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->phy.in_port == DP_MAX_PORTS &&
 137             match->mask && (match->mask->key.phy.in_port == 0xffff))
 138                 mask_allowed |= (1ULL << OVS_KEY_ATTR_IN_PORT);
 139
 140         if (match->key->eth.type == htons(ETH_P_802_2) &&
 141             match->mask && (match->mask->key.eth.type == htons(0xffff)))
 142                 mask_allowed |= (1ULL << OVS_KEY_ATTR_ETHERTYPE);
 143
 144         /* Check key attributes. */
 145         if (match->key->eth.type == htons(ETH_P_ARP)
 146                         || match->key->eth.type == htons(ETH_P_RARP)) {
 147                 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
 148                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 149                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 150         }
 151
 152         if (match->key->eth.type == htons(ETH_P_IP)) {
 153                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 154                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 155                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
 156
 157                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 158                         if (match->key->ip.proto == IPPROTO_UDP) {
 159                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 160                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 161                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 162                         }
 163
 164                         if (match->key->ip.proto == IPPROTO_TCP) {
 165                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 166                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 167                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 168                         }
 169
 170                         if (match->key->ip.proto == IPPROTO_ICMP) {
 171                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 172                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 173                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 174                         }
 175                 }
 176         }
 177
 178         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 179                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 180                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 181                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
 182
 183                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 184                         if (match->key->ip.proto == IPPROTO_UDP) {
 185                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 186                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 187                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 188                         }
 189
 190                         if (match->key->ip.proto == IPPROTO_TCP) {
 191                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 192                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 193                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 194                         }
 195
 196                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 197                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 198                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 199                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 200
 201                                 if (match->key->ipv6.tp.src ==
 202                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 203                                     match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 204                                         key_expected |= 1ULL << OVS_KEY_ATTR_ND;
 205                                         if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
 206                                                 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
 207                                 }
 208                         }
 209                 }
 210         }
 211
 212         if ((key_attrs & key_expected) != key_expected) {
 213                 /* Key attributes check failed. */
 214                 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
 215                                 key_attrs, key_expected);
 216                 return false;
 217         }
 218
 219         if ((mask_attrs & mask_allowed) != mask_attrs) {
 220                 /* Mask attributes check failed. */
 221                 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
 222                                 mask_attrs, mask_allowed);
 223                 return false;
 224         }
 225
 226         return true;
 227 }
 228
 229 static int check_header(struct sk_buff *skb, int len)
 230 {
 231         if (unlikely(skb->len < len))
 232                 return -EINVAL;
 233         if (unlikely(!pskb_may_pull(skb, len)))
 234                 return -ENOMEM;
 235         return 0;
 236 }
 237
 238 static bool arphdr_ok(struct sk_buff *skb)
 239 {
 240         return pskb_may_pull(skb, skb_network_offset(skb) +
 241                                   sizeof(struct arp_eth_header));
 242 }
 243
 244 static int check_iphdr(struct sk_buff *skb)
 245 {
 246         unsigned int nh_ofs = skb_network_offset(skb);
 247         unsigned int ip_len;
 248         int err;
 249
 250         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 251         if (unlikely(err))
 252                 return err;
 253
 254         ip_len = ip_hdrlen(skb);
 255         if (unlikely(ip_len < sizeof(struct iphdr) ||
 256                      skb->len < nh_ofs + ip_len))
 257                 return -EINVAL;
 258
 259         skb_set_transport_header(skb, nh_ofs + ip_len);
 260         return 0;
 261 }
 262
 263 static bool tcphdr_ok(struct sk_buff *skb)
 264 {
 265         int th_ofs = skb_transport_offset(skb);
 266         int tcp_len;
 267
 268         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 269                 return false;
 270
 271         tcp_len = tcp_hdrlen(skb);
 272         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 273                      skb->len < th_ofs + tcp_len))
 274                 return false;
 275
 276         return true;
 277 }
 278
 279 static bool udphdr_ok(struct sk_buff *skb)
 280 {
 281         return pskb_may_pull(skb, skb_transport_offset(skb) +
 282                                   sizeof(struct udphdr));
 283 }
 284
 285 static bool icmphdr_ok(struct sk_buff *skb)
 286 {
 287         return pskb_may_pull(skb, skb_transport_offset(skb) +
 288                                   sizeof(struct icmphdr));
 289 }
 290
 291 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 292 {
 293         struct timespec cur_ts;
 294         u64 cur_ms, idle_ms;
 295
 296         ktime_get_ts(&cur_ts);
 297         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 298         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 299                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 300
 301         return cur_ms - idle_ms;
 302 }
 303
 304 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 305 {
 306         unsigned int nh_ofs = skb_network_offset(skb);
 307         unsigned int nh_len;
 308         int payload_ofs;
 309         struct ipv6hdr *nh;
 310         uint8_t nexthdr;
 311         __be16 frag_off;
 312         int err;
 313
 314         err = check_header(skb, nh_ofs + sizeof(*nh));
 315         if (unlikely(err))
 316                 return err;
 317
 318         nh = ipv6_hdr(skb);
 319         nexthdr = nh->nexthdr;
 320         payload_ofs = (u8 *)(nh + 1) - skb->data;
 321
 322         key->ip.proto = NEXTHDR_NONE;
 323         key->ip.tos = ipv6_get_dsfield(nh);
 324         key->ip.ttl = nh->hop_limit;
 325         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 326         key->ipv6.addr.src = nh->saddr;
 327         key->ipv6.addr.dst = nh->daddr;
 328
 329         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 330         if (unlikely(payload_ofs < 0))
 331                 return -EINVAL;
 332
 333         if (frag_off) {
 334                 if (frag_off & htons(~0x7))
 335                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 336                 else
 337                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 338         }
