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         flow->sf_acts = NULL;
 230
 231         return flow;
 232 }
 233
 234 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 235 {
 236         hash = jhash_1word(hash, table->hash_seed);
 237         return flex_array_get(table->buckets,
 238                                 (hash & (table->n_buckets - 1)));
 239 }
 240
 241 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 242 {
 243         struct flex_array *buckets;
 244         int i, err;
 245
 246         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 247                                    n_buckets, GFP_KERNEL);
 248         if (!buckets)
 249                 return NULL;
 250
 251         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 252         if (err) {
 253                 flex_array_free(buckets);
 254                 return NULL;
 255         }
 256
 257         for (i = 0; i < n_buckets; i++)
 258                 INIT_HLIST_HEAD((struct hlist_head *)
 259                                         flex_array_get(buckets, i));
 260
 261         return buckets;
 262 }
 263
 264 static void free_buckets(struct flex_array *buckets)
 265 {
 266         flex_array_free(buckets);
 267 }
 268
 269 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 270 {
 271         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 272
 273         if (!table)
 274                 return NULL;
 275
 276         table->buckets = alloc_buckets(new_size);
 277
 278         if (!table->buckets) {
 279                 kfree(table);
 280                 return NULL;
 281         }
 282         table->n_buckets = new_size;
 283         table->count = 0;
 284         table->node_ver = 0;
 285         table->keep_flows = false;
 286         get_random_bytes(&table->hash_seed, sizeof(u32));
 287
 288         return table;
 289 }
 290
 291 void ovs_flow_tbl_destroy(struct flow_table *table)
 292 {
 293         int i;
 294
 295         if (!table)
 296                 return;
 297
 298         if (table->keep_flows)
 299                 goto skip_flows;
 300
 301         for (i = 0; i < table->n_buckets; i++) {
 302                 struct sw_flow *flow;
 303                 struct hlist_head *head = flex_array_get(table->buckets, i);
 304                 struct hlist_node *node, *n;
 305                 int ver = table->node_ver;
 306
 307                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 308                         hlist_del_rcu(&flow->hash_node[ver]);
 309                         ovs_flow_free(flow);
 310                 }
 311         }
 312
 313 skip_flows:
 314         free_buckets(table->buckets);
 315         kfree(table);
 316 }
 317
 318 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 319 {
 320         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 321
 322         ovs_flow_tbl_destroy(table);
 323 }
 324
 325 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 326 {
 327         if (!table)
 328                 return;
 329
 330         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 331 }
 332
 333 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 334 {
 335         struct sw_flow *flow;
 336         struct hlist_head *head;
 337         struct hlist_node *n;
 338         int ver;
 339         int i;
 340
 341         ver = table->node_ver;
 342         while (*bucket < table->n_buckets) {
 343                 i = 0;
 344                 head = flex_array_get(table->buckets, *bucket);
 345                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 346                         if (i < *last) {
 347                                 i++;
 348                                 continue;
 349                         }
 350                         *last = i + 1;
 351                         return flow;
 352                 }
 353                 (*bucket)++;
 354                 *last = 0;
 355         }
 356
 357         return NULL;
 358 }
 359
 360 static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 361 {
 362         struct hlist_head *head;
 363         head = find_bucket(table, flow->hash);
 364         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 365         table->count++;
 366 }
 367
 368 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 369 {
 370         int old_ver;
 371         int i;
 372
 373         old_ver = old->node_ver;
 374         new->node_ver = !old_ver;
 375
 376         /* Insert in new table. */
 377         for (i = 0; i < old->n_buckets; i++) {
 378                 struct sw_flow *flow;
 379                 struct hlist_head *head;
 380                 struct hlist_node *n;
 381
 382                 head = flex_array_get(old->buckets, i);
 383
 384                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 385                         __flow_tbl_insert(new, flow);
 386         }
 387         old->keep_flows = true;
 388 }
 389
 390 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 391 {
 392         struct flow_table *new_table;
 393
 394         new_table = ovs_flow_tbl_alloc(n_buckets);
 395         if (!new_table)
 396                 return ERR_PTR(-ENOMEM);
 397
 398         flow_table_copy_flows(table, new_table);
 399
 400         return new_table;
 401 }
 402
 403 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 404 {
 405         return __flow_tbl_rehash(table, table->n_buckets);
 406 }
 407
