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 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
 353                        const struct sw_flow_mask *mask)
 354 {
 355         u8 *m = (u8 *)&mask->key + mask->range.start;
 356         u8 *s = (u8 *)src + mask->range.start;
 357         u8 *d = (u8 *)dst + mask->range.start;
 358         int i;
 359
 360         memset(dst, 0, sizeof(*dst));
 361         for (i = 0; i < ovs_sw_flow_mask_size_roundup(mask); i++) {
 362                 *d = *s & *m;
 363                 d++, s++, m++;
 364         }
 365 }
 366
 367 #define TCP_FLAGS_OFFSET 13
 368 #define TCP_FLAG_MASK 0x3f
 369
 370 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 371 {
 372         u8 tcp_flags = 0;
 373
 374         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 375              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 376             flow->key.ip.proto == IPPROTO_TCP &&
 377             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 378                 u8 *tcp = (u8 *)tcp_hdr(skb);
 379                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 380         }
 381
 382         spin_lock(&flow->lock);
 383         flow->used = jiffies;
 384         flow->packet_count++;
 385         flow->byte_count += skb->len;
 386         flow->tcp_flags |= tcp_flags;
 387         spin_unlock(&flow->lock);
 388 }
 389
 390 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
 391 {
 392         struct sw_flow_actions *sfa;
 393
 394         if (size > MAX_ACTIONS_BUFSIZE)
 395                 return ERR_PTR(-EINVAL);
 396
 397         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 398         if (!sfa)
 399                 return ERR_PTR(-ENOMEM);
 400
 401         sfa->actions_len = 0;
 402         return sfa;
 403 }
 404
 405 struct sw_flow *ovs_flow_alloc(void)
 406 {
 407         struct sw_flow *flow;
 408
 409         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 410         if (!flow)
 411                 return ERR_PTR(-ENOMEM);
 412
 413         spin_lock_init(&flow->lock);
 414         flow->sf_acts = NULL;
 415         flow->mask = NULL;
 416
 417         return flow;
 418 }
 419
 420 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 421 {
 422         hash = jhash_1word(hash, table->hash_seed);
 423         return flex_array_get(table->buckets,
 424                                 (hash & (table->n_buckets - 1)));
 425 }
 426
 427 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 428 {
 429         struct flex_array *buckets;
 430         int i, err;
 431
 432         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 433                                    n_buckets, GFP_KERNEL);
 434         if (!buckets)
 435                 return NULL;
 436
 437         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 438         if (err) {
 439                 flex_array_free(buckets);
 440                 return NULL;
 441         }
 442
 443         for (i = 0; i < n_buckets; i++)
 444                 INIT_HLIST_HEAD((struct hlist_head *)
 445                                         flex_array_get(buckets, i));
 446
 447         return buckets;
 448 }
 449
 450 static void free_buckets(struct flex_array *buckets)
 451 {
 452         flex_array_free(buckets);
 453 }
 454
 455 static struct flow_table *__flow_tbl_alloc(int new_size)
 456 {
 457         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 458
 459         if (!table)
 460                 return NULL;
 461
 462         table->buckets = alloc_buckets(new_size);
 463
 464         if (!table->buckets) {
 465                 kfree(table);
 466                 return NULL;
 467         }
 468         table->n_buckets = new_size;
 469         table->count = 0;
 470         table->node_ver = 0;
 471         table->keep_flows = false;
 472         get_random_bytes(&table->hash_seed, sizeof(u32));
 473         table->mask_list = NULL;
 474
 475         return table;
 476 }
 477
 478 static void __flow_tbl_destroy(struct flow_table *table)
 479 {
 480         int i;
 481
 482         if (table->keep_flows)
 483                 goto skip_flows;
 484
 485         for (i = 0; i < table->n_buckets; i++) {
 486                 struct sw_flow *flow;
 487                 struct hlist_head *head = flex_array_get(table->buckets, i);
 488                 struct hlist_node *n;
 489                 int ver = table->node_ver;
 490
 491                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 492                         hlist_del_rcu(&flow->hash_node[ver]);
 493                         ovs_flow_free(flow, false);
 494                 }
 495         }
 496
 497         BUG_ON(!list_empty(table->mask_list));
 498         kfree(table->mask_list);
 499
 500 skip_flows:
 501         free_buckets(table->buckets);
 502         kfree(table);
 503 }
 504
 505 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 506 {
 507         struct flow_table *table = __flow_tbl_alloc(new_size);
 508
 509         if (!table)
 510                 return NULL;
 511
 512         table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
 513         if (!table->mask_list) {
 514                 table->keep_flows = true;
 515                 __flow_tbl_destroy(table);
 516                 return NULL;
 517         }
 518         INIT_LIST_HEAD(table->mask_list);
 519
 520         return table;
 521 }
 522
 523 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 524 {
 525         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 526
 527         __flow_tbl_destroy(table);
 528 }
 529
 530 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
 531 {
 532         if (!table)
 533                 return;
 534
 535         if (deferred)
 536                 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 537         else
 538                 __flow_tbl_destroy(table);
 539 }
 540
 541 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
 542 {
 543         struct sw_flow *flow;
 544         struct hlist_head *head;
 545         int ver;
 546         int i;
 547
 548         ver = table->node_ver;
 549         while (*bucket < table->n_buckets) {
 550                 i = 0;
 551                 head = flex_array_get(table->buckets, *bucket);
 552                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 553                         if (i < *last) {
 554                                 i++;
 555                                 continue;
 556                         }
 557                         *last = i + 1;
 558                         return flow;
 559                 }
 560                 (*bucket)++;
 561                 *last = 0;
 562         }
 563
 564         return NULL;
 565 }
 566
 567 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
 568 {
