datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira Networks.
   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/if_ether.h>
  36 #include <linux/ip.h>
  37 #include <linux/ipv6.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 int check_header(struct sk_buff *skb, int len)
  52 {
  53         if (unlikely(skb->len < len))
  54                 return -EINVAL;
  55         if (unlikely(!pskb_may_pull(skb, len)))
  56                 return -ENOMEM;
  57         return 0;
  58 }
  59
  60 static bool arphdr_ok(struct sk_buff *skb)
  61 {
  62         return pskb_may_pull(skb, skb_network_offset(skb) +
  63                                   sizeof(struct arp_eth_header));
  64 }
  65
  66 static int check_iphdr(struct sk_buff *skb)
  67 {
  68         unsigned int nh_ofs = skb_network_offset(skb);
  69         unsigned int ip_len;
  70         int err;
  71
  72         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  73         if (unlikely(err))
  74                 return err;
  75
  76         ip_len = ip_hdrlen(skb);
  77         if (unlikely(ip_len < sizeof(struct iphdr) ||
  78                      skb->len < nh_ofs + ip_len))
  79                 return -EINVAL;
  80
  81         skb_set_transport_header(skb, nh_ofs + ip_len);
  82         return 0;
  83 }
  84
  85 static bool tcphdr_ok(struct sk_buff *skb)
  86 {
  87         int th_ofs = skb_transport_offset(skb);
  88         int tcp_len;
  89
  90         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  91                 return false;
  92
  93         tcp_len = tcp_hdrlen(skb);
  94         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  95                      skb->len < th_ofs + tcp_len))
  96                 return false;
  97
  98         return true;
  99 }
 100
 101 static bool udphdr_ok(struct sk_buff *skb)
 102 {
 103         return pskb_may_pull(skb, skb_transport_offset(skb) +
 104                                   sizeof(struct udphdr));
 105 }
 106
 107 static bool icmphdr_ok(struct sk_buff *skb)
 108 {
 109         return pskb_may_pull(skb, skb_transport_offset(skb) +
 110                                   sizeof(struct icmphdr));
 111 }
 112
 113 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 114 {
 115         struct timespec cur_ts;
 116         u64 cur_ms, idle_ms;
 117
 118         ktime_get_ts(&cur_ts);
 119         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 120         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 121                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 122
 123         return cur_ms - idle_ms;
 124 }
 125
 126 #define SW_FLOW_KEY_OFFSET(field)               \
 127         (offsetof(struct sw_flow_key, field) +  \
 128          FIELD_SIZEOF(struct sw_flow_key, field))
 129
 130 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 131                          int *key_lenp)
 132 {
 133         unsigned int nh_ofs = skb_network_offset(skb);
 134         unsigned int nh_len;
 135         int payload_ofs;
 136         struct ipv6hdr *nh;
 137         uint8_t nexthdr;
 138         __be16 frag_off;
 139         int err;
 140
 141         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 142
 143         err = check_header(skb, nh_ofs + sizeof(*nh));
 144         if (unlikely(err))
 145                 return err;
 146
 147         nh = ipv6_hdr(skb);
 148         nexthdr = nh->nexthdr;
 149         payload_ofs = (u8 *)(nh + 1) - skb->data;
 150
 151         key->ip.proto = NEXTHDR_NONE;
 152         key->ip.tos = ipv6_get_dsfield(nh);
 153         key->ip.ttl = nh->hop_limit;
 154         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 155         key->ipv6.addr.src = nh->saddr;
 156         key->ipv6.addr.dst = nh->daddr;
 157
 158         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 159         if (unlikely(payload_ofs < 0))
 160                 return -EINVAL;
 161
 162         if (frag_off) {
 163                 if (frag_off & htons(~0x7))
 164                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 165                 else
 166                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 167         }
 168
 169         nh_len = payload_ofs - nh_ofs;
 170         skb_set_transport_header(skb, nh_ofs + nh_len);
 171         key->ip.proto = nexthdr;
 172         return nh_len;
 173 }
 174
 175 static bool icmp6hdr_ok(struct sk_buff *skb)
 176 {
 177         return pskb_may_pull(skb, skb_transport_offset(skb) +
 178                                   sizeof(struct icmp6hdr));
 179 }
 180
 181 #define TCP_FLAGS_OFFSET 13
 182 #define TCP_FLAG_MASK 0x3f
 183
 184 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 185 {
 186         u8 tcp_flags = 0;
 187
 188         if (flow->key.eth.type == htons(ETH_P_IP) &&
 189             flow->key.ip.proto == IPPROTO_TCP) {
 190                 u8 *tcp = (u8 *)tcp_hdr(skb);
 191                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 192         }
 193
 194         spin_lock(&flow->lock);
 195         flow->used = jiffies;
 196         flow->packet_count++;
 197         flow->byte_count += skb->len;
 198         flow->tcp_flags |= tcp_flags;
 199         spin_unlock(&flow->lock);
 200 }
 201
 202 struct sw_flow_actions *ovs_flow_actions_alloc(const struct nlattr *actions)
 203 {
 204         int actions_len = nla_len(actions);
 205         struct sw_flow_actions *sfa;
 206
 207         /* At least DP_MAX_PORTS actions are required to be able to flood a
 208          * packet to every port.  Factor of 2 allows for setting VLAN tags,
 209          * etc. */
 210         if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
 211                 return ERR_PTR(-EINVAL);
 212
 213         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 214         if (!sfa)
 215                 return ERR_PTR(-ENOMEM);
 216
 217         sfa->actions_len = actions_len;
 218         memcpy(sfa->actions, nla_data(actions), actions_len);
 219         return sfa;
 220 }
 221
 222 struct sw_flow *ovs_flow_alloc(void)
