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/ip.h>
  34 #include <net/ipv6.h>
  35 #include <net/ndisc.h>
  36
  37 #include "vlan.h"
  38
  39 static struct kmem_cache *flow_cache;
  40 static unsigned int hash_seed __read_mostly;
  41
  42 static int check_header(struct sk_buff *skb, int len)
  43 {
  44         if (unlikely(skb->len < len))
  45                 return -EINVAL;
  46         if (unlikely(!pskb_may_pull(skb, len)))
  47                 return -ENOMEM;
  48         return 0;
  49 }
  50
  51 static bool arphdr_ok(struct sk_buff *skb)
  52 {
  53         return pskb_may_pull(skb, skb_network_offset(skb) +
  54                                   sizeof(struct arp_eth_header));
  55 }
  56
  57 static int check_iphdr(struct sk_buff *skb)
  58 {
  59         unsigned int nh_ofs = skb_network_offset(skb);
  60         unsigned int ip_len;
  61         int err;
  62
  63         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  64         if (unlikely(err))
  65                 return err;
  66
  67         ip_len = ip_hdrlen(skb);
  68         if (unlikely(ip_len < sizeof(struct iphdr) ||
  69                      skb->len < nh_ofs + ip_len))
  70                 return -EINVAL;
  71
  72         skb_set_transport_header(skb, nh_ofs + ip_len);
  73         return 0;
  74 }
  75
  76 static bool tcphdr_ok(struct sk_buff *skb)
  77 {
  78         int th_ofs = skb_transport_offset(skb);
  79         int tcp_len;
  80
  81         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  82                 return false;
  83
  84         tcp_len = tcp_hdrlen(skb);
  85         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  86                      skb->len < th_ofs + tcp_len))
  87                 return false;
  88
  89         return true;
  90 }
  91
  92 static bool udphdr_ok(struct sk_buff *skb)
  93 {
  94         return pskb_may_pull(skb, skb_transport_offset(skb) +
  95                                   sizeof(struct udphdr));
  96 }
  97
  98 static bool icmphdr_ok(struct sk_buff *skb)
  99 {
 100         return pskb_may_pull(skb, skb_transport_offset(skb) +
 101                                   sizeof(struct icmphdr));
 102 }
 103
 104 u64 flow_used_time(unsigned long flow_jiffies)
 105 {
 106         struct timespec cur_ts;
 107         u64 cur_ms, idle_ms;
 108
 109         ktime_get_ts(&cur_ts);
 110         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 111         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 112                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 113
 114         return cur_ms - idle_ms;
 115 }
 116
 117 #define SW_FLOW_KEY_OFFSET(field)               \
 118         (offsetof(struct sw_flow_key, field) +  \
 119          FIELD_SIZEOF(struct sw_flow_key, field))
 120
 121 /**
 122  * skip_exthdr - skip any IPv6 extension headers
 123  * @skb: skbuff to parse
 124  * @start: offset of first extension header
 125  * @nexthdrp: Initially, points to the type of the extension header at @start.
 126  * This function updates it to point to the extension header at the final
 127  * offset.
 128  * @frag: Points to the @frag member in a &struct sw_flow_key.  This
 129  * function sets an appropriate %OVS_FRAG_TYPE_* value.
 130  *
 131  * This is based on ipv6_skip_exthdr() but adds the updates to *@frag.
 132  *
 133  * When there is more than one fragment header, this version reports whether
 134  * the final fragment header that it examines is a first fragment.
 135  *
 136  * Returns the final payload offset, or -1 on error.
