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/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 #include "vlan.h"
  47
  48 static struct kmem_cache *flow_cache;
  49
  50 static int check_header(struct sk_buff *skb, int len)
  51 {
  52         if (unlikely(skb->len < len))
  53                 return -EINVAL;
  54         if (unlikely(!pskb_may_pull(skb, len)))
  55                 return -ENOMEM;
  56         return 0;
  57 }
  58
  59 static bool arphdr_ok(struct sk_buff *skb)
  60 {
  61         return pskb_may_pull(skb, skb_network_offset(skb) +
  62                                   sizeof(struct arp_eth_header));
  63 }
  64
  65 static int check_iphdr(struct sk_buff *skb)
  66 {
  67         unsigned int nh_ofs = skb_network_offset(skb);
  68         unsigned int ip_len;
  69         int err;
  70
  71         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  72         if (unlikely(err))
  73                 return err;
  74
  75         ip_len = ip_hdrlen(skb);
  76         if (unlikely(ip_len < sizeof(struct iphdr) ||
  77                      skb->len < nh_ofs + ip_len))
  78                 return -EINVAL;
  79
  80         skb_set_transport_header(skb, nh_ofs + ip_len);
  81         return 0;
  82 }
  83
  84 static bool tcphdr_ok(struct sk_buff *skb)
  85 {
  86         int th_ofs = skb_transport_offset(skb);
  87         int tcp_len;
  88
  89         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  90                 return false;
  91
  92         tcp_len = tcp_hdrlen(skb);
  93         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  94                      skb->len < th_ofs + tcp_len))
  95                 return false;
  96
  97         return true;
  98 }
  99
 100 static bool udphdr_ok(struct sk_buff *skb)
 101 {
 102         return pskb_may_pull(skb, skb_transport_offset(skb) +
 103                                   sizeof(struct udphdr));
 104 }
 105
 106 static bool icmphdr_ok(struct sk_buff *skb)
 107 {
 108         return pskb_may_pull(skb, skb_transport_offset(skb) +
 109                                   sizeof(struct icmphdr));
 110 }
 111
 112 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 113 {
 114         struct timespec cur_ts;
 115         u64 cur_ms, idle_ms;
 116
 117         ktime_get_ts(&cur_ts);
 118         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 119         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 120                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 121
 122         return cur_ms - idle_ms;
 123 }
 124
 125 #define SW_FLOW_KEY_OFFSET(field)               \
 126         (offsetof(struct sw_flow_key, field) +  \
 127          FIELD_SIZEOF(struct sw_flow_key, field))
 128
 129 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 130                          int *key_lenp)
 131 {
 132         unsigned int nh_ofs = skb_network_offset(skb);
 133         unsigned int nh_len;
 134         int payload_ofs;
 135         struct ipv6hdr *nh;
 136         uint8_t nexthdr;
 137         __be16 frag_off;
 138         int err;
 139
 140         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 141
 142         err = check_header(skb, nh_ofs + sizeof(*nh));
 143         if (unlikely(err))
 144                 return err;
 145
 146         nh = ipv6_hdr(skb);
 147         nexthdr = nh->nexthdr;
 148         payload_ofs = (u8 *)(nh + 1) - skb->data;
 149
 150         key->ip.proto = NEXTHDR_NONE;
 151         key->ip.tos = ipv6_get_dsfield(nh);
 152         key->ip.ttl = nh->hop_limit;
 153         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 154         key->ipv6.addr.src = nh->saddr;
 155         key->ipv6.addr.dst = nh->daddr;
 156
 157         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 158         if (unlikely(payload_ofs < 0))
 159                 return -EINVAL;
 160
 161         if (frag_off) {
 162                 if (frag_off & htons(~0x7))
 163                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 164                 else
 165                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 166         }
 167
 168         nh_len = payload_ofs - nh_ofs;
 169         skb_set_transport_header(skb, nh_ofs + nh_len);
 170         key->ip.proto = nexthdr;
 171         return nh_len;
 172 }
 173
 174 static bool icmp6hdr_ok(struct sk_buff *skb)
 175 {
 176         return pskb_may_pull(skb, skb_transport_offset(skb) +
 177                                   sizeof(struct icmp6hdr));
 178 }
 179
 180 #define TCP_FLAGS_OFFSET 13
 181 #define TCP_FLAG_MASK 0x3f
 182
 183 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 184 {
 185         u8 tcp_flags = 0;
 186
 187         if (flow->key.eth.type == htons(ETH_P_IP) &&
 188             flow->key.ip.proto == IPPROTO_TCP &&
 189             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 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         if (actions_len > MAX_ACTIONS_BUFSIZE)
 208                 return ERR_PTR(-EINVAL);
 209
 210         sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
 211         if (!sfa)
 212                 return ERR_PTR(-ENOMEM);
 213
 214         sfa->actions_len = actions_len;
 215         memcpy(sfa->actions, nla_data(actions), actions_len);
 216         return sfa;
 217 }
 218
 219 struct sw_flow *ovs_flow_alloc(void)
 220 {
 221         struct sw_flow *flow;
 222
 223         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 224         if (!flow)
 225                 return ERR_PTR(-ENOMEM);
 226
 227         spin_lock_init(&flow->lock);
 228         atomic_set(&flow->refcnt, 1);
 229         flow->sf_acts = NULL;
 230         flow->dead = false;