 569         struct hlist_head *head;
 570
 571         head = find_bucket(table, flow->hash);
 572         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 573
 574         table->count++;
 575 }
 576
 577 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 578 {
 579         int old_ver;
 580         int i;
 581
 582         old_ver = old->node_ver;
 583         new->node_ver = !old_ver;
 584
 585         /* Insert in new table. */
 586         for (i = 0; i < old->n_buckets; i++) {
 587                 struct sw_flow *flow;
 588                 struct hlist_head *head;
 589
 590                 head = flex_array_get(old->buckets, i);
 591
 592                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 593                         __tbl_insert(new, flow);
 594         }
 595
 596         new->mask_list = old->mask_list;
 597         old->keep_flows = true;
 598 }
 599
 600 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 601 {
 602         struct flow_table *new_table;
 603
 604         new_table = __flow_tbl_alloc(n_buckets);
 605         if (!new_table)
 606                 return ERR_PTR(-ENOMEM);
 607
 608         flow_table_copy_flows(table, new_table);
 609
 610         return new_table;
 611 }
 612
 613 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 614 {
 615         return __flow_tbl_rehash(table, table->n_buckets);
 616 }
 617
 618 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 619 {
 620         return __flow_tbl_rehash(table, table->n_buckets * 2);
 621 }
 622
 623 static void __flow_free(struct sw_flow *flow)
 624 {
 625         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 626         kmem_cache_free(flow_cache, flow);
 627 }
 628
 629 static void rcu_free_flow_callback(struct rcu_head *rcu)
 630 {
 631         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 632
 633         __flow_free(flow);
 634 }
 635
 636 void ovs_flow_free(struct sw_flow *flow, bool deferred)
 637 {
 638         if (!flow)
 639                 return;
 640
 641         ovs_sw_flow_mask_del_ref(flow->mask, deferred);
 642
 643         if (deferred)
 644                 call_rcu(&flow->rcu, rcu_free_flow_callback);
 645         else
 646                 __flow_free(flow);
 647 }
 648
 649 /* RCU callback used by ovs_flow_deferred_free_acts. */
 650 static void rcu_free_acts_callback(struct rcu_head *rcu)
 651 {
 652         struct sw_flow_actions *sf_acts = container_of(rcu,
 653                         struct sw_flow_actions, rcu);
 654         kfree(sf_acts);
 655 }
 656
 657 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 658  * The caller must hold rcu_read_lock for this to be sensible. */
 659 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 660 {
 661         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 662 }
 663
 664 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 665 {
 666         struct qtag_prefix {
 667                 __be16 eth_type; /* ETH_P_8021Q */
 668                 __be16 tci;
 669         };
 670         struct qtag_prefix *qp;
 671
 672         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 673                 return 0;
 674
 675         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 676                                          sizeof(__be16))))
 677                 return -ENOMEM;
 678
 679         qp = (struct qtag_prefix *) skb->data;
 680         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 681         __skb_pull(skb, sizeof(struct qtag_prefix));
 682
 683         return 0;
 684 }
 685
 686 static __be16 parse_ethertype(struct sk_buff *skb)
 687 {
 688         struct llc_snap_hdr {
 689                 u8  dsap;  /* Always 0xAA */
 690                 u8  ssap;  /* Always 0xAA */
 691                 u8  ctrl;
 692                 u8  oui[3];
 693                 __be16 ethertype;
 694         };
 695         struct llc_snap_hdr *llc;
 696         __be16 proto;
 697
 698         proto = *(__be16 *) skb->data;
 699         __skb_pull(skb, sizeof(__be16));
 700
 701         if (ntohs(proto) >= ETH_P_802_3_MIN)
 702                 return proto;
 703
 704         if (skb->len < sizeof(struct llc_snap_hdr))
 705                 return htons(ETH_P_802_2);
 706
 707         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 708                 return htons(0);
 709
 710         llc = (struct llc_snap_hdr *) skb->data;
 711         if (llc->dsap != LLC_SAP_SNAP ||
 712             llc->ssap != LLC_SAP_SNAP ||
 713             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 714                 return htons(ETH_P_802_2);
 715
 716         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 717
 718         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 719                 return llc->ethertype;
 720
 721         return htons(ETH_P_802_2);
 722 }
 723
 724 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 725                         int nh_len)
 726 {
 727         struct icmp6hdr *icmp = icmp6_hdr(skb);
 728
 729         /* The ICMPv6 type and code fields use the 16-bit transport port
 730          * fields, so we need to store them in 16-bit network byte order.
 731          */
 732         key->ipv6.tp.src = htons(icmp->icmp6_type);
 733         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 734
 735         if (icmp->icmp6_code == 0 &&
 736             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 737              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 738                 int icmp_len = skb->len - skb_transport_offset(skb);
 739                 struct nd_msg *nd;
 740                 int offset;
 741
 742                 /* In order to process neighbor discovery options, we need the
 743                  * entire packet.
 744                  */
 745                 if (unlikely(icmp_len < sizeof(*nd)))
 746                         return 0;
 747
 748                 if (unlikely(skb_linearize(skb)))
 749                         return -ENOMEM;
 750
 751                 nd = (struct nd_msg *)skb_transport_header(skb);
 752                 key->ipv6.nd.target = nd->target;
 753
 754                 icmp_len -= sizeof(*nd);
 755                 offset = 0;
 756                 while (icmp_len >= 8) {
 757                         struct nd_opt_hdr *nd_opt =
 758                                  (struct nd_opt_hdr *)(nd->opt + offset);
 759                         int opt_len = nd_opt->nd_opt_len * 8;
 760
 761                         if (unlikely(!opt_len || opt_len > icmp_len))
 762                                 return 0;
 763
 764                         /* Store the link layer address if the appropriate
 765                          * option is provided.  It is considered an error if
 766                          * the same link layer option is specified twice.
