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) {                                              \
  86                         if ((match)->mask)                                  \
  87                                 (match)->mask->key.field = value;           \
  88                 } else {                                                    \
  89                         (match)->key->field = value;                        \
  90                 }                                                           \
  91         } while (0)
  92
  93 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
  94         do { \
  95                 update_range__(match, offsetof(struct sw_flow_key, field),  \
  96                                 len, is_mask);                              \
  97                 if (is_mask) {                                              \
  98                         if ((match)->mask)                                  \
  99                                 memcpy(&(match)->mask->key.field, value_p, len);\
 100                 } else {                                                    \
 101                         memcpy(&(match)->key->field, value_p, len);         \
 102                 }                                                           \
 103         } while (0)
 104
 105 void ovs_match_init(struct sw_flow_match *match,
 106                     struct sw_flow_key *key,
 107                     struct sw_flow_mask *mask)
 108 {
 109         memset(match, 0, sizeof(*match));
 110         match->key = key;
 111         match->mask = mask;
 112
 113         memset(key, 0, sizeof(*key));
 114
 115         if (mask) {
 116                 memset(&mask->key, 0, sizeof(mask->key));
 117                 mask->range.start = mask->range.end = 0;
 118         }
 119 }
 120
 121 static bool ovs_match_validate(const struct sw_flow_match *match,
 122                 u64 key_attrs, u64 mask_attrs)
 123 {
 124         u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
 125         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 126
 127         /* The following mask attributes allowed only if they
 128          * pass the validation tests. */
 129         mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
 130                         | (1ULL << OVS_KEY_ATTR_IPV6)
 131                         | (1ULL << OVS_KEY_ATTR_TCP)
 132                         | (1ULL << OVS_KEY_ATTR_UDP)
 133                         | (1ULL << OVS_KEY_ATTR_ICMP)
 134                         | (1ULL << OVS_KEY_ATTR_ICMPV6)
 135                         | (1ULL << OVS_KEY_ATTR_ARP)
 136                         | (1ULL << OVS_KEY_ATTR_ND));
 137
 138         if (match->key->phy.in_port == DP_MAX_PORTS &&
 139             match->mask && (match->mask->key.phy.in_port == 0xffff))
 140                 mask_allowed |= (1ULL << OVS_KEY_ATTR_IN_PORT);
 141
 142         if (match->key->eth.type == htons(ETH_P_802_2) &&
 143             match->mask && (match->mask->key.eth.type == htons(0xffff)))
 144                 mask_allowed |= (1ULL << OVS_KEY_ATTR_ETHERTYPE);
 145
 146         /* Check key attributes. */
 147         if (match->key->eth.type == htons(ETH_P_ARP)
 148                         || match->key->eth.type == htons(ETH_P_RARP)) {
 149                 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
 150                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 151                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 152         }
 153
 154         if (match->key->eth.type == htons(ETH_P_IP)) {
 155                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 156                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 157                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
 158
 159                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 160                         if (match->key->ip.proto == IPPROTO_UDP) {
 161                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 162                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 163                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 164                         }
 165
 166                         if (match->key->ip.proto == IPPROTO_TCP) {
 167                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 168                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 169                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 170                         }
 171
 172                         if (match->key->ip.proto == IPPROTO_ICMP) {
 173                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 174                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 175                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 176                         }
 177                 }
 178         }
 179
 180         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 181                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 182                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 183                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
 184
 185                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 186                         if (match->key->ip.proto == IPPROTO_UDP) {
 187                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 188                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 189                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 190                         }
 191
 192                         if (match->key->ip.proto == IPPROTO_TCP) {
 193                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 194                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 195                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 196                         }
 197
 198                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 199                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 200                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 201                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 202
 203                                 if (match->key->ipv6.tp.src ==
 204                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 205                                     match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 206                                         key_expected |= 1ULL << OVS_KEY_ATTR_ND;
 207                                         if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
 208                                                 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
 209                                 }
 210                         }
 211                 }
 212         }
 213
 214         if ((key_attrs & key_expected) != key_expected) {
 215                 /* Key attributes check failed. */
 216                 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
 217                                 key_attrs, key_expected);
 218                 return false;
 219         }
 220
 221         if ((mask_attrs & mask_allowed) != mask_attrs) {
 222                 /* Mask attributes check failed. */
 223                 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
 224                                 mask_attrs, mask_allowed);
 225                 return false;
 226         }
 227
 228         return true;
 229 }
 230
 231 static int check_header(struct sk_buff *skb, int len)
 232 {
 233         if (unlikely(skb->len < len))
 234                 return -EINVAL;
 235         if (unlikely(!pskb_may_pull(skb, len)))
 236                 return -ENOMEM;
 237         return 0;
 238 }
 239
 240 static bool arphdr_ok(struct sk_buff *skb)
 241 {
 242         return pskb_may_pull(skb, skb_network_offset(skb) +
 243                                   sizeof(struct arp_eth_header));
