datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira Networks.
   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.ip.proto == IPPROTO_TCP) {
 189                 u8 *tcp = (u8 *)tcp_hdr(skb);
 190                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 191         }
 192
 193         spin_lock(&flow->lock);
 194         flow->used = jiffies;
 195         flow->packet_count++;
 196         flow->byte_count += skb->len;
 197         flow->tcp_flags |= tcp_flags;
 198         spin_unlock(&flow->lock);
 199 }
 200
 201 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 202 {
 203         int actions_len = nla_len(actions);
 204         struct sw_flow_actions *sfa;
 205
 206         if (actions_len > MAX_ACTIONS_BUFSIZE)
 207                 return ERR_PTR(-EINVAL);
 208
 209         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 210         if (!sfa)
 211                 return ERR_PTR(-ENOMEM);
 212
 213         sfa->actions_len = actions_len;
 214         memcpy(sfa->actions, nla_data(actions), actions_len);
 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         atomic_set(&flow->refcnt, 1);
 228         flow->sf_acts = NULL;
 229         flow->dead = false;
 230
 231         return flow;
 232 }
 233
 234 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 235 {
 236         hash = jhash_1word(hash, table->hash_seed);
 237         return flex_array_get(table->buckets,
 238                                 (hash & (table->n_buckets - 1)));
 239 }
 240
 241 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 242 {
 243         struct flex_array *buckets;
 244         int i, err;
 245
 246         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 247                                    n_buckets, GFP_KERNEL);
 248         if (!buckets)
 249                 return NULL;
 250
 251         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 252         if (err) {
 253                 flex_array_free(buckets);
 254                 return NULL;
 255         }
 256
 257         for (i = 0; i < n_buckets; i++)
 258                 INIT_HLIST_HEAD((struct hlist_head *)
 259                                         flex_array_get(buckets, i));
 260
 261         return buckets;
 262 }
 263
 264 static void free_buckets(struct flex_array *buckets)
 265 {
 266         flex_array_free(buckets);
 267 }
 268
 269 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 270 {
 271         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 272
 273         if (!table)
 274                 return NULL;
 275
 276         table->buckets = alloc_buckets(new_size);
 277
 278         if (!table->buckets) {
 279                 kfree(table);
 280                 return NULL;
 281         }
 282         table->n_buckets = new_size;
 283         table->count = 0;
 284         table->node_ver = 0;
 285         table->keep_flows = false;
 286         get_random_bytes(&table->hash_seed, sizeof(u32));
 287
 288         return table;
 289 }
 290
 291 static void flow_free(struct sw_flow *flow)
 292 {
 293         flow->dead = true;
 294         ovs_flow_put(flow);
 295 }
 296
 297 void ovs_flow_tbl_destroy(struct flow_table *table)
 298 {
 299         int i;
 300
 301         if (!table)
 302                 return;
 303
 304         if (table->keep_flows)
 305                 goto skip_flows;
 306
 307         for (i = 0; i < table->n_buckets; i++) {
 308                 struct sw_flow *flow;
 309                 struct hlist_head *head = flex_array_get(table->buckets, i);
 310                 struct hlist_node *node, *n;
 311                 int ver = table->node_ver;
 312
 313                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 314                         hlist_del_rcu(&flow->hash_node[ver]);
 315                         flow_free(flow);
 316                 }
 317         }
 318
 319 skip_flows:
 320         free_buckets(table->buckets);
 321         kfree(table);
 322 }
 323
 324 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 325 {
 326         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 327
 328         ovs_flow_tbl_destroy(table);
 329 }
 330
 331 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 332 {
 333         if (!table)
 334                 return;
 335
 336         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 337 }
 338
 339 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 340 {
 341         struct sw_flow *flow;
 342         struct hlist_head *head;
 343         struct hlist_node *n;
 344         int ver;
 345         int i;
 346
 347         ver = table->node_ver;
 348         while (*bucket < table->n_buckets) {
 349                 i = 0;
 350                 head = flex_array_get(table->buckets, *bucket);
 351                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 352                         if (i < *last) {
 353                                 i++;
 354                                 continue;
 355                         }
 356                         *last = i + 1;
 357                         return flow;
 358                 }
 359                 (*bucket)++;
 360                 *last = 0;
 361         }
 362
 363         return NULL;
 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                         ovs_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 /* RCU callback used by ovs_flow_deferred_free. */
 412 static void rcu_free_flow_callback(struct rcu_head *rcu)
 413 {
 414         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 415
 416         flow->dead = true;
 417         ovs_flow_put(flow);
 418 }
 419
 420 /* Schedules 'flow' to be freed after the next RCU grace period.