 767                          */
 768                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 769                             && opt_len == 8) {
 770                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 771                                         goto invalid;
 772                                 memcpy(key->ipv6.nd.sll,
 773                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 774                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 775                                    && opt_len == 8) {
 776                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 777                                         goto invalid;
 778                                 memcpy(key->ipv6.nd.tll,
 779                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 780                         }
 781
 782                         icmp_len -= opt_len;
 783                         offset += opt_len;
 784                 }
 785         }
 786
 787         return 0;
 788
 789 invalid:
 790         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 791         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 792         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 793
 794         return 0;
 795 }
 796
 797 /**
 798  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 799  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 800  * Ethernet header
 801  * @in_port: port number on which @skb was received.
 802  * @key: output flow key
 803  * @key_lenp: length of output flow key
 804  *
 805  * The caller must ensure that skb->len >= ETH_HLEN.
 806  *
 807  * Returns 0 if successful, otherwise a negative errno value.
 808  *
 809  * Initializes @skb header pointers as follows:
 810  *
 811  *    - skb->mac_header: the Ethernet header.
 812  *
 813  *    - skb->network_header: just past the Ethernet header, or just past the
 814  *      VLAN header, to the first byte of the Ethernet payload.
 815  *
 816  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 817  *      on output, then just past the IP header, if one is present and
 818  *      of a correct length, otherwise the same as skb->network_header.
 819  *      For other key->eth.type values it is left untouched.
 820  */
 821 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
 822 {
 823         int error;
 824         struct ethhdr *eth;
 825
 826         memset(key, 0, sizeof(*key));
 827
 828         key->phy.priority = skb->priority;
 829         if (OVS_CB(skb)->tun_key)
 830                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 831         key->phy.in_port = in_port;
 832         key->phy.skb_mark = skb_get_mark(skb);
 833
 834         skb_reset_mac_header(skb);
 835
 836         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 837          * header in the linear data area.
 838          */
 839         eth = eth_hdr(skb);
 840         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 841         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 842
 843         __skb_pull(skb, 2 * ETH_ALEN);
 844         /* We are going to push all headers that we pull, so no need to
 845          * update skb->csum here. */
 846
 847         if (vlan_tx_tag_present(skb))
 848                 key->eth.tci = htons(vlan_get_tci(skb));
 849         else if (eth->h_proto == htons(ETH_P_8021Q))
 850                 if (unlikely(parse_vlan(skb, key)))
 851                         return -ENOMEM;
 852
 853         key->eth.type = parse_ethertype(skb);
 854         if (unlikely(key->eth.type == htons(0)))
 855                 return -ENOMEM;
 856
 857         skb_reset_network_header(skb);
 858         __skb_push(skb, skb->data - skb_mac_header(skb));
 859
 860         /* Network layer. */
 861         if (key->eth.type == htons(ETH_P_IP)) {
 862                 struct iphdr *nh;
 863                 __be16 offset;
 864
 865                 error = check_iphdr(skb);
 866                 if (unlikely(error)) {
 867                         if (error == -EINVAL) {
 868                                 skb->transport_header = skb->network_header;
 869                                 error = 0;
 870                         }
 871                         return error;
 872                 }
 873
 874                 nh = ip_hdr(skb);
 875                 key->ipv4.addr.src = nh->saddr;
 876                 key->ipv4.addr.dst = nh->daddr;
 877
 878                 key->ip.proto = nh->protocol;
 879                 key->ip.tos = nh->tos;
 880                 key->ip.ttl = nh->ttl;
 881
 882                 offset = nh->frag_off & htons(IP_OFFSET);
 883                 if (offset) {
 884                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 885                         return 0;
 886                 }
 887                 if (nh->frag_off & htons(IP_MF) ||
 888                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 889                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 890
 891                 /* Transport layer. */
 892                 if (key->ip.proto == IPPROTO_TCP) {
 893                         if (tcphdr_ok(skb)) {
 894                                 struct tcphdr *tcp = tcp_hdr(skb);
 895                                 key->ipv4.tp.src = tcp->source;
 896                                 key->ipv4.tp.dst = tcp->dest;
 897                         }
 898                 } else if (key->ip.proto == IPPROTO_UDP) {
 899                         if (udphdr_ok(skb)) {
 900                                 struct udphdr *udp = udp_hdr(skb);
 901                                 key->ipv4.tp.src = udp->source;
 902                                 key->ipv4.tp.dst = udp->dest;
 903                         }
 904                 } else if (key->ip.proto == IPPROTO_ICMP) {
 905                         if (icmphdr_ok(skb)) {
 906                                 struct icmphdr *icmp = icmp_hdr(skb);
 907                                 /* The ICMP type and code fields use the 16-bit
 908                                  * transport port fields, so we need to store
 909                                  * them in 16-bit network byte order. */
 910                                 key->ipv4.tp.src = htons(icmp->type);
 911                                 key->ipv4.tp.dst = htons(icmp->code);
 912                         }
 913                 }
 914
 915         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 916                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 917                 struct arp_eth_header *arp;
 918
 919                 arp = (struct arp_eth_header *)skb_network_header(skb);
 920
 921                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 922                                 && arp->ar_pro == htons(ETH_P_IP)
 923                                 && arp->ar_hln == ETH_ALEN
 924                                 && arp->ar_pln == 4) {
 925
 926                         /* We only match on the lower 8 bits of the opcode. */
 927                         if (ntohs(arp->ar_op) <= 0xff)
 928                                 key->ip.proto = ntohs(arp->ar_op);
 929                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 930                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 931                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 932                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 933                 }
 934         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 935                 int nh_len;             /* IPv6 Header + Extensions */
 936
 937                 nh_len = parse_ipv6hdr(skb, key);
 938                 if (unlikely(nh_len < 0)) {
 939                         if (nh_len == -EINVAL) {
 940                                 skb->transport_header = skb->network_header;
 941                                 error = 0;
 942                         } else {
 943                                 error = nh_len;
 944                         }
 945                         return error;
 946                 }
 947
 948                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 949                         return 0;
 950                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 951                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 952
 953                 /* Transport layer. */
 954                 if (key->ip.proto == NEXTHDR_TCP) {
 955                         if (tcphdr_ok(skb)) {
 956                                 struct tcphdr *tcp = tcp_hdr(skb);
 957                                 key->ipv6.tp.src = tcp->source;
 958                                 key->ipv6.tp.dst = tcp->dest;
 959                         }
 960                 } else if (key->ip.proto == NEXTHDR_UDP) {
 961                         if (udphdr_ok(skb)) {
 962                                 struct udphdr *udp = udp_hdr(skb);