 339
 340         nh_len = payload_ofs - nh_ofs;
 341         skb_set_transport_header(skb, nh_ofs + nh_len);
 342         key->ip.proto = nexthdr;
 343         return nh_len;
 344 }
 345
 346 static bool icmp6hdr_ok(struct sk_buff *skb)
 347 {
 348         return pskb_may_pull(skb, skb_transport_offset(skb) +
 349                                   sizeof(struct icmp6hdr));
 350 }
 351
 352 static void flow_key_mask(struct sw_flow_key *dst,
 353                           const struct sw_flow_key *src,
 354                           const struct sw_flow_mask *mask)
 355 {
 356         u8 *m = (u8 *)&mask->key + mask->range.start;
 357         u8 *s = (u8 *)src + mask->range.start;
 358         u8 *d = (u8 *)dst + mask->range.start;
 359         int i;
 360
 361         memset(dst, 0, sizeof(*dst));
 362         for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
 363                 *d = *s & *m;
 364                 d++, s++, m++;
 365         }
 366 }
 367
 368 #define TCP_FLAGS_OFFSET 13
 369 #define TCP_FLAG_MASK 0x3f
 370
 371 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 372 {
 373         u8 tcp_flags = 0;
 374
 375         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 376              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 377             flow->key.ip.proto == IPPROTO_TCP &&
 378             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 379                 u8 *tcp = (u8 *)tcp_hdr(skb);
 380                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 381         }
 382
 383         spin_lock(&flow->lock);
 384         flow->used = jiffies;
 385         flow->packet_count++;
 386         flow->byte_count += skb->len;
 387         flow->tcp_flags |= tcp_flags;
 388         spin_unlock(&flow->lock);
 389 }
 390
 391 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
 392 {
 393         struct sw_flow_actions *sfa;
 394
 395         if (size > MAX_ACTIONS_BUFSIZE)
 396                 return ERR_PTR(-EINVAL);
 397
 398         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 399         if (!sfa)
 400                 return ERR_PTR(-ENOMEM);
 401
 402         sfa->actions_len = 0;
 403         return sfa;
 404 }
 405
 406 struct sw_flow *ovs_flow_alloc(void)
 407 {
 408         struct sw_flow *flow;
 409
 410         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 411         if (!flow)
 412                 return ERR_PTR(-ENOMEM);
 413
 414         spin_lock_init(&flow->lock);
 415         flow->sf_acts = NULL;
 416         flow->mask = NULL;
 417
 418         return flow;
 419 }
 420
 421 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 422 {
 423         hash = jhash_1word(hash, table->hash_seed);
 424         return flex_array_get(table->buckets,
 425                                 (hash & (table->n_buckets - 1)));
 426 }
 427
 428 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 429 {
 430         struct flex_array *buckets;
 431         int i, err;
 432
 433         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 434                                    n_buckets, GFP_KERNEL);
 435         if (!buckets)
 436                 return NULL;
 437
 438         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 439         if (err) {
 440                 flex_array_free(buckets);
 441                 return NULL;
 442         }
 443
 444         for (i = 0; i < n_buckets; i++)
 445                 INIT_HLIST_HEAD((struct hlist_head *)
 446                                         flex_array_get(buckets, i));
 447
 448         return buckets;
 449 }
 450
 451 static void free_buckets(struct flex_array *buckets)
 452 {
 453         flex_array_free(buckets);
 454 }
 455
 456 static struct flow_table *__flow_tbl_alloc(int new_size)
 457 {
 458         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 459
 460         if (!table)
 461                 return NULL;
 462
 463         table->buckets = alloc_buckets(new_size);
 464
 465         if (!table->buckets) {
 466                 kfree(table);
 467                 return NULL;
 468         }
 469         table->n_buckets = new_size;
 470         table->count = 0;
 471         table->node_ver = 0;
 472         table->keep_flows = false;
 473         get_random_bytes(&table->hash_seed, sizeof(u32));
 474         table->mask_list = NULL;
 475
 476         return table;
 477 }
 478
 479 static void __flow_tbl_destroy(struct flow_table *table)
 480 {
 481         int i;
 482
 483         if (table->keep_flows)
 484                 goto skip_flows;
 485
 486         for (i = 0; i < table->n_buckets; i++) {
 487                 struct sw_flow *flow;
 488                 struct hlist_head *head = flex_array_get(table->buckets, i);
 489                 struct hlist_node *n;
 490                 int ver = table->node_ver;
 491
 492                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 493                         hlist_del_rcu(&flow->hash_node[ver]);
 494                         ovs_flow_free(flow, false);
 495                 }
 496         }
 497
 498         BUG_ON(!list_empty(table->mask_list));
 499         kfree(table->mask_list);
 500
 501 skip_flows:
 502         free_buckets(table->buckets);
 503         kfree(table);
 504 }
 505
 506 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 507 {
 508         struct flow_table *table = __flow_tbl_alloc(new_size);
 509
 510         if (!table)
 511                 return NULL;
 512
 513         table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
 514         if (!table->mask_list) {
 515                 table->keep_flows = true;
 516                 __flow_tbl_destroy(table);
 517                 return NULL;
 518         }
 519         INIT_LIST_HEAD(table->mask_list);
 520
 521         return table;
 522 }
 523
 524 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 525 {
 526         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 527
 528         __flow_tbl_destroy(table);
 529 }
 530
 531 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
 532 {
 533         if (!table)
 534                 return;
 535
 536         if (deferred)
 537                 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 538         else
 539                 __flow_tbl_destroy(table);
 540 }
 541
 542 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
 543 {
 544         struct sw_flow *flow;
 545         struct hlist_head *head;
 546         int ver;
 547         int i;
 548
 549         ver = table->node_ver;
 550         while (*bucket < table->n_buckets) {
 551                 i = 0;
 552                 head = flex_array_get(table->buckets, *bucket);
 553                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 554                         if (i < *last) {
 555                                 i++;
 556                                 continue;
 557                         }
 558                         *last = i + 1;
 559                         return flow;
 560                 }
 561                 (*bucket)++;
 562                 *last = 0;
 563         }
 564
 565         return NULL;
 566 }
 567
 568 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
 569 {
 570         struct hlist_head *head;
 571
 572         head = find_bucket(table, flow->hash);
 573         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 574
 575         table->count++;
 576 }
 577
 578 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 579 {
 580         int old_ver;
 581         int i;
 582
 583         old_ver = old->node_ver;
 584         new->node_ver = !old_ver;
 585
 586         /* Insert in new table. */
 587         for (i = 0; i < old->n_buckets; i++) {
 588                 struct sw_flow *flow;
 589                 struct hlist_head *head;
 590
 591                 head = flex_array_get(old->buckets, i);
 592
 593                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 594                         __tbl_insert(new, flow);
 595         }
 596
 597         new->mask_list = old->mask_list;
 598         old->keep_flows = true;
 599 }
 600
 601 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 602 {
 603         struct flow_table *new_table;
 604
 605         new_table = __flow_tbl_alloc(n_buckets);
 606         if (!new_table)
 607                 return ERR_PTR(-ENOMEM);
 608
 609         flow_table_copy_flows(table, new_table);
 610
 611         return new_table;
 612 }
 613
