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(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 484                                          sizeof(__be16))))
 485                 return -ENOMEM;
 486
 487         qp = (struct qtag_prefix *) skb->data;
 488         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 489         __skb_pull(skb, sizeof(struct qtag_prefix));
 490
 491         return 0;
 492 }
 493
 494 static __be16 parse_ethertype(struct sk_buff *skb)
 495 {
 496         struct llc_snap_hdr {
 497                 u8  dsap;  /* Always 0xAA */
 498                 u8  ssap;  /* Always 0xAA */
 499                 u8  ctrl;
 500                 u8  oui[3];
 501                 __be16 ethertype;
 502         };
 503         struct llc_snap_hdr *llc;
 504         __be16 proto;
 505
 506         proto = *(__be16 *) skb->data;
 507         __skb_pull(skb, sizeof(__be16));
 508
 509         if (ntohs(proto) >= 1536)
 510                 return proto;
 511
 512         if (skb->len < sizeof(struct llc_snap_hdr))
 513                 return htons(ETH_P_802_2);
 514
 515         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 516                 return htons(0);
 517
 518         llc = (struct llc_snap_hdr *) skb->data;
 519         if (llc->dsap != LLC_SAP_SNAP ||
 520             llc->ssap != LLC_SAP_SNAP ||
 521             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 522                 return htons(ETH_P_802_2);
 523
 524         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 525         return llc->ethertype;
 526 }
 527
 528 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 529                         int *key_lenp, int nh_len)
 530 {
 531         struct icmp6hdr *icmp = icmp6_hdr(skb);
 532         int error = 0;
 533         int key_len;
 534
 535         /* The ICMPv6 type and code fields use the 16-bit transport port
 536          * fields, so we need to store them in 16-bit network byte order.
 537          */
 538         key->ipv6.tp.src = htons(icmp->icmp6_type);
 539         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 540         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 541
 542         if (icmp->icmp6_code == 0 &&
 543             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 544              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 545                 int icmp_len = skb->len - skb_transport_offset(skb);
 546                 struct nd_msg *nd;
 547                 int offset;
 548
 549                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 550
 551                 /* In order to process neighbor discovery options, we need the
 552                  * entire packet.
 553                  */
 554                 if (unlikely(icmp_len < sizeof(*nd)))
 555                         goto out;
 556                 if (unlikely(skb_linearize(skb))) {
 557                         error = -ENOMEM;
 558                         goto out;
 559                 }
 560
 561                 nd = (struct nd_msg *)skb_transport_header(skb);
 562                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
 563                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 564
 565                 icmp_len -= sizeof(*nd);
 566                 offset = 0;
 567                 while (icmp_len >= 8) {
 568                         struct nd_opt_hdr *nd_opt =
 569                                  (struct nd_opt_hdr *)(nd->opt + offset);
 570                         int opt_len = nd_opt->nd_opt_len * 8;
 571
 572                         if (unlikely(!opt_len || opt_len > icmp_len))
 573                                 goto invalid;
 574
 575                         /* Store the link layer address if the appropriate
 576                          * option is provided.  It is considered an error if
 577                          * the same link layer option is specified twice.
 578                          */
 579                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 580                             && opt_len == 8) {
 581                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 582                                         goto invalid;
 583                                 memcpy(key->ipv6.nd.sll,
 584                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 585                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 586                                    && opt_len == 8) {
 587                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 588                                         goto invalid;
 589                                 memcpy(key->ipv6.nd.tll,
 590                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 591                         }
 592
 593                         icmp_len -= opt_len;
 594                         offset += opt_len;
 595                 }
 596         }
 597
 598         goto out;
 599
 600 invalid:
 601         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 602         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 603         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 604
 605 out:
 606         *key_lenp = key_len;
 607         return error;
 608 }
 609
 610 /**
 611  * flow_extract - extracts a flow key from an Ethernet frame.
 612  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 613  * Ethernet header
 614  * @in_port: port number on which @skb was received.
 615  * @key: output flow key
 616  * @key_lenp: length of output flow key
 617  *
 618  * The caller must ensure that skb->len >= ETH_HLEN.