 408 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 409 {
 410         return __flow_tbl_rehash(table, table->n_buckets * 2);
 411 }
 412
 413 void ovs_flow_free(struct sw_flow *flow)
 414 {
 415         if (unlikely(!flow))
 416                 return;
 417
 418         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 419         kmem_cache_free(flow_cache, flow);
 420 }
 421
 422 /* RCU callback used by ovs_flow_deferred_free. */
 423 static void rcu_free_flow_callback(struct rcu_head *rcu)
 424 {
 425         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 426
 427         ovs_flow_free(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 /* RCU callback used by ovs_flow_deferred_free_acts. */
 438 static void rcu_free_acts_callback(struct rcu_head *rcu)
 439 {
 440         struct sw_flow_actions *sf_acts = container_of(rcu,
 441                         struct sw_flow_actions, rcu);
 442         kfree(sf_acts);
 443 }
 444
 445 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 446  * The caller must hold rcu_read_lock for this to be sensible. */
 447 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 448 {
 449         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 450 }
 451
 452 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 453 {
 454         struct qtag_prefix {
 455                 __be16 eth_type; /* ETH_P_8021Q */
 456                 __be16 tci;
 457         };
 458         struct qtag_prefix *qp;
 459
 460         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 461                 return 0;
 462
 463         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 464                                          sizeof(__be16))))
 465                 return -ENOMEM;
 466
 467         qp = (struct qtag_prefix *) skb->data;
 468         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 469         __skb_pull(skb, sizeof(struct qtag_prefix));
 470
 471         return 0;
 472 }
 473
 474 static __be16 parse_ethertype(struct sk_buff *skb)
 475 {
 476         struct llc_snap_hdr {
 477                 u8  dsap;  /* Always 0xAA */
 478                 u8  ssap;  /* Always 0xAA */
 479                 u8  ctrl;
 480                 u8  oui[3];
 481                 __be16 ethertype;
 482         };
 483         struct llc_snap_hdr *llc;
 484         __be16 proto;
 485
 486         proto = *(__be16 *) skb->data;
 487         __skb_pull(skb, sizeof(__be16));
 488
 489         if (ntohs(proto) >= 1536)
 490                 return proto;
 491
 492         if (skb->len < sizeof(struct llc_snap_hdr))
 493                 return htons(ETH_P_802_2);
 494
 495         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 496                 return htons(0);
 497
 498         llc = (struct llc_snap_hdr *) skb->data;
 499         if (llc->dsap != LLC_SAP_SNAP ||
 500             llc->ssap != LLC_SAP_SNAP ||
 501             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 502                 return htons(ETH_P_802_2);
 503
 504         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 505         return llc->ethertype;
 506 }
 507
 508 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 509                         int *key_lenp, int nh_len)
 510 {
 511         struct icmp6hdr *icmp = icmp6_hdr(skb);
 512         int error = 0;
 513         int key_len;
 514
 515         /* The ICMPv6 type and code fields use the 16-bit transport port
 516          * fields, so we need to store them in 16-bit network byte order.
 517          */
 518         key->ipv6.tp.src = htons(icmp->icmp6_type);
 519         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 520         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 521
 522         if (icmp->icmp6_code == 0 &&
 523             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 524              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 525                 int icmp_len = skb->len - skb_transport_offset(skb);
 526                 struct nd_msg *nd;
 527                 int offset;
 528
 529                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 530
 531                 /* In order to process neighbor discovery options, we need the
 532                  * entire packet.
 533                  */
 534                 if (unlikely(icmp_len < sizeof(*nd)))
 535                         goto out;
 536                 if (unlikely(skb_linearize(skb))) {
 537                         error = -ENOMEM;
 538                         goto out;
 539                 }
 540
 541                 nd = (struct nd_msg *)skb_transport_header(skb);
 542                 key->ipv6.nd.target = nd->target;
 543                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 544
 545                 icmp_len -= sizeof(*nd);
 546                 offset = 0;
 547                 while (icmp_len >= 8) {
 548                         struct nd_opt_hdr *nd_opt =
 549                                  (struct nd_opt_hdr *)(nd->opt + offset);
 550                         int opt_len = nd_opt->nd_opt_len * 8;
 551
 552                         if (unlikely(!opt_len || opt_len > icmp_len))
 553                                 goto invalid;
 554
 555                         /* Store the link layer address if the appropriate
 556                          * option is provided.  It is considered an error if
 557                          * the same link layer option is specified twice.
