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/sctp.h>
  38 #include <linux/tcp.h>
  39 #include <linux/udp.h>
  40 #include <linux/icmp.h>
  41 #include <linux/icmpv6.h>
  42 #include <linux/rculist.h>
  43 #include <net/ip.h>
  44 #include <net/ipv6.h>
  45 #include <net/ndisc.h>
  46
  47 #include "vlan.h"
  48
  49 static struct kmem_cache *flow_cache;
  50
  51 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
  52                 struct sw_flow_key_range *range, u8 val);
  53
  54 static void update_range__(struct sw_flow_match *match,
  55                           size_t offset, size_t size, bool is_mask)
  56 {
  57         struct sw_flow_key_range *range = NULL;
  58         size_t start = rounddown(offset, sizeof(long));
  59         size_t end = roundup(offset + size, sizeof(long));
  60
  61         if (!is_mask)
  62                 range = &match->range;
  63         else if (match->mask)
  64                 range = &match->mask->range;
  65
  66         if (!range)
  67                 return;
  68
  69         if (range->start == range->end) {
  70                 range->start = start;
  71                 range->end = end;
  72                 return;
  73         }
  74
  75         if (range->start > start)
  76                 range->start = start;
  77
  78         if (range->end < end)
  79                 range->end = end;
  80 }
  81
  82 #define SW_FLOW_KEY_PUT(match, field, value, is_mask) \
  83         do { \
  84                 update_range__(match, offsetof(struct sw_flow_key, field),  \
  85                                      sizeof((match)->key->field), is_mask); \
  86                 if (is_mask) {                                              \
  87                         if ((match)->mask)                                  \
  88                                 (match)->mask->key.field = value;           \
  89                 } else {                                                    \
  90                         (match)->key->field = value;                        \
  91                 }                                                           \
  92         } while (0)
  93
  94 #define SW_FLOW_KEY_MEMCPY(match, field, value_p, len, is_mask) \
  95         do { \
  96                 update_range__(match, offsetof(struct sw_flow_key, field),  \
  97                                 len, is_mask);                              \
  98                 if (is_mask) {                                              \
  99                         if ((match)->mask)                                  \
 100                                 memcpy(&(match)->mask->key.field, value_p, len);\
 101                 } else {                                                    \
 102                         memcpy(&(match)->key->field, value_p, len);         \
 103                 }                                                           \
 104         } while (0)
 105
 106 static u16 range_n_bytes(const struct sw_flow_key_range *range)
 107 {
 108         return range->end - range->start;
 109 }
 110
 111 void ovs_match_init(struct sw_flow_match *match,
 112                     struct sw_flow_key *key,
 113                     struct sw_flow_mask *mask)
 114 {
 115         memset(match, 0, sizeof(*match));
 116         match->key = key;
 117         match->mask = mask;
 118
 119         memset(key, 0, sizeof(*key));
 120
 121         if (mask) {
 122                 memset(&mask->key, 0, sizeof(mask->key));
 123                 mask->range.start = mask->range.end = 0;
 124         }
 125 }
 126
 127 static bool ovs_match_validate(const struct sw_flow_match *match,
 128                 u64 key_attrs, u64 mask_attrs)
 129 {
 130         u64 key_expected = 1ULL << OVS_KEY_ATTR_ETHERNET;
 131         u64 mask_allowed = key_attrs;  /* At most allow all key attributes */
 132
 133         /* The following mask attributes allowed only if they
 134          * pass the validation tests. */
 135         mask_allowed &= ~((1ULL << OVS_KEY_ATTR_IPV4)
 136                         | (1ULL << OVS_KEY_ATTR_IPV6)
 137                         | (1ULL << OVS_KEY_ATTR_TCP)
 138                         | (1ULL << OVS_KEY_ATTR_UDP)
 139                         | (1ULL << OVS_KEY_ATTR_SCTP)
 140                         | (1ULL << OVS_KEY_ATTR_ICMP)
 141                         | (1ULL << OVS_KEY_ATTR_ICMPV6)
 142                         | (1ULL << OVS_KEY_ATTR_ARP)
 143                         | (1ULL << OVS_KEY_ATTR_ND));
 144
 145         /* Always allowed mask fields. */
 146         mask_allowed |= ((1ULL << OVS_KEY_ATTR_TUNNEL)
 147                        | (1ULL << OVS_KEY_ATTR_IN_PORT)
 148                        | (1ULL << OVS_KEY_ATTR_ETHERTYPE));
 149
 150         /* Check key attributes. */
 151         if (match->key->eth.type == htons(ETH_P_ARP)
 152                         || match->key->eth.type == htons(ETH_P_RARP)) {
 153                 key_expected |= 1ULL << OVS_KEY_ATTR_ARP;
 154                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 155                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ARP;
 156         }
 157
 158         if (match->key->eth.type == htons(ETH_P_IP)) {
 159                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV4;
 160                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 161                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV4;
 162
 163                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 164                         if (match->key->ip.proto == IPPROTO_UDP) {
 165                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 166                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 167                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 168                         }
 169
 170                         if (match->key->ip.proto == IPPROTO_SCTP) {
 171                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 172                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 173                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 174                         }
 175
 176                         if (match->key->ip.proto == IPPROTO_TCP) {
 177                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 178                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 179                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 180                         }
 181
 182                         if (match->key->ip.proto == IPPROTO_ICMP) {
 183                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMP;
 184                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 185                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMP;
 186                         }
 187                 }
 188         }
 189
 190         if (match->key->eth.type == htons(ETH_P_IPV6)) {
 191                 key_expected |= 1ULL << OVS_KEY_ATTR_IPV6;
 192                 if (match->mask && (match->mask->key.eth.type == htons(0xffff)))
 193                         mask_allowed |= 1ULL << OVS_KEY_ATTR_IPV6;
 194
 195                 if (match->key->ip.frag != OVS_FRAG_TYPE_LATER) {
 196                         if (match->key->ip.proto == IPPROTO_UDP) {
 197                                 key_expected |= 1ULL << OVS_KEY_ATTR_UDP;
 198                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 199                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_UDP;
 200                         }
 201
 202                         if (match->key->ip.proto == IPPROTO_SCTP) {
 203                                 key_expected |= 1ULL << OVS_KEY_ATTR_SCTP;
 204                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 205                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_SCTP;
 206                         }
 207
 208                         if (match->key->ip.proto == IPPROTO_TCP) {
 209                                 key_expected |= 1ULL << OVS_KEY_ATTR_TCP;
 210                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 211                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_TCP;
 212                         }
 213
 214                         if (match->key->ip.proto == IPPROTO_ICMPV6) {
 215                                 key_expected |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 216                                 if (match->mask && (match->mask->key.ip.proto == 0xff))
 217                                         mask_allowed |= 1ULL << OVS_KEY_ATTR_ICMPV6;
 218
 219                                 if (match->key->ipv6.tp.src ==
 220                                                 htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 221                                     match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 222                                         key_expected |= 1ULL << OVS_KEY_ATTR_ND;
 223                                         if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
 224                                                 mask_allowed |= 1ULL << OVS_KEY_ATTR_ND;
 225                                 }
 226                         }
 227                 }
 228         }
 229
 230         if ((key_attrs & key_expected) != key_expected) {
 231                 /* Key attributes check failed. */
 232                 OVS_NLERR("Missing expected key attributes (key_attrs=%llx, expected=%llx).\n",
 233                                 key_attrs, key_expected);
 234                 return false;
 235         }
 236
 237         if ((mask_attrs & mask_allowed) != mask_attrs) {
 238                 /* Mask attributes check failed. */
 239                 OVS_NLERR("Contain more than allowed mask fields (mask_attrs=%llx, mask_allowed=%llx).\n",
 240                                 mask_attrs, mask_allowed);
 241                 return false;
 242         }
 243
 244         return true;
 245 }
 246
 247 static int check_header(struct sk_buff *skb, int len)
 248 {
 249         if (unlikely(skb->len < len))
 250                 return -EINVAL;
 251         if (unlikely(!pskb_may_pull(skb, len)))
 252                 return -ENOMEM;
 253         return 0;
 254 }
 255
