datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #include "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 #include "vlan.h"
  47
  48 static struct kmem_cache *flow_cache;
  49
  50 static int check_header(struct sk_buff *skb, int len)
  51 {
  52         if (unlikely(skb->len < len))
  53                 return -EINVAL;
  54         if (unlikely(!pskb_may_pull(skb, len)))
  55                 return -ENOMEM;
  56         return 0;
  57 }
  58
  59 static bool arphdr_ok(struct sk_buff *skb)
  60 {
  61         return pskb_may_pull(skb, skb_network_offset(skb) +
  62                                   sizeof(struct arp_eth_header));
  63 }
  64
  65 static int check_iphdr(struct sk_buff *skb)
  66 {
  67         unsigned int nh_ofs = skb_network_offset(skb);
  68         unsigned int ip_len;
  69         int err;
  70
  71         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  72         if (unlikely(err))
  73                 return err;
  74
  75         ip_len = ip_hdrlen(skb);
  76         if (unlikely(ip_len < sizeof(struct iphdr) ||
  77                      skb->len < nh_ofs + ip_len))
  78                 return -EINVAL;
  79
  80         skb_set_transport_header(skb, nh_ofs + ip_len);
  81         return 0;
  82 }
  83
  84 static bool tcphdr_ok(struct sk_buff *skb)
  85 {
  86         int th_ofs = skb_transport_offset(skb);
  87         int tcp_len;
  88
  89         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  90                 return false;
  91
  92         tcp_len = tcp_hdrlen(skb);
  93         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  94                      skb->len < th_ofs + tcp_len))
  95                 return false;
  96
  97         return true;
  98 }
  99
 100 static bool udphdr_ok(struct sk_buff *skb)
 101 {
 102         return pskb_may_pull(skb, skb_transport_offset(skb) +
 103                                   sizeof(struct udphdr));
 104 }
 105
 106 static bool icmphdr_ok(struct sk_buff *skb)
 107 {
 108         return pskb_may_pull(skb, skb_transport_offset(skb) +
 109                                   sizeof(struct icmphdr));
 110 }
 111
 112 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 113 {
 114         struct timespec cur_ts;
 115         u64 cur_ms, idle_ms;
 116
 117         ktime_get_ts(&cur_ts);
 118         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 119         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 120                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 121
 122         return cur_ms - idle_ms;
 123 }
 124
 125 #define SW_FLOW_KEY_OFFSET(field)               \
 126         (offsetof(struct sw_flow_key, field) +  \
 127          FIELD_SIZEOF(struct sw_flow_key, field))
 128
 129 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 130                          int *key_lenp)
 131 {
 132         unsigned int nh_ofs = skb_network_offset(skb);
 133         unsigned int nh_len;
 134         int payload_ofs;
 135         struct ipv6hdr *nh;
 136         uint8_t nexthdr;
 137         __be16 frag_off;
 138         int err;
 139
 140         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 141
 142         err = check_header(skb, nh_ofs + sizeof(*nh));
 143         if (unlikely(err))
 144                 return err;
 145
 146         nh = ipv6_hdr(skb);
 147         nexthdr = nh->nexthdr;
 148         payload_ofs = (u8 *)(nh + 1) - skb->data;
 149
 150         key->ip.proto = NEXTHDR_NONE;
 151         key->ip.tos = ipv6_get_dsfield(nh);
 152         key->ip.ttl = nh->hop_limit;
 153         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 154         key->ipv6.addr.src = nh->saddr;
 155         key->ipv6.addr.dst = nh->daddr;
 156
 157         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 158         if (unlikely(payload_ofs < 0))
 159                 return -EINVAL;
 160
 161         if (frag_off) {
 162                 if (frag_off & htons(~0x7))
 163                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 164                 else
 165                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 166         }
 167
 168         nh_len = payload_ofs - nh_ofs;
 169         skb_set_transport_header(skb, nh_ofs + nh_len);
 170         key->ip.proto = nexthdr;
 171         return nh_len;
 172 }
 173
 174 static bool icmp6hdr_ok(struct sk_buff *skb)
 175 {
 176         return pskb_may_pull(skb, skb_transport_offset(skb) +
 177                                   sizeof(struct icmp6hdr));
 178 }
 179
 180 #define TCP_FLAGS_OFFSET 13
 181 #define TCP_FLAG_MASK 0x3f
 182
 183 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 184 {
 185         u8 tcp_flags = 0;
 186
 187         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 188              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 189             flow->key.ip.proto == IPPROTO_TCP &&
 190             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 191                 u8 *tcp = (u8 *)tcp_hdr(skb);
 192                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 193         }
 194
 195         spin_lock(&flow->lock);
 196         flow->used = jiffies;
 197         flow->packet_count++;
 198         flow->byte_count += skb->len;
 199         flow->tcp_flags |= tcp_flags;
 200         spin_unlock(&flow->lock);
 201 }
 202
 203 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 204 {