 558                          */
 559                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 560                             && opt_len == 8) {
 561                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 562                                         goto invalid;
 563                                 memcpy(key->ipv6.nd.sll,
 564                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 565                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 566                                    && opt_len == 8) {
 567                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 568                                         goto invalid;
 569                                 memcpy(key->ipv6.nd.tll,
 570                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 571                         }
 572
 573                         icmp_len -= opt_len;
 574                         offset += opt_len;
 575                 }
 576         }
 577
 578         goto out;
 579
 580 invalid:
 581         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 582         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 583         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 584
 585 out:
 586         *key_lenp = key_len;
 587         return error;
 588 }
 589
 590 /**
 591  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 592  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 593  * Ethernet header
 594  * @in_port: port number on which @skb was received.
 595  * @key: output flow key
 596  * @key_lenp: length of output flow key
 597  *
 598  * The caller must ensure that skb->len >= ETH_HLEN.
 599  *
 600  * Returns 0 if successful, otherwise a negative errno value.
 601  *
 602  * Initializes @skb header pointers as follows:
 603  *
 604  *    - skb->mac_header: the Ethernet header.
 605  *
 606  *    - skb->network_header: just past the Ethernet header, or just past the
 607  *      VLAN header, to the first byte of the Ethernet payload.
 608  *
 609  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 610  *      on output, then just past the IP header, if one is present and
 611  *      of a correct length, otherwise the same as skb->network_header.
 612  *      For other key->dl_type values it is left untouched.
 613  */
 614 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 615                  int *key_lenp)
 616 {
 617         int error = 0;
 618         int key_len = SW_FLOW_KEY_OFFSET(eth);
 619         struct ethhdr *eth;
 620
 621         memset(key, 0, sizeof(*key));
 622
 623         key->phy.priority = skb->priority;
 624         if (OVS_CB(skb)->tun_key)
 625                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 626         key->phy.in_port = in_port;
 627         key->phy.skb_mark = skb_get_mark(skb);
 628
 629         skb_reset_mac_header(skb);
 630
 631         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 632          * header in the linear data area.
 633          */
 634         eth = eth_hdr(skb);
 635         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 636         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 637
 638         __skb_pull(skb, 2 * ETH_ALEN);
 639
 640         if (vlan_tx_tag_present(skb))
 641                 key->eth.tci = htons(vlan_get_tci(skb));
 642         else if (eth->h_proto == htons(ETH_P_8021Q))
 643                 if (unlikely(parse_vlan(skb, key)))
 644                         return -ENOMEM;
 645
 646         key->eth.type = parse_ethertype(skb);
 647         if (unlikely(key->eth.type == htons(0)))
 648                 return -ENOMEM;
 649
 650         skb_reset_network_header(skb);
 651         __skb_push(skb, skb->data - skb_mac_header(skb));
 652
 653         /* Network layer. */
 654         if (key->eth.type == htons(ETH_P_IP)) {
 655                 struct iphdr *nh;
 656                 __be16 offset;
 657
 658                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 659
 660                 error = check_iphdr(skb);
 661                 if (unlikely(error)) {
 662                         if (error == -EINVAL) {
 663                                 skb->transport_header = skb->network_header;
 664                                 error = 0;
 665                         }
 666                         goto out;
 667                 }
 668
 669                 nh = ip_hdr(skb);
 670                 key->ipv4.addr.src = nh->saddr;
 671                 key->ipv4.addr.dst = nh->daddr;
 672
 673                 key->ip.proto = nh->protocol;
 674                 key->ip.tos = nh->tos;
 675                 key->ip.ttl = nh->ttl;
 676
 677                 offset = nh->frag_off & htons(IP_OFFSET);
 678                 if (offset) {
 679                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 680                         goto out;
 681                 }
 682                 if (nh->frag_off & htons(IP_MF) ||
 683                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 684                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 685
 686                 /* Transport layer. */
 687                 if (key->ip.proto == IPPROTO_TCP) {
 688                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 689                         if (tcphdr_ok(skb)) {
 690                                 struct tcphdr *tcp = tcp_hdr(skb);
 691                                 key->ipv4.tp.src = tcp->source;
 692                                 key->ipv4.tp.dst = tcp->dest;
 693                         }
 694                 } else if (key->ip.proto == IPPROTO_UDP) {
 695                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 696                         if (udphdr_ok(skb)) {
 697                                 struct udphdr *udp = udp_hdr(skb);
 698                                 key->ipv4.tp.src = udp->source;
 699                                 key->ipv4.tp.dst = udp->dest;
 700                         }
 701                 } else if (key->ip.proto == IPPROTO_ICMP) {
 702                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 703                         if (icmphdr_ok(skb)) {
 704                                 struct icmphdr *icmp = icmp_hdr(skb);
 705                                 /* The ICMP type and code fields use the 16-bit
 706                                  * transport port fields, so we need to store