 231
 232         return flow;
 233 }
 234
 235 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 236 {
 237         hash = jhash_1word(hash, table->hash_seed);
 238         return flex_array_get(table->buckets,
 239                                 (hash & (table->n_buckets - 1)));
 240 }
 241
 242 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 243 {
 244         struct flex_array *buckets;
 245         int i, err;
 246
 247         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 248                                    n_buckets, GFP_KERNEL);
 249         if (!buckets)
 250                 return NULL;
 251
 252         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 253         if (err) {
 254                 flex_array_free(buckets);
 255                 return NULL;
 256         }
 257
 258         for (i = 0; i < n_buckets; i++)
 259                 INIT_HLIST_HEAD((struct hlist_head *)
 260                                         flex_array_get(buckets, i));
 261
 262         return buckets;
 263 }
 264
 265 static void free_buckets(struct flex_array *buckets)
 266 {
 267         flex_array_free(buckets);
 268 }
 269
 270 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 271 {
 272         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 273
 274         if (!table)
 275                 return NULL;
 276
 277         table->buckets = alloc_buckets(new_size);
 278
 279         if (!table->buckets) {
 280                 kfree(table);
 281                 return NULL;
 282         }
 283         table->n_buckets = new_size;
 284         table->count = 0;
 285         table->node_ver = 0;
 286         table->keep_flows = false;
 287         get_random_bytes(&table->hash_seed, sizeof(u32));
 288
 289         return table;
 290 }
 291
 292 static void flow_free(struct sw_flow *flow)
 293 {
 294         flow->dead = true;
 295         ovs_flow_put(flow);
 296 }
 297
 298 void ovs_flow_tbl_destroy(struct flow_table *table)
 299 {
 300         int i;
 301
 302         if (!table)
 303                 return;
 304
 305         if (table->keep_flows)
 306                 goto skip_flows;
 307
 308         for (i = 0; i < table->n_buckets; i++) {
 309                 struct sw_flow *flow;
 310                 struct hlist_head *head = flex_array_get(table->buckets, i);
 311                 struct hlist_node *node, *n;
 312                 int ver = table->node_ver;
 313
 314                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 315                         hlist_del_rcu(&flow->hash_node[ver]);
 316                         flow_free(flow);
 317                 }
 318         }
 319
 320 skip_flows:
 321         free_buckets(table->buckets);
 322         kfree(table);
 323 }
 324
 325 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 326 {
 327         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 328
 329         ovs_flow_tbl_destroy(table);
 330 }
 331
 332 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 333 {
 334         if (!table)
 335                 return;
 336
 337         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 338 }
 339
 340 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 341 {
 342         struct sw_flow *flow;
 343         struct hlist_head *head;
 344         struct hlist_node *n;
 345         int ver;
 346         int i;
 347
 348         ver = table->node_ver;
 349         while (*bucket < table->n_buckets) {
 350                 i = 0;
 351                 head = flex_array_get(table->buckets, *bucket);
 352                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 353                         if (i < *last) {
 354                                 i++;
 355                                 continue;
 356                         }
 357                         *last = i + 1;
 358                         return flow;
 359                 }
 360                 (*bucket)++;
 361                 *last = 0;
 362         }
 363
 364         return NULL;
 365 }
 366
 367 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 368 {
 369         int old_ver;
 370         int i;
 371
 372         old_ver = old->node_ver;
 373         new->node_ver = !old_ver;
 374
 375         /* Insert in new table. */
 376         for (i = 0; i < old->n_buckets; i++) {
 377                 struct sw_flow *flow;
 378                 struct hlist_head *head;
 379                 struct hlist_node *n;
 380
 381                 head = flex_array_get(old->buckets, i);
 382
 383                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 384                         ovs_flow_tbl_insert(new, flow);
 385         }
 386         old->keep_flows = true;
 387 }
 388
 389 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 390 {
 391         struct flow_table *new_table;
 392
 393         new_table = ovs_flow_tbl_alloc(n_buckets);
 394         if (!new_table)
 395                 return ERR_PTR(-ENOMEM);
 396
 397         flow_table_copy_flows(table, new_table);
 398
 399         return new_table;
 400 }
 401
 402 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 403 {
 404         return __flow_tbl_rehash(table, table->n_buckets);
 405 }
 406
 407 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 408 {
 409         return __flow_tbl_rehash(table, table->n_buckets * 2);
 410 }
 411
 412 /* RCU callback used by ovs_flow_deferred_free. */
 413 static void rcu_free_flow_callback(struct rcu_head *rcu)
 414 {
 415         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 416
 417         flow->dead = true;
 418         ovs_flow_put(flow);
 419 }
 420
 421 /* Schedules 'flow' to be freed after the next RCU grace period.
