datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #include "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 #include "vlan.h"
  47
  48 static struct kmem_cache *flow_cache;
  49
  50 static int check_header(struct sk_buff *skb, int len)
  51 {
  52         if (unlikely(skb->len < len))
  53                 return -EINVAL;
  54         if (unlikely(!pskb_may_pull(skb, len)))
  55                 return -ENOMEM;
  56         return 0;
  57 }
  58
  59 static bool arphdr_ok(struct sk_buff *skb)
  60 {
  61         return pskb_may_pull(skb, skb_network_offset(skb) +
  62                                   sizeof(struct arp_eth_header));
  63 }
  64
  65 static int check_iphdr(struct sk_buff *skb)
  66 {
  67         unsigned int nh_ofs = skb_network_offset(skb);
  68         unsigned int ip_len;
  69         int err;
  70
  71         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  72         if (unlikely(err))
  73                 return err;
  74
  75         ip_len = ip_hdrlen(skb);
  76         if (unlikely(ip_len < sizeof(struct iphdr) ||
  77                      skb->len < nh_ofs + ip_len))
  78                 return -EINVAL;
  79
  80         skb_set_transport_header(skb, nh_ofs + ip_len);
  81         return 0;
  82 }
  83
  84 static bool tcphdr_ok(struct sk_buff *skb)
  85 {
  86         int th_ofs = skb_transport_offset(skb);
  87         int tcp_len;
  88
  89         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  90                 return false;
  91
  92         tcp_len = tcp_hdrlen(skb);
  93         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  94                      skb->len < th_ofs + tcp_len))
  95                 return false;
  96
  97         return true;
  98 }
  99
 100 static bool udphdr_ok(struct sk_buff *skb)
 101 {
 102         return pskb_may_pull(skb, skb_transport_offset(skb) +
 103                                   sizeof(struct udphdr));
 104 }
 105
 106 static bool icmphdr_ok(struct sk_buff *skb)
 107 {
 108         return pskb_may_pull(skb, skb_transport_offset(skb) +
 109                                   sizeof(struct icmphdr));
 110 }
 111
 112 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 113 {
 114         struct timespec cur_ts;
 115         u64 cur_ms, idle_ms;
 116
 117         ktime_get_ts(&cur_ts);
 118         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 119         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 120                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 121
 122         return cur_ms - idle_ms;
 123 }
 124
 125 #define SW_FLOW_KEY_OFFSET(field)               \
 126         (offsetof(struct sw_flow_key, field) +  \
 127          FIELD_SIZEOF(struct sw_flow_key, field))
 128
 129 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 130                          int *key_lenp)
 131 {
 132         unsigned int nh_ofs = skb_network_offset(skb);
 133         unsigned int nh_len;
 134         int payload_ofs;
 135         struct ipv6hdr *nh;
 136         uint8_t nexthdr;
 137         __be16 frag_off;
 138         int err;
 139
 140         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 141
 142         err = check_header(skb, nh_ofs + sizeof(*nh));
 143         if (unlikely(err))
 144                 return err;
 145
 146         nh = ipv6_hdr(skb);
 147         nexthdr = nh->nexthdr;
 148         payload_ofs = (u8 *)(nh + 1) - skb->data;
 149
 150         key->ip.proto = NEXTHDR_NONE;
 151         key->ip.tos = ipv6_get_dsfield(nh);
 152         key->ip.ttl = nh->hop_limit;
 153         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 154         key->ipv6.addr.src = nh->saddr;
 155         key->ipv6.addr.dst = nh->daddr;
 156
 157         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 158         if (unlikely(payload_ofs < 0))
 159                 return -EINVAL;
 160
 161         if (frag_off) {
 162                 if (frag_off & htons(~0x7))
 163                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 164                 else
 165                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 166         }
 167
 168         nh_len = payload_ofs - nh_ofs;
 169         skb_set_transport_header(skb, nh_ofs + nh_len);
 170         key->ip.proto = nexthdr;
 171         return nh_len;
 172 }
 173
 174 static bool icmp6hdr_ok(struct sk_buff *skb)
 175 {
 176         return pskb_may_pull(skb, skb_transport_offset(skb) +
 177                                   sizeof(struct icmp6hdr));
 178 }
 179
 180 #define TCP_FLAGS_OFFSET 13
 181 #define TCP_FLAG_MASK 0x3f
 182
 183 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 184 {
 185         u8 tcp_flags = 0;
 186
 187         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 188              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 189             flow->key.ip.proto == IPPROTO_TCP &&
 190             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 191                 u8 *tcp = (u8 *)tcp_hdr(skb);
 192                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 193         }
 194
 195         spin_lock(&flow->lock);
 196         flow->used = jiffies;
 197         flow->packet_count++;
 198         flow->byte_count += skb->len;
 199         flow->tcp_flags |= tcp_flags;
 200         spin_unlock(&flow->lock);
 201 }
 202
 203 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
 204 {
 205         struct sw_flow_actions *sfa;
 206
 207         if (size > MAX_ACTIONS_BUFSIZE)
 208                 return ERR_PTR(-EINVAL);
 209
 210         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 211         if (!sfa)
 212                 return ERR_PTR(-ENOMEM);
 213
 214         sfa->actions_len = 0;
 215         return sfa;
 216 }
 217
 218 struct sw_flow *ovs_flow_alloc(void)
 219 {
 220         struct sw_flow *flow;
 221
 222         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 223         if (!flow)
 224                 return ERR_PTR(-ENOMEM);
 225
 226         spin_lock_init(&flow->lock);
 227         flow->sf_acts = NULL;
