pintos-os.org Git - openvswitch/blob - datapath/flow.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include "flow.h"
  10 #include "datapath.h"
  11 #include <linux/uaccess.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/if_ether.h>
  15 #include <linux/if_vlan.h>
  16 #include <net/llc_pdu.h>
  17 #include <linux/kernel.h>
  18 #include <linux/jhash.h>
  19 #include <linux/jiffies.h>
  20 #include <linux/llc.h>
  21 #include <linux/module.h>
  22 #include <linux/in.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/if_arp.h>
  25 #include <linux/if_ether.h>
  26 #include <linux/ip.h>
  27 #include <linux/ipv6.h>
  28 #include <linux/tcp.h>
  29 #include <linux/udp.h>
  30 #include <linux/icmp.h>
  31 #include <linux/icmpv6.h>
  32 #include <linux/rculist.h>
  33 #include <net/inet_ecn.h>
  34 #include <net/ip.h>
  35 #include <net/ipv6.h>
  36 #include <net/ndisc.h>
  37
  38 #include "vlan.h"
  39
  40 static struct kmem_cache *flow_cache;
  41 static unsigned int hash_seed __read_mostly;
  42
  43 static int check_header(struct sk_buff *skb, int len)
  44 {
  45         if (unlikely(skb->len < len))
  46                 return -EINVAL;
  47         if (unlikely(!pskb_may_pull(skb, len)))
  48                 return -ENOMEM;
  49         return 0;
  50 }
  51
  52 static bool arphdr_ok(struct sk_buff *skb)
  53 {
  54         return pskb_may_pull(skb, skb_network_offset(skb) +
  55                                   sizeof(struct arp_eth_header));
  56 }
  57
  58 static int check_iphdr(struct sk_buff *skb)
  59 {
  60         unsigned int nh_ofs = skb_network_offset(skb);
  61         unsigned int ip_len;
  62         int err;
  63
  64         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  65         if (unlikely(err))
  66                 return err;
  67
  68         ip_len = ip_hdrlen(skb);
  69         if (unlikely(ip_len < sizeof(struct iphdr) ||
  70                      skb->len < nh_ofs + ip_len))
  71                 return -EINVAL;
  72
  73         skb_set_transport_header(skb, nh_ofs + ip_len);
  74         return 0;
  75 }
  76
  77 static bool tcphdr_ok(struct sk_buff *skb)
  78 {
  79         int th_ofs = skb_transport_offset(skb);
  80         int tcp_len;
  81
  82         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  83                 return false;
  84
  85         tcp_len = tcp_hdrlen(skb);
  86         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  87                      skb->len < th_ofs + tcp_len))
  88                 return false;
  89
  90         return true;
  91 }
  92
  93 static bool udphdr_ok(struct sk_buff *skb)
  94 {
  95         return pskb_may_pull(skb, skb_transport_offset(skb) +
  96                                   sizeof(struct udphdr));
  97 }
  98
  99 static bool icmphdr_ok(struct sk_buff *skb)
 100 {
 101         return pskb_may_pull(skb, skb_transport_offset(skb) +
 102                                   sizeof(struct icmphdr));
 103 }
 104
 105 u64 flow_used_time(unsigned long flow_jiffies)
 106 {
 107         struct timespec cur_ts;
 108         u64 cur_ms, idle_ms;
 109
 110         ktime_get_ts(&cur_ts);
 111         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 112         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 113                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 114
 115         return cur_ms - idle_ms;
 116 }
 117
 118 #define SW_FLOW_KEY_OFFSET(field)               \
 119         (offsetof(struct sw_flow_key, field) +  \
 120          FIELD_SIZEOF(struct sw_flow_key, field))
 121
 122 /**
 123  * skip_exthdr - skip any IPv6 extension headers
 124  * @skb: skbuff to parse
 125  * @start: offset of first extension header
 126  * @nexthdrp: Initially, points to the type of the extension header at @start.
 127  * This function updates it to point to the extension header at the final
 128  * offset.
 129  * @tos_frag: Points to the @tos_frag member in a &struct sw_flow_key.  This
 130  * function sets an appropriate %OVS_FRAG_TYPE_* value.
 131  *
 132  * This is based on ipv6_skip_exthdr() but adds the updates to *@tos_frag.
 133  *
 134  * When there is more than one fragment header, this version reports whether
 135  * the final fragment header that it examines is a first fragment.
 136  *
 137  * Returns the final payload offset, or -1 on error.