 422  * The caller must hold rcu_read_lock for this to be sensible. */
 423 void ovs_flow_deferred_free(struct sw_flow *flow)
 424 {
 425         call_rcu(&flow->rcu, rcu_free_flow_callback);
 426 }
 427
 428 void ovs_flow_hold(struct sw_flow *flow)
 429 {
 430         atomic_inc(&flow->refcnt);
 431 }
 432
 433 void ovs_flow_put(struct sw_flow *flow)
 434 {
 435         if (unlikely(!flow))
 436                 return;
 437
 438         if (atomic_dec_and_test(&flow->refcnt)) {
 439                 kfree((struct sf_flow_acts __force *)flow->sf_acts);
 440                 kmem_cache_free(flow_cache, flow);
 441         }
 442 }
 443
 444 /* RCU callback used by ovs_flow_deferred_free_acts. */
 445 static void rcu_free_acts_callback(struct rcu_head *rcu)
 446 {
 447         struct sw_flow_actions *sf_acts = container_of(rcu,
 448                         struct sw_flow_actions, rcu);
 449         kfree(sf_acts);
 450 }
 451
 452 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 453  * The caller must hold rcu_read_lock for this to be sensible. */
 454 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 455 {
 456         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 457 }
 458
 459 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 460 {
 461         struct qtag_prefix {
 462                 __be16 eth_type; /* ETH_P_8021Q */
 463                 __be16 tci;
 464         };
 465         struct qtag_prefix *qp;
 466
 467         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 468                 return 0;
 469
 470         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 471                                          sizeof(__be16))))
 472                 return -ENOMEM;
 473
 474         qp = (struct qtag_prefix *) skb->data;
 475         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 476         __skb_pull(skb, sizeof(struct qtag_prefix));
 477
 478         return 0;
 479 }
 480
 481 static __be16 parse_ethertype(struct sk_buff *skb)
 482 {
 483         struct llc_snap_hdr {
 484                 u8  dsap;  /* Always 0xAA */
 485                 u8  ssap;  /* Always 0xAA */
 486                 u8  ctrl;
 487                 u8  oui[3];
 488                 __be16 ethertype;
 489         };
 490         struct llc_snap_hdr *llc;
 491         __be16 proto;
 492
 493         proto = *(__be16 *) skb->data;
 494         __skb_pull(skb, sizeof(__be16));
 495
 496         if (ntohs(proto) >= 1536)
 497                 return proto;
 498
 499         if (skb->len < sizeof(struct llc_snap_hdr))
 500                 return htons(ETH_P_802_2);
 501
 502         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 503                 return htons(0);
 504
 505         llc = (struct llc_snap_hdr *) skb->data;
 506         if (llc->dsap != LLC_SAP_SNAP ||
 507             llc->ssap != LLC_SAP_SNAP ||
 508             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 509                 return htons(ETH_P_802_2);
 510
 511         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 512         return llc->ethertype;
 513 }
 514
 515 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 516                         int *key_lenp, int nh_len)
 517 {
 518         struct icmp6hdr *icmp = icmp6_hdr(skb);
 519         int error = 0;
 520         int key_len;
 521
 522         /* The ICMPv6 type and code fields use the 16-bit transport port
 523          * fields, so we need to store them in 16-bit network byte order.
 524          */
 525         key->ipv6.tp.src = htons(icmp->icmp6_type);
 526         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 527         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 528
 529         if (icmp->icmp6_code == 0 &&
 530             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 531              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 532                 int icmp_len = skb->len - skb_transport_offset(skb);
 533                 struct nd_msg *nd;
 534                 int offset;
 535
 536                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 537
 538                 /* In order to process neighbor discovery options, we need the
 539                  * entire packet.