 619  *
 620  * Returns 0 if successful, otherwise a negative errno value.
 621  *
 622  * Initializes @skb header pointers as follows:
 623  *
 624  *    - skb->mac_header: the Ethernet header.
 625  *
 626  *    - skb->network_header: just past the Ethernet header, or just past the
 627  *      VLAN header, to the first byte of the Ethernet payload.
 628  *
 629  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 630  *      on output, then just past the IP header, if one is present and
 631  *      of a correct length, otherwise the same as skb->network_header.
 632  *      For other key->dl_type values it is left untouched.
 633  */
 634 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 635                  int *key_lenp)
 636 {
 637         int error = 0;
 638         int key_len = SW_FLOW_KEY_OFFSET(eth);
 639         struct ethhdr *eth;
 640
 641         memset(key, 0, sizeof(*key));
 642
 643         key->phy.priority = skb->priority;
 644         key->phy.tun_id = OVS_CB(skb)->tun_id;
 645         key->phy.in_port = in_port;
 646
 647         skb_reset_mac_header(skb);
 648
 649         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 650          * header in the linear data area.
 651          */
 652         eth = eth_hdr(skb);
 653         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 654         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 655
 656         __skb_pull(skb, 2 * ETH_ALEN);
 657
 658         if (vlan_tx_tag_present(skb))
 659                 key->eth.tci = htons(vlan_get_tci(skb));
 660         else if (eth->h_proto == htons(ETH_P_8021Q))
 661                 if (unlikely(parse_vlan(skb, key)))
 662                         return -ENOMEM;
 663
 664         key->eth.type = parse_ethertype(skb);
 665         if (unlikely(key->eth.type == htons(0)))
 666                 return -ENOMEM;
 667
 668         skb_reset_network_header(skb);
 669         __skb_push(skb, skb->data - skb_mac_header(skb));
 670
 671         /* Network layer. */
 672         if (key->eth.type == htons(ETH_P_IP)) {
 673                 struct iphdr *nh;
 674                 __be16 offset;
 675
 676                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 677
 678                 error = check_iphdr(skb);
 679                 if (unlikely(error)) {
 680                         if (error == -EINVAL) {
 681                                 skb->transport_header = skb->network_header;
 682                                 error = 0;
 683                         }
 684                         goto out;
 685                 }
 686
 687                 nh = ip_hdr(skb);
 688                 key->ipv4.addr.src = nh->saddr;
 689                 key->ipv4.addr.dst = nh->daddr;
 690
 691                 key->ip.proto = nh->protocol;
 692                 key->ip.tos = nh->tos;
 693                 key->ip.ttl = nh->ttl;
 694
 695                 offset = nh->frag_off & htons(IP_OFFSET);
 696                 if (offset) {
 697                         key->ip.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.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.frag == OVS_FRAG_TYPE_LATER)
 767                         goto out;
 768                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 769                         key->ip.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 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 846 const u32 ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 847         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 848         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 849         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 850         [OVS_KEY_ATTR_8021Q] = sizeof(struct ovs_key_8021q),
 851         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 852         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 853         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 854         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 855         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 856         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 857         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 858         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 859         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 860
 861         /* Not upstream. */
 862         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 863 };
 864
 865 /**
 866  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 867  * @swkey: receives the extracted flow key.
 868  * @key_lenp: number of bytes used in @swkey.
 869  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 870  * sequence.
 871  *
 872  * This state machine accepts the following forms, with [] for optional
 873  * elements and | for alternatives:
 874  *
 875  * [priority] [tun_id] [in_port] ethernet [8021q] [ethertype \
 876  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 877  *
 878  * except that IPv4 or IPv6 terminates the sequence if its @ipv4_frag or
 879  * @ipv6_frag member, respectively, equals %OVS_FRAG_TYPE_LATER.