 244 }
 245
 246 static int check_iphdr(struct sk_buff *skb)
 247 {
 248         unsigned int nh_ofs = skb_network_offset(skb);
 249         unsigned int ip_len;
 250         int err;
 251
 252         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 253         if (unlikely(err))
 254                 return err;
 255
 256         ip_len = ip_hdrlen(skb);
 257         if (unlikely(ip_len < sizeof(struct iphdr) ||
 258                      skb->len < nh_ofs + ip_len))
 259                 return -EINVAL;
 260
 261         skb_set_transport_header(skb, nh_ofs + ip_len);
 262         return 0;
 263 }
 264
 265 static bool tcphdr_ok(struct sk_buff *skb)
 266 {
 267         int th_ofs = skb_transport_offset(skb);
 268         int tcp_len;
 269
 270         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 271                 return false;
 272
 273         tcp_len = tcp_hdrlen(skb);
 274         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 275                      skb->len < th_ofs + tcp_len))
 276                 return false;
 277
 278         return true;
 279 }
 280
 281 static bool udphdr_ok(struct sk_buff *skb)
 282 {
 283         return pskb_may_pull(skb, skb_transport_offset(skb) +
 284                                   sizeof(struct udphdr));
 285 }
 286
 287 static bool icmphdr_ok(struct sk_buff *skb)
 288 {
 289         return pskb_may_pull(skb, skb_transport_offset(skb) +
 290                                   sizeof(struct icmphdr));
 291 }
 292
 293 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 294 {
 295         struct timespec cur_ts;
 296         u64 cur_ms, idle_ms;
 297
 298         ktime_get_ts(&cur_ts);
 299         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 300         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 301                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 302
 303         return cur_ms - idle_ms;
 304 }
 305
 306 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 307 {
 308         unsigned int nh_ofs = skb_network_offset(skb);
 309         unsigned int nh_len;
 310         int payload_ofs;
 311         struct ipv6hdr *nh;
 312         uint8_t nexthdr;
 313         __be16 frag_off;
 314         int err;
 315
 316         err = check_header(skb, nh_ofs + sizeof(*nh));
 317         if (unlikely(err))
 318                 return err;
 319
 320         nh = ipv6_hdr(skb);
 321         nexthdr = nh->nexthdr;
 322         payload_ofs = (u8 *)(nh + 1) - skb->data;
 323
 324         key->ip.proto = NEXTHDR_NONE;
 325         key->ip.tos = ipv6_get_dsfield(nh);
 326         key->ip.ttl = nh->hop_limit;
 327         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 328         key->ipv6.addr.src = nh->saddr;
 329         key->ipv6.addr.dst = nh->daddr;
 330
 331         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 332         if (unlikely(payload_ofs < 0))
 333                 return -EINVAL;
 334
 335         if (frag_off) {
 336                 if (frag_off & htons(~0x7))
 337                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 338                 else
 339                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 340         }
 341
 342         nh_len = payload_ofs - nh_ofs;
 343         skb_set_transport_header(skb, nh_ofs + nh_len);
 344         key->ip.proto = nexthdr;
 345         return nh_len;
 346 }
 347
 348 static bool icmp6hdr_ok(struct sk_buff *skb)
 349 {
 350         return pskb_may_pull(skb, skb_transport_offset(skb) +
 351                                   sizeof(struct icmp6hdr));
 352 }
 353
 354 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
 355                        const struct sw_flow_mask *mask)
 356 {
 357         u8 *m = (u8 *)&mask->key + mask->range.start;
 358         u8 *s = (u8 *)src + mask->range.start;
 359         u8 *d = (u8 *)dst + mask->range.start;
 360         int i;
 361
 362         memset(dst, 0, sizeof(*dst));
 363         for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
 364                 *d = *s & *m;
 365                 d++, s++, m++;
 366         }
 367 }
 368
 369 #define TCP_FLAGS_OFFSET 13
 370 #define TCP_FLAG_MASK 0x3f
 371
 372 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 373 {
 374         u8 tcp_flags = 0;
 375
 376         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 377              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 378             flow->key.ip.proto == IPPROTO_TCP &&
 379             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 380                 u8 *tcp = (u8 *)tcp_hdr(skb);
 381                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 382         }
 383
 384         spin_lock(&flow->lock);
 385         flow->used = jiffies;
 386         flow->packet_count++;
 387         flow->byte_count += skb->len;
 388         flow->tcp_flags |= tcp_flags;
 389         spin_unlock(&flow->lock);
 390 }
 391
 392 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
 393 {
 394         struct sw_flow_actions *sfa;
 395
 396         if (size > MAX_ACTIONS_BUFSIZE)
 397                 return ERR_PTR(-EINVAL);
 398
 399         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 400         if (!sfa)
 401                 return ERR_PTR(-ENOMEM);
 402
 403         sfa->actions_len = 0;
 404         return sfa;
 405 }
 406
 407 struct sw_flow *ovs_flow_alloc(void)
 408 {
 409         struct sw_flow *flow;
 410
 411         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 412         if (!flow)
 413                 return ERR_PTR(-ENOMEM);
 414
 415         spin_lock_init(&flow->lock);
 416         flow->sf_acts = NULL;
 417         flow->mask = NULL;
 418
 419         return flow;
 420 }
 421
 422 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 423 {
 424         hash = jhash_1word(hash, table->hash_seed);
 425         return flex_array_get(table->buckets,
 426                                 (hash & (table->n_buckets - 1)));
 427 }
 428
 429 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 430 {
 431         struct flex_array *buckets;
 432         int i, err;
 433
 434         buckets = flex_array_alloc(sizeof(struct hlist_head),
 435                                    n_buckets, GFP_KERNEL);
 436         if (!buckets)
 437                 return NULL;
 438
 439         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 440         if (err) {
 441                 flex_array_free(buckets);
 442                 return NULL;
 443         }
 444
 445         for (i = 0; i < n_buckets; i++)
 446                 INIT_HLIST_HEAD((struct hlist_head *)
 447                                         flex_array_get(buckets, i));
 448
 449         return buckets;
 450 }
 451
 452 static void free_buckets(struct flex_array *buckets)
 453 {
 454         flex_array_free(buckets);
 455 }
 456
 457 static struct flow_table *__flow_tbl_alloc(int new_size)
 458 {
 459         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 460
 461         if (!table)
 462                 return NULL;
 463
 464         table->buckets = alloc_buckets(new_size);
 465
 466         if (!table->buckets) {
 467                 kfree(table);
 468                 return NULL;
 469         }
 470         table->n_buckets = new_size;
 471         table->count = 0;
 472         table->node_ver = 0;
 473         table->keep_flows = false;
 474         get_random_bytes(&table->hash_seed, sizeof(u32));
 475         table->mask_list = NULL;
 476
 477         return table;
 478 }
 479
 480 static void __flow_tbl_destroy(struct flow_table *table)
 481 {
 482         int i;
 483
 484         if (table->keep_flows)
 485                 goto skip_flows;
 486
 487         for (i = 0; i < table->n_buckets; i++) {
 488                 struct sw_flow *flow;
 489                 struct hlist_head *head = flex_array_get(table->buckets, i);
 490                 struct hlist_node *n;
 491                 int ver = table->node_ver;
