datapath/flow.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include "flow.h"
  10 #include "datapath.h"
  11 #include <linux/uaccess.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/if_ether.h>
  15 #include <linux/if_vlan.h>
  16 #include <net/llc_pdu.h>
  17 #include <linux/kernel.h>
  18 #include <linux/jhash.h>
  19 #include <linux/jiffies.h>
  20 #include <linux/llc.h>
  21 #include <linux/module.h>
  22 #include <linux/in.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/if_arp.h>
  25 #include <linux/if_ether.h>
  26 #include <linux/ip.h>
  27 #include <linux/ipv6.h>
  28 #include <linux/tcp.h>
  29 #include <linux/udp.h>
  30 #include <linux/icmp.h>
  31 #include <linux/icmpv6.h>
  32 #include <linux/rculist.h>
  33 #include <net/inet_ecn.h>
  34 #include <net/ip.h>
  35 #include <net/ipv6.h>
  36 #include <net/ndisc.h>
  37
  38 #include "vlan.h"
  39
  40 static struct kmem_cache *flow_cache;
  41 static unsigned int hash_seed __read_mostly;
  42
  43 static int check_header(struct sk_buff *skb, int len)
  44 {
  45         if (unlikely(skb->len < len))
  46                 return -EINVAL;
  47         if (unlikely(!pskb_may_pull(skb, len)))
  48                 return -ENOMEM;
  49         return 0;
  50 }
  51
  52 static bool arphdr_ok(struct sk_buff *skb)
  53 {
  54         return pskb_may_pull(skb, skb_network_offset(skb) +
  55                                   sizeof(struct arp_eth_header));
  56 }
  57
  58 static int check_iphdr(struct sk_buff *skb)
  59 {
  60         unsigned int nh_ofs = skb_network_offset(skb);
  61         unsigned int ip_len;
  62         int err;
  63
  64         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  65         if (unlikely(err))
  66                 return err;
  67
  68         ip_len = ip_hdrlen(skb);
  69         if (unlikely(ip_len < sizeof(struct iphdr) ||
  70                      skb->len < nh_ofs + ip_len))
  71                 return -EINVAL;
  72
  73         skb_set_transport_header(skb, nh_ofs + ip_len);
  74         return 0;
  75 }
  76
  77 static bool tcphdr_ok(struct sk_buff *skb)
  78 {
  79         int th_ofs = skb_transport_offset(skb);
  80         int tcp_len;
  81
  82         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  83                 return false;
  84
  85         tcp_len = tcp_hdrlen(skb);
  86         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  87                      skb->len < th_ofs + tcp_len))
  88                 return false;
  89
  90         return true;
  91 }
  92
  93 static bool udphdr_ok(struct sk_buff *skb)
  94 {
  95         return pskb_may_pull(skb, skb_transport_offset(skb) +
  96                                   sizeof(struct udphdr));
  97 }
  98
  99 static bool icmphdr_ok(struct sk_buff *skb)
 100 {
 101         return pskb_may_pull(skb, skb_transport_offset(skb) +
 102                                   sizeof(struct icmphdr));
 103 }
 104
 105 u64 flow_used_time(unsigned long flow_jiffies)
 106 {
 107         struct timespec cur_ts;
 108         u64 cur_ms, idle_ms;
 109
 110         ktime_get_ts(&cur_ts);
 111         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 112         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 113                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 114
 115         return cur_ms - idle_ms;
 116 }
 117
 118 #define SW_FLOW_KEY_OFFSET(field)               \
 119         (offsetof(struct sw_flow_key, field) +  \
 120          FIELD_SIZEOF(struct sw_flow_key, field))
 121
 122 /**
 123  * skip_exthdr - skip any IPv6 extension headers
 124  * @skb: skbuff to parse
 125  * @start: offset of first extension header
 126  * @nexthdrp: Initially, points to the type of the extension header at @start.
 127  * This function updates it to point to the extension header at the final
 128  * offset.
 129  * @tos_frag: Points to the @tos_frag member in a &struct sw_flow_key.  This
 130  * function sets an appropriate %OVS_FRAG_TYPE_* value.
 131  *
 132  * This is based on ipv6_skip_exthdr() but adds the updates to *@tos_frag.
 133  *
 134  * When there is more than one fragment header, this version reports whether
 135  * the final fragment header that it examines is a first fragment.
 136  *
 137  * Returns the final payload offset, or -1 on error.
