datapath/flow.c

   1 /*
   2  * Copyright (c) 2007-2011 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #include "flow.h"
  20 #include "datapath.h"
  21 #include <linux/uaccess.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <net/llc_pdu.h>
  27 #include <linux/kernel.h>
  28 #include <linux/jhash.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/llc.h>
  31 #include <linux/module.h>
  32 #include <linux/in.h>
  33 #include <linux/rcupdate.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/icmp.h>
  40 #include <linux/icmpv6.h>
  41 #include <linux/rculist.h>
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/ndisc.h>
  45
  46 #include "vlan.h"
  47
  48 static struct kmem_cache *flow_cache;
  49
  50 static 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.eth.type == htons(ETH_P_IPV6)) &&
 189             flow->key.ip.proto == IPPROTO_TCP &&
 190             likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
 191                 u8 *tcp = (u8 *)tcp_hdr(skb);
 192                 tcp_flags = *(tcp + TCP_FLAGS_OFFSET) & TCP_FLAG_MASK;
 193         }
 194
 195         spin_lock(&flow->lock);
 196         flow->used = jiffies;
 197         flow->packet_count++;
 198         flow->byte_count += skb->len;
 199         flow->tcp_flags |= tcp_flags;
 200         spin_unlock(&flow->lock);
 201 }
 202
 203 struct sw_flow_actions *ovs_flow_actions_alloc(int size)
 204 {
 205         struct sw_flow_actions *sfa;
 206
 207         if (size > MAX_ACTIONS_BUFSIZE)
 208                 return ERR_PTR(-EINVAL);
 209
 210         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 211         if (!sfa)
 212                 return ERR_PTR(-ENOMEM);
 213
 214         sfa->actions_len = 0;
 215         return sfa;
 216 }
 217
 218 struct sw_flow *ovs_flow_alloc(void)
 219 {
 220         struct sw_flow *flow;
 221
 222         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 223         if (!flow)
 224                 return ERR_PTR(-ENOMEM);
 225
 226         spin_lock_init(&flow->lock);
 227         flow->sf_acts = NULL;
 228
 229         return flow;
 230 }
 231
 232 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 233 {
 234         hash = jhash_1word(hash, table->hash_seed);
 235         return flex_array_get(table->buckets,
 236                                 (hash & (table->n_buckets - 1)));
 237 }
 238
 239 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 240 {
 241         struct flex_array *buckets;
 242         int i, err;
 243
 244         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 245                                    n_buckets, GFP_KERNEL);
 246         if (!buckets)
 247                 return NULL;
 248
 249         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 250         if (err) {
 251                 flex_array_free(buckets);
 252                 return NULL;
 253         }
 254
 255         for (i = 0; i < n_buckets; i++)
 256                 INIT_HLIST_HEAD((struct hlist_head *)
 257                                         flex_array_get(buckets, i));
 258
 259         return buckets;
 260 }
 261
 262 static void free_buckets(struct flex_array *buckets)
 263 {
 264         flex_array_free(buckets);
 265 }
 266
 267 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 268 {
 269         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 270
 271         if (!table)
 272                 return NULL;
 273
 274         table->buckets = alloc_buckets(new_size);
 275
 276         if (!table->buckets) {
 277                 kfree(table);
 278                 return NULL;
 279         }
 280         table->n_buckets = new_size;
 281         table->count = 0;
 282         table->node_ver = 0;
 283         table->keep_flows = false;
 284         get_random_bytes(&table->hash_seed, sizeof(u32));
 285
 286         return table;
 287 }
 288
 289 void ovs_flow_tbl_destroy(struct flow_table *table)
 290 {
 291         int i;
 292
 293         if (!table)
 294                 return;
 295
 296         if (table->keep_flows)
 297                 goto skip_flows;
 298
 299         for (i = 0; i < table->n_buckets; i++) {
 300                 struct sw_flow *flow;
 301                 struct hlist_head *head = flex_array_get(table->buckets, i);
 302                 struct hlist_node *n;
 303                 int ver = table->node_ver;
 304
 305                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 306                         hlist_del_rcu(&flow->hash_node[ver]);
 307                         ovs_flow_free(flow);
 308                 }
 309         }
 310
 311 skip_flows:
 312         free_buckets(table->buckets);
 313         kfree(table);
 314 }
 315
 316 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 317 {
 318         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 319
 320         ovs_flow_tbl_destroy(table);
 321 }
 322
 323 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 324 {
 325         if (!table)
 326                 return;
 327
 328         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 329 }
 330
 331 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 332 {
 333         struct sw_flow *flow;
 334         struct hlist_head *head;
 335         int ver;
 336         int i;
 337
 338         ver = table->node_ver;
 339         while (*bucket < table->n_buckets) {
 340                 i = 0;
 341                 head = flex_array_get(table->buckets, *bucket);
 342                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 343                         if (i < *last) {
 344                                 i++;