 540                  */
 541                 if (unlikely(icmp_len < sizeof(*nd)))
 542                         goto out;
 543                 if (unlikely(skb_linearize(skb))) {
 544                         error = -ENOMEM;
 545                         goto out;
 546                 }
 547
 548                 nd = (struct nd_msg *)skb_transport_header(skb);
 549                 key->ipv6.nd.target = nd->target;
 550                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 551
 552                 icmp_len -= sizeof(*nd);
 553                 offset = 0;
 554                 while (icmp_len >= 8) {
 555                         struct nd_opt_hdr *nd_opt =
 556                                  (struct nd_opt_hdr *)(nd->opt + offset);
 557                         int opt_len = nd_opt->nd_opt_len * 8;
 558
 559                         if (unlikely(!opt_len || opt_len > icmp_len))
 560                                 goto invalid;
 561
 562                         /* Store the link layer address if the appropriate
 563                          * option is provided.  It is considered an error if
 564                          * the same link layer option is specified twice.
 565                          */
 566                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 567                             && opt_len == 8) {
 568                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 569                                         goto invalid;
 570                                 memcpy(key->ipv6.nd.sll,
 571                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 572                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 573                                    && opt_len == 8) {
 574                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 575                                         goto invalid;
 576                                 memcpy(key->ipv6.nd.tll,
 577                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 578                         }
 579
 580                         icmp_len -= opt_len;
 581                         offset += opt_len;
 582                 }
 583         }
 584
 585         goto out;
 586
 587 invalid:
 588         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 589         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 590         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 591
 592 out:
 593         *key_lenp = key_len;
 594         return error;
 595 }
 596
 597 /**
 598  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 599  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 600  * Ethernet header
 601  * @in_port: port number on which @skb was received.
 602  * @key: output flow key
 603  * @key_lenp: length of output flow key
 604  *
 605  * The caller must ensure that skb->len >= ETH_HLEN.
 606  *
 607  * Returns 0 if successful, otherwise a negative errno value.
 608  *
 609  * Initializes @skb header pointers as follows:
 610  *
 611  *    - skb->mac_header: the Ethernet header.
 612  *
 613  *    - skb->network_header: just past the Ethernet header, or just past the
 614  *      VLAN header, to the first byte of the Ethernet payload.
 615  *
 616  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 617  *      on output, then just past the IP header, if one is present and
 618  *      of a correct length, otherwise the same as skb->network_header.
 619  *      For other key->dl_type values it is left untouched.
 620  */
 621 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 622                  int *key_lenp)
 623 {
 624         int error = 0;
 625         int key_len = SW_FLOW_KEY_OFFSET(eth);
 626         struct ethhdr *eth;
 627
 628         memset(key, 0, sizeof(*key));
 629
 630         key->phy.priority = skb->priority;
 631         key->phy.tun_id = OVS_CB(skb)->tun_id;
 632         key->phy.in_port = in_port;
 633
 634         skb_reset_mac_header(skb);
 635
 636         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 637          * header in the linear data area.
 638          */
 639         eth = eth_hdr(skb);
 640         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 641         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 642
 643         __skb_pull(skb, 2 * ETH_ALEN);
 644
 645         if (vlan_tx_tag_present(skb))
 646                 key->eth.tci = htons(vlan_get_tci(skb));
 647         else if (eth->h_proto == htons(ETH_P_8021Q))
 648                 if (unlikely(parse_vlan(skb, key)))
 649                         return -ENOMEM;
 650
 651         key->eth.type = parse_ethertype(skb);
 652         if (unlikely(key->eth.type == htons(0)))
 653                 return -ENOMEM;
 654
 655         skb_reset_network_header(skb);
 656         __skb_push(skb, skb->data - skb_mac_header(skb));
 657
 658         /* Network layer. */
 659         if (key->eth.type == htons(ETH_P_IP)) {
 660                 struct iphdr *nh;
 661                 __be16 offset;
 662
 663                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 664
 665                 error = check_iphdr(skb);
 666                 if (unlikely(error)) {
 667                         if (error == -EINVAL) {
 668                                 skb->transport_header = skb->network_header;
 669                                 error = 0;
 670                         }
 671                         goto out;
 672                 }
 673
 674                 nh = ip_hdr(skb);
 675                 key->ipv4.addr.src = nh->saddr;
 676                 key->ipv4.addr.dst = nh->daddr;
 677
 678                 key->ip.proto = nh->protocol;
 679                 key->ip.tos = nh->tos;
 680                 key->ip.ttl = nh->ttl;
 681
 682                 offset = nh->frag_off & htons(IP_OFFSET);
 683                 if (offset) {
 684                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 685                         goto out;
 686                 }
 687                 if (nh->frag_off & htons(IP_MF) ||
 688                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 689                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 690
 691                 /* Transport layer. */
 692                 if (key->ip.proto == IPPROTO_TCP) {
 693                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 694                         if (tcphdr_ok(skb)) {
 695                                 struct tcphdr *tcp = tcp_hdr(skb);
 696                                 key->ipv4.tp.src = tcp->source;
 697                                 key->ipv4.tp.dst = tcp->dest;
 698                         }
 699                 } else if (key->ip.proto == IPPROTO_UDP) {
 700                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 701                         if (udphdr_ok(skb)) {
 702                                 struct udphdr *udp = udp_hdr(skb);
 703                                 key->ipv4.tp.src = udp->source;
 704                                 key->ipv4.tp.dst = udp->dest;
 705                         }
 706                 } else if (key->ip.proto == IPPROTO_ICMP) {
 707                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 708                         if (icmphdr_ok(skb)) {
 709                                 struct icmphdr *icmp = icmp_hdr(skb);
 710                                 /* The ICMP type and code fields use the 16-bit
 711                                  * transport port fields, so we need to store
 712                                  * them in 16-bit network byte order. */
