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 <asm/uaccess.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/if_ether.h>
  15 #include <linux/if_vlan.h>
  16 #include <net/llc_pdu.h>
  17 #include <linux/kernel.h>
  18 #include <linux/jhash.h>
  19 #include <linux/jiffies.h>
  20 #include <linux/llc.h>
  21 #include <linux/module.h>
  22 #include <linux/in.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/if_arp.h>
  25 #include <linux/if_ether.h>
  26 #include <linux/ip.h>
  27 #include <linux/ipv6.h>
  28 #include <linux/tcp.h>
  29 #include <linux/udp.h>
  30 #include <linux/icmp.h>
  31 #include <linux/icmpv6.h>
  32 #include <linux/rculist.h>
  33 #include <net/inet_ecn.h>
  34 #include <net/ip.h>
  35 #include <net/ipv6.h>
  36 #include <net/ndisc.h>
  37
  38 #include "vlan.h"
  39
  40 static struct kmem_cache *flow_cache;
  41 static unsigned int hash_seed __read_mostly;
  42
  43 static int check_header(struct sk_buff *skb, int len)
  44 {
  45         if (unlikely(skb->len < len))
  46                 return -EINVAL;
  47         if (unlikely(!pskb_may_pull(skb, len)))
  48                 return -ENOMEM;
  49         return 0;
  50 }
  51
  52 static inline bool arphdr_ok(struct sk_buff *skb)
  53 {
  54         return pskb_may_pull(skb, skb_network_offset(skb) +
  55                                   sizeof(struct arp_eth_header));
  56 }
  57
  58 static inline int check_iphdr(struct sk_buff *skb)
  59 {
  60         unsigned int nh_ofs = skb_network_offset(skb);
  61         unsigned int ip_len;
  62         int err;
  63
  64         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  65         if (unlikely(err))
  66                 return err;
  67
  68         ip_len = ip_hdrlen(skb);
  69         if (unlikely(ip_len < sizeof(struct iphdr) ||
  70                      skb->len < nh_ofs + ip_len))
  71                 return -EINVAL;
  72
  73         skb_set_transport_header(skb, nh_ofs + ip_len);
  74         return 0;
  75 }
  76
  77 static inline bool tcphdr_ok(struct sk_buff *skb)
  78 {
  79         int th_ofs = skb_transport_offset(skb);
  80         int tcp_len;
  81
  82         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  83                 return false;
  84
  85         tcp_len = tcp_hdrlen(skb);
  86         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  87                      skb->len < th_ofs + tcp_len))
  88                 return false;
  89
  90         return true;
  91 }
  92
  93 static inline bool udphdr_ok(struct sk_buff *skb)
  94 {
  95         return pskb_may_pull(skb, skb_transport_offset(skb) +
  96                                   sizeof(struct udphdr));
  97 }
  98
  99 static inline bool icmphdr_ok(struct sk_buff *skb)
 100 {
 101         return pskb_may_pull(skb, skb_transport_offset(skb) +
 102                                   sizeof(struct icmphdr));
 103 }
 104
 105 u64 flow_used_time(unsigned long flow_jiffies)
 106 {
 107         struct timespec cur_ts;
 108         u64 cur_ms, idle_ms;
 109
 110         ktime_get_ts(&cur_ts);
 111         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 112         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 113                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 114
 115         return cur_ms - idle_ms;
 116 }
 117
 118 #define SW_FLOW_KEY_OFFSET(field)               \
 119         offsetof(struct sw_flow_key, field) +   \
 120         FIELD_SIZEOF(struct sw_flow_key, field)
 121
 122 /**
 123  * skip_exthdr - skip any IPv6 extension headers
 124  * @skb: skbuff to parse
 125  * @start: offset of first extension header
 126  * @nexthdrp: Initially, points to the type of the extension header at @start.
 127  * This function updates it to point to the extension header at the final
 128  * offset.
 129  * @tos_frag: Points to the @tos_frag member in a &struct sw_flow_key.  This
 130  * function sets an appropriate %OVS_FRAG_TYPE_* value.
 131  *
 132  * This is based on ipv6_skip_exthdr() but adds the updates to *@tos_frag.