 963                                 key->ipv6.tp.src = udp->source;
 964                                 key->ipv6.tp.dst = udp->dest;
 965                         }
 966                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 967                         if (icmp6hdr_ok(skb)) {
 968                                 error = parse_icmpv6(skb, key, nh_len);
 969                                 if (error)
 970                                         return error;
 971                         }
 972                 }
 973         }
 974
 975         return 0;
 976 }
 977
 978 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 979 {
 980         return jhash2((u32 *)((u8 *)key + key_start),
 981                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 982 }
 983
 984 static int flow_key_start(const struct sw_flow_key *key)
 985 {
 986         if (key->tun_key.ipv4_dst)
 987                 return 0;
 988         else
 989                 return offsetof(struct sw_flow_key, phy);
 990 }
 991
 992 static bool __cmp_key(const struct sw_flow_key *key1,
 993                 const struct sw_flow_key *key2,  int key_start, int key_len)
 994 {
 995         return !memcmp((u8 *)key1 + key_start,
 996                         (u8 *)key2 + key_start, (key_len - key_start));
 997 }
 998
 999 static bool __flow_cmp_key(const struct sw_flow *flow,
1000                 const struct sw_flow_key *key, int key_start, int key_len)
1001 {
1002         return __cmp_key(&flow->key, key, key_start, key_len);
1003 }
1004
1005 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1006                   const struct sw_flow_key *key, int key_start, int key_len)
1007 {
1008         return __cmp_key(&flow->unmasked_key, key, key_start, key_len);
1009 }
1010
1011 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1012                 const struct sw_flow_key *key, int key_len)
1013 {
1014         int key_start;
1015         key_start = flow_key_start(key);
1016
1017         return __flow_cmp_unmasked_key(flow, key, key_start, key_len);
1018
1019 }
1020
1021 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1022                                        struct sw_flow_match *match)
1023 {
1024         struct sw_flow_key *unmasked = match->key;
1025         int key_len = match->range.end;
1026         struct sw_flow *flow;
1027
1028         flow = ovs_flow_lookup(table, unmasked);
1029         if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_len)))
1030                 flow = NULL;
1031
1032         return flow;
1033 }
1034
1035 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1036                                     const struct sw_flow_key *flow_key,
1037                                     struct sw_flow_mask *mask)
1038 {
1039         struct sw_flow *flow;
1040         struct hlist_head *head;
1041         int key_start = mask->range.start;
1042         int key_len = mask->range.end;
1043         u32 hash;
1044         struct sw_flow_key masked_key;
1045
1046         ovs_flow_key_mask(&masked_key, flow_key, mask);
1047         hash = ovs_flow_hash(&masked_key, key_start, key_len);
1048         head = find_bucket(table, hash);
1049         hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1050                 if (flow->mask == mask &&
1051                     __flow_cmp_key(flow, &masked_key, key_start, key_len))
1052                         return flow;
1053         }
1054         return NULL;
1055 }
1056
1057 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1058                                 const struct sw_flow_key *key)
1059 {
1060         struct sw_flow *flow = NULL;
1061         struct sw_flow_mask *mask;
1062
1063         list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1064                 flow = ovs_masked_flow_lookup(tbl, key, mask);
1065                 if (flow)  /* Found */
1066                         break;
1067         }
1068
1069         return flow;
1070 }
1071
1072
1073 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1074 {
1075         flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1076                         flow->mask->range.end);
1077         __tbl_insert(table, flow);
1078 }
1079
1080 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1081 {
1082         BUG_ON(table->count == 0);
1083         hlist_del_rcu(&flow->hash_node[table->node_ver]);
1084         table->count--;
1085 }
1086
1087 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
1088 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1089         [OVS_KEY_ATTR_ENCAP] = -1,
1090         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1091         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1092         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1093         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1094         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1095         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1096         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1097         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1098         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1099         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1100         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1101         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1102         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1103         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1104         [OVS_KEY_ATTR_TUNNEL] = -1,
1105 };
1106
1107 static bool is_all_zero(const u8 *fp, size_t size)
1108 {
1109         int i;
1110
1111         if (!fp)
1112                 return false;
1113
1114         for (i = 0; i < size; i++)
1115                 if (fp[i])
1116                         return false;
1117
1118         return true;
1119 }
1120
1121 static int __parse_flow_nlattrs(const struct nlattr *attr,
1122                               const struct nlattr *a[],
1123                               u64 *attrsp, bool nz)
1124 {
1125         const struct nlattr *nla;
1126         u64 attrs;
1127         int rem;
1128
1129         attrs = *attrsp;
1130         nla_for_each_nested(nla, attr, rem) {
1131                 u16 type = nla_type(nla);
1132                 int expected_len;
1133
1134                 if (type > OVS_KEY_ATTR_MAX) {
1135                         OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1136                                   type, OVS_KEY_ATTR_MAX);
1137                 }
1138
1139                 if (attrs & (1ULL << type)) {
1140                         OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1141                         return -EINVAL;
1142                 }
1143
1144                 expected_len = ovs_key_lens[type];
1145                 if (nla_len(nla) != expected_len && expected_len != -1) {
1146                         OVS_NLERR("Key attribute has unexpected length (type=%d"
1147                                   ", length=%d, expected=%d).\n", type,
1148                                   nla_len(nla), expected_len);
1149                         return -EINVAL;
1150                 }
1151
1152                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1153                         attrs |= 1ULL << type;
1154                         a[type] = nla;
1155                 }
1156         }
1157         if (rem) {
1158                 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1159                 return -EINVAL;
1160         }
1161
1162         *attrsp = attrs;
1163         return 0;
1164 }
1165
1166 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1167                               const struct nlattr *a[], u64 *attrsp)
1168 {
1169         return __parse_flow_nlattrs(attr, a, attrsp, true);
1170 }
1171
1172 static int parse_flow_nlattrs(const struct nlattr *attr,
1173                               const struct nlattr *a[], u64 *attrsp)
1174 {
1175         return __parse_flow_nlattrs(attr, a, attrsp, false);
1176 }
1177
1178 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1179                          struct sw_flow_match *match, bool is_mask)
1180 {
1181         struct nlattr *a;
1182         int rem;
1183         bool ttl = false;
1184         __be16 tun_flags = 0;
1185
1186         nla_for_each_nested(a, attr, rem) {
1187                 int type = nla_type(a);
1188                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1189                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1190                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1191                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1192                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1193                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1194                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1195                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1196                 };
