datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #include "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 #include "vlan.h"
  47
  48 static struct kmem_cache *flow_cache;
  49
  50 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
  51                 struct sw_flow_key_range *range, u8 val);
  52
  53 static void update_range__(struct sw_flow_match *match,
  54                           size_t offset, size_t size, bool is_mask)
  55 {
  56         struct sw_flow_key_range *range = NULL;
  57         size_t start = offset;
  58         size_t end = offset + size;
  59
  60         if (!is_mask)
  61                 range = &match->range;
  62         else if (match->mask)
  63                 range = &match->mask->range;
  64
  65         if (!range)
  66                 return;
  67
  68         if (range->start == range->end) {
  69                 range->start = start;
  70                 range->end = end;
  71                 return;
  72         }
  73
  74         if (range->start > start)
  75                 range->start = start;
  76
  77         if (range->end < end)
  78                 range->end = end;
  79 }
  80
  81 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  82         do { \
  83                 update_range__(match, offsetof(struct sw_flow_key, field),  \
  84                                      sizeof((match)->key->field), is_mask); \
  85                 if (is_mask && match->mask != NULL) {                       \
  86                         (match)->mask->key.field = value;                   \
  87                 } else {                                                    \
  88                         (match)->key->field = value;                        \
  89                 }                                                           \
  90         } while (0)
  91
  92 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
  93         do { \
  94                 update_range__(match, offsetof(struct sw_flow_key, field),  \
  95                                 len, is_mask);                              \
  96                 if (is_mask && match->mask != NULL) {                       \
  97                         memcpy(&(match)->mask->key.field, value_p, len);    \
  98                 } else {                                                    \
  99                         memcpy(&(match)->key->field, value_p, len);         \
 100                 }                                                           \
 101         } while (0)
 102
 103 void ovs_match_init(struct sw_flow_match *match,
 104                     struct sw_flow_key *key,
 105                     struct sw_flow_mask *mask)
 106 {
 107         memset(match, 0, sizeof(*match));
 108         match->key = key;
 109         match->mask = mask;
 110
 111         memset(key, 0, sizeof(*key));
 112
 113         if (mask) {
 114                 memset(&mask->key, 0, sizeof(mask->key));
 115                 mask->range.start = mask->range.end = 0;
 116         }
 117 }
 118
 119 static bool ovs_match_validate(const struct sw_flow_match *match,
 120                 u64 key_attrs, u64 mask_attrs)
 121 {
 122         u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
 123         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 124
 125         /* The following mask attributes allowed only if they
 126          * pass the validation tests. */
 127         mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
 128                         | (1ULL << OVS_KEY_ATTR_IPV6)
 129                         | (1ULL << OVS_KEY_ATTR_TCP)
 130                         | (1ULL << OVS_KEY_ATTR_UDP)
 131                         | (1ULL << OVS_KEY_ATTR_ICMP)
 132                         | (1ULL << OVS_KEY_ATTR_ICMPV6)
 133                         | (1ULL << OVS_KEY_ATTR_ARP)
 134                         | (1ULL << OVS_KEY_ATTR_ND));
 135
 136         if (match->key->eth.type == htons(ETH_P_802_2) &&
 137             match->mask && (match->mask->key.eth.type == htons(0xffff)))
 138                 mask_allowed |= (1ULL << OVS_KEY_ATTR_ETHERTYPE);
 139
 140         /* Check key attributes. */
 141         if (match->key->eth.type == htons(ETH_P_ARP)
 142                         || match->key->eth.type == htons(ETH_P_RARP)) {
 143                 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
 144                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 145                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 146         }
 147
 148         if (match->key->eth.type == htons(ETH_P_IP)) {
 149                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 150                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 151                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
 152
 153                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 154                         if (match->key->ip.proto == IPPROTO_UDP) {
 155                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 156                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 157                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 158                         }
 159
 160                         if (match->key->ip.proto == IPPROTO_TCP) {
 161                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 162                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 163                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 164                         }
 165
 166                         if (match->key->ip.proto == IPPROTO_ICMP) {
 167                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 168                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 169                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 170                         }
 171                 }
 172         }
 173
 174         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 175                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 176                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 177                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
 178
 179                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 180                         if (match->key->ip.proto == IPPROTO_UDP) {
 181                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 182                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 183                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 184                         }
 185
 186                         if (match->key->ip.proto == IPPROTO_TCP) {
 187                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 188                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 189                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 190                         }
 191
 192                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 193                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 194                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 195                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 196
 197                                 if (match->key->ipv6.tp.src ==
 198                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 199                                     match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 200                                         key_expected |= 1ULL << OVS_KEY_ATTR_ND;
 201                                         if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
 202                                                 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
 203                                 }
 204                         }
 205                 }
 206         }
 207
 208         if ((key_attrs & key_expected) != key_expected)
 209                 /* Key attributes check failed. */
 210                 return false;
 211
 212         if ((mask_attrs & mask_allowed) != mask_attrs)
 213                 /* Mask attributes check failed. */
 214                 return false;
 215
 216         return true;
 217 }
 218
 219 static int check_header(struct sk_buff *skb, int len)
 220 {
 221         if (unlikely(skb->len < len))
 222                 return -EINVAL;
 223         if (unlikely(!pskb_may_pull(skb, len)))
 224                 return -ENOMEM;