 345                                 continue;
 346                         }
 347                         *last = i + 1;
 348                         return flow;
 349                 }
 350                 (*bucket)++;
 351                 *last = 0;
 352         }
 353
 354         return NULL;
 355 }
 356
 357 static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 358 {
 359         struct hlist_head *head;
 360         head = find_bucket(table, flow->hash);
 361         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 362         table->count++;
 363 }
 364
 365 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 366 {
 367         int old_ver;
 368         int i;
 369
 370         old_ver = old->node_ver;
 371         new->node_ver = !old_ver;
 372
 373         /* Insert in new table. */
 374         for (i = 0; i < old->n_buckets; i++) {
 375                 struct sw_flow *flow;
 376                 struct hlist_head *head;
 377
 378                 head = flex_array_get(old->buckets, i);
 379
 380                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 381                         __flow_tbl_insert(new, flow);
 382         }
 383         old->keep_flows = true;
 384 }
 385
 386 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 387 {
 388         struct flow_table *new_table;
 389
 390         new_table = ovs_flow_tbl_alloc(n_buckets);
 391         if (!new_table)
 392                 return ERR_PTR(-ENOMEM);
 393
 394         flow_table_copy_flows(table, new_table);
 395
 396         return new_table;
 397 }
 398
 399 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 400 {
 401         return __flow_tbl_rehash(table, table->n_buckets);
 402 }
 403
 404 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 405 {
 406         return __flow_tbl_rehash(table, table->n_buckets * 2);
 407 }
 408
 409 void ovs_flow_free(struct sw_flow *flow)
 410 {
 411         if (unlikely(!flow))
 412                 return;
 413
 414         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 415         kmem_cache_free(flow_cache, flow);
 416 }
 417
 418 /* RCU callback used by ovs_flow_deferred_free. */
 419 static void rcu_free_flow_callback(struct rcu_head *rcu)
 420 {
 421         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 422
 423         ovs_flow_free(flow);
 424 }
 425
 426 /* Schedules 'flow' to be freed after the next RCU grace period.
 427  * The caller must hold rcu_read_lock for this to be sensible. */
 428 void ovs_flow_deferred_free(struct sw_flow *flow)
 429 {
 430         call_rcu(&flow->rcu, rcu_free_flow_callback);
 431 }
 432
 433 /* RCU callback used by ovs_flow_deferred_free_acts. */
 434 static void rcu_free_acts_callback(struct rcu_head *rcu)
 435 {
 436         struct sw_flow_actions *sf_acts = container_of(rcu,
 437                         struct sw_flow_actions, rcu);
 438         kfree(sf_acts);
 439 }
 440
 441 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 442  * The caller must hold rcu_read_lock for this to be sensible. */
 443 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 444 {
 445         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 446 }
 447
 448 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 449 {
 450         struct qtag_prefix {
 451                 __be16 eth_type; /* ETH_P_8021Q */
 452                 __be16 tci;
 453         };
 454         struct qtag_prefix *qp;
 455
 456         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 457                 return 0;
 458
 459         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 460                                          sizeof(__be16))))
 461                 return -ENOMEM;
 462
 463         qp = (struct qtag_prefix *) skb->data;
 464         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 465         __skb_pull(skb, sizeof(struct qtag_prefix));
 466
 467         return 0;
 468 }
 469
 470 static __be16 parse_ethertype(struct sk_buff *skb)
 471 {
 472         struct llc_snap_hdr {
 473                 u8  dsap;  /* Always 0xAA */
 474                 u8  ssap;  /* Always 0xAA */
 475                 u8  ctrl;
 476                 u8  oui[3];
 477                 __be16 ethertype;
 478         };
 479         struct llc_snap_hdr *llc;
 480         __be16 proto;
 481
 482         proto = *(__be16 *) skb->data;
 483         __skb_pull(skb, sizeof(__be16));
 484
 485         if (ntohs(proto) >= ETH_P_802_3_MIN)
 486                 return proto;
 487
 488         if (skb->len < sizeof(struct llc_snap_hdr))
 489                 return htons(ETH_P_802_2);
 490
 491         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 492                 return htons(0);
 493
 494         llc = (struct llc_snap_hdr *) skb->data;
 495         if (llc->dsap != LLC_SAP_SNAP ||
 496             llc->ssap != LLC_SAP_SNAP ||
 497             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 498                 return htons(ETH_P_802_2);
 499
 500         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 501
 502         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 503                 return llc->ethertype;
 504
 505         return htons(ETH_P_802_2);
 506 }
 507
 508 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 509                         int *key_lenp, int nh_len)
 510 {
 511         struct icmp6hdr *icmp = icmp6_hdr(skb);
 512         int error = 0;
 513         int key_len;
 514
 515         /* The ICMPv6 type and code fields use the 16-bit transport port
 516          * fields, so we need to store them in 16-bit network byte order.