 137  */
 138 static int skip_exthdr(const struct sk_buff *skb, int start, u8 *nexthdrp,
 139                        u8 *frag)
 140 {
 141         u8 nexthdr = *nexthdrp;
 142
 143         while (ipv6_ext_hdr(nexthdr)) {
 144                 struct ipv6_opt_hdr _hdr, *hp;
 145                 int hdrlen;
 146
 147                 if (nexthdr == NEXTHDR_NONE)
 148                         return -1;
 149                 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
 150                 if (hp == NULL)
 151                         return -1;
 152                 if (nexthdr == NEXTHDR_FRAGMENT) {
 153                         __be16 _frag_off, *fp;
 154                         fp = skb_header_pointer(skb,
 155                                                 start+offsetof(struct frag_hdr,
 156                                                                frag_off),
 157                                                 sizeof(_frag_off),
 158                                                 &_frag_off);
 159                         if (fp == NULL)
 160                                 return -1;
 161
 162                         if (ntohs(*fp) & ~0x7) {
 163                                 *frag = OVS_FRAG_TYPE_LATER;
 164                                 break;
 165                         }
 166                         *frag = OVS_FRAG_TYPE_FIRST;
 167                         hdrlen = 8;
 168                 } else if (nexthdr == NEXTHDR_AUTH)
 169                         hdrlen = (hp->hdrlen+2)<<2;
 170                 else
 171                         hdrlen = ipv6_optlen(hp);
 172
 173                 nexthdr = hp->nexthdr;
 174                 start += hdrlen;
 175         }
 176
 177         *nexthdrp = nexthdr;
 178         return start;
 179 }
 180
 181 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 182                          int *key_lenp)
 183 {
 184         unsigned int nh_ofs = skb_network_offset(skb);
 185         unsigned int nh_len;
 186         int payload_ofs;
 187         struct ipv6hdr *nh;
 188         uint8_t nexthdr;
 189         int err;
 190
 191         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 192
 193         err = check_header(skb, nh_ofs + sizeof(*nh));
 194         if (unlikely(err))
 195                 return err;
 196
 197         nh = ipv6_hdr(skb);
 198         nexthdr = nh->nexthdr;
 199         payload_ofs = (u8 *)(nh + 1) - skb->data;
 200
 201         key->ip.proto = NEXTHDR_NONE;
 202         key->ip.tos = ipv6_get_dsfield(nh);
 203         key->ip.ttl = nh->hop_limit;
 204         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 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, &nexthdr, &key->ip.frag);
 209         if (unlikely(payload_ofs < 0))
 210                 return -EINVAL;
 211
 212         nh_len = payload_ofs - nh_ofs;
 213         skb_set_transport_header(skb, nh_ofs + nh_len);
 214         key->ip.proto = nexthdr;
 215         return nh_len;
 216 }
 217
 218 static bool icmp6hdr_ok(struct sk_buff *skb)
 219 {
 220         return pskb_may_pull(skb, skb_transport_offset(skb) +
 221                                   sizeof(struct icmp6hdr));
 222 }
 223
 224 #define TCP_FLAGS_OFFSET 13
 225 #define TCP_FLAG_MASK 0x3f
 226
 227 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 228 {
 229         u8 tcp_flags = 0;
 230
 231         if (flow->key.eth.type == htons(ETH_P_IP) &&
 232             flow->key.ip.proto == IPPROTO_TCP) {
 233                 u8 *tcp = (u8 *)tcp_hdr(skb);
 234                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 235         }
 236
 237         spin_lock(&flow->lock);
 238         flow->used = jiffies;
 239         flow->packet_count++;
 240         flow->byte_count += skb->len;
 241         flow->tcp_flags |= tcp_flags;
 242         spin_unlock(&flow->lock);
 243 }
 244
 245 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 246 {
 247         int actions_len = nla_len(actions);
 248         struct sw_flow_actions *sfa;
 249
 250         /* At least DP_MAX_PORTS actions are required to be able to flood a
 251          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 252          * etc. */
 253         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 254                 return ERR_PTR(-EINVAL);
 255
 256         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 257         if (!sfa)
 258                 return ERR_PTR(-ENOMEM);
 259
 260         sfa->actions_len = actions_len;
 261         memcpy(sfa->actions, nla_data(actions), actions_len);
 262         return sfa;
 263 }
 264
 265 struct sw_flow *flow_alloc(void)
 266 {
 267         struct sw_flow *flow;
 268
 269         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 270         if (!flow)
 271                 return ERR_PTR(-ENOMEM);
 272
 273         spin_lock_init(&flow->lock);
 274         atomic_set(&flow->refcnt, 1);
 275         flow->sf_acts = NULL;
 276         flow->dead = false;
 277
 278         return flow;
 279 }
 280
 281 static struct hlist_head __rcu *find_bucket(struct flow_table * table, u32 hash)
 282 {
 283         return flex_array_get(table->buckets,
 284                                 (hash & (table->n_buckets - 1)));
 285 }
 286
 287 static struct flex_array  __rcu *alloc_buckets(unsigned int n_buckets)
 288 {
 289         struct flex_array  __rcu *buckets;
 290         int i, err;
 291
 292         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 293                                    n_buckets, GFP_KERNEL);
 294         if (!buckets)
 295                 return NULL;
 296
 297         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 298         if (err) {
 299                 flex_array_free(buckets);
 300                 return NULL;
 301         }
 302
 303         for (i = 0; i < n_buckets; i++)
 304                 INIT_HLIST_HEAD((struct hlist_head *)
 305                                         flex_array_get(buckets, i));
 306
 307         return buckets;
 308 }
 309
 310 static void free_buckets(struct flex_array *buckets)
 311 {
 312         flex_array_free(buckets);
 313 }
 314
 315 struct flow_table *flow_tbl_alloc(int new_size)
 316 {
 317         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 318
 319         if (!table)
 320                 return NULL;
 321
 322         table->buckets = alloc_buckets(new_size);
 323
 324         if (!table->buckets) {
 325                 kfree(table);
 326                 return NULL;
 327         }
 328         table->n_buckets = new_size;
 329         table->count = 0;
 330
 331         return table;
 332 }
 333
 334 static void flow_free(struct sw_flow *flow)
 335 {