 228
 229         return flow;
 230 }
 231
 232 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 233 {
 234         hash = jhash_1word(hash, table->hash_seed);
 235         return flex_array_get(table->buckets,
 236                                 (hash & (table->n_buckets - 1)));
 237 }
 238
 239 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 240 {
 241         struct flex_array *buckets;
 242         int i, err;
 243
 244         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 245                                    n_buckets, GFP_KERNEL);
 246         if (!buckets)
 247                 return NULL;
 248
 249         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 250         if (err) {
 251                 flex_array_free(buckets);
 252                 return NULL;
 253         }
 254
 255         for (i = 0; i < n_buckets; i++)
 256                 INIT_HLIST_HEAD((struct hlist_head *)
 257                                         flex_array_get(buckets, i));
 258
 259         return buckets;
 260 }
 261
 262 static void free_buckets(struct flex_array *buckets)
 263 {
 264         flex_array_free(buckets);
 265 }
 266
 267 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 268 {
 269         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 270
 271         if (!table)
 272                 return NULL;
 273
 274         table->buckets = alloc_buckets(new_size);
 275
 276         if (!table->buckets) {
 277                 kfree(table);
 278                 return NULL;
 279         }
 280         table->n_buckets = new_size;
 281         table->count = 0;
 282         table->node_ver = 0;
 283         table->keep_flows = false;
 284         get_random_bytes(&table->hash_seed, sizeof(u32));
 285
 286         return table;
 287 }
 288
 289 void ovs_flow_tbl_destroy(struct flow_table *table)
 290 {
 291         int i;
 292
 293         if (!table)
 294                 return;
 295
 296         if (table->keep_flows)
 297                 goto skip_flows;
 298
 299         for (i = 0; i < table->n_buckets; i++) {
 300                 struct sw_flow *flow;
 301                 struct hlist_head *head = flex_array_get(table->buckets, i);
 302                 struct hlist_node *node, *n;
 303                 int ver = table->node_ver;
 304
 305                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 306                         hlist_del_rcu(&flow->hash_node[ver]);
 307                         ovs_flow_free(flow);
 308                 }
 309         }
 310
 311 skip_flows:
 312         free_buckets(table->buckets);
 313         kfree(table);
 314 }
 315
 316 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 317 {
 318         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 319
 320         ovs_flow_tbl_destroy(table);
 321 }
 322
 323 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 324 {
 325         if (!table)
 326                 return;
 327
 328         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 329 }
 330
 331 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 332 {
 333         struct sw_flow *flow;
 334         struct hlist_head *head;
 335         struct hlist_node *n;
 336         int ver;
 337         int i;
 338
 339         ver = table->node_ver;
 340         while (*bucket < table->n_buckets) {
 341                 i = 0;
 342                 head = flex_array_get(table->buckets, *bucket);
 343                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 344                         if (i < *last) {
 345                                 i++;
 346                                 continue;
 347                         }
 348                         *last = i + 1;
 349                         return flow;
 350                 }
 351                 (*bucket)++;
 352                 *last = 0;
 353         }
 354
 355         return NULL;
 356 }
 357
 358 static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 359 {
 360         struct hlist_head *head;
 361         head = find_bucket(table, flow->hash);
 362         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 363         table->count++;
 364 }
 365
 366 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 367 {
 368         int old_ver;
 369         int i;
 370
 371         old_ver = old->node_ver;
 372         new->node_ver = !old_ver;
 373
 374         /* Insert in new table. */
 375         for (i = 0; i < old->n_buckets; i++) {
 376                 struct sw_flow *flow;
 377                 struct hlist_head *head;
 378                 struct hlist_node *n;
 379
 380                 head = flex_array_get(old->buckets, i);
 381
 382                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 383                         __flow_tbl_insert(new, flow);
 384         }
 385         old->keep_flows = true;
 386 }
 387
 388 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 389 {
 390         struct flow_table *new_table;
 391
 392         new_table = ovs_flow_tbl_alloc(n_buckets);
 393         if (!new_table)
 394                 return ERR_PTR(-ENOMEM);
 395
 396         flow_table_copy_flows(table, new_table);
 397
 398         return new_table;
 399 }
 400
 401 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 402 {
 403         return __flow_tbl_rehash(table, table->n_buckets);
 404 }
 405
 406 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 407 {
 408         return __flow_tbl_rehash(table, table->n_buckets * 2);
 409 }
 410
 411 void ovs_flow_free(struct sw_flow *flow)
 412 {
 413         if (unlikely(!flow))
 414                 return;
 415
 416         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 417         kmem_cache_free(flow_cache, flow);
 418 }
 419
 420 /* RCU callback used by ovs_flow_deferred_free. */
 421 static void rcu_free_flow_callback(struct rcu_head *rcu)
 422 {
 423         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 424
 425         ovs_flow_free(flow);
 426 }
 427
 428 /* Schedules 'flow' to be freed after the next RCU grace period.