 614 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 615 {
 616         return __flow_tbl_rehash(table, table->n_buckets);
 617 }
 618
 619 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 620 {
 621         return __flow_tbl_rehash(table, table->n_buckets * 2);
 622 }
 623
 624 static void __flow_free(struct sw_flow *flow)
 625 {
 626         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 627         kmem_cache_free(flow_cache, flow);
 628 }
 629
 630 static void rcu_free_flow_callback(struct rcu_head *rcu)
 631 {
 632         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 633
 634         __flow_free(flow);
 635 }
 636
 637 void ovs_flow_free(struct sw_flow *flow, bool deferred)
 638 {
 639         if (!flow)
 640                 return;
 641
 642         ovs_sw_flow_mask_del_ref((struct sw_flow_mask __force *)flow->mask,
 643                                  deferred);
 644
 645         if (deferred)
 646                 call_rcu(&flow->rcu, rcu_free_flow_callback);
 647         else
 648                 __flow_free(flow);
 649 }
 650
 651 /* RCU callback used by ovs_flow_deferred_free_acts. */
 652 static void rcu_free_acts_callback(struct rcu_head *rcu)
 653 {
 654         struct sw_flow_actions *sf_acts = container_of(rcu,
 655                         struct sw_flow_actions, rcu);
 656         kfree(sf_acts);
 657 }
 658
 659 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 660  * The caller must hold rcu_read_lock for this to be sensible. */
 661 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 662 {
 663         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 664 }
 665
 666 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 667 {
 668         struct qtag_prefix {
 669                 __be16 eth_type; /* ETH_P_8021Q */
 670                 __be16 tci;
 671         };
 672         struct qtag_prefix *qp;
 673
 674         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 675                 return 0;
 676
 677         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 678                                          sizeof(__be16))))
 679                 return -ENOMEM;
 680
 681         qp = (struct qtag_prefix *) skb->data;
 682         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 683         __skb_pull(skb, sizeof(struct qtag_prefix));
 684
 685         return 0;
 686 }
 687
 688 static __be16 parse_ethertype(struct sk_buff *skb)
 689 {
 690         struct llc_snap_hdr {
 691                 u8  dsap;  /* Always 0xAA */
 692                 u8  ssap;  /* Always 0xAA */
 693                 u8  ctrl;
 694                 u8  oui[3];
 695                 __be16 ethertype;
 696         };
 697         struct llc_snap_hdr *llc;
 698         __be16 proto;
 699
 700         proto = *(__be16 *) skb->data;
 701         __skb_pull(skb, sizeof(__be16));
 702
 703         if (ntohs(proto) >= ETH_P_802_3_MIN)
 704                 return proto;
 705
 706         if (skb->len < sizeof(struct llc_snap_hdr))
 707                 return htons(ETH_P_802_2);
 708
 709         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 710                 return htons(0);
 711
 712         llc = (struct llc_snap_hdr *) skb->data;
 713         if (llc->dsap != LLC_SAP_SNAP ||
 714             llc->ssap != LLC_SAP_SNAP ||
 715             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 716                 return htons(ETH_P_802_2);
 717
 718         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 719
 720         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 721                 return llc->ethertype;
 722
 723         return htons(ETH_P_802_2);
 724 }
 725
 726 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 727                         int nh_len)
 728 {
 729         struct icmp6hdr *icmp = icmp6_hdr(skb);
 730
 731         /* The ICMPv6 type and code fields use the 16-bit transport port
 732          * fields, so we need to store them in 16-bit network byte order.
 733          */
 734         key->ipv6.tp.src = htons(icmp->icmp6_type);
 735         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 736
 737         if (icmp->icmp6_code == 0 &&
 738             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 739              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 740                 int icmp_len = skb->len - skb_transport_offset(skb);
 741                 struct nd_msg *nd;
 742                 int offset;
 743
 744                 /* In order to process neighbor discovery options, we need the
 745                  * entire packet.
 746                  */
 747                 if (unlikely(icmp_len < sizeof(*nd)))
 748                         return 0;
 749
 750                 if (unlikely(skb_linearize(skb)))
 751                         return -ENOMEM;
 752
 753                 nd = (struct nd_msg *)skb_transport_header(skb);
 754                 key->ipv6.nd.target = nd->target;
 755
 756                 icmp_len -= sizeof(*nd);
 757                 offset = 0;
 758                 while (icmp_len >= 8) {
 759                         struct nd_opt_hdr *nd_opt =
 760                                  (struct nd_opt_hdr *)(nd->opt + offset);
 761                         int opt_len = nd_opt->nd_opt_len * 8;
 762
 763                         if (unlikely(!opt_len || opt_len > icmp_len))
 764                                 return 0;
 765
 766                         /* Store the link layer address if the appropriate
 767                          * option is provided.  It is considered an error if
 768                          * the same link layer option is specified twice.
 769                          */
 770                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 771                             && opt_len == 8) {
 772                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 773                                         goto invalid;
 774                                 memcpy(key->ipv6.nd.sll,
 775                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 776                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 777                                    && opt_len == 8) {
 778                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 779                                         goto invalid;
 780                                 memcpy(key->ipv6.nd.tll,
 781                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 782                         }
 783
 784                         icmp_len -= opt_len;
 785                         offset += opt_len;
 786                 }
 787         }
 788
 789         return 0;
 790
 791 invalid:
 792         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 793         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 794         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 795
 796         return 0;
 797 }
 798
 799 /**
 800  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 801  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 802  * Ethernet header
 803  * @in_port: port number on which @skb was received.
 804  * @key: output flow key
 805  * @key_lenp: length of output flow key
 806  *
 807  * The caller must ensure that skb->len >= ETH_HLEN.
 808  *
 809  * Returns 0 if successful, otherwise a negative errno value.
 810  *
 811  * Initializes @skb header pointers as follows:
 812  *
 813  *    - skb->mac_header: the Ethernet header.
 814  *
 815  *    - skb->network_header: just past the Ethernet header, or just past the
 816  *      VLAN header, to the first byte of the Ethernet payload.
 817  *
 818  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 819  *      on output, then just past the IP header, if one is present and
 820  *      of a correct length, otherwise the same as skb->network_header.
 821  *      For other key->eth.type values it is left untouched.
 822  */
 823 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
 824 {
 825         int error;
 826         struct ethhdr *eth;
 827
 828         memset(key, 0, sizeof(*key));
 829
 830         key->phy.priority = skb->priority;
 831         if (OVS_CB(skb)->tun_key)
 832                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 833         key->phy.in_port = in_port;
 834         key->phy.skb_mark = skb_get_mark(skb);
 835
 836         skb_reset_mac_header(skb);
 837
 838         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 839          * header in the linear data area.
