));
 620
 621         key->phy.priority = skb->priority;
 622         if (OVS_CB(skb)->tun_key)
 623                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 624         key->phy.in_port = in_port;
 625         key->phy.skb_mark = skb_get_mark(skb);
 626
 627         skb_reset_mac_header(skb);
 628
 629         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 630          * header in the linear data area.
 631          */
 632         eth = eth_hdr(skb);
 633         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 634         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 635
 636         __skb_pull(skb, 2 * ETH_ALEN);
 637
 638         if (vlan_tx_tag_present(skb))
 639                 key->eth.tci = htons(vlan_get_tci(skb));
 640         else if (eth->h_proto == htons(ETH_P_8021Q))
 641                 if (unlikely(parse_vlan(skb, key)))
 642                         return -ENOMEM;
 643
 644         key->eth.type = parse_ethertype(skb);
 645         if (unlikely(key->eth.type == htons(0)))
 646                 return -ENOMEM;
 647
 648         skb_reset_network_header(skb);
 649         __skb_push(skb, skb->data - skb_mac_header(skb));
 650
 651         /* Network layer. */
 652         if (key->eth.type == htons(ETH_P_IP)) {
 653                 struct iphdr *nh;
 654                 __be16 offset;
 655
 656                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 657
 658                 error = check_iphdr(skb);
 659                 if (unlikely(error)) {
 660                         if (error == -EINVAL) {
 661                                 skb->transport_header = skb->network_header;
 662                                 error = 0;
 663                         }
 664                         goto out;
 665                 }
 666
 667                 nh = ip_hdr(skb);
 668                 key->ipv4.addr.src = nh->saddr;
 669                 key->ipv4.addr.dst = nh->daddr;
 670
 671                 key->ip.proto = nh->protocol;
 672                 key->ip.tos = nh->tos;
 673                 key->ip.ttl = nh->ttl;
 674
 675                 offset = nh->frag_off & htons(IP_OFFSET);
 676                 if (offset) {
 677                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 678                         goto out;
 679                 }
 680                 if (nh->frag_off & htons(IP_MF) ||
 681                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 682                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 683
 684                 /* Transport layer. */
 685                 if (key->ip.proto == IPPROTO_TCP) {
 686                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 687                         if (tcphdr_ok(skb)) {
 688                                 struct tcphdr *tcp = tcp_hdr(skb);
 689                                 key->ipv4.tp.src = tcp->source;
 690                                 key->ipv4.tp.dst = tcp->dest;
 691                         }
 692                 } else if (key->ip.proto == IPPROTO_UDP) {
 693                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 694                         if (udphdr_ok(skb)) {
 695                                 struct udphdr *udp = udp_hdr(skb);
 696                                 key->ipv4.tp.src = udp->source;
 697                                 key->ipv4.tp.dst = udp->dest;
 698                         }
 699                 } else if (key->ip.proto == IPPROTO_ICMP) {
 700                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 701                         if (icmphdr_ok(skb)) {
 702                                 struct icmphdr *icmp = icmp_hdr(skb);
 703                                 /* The ICMP type and code fields use the 16-bit
 704                                  * transport port fields, so we need to store
 705                                  * them in 16-bit network byte order. */
 706                                 key->ipv4.tp.src = htons(icmp->type);
 707                                 key->ipv4.tp.dst = htons(icmp->code);
 708                         }
 709                 }
 710
 711         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 712                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 713                 struct arp_eth_header *arp;
 714
 715                 arp = (struct arp_eth_header *)skb_network_header(skb);
 716
 717                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 718                                 && arp->ar_pro == htons(ETH_P_IP)
 719                                 && arp->ar_hln == ETH_ALEN
 720                                 && arp->ar_pln == 4) {
 721
 722                         /* We only match on the lower 8 bits of the opcode. */
 723                         if (ntohs(arp->ar_op) <= 0xff)
 724                                 key->ip.proto = ntohs(arp->ar_op);
 725                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 726                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 727                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 728                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 729                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 730                 }
 731         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 732                 int nh_len;             /* IPv6 Header + Extensions */
 733
 734                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 735                 if (unlikely(nh_len < 0)) {
 736                         if (nh_len == -EINVAL)
 737                                 skb->transport_header = skb->network_header;
 738                         else
 739                                 error = nh_len;
 740                         goto out;
 741                 }
 742
 743                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 744                         goto out;
 745                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 746                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 747