 223 {
 224         struct sw_flow *flow;
 225
 226         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 227         if (!flow)
 228                 return ERR_PTR(-ENOMEM);
 229
 230         spin_lock_init(&flow->lock);
 231         atomic_set(&flow->refcnt, 1);
 232         flow->sf_acts = NULL;
 233         flow->dead = false;
 234
 235         return flow;
 236 }
 237
 238 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 239 {
 240         hash = jhash_1word(hash, table->hash_seed);
 241         return flex_array_get(table->buckets,
 242                                 (hash & (table->n_buckets - 1)));
 243 }
 244
 245 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 246 {
 247         struct flex_array *buckets;
 248         int i, err;
 249
 250         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 251                                    n_buckets, GFP_KERNEL);
 252         if (!buckets)
 253                 return NULL;
 254
 255         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 256         if (err) {
 257                 flex_array_free(buckets);
 258                 return NULL;
 259         }
 260
 261         for (i = 0; i < n_buckets; i++)
 262                 INIT_HLIST_HEAD((struct hlist_head *)
 263                                         flex_array_get(buckets, i));
 264
 265         return buckets;
 266 }
 267
 268 static void free_buckets(struct flex_array *buckets)
 269 {
 270         flex_array_free(buckets);
 271 }
 272
 273 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 274 {
 275         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 276
 277         if (!table)
 278                 return NULL;
 279
 280         table->buckets = alloc_buckets(new_size);
 281
 282         if (!table->buckets) {
 283                 kfree(table);
 284                 return NULL;
 285         }
 286         table->n_buckets = new_size;
 287         table->count = 0;
 288         table->node_ver = 0;
 289         table->keep_flows = false;
 290         get_random_bytes(&table->hash_seed, sizeof(u32));
 291
 292         return table;
 293 }
 294
 295 static void flow_free(struct sw_flow *flow)
 296 {
 297         flow->dead = true;
 298         ovs_flow_put(flow);
 299 }
 300
 301 void ovs_flow_tbl_destroy(struct flow_table *table)
 302 {
 303         int i;
 304
 305         if (!table)
 306                 return;
 307
 308         if (table->keep_flows)
 309                 goto skip_flows;
 310
 311         for (i = 0; i < table->n_buckets; i++) {
 312                 struct sw_flow *flow;
 313                 struct hlist_head *head = flex_array_get(table->buckets, i);
 314                 struct hlist_node *node, *n;
 315                 int ver = table->node_ver;
 316
 317                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 318                         hlist_del_rcu(&flow->hash_node[ver]);
 319                         flow_free(flow);
 320                 }
 321         }
 322
 323 skip_flows:
 324         free_buckets(table->buckets);
 325         kfree(table);
 326 }
 327
 328 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 329 {
 330         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 331
 332         ovs_flow_tbl_destroy(table);
 333 }
 334
 335 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 336 {
 337         if (!table)
 338                 return;
 339
 340         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 341 }
 342
 343 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 344 {
 345         struct sw_flow *flow;
 346         struct hlist_head *head;
 347         struct hlist_node *n;
 348         int ver;
 349         int i;
 350
 351         ver = table->node_ver;
 352         while (*bucket < table->n_buckets) {
 353                 i = 0;
 354                 head = flex_array_get(table->buckets, *bucket);
 355                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 356                         if (i < *last) {
 357                                 i++;
 358                                 continue;
 359                         }
 360                         *last = i + 1;
 361                         return flow;
 362                 }
 363                 (*bucket)++;
 364                 *last = 0;
 365         }
 366
 367         return NULL;
 368 }
 369
 370 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 371 {
 372         int old_ver;
 373         int i;
 374
 375         old_ver = old->node_ver;
 376         new->node_ver = !old_ver;
 377
 378         /* Insert in new table. */
 379         for (i = 0; i < old->n_buckets; i++) {
 380                 struct sw_flow *flow;
 381                 struct hlist_head *head;
 382                 struct hlist_node *n;
 383
 384                 head = flex_array_get(old->buckets, i);
 385
 386                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 387                         ovs_flow_tbl_insert(new, flow);
 388         }
 389         old->keep_flows = true;
 390 }
 391
 392 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 393 {
 394         struct flow_table *new_table;
 395
 396         new_table = ovs_flow_tbl_alloc(n_buckets);
 397         if (!new_table)
 398                 return ERR_PTR(-ENOMEM);
 399
 400         flow_table_copy_flows(table, new_table);
 401
 402         return new_table;
 403 }
 404
 405 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 406 {
 407         return __flow_tbl_rehash(table, table->n_buckets);
 408 }
 409
 410 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 411 {
 412         return __flow_tbl_rehash(table, table->n_buckets * 2);
 413 }
 414
 415 /* RCU callback used by ovs_flow_deferred_free. */
 416 static void rcu_free_flow_callback(struct rcu_head *rcu)
 417 {
 418         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 419
 420         flow->dead = true;
 421         ovs_flow_put(flow);
 422 }
 423
 424 /* Schedules 'flow' to be freed after the next RCU grace period.