 421  * The caller must hold rcu_read_lock for this to be sensible. */
 422 void ovs_flow_deferred_free(struct sw_flow *flow)
 423 {
 424         call_rcu(&flow->rcu, rcu_free_flow_callback);
 425 }
 426
 427 void ovs_flow_hold(struct sw_flow *flow)
 428 {
 429         atomic_inc(&flow->refcnt);
 430 }
 431
 432 void ovs_flow_put(struct sw_flow *flow)
 433 {
 434         if (unlikely(!flow))
 435                 return;
 436
 437         if (atomic_dec_and_test(&flow->refcnt)) {
 438                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 439                 kmem_cache_free(flow_cache, flow);
 440         }
 441 }
 442
 443 /* RCU callback used by ovs_flow_deferred_free_acts. */
 444 static void rcu_free_acts_callback(struct rcu_head *rcu)
 445 {
 446         struct sw_flow_actions *sf_acts = container_of(rcu,
 447                         struct sw_flow_actions, rcu);
 448         kfree(sf_acts);
 449 }
 450
 451 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 452  * The caller must hold rcu_read_lock for this to be sensible. */
 453 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 454 {
 455         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 456 }
 457
 458 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 459 {
 460         struct qtag_prefix {
 461                 __be16 eth_type; /* ETH_P_8021Q */
 462                 __be16 tci;
 463         };
 464         struct qtag_prefix *qp;
 465
 466         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 467                 return 0;
 468
 469         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 470                                          sizeof(__be16))))
 471                 return -ENOMEM;
 472
 473         qp = (struct qtag_prefix *) skb->data;
 474         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 475         __skb_pull(skb, sizeof(struct qtag_prefix));
 476
 477         return 0;
 478 }
 479
 480 static __be16 parse_ethertype(struct sk_buff *skb)
 481 {
 482         struct llc_snap_hdr {
 483                 u8  dsap;  /* Always 0xAA */
 484                 u8  ssap;  /* Always 0xAA */
 485                 u8  ctrl;
 486                 u8  oui[3];
 487                 __be16 ethertype;
 488         };
 489         struct llc_snap_hdr *llc;
 490         __be16 proto;
 491
 492         proto = *(__be16 *) skb->data;
 493         __skb_pull(skb, sizeof(__be16));
 494
 495         if (ntohs(proto) >= 1536)
 496                 return proto;
 497
 498         if (skb->len < sizeof(struct llc_snap_hdr))
 499                 return htons(ETH_P_802_2);
 500
 501         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 502                 return htons(0);
 503
 504         llc = (struct llc_snap_hdr *) skb->data;
 505         if (llc->dsap != LLC_SAP_SNAP ||
 506             llc->ssap != LLC_SAP_SNAP ||
 507             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 508                 return htons(ETH_P_802_2);
 509
 510         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 511         return llc->ethertype;
 512 }
 513
 514 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 515                         int *key_lenp, int nh_len)
 516 {
 517         struct icmp6hdr *icmp = icmp6_hdr(skb);
 518         int error = 0;
 519         int key_len;
 520
 521         /* The ICMPv6 type and code fields use the 16-bit transport port
 522          * fields, so we need to store them in 16-bit network byte order.