 256 static bool arphdr_ok(struct sk_buff *skb)
 257 {
 258         return pskb_may_pull(skb, skb_network_offset(skb) +
 259                                   sizeof(struct arp_eth_header));
 260 }
 261
 262 static int check_iphdr(struct sk_buff *skb)
 263 {
 264         unsigned int nh_ofs = skb_network_offset(skb);
 265         unsigned int ip_len;
 266         int err;
 267
 268         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
 269         if (unlikely(err))
 270                 return err;
 271
 272         ip_len = ip_hdrlen(skb);
 273         if (unlikely(ip_len < sizeof(struct iphdr) ||
 274                      skb->len < nh_ofs + ip_len))
 275                 return -EINVAL;
 276
 277         skb_set_transport_header(skb, nh_ofs + ip_len);
 278         return 0;
 279 }
 280
 281 static bool tcphdr_ok(struct sk_buff *skb)
 282 {
 283         int th_ofs = skb_transport_offset(skb);
 284         int tcp_len;
 285
 286         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
 287                 return false;
 288
 289         tcp_len = tcp_hdrlen(skb);
 290         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
 291                      skb->len < th_ofs + tcp_len))
 292                 return false;
 293
 294         return true;
 295 }
 296
 297 static bool udphdr_ok(struct sk_buff *skb)
 298 {
 299         return pskb_may_pull(skb, skb_transport_offset(skb) +
 300                                   sizeof(struct udphdr));
 301 }
 302
 303 static bool sctphdr_ok(struct sk_buff *skb)
 304 {
 305         return pskb_may_pull(skb, skb_transport_offset(skb) +
 306                                   sizeof(struct sctphdr));
 307 }
 308
 309 static bool icmphdr_ok(struct sk_buff *skb)
 310 {
 311         return pskb_may_pull(skb, skb_transport_offset(skb) +
 312                                   sizeof(struct icmphdr));
 313 }
 314
 315 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 316 {
 317         struct timespec cur_ts;
 318         u64 cur_ms, idle_ms;
 319
 320         ktime_get_ts(&cur_ts);
 321         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 322         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 323                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 324
 325         return cur_ms - idle_ms;
 326 }
 327
 328 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key)
 329 {
 330         unsigned int nh_ofs = skb_network_offset(skb);
 331         unsigned int nh_len;
 332         int payload_ofs;
 333         struct ipv6hdr *nh;
 334         uint8_t nexthdr;
 335         __be16 frag_off;
 336         int err;
 337
 338         err = check_header(skb, nh_ofs + sizeof(*nh));
 339         if (unlikely(err))
 340                 return err;
 341
 342         nh = ipv6_hdr(skb);
 343         nexthdr = nh->nexthdr;
 344         payload_ofs = (u8 *)(nh + 1) - skb->data;
 345
 346         key->ip.proto = NEXTHDR_NONE;
 347         key->ip.tos = ipv6_get_dsfield(nh);
 348         key->ip.ttl = nh->hop_limit;
 349         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 350         key->ipv6.addr.src = nh->saddr;
 351         key->ipv6.addr.dst = nh->daddr;
 352
 353         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 354         if (unlikely(payload_ofs < 0))
 355                 return -EINVAL;
 356
 357         if (frag_off) {
 358                 if (frag_off & htons(~0x7))
 359                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 360                 else
 361                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 362         }
 363
 364         nh_len = payload_ofs - nh_ofs;
 365         skb_set_transport_header(skb, nh_ofs + nh_len);
 366         key->ip.proto = nexthdr;
 367         return nh_len;
 368 }
 369
 370 static bool icmp6hdr_ok(struct sk_buff *skb)
 371 {
 372         return pskb_may_pull(skb, skb_transport_offset(skb) +
 373                                   sizeof(struct icmp6hdr));
 374 }
 375
 376 void ovs_flow_key_mask(struct sw_flow_key *dst, const struct sw_flow_key *src,
 377                        const struct sw_flow_mask *mask)
 378 {
 379         const long *m = (long *)((u8 *)&mask->key + mask->range.start);
 380         const long *s = (long *)((u8 *)src + mask->range.start);
 381         long *d = (long *)((u8 *)dst + mask->range.start);
 382         int i;
 383
 384         /* The memory outside of the 'mask->range' are not set since
 385          * further operations on 'dst' only uses contents within
 386          * 'mask->range'.
 387          */
 388         for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
 389                 *d++ = *s++ & *m++;
 390 }
 391
 392 #define TCP_FLAGS_OFFSET 13
 393 #define TCP_FLAG_MASK 0x3f
 394
 395 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 396 {
 397         u8 tcp_flags = 0;
 398
 399         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 400              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 401             flow->key.ip.proto == IPPROTO_TCP &&
 402             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 403                 u8 *tcp = (u8 *)tcp_hdr(skb);
 404                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 405         }
 406
 407         spin_lock(&flow->lock);
 408         flow->used = jiffies;
 409         flow->packet_count++;
 410         flow->byte_count += skb->len;
 411         flow->tcp_flags |= tcp_flags;
 412         spin_unlock(&flow->lock);
 413 }
 414
 415 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
 416 {
 417         struct sw_flow_actions *sfa;
 418
 419         if (size > MAX_ACTIONS_BUFSIZE)
 420                 return ERR_PTR(-EINVAL);
 421
 422         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 423         if (!sfa)
 424                 return ERR_PTR(-ENOMEM);
 425
 426         sfa->actions_len = 0;
 427         return sfa;
 428 }
 429
 430 struct sw_flow *ovs_flow_alloc(void)
 431 {
 432         struct sw_flow *flow;
 433
 434         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 435         if (!flow)
 436                 return ERR_PTR(-ENOMEM);
 437
 438         spin_lock_init(&flow->lock);
 439         flow->sf_acts = NULL;
 440         flow->mask = NULL;
 441
 442         return flow;
 443 }
 444
 445 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 446 {
 447         hash = jhash_1word(hash, table->hash_seed);
 448         return flex_array_get(table->buckets,
 449                                 (hash & (table->n_buckets - 1)));
 450 }
 451
 452 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 453 {
 454         struct flex_array *buckets;
 455         int i, err;
 456
 457         buckets = flex_array_alloc(sizeof(struct hlist_head),
 458                                    n_buckets, GFP_KERNEL);
 459         if (!buckets)
 460                 return NULL;
 461
 462         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 463         if (err) {
 464                 flex_array_free(buckets);
 465                 return NULL;
 466         }
 467
 468         for (i = 0; i < n_buckets; i++)
 469                 INIT_HLIST_HEAD((struct hlist_head *)
 470                                         flex_array_get(buckets, i));
 471
 472         return buckets;
 473 }
 474
 475 static void free_buckets(struct flex_array *buckets)
 476 {
 477         flex_array_free(buckets);
 478 }
 479
 480 static struct flow_table *__flow_tbl_alloc(int new_size)
 481 {
 482         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 483
 484         if (!table)
 485                 return NULL;
 486
 487         table->buckets = alloc_buckets(new_size);
 488
 489         if (!table->buckets) {
 490                 kfree(table);
 491                 return NULL;
 492         }
 493         table->n_buckets = new_size;
 494         table->count = 0;
 495         table->node_ver = 0;
 496         table->keep_flows = false;
 497         get_random_bytes(&table->hash_seed, sizeof(u32));
 498         table->mask_list = NULL;
 499
 500         return table;
 501 }
 502
 503 static void __flow_tbl_destroy(struct flow_table *table)
 504 {
 505         int i;
 506
 507         if (table->keep_flows)
 508                 goto skip_flows;
 509
 510         for (i = 0; i < table->n_buckets; i++) {
 511                 struct sw_flow *flow;
 512                 struct hlist_head *head = flex_array_get(table->buckets, i);
 513                 struct hlist_node *n;
 514                 int ver = table->node_ver;
 515
 516                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 517                         hlist_del(&flow->hash_node[ver]);
 518                         ovs_flow_free(flow, false);
 519                 }
 520         }
 521
 522         BUG_ON(!list_empty(table->mask_list));
 523         kfree(table->mask_list);
 524
 525 skip_flows:
 526         free_buckets(table->buckets);
 527         kfree(table);
 528 }
 529
 530 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 531 {
 532         struct flow_table *table = __flow_tbl_alloc(new_size);
 533
 534         if (!table)
 535                 return NULL;
 536
 537         table->mask_list = kmalloc(sizeof(struct list_head), GFP_KERNEL);
 538         if (!table->mask_list) {
 539                 table->keep_flows = true;
 540                 __flow_tbl_destroy(table);
 541                 return NULL;
 542         }
 543         INIT_LIST_HEAD(table->mask_list);
 544
 545         return table;
 546 }
 547
 548 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 549 {
 550         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 551
 552         __flow_tbl_destroy(table);
 553 }
 554
 555 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
 556 {
 557         if (!table)
 558                 return;
 559
 560         if (deferred)
 561                 call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 562         else
 563                 __flow_tbl_destroy(table);
 564 }
 565
 566 struct sw_flow *ovs_flow_dump_next(struct flow_table *table, u32 *bucket, u32 *last)
 567 {
 568         struct sw_flow *flow;
 569         struct hlist_head *head;
 570         int ver;
 571         int i;
 572
 573         ver = table->node_ver;
 574         while (*bucket < table->n_buckets) {