 336         flow->dead = true;
 337         flow_put(flow);
 338 }
 339
 340 void flow_tbl_destroy(struct flow_table *table)
 341 {
 342         int i;
 343
 344         if (!table)
 345                 return;
 346
 347         for (i = 0; i < table->n_buckets; i++) {
 348                 struct sw_flow *flow;
 349                 struct hlist_head *head = flex_array_get(table->buckets, i);
 350                 struct hlist_node *node, *n;
 351
 352                 hlist_for_each_entry_safe(flow, node, n, head, hash_node) {
 353                         hlist_del_init_rcu(&flow->hash_node);
 354                         flow_free(flow);
 355                 }
 356         }
 357
 358         free_buckets(table->buckets);
 359         kfree(table);
 360 }
 361
 362 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 363 {
 364         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 365
 366         flow_tbl_destroy(table);
 367 }
 368
 369 void flow_tbl_deferred_destroy(struct flow_table *table)
 370 {
 371         if (!table)
 372                 return;
 373
 374         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 375 }
 376
 377 struct sw_flow *flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 378 {
 379         struct sw_flow *flow;
 380         struct hlist_head *head;
 381         struct hlist_node *n;
 382         int i;
 383
 384         while (*bucket < table->n_buckets) {
 385                 i = 0;
 386                 head = flex_array_get(table->buckets, *bucket);
 387                 hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 388                         if (i < *last) {
 389                                 i++;
 390                                 continue;
 391                         }
 392                         *last = i + 1;
 393                         return flow;
 394                 }
 395                 (*bucket)++;
 396                 *last = 0;
 397         }
 398
 399         return NULL;
 400 }
 401
 402 struct flow_table *flow_tbl_expand(struct flow_table *table)
 403 {
 404         struct flow_table *new_table;
 405         int n_buckets = table->n_buckets * 2;
 406         int i;
 407
 408         new_table = flow_tbl_alloc(n_buckets);
 409         if (!new_table)
 410                 return ERR_PTR(-ENOMEM);
 411
 412         for (i = 0; i < table->n_buckets; i++) {
 413                 struct sw_flow *flow;
 414                 struct hlist_head *head;
 415                 struct hlist_node *n, *pos;
 416
 417                 head = flex_array_get(table->buckets, i);
 418
 419                 hlist_for_each_entry_safe(flow, n, pos, head, hash_node) {
 420                         hlist_del_init_rcu(&flow->hash_node);
 421                         flow_tbl_insert(new_table, flow);
 422                 }
 423         }
 424
 425         return new_table;
 426 }
 427
 428 /* RCU callback used by flow_deferred_free. */
 429 static void rcu_free_flow_callback(struct rcu_head *rcu)
 430 {
 431         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 432
 433         flow->dead = true;
 434         flow_put(flow);
 435 }
 436
 437 /* Schedules 'flow' to be freed after the next RCU grace period.
 438  * The caller must hold rcu_read_lock for this to be sensible. */
 439 void flow_deferred_free(struct sw_flow *flow)
 440 {
 441         call_rcu(&flow->rcu, rcu_free_flow_callback);
 442 }
 443
 444 void flow_hold(struct sw_flow *flow)
 445 {
 446         atomic_inc(&flow->refcnt);
 447 }
 448
 449 void flow_put(struct sw_flow *flow)
 450 {
 451         if (unlikely(!flow))
 452                 return;
 453
 454         if (atomic_dec_and_test(&flow->refcnt)) {
 455                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 456                 kmem_cache_free(flow_cache, flow);
 457         }
 458 }
 459
 460 /* RCU callback used by flow_deferred_free_acts. */
 461 static void rcu_free_acts_callback(struct rcu_head *rcu)
 462 {
 463         struct sw_flow_actions *sf_acts = container_of(rcu,
 464                         struct sw_flow_actions, rcu);
 465         kfree(sf_acts);
 466 }
 467
 468 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 469  * The caller must hold rcu_read_lock for this to be sensible. */
 470 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 471 {
 472         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 473 }
 474
 475 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 476 {
 477         struct qtag_prefix {
 478                 __be16 eth_type; /* ETH_P_8021Q */
 479                 __be16 tci;
 480         };
 481         struct qtag_prefix *qp;
 482
 483         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 484                 return 0;
 485
 486         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 487                                          sizeof(__be16))))
 488                 return -ENOMEM;
 489
 490         qp = (struct qtag_prefix *) skb->data;
 491         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 492         __skb_pull(skb, sizeof(struct qtag_prefix));
 493
 494         return 0;
 495 }
 496
 497 static __be16 parse_ethertype(struct sk_buff *skb)
 498 {
 499         struct llc_snap_hdr {
 500                 u8  dsap;  /* Always 0xAA */
 501                 u8  ssap;  /* Always 0xAA */
 502                 u8  ctrl;
 503                 u8  oui[3];
 504                 __be16 ethertype;
 505         };
 506         struct llc_snap_hdr *llc;
 507         __be16 proto;
 508
 509         proto = *(__be16 *) skb->data;
 510         __skb_pull(skb, sizeof(__be16));
 511
 512         if (ntohs(proto) >= 1536)
 513                 return proto;
 514
 515         if (skb->len < sizeof(struct llc_snap_hdr))
 516                 return htons(ETH_P_802_2);
 517
 518         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 519                 return htons(0);
 520
 521         llc = (struct llc_snap_hdr *) skb->data;
 522         if (llc->dsap != LLC_SAP_SNAP ||
 523             llc->ssap != LLC_SAP_SNAP ||
 524             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 525                 return htons(ETH_P_802_2);
 526
 527         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 528         return llc->ethertype;
 529 }
 530
 531 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 532                         int *key_lenp, int nh_len)
 533 {
 534         struct icmp6hdr *icmp = icmp6_hdr(skb);
 535         int error = 0;
 536         int key_len;
 537
 538         /* The ICMPv6 type and code fields use the 16-bit transport port
 539          * fields, so we need to store them in 16-bit network byte order.