 492
 493                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 494                         hlist_del(&flow->hash_node[ver]);
 495                         ovs_flow_free(flow, false);
 496                 }
 497         }
 498
 499         BUG_ON(!list_empty(table->mask_list));
 500         kfree(table->mask_list);
 501
 502 skip_flows:
 503         free_buckets(table->buckets);
 504         kfree(table);
 505 }
 506
 507 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 508 {
 509         struct flow_table *table = __flow_tbl_alloc(new_size);
 510
 511         if (!table)
 512                 return NULL;
 513
 514         table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
 515         if (!table->mask_list) {
 516                 table->keep_flows = true;
 517                 __flow_tbl_destroy(table);
 518                 return NULL;
 519         }
 520         INIT_LIST_HEAD(table->mask_list);
 521
 522         return table;
 523 }
 524
 525 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 526 {
 527         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 528
 529         __flow_tbl_destroy(table);
 530 }
 531
 532 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
 533 {
 534         if (!table)
 535                 return;
 536
 537         if (deferred)
 538                 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 539         else
 540                 __flow_tbl_destroy(table);
 541 }
 542
 543 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
 544 {
 545         struct sw_flow *flow;
 546         struct hlist_head *head;
 547         int ver;
 548         int i;
 549
 550         ver = table->node_ver;
 551         while (*bucket < table->n_buckets) {
 552                 i = 0;
 553                 head = flex_array_get(table->buckets, *bucket);
 554                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 555                         if (i < *last) {
 556                                 i++;
 557                                 continue;
 558                         }
 559                         *last = i + 1;
 560                         return flow;
 561                 }
 562                 (*bucket)++;
 563                 *last = 0;
 564         }
 565
 566         return NULL;
 567 }
 568
 569 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
 570 {
 571         struct hlist_head *head;
 572
 573         head = find_bucket(table, flow->hash);
 574         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 575
 576         table->count++;
 577 }
 578
 579 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 580 {
 581         int old_ver;
 582         int i;
 583
 584         old_ver = old->node_ver;
 585         new->node_ver = !old_ver;
 586
 587         /* Insert in new table. */
 588         for (i = 0; i < old->n_buckets; i++) {
 589                 struct sw_flow *flow;
 590                 struct hlist_head *head;
 591
 592                 head = flex_array_get(old->buckets, i);
 593
 594                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 595                         __tbl_insert(new, flow);
 596         }
 597
 598         new->mask_list = old->mask_list;
 599         old->keep_flows = true;
 600 }
 601
 602 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 603 {
 604         struct flow_table *new_table;
 605
 606         new_table = __flow_tbl_alloc(n_buckets);
 607         if (!new_table)
 608                 return ERR_PTR(-ENOMEM);
 609
 610         flow_table_copy_flows(table, new_table);
 611
 612         return new_table;
 613 }
 614
 615 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 616 {
 617         return __flow_tbl_rehash(table, table->n_buckets);
 618 }
 619
 620 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 621 {
 622         return __flow_tbl_rehash(table, table->n_buckets * 2);
 623 }
 624
 625 static void __flow_free(struct sw_flow *flow)
 626 {
 627         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 628         kmem_cache_free(flow_cache, flow);
 629 }
 630
 631 static void rcu_free_flow_callback(struct rcu_head *rcu)
 632 {
 633         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 634
 635         __flow_free(flow);
 636 }
 637
 638 void ovs_flow_free(struct sw_flow *flow, bool deferred)
 639 {
 640         if (!flow)
 641                 return;
 642
 643         ovs_sw_flow_mask_del_ref(flow->mask, 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         ovs_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->mask == mask &&
1053                     __flow_cmp_key(flow, &masked_key, key_start, key_len))
1054                         return flow;
1055         }
1056         return NULL;
1057 }
1058
1059 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1060                                 const struct sw_flow_key *key)
1061 {
1062         struct sw_flow *flow = NULL;
1063         struct sw_flow_mask *mask;
1064
1065         list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1066                 flow = ovs_masked_flow_lookup(tbl, key, mask);
1067                 if (flow)  /* Found */
1068                         break;
1069         }
1070
1071         return flow;
1072 }
1073
1074
1075 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1076 {
1077         flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1078                         flow->mask->range.end);
1079         __tbl_insert(table, flow);
1080 }
1081
1082 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1083 {
1084         BUG_ON(table->count == 0);
1085         hlist_del_rcu(&flow->hash_node[table->node_ver]);
1086         table->count--;
1087 }
1088
1089 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
1090 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1091         [OVS_KEY_ATTR_ENCAP] = -1,
1092         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1093         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1094         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1095         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1096         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1097         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1098         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1099         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1100         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1101         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1102         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1103         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1104         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1105         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1106         [OVS_KEY_ATTR_TUNNEL] = -1,
1107 };
1108
1109 static bool is_all_zero(const u8 *fp, size_t size)
1110 {
1111         int i;
1112
1113         if (!fp)
1114                 return false;
1115
1116         for (i = 0; i < size; i++)
1117                 if (fp[i])
1118                         return false;
1119
1120         return true;
1121 }
1122
1123 static int __parse_flow_nlattrs(const struct nlattr *attr,
1124                               const struct nlattr *a[],
1125                               u64 *attrsp, bool nz)
1126 {
1127         const struct nlattr *nla;
1128         u64 attrs;
1129         int rem;
1130
1131         attrs = *attrsp;
1132         nla_for_each_nested(nla, attr, rem) {
1133                 u16 type = nla_type(nla);
1134                 int expected_len;
1135
1136                 if (type > OVS_KEY_ATTR_MAX) {
1137                         OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1138                                   type, OVS_KEY_ATTR_MAX);
1139                 }
1140
1141                 if (attrs & (1ULL << type)) {
1142                         OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1143                         return -EINVAL;