 575                 i = 0;
 576                 head = flex_array_get(table->buckets, *bucket);
 577                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 578                         if (i < *last) {
 579                                 i++;
 580                                 continue;
 581                         }
 582                         *last = i + 1;
 583                         return flow;
 584                 }
 585                 (*bucket)++;
 586                 *last = 0;
 587         }
 588
 589         return NULL;
 590 }
 591
 592 static void __tbl_insert(struct flow_table *table, struct sw_flow *flow)
 593 {
 594         struct hlist_head *head;
 595
 596         head = find_bucket(table, flow->hash);
 597         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 598
 599         table->count++;
 600 }
 601
 602 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 603 {
 604         int old_ver;
 605         int i;
 606
 607         old_ver = old->node_ver;
 608         new->node_ver = !old_ver;
 609
 610         /* Insert in new table. */
 611         for (i = 0; i < old->n_buckets; i++) {
 612                 struct sw_flow *flow;
 613                 struct hlist_head *head;
 614
 615                 head = flex_array_get(old->buckets, i);
 616
 617                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 618                         __tbl_insert(new, flow);
 619         }
 620
 621         new->mask_list = old->mask_list;
 622         old->keep_flows = true;
 623 }
 624
 625 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 626 {
 627         struct flow_table *new_table;
 628
 629         new_table = __flow_tbl_alloc(n_buckets);
 630         if (!new_table)
 631                 return ERR_PTR(-ENOMEM);
 632
 633         flow_table_copy_flows(table, new_table);
 634
 635         return new_table;
 636 }
 637
 638 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 639 {
 640         return __flow_tbl_rehash(table, table->n_buckets);
 641 }
 642
 643 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 644 {
 645         return __flow_tbl_rehash(table, table->n_buckets * 2);
 646 }
 647
 648 static void __flow_free(struct sw_flow *flow)
 649 {
 650         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 651         kmem_cache_free(flow_cache, flow);
 652 }
 653
 654 static void rcu_free_flow_callback(struct rcu_head *rcu)
 655 {
 656         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 657
 658         __flow_free(flow);
 659 }
 660
 661 void ovs_flow_free(struct sw_flow *flow, bool deferred)
 662 {
 663         if (!flow)
 664                 return;
 665
 666         ovs_sw_flow_mask_del_ref(flow->mask, deferred);
 667
 668         if (deferred)
 669                 call_rcu(&flow->rcu, rcu_free_flow_callback);
 670         else
 671                 __flow_free(flow);
 672 }
 673
 674 /* RCU callback used by ovs_flow_deferred_free_acts. */
 675 static void rcu_free_acts_callback(struct rcu_head *rcu)
 676 {
 677         struct sw_flow_actions *sf_acts = container_of(rcu,
 678                         struct sw_flow_actions, rcu);
 679         kfree(sf_acts);
 680 }
 681
 682 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 683  * The caller must hold rcu_read_lock for this to be sensible. */
 684 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 685 {
 686         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 687 }
 688
 689 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 690 {
 691         struct qtag_prefix {
 692                 __be16 eth_type; /* ETH_P_8021Q */
 693                 __be16 tci;
 694         };
 695         struct qtag_prefix *qp;
 696
 697         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 698                 return 0;
 699
 700         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 701                                          sizeof(__be16))))
 702                 return -ENOMEM;
 703
 704         qp = (struct qtag_prefix *) skb->data;
 705         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 706         __skb_pull(skb, sizeof(struct qtag_prefix));
 707
 708         return 0;
 709 }
 710
 711 static __be16 parse_ethertype(struct sk_buff *skb)
 712 {
 713         struct llc_snap_hdr {
 714                 u8  dsap;  /* Always 0xAA */
 715                 u8  ssap;  /* Always 0xAA */
 716                 u8  ctrl;
 717                 u8  oui[3];
 718                 __be16 ethertype;
 719         };
 720         struct llc_snap_hdr *llc;
 721         __be16 proto;
 722
 723         proto = *(__be16 *) skb->data;
 724         __skb_pull(skb, sizeof(__be16));
 725
 726         if (ntohs(proto) >= ETH_P_802_3_MIN)
 727                 return proto;
 728
 729         if (skb->len < sizeof(struct llc_snap_hdr))
 730                 return htons(ETH_P_802_2);
 731
 732         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 733                 return htons(0);
 734
 735         llc = (struct llc_snap_hdr *) skb->data;
 736         if (llc->dsap != LLC_SAP_SNAP ||
 737             llc->ssap != LLC_SAP_SNAP ||
 738             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 739                 return htons(ETH_P_802_2);
 740
 741         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 742
 743         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 744                 return llc->ethertype;
 745
 746         return htons(ETH_P_802_2);
 747 }
 748
 749 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 750                         int nh_len)
 751 {
 752         struct icmp6hdr *icmp = icmp6_hdr(skb);
 753
 754         /* The ICMPv6 type and code fields use the 16-bit transport port
 755          * fields, so we need to store them in 16-bit network byte order.
 756          */
 757         key->ipv6.tp.src = htons(icmp->icmp6_type);
 758         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 759
 760         if (icmp->icmp6_code == 0 &&
 761             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 762              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 763                 int icmp_len = skb->len - skb_transport_offset(skb);
 764                 struct nd_msg *nd;
 765                 int offset;
 766
 767                 /* In order to process neighbor discovery options, we need the
 768                  * entire packet.
 769                  */
 770                 if (unlikely(icmp_len < sizeof(*nd)))
 771                         return 0;
 772
 773                 if (unlikely(skb_linearize(skb)))
 774                         return -ENOMEM;
 775
 776                 nd = (struct nd_msg *)skb_transport_header(skb);
 777                 key->ipv6.nd.target = nd->target;
 778
 779                 icmp_len -= sizeof(*nd);
 780                 offset = 0;
 781                 while (icmp_len >= 8) {
 782                         struct nd_opt_hdr *nd_opt =
 783                                  (struct nd_opt_hdr *)(nd->opt + offset);
 784                         int opt_len = nd_opt->nd_opt_len * 8;
 785
 786                         if (unlikely(!opt_len || opt_len > icmp_len))
 787                                 return 0;
 788
 789                         /* Store the link layer address if the appropriate
 790                          * option is provided.  It is considered an error if
 791                          * the same link layer option is specified twice.
 792                          */
 793                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 794                             && opt_len == 8) {
 795                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 796                                         goto invalid;
 797                                 memcpy(key->ipv6.nd.sll,
 798                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 799                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 800                                    && opt_len == 8) {
 801                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 802                                         goto invalid;
 803                                 memcpy(key->ipv6.nd.tll,
 804                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 805                         }
 806
 807                         icmp_len -= opt_len;
 808                         offset += opt_len;
 809                 }
 810         }
 811
 812         return 0;
 813
 814 invalid:
 815         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 816         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 817         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 818
 819         return 0;
 820 }
 821
 822 /**
 823  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 824  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 825  * Ethernet header
 826  * @in_port: port number on which @skb was received.
 827  * @key: output flow key
 828  *
 829  * The caller must ensure that skb->len >= ETH_HLEN.
 830  *
 831  * Returns 0 if successful, otherwise a negative errno value.
 832  *
 833  * Initializes @skb header pointers as follows:
 834  *
 835  *    - skb->mac_header: the Ethernet header.
 836  *
 837  *    - skb->network_header: just past the Ethernet header, or just past the
 838  *      VLAN header, to the first byte of the Ethernet payload.
 839  *
 840  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 841  *      on output, then just past the IP header, if one is present and
 842  *      of a correct length, otherwise the same as skb->network_header.
 843  *      For other key->eth.type values it is left untouched.
 844  */
 845 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key)
 846 {
 847         int error;
 848         struct ethhdr *eth;
 849
 850         memset(key, 0, sizeof(*key));
 851
 852         key->phy.priority = skb->priority;
 853         if (OVS_CB(skb)->tun_key)
 854                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 855         key->phy.in_port = in_port;
 856         key->phy.skb_mark = skb_get_mark(skb);
 857
 858         skb_reset_mac_header(skb);
 859
 860         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 861          * header in the linear data area.