 540          */
 541         key->ipv6.tp.src = htons(icmp->icmp6_type);
 542         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 543         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 544
 545         if (icmp->icmp6_code == 0 &&
 546             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 547              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 548                 int icmp_len = skb->len - skb_transport_offset(skb);
 549                 struct nd_msg *nd;
 550                 int offset;
 551
 552                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 553
 554                 /* In order to process neighbor discovery options, we need the
 555                  * entire packet.
 556                  */
 557                 if (unlikely(icmp_len < sizeof(*nd)))
 558                         goto out;
 559                 if (unlikely(skb_linearize(skb))) {
 560                         error = -ENOMEM;
 561                         goto out;
 562                 }
 563
 564                 nd = (struct nd_msg *)skb_transport_header(skb);
 565                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
 566                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 567
 568                 icmp_len -= sizeof(*nd);
 569                 offset = 0;
 570                 while (icmp_len >= 8) {
 571                         struct nd_opt_hdr *nd_opt =
 572                                  (struct nd_opt_hdr *)(nd->opt + offset);
 573                         int opt_len = nd_opt->nd_opt_len * 8;
 574
 575                         if (unlikely(!opt_len || opt_len > icmp_len))
 576                                 goto invalid;
 577
 578                         /* Store the link layer address if the appropriate
 579                          * option is provided.  It is considered an error if
 580                          * the same link layer option is specified twice.
 581                          */
 582                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 583                             && opt_len == 8) {
 584                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 585                                         goto invalid;
 586                                 memcpy(key->ipv6.nd.sll,
 587                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 588                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 589                                    && opt_len == 8) {
 590                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 591                                         goto invalid;
 592                                 memcpy(key->ipv6.nd.tll,
 593                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 594                         }
 595
 596                         icmp_len -= opt_len;
 597                         offset += opt_len;
 598                 }
 599         }
 600
 601         goto out;
 602
 603 invalid:
 604         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 605         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 606         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 607
 608 out:
 609         *key_lenp = key_len;
 610         return error;
 611 }
 612
 613 /**
 614  * flow_extract - extracts a flow key from an Ethernet frame.
 615  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 616  * Ethernet header
 617  * @in_port: port number on which @skb was received.
 618  * @key: output flow key
 619  * @key_lenp: length of output flow key
 620  *
 621  * The caller must ensure that skb->len >= ETH_HLEN.
 622  *
 623  * Returns 0 if successful, otherwise a negative errno value.
 624  *
 625  * Initializes @skb header pointers as follows:
 626  *
 627  *    - skb->mac_header: the Ethernet header.
 628  *
 629  *    - skb->network_header: just past the Ethernet header, or just past the
 630  *      VLAN header, to the first byte of the Ethernet payload.
 631  *
 632  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 633  *      on output, then just past the IP header, if one is present and
 634  *      of a correct length, otherwise the same as skb->network_header.
 635  *      For other key->dl_type values it is left untouched.
 636  */
 637 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 638                  int *key_lenp)
 639 {
 640         int error = 0;
 641         int key_len = SW_FLOW_KEY_OFFSET(eth);
 642         struct ethhdr *eth;
 643
 644         memset(key, 0, sizeof(*key));
 645
 646         key->phy.priority = skb->priority;
 647         key->phy.tun_id = OVS_CB(skb)->tun_id;
 648         key->phy.in_port = in_port;
 649
 650         skb_reset_mac_header(skb);
 651
 652         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 653          * header in the linear data area.