 429  * The caller must hold rcu_read_lock for this to be sensible. */
 430 void ovs_flow_deferred_free(struct sw_flow *flow)
 431 {
 432         call_rcu(&flow->rcu, rcu_free_flow_callback);
 433 }
 434
 435 /* RCU callback used by ovs_flow_deferred_free_acts. */
 436 static void rcu_free_acts_callback(struct rcu_head *rcu)
 437 {
 438         struct sw_flow_actions *sf_acts = container_of(rcu,
 439                         struct sw_flow_actions, rcu);
 440         kfree(sf_acts);
 441 }
 442
 443 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 444  * The caller must hold rcu_read_lock for this to be sensible. */
 445 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 446 {
 447         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 448 }
 449
 450 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 451 {
 452         struct qtag_prefix {
 453                 __be16 eth_type; /* ETH_P_8021Q */
 454                 __be16 tci;
 455         };
 456         struct qtag_prefix *qp;
 457
 458         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 459                 return 0;
 460
 461         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 462                                          sizeof(__be16))))
 463                 return -ENOMEM;
 464
 465         qp = (struct qtag_prefix *) skb->data;
 466         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 467         __skb_pull(skb, sizeof(struct qtag_prefix));
 468
 469         return 0;
 470 }
 471
 472 static __be16 parse_ethertype(struct sk_buff *skb)
 473 {
 474         struct llc_snap_hdr {
 475                 u8  dsap;  /* Always 0xAA */
 476                 u8  ssap;  /* Always 0xAA */
 477                 u8  ctrl;
 478                 u8  oui[3];
 479                 __be16 ethertype;
 480         };
 481         struct llc_snap_hdr *llc;
 482         __be16 proto;
 483
 484         proto = *(__be16 *) skb->data;
 485         __skb_pull(skb, sizeof(__be16));
 486
 487         if (ntohs(proto) >= 1536)
 488                 return proto;
 489
 490         if (skb->len < sizeof(struct llc_snap_hdr))
 491                 return htons(ETH_P_802_2);
 492
 493         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 494                 return htons(0);
 495
 496         llc = (struct llc_snap_hdr *) skb->data;
 497         if (llc->dsap != LLC_SAP_SNAP ||
 498             llc->ssap != LLC_SAP_SNAP ||
 499             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 500                 return htons(ETH_P_802_2);
 501
 502         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 503
 504         if (ntohs(llc->ethertype) >= 1536)
 505                 return llc->ethertype;
 506
 507         return htons(ETH_P_802_2);
 508 }
 509
 510 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 511                         int *key_lenp, int nh_len)
 512 {
 513         struct icmp6hdr *icmp = icmp6_hdr(skb);
 514         int error = 0;
 515         int key_len;
 516
 517         /* The ICMPv6 type and code fields use the 16-bit transport port
 518          * fields, so we need to store them in 16-bit network byte order.
 519          */
 520         key->ipv6.tp.src = htons(icmp->icmp6_type);
 521         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 522         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 523
 524         if (icmp->icmp6_code == 0 &&
 525             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 526              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 527                 int icmp_len = skb->len - skb_transport_offset(skb);
 528                 struct nd_msg *nd;
 529                 int offset;
 530
 531                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 532
 533                 /* In order to process neighbor discovery options, we need the
 534                  * entire packet.
 535                  */
 536                 if (unlikely(icmp_len < sizeof(*nd)))
 537                         goto out;
 538                 if (unlikely(skb_linearize(skb))) {
 539                         error = -ENOMEM;
 540                         goto out;
 541                 }
 542
 543                 nd = (struct nd_msg *)skb_transport_header(skb);
 544                 key->ipv6.nd.target = nd->target;
 545                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 546
 547                 icmp_len -= sizeof(*nd);
 548                 offset = 0;
 549                 while (icmp_len >= 8) {
 550                         struct nd_opt_hdr *nd_opt =
 551                                  (struct nd_opt_hdr *)(nd->opt + offset);
 552                         int opt_len = nd_opt->nd_opt_len * 8;
 553
 554                         if (unlikely(!opt_len || opt_len > icmp_len))
 555                                 goto invalid;
 556
 557                         /* Store the link layer address if the appropriate
 558                          * option is provided.  It is considered an error if
 559                          * the same link layer option is specified twice.
 560                          */
 561                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 562                             && opt_len == 8) {
 563                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 564                                         goto invalid;
 565                                 memcpy(key->ipv6.nd.sll,
 566                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 567                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 568                                    && opt_len == 8) {
 569                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 570                                         goto invalid;
 571                                 memcpy(key->ipv6.nd.tll,
 572                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 573                         }
 574
 575                         icmp_len -= opt_len;
 576                         offset += opt_len;
 577                 }
 578         }
 579
 580         goto out;
 581
 582 invalid:
 583         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 584         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 585         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 586
 587 out:
 588         *key_lenp = key_len;
 589         return error;
 590 }
 591
 592 /**
 593  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 594  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 595  * Ethernet header
 596  * @in_port: port number on which @skb was received.
 597  * @key: output flow key
 598  * @key_lenp: length of output flow key
 599  *
 600  * The caller must ensure that skb->len >= ETH_HLEN.
 601  *
 602  * Returns 0 if successful, otherwise a negative errno value.
