Pileus Git - ~andy/linux/blob - net/openvswitch/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/ip_tunnels.h>
  44 #include <net/ipv6.h>
  45 #include <net/ndisc.h>
  46
  47 static struct kmem_cache *flow_cache;
  48
  49 static int check_header(struct sk_buff *skb, int len)
  50 {
  51         if (unlikely(skb->len < len))
  52                 return -EINVAL;
  53         if (unlikely(!pskb_may_pull(skb, len)))
  54                 return -ENOMEM;
  55         return 0;
  56 }
  57
  58 static bool arphdr_ok(struct sk_buff *skb)
  59 {
  60         return pskb_may_pull(skb, skb_network_offset(skb) +
  61                                   sizeof(struct arp_eth_header));
  62 }
  63
  64 static int check_iphdr(struct sk_buff *skb)
  65 {
  66         unsigned int nh_ofs = skb_network_offset(skb);
  67         unsigned int ip_len;
  68         int err;
  69
  70         err = check_header(skb, nh_ofs + sizeof(struct iphdr));
  71         if (unlikely(err))
  72                 return err;
  73
  74         ip_len = ip_hdrlen(skb);
  75         if (unlikely(ip_len < sizeof(struct iphdr) ||
  76                      skb->len < nh_ofs + ip_len))
  77                 return -EINVAL;
  78
  79         skb_set_transport_header(skb, nh_ofs + ip_len);
  80         return 0;
  81 }
  82
  83 static bool tcphdr_ok(struct sk_buff *skb)
  84 {
  85         int th_ofs = skb_transport_offset(skb);
  86         int tcp_len;
  87
  88         if (unlikely(!pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))))
  89                 return false;
  90
  91         tcp_len = tcp_hdrlen(skb);
  92         if (unlikely(tcp_len < sizeof(struct tcphdr) ||
  93                      skb->len < th_ofs + tcp_len))
  94                 return false;
  95
  96         return true;
  97 }
  98
  99 static bool udphdr_ok(struct sk_buff *skb)
 100 {
 101         return pskb_may_pull(skb, skb_transport_offset(skb) +
 102                                   sizeof(struct udphdr));
 103 }
 104
 105 static bool icmphdr_ok(struct sk_buff *skb)
 106 {
 107         return pskb_may_pull(skb, skb_transport_offset(skb) +
 108                                   sizeof(struct icmphdr));
 109 }
 110
 111 u64 ovs_flow_used_time(unsigned long flow_jiffies)
 112 {
 113         struct timespec cur_ts;
 114         u64 cur_ms, idle_ms;
 115
 116         ktime_get_ts(&cur_ts);
 117         idle_ms = jiffies_to_msecs(jiffies - flow_jiffies);
 118         cur_ms = (u64)cur_ts.tv_sec * MSEC_PER_SEC +
 119                  cur_ts.tv_nsec / NSEC_PER_MSEC;
 120
 121         return cur_ms - idle_ms;
 122 }
 123
 124 #define SW_FLOW_KEY_OFFSET(field)               \
 125         (offsetof(struct sw_flow_key, field) +  \
 126          FIELD_SIZEOF(struct sw_flow_key, field))
 127
 128 static int parse_ipv6hdr(struct sk_buff *skb, struct sw_flow_key *key,
 129                          int *key_lenp)
 130 {
 131         unsigned int nh_ofs = skb_network_offset(skb);
 132         unsigned int nh_len;
 133         int payload_ofs;
 134         struct ipv6hdr *nh;
 135         uint8_t nexthdr;
 136         __be16 frag_off;
 137         int err;
 138
 139         *key_lenp = SW_FLOW_KEY_OFFSET(ipv6.label);
 140
 141         err = check_header(skb, nh_ofs + sizeof(*nh));
 142         if (unlikely(err))
 143                 return err;
 144
 145         nh = ipv6_hdr(skb);
 146         nexthdr = nh->nexthdr;
 147         payload_ofs = (u8 *)(nh + 1) - skb->data;
 148
 149         key->ip.proto = NEXTHDR_NONE;
 150         key->ip.tos = ipv6_get_dsfield(nh);
 151         key->ip.ttl = nh->hop_limit;
 152         key->ipv6.label = *(__be32 *)nh & htonl(IPV6_FLOWINFO_FLOWLABEL);
 153         key->ipv6.addr.src = nh->saddr;
 154         key->ipv6.addr.dst = nh->daddr;
 155
 156         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 157         if (unlikely(payload_ofs < 0))
 158                 return -EINVAL;
 159
 160         if (frag_off) {
 161                 if (frag_off & htons(~0x7))
 162                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 163                 else
 164                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 165         }
 166
 167         nh_len = payload_ofs - nh_ofs;
 168         skb_set_transport_header(skb, nh_ofs + nh_len);
 169         key->ip.proto = nexthdr;
 170         return nh_len;
 171 }
 172
 173 static bool icmp6hdr_ok(struct sk_buff *skb)
 174 {