 205         int actions_len = nla_len(actions);
 206         struct sw_flow_actions *sfa;
 207
 208         if (actions_len > MAX_ACTIONS_BUFSIZE)
 209                 return ERR_PTR(-EINVAL);
 210
 211         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 212         if (!sfa)
 213                 return ERR_PTR(-ENOMEM);
 214
 215         sfa->actions_len = actions_len;
 216         memcpy(sfa->actions, nla_data(actions), actions_len);
 217         return sfa;
 218 }
 219
 220 struct sw_flow *ovs_flow_alloc(void)
 221 {
 222         struct sw_flow *flow;
 223
 224         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 225         if (!flow)
 226                 return ERR_PTR(-ENOMEM);
 227
 228         spin_lock_init(&flow->lock);
 229         atomic_set(&flow->refcnt, 1);
 230         flow->sf_acts = NULL;
 231         flow->dead = false;
 232
 233         return flow;
 234 }
 235
 236 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 237 {
 238         hash = jhash_1word(hash, table->hash_seed);
 239         return flex_array_get(table->buckets,
 240                                 (hash & (table->n_buckets - 1)));
 241 }
 242
 243 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 244 {
 245         struct flex_array *buckets;
 246         int i, err;
 247
 248         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 249                                    n_buckets, GFP_KERNEL);
 250         if (!buckets)
 251                 return NULL;
 252
 253         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 254         if (err) {
 255                 flex_array_free(buckets);
 256                 return NULL;
 257         }
 258
 259         for (i = 0; i < n_buckets; i++)
 260                 INIT_HLIST_HEAD((struct hlist_head *)
 261                                         flex_array_get(buckets, i));
 262
 263         return buckets;
 264 }
 265
 266 static void free_buckets(struct flex_array *buckets)
 267 {
 268         flex_array_free(buckets);
 269 }
 270
 271 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 272 {
 273         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 274
 275         if (!table)
 276                 return NULL;
 277
 278         table->buckets = alloc_buckets(new_size);
 279
 280         if (!table->buckets) {
 281                 kfree(table);
 282                 return NULL;
 283         }
 284         table->n_buckets = new_size;
 285         table->count = 0;
 286         table->node_ver = 0;
 287         table->keep_flows = false;
 288         get_random_bytes(&table->hash_seed, sizeof(u32));
 289
 290         return table;
 291 }
 292
 293 static void flow_free(struct sw_flow *flow)
 294 {
 295         flow->dead = true;
 296         ovs_flow_put(flow);
 297 }
 298
 299 void ovs_flow_tbl_destroy(struct flow_table *table)
 300 {
 301         int i;
 302
 303         if (!table)
 304                 return;
 305
 306         if (table->keep_flows)
 307                 goto skip_flows;
 308
 309         for (i = 0; i < table->n_buckets; i++) {
 310                 struct sw_flow *flow;
 311                 struct hlist_head *head = flex_array_get(table->buckets, i);
 312                 struct hlist_node *node, *n;
 313                 int ver = table->node_ver;
 314
 315                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 316                         hlist_del_rcu(&flow->hash_node[ver]);
 317                         flow_free(flow);
 318                 }
 319         }
 320
 321 skip_flows:
 322         free_buckets(table->buckets);
 323         kfree(table);
 324 }
 325
 326 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 327 {
 328         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 329
 330         ovs_flow_tbl_destroy(table);
 331 }
 332
 333 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 334 {
 335         if (!table)
 336                 return;
 337
 338         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 339 }
 340
 341 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 342 {
 343         struct sw_flow *flow;
 344         struct hlist_head *head;
 345         struct hlist_node *n;
 346         int ver;
 347         int i;
 348
 349         ver = table->node_ver;
 350         while (*bucket < table->n_buckets) {
 351                 i = 0;
 352                 head = flex_array_get(table->buckets, *bucket);
 353                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 354                         if (i < *last) {
 355                                 i++;
 356                                 continue;
 357                         }
 358                         *last = i + 1;
 359                         return flow;
 360                 }
 361                 (*bucket)++;
 362                 *last = 0;
 363         }
 364
 365         return NULL;
 366 }
 367
 368 static void flow_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                         ovs_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 /* RCU callback used by ovs_flow_deferred_free. */
 414 static void rcu_free_flow_callback(struct rcu_head *rcu)
 415 {
 416         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 417
 418         flow->dead = true;
 419         ovs_flow_put(flow);
 420 }
 421
 422 /* Schedules 'flow' to be freed after the next RCU grace period.