 880  */
 881 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 882                       const struct nlattr *attr)
 883 {
 884         int error = 0;
 885         const struct nlattr *nla;
 886         u16 prev_type;
 887         int rem;
 888         int key_len;
 889
 890         memset(swkey, 0, sizeof(*swkey));
 891         swkey->phy.in_port = USHRT_MAX;
 892         swkey->eth.type = htons(ETH_P_802_2);
 893         key_len = SW_FLOW_KEY_OFFSET(eth);
 894
 895         prev_type = OVS_KEY_ATTR_UNSPEC;
 896         nla_for_each_nested(nla, attr, rem) {
 897                 const struct ovs_key_ethernet *eth_key;
 898                 const struct ovs_key_8021q *q_key;
 899                 const struct ovs_key_ipv4 *ipv4_key;
 900                 const struct ovs_key_ipv6 *ipv6_key;
 901                 const struct ovs_key_tcp *tcp_key;
 902                 const struct ovs_key_udp *udp_key;
 903                 const struct ovs_key_icmp *icmp_key;
 904                 const struct ovs_key_icmpv6 *icmpv6_key;
 905                 const struct ovs_key_arp *arp_key;
 906                 const struct ovs_key_nd *nd_key;
 907
 908                 int type = nla_type(nla);
 909
 910                 if (type > OVS_KEY_ATTR_MAX ||
 911                     nla_len(nla) != ovs_key_lens[type])
 912                         goto invalid;
 913
 914 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 915                 switch (TRANSITION(prev_type, type)) {
 916                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_PRIORITY):
 917                         swkey->phy.priority = nla_get_u32(nla);
 918                         break;
 919
 920                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
 921                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_TUN_ID):
 922                         swkey->phy.tun_id = nla_get_be64(nla);
 923                         break;
 924
 925                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
 926                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_IN_PORT):
 927                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
 928                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 929                                 goto invalid;
 930                         swkey->phy.in_port = nla_get_u32(nla);
 931                         break;
 932
 933                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_ETHERNET):
 934                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_ETHERNET):
 935                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_ETHERNET):
 936                 case TRANSITION(OVS_KEY_ATTR_IN_PORT, OVS_KEY_ATTR_ETHERNET):
 937                         eth_key = nla_data(nla);
 938                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
 939                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
 940                         break;
 941
 942                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_8021Q):
 943                         q_key = nla_data(nla);
 944                         /* Only standard 0x8100 VLANs currently supported. */
 945                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 946                                 goto invalid;
 947                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 948                                 goto invalid;
 949                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 950                         break;
 951
 952                 case TRANSITION(OVS_KEY_ATTR_8021Q, OVS_KEY_ATTR_ETHERTYPE):
 953                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_ETHERTYPE):
 954                         swkey->eth.type = nla_get_be16(nla);
 955                         if (ntohs(swkey->eth.type) < 1536)
 956                                 goto invalid;
 957                         break;
 958
 959                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV4):
 960                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 961                         if (swkey->eth.type != htons(ETH_P_IP))
 962                                 goto invalid;
 963                         ipv4_key = nla_data(nla);
 964                         if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
 965                                 goto invalid;
 966                         swkey->ip.proto = ipv4_key->ipv4_proto;
 967                         swkey->ip.tos = ipv4_key->ipv4_tos;
 968                         swkey->ip.ttl = ipv4_key->ipv4_ttl;
 969                         swkey->ip.frag = ipv4_key->ipv4_frag;
 970                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
 971                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
 972                         break;
 973
 974                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV6):
 975                         key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
 976                         if (swkey->eth.type != htons(ETH_P_IPV6))
 977                                 goto invalid;
 978                         ipv6_key = nla_data(nla);
 979                         if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
 980                                 goto invalid;
 981                         swkey->ipv6.label = ipv6_key->ipv6_label;
 982                         swkey->ip.proto = ipv6_key->ipv6_proto;
 983                         swkey->ip.tos = ipv6_key->ipv6_tclass;
 984                         swkey->ip.ttl = ipv6_key->ipv6_hlimit;
 985                         swkey->ip.frag = ipv6_key->ipv6_frag;
 986                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
 987                                         sizeof(swkey->ipv6.addr.src));
 988                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