 425  * The caller must hold rcu_read_lock for this to be sensible. */
 426 void ovs_flow_deferred_free(struct sw_flow *flow)
 427 {
 428         call_rcu(&flow->rcu, rcu_free_flow_callback);
 429 }
 430
 431 void ovs_flow_hold(struct sw_flow *flow)
 432 {
 433         atomic_inc(&flow->refcnt);
 434 }
 435
 436 void ovs_flow_put(struct sw_flow *flow)
 437 {
 438         if (unlikely(!flow))
 439                 return;
 440
 441         if (atomic_dec_and_test(&flow->refcnt)) {
 442                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 443                 kmem_cache_free(flow_cache, flow);
 444         }
 445 }
 446
 447 /* RCU callback used by ovs_flow_deferred_free_acts. */
 448 static void rcu_free_acts_callback(struct rcu_head *rcu)
 449 {
 450         struct sw_flow_actions *sf_acts = container_of(rcu,
 451                         struct sw_flow_actions, rcu);
 452         kfree(sf_acts);
 453 }
 454
 455 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 456  * The caller must hold rcu_read_lock for this to be sensible. */
 457 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 458 {
 459         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 460 }
 461
 462 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 463 {
 464         struct qtag_prefix {
 465                 __be16 eth_type; /* ETH_P_8021Q */
 466                 __be16 tci;
 467         };
 468         struct qtag_prefix *qp;
 469
 470         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 471                 return 0;
 472
 473         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 474                                          sizeof(__be16))))
 475                 return -ENOMEM;
 476
 477         qp = (struct qtag_prefix *) skb->data;
 478         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 479         __skb_pull(skb, sizeof(struct qtag_prefix));
 480
 481         return 0;
 482 }
 483
 484 static __be16 parse_ethertype(struct sk_buff *skb)
 485 {
 486         struct llc_snap_hdr {
 487                 u8  dsap;  /* Always 0xAA */
 488                 u8  ssap;  /* Always 0xAA */
 489                 u8  ctrl;
 490                 u8  oui[3];
 491                 __be16 ethertype;
 492         };
 493         struct llc_snap_hdr *llc;
 494         __be16 proto;
 495
 496         proto = *(__be16 *) skb->data;
 497         __skb_pull(skb, sizeof(__be16));
 498
 499         if (ntohs(proto) >= 1536)
 500                 return proto;
 501
 502         if (skb->len < sizeof(struct llc_snap_hdr))
 503                 return htons(ETH_P_802_2);
 504
 505         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 506                 return htons(0);
 507
 508         llc = (struct llc_snap_hdr *) skb->data;
 509         if (llc->dsap != LLC_SAP_SNAP ||
 510             llc->ssap != LLC_SAP_SNAP ||
 511             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 512                 return htons(ETH_P_802_2);
 513
 514         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 515         return llc->ethertype;
 516 }
 517
 518 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 519                         int *key_lenp, int nh_len)
 520 {
 521         struct icmp6hdr *icmp = icmp6_hdr(skb);
 522         int error = 0;
 523         int key_len;
 524
 525         /* The ICMPv6 type and code fields use the 16-bit transport port
 526          * fields, so we need to store them in 16-bit network byte order.