 603  *
 604  * Initializes @skb header pointers as follows:
 605  *
 606  *    - skb->mac_header: the Ethernet header.
 607  *
 608  *    - skb->network_header: just past the Ethernet header, or just past the
 609  *      VLAN header, to the first byte of the Ethernet payload.
 610  *
 611  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 612  *      on output, then just past the IP header, if one is present and
 613  *      of a correct length, otherwise the same as skb->network_header.
 614  *      For other key->dl_type values it is left untouched.
 615  */
 616 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 617                  int *key_lenp)
 618 {
 619         int error = 0;
 620         int key_len = SW_FLOW_KEY_OFFSET(eth);
 621         struct ethhdr *eth;
 622
 623         memset(key, 0, sizeof(*key));
 624
 625         key->phy.priority = skb->priority;
 626         if (OVS_CB(skb)->tun_key)
 627                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 628         key->phy.in_port = in_port;
 629         key->phy.skb_mark = skb_get_mark(skb);
 630
 631         skb_reset_mac_header(skb);
 632
 633         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 634          * header in the linear data area.
 635          */
 636         eth = eth_hdr(skb);
 637         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 638         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 639
 640         __skb_pull(skb, 2 * ETH_ALEN);
 641
 642         if (vlan_tx_tag_present(skb))
 643                 key->eth.tci = htons(vlan_get_tci(skb));
 644         else if (eth->h_proto == htons(ETH_P_8021Q))
 645                 if (unlikely(parse_vlan(skb, key)))
 646                         return -ENOMEM;
 647
 648         key->eth.type = parse_ethertype(skb);
 649         if (unlikely(key->eth.type == htons(0)))
 650                 return -ENOMEM;
 651
 652         skb_reset_network_header(skb);
 653         __skb_push(skb, skb->data - skb_mac_header(skb));
 654
 655         /* Network layer. */
 656         if (key->eth.type == htons(ETH_P_IP)) {
 657                 struct iphdr *nh;
 658                 __be16 offset;
 659
 660                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 661
 662                 error = check_iphdr(skb);
 663                 if (unlikely(error)) {
 664                         if (error == -EINVAL) {
 665                                 skb->transport_header = skb->network_header;
 666                                 error = 0;
 667                         }
 668                         goto out;
 669                 }
 670
 671                 nh = ip_hdr(skb);
 672                 key->ipv4.addr.src = nh->saddr;
 673                 key->ipv4.addr.dst = nh->daddr;
 674
 675                 key->ip.proto = nh->protocol;
 676                 key->ip.tos = nh->tos;
 677                 key->ip.ttl = nh->ttl;
 678
 679                 offset = nh->frag_off & htons(IP_OFFSET);
 680                 if (offset) {
 681                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 682                         goto out;
 683                 }
 684                 if (nh->frag_off & htons(IP_MF) ||
 685                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 686                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 687
 688                 /* Transport layer. */
 689                 if (key->ip.proto == IPPROTO_TCP) {
 690                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 691                         if (tcphdr_ok(skb)) {
 692                                 struct tcphdr *tcp = tcp_hdr(skb);
 693                                 key->ipv4.tp.src = tcp->source;
 694                                 key->ipv4.tp.dst = tcp->dest;
 695                         }
 696                 } else if (key->ip.proto == IPPROTO_UDP) {
 697                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 698                         if (udphdr_ok(skb)) {
 699                                 struct udphdr *udp = udp_hdr(skb);
 700                                 key->ipv4.tp.src = udp->source;
 701                                 key->ipv4.tp.dst = udp->dest;
 702                         }
 703                 } else if (key->ip.proto == IPPROTO_ICMP) {
 704                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 705                         if (icmphdr_ok(skb)) {
 706                                 struct icmphdr *icmp = icmp_hdr(skb);
 707                                 /* The ICMP type and code fields use the 16-bit
 708                                  * transport port fields, so we need to store
 709                                  * them in 16-bit network byte order. */
 710                                 key->ipv4.tp.src = htons(icmp->type);
 711                                 key->ipv4.tp.dst = htons(icmp->code);
 712                         }
 713                 }
 714
 715         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 716                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 717                 struct arp_eth_header *arp;
 718
 719                 arp = (struct arp_eth_header *)skb_network_header(skb);
 720
 721                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 722                                 && arp->ar_pro == htons(ETH_P_IP)
 723                                 && arp->ar_hln == ETH_ALEN
 724                                 && arp->ar_pln == 4) {
 725
 726                         /* We only match on the lower 8 bits of the opcode. */
 727                         if (ntohs(arp->ar_op) <= 0xff)
 728                                 key->ip.proto = ntohs(arp->ar_op);
 729                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 730                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 731                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 732                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 733                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 734                 }
 735         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 736                 int nh_len;             /* IPv6 Header + Extensions */
 737
 738                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 739                 if (unlikely(nh_len < 0)) {
 740                         if (nh_len == -EINVAL)
 741                                 skb->transport_header = skb->network_header;
 742                         else
 743                                 error = nh_len;
 744                         goto out;
 745                 }
 746
 747                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 748                         goto out;
 749                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 750                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 751
 752                 /* Transport layer. */
 753                 if (key->ip.proto == NEXTHDR_TCP) {
 754                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 755                         if (tcphdr_ok(skb)) {
 756                                 struct tcphdr *tcp = tcp_hdr(skb);
 757                                 key->ipv6.tp.src = tcp->source;
 758                                 key->ipv6.tp.dst = tcp->dest;
 759                         }
 760                 } else if (key->ip.proto == NEXTHDR_UDP) {
 761                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 762                         if (udphdr_ok(skb)) {
 763                                 struct udphdr *udp = udp_hdr(skb);
 764                                 key->ipv6.tp.src = udp->source;
 765                                 key->ipv6.tp.dst = udp->dest;
 766                         }
 767                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 768                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 769                         if (icmp6hdr_ok(skb)) {
 770                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 771                                 if (error < 0)
 772                                         goto out;
 773                         }
 774                 }
 775         }