 989                                         sizeof(swkey->ipv6.addr.dst));
 990                         break;
 991
 992                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_TCP):
 993                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 994                         if (swkey->ip.proto != IPPROTO_TCP)
 995                                 goto invalid;
 996                         tcp_key = nla_data(nla);
 997                         swkey->ipv4.tp.src = tcp_key->tcp_src;
 998                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 999                         break;
1000
1001                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_TCP):
1002                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1003                         if (swkey->ip.proto != IPPROTO_TCP)
1004                                 goto invalid;
1005                         tcp_key = nla_data(nla);
1006                         swkey->ipv6.tp.src = tcp_key->tcp_src;
1007                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
1008                         break;
1009
1010                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_UDP):
1011                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1012                         if (swkey->ip.proto != IPPROTO_UDP)
1013                                 goto invalid;
1014                         udp_key = nla_data(nla);
1015                         swkey->ipv4.tp.src = udp_key->udp_src;
1016                         swkey->ipv4.tp.dst = udp_key->udp_dst;
1017                         break;
1018
1019                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_UDP):
1020                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1021                         if (swkey->ip.proto != IPPROTO_UDP)
1022                                 goto invalid;
1023                         udp_key = nla_data(nla);
1024                         swkey->ipv6.tp.src = udp_key->udp_src;
1025                         swkey->ipv6.tp.dst = udp_key->udp_dst;
1026                         break;
1027
1028                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_ICMP):
1029                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1030                         if (swkey->ip.proto != IPPROTO_ICMP)
1031                                 goto invalid;
1032                         icmp_key = nla_data(nla);
1033                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
1034                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
1035                         break;
1036
1037                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_ICMPV6):
1038                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1039                         if (swkey->ip.proto != IPPROTO_ICMPV6)
1040                                 goto invalid;
1041                         icmpv6_key = nla_data(nla);
1042                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
1043                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
1044                         break;
1045
1046                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_ARP):
1047                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1048                         if (swkey->eth.type != htons(ETH_P_ARP))
1049                                 goto invalid;
1050                         arp_key = nla_data(nla);
1051                         swkey->ipv4.addr.src = arp_key->arp_sip;
1052                         swkey->ipv4.addr.dst = arp_key->arp_tip;
1053                         if (arp_key->arp_op & htons(0xff00))
1054                                 goto invalid;
1055                         swkey->ip.proto = ntohs(arp_key->arp_op);
1056                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1057                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1058                         break;
1059
1060                 case TRANSITION(OVS_KEY_ATTR_ICMPV6, OVS_KEY_ATTR_ND):
1061                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
1062                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
1063                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
1064                                 goto invalid;
1065                         nd_key = nla_data(nla);
1066                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
1067                                         sizeof(swkey->ipv6.nd.target));
1068                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
1069                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
1070                         break;
1071
1072                 default:
1073                         goto invalid;
1074                 }
1075
1076                 prev_type = type;
1077         }
1078         if (rem)
1079                 goto invalid;
1080
1081         switch (prev_type) {
1082         case OVS_KEY_ATTR_UNSPEC:
1083                 goto invalid;
1084
1085         case OVS_KEY_ATTR_PRIORITY:
1086         case OVS_KEY_ATTR_TUN_ID:
1087         case OVS_KEY_ATTR_IN_PORT:
1088                 goto invalid;
1089
1090         case OVS_KEY_ATTR_ETHERNET:
1091         case OVS_KEY_ATTR_8021Q:
1092                 goto ok;
1093
1094         case OVS_KEY_ATTR_ETHERTYPE:
1095                 if (swkey->eth.type == htons(ETH_P_IP) ||
1096                     swkey->eth.type == htons(ETH_P_IPV6) ||
1097                     swkey->eth.type == htons(ETH_P_ARP))
1098                         goto invalid;
1099                 goto ok;
1100
1101         case OVS_KEY_ATTR_IPV4:
1102                 if (swkey->ip.frag == OVS_FRAG_TYPE_LATER)
1103                         goto ok;
1104                 if (swkey->ip.proto == IPPROTO_TCP ||