1197
1198                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1199                         OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1200                         type, OVS_TUNNEL_KEY_ATTR_MAX);
1201                         return -EINVAL;
1202                 }
1203
1204                 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1205                         OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1206                                   " legnth (type=%d, length=%d, expected=%d).\n",
1207                                   type, nla_len(a), ovs_tunnel_key_lens[type]);
1208                         return -EINVAL;
1209                 }
1210
1211                 switch (type) {
1212                 case OVS_TUNNEL_KEY_ATTR_ID:
1213                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1214                                         nla_get_be64(a), is_mask);
1215                         tun_flags |= TUNNEL_KEY;
1216                         break;
1217                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1218                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1219                                         nla_get_be32(a), is_mask);
1220                         break;
1221                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1222                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1223                                         nla_get_be32(a), is_mask);
1224                         break;
1225                 case OVS_TUNNEL_KEY_ATTR_TOS:
1226                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1227                                         nla_get_u8(a), is_mask);
1228                         break;
1229                 case OVS_TUNNEL_KEY_ATTR_TTL:
1230                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1231                                         nla_get_u8(a), is_mask);
1232                         ttl = true;
1233                         break;
1234                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1235                         tun_flags |= TUNNEL_DONT_FRAGMENT;
1236                         break;
1237                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1238                         tun_flags |= TUNNEL_CSUM;
1239                         break;
1240                 default:
1241                         return -EINVAL;
1242                 }
1243         }
1244
1245         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1246
1247         if (rem > 0) {
1248                 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1249                 return -EINVAL;
1250         }
1251
1252         if (!match->key->tun_key.ipv4_dst) {
1253                 OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1254                 return -EINVAL;
1255         }
1256
1257         if (!ttl) {
1258                 OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1259                 return -EINVAL;
1260         }
1261
1262         return 0;
1263 }
1264
1265 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1266                         const struct ovs_key_ipv4_tunnel *tun_key,
1267                         const struct ovs_key_ipv4_tunnel *output)
1268 {
1269         struct nlattr *nla;
1270
1271         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1272         if (!nla)
1273                 return -EMSGSIZE;
1274
1275         if (output->tun_flags & TUNNEL_KEY &&
1276             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1277                 return -EMSGSIZE;
1278         if (output->ipv4_src &&
1279                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1280                 return -EMSGSIZE;
1281         if (output->ipv4_dst &&
1282                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1283                 return -EMSGSIZE;
1284         if (output->ipv4_tos &&
1285                 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1286                 return -EMSGSIZE;
1287         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1288                 return -EMSGSIZE;
1289         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1290                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1291                 return -EMSGSIZE;
1292         if ((output->tun_flags & TUNNEL_CSUM) &&
1293                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1294                 return -EMSGSIZE;
1295
1296         nla_nest_end(skb, nla);
1297         return 0;
1298 }
1299
1300
1301 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
1302                 const struct nlattr **a, bool is_mask)
1303 {
1304         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1305                 SW_FLOW_KEY_PUT(match, phy.priority,
1306                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1307                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1308         }
1309
1310         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1311                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1312
1313                 if (!is_mask && in_port >= DP_MAX_PORTS)
1314                         return -EINVAL;
1315                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1316                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1317         } else if (!is_mask) {
1318                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1319         }
1320
1321         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1322                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1323 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1324                 if (!is_mask && mark != 0) {
1325                         OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1326                         return -EINVAL;
1327                 }
1328 #endif
1329                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1330                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1331         }
1332         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1333                 if (ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1334                                         is_mask))
1335                         return -EINVAL;
1336                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1337         }
1338         return 0;
1339 }
1340
1341 static int ovs_key_from_nlattrs(struct sw_flow_match *match,  u64 attrs,
1342                 const struct nlattr **a, bool is_mask)
1343 {
1344         int err;
1345         u64 orig_attrs = attrs;
1346
1347         err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1348         if (err)
1349                 return err;
1350
1351         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1352                 const struct ovs_key_ethernet *eth_key;
1353
1354                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1355                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1356                                 eth_key->eth_src, ETH_ALEN, is_mask);
1357                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1358                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1359                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1360         }
1361
1362         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1363                 __be16 tci;
1364
1365                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1366                 if (!is_mask)
1367                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1368                                 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1369                                 return -EINVAL;
1370                         }
1371
1372                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1373                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1374         }
1375
1376         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1377                 __be16 eth_type;
1378
1379                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1380                 if (!is_mask && ntohs(eth_type) < ETH_P_802_3_MIN) {
1381                         OVS_NLERR("EtherType is less than mimimum (type=%x, min=%x).\n",
1382                                         ntohs(eth_type), ETH_P_802_3_MIN);
1383                         return -EINVAL;
1384                 }
1385
1386                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1387                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1388         } else if (!is_mask) {
1389                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1390         }
1391
1392         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1393                 const struct ovs_key_ipv4 *ipv4_key;