1144                 }
1145
1146                 expected_len = ovs_key_lens[type];
1147                 if (nla_len(nla) != expected_len && expected_len != -1) {
1148                         OVS_NLERR("Key attribute has unexpected length (type=%d"
1149                                   ", length=%d, expected=%d).\n", type,
1150                                   nla_len(nla), expected_len);
1151                         return -EINVAL;
1152                 }
1153
1154                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1155                         attrs |= 1ULL << type;
1156                         a[type] = nla;
1157                 }
1158         }
1159         if (rem) {
1160                 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1161                 return -EINVAL;
1162         }
1163
1164         *attrsp = attrs;
1165         return 0;
1166 }
1167
1168 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1169                               const struct nlattr *a[], u64 *attrsp)
1170 {
1171         return __parse_flow_nlattrs(attr, a, attrsp, true);
1172 }
1173
1174 static int parse_flow_nlattrs(const struct nlattr *attr,
1175                               const struct nlattr *a[], u64 *attrsp)
1176 {
1177         return __parse_flow_nlattrs(attr, a, attrsp, false);
1178 }
1179
1180 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1181                          struct sw_flow_match *match, bool is_mask)
1182 {
1183         struct nlattr *a;
1184         int rem;
1185         bool ttl = false;
1186         __be16 tun_flags = 0;
1187
1188         nla_for_each_nested(a, attr, rem) {
1189                 int type = nla_type(a);
1190                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1191                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1192                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1193                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1194                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1195                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1196                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1197                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1198                 };
1199
1200                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1201                         OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1202                         type, OVS_TUNNEL_KEY_ATTR_MAX);
1203                         return -EINVAL;
1204                 }
1205
1206                 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1207                         OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1208                                   " legnth (type=%d, length=%d, expected=%d).\n",
1209                                   type, nla_len(a), ovs_tunnel_key_lens[type]);
1210                         return -EINVAL;
1211                 }
1212
1213                 switch (type) {
1214                 case OVS_TUNNEL_KEY_ATTR_ID:
1215                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1216                                         nla_get_be64(a), is_mask);
1217                         tun_flags |= TUNNEL_KEY;
1218                         break;
1219                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1220                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1221                                         nla_get_be32(a), is_mask);
1222                         break;
1223                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1224                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1225                                         nla_get_be32(a), is_mask);
1226                         break;
1227                 case OVS_TUNNEL_KEY_ATTR_TOS:
1228                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1229                                         nla_get_u8(a), is_mask);
1230                         break;
1231                 case OVS_TUNNEL_KEY_ATTR_TTL:
1232                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1233                                         nla_get_u8(a), is_mask);
1234                         ttl = true;
1235                         break;
1236                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1237                         tun_flags |= TUNNEL_DONT_FRAGMENT;
1238                         break;
1239                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1240                         tun_flags |= TUNNEL_CSUM;
1241                         break;
1242                 default:
1243                         return -EINVAL;
1244                 }
1245         }
1246
1247         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1248
1249         if (rem > 0) {
1250                 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1251                 return -EINVAL;
1252         }
1253
1254         if (!match->key->tun_key.ipv4_dst) {
1255                 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1256                 return -EINVAL;
1257         }
1258
1259         if (!ttl) {
1260                 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1261                 return -EINVAL;
1262         }
1263
1264         return 0;
1265 }
1266
1267 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1268                         const struct ovs_key_ipv4_tunnel *tun_key,
1269                         const struct ovs_key_ipv4_tunnel *output)
1270 {
1271         struct nlattr *nla;
1272
1273         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1274         if (!nla)
1275                 return -EMSGSIZE;
1276
1277         if (output->tun_flags & TUNNEL_KEY &&
1278             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1279                 return -EMSGSIZE;
1280         if (output->ipv4_src &&
1281                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1282                 return -EMSGSIZE;
1283         if (output->ipv4_dst &&
1284                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1285                 return -EMSGSIZE;
1286         if (output->ipv4_tos &&
1287                 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1288                 return -EMSGSIZE;
1289         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1290                 return -EMSGSIZE;
1291         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1292                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1293                 return -EMSGSIZE;
1294         if ((output->tun_flags & TUNNEL_CSUM) &&
1295                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1296                 return -EMSGSIZE;
1297
1298         nla_nest_end(skb, nla);
1299         return 0;
1300 }
1301
1302
1303 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
1304                 const struct nlattr **a, bool is_mask)
1305 {
1306         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1307                 SW_FLOW_KEY_PUT(match, phy.priority,
1308                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1309                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1310         }
1311
1312         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1313                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1314
1315                 if (!is_mask && in_port >= DP_MAX_PORTS)
1316                         return -EINVAL;
1317                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1318                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1319         } else if (!is_mask) {
1320                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1321         }
1322
1323         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1324                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1325 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1326                 if (!is_mask && mark != 0) {