 748                 /* Transport layer. */
 749                 if (key->ip.proto == NEXTHDR_TCP) {
 750                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 751                         if (tcphdr_ok(skb)) {
 752                                 struct tcphdr *tcp = tcp_hdr(skb);
 753                                 key->ipv6.tp.src = tcp->source;
 754                                 key->ipv6.tp.dst = tcp->dest;
 755                         }
 756                 } else if (key->ip.proto == NEXTHDR_UDP) {
 757                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 758                         if (udphdr_ok(skb)) {
 759                                 struct udphdr *udp = udp_hdr(skb);
 760                                 key->ipv6.tp.src = udp->source;
 761                                 key->ipv6.tp.dst = udp->dest;
 762                         }
 763                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 764                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 765                         if (icmp6hdr_ok(skb)) {
 766                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 767                                 if (error < 0)
 768                                         goto out;
 769                         }
 770                 }
 771         }
 772
 773 out:
 774         *key_lenp = key_len;
 775         return error;
 776 }
 777
 778 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 779 {
 780         return jhash2((u32 *)((u8 *)key + key_start),
 781                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 782 }
 783
 784 static int flow_key_start(struct sw_flow_key *key)
 785 {
 786         if (key->tun_key.ipv4_dst)
 787                 return 0;
 788         else
 789                 return offsetof(struct sw_flow_key, phy);
 790 }
 791
 792 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 793                                 struct sw_flow_key *key, int key_len)
 794 {
 795         struct sw_flow *flow;
 796         struct hlist_node *n;
 797         struct hlist_head *head;
 798         u8 *_key;
 799         int key_start;
 800         u32 hash;
 801
 802         key_start = flow_key_start(key);
 803         hash = ovs_flow_hash(key, key_start, key_len);
 804
 805         _key = (u8 *) key + key_start;
 806         head = find_bucket(table, hash);
 807         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 808
 809                 if (flow->hash == hash &&
 810                     !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
 811                         return flow;
 812                 }
 813         }
 814         return NULL;
 815 }
 816
 817 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 818                          struct sw_flow_key *key, int key_len)
 819 {
 820         flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
 821         memcpy(&flow->key, key, sizeof(flow->key));
 822         __flow_tbl_insert(table, flow);
 823 }
 824
 825 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 826 {
 827         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 828         table->count--;
 829         BUG_ON(table->count < 0);
 830 }
 831
 832 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 833 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 834         [OVS_KEY_ATTR_ENCAP] = -1,
 835         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 836         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 837         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 838         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 839         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 840         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 841         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 842         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 843         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 844         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 845         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 846         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 847         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 848         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 849         [OVS_KEY_ATTR_TUNNEL] = -1,
 850
 851         /* Not upstream. */
 852         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 853 };
 854
 855 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 856                                   const struct nlattr *a[], u64 *attrs)
 857 {
 858         const struct ovs_key_icmp *icmp_key;
 859         const struct ovs_key_tcp *tcp_key;
 860         const struct ovs_key_udp *udp_key;
 861
 862         switch (swkey->ip.proto) {
 863         case IPPROTO_TCP:
 864                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 865                         return -EINVAL;
 866                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 867
 868                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 869                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 870                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 871                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 872                 break;
 873
 874         case IPPROTO_UDP:
 875                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 876                         return -EINVAL;
 877                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 878
 879                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 880                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 881                 swkey->ipv4.tp.src = udp_key->udp_src;
 882                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 883                 break;
 884
 885         case IPPROTO_ICMP:
 886                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 887                         return -EINVAL;