 423  * The caller must hold rcu_read_lock for this to be sensible. */
 424 void ovs_flow_deferred_free(struct sw_flow *flow)
 425 {
 426         call_rcu(&flow->rcu, rcu_free_flow_callback);
 427 }
 428
 429 void ovs_flow_hold(struct sw_flow *flow)
 430 {
 431         atomic_inc(&flow->refcnt);
 432 }
 433
 434 void ovs_flow_put(struct sw_flow *flow)
 435 {
 436         if (unlikely(!flow))
 437                 return;
 438
 439         if (atomic_dec_and_test(&flow->refcnt)) {
 440                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 441                 kmem_cache_free(flow_cache, flow);
 442         }
 443 }
 444
 445 /* RCU callback used by ovs_flow_deferred_free_acts. */
 446 static void rcu_free_acts_callback(struct rcu_head *rcu)
 447 {
 448         struct sw_flow_actions *sf_acts = container_of(rcu,
 449                         struct sw_flow_actions, rcu);
 450         kfree(sf_acts);
 451 }
 452
 453 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 454  * The caller must hold rcu_read_lock for this to be sensible. */
 455 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 456 {
 457         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 458 }
 459
 460 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 461 {
 462         struct qtag_prefix {
 463                 __be16 eth_type; /* ETH_P_8021Q */
 464                 __be16 tci;
 465         };
 466         struct qtag_prefix *qp;
 467
 468         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 469                 return 0;
 470
 471         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 472                                          sizeof(__be16))))
 473                 return -ENOMEM;
 474
 475         qp = (struct qtag_prefix *) skb->data;
 476         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 477         __skb_pull(skb, sizeof(struct qtag_prefix));
 478
 479         return 0;
 480 }
 481
 482 static __be16 parse_ethertype(struct sk_buff *skb)
 483 {
 484         struct llc_snap_hdr {
 485                 u8  dsap;  /* Always 0xAA */
 486                 u8  ssap;  /* Always 0xAA */
 487                 u8  ctrl;
 488                 u8  oui[3];
 489                 __be16 ethertype;
 490         };
 491         struct llc_snap_hdr *llc;
 492         __be16 proto;
 493
 494         proto = *(__be16 *) skb->data;
 495         __skb_pull(skb, sizeof(__be16));
 496
 497         if (ntohs(proto) >= 1536)
 498                 return proto;
 499
 500         if (skb->len < sizeof(struct llc_snap_hdr))
 501                 return htons(ETH_P_802_2);
 502
 503         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 504                 return htons(0);
 505
 506         llc = (struct llc_snap_hdr *) skb->data;
 507         if (llc->dsap != LLC_SAP_SNAP ||
 508             llc->ssap != LLC_SAP_SNAP ||
 509             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 510                 return htons(ETH_P_802_2);
 511
 512         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 513         return llc->ethertype;
 514 }
 515
 516 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 517                         int *key_lenp, int nh_len)
 518 {
 519         struct icmp6hdr *icmp = icmp6_hdr(skb);
 520         int error = 0;
 521         int key_len;
 522
 523         /* The ICMPv6 type and code fields use the 16-bit transport port
 524          * fields, so we need to store them in 16-bit network byte order.
 525          */
 526         key->ipv6.tp.src = htons(icmp->icmp6_type);
 527         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 528         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 529
 530         if (icmp->icmp6_code == 0 &&
 531             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 532              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 533                 int icmp_len = skb->len - skb_transport_offset(skb);
 534                 struct nd_msg *nd;
 535                 int offset;
 536
 537                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 538
 539                 /* In order to process neighbor discovery options, we need the
 540                  * entire packet.
 541                  */
 542                 if (unlikely(icmp_len < sizeof(*nd)))
 543                         goto out;
 544                 if (unlikely(skb_linearize(skb))) {
 545                         error = -ENOMEM;
 546                         goto out;
 547                 }
 548
 549                 nd = (struct nd_msg *)skb_transport_header(skb);
 550                 key->ipv6.nd.target = nd->target;
 551                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 552
 553                 icmp_len -= sizeof(*nd);
 554                 offset = 0;
 555                 while (icmp_len >= 8) {
 556                         struct nd_opt_hdr *nd_opt =
 557                                  (struct nd_opt_hdr *)(nd->opt + offset);
 558                         int opt_len = nd_opt->nd_opt_len * 8;
 559
 560                         if (unlikely(!opt_len || opt_len > icmp_len))
 561                                 goto invalid;
 562
 563                         /* Store the link layer address if the appropriate
 564                          * option is provided.  It is considered an error if
 565                          * the same link layer option is specified twice.
 566                          */
 567                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 568                             && opt_len == 8) {
 569                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 570                                         goto invalid;
 571                                 memcpy(key->ipv6.nd.sll,
 572                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 573                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 574                                    && opt_len == 8) {
 575                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 576                                         goto invalid;
 577                                 memcpy(key->ipv6.nd.tll,
 578                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 579                         }
 580
 581                         icmp_len -= opt_len;
 582                         offset += opt_len;
 583                 }
 584         }
 585
 586         goto out;
 587
 588 invalid:
 589         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 590         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 591         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 592
 593 out:
 594         *key_lenp = key_len;
 595         return error;
 596 }
 597
 598 /**
 599  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 600  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 601  * Ethernet header
 602  * @in_port: port number on which @skb was received.
 603  * @key: output flow key
 604  * @key_lenp: length of output flow key
 605  *
 606  * The caller must ensure that skb->len >= ETH_HLEN.