1327                         OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1328                         return -EINVAL;
1329                 }
1330 #endif
1331                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1332                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1333         }
1334         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1335                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1336                                         is_mask))
1337                         return -EINVAL;
1338                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1339         }
1340         return 0;
1341 }
1342
1343 static int ovs_key_from_nlattrs(struct sw_flow_match *match,  u64 attrs,
1344                 const struct nlattr **a, bool is_mask)
1345 {
1346         int err;
1347         u64 orig_attrs = attrs;
1348
1349         err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1350         if (err)
1351                 return err;
1352
1353         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1354                 const struct ovs_key_ethernet *eth_key;
1355
1356                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1357                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1358                                 eth_key->eth_src, ETH_ALEN, is_mask);
1359                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1360                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1361                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1362         }
1363
1364         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1365                 __be16 tci;
1366
1367                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1368                 if (!is_mask)
1369                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1370                                 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1371                                 return -EINVAL;
1372                         }
1373
1374                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1375                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1376         }
1377
1378         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1379                 __be16 eth_type;
1380
1381                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1382                 if (!is_mask && ntohs(eth_type) < ETH_P_802_3_MIN) {
1383                         OVS_NLERR("EtherType is less than mimimum (type=%x, min=%x).\n",
1384                                         ntohs(eth_type), ETH_P_802_3_MIN);
1385                         return -EINVAL;
1386                 }
1387
1388                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1389                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1390         } else if (!is_mask) {
1391                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1392         }
1393
1394         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1395                 const struct ovs_key_ipv4 *ipv4_key;
1396
1397                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1398                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1399                         OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1400                                 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1401                         return -EINVAL;
1402                 }
1403                 SW_FLOW_KEY_PUT(match, ip.proto,
1404                                 ipv4_key->ipv4_proto, is_mask);
1405                 SW_FLOW_KEY_PUT(match, ip.tos,
1406                                 ipv4_key->ipv4_tos, is_mask);
1407                 SW_FLOW_KEY_PUT(match, ip.ttl,
1408                                 ipv4_key->ipv4_ttl, is_mask);
1409                 SW_FLOW_KEY_PUT(match, ip.frag,
1410                                 ipv4_key->ipv4_frag, is_mask);
1411                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1412                                 ipv4_key->ipv4_src, is_mask);
1413                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1414                                 ipv4_key->ipv4_dst, is_mask);
1415                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1416         }
1417
1418         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1419                 const struct ovs_key_ipv6 *ipv6_key;
1420
1421                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1422                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1423                         OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1424                                 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1425                         return -EINVAL;
1426                 }
1427                 SW_FLOW_KEY_PUT(match, ipv6.label,
1428                                 ipv6_key->ipv6_label, is_mask);
1429                 SW_FLOW_KEY_PUT(match, ip.proto,
1430                                 ipv6_key->ipv6_proto, is_mask);
1431                 SW_FLOW_KEY_PUT(match, ip.tos,
1432                                 ipv6_key->ipv6_tclass, is_mask);
1433                 SW_FLOW_KEY_PUT(match, ip.ttl,
1434                                 ipv6_key->ipv6_hlimit, is_mask);
1435                 SW_FLOW_KEY_PUT(match, ip.frag,
1436                                 ipv6_key->ipv6_frag, is_mask);
1437                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1438                                 ipv6_key->ipv6_src,
1439                                 sizeof(match->key->ipv6.addr.src),
1440                                 is_mask);
1441                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1442                                 ipv6_key->ipv6_dst,
1443                                 sizeof(match->key->ipv6.addr.dst),
1444                                 is_mask);
1445
1446                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1447         }
1448
1449         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1450                 const struct ovs_key_arp *arp_key;
1451
1452                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1453                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1454                         OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1455                                   arp_key->arp_op);
1456                         return -EINVAL;
1457                 }
1458
1459                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1460                                 arp_key->arp_sip, is_mask);
1461                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1462                         arp_key->arp_tip, is_mask);
1463                 SW_FLOW_KEY_PUT(match, ip.proto,
1464                                 ntohs(arp_key->arp_op), is_mask);
1465                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1466                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1467                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1468                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1469
1470                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1471         }
1472
1473         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1474                 const struct ovs_key_tcp *tcp_key;
1475
1476                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1477                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1478                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1479                                         tcp_key->tcp_src, is_mask);
1480                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1481                                         tcp_key->tcp_dst, is_mask);