 138  */
 139 static int skip_exthdr(const struct sk_buff *skb, int start, u8 *nexthdrp,
 140                        u8 *tos_frag)
 141 {
 142         u8 nexthdr = *nexthdrp;
 143
 144         while (ipv6_ext_hdr(nexthdr)) {
 145                 struct ipv6_opt_hdr _hdr, *hp;
 146                 int hdrlen;
 147
 148                 if (nexthdr == NEXTHDR_NONE)
 149                         return -1;
 150                 hp = skb_header_pointer(skb, start, sizeof(_hdr), &_hdr);
 151                 if (hp == NULL)
 152                         return -1;
 153                 if (nexthdr == NEXTHDR_FRAGMENT) {
 154                         __be16 _frag_off, *fp;
 155                         fp = skb_header_pointer(skb,
 156                                                 start+offsetof(struct frag_hdr,
 157                                                                frag_off),
 158                                                 sizeof(_frag_off),
 159                                                 &_frag_off);
 160                         if (fp == NULL)
 161                                 return -1;
 162
 163                         *tos_frag &= ~OVS_FRAG_TYPE_MASK;
 164                         if (ntohs(*fp) & ~0x7) {
 165                                 *tos_frag |= OVS_FRAG_TYPE_LATER;
 166                                 break;
 167                         }
 168                         *tos_frag |= OVS_FRAG_TYPE_FIRST;
 169                         hdrlen = 8;
 170                 } else if (nexthdr == NEXTHDR_AUTH)
 171                         hdrlen = (hp->hdrlen+2)<<2;
 172                 else
 173                         hdrlen = ipv6_optlen(hp);
 174
 175                 nexthdr = hp->nexthdr;
 176                 start += hdrlen;
 177         }
 178
 179         *nexthdrp = nexthdr;
 180         return start;
 181 }
 182
 183 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 184                          int *key_lenp)
 185 {
 186         unsigned int nh_ofs = skb_network_offset(skb);
 187         unsigned int nh_len;
 188         int payload_ofs;
 189         struct ipv6hdr *nh;
 190         uint8_t nexthdr;
 191         int err;
 192
 193         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 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         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 206         ipv6_addr_copy(&key->ipv6.addr.src, &nh->saddr);
 207         ipv6_addr_copy(&key->ipv6.addr.dst, &nh->daddr);
 208
 209         payload_ofs = skip_exthdr(skb, payload_ofs,
 210                                   &nexthdr, &key->ip.tos_frag);
 211         if (unlikely(payload_ofs < 0))
 212                 return -EINVAL;
 213
 214         nh_len = payload_ofs - nh_ofs;
 215         skb_set_transport_header(skb, nh_ofs + nh_len);
 216         key->ip.proto = nexthdr;
 217         return nh_len;
 218 }
 219
 220 static bool icmp6hdr_ok(struct sk_buff *skb)
 221 {
 222         return pskb_may_pull(skb, skb_transport_offset(skb) +
 223                                   sizeof(struct icmp6hdr));
 224 }
 225
 226 #define TCP_FLAGS_OFFSET 13
 227 #define TCP_FLAG_MASK 0x3f
 228
 229 void flow_used(struct sw_flow *flow, struct sk_buff *skb)
 230 {
 231         u8 tcp_flags = 0;
 232
 233         if (flow->key.eth.type == htons(ETH_P_IP) &&
 234             flow->key.ip.proto == IPPROTO_TCP) {
 235                 u8 *tcp = (u8 *)tcp_hdr(skb);
 236                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 237         }
 238
 239         spin_lock(&flow->lock);
 240         flow->used = jiffies;
 241         flow->packet_count++;
 242         flow->byte_count += skb->len;
 243         flow->tcp_flags |= tcp_flags;
 244         spin_unlock(&flow->lock);
 245 }
 246
 247 struct sw_flow_actions *flow_actions_alloc(const struct nlattr *actions)
 248 {
 249         int actions_len = nla_len(actions);
 250         struct sw_flow_actions *sfa;
 251
 252         /* At least DP_MAX_PORTS actions are required to be able to flood a
 253          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 254          * etc. */
 255         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 256                 return ERR_PTR(-EINVAL);
 257
 258         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 259         if (!sfa)
 260                 return ERR_PTR(-ENOMEM);
 261
 262         sfa->actions_len = actions_len;
 263         memcpy(sfa->actions, nla_data(actions), actions_len);
 264         return sfa;
 265 }
 266
 267 struct sw_flow *flow_alloc(void)
 268 {
 269         struct sw_flow *flow;
 270
 271         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 272         if (!flow)
 273                 return ERR_PTR(-ENOMEM);
 274
 275         spin_lock_init(&flow->lock);
 276         atomic_set(&flow->refcnt, 1);
 277         flow->sf_acts = NULL;
 278         flow->dead = false;
 279
 280         return flow;
 281 }
 282
 283 static struct hlist_head __rcu *find_bucket(struct flow_table * table, u32 hash)
 284 {
 285         return flex_array_get(table->buckets,
 286                                 (hash & (table->n_buckets - 1)));
 287 }
 288
 289 static struct flex_array  __rcu *alloc_buckets(unsigned int n_buckets)
 290 {
 291         struct flex_array  __rcu *buckets;
 292         int i, err;
 293
 294         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 295                                    n_buckets, GFP_KERNEL);
 296         if (!buckets)
 297                 return NULL;
 298
 299         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 300         if (err) {
 301                 flex_array_free(buckets);
 302                 return NULL;
 303         }
 304
 305         for (i = 0; i < n_buckets; i++)
 306                 INIT_HLIST_HEAD((struct hlist_head *)
 307                                         flex_array_get(buckets, i));
 308
 309         return buckets;
 310 }
 311
 312 static void free_buckets(struct flex_array *buckets)
 313 {
 314         flex_array_free(buckets);
 315 }
 316
 317 struct flow_table *flow_tbl_alloc(int new_size)
 318 {
 319         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 320
 321         if (!table)
 322                 return NULL;
 323
 324         table->buckets = alloc_buckets(new_size);
 325
 326         if (!table->buckets) {
 327                 kfree(table);
 328                 return NULL;
 329         }
 330         table->n_buckets = new_size;
 331         table->count = 0;
 332
 333         return table;
 334 }
 335
 336 static void flow_free(struct sw_flow *flow)
 337 {
 338         flow->dead = true;
 339         flow_put(flow);
 340 }
 341
 342 void flow_tbl_destroy(struct flow_table *table)
 343 {
 344         int i;
 345
 346         if (!table)
 347                 return;
 348
 349         for (i = 0; i < table->n_buckets; i++) {
 350                 struct sw_flow *flow;
 351                 struct hlist_head *head = flex_array_get(table->buckets, i);