 862          */
 863         eth = eth_hdr(skb);
 864         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 865         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 866
 867         __skb_pull(skb, 2 * ETH_ALEN);
 868         /* We are going to push all headers that we pull, so no need to
 869          * update skb->csum here. */
 870
 871         if (vlan_tx_tag_present(skb))
 872                 key->eth.tci = htons(vlan_get_tci(skb));
 873         else if (eth->h_proto == htons(ETH_P_8021Q))
 874                 if (unlikely(parse_vlan(skb, key)))
 875                         return -ENOMEM;
 876
 877         key->eth.type = parse_ethertype(skb);
 878         if (unlikely(key->eth.type == htons(0)))
 879                 return -ENOMEM;
 880
 881         skb_reset_network_header(skb);
 882         __skb_push(skb, skb->data - skb_mac_header(skb));
 883
 884         /* Network layer. */
 885         if (key->eth.type == htons(ETH_P_IP)) {
 886                 struct iphdr *nh;
 887                 __be16 offset;
 888
 889                 error = check_iphdr(skb);
 890                 if (unlikely(error)) {
 891                         if (error == -EINVAL) {
 892                                 skb->transport_header = skb->network_header;
 893                                 error = 0;
 894                         }
 895                         return error;
 896                 }
 897
 898                 nh = ip_hdr(skb);
 899                 key->ipv4.addr.src = nh->saddr;
 900                 key->ipv4.addr.dst = nh->daddr;
 901
 902                 key->ip.proto = nh->protocol;
 903                 key->ip.tos = nh->tos;
 904                 key->ip.ttl = nh->ttl;
 905
 906                 offset = nh->frag_off & htons(IP_OFFSET);
 907                 if (offset) {
 908                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 909                         return 0;
 910                 }
 911                 if (nh->frag_off & htons(IP_MF) ||
 912                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 913                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 914
 915                 /* Transport layer. */
 916                 if (key->ip.proto == IPPROTO_TCP) {
 917                         if (tcphdr_ok(skb)) {
 918                                 struct tcphdr *tcp = tcp_hdr(skb);
 919                                 key->ipv4.tp.src = tcp->source;
 920                                 key->ipv4.tp.dst = tcp->dest;
 921                         }
 922                 } else if (key->ip.proto == IPPROTO_UDP) {
 923                         if (udphdr_ok(skb)) {
 924                                 struct udphdr *udp = udp_hdr(skb);
 925                                 key->ipv4.tp.src = udp->source;
 926                                 key->ipv4.tp.dst = udp->dest;
 927                         }
 928                 } else if (key->ip.proto == IPPROTO_SCTP) {
 929                         if (sctphdr_ok(skb)) {
 930                                 struct sctphdr *sctp = sctp_hdr(skb);
 931                                 key->ipv4.tp.src = sctp->source;
 932                                 key->ipv4.tp.dst = sctp->dest;
 933                         }
 934                 } else if (key->ip.proto == IPPROTO_ICMP) {
 935                         if (icmphdr_ok(skb)) {
 936                                 struct icmphdr *icmp = icmp_hdr(skb);
 937                                 /* The ICMP type and code fields use the 16-bit
 938                                  * transport port fields, so we need to store
 939                                  * them in 16-bit network byte order. */
 940                                 key->ipv4.tp.src = htons(icmp->type);
 941                                 key->ipv4.tp.dst = htons(icmp->code);
 942                         }
 943                 }
 944
 945         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 946                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 947                 struct arp_eth_header *arp;
 948
 949                 arp = (struct arp_eth_header *)skb_network_header(skb);
 950
 951                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 952                                 && arp->ar_pro == htons(ETH_P_IP)
 953                                 && arp->ar_hln == ETH_ALEN
 954                                 && arp->ar_pln == 4) {
 955
 956                         /* We only match on the lower 8 bits of the opcode. */
 957                         if (ntohs(arp->ar_op) <= 0xff)
 958                                 key->ip.proto = ntohs(arp->ar_op);
 959                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 960                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 961                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 962                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 963                 }
 964         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 965                 int nh_len;             /* IPv6 Header + Extensions */
 966
 967                 nh_len = parse_ipv6hdr(skb, key);
 968                 if (unlikely(nh_len < 0)) {
 969                         if (nh_len == -EINVAL) {
 970                                 skb->transport_header = skb->network_header;
 971                                 error = 0;
 972                         } else {
 973                                 error = nh_len;
 974                         }
 975                         return error;
 976                 }
 977
 978                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 979                         return 0;
 980                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 981                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 982
 983                 /* Transport layer. */
 984                 if (key->ip.proto == NEXTHDR_TCP) {
 985                         if (tcphdr_ok(skb)) {
 986                                 struct tcphdr *tcp = tcp_hdr(skb);
 987                                 key->ipv6.tp.src = tcp->source;
 988                                 key->ipv6.tp.dst = tcp->dest;
 989                         }
 990                 } else if (key->ip.proto == NEXTHDR_UDP) {
 991                         if (udphdr_ok(skb)) {
 992                                 struct udphdr *udp = udp_hdr(skb);
 993                                 key->ipv6.tp.src = udp->source;
 994                                 key->ipv6.tp.dst = udp->dest;
 995                         }
 996                 } else if (key->ip.proto == NEXTHDR_SCTP) {
 997                         if (sctphdr_ok(skb)) {
 998                                 struct sctphdr *sctp = sctp_hdr(skb);
 999                                 key->ipv6.tp.src = sctp->source;
1000                                 key->ipv6.tp.dst = sctp->dest;
1001                         }
1002                 } else if (key->ip.proto == NEXTHDR_ICMP) {
1003                         if (icmp6hdr_ok(skb)) {
1004                                 error = parse_icmpv6(skb, key, nh_len);
1005                                 if (error)
1006                                         return error;
1007                         }
1008                 }
1009         }
1010
1011         return 0;
1012 }
1013
1014 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start,
1015                          int key_end)
1016 {
1017         u32 *hash_key = (u32 *)((u8 *)key + key_start);
1018         int hash_u32s = (key_end - key_start) >> 2;
1019
1020         /* Make sure number of hash bytes are multiple of u32. */
1021         BUILD_BUG_ON(sizeof(long) % sizeof(u32));
1022
1023         return jhash2(hash_key, hash_u32s, 0);
1024 }
1025
1026 static int flow_key_start(const struct sw_flow_key *key)
1027 {
1028         if (key->tun_key.ipv4_dst)
1029                 return 0;
1030         else
1031                 return rounddown(offsetof(struct sw_flow_key, phy),
1032                                           sizeof(long));
1033 }
1034
1035 static bool __cmp_key(const struct sw_flow_key *key1,
1036                 const struct sw_flow_key *key2,  int key_start, int key_end)
1037 {
1038         const long *cp1 = (long *)((u8 *)key1 + key_start);
1039         const long *cp2 = (long *)((u8 *)key2 + key_start);
1040         long diffs = 0;
1041         int i;
1042
1043         for (i = key_start; i < key_end;  i += sizeof(long))
1044                 diffs |= *cp1++ ^ *cp2++;
1045
1046         return diffs == 0;
1047 }
1048
1049 static bool __flow_cmp_masked_key(const struct sw_flow *flow,
1050                 const struct sw_flow_key *key, int key_start, int key_end)
1051 {
1052         return __cmp_key(&flow->key, key, key_start, key_end);
1053 }
1054
1055 static bool __flow_cmp_unmasked_key(const struct sw_flow *flow,
1056                   const struct sw_flow_key *key, int key_start, int key_end)
1057 {
1058         return __cmp_key(&flow->unmasked_key, key, key_start, key_end);
1059 }
1060
1061 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
1062                 const struct sw_flow_key *key, int key_end)
1063 {
1064         int key_start;
1065         key_start = flow_key_start(key);
1066
1067         return __flow_cmp_unmasked_key(flow, key, key_start, key_end);
1068
1069 }
1070
1071 struct sw_flow *ovs_flow_lookup_unmasked_key(struct flow_table *table,
1072                                        struct sw_flow_match *match)
1073 {
1074         struct sw_flow_key *unmasked = match->key;
1075         int key_end = match->range.end;
1076         struct sw_flow *flow;
1077
1078         flow = ovs_flow_lookup(table, unmasked);
1079         if (flow && (!ovs_flow_cmp_unmasked_key(flow, unmasked, key_end)))
1080                 flow = NULL;
1081
1082         return flow;
1083 }
1084
1085 static struct sw_flow *ovs_masked_flow_lookup(struct flow_table *table,
1086                                     const struct sw_flow_key *unmasked,
1087                                     struct sw_flow_mask *mask)
1088 {
1089         struct sw_flow *flow;
1090         struct hlist_head *head;
1091         int key_start = mask->range.start;
1092         int key_end = mask->range.end;
1093         u32 hash;
1094         struct sw_flow_key masked_key;
1095
1096         ovs_flow_key_mask(&masked_key, unmasked, mask);
1097         hash = ovs_flow_hash(&masked_key, key_start, key_end);
1098         head = find_bucket(table, hash);
1099         hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
1100                 if (flow->mask == mask &&
1101                     __flow_cmp_masked_key(flow, &masked_key,
1102                                           key_start, key_end))
1103                         return flow;
1104         }
1105         return NULL;
1106 }
1107
1108 struct sw_flow *ovs_flow_lookup(struct flow_table *tbl,
1109                                 const struct sw_flow_key *key)
1110 {
1111         struct sw_flow *flow = NULL;
1112         struct sw_flow_mask *mask;
1113
1114         list_for_each_entry_rcu(mask, tbl->mask_list, list) {
1115                 flow = ovs_masked_flow_lookup(tbl, key, mask);
1116                 if (flow)  /* Found */
1117                         break;
1118         }
1119
1120         return flow;
1121 }
1122
1123
1124 void ovs_flow_insert(struct flow_table *table, struct sw_flow *flow)
1125 {
1126         flow->hash = ovs_flow_hash(&flow->key, flow->mask->range.start,
1127                         flow->mask->range.end);
1128         __tbl_insert(table, flow);
1129 }
1130
1131 void ovs_flow_remove(struct flow_table *table, struct sw_flow *flow)
1132 {
1133         BUG_ON(table->count == 0);
1134         hlist_del_rcu(&flow->hash_node[table->node_ver]);
1135         table->count--;
1136 }
1137
1138 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