 527          */
 528         key->ipv6.tp.src = htons(icmp->icmp6_type);
 529         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 530         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 531
 532         if (icmp->icmp6_code == 0 &&
 533             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 534              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 535                 int icmp_len = skb->len - skb_transport_offset(skb);
 536                 struct nd_msg *nd;
 537                 int offset;
 538
 539                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 540
 541                 /* In order to process neighbor discovery options, we need the
 542                  * entire packet.
 543                  */
 544                 if (unlikely(icmp_len < sizeof(*nd)))
 545                         goto out;
 546                 if (unlikely(skb_linearize(skb))) {
 547                         error = -ENOMEM;
 548                         goto out;
 549                 }
 550
 551                 nd = (struct nd_msg *)skb_transport_header(skb);
 552                 key->ipv6.nd.target = nd->target;
 553                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 554
 555                 icmp_len -= sizeof(*nd);
 556                 offset = 0;
 557                 while (icmp_len >= 8) {
 558                         struct nd_opt_hdr *nd_opt =
 559                                  (struct nd_opt_hdr *)(nd->opt + offset);
 560                         int opt_len = nd_opt->nd_opt_len * 8;
 561
 562                         if (unlikely(!opt_len || opt_len > icmp_len))
 563                                 goto invalid;
 564
 565                         /* Store the link layer address if the appropriate
 566                          * option is provided.  It is considered an error if
 567                          * the same link layer option is specified twice.
 568                          */
 569                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 570                             && opt_len == 8) {
 571                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 572                                         goto invalid;
 573                                 memcpy(key->ipv6.nd.sll,
 574                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 575                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 576                                    && opt_len == 8) {
 577                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 578                                         goto invalid;
 579                                 memcpy(key->ipv6.nd.tll,
 580                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 581                         }
 582
 583                         icmp_len -= opt_len;
 584                         offset += opt_len;
 585                 }
 586         }
 587
 588         goto out;
 589
 590 invalid:
 591         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 592         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 593         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 594
 595 out:
 596         *key_lenp = key_len;
 597         return error;
 598 }
 599
 600 /**
 601  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 602  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 603  * Ethernet header
 604  * @in_port: port number on which @skb was received.
 605  * @key: output flow key
 606  * @key_lenp: length of output flow key
 607  *
 608  * The caller must ensure that skb->len >= ETH_HLEN.
 609  *
 610  * Returns 0 if successful, otherwise a negative errno value.
 611  *
 612  * Initializes @skb header pointers as follows:
 613  *
 614  *    - skb->mac_header: the Ethernet header.
 615  *
 616  *    - skb->network_header: just past the Ethernet header, or just past the
 617  *      VLAN header, to the first byte of the Ethernet payload.
 618  *
 619  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 620  *      on output, then just past the IP header, if one is present and
 621  *      of a correct length, otherwise the same as skb->network_header.
 622  *      For other key->dl_type values it is left untouched.
 623  */
 624 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 625                  int *key_lenp)
 626 {
 627         int error = 0;
 628         int key_len = SW_FLOW_KEY_OFFSET(eth);
 629         struct ethhdr *eth;
 630
 631         memset(key, 0, sizeof(*key));
 632
 633         key->phy.priority = skb->priority;
 634         key->phy.tun_id = OVS_CB(skb)->tun_id;
 635         key->phy.in_port = in_port;
 636
 637         skb_reset_mac_header(skb);
 638
 639         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 640          * header in the linear data area.
 641          */
 642         eth = eth_hdr(skb);
 643         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 644         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 645
 646         __skb_pull(skb, 2 * ETH_ALEN);
 647
 648         if (vlan_tx_tag_present(skb))
 649                 key->eth.tci = htons(vlan_get_tci(skb));
 650         else if (eth->h_proto == htons(ETH_P_8021Q))
 651                 if (unlikely(parse_vlan(skb, key)))
 652                         return -ENOMEM;
 653
 654         key->eth.type = parse_ethertype(skb);
 655         if (unlikely(key->eth.type == htons(0)))
 656                 return -ENOMEM;
 657
 658         skb_reset_network_header(skb);
 659         __skb_push(skb, skb->data - skb_mac_header(skb));
 660
 661         /* Network layer. */
 662         if (key->eth.type == htons(ETH_P_IP)) {
 663                 struct iphdr *nh;
 664                 __be16 offset;
 665
 666                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 667
 668                 error = check_iphdr(skb);
 669                 if (unlikely(error)) {
 670                         if (error == -EINVAL) {
 671                                 skb->transport_header = skb->network_header;
 672                                 error = 0;
 673                         }
 674                         goto out;
 675                 }
 676
 677                 nh = ip_hdr(skb);
 678                 key->ipv4.addr.src = nh->saddr;
 679                 key->ipv4.addr.dst = nh->daddr;
 680
 681                 key->ip.proto = nh->protocol;
 682                 key->ip.tos = nh->tos;
 683                 key->ip.ttl = nh->ttl;
 684
 685                 offset = nh->frag_off & htons(IP_OFFSET);
 686                 if (offset) {
 687                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 688                         goto out;
 689                 }
 690                 if (nh->frag_off & htons(IP_MF) ||