 352                 struct hlist_node *node, *n;
 353
 354                 hlist_for_each_entry_safe(flow, node, n, head, hash_node) {
 355                         hlist_del_init_rcu(&flow->hash_node);
 356                         flow_free(flow);
 357                 }
 358         }
 359
 360         free_buckets(table->buckets);
 361         kfree(table);
 362 }
 363
 364 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 365 {
 366         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 367
 368         flow_tbl_destroy(table);
 369 }
 370
 371 void flow_tbl_deferred_destroy(struct flow_table *table)
 372 {
 373         if (!table)
 374                 return;
 375
 376         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 377 }
 378
 379 struct sw_flow *flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 380 {
 381         struct sw_flow *flow;
 382         struct hlist_head *head;
 383         struct hlist_node *n;
 384         int i;
 385
 386         while (*bucket < table->n_buckets) {
 387                 i = 0;
 388                 head = flex_array_get(table->buckets, *bucket);
 389                 hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 390                         if (i < *last) {
 391                                 i++;
 392                                 continue;
 393                         }
 394                         *last = i + 1;
 395                         return flow;
 396                 }
 397                 (*bucket)++;
 398                 *last = 0;
 399         }
 400
 401         return NULL;
 402 }
 403
 404 struct flow_table *flow_tbl_expand(struct flow_table *table)
 405 {
 406         struct flow_table *new_table;
 407         int n_buckets = table->n_buckets * 2;
 408         int i;
 409
 410         new_table = flow_tbl_alloc(n_buckets);
 411         if (!new_table)
 412                 return ERR_PTR(-ENOMEM);
 413
 414         for (i = 0; i < table->n_buckets; i++) {
 415                 struct sw_flow *flow;
 416                 struct hlist_head *head;
 417                 struct hlist_node *n, *pos;
 418
 419                 head = flex_array_get(table->buckets, i);
 420
 421                 hlist_for_each_entry_safe(flow, n, pos, head, hash_node) {
 422                         hlist_del_init_rcu(&flow->hash_node);
 423                         flow_tbl_insert(new_table, flow);
 424                 }
 425         }
 426
 427         return new_table;
 428 }
 429
 430 /* RCU callback used by flow_deferred_free. */
 431 static void rcu_free_flow_callback(struct rcu_head *rcu)
 432 {
 433         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 434
 435         flow->dead = true;
 436         flow_put(flow);
 437 }
 438
 439 /* Schedules 'flow' to be freed after the next RCU grace period.
 440  * The caller must hold rcu_read_lock for this to be sensible. */
 441 void flow_deferred_free(struct sw_flow *flow)
 442 {
 443         call_rcu(&flow->rcu, rcu_free_flow_callback);
 444 }
 445
 446 void flow_hold(struct sw_flow *flow)
 447 {
 448         atomic_inc(&flow->refcnt);
 449 }
 450
 451 void flow_put(struct sw_flow *flow)
 452 {
 453         if (unlikely(!flow))
 454                 return;
 455
 456         if (atomic_dec_and_test(&flow->refcnt)) {
 457                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 458                 kmem_cache_free(flow_cache, flow);
 459         }
 460 }
 461
 462 /* RCU callback used by flow_deferred_free_acts. */
 463 static void rcu_free_acts_callback(struct rcu_head *rcu)
 464 {
 465         struct sw_flow_actions *sf_acts = container_of(rcu,
 466                         struct sw_flow_actions, rcu);
 467         kfree(sf_acts);
 468 }
 469
 470 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 471  * The caller must hold rcu_read_lock for this to be sensible. */
 472 void flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 473 {
 474         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 475 }
 476
 477 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 478 {
 479         struct qtag_prefix {
 480                 __be16 eth_type; /* ETH_P_8021Q */
 481                 __be16 tci;
 482         };
 483         struct qtag_prefix *qp;
 484
 485         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 486                                          sizeof(__be16))))
 487                 return -ENOMEM;
 488
 489         qp = (struct qtag_prefix *) skb->data;
 490         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 491         __skb_pull(skb, sizeof(struct qtag_prefix));
 492
 493         return 0;
 494 }
 495
 496 static __be16 parse_ethertype(struct sk_buff *skb)
 497 {
 498         struct llc_snap_hdr {
 499                 u8  dsap;  /* Always 0xAA */
 500                 u8  ssap;  /* Always 0xAA */
 501                 u8  ctrl;
 502                 u8  oui[3];
 503                 __be16 ethertype;
 504         };
 505         struct llc_snap_hdr *llc;
 506         __be16 proto;
 507
 508         proto = *(__be16 *) skb->data;
 509         __skb_pull(skb, sizeof(__be16));
 510
 511         if (ntohs(proto) >= 1536)
 512                 return proto;
 513
 514         if (skb->len < sizeof(struct llc_snap_hdr))
 515                 return htons(ETH_P_802_2);
 516
 517         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 518                 return htons(0);
 519
 520         llc = (struct llc_snap_hdr *) skb->data;
 521         if (llc->dsap != LLC_SAP_SNAP ||
 522             llc->ssap != LLC_SAP_SNAP ||
 523             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 524                 return htons(ETH_P_802_2);
 525
 526         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 527         return llc->ethertype;
 528 }
 529
 530 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 531                         int *key_lenp, int nh_len)
 532 {
 533         struct icmp6hdr *icmp = icmp6_hdr(skb);
 534         int error = 0;
 535         int key_len;
 536
 537         /* The ICMPv6 type and code fields use the 16-bit transport port
 538          * fields, so we need to store them in 16-bit network byte order.
 539          */
 540         key->ipv6.tp.src = htons(icmp->icmp6_type);
 541         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 542         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 543
 544         if (icmp->icmp6_code == 0 &&
 545             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 546              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 547                 int icmp_len = skb->len - skb_transport_offset(skb);
 548                 struct nd_msg *nd;
 549                 int offset;
 550
 551                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 552
 553                 /* In order to process neighbor discovery options, we need the
 554                  * entire packet.