 133  *
 134  * When there is more than one fragment header, this version reports whether
 135  * the final fragment header that it examines is a first fragment.
 136  *
 137  * Returns the final payload offset, or -1 on error.
 138  */
 139 static int skip_exthdr(const struct sk_buff *skb, int start, u8 *nexthdrp,
 140                        u8 *tos_frag)
 141 {
 142         u8 nexthdr = *nexthdrp;
 143
 144         while (ipv6_ext_hdr(nexthdr)) {
 145                 struct ipv6_opt_hdr _hdr, *hp;
 146                 int hdrlen;
 147
 148                 if (nexthdr == NEXTHDR_NONE)
 149                         return -1;
 150                 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
 151                 if (hp == NULL)
 152                         return -1;
 153                 if (nexthdr == NEXTHDR_FRAGMENT) {
 154                         __be16 _frag_off, *fp;
 155                         fp = skb_header_pointer(skb,
 156                                                 start+offsetof(struct frag_hdr,
 157                                                                frag_off),
 158                                                 sizeof(_frag_off),
 159                                                 &_frag_off);
 160                         if (fp == NULL)
 161                                 return -1;
 162
 163                         *tos_frag &= ~OVS_FRAG_TYPE_MASK;
 164                         if (ntohs(*fp) & ~0x7) {
 165                                 *tos_frag |= OVS_FRAG_TYPE_LATER;
 166                                 break;
 167                         }
 168                         *tos_frag |= OVS_FRAG_TYPE_FIRST;
 169                         hdrlen = 8;
 170                 } else if (nexthdr == NEXTHDR_AUTH)
 171                         hdrlen = (hp->hdrlen+2)<<2;
 172                 else
 173                         hdrlen = ipv6_optlen(hp);
 174
 175                 nexthdr = hp->nexthdr;
 176                 start += hdrlen;
 177         }
 178
 179         *nexthdrp = nexthdr;
 180         return start;
 181 }
 182
 183 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 184                          int *key_lenp)
 185 {
 186         unsigned int nh_ofs = skb_network_offset(skb);
 187         unsigned int nh_len;
 188         int payload_ofs;
 189         struct ipv6hdr *nh;
 190         uint8_t nexthdr;
 191         int err;
 192
 193         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.addr);
 194
 195         err = check_header(skb, nh_ofs + sizeof(*nh));
 196         if (unlikely(err))
 197                 return err;
 198
 199         nh = ipv6_hdr(skb);
 200         nexthdr = nh->nexthdr;
 201         payload_ofs = (u8 *)(nh + 1) - skb->data;
 202
 203         key->ip.proto = NEXTHDR_NONE;
 204         key->ip.tos_frag = ipv6_get_dsfield(nh) & ~INET_ECN_MASK;
 205         ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
 206         ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
 207
 208         payload_ofs = skip_exthdr(skb, payload_ofs, &nexthdr, &key->ip.tos_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 = (struct nd_opt_hdr *)(nd->opt + offset);
 569                         int opt_len = nd_opt->nd_opt_len * 8;
 570
 571                         if (unlikely(!opt_len || opt_len > icmp_len))
 572                                 goto invalid;
 573
 574                         /* Store the link layer address if the appropriate
 575                          * option is provided.  It is considered an error if
 576                          * the same link layer option is specified twice.
 577                          */
 578                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 579                             && opt_len == 8) {
 580                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 581                                         goto invalid;
 582                                 memcpy(key->ipv6.nd.sll,
 583                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 584                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 585                                    && opt_len == 8) {
 586                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 587                                         goto invalid;
 588                                 memcpy(key->ipv6.nd.tll,
 589                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 590                         }
 591
 592                         icmp_len -= opt_len;
 593                         offset += opt_len;
 594                 }
 595         }
 596
 597         goto out;
 598
 599 invalid:
 600         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 601         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 602         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 603
 604 out:
 605         *key_lenp = key_len;
 606         return error;
 607 }
 608
 609 /**
 610  * flow_extract - extracts a flow key from an Ethernet frame.
 611  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 612  * Ethernet header
 613  * @in_port: port number on which @skb was received.
 614  * @key: output flow key
 615  * @key_lenp: length of output flow key
 616  *
 617  * The caller must ensure that skb->len >= ETH_HLEN.