 138  */
 139 static int skip_exthdr(const struct sk_buff *skb, int start, u8 *nexthdrp,
 140                        u8 *tos_frag)
 141 {
 142         u8 nexthdr = *nexthdrp;
 143
 144         while (ipv6_ext_hdr(nexthdr)) {
 145                 struct ipv6_opt_hdr _hdr, *hp;
 146                 int hdrlen;
 147
 148                 if (nexthdr == NEXTHDR_NONE)
 149                         return -1;
 150                 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
 151                 if (hp == NULL)
 152                         return -1;
 153                 if (nexthdr == NEXTHDR_FRAGMENT) {
 154                         __be16 _frag_off, *fp;
 155                         fp = skb_header_pointer(skb,
 156                                                 start+offsetof(struct frag_hdr,
 157                                                                frag_off),
 158                                                 sizeof(_frag_off),
 159                                                 &_frag_off);
 160                         if (fp == NULL)
 161                                 return -1;
 162
 163                         *tos_frag &= ~OVS_FRAG_TYPE_MASK;
 164                         if (ntohs(*fp) & ~0x7) {
 165                                 *tos_frag |= OVS_FRAG_TYPE_LATER;
 166                                 break;
 167                         }
 168                         *tos_frag |= OVS_FRAG_TYPE_FIRST;
 169                         hdrlen = 8;
 170                 } else if (nexthdr == NEXTHDR_AUTH)
 171                         hdrlen = (hp->hdrlen+2)<<2;
 172                 else
 173                         hdrlen = ipv6_optlen(hp);
 174
 175                 nexthdr = hp->nexthdr;
 176                 start += hdrlen;
 177         }
 178
 179         *nexthdrp = nexthdr;
 180         return start;
 181 }
 182
 183 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 184                          int *key_lenp)
 185 {
 186         unsigned int nh_ofs = skb_network_offset(skb);
 187         unsigned int nh_len;
 188         int payload_ofs;
 189         struct ipv6hdr *nh;
 190         uint8_t nexthdr;
 191         int err;
 192
 193         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.addr);
 194
 195         err = check_header(skb, nh_ofs + sizeof(*nh));
 196         if (unlikely(err))
 197                 return err;
 198
 199         nh = ipv6_hdr(skb);
 200         nexthdr = nh->nexthdr;
 201         payload_ofs = (u8 *)(nh + 1) - skb->data;
 202
 203         key->ip.proto = NEXTHDR_NONE;
 204         key->ip.tos_frag = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
 205         ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
 206         ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
 207
 208         payload_ofs = skip_exthdr(skb, payload_ofs,
 209                                   &nexthdr, &key->ip.tos_frag);
 210         if (unlikely(payload_ofs < 0))
 211                 return -EINVAL;
 212
 213         nh_len = payload_ofs - nh_ofs;
 214         skb_set_transport_header(skb, nh_ofs + nh_len);
 215         key->ip.proto = nexthdr;
 216         return nh_len;
 217 }
 218
 219 static bool icmp6hdr_ok(struct sk_buff *skb)
 220 {
 221         return pskb_may_pull(skb, skb_transport_offset(skb) +
 222                                   sizeof(struct icmp6hdr));
 223 }
 224
 225 #define TCP_FLAGS_OFFSET 13
 226 #define TCP_FLAG_MASK 0x3f
 227
 228 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 229 {
 230         u8 tcp_flags = 0;
 231
 232         if (flow->key.eth.type == htons(ETH_P_IP) &&
 233             flow->key.ip.proto == IPPROTO_TCP) {
 234                 u8 *tcp = (u8 *)tcp_hdr(skb);
 235                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 236         }
 237
 238         spin_lock(&flow->lock);
 239         flow->used = jiffies;
 240         flow->packet_count++;
 241         flow->byte_count += skb->len;
 242         flow->tcp_flags |= tcp_flags;
 243         spin_unlock(&flow->lock);
 244 }
 245
 246 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 247 {
 248         int actions_len = nla_len(actions);
 249         struct sw_flow_actions *sfa;
 250
 251         /* At least DP_MAX_PORTS actions are required to be able to flood a
 252          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 253          * etc. */
 254         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 255                 return ERR_PTR(-EINVAL);
 256
 257         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 258         if (!sfa)
 259                 return ERR_PTR(-ENOMEM);
 260
 261         sfa->actions_len = actions_len;
 262         memcpy(sfa->actions, nla_data(actions), actions_len);
 263         return sfa;
 264 }
 265
 266 struct sw_flow *flow_alloc(void)
 267 {
 268         struct sw_flow *flow;
 269
 270         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 271         if (!flow)
 272                 return ERR_PTR(-ENOMEM);
 273
 274         spin_lock_init(&flow->lock);
 275         atomic_set(&flow->refcnt, 1);
 276         flow->sf_acts = NULL;
 277         flow->dead = false;
 278
 279         return flow;
 280 }
 281
 282 static struct hlist_head __rcu *find_bucket(struct flow_table * table, u32 hash)
 283 {
 284         return flex_array_get(table->buckets,
 285                                 (hash & (table->n_buckets - 1)));
 286 }
 287
 288 static struct flex_array  __rcu *alloc_buckets(unsigned int n_buckets)
 289 {
 290         struct flex_array  __rcu *buckets;
 291         int i, err;
 292
 293         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 294                                    n_buckets, GFP_KERNEL);
 295         if (!buckets)
 296                 return NULL;
 297
 298         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 299         if (err) {
 300                 flex_array_free(buckets);
 301                 return NULL;
 302         }
 303
 304         for (i = 0; i < n_buckets; i++)
 305                 INIT_HLIST_HEAD((struct hlist_head *)
 306                                         flex_array_get(buckets, i));
 307
 308         return buckets;
 309 }
 310
 311 static void free_buckets(struct flex_array *buckets)
 312 {
 313         flex_array_free(buckets);
 314 }
 315
 316 struct flow_table *flow_tbl_alloc(int new_size)
 317 {
 318         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 319
 320         if (!table)
 321                 return NULL;
 322
 323         table->buckets = alloc_buckets(new_size);
 324
 325         if (!table->buckets) {
 326                 kfree(table);
 327                 return NULL;
 328         }
 329         table->n_buckets = new_size;