 840          */
 841         eth = eth_hdr(skb);
 842         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 843         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 844
 845         __skb_pull(skb, 2 * ETH_ALEN);
 846         /* We are going to push all headers that we pull, so no need to
 847          * update skb->csum here. */
 848
 849         if (vlan_tx_tag_present(skb))
 850                 key->eth.tci = htons(vlan_get_tci(skb));
 851         else if (eth->h_proto == htons(ETH_P_8021Q))
 852                 if (unlikely(parse_vlan(skb, key)))
 853                         return -ENOMEM;
 854
 855         key->eth.type = parse_ethertype(skb);
 856         if (unlikely(key->eth.type == htons(0)))
 857                 return -ENOMEM;
 858
 859         skb_reset_network_header(skb);
 860         __skb_push(skb, skb->data - skb_mac_header(skb));
 861
 862         /* Network layer. */
 863         if (key->eth.type == htons(ETH_P_IP)) {
 864                 struct iphdr *nh;
 865                 __be16 offset;
 866
 867                 error = check_iphdr(skb);
 868                 if (unlikely(error)) {
 869                         if (error == -EINVAL) {
 870                                 skb->transport_header = skb->network_header;
 871                                 error = 0;
 872                         }
 873                         return error;
 874                 }
 875
 876                 nh = ip_hdr(skb);
 877                 key->ipv4.addr.src = nh->saddr;
 878                 key->ipv4.addr.dst = nh->daddr;
 879
 880                 key->ip.proto = nh->protocol;
 881                 key->ip.tos = nh->tos;
 882                 key->ip.ttl = nh->ttl;
 883
 884                 offset = nh->frag_off & htons(IP_OFFSET);
 885                 if (offset) {
 886                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 887                         return 0;
 888                 }
 889                 if (nh->frag_off & htons(IP_MF) ||
 890                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 891                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 892
 893                 /* Transport layer. */
 894                 if (key->ip.proto == IPPROTO_TCP) {
 895                         if (tcphdr_ok(skb)) {
 896                                 struct tcphdr *tcp = tcp_hdr(skb);
 897                                 key->ipv4.tp.src = tcp->source;
 898                                 key->ipv4.tp.dst = tcp->dest;
 899                         }
 900                 } else if (key->ip.proto == IPPROTO_UDP) {
 901                         if (udphdr_ok(skb)) {
 902                                 struct udphdr *udp = udp_hdr(skb);
 903                                 key->ipv4.tp.src = udp->source;
 904                                 key->ipv4.tp.dst = udp->dest;
 905                         }
 906                 } else if (key->ip.proto == IPPROTO_ICMP) {
 907                         if (icmphdr_ok(skb)) {
 908                                 struct icmphdr *icmp = icmp_hdr(skb);
 909                                 /* The ICMP type and code fields use the 16-bit
 910                                  * transport port fields, so we need to store
 911                                  * them in 16-bit network byte order. */
 912                                 key->ipv4.tp.src = htons(icmp->type);
 913                                 key->ipv4.tp.dst = htons(icmp->code);
 914                         }
 915                 }
 916
 917         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 918                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 919                 struct arp_eth_header *arp;
 920
 921                 arp = (struct arp_eth_header *)skb_network_header(skb);
 922
 923                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 924                                 && arp->ar_pro == htons(ETH_P_IP)
 925                                 && arp->ar_hln == ETH_ALEN
 926                                 && arp->ar_pln == 4) {
 927
 928                         /* We only match on the lower 8 bits of the opcode. */
 929                         if (ntohs(arp->ar_op) <= 0xff)
 930                                 key->ip.proto = ntohs(arp->ar_op);
 931                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 932                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 933                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 934                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 935                 }
 936         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 937                 int nh_len;             /* IPv6 Header + Extensions */
 938
 939                 nh_len = parse_ipv6hdr(skb, key);
 940                 if (unlikely(nh_len < 0)) {
 941                         if (nh_len == -EINVAL) {
 942                                 skb->transport_header = skb->network_header;
 943                                 error = 0;
 944                         } else {
 945                                 error = nh_len;
 946                         }
 947                         return error;
 948                 }
 949
 950                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 951                         return 0;
 952                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 953                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 954
 955                 /* Transport layer. */
 956                 if (key->ip.proto == NEXTHDR_TCP) {
 957                         if (tcphdr_ok(skb)) {
 958                                 struct tcphdr *tcp = tcp_hdr(skb);
 959                                 key->ipv6.tp.src = tcp->source;
 960                                 key->ipv6.tp.dst = tcp->dest;
 961                         }
 962                 } else if (key->ip.proto == NEXTHDR_UDP) {
 963                         if (udphdr_ok(skb)) {
 964                                 struct udphdr *udp = udp_hdr(skb);
 965                                 key->ipv6.tp.src = udp->source;
 966                                 key->ipv6.tp.dst = udp->dest;
 967                         }
 968                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 969                         if (icmp6hdr_ok(skb)) {
 970                                 error = parse_icmpv6(skb, key, nh_len);
 971                                 if (error)
 972                                         return error;
 973                         }
 974                 }
 975         }
 976
 977         return 0;
 978 }
 979
 980 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 981 {
 982         return jhash2((u32 *)((u8 *)key + key_start),
 983                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 984 }
 985
 986 static int flow_key_start(const struct sw_flow_key *key)
 987 {
 988         if (key->tun_key.ipv4_dst)
 989                 return 0;
 990         else
 991                 return offsetof(struct sw_flow_key, phy);
 992 }
 993
 994 static bool __cmp_key(const struct sw_flow_key *key1,
 995                 const struct sw_flow_key *key2,  int key_start, int key_len)
 996 {
 997         return !memcmp((u8 *)key1 + key_start,
 998                         (u8 *)key2 + key_start, (key_len - key_start));
 999 }
1000
1001 static bool __flow_cmp_key(const struct sw_flow *flow,
1002                 const struct sw_flow_key *key, int key_start, int key_len)
1003 {
1004         return __cmp_key(&flow->key, key, key_start, key_len);
1005 }
1006
1007 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1008                   const struct sw_flow_key *key, int key_start, int key_len)
1009 {
1010         return __cmp_key(&flow->unmasked_key, key, key_start, key_len);
1011 }
1012
1013 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1014                 const struct sw_flow_key *key, int key_len)
1015 {
1016         int key_start;
1017         key_start = flow_key_start(key);
1018
1019         return __flow_cmp_unmasked_key(flow, key, key_start, key_len);
1020
1021 }
1022
1023 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1024                                        struct sw_flow_match *match)
1025 {
1026         struct sw_flow_key *unmasked = match->key;
1027         int key_len = match->range.end;
1028         struct sw_flow *flow;
1029
1030         flow = ovs_flow_lookup(table, unmasked);
1031         if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_len)))
1032                 flow = NULL;
1033
1034         return flow;
1035 }
1036
1037 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1038                                     const struct sw_flow_key *flow_key,
1039                                     struct sw_flow_mask *mask)
1040 {
1041         struct sw_flow *flow;
1042         struct hlist_head *head;
1043         int key_start = mask->range.start;
1044         int key_len = mask->range.end;
1045         u32 hash;
1046         struct sw_flow_key masked_key;
1047
1048         flow_key_mask(&masked_key, flow_key, mask);
1049         hash = ovs_flow_hash(&masked_key, key_start, key_len);
1050         head = find_bucket(table, hash);
1051         hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1052                 if (__flow_cmp_key(flow, &masked_key, key_start, key_len))
1053                         return flow;
1054         }
1055         return NULL;
1056 }
1057
1058 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1059                                 const struct sw_flow_key *key)
1060 {
1061         struct sw_flow *flow = NULL;
1062         struct sw_flow_mask *mask;
1063
1064         list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1065                 flow = ovs_masked_flow_lookup(tbl, key, mask);
1066                 if (flow)  /* Found */
1067                         break;
1068         }
1069
1070         return flow;
1071 }
1072
1073
1074 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow,
1075                          const struct sw_flow_key *key, int key_len)
1076 {
1077         flow->unmasked_key = *key;
1078         flow_key_mask(&flow->key, &flow->unmasked_key, ovsl_dereference(flow->mask));
1079         flow->hash = ovs_flow_hash(&flow->key,
1080                         ovsl_dereference(flow->mask)->range.start,
1081                         ovsl_dereference(flow->mask)->range.end);
1082         __tbl_insert(table, flow);
1083 }
1084
1085 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1086 {
1087         BUG_ON(table->count == 0);
1088         hlist_del_rcu(&flow->hash_node[table->node_ver]);
1089         table->count--;
1090 }
1091
1092 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
1093 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1094         [OVS_KEY_ATTR_ENCAP] = -1,