 888                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 889
 890                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 891                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 892                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 893                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 894                 break;
 895         }
 896
 897         return 0;
 898 }
 899
 900 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 901                                   const struct nlattr *a[], u64 *attrs)
 902 {
 903         const struct ovs_key_icmpv6 *icmpv6_key;
 904         const struct ovs_key_tcp *tcp_key;
 905         const struct ovs_key_udp *udp_key;
 906
 907         switch (swkey->ip.proto) {
 908         case IPPROTO_TCP:
 909                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 910                         return -EINVAL;
 911                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 912
 913                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 914                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 915                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 916                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 917                 break;
 918
 919         case IPPROTO_UDP:
 920                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 921                         return -EINVAL;
 922                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 923
 924                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 925                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 926                 swkey->ipv6.tp.src = udp_key->udp_src;
 927                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 928                 break;
 929
 930         case IPPROTO_ICMPV6:
 931                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 932                         return -EINVAL;
 933                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 934
 935                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 936                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 937                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 938                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 939
 940                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 941                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 942                         const struct ovs_key_nd *nd_key;
 943
 944                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 945                                 return -EINVAL;
 946                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 947
 948                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 949                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 950                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 951                                sizeof(swkey->ipv6.nd.target));
 952                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 953                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 954                 }
 955                 break;
 956         }
 957
 958         return 0;
 959 }
 960
 961 static int parse_flow_nlattrs(const struct nlattr *attr,
 962                               const struct nlattr *a[], u64 *attrsp)
 963 {
 964         const struct nlattr *nla;
 965         u64 attrs;
 966         int rem;
 967
 968         attrs = 0;
 969         nla_for_each_nested(nla, attr, rem) {
 970                 u16 type = nla_type(nla);
 971                 int expected_len;
 972
 973                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 974                         return -EINVAL;
 975
 976                 expected_len = ovs_key_lens[type];
 977                 if (nla_len(nla) != expected_len && expected_len != -1)
 978                         return -EINVAL;
 979
 980                 attrs |= 1ULL << type;
 981                 a[type] = nla;
 982         }
 983         if (rem)
 984                 return -EINVAL;
 985
 986         *attrsp = attrs;
 987         return 0;
 988 }
 989
 990 int ipv4_tun_from_nlattr(const struct nlattr *attr,
 991                          struct ovs_key_ipv4_tunnel *tun_key)
 992 {
 993         struct nlattr *a;
 994         int rem;
 995         bool ttl = false;
 996
 997         memset(tun_key, 0, sizeof(*tun_key));
 998
 999         nla_for_each_nested(a, attr, rem) {
1000                 int type = nla_type(a);
1001                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1002                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1003                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1004                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1005                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1006                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1007                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1008                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1009                 };
1010
1011                 if (type > OVS_TUNNEL_KEY_ATTR_MAX ||
1012                         ovs_tunnel_key_lens[type] != nla_len(a))
1013                         return -EINVAL;
1014
1015                 switch (type) {
1016                 case OVS_TUNNEL_KEY_ATTR_ID:
1017                         tun_key->tun_id = nla_get_be64(a);
1018                         tun_key->tun_flags |= OVS_TNL_F_KEY;
1019                         break;
1020                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1021                         tun_key->ipv4_src = nla_get_be32(a);
1022                         break;
1023                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1024                         tun_key->ipv4_dst = nla_get_be32(a);
1025                         break;