1139 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
1140         [OVS_KEY_ATTR_ENCAP] = -1,
1141         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
1142         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
1143         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
1144         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
1145         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
1146         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
1147         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
1148         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
1149         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
1150         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
1151         [OVS_KEY_ATTR_SCTP] = sizeof(struct ovs_key_sctp),
1152         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
1153         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
1154         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
1155         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
1156         [OVS_KEY_ATTR_TUNNEL] = -1,
1157 };
1158
1159 static bool is_all_zero(const u8 *fp, size_t size)
1160 {
1161         int i;
1162
1163         if (!fp)
1164                 return false;
1165
1166         for (i = 0; i < size; i++)
1167                 if (fp[i])
1168                         return false;
1169
1170         return true;
1171 }
1172
1173 static int __parse_flow_nlattrs(const struct nlattr *attr,
1174                               const struct nlattr *a[],
1175                               u64 *attrsp, bool nz)
1176 {
1177         const struct nlattr *nla;
1178         u64 attrs;
1179         int rem;
1180
1181         attrs = *attrsp;
1182         nla_for_each_nested(nla, attr, rem) {
1183                 u16 type = nla_type(nla);
1184                 int expected_len;
1185
1186                 if (type > OVS_KEY_ATTR_MAX) {
1187                         OVS_NLERR("Unknown key attribute (type=%d, max=%d).\n",
1188                                   type, OVS_KEY_ATTR_MAX);
1189                 }
1190
1191                 if (attrs & (1ULL << type)) {
1192                         OVS_NLERR("Duplicate key attribute (type %d).\n", type);
1193                         return -EINVAL;
1194                 }
1195
1196                 expected_len = ovs_key_lens[type];
1197                 if (nla_len(nla) != expected_len && expected_len != -1) {
1198                         OVS_NLERR("Key attribute has unexpected length (type=%d"
1199                                   ", length=%d, expected=%d).\n", type,
1200                                   nla_len(nla), expected_len);
1201                         return -EINVAL;
1202                 }
1203
1204                 if (!nz || !is_all_zero(nla_data(nla), expected_len)) {
1205                         attrs |= 1ULL << type;
1206                         a[type] = nla;
1207                 }
1208         }
1209         if (rem) {
1210                 OVS_NLERR("Message has %d unknown bytes.\n", rem);
1211                 return -EINVAL;
1212         }
1213
1214         *attrsp = attrs;
1215         return 0;
1216 }
1217
1218 static int parse_flow_mask_nlattrs(const struct nlattr *attr,
1219                               const struct nlattr *a[], u64 *attrsp)
1220 {
1221         return __parse_flow_nlattrs(attr, a, attrsp, true);
1222 }
1223
1224 static int parse_flow_nlattrs(const struct nlattr *attr,
1225                               const struct nlattr *a[], u64 *attrsp)
1226 {
1227         return __parse_flow_nlattrs(attr, a, attrsp, false);
1228 }
1229
1230 int ovs_ipv4_tun_from_nlattr(const struct nlattr *attr,
1231                              struct sw_flow_match *match, bool is_mask)
1232 {
1233         struct nlattr *a;
1234         int rem;
1235         bool ttl = false;
1236         __be16 tun_flags = 0;
1237
1238         nla_for_each_nested(a, attr, rem) {
1239                 int type = nla_type(a);
1240                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1241                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1242                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1243                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1244                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1245                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1246                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1247                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1248                 };
1249
1250                 if (type > OVS_TUNNEL_KEY_ATTR_MAX) {
1251                         OVS_NLERR("Unknown IPv4 tunnel attribute (type=%d, max=%d).\n",
1252                         type, OVS_TUNNEL_KEY_ATTR_MAX);
1253                         return -EINVAL;
1254                 }
1255
1256                 if (ovs_tunnel_key_lens[type] != nla_len(a)) {
1257                         OVS_NLERR("IPv4 tunnel attribute type has unexpected "
1258                                   " length (type=%d, length=%d, expected=%d).\n",
1259                                   type, nla_len(a), ovs_tunnel_key_lens[type]);
1260                         return -EINVAL;
1261                 }
1262
1263                 switch (type) {
1264                 case OVS_TUNNEL_KEY_ATTR_ID:
1265                         SW_FLOW_KEY_PUT(match, tun_key.tun_id,
1266                                         nla_get_be64(a), is_mask);
1267                         tun_flags |= TUNNEL_KEY;
1268                         break;
1269                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1270                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_src,
1271                                         nla_get_be32(a), is_mask);
1272                         break;
1273                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1274                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_dst,
1275                                         nla_get_be32(a), is_mask);
1276                         break;
1277                 case OVS_TUNNEL_KEY_ATTR_TOS:
1278                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_tos,
1279                                         nla_get_u8(a), is_mask);
1280                         break;
1281                 case OVS_TUNNEL_KEY_ATTR_TTL:
1282                         SW_FLOW_KEY_PUT(match, tun_key.ipv4_ttl,
1283                                         nla_get_u8(a), is_mask);
1284                         ttl = true;
1285                         break;
1286                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1287                         tun_flags |= TUNNEL_DONT_FRAGMENT;
1288                         break;
1289                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1290                         tun_flags |= TUNNEL_CSUM;
1291                         break;
1292                 default:
1293                         return -EINVAL;
1294                 }
1295         }
1296
1297         SW_FLOW_KEY_PUT(match, tun_key.tun_flags, tun_flags, is_mask);
1298
1299         if (rem > 0) {
1300                 OVS_NLERR("IPv4 tunnel attribute has %d unknown bytes.\n", rem);
1301                 return -EINVAL;
1302         }
1303
1304         if (!is_mask) {
1305                 if (!match->key->tun_key.ipv4_dst) {
1306                         OVS_NLERR("IPv4 tunnel destination address is zero.\n");
1307                         return -EINVAL;
1308                 }
1309
1310                 if (!ttl) {
1311                         OVS_NLERR("IPv4 tunnel TTL not specified.\n");
1312                         return -EINVAL;
1313                 }
1314         }
1315
1316         return 0;
1317 }
1318
1319 int ovs_ipv4_tun_to_nlattr(struct sk_buff *skb,
1320                            const struct ovs_key_ipv4_tunnel *tun_key,
1321                            const struct ovs_key_ipv4_tunnel *output)
1322 {
1323         struct nlattr *nla;
1324
1325         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1326         if (!nla)
1327                 return -EMSGSIZE;
1328
1329         if (output->tun_flags & TUNNEL_KEY &&
1330             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, output->tun_id))
1331                 return -EMSGSIZE;
1332         if (output->ipv4_src &&
1333                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, output->ipv4_src))
1334                 return -EMSGSIZE;
1335         if (output->ipv4_dst &&
1336                 nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, output->ipv4_dst))
1337                 return -EMSGSIZE;
1338         if (output->ipv4_tos &&
1339                 nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, output->ipv4_tos))
1340                 return -EMSGSIZE;
1341         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, output->ipv4_ttl))
1342                 return -EMSGSIZE;
1343         if ((output->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1344                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1345                 return -EMSGSIZE;
1346         if ((output->tun_flags & TUNNEL_CSUM) &&
1347                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1348                 return -EMSGSIZE;
1349
1350         nla_nest_end(skb, nla);
1351         return 0;
1352 }
1353
1354
1355 static int metadata_from_nlattrs(struct sw_flow_match *match,  u64 *attrs,
1356                 const struct nlattr **a, bool is_mask)
1357 {
1358         if (*attrs & (1ULL << OVS_KEY_ATTR_PRIORITY)) {
1359                 SW_FLOW_KEY_PUT(match, phy.priority,
1360                           nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]), is_mask);
1361                 *attrs &= ~(1ULL << OVS_KEY_ATTR_PRIORITY);
1362         }
1363
1364         if (*attrs & (1ULL << OVS_KEY_ATTR_IN_PORT)) {
1365                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1366
1367                 if (is_mask)
1368                         in_port = 0xffffffff; /* Always exact match in_port. */
1369                 else if (in_port >= DP_MAX_PORTS)
1370                         return -EINVAL;
1371
1372                 SW_FLOW_KEY_PUT(match, phy.in_port, in_port, is_mask);
1373                 *attrs &= ~(1ULL << OVS_KEY_ATTR_IN_PORT);
1374         } else if (!is_mask) {
1375                 SW_FLOW_KEY_PUT(match, phy.in_port, DP_MAX_PORTS, is_mask);
1376         }
1377
1378         if (*attrs & (1ULL << OVS_KEY_ATTR_SKB_MARK)) {
1379                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1380 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1381                 if (!is_mask && mark != 0) {
1382                         OVS_NLERR("skb->mark must be zero on this kernel (mark=%d).\n", mark);
1383                         return -EINVAL;
1384                 }
1385 #endif
1386                 SW_FLOW_KEY_PUT(match, phy.skb_mark, mark, is_mask);
1387                 *attrs &= ~(1ULL << OVS_KEY_ATTR_SKB_MARK);
1388         }
1389         if (*attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1390                 if (ovs_ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], match,
1391                                         is_mask))
1392                         return -EINVAL;
1393                 *attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1394         }
1395         return 0;
1396 }
1397
1398 static int ovs_key_from_nlattrs(struct sw_flow_match *match,  u64 attrs,
1399                 const struct nlattr **a, bool is_mask)
1400 {
1401         int err;
1402         u64 orig_attrs = attrs;
1403
1404         err = metadata_from_nlattrs(match, &attrs, a, is_mask);
1405         if (err)
1406                 return err;
1407