 555                  */
 556                 if (unlikely(icmp_len < sizeof(*nd)))
 557                         goto out;
 558                 if (unlikely(skb_linearize(skb))) {
 559                         error = -ENOMEM;
 560                         goto out;
 561                 }
 562
 563                 nd = (struct nd_msg *)skb_transport_header(skb);
 564                 ipv6_addr_copy(&key->ipv6.nd.target, &nd->target);
 565                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 566
 567                 icmp_len -= sizeof(*nd);
 568                 offset = 0;
 569                 while (icmp_len >= 8) {
 570                         struct nd_opt_hdr *nd_opt =
 571                                  (struct nd_opt_hdr *)(nd->opt + offset);
 572                         int opt_len = nd_opt->nd_opt_len * 8;
 573
 574                         if (unlikely(!opt_len || opt_len > icmp_len))
 575                                 goto invalid;
 576
 577                         /* Store the link layer address if the appropriate
 578                          * option is provided.  It is considered an error if
 579                          * the same link layer option is specified twice.
 580                          */
 581                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 582                             && opt_len == 8) {
 583                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 584                                         goto invalid;
 585                                 memcpy(key->ipv6.nd.sll,
 586                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 587                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 588                                    && opt_len == 8) {
 589                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 590                                         goto invalid;
 591                                 memcpy(key->ipv6.nd.tll,
 592                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 593                         }
 594
 595                         icmp_len -= opt_len;
 596                         offset += opt_len;
 597                 }
 598         }
 599
 600         goto out;
 601
 602 invalid:
 603         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 604         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 605         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 606
 607 out:
 608         *key_lenp = key_len;
 609         return error;
 610 }
 611
 612 /**
 613  * flow_extract - extracts a flow key from an Ethernet frame.
 614  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 615  * Ethernet header
 616  * @in_port: port number on which @skb was received.
 617  * @key: output flow key
 618  * @key_lenp: length of output flow key
 619  *
 620  * The caller must ensure that skb->len >= ETH_HLEN.
 621  *
 622  * Returns 0 if successful, otherwise a negative errno value.
 623  *
 624  * Initializes @skb header pointers as follows:
 625  *
 626  *    - skb->mac_header: the Ethernet header.
 627  *
 628  *    - skb->network_header: just past the Ethernet header, or just past the
 629  *      VLAN header, to the first byte of the Ethernet payload.
 630  *
 631  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 632  *      on output, then just past the IP header, if one is present and
 633  *      of a correct length, otherwise the same as skb->network_header.
 634  *      For other key->dl_type values it is left untouched.
 635  */
 636 int flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 637                  int *key_lenp)
 638 {
 639         int error = 0;
 640         int key_len = SW_FLOW_KEY_OFFSET(eth);
 641         struct ethhdr *eth;
 642
 643         memset(key, 0, sizeof(*key));
 644
 645         key->phy.priority = skb->priority;
 646         key->phy.tun_id = OVS_CB(skb)->tun_id;
 647         key->phy.in_port = in_port;
 648
 649         skb_reset_mac_header(skb);
 650
 651         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 652          * header in the linear data area.
 653          */
 654         eth = eth_hdr(skb);
 655         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 656         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 657
 658         __skb_pull(skb, 2 * ETH_ALEN);
 659
 660         if (vlan_tx_tag_present(skb))
 661                 key->eth.tci = htons(vlan_get_tci(skb));
 662         else if (eth->h_proto == htons(ETH_P_8021Q))
 663                 if (unlikely(parse_vlan(skb, key)))
 664                         return -ENOMEM;
 665
 666         key->eth.type = parse_ethertype(skb);
 667         if (unlikely(key->eth.type == htons(0)))
 668                 return -ENOMEM;
 669
 670         skb_reset_network_header(skb);
 671         __skb_push(skb, skb->data - skb_mac_header(skb));
 672
 673         /* Network layer. */
 674         if (key->eth.type == htons(ETH_P_IP)) {
 675                 struct iphdr *nh;
 676                 __be16 offset;
 677
 678                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 679
 680                 error = check_iphdr(skb);
 681                 if (unlikely(error)) {
 682                         if (error == -EINVAL) {
 683                                 skb->transport_header = skb->network_header;
 684                                 error = 0;
 685                         }
 686                         goto out;
 687                 }
 688
 689                 nh = ip_hdr(skb);
 690                 key->ipv4.addr.src = nh->saddr;
 691                 key->ipv4.addr.dst = nh->daddr;
 692
 693                 key->ip.proto = nh->protocol;
 694                 key->ip.tos_frag = nh->tos & ~INET_ECN_MASK;
 695
 696                 offset = nh->frag_off & htons(IP_OFFSET);
 697                 if (offset) {
 698                         key->ip.tos_frag |= OVS_FRAG_TYPE_LATER;
 699                         goto out;
 700                 }
 701                 if (nh->frag_off & htons(IP_MF) ||
 702                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 703                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
 704
 705                 /* Transport layer. */
 706                 if (key->ip.proto == IPPROTO_TCP) {
 707                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 708                         if (tcphdr_ok(skb)) {
 709                                 struct tcphdr *tcp = tcp_hdr(skb);
 710                                 key->ipv4.tp.src = tcp->source;
 711                                 key->ipv4.tp.dst = tcp->dest;
 712                         }
 713                 } else if (key->ip.proto == IPPROTO_UDP) {
 714                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 715                         if (udphdr_ok(skb)) {
 716                                 struct udphdr *udp = udp_hdr(skb);
 717                                 key->ipv4.tp.src = udp->source;
 718                                 key->ipv4.tp.dst = udp->dest;
 719                         }
 720                 } else if (key->ip.proto == IPPROTO_ICMP) {
 721                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 722                         if (icmphdr_ok(skb)) {