 225         return 0;
 226 }
 227
 228 static bool arphdr_ok(struct sk_buff *skb)
 229 {
 230         return pskb_may_pull(skb, skb_network_offset(skb) +
 231                                   sizeof(struct arp_eth_header));
 232 }
 233
 234 static int check_iphdr(struct sk_buff *skb)
 235 {
 236         unsigned int nh_ofs = skb_network_offset(skb);
 237         unsigned int ip_len;
 238         int err;
 239
 240         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 241         if (unlikely(err))
 242                 return err;
 243
 244         ip_len = ip_hdrlen(skb);
 245         if (unlikely(ip_len < sizeof(struct iphdr) ||
 246                      skb->len < nh_ofs + ip_len))
 247                 return -EINVAL;
 248
 249         skb_set_transport_header(skb, nh_ofs + ip_len);
 250         return 0;
 251 }
 252
 253 static bool tcphdr_ok(struct sk_buff *skb)
 254 {
 255         int th_ofs = skb_transport_offset(skb);
 256         int tcp_len;
 257
 258         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 259                 return false;
 260
 261         tcp_len = tcp_hdrlen(skb);
 262         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 263                      skb->len < th_ofs + tcp_len))
 264                 return false;
 265
 266         return true;
 267 }
 268
 269 static bool udphdr_ok(struct sk_buff *skb)
 270 {
 271         return pskb_may_pull(skb, skb_transport_offset(skb) +
 272                                   sizeof(struct udphdr));
 273 }
 274
 275 static bool icmphdr_ok(struct sk_buff *skb)
 276 {
 277         return pskb_may_pull(skb, skb_transport_offset(skb) +
 278                                   sizeof(struct icmphdr));
 279 }
 280
 281 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 282 {
 283         struct timespec cur_ts;
 284         u64 cur_ms, idle_ms;
 285
 286         ktime_get_ts(&cur_ts);
 287         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 288         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 289                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 290
 291         return cur_ms - idle_ms;
 292 }
 293
 294 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 295 {
 296         unsigned int nh_ofs = skb_network_offset(skb);
 297         unsigned int nh_len;
 298         int payload_ofs;
 299         struct ipv6hdr *nh;
 300         uint8_t nexthdr;
 301         __be16 frag_off;
 302         int err;
 303
 304         err = check_header(skb, nh_ofs + sizeof(*nh));
 305         if (unlikely(err))
 306                 return err;
 307
 308         nh = ipv6_hdr(skb);
 309         nexthdr = nh->nexthdr;
 310         payload_ofs = (u8 *)(nh + 1) - skb->data;
 311
 312         key->ip.proto = NEXTHDR_NONE;
 313         key->ip.tos = ipv6_get_dsfield(nh);
 314         key->ip.ttl = nh->hop_limit;
 315         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 316         key->ipv6.addr.src = nh->saddr;
 317         key->ipv6.addr.dst = nh->daddr;
 318
 319         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 320         if (unlikely(payload_ofs < 0))
 321                 return -EINVAL;
 322
 323         if (frag_off) {
 324                 if (frag_off & htons(~0x7))
 325                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 326                 else
 327                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 328         }
 329
 330         nh_len = payload_ofs - nh_ofs;
 331         skb_set_transport_header(skb, nh_ofs + nh_len);
 332         key->ip.proto = nexthdr;
 333         return nh_len;
 334 }
 335
 336 static bool icmp6hdr_ok(struct sk_buff *skb)
 337 {
 338         return pskb_may_pull(skb, skb_transport_offset(skb) +
 339                                   sizeof(struct icmp6hdr));
 340 }
 341
 342 static void flow_key_mask(struct sw_flow_key *dst,
 343                           const struct sw_flow_key *src,
 344                           const struct sw_flow_mask *mask)
 345 {
 346         u8 *m = (u8 *)&mask->key + mask->range.start;
 347         u8 *s = (u8 *)src + mask->range.start;
 348         u8 *d = (u8 *)dst + mask->range.start;
 349         int i;
 350
 351         memset(dst, 0, sizeof(*dst));
 352         for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
 353                 *d = *s & *m;
 354                 d++, s++, m++;
 355         }
 356 }
 357
 358 #define TCP_FLAGS_OFFSET 13
 359 #define TCP_FLAG_MASK 0x3f
 360
 361 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 362 {
 363         u8 tcp_flags = 0;
 364
 365         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 366              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 367             flow->key.ip.proto == IPPROTO_TCP &&
 368             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 369                 u8 *tcp = (u8 *)tcp_hdr(skb);
 370                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 371         }
 372
 373         spin_lock(&flow->lock);
 374         flow->used = jiffies;
 375         flow->packet_count++;
 376         flow->byte_count += skb->len;
 377         flow->tcp_flags |= tcp_flags;
 378         spin_unlock(&flow->lock);
 379 }
 380
 381 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
 382 {
 383         struct sw_flow_actions *sfa;
 384
 385         if (size > MAX_ACTIONS_BUFSIZE)
 386                 return ERR_PTR(-EINVAL);
 387
 388         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 389         if (!sfa)
 390                 return ERR_PTR(-ENOMEM);
 391
 392         sfa->actions_len = 0;
 393         return sfa;
 394 }
 395
 396 struct sw_flow *ovs_flow_alloc(void)
 397 {
 398         struct sw_flow *flow;
 399
 400         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 401         if (!flow)
 402                 return ERR_PTR(-ENOMEM);
 403
 404         spin_lock_init(&flow->lock);
 405         flow->sf_acts = NULL;
 406         flow->mask = NULL;
 407
 408         return flow;
 409 }
 410
 411 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 412 {
 413         hash = jhash_1word(hash, table->hash_seed);
 414         return flex_array_get(table->buckets,
 415                                 (hash & (table->n_buckets - 1)));
 416 }
 417
 418 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 419 {
 420         struct flex_array *buckets;
 421         int i, err;
 422
 423         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 424                                    n_buckets, GFP_KERNEL);
 425         if (!buckets)
 426                 return NULL;
 427
 428         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 429         if (err) {
 430                 flex_array_free(buckets);
 431                 return NULL;
 432         }
 433
 434         for (i = 0; i < n_buckets; i++)
 435                 INIT_HLIST_HEAD((struct hlist_head *)
 436                                         flex_array_get(buckets, i));
 437
 438         return buckets;
 439 }
 440
 441 static void free_buckets(struct flex_array *buckets)
 442 {
 443         flex_array_free(buckets);
 444 }
 445
 446 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 447 {
 448         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 449
 450         if (!table)
 451                 return NULL;
 452
 453         table->buckets = alloc_buckets(new_size);
 454
 455         if (!table->buckets) {
 456                 kfree(table);
 457                 return NULL;
 458         }
 459         table->n_buckets = new_size;
 460         table->count = 0;
 461         table->node_ver = 0;
 462         table->keep_flows = false;
 463         get_random_bytes(&table->hash_seed, sizeof(u32));
 464         INIT_LIST_HEAD(&table->mask_list);
 465
 466         return table;
 467 }
 468
 469 static void __flow_tbl_destroy(struct flow_table *table)
 470 {
 471         int i;
 472
 473         if (table->keep_flows)
 474                 goto skip_flows;
 475
 476         for (i = 0; i < table->n_buckets; i++) {
 477                 struct sw_flow *flow;
 478                 struct hlist_head *head = flex_array_get(table->buckets, i);
 479                 struct hlist_node *n;
 480                 int ver = table->node_ver;
 481
 482                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 483                         hlist_del_rcu(&flow->hash_node[ver]);
 484                         ovs_flow_free(flow, false);