1095         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1096         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1097         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1098         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1099         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1100         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1101         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1102         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1103         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1104         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1105         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1106         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1107         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1108         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1109         [OVS_KEY_ATTR_TUNNEL] = -1,
1110 };
1111
1112 static bool is_all_zero(const u8 *fp, size_t size)
1113 {
1114         int i;
1115
1116         if (!fp)
1117                 return false;
1118
1119         for (i = 0; i < size; i++)
1120                 if (fp[i])
1121                         return false;
1122
1123         return true;
1124 }
1125
1126 static int __parse_flow_nlattrs(const struct nlattr *attr,
1127                               const struct nlattr *a[],
1128                               u64 *attrsp, bool nz)
1129 {
1130         const struct nlattr *nla;
1131         u64 attrs;
1132         int rem;
1133
1134         attrs = *attrsp;
1135         nla_for_each_nested(nla, attr, rem) {
1136                 u16 type = nla_type(nla);
1137                 int expected_len;
1138
1139                 if (type > OVS_KEY_ATTR_MAX) {
1140                         OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1141                                   type, OVS_KEY_ATTR_MAX);
1142                 }
1143
1144                 if (attrs & (1ULL << type)) {
1145                         OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1146                         return -EINVAL;
1147                 }
1148
1149                 expected_len = ovs_key_lens[type];
1150                 if (nla_len(nla) != expected_len && expected_len != -1) {
1151                         OVS_NLERR("Key attribute has unexpected length (type=%d"
1152                                   ", length=%d, expected=%d).\n", type,
1153                                   nla_len(nla), expected_len);
1154                         return -EINVAL;
1155                 }
1156
1157                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1158                         attrs |= 1ULL << type;
1159                         a[type] = nla;
1160                 }
1161         }
1162         if (rem) {
1163                 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1164                 return -EINVAL;
1165         }
1166
1167         *attrsp = attrs;
1168         return 0;
1169 }
1170
1171 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1172                               const struct nlattr *a[], u64 *attrsp)
1173 {
1174         return __parse_flow_nlattrs(attr, a, attrsp, true);
1175 }
1176
1177 static int parse_flow_nlattrs(const struct nlattr *attr,
1178                               const struct nlattr *a[], u64 *attrsp)
1179 {
1180         return __parse_flow_nlattrs(attr, a, attrsp, false);
1181 }
1182
1183 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1184                          struct sw_flow_match *match, bool is_mask)
1185 {
1186         struct nlattr *a;
1187         int rem;
1188         bool ttl = false;
1189         __be16 tun_flags = 0;
1190
1191         nla_for_each_nested(a, attr, rem) {
1192                 int type = nla_type(a);
1193                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1194                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1195                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1196                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1197                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1198                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1199                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1200                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1201                 };
1202
1203                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1204                         OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1205                         type, OVS_TUNNEL_KEY_ATTR_MAX);
1206                         return -EINVAL;
1207                 }
1208
1209                 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1210                         OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1211                                   " legnth (type=%d, length=%d, expected=%d).\n",
1212                                   type, nla_len(a), ovs_tunnel_key_lens[type]);
1213                         return -EINVAL;
1214                 }
1215
1216                 switch (type) {
1217                 case OVS_TUNNEL_KEY_ATTR_ID:
1218                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1219                                         nla_get_be64(a), is_mask);
1220                         tun_flags |= TUNNEL_KEY;
1221                         break;
1222                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1223                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1224                                         nla_get_be32(a), is_mask);
1225                         break;
1226                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1227                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1228                                         nla_get_be32(a), is_mask);
1229                         break;
1230                 case OVS_TUNNEL_KEY_ATTR_TOS:
1231                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1232                                         nla_get_u8(a), is_mask);
1233                         break;
1234                 case OVS_TUNNEL_KEY_ATTR_TTL:
1235                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1236                                         nla_get_u8(a), is_mask);
1237                         ttl = true;
1238                         break;
1239                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1240                         tun_flags |= TUNNEL_DONT_FRAGMENT;
1241                         break;
1242                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1243                         tun_flags |= TUNNEL_CSUM;
1244                         break;
1245                 default:
1246                         return -EINVAL;
1247                 }
1248         }
1249
1250         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1251
1252         if (rem > 0) {
1253                 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1254                 return -EINVAL;
1255         }
1256
1257         if (!match->key->tun_key.ipv4_dst) {
1258                 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1259                 return -EINVAL;
1260         }
1261
1262         if (!ttl) {
1263                 OVS_NLERR("IPv4 tunnel TTL is zero.\n");
1264                 return -EINVAL;
1265         }
1266
1267         return 0;
1268 }
1269
1270 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1271                         const struct ovs_key_ipv4_tunnel *tun_key,
1272                         const struct ovs_key_ipv4_tunnel *output)
1273 {
1274         struct nlattr *nla;
1275
1276         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1277         if (!nla)
1278                 return -EMSGSIZE;
1279
1280         if (tun_key->tun_flags & TUNNEL_KEY &&
1281             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1282                 return -EMSGSIZE;
1283         if (tun_key->ipv4_src &&
1284             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1285                 return -EMSGSIZE;
1286         if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1287                 return -EMSGSIZE;
1288         if (tun_key->ipv4_tos &&
1289             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1290                 return -EMSGSIZE;
1291         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1292                 return -EMSGSIZE;
1293         if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1294                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1295                 return -EMSGSIZE;
1296         if ((tun_key->tun_flags & TUNNEL_CSUM) &&
1297                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1298                 return -EMSGSIZE;
1299
1300         nla_nest_end(skb, nla);
1301         return 0;
1302 }
1303
1304
1305 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
1306                 const struct nlattr **a, bool is_mask)
1307 {
1308         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1309                 SW_FLOW_KEY_PUT(match, phy.priority,
1310                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1311                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1312         }
1313
1314         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1315                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1316
1317                 if (!is_mask && in_port >= DP_MAX_PORTS)
1318                         return -EINVAL;
1319                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1320                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1321         } else if (!is_mask) {
1322                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1323         }
1324
1325         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1326                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1327 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1328                 if (!is_mask && mark != 0) {
1329                         OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1330                         return -EINVAL;
1331                 }
1332 #endif
1333                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1334                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1335         }
1336         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1337                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1338                                         is_mask))
1339                         return -EINVAL;
1340                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1341         }
1342         return 0;
1343 }
1344
1345 static int ovs_key_from_nlattrs(struct sw_flow_match *match,  u64 attrs,