 517          */
 518         key->ipv6.tp.src = htons(icmp->icmp6_type);
 519         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 520         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 521
 522         if (icmp->icmp6_code == 0 &&
 523             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 524              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 525                 int icmp_len = skb->len - skb_transport_offset(skb);
 526                 struct nd_msg *nd;
 527                 int offset;
 528
 529                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 530
 531                 /* In order to process neighbor discovery options, we need the
 532                  * entire packet.
 533                  */
 534                 if (unlikely(icmp_len < sizeof(*nd)))
 535                         goto out;
 536                 if (unlikely(skb_linearize(skb))) {
 537                         error = -ENOMEM;
 538                         goto out;
 539                 }
 540
 541                 nd = (struct nd_msg *)skb_transport_header(skb);
 542                 key->ipv6.nd.target = nd->target;
 543                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 544
 545                 icmp_len -= sizeof(*nd);
 546                 offset = 0;
 547                 while (icmp_len >= 8) {
 548                         struct nd_opt_hdr *nd_opt =
 549                                  (struct nd_opt_hdr *)(nd->opt + offset);
 550                         int opt_len = nd_opt->nd_opt_len * 8;
 551
 552                         if (unlikely(!opt_len || opt_len > icmp_len))
 553                                 goto invalid;
 554
 555                         /* Store the link layer address if the appropriate
 556                          * option is provided.  It is considered an error if
 557                          * the same link layer option is specified twice.
 558                          */
 559                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 560                             && opt_len == 8) {
 561                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 562                                         goto invalid;
 563                                 memcpy(key->ipv6.nd.sll,
 564                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 565                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 566                                    && opt_len == 8) {
 567                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 568                                         goto invalid;
 569                                 memcpy(key->ipv6.nd.tll,
 570                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 571                         }
 572
 573                         icmp_len -= opt_len;
 574                         offset += opt_len;
 575                 }
 576         }
 577
 578         goto out;
 579
 580 invalid:
 581         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 582         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 583         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 584
 585 out:
 586         *key_lenp = key_len;
 587         return error;
 588 }
 589
 590 /**
 591  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 592  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 593  * Ethernet header
 594  * @in_port: port number on which @skb was received.
 595  * @key: output flow key
 596  * @key_lenp: length of output flow key
 597  *
 598  * The caller must ensure that skb->len >= ETH_HLEN.
 599  *
 600  * Returns 0 if successful, otherwise a negative errno value.
 601  *
 602  * Initializes @skb header pointers as follows:
 603  *
 604  *    - skb->mac_header: the Ethernet header.
 605  *
 606  *    - skb->network_header: just past the Ethernet header, or just past the
 607  *      VLAN header, to the first byte of the Ethernet payload.
 608  *
 609  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 610  *      on output, then just past the IP header, if one is present and
 611  *      of a correct length, otherwise the same as skb->network_header.
 612  *      For other key->eth.type values it is left untouched.
 613  */
 614 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 615                  int *key_lenp)
 616 {
 617         int error = 0;
 618         int key_len = SW_FLOW_KEY_OFFSET(eth);
 619         struct ethhdr *eth;
 620
 621         memset(key, 0, sizeof(*key));
 622
 623         key->phy.priority = skb->priority;
 624         if (OVS_CB(skb)->tun_key)
 625                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 626         key->phy.in_port = in_port;
 627         key->phy.skb_mark = skb_get_mark(skb);
 628
 629         skb_reset_mac_header(skb);
 630
 631         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 632          * header in the linear data area.