 485                 }
 486         }
 487
 488 skip_flows:
 489         free_buckets(table->buckets);
 490         kfree(table);
 491 }
 492
 493 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 494 {
 495         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 496
 497         __flow_tbl_destroy(table);
 498 }
 499
 500 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
 501 {
 502         if (!table)
 503                 return;
 504
 505         if (deferred)
 506                 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 507         else
 508                 __flow_tbl_destroy(table);
 509 }
 510
 511 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
 512 {
 513         struct sw_flow *flow;
 514         struct hlist_head *head;
 515         int ver;
 516         int i;
 517
 518         ver = table->node_ver;
 519         while (*bucket < table->n_buckets) {
 520                 i = 0;
 521                 head = flex_array_get(table->buckets, *bucket);
 522                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 523                         if (i < *last) {
 524                                 i++;
 525                                 continue;
 526                         }
 527                         *last = i + 1;
 528                         return flow;
 529                 }
 530                 (*bucket)++;
 531                 *last = 0;
 532         }
 533
 534         return NULL;
 535 }
 536
 537 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
 538 {
 539         struct hlist_head *head;
 540
 541         head = find_bucket(table, flow->hash);
 542         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 543
 544         table->count++;
 545 }
 546
 547 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 548 {
 549         int old_ver;
 550         int i;
 551
 552         old_ver = old->node_ver;
 553         new->node_ver = !old_ver;
 554
 555         /* Insert in new table. */
 556         for (i = 0; i < old->n_buckets; i++) {
 557                 struct sw_flow *flow;
 558                 struct hlist_head *head;
 559
 560                 head = flex_array_get(old->buckets, i);
 561
 562                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 563                         __tbl_insert(new, flow);
 564         }
 565
 566         new->mask_list = old->mask_list;
 567         old->keep_flows = true;
 568 }
 569
 570 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 571 {
 572         struct flow_table *new_table;
 573
 574         new_table = ovs_flow_tbl_alloc(n_buckets);
 575         if (!new_table)
 576                 return ERR_PTR(-ENOMEM);
 577
 578         flow_table_copy_flows(table, new_table);
 579
 580         return new_table;
 581 }
 582
 583 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 584 {
 585         return __flow_tbl_rehash(table, table->n_buckets);
 586 }
 587
 588 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 589 {
 590         return __flow_tbl_rehash(table, table->n_buckets * 2);
 591 }
 592
 593 static void __flow_free(struct sw_flow *flow)
 594 {
 595         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 596         kmem_cache_free(flow_cache, flow);
 597 }
 598
 599 static void rcu_free_flow_callback(struct rcu_head *rcu)
 600 {
 601         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 602
 603         __flow_free(flow);
 604 }
 605
 606 void ovs_flow_free(struct sw_flow *flow, bool deferred)
 607 {
 608         if (!flow)
 609                 return;
 610
 611         ovs_sw_flow_mask_del_ref((struct sw_flow_mask __force *)flow->mask,
 612                                  deferred);
 613
 614         if (deferred)
 615                 call_rcu(&flow->rcu, rcu_free_flow_callback);
 616         else
 617                 __flow_free(flow);
 618 }
 619
 620 /* RCU callback used by ovs_flow_deferred_free_acts. */
 621 static void rcu_free_acts_callback(struct rcu_head *rcu)
 622 {
 623         struct sw_flow_actions *sf_acts = container_of(rcu,
 624                         struct sw_flow_actions, rcu);
 625         kfree(sf_acts);
 626 }
 627
 628 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 629  * The caller must hold rcu_read_lock for this to be sensible. */
 630 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 631 {
 632         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 633 }
 634
 635 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 636 {
 637         struct qtag_prefix {
 638                 __be16 eth_type; /* ETH_P_8021Q */
 639                 __be16 tci;
 640         };
 641         struct qtag_prefix *qp;
 642
 643         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 644                 return 0;
 645
 646         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 647                                          sizeof(__be16))))
 648                 return -ENOMEM;
 649
 650         qp = (struct qtag_prefix *) skb->data;
 651         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 652         __skb_pull(skb, sizeof(struct qtag_prefix));
 653
 654         return 0;
 655 }
 656
 657 static __be16 parse_ethertype(struct sk_buff *skb)
 658 {
 659         struct llc_snap_hdr {
 660                 u8  dsap;  /* Always 0xAA */
 661                 u8  ssap;  /* Always 0xAA */
 662                 u8  ctrl;
 663                 u8  oui[3];
 664                 __be16 ethertype;
 665         };
 666         struct llc_snap_hdr *llc;
 667         __be16 proto;
 668
 669         proto = *(__be16 *) skb->data;
 670         __skb_pull(skb, sizeof(__be16));
 671
 672         if (ntohs(proto) >= ETH_P_802_3_MIN)
 673                 return proto;
 674
 675         if (skb->len < sizeof(struct llc_snap_hdr))
 676                 return htons(ETH_P_802_2);
 677
 678         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 679                 return htons(0);
 680
 681         llc = (struct llc_snap_hdr *) skb->data;
 682         if (llc->dsap != LLC_SAP_SNAP ||
 683             llc->ssap != LLC_SAP_SNAP ||
 684             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 685                 return htons(ETH_P_802_2);
 686
 687         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 688
 689         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 690                 return llc->ethertype;
 691
 692         return htons(ETH_P_802_2);
 693 }
 694
 695 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 696                         int nh_len)
 697 {
 698         struct icmp6hdr *icmp = icmp6_hdr(skb);
 699
 700         /* The ICMPv6 type and code fields use the 16-bit transport port
 701          * fields, so we need to store them in 16-bit network byte order.
 702          */
 703         key->ipv6.tp.src = htons(icmp->icmp6_type);
 704         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 705
 706         if (icmp->icmp6_code == 0 &&
 707             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 708              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 709                 int icmp_len = skb->len - skb_transport_offset(skb);
 710                 struct nd_msg *nd;
 711                 int offset;
 712
 713                 /* In order to process neighbor discovery options, we need the
 714                  * entire packet.
 715                  */
 716                 if (unlikely(icmp_len < sizeof(*nd)))
 717                         return 0;
 718
 719                 if (unlikely(skb_linearize(skb)))
 720                         return -ENOMEM;
 721
 722                 nd = (struct nd_msg *)skb_transport_header(skb);
 723                 key->ipv6.nd.target = nd->target;
 724
 725                 icmp_len -= sizeof(*nd);
 726                 offset = 0;
 727                 while (icmp_len >= 8) {
 728                         struct nd_opt_hdr *nd_opt =
 729                                  (struct nd_opt_hdr *)(nd->opt + offset);
 730                         int opt_len = nd_opt->nd_opt_len * 8;
 731
 732                         if (unlikely(!opt_len || opt_len > icmp_len))
 733                                 return 0;
 734
 735                         /* Store the link layer address if the appropriate
 736                          * option is provided.  It is considered an error if
 737                          * the same link layer option is specified twice.