 723                                 struct icmphdr *icmp = icmp_hdr(skb);
 724                                 /* The ICMP type and code fields use the 16-bit
 725                                  * transport port fields, so we need to store
 726                                  * them in 16-bit network byte order. */
 727                                 key->ipv4.tp.src = htons(icmp->type);
 728                                 key->ipv4.tp.dst = htons(icmp->code);
 729                         }
 730                 }
 731
 732         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 733                 struct arp_eth_header *arp;
 734
 735                 arp = (struct arp_eth_header *)skb_network_header(skb);
 736
 737                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 738                                 && arp->ar_pro == htons(ETH_P_IP)
 739                                 && arp->ar_hln == ETH_ALEN
 740                                 && arp->ar_pln == 4) {
 741
 742                         /* We only match on the lower 8 bits of the opcode. */
 743                         if (ntohs(arp->ar_op) <= 0xff)
 744                                 key->ip.proto = ntohs(arp->ar_op);
 745
 746                         if (key->ip.proto == ARPOP_REQUEST
 747                                         || key->ip.proto == ARPOP_REPLY) {
 748                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 749                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 750                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 751                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 752                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 753                         }
 754                 }
 755         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 756                 int nh_len;             /* IPv6 Header + Extensions */
 757
 758                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 759                 if (unlikely(nh_len < 0)) {
 760                         if (nh_len == -EINVAL)
 761                                 skb->transport_header = skb->network_header;
 762                         else
 763                                 error = nh_len;
 764                         goto out;
 765                 }
 766
 767                 if ((key->ip.tos_frag & OVS_FRAG_TYPE_MASK) == OVS_FRAG_TYPE_LATER)
 768                         goto out;
 769                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 770                         key->ip.tos_frag |= OVS_FRAG_TYPE_FIRST;
 771
 772                 /* Transport layer. */
 773                 if (key->ip.proto == NEXTHDR_TCP) {
 774                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 775                         if (tcphdr_ok(skb)) {
 776                                 struct tcphdr *tcp = tcp_hdr(skb);
 777                                 key->ipv6.tp.src = tcp->source;
 778                                 key->ipv6.tp.dst = tcp->dest;
 779                         }
 780                 } else if (key->ip.proto == NEXTHDR_UDP) {
 781                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 782                         if (udphdr_ok(skb)) {
 783                                 struct udphdr *udp = udp_hdr(skb);
 784                                 key->ipv6.tp.src = udp->source;
 785                                 key->ipv6.tp.dst = udp->dest;
 786                         }
 787                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 788                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 789                         if (icmp6hdr_ok(skb)) {
 790                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 791                                 if (error < 0)
 792                                         goto out;
 793                         }
 794                 }
 795         }
 796
 797 out:
 798         *key_lenp = key_len;
 799         return error;
 800 }
 801
 802 u32 flow_hash(const struct sw_flow_key *key, int key_len)
 803 {
 804         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), hash_seed);
 805 }
 806
 807 struct sw_flow *flow_tbl_lookup(struct flow_table *table,
 808                                 struct sw_flow_key *key, int key_len)
 809 {
 810         struct sw_flow *flow;
 811         struct hlist_node *n;
 812         struct hlist_head *head;
 813         u32 hash;
 814
 815         hash = flow_hash(key, key_len);
 816
 817         head = find_bucket(table, hash);
 818         hlist_for_each_entry_rcu(flow, n, head, hash_node) {
 819
 820                 if (flow->hash == hash &&
 821                     !memcmp(&flow->key, key, key_len)) {
 822                         return flow;
 823                 }
 824         }
 825         return NULL;
 826 }
 827
 828 void flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 829 {
 830         struct hlist_head *head;
 831
 832         head = find_bucket(table, flow->hash);
 833         hlist_add_head_rcu(&flow->hash_node, head);
 834         table->count++;
 835 }
 836
 837 void flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 838 {
 839         if (!hlist_unhashed(&flow->hash_node)) {
 840                 hlist_del_init_rcu(&flow->hash_node);
 841                 table->count--;
 842                 BUG_ON(table->count < 0);
 843         }
 844 }
 845
 846 static int parse_tos_frag(struct sw_flow_key *swkey, u8 tos, u8 frag)
 847 {
 848         if (tos & INET_ECN_MASK || frag > OVS_FRAG_TYPE_MAX)
 849                 return -EINVAL;
 850
 851         swkey->ip.tos_frag = tos | frag;
 852         return 0;
 853 }
 854
 855 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 856 const u32 ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 857         [OVS_KEY_ATTR_PRIORITY] = 4,
 858         [OVS_KEY_ATTR_TUN_ID] = 8,
 859         [OVS_KEY_ATTR_IN_PORT] = 4,
 860         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 861         [OVS_KEY_ATTR_8021Q] = sizeof(struct ovs_key_8021q),
 862         [OVS_KEY_ATTR_ETHERTYPE] = 2,
 863         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 864         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 865         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 866         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 867         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 868         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 869         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 870         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 871 };
 872
 873 /**
 874  * flow_from_nlattrs - parses Netlink attributes into a flow key.
 875  * @swkey: receives the extracted flow key.
 876  * @key_lenp: number of bytes used in @swkey.
 877  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 878  * sequence.
 879  *
 880  * This state machine accepts the following forms, with [] for optional
 881  * elements and | for alternatives:
 882  *
 883  * [priority] [tun_id] [in_port] ethernet [8021q] [ethertype \
 884  *              [IPv4 [TCP|UDP|ICMP] | IPv6 [TCP|UDP|ICMPv6 [ND]] | ARP]]
 885  *
 886  * except that IPv4 or IPv6 terminates the sequence if its @ipv4_frag or
 887  * @ipv6_frag member, respectively, equals %OVS_FRAG_TYPE_LATER.