1394
1395                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1396                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1397                         OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1398                                 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1399                         return -EINVAL;
1400                 }
1401                 SW_FLOW_KEY_PUT(match, ip.proto,
1402                                 ipv4_key->ipv4_proto, is_mask);
1403                 SW_FLOW_KEY_PUT(match, ip.tos,
1404                                 ipv4_key->ipv4_tos, is_mask);
1405                 SW_FLOW_KEY_PUT(match, ip.ttl,
1406                                 ipv4_key->ipv4_ttl, is_mask);
1407                 SW_FLOW_KEY_PUT(match, ip.frag,
1408                                 ipv4_key->ipv4_frag, is_mask);
1409                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1410                                 ipv4_key->ipv4_src, is_mask);
1411                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1412                                 ipv4_key->ipv4_dst, is_mask);
1413                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1414         }
1415
1416         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1417                 const struct ovs_key_ipv6 *ipv6_key;
1418
1419                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1420                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1421                         OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1422                                 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1423                         return -EINVAL;
1424                 }
1425                 SW_FLOW_KEY_PUT(match, ipv6.label,
1426                                 ipv6_key->ipv6_label, is_mask);
1427                 SW_FLOW_KEY_PUT(match, ip.proto,
1428                                 ipv6_key->ipv6_proto, is_mask);
1429                 SW_FLOW_KEY_PUT(match, ip.tos,
1430                                 ipv6_key->ipv6_tclass, is_mask);
1431                 SW_FLOW_KEY_PUT(match, ip.ttl,
1432                                 ipv6_key->ipv6_hlimit, is_mask);
1433                 SW_FLOW_KEY_PUT(match, ip.frag,
1434                                 ipv6_key->ipv6_frag, is_mask);
1435                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1436                                 ipv6_key->ipv6_src,
1437                                 sizeof(match->key->ipv6.addr.src),
1438                                 is_mask);
1439                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1440                                 ipv6_key->ipv6_dst,
1441                                 sizeof(match->key->ipv6.addr.dst),
1442                                 is_mask);
1443
1444                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1445         }
1446
1447         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1448                 const struct ovs_key_arp *arp_key;
1449
1450                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1451                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1452                         OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1453                                   arp_key->arp_op);
1454                         return -EINVAL;
1455                 }
1456
1457                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1458                                 arp_key->arp_sip, is_mask);
1459                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1460                         arp_key->arp_tip, is_mask);
1461                 SW_FLOW_KEY_PUT(match, ip.proto,
1462                                 ntohs(arp_key->arp_op), is_mask);
1463                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1464                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1465                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1466                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1467
1468                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1469         }
1470
1471         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1472                 const struct ovs_key_tcp *tcp_key;
1473
1474                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1475                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1476                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1477                                         tcp_key->tcp_src, is_mask);
1478                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1479                                         tcp_key->tcp_dst, is_mask);
1480                 } else {
1481                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1482                                         tcp_key->tcp_src, is_mask);
1483                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1484                                         tcp_key->tcp_dst, is_mask);
1485                 }
1486                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1487         }
1488
1489         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1490                 const struct ovs_key_udp *udp_key;
1491
1492                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1493                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1494                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1495                                         udp_key->udp_src, is_mask);
1496                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1497                                         udp_key->udp_dst, is_mask);
1498                 } else {
1499                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1500                                         udp_key->udp_src, is_mask);
1501                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1502                                         udp_key->udp_dst, is_mask);
1503                 }
1504                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1505         }
1506
1507         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1508                 const struct ovs_key_icmp *icmp_key;
1509
1510                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1511                 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1512                                 htons(icmp_key->icmp_type), is_mask);
1513                 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1514                                 htons(icmp_key->icmp_code), is_mask);
1515                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1516         }
1517
1518         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1519                 const struct ovs_key_icmpv6 *icmpv6_key;
1520
1521                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1522                 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1523                                 htons(icmpv6_key->icmpv6_type), is_mask);
1524                 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1525                                 htons(icmpv6_key->icmpv6_code), is_mask);
1526                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1527         }
1528
1529         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1530                 const struct ovs_key_nd *nd_key;
1531
1532                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1533                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1534                         nd_key->nd_target,
1535                         sizeof(match->key->ipv6.nd.target),
1536                         is_mask);
1537                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1538                         nd_key->nd_sll, ETH_ALEN, is_mask);
1539                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1540                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1541                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1542         }
1543
1544         if (attrs != 0)
1545                 return -EINVAL;
1546
1547         return 0;
1548 }
1549
1550 /**
1551  * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1552  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1553  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1554  * does not include any don't care bit.
1555  * @match: receives the extracted flow match information.
1556  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1557  * sequence. The fields should of the packet that triggered the creation
1558  * of this flow.
1559  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1560  * attribute specifies the mask field of the wildcarded flow.