 330         table->count = 0;
 331
 332         return table;
 333 }
 334
 335 static void flow_free(struct sw_flow *flow)
 336 {
 337         flow->dead = true;
 338         flow_put(flow);
 339 }
 340
 341 void flow_tbl_destroy(struct flow_table *table)
 342 {
 343         int i;
 344
 345         if (!table)
 346                 return;
 347
 348         for (i = 0; i < table->n_buckets; i++) {
 349                 struct sw_flow *flow;
 350                 struct hlist_head *head = flex_array_get(table->buckets, i);
 351                 struct hlist_node *node, *n;
 352
 353                 hlist_for_each_entry_safe(flow, node, n, head, hash_node) {
 354                         hlist_del_init_rcu(&flow->hash_node);
 355                         flow_free(flow);
 356                 }
 357         }
 358
 359         free_buckets(table->buckets);
 360         kfree(table);
 361 }
 362
 363 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 364 {
 365         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 366
 367         flow_tbl_destroy(table);
 368 }
 369
 370 void flow_tbl_deferred_destroy(struct flow_table *table)
 371 {
 372         if (!table)
 373                 return;
 374
 375         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 376 }
 377
 378 struct sw_flow *flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 379 {
 380         struct sw_flow *flow;
 381         struct hlist_head *head;
 382         struct hlist_node *n;
 383         int i;
 384
 385         while (*bucket < table->n_buckets) {
 386                 i = 0;
 387                 head = flex_array_get(table->buckets, *bucket);
 388                 hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 389                         if (i < *last) {
 390                                 i++;
 391                                 continue;
 392                         }
 393                         *last = i + 1;
 394                         return flow;
 395                 }
 396                 (*bucket)++;
 397                 *last = 0;
 398         }
 399
 400         return NULL;
 401 }
 402
 403 struct flow_table *flow_tbl_expand(struct flow_table *table)
 404 {
 405         struct flow_table *new_table;
 406         int n_buckets = table->n_buckets * 2;
 407         int i;
 408
 409         new_table = flow_tbl_alloc(n_buckets);
 410         if (!new_table)
 411                 return ERR_PTR(-ENOMEM);
 412
 413         for (i = 0; i < table->n_buckets; i++) {
 414                 struct sw_flow *flow;
 415                 struct hlist_head *head;
 416                 struct hlist_node *n, *pos;
 417
 418                 head = flex_array_get(table->buckets, i);
 419
 420                 hlist_for_each_entry_safe(flow, n, pos, head, hash_node) {
 421                         hlist_del_init_rcu(&flow->hash_node);
 422                         flow_tbl_insert(new_table, flow);
 423                 }
 424         }
 425
 426         return new_table;
 427 }
 428
 429 /* RCU callback used by flow_deferred_free. */
 430 static void rcu_free_flow_callback(struct rcu_head *rcu)
 431 {
 432         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 433
 434         flow->dead = true;
 435         flow_put(flow);
 436 }
 437
 438 /* Schedules 'flow' to be freed after the next RCU grace period.
 439  * The caller must hold rcu_read_lock for this to be sensible. */
 440 void flow_deferred_free(struct sw_flow *flow)
 441 {
 442         call_rcu(&flow->rcu, rcu_free_flow_callback);
 443 }
 444
 445 void flow_hold(struct sw_flow *flow)
 446 {
 447         atomic_inc(&flow->refcnt);
 448 }
 449
 450 void flow_put(struct sw_flow *flow)
 451 {
 452         if (unlikely(!flow))
 453                 return;
 454
 455         if (atomic_dec_and_test(&flow->refcnt)) {
 456                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 457                 kmem_cache_free(flow_cache, flow);
 458         }
 459 }
 460
 461 /* RCU callback used by flow_deferred_free_acts. */
 462 static void rcu_free_acts_callback(struct rcu_head *rcu)
 463 {
 464         struct sw_flow_actions *sf_acts = container_of(rcu,
 465                         struct sw_flow_actions, rcu);
 466         kfree(sf_acts);
 467 }
 468
 469 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 470  * The caller must hold rcu_read_lock for this to be sensible. */
 471 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 472 {
 473         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 474 }
 475
 476 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 477 {
 478         struct qtag_prefix {
 479                 __be16 eth_type; /* ETH_P_8021Q */
 480                 __be16 tci;
 481         };
 482         struct qtag_prefix *qp;
 483
 484         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 485                                          sizeof(__be16))))
 486                 return -ENOMEM;
 487
 488         qp = (struct qtag_prefix *) skb->data;
 489         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 490         __skb_pull(skb, sizeof(struct qtag_prefix));
 491
 492         return 0;
 493 }
 494
 495 static __be16 parse_ethertype(struct sk_buff *skb)
 496 {
 497         struct llc_snap_hdr {
 498                 u8  dsap;  /* Always 0xAA */
 499                 u8  ssap;  /* Always 0xAA */
 500                 u8  ctrl;
 501                 u8  oui[3];
 502                 __be16 ethertype;
 503         };
 504         struct llc_snap_hdr *llc;
 505         __be16 proto;
 506
 507         proto = *(__be16 *) skb->data;
 508         __skb_pull(skb, sizeof(__be16));
 509
 510         if (ntohs(proto) >= 1536)
 511                 return proto;
 512
 513         if (skb->len < sizeof(struct llc_snap_hdr))
 514                 return htons(ETH_P_802_2);
 515
 516         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 517                 return htons(0);
 518
 519         llc = (struct llc_snap_hdr *) skb->data;
 520         if (llc->dsap != LLC_SAP_SNAP ||
 521             llc->ssap != LLC_SAP_SNAP ||
 522             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 523                 return htons(ETH_P_802_2);
 524
 525         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 526         return llc->ethertype;
 527 }
 528
 529 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 530                         int *key_lenp, int nh_len)
 531 {
 532         struct icmp6hdr *icmp = icmp6_hdr(skb);
 533         int error = 0;
 534         int key_len;
 535
 536         /* The ICMPv6 type and code fields use the 16-bit transport port
 537          * fields, so we need to store them in 16-bit network byte order.