1026                 case OVS_TUNNEL_KEY_ATTR_TOS:
1027                         tun_key->ipv4_tos = nla_get_u8(a);
1028                         break;
1029                 case OVS_TUNNEL_KEY_ATTR_TTL:
1030                         tun_key->ipv4_ttl = nla_get_u8(a);
1031                         ttl = true;
1032                         break;
1033                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1034                         tun_key->tun_flags |= OVS_TNL_F_DONT_FRAGMENT;
1035                         break;
1036                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1037                         tun_key->tun_flags |= OVS_TNL_F_CSUM;
1038                         break;
1039                 default:
1040                         return -EINVAL;
1041
1042                 }
1043         }
1044         if (rem > 0)
1045                 return -EINVAL;
1046
1047         if (!tun_key->ipv4_dst)
1048                 return -EINVAL;
1049
1050         if (!ttl)
1051                 return -EINVAL;
1052
1053         return 0;
1054 }
1055
1056 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1057                         const struct ovs_key_ipv4_tunnel *tun_key)
1058 {
1059         struct nlattr *nla;
1060
1061         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1062         if (!nla)
1063                 return -EMSGSIZE;
1064
1065         if (tun_key->tun_flags & OVS_TNL_F_KEY &&
1066             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id))
1067                 return -EMSGSIZE;
1068         if (tun_key->ipv4_src &&
1069             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ipv4_src))
1070                 return -EMSGSIZE;
1071         if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ipv4_dst))
1072                 return -EMSGSIZE;
1073         if (tun_key->ipv4_tos &&
1074             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ipv4_tos))
1075                 return -EMSGSIZE;
1076         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ipv4_ttl))
1077                 return -EMSGSIZE;
1078         if ((tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT) &&
1079                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1080                 return -EMSGSIZE;
1081         if ((tun_key->tun_flags & OVS_TNL_F_CSUM) &&
1082                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1083                 return -EMSGSIZE;
1084
1085         nla_nest_end(skb, nla);
1086         return 0;
1087 }
1088
1089 /**
1090  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
1091  * @swkey: receives the extracted flow key.
1092  * @key_lenp: number of bytes used in @swkey.
1093  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1094  * sequence.
1095  */
1096 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1097                       const struct nlattr *attr)
1098 {
1099         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1100         const struct ovs_key_ethernet *eth_key;
1101         int key_len;
1102         u64 attrs;
1103         int err;
1104
1105         memset(swkey, 0, sizeof(struct sw_flow_key));
1106         key_len = SW_FLOW_KEY_OFFSET(eth);
1107
1108         err = parse_flow_nlattrs(attr, a, &attrs);
1109         if (err)
1110                 return err;
1111
1112         /* Metadata attributes. */
1113         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1114                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1115                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1116         }
1117         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1118                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1119                 if (in_port >= DP_MAX_PORTS)
1120                         return -EINVAL;
1121                 swkey->phy.in_port = in_port;
1122                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1123         } else {
1124                 swkey->phy.in_port = DP_MAX_PORTS;
1125         }
1126         if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1127                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1128 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1129                 if (mark != 0)
1130                         return -EINVAL;
1131 #endif
1132                 swkey->phy.skb_mark = mark;
1133                 attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1134         }
1135
1136         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID) &&
1137             attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1138                 __be64 tun_id;
1139
1140                 err = ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1141                 if (err)
1142                         return err;
1143
1144                 if (!(swkey->tun_key.tun_flags & OVS_TNL_F_KEY))
1145                         return -EINVAL;
1146
1147                 tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1148                 if (tun_id != swkey->tun_key.tun_id)
1149                         return -EINVAL;
1150
1151                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1152                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1153         } else if (attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1154
1155                 err = ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1156                 if (err)
1157                         return err;
1158
1159                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1160         }
1161
1162         /* Data attributes. */
1163         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1164                 return -EINVAL;
1165         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1166
1167         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1168         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1169         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1170