1561  */
1562 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1563                            const struct nlattr *key,
1564                            const struct nlattr *mask)
1565 {
1566         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1567         const struct nlattr *encap;
1568         u64 key_attrs = 0;
1569         u64 mask_attrs = 0;
1570         bool encap_valid = false;
1571         int err;
1572
1573         err = parse_flow_nlattrs(key, a, &key_attrs);
1574         if (err)
1575                 return err;
1576
1577         if (key_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) {
1578                 encap = a[OVS_KEY_ATTR_ENCAP];
1579                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1580                 if (nla_len(encap)) {
1581                         __be16 eth_type = 0; /* ETH_P_8021Q */
1582
1583                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1584                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1585
1586                         if  ((eth_type == htons(ETH_P_8021Q)) && (a[OVS_KEY_ATTR_VLAN])) {
1587                                 encap_valid = true;
1588                                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1589                                 err = parse_flow_nlattrs(encap, a, &key_attrs);
1590                         } else {
1591                                 OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1592                                 err = -EINVAL;
1593                         }
1594
1595                         if (err)
1596                                 return err;
1597                 }
1598         }
1599
1600         err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1601         if (err)
1602                 return err;
1603
1604         if (mask) {
1605                 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1606                 if (err)
1607                         return err;
1608
1609                 if ((mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP) && encap_valid) {
1610                         __be16 eth_type = 0;
1611
1612                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1613                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1614                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1615                         if (eth_type == htons(0xffff)) {
1616                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1617                                 encap = a[OVS_KEY_ATTR_ENCAP];
1618                                 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1619                         } else {
1620                                 OVS_NLERR("VLAN frames must have an exact match"
1621                                          " on the TPID (mask=%x).\n",
1622                                          ntohs(eth_type));
1623                                 err = -EINVAL;
1624                         }
1625
1626                         if (err)
1627                                 return err;
1628                 }
1629
1630                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1631                 if (err)
1632                         return err;
1633         } else {
1634                 /* Populate exact match flow's key mask. */
1635                 if (match->mask)
1636                         ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1637         }
1638
1639         if (!ovs_match_validate(match, key_attrs, mask_attrs))
1640                 return -EINVAL;
1641
1642         return 0;
1643 }
1644
1645 /**
1646  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1647  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1648  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1649  * sequence.
1650  *
1651  * This parses a series of Netlink attributes that form a flow key, which must
1652  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1653  * get the metadata, that is, the parts of the flow key that cannot be
1654  * extracted from the packet itself.
1655  */
1656
1657 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1658                 const struct nlattr *attr)
1659 {
1660         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1661         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1662         u64 attrs = 0;
1663         int err;
1664         struct sw_flow_match match;
1665
1666         flow->key.phy.in_port = DP_MAX_PORTS;
1667         flow->key.phy.priority = 0;
1668         flow->key.phy.skb_mark = 0;
1669         memset(tun_key, 0, sizeof(flow->key.tun_key));
1670
1671         err = parse_flow_nlattrs(attr, a, &attrs);
1672         if (err)
1673                 return -EINVAL;
1674
1675         memset(&match, 0, sizeof(match));
1676         match.key = &flow->key;
1677
1678         err = metadata_from_nlattrs(&match, &attrs, a, false);
1679         if (err)
1680                 return err;
1681
1682         return 0;
1683 }
1684
1685 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1686                 const struct sw_flow_key *output, struct sk_buff *skb)
1687 {
1688         struct ovs_key_ethernet *eth_key;
1689         struct nlattr *nla, *encap;
1690
1691         if (output->phy.priority &&
1692                 nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1693                 goto nla_put_failure;
1694
1695         if (swkey->tun_key.ipv4_dst &&
1696             ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1697                 goto nla_put_failure;
1698
1699         if (swkey->phy.in_port == DP_MAX_PORTS) {
1700                 if ((swkey != output) && (output->phy.in_port == 0xffff))
1701                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1702                                 goto nla_put_failure;
1703         } else {
1704                 u16 upper_u16;
1705                 upper_u16 = (swkey == output) ? 0 : 0xffff;
1706
1707                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1708                                 (upper_u16 << 16) | output->phy.in_port))
1709                         goto nla_put_failure;
1710         }
1711
1712         if (output->phy.skb_mark &&
1713                 nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1714                 goto nla_put_failure;
1715
1716         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1717         if (!nla)
1718                 goto nla_put_failure;
1719
1720         eth_key = nla_data(nla);
1721         memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1722         memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1723
1724         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1725                 __be16 eth_type;
1726                 eth_type = (swkey == output) ? htons(ETH_P_8021Q) : htons(0xffff) ;
1727                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1728                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1729                         goto nla_put_failure;
1730                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1731                 if (!swkey->eth.tci)
1732                         goto unencap;
1733         } else
1734                 encap = NULL;
1735
1736         if (swkey->eth.type == htons(ETH_P_802_2)) {
1737                 /*
1738                  * Ethertype 802.2 is represented in the netlink with omitted
1739                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1740                  * 0xffff in the mask attribute.  Ethertype can also
1741                  * be wildcarded.
