pintos-os.org Git - openvswitch/blob - 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
 628         skb_reset_mac_header(skb);
 629
 630         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 631          * header in the linear data area.
 632          */
 633         eth = eth_hdr(skb);
 634         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 635         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 636
 637         __skb_pull(skb, 2 * ETH_ALEN);
 638
 639         if (vlan_tx_tag_present(skb))
 640                 key->eth.tci = htons(vlan_get_tci(skb));
 641         else if (eth->h_proto == htons(ETH_P_8021Q))
 642                 if (unlikely(parse_vlan(skb, key)))
 643                         return -ENOMEM;
 644
 645         key->eth.type = parse_ethertype(skb);
 646         if (unlikely(key->eth.type == htons(0)))
 647                 return -ENOMEM;
 648
 649         skb_reset_network_header(skb);
 650         __skb_push(skb, skb->data - skb_mac_header(skb));
 651
 652         /* Network layer. */
 653         if (key->eth.type == htons(ETH_P_IP)) {
 654                 struct iphdr *nh;
 655                 __be16 offset;
 656
 657                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 658
 659                 error = check_iphdr(skb);
 660                 if (unlikely(error)) {
 661                         if (error == -EINVAL) {
 662                                 skb->transport_header = skb->network_header;
 663                                 error = 0;
 664                         }
 665                         goto out;
 666                 }
 667
 668                 nh = ip_hdr(skb);
 669                 key->ipv4.addr.src = nh->saddr;
 670                 key->ipv4.addr.dst = nh->daddr;
 671
 672                 key->ip.proto = nh->protocol;
 673                 key->ip.tos = nh->tos;
 674                 key->ip.ttl = nh->ttl;
 675
 676                 offset = nh->frag_off & htons(IP_OFFSET);
 677                 if (offset) {
 678                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 679                         goto out;
 680                 }
 681                 if (nh->frag_off & htons(IP_MF) ||
 682                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 683                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 684
 685                 /* Transport layer. */
 686                 if (key->ip.proto == IPPROTO_TCP) {
 687                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 688                         if (tcphdr_ok(skb)) {
 689                                 struct tcphdr *tcp = tcp_hdr(skb);
 690                                 key->ipv4.tp.src = tcp->source;
 691                                 key->ipv4.tp.dst = tcp->dest;
 692                         }
 693                 } else if (key->ip.proto == IPPROTO_UDP) {
 694                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 695                         if (udphdr_ok(skb)) {
 696                                 struct udphdr *udp = udp_hdr(skb);
 697                                 key->ipv4.tp.src = udp->source;
 698                                 key->ipv4.tp.dst = udp->dest;
 699                         }
 700                 } else if (key->ip.proto == IPPROTO_ICMP) {
 701                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 702                         if (icmphdr_ok(skb)) {
 703                                 struct icmphdr *icmp = icmp_hdr(skb);
 704                                 /* The ICMP type and code fields use the 16-bit
 705                                  * transport port fields, so we need to store
 706                                  * them in 16-bit network byte order. */
 707                                 key->ipv4.tp.src = htons(icmp->type);
 708                                 key->ipv4.tp.dst = htons(icmp->code);
 709                         }
 710                 }
 711
 712         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 713                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 714                 struct arp_eth_header *arp;
 715
 716                 arp = (struct arp_eth_header *)skb_network_header(skb);
 717
 718                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 719                                 && arp->ar_pro == htons(ETH_P_IP)
 720                                 && arp->ar_hln == ETH_ALEN
 721                                 && arp->ar_pln == 4) {
 722
 723                         /* We only match on the lower 8 bits of the opcode. */
 724                         if (ntohs(arp->ar_op) <= 0xff)
 725                                 key->ip.proto = ntohs(arp->ar_op);