 776
 777 out:
 778         *key_lenp = key_len;
 779         return error;
 780 }
 781
 782 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 783 {
 784         return jhash2((u32 *)((u8 *)key + key_start),
 785                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 786 }
 787
 788 static int flow_key_start(struct sw_flow_key *key)
 789 {
 790         if (key->tun_key.ipv4_dst)
 791                 return 0;
 792         else
 793                 return offsetof(struct sw_flow_key, phy);
 794 }
 795
 796 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 797                                 struct sw_flow_key *key, int key_len)
 798 {
 799         struct sw_flow *flow;
 800         struct hlist_node *n;
 801         struct hlist_head *head;
 802         u8 *_key;
 803         int key_start;
 804         u32 hash;
 805
 806         key_start = flow_key_start(key);
 807         hash = ovs_flow_hash(key, key_start, key_len);
 808
 809         _key = (u8 *) key + key_start;
 810         head = find_bucket(table, hash);
 811         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 812
 813                 if (flow->hash == hash &&
 814                     !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
 815                         return flow;
 816                 }
 817         }
 818         return NULL;
 819 }
 820
 821 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 822                          struct sw_flow_key *key, int key_len)
 823 {
 824         flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
 825         memcpy(&flow->key, key, sizeof(flow->key));
 826         __flow_tbl_insert(table, flow);
 827 }
 828
 829 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 830 {
 831         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 832         table->count--;
 833         BUG_ON(table->count < 0);
 834 }
 835
 836 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 837 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 838         [OVS_KEY_ATTR_ENCAP] = -1,
 839         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 840         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 841         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 842         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 843         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 844         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 845         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 846         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 847         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 848         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 849         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 850         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 851         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 852         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 853         [OVS_KEY_ATTR_TUNNEL] = -1,
 854 };
 855
 856 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 857                                   const struct nlattr *a[], u64 *attrs)
 858 {
 859         const struct ovs_key_icmp *icmp_key;
 860         const struct ovs_key_tcp *tcp_key;
 861         const struct ovs_key_udp *udp_key;
 862
 863         switch (swkey->ip.proto) {
 864         case IPPROTO_TCP:
 865                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 866                         return -EINVAL;
 867                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 868
 869                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 870                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 871                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 872                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 873                 break;
 874
 875         case IPPROTO_UDP:
 876                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 877                         return -EINVAL;
 878                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 879
 880                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 881                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 882                 swkey->ipv4.tp.src = udp_key->udp_src;
 883                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 884                 break;
 885
 886         case IPPROTO_ICMP:
 887                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 888                         return -EINVAL;
 889                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 890
 891                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 892                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 893                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 894                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 895                 break;
 896         }
 897
 898         return 0;
 899 }
 900
 901 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 902                                   const struct nlattr *a[], u64 *attrs)
 903 {
 904         const struct ovs_key_icmpv6 *icmpv6_key;
 905         const struct ovs_key_tcp *tcp_key;
 906         const struct ovs_key_udp *udp_key;
 907
 908         switch (swkey->ip.proto) {
 909         case IPPROTO_TCP:
 910                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 911                         return -EINVAL;
 912                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 913
 914                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 915                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 916                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 917                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 918                 break;
 919
 920         case IPPROTO_UDP:
 921                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 922                         return -EINVAL;
 923                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 924
 925                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 926                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 927                 swkey->ipv6.tp.src = udp_key->udp_src;
 928                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 929                 break;
 930
 931         case IPPROTO_ICMPV6:
 932                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 933                         return -EINVAL;
 934                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 935
 936                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 937                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 938                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 939                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 940
 941                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 942                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 943                         const struct ovs_key_nd *nd_key;
 944
 945                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 946                                 return -EINVAL;
 947                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 948
 949                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 950                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 951                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 952                                sizeof(swkey->ipv6.nd.target));
 953                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 954                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 955                 }
 956                 break;
 957         }
 958
 959         return 0;
 960 }
 961
 962 static int parse_flow_nlattrs(const struct nlattr *attr,
 963                               const struct nlattr *a[], u64 *attrsp)
 964 {
 965         const struct nlattr *nla;
 966         u64 attrs;
 967         int rem;
 968
 969         attrs = 0;
 970         nla_for_each_nested(nla, attr, rem) {