1346                 const struct nlattr **a, bool is_mask)
1347 {
1348         int err;
1349         u64 orig_attrs = attrs;
1350
1351         err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1352         if (err)
1353                 return err;
1354
1355         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1356                 const struct ovs_key_ethernet *eth_key;
1357
1358                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1359                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1360                                 eth_key->eth_src, ETH_ALEN, is_mask);
1361                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1362                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1363                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1364         }
1365
1366         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1367                 __be16 tci;
1368
1369                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1370                 if (!is_mask)
1371                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1372                                 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1373                                 return -EINVAL;
1374                         }
1375
1376                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1377                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1378         }
1379
1380         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1381                 __be16 eth_type;
1382
1383                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1384                 if (!is_mask && ntohs(eth_type) < ETH_P_802_3_MIN) {
1385                         OVS_NLERR("EtherType is less than mimimum (type=%x, min=%x).\n",
1386                                         ntohs(eth_type), ETH_P_802_3_MIN);
1387                         return -EINVAL;
1388                 }
1389
1390                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1391                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1392         } else if (!is_mask) {
1393                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1394         }
1395
1396         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1397                 const struct ovs_key_ipv4 *ipv4_key;
1398
1399                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1400                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1401                         OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1402                                 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1403                         return -EINVAL;
1404                 }
1405                 SW_FLOW_KEY_PUT(match, ip.proto,
1406                                 ipv4_key->ipv4_proto, is_mask);
1407                 SW_FLOW_KEY_PUT(match, ip.tos,
1408                                 ipv4_key->ipv4_tos, is_mask);
1409                 SW_FLOW_KEY_PUT(match, ip.ttl,
1410                                 ipv4_key->ipv4_ttl, is_mask);
1411                 SW_FLOW_KEY_PUT(match, ip.frag,
1412                                 ipv4_key->ipv4_frag, is_mask);
1413                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1414                                 ipv4_key->ipv4_src, is_mask);
1415                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1416                                 ipv4_key->ipv4_dst, is_mask);
1417                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1418         }
1419
1420         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1421                 const struct ovs_key_ipv6 *ipv6_key;
1422
1423                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1424                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1425                         OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1426                                 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1427                         return -EINVAL;
1428                 }
1429                 SW_FLOW_KEY_PUT(match, ipv6.label,
1430                                 ipv6_key->ipv6_label, is_mask);
1431                 SW_FLOW_KEY_PUT(match, ip.proto,
1432                                 ipv6_key->ipv6_proto, is_mask);
1433                 SW_FLOW_KEY_PUT(match, ip.tos,
1434                                 ipv6_key->ipv6_tclass, is_mask);
1435                 SW_FLOW_KEY_PUT(match, ip.ttl,
1436                                 ipv6_key->ipv6_hlimit, is_mask);
1437                 SW_FLOW_KEY_PUT(match, ip.frag,
1438                                 ipv6_key->ipv6_frag, is_mask);
1439                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1440                                 ipv6_key->ipv6_src,
1441                                 sizeof(match->key->ipv6.addr.src),
1442                                 is_mask);
1443                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1444                                 ipv6_key->ipv6_dst,
1445                                 sizeof(match->key->ipv6.addr.dst),
1446                                 is_mask);
1447
1448                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1449         }
1450
1451         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1452                 const struct ovs_key_arp *arp_key;
1453
1454                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1455                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1456                         OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1457                                   arp_key->arp_op);
1458                         return -EINVAL;
1459                 }
1460
1461                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1462                                 arp_key->arp_sip, is_mask);
1463                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1464                         arp_key->arp_tip, is_mask);
1465                 SW_FLOW_KEY_PUT(match, ip.proto,
1466                                 ntohs(arp_key->arp_op), is_mask);
1467                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1468                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1469                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1470                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1471
1472                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1473         }
1474
1475         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1476                 const struct ovs_key_tcp *tcp_key;
1477
1478                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1479                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1480                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1481                                         tcp_key->tcp_src, is_mask);
1482                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1483                                         tcp_key->tcp_dst, is_mask);
1484                 } else {
1485                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1486                                         tcp_key->tcp_src, is_mask);
1487                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1488                                         tcp_key->tcp_dst, is_mask);
1489                 }
1490                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1491         }
1492
1493         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1494                 const struct ovs_key_udp *udp_key;
1495
1496                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1497                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1498                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1499                                         udp_key->udp_src, is_mask);
1500                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1501                                         udp_key->udp_dst, is_mask);
1502                 } else {
1503                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1504                                         udp_key->udp_src, is_mask);
1505                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1506                                         udp_key->udp_dst, is_mask);
1507                 }
1508                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1509         }
1510
1511         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1512                 const struct ovs_key_icmp *icmp_key;
1513
1514                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1515                 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1516                                 htons(icmp_key->icmp_type), is_mask);
1517                 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1518                                 htons(icmp_key->icmp_code), is_mask);
1519                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1520         }
1521
1522         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1523                 const struct ovs_key_icmpv6 *icmpv6_key;
1524
1525                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1526                 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1527                                 htons(icmpv6_key->icmpv6_type), is_mask);
1528                 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1529                                 htons(icmpv6_key->icmpv6_code), is_mask);
1530                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1531         }
1532
1533         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1534                 const struct ovs_key_nd *nd_key;
1535
1536                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1537                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1538                         nd_key->nd_target,
1539                         sizeof(match->key->ipv6.nd.target),
1540                         is_mask);
1541                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1542                         nd_key->nd_sll, ETH_ALEN, is_mask);
1543                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1544                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1545                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1546         }
1547
1548         if (attrs != 0)
1549                 return -EINVAL;
1550
1551         return 0;
1552 }
1553
1554 /**
1555  * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1556  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1557  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1558  * does not include any don't care bit.