 654          */
 655         eth = eth_hdr(skb);
 656         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 657         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 658
 659         __skb_pull(skb, 2 * ETH_ALEN);
 660
 661         if (vlan_tx_tag_present(skb))
 662                 key->eth.tci = htons(vlan_get_tci(skb));
 663         else if (eth->h_proto == htons(ETH_P_8021Q))
 664                 if (unlikely(parse_vlan(skb, key)))
 665                         return -ENOMEM;
 666
 667         key->eth.type = parse_ethertype(skb);
 668         if (unlikely(key->eth.type == htons(0)))
 669                 return -ENOMEM;
 670
 671         skb_reset_network_header(skb);
 672         __skb_push(skb, skb->data - skb_mac_header(skb));
 673
 674         /* Network layer. */
 675         if (key->eth.type == htons(ETH_P_IP)) {
 676                 struct iphdr *nh;
 677                 __be16 offset;
 678
 679                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 680
 681                 error = check_iphdr(skb);
 682                 if (unlikely(error)) {
 683                         if (error == -EINVAL) {
 684                                 skb->transport_header = skb->network_header;
 685                                 error = 0;
 686                         }
 687                         goto out;
 688                 }
 689
 690                 nh = ip_hdr(skb);
 691                 key->ipv4.addr.src = nh->saddr;
 692                 key->ipv4.addr.dst = nh->daddr;
 693
 694                 key->ip.proto = nh->protocol;
 695                 key->ip.tos = nh->tos;
 696                 key->ip.ttl = nh->ttl;
 697
 698                 offset = nh->frag_off & htons(IP_OFFSET);
 699                 if (offset) {
 700                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 701                         goto out;
 702                 }
 703                 if (nh->frag_off & htons(IP_MF) ||
 704                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 705                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 706
 707                 /* Transport layer. */
 708                 if (key->ip.proto == IPPROTO_TCP) {
 709                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 710                         if (tcphdr_ok(skb)) {
 711                                 struct tcphdr *tcp = tcp_hdr(skb);
 712                                 key->ipv4.tp.src = tcp->source;
 713                                 key->ipv4.tp.dst = tcp->dest;
 714                         }
 715                 } else if (key->ip.proto == IPPROTO_UDP) {
 716                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 717                         if (udphdr_ok(skb)) {
 718                                 struct udphdr *udp = udp_hdr(skb);
 719                                 key->ipv4.tp.src = udp->source;
 720                                 key->ipv4.tp.dst = udp->dest;
 721                         }
 722                 } else if (key->ip.proto == IPPROTO_ICMP) {
 723                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 724                         if (icmphdr_ok(skb)) {
 725                                 struct icmphdr *icmp = icmp_hdr(skb);
 726                                 /* The ICMP type and code fields use the 16-bit
 727                                  * transport port fields, so we need to store
 728                                  * them in 16-bit network byte order. */
 729                                 key->ipv4.tp.src = htons(icmp->type);
 730                                 key->ipv4.tp.dst = htons(icmp->code);
 731                         }
 732                 }
 733
 734         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 735                 struct arp_eth_header *arp;
 736
 737                 arp = (struct arp_eth_header *)skb_network_header(skb);
 738
 739                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 740                                 && arp->ar_pro == htons(ETH_P_IP)
 741                                 && arp->ar_hln == ETH_ALEN
 742                                 && arp->ar_pln == 4) {
 743
 744                         /* We only match on the lower 8 bits of the opcode. */
 745                         if (ntohs(arp->ar_op) <= 0xff)
 746                                 key->ip.proto = ntohs(arp->ar_op);
 747
 748                         if (key->ip.proto == ARPOP_REQUEST
 749                                         || key->ip.proto == ARPOP_REPLY) {
 750                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 751                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 752                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 753                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 754                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 755                         }
 756                 }
 757         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 758                 int nh_len;             /* IPv6 Header + Extensions */
 759
 760                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 761                 if (unlikely(nh_len < 0)) {
 762                         if (nh_len == -EINVAL)
 763                                 skb->transport_header = skb->network_header;
 764                         else
 765                                 error = nh_len;
 766                         goto out;
 767                 }
 768
 769                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 770                         goto out;
 771                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 772                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 773
 774                 /* Transport layer. */
 775                 if (key->ip.proto == NEXTHDR_TCP) {
 776                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 777                         if (tcphdr_ok(skb)) {
 778                                 struct tcphdr *tcp = tcp_hdr(skb);
 779                                 key->ipv6.tp.src = tcp->source;
 780                                 key->ipv6.tp.dst = tcp->dest;
 781                         }
 782                 } else if (key->ip.proto == NEXTHDR_UDP) {
 783                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 784                         if (udphdr_ok(skb)) {
 785                                 struct udphdr *udp = udp_hdr(skb);
 786                                 key->ipv6.tp.src = udp->source;
 787                                 key->ipv6.tp.dst = udp->dest;
 788                         }
 789                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 790                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 791                         if (icmp6hdr_ok(skb)) {
 792                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 793                                 if (error < 0)
 794                                         goto out;
 795                         }
 796                 }
 797         }
 798
 799 out:
 800         *key_lenp = key_len;
 801         return error;
 802 }
 803
 804 u32 flow_hash(const struct sw_flow_key *key, int key_len)
 805 {
 806         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
 807 }
 808
 809 struct sw_flow *flow_tbl_lookup(struct flow_table *table,
 810                                 struct sw_flow_key *key, int key_len)
 811 {
 812         struct sw_flow *flow;
 813         struct hlist_node *n;
 814         struct hlist_head *head;
 815         u32 hash;
 816
 817         hash = flow_hash(key, key_len);
 818
 819         head = find_bucket(table, hash);
 820         hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 821
 822                 if (flow->hash == hash &&
 823                     !memcmp(&flow->key, key, key_len)) {