 691                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 692                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 693
 694                 /* Transport layer. */
 695                 if (key->ip.proto == IPPROTO_TCP) {
 696                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 697                         if (tcphdr_ok(skb)) {
 698                                 struct tcphdr *tcp = tcp_hdr(skb);
 699                                 key->ipv4.tp.src = tcp->source;
 700                                 key->ipv4.tp.dst = tcp->dest;
 701                         }
 702                 } else if (key->ip.proto == IPPROTO_UDP) {
 703                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 704                         if (udphdr_ok(skb)) {
 705                                 struct udphdr *udp = udp_hdr(skb);
 706                                 key->ipv4.tp.src = udp->source;
 707                                 key->ipv4.tp.dst = udp->dest;
 708                         }
 709                 } else if (key->ip.proto == IPPROTO_ICMP) {
 710                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 711                         if (icmphdr_ok(skb)) {
 712                                 struct icmphdr *icmp = icmp_hdr(skb);
 713                                 /* The ICMP type and code fields use the 16-bit
 714                                  * transport port fields, so we need to store
 715                                  * them in 16-bit network byte order. */
 716                                 key->ipv4.tp.src = htons(icmp->type);
 717                                 key->ipv4.tp.dst = htons(icmp->code);
 718                         }
 719                 }
 720
 721         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 722                 struct arp_eth_header *arp;
 723
 724                 arp = (struct arp_eth_header *)skb_network_header(skb);
 725
 726                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 727                                 && arp->ar_pro == htons(ETH_P_IP)
 728                                 && arp->ar_hln == ETH_ALEN
 729                                 && arp->ar_pln == 4) {
 730
 731                         /* We only match on the lower 8 bits of the opcode. */
 732                         if (ntohs(arp->ar_op) <= 0xff)
 733                                 key->ip.proto = ntohs(arp->ar_op);
 734
 735                         if (key->ip.proto == ARPOP_REQUEST
 736                                         || key->ip.proto == ARPOP_REPLY) {
 737                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 738                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 739                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 740                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 741                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 742                         }
 743                 }
 744         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 745                 int nh_len;             /* IPv6 Header + Extensions */
 746
 747                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 748                 if (unlikely(nh_len < 0)) {
 749                         if (nh_len == -EINVAL)
 750                                 skb->transport_header = skb->network_header;
 751                         else
 752                                 error = nh_len;
 753                         goto out;
 754                 }
 755
 756                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 757                         goto out;
 758                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 759                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 760
 761                 /* Transport layer. */
 762                 if (key->ip.proto == NEXTHDR_TCP) {
 763                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 764                         if (tcphdr_ok(skb)) {
 765                                 struct tcphdr *tcp = tcp_hdr(skb);
 766                                 key->ipv6.tp.src = tcp->source;
 767                                 key->ipv6.tp.dst = tcp->dest;
 768                         }
 769                 } else if (key->ip.proto == NEXTHDR_UDP) {
 770                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 771                         if (udphdr_ok(skb)) {
 772                                 struct udphdr *udp = udp_hdr(skb);
 773                                 key->ipv6.tp.src = udp->source;
 774                                 key->ipv6.tp.dst = udp->dest;
 775                         }
 776                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 777                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 778                         if (icmp6hdr_ok(skb)) {
 779                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 780                                 if (error < 0)
 781                                         goto out;
 782                         }
 783                 }
 784         }
 785
 786 out:
 787         *key_lenp = key_len;
 788         return error;
 789 }
 790
 791 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 792 {
 793         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 794 }
 795
 796 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 797                                 struct sw_flow_key *key, int key_len)
 798 {
 799         struct sw_flow *flow;
 800         struct hlist_node *n;
 801         struct hlist_head *head;
 802         u32 hash;
 803
 804         hash = ovs_flow_hash(key, key_len);
 805
 806         head = find_bucket(table, hash);
 807         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 808
 809                 if (flow->hash == hash &&
 810                     !memcmp(&flow->key, key, key_len)) {
 811                         return flow;
 812                 }
 813         }
 814         return NULL;
 815 }
 816
 817 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 818 {
 819         struct hlist_head *head;
 820
 821         head = find_bucket(table, flow->hash);
 822         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 823         table->count++;
 824 }
 825
 826 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 827 {
 828         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 829         table->count--;
 830         BUG_ON(table->count < 0);
 831 }
 832
 833 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 834 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 835         [OVS_KEY_ATTR_ENCAP] = -1,