 633          */
 634         eth = eth_hdr(skb);
 635         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 636         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 637
 638         __skb_pull(skb, 2 * ETH_ALEN);
 639         /* We are going to push all headers that we pull, so no need to
 640          * update skb->csum here. */
 641
 642         if (vlan_tx_tag_present(skb))
 643                 key->eth.tci = htons(vlan_get_tci(skb));
 644         else if (eth->h_proto == htons(ETH_P_8021Q))
 645                 if (unlikely(parse_vlan(skb, key)))
 646                         return -ENOMEM;
 647
 648         key->eth.type = parse_ethertype(skb);
 649         if (unlikely(key->eth.type == htons(0)))
 650                 return -ENOMEM;
 651
 652         skb_reset_network_header(skb);
 653         __skb_push(skb, skb->data - skb_mac_header(skb));
 654
 655         /* Network layer. */
 656         if (key->eth.type == htons(ETH_P_IP)) {
 657                 struct iphdr *nh;
 658                 __be16 offset;
 659
 660                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 661
 662                 error = check_iphdr(skb);
 663                 if (unlikely(error)) {
 664                         if (error == -EINVAL) {
 665                                 skb->transport_header = skb->network_header;
 666                                 error = 0;
 667                         }
 668                         goto out;
 669                 }
 670
 671                 nh = ip_hdr(skb);
 672                 key->ipv4.addr.src = nh->saddr;
 673                 key->ipv4.addr.dst = nh->daddr;
 674
 675                 key->ip.proto = nh->protocol;
 676                 key->ip.tos = nh->tos;
 677                 key->ip.ttl = nh->ttl;
 678
 679                 offset = nh->frag_off & htons(IP_OFFSET);
 680                 if (offset) {
 681                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 682                         goto out;
 683                 }
 684                 if (nh->frag_off & htons(IP_MF) ||
 685                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 686                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 687
 688                 /* Transport layer. */
 689                 if (key->ip.proto == IPPROTO_TCP) {
 690                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 691                         if (tcphdr_ok(skb)) {
 692                                 struct tcphdr *tcp = tcp_hdr(skb);
 693                                 key->ipv4.tp.src = tcp->source;
 694                                 key->ipv4.tp.dst = tcp->dest;
 695                         }
 696                 } else if (key->ip.proto == IPPROTO_UDP) {
 697                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 698                         if (udphdr_ok(skb)) {
 699                                 struct udphdr *udp = udp_hdr(skb);
 700                                 key->ipv4.tp.src = udp->source;
 701                                 key->ipv4.tp.dst = udp->dest;
 702                         }
 703                 } else if (key->ip.proto == IPPROTO_ICMP) {
 704                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 705                         if (icmphdr_ok(skb)) {
 706                                 struct icmphdr *icmp = icmp_hdr(skb);
 707                                 /* The ICMP type and code fields use the 16-bit
 708                                  * transport port fields, so we need to store
 709                                  * them in 16-bit network byte order. */
 710                                 key->ipv4.tp.src = htons(icmp->type);
 711                                 key->ipv4.tp.dst = htons(icmp->code);
 712                         }
 713                 }
 714
 715         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 716                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 717                 struct arp_eth_header *arp;
 718
 719                 arp = (struct arp_eth_header *)skb_network_header(skb);
 720
 721                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 722                                 && arp->ar_pro == htons(ETH_P_IP)
 723                                 && arp->ar_hln == ETH_ALEN
 724                                 && arp->ar_pln == 4) {
 725
 726                         /* We only match on the lower 8 bits of the opcode. */
 727                         if (ntohs(arp->ar_op) <= 0xff)
 728                                 key->ip.proto = ntohs(arp->ar_op);
 729                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 730                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 731                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 732                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 733                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 734                 }
 735         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 736                 int nh_len;             /* IPv6 Header + Extensions */
 737
 738                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 739                 if (unlikely(nh_len < 0)) {
 740                         if (nh_len == -EINVAL)
 741                                 skb->transport_header = skb->network_header;
 742                         else
 743                                 error = nh_len;
 744                         goto out;
 745                 }
 746
 747                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 748                         goto out;
 749                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 750                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 751
 752                 /* Transport layer. */
 753                 if (key->ip.proto == NEXTHDR_TCP) {
 754                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 755                         if (tcphdr_ok(skb)) {
 756                                 struct tcphdr *tcp = tcp_hdr(skb);
 757                                 key->ipv6.tp.src = tcp->source;
 758                                 key->ipv6.tp.dst = tcp->dest;
 759                         }
 760                 } else if (key->ip.proto == NEXTHDR_UDP) {
 761                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 762                         if (udphdr_ok(skb)) {
 763                                 struct udphdr *udp = udp_hdr(skb);
 764                                 key->ipv6.tp.src = udp->source;
 765                                 key->ipv6.tp.dst = udp->dest;
 766                         }
 767                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 768                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 769                         if (icmp6hdr_ok(skb)) {
 770                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 771                                 if (error < 0)
 772                                         goto out;
 773                         }
 774                 }
 775         }
 776
 777 out:
 778         *key_lenp = key_len;
 779         return error;
 780 }
 781
 782 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 783 {
 784         return jhash2((u32 *)((u8 *)key + key_start),
 785                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 786 }
 787
 788 static int flow_key_start(struct sw_flow_key *key)
 789 {
 790         if (key->tun_key.ipv4_dst)
 791                 return 0;
 792         else
 793                 return offsetof(struct sw_flow_key, phy);
 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_head *head;
 801         u8 *_key;
 802         int key_start;
 803         u32 hash;
 804
 805         key_start = flow_key_start(key);
 806         hash = ovs_flow_hash(key, key_start, key_len);
 807
 808         _key = (u8 *) key + key_start;
 809         head = find_bucket(table, hash);