 523          */
 524         key->ipv6.tp.src = htons(icmp->icmp6_type);
 525         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 526         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 527
 528         if (icmp->icmp6_code == 0 &&
 529             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 530              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 531                 int icmp_len = skb->len - skb_transport_offset(skb);
 532                 struct nd_msg *nd;
 533                 int offset;
 534
 535                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 536
 537                 /* In order to process neighbor discovery options, we need the
 538                  * entire packet.
 539                  */
 540                 if (unlikely(icmp_len < sizeof(*nd)))
 541                         goto out;
 542                 if (unlikely(skb_linearize(skb))) {
 543                         error = -ENOMEM;
 544                         goto out;
 545                 }
 546
 547                 nd = (struct nd_msg *)skb_transport_header(skb);
 548                 key->ipv6.nd.target = nd->target;
 549                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 550
 551                 icmp_len -= sizeof(*nd);
 552                 offset = 0;
 553                 while (icmp_len >= 8) {
 554                         struct nd_opt_hdr *nd_opt =
 555                                  (struct nd_opt_hdr *)(nd->opt + offset);
 556                         int opt_len = nd_opt->nd_opt_len * 8;
 557
 558                         if (unlikely(!opt_len || opt_len > icmp_len))
 559                                 goto invalid;
 560
 561                         /* Store the link layer address if the appropriate
 562                          * option is provided.  It is considered an error if
 563                          * the same link layer option is specified twice.
 564                          */
 565                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 566                             && opt_len == 8) {
 567                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 568                                         goto invalid;
 569                                 memcpy(key->ipv6.nd.sll,
 570                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 571                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 572                                    && opt_len == 8) {
 573                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 574                                         goto invalid;
 575                                 memcpy(key->ipv6.nd.tll,
 576                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 577                         }
 578
 579                         icmp_len -= opt_len;
 580                         offset += opt_len;
 581                 }
 582         }
 583
 584         goto out;
 585
 586 invalid:
 587         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 588         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 589         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 590
 591 out:
 592         *key_lenp = key_len;
 593         return error;
 594 }
 595
 596 /**
 597  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 598  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 599  * Ethernet header
 600  * @in_port: port number on which @skb was received.
 601  * @key: output flow key
 602  * @key_lenp: length of output flow key
 603  *
 604  * The caller must ensure that skb->len >= ETH_HLEN.
 605  *
 606  * Returns 0 if successful, otherwise a negative errno value.
 607  *
 608  * Initializes @skb header pointers as follows:
 609  *
 610  *    - skb->mac_header: the Ethernet header.
 611  *
 612  *    - skb->network_header: just past the Ethernet header, or just past the
 613  *      VLAN header, to the first byte of the Ethernet payload.
 614  *
 615  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 616  *      on output, then just past the IP header, if one is present and
 617  *      of a correct length, otherwise the same as skb->network_header.
 618  *      For other key->dl_type values it is left untouched.
 619  */
 620 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 621                  int *key_lenp)
 622 {
 623         int error = 0;
 624         int key_len = SW_FLOW_KEY_OFFSET(eth);
 625         struct ethhdr *eth;
 626
 627         memset(key, 0, sizeof(*key));
 628
 629         key->phy.priority = skb->priority;
 630         key->phy.tun_id = OVS_CB(skb)->tun_id;
 631         key->phy.in_port = in_port;
 632
 633         skb_reset_mac_header(skb);
 634
 635         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 636          * header in the linear data area.