1482                 } else {
1483                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1484                                         tcp_key->tcp_src, is_mask);
1485                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1486                                         tcp_key->tcp_dst, is_mask);
1487                 }
1488                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1489         }
1490
1491         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1492                 const struct ovs_key_udp *udp_key;
1493
1494                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1495                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1496                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1497                                         udp_key->udp_src, is_mask);
1498                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1499                                         udp_key->udp_dst, is_mask);
1500                 } else {
1501                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1502                                         udp_key->udp_src, is_mask);
1503                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1504                                         udp_key->udp_dst, is_mask);
1505                 }
1506                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1507         }
1508
1509         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1510                 const struct ovs_key_icmp *icmp_key;
1511
1512                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1513                 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1514                                 htons(icmp_key->icmp_type), is_mask);
1515                 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1516                                 htons(icmp_key->icmp_code), is_mask);
1517                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1518         }
1519
1520         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1521                 const struct ovs_key_icmpv6 *icmpv6_key;
1522
1523                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1524                 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1525                                 htons(icmpv6_key->icmpv6_type), is_mask);
1526                 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1527                                 htons(icmpv6_key->icmpv6_code), is_mask);
1528                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1529         }
1530
1531         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1532                 const struct ovs_key_nd *nd_key;
1533
1534                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1535                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1536                         nd_key->nd_target,
1537                         sizeof(match->key->ipv6.nd.target),
1538                         is_mask);
1539                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1540                         nd_key->nd_sll, ETH_ALEN, is_mask);
1541                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1542                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1543                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1544         }
1545
1546         if (attrs != 0)
1547                 return -EINVAL;
1548
1549         return 0;
1550 }
1551
1552 /**
1553  * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1554  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1555  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1556  * does not include any don't care bit.
1557  * @match: receives the extracted flow match information.
1558  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1559  * sequence. The fields should of the packet that triggered the creation
1560  * of this flow.
1561  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1562  * attribute specifies the mask field of the wildcarded flow.
1563  */
1564 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1565                            const struct nlattr *key,
1566                            const struct nlattr *mask)
1567 {
1568         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1569         const struct nlattr *encap;
1570         u64 key_attrs = 0;
1571         u64 mask_attrs = 0;
1572         bool encap_valid = false;
1573         int err;
1574
1575         err = parse_flow_nlattrs(key, a, &key_attrs);
1576         if (err)
1577                 return err;
1578
1579         if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1580                 encap = a[OVS_KEY_ATTR_ENCAP];
1581                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1582                 if (nla_len(encap)) {
1583                         __be16 eth_type = 0; /* ETH_P_8021Q */
1584
1585                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1586                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1587
1588                         if  ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1589                                 encap_valid = true;
1590                                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1591                                 err = parse_flow_nlattrs(encap, a, &key_attrs);
1592                         } else {
1593                                 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1594                                 err = -EINVAL;
1595                         }
1596
1597                         if (err)
1598                                 return err;
1599                 }
1600         }
1601
1602         err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1603         if (err)
1604                 return err;
1605
1606         if (mask) {
1607                 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1608                 if (err)
1609                         return err;
1610
1611                 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1612                         __be16 eth_type = 0;
1613
1614                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1615                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1616                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1617                         if (eth_type == htons(0xffff)) {
1618                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1619                                 encap = a[OVS_KEY_ATTR_ENCAP];
1620                                 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1621                         } else {
1622                                 OVS_NLERR("VLAN frames must have an exact match"
1623                                          " on the TPID (mask=%x).\n",
1624                                          ntohs(eth_type));
1625                                 err = -EINVAL;
1626                         }
1627
1628                         if (err)
1629                                 return err;
1630                 }
1631
1632                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1633                 if (err)
1634                         return err;
1635         } else {
1636                 /* Populate exact match flow's key mask. */
1637                 if (match->mask)
1638                         ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1639         }
1640
1641         if (!ovs_match_validate(match, key_attrs, mask_attrs))
1642                 return -EINVAL;
1643
1644         return 0;
1645 }
1646
1647 /**
1648  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1649  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1650  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1651  * sequence.