 607  *
 608  * Returns 0 if successful, otherwise a negative errno value.
 609  *
 610  * Initializes @skb header pointers as follows:
 611  *
 612  *    - skb->mac_header: the Ethernet header.
 613  *
 614  *    - skb->network_header: just past the Ethernet header, or just past the
 615  *      VLAN header, to the first byte of the Ethernet payload.
 616  *
 617  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 618  *      on output, then just past the IP header, if one is present and
 619  *      of a correct length, otherwise the same as skb->network_header.
 620  *      For other key->dl_type values it is left untouched.
 621  */
 622 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 623                  int *key_lenp)
 624 {
 625         int error = 0;
 626         int key_len = SW_FLOW_KEY_OFFSET(eth);
 627         struct ethhdr *eth;
 628
 629         memset(key, 0, sizeof(*key));
 630
 631         key->phy.priority = skb->priority;
 632         if (OVS_CB(skb)->tun_key)
 633                 memcpy(&key->tun.tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun.tun_key));
 634         key->phy.in_port = in_port;
 635
 636         skb_reset_mac_header(skb);
 637
 638         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 639          * header in the linear data area.
 640          */
 641         eth = eth_hdr(skb);
 642         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 643         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 644
 645         __skb_pull(skb, 2 * ETH_ALEN);
 646
 647         if (vlan_tx_tag_present(skb))
 648                 key->eth.tci = htons(vlan_get_tci(skb));
 649         else if (eth->h_proto == htons(ETH_P_8021Q))
 650                 if (unlikely(parse_vlan(skb, key)))
 651                         return -ENOMEM;
 652
 653         key->eth.type = parse_ethertype(skb);
 654         if (unlikely(key->eth.type == htons(0)))
 655                 return -ENOMEM;
 656
 657         skb_reset_network_header(skb);
 658         __skb_push(skb, skb->data - skb_mac_header(skb));
 659
 660         /* Network layer. */
 661         if (key->eth.type == htons(ETH_P_IP)) {
 662                 struct iphdr *nh;
 663                 __be16 offset;
 664
 665                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 666
 667                 error = check_iphdr(skb);
 668                 if (unlikely(error)) {
 669                         if (error == -EINVAL) {
 670                                 skb->transport_header = skb->network_header;
 671                                 error = 0;
 672                         }
 673                         goto out;
 674                 }
 675
 676                 nh = ip_hdr(skb);
 677                 key->ipv4.addr.src = nh->saddr;
 678                 key->ipv4.addr.dst = nh->daddr;
 679
 680                 key->ip.proto = nh->protocol;
 681                 key->ip.tos = nh->tos;
 682                 key->ip.ttl = nh->ttl;
 683
 684                 offset = nh->frag_off & htons(IP_OFFSET);
 685                 if (offset) {
 686                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 687                         goto out;
 688                 }
 689                 if (nh->frag_off & htons(IP_MF) ||
 690                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 691                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 692
 693                 /* Transport layer. */
 694                 if (key->ip.proto == IPPROTO_TCP) {
 695                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 696                         if (tcphdr_ok(skb)) {
 697                                 struct tcphdr *tcp = tcp_hdr(skb);
 698                                 key->ipv4.tp.src = tcp->source;
 699                                 key->ipv4.tp.dst = tcp->dest;
 700                         }
 701                 } else if (key->ip.proto == IPPROTO_UDP) {
 702                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 703                         if (udphdr_ok(skb)) {
 704                                 struct udphdr *udp = udp_hdr(skb);
 705                                 key->ipv4.tp.src = udp->source;
 706                                 key->ipv4.tp.dst = udp->dest;
 707                         }
 708                 } else if (key->ip.proto == IPPROTO_ICMP) {
 709                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 710                         if (icmphdr_ok(skb)) {
 711                                 struct icmphdr *icmp = icmp_hdr(skb);
 712                                 /* The ICMP type and code fields use the 16-bit
 713                                  * transport port fields, so we need to store
 714                                  * them in 16-bit network byte order. */
 715                                 key->ipv4.tp.src = htons(icmp->type);
 716                                 key->ipv4.tp.dst = htons(icmp->code);
 717                         }
 718                 }
 719
 720         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 721                 struct arp_eth_header *arp;
 722
 723                 arp = (struct arp_eth_header *)skb_network_header(skb);
 724
 725                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 726                                 && arp->ar_pro == htons(ETH_P_IP)
 727                                 && arp->ar_hln == ETH_ALEN
 728                                 && arp->ar_pln == 4) {
 729
 730                         /* We only match on the lower 8 bits of the opcode. */
 731                         if (ntohs(arp->ar_op) <= 0xff)
 732                                 key->ip.proto = ntohs(arp->ar_op);
 733
 734                         if (key->ip.proto == ARPOP_REQUEST
 735                                         || key->ip.proto == ARPOP_REPLY) {
 736                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 737                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 738                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 739                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 740                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 741                         }
 742                 }
 743         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 744                 int nh_len;             /* IPv6 Header + Extensions */
 745
 746                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 747                 if (unlikely(nh_len < 0)) {
 748                         if (nh_len == -EINVAL)
 749                                 skb->transport_header = skb->network_header;
 750                         else
 751                                 error = nh_len;
 752                         goto out;
 753                 }
 754
 755                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 756                         goto out;
 757                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 758                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 759
 760                 /* Transport layer. */
 761                 if (key->ip.proto == NEXTHDR_TCP) {
 762                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 763                         if (tcphdr_ok(skb)) {
 764                                 struct tcphdr *tcp = tcp_hdr(skb);
 765                                 key->ipv6.tp.src = tcp->source;
 766                                 key->ipv6.tp.dst = tcp->dest;
 767                         }
 768                 } else if (key->ip.proto == NEXTHDR_UDP) {
 769                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 770                         if (udphdr_ok(skb)) {
 771                                 struct udphdr *udp = udp_hdr(skb);
 772                                 key->ipv6.tp.src = udp->source;
 773                                 key->ipv6.tp.dst = udp->dest;
 774                         }
 775                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 776                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 777                         if (icmp6hdr_ok(skb)) {
 778                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 779                                 if (error < 0)
 780                                         goto out;
 781                         }
 782                 }
 783         }
 784
 785 out:
 786         *key_lenp = key_len;