 538          */
 539         key->ipv6.tp.src = htons(icmp->icmp6_type);
 540         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 541         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 542
 543         if (icmp->icmp6_code == 0 &&
 544             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 545              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 546                 int icmp_len = skb->len - skb_transport_offset(skb);
 547                 struct nd_msg *nd;
 548                 int offset;
 549
 550                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 551
 552                 /* In order to process neighbor discovery options, we need the
 553                  * entire packet.
 554                  */
 555                 if (unlikely(icmp_len < sizeof(*nd)))
 556                         goto out;
 557                 if (unlikely(skb_linearize(skb))) {
 558                         error = -ENOMEM;
 559                         goto out;
 560                 }
 561
 562                 nd = (struct nd_msg *)skb_transport_header(skb);
 563                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
 564                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 565
 566                 icmp_len -= sizeof(*nd);
 567                 offset = 0;
 568                 while (icmp_len >= 8) {
 569                         struct nd_opt_hdr *nd_opt =
 570                                  (struct nd_opt_hdr *)(nd->opt + offset);
 571                         int opt_len = nd_opt->nd_opt_len * 8;
 572
 573                         if (unlikely(!opt_len || opt_len > icmp_len))
 574                                 goto invalid;
 575
 576                         /* Store the link layer address if the appropriate
 577                          * option is provided.  It is considered an error if
 578                          * the same link layer option is specified twice.
 579                          */
 580                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 581                             && opt_len == 8) {
 582                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 583                                         goto invalid;
 584                                 memcpy(key->ipv6.nd.sll,
 585                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 586                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 587                                    && opt_len == 8) {
 588                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 589                                         goto invalid;
 590                                 memcpy(key->ipv6.nd.tll,
 591                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 592                         }
 593
 594                         icmp_len -= opt_len;
 595                         offset += opt_len;
 596                 }
 597         }
 598
 599         goto out;
 600
 601 invalid:
 602         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 603         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 604         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 605
 606 out:
 607         *key_lenp = key_len;
 608         return error;
 609 }
 610
 611 /**
 612  * flow_extract - extracts a flow key from an Ethernet frame.
 613  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 614  * Ethernet header
 615  * @in_port: port number on which @skb was received.
 616  * @key: output flow key
 617  * @key_lenp: length of output flow key
 618  *
 619  * The caller must ensure that skb->len >= ETH_HLEN.
 620  *
 621  * Returns 0 if successful, otherwise a negative errno value.
 622  *
 623  * Initializes @skb header pointers as follows:
 624  *
 625  *    - skb->mac_header: the Ethernet header.
 626  *
 627  *    - skb->network_header: just past the Ethernet header, or just past the
 628  *      VLAN header, to the first byte of the Ethernet payload.
 629  *
 630  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 631  *      on output, then just past the IP header, if one is present and
 632  *      of a correct length, otherwise the same as skb->network_header.
 633  *      For other key->dl_type values it is left untouched.
 634  */
 635 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 636                  int *key_lenp)
 637 {
 638         int error = 0;
 639         int key_len = SW_FLOW_KEY_OFFSET(eth);
 640         struct ethhdr *eth;
 641
 642         memset(key, 0, sizeof(*key));
 643
 644         key->phy.priority = skb->priority;
 645         key->phy.tun_id = OVS_CB(skb)->tun_id;
 646         key->phy.in_port = in_port;
 647
 648         skb_reset_mac_header(skb);
 649
 650         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 651          * header in the linear data area.
 652          */
 653         eth = eth_hdr(skb);
 654         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 655         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 656
 657         __skb_pull(skb, 2 * ETH_ALEN);
 658
 659         if (vlan_tx_tag_present(skb))
 660                 key->eth.tci = htons(vlan_get_tci(skb));
 661         else if (eth->h_proto == htons(ETH_P_8021Q))
 662                 if (unlikely(parse_vlan(skb, key)))
 663                         return -ENOMEM;
 664
 665         key->eth.type = parse_ethertype(skb);
 666         if (unlikely(key->eth.type == htons(0)))
 667                 return -ENOMEM;
 668
 669         skb_reset_network_header(skb);
 670         __skb_push(skb, skb->data - skb_mac_header(skb));
 671
 672         /* Network layer. */
 673         if (key->eth.type == htons(ETH_P_IP)) {
 674                 struct iphdr *nh;
 675                 __be16 offset;
 676
 677                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 678
 679                 error = check_iphdr(skb);
 680                 if (unlikely(error)) {
 681                         if (error == -EINVAL) {
 682                                 skb->transport_header = skb->network_header;
 683                                 error = 0;
 684                         }
 685                         goto out;
 686                 }
 687
 688                 nh = ip_hdr(skb);
 689                 key->ipv4.addr.src = nh->saddr;
 690                 key->ipv4.addr.dst = nh->daddr;
 691
 692                 key->ip.proto = nh->protocol;
 693                 key->ip.tos_frag = nh->tos & ~INET_ECN_MASK;
 694
 695                 offset = nh->frag_off & htons(IP_OFFSET);
 696                 if (offset) {
 697                         key->ip.tos_frag |= OVS_FRAG_TYPE_LATER;
 698                         goto out;
 699                 }
 700                 if (nh->frag_off & htons(IP_MF) ||
 701                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 702                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
 703
 704                 /* Transport layer. */
 705                 if (key->ip.proto == IPPROTO_TCP) {
 706                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 707                         if (tcphdr_ok(skb)) {
 708                                 struct tcphdr *tcp = tcp_hdr(skb);
 709                                 key->ipv4.tp.src = tcp->source;
 710                                 key->ipv4.tp.dst = tcp->dest;
 711                         }
 712                 } else if (key->ip.proto == IPPROTO_UDP) {
