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