 824                         return flow;
 825                 }
 826         }
 827         return NULL;
 828 }
 829
 830 void flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 831 {
 832         struct hlist_head *head;
 833
 834         head = find_bucket(table, flow->hash);
 835         hlist_add_head_rcu(&flow->hash_node, head);
 836         table->count++;
 837 }
 838
 839 void flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 840 {
 841         if (!hlist_unhashed(&flow->hash_node)) {
 842                 hlist_del_init_rcu(&flow->hash_node);
 843                 table->count--;
 844                 BUG_ON(table->count < 0);
 845         }
 846 }
 847
 848 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 849 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 850         [OVS_KEY_ATTR_ENCAP] = -1,
 851         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 852         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 853         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 854         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 855         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 856         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 857         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 858         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 859         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 860         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 861         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 862         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 863         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 864
 865         /* Not upstream. */
 866         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 867 };
 868
 869 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 870                                   const struct nlattr *a[], u64 *attrs)
 871 {
 872         const struct ovs_key_icmp *icmp_key;
 873         const struct ovs_key_tcp *tcp_key;
 874         const struct ovs_key_udp *udp_key;
 875
 876         switch (swkey->ip.proto) {
 877         case IPPROTO_TCP:
 878                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 879                         return -EINVAL;
 880                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 881
 882                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 883                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 884                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 885                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 886                 break;
 887
 888         case IPPROTO_UDP:
 889                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 890                         return -EINVAL;
 891                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 892
 893                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 894                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 895                 swkey->ipv4.tp.src = udp_key->udp_src;
 896                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 897                 break;
 898
 899         case IPPROTO_ICMP:
 900                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 901                         return -EINVAL;
 902                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 903
 904                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 905                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 906                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 907                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 908                 break;
 909         }
 910
 911         return 0;
 912 }
 913
 914 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 915                                   const struct nlattr *a[], u64 *attrs)
 916 {
 917         const struct ovs_key_icmpv6 *icmpv6_key;
 918         const struct ovs_key_tcp *tcp_key;
 919         const struct ovs_key_udp *udp_key;
 920
 921         switch (swkey->ip.proto) {
 922         case IPPROTO_TCP:
 923                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 924                         return -EINVAL;
 925                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 926
 927                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 928                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 929                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 930                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 931                 break;
 932
 933         case IPPROTO_UDP:
 934                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 935                         return -EINVAL;
 936                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 937
 938                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 939                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 940                 swkey->ipv6.tp.src = udp_key->udp_src;
 941                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 942                 break;
 943
 944         case IPPROTO_ICMPV6:
 945                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 946                         return -EINVAL;
 947                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 948
 949                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 950                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 951                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 952                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 953
 954                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 955                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 956                         const struct ovs_key_nd *nd_key;
 957
 958                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 959                                 return -EINVAL;
 960                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 961
 962                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 963                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 964                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 965                                sizeof(swkey->ipv6.nd.target));
 966                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 967                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 968                 }
 969                 break;
 970         }
 971
 972         return 0;
 973 }
 974
 975 static int parse_flow_nlattrs(const struct nlattr *attr,
 976                               const struct nlattr *a[], u64 *attrsp)
 977 {
 978         const struct nlattr *nla;
 979         u64 attrs;
 980         int rem;
 981
 982         attrs = 0;
 983         nla_for_each_nested(nla, attr, rem) {
 984                 u16 type = nla_type(nla);
 985                 int expected_len;
 986
 987                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 988                         return -EINVAL;
 989
 990                 expected_len = ovs_key_lens[type];
 991                 if (nla_len(nla) != expected_len && expected_len != -1)
 992                         return -EINVAL;
 993
 994                 attrs |= 1ULL << type;
 995                 a[type] = nla;
 996         }
 997         if (rem)
 998                 return -EINVAL;
 999
1000         *attrsp = attrs;
1001         return 0;
1002 }
1003
1004 /**
1005  * flow_from_nlattrs - parses Netlink attributes into a flow key.
1006  * @swkey: receives the extracted flow key.
1007  * @key_lenp: number of bytes used in @swkey.
1008  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1009  * sequence.