 713                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 714                         if (udphdr_ok(skb)) {
 715                                 struct udphdr *udp = udp_hdr(skb);
 716                                 key->ipv4.tp.src = udp->source;
 717                                 key->ipv4.tp.dst = udp->dest;
 718                         }
 719                 } else if (key->ip.proto == IPPROTO_ICMP) {
 720                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 721                         if (icmphdr_ok(skb)) {
 722                                 struct icmphdr *icmp = icmp_hdr(skb);
 723                                 /* The ICMP type and code fields use the 16-bit
 724                                  * transport port fields, so we need to store
 725                                  * them in 16-bit network byte order. */
 726                                 key->ipv4.tp.src = htons(icmp->type);
 727                                 key->ipv4.tp.dst = htons(icmp->code);
 728                         }
 729                 }
 730
 731         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 732                 struct arp_eth_header *arp;
 733
 734                 arp = (struct arp_eth_header *)skb_network_header(skb);
 735
 736                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 737                                 && arp->ar_pro == htons(ETH_P_IP)
 738                                 && arp->ar_hln == ETH_ALEN
 739                                 && arp->ar_pln == 4) {
 740
 741                         /* We only match on the lower 8 bits of the opcode. */
 742                         if (ntohs(arp->ar_op) <= 0xff)
 743                                 key->ip.proto = ntohs(arp->ar_op);
 744
 745                         if (key->ip.proto == ARPOP_REQUEST
 746                                         || key->ip.proto == ARPOP_REPLY) {
 747                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 748                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 749                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 750                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 751                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 752                         }
 753                 }
 754         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 755                 int nh_len;             /* IPv6 Header + Extensions */
 756
 757                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 758                 if (unlikely(nh_len < 0)) {
 759                         if (nh_len == -EINVAL)
 760                                 skb->transport_header = skb->network_header;
 761                         else
 762                                 error = nh_len;
 763                         goto out;
 764                 }
 765
 766                 if ((key->ip.tos_frag & OVS_FRAG_TYPE_MASK) == OVS_FRAG_TYPE_LATER)
 767                         goto out;
 768                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 769                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
 770
 771                 /* Transport layer. */
 772                 if (key->ip.proto == NEXTHDR_TCP) {
 773                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 774                         if (tcphdr_ok(skb)) {
 775                                 struct tcphdr *tcp = tcp_hdr(skb);
 776                                 key->ipv6.tp.src = tcp->source;
 777                                 key->ipv6.tp.dst = tcp->dest;
 778                         }
 779                 } else if (key->ip.proto == NEXTHDR_UDP) {
 780                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 781                         if (udphdr_ok(skb)) {
 782                                 struct udphdr *udp = udp_hdr(skb);
 783                                 key->ipv6.tp.src = udp->source;
 784                                 key->ipv6.tp.dst = udp->dest;
 785                         }
 786                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 787                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 788                         if (icmp6hdr_ok(skb)) {
 789                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 790                                 if (error < 0)
 791                                         goto out;
 792                         }
 793                 }
 794         }
 795
 796 out:
 797         *key_lenp = key_len;
 798         return error;
 799 }
 800
 801 u32 flow_hash(const struct sw_flow_key *key, int key_len)
 802 {
 803         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
 804 }
 805
 806 struct sw_flow *flow_tbl_lookup(struct flow_table *table,
 807                                 struct sw_flow_key *key, int key_len)
 808 {
 809         struct sw_flow *flow;
 810         struct hlist_node *n;
 811         struct hlist_head *head;
 812         u32 hash;
 813
 814         hash = flow_hash(key, key_len);
 815
 816         head = find_bucket(table, hash);
 817         hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 818
 819                 if (flow->hash == hash &&
 820                     !memcmp(&flow->key, key, key_len)) {
 821                         return flow;
 822                 }
 823         }
 824         return NULL;
 825 }
 826
 827 void flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 828 {
 829         struct hlist_head *head;
 830
 831         head = find_bucket(table, flow->hash);
 832         hlist_add_head_rcu(&flow->hash_node, head);
 833         table->count++;
 834 }
 835
 836 void flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 837 {
 838         if (!hlist_unhashed(&flow->hash_node)) {
 839                 hlist_del_init_rcu(&flow->hash_node);
 840                 table->count--;
 841                 BUG_ON(table->count < 0);
 842         }
 843 }
 844
 845 static int parse_tos_frag(struct sw_flow_key *swkey, u8 tos, u8 frag)
 846 {
 847         if (tos & INET_ECN_MASK || frag > OVS_FRAG_TYPE_MAX)
 848                 return -EINVAL;
 849
 850         swkey->ip.tos_frag = tos | frag;
 851         return 0;
 852 }
 853
 854 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 855 const u32 ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 856         [OVS_KEY_ATTR_PRIORITY] = 4,
 857         [OVS_KEY_ATTR_TUN_ID] = 8,
 858         [OVS_KEY_ATTR_IN_PORT] = 4,
 859         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 860         [OVS_KEY_ATTR_8021Q] = sizeof(struct ovs_key_8021q),
 861         [OVS_KEY_ATTR_ETHERTYPE] = 2,
 862         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 863         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 864         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 865         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 866         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 867         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 868         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 869         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 870 };
 871
 872 /**
 873  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 874  * @swkey: receives the extracted flow key.
 875  * @key_lenp: number of bytes used in @swkey.
 876  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 877  * sequence.