 738                          */
 739                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 740                             && opt_len == 8) {
 741                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 742                                         goto invalid;
 743                                 memcpy(key->ipv6.nd.sll,
 744                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 745                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 746                                    && opt_len == 8) {
 747                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 748                                         goto invalid;
 749                                 memcpy(key->ipv6.nd.tll,
 750                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 751                         }
 752
 753                         icmp_len -= opt_len;
 754                         offset += opt_len;
 755                 }
 756         }
 757
 758         return 0;
 759
 760 invalid:
 761         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 762         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 763         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 764
 765         return 0;
 766 }
 767
 768 /**
 769  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 770  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 771  * Ethernet header
 772  * @in_port: port number on which @skb was received.
 773  * @key: output flow key
 774  * @key_lenp: length of output flow key
 775  *
 776  * The caller must ensure that skb->len >= ETH_HLEN.
 777  *
 778  * Returns 0 if successful, otherwise a negative errno value.
 779  *
 780  * Initializes @skb header pointers as follows:
 781  *
 782  *    - skb->mac_header: the Ethernet header.
 783  *
 784  *    - skb->network_header: just past the Ethernet header, or just past the
 785  *      VLAN header, to the first byte of the Ethernet payload.
 786  *
 787  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 788  *      on output, then just past the IP header, if one is present and
 789  *      of a correct length, otherwise the same as skb->network_header.
 790  *      For other key->eth.type values it is left untouched.
 791  */
 792 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
 793 {
 794         int error;
 795         struct ethhdr *eth;
 796
 797         memset(key, 0, sizeof(*key));
 798
 799         key->phy.priority = skb->priority;
 800         if (OVS_CB(skb)->tun_key)
 801                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 802         key->phy.in_port = in_port;
 803         key->phy.skb_mark = skb_get_mark(skb);
 804
 805         skb_reset_mac_header(skb);
 806
 807         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 808          * header in the linear data area.
 809          */
 810         eth = eth_hdr(skb);
 811         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 812         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 813
 814         __skb_pull(skb, 2 * ETH_ALEN);
 815         /* We are going to push all headers that we pull, so no need to
 816          * update skb->csum here. */
 817
 818         if (vlan_tx_tag_present(skb))
 819                 key->eth.tci = htons(vlan_get_tci(skb));
 820         else if (eth->h_proto == htons(ETH_P_8021Q))
 821                 if (unlikely(parse_vlan(skb, key)))
 822                         return -ENOMEM;
 823
 824         key->eth.type = parse_ethertype(skb);
 825         if (unlikely(key->eth.type == htons(0)))
 826                 return -ENOMEM;
 827
 828         skb_reset_network_header(skb);
 829         __skb_push(skb, skb->data - skb_mac_header(skb));
 830
 831         /* Network layer. */
 832         if (key->eth.type == htons(ETH_P_IP)) {
 833                 struct iphdr *nh;
 834                 __be16 offset;
 835
 836                 error = check_iphdr(skb);
 837                 if (unlikely(error)) {
 838                         if (error == -EINVAL) {
 839                                 skb->transport_header = skb->network_header;
 840                                 error = 0;
 841                         }
 842                         return error;
 843                 }
 844
 845                 nh = ip_hdr(skb);
 846                 key->ipv4.addr.src = nh->saddr;
 847                 key->ipv4.addr.dst = nh->daddr;
 848
 849                 key->ip.proto = nh->protocol;
 850                 key->ip.tos = nh->tos;
 851                 key->ip.ttl = nh->ttl;
 852
 853                 offset = nh->frag_off & htons(IP_OFFSET);
 854                 if (offset) {
 855                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 856                         return 0;
 857                 }
 858                 if (nh->frag_off & htons(IP_MF) ||
 859                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 860                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 861
 862                 /* Transport layer. */
 863                 if (key->ip.proto == IPPROTO_TCP) {
 864                         if (tcphdr_ok(skb)) {
 865                                 struct tcphdr *tcp = tcp_hdr(skb);
 866                                 key->ipv4.tp.src = tcp->source;
 867                                 key->ipv4.tp.dst = tcp->dest;
 868                         }
 869                 } else if (key->ip.proto == IPPROTO_UDP) {
 870                         if (udphdr_ok(skb)) {
 871                                 struct udphdr *udp = udp_hdr(skb);
 872                                 key->ipv4.tp.src = udp->source;
 873                                 key->ipv4.tp.dst = udp->dest;
 874                         }
 875                 } else if (key->ip.proto == IPPROTO_ICMP) {
 876                         if (icmphdr_ok(skb)) {
 877                                 struct icmphdr *icmp = icmp_hdr(skb);
 878                                 /* The ICMP type and code fields use the 16-bit
 879                                  * transport port fields, so we need to store
 880                                  * them in 16-bit network byte order. */
 881                                 key->ipv4.tp.src = htons(icmp->type);
 882                                 key->ipv4.tp.dst = htons(icmp->code);
 883                         }
 884                 }
 885
 886         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 887                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 888                 struct arp_eth_header *arp;
 889
 890                 arp = (struct arp_eth_header *)skb_network_header(skb);
 891
 892                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 893                                 && arp->ar_pro == htons(ETH_P_IP)
 894                                 && arp->ar_hln == ETH_ALEN
 895                                 && arp->ar_pln == 4) {
 896
 897                         /* We only match on the lower 8 bits of the opcode. */
 898                         if (ntohs(arp->ar_op) <= 0xff)
 899                                 key->ip.proto = ntohs(arp->ar_op);
 900                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 901                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 902                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 903                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 904                 }
 905         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 906                 int nh_len;             /* IPv6 Header + Extensions */
 907
 908                 nh_len = parse_ipv6hdr(skb, key);
 909                 if (unlikely(nh_len < 0)) {
 910                         if (nh_len == -EINVAL) {
 911                                 skb->transport_header = skb->network_header;
 912                                 error = 0;
 913                         } else {
 914                                 error = nh_len;
 915                         }
 916                         return error;
 917                 }
 918
 919                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 920                         return 0;
 921                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 922                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 923
 924                 /* Transport layer. */
 925                 if (key->ip.proto == NEXTHDR_TCP) {
 926                         if (tcphdr_ok(skb)) {
 927                                 struct tcphdr *tcp = tcp_hdr(skb);
 928                                 key->ipv6.tp.src = tcp->source;
 929                                 key->ipv6.tp.dst = tcp->dest;
 930                         }
 931                 } else if (key->ip.proto == NEXTHDR_UDP) {
 932                         if (udphdr_ok(skb)) {
 933                                 struct udphdr *udp = udp_hdr(skb);
 934                                 key->ipv6.tp.src = udp->source;
 935                                 key->ipv6.tp.dst = udp->dest;
 936                         }
 937                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 938                         if (icmp6hdr_ok(skb)) {
 939                                 error = parse_icmpv6(skb, key, nh_len);
 940                                 if (error)
 941                                         return error;
 942                         }
 943                 }
 944         }
 945
 946         return 0;
 947 }
 948
 949 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 950 {
 951         return jhash2((u32 *)((u8 *)key + key_start),
 952                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 953 }
 954
 955 static int flow_key_start(const struct sw_flow_key *key)
 956 {
 957         if (key->tun_key.ipv4_dst)
 958                 return 0;
 959         else
 960                 return offsetof(struct sw_flow_key, phy);
 961 }
 962
 963 static bool __cmp_key(const struct sw_flow_key *key1,
 964                 const struct sw_flow_key *key2,  int key_start, int key_len)
 965 {
 966         return !memcmp((u8 *)key1 + key_start,
 967                         (u8 *)key2 + key_start, (key_len - key_start));
 968 }
 969
 970 static bool __flow_cmp_key(const struct sw_flow *flow,
 971                 const struct sw_flow_key *key, int key_start, int key_len)
 972 {
 973         return __cmp_key(&flow->key, key, key_start, key_len);
 974 }
 975
 976 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
 977                   const struct sw_flow_key *key, int key_start, int key_len)
 978 {
 979         return __cmp_key(&flow->unmasked_key, key, key_start, key_len);
 980 }
 981
 982 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