1408         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) {
1409                 const struct ovs_key_ethernet *eth_key;
1410
1411                 eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1412                 SW_FLOW_KEY_MEMCPY(match, eth.src,
1413                                 eth_key->eth_src, ETH_ALEN, is_mask);
1414                 SW_FLOW_KEY_MEMCPY(match, eth.dst,
1415                                 eth_key->eth_dst, ETH_ALEN, is_mask);
1416                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERNET);
1417         }
1418
1419         if (attrs & (1ULL << OVS_KEY_ATTR_VLAN)) {
1420                 __be16 tci;
1421
1422                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1423                 if (!(tci & htons(VLAN_TAG_PRESENT))) {
1424                         if (is_mask)
1425                                 OVS_NLERR("VLAN TCI mask does not have exact match for VLAN_TAG_PRESENT bit.\n");
1426                         else
1427                                 OVS_NLERR("VLAN TCI does not have VLAN_TAG_PRESENT bit set.\n");
1428
1429                         return -EINVAL;
1430                 }
1431
1432                 SW_FLOW_KEY_PUT(match, eth.tci, tci, is_mask);
1433                 attrs &= ~(1ULL << OVS_KEY_ATTR_VLAN);
1434         } else if (!is_mask)
1435                 SW_FLOW_KEY_PUT(match, eth.tci, htons(0xffff), true);
1436
1437         if (attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) {
1438                 __be16 eth_type;
1439
1440                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1441                 if (is_mask) {
1442                         /* Always exact match EtherType. */
1443                         eth_type = htons(0xffff);
1444                 } else if (ntohs(eth_type) < ETH_P_802_3_MIN) {
1445                         OVS_NLERR("EtherType is less than minimum (type=%x, min=%x).\n",
1446                                         ntohs(eth_type), ETH_P_802_3_MIN);
1447                         return -EINVAL;
1448                 }
1449
1450                 SW_FLOW_KEY_PUT(match, eth.type, eth_type, is_mask);
1451                 attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1452         } else if (!is_mask) {
1453                 SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask);
1454         }
1455
1456         if (attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1457                 const struct ovs_key_ipv4 *ipv4_key;
1458
1459                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1460                 if (!is_mask && ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX) {
1461                         OVS_NLERR("Unknown IPv4 fragment type (value=%d, max=%d).\n",
1462                                 ipv4_key->ipv4_frag, OVS_FRAG_TYPE_MAX);
1463                         return -EINVAL;
1464                 }
1465                 SW_FLOW_KEY_PUT(match, ip.proto,
1466                                 ipv4_key->ipv4_proto, is_mask);
1467                 SW_FLOW_KEY_PUT(match, ip.tos,
1468                                 ipv4_key->ipv4_tos, is_mask);
1469                 SW_FLOW_KEY_PUT(match, ip.ttl,
1470                                 ipv4_key->ipv4_ttl, is_mask);
1471                 SW_FLOW_KEY_PUT(match, ip.frag,
1472                                 ipv4_key->ipv4_frag, is_mask);
1473                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1474                                 ipv4_key->ipv4_src, is_mask);
1475                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1476                                 ipv4_key->ipv4_dst, is_mask);
1477                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV4);
1478         }
1479
1480         if (attrs & (1ULL << OVS_KEY_ATTR_IPV6)) {
1481                 const struct ovs_key_ipv6 *ipv6_key;
1482
1483                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1484                 if (!is_mask && ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX) {
1485                         OVS_NLERR("Unknown IPv6 fragment type (value=%d, max=%d).\n",
1486                                 ipv6_key->ipv6_frag, OVS_FRAG_TYPE_MAX);
1487                         return -EINVAL;
1488                 }
1489                 SW_FLOW_KEY_PUT(match, ipv6.label,
1490                                 ipv6_key->ipv6_label, is_mask);
1491                 SW_FLOW_KEY_PUT(match, ip.proto,
1492                                 ipv6_key->ipv6_proto, is_mask);
1493                 SW_FLOW_KEY_PUT(match, ip.tos,
1494                                 ipv6_key->ipv6_tclass, is_mask);
1495                 SW_FLOW_KEY_PUT(match, ip.ttl,
1496                                 ipv6_key->ipv6_hlimit, is_mask);
1497                 SW_FLOW_KEY_PUT(match, ip.frag,
1498                                 ipv6_key->ipv6_frag, is_mask);
1499                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.src,
1500                                 ipv6_key->ipv6_src,
1501                                 sizeof(match->key->ipv6.addr.src),
1502                                 is_mask);
1503                 SW_FLOW_KEY_MEMCPY(match, ipv6.addr.dst,
1504                                 ipv6_key->ipv6_dst,
1505                                 sizeof(match->key->ipv6.addr.dst),
1506                                 is_mask);
1507
1508                 attrs &= ~(1ULL << OVS_KEY_ATTR_IPV6);
1509         }
1510
1511         if (attrs & (1ULL << OVS_KEY_ATTR_ARP)) {
1512                 const struct ovs_key_arp *arp_key;
1513
1514                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1515                 if (!is_mask && (arp_key->arp_op & htons(0xff00))) {
1516                         OVS_NLERR("Unknown ARP opcode (opcode=%d).\n",
1517                                   arp_key->arp_op);
1518                         return -EINVAL;
1519                 }
1520
1521                 SW_FLOW_KEY_PUT(match, ipv4.addr.src,
1522                                 arp_key->arp_sip, is_mask);
1523                 SW_FLOW_KEY_PUT(match, ipv4.addr.dst,
1524                         arp_key->arp_tip, is_mask);
1525                 SW_FLOW_KEY_PUT(match, ip.proto,
1526                                 ntohs(arp_key->arp_op), is_mask);
1527                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.sha,
1528                                 arp_key->arp_sha, ETH_ALEN, is_mask);
1529                 SW_FLOW_KEY_MEMCPY(match, ipv4.arp.tha,
1530                                 arp_key->arp_tha, ETH_ALEN, is_mask);
1531
1532                 attrs &= ~(1ULL << OVS_KEY_ATTR_ARP);
1533         }
1534
1535         if (attrs & (1ULL << OVS_KEY_ATTR_TCP)) {
1536                 const struct ovs_key_tcp *tcp_key;
1537
1538                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
1539                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1540                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1541                                         tcp_key->tcp_src, is_mask);
1542                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1543                                         tcp_key->tcp_dst, is_mask);
1544                 } else {
1545                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1546                                         tcp_key->tcp_src, is_mask);
1547                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1548                                         tcp_key->tcp_dst, is_mask);
1549                 }
1550                 attrs &= ~(1ULL << OVS_KEY_ATTR_TCP);
1551         }
1552
1553         if (attrs & (1ULL << OVS_KEY_ATTR_UDP)) {
1554                 const struct ovs_key_udp *udp_key;
1555
1556                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
1557                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1558                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1559                                         udp_key->udp_src, is_mask);
1560                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1561                                         udp_key->udp_dst, is_mask);
1562                 } else {
1563                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1564                                         udp_key->udp_src, is_mask);
1565                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1566                                         udp_key->udp_dst, is_mask);
1567                 }
1568                 attrs &= ~(1ULL << OVS_KEY_ATTR_UDP);
1569         }
1570
1571         if (attrs & (1ULL << OVS_KEY_ATTR_SCTP)) {
1572                 const struct ovs_key_sctp *sctp_key;
1573
1574                 sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]);
1575                 if (orig_attrs & (1ULL << OVS_KEY_ATTR_IPV4)) {
1576                         SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1577                                         sctp_key->sctp_src, is_mask);
1578                         SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1579                                         sctp_key->sctp_dst, is_mask);
1580                 } else {
1581                         SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1582                                         sctp_key->sctp_src, is_mask);
1583                         SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1584                                         sctp_key->sctp_dst, is_mask);
1585                 }
1586                 attrs &= ~(1ULL << OVS_KEY_ATTR_SCTP);
1587         }
1588
1589         if (attrs & (1ULL << OVS_KEY_ATTR_ICMP)) {
1590                 const struct ovs_key_icmp *icmp_key;
1591
1592                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
1593                 SW_FLOW_KEY_PUT(match, ipv4.tp.src,
1594                                 htons(icmp_key->icmp_type), is_mask);
1595                 SW_FLOW_KEY_PUT(match, ipv4.tp.dst,
1596                                 htons(icmp_key->icmp_code), is_mask);
1597                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMP);
1598         }
1599
1600         if (attrs & (1ULL << OVS_KEY_ATTR_ICMPV6)) {
1601                 const struct ovs_key_icmpv6 *icmpv6_key;
1602
1603                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
1604                 SW_FLOW_KEY_PUT(match, ipv6.tp.src,
1605                                 htons(icmpv6_key->icmpv6_type), is_mask);
1606                 SW_FLOW_KEY_PUT(match, ipv6.tp.dst,
1607                                 htons(icmpv6_key->icmpv6_code), is_mask);
1608                 attrs &= ~(1ULL << OVS_KEY_ATTR_ICMPV6);
1609         }
1610
1611         if (attrs & (1ULL << OVS_KEY_ATTR_ND)) {
1612                 const struct ovs_key_nd *nd_key;
1613
1614                 nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
1615                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.target,
1616                         nd_key->nd_target,
1617                         sizeof(match->key->ipv6.nd.target),
1618                         is_mask);
1619                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.sll,
1620                         nd_key->nd_sll, ETH_ALEN, is_mask);
1621                 SW_FLOW_KEY_MEMCPY(match, ipv6.nd.tll,
1622                                 nd_key->nd_tll, ETH_ALEN, is_mask);
1623                 attrs &= ~(1ULL << OVS_KEY_ATTR_ND);
1624         }
1625
1626         if (attrs != 0)
1627                 return -EINVAL;
1628
1629         return 0;
1630 }
1631
1632 /**
1633  * ovs_match_from_nlattrs - parses Netlink attributes into a flow key and
1634  * mask. In case the 'mask' is NULL, the flow is treated as exact match
1635  * flow. Otherwise, it is treated as a wildcarded flow, except the mask
1636  * does not include any don't care bit.