 175         return pskb_may_pull(skb, skb_transport_offset(skb) +
 176                                   sizeof(struct icmp6hdr));
 177 }
 178
 179 #define TCP_FLAGS_OFFSET 13
 180 #define TCP_FLAG_MASK 0x3f
 181
 182 void ovs_flow_used(struct sw_flow *flow, struct sk_buff *skb)
 183 {
 184         u8 tcp_flags = 0;
 185
 186         if ((flow->key.eth.type == htons(ETH_P_IP) ||
 187              flow->key.eth.type == htons(ETH_P_IPV6)) &&
 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(int size)
 203 {
 204         struct sw_flow_actions *sfa;
 205
 206         if (size > MAX_ACTIONS_BUFSIZE)
 207                 return ERR_PTR(-EINVAL);
 208
 209         sfa = kmalloc(sizeof(*sfa) + size, GFP_KERNEL);
 210         if (!sfa)
 211                 return ERR_PTR(-ENOMEM);
 212
 213         sfa->actions_len = 0;
 214         return sfa;
 215 }
 216
 217 struct sw_flow *ovs_flow_alloc(void)
 218 {
 219         struct sw_flow *flow;
 220
 221         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 222         if (!flow)
 223                 return ERR_PTR(-ENOMEM);
 224
 225         spin_lock_init(&flow->lock);
 226         flow->sf_acts = NULL;
 227
 228         return flow;
 229 }
 230
 231 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 232 {
 233         hash = jhash_1word(hash, table->hash_seed);
 234         return flex_array_get(table->buckets,
 235                                 (hash & (table->n_buckets - 1)));
 236 }
 237
 238 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 239 {
 240         struct flex_array *buckets;
 241         int i, err;
 242
 243         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 244                                    n_buckets, GFP_KERNEL);
 245         if (!buckets)
 246                 return NULL;
 247
 248         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 249         if (err) {
 250                 flex_array_free(buckets);
 251                 return NULL;
 252         }
 253
 254         for (i = 0; i < n_buckets; i++)
 255                 INIT_HLIST_HEAD((struct hlist_head *)
 256                                         flex_array_get(buckets, i));
 257
 258         return buckets;
 259 }
 260
 261 static void free_buckets(struct flex_array *buckets)
 262 {
 263         flex_array_free(buckets);
 264 }
 265
 266 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 267 {
 268         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 269
 270         if (!table)
 271                 return NULL;
 272
 273         table->buckets = alloc_buckets(new_size);
 274
 275         if (!table->buckets) {
 276                 kfree(table);
 277                 return NULL;
 278         }
 279         table->n_buckets = new_size;
 280         table->count = 0;
 281         table->node_ver = 0;
 282         table->keep_flows = false;
 283         get_random_bytes(&table->hash_seed, sizeof(u32));
 284
 285         return table;
 286 }
 287
 288 void ovs_flow_tbl_destroy(struct flow_table *table)
 289 {
 290         int i;
 291
 292         if (!table)
 293                 return;
 294
 295         if (table->keep_flows)
 296                 goto skip_flows;
 297
 298         for (i = 0; i < table->n_buckets; i++) {
 299                 struct sw_flow *flow;
 300                 struct hlist_head *head = flex_array_get(table->buckets, i);
 301                 struct hlist_node *n;
 302                 int ver = table->node_ver;
 303
 304                 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
 305                         hlist_del_rcu(&flow->hash_node[ver]);
 306                         ovs_flow_free(flow);
 307                 }
 308         }
 309
 310 skip_flows:
 311         free_buckets(table->buckets);
 312         kfree(table);
 313 }
 314
 315 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 316 {
 317         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 318
 319         ovs_flow_tbl_destroy(table);
 320 }
 321
 322 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 323 {
 324         if (!table)
 325                 return;
 326
 327         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 328 }
 329
 330 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 331 {
 332         struct sw_flow *flow;
 333         struct hlist_head *head;
 334         int ver;
 335         int i;
 336
 337         ver = table->node_ver;
 338         while (*bucket < table->n_buckets) {
 339                 i = 0;