 637          */
 638         eth = eth_hdr(skb);
 639         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 640         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 641
 642         __skb_pull(skb, 2 * ETH_ALEN);
 643
 644         if (vlan_tx_tag_present(skb))
 645                 key->eth.tci = htons(vlan_get_tci(skb));
 646         else if (eth->h_proto == htons(ETH_P_8021Q))
 647                 if (unlikely(parse_vlan(skb, key)))
 648                         return -ENOMEM;
 649
 650         key->eth.type = parse_ethertype(skb);
 651         if (unlikely(key->eth.type == htons(0)))
 652                 return -ENOMEM;
 653
 654         skb_reset_network_header(skb);
 655         __skb_push(skb, skb->data - skb_mac_header(skb));
 656
 657         /* Network layer. */
 658         if (key->eth.type == htons(ETH_P_IP)) {
 659                 struct iphdr *nh;
 660                 __be16 offset;
 661
 662                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 663
 664                 error = check_iphdr(skb);
 665                 if (unlikely(error)) {
 666                         if (error == -EINVAL) {
 667                                 skb->transport_header = skb->network_header;
 668                                 error = 0;
 669                         }
 670                         goto out;
 671                 }
 672
 673                 nh = ip_hdr(skb);
 674                 key->ipv4.addr.src = nh->saddr;
 675                 key->ipv4.addr.dst = nh->daddr;
 676
 677                 key->ip.proto = nh->protocol;
 678                 key->ip.tos = nh->tos;
 679                 key->ip.ttl = nh->ttl;
 680
 681                 offset = nh->frag_off & htons(IP_OFFSET);
 682                 if (offset) {
 683                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 684                         goto out;
 685                 }
 686                 if (nh->frag_off & htons(IP_MF) ||
 687                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 688                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 689
 690                 /* Transport layer. */
 691                 if (key->ip.proto == IPPROTO_TCP) {
 692                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 693                         if (tcphdr_ok(skb)) {
 694                                 struct tcphdr *tcp = tcp_hdr(skb);
 695                                 key->ipv4.tp.src = tcp->source;
 696                                 key->ipv4.tp.dst = tcp->dest;
 697                         }
 698                 } else if (key->ip.proto == IPPROTO_UDP) {
 699                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 700                         if (udphdr_ok(skb)) {
 701                                 struct udphdr *udp = udp_hdr(skb);
 702                                 key->ipv4.tp.src = udp->source;
 703                                 key->ipv4.tp.dst = udp->dest;
 704                         }
 705                 } else if (key->ip.proto == IPPROTO_ICMP) {
 706                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 707                         if (icmphdr_ok(skb)) {
 708                                 struct icmphdr *icmp = icmp_hdr(skb);
 709                                 /* The ICMP type and code fields use the 16-bit
 710                                  * transport port fields, so we need to store
 711                                  * them in 16-bit network byte order. */
 712                                 key->ipv4.tp.src = htons(icmp->type);
 713                                 key->ipv4.tp.dst = htons(icmp->code);
 714                         }
 715                 }
 716
 717         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 718                 struct arp_eth_header *arp;
 719
 720                 arp = (struct arp_eth_header *)skb_network_header(skb);
 721
 722                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 723                                 && arp->ar_pro == htons(ETH_P_IP)
 724                                 && arp->ar_hln == ETH_ALEN
 725                                 && arp->ar_pln == 4) {
 726
 727                         /* We only match on the lower 8 bits of the opcode. */
 728                         if (ntohs(arp->ar_op) <= 0xff)
 729                                 key->ip.proto = ntohs(arp->ar_op);
 730
 731                         if (key->ip.proto == ARPOP_REQUEST
 732                                         || key->ip.proto == ARPOP_REPLY) {
 733                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 734                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 735                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 736                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 737                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 738                         }
 739                 }
 740         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 741                 int nh_len;             /* IPv6 Header + Extensions */
 742
 743                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 744                 if (unlikely(nh_len < 0)) {
 745                         if (nh_len == -EINVAL)
 746                                 skb->transport_header = skb->network_header;
 747                         else
 748                                 error = nh_len;
 749                         goto out;
 750                 }
 751
 752                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 753                         goto out;
 754                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 755                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 756
 757                 /* Transport layer. */
 758                 if (key->ip.proto == NEXTHDR_TCP) {
 759                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 760                         if (tcphdr_ok(skb)) {
 761                                 struct tcphdr *tcp = tcp_hdr(skb);
 762                                 key->ipv6.tp.src = tcp->source;
 763                                 key->ipv6.tp.dst = tcp->dest;
 764                         }
 765                 } else if (key->ip.proto == NEXTHDR_UDP) {
 766                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 767                         if (udphdr_ok(skb)) {
 768                                 struct udphdr *udp = udp_hdr(skb);
 769                                 key->ipv6.tp.src = udp->source;
 770                                 key->ipv6.tp.dst = udp->dest;
 771                         }
 772                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 773                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 774                         if (icmp6hdr_ok(skb)) {
 775                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 776                                 if (error < 0)
 777                                         goto out;
 778                         }
 779                 }
 780         }
 781
 782 out:
 783         *key_lenp = key_len;
 784         return error;
 785 }
 786
 787 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 788 {
 789         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 790 }
 791
 792 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 793                                 struct sw_flow_key *key, int key_len)
 794 {
 795         struct sw_flow *flow;
 796         struct hlist_node *n;
 797         struct hlist_head *head;
 798         u32 hash;
 799
 800         hash = ovs_flow_hash(key, key_len);
 801
 802         head = find_bucket(table, hash);
 803         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 804