 707                                  * them in 16-bit network byte order. */
 708                                 key->ipv4.tp.src = htons(icmp->type);
 709                                 key->ipv4.tp.dst = htons(icmp->code);
 710                         }
 711                 }
 712
 713         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 714                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 715                 struct arp_eth_header *arp;
 716
 717                 arp = (struct arp_eth_header *)skb_network_header(skb);
 718
 719                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 720                                 && arp->ar_pro == htons(ETH_P_IP)
 721                                 && arp->ar_hln == ETH_ALEN
 722                                 && arp->ar_pln == 4) {
 723
 724                         /* We only match on the lower 8 bits of the opcode. */
 725                         if (ntohs(arp->ar_op) <= 0xff)
 726                                 key->ip.proto = ntohs(arp->ar_op);
 727                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 728                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 729                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 730                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 731                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 732                 }
 733         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 734                 int nh_len;             /* IPv6 Header + Extensions */
 735
 736                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 737                 if (unlikely(nh_len < 0)) {
 738                         if (nh_len == -EINVAL)
 739                                 skb->transport_header = skb->network_header;
 740                         else
 741                                 error = nh_len;
 742                         goto out;
 743                 }
 744
 745                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 746                         goto out;
 747                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 748                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 749
 750                 /* Transport layer. */
 751                 if (key->ip.proto == NEXTHDR_TCP) {
 752                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 753                         if (tcphdr_ok(skb)) {
 754                                 struct tcphdr *tcp = tcp_hdr(skb);
 755                                 key->ipv6.tp.src = tcp->source;
 756                                 key->ipv6.tp.dst = tcp->dest;
 757                         }
 758                 } else if (key->ip.proto == NEXTHDR_UDP) {
 759                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 760                         if (udphdr_ok(skb)) {
 761                                 struct udphdr *udp = udp_hdr(skb);
 762                                 key->ipv6.tp.src = udp->source;
 763                                 key->ipv6.tp.dst = udp->dest;
 764                         }
 765                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 766                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 767                         if (icmp6hdr_ok(skb)) {
 768                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 769                                 if (error < 0)
 770                                         goto out;
 771                         }
 772                 }
 773         }
 774
 775 out:
 776         *key_lenp = key_len;
 777         return error;
 778 }
 779
 780 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 781 {
 782         return jhash2((u32 *)((u8 *)key + key_start),
 783                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 784 }
 785
 786 static int flow_key_start(struct sw_flow_key *key)
 787 {
 788         if (key->tun_key.ipv4_dst)
 789                 return 0;
 790         else
 791                 return offsetof(struct sw_flow_key, phy);
 792 }
 793
 794 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 795                                 struct sw_flow_key *key, int key_len)
 796 {
 797         struct sw_flow *flow;
 798         struct hlist_node *n;
 799         struct hlist_head *head;
 800         u8 *_key;
 801         int key_start;
 802         u32 hash;
 803
 804         key_start = flow_key_start(key);
 805         hash = ovs_flow_hash(key, key_start, key_len);
 806
 807         _key = (u8 *) key + key_start;
 808         head = find_bucket(table, hash);
 809         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 810
 811                 if (flow->hash == hash &&
 812                     !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
 813                         return flow;
 814                 }
 815         }
 816         return NULL;
 817 }
 818
 819 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 820                          struct sw_flow_key *key, int key_len)
 821 {
 822         flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
 823         memcpy(&flow->key, key, sizeof(flow->key));
 824         __flow_tbl_insert(table, flow);
 825 }
 826
 827 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 828 {
 829         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 830         table->count--;
 831         BUG_ON(table->count < 0);
 832 }
 833
 834 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 835 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 836         [OVS_KEY_ATTR_ENCAP] = -1,
 837         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 838         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 839         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 840         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 841         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 842         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 843         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 844         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 845         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 846         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 847         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 848         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 849         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 850         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 851         [OVS_KEY_ATTR_IPV4_TUNNEL] = sizeof(struct ovs_key_ipv4_tunnel),