 836         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 837         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 838         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 839         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 840         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 841         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 842         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 843         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 844         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 845         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 846         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 847         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 848         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 849
 850         /* Not upstream. */
 851         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 852 };
 853
 854 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 855                                   const struct nlattr *a[], u64 *attrs)
 856 {
 857         const struct ovs_key_icmp *icmp_key;
 858         const struct ovs_key_tcp *tcp_key;
 859         const struct ovs_key_udp *udp_key;
 860
 861         switch (swkey->ip.proto) {
 862         case IPPROTO_TCP:
 863                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 864                         return -EINVAL;
 865                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 866
 867                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 868                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 869                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 870                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 871                 break;
 872
 873         case IPPROTO_UDP:
 874                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 875                         return -EINVAL;
 876                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 877
 878                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 879                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 880                 swkey->ipv4.tp.src = udp_key->udp_src;
 881                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 882                 break;
 883
 884         case IPPROTO_ICMP:
 885                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 886                         return -EINVAL;
 887                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 888
 889                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 890                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 891                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 892                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 893                 break;
 894         }
 895
 896         return 0;
 897 }
 898
 899 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 900                                   const struct nlattr *a[], u64 *attrs)
 901 {
 902         const struct ovs_key_icmpv6 *icmpv6_key;
 903         const struct ovs_key_tcp *tcp_key;
 904         const struct ovs_key_udp *udp_key;
 905
 906         switch (swkey->ip.proto) {
 907         case IPPROTO_TCP:
 908                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 909                         return -EINVAL;
 910                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 911
 912                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 913                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 914                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 915                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 916                 break;
 917
 918         case IPPROTO_UDP:
 919                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 920                         return -EINVAL;
 921                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 922
 923                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 924                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 925                 swkey->ipv6.tp.src = udp_key->udp_src;
 926                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 927                 break;
 928
 929         case IPPROTO_ICMPV6:
 930                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 931                         return -EINVAL;
 932                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 933
 934                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 935                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 936                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 937                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 938
 939                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 940                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 941                         const struct ovs_key_nd *nd_key;
 942
 943                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 944                                 return -EINVAL;
 945                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 946
 947                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 948                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 949                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 950                                sizeof(swkey->ipv6.nd.target));
 951                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 952                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 953                 }
 954                 break;
 955         }
 956
 957         return 0;
 958 }
 959
 960 static int parse_flow_nlattrs(const struct nlattr *attr,
 961                               const struct nlattr *a[], u64 *attrsp)
 962 {
 963         const struct nlattr *nla;
 964         u64 attrs;
 965         int rem;
 966
 967         attrs = 0;
 968         nla_for_each_nested(nla, attr, rem) {
 969                 u16 type = nla_type(nla);
 970                 int expected_len;
 971
 972                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 973                         return -EINVAL;
 974
 975                 expected_len = ovs_key_lens[type];
 976                 if (nla_len(nla) != expected_len && expected_len != -1)
 977                         return -EINVAL;
 978
 979                 attrs |= 1ULL << type;
 980                 a[type] = nla;
 981         }
 982         if (rem)
 983                 return -EINVAL;
 984
 985         *attrsp = attrs;
 986         return 0;
 987 }
 988
 989 /**
 990  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 991  * @swkey: receives the extracted flow key.
 992  * @key_lenp: number of bytes used in @swkey.
 993  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 994  * sequence.