 983                 const struct sw_flow_key *key, int key_len)
 984 {
 985         int key_start;
 986         key_start = flow_key_start(key);
 987
 988         return __flow_cmp_unmasked_key(flow, key, key_start, key_len);
 989
 990 }
 991
 992 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
 993                                        struct sw_flow_match *match)
 994 {
 995         struct sw_flow_key *unmasked = match->key;
 996         int key_len = match->range.end;
 997         struct sw_flow *flow;
 998
 999         flow = ovs_flow_lookup(table, unmasked);
1000         if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_len)))
1001                 flow = NULL;
1002
1003         return flow;
1004 }
1005
1006 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1007                                     const struct sw_flow_key *flow_key,
1008                                     struct sw_flow_mask *mask)
1009 {
1010         struct sw_flow *flow;
1011         struct hlist_head *head;
1012         int key_start = mask->range.start;
1013         int key_len = mask->range.end;
1014         u32 hash;
1015         struct sw_flow_key masked_key;
1016
1017         flow_key_mask(&masked_key, flow_key, mask);
1018         hash = ovs_flow_hash(&masked_key, key_start, key_len);
1019         head = find_bucket(table, hash);
1020         hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1021                 if (__flow_cmp_key(flow, &masked_key, key_start, key_len))
1022                         return flow;
1023         }
1024         return NULL;
1025 }
1026
1027 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1028                                 const struct sw_flow_key *key)
1029 {
1030         struct sw_flow *flow = NULL;
1031         struct sw_flow_mask *mask;
1032
1033         list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
1034                 flow = ovs_masked_flow_lookup(tbl, key, mask);
1035                 if (flow)  /* Found */
1036                         break;
1037         }
1038
1039         return flow;
1040 }
1041
1042
1043 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow,
1044                          const struct sw_flow_key *key, int key_len)
1045 {
1046         flow->unmasked_key = *key;
1047         flow_key_mask(&flow->key, &flow->unmasked_key, ovsl_dereference(flow->mask));
1048         flow->hash = ovs_flow_hash(&flow->key,
1049                         ovsl_dereference(flow->mask)->range.start,
1050                         ovsl_dereference(flow->mask)->range.end);
1051         __tbl_insert(table, flow);
1052 }
1053
1054 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1055 {
1056         BUG_ON(table->count == 0);
1057         hlist_del_rcu(&flow->hash_node[table->node_ver]);
1058         table->count--;
1059 }
1060
1061 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
1062 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1063         [OVS_KEY_ATTR_ENCAP] = -1,
1064         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1065         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1066         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1067         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1068         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1069         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1070         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1071         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1072         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1073         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1074         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1075         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1076         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1077         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1078         [OVS_KEY_ATTR_TUNNEL] = -1,
1079 };
1080
1081 static bool is_all_zero(const u8 *fp, size_t size)
1082 {
1083         int i;
1084
1085         if (!fp)
1086                 return false;
1087
1088         for (i = 0; i < size; i++)
1089                 if (fp[i])
1090                         return false;
1091
1092         return true;
1093 }
1094
1095 static int __parse_flow_nlattrs(const struct nlattr *attr,
1096                               const struct nlattr *a[],
1097                               u64 *attrsp, bool nz)
1098 {
1099         const struct nlattr *nla;
1100         u64 attrs;
1101         int rem;
1102
1103         attrs = *attrsp;
1104         nla_for_each_nested(nla, attr, rem) {
1105                 u16 type = nla_type(nla);
1106                 int expected_len;
1107
1108                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
1109                         return -EINVAL;
1110
1111                 expected_len = ovs_key_lens[type];
1112                 if (nla_len(nla) != expected_len && expected_len != -1)
1113                         return -EINVAL;
1114
1115                 if (attrs & (1ULL << type))
1116                         /* Duplicated field. */
1117                         return -EINVAL;
1118
1119                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1120                         attrs |= 1ULL << type;
1121                         a[type] = nla;
1122                 }
1123         }
1124         if (rem)
1125                 return -EINVAL;
1126
1127         *attrsp = attrs;
1128         return 0;
1129 }
1130
1131 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1132                               const struct nlattr *a[], u64 *attrsp)
1133 {
1134         return __parse_flow_nlattrs(attr, a, attrsp, true);
1135 }
1136
1137 static int parse_flow_nlattrs(const struct nlattr *attr,
1138                               const struct nlattr *a[], u64 *attrsp)
1139 {
1140         return __parse_flow_nlattrs(attr, a, attrsp, false);
1141 }
1142
1143 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1144                          struct sw_flow_match *match, bool is_mask)
1145 {
1146         struct nlattr *a;
1147         int rem;
1148         bool ttl = false;
1149         u16 tun_flags = 0;
1150
1151         nla_for_each_nested(a, attr, rem) {
1152                 int type = nla_type(a);
1153                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1154                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1155                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1156                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1157                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1158                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1159                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1160                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1161                 };
1162
1163                 if (type > OVS_TUNNEL_KEY_ATTR_MAX ||
1164                         ovs_tunnel_key_lens[type] != nla_len(a))
1165                         return -EINVAL;
1166
1167                 switch (type) {
1168                 case OVS_TUNNEL_KEY_ATTR_ID:
1169                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1170                                         nla_get_be64(a), is_mask);
1171                         tun_flags |= OVS_TNL_F_KEY;
1172                         break;
1173                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1174                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1175                                         nla_get_be32(a), is_mask);
1176                         break;
1177                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1178                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1179                                         nla_get_be32(a), is_mask);
1180                         break;
1181                 case OVS_TUNNEL_KEY_ATTR_TOS:
1182                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1183                                         nla_get_u8(a), is_mask);
1184                         break;
1185                 case OVS_TUNNEL_KEY_ATTR_TTL:
1186                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1187                                         nla_get_u8(a), is_mask);
1188                         ttl = true;
1189                         break;
1190                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1191                         tun_flags |= OVS_TNL_F_DONT_FRAGMENT;
1192                         break;
1193                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1194                         tun_flags |= OVS_TNL_F_CSUM;
1195                         break;
1196                 default:
1197                         return -EINVAL;
1198                 }
1199         }
1200
1201         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1202
1203         if (rem > 0)
1204                 return -EINVAL;
1205
1206         if (!match->key->tun_key.ipv4_dst)
1207                 return -EINVAL;
1208
1209         if (!ttl)
1210                 return -EINVAL;
1211
1212         return 0;
1213 }
1214
1215 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1216                         const struct ovs_key_ipv4_tunnel *tun_key,
1217                         const struct ovs_key_ipv4_tunnel *output)
1218 {
1219         struct nlattr *nla;
1220
1221         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1222         if (!nla)
1223                 return -EMSGSIZE;
1224
1225         if (tun_key->tun_flags & OVS_TNL_F_KEY &&
1226             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1227                 return -EMSGSIZE;
1228         if (tun_key->ipv4_src &&
1229             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1230                 return -EMSGSIZE;
1231         if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1232                 return -EMSGSIZE;
1233         if (tun_key->ipv4_tos &&
1234             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1235                 return -EMSGSIZE;
1236         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1237                 return -EMSGSIZE;
1238         if ((tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT) &&
1239                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1240                 return -EMSGSIZE;
1241         if ((tun_key->tun_flags & OVS_TNL_F_CSUM) &&
1242                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1243                 return -EMSGSIZE;
1244
1245         nla_nest_end(skb, nla);
1246         return 0;
1247 }
1248
1249
1250 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