 852
 853         /* Not upstream. */
 854         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 855 };
 856
 857 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 858                                   const struct nlattr *a[], u64 *attrs)
 859 {
 860         const struct ovs_key_icmp *icmp_key;
 861         const struct ovs_key_tcp *tcp_key;
 862         const struct ovs_key_udp *udp_key;
 863
 864         switch (swkey->ip.proto) {
 865         case IPPROTO_TCP:
 866                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 867                         return -EINVAL;
 868                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 869
 870                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 871                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 872                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 873                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 874                 break;
 875
 876         case IPPROTO_UDP:
 877                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 878                         return -EINVAL;
 879                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 880
 881                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 882                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 883                 swkey->ipv4.tp.src = udp_key->udp_src;
 884                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 885                 break;
 886
 887         case IPPROTO_ICMP:
 888                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 889                         return -EINVAL;
 890                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 891
 892                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 893                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 894                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 895                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 896                 break;
 897         }
 898
 899         return 0;
 900 }
 901
 902 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 903                                   const struct nlattr *a[], u64 *attrs)
 904 {
 905         const struct ovs_key_icmpv6 *icmpv6_key;
 906         const struct ovs_key_tcp *tcp_key;
 907         const struct ovs_key_udp *udp_key;
 908
 909         switch (swkey->ip.proto) {
 910         case IPPROTO_TCP:
 911                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 912                         return -EINVAL;
 913                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 914
 915                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 916                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 917                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 918                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 919                 break;
 920
 921         case IPPROTO_UDP:
 922                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 923                         return -EINVAL;
 924                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 925
 926                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 927                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 928                 swkey->ipv6.tp.src = udp_key->udp_src;
 929                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 930                 break;
 931
 932         case IPPROTO_ICMPV6:
 933                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 934                         return -EINVAL;
 935                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 936
 937                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 938                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 939                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 940                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 941
 942                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 943                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 944                         const struct ovs_key_nd *nd_key;
 945
 946                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 947                                 return -EINVAL;
 948                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 949
 950                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 951                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 952                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 953                                sizeof(swkey->ipv6.nd.target));
 954                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 955                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 956                 }
 957                 break;
 958         }
 959
 960         return 0;
 961 }
 962
 963 static int parse_flow_nlattrs(const struct nlattr *attr,
 964                               const struct nlattr *a[], u64 *attrsp)
 965 {
 966         const struct nlattr *nla;
 967         u64 attrs;
 968         int rem;
 969
 970         attrs = 0;
 971         nla_for_each_nested(nla, attr, rem) {
 972                 u16 type = nla_type(nla);
 973                 int expected_len;
 974
 975                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 976                         return -EINVAL;
 977
 978                 expected_len = ovs_key_lens[type];
 979                 if (nla_len(nla) != expected_len && expected_len != -1)
 980                         return -EINVAL;
 981
 982                 attrs |= 1ULL << type;
 983                 a[type] = nla;
 984         }
 985         if (rem)
 986                 return -EINVAL;
 987
 988         *attrsp = attrs;
 989         return 0;
 990 }
 991
 992 /**
 993  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 994  * @swkey: receives the extracted flow key.
 995  * @key_lenp: number of bytes used in @swkey.
 996  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 997  * sequence.