1010  */
1011 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1012                       const struct nlattr *attr)
1013 {
1014         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1015         const struct ovs_key_ethernet *eth_key;
1016         int key_len;
1017         u64 attrs;
1018         int err;
1019
1020         memset(swkey, 0, sizeof(struct sw_flow_key));
1021         key_len = SW_FLOW_KEY_OFFSET(eth);
1022
1023         err = parse_flow_nlattrs(attr, a, &attrs);
1024         if (err)
1025                 return err;
1026
1027         /* Metadata attributes. */
1028         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1029                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1030                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1031         }
1032         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1033                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1034                 if (in_port >= DP_MAX_PORTS)
1035                         return -EINVAL;
1036                 swkey->phy.in_port = in_port;
1037                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1038         } else {
1039                 swkey->phy.in_port = USHRT_MAX;
1040         }
1041
1042         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
1043                 swkey->phy.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1044                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1045         }
1046
1047         /* Data attributes. */
1048         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1049                 return -EINVAL;
1050         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1051
1052         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1053         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1054         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1055
1056         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1057             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1058                 const struct nlattr *encap;
1059                 __be16 tci;
1060
1061                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1062                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1063                               (1 << OVS_KEY_ATTR_ENCAP)))
1064                         return -EINVAL;
1065
1066                 encap = a[OVS_KEY_ATTR_ENCAP];
1067                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1068                 if (tci & htons(VLAN_TAG_PRESENT)) {
1069                         swkey->eth.tci = tci;
1070
1071                         err = parse_flow_nlattrs(encap, a, &attrs);
1072                         if (err)
1073                                 return err;
1074                 } else if (!tci) {
1075                         /* Corner case for truncated 802.1Q header. */
1076                         if (nla_len(encap))
1077                                 return -EINVAL;
1078
1079                         swkey->eth.type = htons(ETH_P_8021Q);
1080                         *key_lenp = key_len;
1081                         return 0;
1082                 } else {
1083                         return -EINVAL;
1084                 }
1085         }
1086
1087         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1088                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1089                 if (ntohs(swkey->eth.type) < 1536)
1090                         return -EINVAL;
1091                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1092         } else {
1093                 swkey->eth.type = htons(ETH_P_802_2);
1094         }
1095
1096         if (swkey->eth.type == htons(ETH_P_IP)) {
1097                 const struct ovs_key_ipv4 *ipv4_key;
1098
1099                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1100                         return -EINVAL;
1101                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1102
1103                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1104                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1105                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1106                         return -EINVAL;
1107                 swkey->ip.proto = ipv4_key->ipv4_proto;
1108                 swkey->ip.tos = ipv4_key->ipv4_tos;
1109                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1110                 swkey->ip.frag = ipv4_key->ipv4_frag;
1111                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1112                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1113
1114                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1115                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1116                         if (err)
1117                                 return err;
1118                 }
1119         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1120                 const struct ovs_key_ipv6 *ipv6_key;
1121
1122                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1123                         return -EINVAL;
1124                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1125
1126                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1127                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1128                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1129                         return -EINVAL;
1130                 swkey->ipv6.label = ipv6_key->ipv6_label;
1131                 swkey->ip.proto = ipv6_key->ipv6_proto;
1132                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1133                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1134                 swkey->ip.frag = ipv6_key->ipv6_frag;
1135                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1136                        sizeof(swkey->ipv6.addr.src));
1137                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1138                        sizeof(swkey->ipv6.addr.dst));
1139
1140                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1141                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1142                         if (err)
1143                                 return err;
1144                 }
1145         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1146                 const struct ovs_key_arp *arp_key;
1147
1148                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1149                         return -EINVAL;
1150                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1151
1152                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1153                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1154                 swkey->ipv4.addr.src = arp_key->arp_sip;
1155                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1156                 if (arp_key->arp_op & htons(0xff00))
1157                         return -EINVAL;
1158                 swkey->ip.proto = ntohs(arp_key->arp_op);
1159                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1160                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1161         }
1162
1163         if (attrs)
1164                 return -EINVAL;
1165         *key_lenp = key_len;
1166
1167         return 0;
1168 }
1169
1170 /**
1171  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1172  * @in_port: receives the extracted input port.
1173  * @tun_id: receives the extracted tunnel ID.
1174  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1175  * sequence.
1176  *
1177  * This parses a series of Netlink attributes that form a flow key, which must
1178  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1179  * get the metadata, that is, the parts of the flow key that cannot be
1180  * extracted from the packet itself.