1251                 const struct nlattr **a, bool is_mask)
1252 {
1253         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1254                 SW_FLOW_KEY_PUT(match, phy.priority,
1255                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1256                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1257         }
1258
1259         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1260                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1261
1262                 if (!is_mask && in_port >= DP_MAX_PORTS)
1263                         return -EINVAL;
1264                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1265                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1266         }
1267
1268         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1269                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1270 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1271                 if (!is_mask && mark != 0)
1272                         return -EINVAL;
1273 #endif
1274                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1275                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1276         }
1277         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1278                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1279                                         is_mask))
1280                         return -EINVAL;
1281                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1282         }
1283         return 0;
1284 }
1285
1286 static int ovs_key_from_nlattrs(struct sw_flow_match *match,  u64 attrs,
1287                 const struct nlattr **a, bool is_mask)
1288 {
1289         int err;
1290
1291         err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1292         if (err)
1293                 return err;
1294
1295         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1296                 const struct ovs_key_ethernet *eth_key;
1297
1298                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1299                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1300                                 eth_key->eth_src, ETH_ALEN, is_mask);
1301                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1302                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1303                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1304         }
1305
1306         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1307                 __be16 tci;
1308
1309                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1310                 if (!is_mask && (tci & htons(VLAN_TAG_PRESENT)))
1311                         return -EINVAL;
1312
1313                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1314                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1315         }
1316
1317         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1318                 __be16 eth_type;
1319
1320                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1321                 if (!is_mask && ntohs(eth_type) < ETH_P_802_3_MIN)
1322                         return -EINVAL;
1323
1324                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1325                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1326         } else if (!is_mask) {
1327                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1328         }
1329
1330         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1331                 const struct ovs_key_ipv4 *ipv4_key;
1332
1333                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1334                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1335                         return -EINVAL;
1336                 SW_FLOW_KEY_PUT(match, ip.proto,
1337                                 ipv4_key->ipv4_proto, is_mask);
1338                 SW_FLOW_KEY_PUT(match, ip.tos,
1339                                 ipv4_key->ipv4_tos, is_mask);
1340                 SW_FLOW_KEY_PUT(match, ip.ttl,
1341                                 ipv4_key->ipv4_ttl, is_mask);
1342                 SW_FLOW_KEY_PUT(match, ip.frag,
1343                                 ipv4_key->ipv4_frag, is_mask);
1344                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1345                                 ipv4_key->ipv4_src, is_mask);
1346                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1347                                 ipv4_key->ipv4_dst, is_mask);
1348                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1349         }
1350
1351         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1352                 const struct ovs_key_ipv6 *ipv6_key;
1353
1354                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1355                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1356                         return -EINVAL;
1357                 SW_FLOW_KEY_PUT(match, ipv6.label,
1358                                 ipv6_key->ipv6_label, is_mask);
1359                 SW_FLOW_KEY_PUT(match, ip.proto,
1360                                 ipv6_key->ipv6_proto, is_mask);
1361                 SW_FLOW_KEY_PUT(match, ip.tos,
1362                                 ipv6_key->ipv6_tclass, is_mask);
1363                 SW_FLOW_KEY_PUT(match, ip.ttl,
1364                                 ipv6_key->ipv6_hlimit, is_mask);
1365                 SW_FLOW_KEY_PUT(match, ip.frag,
1366                                 ipv6_key->ipv6_frag, is_mask);
1367                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1368                                 ipv6_key->ipv6_src,
1369                                 sizeof(match->key->ipv6.addr.src),
1370                                 is_mask);
1371                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1372                                 ipv6_key->ipv6_dst,
1373                                 sizeof(match->key->ipv6.addr.dst),
1374                                 is_mask);
1375
1376                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1377         }
1378
1379         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1380                 const struct ovs_key_arp *arp_key;
1381
1382                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1383                 if (!is_mask && (arp_key->arp_op & htons(0xff00)))
1384                         return -EINVAL;
1385
1386                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1387                                 arp_key->arp_sip, is_mask);
1388                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1389                         arp_key->arp_tip, is_mask);
1390                 SW_FLOW_KEY_PUT(match, ip.proto,
1391                                 ntohs(arp_key->arp_op), is_mask);
1392                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1393                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1394                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1395                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1396
1397                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1398         }
1399
1400         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1401                 const struct ovs_key_tcp *tcp_key;
1402
1403                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1404                 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1405                                 tcp_key->tcp_src, is_mask);
1406                 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1407                                 tcp_key->tcp_dst, is_mask);
1408                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1409         }
1410
1411         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1412                 const struct ovs_key_udp *udp_key;
1413
1414                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1415                 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1416                                 udp_key->udp_src, is_mask);
1417                 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1418                                 udp_key->udp_dst, is_mask);
1419                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1420         }
1421
1422         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1423                 const struct ovs_key_icmp *icmp_key;
1424
1425                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1426                 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1427                                 htons(icmp_key->icmp_type), is_mask);
1428                 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1429                                 htons(icmp_key->icmp_code), is_mask);
1430                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1431         }
1432
1433         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1434                 const struct ovs_key_icmpv6 *icmpv6_key;
1435
1436                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1437                 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1438                                 htons(icmpv6_key->icmpv6_type), is_mask);
1439                 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1440                                 htons(icmpv6_key->icmpv6_code), is_mask);
1441                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1442         }
1443
1444         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1445                 const struct ovs_key_nd *nd_key;
1446
1447                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1448                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1449                         nd_key->nd_target,
1450                         sizeof(match->key->ipv6.nd.target),
1451                         is_mask);
1452                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1453                         nd_key->nd_sll, ETH_ALEN, is_mask);
1454                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1455                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1456                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1457         }
1458
1459         if (attrs != 0)
1460                 return -EINVAL;
1461
1462         return 0;
1463 }
1464
1465 /**
1466  * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1467  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1468  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1469  * does not include any don't care bit.