 878  *
 879  * This state machine accepts the following forms, with [] for optional
 880  * elements and | for alternatives:
 881  *
 882  * [priority] [tun_id] [in_port] ethernet [8021q] [ethertype \
 883  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 884  *
 885  * except that IPv4 or IPv6 terminates the sequence if its @ipv4_frag or
 886  * @ipv6_frag member, respectively, equals %OVS_FRAG_TYPE_LATER.
 887  */
 888 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 889                       const struct nlattr *attr)
 890 {
 891         int error = 0;
 892         enum ovs_frag_type frag_type;
 893         const struct nlattr *nla;
 894         u16 prev_type;
 895         int rem;
 896         int key_len;
 897
 898         memset(swkey, 0, sizeof(*swkey));
 899         swkey->phy.in_port = USHRT_MAX;
 900         swkey->eth.type = htons(ETH_P_802_2);
 901         key_len = SW_FLOW_KEY_OFFSET(eth);
 902
 903         prev_type = OVS_KEY_ATTR_UNSPEC;
 904         nla_for_each_nested(nla, attr, rem) {
 905                 const struct ovs_key_ethernet *eth_key;
 906                 const struct ovs_key_8021q *q_key;
 907                 const struct ovs_key_ipv4 *ipv4_key;
 908                 const struct ovs_key_ipv6 *ipv6_key;
 909                 const struct ovs_key_tcp *tcp_key;
 910                 const struct ovs_key_udp *udp_key;
 911                 const struct ovs_key_icmp *icmp_key;
 912                 const struct ovs_key_icmpv6 *icmpv6_key;
 913                 const struct ovs_key_arp *arp_key;
 914                 const struct ovs_key_nd *nd_key;
 915
 916                 int type = nla_type(nla);
 917
 918                 if (type > OVS_KEY_ATTR_MAX ||
 919                     nla_len(nla) != ovs_key_lens[type])
 920                         goto invalid;
 921
 922 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 923                 switch (TRANSITION(prev_type, type)) {
 924                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_PRIORITY):
 925                         swkey->phy.priority = nla_get_u32(nla);
 926                         break;
 927
 928                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
 929                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_TUN_ID):
 930                         swkey->phy.tun_id = nla_get_be64(nla);
 931                         break;
 932
 933                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
 934                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_IN_PORT):
 935                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
 936                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 937                                 goto invalid;
 938                         swkey->phy.in_port = nla_get_u32(nla);
 939                         break;
 940
 941                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_ETHERNET):
 942                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_ETHERNET):
 943                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_ETHERNET):
 944                 case TRANSITION(OVS_KEY_ATTR_IN_PORT, OVS_KEY_ATTR_ETHERNET):
 945                         eth_key = nla_data(nla);
 946                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
 947                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
 948                         break;
 949
 950                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_8021Q):
 951                         q_key = nla_data(nla);
 952                         /* Only standard 0x8100 VLANs currently supported. */
 953                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 954                                 goto invalid;
 955                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 956                                 goto invalid;
 957                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 958                         break;
 959
 960                 case TRANSITION(OVS_KEY_ATTR_8021Q, OVS_KEY_ATTR_ETHERTYPE):
 961                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_ETHERTYPE):
 962                         swkey->eth.type = nla_get_be16(nla);
 963                         if (ntohs(swkey->eth.type) < 1536)
 964                                 goto invalid;
 965                         break;
 966
 967                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV4):
 968                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 969                         if (swkey->eth.type != htons(ETH_P_IP))
 970                                 goto invalid;
 971                         ipv4_key = nla_data(nla);
 972                         swkey->ip.proto = ipv4_key->ipv4_proto;
 973                         if (parse_tos_frag(swkey, ipv4_key->ipv4_tos,
 974                                            ipv4_key->ipv4_frag))
 975                                 goto invalid;
 976                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
 977                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
 978                         break;
 979
 980                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV6):
 981                         key_len = SW_FLOW_KEY_OFFSET(ipv6.addr);
 982                         if (swkey->eth.type != htons(ETH_P_IPV6))
 983                                 goto invalid;
 984                         ipv6_key = nla_data(nla);
 985                         swkey->ip.proto = ipv6_key->ipv6_proto;
 986                         if (parse_tos_frag(swkey, ipv6_key->ipv6_tos,
 987                                            ipv6_key->ipv6_frag))
 988                                 goto invalid;
 989                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
 990                                         sizeof(swkey->ipv6.addr.src));
 991                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
 992                                         sizeof(swkey->ipv6.addr.dst));
 993                         break;
 994
 995                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_TCP):
 996                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 997                         if (swkey->ip.proto != IPPROTO_TCP)
 998                                 goto invalid;
 999                         tcp_key = nla_data(nla);
1000                         swkey->ipv4.tp.src = tcp_key->tcp_src;
1001                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
1002                         break;
1003
1004                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_TCP):
1005                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1006                         if (swkey->ip.proto != IPPROTO_TCP)
1007                                 goto invalid;
1008                         tcp_key = nla_data(nla);