 805                 if (flow->hash == hash &&
 806                     !memcmp(&flow->key, key, key_len)) {
 807                         return flow;
 808                 }
 809         }
 810         return NULL;
 811 }
 812
 813 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 814 {
 815         struct hlist_head *head;
 816
 817         head = find_bucket(table, flow->hash);
 818         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 819         table->count++;
 820 }
 821
 822 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 823 {
 824         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 825         table->count--;
 826         BUG_ON(table->count < 0);
 827 }
 828
 829 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 830 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 831         [OVS_KEY_ATTR_ENCAP] = -1,
 832         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 833         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 834         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 835         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 836         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 837         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 838         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 839         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 840         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 841         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 842         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 843         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 844         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 845
 846         /* Not upstream. */
 847         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 848 };
 849
 850 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 851                                   const struct nlattr *a[], u64 *attrs)
 852 {
 853         const struct ovs_key_icmp *icmp_key;
 854         const struct ovs_key_tcp *tcp_key;
 855         const struct ovs_key_udp *udp_key;
 856
 857         switch (swkey->ip.proto) {
 858         case IPPROTO_TCP:
 859                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 860                         return -EINVAL;
 861                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 862
 863                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 864                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 865                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 866                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 867                 break;
 868
 869         case IPPROTO_UDP:
 870                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 871                         return -EINVAL;
 872                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 873
 874                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 875                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 876                 swkey->ipv4.tp.src = udp_key->udp_src;
 877                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 878                 break;
 879
 880         case IPPROTO_ICMP:
 881                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 882                         return -EINVAL;
 883                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 884
 885                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 886                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 887                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 888                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 889                 break;
 890         }
 891
 892         return 0;
 893 }
 894
 895 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 896                                   const struct nlattr *a[], u64 *attrs)
 897 {
 898         const struct ovs_key_icmpv6 *icmpv6_key;
 899         const struct ovs_key_tcp *tcp_key;
 900         const struct ovs_key_udp *udp_key;
 901
 902         switch (swkey->ip.proto) {
 903         case IPPROTO_TCP:
 904                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 905                         return -EINVAL;
 906                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 907
 908                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 909                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 910                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 911                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 912                 break;
 913
 914         case IPPROTO_UDP:
 915                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 916                         return -EINVAL;
 917                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 918
 919                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 920                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 921                 swkey->ipv6.tp.src = udp_key->udp_src;
 922                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 923                 break;
 924
 925         case IPPROTO_ICMPV6:
 926                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 927                         return -EINVAL;
 928                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 929
 930                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 931                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 932                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 933                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 934
 935                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 936                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 937                         const struct ovs_key_nd *nd_key;
 938
 939                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 940                                 return -EINVAL;
 941                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 942
 943                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 944                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 945                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 946                                sizeof(swkey->ipv6.nd.target));
 947                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 948                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 949                 }
 950                 break;
 951         }
 952
 953         return 0;
 954 }
 955
 956 static int parse_flow_nlattrs(const struct nlattr *attr,
 957                               const struct nlattr *a[], u64 *attrsp)
 958 {
 959         const struct nlattr *nla;
 960         u64 attrs;
 961         int rem;
 962
 963         attrs = 0;
 964         nla_for_each_nested(nla, attr, rem) {
 965                 u16 type = nla_type(nla);
 966                 int expected_len;
 967
 968                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 969                         return -EINVAL;
 970
 971                 expected_len = ovs_key_lens[type];
 972                 if (nla_len(nla) != expected_len && expected_len != -1)
 973                         return -EINVAL;
 974
 975                 attrs |= 1ULL << type;
 976                 a[type] = nla;
 977         }
 978         if (rem)
 979                 return -EINVAL;
 980
 981         *attrsp = attrs;
 982         return 0;
 983 }
 984
 985 /**
 986  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 987  * @swkey: receives the extracted flow key.