1470  * @match: receives the extracted flow match information.
1471  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1472  * sequence. The fields should of the packet that triggered the creation
1473  * of this flow.
1474  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1475  * attribute specifies the mask field of the wildcarded flow.
1476  */
1477 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1478                            const struct nlattr *key,
1479                            const struct nlattr *mask)
1480 {
1481         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1482         const struct nlattr *m[OVS_KEY_ATTR_MAX + 1];
1483         const struct nlattr *encap;
1484         u64 key_attrs = 0;
1485         u64 mask_attrs = 0;
1486         bool encap_valid = false;
1487         int err;
1488
1489         err = parse_flow_nlattrs(key, a, &key_attrs);
1490         if (err)
1491                 return err;
1492
1493         if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1494                 encap = a[OVS_KEY_ATTR_ENCAP];
1495                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1496                 if (nla_len(encap)) {
1497                         __be16 eth_type = 0; /* ETH_P_8021Q */
1498
1499                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1500                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1501
1502                         if  ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1503                                 encap_valid = true;
1504                                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1505                                 err = parse_flow_nlattrs(encap, a, &key_attrs);
1506                         } else
1507                                 err = -EINVAL;
1508
1509                         if (err)
1510                                 return err;
1511                 }
1512         }
1513
1514         err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1515         if (err)
1516                 return err;
1517
1518         if (mask) {
1519                 err = parse_flow_mask_nlattrs(mask, m, &mask_attrs);
1520                 if (err)
1521                         return err;
1522
1523                 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1524                         __be16 eth_type = 0;
1525
1526                         if (m[OVS_KEY_ATTR_ETHERTYPE])
1527                                 eth_type = nla_get_be16(m[OVS_KEY_ATTR_ETHERTYPE]);
1528                         if (eth_type == htons(0xffff)) {
1529                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1530                                 encap = m[OVS_KEY_ATTR_ENCAP];
1531                                 err = parse_flow_mask_nlattrs(encap, m, &mask_attrs);
1532                         } else
1533                                 err = -EINVAL;
1534
1535                         if (err)
1536                                 return err;
1537                 }
1538
1539                 err = ovs_key_from_nlattrs(match,  mask_attrs, m, true);
1540                 if (err)
1541                         return err;
1542         } else {
1543                 /* Populate exact match flow's key mask. */
1544                 if (match->mask)
1545                         ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1546         }
1547
1548         if (!ovs_match_validate(match, key_attrs, mask_attrs))
1549                 return -EINVAL;
1550
1551         return 0;
1552 }
1553
1554 /**
1555  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1556  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1557  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1558  * sequence.
1559  *
1560  * This parses a series of Netlink attributes that form a flow key, which must
1561  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1562  * get the metadata, that is, the parts of the flow key that cannot be
1563  * extracted from the packet itself.
1564  */
1565
1566 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1567                 const struct nlattr *attr)
1568 {
1569         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1570         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1571         u64 attrs = 0;
1572         int err;
1573         struct sw_flow_match match;
1574
1575         flow->key.phy.in_port = DP_MAX_PORTS;
1576         flow->key.phy.priority = 0;
1577         flow->key.phy.skb_mark = 0;
1578         memset(tun_key, 0, sizeof(flow->key.tun_key));
1579
1580         err = parse_flow_nlattrs(attr, a, &attrs);
1581         if (err)
1582                 return -EINVAL;
1583
1584         ovs_match_init(&match, &flow->key, NULL);
1585
1586         err = metadata_from_nlattrs(&match, &attrs, a, false);
1587         if (err)
1588                 return err;
1589
1590         return 0;
1591 }
1592
1593 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1594                 const struct sw_flow_key *output, struct sk_buff *skb)
1595 {
1596         struct ovs_key_ethernet *eth_key;
1597         struct nlattr *nla, *encap;
1598
1599         if (swkey->phy.priority &&
1600             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1601                 goto nla_put_failure;
1602
1603         if (swkey->tun_key.ipv4_dst &&
1604             ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1605                 goto nla_put_failure;
1606
1607         if (swkey->phy.in_port != DP_MAX_PORTS) {
1608                 /* Exact match upper 16 bits. */
1609                 u16 upper_u16;
1610                 upper_u16 = (swkey == output) ? 0 : 0xffff;
1611
1612                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1613                                         (upper_u16 << 16) | output->phy.in_port))
1614                         goto nla_put_failure;
1615         }
1616
1617         if (swkey->phy.skb_mark &&
1618             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1619                 goto nla_put_failure;
1620
1621         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1622         if (!nla)
1623                 goto nla_put_failure;
1624
1625         eth_key = nla_data(nla);
1626         memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1627         memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1628
1629         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1630                 __be16 eth_type;
1631                 eth_type = (swkey == output) ? htons(ETH_P_8021Q) : htons(0xffff) ;
1632                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1633                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1634                         goto nla_put_failure;
1635                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1636                 if (!swkey->eth.tci)
1637                         goto unencap;
1638         } else
1639                 encap = NULL;
1640
1641         if ((swkey == output) && (swkey->eth.type == htons(ETH_P_802_2)))
1642                 goto unencap;
1643
1644         if (output->eth.type != 0)
1645                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1646                         goto nla_put_failure;
1647
1648         if (swkey->eth.type == htons(ETH_P_IP)) {
1649                 struct ovs_key_ipv4 *ipv4_key;
1650
1651                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1652                 if (!nla)
1653                         goto nla_put_failure;
1654                 ipv4_key = nla_data(nla);
1655                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1656                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1657                 ipv4_key->ipv4_proto = output->ip.proto;
1658                 ipv4_key->ipv4_tos = output->ip.tos;
1659                 ipv4_key->ipv4_ttl = output->ip.ttl;
1660                 ipv4_key->ipv4_frag = output->ip.frag;
1661         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1662                 struct ovs_key_ipv6 *ipv6_key;
1663
1664                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1665                 if (!nla)
1666                         goto nla_put_failure;
1667                 ipv6_key = nla_data(nla);
1668                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1669                                 sizeof(ipv6_key->ipv6_src));
1670                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1671                                 sizeof(ipv6_key->ipv6_dst));