 787         return error;
 788 }
 789
 790 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 791 {
 792         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 793 }
 794
 795 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 796                                 struct sw_flow_key *key, int key_len)
 797 {
 798         struct sw_flow *flow;
 799         struct hlist_node *n;
 800         struct hlist_head *head;
 801         u32 hash;
 802
 803         hash = ovs_flow_hash(key, key_len);
 804
 805         head = find_bucket(table, hash);
 806         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 807
 808                 if (flow->hash == hash &&
 809                     !memcmp(&flow->key, key, key_len)) {
 810                         return flow;
 811                 }
 812         }
 813         return NULL;
 814 }
 815
 816 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 817 {
 818         struct hlist_head *head;
 819
 820         head = find_bucket(table, flow->hash);
 821         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 822         table->count++;
 823 }
 824
 825 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 826 {
 827         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 828         table->count--;
 829         BUG_ON(table->count < 0);
 830 }
 831
 832 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 833 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 834         [OVS_KEY_ATTR_ENCAP] = -1,
 835         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 836         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 837         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 838         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 839         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 840         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 841         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 842         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 843         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 844         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 845         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 846         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 847         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 848
 849         /* Not upstream. */
 850         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 851         [OVS_KEY_ATTR_IPV4_TUNNEL] = sizeof(struct ovs_key_ipv4_tunnel),
 852 };
 853
 854 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 855                                   const struct nlattr *a[], u64 *attrs)
 856 {
 857         const struct ovs_key_icmp *icmp_key;
 858         const struct ovs_key_tcp *tcp_key;
 859         const struct ovs_key_udp *udp_key;
 860
 861         switch (swkey->ip.proto) {
 862         case IPPROTO_TCP:
 863                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 864                         return -EINVAL;
 865                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 866
 867                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 868                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 869                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 870                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 871                 break;
 872
 873         case IPPROTO_UDP:
 874                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 875                         return -EINVAL;
 876                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 877
 878                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 879                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 880                 swkey->ipv4.tp.src = udp_key->udp_src;
 881                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 882                 break;
 883
 884         case IPPROTO_ICMP:
 885                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 886                         return -EINVAL;
 887                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 888
 889                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 890                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 891                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 892                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 893                 break;
 894         }
 895
 896         return 0;
 897 }
 898
 899 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 900                                   const struct nlattr *a[], u64 *attrs)
 901 {
 902         const struct ovs_key_icmpv6 *icmpv6_key;
 903         const struct ovs_key_tcp *tcp_key;
 904         const struct ovs_key_udp *udp_key;
 905
 906         switch (swkey->ip.proto) {
 907         case IPPROTO_TCP:
 908                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 909                         return -EINVAL;
 910                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 911
 912                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 913                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 914                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 915                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 916                 break;
 917
 918         case IPPROTO_UDP:
 919                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 920                         return -EINVAL;
 921                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 922
 923                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 924                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 925                 swkey->ipv6.tp.src = udp_key->udp_src;
 926                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 927                 break;
 928
 929         case IPPROTO_ICMPV6:
 930                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 931                         return -EINVAL;
 932                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 933
 934                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 935                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 936                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 937                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 938
 939                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 940                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 941                         const struct ovs_key_nd *nd_key;
 942
 943                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 944                                 return -EINVAL;
 945                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 946
 947                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 948                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 949                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 950                                sizeof(swkey->ipv6.nd.target));
 951                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 952                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 953                 }
 954                 break;
 955         }
 956
 957         return 0;
 958 }
 959
 960 static int parse_flow_nlattrs(const struct nlattr *attr,
 961                               const struct nlattr *a[], u64 *attrsp)
 962 {
 963         const struct nlattr *nla;
 964         u64 attrs;
 965         int rem;
 966
 967         attrs = 0;
 968         nla_for_each_nested(nla, attr, rem) {
 969                 u16 type = nla_type(nla);
 970                 int expected_len;
 971
 972                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 973                         return -EINVAL;
 974
 975                 expected_len = ovs_key_lens[type];
 976                 if (nla_len(nla) != expected_len && expected_len != -1)
 977                         return -EINVAL;
 978
 979                 attrs |= 1ULL << type;
 980                 a[type] = nla;
 981         }
 982         if (rem)
 983                 return -EINVAL;
 984
 985         *attrsp = attrs;
 986         return 0;
 987 }
 988
 989 /**
 990  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 991  * @swkey: receives the extracted flow key.