1181  */
1182 int flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1183                                const struct nlattr *attr)
1184 {
1185         const struct nlattr *nla;
1186         int rem;
1187
1188         *in_port = USHRT_MAX;
1189         *tun_id = 0;
1190         *priority = 0;
1191
1192         nla_for_each_nested(nla, attr, rem) {
1193                 int type = nla_type(nla);
1194
1195                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1196                         if (nla_len(nla) != ovs_key_lens[type])
1197                                 return -EINVAL;
1198
1199                         switch (type) {
1200                         case OVS_KEY_ATTR_PRIORITY:
1201                                 *priority = nla_get_u32(nla);
1202                                 break;
1203
1204                         case OVS_KEY_ATTR_TUN_ID:
1205                                 *tun_id = nla_get_be64(nla);
1206                                 break;
1207
1208                         case OVS_KEY_ATTR_IN_PORT:
1209                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1210                                         return -EINVAL;
1211                                 *in_port = nla_get_u32(nla);
1212                                 break;
1213                         }
1214                 }
1215         }
1216         if (rem)
1217                 return -EINVAL;
1218         return 0;
1219 }
1220
1221 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1222 {
1223         struct ovs_key_ethernet *eth_key;
1224         struct nlattr *nla, *encap;
1225
1226         if (swkey->phy.priority)
1227                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1228
1229         if (swkey->phy.tun_id != cpu_to_be64(0))
1230                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1231
1232         if (swkey->phy.in_port != USHRT_MAX)
1233                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1234
1235         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1236         if (!nla)
1237                 goto nla_put_failure;
1238         eth_key = nla_data(nla);
1239         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1240         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1241
1242         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1243                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q));
1244                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci);
1245                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1246                 if (!swkey->eth.tci)
1247                         goto unencap;
1248         } else {
1249                 encap = NULL;
1250         }
1251
1252         if (swkey->eth.type == htons(ETH_P_802_2))
1253                 goto unencap;
1254
1255         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1256
1257         if (swkey->eth.type == htons(ETH_P_IP)) {
1258                 struct ovs_key_ipv4 *ipv4_key;
1259
1260                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1261                 if (!nla)
1262                         goto nla_put_failure;
1263                 ipv4_key = nla_data(nla);
1264                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1265                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1266                 ipv4_key->ipv4_proto = swkey->ip.proto;
1267                 ipv4_key->ipv4_tos = swkey->ip.tos;
1268                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1269                 ipv4_key->ipv4_frag = swkey->ip.frag;
1270         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1271                 struct ovs_key_ipv6 *ipv6_key;
1272
1273                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1274                 if (!nla)
1275                         goto nla_put_failure;
1276                 ipv6_key = nla_data(nla);
1277                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1278                                 sizeof(ipv6_key->ipv6_src));
1279                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1280                                 sizeof(ipv6_key->ipv6_dst));
1281                 ipv6_key->ipv6_label = swkey->ipv6.label;
1282                 ipv6_key->ipv6_proto = swkey->ip.proto;
1283                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1284                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1285                 ipv6_key->ipv6_frag = swkey->ip.frag;
1286         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1287                 struct ovs_key_arp *arp_key;
1288
1289                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1290                 if (!nla)
1291                         goto nla_put_failure;
1292                 arp_key = nla_data(nla);
1293                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1294                 arp_key->arp_sip = swkey->ipv4.addr.src;
1295                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1296                 arp_key->arp_op = htons(swkey->ip.proto);
1297                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1298                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1299         }
1300
1301         if ((swkey->eth.type == htons(ETH_P_IP) ||
1302              swkey->eth.type == htons(ETH_P_IPV6)) &&
1303              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1304
1305                 if (swkey->ip.proto == IPPROTO_TCP) {
1306                         struct ovs_key_tcp *tcp_key;
1307
1308                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1309                         if (!nla)
1310                                 goto nla_put_failure;
1311                         tcp_key = nla_data(nla);
1312                         if (swkey->eth.type == htons(ETH_P_IP)) {
1313                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1314                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1315                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1316                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1317                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1318                         }
1319                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1320                         struct ovs_key_udp *udp_key;
1321
1322                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1323                         if (!nla)
1324                                 goto nla_put_failure;
1325                         udp_key = nla_data(nla);
1326                         if (swkey->eth.type == htons(ETH_P_IP)) {
1327                                 udp_key->udp_src = swkey->ipv4.tp.src;
1328                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1329                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1330                                 udp_key->udp_src = swkey->ipv6.tp.src;
1331                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1332                         }
1333                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1334                            swkey->ip.proto == IPPROTO_ICMP) {
1335                         struct ovs_key_icmp *icmp_key;
1336
1337                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1338                         if (!nla)
1339                                 goto nla_put_failure;
1340                         icmp_key = nla_data(nla);
1341                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1342                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1343                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1344                            swkey->ip.proto == IPPROTO_ICMPV6) {
1345                         struct ovs_key_icmpv6 *icmpv6_key;
1346
1347                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1348                                                 sizeof(*icmpv6_key));
1349                         if (!nla)
1350                                 goto nla_put_failure;
1351                         icmpv6_key = nla_data(nla);
1352                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1353                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1354
1355                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1356                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1357                                 struct ovs_key_nd *nd_key;
1358
1359                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1360                                 if (!nla)
1361                                         goto nla_put_failure;
1362                                 nd_key = nla_data(nla);
1363                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1364                                                         sizeof(nd_key->nd_target));
1365                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1366                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1367                         }
1368                 }
1369         }
1370
1371 unencap:
1372         if (encap)
1373                 nla_nest_end(skb, encap);
1374
1375         return 0;
1376
1377 nla_put_failure:
1378         return -EMSGSIZE;
1379 }
1380
1381 /* Initializes the flow module.
1382  * Returns zero if successful or a negative error code. */
1383 int flow_init(void)
1384 {
1385         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1386                                         0, NULL);
1387         if (flow_cache == NULL)
1388                 return -ENOMEM;
1389
1390         get_random_bytes(&hash_seed, sizeof(hash_seed));
1391
1392         return 0;
1393 }
1394
1395 /* Uninitializes the flow module. */
1396 void flow_exit(void)
1397 {
1398         kmem_cache_destroy(flow_cache);
1399 }