1672                 ipv6_key->ipv6_label = output->ipv6.label;
1673                 ipv6_key->ipv6_proto = output->ip.proto;
1674                 ipv6_key->ipv6_tclass = output->ip.tos;
1675                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1676                 ipv6_key->ipv6_frag = output->ip.frag;
1677         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1678                    swkey->eth.type == htons(ETH_P_RARP)) {
1679                 struct ovs_key_arp *arp_key;
1680
1681                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1682                 if (!nla)
1683                         goto nla_put_failure;
1684                 arp_key = nla_data(nla);
1685                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1686                 arp_key->arp_sip = output->ipv4.addr.src;
1687                 arp_key->arp_tip = output->ipv4.addr.dst;
1688                 arp_key->arp_op = htons(output->ip.proto);
1689                 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1690                 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1691         }
1692
1693         if ((swkey->eth.type == htons(ETH_P_IP) ||
1694              swkey->eth.type == htons(ETH_P_IPV6)) &&
1695              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1696
1697                 if (swkey->ip.proto == IPPROTO_TCP) {
1698                         struct ovs_key_tcp *tcp_key;
1699
1700                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1701                         if (!nla)
1702                                 goto nla_put_failure;
1703                         tcp_key = nla_data(nla);
1704                         if (swkey->eth.type == htons(ETH_P_IP)) {
1705                                 tcp_key->tcp_src = output->ipv4.tp.src;
1706                                 tcp_key->tcp_dst = output->ipv4.tp.dst;
1707                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1708                                 tcp_key->tcp_src = output->ipv6.tp.src;
1709                                 tcp_key->tcp_dst = output->ipv6.tp.dst;
1710                         }
1711                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1712                         struct ovs_key_udp *udp_key;
1713
1714                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1715                         if (!nla)
1716                                 goto nla_put_failure;
1717                         udp_key = nla_data(nla);
1718                         if (swkey->eth.type == htons(ETH_P_IP)) {
1719                                 udp_key->udp_src = output->ipv4.tp.src;
1720                                 udp_key->udp_dst = output->ipv4.tp.dst;
1721                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1722                                 udp_key->udp_src = output->ipv6.tp.src;
1723                                 udp_key->udp_dst = output->ipv6.tp.dst;
1724                         }
1725                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1726                            swkey->ip.proto == IPPROTO_ICMP) {
1727                         struct ovs_key_icmp *icmp_key;
1728
1729                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1730                         if (!nla)
1731                                 goto nla_put_failure;
1732                         icmp_key = nla_data(nla);
1733                         icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1734                         icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1735                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1736                            swkey->ip.proto == IPPROTO_ICMPV6) {
1737                         struct ovs_key_icmpv6 *icmpv6_key;
1738
1739                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1740                                                 sizeof(*icmpv6_key));
1741                         if (!nla)
1742                                 goto nla_put_failure;
1743                         icmpv6_key = nla_data(nla);
1744                         icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1745                         icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1746
1747                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1748                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1749                                 struct ovs_key_nd *nd_key;
1750
1751                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1752                                 if (!nla)
1753                                         goto nla_put_failure;
1754                                 nd_key = nla_data(nla);
1755                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1756                                                         sizeof(nd_key->nd_target));
1757                                 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1758                                 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1759                         }
1760                 }
1761         }
1762
1763 unencap:
1764         if (encap)
1765                 nla_nest_end(skb, encap);
1766
1767         return 0;
1768
1769 nla_put_failure:
1770         return -EMSGSIZE;
1771 }
1772
1773 /* Initializes the flow module.
1774  * Returns zero if successful or a negative error code. */
1775 int ovs_flow_init(void)
1776 {
1777         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1778                                         0, NULL);
1779         if (flow_cache == NULL)
1780                 return -ENOMEM;
1781
1782         return 0;
1783 }
1784
1785 /* Uninitializes the flow module. */
1786 void ovs_flow_exit(void)
1787 {
1788         kmem_cache_destroy(flow_cache);
1789 }
1790
1791 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1792 {
1793         struct sw_flow_mask *mask;
1794
1795         mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1796         if (mask)
1797                 mask->ref_count = 0;
1798
1799         return mask;
1800 }
1801
1802 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1803 {
1804         mask->ref_count++;
1805 }
1806
1807 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1808 {
1809         struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1810
1811         kfree(mask);
1812 }
1813
1814 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1815 {
1816         if (!mask)
1817                 return;
1818
1819         BUG_ON(!mask->ref_count);
1820         mask->ref_count--;
1821
1822         if (!mask->ref_count) {
1823                 list_del_rcu(&mask->list);
1824                 if (deferred)
1825                         call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1826                 else
1827                         kfree(mask);
1828         }
1829 }
1830
1831 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1832                 const struct sw_flow_mask *b)
1833 {
1834         u8 *a_ = (u8 *)&a->key + a->range.start;
1835         u8 *b_ = (u8 *)&b->key + b->range.start;
1836
1837         return  (a->range.end == b->range.end)
1838                 && (a->range.start == b->range.start)
1839                 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1840 }
1841
1842 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1843                                            const struct sw_flow_mask *mask)
1844 {
1845         struct list_head *ml;
1846
1847         list_for_each(ml, &tbl->mask_list) {
1848                 struct sw_flow_mask *m;
1849                 m = container_of(ml, struct sw_flow_mask, list);
1850                 if (ovs_sw_flow_mask_equal(mask, m))
1851                         return m;
1852         }
1853
1854         return NULL;
1855 }
1856
1857 /**
1858  * add a new mask into the mask list.
1859  * The caller needs to make sure that 'mask' is not the same
1860  * as any masks that are already on the list.
1861  */
1862 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1863 {
1864         list_add_rcu(&mask->list, &tbl->mask_list);
1865 }
1866
1867 /**
1868  * Set 'range' fields in the mask to the value of 'val'.
1869  */
1870 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1871                 struct sw_flow_key_range *range, u8 val)
1872 {
1873         u8 *m = (u8 *)&mask->key + range->start;
1874
1875         mask->range = *range;
1876         memset(m, val, ovs_sw_flow_mask_size_roundup(mask));
1877 }