1637  * @match: receives the extracted flow match information.
1638  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1639  * sequence. The fields should of the packet that triggered the creation
1640  * of this flow.
1641  * @mask: Optional. Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink
1642  * attribute specifies the mask field of the wildcarded flow.
1643  */
1644 int ovs_match_from_nlattrs(struct sw_flow_match *match,
1645                            const struct nlattr *key,
1646                            const struct nlattr *mask)
1647 {
1648         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1649         const struct nlattr *encap;
1650         u64 key_attrs = 0;
1651         u64 mask_attrs = 0;
1652         bool encap_valid = false;
1653         int err;
1654
1655         err = parse_flow_nlattrs(key, a, &key_attrs);
1656         if (err)
1657                 return err;
1658
1659         if ((key_attrs & (1ULL << OVS_KEY_ATTR_ETHERNET)) &&
1660             (key_attrs & (1ULL << OVS_KEY_ATTR_ETHERTYPE)) &&
1661             (nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q))) {
1662                 __be16 tci;
1663
1664                 if (!((key_attrs & (1ULL << OVS_KEY_ATTR_VLAN)) &&
1665                       (key_attrs & (1ULL << OVS_KEY_ATTR_ENCAP)))) {
1666                         OVS_NLERR("Invalid Vlan frame.\n");
1667                         return -EINVAL;
1668                 }
1669
1670                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1671                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1672                 encap = a[OVS_KEY_ATTR_ENCAP];
1673                 key_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1674                 encap_valid = true;
1675
1676                 if (tci & htons(VLAN_TAG_PRESENT)) {
1677                         err = parse_flow_nlattrs(encap, a, &key_attrs);
1678                         if (err)
1679                                 return err;
1680                 } else if (!tci) {
1681                         /* Corner case for truncated 802.1Q header. */
1682                         if (nla_len(encap)) {
1683                                 OVS_NLERR("Truncated 802.1Q header has non-zero encap attribute.\n");
1684                                 return -EINVAL;
1685                         }
1686                 } else {
1687                         OVS_NLERR("Encap attribute is set for a non-VLAN frame.\n");
1688                         return  -EINVAL;
1689                 }
1690         }
1691
1692         err = ovs_key_from_nlattrs(match, key_attrs, a, false);
1693         if (err)
1694                 return err;
1695
1696         if (mask) {
1697                 err = parse_flow_mask_nlattrs(mask, a, &mask_attrs);
1698                 if (err)
1699                         return err;
1700
1701                 if (mask_attrs & 1ULL << OVS_KEY_ATTR_ENCAP)  {
1702                         __be16 eth_type = 0;
1703                         __be16 tci = 0;
1704
1705                         if (!encap_valid) {
1706                                 OVS_NLERR("Encap mask attribute is set for non-VLAN frame.\n");
1707                                 return  -EINVAL;
1708                         }
1709
1710                         mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ENCAP);
1711                         if (a[OVS_KEY_ATTR_ETHERTYPE])
1712                                 eth_type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1713
1714                         if (eth_type == htons(0xffff)) {
1715                                 mask_attrs &= ~(1ULL << OVS_KEY_ATTR_ETHERTYPE);
1716                                 encap = a[OVS_KEY_ATTR_ENCAP];
1717                                 err = parse_flow_mask_nlattrs(encap, a, &mask_attrs);
1718                         } else {
1719                                 OVS_NLERR("VLAN frames must have an exact match on the TPID (mask=%x).\n",
1720                                                 ntohs(eth_type));
1721                                 return -EINVAL;
1722                         }
1723
1724                         if (a[OVS_KEY_ATTR_VLAN])
1725                                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1726
1727                         if (!(tci & htons(VLAN_TAG_PRESENT))) {
1728                                 OVS_NLERR("VLAN tag present bit must have an exact match (tci_mask=%x).\n", ntohs(tci));
1729                                 return -EINVAL;
1730                         }
1731                 }
1732
1733                 err = ovs_key_from_nlattrs(match, mask_attrs, a, true);
1734                 if (err)
1735                         return err;
1736         } else {
1737                 /* Populate exact match flow's key mask. */
1738                 if (match->mask)
1739                         ovs_sw_flow_mask_set(match->mask, &match->range, 0xff);
1740         }
1741
1742         if (!ovs_match_validate(match, key_attrs, mask_attrs))
1743                 return -EINVAL;
1744
1745         return 0;
1746 }
1747
1748 /**
1749  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1750  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1751  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1752  * sequence.
1753  *
1754  * This parses a series of Netlink attributes that form a flow key, which must
1755  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1756  * get the metadata, that is, the parts of the flow key that cannot be
1757  * extracted from the packet itself.
1758  */
1759
1760 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1761                 const struct nlattr *attr)
1762 {
1763         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1764         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1765         u64 attrs = 0;
1766         int err;
1767         struct sw_flow_match match;
1768
1769         flow->key.phy.in_port = DP_MAX_PORTS;
1770         flow->key.phy.priority = 0;
1771         flow->key.phy.skb_mark = 0;
1772         memset(tun_key, 0, sizeof(flow->key.tun_key));
1773
1774         err = parse_flow_nlattrs(attr, a, &attrs);
1775         if (err)
1776                 return -EINVAL;
1777
1778         memset(&match, 0, sizeof(match));
1779         match.key = &flow->key;
1780
1781         err = metadata_from_nlattrs(&match, &attrs, a, false);
1782         if (err)
1783                 return err;
1784
1785         return 0;
1786 }
1787
1788 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey,
1789                 const struct sw_flow_key *output, struct sk_buff *skb)
1790 {
1791         struct ovs_key_ethernet *eth_key;
1792         struct nlattr *nla, *encap;
1793         bool is_mask = (swkey != output);
1794
1795         if (nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, output->phy.priority))
1796                 goto nla_put_failure;
1797
1798         if ((swkey->tun_key.ipv4_dst || is_mask) &&
1799             ovs_ipv4_tun_to_nlattr(skb, &swkey->tun_key, &output->tun_key))
1800                 goto nla_put_failure;
1801
1802         if (swkey->phy.in_port == DP_MAX_PORTS) {
1803                 if (is_mask && (output->phy.in_port == 0xffff))
1804                         if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, 0xffffffff))
1805                                 goto nla_put_failure;
1806         } else {
1807                 u16 upper_u16;
1808                 upper_u16 = !is_mask ? 0 : 0xffff;
1809
1810                 if (nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT,
1811                                 (upper_u16 << 16) | output->phy.in_port))
1812                         goto nla_put_failure;
1813         }
1814
1815         if (nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, output->phy.skb_mark))
1816                 goto nla_put_failure;
1817
1818         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1819         if (!nla)
1820                 goto nla_put_failure;
1821
1822         eth_key = nla_data(nla);
1823         memcpy(eth_key->eth_src, output->eth.src, ETH_ALEN);
1824         memcpy(eth_key->eth_dst, output->eth.dst, ETH_ALEN);
1825
1826         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1827                 __be16 eth_type;
1828                 eth_type = !is_mask ? htons(ETH_P_8021Q) : htons(0xffff);
1829                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, eth_type) ||
1830                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, output->eth.tci))
1831                         goto nla_put_failure;
1832                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1833                 if (!swkey->eth.tci)
1834                         goto unencap;
1835         } else
1836                 encap = NULL;
1837
1838         if (swkey->eth.type == htons(ETH_P_802_2)) {
1839                 /*
1840                  * Ethertype 802.2 is represented in the netlink with omitted
1841                  * OVS_KEY_ATTR_ETHERTYPE in the flow key attribute, and
1842                  * 0xffff in the mask attribute.  Ethertype can also
1843                  * be wildcarded.