 988  * @key_lenp: number of bytes used in @swkey.
 989  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 990  * sequence.
 991  */
 992 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 993                       const struct nlattr *attr)
 994 {
 995         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 996         const struct ovs_key_ethernet *eth_key;
 997         int key_len;
 998         u64 attrs;
 999         int err;
1000
1001         memset(swkey, 0, sizeof(struct sw_flow_key));
1002         key_len = SW_FLOW_KEY_OFFSET(eth);
1003
1004         err = parse_flow_nlattrs(attr, a, &attrs);
1005         if (err)
1006                 return err;
1007
1008         /* Metadata attributes. */
1009         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1010                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1011                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1012         }
1013         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1014                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1015                 if (in_port >= DP_MAX_PORTS)
1016                         return -EINVAL;
1017                 swkey->phy.in_port = in_port;
1018                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1019         } else {
1020                 swkey->phy.in_port = DP_MAX_PORTS;
1021         }
1022
1023         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
1024                 swkey->phy.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1025                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1026         }
1027
1028         /* Data attributes. */
1029         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1030                 return -EINVAL;
1031         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1032
1033         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1034         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1035         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1036
1037         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1038             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1039                 const struct nlattr *encap;
1040                 __be16 tci;
1041
1042                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1043                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1044                               (1 << OVS_KEY_ATTR_ENCAP)))
1045                         return -EINVAL;
1046
1047                 encap = a[OVS_KEY_ATTR_ENCAP];
1048                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1049                 if (tci & htons(VLAN_TAG_PRESENT)) {
1050                         swkey->eth.tci = tci;
1051
1052                         err = parse_flow_nlattrs(encap, a, &attrs);
1053                         if (err)
1054                                 return err;
1055                 } else if (!tci) {
1056                         /* Corner case for truncated 802.1Q header. */
1057                         if (nla_len(encap))
1058                                 return -EINVAL;
1059
1060                         swkey->eth.type = htons(ETH_P_8021Q);
1061                         *key_lenp = key_len;
1062                         return 0;
1063                 } else {
1064                         return -EINVAL;
1065                 }
1066         }
1067
1068         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1069                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1070                 if (ntohs(swkey->eth.type) < 1536)
1071                         return -EINVAL;
1072                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1073         } else {
1074                 swkey->eth.type = htons(ETH_P_802_2);
1075         }
1076
1077         if (swkey->eth.type == htons(ETH_P_IP)) {
1078                 const struct ovs_key_ipv4 *ipv4_key;
1079
1080                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1081                         return -EINVAL;
1082                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1083
1084                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1085                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1086                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1087                         return -EINVAL;
1088                 swkey->ip.proto = ipv4_key->ipv4_proto;
1089                 swkey->ip.tos = ipv4_key->ipv4_tos;
1090                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1091                 swkey->ip.frag = ipv4_key->ipv4_frag;
1092                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1093                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1094
1095                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1096                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1097                         if (err)
1098                                 return err;
1099                 }
1100         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1101                 const struct ovs_key_ipv6 *ipv6_key;
1102
1103                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1104                         return -EINVAL;
1105                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1106
1107                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1108                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1109                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1110                         return -EINVAL;
1111                 swkey->ipv6.label = ipv6_key->ipv6_label;
1112                 swkey->ip.proto = ipv6_key->ipv6_proto;
1113                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1114                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1115                 swkey->ip.frag = ipv6_key->ipv6_frag;