 998  */
 999 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1000                       const struct nlattr *attr)
1001 {
1002         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1003         const struct ovs_key_ethernet *eth_key;
1004         int key_len;
1005         u64 attrs;
1006         int err;
1007
1008         memset(swkey, 0, sizeof(struct sw_flow_key));
1009         key_len = SW_FLOW_KEY_OFFSET(eth);
1010
1011         err = parse_flow_nlattrs(attr, a, &attrs);
1012         if (err)
1013                 return err;
1014
1015         /* Metadata attributes. */
1016         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1017                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1018                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1019         }
1020         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1021                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1022                 if (in_port >= DP_MAX_PORTS)
1023                         return -EINVAL;
1024                 swkey->phy.in_port = in_port;
1025                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1026         } else {
1027                 swkey->phy.in_port = DP_MAX_PORTS;
1028         }
1029         if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1030                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1031 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1032                 if (mark != 0)
1033                         return -EINVAL;
1034 #endif
1035                 swkey->phy.skb_mark = mark;
1036                 attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1037         }
1038
1039         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID) &&
1040             attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1041                 struct ovs_key_ipv4_tunnel *tun_key;
1042                 __be64 tun_id;
1043
1044                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1045
1046                 if (!tun_key->ipv4_dst)
1047                         return -EINVAL;
1048                 if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1049                         return -EINVAL;
1050
1051                 tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1052                 if (tun_id != tun_key->tun_id)
1053                         return -EINVAL;
1054
1055                 memcpy(&swkey->tun_key, tun_key, sizeof(swkey->tun_key));
1056                 memset(swkey->tun_key.pad, 0, sizeof(swkey->tun_key.pad));
1057
1058                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1059                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1060         } else if (attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1061                 struct ovs_key_ipv4_tunnel *tun_key;
1062                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1063
1064                 if (!tun_key->ipv4_dst)
1065                         return -EINVAL;
1066
1067                 memcpy(&swkey->tun_key, tun_key, sizeof(swkey->tun_key));
1068                 memset(swkey->tun_key.pad, 0, sizeof(swkey->tun_key.pad));
1069
1070                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1071         }
1072
1073         /* Data attributes. */
1074         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1075                 return -EINVAL;
1076         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1077
1078         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1079         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1080         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1081
1082         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1083             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1084                 const struct nlattr *encap;
1085                 __be16 tci;
1086
1087                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1088                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1089                               (1 << OVS_KEY_ATTR_ENCAP)))
1090                         return -EINVAL;
1091
1092                 encap = a[OVS_KEY_ATTR_ENCAP];
1093                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1094                 if (tci & htons(VLAN_TAG_PRESENT)) {
1095                         swkey->eth.tci = tci;
1096
1097                         err = parse_flow_nlattrs(encap, a, &attrs);
1098                         if (err)
1099                                 return err;
1100                 } else if (!tci) {
1101                         /* Corner case for truncated 802.1Q header. */
1102                         if (nla_len(encap))
1103                                 return -EINVAL;
1104
1105                         swkey->eth.type = htons(ETH_P_8021Q);
1106                         *key_lenp = key_len;
1107                         return 0;
1108                 } else {
1109                         return -EINVAL;
1110                 }
1111         }
1112
1113         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1114                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1115                 if (ntohs(swkey->eth.type) < 1536)
1116                         return -EINVAL;
1117                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1118         } else {
1119                 swkey->eth.type = htons(ETH_P_802_2);
1120         }
1121
1122         if (swkey->eth.type == htons(ETH_P_IP)) {
1123                 const struct ovs_key_ipv4 *ipv4_key;
1124
1125                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1126                         return -EINVAL;
1127                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1128
1129                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1130                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1131                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1132                         return -EINVAL;
1133                 swkey->ip.proto = ipv4_key->ipv4_proto;
1134                 swkey->ip.tos = ipv4_key->ipv4_tos;
1135                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1136                 swkey->ip.frag = ipv4_key->ipv4_frag;
1137                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1138                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1139
1140                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1141                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1142                         if (err)
1143                                 return err;
1144                 }
1145         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1146                 const struct ovs_key_ipv6 *ipv6_key;
1147
1148                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1149                         return -EINVAL;
1150                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1151
1152                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1153                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1154                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1155                         return -EINVAL;
1156                 swkey->ipv6.label = ipv6_key->ipv6_label;
1157                 swkey->ip.proto = ipv6_key->ipv6_proto;
1158                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1159                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1160                 swkey->ip.frag = ipv6_key->ipv6_frag;
1161                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1162                        sizeof(swkey->ipv6.addr.src));
1163                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1164                        sizeof(swkey->ipv6.addr.dst));
1165
1166                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1167                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1168                         if (err)
1169                                 return err;
1170                 }
1171         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1172                    swkey->eth.type == htons(ETH_P_RARP)) {
1173                 const struct ovs_key_arp *arp_key;
1174
1175                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1176                         return -EINVAL;
1177                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1178
1179                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1180                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1181                 swkey->ipv4.addr.src = arp_key->arp_sip;
1182                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1183                 if (arp_key->arp_op & htons(0xff00))
1184                         return -EINVAL;
1185                 swkey->ip.proto = ntohs(arp_key->arp_op);
1186                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1187                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1188         }
1189
1190         if (attrs)
1191                 return -EINVAL;
1192         *key_lenp = key_len;
1193
1194         return 0;
1195 }
1196
1197 /**
1198  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1199  * @in_port: receives the extracted input port.
1200  * @tun_id: receives the extracted tunnel ID.