1652  *
1653  * This parses a series of Netlink attributes that form a flow key, which must
1654  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1655  * get the metadata, that is, the parts of the flow key that cannot be
1656  * extracted from the packet itself.
1657  */
1658
1659 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1660                 const struct nlattr *attr)
1661 {
1662         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1663         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1664         u64 attrs = 0;
1665         int err;
1666         struct sw_flow_match match;
1667
1668         flow->key.phy.in_port = DP_MAX_PORTS;
1669         flow->key.phy.priority = 0;
1670         flow->key.phy.skb_mark = 0;
1671         memset(tun_key, 0, sizeof(flow->key.tun_key));
1672
1673         err = parse_flow_nlattrs(attr, a, &attrs);
1674         if (err)
1675                 return -EINVAL;
1676
1677         memset(&match, 0, sizeof(match));
1678         match.key = &flow->key;
1679
1680         err = metadata_from_nlattrs(&match, &attrs, a, false);
1681         if (err)
1682                 return err;
1683
1684         return 0;
1685 }
1686
1687 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1688                 const struct sw_flow_key *output, struct sk_buff *skb)
1689 {
1690         struct ovs_key_ethernet *eth_key;
1691         struct nlattr *nla, *encap;
1692
1693         if (output->phy.priority &&
1694                 nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1695                 goto nla_put_failure;
1696
1697         if (swkey->tun_key.ipv4_dst &&
1698             ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1699                 goto nla_put_failure;
1700
1701         if (swkey->phy.in_port == DP_MAX_PORTS) {
1702                 if ((swkey != output) && (output->phy.in_port == 0xffff))
1703                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1704                                 goto nla_put_failure;
1705         } else {
1706                 u16 upper_u16;
1707                 upper_u16 = (swkey == output) ? 0 : 0xffff;
1708
1709                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1710                                 (upper_u16 << 16) | output->phy.in_port))
1711                         goto nla_put_failure;
1712         }
1713
1714         if (output->phy.skb_mark &&
1715                 nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1716                 goto nla_put_failure;
1717
1718         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1719         if (!nla)
1720                 goto nla_put_failure;
1721
1722         eth_key = nla_data(nla);
1723         memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1724         memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1725
1726         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1727                 __be16 eth_type;
1728                 eth_type = (swkey == output) ? htons(ETH_P_8021Q) : htons(0xffff) ;
1729                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1730                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1731                         goto nla_put_failure;
1732                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1733                 if (!swkey->eth.tci)
1734                         goto unencap;
1735         } else
1736                 encap = NULL;
1737
1738         if (swkey->eth.type == htons(ETH_P_802_2)) {
1739                 /*
1740                  * Ethertype 802.2 is represented in the netlink with omitted
1741                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1742                  * 0xffff in the mask attribute.  Ethertype can also
1743                  * be wildcarded.
1744                  */
1745                 if (swkey != output && output->eth.type)
1746                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1747                                                 output->eth.type))
1748                                 goto nla_put_failure;
1749                 goto unencap;
1750         }
1751
1752         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1753                 goto nla_put_failure;
1754
1755         if (swkey->eth.type == htons(ETH_P_IP)) {
1756                 struct ovs_key_ipv4 *ipv4_key;
1757
1758                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1759                 if (!nla)
1760                         goto nla_put_failure;
1761                 ipv4_key = nla_data(nla);
1762                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1763                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1764                 ipv4_key->ipv4_proto = output->ip.proto;
1765                 ipv4_key->ipv4_tos = output->ip.tos;
1766                 ipv4_key->ipv4_ttl = output->ip.ttl;
1767                 ipv4_key->ipv4_frag = output->ip.frag;
1768         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1769                 struct ovs_key_ipv6 *ipv6_key;
1770
1771                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1772                 if (!nla)
1773                         goto nla_put_failure;
1774                 ipv6_key = nla_data(nla);
1775                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1776                                 sizeof(ipv6_key->ipv6_src));
1777                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1778                                 sizeof(ipv6_key->ipv6_dst));
1779                 ipv6_key->ipv6_label = output->ipv6.label;
1780                 ipv6_key->ipv6_proto = output->ip.proto;
1781                 ipv6_key->ipv6_tclass = output->ip.tos;
1782                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1783                 ipv6_key->ipv6_frag = output->ip.frag;
1784         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1785                    swkey->eth.type == htons(ETH_P_RARP)) {
1786                 struct ovs_key_arp *arp_key;
1787
1788                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1789                 if (!nla)
1790                         goto nla_put_failure;
1791                 arp_key = nla_data(nla);
1792                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1793                 arp_key->arp_sip = output->ipv4.addr.src;
1794                 arp_key->arp_tip = output->ipv4.addr.dst;
1795                 arp_key->arp_op = htons(output->ip.proto);
1796                 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1797                 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1798         }
1799
1800         if ((swkey->eth.type == htons(ETH_P_IP) ||
1801              swkey->eth.type == htons(ETH_P_IPV6)) &&
1802              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1803
1804                 if (swkey->ip.proto == IPPROTO_TCP) {
1805                         struct ovs_key_tcp *tcp_key;
1806