 971                 u16 type = nla_type(nla);
 972                 int expected_len;
 973
 974                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 975                         return -EINVAL;
 976
 977                 expected_len = ovs_key_lens[type];
 978                 if (nla_len(nla) != expected_len && expected_len != -1)
 979                         return -EINVAL;
 980
 981                 attrs |= 1ULL << type;
 982                 a[type] = nla;
 983         }
 984         if (rem)
 985                 return -EINVAL;
 986
 987         *attrsp = attrs;
 988         return 0;
 989 }
 990
 991 int ipv4_tun_from_nlattr(const struct nlattr *attr,
 992                          struct ovs_key_ipv4_tunnel *tun_key)
 993 {
 994         struct nlattr *a;
 995         int rem;
 996         bool ttl = false;
 997
 998         memset(tun_key, 0, sizeof(*tun_key));
 999
1000         nla_for_each_nested(a, attr, rem) {
1001                 int type = nla_type(a);
1002                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1003                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1004                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1005                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1006                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1007                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1008                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1009                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1010                 };
1011
1012                 if (type > OVS_TUNNEL_KEY_ATTR_MAX ||
1013                         ovs_tunnel_key_lens[type] != nla_len(a))
1014                         return -EINVAL;
1015
1016                 switch (type) {
1017                 case OVS_TUNNEL_KEY_ATTR_ID:
1018                         tun_key->tun_id = nla_get_be64(a);
1019                         tun_key->tun_flags |= OVS_TNL_F_KEY;
1020                         break;
1021                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1022                         tun_key->ipv4_src = nla_get_be32(a);
1023                         break;
1024                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1025                         tun_key->ipv4_dst = nla_get_be32(a);
1026                         break;
1027                 case OVS_TUNNEL_KEY_ATTR_TOS:
1028                         tun_key->ipv4_tos = nla_get_u8(a);
1029                         break;
1030                 case OVS_TUNNEL_KEY_ATTR_TTL:
1031                         tun_key->ipv4_ttl = nla_get_u8(a);
1032                         ttl = true;
1033                         break;
1034                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1035                         tun_key->tun_flags |= OVS_TNL_F_DONT_FRAGMENT;
1036                         break;
1037                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1038                         tun_key->tun_flags |= OVS_TNL_F_CSUM;
1039                         break;
1040                 default:
1041                         return -EINVAL;
1042
1043                 }
1044         }
1045         if (rem > 0)
1046                 return -EINVAL;
1047
1048         if (!tun_key->ipv4_dst)
1049                 return -EINVAL;
1050
1051         if (!ttl)
1052                 return -EINVAL;
1053
1054         return 0;
1055 }
1056
1057 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1058                         const struct ovs_key_ipv4_tunnel *tun_key)
1059 {
1060         struct nlattr *nla;
1061
1062         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1063         if (!nla)
1064                 return -EMSGSIZE;
1065
1066         if (tun_key->tun_flags & OVS_TNL_F_KEY &&
1067             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id))
1068                 return -EMSGSIZE;
1069         if (tun_key->ipv4_src &&
1070             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ipv4_src))
1071                 return -EMSGSIZE;
1072         if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ipv4_dst))
1073                 return -EMSGSIZE;
1074         if (tun_key->ipv4_tos &&
1075             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ipv4_tos))
1076                 return -EMSGSIZE;
1077         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ipv4_ttl))
1078                 return -EMSGSIZE;
1079         if ((tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT) &&
1080                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1081                 return -EMSGSIZE;
1082         if ((tun_key->tun_flags & OVS_TNL_F_CSUM) &&
1083                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1084                 return -EMSGSIZE;
1085
1086         nla_nest_end(skb, nla);
1087         return 0;
1088 }
1089
1090 /**
1091  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
1092  * @swkey: receives the extracted flow key.
1093  * @key_lenp: number of bytes used in @swkey.
1094  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1095  * sequence.
1096  */
1097 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1098                       const struct nlattr *attr)
1099 {
1100         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1101         const struct ovs_key_ethernet *eth_key;
1102         int key_len;
1103         u64 attrs;
1104         int err;
1105
1106         memset(swkey, 0, sizeof(struct sw_flow_key));
1107         key_len = SW_FLOW_KEY_OFFSET(eth);
1108
1109         err = parse_flow_nlattrs(attr, a, &attrs);
1110         if (err)
1111                 return err;
1112
1113         /* Metadata attributes. */
1114         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1115                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1116                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1117         }
1118         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1119                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1120                 if (in_port >= DP_MAX_PORTS)
1121                         return -EINVAL;
1122                 swkey->phy.in_port = in_port;
1123                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1124         } else {
1125                 swkey->phy.in_port = DP_MAX_PORTS;
1126         }
1127         if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1128                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1129 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1130                 if (mark != 0)
1131                         return -EINVAL;
1132 #endif
1133                 swkey->phy.skb_mark = mark;
1134                 attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1135         }
1136
1137         if (attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1138                 err = ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1139                 if (err)
1140                         return err;
1141
1142                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1143         }
1144
1145         /* Data attributes. */
1146         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1147                 return -EINVAL;
1148         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1149
1150         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1151         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1152         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1153
1154         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1155             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1156                 const struct nlattr *encap;
1157                 __be16 tci;
1158
1159                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1160                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1161                               (1 << OVS_KEY_ATTR_ENCAP)))
1162                         return -EINVAL;
1163
1164                 encap = a[OVS_KEY_ATTR_ENCAP];
1165                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1166                 if (tci & htons(VLAN_TAG_PRESENT)) {
1167                         swkey->eth.tci = tci;