 618  *
 619  * Returns 0 if successful, otherwise a negative errno value.
 620  *
 621  * Initializes @skb header pointers as follows:
 622  *
 623  *    - skb->mac_header: the Ethernet header.
 624  *
 625  *    - skb->network_header: just past the Ethernet header, or just past the
 626  *      VLAN header, to the first byte of the Ethernet payload.
 627  *
 628  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 629  *      on output, then just past the IP header, if one is present and
 630  *      of a correct length, otherwise the same as skb->network_header.
 631  *      For other key->dl_type values it is left untouched.
 632  */
 633 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 634                  int *key_lenp)
 635 {
 636         int error = 0;
 637         int key_len = SW_FLOW_KEY_OFFSET(eth);
 638         struct ethhdr *eth;
 639
 640         memset(key, 0, sizeof(*key));
 641         key->eth.tun_id = OVS_CB(skb)->tun_id;
 642         key->eth.in_port = in_port;
 643
 644         skb_reset_mac_header(skb);
 645
 646         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 647          * header in the linear data area.
 648          */
 649         eth = eth_hdr(skb);
 650         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 651         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 652
 653         __skb_pull(skb, 2 * ETH_ALEN);
 654
 655         if (vlan_tx_tag_present(skb))
 656                 key->eth.tci = htons(vlan_get_tci(skb));
 657         else if (eth->h_proto == htons(ETH_P_8021Q))
 658                 if (unlikely(parse_vlan(skb, key)))
 659                         return -ENOMEM;
 660
 661         key->eth.type = parse_ethertype(skb);
 662         if (unlikely(key->eth.type == htons(0)))
 663                 return -ENOMEM;
 664
 665         skb_reset_network_header(skb);
 666         __skb_push(skb, skb->data - skb_mac_header(skb));
 667
 668         /* Network layer. */
 669         if (key->eth.type == htons(ETH_P_IP)) {
 670                 struct iphdr *nh;
 671                 __be16 offset;
 672
 673                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 674
 675                 error = check_iphdr(skb);
 676                 if (unlikely(error)) {
 677                         if (error == -EINVAL) {
 678                                 skb->transport_header = skb->network_header;
 679                                 error = 0;
 680                         }
 681                         goto out;
 682                 }
 683
 684                 nh = ip_hdr(skb);
 685                 key->ipv4.addr.src = nh->saddr;
 686                 key->ipv4.addr.dst = nh->daddr;
 687
 688                 key->ip.proto = nh->protocol;
 689                 key->ip.tos_frag = nh->tos & ~INET_ECN_MASK;
 690
 691                 offset = nh->frag_off & htons(IP_OFFSET);
 692                 if (offset) {
 693                         key->ip.tos_frag |= OVS_FRAG_TYPE_LATER;
 694                         goto out;
 695                 }
 696                 if (nh->frag_off & htons(IP_MF) ||
 697                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 698                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
 699
 700                 /* Transport layer. */
 701                 if (key->ip.proto == IPPROTO_TCP) {
 702                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 703                         if (tcphdr_ok(skb)) {
 704                                 struct tcphdr *tcp = tcp_hdr(skb);
 705                                 key->ipv4.tp.src = tcp->source;
 706                                 key->ipv4.tp.dst = tcp->dest;
 707                         }
 708                 } else if (key->ip.proto == IPPROTO_UDP) {
 709                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 710                         if (udphdr_ok(skb)) {
 711                                 struct udphdr *udp = udp_hdr(skb);
 712                                 key->ipv4.tp.src = udp->source;
 713                                 key->ipv4.tp.dst = udp->dest;
 714                         }
 715                 } else if (key->ip.proto == IPPROTO_ICMP) {
 716                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 717                         if (icmphdr_ok(skb)) {
 718                                 struct icmphdr *icmp = icmp_hdr(skb);
 719                                 /* The ICMP type and code fields use the 16-bit
 720                                  * transport port fields, so we need to store them
 721                                  * in 16-bit network byte order. */
 722                                 key->ipv4.tp.src = htons(icmp->type);
 723                                 key->ipv4.tp.dst = htons(icmp->code);
 724                         }
 725                 }
 726
 727         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 728                 struct arp_eth_header *arp;
 729
 730                 arp = (struct arp_eth_header *)skb_network_header(skb);
 731
 732                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 733                                 && arp->ar_pro == htons(ETH_P_IP)
 734                                 && arp->ar_hln == ETH_ALEN
 735                                 && arp->ar_pln == 4) {
 736
 737                         /* We only match on the lower 8 bits of the opcode. */
 738                         if (ntohs(arp->ar_op) <= 0xff)
 739                                 key->ip.proto = ntohs(arp->ar_op);