 810         hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
 811
 812                 if (flow->hash == hash &&
 813                     !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
 814                         return flow;
 815                 }
 816         }
 817         return NULL;
 818 }
 819
 820 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 821                          struct sw_flow_key *key, int key_len)
 822 {
 823         flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
 824         memcpy(&flow->key, key, sizeof(flow->key));
 825         __flow_tbl_insert(table, flow);
 826 }
 827
 828 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 829 {
 830         BUG_ON(table->count == 0);
 831         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 832         table->count--;
 833 }
 834
 835 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 836 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 837         [OVS_KEY_ATTR_ENCAP] = -1,
 838         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 839         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 840         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 841         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 842         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 843         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 844         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 845         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 846         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 847         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 848         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 849         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 850         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 851         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 852         [OVS_KEY_ATTR_TUNNEL] = -1,
 853 };
 854
 855 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 856                                   const struct nlattr *a[], u64 *attrs)
 857 {
 858         const struct ovs_key_icmp *icmp_key;
 859         const struct ovs_key_tcp *tcp_key;
 860         const struct ovs_key_udp *udp_key;
 861
 862         switch (swkey->ip.proto) {
 863         case IPPROTO_TCP:
 864                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 865                         return -EINVAL;
 866                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 867
 868                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 869                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 870                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 871                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 872                 break;
 873
 874         case IPPROTO_UDP:
 875                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 876                         return -EINVAL;
 877                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 878
 879                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 880                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 881                 swkey->ipv4.tp.src = udp_key->udp_src;
 882                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 883                 break;
 884
 885         case IPPROTO_ICMP:
 886                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 887                         return -EINVAL;
 888                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 889
 890                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 891                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 892                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 893                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 894                 break;
 895         }
 896
 897         return 0;
 898 }
 899
 900 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 901                                   const struct nlattr *a[], u64 *attrs)
 902 {
 903         const struct ovs_key_icmpv6 *icmpv6_key;
 904         const struct ovs_key_tcp *tcp_key;
 905         const struct ovs_key_udp *udp_key;
 906
 907         switch (swkey->ip.proto) {
 908         case IPPROTO_TCP:
 909                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 910                         return -EINVAL;
 911                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 912
 913                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 914                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 915                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 916                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 917                 break;
 918
 919         case IPPROTO_UDP:
 920                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 921                         return -EINVAL;
 922                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 923
 924                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 925                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 926                 swkey->ipv6.tp.src = udp_key->udp_src;
 927                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 928                 break;
 929
 930         case IPPROTO_ICMPV6:
 931                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 932                         return -EINVAL;
 933                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 934
 935                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 936                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 937                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 938                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 939
 940                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 941                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 942                         const struct ovs_key_nd *nd_key;
 943
 944                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 945                                 return -EINVAL;
 946                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 947
 948                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 949                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 950                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 951                                sizeof(swkey->ipv6.nd.target));
 952                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 953                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 954                 }
 955                 break;
 956         }
 957
 958         return 0;
 959 }
 960
 961 static int parse_flow_nlattrs(const struct nlattr *attr,
 962                               const struct nlattr *a[], u64 *attrsp)
 963 {
 964         const struct nlattr *nla;
 965         u64 attrs;
 966         int rem;
 967
 968         attrs = 0;
 969         nla_for_each_nested(nla, attr, rem) {
 970                 u16 type = nla_type(nla);
 971                 int expected_len;
 972
 973                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 974                         return -EINVAL;
 975
 976                 expected_len = ovs_key_lens[type];
 977                 if (nla_len(nla) != expected_len && expected_len != -1)
 978                         return -EINVAL;
 979
 980                 attrs |= 1ULL << type;
 981                 a[type] = nla;
 982         }
 983         if (rem)
 984                 return -EINVAL;
 985
 986         *attrsp = attrs;
 987         return 0;
 988 }
 989
 990 int ipv4_tun_from_nlattr(const struct nlattr *attr,
 991                          struct ovs_key_ipv4_tunnel *tun_key)
 992 {
 993         struct nlattr *a;
 994         int rem;
 995         bool ttl = false;