 713                                 key->ipv4.tp.src = htons(icmp->type);
 714                                 key->ipv4.tp.dst = htons(icmp->code);
 715                         }
 716                 }
 717
 718         } else if (key->eth.type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
 719                 struct arp_eth_header *arp;
 720
 721                 arp = (struct arp_eth_header *)skb_network_header(skb);
 722
 723                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 724                                 && arp->ar_pro == htons(ETH_P_IP)
 725                                 && arp->ar_hln == ETH_ALEN
 726                                 && arp->ar_pln == 4) {
 727
 728                         /* We only match on the lower 8 bits of the opcode. */
 729                         if (ntohs(arp->ar_op) <= 0xff)
 730                                 key->ip.proto = ntohs(arp->ar_op);
 731
 732                         if (key->ip.proto == ARPOP_REQUEST
 733                                         || key->ip.proto == ARPOP_REPLY) {
 734                                 memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 735                                 memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 736                                 memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 737                                 memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 738                                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 739                         }
 740                 }
 741         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 742                 int nh_len;             /* IPv6 Header + Extensions */
 743
 744                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 745                 if (unlikely(nh_len < 0)) {
 746                         if (nh_len == -EINVAL)
 747                                 skb->transport_header = skb->network_header;
 748                         else
 749                                 error = nh_len;
 750                         goto out;
 751                 }
 752
 753                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 754                         goto out;
 755                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 756                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 757
 758                 /* Transport layer. */
 759                 if (key->ip.proto == NEXTHDR_TCP) {
 760                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 761                         if (tcphdr_ok(skb)) {
 762                                 struct tcphdr *tcp = tcp_hdr(skb);
 763                                 key->ipv6.tp.src = tcp->source;
 764                                 key->ipv6.tp.dst = tcp->dest;
 765                         }
 766                 } else if (key->ip.proto == NEXTHDR_UDP) {
 767                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 768                         if (udphdr_ok(skb)) {
 769                                 struct udphdr *udp = udp_hdr(skb);
 770                                 key->ipv6.tp.src = udp->source;
 771                                 key->ipv6.tp.dst = udp->dest;
 772                         }
 773                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 774                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 775                         if (icmp6hdr_ok(skb)) {
 776                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 777                                 if (error < 0)
 778                                         goto out;
 779                         }
 780                 }
 781         }
 782
 783 out:
 784         *key_lenp = key_len;
 785         return error;
 786 }
 787
 788 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 789 {
 790         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 791 }
 792
 793 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 794                                 struct sw_flow_key *key, int key_len)
 795 {
 796         struct sw_flow *flow;
 797         struct hlist_node *n;
 798         struct hlist_head *head;
 799         u32 hash;
 800
 801         hash = ovs_flow_hash(key, key_len);
 802
 803         head = find_bucket(table, hash);
 804         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 805
 806                 if (flow->hash == hash &&
 807                     !memcmp(&flow->key, key, key_len)) {
 808                         return flow;
 809                 }
 810         }
 811         return NULL;
 812 }
 813
 814 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 815 {
 816         struct hlist_head *head;
 817
 818         head = find_bucket(table, flow->hash);
 819         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 820         table->count++;
 821 }
 822
 823 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 824 {
 825         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 826         table->count--;
 827         BUG_ON(table->count < 0);
 828 }
 829
 830 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 831 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 832         [OVS_KEY_ATTR_ENCAP] = -1,
 833         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 834         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 835         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 836         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 837         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 838         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 839         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 840         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 841         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 842         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 843         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 844         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 845         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 846
 847         /* Not upstream. */
 848         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 849 };
 850
 851 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 852                                   const struct nlattr *a[], u64 *attrs)
 853 {
 854         const struct ovs_key_icmp *icmp_key;
 855         const struct ovs_key_tcp *tcp_key;
 856         const struct ovs_key_udp *udp_key;
 857
 858         switch (swkey->ip.proto) {
 859         case IPPROTO_TCP:
 860                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 861                         return -EINVAL;
 862                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 863
 864                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 865                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 866                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 867                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 868                 break;
 869
 870         case IPPROTO_UDP:
 871                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 872                         return -EINVAL;
 873                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 874
 875                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 876                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 877                 swkey->ipv4.tp.src = udp_key->udp_src;