1168
1169                         err = parse_flow_nlattrs(encap, a, &attrs);
1170                         if (err)
1171                                 return err;
1172                 } else if (!tci) {
1173                         /* Corner case for truncated 802.1Q header. */
1174                         if (nla_len(encap))
1175                                 return -EINVAL;
1176
1177                         swkey->eth.type = htons(ETH_P_8021Q);
1178                         *key_lenp = key_len;
1179                         return 0;
1180                 } else {
1181                         return -EINVAL;
1182                 }
1183         }
1184
1185         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1186                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1187                 if (ntohs(swkey->eth.type) < 1536)
1188                         return -EINVAL;
1189                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1190         } else {
1191                 swkey->eth.type = htons(ETH_P_802_2);
1192         }
1193
1194         if (swkey->eth.type == htons(ETH_P_IP)) {
1195                 const struct ovs_key_ipv4 *ipv4_key;
1196
1197                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1198                         return -EINVAL;
1199                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1200
1201                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1202                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1203                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1204                         return -EINVAL;
1205                 swkey->ip.proto = ipv4_key->ipv4_proto;
1206                 swkey->ip.tos = ipv4_key->ipv4_tos;
1207                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1208                 swkey->ip.frag = ipv4_key->ipv4_frag;
1209                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1210                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1211
1212                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1213                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1214                         if (err)
1215                                 return err;
1216                 }
1217         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1218                 const struct ovs_key_ipv6 *ipv6_key;
1219
1220                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1221                         return -EINVAL;
1222                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1223
1224                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1225                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1226                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1227                         return -EINVAL;
1228                 swkey->ipv6.label = ipv6_key->ipv6_label;
1229                 swkey->ip.proto = ipv6_key->ipv6_proto;
1230                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1231                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1232                 swkey->ip.frag = ipv6_key->ipv6_frag;
1233                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1234                        sizeof(swkey->ipv6.addr.src));
1235                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1236                        sizeof(swkey->ipv6.addr.dst));
1237
1238                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1239                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1240                         if (err)
1241                                 return err;
1242                 }
1243         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1244                    swkey->eth.type == htons(ETH_P_RARP)) {
1245                 const struct ovs_key_arp *arp_key;
1246
1247                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1248                         return -EINVAL;
1249                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1250
1251                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1252                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1253                 swkey->ipv4.addr.src = arp_key->arp_sip;
1254                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1255                 if (arp_key->arp_op & htons(0xff00))
1256                         return -EINVAL;
1257                 swkey->ip.proto = ntohs(arp_key->arp_op);
1258                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1259                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1260         }
1261
1262         if (attrs)
1263                 return -EINVAL;
1264         *key_lenp = key_len;
1265
1266         return 0;
1267 }
1268
1269 /**
1270  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1271  * @in_port: receives the extracted input port.
1272  * @tun_id: receives the extracted tunnel ID.
1273  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1274  * sequence.
1275  *
1276  * This parses a series of Netlink attributes that form a flow key, which must
1277  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1278  * get the metadata, that is, the parts of the flow key that cannot be
1279  * extracted from the packet itself.
1280  */
1281
1282 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
1283 {
1284         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1285         const struct nlattr *nla;
1286         int rem;
1287
1288         flow->key.phy.in_port = DP_MAX_PORTS;
1289         flow->key.phy.priority = 0;
1290         flow->key.phy.skb_mark = 0;
1291         memset(tun_key, 0, sizeof(flow->key.tun_key));
1292
1293         nla_for_each_nested(nla, attr, rem) {
1294                 int type = nla_type(nla);
1295
1296                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1297                         int err;
1298
1299                         if (nla_len(nla) != ovs_key_lens[type])
1300                                 return -EINVAL;
1301
1302                         switch (type) {
1303                         case OVS_KEY_ATTR_PRIORITY:
1304                                 flow->key.phy.priority = nla_get_u32(nla);
1305                                 break;
1306
1307                         case OVS_KEY_ATTR_TUNNEL:
1308                                 err = ipv4_tun_from_nlattr(nla, tun_key);
1309                                 if (err)
1310                                         return err;
1311                                 break;
1312
1313                         case OVS_KEY_ATTR_IN_PORT:
1314                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1315                                         return -EINVAL;
1316                                 flow->key.phy.in_port = nla_get_u32(nla);
1317                                 break;
1318
1319                         case OVS_KEY_ATTR_SKB_MARK:
1320 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1321                                 if (nla_get_u32(nla) != 0)
1322                                         return -EINVAL;
1323 #endif
1324                                 flow->key.phy.skb_mark = nla_get_u32(nla);
1325                                 break;
1326                         }
1327                 }
1328         }
1329         if (rem)
1330                 return -EINVAL;
1331
1332         flow->hash = ovs_flow_hash(&flow->key,
1333                                    flow_key_start(&flow->key), key_len);
1334
1335         return 0;
1336 }
1337
1338 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1339 {
1340         struct ovs_key_ethernet *eth_key;
1341         struct nlattr *nla, *encap;
1342
1343         if (swkey->phy.priority &&
1344             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1345                 goto nla_put_failure;
1346
1347         if (swkey->tun_key.ipv4_dst &&
1348             ipv4_tun_to_nlattr(skb, &swkey->tun_key))
1349                 goto nla_put_failure;
1350
1351         if (swkey->phy.in_port != DP_MAX_PORTS &&
1352             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1353                 goto nla_put_failure;