1009                         swkey->ipv6.tp.src = tcp_key->tcp_src;
1010                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
1011                         break;
1012
1013                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_UDP):
1014                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1015                         if (swkey->ip.proto != IPPROTO_UDP)
1016                                 goto invalid;
1017                         udp_key = nla_data(nla);
1018                         swkey->ipv4.tp.src = udp_key->udp_src;
1019                         swkey->ipv4.tp.dst = udp_key->udp_dst;
1020                         break;
1021
1022                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_UDP):
1023                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1024                         if (swkey->ip.proto != IPPROTO_UDP)
1025                                 goto invalid;
1026                         udp_key = nla_data(nla);
1027                         swkey->ipv6.tp.src = udp_key->udp_src;
1028                         swkey->ipv6.tp.dst = udp_key->udp_dst;
1029                         break;
1030
1031                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_ICMP):
1032                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1033                         if (swkey->ip.proto != IPPROTO_ICMP)
1034                                 goto invalid;
1035                         icmp_key = nla_data(nla);
1036                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
1037                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
1038                         break;
1039
1040                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_ICMPV6):
1041                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1042                         if (swkey->ip.proto != IPPROTO_ICMPV6)
1043                                 goto invalid;
1044                         icmpv6_key = nla_data(nla);
1045                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
1046                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
1047                         break;
1048
1049                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_ARP):
1050                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1051                         if (swkey->eth.type != htons(ETH_P_ARP))
1052                                 goto invalid;
1053                         arp_key = nla_data(nla);
1054                         swkey->ipv4.addr.src = arp_key->arp_sip;
1055                         swkey->ipv4.addr.dst = arp_key->arp_tip;
1056                         if (arp_key->arp_op & htons(0xff00))
1057                                 goto invalid;
1058                         swkey->ip.proto = ntohs(arp_key->arp_op);
1059                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1060                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1061                         break;
1062
1063                 case TRANSITION(OVS_KEY_ATTR_ICMPV6, OVS_KEY_ATTR_ND):
1064                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
1065                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
1066                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
1067                                 goto invalid;
1068                         nd_key = nla_data(nla);
1069                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
1070                                         sizeof(swkey->ipv6.nd.target));
1071                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
1072                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
1073                         break;
1074
1075                 default:
1076                         goto invalid;
1077                 }
1078
1079                 prev_type = type;
1080         }
1081         if (rem)
1082                 goto invalid;
1083
1084         frag_type = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1085         switch (prev_type) {
1086         case OVS_KEY_ATTR_UNSPEC:
1087                 goto invalid;
1088
1089         case OVS_KEY_ATTR_PRIORITY:
1090         case OVS_KEY_ATTR_TUN_ID:
1091         case OVS_KEY_ATTR_IN_PORT:
1092                 goto invalid;
1093
1094         case OVS_KEY_ATTR_ETHERNET:
1095         case OVS_KEY_ATTR_8021Q:
1096                 goto ok;
1097
1098         case OVS_KEY_ATTR_ETHERTYPE:
1099                 if (swkey->eth.type == htons(ETH_P_IP) ||
1100                     swkey->eth.type == htons(ETH_P_IPV6) ||
1101                     swkey->eth.type == htons(ETH_P_ARP))
1102                         goto invalid;
1103                 goto ok;
1104
1105         case OVS_KEY_ATTR_IPV4:
1106                 if (frag_type == OVS_FRAG_TYPE_LATER)
1107                         goto ok;
1108                 if (swkey->ip.proto == IPPROTO_TCP ||
1109                     swkey->ip.proto == IPPROTO_UDP ||
1110                     swkey->ip.proto == IPPROTO_ICMP)
1111                         goto invalid;
1112                 goto ok;
1113
1114         case OVS_KEY_ATTR_IPV6:
1115                 if (frag_type == OVS_FRAG_TYPE_LATER)
1116                         goto ok;
1117                 if (swkey->ip.proto == IPPROTO_TCP ||
1118                     swkey->ip.proto == IPPROTO_UDP ||
1119                     swkey->ip.proto == IPPROTO_ICMPV6)
1120                         goto invalid;
1121                 goto ok;
1122
1123         case OVS_KEY_ATTR_ICMPV6:
1124                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1125                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT) ||
1126                     frag_type == OVS_FRAG_TYPE_LATER)
1127                         goto invalid;
1128                 goto ok;
1129
1130         case OVS_KEY_ATTR_TCP:
1131         case OVS_KEY_ATTR_UDP:
1132         case OVS_KEY_ATTR_ICMP:
1133         case OVS_KEY_ATTR_ND:
1134                 if (frag_type == OVS_FRAG_TYPE_LATER)
1135                         goto invalid;
1136                 goto ok;
1137
1138         case OVS_KEY_ATTR_ARP:
1139                 goto ok;
1140
1141         default:
1142                 WARN_ON_ONCE(1);
1143         }
1144
1145 invalid:
1146         error = -EINVAL;
1147
1148 ok:
1149         *key_lenp = key_len;
1150         return error;
1151 }
1152
1153 /**
1154  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1155  * @in_port: receives the extracted input port.
1156  * @tun_id: receives the extracted tunnel ID.
1157  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1158  * sequence.
1159  *
1160  * This parses a series of Netlink attributes that form a flow key, which must
1161  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1162  * get the metadata, that is, the parts of the flow key that cannot be
1163  * extracted from the packet itself.