 726                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 727                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 728                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 729                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 730                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 731                 }
 732         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 733                 int nh_len;             /* IPv6 Header + Extensions */
 734
 735                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 736                 if (unlikely(nh_len < 0)) {
 737                         if (nh_len == -EINVAL)
 738                                 skb->transport_header = skb->network_header;
 739                         else
 740                                 error = nh_len;
 741                         goto out;
 742                 }
 743
 744                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 745                         goto out;
 746                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 747                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 748
 749                 /* Transport layer. */
 750                 if (key->ip.proto == NEXTHDR_TCP) {
 751                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 752                         if (tcphdr_ok(skb)) {
 753                                 struct tcphdr *tcp = tcp_hdr(skb);
 754                                 key->ipv6.tp.src = tcp->source;
 755                                 key->ipv6.tp.dst = tcp->dest;
 756                         }
 757                 } else if (key->ip.proto == NEXTHDR_UDP) {
 758                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 759                         if (udphdr_ok(skb)) {
 760                                 struct udphdr *udp = udp_hdr(skb);
 761                                 key->ipv6.tp.src = udp->source;
 762                                 key->ipv6.tp.dst = udp->dest;
 763                         }
 764                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 765                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 766                         if (icmp6hdr_ok(skb)) {
 767                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 768                                 if (error < 0)
 769                                         goto out;
 770                         }
 771                 }
 772         }
 773
 774 out:
 775         *key_lenp = key_len;
 776         return error;
 777 }
 778
 779 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 780 {
 781         return jhash2((u32 *)((u8 *)key + key_start),
 782                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 783 }
 784
 785 static int flow_key_start(struct sw_flow_key *key)
 786 {
 787         if (key->tun_key.ipv4_dst)
 788                 return 0;
 789         else
 790                 return offsetof(struct sw_flow_key, phy);
 791 }
 792
 793 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 794                                 struct sw_flow_key *key, int key_len)
 795 {
 796         struct sw_flow *flow;
 797         struct hlist_node *n;
 798         struct hlist_head *head;
 799         u8 *_key;
 800         int key_start;
 801         u32 hash;
 802
 803         key_start = flow_key_start(key);
 804         hash = ovs_flow_hash(key, key_start, key_len);
 805
 806         _key = (u8 *) key + key_start;
 807         head = find_bucket(table, hash);
 808         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 809
 810                 if (flow->hash == hash &&
 811                     !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
 812                         return flow;
 813                 }
 814         }
 815         return NULL;
 816 }
 817
 818 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 819                          struct sw_flow_key *key, int key_len)
 820 {
 821         flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
 822         memcpy(&flow->key, key, sizeof(flow->key));
 823         __flow_tbl_insert(table, flow);
 824 }
 825
 826 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 827 {
 828         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 829         table->count--;
 830         BUG_ON(table->count < 0);
 831 }
 832
 833 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 834 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 835         [OVS_KEY_ATTR_ENCAP] = -1,
 836         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 837         [OVS_KEY_ATTR_IN_PORT] = 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_IPV4_TUNNEL] = sizeof(struct ovs_key_ipv4_tunnel),
 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 /**
 991  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 992  * @swkey: receives the extracted flow key.
 993  * @key_lenp: number of bytes used in @swkey.
 994  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 995  * sequence.