1354
1355         if (swkey->phy.skb_mark &&
1356             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1357                 goto nla_put_failure;
1358
1359         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1360         if (!nla)
1361                 goto nla_put_failure;
1362         eth_key = nla_data(nla);
1363         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1364         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1365
1366         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1367                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1368                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1369                         goto nla_put_failure;
1370                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1371                 if (!swkey->eth.tci)
1372                         goto unencap;
1373         } else {
1374                 encap = NULL;
1375         }
1376
1377         if (swkey->eth.type == htons(ETH_P_802_2))
1378                 goto unencap;
1379
1380         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1381                 goto nla_put_failure;
1382
1383         if (swkey->eth.type == htons(ETH_P_IP)) {
1384                 struct ovs_key_ipv4 *ipv4_key;
1385
1386                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1387                 if (!nla)
1388                         goto nla_put_failure;
1389                 ipv4_key = nla_data(nla);
1390                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1391                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1392                 ipv4_key->ipv4_proto = swkey->ip.proto;
1393                 ipv4_key->ipv4_tos = swkey->ip.tos;
1394                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1395                 ipv4_key->ipv4_frag = swkey->ip.frag;
1396         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1397                 struct ovs_key_ipv6 *ipv6_key;
1398
1399                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1400                 if (!nla)
1401                         goto nla_put_failure;
1402                 ipv6_key = nla_data(nla);
1403                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1404                                 sizeof(ipv6_key->ipv6_src));
1405                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1406                                 sizeof(ipv6_key->ipv6_dst));
1407                 ipv6_key->ipv6_label = swkey->ipv6.label;
1408                 ipv6_key->ipv6_proto = swkey->ip.proto;
1409                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1410                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1411                 ipv6_key->ipv6_frag = swkey->ip.frag;
1412         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1413                    swkey->eth.type == htons(ETH_P_RARP)) {
1414                 struct ovs_key_arp *arp_key;
1415
1416                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1417                 if (!nla)
1418                         goto nla_put_failure;
1419                 arp_key = nla_data(nla);
1420                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1421                 arp_key->arp_sip = swkey->ipv4.addr.src;
1422                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1423                 arp_key->arp_op = htons(swkey->ip.proto);
1424                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1425                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1426         }
1427
1428         if ((swkey->eth.type == htons(ETH_P_IP) ||
1429              swkey->eth.type == htons(ETH_P_IPV6)) &&
1430              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1431
1432                 if (swkey->ip.proto == IPPROTO_TCP) {
1433                         struct ovs_key_tcp *tcp_key;
1434
1435                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1436                         if (!nla)
1437                                 goto nla_put_failure;
1438                         tcp_key = nla_data(nla);
1439                         if (swkey->eth.type == htons(ETH_P_IP)) {
1440                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1441                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1442                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1443                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1444                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1445                         }
1446                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1447                         struct ovs_key_udp *udp_key;
1448
1449                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1450                         if (!nla)
1451                                 goto nla_put_failure;
1452                         udp_key = nla_data(nla);
1453                         if (swkey->eth.type == htons(ETH_P_IP)) {
1454                                 udp_key->udp_src = swkey->ipv4.tp.src;
1455                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1456                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1457                                 udp_key->udp_src = swkey->ipv6.tp.src;
1458                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1459                         }
1460                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1461                            swkey->ip.proto == IPPROTO_ICMP) {
1462                         struct ovs_key_icmp *icmp_key;
1463
1464                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1465                         if (!nla)
1466                                 goto nla_put_failure;
1467                         icmp_key = nla_data(nla);
1468                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1469                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1470                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1471                            swkey->ip.proto == IPPROTO_ICMPV6) {
1472                         struct ovs_key_icmpv6 *icmpv6_key;
1473
1474                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1475                                                 sizeof(*icmpv6_key));
1476                         if (!nla)
1477                                 goto nla_put_failure;
1478                         icmpv6_key = nla_data(nla);
1479                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1480                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1481
1482                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1483                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1484                                 struct ovs_key_nd *nd_key;
1485
1486                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1487                                 if (!nla)
1488                                         goto nla_put_failure;
1489                                 nd_key = nla_data(nla);
1490                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1491                                                         sizeof(nd_key->nd_target));
1492                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1493                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1494                         }
1495                 }
1496         }
1497
1498 unencap:
1499         if (encap)
1500                 nla_nest_end(skb, encap);
1501
1502         return 0;
1503
1504 nla_put_failure:
1505         return -EMSGSIZE;
1506 }
1507
1508 /* Initializes the flow module.
1509  * Returns zero if successful or a negative error code. */
1510 int ovs_flow_init(void)
1511 {
1512         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1513                                         0, NULL);
1514         if (flow_cache == NULL)
1515                 return -ENOMEM;
1516
1517         return 0;
1518 }
1519
1520 /* Uninitializes the flow module. */
1521 void ovs_flow_exit(void)
1522 {
1523         kmem_cache_destroy(flow_cache);
1524 }