1171         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1172             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1173                 const struct nlattr *encap;
1174                 __be16 tci;
1175
1176                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1177                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1178                               (1 << OVS_KEY_ATTR_ENCAP)))
1179                         return -EINVAL;
1180
1181                 encap = a[OVS_KEY_ATTR_ENCAP];
1182                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1183                 if (tci & htons(VLAN_TAG_PRESENT)) {
1184                         swkey->eth.tci = tci;
1185
1186                         err = parse_flow_nlattrs(encap, a, &attrs);
1187                         if (err)
1188                                 return err;
1189                 } else if (!tci) {
1190                         /* Corner case for truncated 802.1Q header. */
1191                         if (nla_len(encap))
1192                                 return -EINVAL;
1193
1194                         swkey->eth.type = htons(ETH_P_8021Q);
1195                         *key_lenp = key_len;
1196                         return 0;
1197                 } else {
1198                         return -EINVAL;
1199                 }
1200         }
1201
1202         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1203                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1204                 if (ntohs(swkey->eth.type) < 1536)
1205                         return -EINVAL;
1206                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1207         } else {
1208                 swkey->eth.type = htons(ETH_P_802_2);
1209         }
1210
1211         if (swkey->eth.type == htons(ETH_P_IP)) {
1212                 const struct ovs_key_ipv4 *ipv4_key;
1213
1214                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1215                         return -EINVAL;
1216                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1217
1218                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1219                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1220                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1221                         return -EINVAL;
1222                 swkey->ip.proto = ipv4_key->ipv4_proto;
1223                 swkey->ip.tos = ipv4_key->ipv4_tos;
1224                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1225                 swkey->ip.frag = ipv4_key->ipv4_frag;
1226                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1227                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1228
1229                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1230                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1231                         if (err)
1232                                 return err;
1233                 }
1234         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1235                 const struct ovs_key_ipv6 *ipv6_key;
1236
1237                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1238                         return -EINVAL;
1239                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1240
1241                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1242                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1243                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1244                         return -EINVAL;
1245                 swkey->ipv6.label = ipv6_key->ipv6_label;
1246                 swkey->ip.proto = ipv6_key->ipv6_proto;
1247                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1248                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1249                 swkey->ip.frag = ipv6_key->ipv6_frag;
1250                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1251                        sizeof(swkey->ipv6.addr.src));
1252                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1253                        sizeof(swkey->ipv6.addr.dst));
1254
1255                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1256                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1257                         if (err)
1258                                 return err;
1259                 }
1260         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1261                    swkey->eth.type == htons(ETH_P_RARP)) {
1262                 const struct ovs_key_arp *arp_key;
1263
1264                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1265                         return -EINVAL;
1266                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1267
1268                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1269                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1270                 swkey->ipv4.addr.src = arp_key->arp_sip;
1271                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1272                 if (arp_key->arp_op & htons(0xff00))
1273                         return -EINVAL;
1274                 swkey->ip.proto = ntohs(arp_key->arp_op);
1275                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1276                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1277         }
1278
1279         if (attrs)
1280                 return -EINVAL;
1281         *key_lenp = key_len;
1282
1283         return 0;
1284 }
1285
1286 /**
1287  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1288  * @in_port: receives the extracted input port.
1289  * @tun_id: receives the extracted tunnel ID.
1290  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1291  * sequence.
1292  *
1293  * This parses a series of Netlink attributes that form a flow key, which must
1294  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1295  * get the metadata, that is, the parts of the flow key that cannot be
1296  * extracted from the packet itself.