 878                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 879                 break;
 880
 881         case IPPROTO_ICMP:
 882                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 883                         return -EINVAL;
 884                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 885
 886                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 887                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 888                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 889                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 890                 break;
 891         }
 892
 893         return 0;
 894 }
 895
 896 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 897                                   const struct nlattr *a[], u64 *attrs)
 898 {
 899         const struct ovs_key_icmpv6 *icmpv6_key;
 900         const struct ovs_key_tcp *tcp_key;
 901         const struct ovs_key_udp *udp_key;
 902
 903         switch (swkey->ip.proto) {
 904         case IPPROTO_TCP:
 905                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 906                         return -EINVAL;
 907                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 908
 909                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 910                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 911                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 912                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 913                 break;
 914
 915         case IPPROTO_UDP:
 916                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 917                         return -EINVAL;
 918                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 919
 920                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 921                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 922                 swkey->ipv6.tp.src = udp_key->udp_src;
 923                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 924                 break;
 925
 926         case IPPROTO_ICMPV6:
 927                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 928                         return -EINVAL;
 929                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 930
 931                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 932                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 933                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 934                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 935
 936                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 937                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 938                         const struct ovs_key_nd *nd_key;
 939
 940                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 941                                 return -EINVAL;
 942                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 943
 944                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 945                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 946                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 947                                sizeof(swkey->ipv6.nd.target));
 948                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 949                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 950                 }
 951                 break;
 952         }
 953
 954         return 0;
 955 }
 956
 957 static int parse_flow_nlattrs(const struct nlattr *attr,
 958                               const struct nlattr *a[], u64 *attrsp)
 959 {
 960         const struct nlattr *nla;
 961         u64 attrs;
 962         int rem;
 963
 964         attrs = 0;
 965         nla_for_each_nested(nla, attr, rem) {
 966                 u16 type = nla_type(nla);
 967                 int expected_len;
 968
 969                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 970                         return -EINVAL;
 971
 972                 expected_len = ovs_key_lens[type];
 973                 if (nla_len(nla) != expected_len && expected_len != -1)
 974                         return -EINVAL;
 975
 976                 attrs |= 1ULL << type;
 977                 a[type] = nla;
 978         }
 979         if (rem)
 980                 return -EINVAL;
 981
 982         *attrsp = attrs;
 983         return 0;
 984 }
 985
 986 /**
 987  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
 988  * @swkey: receives the extracted flow key.
 989  * @key_lenp: number of bytes used in @swkey.
 990  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
 991  * sequence.
 992  */
 993 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
 994                       const struct nlattr *attr)
 995 {
 996         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
 997         const struct ovs_key_ethernet *eth_key;
 998         int key_len;
 999         u64 attrs;
1000         int err;
1001
1002         memset(swkey, 0, sizeof(struct sw_flow_key));
1003         key_len = SW_FLOW_KEY_OFFSET(eth);
1004
1005         err = parse_flow_nlattrs(attr, a, &attrs);
1006         if (err)
1007                 return err;
1008
1009         /* Metadata attributes. */
1010         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1011                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1012                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1013         }
1014         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1015                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1016                 if (in_port >= DP_MAX_PORTS)
1017                         return -EINVAL;
1018                 swkey->phy.in_port = in_port;
1019                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1020         } else {
1021                 swkey->phy.in_port = DP_MAX_PORTS;
1022         }
1023
1024         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID)) {
1025                 swkey->phy.tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1026                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1027         }
1028
1029         /* Data attributes. */
1030         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1031                 return -EINVAL;
1032         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1033
1034         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1035         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1036         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1037
1038         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1039             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1040                 const struct nlattr *encap;
1041                 __be16 tci;
1042
1043                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1044                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1045                               (1 << OVS_KEY_ATTR_ENCAP)))
1046                         return -EINVAL;
1047
1048                 encap = a[OVS_KEY_ATTR_ENCAP];
1049                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1050                 if (tci & htons(VLAN_TAG_PRESENT)) {
1051                         swkey->eth.tci = tci;
1052
1053                         err = parse_flow_nlattrs(encap, a, &attrs);
1054                         if (err)
1055                                 return err;