 996
 997         memset(tun_key, 0, sizeof(*tun_key));
 998
 999         nla_for_each_nested(a, attr, rem) {
1000                 int type = nla_type(a);
1001                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1002                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1003                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1004                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1005                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1006                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1007                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1008                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1009                 };
1010
1011                 if (type > OVS_TUNNEL_KEY_ATTR_MAX ||
1012                         ovs_tunnel_key_lens[type] != nla_len(a))
1013                         return -EINVAL;
1014
1015                 switch (type) {
1016                 case OVS_TUNNEL_KEY_ATTR_ID:
1017                         tun_key->tun_id = nla_get_be64(a);
1018                         tun_key->tun_flags |= OVS_TNL_F_KEY;
1019                         break;
1020                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1021                         tun_key->ipv4_src = nla_get_be32(a);
1022                         break;
1023                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1024                         tun_key->ipv4_dst = nla_get_be32(a);
1025                         break;
1026                 case OVS_TUNNEL_KEY_ATTR_TOS:
1027                         tun_key->ipv4_tos = nla_get_u8(a);
1028                         break;
1029                 case OVS_TUNNEL_KEY_ATTR_TTL:
1030                         tun_key->ipv4_ttl = nla_get_u8(a);
1031                         ttl = true;
1032                         break;
1033                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1034                         tun_key->tun_flags |= OVS_TNL_F_DONT_FRAGMENT;
1035                         break;
1036                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1037                         tun_key->tun_flags |= OVS_TNL_F_CSUM;
1038                         break;
1039                 default:
1040                         return -EINVAL;
1041
1042                 }
1043         }
1044         if (rem > 0)
1045                 return -EINVAL;
1046
1047         if (!tun_key->ipv4_dst)
1048                 return -EINVAL;
1049
1050         if (!ttl)
1051                 return -EINVAL;
1052
1053         return 0;
1054 }
1055
1056 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1057                         const struct ovs_key_ipv4_tunnel *tun_key)
1058 {
1059         struct nlattr *nla;
1060
1061         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1062         if (!nla)
1063                 return -EMSGSIZE;
1064
1065         if (tun_key->tun_flags & OVS_TNL_F_KEY &&
1066             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id))
1067                 return -EMSGSIZE;
1068         if (tun_key->ipv4_src &&
1069             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ipv4_src))
1070                 return -EMSGSIZE;
1071         if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ipv4_dst))
1072                 return -EMSGSIZE;
1073         if (tun_key->ipv4_tos &&
1074             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ipv4_tos))
1075                 return -EMSGSIZE;
1076         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ipv4_ttl))
1077                 return -EMSGSIZE;
1078         if ((tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT) &&
1079                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1080                 return -EMSGSIZE;
1081         if ((tun_key->tun_flags & OVS_TNL_F_CSUM) &&
1082                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1083                 return -EMSGSIZE;
1084
1085         nla_nest_end(skb, nla);
1086         return 0;
1087 }
1088
1089 /**
1090  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
1091  * @swkey: receives the extracted flow key.
1092  * @key_lenp: number of bytes used in @swkey.
1093  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1094  * sequence.
1095  */
1096 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1097                       const struct nlattr *attr)
1098 {
1099         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1100         const struct ovs_key_ethernet *eth_key;
1101         int key_len;
1102         u64 attrs;
1103         int err;
1104
1105         memset(swkey, 0, sizeof(struct sw_flow_key));
1106         key_len = SW_FLOW_KEY_OFFSET(eth);
1107
1108         err = parse_flow_nlattrs(attr, a, &attrs);
1109         if (err)
1110                 return err;
1111
1112         /* Metadata attributes. */
1113         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1114                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1115                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1116         }
1117         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1118                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1119                 if (in_port >= DP_MAX_PORTS)
1120                         return -EINVAL;
1121                 swkey->phy.in_port = in_port;
1122                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1123         } else {
1124                 swkey->phy.in_port = DP_MAX_PORTS;
1125         }
1126         if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1127                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1128 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1129                 if (mark != 0)
1130                         return -EINVAL;
1131 #endif
1132                 swkey->phy.skb_mark = mark;
1133                 attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1134         }
1135
1136         if (attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1137                 err = ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1138                 if (err)
1139                         return err;
1140
1141                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1142         }
1143
1144         /* Data attributes. */
1145         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1146                 return -EINVAL;
1147         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1148
1149         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1150         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1151         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1152
1153         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1154             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1155                 const struct nlattr *encap;
1156                 __be16 tci;
1157
1158                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1159                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1160                               (1 << OVS_KEY_ATTR_ENCAP)))
1161                         return -EINVAL;
1162
1163                 encap = a[OVS_KEY_ATTR_ENCAP];
1164                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1165                 if (tci & htons(VLAN_TAG_PRESENT)) {
1166                         swkey->eth.tci = tci;
1167
1168                         err = parse_flow_nlattrs(encap, a, &attrs);
1169                         if (err)
1170                                 return err;
1171                 } else if (!tci) {
1172                         /* Corner case for truncated 802.1Q header. */
1173                         if (nla_len(encap))
1174                                 return -EINVAL;
1175
1176                         swkey->eth.type = htons(ETH_P_8021Q);
1177                         *key_lenp = key_len;
1178                         return 0;