1807                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1808                         if (!nla)
1809                                 goto nla_put_failure;
1810                         tcp_key = nla_data(nla);
1811                         if (swkey->eth.type == htons(ETH_P_IP)) {
1812                                 tcp_key->tcp_src = output->ipv4.tp.src;
1813                                 tcp_key->tcp_dst = output->ipv4.tp.dst;
1814                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1815                                 tcp_key->tcp_src = output->ipv6.tp.src;
1816                                 tcp_key->tcp_dst = output->ipv6.tp.dst;
1817                         }
1818                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1819                         struct ovs_key_udp *udp_key;
1820
1821                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1822                         if (!nla)
1823                                 goto nla_put_failure;
1824                         udp_key = nla_data(nla);
1825                         if (swkey->eth.type == htons(ETH_P_IP)) {
1826                                 udp_key->udp_src = output->ipv4.tp.src;
1827                                 udp_key->udp_dst = output->ipv4.tp.dst;
1828                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1829                                 udp_key->udp_src = output->ipv6.tp.src;
1830                                 udp_key->udp_dst = output->ipv6.tp.dst;
1831                         }
1832                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1833                            swkey->ip.proto == IPPROTO_ICMP) {
1834                         struct ovs_key_icmp *icmp_key;
1835
1836                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1837                         if (!nla)
1838                                 goto nla_put_failure;
1839                         icmp_key = nla_data(nla);
1840                         icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1841                         icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1842                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1843                            swkey->ip.proto == IPPROTO_ICMPV6) {
1844                         struct ovs_key_icmpv6 *icmpv6_key;
1845
1846                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1847                                                 sizeof(*icmpv6_key));
1848                         if (!nla)
1849                                 goto nla_put_failure;
1850                         icmpv6_key = nla_data(nla);
1851                         icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1852                         icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1853
1854                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1855                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1856                                 struct ovs_key_nd *nd_key;
1857
1858                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1859                                 if (!nla)
1860                                         goto nla_put_failure;
1861                                 nd_key = nla_data(nla);
1862                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1863                                                         sizeof(nd_key->nd_target));
1864                                 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1865                                 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1866                         }
1867                 }
1868         }
1869
1870 unencap:
1871         if (encap)
1872                 nla_nest_end(skb, encap);
1873
1874         return 0;
1875
1876 nla_put_failure:
1877         return -EMSGSIZE;
1878 }
1879
1880 /* Initializes the flow module.
1881  * Returns zero if successful or a negative error code. */
1882 int ovs_flow_init(void)
1883 {
1884         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1885                                         0, NULL);
1886         if (flow_cache == NULL)
1887                 return -ENOMEM;
1888
1889         return 0;
1890 }
1891
1892 /* Uninitializes the flow module. */
1893 void ovs_flow_exit(void)
1894 {
1895         kmem_cache_destroy(flow_cache);
1896 }
1897
1898 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1899 {
1900         struct sw_flow_mask *mask;
1901
1902         mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1903         if (mask)
1904                 mask->ref_count = 0;
1905
1906         return mask;
1907 }
1908
1909 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1910 {
1911         mask->ref_count++;
1912 }
1913
1914 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1915 {
1916         struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1917
1918         kfree(mask);
1919 }
1920
1921 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1922 {
1923         if (!mask)
1924                 return;
1925
1926         BUG_ON(!mask->ref_count);
1927         mask->ref_count--;
1928
1929         if (!mask->ref_count) {
1930                 list_del_rcu(&mask->list);
1931                 if (deferred)
1932                         call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1933                 else
1934                         kfree(mask);
1935         }
1936 }
1937
1938 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1939                 const struct sw_flow_mask *b)
1940 {
1941         u8 *a_ = (u8 *)&a->key + a->range.start;
1942         u8 *b_ = (u8 *)&b->key + b->range.start;
1943
1944         return  (a->range.end == b->range.end)
1945                 && (a->range.start == b->range.start)
1946                 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1947 }
1948
1949 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1950                                            const struct sw_flow_mask *mask)
1951 {
1952         struct list_head *ml;
1953
1954         list_for_each(ml, tbl->mask_list) {
1955                 struct sw_flow_mask *m;
1956                 m = container_of(ml, struct sw_flow_mask, list);
1957                 if (ovs_sw_flow_mask_equal(mask, m))
1958                         return m;
1959         }
1960
1961         return NULL;
1962 }
1963
1964 /**
1965  * add a new mask into the mask list.
1966  * The caller needs to make sure that 'mask' is not the same
1967  * as any masks that are already on the list.
1968  */
1969 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1970 {
1971         list_add_rcu(&mask->list, tbl->mask_list);
1972 }
1973
1974 /**
1975  * Set 'range' fields in the mask to the value of 'val'.
1976  */
1977 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1978                 struct sw_flow_key_range *range, u8 val)
1979 {
1980         u8 *m = (u8 *)&mask->key + range->start;
1981
1982         mask->range = *range;
1983         memset(m, val, ovs_sw_flow_mask_size_roundup(mask));
1984 }