 992  * @key_lenp: number of bytes used in @swkey.
 993  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 994  * sequence.
 995  */
 996 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 997                       const struct nlattr *attr)
 998 {
 999         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1000         const struct ovs_key_ethernet *eth_key;
1001         int key_len;
1002         u64 attrs;
1003         int err;
1004
1005         memset(swkey, 0, sizeof(struct sw_flow_key));
1006         key_len = SW_FLOW_KEY_OFFSET(eth);
1007
1008         err = parse_flow_nlattrs(attr, a, &attrs);
1009         if (err)
1010                 return err;
1011
1012         /* Metadata attributes. */
1013         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1014                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1015                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1016         }
1017         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1018                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1019                 if (in_port >= DP_MAX_PORTS)
1020                         return -EINVAL;
1021                 swkey->phy.in_port = in_port;
1022                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1023         } else {
1024                 swkey->phy.in_port = DP_MAX_PORTS;
1025         }
1026
1027         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID) &&
1028             attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1029                 struct ovs_key_ipv4_tunnel *tun_key;
1030                 __be64 tun_id;
1031
1032                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1033
1034                 if (!tun_key->ipv4_dst)
1035                         return -EINVAL;
1036                 if (!(tun_key->tun_flags & OVS_FLOW_TNL_F_KEY))
1037                         return -EINVAL;
1038
1039                 tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1040                 if (tun_id != tun_key->tun_id)
1041                         return -EINVAL;
1042
1043                 memcpy(&swkey->tun.tun_key, tun_key, sizeof(swkey->tun.tun_key));
1044                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1045                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1046         } else if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
1047                 swkey->tun.tun_key.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1048                 swkey->tun.tun_key.tun_flags |= OVS_FLOW_TNL_F_KEY;
1049
1050                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1051         } else if (attrs & (1ULL << OVS_KEY_ATTR_IPV4_TUNNEL)) {
1052                 struct ovs_key_ipv4_tunnel *tun_key;
1053                 tun_key = nla_data(a[OVS_KEY_ATTR_IPV4_TUNNEL]);
1054
1055                 if (!tun_key->ipv4_dst)
1056                         return -EINVAL;
1057
1058                 memcpy(&swkey->tun.tun_key, tun_key, sizeof(swkey->tun.tun_key));
1059                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4_TUNNEL);
1060         }
1061
1062         /* Data attributes. */
1063         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1064                 return -EINVAL;
1065         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1066
1067         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1068         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1069         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1070
1071         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1072             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1073                 const struct nlattr *encap;
1074                 __be16 tci;
1075
1076                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1077                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1078                               (1 << OVS_KEY_ATTR_ENCAP)))
1079                         return -EINVAL;
1080
1081                 encap = a[OVS_KEY_ATTR_ENCAP];
1082                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1083                 if (tci & htons(VLAN_TAG_PRESENT)) {
1084                         swkey->eth.tci = tci;
1085
1086                         err = parse_flow_nlattrs(encap, a, &attrs);
1087                         if (err)
1088                                 return err;
1089                 } else if (!tci) {
1090                         /* Corner case for truncated 802.1Q header. */
1091                         if (nla_len(encap))
1092                                 return -EINVAL;
1093
1094                         swkey->eth.type = htons(ETH_P_8021Q);
1095                         *key_lenp = key_len;
1096                         return 0;
1097                 } else {
1098                         return -EINVAL;
1099                 }
1100         }
1101
1102         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1103                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1104                 if (ntohs(swkey->eth.type) < 1536)
1105                         return -EINVAL;
1106                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1107         } else {
1108                 swkey->eth.type = htons(ETH_P_802_2);
1109         }
1110
1111         if (swkey->eth.type == htons(ETH_P_IP)) {
1112                 const struct ovs_key_ipv4 *ipv4_key;
1113
1114                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1115                         return -EINVAL;
1116                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1117
1118                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1119                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1120                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1121                         return -EINVAL;
1122                 swkey->ip.proto = ipv4_key->ipv4_proto;
1123                 swkey->ip.tos = ipv4_key->ipv4_tos;
1124                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1125                 swkey->ip.frag = ipv4_key->ipv4_frag;
1126                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1127                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1128
1129                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1130                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1131                         if (err)
1132                                 return err;
1133                 }
1134         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1135                 const struct ovs_key_ipv6 *ipv6_key;
1136
1137                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1138                         return -EINVAL;
1139                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1140
1141                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1142                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1143                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1144                         return -EINVAL;
1145                 swkey->ipv6.label = ipv6_key->ipv6_label;
1146                 swkey->ip.proto = ipv6_key->ipv6_proto;
1147                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1148                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1149                 swkey->ip.frag = ipv6_key->ipv6_frag;
1150                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1151                        sizeof(swkey->ipv6.addr.src));
1152                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1153                        sizeof(swkey->ipv6.addr.dst));
1154
1155                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1156                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1157                         if (err)
1158                                 return err;
1159                 }
1160         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1161                 const struct ovs_key_arp *arp_key;
1162
1163                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1164                         return -EINVAL;
1165                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1166
1167                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1168                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1169                 swkey->ipv4.addr.src = arp_key->arp_sip;
1170                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1171                 if (arp_key->arp_op & htons(0xff00))
1172                         return -EINVAL;
1173                 swkey->ip.proto = ntohs(arp_key->arp_op);
1174                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1175                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1176         }
1177
1178         if (attrs)
1179                 return -EINVAL;
1180         *key_lenp = key_len;
1181
1182         return 0;
1183 }
1184
1185 /**
1186  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1187  * @in_port: receives the extracted input port.