1105                     swkey->ip.proto == IPPROTO_UDP ||
1106                     swkey->ip.proto == IPPROTO_ICMP)
1107                         goto invalid;
1108                 goto ok;
1109
1110         case OVS_KEY_ATTR_IPV6:
1111                 if (swkey->ip.frag == OVS_FRAG_TYPE_LATER)
1112                         goto ok;
1113                 if (swkey->ip.proto == IPPROTO_TCP ||
1114                     swkey->ip.proto == IPPROTO_UDP ||
1115                     swkey->ip.proto == IPPROTO_ICMPV6)
1116                         goto invalid;
1117                 goto ok;
1118
1119         case OVS_KEY_ATTR_ICMPV6:
1120                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1121                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT) ||
1122                     swkey->ip.frag == OVS_FRAG_TYPE_LATER)
1123                         goto invalid;
1124                 goto ok;
1125
1126         case OVS_KEY_ATTR_TCP:
1127         case OVS_KEY_ATTR_UDP:
1128         case OVS_KEY_ATTR_ICMP:
1129         case OVS_KEY_ATTR_ND:
1130                 if (swkey->ip.frag == OVS_FRAG_TYPE_LATER)
1131                         goto invalid;
1132                 goto ok;
1133
1134         case OVS_KEY_ATTR_ARP:
1135                 goto ok;
1136
1137         default:
1138                 WARN_ON_ONCE(1);
1139         }
1140
1141 invalid:
1142         error = -EINVAL;
1143
1144 ok:
1145         *key_lenp = key_len;
1146         return error;
1147 }
1148
1149 /**
1150  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1151  * @in_port: receives the extracted input port.
1152  * @tun_id: receives the extracted tunnel ID.
1153  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1154  * sequence.
1155  *
1156  * This parses a series of Netlink attributes that form a flow key, which must
1157  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1158  * get the metadata, that is, the parts of the flow key that cannot be
1159  * extracted from the packet itself.
1160  */
1161 int flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1162                                const struct nlattr *attr)
1163 {
1164         const struct nlattr *nla;
1165         int rem;
1166
1167         *in_port = USHRT_MAX;
1168         *tun_id = 0;
1169         *priority = 0;
1170
1171         nla_for_each_nested(nla, attr, rem) {
1172                 int type = nla_type(nla);
1173
1174                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] != 0) {
1175                         if (nla_len(nla) != ovs_key_lens[type])
1176                                 return -EINVAL;
1177
1178                         switch (type) {
1179                         case OVS_KEY_ATTR_PRIORITY:
1180                                 *priority = nla_get_u32(nla);
1181                                 break;
1182
1183                         case OVS_KEY_ATTR_TUN_ID:
1184                                 *tun_id = nla_get_be64(nla);
1185                                 break;
1186
1187                         case OVS_KEY_ATTR_IN_PORT:
1188                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1189                                         return -EINVAL;
1190                                 *in_port = nla_get_u32(nla);
1191                                 break;
1192                         }
1193                 }
1194         }
1195         if (rem)
1196                 return -EINVAL;
1197         return 0;
1198 }
1199
1200 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1201 {
1202         struct ovs_key_ethernet *eth_key;
1203         struct nlattr *nla;
1204
1205         if (swkey->phy.priority)
1206                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1207
1208         if (swkey->phy.tun_id != cpu_to_be64(0))
1209                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1210
1211         if (swkey->phy.in_port != USHRT_MAX)
1212                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1213
1214         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1215         if (!nla)
1216                 goto nla_put_failure;
1217         eth_key = nla_data(nla);
1218         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1219         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1220
1221         if (swkey->eth.tci != htons(0)) {
1222                 struct ovs_key_8021q q_key;
1223
1224                 q_key.q_tpid = htons(ETH_P_8021Q);
1225                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
1226                 NLA_PUT(skb, OVS_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
1227         }
1228
1229         if (swkey->eth.type == htons(ETH_P_802_2))
1230                 return 0;
1231
1232         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1233
1234         if (swkey->eth.type == htons(ETH_P_IP)) {
1235                 struct ovs_key_ipv4 *ipv4_key;
1236
1237                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1238                 if (!nla)
1239                         goto nla_put_failure;
1240                 ipv4_key = nla_data(nla);
1241                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1242                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1243                 ipv4_key->ipv4_proto = swkey->ip.proto;
1244                 ipv4_key->ipv4_tos = swkey->ip.tos;
1245                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1246                 ipv4_key->ipv4_frag = swkey->ip.frag;
1247         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1248                 struct ovs_key_ipv6 *ipv6_key;
1249
1250                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1251                 if (!nla)
1252                         goto nla_put_failure;
1253                 ipv6_key = nla_data(nla);