 740
 741                         if (key->ip.proto == ARPOP_REQUEST
 742                                         || key->ip.proto == ARPOP_REPLY) {
 743                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 744                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 745                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 746                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 747                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 748                         }
 749                 }
 750         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 751                 int nh_len;             /* IPv6 Header + Extensions */
 752
 753                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 754                 if (unlikely(nh_len < 0)) {
 755                         if (nh_len == -EINVAL)
 756                                 skb->transport_header = skb->network_header;
 757                         else
 758                                 error = nh_len;
 759                         goto out;
 760                 }
 761
 762                 if ((key->ip.tos_frag & OVS_FRAG_TYPE_MASK) == OVS_FRAG_TYPE_LATER)
 763                         goto out;
 764                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 765                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
 766
 767                 /* Transport layer. */
 768                 if (key->ip.proto == NEXTHDR_TCP) {
 769                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 770                         if (tcphdr_ok(skb)) {
 771                                 struct tcphdr *tcp = tcp_hdr(skb);
 772                                 key->ipv6.tp.src = tcp->source;
 773                                 key->ipv6.tp.dst = tcp->dest;
 774                         }
 775                 } else if (key->ip.proto == NEXTHDR_UDP) {
 776                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 777                         if (udphdr_ok(skb)) {
 778                                 struct udphdr *udp = udp_hdr(skb);
 779                                 key->ipv6.tp.src = udp->source;
 780                                 key->ipv6.tp.dst = udp->dest;
 781                         }
 782                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 783                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 784                         if (icmp6hdr_ok(skb)) {
 785                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 786                                 if (error < 0)
 787                                         goto out;
 788                         }
 789                 }
 790         }
 791
 792 out:
 793         *key_lenp = key_len;
 794         return error;
 795 }
 796
 797 u32 flow_hash(const struct sw_flow_key *key, int key_len)
 798 {
 799         return jhash2((u32*)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
 800 }
 801
 802 struct sw_flow * flow_tbl_lookup(struct flow_table *table,
 803                                 struct sw_flow_key *key, int key_len)
 804 {
 805         struct sw_flow *flow;
 806         struct hlist_node *n;
 807         struct hlist_head *head;
 808         u32 hash;
 809
 810         hash = flow_hash(key, key_len);
 811
 812         head = find_bucket(table, hash);
 813         hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 814
 815                 if (flow->hash == hash &&
 816                     !memcmp(&flow->key, key, key_len)) {
 817                         return flow;
 818                 }
 819         }
 820         return NULL;
 821 }
 822
 823 void flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 824 {
 825         struct hlist_head *head;
 826
 827         head = find_bucket(table, flow->hash);
 828         hlist_add_head_rcu(&flow->hash_node, head);
 829         table->count++;
 830 }
 831
 832 void flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 833 {
 834         if (!hlist_unhashed(&flow->hash_node)) {
 835                 hlist_del_init_rcu(&flow->hash_node);
 836                 table->count--;
 837                 BUG_ON(table->count < 0);
 838         }
 839 }
 840
 841 static int parse_tos_frag(struct sw_flow_key *swkey, u8 tos, u8 frag)
 842 {
 843         if (tos & INET_ECN_MASK || frag > OVS_FRAG_TYPE_MAX)
 844                 return -EINVAL;
 845
 846         swkey->ip.tos_frag = tos | frag;
 847         return 0;
 848 }
 849
 850 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 851 const u32 ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 852         [OVS_KEY_ATTR_TUN_ID] = 8,
 853         [OVS_KEY_ATTR_IN_PORT] = 4,
 854         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 855         [OVS_KEY_ATTR_8021Q] = sizeof(struct ovs_key_8021q),
 856         [OVS_KEY_ATTR_ETHERTYPE] = 2,
 857         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 858         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 859         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 860         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 861         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 862         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 863         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 864         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 865 };
 866
 867 /**
 868  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 869  * @swkey: receives the extracted flow key.