 888  */
 889 int flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 890                       const struct nlattr *attr)
 891 {
 892         int error = 0;
 893         enum ovs_frag_type frag_type;
 894         const struct nlattr *nla;
 895         u16 prev_type;
 896         int rem;
 897         int key_len;
 898
 899         memset(swkey, 0, sizeof(*swkey));
 900         swkey->phy.in_port = USHRT_MAX;
 901         swkey->eth.type = htons(ETH_P_802_2);
 902         key_len = SW_FLOW_KEY_OFFSET(eth);
 903
 904         prev_type = OVS_KEY_ATTR_UNSPEC;
 905         nla_for_each_nested(nla, attr, rem) {
 906                 const struct ovs_key_ethernet *eth_key;
 907                 const struct ovs_key_8021q *q_key;
 908                 const struct ovs_key_ipv4 *ipv4_key;
 909                 const struct ovs_key_ipv6 *ipv6_key;
 910                 const struct ovs_key_tcp *tcp_key;
 911                 const struct ovs_key_udp *udp_key;
 912                 const struct ovs_key_icmp *icmp_key;
 913                 const struct ovs_key_icmpv6 *icmpv6_key;
 914                 const struct ovs_key_arp *arp_key;
 915                 const struct ovs_key_nd *nd_key;
 916
 917                 int type = nla_type(nla);
 918
 919                 if (type > OVS_KEY_ATTR_MAX ||
 920                     nla_len(nla) != ovs_key_lens[type])
 921                         goto invalid;
 922
 923 #define TRANSITION(PREV_TYPE, TYPE) (((PREV_TYPE) << 16) | (TYPE))
 924                 switch (TRANSITION(prev_type, type)) {
 925                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_PRIORITY):
 926                         swkey->phy.priority = nla_get_u32(nla);
 927                         break;
 928
 929                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
 930                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_TUN_ID):
 931                         swkey->phy.tun_id = nla_get_be64(nla);
 932                         break;
 933
 934                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
 935                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_IN_PORT):
 936                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
 937                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
 938                                 goto invalid;
 939                         swkey->phy.in_port = nla_get_u32(nla);
 940                         break;
 941
 942                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_ETHERNET):
 943                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_ETHERNET):
 944                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_ETHERNET):
 945                 case TRANSITION(OVS_KEY_ATTR_IN_PORT, OVS_KEY_ATTR_ETHERNET):
 946                         eth_key = nla_data(nla);
 947                         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
 948                         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
 949                         break;
 950
 951                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_8021Q):
 952                         q_key = nla_data(nla);
 953                         /* Only standard 0x8100 VLANs currently supported. */
 954                         if (q_key->q_tpid != htons(ETH_P_8021Q))
 955                                 goto invalid;
 956                         if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
 957                                 goto invalid;
 958                         swkey->eth.tci = q_key->q_tci | htons(VLAN_TAG_PRESENT);
 959                         break;
 960
 961                 case TRANSITION(OVS_KEY_ATTR_8021Q, OVS_KEY_ATTR_ETHERTYPE):
 962                 case TRANSITION(OVS_KEY_ATTR_ETHERNET, OVS_KEY_ATTR_ETHERTYPE):
 963                         swkey->eth.type = nla_get_be16(nla);
 964                         if (ntohs(swkey->eth.type) < 1536)
 965                                 goto invalid;
 966                         break;
 967
 968                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV4):
 969                         key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 970                         if (swkey->eth.type != htons(ETH_P_IP))
 971                                 goto invalid;
 972                         ipv4_key = nla_data(nla);
 973                         swkey->ip.proto = ipv4_key->ipv4_proto;
 974                         if (parse_tos_frag(swkey, ipv4_key->ipv4_tos,
 975                                            ipv4_key->ipv4_frag))
 976                                 goto invalid;
 977                         swkey->ipv4.addr.src = ipv4_key->ipv4_src;
 978                         swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
 979                         break;
 980
 981                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_IPV6):
 982                         key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
 983                         if (swkey->eth.type != htons(ETH_P_IPV6))
 984                                 goto invalid;
 985                         ipv6_key = nla_data(nla);
 986                         swkey->ipv6.label = ipv6_key->ipv6_label;
 987                         swkey->ip.proto = ipv6_key->ipv6_proto;
 988                         if (parse_tos_frag(swkey, ipv6_key->ipv6_tos,
 989                                            ipv6_key->ipv6_frag))
 990                                 goto invalid;
 991                         memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
 992                                         sizeof(swkey->ipv6.addr.src));
 993                         memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
 994                                         sizeof(swkey->ipv6.addr.dst));
 995                         break;
 996
 997                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_TCP):
 998                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 999                         if (swkey->ip.proto != IPPROTO_TCP)
1000                                 goto invalid;
1001                         tcp_key = nla_data(nla);
1002                         swkey->ipv4.tp.src = tcp_key->tcp_src;
1003                         swkey->ipv4.tp.dst = tcp_key->tcp_dst;
1004                         break;
1005
1006                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_TCP):
1007                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1008                         if (swkey->ip.proto != IPPROTO_TCP)
1009                                 goto invalid;
1010                         tcp_key = nla_data(nla);
1011                         swkey->ipv6.tp.src = tcp_key->tcp_src;
1012                         swkey->ipv6.tp.dst = tcp_key->tcp_dst;
1013                         break;
1014