1254                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1255                                 sizeof(ipv6_key->ipv6_src));
1256                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1257                                 sizeof(ipv6_key->ipv6_dst));
1258                 ipv6_key->ipv6_label = swkey->ipv6.label;
1259                 ipv6_key->ipv6_proto = swkey->ip.proto;
1260                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1261                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1262                 ipv6_key->ipv6_frag = swkey->ip.frag;
1263         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1264                 struct ovs_key_arp *arp_key;
1265
1266                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1267                 if (!nla)
1268                         goto nla_put_failure;
1269                 arp_key = nla_data(nla);
1270                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1271                 arp_key->arp_sip = swkey->ipv4.addr.src;
1272                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1273                 arp_key->arp_op = htons(swkey->ip.proto);
1274                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1275                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1276         }
1277
1278         if ((swkey->eth.type == htons(ETH_P_IP) ||
1279              swkey->eth.type == htons(ETH_P_IPV6)) &&
1280              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1281
1282                 if (swkey->ip.proto == IPPROTO_TCP) {
1283                         struct ovs_key_tcp *tcp_key;
1284
1285                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1286                         if (!nla)
1287                                 goto nla_put_failure;
1288                         tcp_key = nla_data(nla);
1289                         if (swkey->eth.type == htons(ETH_P_IP)) {
1290                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1291                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1292                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1293                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1294                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1295                         }
1296                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1297                         struct ovs_key_udp *udp_key;
1298
1299                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1300                         if (!nla)
1301                                 goto nla_put_failure;
1302                         udp_key = nla_data(nla);
1303                         if (swkey->eth.type == htons(ETH_P_IP)) {
1304                                 udp_key->udp_src = swkey->ipv4.tp.src;
1305                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1306                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1307                                 udp_key->udp_src = swkey->ipv6.tp.src;
1308                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1309                         }
1310                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1311                            swkey->ip.proto == IPPROTO_ICMP) {
1312                         struct ovs_key_icmp *icmp_key;
1313
1314                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1315                         if (!nla)
1316                                 goto nla_put_failure;
1317                         icmp_key = nla_data(nla);
1318                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1319                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1320                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1321                            swkey->ip.proto == IPPROTO_ICMPV6) {
1322                         struct ovs_key_icmpv6 *icmpv6_key;
1323
1324                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1325                                                 sizeof(*icmpv6_key));
1326                         if (!nla)
1327                                 goto nla_put_failure;
1328                         icmpv6_key = nla_data(nla);
1329                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1330                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1331
1332                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1333                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1334                                 struct ovs_key_nd *nd_key;
1335
1336                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1337                                 if (!nla)
1338                                         goto nla_put_failure;
1339                                 nd_key = nla_data(nla);
1340                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1341                                                         sizeof(nd_key->nd_target));
1342                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1343                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1344                         }
1345                 }
1346         }
1347
1348         return 0;
1349
1350 nla_put_failure:
1351         return -EMSGSIZE;
1352 }
1353
1354 /* Initializes the flow module.
1355  * Returns zero if successful or a negative error code. */
1356 int flow_init(void)
1357 {
1358         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1359                                         0, NULL);
1360         if (flow_cache == NULL)
1361                 return -ENOMEM;
1362
1363         get_random_bytes(&hash_seed, sizeof(hash_seed));
1364
1365         return 0;
1366 }
1367
1368 /* Uninitializes the flow module. */
1369 void flow_exit(void)
1370 {
1371         kmem_cache_destroy(flow_cache);
1372 }