 870  * @key_lenp: number of bytes used in @swkey.
 871  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 872  * sequence.
 873  *
 874  * This state machine accepts the following forms, with [] for optional
 875  * elements and | for alternatives:
 876  *
 877  * [tun_id] [in_port] ethernet [8021q] [ethertype \
 878  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 879  *
 880  * except that IPv4 or IPv6 terminates the sequence if its @ipv4_frag or
 881  * @ipv6_frag member, respectively, equals %OVS_FRAG_TYPE_LATER.
 882  */
 883 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 884                       const struct nlattr *attr)
 885 {
 886         int error = 0;
 887         enum ovs_frag_type frag_type;
 888         const struct nlattr *nla;
 889         u16 prev_type;
 890         int rem;
 891         int key_len;
 892
 893         memset(swkey, 0, sizeof(*swkey));
 894         swkey->eth.in_port = USHRT_MAX;
 895         swkey->eth.type = htons(ETH_P_802_2);
 896         key_len = SW_FLOW_KEY_OFFSET(eth);
 897
 898         prev_type = OVS_KEY_ATTR_UNSPEC;
 899         nla_for_each_nested(nla, attr, rem) {
 900                 const struct ovs_key_ethernet *eth_key;
 901                 const struct ovs_key_8021q *q_key;
 902                 const struct ovs_key_ipv4 *ipv4_key;
 903                 const struct ovs_key_ipv6 *ipv6_key;
 904                 const struct ovs_key_tcp *tcp_key;
 905                 const struct ovs_key_udp *udp_key;
 906                 const struct ovs_key_icmp *icmp_key;
 907                 const struct ovs_key_icmpv6 *icmpv6_key;
 908                 const struct ovs_key_arp *arp_key;
 909                 const struct ovs_key_nd *nd_key;
 910
 911                 int type = nla_type(nla);
 912
 913                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
 914                         goto invalid;
 915
 916 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 917                 switch (TRANSITION(prev_type, type)) {
 918                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
 919                         swkey->eth.tun_id = nla_get_be64(nla);
 920                         break;
 921
 922                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
 923                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
 924                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 925                                 goto invalid;
 926                         swkey->eth.in_port = nla_get_u32(nla);
 927                         break;
 928
 929                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_ETHERNET):
 930                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_ETHERNET):
 931                 case TRANSITION(OVS_KEY_ATTR_IN_PORT, OVS_KEY_ATTR_ETHERNET):
 932                         eth_key = nla_data(nla);
 933                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
 934                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
 935                         break;
 936
 937                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_8021Q):
 938                         q_key = nla_data(nla);
 939                         /* Only standard 0x8100 VLANs currently supported. */
 940                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 941                                 goto invalid;
 942                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 943                                 goto invalid;
 944                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 945                         break;
 946
 947                 case TRANSITION(OVS_KEY_ATTR_8021Q, OVS_KEY_ATTR_ETHERTYPE):
 948                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_ETHERTYPE):
 949                         swkey->eth.type = nla_get_be16(nla);
 950                         if (ntohs(swkey->eth.type) < 1536)
 951                                 goto invalid;
 952                         break;
 953
 954                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV4):
 955                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 956                         if (swkey->eth.type != htons(ETH_P_IP))
 957                                 goto invalid;
 958                         ipv4_key = nla_data(nla);
 959                         swkey->ip.proto = ipv4_key->ipv4_proto;
 960                         if (parse_tos_frag(swkey, ipv4_key->ipv4_tos,
 961                                            ipv4_key->ipv4_frag))
 962                                 goto invalid;
 963                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
 964                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
 965                         break;
 966
 967                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV6):
 968                         key_len = SW_FLOW_KEY_OFFSET(ipv6.addr);
 969                         if (swkey->eth.type != htons(ETH_P_IPV6))
 970                                 goto invalid;
 971                         ipv6_key = nla_data(nla);
 972                         swkey->ip.proto = ipv6_key->ipv6_proto;
 973                         if (parse_tos_frag(swkey, ipv6_key->ipv6_tos,
 974                                            ipv6_key->ipv6_frag))
 975                                 goto invalid;
 976                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
 977                                         sizeof(swkey->ipv6.addr.src));
 978                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
 979                                         sizeof(swkey->ipv6.addr.dst));
 980                         break;
 981
 982                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_TCP):