 996  */
 997 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 998                       const struct nlattr *attr)
 999 {
1000         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1001         const struct ovs_key_ethernet *eth_key;
1002         int key_len;
1003         u64 attrs;
1004         int err;
1005
1006         memset(swkey, 0, sizeof(struct sw_flow_key));
1007         key_len = SW_FLOW_KEY_OFFSET(eth);
1008
1009         err = parse_flow_nlattrs(attr, a, &attrs);
1010         if (err)
1011                 return err;
1012
1013         /* Metadata attributes. */
1014         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1015                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1016                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1017         }
1018         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1019                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1020                 if (in_port >= DP_MAX_PORTS)
1021                         return -EINVAL;
1022                 swkey->phy.in_port = in_port;
1023                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1024         } else {
1025                 swkey->phy.in_port = DP_MAX_PORTS;
1026         }
1027
1028         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID) &&
1029             attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1030                 struct ovs_key_ipv4_tunnel *tun_key;
1031                 __be64 tun_id;
1032
1033                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1034
1035                 if (!tun_key->ipv4_dst)
1036                         return -EINVAL;
1037                 if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1038                         return -EINVAL;
1039
1040                 tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1041                 if (tun_id != tun_key->tun_id)
1042                         return -EINVAL;
1043
1044                 memcpy(&swkey->tun_key, tun_key, sizeof(swkey->tun_key));
1045
1046                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1047                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1048         } else if (attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1049                 struct ovs_key_ipv4_tunnel *tun_key;
1050                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1051
1052                 if (!tun_key->ipv4_dst)
1053                         return -EINVAL;
1054
1055                 memcpy(&swkey->tun_key, tun_key, sizeof(swkey->tun_key));
1056
1057                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1058         }
1059
1060         /* Data attributes. */
1061         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1062                 return -EINVAL;
1063         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1064
1065         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1066         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1067         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1068
1069         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1070             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1071                 const struct nlattr *encap;
1072                 __be16 tci;
1073
1074                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1075                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1076                               (1 << OVS_KEY_ATTR_ENCAP)))
1077                         return -EINVAL;
1078
1079                 encap = a[OVS_KEY_ATTR_ENCAP];
1080                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1081                 if (tci & htons(VLAN_TAG_PRESENT)) {
1082                         swkey->eth.tci = tci;
1083
1084                         err = parse_flow_nlattrs(encap, a, &attrs);
1085                         if (err)
1086                                 return err;
1087                 } else if (!tci) {
1088                         /* Corner case for truncated 802.1Q header. */
1089                         if (nla_len(encap))
1090                                 return -EINVAL;
1091
1092                         swkey->eth.type = htons(ETH_P_8021Q);
1093                         *key_lenp = key_len;
1094                         return 0;
1095                 } else {
1096                         return -EINVAL;
1097                 }
1098         }
1099
1100         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1101                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1102                 if (ntohs(swkey->eth.type) < 1536)
1103                         return -EINVAL;
1104                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1105         } else {
1106                 swkey->eth.type = htons(ETH_P_802_2);
1107         }
1108
1109         if (swkey->eth.type == htons(ETH_P_IP)) {
1110                 const struct ovs_key_ipv4 *ipv4_key;
1111
1112                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1113                         return -EINVAL;
1114                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1115
1116                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1117                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1118                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1119                         return -EINVAL;
1120                 swkey->ip.proto = ipv4_key->ipv4_proto;
1121                 swkey->ip.tos = ipv4_key->ipv4_tos;
1122                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1123                 swkey->ip.frag = ipv4_key->ipv4_frag;
1124                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1125                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1126
1127                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1128                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1129                         if (err)
1130                                 return err;
1131                 }
1132         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1133                 const struct ovs_key_ipv6 *ipv6_key;
1134
1135                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1136                         return -EINVAL;
1137                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1138
1139                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1140                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1141                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1142                         return -EINVAL;
1143                 swkey->ipv6.label = ipv6_key->ipv6_label;
1144                 swkey->ip.proto = ipv6_key->ipv6_proto;
1145                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1146                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1147                 swkey->ip.frag = ipv6_key->ipv6_frag;
1148                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1149                        sizeof(swkey->ipv6.addr.src));
1150                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1151                        sizeof(swkey->ipv6.addr.dst));
1152
1153                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1154                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1155                         if (err)
1156                                 return err;
1157                 }
1158         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1159                    swkey->eth.type == htons(ETH_P_RARP)) {
1160                 const struct ovs_key_arp *arp_key;
1161
1162                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1163                         return -EINVAL;
1164                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1165
1166                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1167                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1168                 swkey->ipv4.addr.src = arp_key->arp_sip;
1169                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1170                 if (arp_key->arp_op & htons(0xff00))
1171                         return -EINVAL;
1172                 swkey->ip.proto = ntohs(arp_key->arp_op);
1173                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1174                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1175         }
1176
1177         if (attrs)
1178                 return -EINVAL;
1179         *key_lenp = key_len;
1180
1181         return 0;
1182 }
1183
1184 /**
1185  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1186  * @in_port: receives the extracted input port.
1187  * @tun_id: receives the extracted tunnel ID.
1188  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1189  * sequence.
1190  *
1191  * This parses a series of Netlink attributes that form a flow key, which must
1192  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1193  * get the metadata, that is, the parts of the flow key that cannot be
1194  * extracted from the packet itself.