1015                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_UDP):
1016                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1017                         if (swkey->ip.proto != IPPROTO_UDP)
1018                                 goto invalid;
1019                         udp_key = nla_data(nla);
1020                         swkey->ipv4.tp.src = udp_key->udp_src;
1021                         swkey->ipv4.tp.dst = udp_key->udp_dst;
1022                         break;
1023
1024                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_UDP):
1025                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1026                         if (swkey->ip.proto != IPPROTO_UDP)
1027                                 goto invalid;
1028                         udp_key = nla_data(nla);
1029                         swkey->ipv6.tp.src = udp_key->udp_src;
1030                         swkey->ipv6.tp.dst = udp_key->udp_dst;
1031                         break;
1032
1033                 case TRANSITION(OVS_KEY_ATTR_IPV4, OVS_KEY_ATTR_ICMP):
1034                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
1035                         if (swkey->ip.proto != IPPROTO_ICMP)
1036                                 goto invalid;
1037                         icmp_key = nla_data(nla);
1038                         swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
1039                         swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
1040                         break;
1041
1042                 case TRANSITION(OVS_KEY_ATTR_IPV6, OVS_KEY_ATTR_ICMPV6):
1043                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
1044                         if (swkey->ip.proto != IPPROTO_ICMPV6)
1045                                 goto invalid;
1046                         icmpv6_key = nla_data(nla);
1047                         swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
1048                         swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
1049                         break;
1050
1051                 case TRANSITION(OVS_KEY_ATTR_ETHERTYPE, OVS_KEY_ATTR_ARP):
1052                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1053                         if (swkey->eth.type != htons(ETH_P_ARP))
1054                                 goto invalid;
1055                         arp_key = nla_data(nla);
1056                         swkey->ipv4.addr.src = arp_key->arp_sip;
1057                         swkey->ipv4.addr.dst = arp_key->arp_tip;
1058                         if (arp_key->arp_op & htons(0xff00))
1059                                 goto invalid;
1060                         swkey->ip.proto = ntohs(arp_key->arp_op);
1061                         memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1062                         memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1063                         break;
1064
1065                 case TRANSITION(OVS_KEY_ATTR_ICMPV6, OVS_KEY_ATTR_ND):
1066                         key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
1067                         if (swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_SOLICITATION)
1068                             && swkey->ipv6.tp.src != htons(NDISC_NEIGHBOUR_ADVERTISEMENT))
1069                                 goto invalid;
1070                         nd_key = nla_data(nla);
1071                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
1072                                         sizeof(swkey->ipv6.nd.target));
1073                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
1074                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
1075                         break;
1076
1077                 default:
1078                         goto invalid;
1079                 }
1080
1081                 prev_type = type;
1082         }
1083         if (rem)
1084                 goto invalid;
1085
1086         frag_type = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1087         switch (prev_type) {
1088         case OVS_KEY_ATTR_UNSPEC:
1089                 goto invalid;
1090
1091         case OVS_KEY_ATTR_PRIORITY:
1092         case OVS_KEY_ATTR_TUN_ID:
1093         case OVS_KEY_ATTR_IN_PORT:
1094                 goto invalid;
1095
1096         case OVS_KEY_ATTR_ETHERNET:
1097         case OVS_KEY_ATTR_8021Q:
1098                 goto ok;
1099
1100         case OVS_KEY_ATTR_ETHERTYPE:
1101                 if (swkey->eth.type == htons(ETH_P_IP) ||
1102                     swkey->eth.type == htons(ETH_P_IPV6) ||
1103                     swkey->eth.type == htons(ETH_P_ARP))
1104                         goto invalid;
1105                 goto ok;
1106
1107         case OVS_KEY_ATTR_IPV4:
1108                 if (frag_type == OVS_FRAG_TYPE_LATER)
1109                         goto ok;
1110                 if (swkey->ip.proto == IPPROTO_TCP ||
1111                     swkey->ip.proto == IPPROTO_UDP ||
1112                     swkey->ip.proto == IPPROTO_ICMP)
1113                         goto invalid;
1114                 goto ok;
1115
1116         case OVS_KEY_ATTR_IPV6:
1117                 if (frag_type == OVS_FRAG_TYPE_LATER)
1118                         goto ok;
1119                 if (swkey->ip.proto == IPPROTO_TCP ||
1120                     swkey->ip.proto == IPPROTO_UDP ||
1121                     swkey->ip.proto == IPPROTO_ICMPV6)
1122                         goto invalid;
1123                 goto ok;
1124
1125         case OVS_KEY_ATTR_ICMPV6:
1126                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
1127                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT) ||
1128                     frag_type == OVS_FRAG_TYPE_LATER)
1129                         goto invalid;
1130                 goto ok;
1131
1132         case OVS_KEY_ATTR_TCP:
1133         case OVS_KEY_ATTR_UDP:
1134         case OVS_KEY_ATTR_ICMP:
1135         case OVS_KEY_ATTR_ND:
1136                 if (frag_type == OVS_FRAG_TYPE_LATER)
1137                         goto invalid;
1138                 goto ok;
1139
1140         case OVS_KEY_ATTR_ARP:
1141                 goto ok;
1142
1143         default:
1144                 WARN_ON_ONCE(1);
1145         }
1146
1147 invalid:
1148         error = -EINVAL;
1149
1150 ok:
1151         *key_lenp = key_len;
1152         return error;
1153 }
1154
1155 /**
1156  * flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1157  * @in_port: receives the extracted input port.
1158  * @tun_id: receives the extracted tunnel ID.
1159  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1160  * sequence.
1161  *
1162  * This parses a series of Netlink attributes that form a flow key, which must
1163  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1164  * get the metadata, that is, the parts of the flow key that cannot be
1165  * extracted from the packet itself.