1164  */
1165 int flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1166                                const struct nlattr *attr)
1167 {
1168         const struct nlattr *nla;
1169         u16 prev_type;
1170         int rem;
1171
1172         *in_port = USHRT_MAX;
1173         *tun_id = 0;
1174         *priority = 0;
1175
1176         prev_type = OVS_KEY_ATTR_UNSPEC;
1177         nla_for_each_nested(nla, attr, rem) {
1178                 int type = nla_type(nla);
1179
1180                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
1181                         return -EINVAL;
1182
1183                 switch (TRANSITION(prev_type, type)) {
1184                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_PRIORITY):
1185                         *priority = nla_get_u32(nla);
1186                         break;
1187
1188                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
1189                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_TUN_ID):
1190                         *tun_id = nla_get_be64(nla);
1191                         break;
1192
1193                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
1194                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_IN_PORT):
1195                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
1196                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
1197                                 return -EINVAL;
1198                         *in_port = nla_get_u32(nla);
1199                         break;
1200
1201                 default:
1202                         return 0;
1203                 }
1204
1205                 prev_type = type;
1206         }
1207         if (rem)
1208                 return -EINVAL;
1209         return 0;
1210 }
1211
1212 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1213 {
1214         struct ovs_key_ethernet *eth_key;
1215         struct nlattr *nla;
1216
1217         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
1218          * to be updated, but will at least raise awareness when new
1219          * datapath key types are added. */
1220         BUILD_BUG_ON(__OVS_KEY_ATTR_MAX != 15);
1221
1222         if (swkey->phy.priority)
1223                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1224
1225         if (swkey->phy.tun_id != cpu_to_be64(0))
1226                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1227
1228         if (swkey->phy.in_port != USHRT_MAX)
1229                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1230
1231         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1232         if (!nla)
1233                 goto nla_put_failure;
1234         eth_key = nla_data(nla);
1235         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1236         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1237
1238         if (swkey->eth.tci != htons(0)) {
1239                 struct ovs_key_8021q q_key;
1240
1241                 q_key.q_tpid = htons(ETH_P_8021Q);
1242                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
1243                 NLA_PUT(skb, OVS_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
1244         }
1245
1246         if (swkey->eth.type == htons(ETH_P_802_2))
1247                 return 0;
1248
1249         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1250
1251         if (swkey->eth.type == htons(ETH_P_IP)) {
1252                 struct ovs_key_ipv4 *ipv4_key;
1253
1254                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1255                 if (!nla)
1256                         goto nla_put_failure;
1257                 ipv4_key = nla_data(nla);
1258                 memset(ipv4_key, 0, sizeof(struct ovs_key_ipv4));
1259                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1260                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1261                 ipv4_key->ipv4_proto = swkey->ip.proto;
1262                 ipv4_key->ipv4_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1263                 ipv4_key->ipv4_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1264         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1265                 struct ovs_key_ipv6 *ipv6_key;
1266
1267                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1268                 if (!nla)
1269                         goto nla_put_failure;
1270                 ipv6_key = nla_data(nla);
1271                 memset(ipv6_key, 0, sizeof(struct ovs_key_ipv6));
1272                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1273                                 sizeof(ipv6_key->ipv6_src));
1274                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1275                                 sizeof(ipv6_key->ipv6_dst));
1276                 ipv6_key->ipv6_proto = swkey->ip.proto;
1277                 ipv6_key->ipv6_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1278                 ipv6_key->ipv6_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1279         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1280                 struct ovs_key_arp *arp_key;
1281
1282                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1283                 if (!nla)
1284                         goto nla_put_failure;
1285                 arp_key = nla_data(nla);
1286                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1287                 arp_key->arp_sip = swkey->ipv4.addr.src;
1288                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1289                 arp_key->arp_op = htons(swkey->ip.proto);
1290                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1291                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1292         }
1293
1294         if ((swkey->eth.type == htons(ETH_P_IP) ||
1295              swkey->eth.type == htons(ETH_P_IPV6)) &&
1296             (swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK) != OVS_FRAG_TYPE_LATER) {
1297
1298                 if (swkey->ip.proto == IPPROTO_TCP) {
1299                         struct ovs_key_tcp *tcp_key;
1300
1301                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1302                         if (!nla)
1303                                 goto nla_put_failure;
1304                         tcp_key = nla_data(nla);
1305                         if (swkey->eth.type == htons(ETH_P_IP)) {
1306                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1307                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1308                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1309                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1310                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1311                         }
1312                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1313                         struct ovs_key_udp *udp_key;
1314
1315                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1316                         if (!nla)
1317                                 goto nla_put_failure;
1318                         udp_key = nla_data(nla);
1319                         if (swkey->eth.type == htons(ETH_P_IP)) {
1320                                 udp_key->udp_src = swkey->ipv4.tp.src;
1321                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1322                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1323                                 udp_key->udp_src = swkey->ipv6.tp.src;
1324                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1325                         }
1326                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1327                            swkey->ip.proto == IPPROTO_ICMP) {
1328                         struct ovs_key_icmp *icmp_key;
1329
1330                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1331                         if (!nla)
1332                                 goto nla_put_failure;
1333                         icmp_key = nla_data(nla);
1334                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1335                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1336                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1337                            swkey->ip.proto == IPPROTO_ICMPV6) {
1338                         struct ovs_key_icmpv6 *icmpv6_key;
1339
1340                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1341                                                 sizeof(*icmpv6_key));
1342                         if (!nla)
1343                                 goto nla_put_failure;
1344                         icmpv6_key = nla_data(nla);
1345                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1346                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1347
1348                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1349                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1350                                 struct ovs_key_nd *nd_key;
1351
1352                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1353                                 if (!nla)
1354                                         goto nla_put_failure;
1355                                 nd_key = nla_data(nla);
1356                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1357                                                         sizeof(nd_key->nd_target));
1358                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1359                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1360                         }
1361                 }
1362         }
1363
1364         return 0;
1365
1366 nla_put_failure:
1367         return -EMSGSIZE;
1368 }
1369
1370 /* Initializes the flow module.
1371  * Returns zero if successful or a negative error code. */
1372 int flow_init(void)
1373 {
1374         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1375                                         0, NULL);
1376         if (flow_cache == NULL)
1377                 return -ENOMEM;
1378
1379         get_random_bytes(&hash_seed, sizeof(hash_seed));
1380
1381         return 0;
1382 }
1383
1384 /* Uninitializes the flow module. */
1385 void flow_exit(void)
1386 {
1387         kmem_cache_destroy(flow_cache);
1388 }