1188  * @tun_id: receives the extracted tunnel ID.
1189  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1190  * sequence.
1191  *
1192  * This parses a series of Netlink attributes that form a flow key, which must
1193  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1194  * get the metadata, that is, the parts of the flow key that cannot be
1195  * extracted from the packet itself.
1196  */
1197 int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port,
1198                                    struct ovs_key_ipv4_tunnel *tun_key,
1199                                    const struct nlattr *attr)
1200 {
1201         const struct nlattr *nla;
1202         int rem;
1203         __be64 tun_id;
1204
1205         *in_port = DP_MAX_PORTS;
1206         memset(tun_key, 0, sizeof(*tun_key));
1207         *priority = 0;
1208
1209         nla_for_each_nested(nla, attr, rem) {
1210                 int type = nla_type(nla);
1211
1212                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1213                         if (nla_len(nla) != ovs_key_lens[type])
1214                                 return -EINVAL;
1215
1216                         switch (type) {
1217                         case OVS_KEY_ATTR_PRIORITY:
1218                                 *priority = nla_get_u32(nla);
1219                                 break;
1220
1221                         case OVS_KEY_ATTR_TUN_ID:
1222                                 tun_id = nla_get_be64(nla);
1223
1224                                 if (tun_key->ipv4_dst) {
1225                                         if (!(tun_key->tun_flags & OVS_FLOW_TNL_F_KEY))
1226                                                 return -EINVAL;
1227                                         if (tun_key->tun_id != tun_id)
1228                                                 return -EINVAL;
1229                                         break;
1230                                 }
1231                                 tun_key->tun_id = tun_id;
1232                                 tun_key->tun_flags |= OVS_FLOW_TNL_F_KEY;
1233
1234                                 break;
1235
1236                         case OVS_KEY_ATTR_IPV4_TUNNEL:
1237                                 if (tun_key->tun_flags & OVS_FLOW_TNL_F_KEY) {
1238                                         tun_id = tun_key->tun_id;
1239
1240                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1241                                         if (!(tun_key->tun_flags & OVS_FLOW_TNL_F_KEY))
1242                                                 return -EINVAL;
1243
1244                                         if (tun_key->tun_id != tun_id)
1245                                                 return -EINVAL;
1246                                 } else
1247                                         memcpy(tun_key, nla_data(nla), sizeof(*tun_key));
1248
1249                                 if (!tun_key->ipv4_dst)
1250                                         return -EINVAL;
1251                                 break;
1252
1253                         case OVS_KEY_ATTR_IN_PORT:
1254                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1255                                         return -EINVAL;
1256                                 *in_port = nla_get_u32(nla);
1257                                 break;
1258                         }
1259                 }
1260         }
1261         if (rem)
1262                 return -EINVAL;
1263         return 0;
1264 }
1265
1266 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1267 {
1268         struct ovs_key_ethernet *eth_key;
1269         struct nlattr *nla, *encap;
1270
1271         if (swkey->phy.priority &&
1272             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1273                 goto nla_put_failure;
1274
1275         if (swkey->tun.tun_key.ipv4_dst) {
1276                 struct ovs_key_ipv4_tunnel *tun_key;
1277                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4_TUNNEL, sizeof(*tun_key));
1278                 if (!nla)
1279                         goto nla_put_failure;
1280                 tun_key = nla_data(nla);
1281                 memcpy(tun_key, &swkey->tun.tun_key, sizeof(*tun_key));
1282         }
1283         if ((swkey->tun.tun_key.tun_flags & OVS_FLOW_TNL_F_KEY) &&
1284             nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->tun.tun_key.tun_id))
1285                 goto nla_put_failure;
1286
1287         if (swkey->phy.in_port != DP_MAX_PORTS &&
1288             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1289                 goto nla_put_failure;
1290
1291         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1292         if (!nla)
1293                 goto nla_put_failure;
1294         eth_key = nla_data(nla);
1295         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1296         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1297
1298         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1299                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1300                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1301                         goto nla_put_failure;
1302                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1303                 if (!swkey->eth.tci)
1304                         goto unencap;
1305         } else {
1306                 encap = NULL;
1307         }
1308
1309         if (swkey->eth.type == htons(ETH_P_802_2))
1310                 goto unencap;
1311
1312         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1313                 goto nla_put_failure;
1314
1315         if (swkey->eth.type == htons(ETH_P_IP)) {
1316                 struct ovs_key_ipv4 *ipv4_key;
1317
1318                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1319                 if (!nla)
1320                         goto nla_put_failure;
1321                 ipv4_key = nla_data(nla);
1322                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1323                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1324                 ipv4_key->ipv4_proto = swkey->ip.proto;
1325                 ipv4_key->ipv4_tos = swkey->ip.tos;
1326                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1327                 ipv4_key->ipv4_frag = swkey->ip.frag;