1179                 } else {
1180                         return -EINVAL;
1181                 }
1182         }
1183
1184         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1185                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1186                 if (ntohs(swkey->eth.type) < ETH_P_802_3_MIN)
1187                         return -EINVAL;
1188                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1189         } else {
1190                 swkey->eth.type = htons(ETH_P_802_2);
1191         }
1192
1193         if (swkey->eth.type == htons(ETH_P_IP)) {
1194                 const struct ovs_key_ipv4 *ipv4_key;
1195
1196                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1197                         return -EINVAL;
1198                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1199
1200                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1201                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1202                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1203                         return -EINVAL;
1204                 swkey->ip.proto = ipv4_key->ipv4_proto;
1205                 swkey->ip.tos = ipv4_key->ipv4_tos;
1206                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1207                 swkey->ip.frag = ipv4_key->ipv4_frag;
1208                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1209                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1210
1211                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1212                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1213                         if (err)
1214                                 return err;
1215                 }
1216         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1217                 const struct ovs_key_ipv6 *ipv6_key;
1218
1219                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1220                         return -EINVAL;
1221                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1222
1223                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1224                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1225                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1226                         return -EINVAL;
1227                 swkey->ipv6.label = ipv6_key->ipv6_label;
1228                 swkey->ip.proto = ipv6_key->ipv6_proto;
1229                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1230                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1231                 swkey->ip.frag = ipv6_key->ipv6_frag;
1232                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1233                        sizeof(swkey->ipv6.addr.src));
1234                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1235                        sizeof(swkey->ipv6.addr.dst));
1236
1237                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1238                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1239                         if (err)
1240                                 return err;
1241                 }
1242         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1243                    swkey->eth.type == htons(ETH_P_RARP)) {
1244                 const struct ovs_key_arp *arp_key;
1245
1246                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1247                         return -EINVAL;
1248                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1249
1250                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1251                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1252                 swkey->ipv4.addr.src = arp_key->arp_sip;
1253                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1254                 if (arp_key->arp_op & htons(0xff00))
1255                         return -EINVAL;
1256                 swkey->ip.proto = ntohs(arp_key->arp_op);
1257                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1258                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1259         }
1260
1261         if (attrs)
1262                 return -EINVAL;
1263         *key_lenp = key_len;
1264
1265         return 0;
1266 }
1267
1268 /**
1269  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1270  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1271  * @key_len: Length of key in @flow.  Used for calculating flow hash.
1272  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1273  * sequence.
1274  *
1275  * This parses a series of Netlink attributes that form a flow key, which must
1276  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1277  * get the metadata, that is, the parts of the flow key that cannot be
1278  * extracted from the packet itself.
1279  */
1280
1281 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
1282 {
1283         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1284         const struct nlattr *nla;
1285         int rem;
1286
1287         flow->key.phy.in_port = DP_MAX_PORTS;
1288         flow->key.phy.priority = 0;
1289         flow->key.phy.skb_mark = 0;
1290         memset(tun_key, 0, sizeof(flow->key.tun_key));
1291
1292         nla_for_each_nested(nla, attr, rem) {
1293                 int type = nla_type(nla);
1294
1295                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1296                         int err;
1297
1298                         if (nla_len(nla) != ovs_key_lens[type])
1299                                 return -EINVAL;
1300
1301                         switch (type) {
1302                         case OVS_KEY_ATTR_PRIORITY:
1303                                 flow->key.phy.priority = nla_get_u32(nla);
1304                                 break;
1305
1306                         case OVS_KEY_ATTR_TUNNEL:
1307                                 err = ipv4_tun_from_nlattr(nla, tun_key);
1308                                 if (err)
1309                                         return err;
1310                                 break;
1311
1312                         case OVS_KEY_ATTR_IN_PORT:
1313                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1314                                         return -EINVAL;
1315                                 flow->key.phy.in_port = nla_get_u32(nla);
1316                                 break;
1317
1318                         case OVS_KEY_ATTR_SKB_MARK:
1319 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1320                                 if (nla_get_u32(nla) != 0)
1321                                         return -EINVAL;
1322 #endif
1323                                 flow->key.phy.skb_mark = nla_get_u32(nla);
1324                                 break;
1325                         }
1326                 }
1327         }
1328         if (rem)
1329                 return -EINVAL;
1330
1331         flow->hash = ovs_flow_hash(&flow->key,
1332                                    flow_key_start(&flow->key), key_len);
1333
1334         return 0;
1335 }
1336
1337 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1338 {
1339         struct ovs_key_ethernet *eth_key;
1340         struct nlattr *nla, *encap;
1341
1342         if (swkey->phy.priority &&
1343             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1344                 goto nla_put_failure;
1345
1346         if (swkey->tun_key.ipv4_dst &&
1347             ipv4_tun_to_nlattr(skb, &swkey->tun_key))
1348                 goto nla_put_failure;
1349
1350         if (swkey->phy.in_port != DP_MAX_PORTS &&
1351             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1352                 goto nla_put_failure;
1353
1354         if (swkey->phy.skb_mark &&
1355             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1356                 goto nla_put_failure;
1357
1358         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1359         if (!nla)