1559  * @match: receives the extracted flow match information.
1560  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1561  * sequence. The fields should of the packet that triggered the creation
1562  * of this flow.
1563  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1564  * attribute specifies the mask field of the wildcarded flow.
1565  */
1566 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1567                            const struct nlattr *key,
1568                            const struct nlattr *mask)
1569 {
1570         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1571         const struct nlattr *encap;
1572         u64 key_attrs = 0;
1573         u64 mask_attrs = 0;
1574         bool encap_valid = false;
1575         int err;
1576
1577         err = parse_flow_nlattrs(key, a, &key_attrs);
1578         if (err)
1579                 return err;
1580
1581         if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1582                 encap = a[OVS_KEY_ATTR_ENCAP];
1583                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1584                 if (nla_len(encap)) {
1585                         __be16 eth_type = 0; /* ETH_P_8021Q */
1586
1587                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1588                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1589
1590                         if  ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1591                                 encap_valid = true;
1592                                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1593                                 err = parse_flow_nlattrs(encap, a, &key_attrs);
1594                         } else {
1595                                 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1596                                 err = -EINVAL;
1597                         }
1598
1599                         if (err)
1600                                 return err;
1601                 }
1602         }
1603
1604         err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1605         if (err)
1606                 return err;
1607
1608         if (mask) {
1609                 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1610                 if (err)
1611                         return err;
1612
1613                 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1614                         __be16 eth_type = 0;
1615
1616                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1617                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1618                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1619                         if (eth_type == htons(0xffff)) {
1620                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1621                                 encap = a[OVS_KEY_ATTR_ENCAP];
1622                                 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1623                         } else {
1624                                 OVS_NLERR("VLAN frames must have an exact match"
1625                                          " on the TPID (mask=%x).\n",
1626                                          ntohs(eth_type));
1627                                 err = -EINVAL;
1628                         }
1629
1630                         if (err)
1631                                 return err;
1632                 }
1633
1634                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1635                 if (err)
1636                         return err;
1637         } else {
1638                 /* Populate exact match flow's key mask. */
1639                 if (match->mask)
1640                         ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1641         }
1642
1643         if (!ovs_match_validate(match, key_attrs, mask_attrs))
1644                 return -EINVAL;
1645
1646         return 0;
1647 }
1648
1649 /**
1650  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1651  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1652  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1653  * sequence.
1654  *
1655  * This parses a series of Netlink attributes that form a flow key, which must
1656  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1657  * get the metadata, that is, the parts of the flow key that cannot be
1658  * extracted from the packet itself.
1659  */
1660
1661 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1662                 const struct nlattr *attr)
1663 {
1664         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1665         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1666         u64 attrs = 0;
1667         int err;
1668         struct sw_flow_match match;
1669
1670         flow->key.phy.in_port = DP_MAX_PORTS;
1671         flow->key.phy.priority = 0;
1672         flow->key.phy.skb_mark = 0;
1673         memset(tun_key, 0, sizeof(flow->key.tun_key));
1674
1675         err = parse_flow_nlattrs(attr, a, &attrs);
1676         if (err)
1677                 return -EINVAL;
1678
1679         memset(&match, 0, sizeof(match));
1680         match.key = &flow->key;
1681
1682         err = metadata_from_nlattrs(&match, &attrs, a, false);
1683         if (err)
1684                 return err;
1685
1686         return 0;
1687 }
1688
1689 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1690                 const struct sw_flow_key *output, struct sk_buff *skb)
1691 {
1692         struct ovs_key_ethernet *eth_key;
1693         struct nlattr *nla, *encap;
1694
1695         if (swkey->phy.priority &&
1696             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1697                 goto nla_put_failure;
1698
1699         if (swkey->tun_key.ipv4_dst &&
1700             ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1701                 goto nla_put_failure;
1702
1703         if (swkey->phy.in_port != DP_MAX_PORTS) {
1704                 /* Exact match upper 16 bits. */
1705                 u16 upper_u16;
1706                 upper_u16 = (swkey == output) ? 0 : 0xffff;
1707
1708                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1709                                         (upper_u16 << 16) | output->phy.in_port))
1710                         goto nla_put_failure;
1711         }
1712
1713         if (swkey->phy.skb_mark &&
1714             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1715                 goto nla_put_failure;
1716
1717         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1718         if (!nla)
1719                 goto nla_put_failure;
1720
1721         eth_key = nla_data(nla);
1722         memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1723         memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1724
1725         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1726                 __be16 eth_type;
1727                 eth_type = (swkey == output) ? htons(ETH_P_8021Q) : htons(0xffff) ;
1728                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1729                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1730                         goto nla_put_failure;
1731                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1732                 if (!swkey->eth.tci)
1733                         goto unencap;
1734         } else
1735                 encap = NULL;
1736
1737         if ((swkey == output) && (swkey->eth.type == htons(ETH_P_802_2)))
1738                 goto unencap;
1739
1740         if (output->eth.type != 0)
1741                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1742                         goto nla_put_failure;
1743
1744         if (swkey->eth.type == htons(ETH_P_IP)) {
1745                 struct ovs_key_ipv4 *ipv4_key;
1746
1747                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1748                 if (!nla)
1749                         goto nla_put_failure;
1750                 ipv4_key = nla_data(nla);
1751                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1752                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1753                 ipv4_key->ipv4_proto = output->ip.proto;
1754                 ipv4_key->ipv4_tos = output->ip.tos;
1755                 ipv4_key->ipv4_ttl = output->ip.ttl;
1756                 ipv4_key->ipv4_frag = output->ip.frag;
1757         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1758                 struct ovs_key_ipv6 *ipv6_key;
1759
1760                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1761                 if (!nla)
1762                         goto nla_put_failure;
1763                 ipv6_key = nla_data(nla);