1328         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1329                 struct ovs_key_ipv6 *ipv6_key;
1330
1331                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1332                 if (!nla)
1333                         goto nla_put_failure;
1334                 ipv6_key = nla_data(nla);
1335                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1336                                 sizeof(ipv6_key->ipv6_src));
1337                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1338                                 sizeof(ipv6_key->ipv6_dst));
1339                 ipv6_key->ipv6_label = swkey->ipv6.label;
1340                 ipv6_key->ipv6_proto = swkey->ip.proto;
1341                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1342                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1343                 ipv6_key->ipv6_frag = swkey->ip.frag;
1344         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1345                 struct ovs_key_arp *arp_key;
1346
1347                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1348                 if (!nla)
1349                         goto nla_put_failure;
1350                 arp_key = nla_data(nla);
1351                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1352                 arp_key->arp_sip = swkey->ipv4.addr.src;
1353                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1354                 arp_key->arp_op = htons(swkey->ip.proto);
1355                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1356                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1357         }
1358
1359         if ((swkey->eth.type == htons(ETH_P_IP) ||
1360              swkey->eth.type == htons(ETH_P_IPV6)) &&
1361              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1362
1363                 if (swkey->ip.proto == IPPROTO_TCP) {
1364                         struct ovs_key_tcp *tcp_key;
1365
1366                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1367                         if (!nla)
1368                                 goto nla_put_failure;
1369                         tcp_key = nla_data(nla);
1370                         if (swkey->eth.type == htons(ETH_P_IP)) {
1371                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1372                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1373                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1374                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1375                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1376                         }
1377                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1378                         struct ovs_key_udp *udp_key;
1379
1380                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1381                         if (!nla)
1382                                 goto nla_put_failure;
1383                         udp_key = nla_data(nla);
1384                         if (swkey->eth.type == htons(ETH_P_IP)) {
1385                                 udp_key->udp_src = swkey->ipv4.tp.src;
1386                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1387                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1388                                 udp_key->udp_src = swkey->ipv6.tp.src;
1389                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1390                         }
1391                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1392                            swkey->ip.proto == IPPROTO_ICMP) {
1393                         struct ovs_key_icmp *icmp_key;
1394
1395                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1396                         if (!nla)
1397                                 goto nla_put_failure;
1398                         icmp_key = nla_data(nla);
1399                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1400                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1401                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1402                            swkey->ip.proto == IPPROTO_ICMPV6) {
1403                         struct ovs_key_icmpv6 *icmpv6_key;
1404
1405                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1406                                                 sizeof(*icmpv6_key));
1407                         if (!nla)
1408                                 goto nla_put_failure;
1409                         icmpv6_key = nla_data(nla);
1410                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1411                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1412
1413                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1414                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1415                                 struct ovs_key_nd *nd_key;
1416
1417                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1418                                 if (!nla)
1419                                         goto nla_put_failure;
1420                                 nd_key = nla_data(nla);
1421                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1422                                                         sizeof(nd_key->nd_target));
1423                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1424                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1425                         }
1426                 }
1427         }
1428
1429 unencap:
1430         if (encap)
1431                 nla_nest_end(skb, encap);
1432
1433         return 0;
1434
1435 nla_put_failure:
1436         return -EMSGSIZE;
1437 }
1438
1439 /* Initializes the flow module.
1440  * Returns zero if successful or a negative error code. */
1441 int ovs_flow_init(void)
1442 {
1443         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1444                                         0, NULL);
1445         if (flow_cache == NULL)
1446                 return -ENOMEM;
1447
1448         return 0;
1449 }
1450
1451 /* Uninitializes the flow module. */
1452 void ovs_flow_exit(void)
1453 {
1454         kmem_cache_destroy(flow_cache);
1455 }