1360                 goto nla_put_failure;
1361         eth_key = nla_data(nla);
1362         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1363         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1364
1365         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1366                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1367                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1368                         goto nla_put_failure;
1369                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1370                 if (!swkey->eth.tci)
1371                         goto unencap;
1372         } else {
1373                 encap = NULL;
1374         }
1375
1376         if (swkey->eth.type == htons(ETH_P_802_2))
1377                 goto unencap;
1378
1379         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1380                 goto nla_put_failure;
1381
1382         if (swkey->eth.type == htons(ETH_P_IP)) {
1383                 struct ovs_key_ipv4 *ipv4_key;
1384
1385                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1386                 if (!nla)
1387                         goto nla_put_failure;
1388                 ipv4_key = nla_data(nla);
1389                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1390                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1391                 ipv4_key->ipv4_proto = swkey->ip.proto;
1392                 ipv4_key->ipv4_tos = swkey->ip.tos;
1393                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1394                 ipv4_key->ipv4_frag = swkey->ip.frag;
1395         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1396                 struct ovs_key_ipv6 *ipv6_key;
1397
1398                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1399                 if (!nla)
1400                         goto nla_put_failure;
1401                 ipv6_key = nla_data(nla);
1402                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1403                                 sizeof(ipv6_key->ipv6_src));
1404                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1405                                 sizeof(ipv6_key->ipv6_dst));
1406                 ipv6_key->ipv6_label = swkey->ipv6.label;
1407                 ipv6_key->ipv6_proto = swkey->ip.proto;
1408                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1409                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1410                 ipv6_key->ipv6_frag = swkey->ip.frag;
1411         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1412                    swkey->eth.type == htons(ETH_P_RARP)) {
1413                 struct ovs_key_arp *arp_key;
1414
1415                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1416                 if (!nla)
1417                         goto nla_put_failure;
1418                 arp_key = nla_data(nla);
1419                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1420                 arp_key->arp_sip = swkey->ipv4.addr.src;
1421                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1422                 arp_key->arp_op = htons(swkey->ip.proto);
1423                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1424                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1425         }
1426
1427         if ((swkey->eth.type == htons(ETH_P_IP) ||
1428              swkey->eth.type == htons(ETH_P_IPV6)) &&
1429              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1430
1431                 if (swkey->ip.proto == IPPROTO_TCP) {
1432                         struct ovs_key_tcp *tcp_key;
1433
1434                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1435                         if (!nla)
1436                                 goto nla_put_failure;
1437                         tcp_key = nla_data(nla);
1438                         if (swkey->eth.type == htons(ETH_P_IP)) {
1439                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1440                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1441                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1442                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1443                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1444                         }
1445                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1446                         struct ovs_key_udp *udp_key;
1447
1448                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1449                         if (!nla)
1450                                 goto nla_put_failure;
1451                         udp_key = nla_data(nla);
1452                         if (swkey->eth.type == htons(ETH_P_IP)) {
1453                                 udp_key->udp_src = swkey->ipv4.tp.src;
1454                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1455                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1456                                 udp_key->udp_src = swkey->ipv6.tp.src;
1457                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1458                         }
1459                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1460                            swkey->ip.proto == IPPROTO_ICMP) {
1461                         struct ovs_key_icmp *icmp_key;
1462
1463                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1464                         if (!nla)
1465                                 goto nla_put_failure;
1466                         icmp_key = nla_data(nla);
1467                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1468                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1469                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1470                            swkey->ip.proto == IPPROTO_ICMPV6) {
1471                         struct ovs_key_icmpv6 *icmpv6_key;
1472
1473                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1474                                                 sizeof(*icmpv6_key));
1475                         if (!nla)
1476                                 goto nla_put_failure;
1477                         icmpv6_key = nla_data(nla);
1478                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1479                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1480
1481                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1482                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1483                                 struct ovs_key_nd *nd_key;
1484
1485                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1486                                 if (!nla)
1487                                         goto nla_put_failure;
1488                                 nd_key = nla_data(nla);
1489                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1490                                                         sizeof(nd_key->nd_target));
1491                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1492                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1493                         }
1494                 }
1495         }
1496
1497 unencap:
1498         if (encap)
1499                 nla_nest_end(skb, encap);
1500
1501         return 0;
1502
1503 nla_put_failure:
1504         return -EMSGSIZE;
1505 }
1506
1507 /* Initializes the flow module.
1508  * Returns zero if successful or a negative error code. */
1509 int ovs_flow_init(void)
1510 {
1511         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1512                                         0, NULL);
1513         if (flow_cache == NULL)
1514                 return -ENOMEM;
1515
1516         return 0;
1517 }
1518
1519 /* Uninitializes the flow module. */
1520 void ovs_flow_exit(void)
1521 {
1522         kmem_cache_destroy(flow_cache);
1523 }