1764                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1765                                 sizeof(ipv6_key->ipv6_src));
1766                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1767                                 sizeof(ipv6_key->ipv6_dst));
1768                 ipv6_key->ipv6_label = output->ipv6.label;
1769                 ipv6_key->ipv6_proto = output->ip.proto;
1770                 ipv6_key->ipv6_tclass = output->ip.tos;
1771                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1772                 ipv6_key->ipv6_frag = output->ip.frag;
1773         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1774                    swkey->eth.type == htons(ETH_P_RARP)) {
1775                 struct ovs_key_arp *arp_key;
1776
1777                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1778                 if (!nla)
1779                         goto nla_put_failure;
1780                 arp_key = nla_data(nla);
1781                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1782                 arp_key->arp_sip = output->ipv4.addr.src;
1783                 arp_key->arp_tip = output->ipv4.addr.dst;
1784                 arp_key->arp_op = htons(output->ip.proto);
1785                 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1786                 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1787         }
1788
1789         if ((swkey->eth.type == htons(ETH_P_IP) ||
1790              swkey->eth.type == htons(ETH_P_IPV6)) &&
1791              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1792
1793                 if (swkey->ip.proto == IPPROTO_TCP) {
1794                         struct ovs_key_tcp *tcp_key;
1795
1796                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1797                         if (!nla)
1798                                 goto nla_put_failure;
1799                         tcp_key = nla_data(nla);
1800                         if (swkey->eth.type == htons(ETH_P_IP)) {
1801                                 tcp_key->tcp_src = output->ipv4.tp.src;
1802                                 tcp_key->tcp_dst = output->ipv4.tp.dst;
1803                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1804                                 tcp_key->tcp_src = output->ipv6.tp.src;
1805                                 tcp_key->tcp_dst = output->ipv6.tp.dst;
1806                         }
1807                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1808                         struct ovs_key_udp *udp_key;
1809
1810                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1811                         if (!nla)
1812                                 goto nla_put_failure;
1813                         udp_key = nla_data(nla);
1814                         if (swkey->eth.type == htons(ETH_P_IP)) {
1815                                 udp_key->udp_src = output->ipv4.tp.src;
1816                                 udp_key->udp_dst = output->ipv4.tp.dst;
1817                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1818                                 udp_key->udp_src = output->ipv6.tp.src;
1819                                 udp_key->udp_dst = output->ipv6.tp.dst;
1820                         }
1821                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1822                            swkey->ip.proto == IPPROTO_ICMP) {
1823                         struct ovs_key_icmp *icmp_key;
1824
1825                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1826                         if (!nla)
1827                                 goto nla_put_failure;
1828                         icmp_key = nla_data(nla);
1829                         icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1830                         icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1831                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1832                            swkey->ip.proto == IPPROTO_ICMPV6) {
1833                         struct ovs_key_icmpv6 *icmpv6_key;
1834
1835                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1836                                                 sizeof(*icmpv6_key));
1837                         if (!nla)
1838                                 goto nla_put_failure;
1839                         icmpv6_key = nla_data(nla);
1840                         icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1841                         icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1842
1843                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1844                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1845                                 struct ovs_key_nd *nd_key;
1846
1847                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1848                                 if (!nla)
1849                                         goto nla_put_failure;
1850                                 nd_key = nla_data(nla);
1851                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1852                                                         sizeof(nd_key->nd_target));
1853                                 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1854                                 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1855                         }
1856                 }
1857         }
1858
1859 unencap:
1860         if (encap)
1861                 nla_nest_end(skb, encap);
1862
1863         return 0;
1864
1865 nla_put_failure:
1866         return -EMSGSIZE;
1867 }
1868
1869 /* Initializes the flow module.
1870  * Returns zero if successful or a negative error code. */
1871 int ovs_flow_init(void)
1872 {
1873         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1874                                         0, NULL);
1875         if (flow_cache == NULL)
1876                 return -ENOMEM;
1877
1878         return 0;
1879 }
1880
1881 /* Uninitializes the flow module. */
1882 void ovs_flow_exit(void)
1883 {
1884         kmem_cache_destroy(flow_cache);
1885 }
1886
1887 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1888 {
1889         struct sw_flow_mask *mask;
1890
1891         mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1892         if (mask)
1893                 mask->ref_count = 0;
1894
1895         return mask;
1896 }
1897
1898 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1899 {
1900         mask->ref_count++;
1901 }
1902
1903 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1904 {
1905         struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1906
1907         kfree(mask);
1908 }
1909
1910 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1911 {
1912         if (!mask)
1913                 return;
1914
1915         BUG_ON(!mask->ref_count);
1916         mask->ref_count--;
1917
1918         if (!mask->ref_count) {
1919                 list_del_rcu(&mask->list);
1920                 if (deferred)
1921                         call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1922                 else
1923                         kfree(mask);
1924         }
1925 }
1926
1927 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1928                 const struct sw_flow_mask *b)
1929 {
1930         u8 *a_ = (u8 *)&a->key + a->range.start;
1931         u8 *b_ = (u8 *)&b->key + b->range.start;
1932
1933         return  (a->range.end == b->range.end)
1934                 && (a->range.start == b->range.start)
1935                 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1936 }
1937
1938 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1939                                            const struct sw_flow_mask *mask)
1940 {
1941         struct list_head *ml;
1942
1943         list_for_each(ml, tbl->mask_list) {
1944                 struct sw_flow_mask *m;
1945                 m = container_of(ml, struct sw_flow_mask, list);
1946                 if (ovs_sw_flow_mask_equal(mask, m))
1947                         return m;
1948         }
1949
1950         return NULL;
1951 }
1952
1953 /**
1954  * add a new mask into the mask list.
1955  * The caller needs to make sure that 'mask' is not the same
1956  * as any masks that are already on the list.
1957  */
1958 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1959 {
1960         list_add_rcu(&mask->list, tbl->mask_list);
1961 }
1962
1963 /**
1964  * Set 'range' fields in the mask to the value of 'val'.
1965  */
1966 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1967                 struct sw_flow_key_range *range, u8 val)
1968 {
1969         u8 *m = (u8 *)&mask->key + range->start;
1970
1971         mask->range = *range;
1972         memset(m, val, ovs_sw_flow_mask_size_roundup(mask));
1973 }