1116                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1117                        sizeof(swkey->ipv6.addr.src));
1118                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1119                        sizeof(swkey->ipv6.addr.dst));
1120
1121                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1122                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1123                         if (err)
1124                                 return err;
1125                 }
1126         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1127                 const struct ovs_key_arp *arp_key;
1128
1129                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1130                         return -EINVAL;
1131                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1132
1133                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1134                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1135                 swkey->ipv4.addr.src = arp_key->arp_sip;
1136                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1137                 if (arp_key->arp_op & htons(0xff00))
1138                         return -EINVAL;
1139                 swkey->ip.proto = ntohs(arp_key->arp_op);
1140                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1141                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1142         }
1143
1144         if (attrs)
1145                 return -EINVAL;
1146         *key_lenp = key_len;
1147
1148         return 0;
1149 }
1150
1151 /**
1152  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1153  * @in_port: receives the extracted input port.
1154  * @tun_id: receives the extracted tunnel ID.
1155  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1156  * sequence.
1157  *
1158  * This parses a series of Netlink attributes that form a flow key, which must
1159  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1160  * get the metadata, that is, the parts of the flow key that cannot be
1161  * extracted from the packet itself.
1162  */
1163 int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1164                                    const struct nlattr *attr)
1165 {
1166         const struct nlattr *nla;
1167         int rem;
1168
1169         *in_port = DP_MAX_PORTS;
1170         *tun_id = 0;
1171         *priority = 0;
1172
1173         nla_for_each_nested(nla, attr, rem) {
1174                 int type = nla_type(nla);
1175
1176                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1177                         if (nla_len(nla) != ovs_key_lens[type])
1178                                 return -EINVAL;
1179
1180                         switch (type) {
1181                         case OVS_KEY_ATTR_PRIORITY:
1182                                 *priority = nla_get_u32(nla);
1183                                 break;
1184
1185                         case OVS_KEY_ATTR_TUN_ID:
1186                                 *tun_id = nla_get_be64(nla);
1187                                 break;
1188
1189                         case OVS_KEY_ATTR_IN_PORT:
1190                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1191                                         return -EINVAL;
1192                                 *in_port = nla_get_u32(nla);
1193                                 break;
1194                         }
1195                 }
1196         }
1197         if (rem)
1198                 return -EINVAL;
1199         return 0;
1200 }
1201
1202 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1203 {
1204         struct ovs_key_ethernet *eth_key;
1205         struct nlattr *nla, *encap;
1206
1207         if (swkey->phy.priority)
1208                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1209
1210         if (swkey->phy.tun_id != cpu_to_be64(0))
1211                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1212
1213         if (swkey->phy.in_port != DP_MAX_PORTS)
1214                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1215
1216         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1217         if (!nla)
1218                 goto nla_put_failure;
1219         eth_key = nla_data(nla);
1220         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1221         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1222
1223         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1224                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q));
1225                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci);
1226                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1227                 if (!swkey->eth.tci)
1228                         goto unencap;
1229         } else {
1230                 encap = NULL;
1231         }
1232
1233         if (swkey->eth.type == htons(ETH_P_802_2))
1234                 goto unencap;
1235
1236         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1237
1238         if (swkey->eth.type == htons(ETH_P_IP)) {
1239                 struct ovs_key_ipv4 *ipv4_key;
1240
1241                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1242                 if (!nla)
1243                         goto nla_put_failure;
1244                 ipv4_key = nla_data(nla);
1245                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1246                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1247                 ipv4_key->ipv4_proto = swkey->ip.proto;
1248                 ipv4_key->ipv4_tos = swkey->ip.tos;
1249                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1250                 ipv4_key->ipv4_frag = swkey->ip.frag;
1251         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1252                 struct ovs_key_ipv6 *ipv6_key;
1253
1254                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1255                 if (!nla)
1256                         goto nla_put_failure;