 995  */
 996 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 997                       const struct nlattr *attr)
 998 {
 999         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1000         const struct ovs_key_ethernet *eth_key;
1001         int key_len;
1002         u64 attrs;
1003         int err;
1004
1005         memset(swkey, 0, sizeof(struct sw_flow_key));
1006         key_len = SW_FLOW_KEY_OFFSET(eth);
1007
1008         err = parse_flow_nlattrs(attr, a, &attrs);
1009         if (err)
1010                 return err;
1011
1012         /* Metadata attributes. */
1013         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1014                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1015                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1016         }
1017         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1018                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1019                 if (in_port >= DP_MAX_PORTS)
1020                         return -EINVAL;
1021                 swkey->phy.in_port = in_port;
1022                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1023         } else {
1024                 swkey->phy.in_port = USHRT_MAX;
1025         }
1026
1027         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
1028                 swkey->phy.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1029                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1030         }
1031
1032         /* Data attributes. */
1033         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1034                 return -EINVAL;
1035         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1036
1037         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1038         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1039         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1040
1041         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1042             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1043                 const struct nlattr *encap;
1044                 __be16 tci;
1045
1046                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1047                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1048                               (1 << OVS_KEY_ATTR_ENCAP)))
1049                         return -EINVAL;
1050
1051                 encap = a[OVS_KEY_ATTR_ENCAP];
1052                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1053                 if (tci & htons(VLAN_TAG_PRESENT)) {
1054                         swkey->eth.tci = tci;
1055
1056                         err = parse_flow_nlattrs(encap, a, &attrs);
1057                         if (err)
1058                                 return err;
1059                 } else if (!tci) {
1060                         /* Corner case for truncated 802.1Q header. */
1061                         if (nla_len(encap))
1062                                 return -EINVAL;
1063
1064                         swkey->eth.type = htons(ETH_P_8021Q);
1065                         *key_lenp = key_len;
1066                         return 0;
1067                 } else {
1068                         return -EINVAL;
1069                 }
1070         }
1071
1072         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1073                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1074                 if (ntohs(swkey->eth.type) < 1536)
1075                         return -EINVAL;
1076                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1077         } else {
1078                 swkey->eth.type = htons(ETH_P_802_2);
1079         }
1080
1081         if (swkey->eth.type == htons(ETH_P_IP)) {
1082                 const struct ovs_key_ipv4 *ipv4_key;
1083
1084                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1085                         return -EINVAL;
1086                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1087
1088                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1089                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1090                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1091                         return -EINVAL;
1092                 swkey->ip.proto = ipv4_key->ipv4_proto;
1093                 swkey->ip.tos = ipv4_key->ipv4_tos;
1094                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1095                 swkey->ip.frag = ipv4_key->ipv4_frag;
1096                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1097                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1098
1099                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1100                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1101                         if (err)
1102                                 return err;
1103                 }
1104         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1105                 const struct ovs_key_ipv6 *ipv6_key;
1106
1107                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1108                         return -EINVAL;
1109                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1110
1111                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1112                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1113                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1114                         return -EINVAL;
1115                 swkey->ipv6.label = ipv6_key->ipv6_label;
1116                 swkey->ip.proto = ipv6_key->ipv6_proto;
1117                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1118                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1119                 swkey->ip.frag = ipv6_key->ipv6_frag;
1120                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1121                        sizeof(swkey->ipv6.addr.src));
1122                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1123                        sizeof(swkey->ipv6.addr.dst));
1124
1125                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1126                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1127                         if (err)
1128                                 return err;
1129                 }
1130         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1131                 const struct ovs_key_arp *arp_key;
1132
1133                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1134                         return -EINVAL;
1135                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1136
1137                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1138                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1139                 swkey->ipv4.addr.src = arp_key->arp_sip;
1140                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1141                 if (arp_key->arp_op & htons(0xff00))
1142                         return -EINVAL;
1143                 swkey->ip.proto = ntohs(arp_key->arp_op);
1144                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1145                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1146         }
1147
1148         if (attrs)
1149                 return -EINVAL;
1150         *key_lenp = key_len;
1151
1152         return 0;
1153 }
1154
1155 /**
1156  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1157  * @in_port: receives the extracted input port.
1158  * @tun_id: receives the extracted tunnel ID.
1159  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1160  * sequence.
1161  *
1162  * This parses a series of Netlink attributes that form a flow key, which must
1163  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1164  * get the metadata, that is, the parts of the flow key that cannot be
1165  * extracted from the packet itself.
1166  */
1167 int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1168                                    const struct nlattr *attr)
1169 {
1170         const struct nlattr *nla;
1171         int rem;
1172
1173         *in_port = USHRT_MAX;
1174         *tun_id = 0;
1175         *priority = 0;
1176
1177         nla_for_each_nested(nla, attr, rem) {
1178                 int type = nla_type(nla);
1179
1180                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1181                         if (nla_len(nla) != ovs_key_lens[type])
1182                                 return -EINVAL;
1183
1184                         switch (type) {
1185                         case OVS_KEY_ATTR_PRIORITY:
1186                                 *priority = nla_get_u32(nla);
1187                                 break;
1188
1189                         case OVS_KEY_ATTR_TUN_ID:
1190                                 *tun_id = nla_get_be64(nla);
1191                                 break;