1297  */
1298
1299 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
1300 {
1301         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1302         const struct nlattr *nla;
1303         int rem;
1304         __be64 tun_id = 0;
1305
1306         flow->key.phy.in_port = DP_MAX_PORTS;
1307         flow->key.phy.priority = 0;
1308         flow->key.phy.skb_mark = 0;
1309         memset(tun_key, 0, sizeof(flow->key.tun_key));
1310
1311         nla_for_each_nested(nla, attr, rem) {
1312                 int type = nla_type(nla);
1313
1314                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1315                         int err;
1316
1317                         if (nla_len(nla) != ovs_key_lens[type])
1318                                 return -EINVAL;
1319
1320                         switch (type) {
1321                         case OVS_KEY_ATTR_PRIORITY:
1322                                 flow->key.phy.priority = nla_get_u32(nla);
1323                                 break;
1324
1325                         case OVS_KEY_ATTR_TUN_ID:
1326                                 tun_id = nla_get_be64(nla);
1327
1328                                 if (tun_key->ipv4_dst) {
1329                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1330                                                 return -EINVAL;
1331                                         if (tun_key->tun_id != tun_id)
1332                                                 return -EINVAL;
1333                                         break;
1334                                 }
1335                                 tun_key->tun_id = tun_id;
1336                                 tun_key->tun_flags |= OVS_TNL_F_KEY;
1337
1338                                 break;
1339
1340                         case OVS_KEY_ATTR_TUNNEL:
1341                                 if (tun_key->tun_flags & OVS_TNL_F_KEY) {
1342                                         tun_id = tun_key->tun_id;
1343                                         err = ipv4_tun_from_nlattr(nla, tun_key);
1344                                         if (err)
1345                                                 return err;
1346
1347                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1348                                                 return -EINVAL;
1349
1350                                         if (tun_key->tun_id != tun_id)
1351                                                 return -EINVAL;
1352                                 } else {
1353                                         err = ipv4_tun_from_nlattr(nla, tun_key);
1354                                         if (err)
1355                                                 return err;
1356                                 }
1357                                 break;
1358
1359                         case OVS_KEY_ATTR_IN_PORT:
1360                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1361                                         return -EINVAL;
1362                                 flow->key.phy.in_port = nla_get_u32(nla);
1363                                 break;
1364
1365                         case OVS_KEY_ATTR_SKB_MARK:
1366 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1367                                 if (nla_get_u32(nla) != 0)
1368                                         return -EINVAL;
1369 #endif
1370                                 flow->key.phy.skb_mark = nla_get_u32(nla);
1371                                 break;
1372                         }
1373                 }
1374         }
1375         if (rem)
1376                 return -EINVAL;
1377
1378         flow->hash = ovs_flow_hash(&flow->key,
1379                                    flow_key_start(&flow->key), key_len);
1380
1381         return 0;
1382 }
1383
1384 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1385 {
1386         struct ovs_key_ethernet *eth_key;
1387         struct nlattr *nla, *encap;
1388
1389         if (swkey->phy.priority &&
1390             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1391                 goto nla_put_failure;
1392
1393         if (swkey->tun_key.ipv4_dst &&
1394             ipv4_tun_to_nlattr(skb, &swkey->tun_key))
1395                 goto nla_put_failure;
1396
1397         if ((swkey->tun_key.tun_flags & OVS_TNL_F_KEY) &&
1398             nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->tun_key.tun_id))
1399                 goto nla_put_failure;
1400
1401         if (swkey->phy.in_port != DP_MAX_PORTS &&
1402             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1403                 goto nla_put_failure;
1404
1405         if (swkey->phy.skb_mark &&
1406             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1407                 goto nla_put_failure;
1408
1409         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1410         if (!nla)
1411                 goto nla_put_failure;
1412         eth_key = nla_data(nla);
1413         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1414         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1415
1416         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1417                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1418                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1419                         goto nla_put_failure;
1420                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1421                 if (!swkey->eth.tci)
1422                         goto unencap;
1423         } else {
1424                 encap = NULL;
1425         }
1426
1427         if (swkey->eth.type == htons(ETH_P_802_2))
1428                 goto unencap;
1429
1430         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1431                 goto nla_put_failure;
1432
1433         if (swkey->eth.type == htons(ETH_P_IP)) {
1434                 struct ovs_key_ipv4 *ipv4_key;
1435
1436                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1437                 if (!nla)
1438                         goto nla_put_failure;
1439                 ipv4_key = nla_data(nla);
1440                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1441                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1442                 ipv4_key->ipv4_proto = swkey->ip.proto;
1443                 ipv4_key->ipv4_tos = swkey->ip.tos;