 983                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 984                         if (swkey->ip.proto != IPPROTO_TCP)
 985                                 goto invalid;
 986                         tcp_key = nla_data(nla);
 987                         swkey->ipv4.tp.src = tcp_key->tcp_src;
 988                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 989                         break;
 990
 991                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_TCP):
 992                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 993                         if (swkey->ip.proto != IPPROTO_TCP)
 994                                 goto invalid;
 995                         tcp_key = nla_data(nla);
 996                         swkey->ipv6.tp.src = tcp_key->tcp_src;
 997                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 998                         break;
 999
1000                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_UDP):
1001                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1002                         if (swkey->ip.proto != IPPROTO_UDP)
1003                                 goto invalid;
1004                         udp_key = nla_data(nla);
1005                         swkey->ipv4.tp.src = udp_key->udp_src;
1006                         swkey->ipv4.tp.dst = udp_key->udp_dst;
1007                         break;
1008
1009                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_UDP):
1010                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1011                         if (swkey->ip.proto != IPPROTO_UDP)
1012                                 goto invalid;
1013                         udp_key = nla_data(nla);
1014                         swkey->ipv6.tp.src = udp_key->udp_src;
1015                         swkey->ipv6.tp.dst = udp_key->udp_dst;
1016                         break;
1017
1018                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_ICMP):
1019                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1020                         if (swkey->ip.proto != IPPROTO_ICMP)
1021                                 goto invalid;
1022                         icmp_key = nla_data(nla);
1023                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
1024                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
1025                         break;
1026
1027                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_ICMPV6):
1028                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1029                         if (swkey->ip.proto != IPPROTO_ICMPV6)
1030                                 goto invalid;
1031                         icmpv6_key = nla_data(nla);
1032                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
1033                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
1034                         break;
1035
1036                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_ARP):
1037                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1038                         if (swkey->eth.type != htons(ETH_P_ARP))
1039                                 goto invalid;
1040                         arp_key = nla_data(nla);
1041                         swkey->ipv4.addr.src = arp_key->arp_sip;
1042                         swkey->ipv4.addr.dst = arp_key->arp_tip;
1043                         if (arp_key->arp_op & htons(0xff00))
1044                                 goto invalid;
1045                         swkey->ip.proto = ntohs(arp_key->arp_op);
1046                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1047                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1048                         break;
1049
1050                 case TRANSITION(OVS_KEY_ATTR_ICMPV6, OVS_KEY_ATTR_ND):
1051                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
1052                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
1053                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
1054                                 goto invalid;
1055                         nd_key = nla_data(nla);
1056                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
1057                                         sizeof(swkey->ipv6.nd.target));
1058                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
1059                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
1060                         break;
1061
1062                 default:
1063                         goto invalid;
1064                 }
1065
1066                 prev_type = type;
1067         }
1068         if (rem)
1069                 goto invalid;
1070
1071         frag_type = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1072         switch (prev_type) {
1073         case OVS_KEY_ATTR_UNSPEC:
1074                 goto invalid;
1075
1076         case OVS_KEY_ATTR_TUN_ID:
1077         case OVS_KEY_ATTR_IN_PORT:
1078                 goto invalid;
1079
1080         case OVS_KEY_ATTR_ETHERNET:
1081         case OVS_KEY_ATTR_8021Q:
1082                 goto ok;
1083
1084         case OVS_KEY_ATTR_ETHERTYPE:
1085                 if (swkey->eth.type == htons(ETH_P_IP) ||
1086                     swkey->eth.type == htons(ETH_P_ARP))
1087                         goto invalid;
1088                 goto ok;
1089
1090         case OVS_KEY_ATTR_IPV4:
1091                 if (frag_type == OVS_FRAG_TYPE_LATER)
1092                         goto ok;
1093                 if (swkey->ip.proto == IPPROTO_TCP ||
1094                     swkey->ip.proto == IPPROTO_UDP ||
1095                     swkey->ip.proto == IPPROTO_ICMP)
1096                         goto invalid;
1097                 goto ok;
1098
1099         case OVS_KEY_ATTR_IPV6:
1100                 if (frag_type == OVS_FRAG_TYPE_LATER)