1056                 } else if (!tci) {
1057                         /* Corner case for truncated 802.1Q header. */
1058                         if (nla_len(encap))
1059                                 return -EINVAL;
1060
1061                         swkey->eth.type = htons(ETH_P_8021Q);
1062                         *key_lenp = key_len;
1063                         return 0;
1064                 } else {
1065                         return -EINVAL;
1066                 }
1067         }
1068
1069         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1070                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1071                 if (ntohs(swkey->eth.type) < 1536)
1072                         return -EINVAL;
1073                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1074         } else {
1075                 swkey->eth.type = htons(ETH_P_802_2);
1076         }
1077
1078         if (swkey->eth.type == htons(ETH_P_IP)) {
1079                 const struct ovs_key_ipv4 *ipv4_key;
1080
1081                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1082                         return -EINVAL;
1083                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1084
1085                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1086                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1087                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1088                         return -EINVAL;
1089                 swkey->ip.proto = ipv4_key->ipv4_proto;
1090                 swkey->ip.tos = ipv4_key->ipv4_tos;
1091                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1092                 swkey->ip.frag = ipv4_key->ipv4_frag;
1093                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1094                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1095
1096                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1097                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1098                         if (err)
1099                                 return err;
1100                 }
1101         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1102                 const struct ovs_key_ipv6 *ipv6_key;
1103
1104                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1105                         return -EINVAL;
1106                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1107
1108                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1109                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1110                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1111                         return -EINVAL;
1112                 swkey->ipv6.label = ipv6_key->ipv6_label;
1113                 swkey->ip.proto = ipv6_key->ipv6_proto;
1114                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1115                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1116                 swkey->ip.frag = ipv6_key->ipv6_frag;
1117                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1118                        sizeof(swkey->ipv6.addr.src));
1119                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1120                        sizeof(swkey->ipv6.addr.dst));
1121
1122                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1123                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1124                         if (err)
1125                                 return err;
1126                 }
1127         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1128                 const struct ovs_key_arp *arp_key;
1129
1130                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1131                         return -EINVAL;
1132                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1133
1134                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1135                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1136                 swkey->ipv4.addr.src = arp_key->arp_sip;
1137                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1138                 if (arp_key->arp_op & htons(0xff00))
1139                         return -EINVAL;
1140                 swkey->ip.proto = ntohs(arp_key->arp_op);
1141                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1142                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1143         }
1144
1145         if (attrs)
1146                 return -EINVAL;
1147         *key_lenp = key_len;
1148
1149         return 0;
1150 }
1151
1152 /**
1153  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1154  * @in_port: receives the extracted input port.
1155  * @tun_id: receives the extracted tunnel ID.
1156  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1157  * sequence.
1158  *
1159  * This parses a series of Netlink attributes that form a flow key, which must
1160  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1161  * get the metadata, that is, the parts of the flow key that cannot be
1162  * extracted from the packet itself.
1163  */
1164 int ovs_flow_metadata_from_nlattrs(u32 *priority, u16 *in_port, __be64 *tun_id,
1165                                    const struct nlattr *attr)
1166 {
1167         const struct nlattr *nla;
1168         int rem;
1169
1170         *in_port = DP_MAX_PORTS;
1171         *tun_id = 0;
1172         *priority = 0;
1173
1174         nla_for_each_nested(nla, attr, rem) {
1175                 int type = nla_type(nla);
1176
1177                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1178                         if (nla_len(nla) != ovs_key_lens[type])
1179                                 return -EINVAL;
1180
1181                         switch (type) {
1182                         case OVS_KEY_ATTR_PRIORITY:
1183                                 *priority = nla_get_u32(nla);
1184                                 break;
1185
1186                         case OVS_KEY_ATTR_TUN_ID:
1187                                 *tun_id = nla_get_be64(nla);
1188                                 break;
1189
1190                         case OVS_KEY_ATTR_IN_PORT:
1191                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1192                                         return -EINVAL;
1193                                 *in_port = nla_get_u32(nla);
1194                                 break;
1195                         }
1196                 }
1197         }
1198         if (rem)
1199                 return -EINVAL;
1200         return 0;
1201 }
1202
1203 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1204 {
1205         struct ovs_key_ethernet *eth_key;
1206         struct nlattr *nla, *encap;
1207
1208         if (swkey->phy.priority)
1209                 NLA_PUT_U32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority);
1210
1211         if (swkey->phy.tun_id != cpu_to_be64(0))
1212                 NLA_PUT_BE64(skb, OVS_KEY_ATTR_TUN_ID, swkey->phy.tun_id);
1213
1214         if (swkey->phy.in_port != DP_MAX_PORTS)
1215                 NLA_PUT_U32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port);
1216
1217         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1218         if (!nla)
1219                 goto nla_put_failure;
1220         eth_key = nla_data(nla);
1221         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1222         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1223