1257                 ipv6_key = nla_data(nla);
1258                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1259                                 sizeof(ipv6_key->ipv6_src));
1260                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1261                                 sizeof(ipv6_key->ipv6_dst));
1262                 ipv6_key->ipv6_label = swkey->ipv6.label;
1263                 ipv6_key->ipv6_proto = swkey->ip.proto;
1264                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1265                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1266                 ipv6_key->ipv6_frag = swkey->ip.frag;
1267         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1268                 struct ovs_key_arp *arp_key;
1269
1270                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1271                 if (!nla)
1272                         goto nla_put_failure;
1273                 arp_key = nla_data(nla);
1274                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1275                 arp_key->arp_sip = swkey->ipv4.addr.src;
1276                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1277                 arp_key->arp_op = htons(swkey->ip.proto);
1278                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1279                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1280         }
1281
1282         if ((swkey->eth.type == htons(ETH_P_IP) ||
1283              swkey->eth.type == htons(ETH_P_IPV6)) &&
1284              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1285
1286                 if (swkey->ip.proto == IPPROTO_TCP) {
1287                         struct ovs_key_tcp *tcp_key;
1288
1289                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1290                         if (!nla)
1291                                 goto nla_put_failure;
1292                         tcp_key = nla_data(nla);
1293                         if (swkey->eth.type == htons(ETH_P_IP)) {
1294                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1295                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1296                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1297                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1298                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1299                         }
1300                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1301                         struct ovs_key_udp *udp_key;
1302
1303                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1304                         if (!nla)
1305                                 goto nla_put_failure;
1306                         udp_key = nla_data(nla);
1307                         if (swkey->eth.type == htons(ETH_P_IP)) {
1308                                 udp_key->udp_src = swkey->ipv4.tp.src;
1309                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1310                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1311                                 udp_key->udp_src = swkey->ipv6.tp.src;
1312                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1313                         }
1314                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1315                            swkey->ip.proto == IPPROTO_ICMP) {
1316                         struct ovs_key_icmp *icmp_key;
1317
1318                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1319                         if (!nla)
1320                                 goto nla_put_failure;
1321                         icmp_key = nla_data(nla);
1322                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1323                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1324                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1325                            swkey->ip.proto == IPPROTO_ICMPV6) {
1326                         struct ovs_key_icmpv6 *icmpv6_key;
1327
1328                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1329                                                 sizeof(*icmpv6_key));
1330                         if (!nla)
1331                                 goto nla_put_failure;
1332                         icmpv6_key = nla_data(nla);
1333                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1334                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1335
1336                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1337                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1338                                 struct ovs_key_nd *nd_key;
1339
1340                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1341                                 if (!nla)
1342                                         goto nla_put_failure;
1343                                 nd_key = nla_data(nla);
1344                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1345                                                         sizeof(nd_key->nd_target));
1346                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1347                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1348                         }
1349                 }
1350         }
1351
1352 unencap:
1353         if (encap)
1354                 nla_nest_end(skb, encap);
1355
1356         return 0;
1357
1358 nla_put_failure:
1359         return -EMSGSIZE;
1360 }
1361
1362 /* Initializes the flow module.
1363  * Returns zero if successful or a negative error code. */
1364 int ovs_flow_init(void)
1365 {
1366         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1367                                         0, NULL);
1368         if (flow_cache == NULL)
1369                 return -ENOMEM;
1370
1371         return 0;
1372 }
1373
1374 /* Uninitializes the flow module. */
1375 void ovs_flow_exit(void)
1376 {
1377         kmem_cache_destroy(flow_cache);
1378 }