1166  */
1167 int flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1168                                const struct nlattr *attr)
1169 {
1170         const struct nlattr *nla;
1171         u16 prev_type;
1172         int rem;
1173
1174         *in_port = USHRT_MAX;
1175         *tun_id = 0;
1176         *priority = 0;
1177
1178         prev_type = OVS_KEY_ATTR_UNSPEC;
1179         nla_for_each_nested(nla, attr, rem) {
1180                 int type = nla_type(nla);
1181
1182                 if (type > OVS_KEY_ATTR_MAX || nla_len(nla) != ovs_key_lens[type])
1183                         return -EINVAL;
1184
1185                 switch (TRANSITION(prev_type, type)) {
1186                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_PRIORITY):
1187                         *priority = nla_get_u32(nla);
1188                         break;
1189
1190                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_TUN_ID):
1191                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_TUN_ID):
1192                         *tun_id = nla_get_be64(nla);
1193                         break;
1194
1195                 case TRANSITION(OVS_KEY_ATTR_UNSPEC, OVS_KEY_ATTR_IN_PORT):
1196                 case TRANSITION(OVS_KEY_ATTR_PRIORITY, OVS_KEY_ATTR_IN_PORT):
1197                 case TRANSITION(OVS_KEY_ATTR_TUN_ID, OVS_KEY_ATTR_IN_PORT):
1198                         if (nla_get_u32(nla) >= DP_MAX_PORTS)
1199                                 return -EINVAL;
1200                         *in_port = nla_get_u32(nla);
1201                         break;
1202
1203                 default:
1204                         return 0;
1205                 }
1206
1207                 prev_type = type;
1208         }
1209         if (rem)
1210                 return -EINVAL;
1211         return 0;
1212 }
1213
1214 int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1215 {
1216         struct ovs_key_ethernet *eth_key;
1217         struct nlattr *nla;
1218
1219         if (swkey->phy.priority)
1220                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1221
1222         if (swkey->phy.tun_id != cpu_to_be64(0))
1223                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1224
1225         if (swkey->phy.in_port != USHRT_MAX)
1226                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1227
1228         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1229         if (!nla)
1230                 goto nla_put_failure;
1231         eth_key = nla_data(nla);
1232         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1233         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1234
1235         if (swkey->eth.tci != htons(0)) {
1236                 struct ovs_key_8021q q_key;
1237
1238                 q_key.q_tpid = htons(ETH_P_8021Q);
1239                 q_key.q_tci = swkey->eth.tci & ~htons(VLAN_TAG_PRESENT);
1240                 NLA_PUT(skb, OVS_KEY_ATTR_8021Q, sizeof(q_key), &q_key);
1241         }
1242
1243         if (swkey->eth.type == htons(ETH_P_802_2))
1244                 return 0;
1245
1246         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1247
1248         if (swkey->eth.type == htons(ETH_P_IP)) {
1249                 struct ovs_key_ipv4 *ipv4_key;
1250
1251                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1252                 if (!nla)
1253                         goto nla_put_failure;
1254                 ipv4_key = nla_data(nla);
1255                 memset(ipv4_key, 0, sizeof(struct ovs_key_ipv4));
1256                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1257                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1258                 ipv4_key->ipv4_proto = swkey->ip.proto;
1259                 ipv4_key->ipv4_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1260                 ipv4_key->ipv4_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1261         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1262                 struct ovs_key_ipv6 *ipv6_key;
1263
1264                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1265                 if (!nla)
1266                         goto nla_put_failure;
1267                 ipv6_key = nla_data(nla);
1268                 memset(ipv6_key, 0, sizeof(struct ovs_key_ipv6));
1269                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1270                                 sizeof(ipv6_key->ipv6_src));
1271                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1272                                 sizeof(ipv6_key->ipv6_dst));
1273                 ipv6_key->ipv6_label = swkey->ipv6.label;
1274                 ipv6_key->ipv6_proto = swkey->ip.proto;
1275                 ipv6_key->ipv6_tos = swkey->ip.tos_frag & ~INET_ECN_MASK;
1276                 ipv6_key->ipv6_frag = swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK;
1277         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1278                 struct ovs_key_arp *arp_key;
1279
1280                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1281                 if (!nla)
1282                         goto nla_put_failure;
1283                 arp_key = nla_data(nla);
1284                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1285                 arp_key->arp_sip = swkey->ipv4.addr.src;
1286                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1287                 arp_key->arp_op = htons(swkey->ip.proto);
1288                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1289                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1290         }
1291
1292         if ((swkey->eth.type == htons(ETH_P_IP) ||
1293              swkey->eth.type == htons(ETH_P_IPV6)) &&
1294             (swkey->ip.tos_frag & OVS_FRAG_TYPE_MASK) != OVS_FRAG_TYPE_LATER) {
1295
1296                 if (swkey->ip.proto == IPPROTO_TCP) {
1297                         struct ovs_key_tcp *tcp_key;
1298
1299                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1300                         if (!nla)
1301                                 goto nla_put_failure;
1302                         tcp_key = nla_data(nla);
1303                         if (swkey->eth.type == htons(ETH_P_IP)) {
1304                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1305                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1306                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1307                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1308                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1309                         }
1310                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1311                         struct ovs_key_udp *udp_key;
1312
1313                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1314                         if (!nla)
1315                                 goto nla_put_failure;
1316                         udp_key = nla_data(nla);
1317                         if (swkey->eth.type == htons(ETH_P_IP)) {
1318                                 udp_key->udp_src = swkey->ipv4.tp.src;
1319                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1320                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1321                                 udp_key->udp_src = swkey->ipv6.tp.src;
1322                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1323                         }
1324                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1325                            swkey->ip.proto == IPPROTO_ICMP) {
1326                         struct ovs_key_icmp *icmp_key;
1327
1328                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1329                         if (!nla)
1330                                 goto nla_put_failure;
1331                         icmp_key = nla_data(nla);
1332                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1333                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1334                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1335                            swkey->ip.proto == IPPROTO_ICMPV6) {
1336                         struct ovs_key_icmpv6 *icmpv6_key;
1337
1338                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1339                                                 sizeof(*icmpv6_key));
1340                         if (!nla)
1341                                 goto nla_put_failure;
1342                         icmpv6_key = nla_data(nla);
1343                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1344                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1345
1346                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1347                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1348                                 struct ovs_key_nd *nd_key;
1349
1350                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1351                                 if (!nla)
1352                                         goto nla_put_failure;
1353                                 nd_key = nla_data(nla);
1354                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1355                                                         sizeof(nd_key->nd_target));
1356                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1357                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1358                         }
1359                 }
1360         }
1361
1362         return 0;
1363
1364 nla_put_failure:
1365         return -EMSGSIZE;
1366 }
1367
1368 /* Initializes the flow module.
1369  * Returns zero if successful or a negative error code. */
1370 int flow_init(void)
1371 {
1372         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1373                                         0, NULL);
1374         if (flow_cache == NULL)
1375                 return -ENOMEM;
1376
1377         get_random_bytes(&hash_seed, sizeof(hash_seed));
1378
1379         return 0;
1380 }
1381
1382 /* Uninitializes the flow module. */
1383 void flow_exit(void)
1384 {
1385         kmem_cache_destroy(flow_cache);
1386 }