1195  */
1196
1197 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
1198 {
1199         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1200         const struct nlattr *nla;
1201         int rem;
1202         __be64 tun_id = 0;
1203
1204         flow->key.phy.in_port = DP_MAX_PORTS;
1205         flow->key.phy.priority = 0;
1206         memset(tun_key, 0, sizeof(flow->key.tun_key));
1207
1208         nla_for_each_nested(nla, attr, rem) {
1209                 int type = nla_type(nla);
1210
1211                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1212                         if (nla_len(nla) != ovs_key_lens[type])
1213                                 return -EINVAL;
1214
1215                         switch (type) {
1216                         case OVS_KEY_ATTR_PRIORITY:
1217                                 flow->key.phy.priority = nla_get_u32(nla);
1218                                 break;
1219
1220                         case OVS_KEY_ATTR_TUN_ID:
1221                                 tun_id = nla_get_be64(nla);
1222
1223                                 if (tun_key->ipv4_dst) {
1224                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1225                                                 return -EINVAL;
1226                                         if (tun_key->tun_id != tun_id)
1227                                                 return -EINVAL;
1228                                         break;
1229                                 }
1230                                 tun_key->tun_id = tun_id;
1231                                 tun_key->tun_flags |= OVS_TNL_F_KEY;
1232
1233                                 break;
1234
1235                         case OVS_KEY_ATTR_IPV4_TUNNEL:
1236                                 if (tun_key->tun_flags & OVS_TNL_F_KEY) {
1237                                         tun_id = tun_key->tun_id;
1238
1239                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1240                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1241                                                 return -EINVAL;
1242
1243                                         if (tun_key->tun_id != tun_id)
1244                                                 return -EINVAL;
1245                                 } else
1246                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1247
1248                                 if (!tun_key->ipv4_dst)
1249                                         return -EINVAL;
1250                                 break;
1251
1252                         case OVS_KEY_ATTR_IN_PORT:
1253                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1254                                         return -EINVAL;
1255                                 flow->key.phy.in_port = nla_get_u32(nla);
1256                                 break;
1257                         }
1258                 }
1259         }
1260         if (rem)
1261                 return -EINVAL;
1262
1263         flow->hash = ovs_flow_hash(&flow->key,
1264                                    flow_key_start(&flow->key), key_len);
1265
1266         return 0;
1267 }
1268
1269 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1270 {
1271         struct ovs_key_ethernet *eth_key;
1272         struct nlattr *nla, *encap;
1273
1274         if (swkey->phy.priority &&
1275             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1276                 goto nla_put_failure;
1277
1278         if (swkey->tun_key.ipv4_dst) {
1279                 struct ovs_key_ipv4_tunnel *tun_key;
1280                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4_TUNNEL, sizeof(*tun_key));
1281                 if (!nla)
1282                         goto nla_put_failure;
1283                 tun_key = nla_data(nla);
1284                 memcpy(tun_key, &swkey->tun_key, sizeof(*tun_key));
1285         }
1286         if ((swkey->tun_key.tun_flags & OVS_TNL_F_KEY) &&
1287             nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->tun_key.tun_id))
1288                 goto nla_put_failure;
1289
1290         if (swkey->phy.in_port != DP_MAX_PORTS &&
1291             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1292                 goto nla_put_failure;
1293
1294         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1295         if (!nla)
1296                 goto nla_put_failure;
1297         eth_key = nla_data(nla);
1298         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1299         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1300
1301         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1302                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1303                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1304                         goto nla_put_failure;
1305                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1306                 if (!swkey->eth.tci)
1307                         goto unencap;
1308         } else {
1309                 encap = NULL;
1310         }
1311
1312         if (swkey->eth.type == htons(ETH_P_802_2))
1313                 goto unencap;
1314
1315         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1316                 goto nla_put_failure;
1317
1318         if (swkey->eth.type == htons(ETH_P_IP)) {
1319                 struct ovs_key_ipv4 *ipv4_key;
1320
1321                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1322                 if (!nla)
1323                         goto nla_put_failure;
1324                 ipv4_key = nla_data(nla);
1325                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1326                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1327                 ipv4_key->ipv4_proto = swkey->ip.proto;
1328                 ipv4_key->ipv4_tos = swkey->ip.tos;
1329                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1330                 ipv4_key->ipv4_frag = swkey->ip.frag;
1331         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1332                 struct ovs_key_ipv6 *ipv6_key;
1333
1334                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1335                 if (!nla)