1192
1193                         case OVS_KEY_ATTR_IN_PORT:
1194                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1195                                         return -EINVAL;
1196                                 *in_port = nla_get_u32(nla);
1197                                 break;
1198                         }
1199                 }
1200         }
1201         if (rem)
1202                 return -EINVAL;
1203         return 0;
1204 }
1205
1206 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1207 {
1208         struct ovs_key_ethernet *eth_key;
1209         struct nlattr *nla, *encap;
1210
1211         if (swkey->phy.priority)
1212                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1213
1214         if (swkey->phy.tun_id != cpu_to_be64(0))
1215                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1216
1217         if (swkey->phy.in_port != USHRT_MAX)
1218                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1219
1220         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1221         if (!nla)
1222                 goto nla_put_failure;
1223         eth_key = nla_data(nla);
1224         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1225         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1226
1227         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1228                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q));
1229                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci);
1230                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1231                 if (!swkey->eth.tci)
1232                         goto unencap;
1233         } else {
1234                 encap = NULL;
1235         }
1236
1237         if (swkey->eth.type == htons(ETH_P_802_2))
1238                 goto unencap;
1239
1240         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1241
1242         if (swkey->eth.type == htons(ETH_P_IP)) {
1243                 struct ovs_key_ipv4 *ipv4_key;
1244
1245                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1246                 if (!nla)
1247                         goto nla_put_failure;
1248                 ipv4_key = nla_data(nla);
1249                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1250                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1251                 ipv4_key->ipv4_proto = swkey->ip.proto;
1252                 ipv4_key->ipv4_tos = swkey->ip.tos;
1253                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1254                 ipv4_key->ipv4_frag = swkey->ip.frag;
1255         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1256                 struct ovs_key_ipv6 *ipv6_key;
1257
1258                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1259                 if (!nla)
1260                         goto nla_put_failure;
1261                 ipv6_key = nla_data(nla);
1262                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1263                                 sizeof(ipv6_key->ipv6_src));
1264                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1265                                 sizeof(ipv6_key->ipv6_dst));
1266                 ipv6_key->ipv6_label = swkey->ipv6.label;
1267                 ipv6_key->ipv6_proto = swkey->ip.proto;
1268                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1269                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1270                 ipv6_key->ipv6_frag = swkey->ip.frag;
1271         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1272                 struct ovs_key_arp *arp_key;
1273
1274                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1275                 if (!nla)
1276                         goto nla_put_failure;
1277                 arp_key = nla_data(nla);
1278                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1279                 arp_key->arp_sip = swkey->ipv4.addr.src;
1280                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1281                 arp_key->arp_op = htons(swkey->ip.proto);
1282                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1283                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1284         }
1285
1286         if ((swkey->eth.type == htons(ETH_P_IP) ||
1287              swkey->eth.type == htons(ETH_P_IPV6)) &&
1288              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1289
1290                 if (swkey->ip.proto == IPPROTO_TCP) {
1291                         struct ovs_key_tcp *tcp_key;
1292
1293                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1294                         if (!nla)
1295                                 goto nla_put_failure;
1296                         tcp_key = nla_data(nla);
1297                         if (swkey->eth.type == htons(ETH_P_IP)) {
1298                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1299                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1300                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1301                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1302                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1303                         }
1304                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1305                         struct ovs_key_udp *udp_key;
1306
1307                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1308                         if (!nla)
1309                                 goto nla_put_failure;
1310                         udp_key = nla_data(nla);
1311                         if (swkey->eth.type == htons(ETH_P_IP)) {
1312                                 udp_key->udp_src = swkey->ipv4.tp.src;
1313                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1314                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1315                                 udp_key->udp_src = swkey->ipv6.tp.src;
1316                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1317                         }
1318                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1319                            swkey->ip.proto == IPPROTO_ICMP) {
1320                         struct ovs_key_icmp *icmp_key;
1321
1322                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1323                         if (!nla)
1324                                 goto nla_put_failure;
1325                         icmp_key = nla_data(nla);
1326                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1327                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1328                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1329                            swkey->ip.proto == IPPROTO_ICMPV6) {
1330                         struct ovs_key_icmpv6 *icmpv6_key;
1331
1332                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1333                                                 sizeof(*icmpv6_key));
1334                         if (!nla)
1335                                 goto nla_put_failure;
1336                         icmpv6_key = nla_data(nla);
1337                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1338                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1339
1340                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1341                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1342                                 struct ovs_key_nd *nd_key;
1343
1344                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1345                                 if (!nla)
1346                                         goto nla_put_failure;
1347                                 nd_key = nla_data(nla);
1348                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1349                                                         sizeof(nd_key->nd_target));
1350                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1351                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1352                         }
1353                 }
1354         }
1355
1356 unencap:
1357         if (encap)
1358                 nla_nest_end(skb, encap);
1359
1360         return 0;
1361
1362 nla_put_failure:
1363         return -EMSGSIZE;
1364 }
1365
1366 /* Initializes the flow module.
1367  * Returns zero if successful or a negative error code. */
1368 int ovs_flow_init(void)
1369 {
1370         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1371                                         0, NULL);
1372         if (flow_cache == NULL)
1373                 return -ENOMEM;
1374
1375         return 0;
1376 }
1377
1378 /* Uninitializes the flow module. */
1379 void ovs_flow_exit(void)
1380 {
1381         kmem_cache_destroy(flow_cache);
1382 }