1444                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1445                 ipv4_key->ipv4_frag = swkey->ip.frag;
1446         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1447                 struct ovs_key_ipv6 *ipv6_key;
1448
1449                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1450                 if (!nla)
1451                         goto nla_put_failure;
1452                 ipv6_key = nla_data(nla);
1453                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1454                                 sizeof(ipv6_key->ipv6_src));
1455                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1456                                 sizeof(ipv6_key->ipv6_dst));
1457                 ipv6_key->ipv6_label = swkey->ipv6.label;
1458                 ipv6_key->ipv6_proto = swkey->ip.proto;
1459                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1460                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1461                 ipv6_key->ipv6_frag = swkey->ip.frag;
1462         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1463                    swkey->eth.type == htons(ETH_P_RARP)) {
1464                 struct ovs_key_arp *arp_key;
1465
1466                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1467                 if (!nla)
1468                         goto nla_put_failure;
1469                 arp_key = nla_data(nla);
1470                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1471                 arp_key->arp_sip = swkey->ipv4.addr.src;
1472                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1473                 arp_key->arp_op = htons(swkey->ip.proto);
1474                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1475                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1476         }
1477
1478         if ((swkey->eth.type == htons(ETH_P_IP) ||
1479              swkey->eth.type == htons(ETH_P_IPV6)) &&
1480              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1481
1482                 if (swkey->ip.proto == IPPROTO_TCP) {
1483                         struct ovs_key_tcp *tcp_key;
1484
1485                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1486                         if (!nla)
1487                                 goto nla_put_failure;
1488                         tcp_key = nla_data(nla);
1489                         if (swkey->eth.type == htons(ETH_P_IP)) {
1490                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1491                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1492                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1493                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1494                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1495                         }
1496                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1497                         struct ovs_key_udp *udp_key;
1498
1499                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1500                         if (!nla)
1501                                 goto nla_put_failure;
1502                         udp_key = nla_data(nla);
1503                         if (swkey->eth.type == htons(ETH_P_IP)) {
1504                                 udp_key->udp_src = swkey->ipv4.tp.src;
1505                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1506                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1507                                 udp_key->udp_src = swkey->ipv6.tp.src;
1508                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1509                         }
1510                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1511                            swkey->ip.proto == IPPROTO_ICMP) {
1512                         struct ovs_key_icmp *icmp_key;
1513
1514                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1515                         if (!nla)
1516                                 goto nla_put_failure;
1517                         icmp_key = nla_data(nla);
1518                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1519                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1520                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1521                            swkey->ip.proto == IPPROTO_ICMPV6) {
1522                         struct ovs_key_icmpv6 *icmpv6_key;
1523
1524                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1525                                                 sizeof(*icmpv6_key));
1526                         if (!nla)
1527                                 goto nla_put_failure;
1528                         icmpv6_key = nla_data(nla);
1529                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1530                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1531
1532                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1533                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1534                                 struct ovs_key_nd *nd_key;
1535
1536                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1537                                 if (!nla)
1538                                         goto nla_put_failure;
1539                                 nd_key = nla_data(nla);
1540                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1541                                                         sizeof(nd_key->nd_target));
1542                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1543                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1544                         }
1545                 }
1546         }
1547
1548 unencap:
1549         if (encap)
1550                 nla_nest_end(skb, encap);
1551
1552         return 0;
1553
1554 nla_put_failure:
1555         return -EMSGSIZE;
1556 }
1557
1558 /* Initializes the flow module.
1559  * Returns zero if successful or a negative error code. */
1560 int ovs_flow_init(void)
1561 {
1562         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1563                                         0, NULL);
1564         if (flow_cache == NULL)
1565                 return -ENOMEM;
1566
1567         return 0;
1568 }
1569
1570 /* Uninitializes the flow module. */
1571 void ovs_flow_exit(void)
1572 {
1573         kmem_cache_destroy(flow_cache);
1574 }