1224         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1225                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q));
1226                 NLA_PUT_BE16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci);
1227                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1228                 if (!swkey->eth.tci)
1229                         goto unencap;
1230         } else {
1231                 encap = NULL;
1232         }
1233
1234         if (swkey->eth.type == htons(ETH_P_802_2))
1235                 goto unencap;
1236
1237         NLA_PUT_BE16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type);
1238
1239         if (swkey->eth.type == htons(ETH_P_IP)) {
1240                 struct ovs_key_ipv4 *ipv4_key;
1241
1242                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1243                 if (!nla)
1244                         goto nla_put_failure;
1245                 ipv4_key = nla_data(nla);
1246                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1247                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1248                 ipv4_key->ipv4_proto = swkey->ip.proto;
1249                 ipv4_key->ipv4_tos = swkey->ip.tos;
1250                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1251                 ipv4_key->ipv4_frag = swkey->ip.frag;
1252         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1253                 struct ovs_key_ipv6 *ipv6_key;
1254
1255                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1256                 if (!nla)
1257                         goto nla_put_failure;
1258                 ipv6_key = nla_data(nla);
1259                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1260                                 sizeof(ipv6_key->ipv6_src));
1261                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1262                                 sizeof(ipv6_key->ipv6_dst));
1263                 ipv6_key->ipv6_label = swkey->ipv6.label;
1264                 ipv6_key->ipv6_proto = swkey->ip.proto;
1265                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1266                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1267                 ipv6_key->ipv6_frag = swkey->ip.frag;
1268         } else if (swkey->eth.type == htons(ETH_P_ARP)) {
1269                 struct ovs_key_arp *arp_key;
1270
1271                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1272                 if (!nla)
1273                         goto nla_put_failure;
1274                 arp_key = nla_data(nla);
1275                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1276                 arp_key->arp_sip = swkey->ipv4.addr.src;
1277                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1278                 arp_key->arp_op = htons(swkey->ip.proto);
1279                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1280                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1281         }
1282
1283         if ((swkey->eth.type == htons(ETH_P_IP) ||
1284              swkey->eth.type == htons(ETH_P_IPV6)) &&
1285              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1286
1287                 if (swkey->ip.proto == IPPROTO_TCP) {
1288                         struct ovs_key_tcp *tcp_key;
1289
1290                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1291                         if (!nla)
1292                                 goto nla_put_failure;
1293                         tcp_key = nla_data(nla);
1294                         if (swkey->eth.type == htons(ETH_P_IP)) {
1295                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1296                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1297                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1298                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1299                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1300                         }
1301                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1302                         struct ovs_key_udp *udp_key;
1303
1304                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1305                         if (!nla)
1306                                 goto nla_put_failure;
1307                         udp_key = nla_data(nla);
1308                         if (swkey->eth.type == htons(ETH_P_IP)) {
1309                                 udp_key->udp_src = swkey->ipv4.tp.src;
1310                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1311                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1312                                 udp_key->udp_src = swkey->ipv6.tp.src;
1313                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1314                         }
1315                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1316                            swkey->ip.proto == IPPROTO_ICMP) {
1317                         struct ovs_key_icmp *icmp_key;
1318
1319                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1320                         if (!nla)
1321                                 goto nla_put_failure;
1322                         icmp_key = nla_data(nla);
1323                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1324                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1325                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1326                            swkey->ip.proto == IPPROTO_ICMPV6) {
1327                         struct ovs_key_icmpv6 *icmpv6_key;
1328
1329                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1330                                                 sizeof(*icmpv6_key));
1331                         if (!nla)
1332                                 goto nla_put_failure;
1333                         icmpv6_key = nla_data(nla);
1334                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1335                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1336
1337                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1338                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1339                                 struct ovs_key_nd *nd_key;
1340
1341                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1342                                 if (!nla)
1343                                         goto nla_put_failure;
1344                                 nd_key = nla_data(nla);
1345                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1346                                                         sizeof(nd_key->nd_target));
1347                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1348                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1349                         }
1350                 }
1351         }
1352
1353 unencap:
1354         if (encap)
1355                 nla_nest_end(skb, encap);
1356
1357         return 0;
1358
1359 nla_put_failure:
1360         return -EMSGSIZE;
1361 }
1362
1363 /* Initializes the flow module.
1364  * Returns zero if successful or a negative error code. */
1365 int ovs_flow_init(void)
1366 {
1367         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1368                                         0, NULL);
1369         if (flow_cache == NULL)
1370                 return -ENOMEM;
1371
1372         return 0;
1373 }
1374
1375 /* Uninitializes the flow module. */
1376 void ovs_flow_exit(void)
1377 {
1378         kmem_cache_destroy(flow_cache);
1379 }