1101                         goto ok;
1102                 if (swkey->ip.proto == IPPROTO_TCP ||
1103                     swkey->ip.proto == IPPROTO_UDP ||
1104                     swkey->ip.proto == IPPROTO_ICMPV6)
1105                         goto invalid;
1106                 goto ok;
1107
1108         case OVS_KEY_ATTR_ICMPV6:
1109                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1110                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT) ||
1111                     frag_type == OVS_FRAG_TYPE_LATER)
1112                         goto invalid;
1113                 goto ok;
1114
1115         case OVS_KEY_ATTR_TCP:
1116         case OVS_KEY_ATTR_UDP:
1117         case OVS_KEY_ATTR_ICMP:
1118         case OVS_KEY_ATTR_ND:
1119                 if (frag_type == OVS_FRAG_TYPE_LATER)
1120                         goto invalid;
1121                 goto ok;
1122
1123         case OVS_KEY_ATTR_ARP:
1124                 goto ok;
1125
1126         default:
1127                 WARN_ON_ONCE(1);
1128         }
1129
1130 invalid:
1131         error = -EINVAL;
1132
1133 ok:
1134         WARN_ON_ONCE(!key_len && !error);
1135         *key_lenp = key_len;
1136         return error;
1137 }
1138
1139 /**
1140  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1141  * @in_port: receives the extracted input port.
1142  * @tun_id: receives the extracted tunnel ID.
1143  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1144  * sequence.
1145  *
1146  * This parses a series of Netlink attributes that form a flow key, which must
1147  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1148  * get the metadata, that is, the parts of the flow key that cannot be
1149  * extracted from the packet itself.
1150  */
1151 int flow_metadata_from_nlattrs(u16 *in_port, __be64 *tun_id,
1152                                const struct nlattr *attr)
1153 {
1154         const struct nlattr *nla;
1155         u16 prev_type;
1156         int rem;
1157
1158         *in_port = USHRT_MAX;
1159         *tun_id = 0;
1160
1161         prev_type = OVS_KEY_ATTR_UNSPEC;
1162         nla_for_each_nested(nla, attr, rem) {
1163                 int type = nla_type(nla);
1164
1165                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
1166                         return -EINVAL;
1167
1168                 switch (TRANSITION(prev_type, type)) {
1169                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
1170                         *tun_id = nla_get_be64(nla);
1171                         break;
1172
1173                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
1174                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
1175                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
1176                                 return -EINVAL;
1177                         *in_port = nla_get_u32(nla);
1178                         break;
1179
1180                 default:
1181                         return 0;
1182                 }
1183
1184                 prev_type = type;
1185         }
1186         if (rem)
1187                 return -EINVAL;
1188         return 0;
1189 }
1190
1191 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1192 {
1193         struct ovs_key_ethernet *eth_key;
1194         struct nlattr *nla;
1195
1196         /* This is an imperfect sanity-check that FLOW_BUFSIZE doesn't need
1197          * to be updated, but will at least raise awareness when new
1198          * datapath key types are added. */
1199         BUILD_BUG_ON(__OVS_KEY_ATTR_MAX != 14);
1200
1201         if (swkey->eth.tun_id != cpu_to_be64(0))
1202                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->eth.tun_id);
1203
1204         if (swkey->eth.in_port != USHRT_MAX)
1205                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->eth.in_port);
1206
1207         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1208         if (!nla)
1209                 goto nla_put_failure;
1210         eth_key = nla_data(nla);
1211         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1212         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1213
1214         if (swkey->eth.tci != htons(0)) {
1215                 struct ovs_key_8021q q_key;
1216
1217                 q_key.q_tpid = htons(ETH_P_8021Q);
1218                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
1219                 NLA_PUT(skb, OVS_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
1220         }
1221
1222         if (swkey->eth.type == htons(ETH_P_802_2))
1223                 return 0;
1224
1225         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1226
1227         if (swkey->eth.type == htons(ETH_P_IP)) {
1228                 struct ovs_key_ipv4 *ipv4_key;
1229
1230                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1231                 if (!nla)
1232                         goto nla_put_failure;
1233                 ipv4_key = nla_data(nla);
1234                 memset(ipv4_key, 0, sizeof(struct ovs_key_ipv4));
1235                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1236                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1237                 ipv4_key->ipv4_proto = swkey->ip.proto;
1238                 ipv4_key->ipv4_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1239                 ipv4_key->ipv4_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1240         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1241                 struct ovs_key_ipv6 *ipv6_key;
1242
1243                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1244                 if (!nla)
1245                         goto nla_put_failure;
1246                 ipv6_key = nla_data(nla);
1247                 memset(ipv6_key, 0, sizeof(struct ovs_key_ipv6));