1742                  */
1743                 if (swkey != output && output->eth.type)
1744                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1745                                                 output->eth.type))
1746                                 goto nla_put_failure;
1747                 goto unencap;
1748         }
1749
1750         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1751                 goto nla_put_failure;
1752
1753         if (swkey->eth.type == htons(ETH_P_IP)) {
1754                 struct ovs_key_ipv4 *ipv4_key;
1755
1756                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1757                 if (!nla)
1758                         goto nla_put_failure;
1759                 ipv4_key = nla_data(nla);
1760                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1761                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1762                 ipv4_key->ipv4_proto = output->ip.proto;
1763                 ipv4_key->ipv4_tos = output->ip.tos;
1764                 ipv4_key->ipv4_ttl = output->ip.ttl;
1765                 ipv4_key->ipv4_frag = output->ip.frag;
1766         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1767                 struct ovs_key_ipv6 *ipv6_key;
1768
1769                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1770                 if (!nla)
1771                         goto nla_put_failure;
1772                 ipv6_key = nla_data(nla);
1773                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1774                                 sizeof(ipv6_key->ipv6_src));
1775                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1776                                 sizeof(ipv6_key->ipv6_dst));
1777                 ipv6_key->ipv6_label = output->ipv6.label;
1778                 ipv6_key->ipv6_proto = output->ip.proto;
1779                 ipv6_key->ipv6_tclass = output->ip.tos;
1780                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1781                 ipv6_key->ipv6_frag = output->ip.frag;
1782         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1783                    swkey->eth.type == htons(ETH_P_RARP)) {
1784                 struct ovs_key_arp *arp_key;
1785
1786                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1787                 if (!nla)
1788                         goto nla_put_failure;
1789                 arp_key = nla_data(nla);
1790                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1791                 arp_key->arp_sip = output->ipv4.addr.src;
1792                 arp_key->arp_tip = output->ipv4.addr.dst;
1793                 arp_key->arp_op = htons(output->ip.proto);
1794                 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1795                 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1796         }
1797
1798         if ((swkey->eth.type == htons(ETH_P_IP) ||
1799              swkey->eth.type == htons(ETH_P_IPV6)) &&
1800              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1801
1802                 if (swkey->ip.proto == IPPROTO_TCP) {
1803                         struct ovs_key_tcp *tcp_key;
1804
1805                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1806                         if (!nla)
1807                                 goto nla_put_failure;
1808                         tcp_key = nla_data(nla);
1809                         if (swkey->eth.type == htons(ETH_P_IP)) {
1810                                 tcp_key->tcp_src = output->ipv4.tp.src;
1811                                 tcp_key->tcp_dst = output->ipv4.tp.dst;
1812                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1813                                 tcp_key->tcp_src = output->ipv6.tp.src;
1814                                 tcp_key->tcp_dst = output->ipv6.tp.dst;
1815                         }
1816                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1817                         struct ovs_key_udp *udp_key;
1818
1819                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1820                         if (!nla)
1821                                 goto nla_put_failure;
1822                         udp_key = nla_data(nla);
1823                         if (swkey->eth.type == htons(ETH_P_IP)) {
1824                                 udp_key->udp_src = output->ipv4.tp.src;
1825                                 udp_key->udp_dst = output->ipv4.tp.dst;
1826                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1827                                 udp_key->udp_src = output->ipv6.tp.src;
1828                                 udp_key->udp_dst = output->ipv6.tp.dst;
1829                         }
1830                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1831                            swkey->ip.proto == IPPROTO_ICMP) {
1832                         struct ovs_key_icmp *icmp_key;
1833
1834                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1835                         if (!nla)
1836                                 goto nla_put_failure;
1837                         icmp_key = nla_data(nla);
1838                         icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1839                         icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1840                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1841                            swkey->ip.proto == IPPROTO_ICMPV6) {
1842                         struct ovs_key_icmpv6 *icmpv6_key;
1843
1844                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1845                                                 sizeof(*icmpv6_key));
1846                         if (!nla)
1847                                 goto nla_put_failure;
1848                         icmpv6_key = nla_data(nla);
1849                         icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1850                         icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1851
1852                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1853                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1854                                 struct ovs_key_nd *nd_key;
1855
1856                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1857                                 if (!nla)
1858                                         goto nla_put_failure;
1859                                 nd_key = nla_data(nla);
1860                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1861                                                         sizeof(nd_key->nd_target));
1862                                 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1863                                 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1864                         }
1865                 }
1866         }
1867
1868 unencap:
1869         if (encap)
1870                 nla_nest_end(skb, encap);
1871
1872         return 0;
1873
1874 nla_put_failure:
1875         return -EMSGSIZE;
1876 }
1877
1878 /* Initializes the flow module.
1879  * Returns zero if successful or a negative error code. */
1880 int ovs_flow_init(void)
1881 {
1882         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1883                                         0, NULL);
1884         if (flow_cache == NULL)
1885                 return -ENOMEM;
1886
1887         return 0;
1888 }
1889
1890 /* Uninitializes the flow module. */
1891 void ovs_flow_exit(void)
1892 {
1893         kmem_cache_destroy(flow_cache);
1894 }
1895
1896 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
1897 {
1898         struct sw_flow_mask *mask;
1899
1900         mask = kmalloc(sizeof(*mask), GFP_KERNEL);
1901         if (mask)
1902                 mask->ref_count = 0;
1903
1904         return mask;
1905 }
1906
1907 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
1908 {
1909         mask->ref_count++;
1910 }
1911
1912 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
1913 {
1914         struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
1915
1916         kfree(mask);
1917 }
1918
1919 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
1920 {
1921         if (!mask)
1922                 return;
1923
1924         BUG_ON(!mask->ref_count);
1925         mask->ref_count--;
1926
1927         if (!mask->ref_count) {
1928                 list_del_rcu(&mask->list);
1929                 if (deferred)
1930                         call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
1931                 else
1932                         kfree(mask);
1933         }
1934 }
1935
1936 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
1937                 const struct sw_flow_mask *b)
1938 {
1939         u8 *a_ = (u8 *)&a->key + a->range.start;
1940         u8 *b_ = (u8 *)&b->key + b->range.start;
1941
1942         return  (a->range.end == b->range.end)
1943                 && (a->range.start == b->range.start)
1944                 && (memcmp(a_, b_, ovs_sw_flow_mask_actual_size(a)) == 0);
1945 }
1946
1947 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
1948                                            const struct sw_flow_mask *mask)
1949 {
1950         struct list_head *ml;
1951
1952         list_for_each(ml, tbl->mask_list) {
1953                 struct sw_flow_mask *m;
1954                 m = container_of(ml, struct sw_flow_mask, list);
1955                 if (ovs_sw_flow_mask_equal(mask, m))
1956                         return m;
1957         }
1958
1959         return NULL;
1960 }
1961
1962 /**
1963  * add a new mask into the mask list.
1964  * The caller needs to make sure that 'mask' is not the same
1965  * as any masks that are already on the list.
1966  */
1967 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
1968 {
1969         list_add_rcu(&mask->list, tbl->mask_list);
1970 }
1971
1972 /**
1973  * Set 'range' fields in the mask to the value of 'val'.
1974  */
1975 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
1976                 struct sw_flow_key_range *range, u8 val)
1977 {
1978         u8 *m = (u8 *)&mask->key + range->start;
1979
1980         mask->range = *range;
1981         memset(m, val, ovs_sw_flow_mask_size_roundup(mask));
1982 }