1201  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1202  * sequence.
1203  *
1204  * This parses a series of Netlink attributes that form a flow key, which must
1205  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1206  * get the metadata, that is, the parts of the flow key that cannot be
1207  * extracted from the packet itself.
1208  */
1209
1210 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
1211 {
1212         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1213         const struct nlattr *nla;
1214         int rem;
1215         __be64 tun_id = 0;
1216
1217         flow->key.phy.in_port = DP_MAX_PORTS;
1218         flow->key.phy.priority = 0;
1219         flow->key.phy.skb_mark = 0;
1220         memset(tun_key, 0, sizeof(flow->key.tun_key));
1221
1222         nla_for_each_nested(nla, attr, rem) {
1223                 int type = nla_type(nla);
1224
1225                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1226                         if (nla_len(nla) != ovs_key_lens[type])
1227                                 return -EINVAL;
1228
1229                         switch (type) {
1230                         case OVS_KEY_ATTR_PRIORITY:
1231                                 flow->key.phy.priority = nla_get_u32(nla);
1232                                 break;
1233
1234                         case OVS_KEY_ATTR_TUN_ID:
1235                                 tun_id = nla_get_be64(nla);
1236
1237                                 if (tun_key->ipv4_dst) {
1238                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1239                                                 return -EINVAL;
1240                                         if (tun_key->tun_id != tun_id)
1241                                                 return -EINVAL;
1242                                         break;
1243                                 }
1244                                 tun_key->tun_id = tun_id;
1245                                 tun_key->tun_flags |= OVS_TNL_F_KEY;
1246
1247                                 break;
1248
1249                         case OVS_KEY_ATTR_IPV4_TUNNEL:
1250                                 if (tun_key->tun_flags & OVS_TNL_F_KEY) {
1251                                         tun_id = tun_key->tun_id;
1252
1253                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1254                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1255                                                 return -EINVAL;
1256
1257                                         if (tun_key->tun_id != tun_id)
1258                                                 return -EINVAL;
1259                                 } else
1260                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1261
1262                                 if (!tun_key->ipv4_dst)
1263                                         return -EINVAL;
1264                                 break;
1265
1266                         case OVS_KEY_ATTR_IN_PORT:
1267                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1268                                         return -EINVAL;
1269                                 flow->key.phy.in_port = nla_get_u32(nla);
1270                                 break;
1271
1272                         case OVS_KEY_ATTR_SKB_MARK:
1273 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1274                                 if (nla_get_u32(nla) != 0)
1275                                         return -EINVAL;
1276 #endif
1277                                 flow->key.phy.skb_mark = nla_get_u32(nla);
1278                                 break;
1279                         }
1280                 }
1281         }
1282         if (rem)
1283                 return -EINVAL;
1284
1285         flow->hash = ovs_flow_hash(&flow->key,
1286                                    flow_key_start(&flow->key), key_len);
1287
1288         return 0;
1289 }
1290
1291 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1292 {
1293         struct ovs_key_ethernet *eth_key;
1294         struct nlattr *nla, *encap;
1295
1296         if (swkey->phy.priority &&
1297             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1298                 goto nla_put_failure;
1299
1300         if (swkey->tun_key.ipv4_dst) {
1301                 struct ovs_key_ipv4_tunnel *tun_key;
1302                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4_TUNNEL, sizeof(*tun_key));
1303                 if (!nla)
1304                         goto nla_put_failure;
1305                 tun_key = nla_data(nla);
1306                 memcpy(tun_key, &swkey->tun_key, sizeof(*tun_key));
1307         }
1308         if ((swkey->tun_key.tun_flags & OVS_TNL_F_KEY) &&
1309             nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->tun_key.tun_id))
1310                 goto nla_put_failure;
1311
1312         if (swkey->phy.in_port != DP_MAX_PORTS &&
1313             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1314                 goto nla_put_failure;
1315
1316         if (swkey->phy.skb_mark &&
1317             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1318                 goto nla_put_failure;
1319
1320         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1321         if (!nla)
1322                 goto nla_put_failure;
1323         eth_key = nla_data(nla);
1324         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1325         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1326
1327         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1328                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1329                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1330                         goto nla_put_failure;
1331                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1332                 if (!swkey->eth.tci)
1333                         goto unencap;
1334         } else {
1335                 encap = NULL;
1336         }
1337
1338         if (swkey->eth.type == htons(ETH_P_802_2))
1339                 goto unencap;
1340
1341         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1342                 goto nla_put_failure;
1343
1344         if (swkey->eth.type == htons(ETH_P_IP)) {
1345                 struct ovs_key_ipv4 *ipv4_key;
1346
1347                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1348                 if (!nla)
1349                         goto nla_put_failure;
1350                 ipv4_key = nla_data(nla);
1351                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1352                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1353                 ipv4_key->ipv4_proto = swkey->ip.proto;