1336                         goto nla_put_failure;
1337                 ipv6_key = nla_data(nla);
1338                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1339                                 sizeof(ipv6_key->ipv6_src));
1340                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1341                                 sizeof(ipv6_key->ipv6_dst));
1342                 ipv6_key->ipv6_label = swkey->ipv6.label;
1343                 ipv6_key->ipv6_proto = swkey->ip.proto;
1344                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1345                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1346                 ipv6_key->ipv6_frag = swkey->ip.frag;
1347         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1348                    swkey->eth.type == htons(ETH_P_RARP)) {
1349                 struct ovs_key_arp *arp_key;
1350
1351                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1352                 if (!nla)
1353                         goto nla_put_failure;
1354                 arp_key = nla_data(nla);
1355                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1356                 arp_key->arp_sip = swkey->ipv4.addr.src;
1357                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1358                 arp_key->arp_op = htons(swkey->ip.proto);
1359                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1360                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1361         }
1362
1363         if ((swkey->eth.type == htons(ETH_P_IP) ||
1364              swkey->eth.type == htons(ETH_P_IPV6)) &&
1365              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1366
1367                 if (swkey->ip.proto == IPPROTO_TCP) {
1368                         struct ovs_key_tcp *tcp_key;
1369
1370                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1371                         if (!nla)
1372                                 goto nla_put_failure;
1373                         tcp_key = nla_data(nla);
1374                         if (swkey->eth.type == htons(ETH_P_IP)) {
1375                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1376                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1377                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1378                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1379                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1380                         }
1381                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1382                         struct ovs_key_udp *udp_key;
1383
1384                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1385                         if (!nla)
1386                                 goto nla_put_failure;
1387                         udp_key = nla_data(nla);
1388                         if (swkey->eth.type == htons(ETH_P_IP)) {
1389                                 udp_key->udp_src = swkey->ipv4.tp.src;
1390                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1391                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1392                                 udp_key->udp_src = swkey->ipv6.tp.src;
1393                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1394                         }
1395                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1396                            swkey->ip.proto == IPPROTO_ICMP) {
1397                         struct ovs_key_icmp *icmp_key;
1398
1399                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1400                         if (!nla)
1401                                 goto nla_put_failure;
1402                         icmp_key = nla_data(nla);
1403                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1404                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1405                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1406                            swkey->ip.proto == IPPROTO_ICMPV6) {
1407                         struct ovs_key_icmpv6 *icmpv6_key;
1408
1409                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1410                                                 sizeof(*icmpv6_key));
1411                         if (!nla)
1412                                 goto nla_put_failure;
1413                         icmpv6_key = nla_data(nla);
1414                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1415                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1416
1417                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1418                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1419                                 struct ovs_key_nd *nd_key;
1420
1421                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1422                                 if (!nla)
1423                                         goto nla_put_failure;
1424                                 nd_key = nla_data(nla);
1425                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1426                                                         sizeof(nd_key->nd_target));
1427                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1428                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1429                         }
1430                 }
1431         }
1432
1433 unencap:
1434         if (encap)
1435                 nla_nest_end(skb, encap);
1436
1437         return 0;
1438
1439 nla_put_failure:
1440         return -EMSGSIZE;
1441 }
1442
1443 /* Initializes the flow module.
1444  * Returns zero if successful or a negative error code. */
1445 int ovs_flow_init(void)
1446 {
1447         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1448                                         0, NULL);
1449         if (flow_cache == NULL)
1450                 return -ENOMEM;
1451
1452         return 0;
1453 }
1454
1455 /* Uninitializes the flow module. */
1456 void ovs_flow_exit(void)
1457 {
1458         kmem_cache_destroy(flow_cache);
1459 }