1844                  */
1845                 if (is_mask && output->eth.type)
1846                         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE,
1847                                                 output->eth.type))
1848                                 goto nla_put_failure;
1849                 goto unencap;
1850         }
1851
1852         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, output->eth.type))
1853                 goto nla_put_failure;
1854
1855         if (swkey->eth.type == htons(ETH_P_IP)) {
1856                 struct ovs_key_ipv4 *ipv4_key;
1857
1858                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1859                 if (!nla)
1860                         goto nla_put_failure;
1861                 ipv4_key = nla_data(nla);
1862                 ipv4_key->ipv4_src = output->ipv4.addr.src;
1863                 ipv4_key->ipv4_dst = output->ipv4.addr.dst;
1864                 ipv4_key->ipv4_proto = output->ip.proto;
1865                 ipv4_key->ipv4_tos = output->ip.tos;
1866                 ipv4_key->ipv4_ttl = output->ip.ttl;
1867                 ipv4_key->ipv4_frag = output->ip.frag;
1868         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1869                 struct ovs_key_ipv6 *ipv6_key;
1870
1871                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1872                 if (!nla)
1873                         goto nla_put_failure;
1874                 ipv6_key = nla_data(nla);
1875                 memcpy(ipv6_key->ipv6_src, &output->ipv6.addr.src,
1876                                 sizeof(ipv6_key->ipv6_src));
1877                 memcpy(ipv6_key->ipv6_dst, &output->ipv6.addr.dst,
1878                                 sizeof(ipv6_key->ipv6_dst));
1879                 ipv6_key->ipv6_label = output->ipv6.label;
1880                 ipv6_key->ipv6_proto = output->ip.proto;
1881                 ipv6_key->ipv6_tclass = output->ip.tos;
1882                 ipv6_key->ipv6_hlimit = output->ip.ttl;
1883                 ipv6_key->ipv6_frag = output->ip.frag;
1884         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1885                    swkey->eth.type == htons(ETH_P_RARP)) {
1886                 struct ovs_key_arp *arp_key;
1887
1888                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1889                 if (!nla)
1890                         goto nla_put_failure;
1891                 arp_key = nla_data(nla);
1892                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1893                 arp_key->arp_sip = output->ipv4.addr.src;
1894                 arp_key->arp_tip = output->ipv4.addr.dst;
1895                 arp_key->arp_op = htons(output->ip.proto);
1896                 memcpy(arp_key->arp_sha, output->ipv4.arp.sha, ETH_ALEN);
1897                 memcpy(arp_key->arp_tha, output->ipv4.arp.tha, ETH_ALEN);
1898         }
1899
1900         if ((swkey->eth.type == htons(ETH_P_IP) ||
1901              swkey->eth.type == htons(ETH_P_IPV6)) &&
1902              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1903
1904                 if (swkey->ip.proto == IPPROTO_TCP) {
1905                         struct ovs_key_tcp *tcp_key;
1906
1907                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1908                         if (!nla)
1909                                 goto nla_put_failure;
1910                         tcp_key = nla_data(nla);
1911                         if (swkey->eth.type == htons(ETH_P_IP)) {
1912                                 tcp_key->tcp_src = output->ipv4.tp.src;
1913                                 tcp_key->tcp_dst = output->ipv4.tp.dst;
1914                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1915                                 tcp_key->tcp_src = output->ipv6.tp.src;
1916                                 tcp_key->tcp_dst = output->ipv6.tp.dst;
1917                         }
1918                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1919                         struct ovs_key_udp *udp_key;
1920
1921                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1922                         if (!nla)
1923                                 goto nla_put_failure;
1924                         udp_key = nla_data(nla);
1925                         if (swkey->eth.type == htons(ETH_P_IP)) {
1926                                 udp_key->udp_src = output->ipv4.tp.src;
1927                                 udp_key->udp_dst = output->ipv4.tp.dst;
1928                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1929                                 udp_key->udp_src = output->ipv6.tp.src;
1930                                 udp_key->udp_dst = output->ipv6.tp.dst;
1931                         }
1932                 } else if (swkey->ip.proto == IPPROTO_SCTP) {
1933                         struct ovs_key_sctp *sctp_key;
1934
1935                         nla = nla_reserve(skb, OVS_KEY_ATTR_SCTP, sizeof(*sctp_key));
1936                         if (!nla)
1937                                 goto nla_put_failure;
1938                         sctp_key = nla_data(nla);
1939                         if (swkey->eth.type == htons(ETH_P_IP)) {
1940                                 sctp_key->sctp_src = swkey->ipv4.tp.src;
1941                                 sctp_key->sctp_dst = swkey->ipv4.tp.dst;
1942                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1943                                 sctp_key->sctp_src = swkey->ipv6.tp.src;
1944                                 sctp_key->sctp_dst = swkey->ipv6.tp.dst;
1945                         }
1946                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1947                            swkey->ip.proto == IPPROTO_ICMP) {
1948                         struct ovs_key_icmp *icmp_key;
1949
1950                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1951                         if (!nla)
1952                                 goto nla_put_failure;
1953                         icmp_key = nla_data(nla);
1954                         icmp_key->icmp_type = ntohs(output->ipv4.tp.src);
1955                         icmp_key->icmp_code = ntohs(output->ipv4.tp.dst);
1956                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1957                            swkey->ip.proto == IPPROTO_ICMPV6) {
1958                         struct ovs_key_icmpv6 *icmpv6_key;
1959
1960                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1961                                                 sizeof(*icmpv6_key));
1962                         if (!nla)
1963                                 goto nla_put_failure;
1964                         icmpv6_key = nla_data(nla);
1965                         icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src);
1966                         icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst);
1967
1968                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1969                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1970                                 struct ovs_key_nd *nd_key;
1971
1972                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1973                                 if (!nla)
1974                                         goto nla_put_failure;
1975                                 nd_key = nla_data(nla);
1976                                 memcpy(nd_key->nd_target, &output->ipv6.nd.target,
1977                                                         sizeof(nd_key->nd_target));
1978                                 memcpy(nd_key->nd_sll, output->ipv6.nd.sll, ETH_ALEN);
1979                                 memcpy(nd_key->nd_tll, output->ipv6.nd.tll, ETH_ALEN);
1980                         }
1981                 }
1982         }
1983
1984 unencap:
1985         if (encap)
1986                 nla_nest_end(skb, encap);
1987
1988         return 0;
1989
1990 nla_put_failure:
1991         return -EMSGSIZE;
1992 }
1993
1994 /* Initializes the flow module.
1995  * Returns zero if successful or a negative error code. */
1996 int ovs_flow_init(void)
1997 {
1998         BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
1999
2000         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
2001                                         0, NULL);
2002         if (flow_cache == NULL)
2003                 return -ENOMEM;
2004
2005         return 0;
2006 }
2007
2008 /* Uninitializes the flow module. */
2009 void ovs_flow_exit(void)
2010 {
2011         kmem_cache_destroy(flow_cache);
2012 }
2013
2014 struct sw_flow_mask *ovs_sw_flow_mask_alloc(void)
2015 {
2016         struct sw_flow_mask *mask;
2017
2018         mask = kmalloc(sizeof(*mask), GFP_KERNEL);
2019         if (mask)
2020                 mask->ref_count = 0;
2021
2022         return mask;
2023 }
2024
2025 void ovs_sw_flow_mask_add_ref(struct sw_flow_mask *mask)
2026 {
2027         mask->ref_count++;
2028 }
2029
2030 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
2031 {
2032         struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
2033
2034         kfree(mask);
2035 }
2036
2037 void ovs_sw_flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
2038 {
2039         if (!mask)
2040                 return;
2041
2042         BUG_ON(!mask->ref_count);
2043         mask->ref_count--;
2044
2045         if (!mask->ref_count) {
2046                 list_del_rcu(&mask->list);
2047                 if (deferred)
2048                         call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
2049                 else
2050                         kfree(mask);
2051         }
2052 }
2053
2054 static bool ovs_sw_flow_mask_equal(const struct sw_flow_mask *a,
2055                 const struct sw_flow_mask *b)
2056 {
2057         u8 *a_ = (u8 *)&a->key + a->range.start;
2058         u8 *b_ = (u8 *)&b->key + b->range.start;
2059
2060         return  (a->range.end == b->range.end)
2061                 && (a->range.start == b->range.start)
2062                 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
2063 }
2064
2065 struct sw_flow_mask *ovs_sw_flow_mask_find(const struct flow_table *tbl,
2066                                            const struct sw_flow_mask *mask)
2067 {
2068         struct list_head *ml;
2069
2070         list_for_each(ml, tbl->mask_list) {
2071                 struct sw_flow_mask *m;
2072                 m = container_of(ml, struct sw_flow_mask, list);
2073                 if (ovs_sw_flow_mask_equal(mask, m))
2074                         return m;
2075         }
2076
2077         return NULL;
2078 }
2079
2080 /**
2081  * add a new mask into the mask list.
2082  * The caller needs to make sure that 'mask' is not the same
2083  * as any masks that are already on the list.
2084  */
2085 void ovs_sw_flow_mask_insert(struct flow_table *tbl, struct sw_flow_mask *mask)
2086 {
2087         list_add_rcu(&mask->list, tbl->mask_list);
2088 }
2089
2090 /**
2091  * Set 'range' fields in the mask to the value of 'val'.
2092  */
2093 static void ovs_sw_flow_mask_set(struct sw_flow_mask *mask,
2094                 struct sw_flow_key_range *range, u8 val)
2095 {
2096         u8 *m = (u8 *)&mask->key + range->start;
2097
2098         mask->range = *range;
2099         memset(m, val, range_n_bytes(range));
2100 }