1354                 ipv4_key->ipv4_tos = swkey->ip.tos;
1355                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1356                 ipv4_key->ipv4_frag = swkey->ip.frag;
1357         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1358                 struct ovs_key_ipv6 *ipv6_key;
1359
1360                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1361                 if (!nla)
1362                         goto nla_put_failure;
1363                 ipv6_key = nla_data(nla);
1364                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1365                                 sizeof(ipv6_key->ipv6_src));
1366                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1367                                 sizeof(ipv6_key->ipv6_dst));
1368                 ipv6_key->ipv6_label = swkey->ipv6.label;
1369                 ipv6_key->ipv6_proto = swkey->ip.proto;
1370                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1371                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1372                 ipv6_key->ipv6_frag = swkey->ip.frag;
1373         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1374                    swkey->eth.type == htons(ETH_P_RARP)) {
1375                 struct ovs_key_arp *arp_key;
1376
1377                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1378                 if (!nla)
1379                         goto nla_put_failure;
1380                 arp_key = nla_data(nla);
1381                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1382                 arp_key->arp_sip = swkey->ipv4.addr.src;
1383                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1384                 arp_key->arp_op = htons(swkey->ip.proto);
1385                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1386                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1387         }
1388
1389         if ((swkey->eth.type == htons(ETH_P_IP) ||
1390              swkey->eth.type == htons(ETH_P_IPV6)) &&
1391              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1392
1393                 if (swkey->ip.proto == IPPROTO_TCP) {
1394                         struct ovs_key_tcp *tcp_key;
1395
1396                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1397                         if (!nla)
1398                                 goto nla_put_failure;
1399                         tcp_key = nla_data(nla);
1400                         if (swkey->eth.type == htons(ETH_P_IP)) {
1401                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1402                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1403                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1404                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1405                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1406                         }
1407                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1408                         struct ovs_key_udp *udp_key;
1409
1410                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1411                         if (!nla)
1412                                 goto nla_put_failure;
1413                         udp_key = nla_data(nla);
1414                         if (swkey->eth.type == htons(ETH_P_IP)) {
1415                                 udp_key->udp_src = swkey->ipv4.tp.src;
1416                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1417                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1418                                 udp_key->udp_src = swkey->ipv6.tp.src;
1419                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1420                         }
1421                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1422                            swkey->ip.proto == IPPROTO_ICMP) {
1423                         struct ovs_key_icmp *icmp_key;
1424
1425                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1426                         if (!nla)
1427                                 goto nla_put_failure;
1428                         icmp_key = nla_data(nla);
1429                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1430                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1431                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1432                            swkey->ip.proto == IPPROTO_ICMPV6) {
1433                         struct ovs_key_icmpv6 *icmpv6_key;
1434
1435                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1436                                                 sizeof(*icmpv6_key));
1437                         if (!nla)
1438                                 goto nla_put_failure;
1439                         icmpv6_key = nla_data(nla);
1440                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1441                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1442
1443                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1444                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1445                                 struct ovs_key_nd *nd_key;
1446
1447                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1448                                 if (!nla)
1449                                         goto nla_put_failure;
1450                                 nd_key = nla_data(nla);
1451                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1452                                                         sizeof(nd_key->nd_target));
1453                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1454                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1455                         }
1456                 }
1457         }
1458
1459 unencap:
1460         if (encap)
1461                 nla_nest_end(skb, encap);
1462
1463         return 0;
1464
1465 nla_put_failure:
1466         return -EMSGSIZE;
1467 }
1468
1469 /* Initializes the flow module.
1470  * Returns zero if successful or a negative error code. */
1471 int ovs_flow_init(void)
1472 {
1473         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1474                                         0, NULL);
1475         if (flow_cache == NULL)
1476                 return -ENOMEM;
1477
1478         return 0;
1479 }
1480
1481 /* Uninitializes the flow module. */
1482 void ovs_flow_exit(void)
1483 {
1484         kmem_cache_destroy(flow_cache);
1485 }