1248                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1249                                 sizeof(ipv6_key->ipv6_src));
1250                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1251                                 sizeof(ipv6_key->ipv6_dst));
1252                 ipv6_key->ipv6_proto = swkey->ip.proto;
1253                 ipv6_key->ipv6_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1254                 ipv6_key->ipv6_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1255         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1256                 struct ovs_key_arp *arp_key;
1257
1258                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1259                 if (!nla)
1260                         goto nla_put_failure;
1261                 arp_key = nla_data(nla);
1262                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1263                 arp_key->arp_sip = swkey->ipv4.addr.src;
1264                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1265                 arp_key->arp_op = htons(swkey->ip.proto);
1266                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1267                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1268         }
1269
1270         if ((swkey->eth.type == htons(ETH_P_IP) ||
1271              swkey->eth.type == htons(ETH_P_IPV6)) &&
1272             (swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK) != OVS_FRAG_TYPE_LATER) {
1273
1274                 if (swkey->ip.proto == IPPROTO_TCP) {
1275                         struct ovs_key_tcp *tcp_key;
1276
1277                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1278                         if (!nla)
1279                                 goto nla_put_failure;
1280                         tcp_key = nla_data(nla);
1281                         if (swkey->eth.type == htons(ETH_P_IP)) {
1282                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1283                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1284                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1285                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1286                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1287                         }
1288                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1289                         struct ovs_key_udp *udp_key;
1290
1291                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1292                         if (!nla)
1293                                 goto nla_put_failure;
1294                         udp_key = nla_data(nla);
1295                         if (swkey->eth.type == htons(ETH_P_IP)) {
1296                                 udp_key->udp_src = swkey->ipv4.tp.src;
1297                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1298                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1299                                 udp_key->udp_src = swkey->ipv6.tp.src;
1300                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1301                         }
1302                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1303                            swkey->ip.proto == IPPROTO_ICMP) {
1304                         struct ovs_key_icmp *icmp_key;
1305
1306                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1307                         if (!nla)
1308                                 goto nla_put_failure;
1309                         icmp_key = nla_data(nla);
1310                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1311                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1312                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1313                            swkey->ip.proto == IPPROTO_ICMPV6) {
1314                         struct ovs_key_icmpv6 *icmpv6_key;
1315
1316                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1317                                                 sizeof(*icmpv6_key));
1318                         if (!nla)
1319                                 goto nla_put_failure;
1320                         icmpv6_key = nla_data(nla);
1321                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1322                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1323
1324                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1325                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1326                                 struct ovs_key_nd *nd_key;
1327
1328                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1329                                 if (!nla)
1330                                         goto nla_put_failure;
1331                                 nd_key = nla_data(nla);
1332                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1333                                                         sizeof(nd_key->nd_target));
1334                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1335                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1336                         }
1337                 }
1338         }
1339
1340         return 0;
1341
1342 nla_put_failure:
1343         return -EMSGSIZE;
1344 }
1345
1346 /* Initializes the flow module.
1347  * Returns zero if successful or a negative error code. */
1348 int flow_init(void)
1349 {
1350         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1351                                         0, NULL);
1352         if (flow_cache == NULL)
1353                 return -ENOMEM;
1354
1355         get_random_bytes(&hash_seed, sizeof(hash_seed));
1356
1357         return 0;
1358 }
1359
1360 /* Uninitializes the flow module. */
1361 void flow_exit(void)
1362 {
1363         kmem_cache_destroy(flow_cache);
1364 }