 340                 head = flex_array_get(table->buckets, *bucket);
 341                 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
 342                         if (i < *last) {
 343                                 i++;
 344                                 continue;
 345                         }
 346                         *last = i + 1;
 347                         return flow;
 348                 }
 349                 (*bucket)++;
 350                 *last = 0;
 351         }
 352
 353         return NULL;
 354 }
 355
 356 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 357 {
 358         int old_ver;
 359         int i;
 360
 361         old_ver = old->node_ver;
 362         new->node_ver = !old_ver;
 363
 364         /* Insert in new table. */
 365         for (i = 0; i < old->n_buckets; i++) {
 366                 struct sw_flow *flow;
 367                 struct hlist_head *head;
 368
 369                 head = flex_array_get(old->buckets, i);
 370
 371                 hlist_for_each_entry(flow, head, hash_node[old_ver])
 372                         ovs_flow_tbl_insert(new, flow);
 373         }
 374         old->keep_flows = true;
 375 }
 376
 377 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 378 {
 379         struct flow_table *new_table;
 380
 381         new_table = ovs_flow_tbl_alloc(n_buckets);
 382         if (!new_table)
 383                 return ERR_PTR(-ENOMEM);
 384
 385         flow_table_copy_flows(table, new_table);
 386
 387         return new_table;
 388 }
 389
 390 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 391 {
 392         return __flow_tbl_rehash(table, table->n_buckets);
 393 }
 394
 395 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 396 {
 397         return __flow_tbl_rehash(table, table->n_buckets * 2);
 398 }
 399
 400 void ovs_flow_free(struct sw_flow *flow)
 401 {
 402         if (unlikely(!flow))
 403                 return;
 404
 405         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 406         kmem_cache_free(flow_cache, flow);
 407 }
 408
 409 /* RCU callback used by ovs_flow_deferred_free. */
 410 static void rcu_free_flow_callback(struct rcu_head *rcu)
 411 {
 412         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 413
 414         ovs_flow_free(flow);
 415 }
 416
 417 /* Schedules 'flow' to be freed after the next RCU grace period.
 418  * The caller must hold rcu_read_lock for this to be sensible. */
 419 void ovs_flow_deferred_free(struct sw_flow *flow)
 420 {
 421         call_rcu(&flow->rcu, rcu_free_flow_callback);
 422 }
 423
 424 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 425  * The caller must hold rcu_read_lock for this to be sensible. */
 426 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 427 {
 428         kfree_rcu(sf_acts, rcu);
 429 }
 430
 431 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 432 {
 433         struct qtag_prefix {
 434                 __be16 eth_type; /* ETH_P_8021Q */
 435                 __be16 tci;
 436         };
 437         struct qtag_prefix *qp;
 438
 439         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 440                 return 0;
 441
 442         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 443                                          sizeof(__be16))))
 444                 return -ENOMEM;
 445
 446         qp = (struct qtag_prefix *) skb->data;
 447         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 448         __skb_pull(skb, sizeof(struct qtag_prefix));
 449
 450         return 0;
 451 }
 452
 453 static __be16 parse_ethertype(struct sk_buff *skb)
 454 {
 455         struct llc_snap_hdr {
 456                 u8  dsap;  /* Always 0xAA */
 457                 u8  ssap;  /* Always 0xAA */
 458                 u8  ctrl;
 459                 u8  oui[3];
 460                 __be16 ethertype;
 461         };
 462         struct llc_snap_hdr *llc;
 463         __be16 proto;
 464
 465         proto = *(__be16 *) skb->data;
 466         __skb_pull(skb, sizeof(__be16));
 467
 468         if (ntohs(proto) >= ETH_P_802_3_MIN)
 469                 return proto;
 470
 471         if (skb->len < sizeof(struct llc_snap_hdr))
 472                 return htons(ETH_P_802_2);
 473
 474         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 475                 return htons(0);
 476
 477         llc = (struct llc_snap_hdr *) skb->data;
 478         if (llc->dsap != LLC_SAP_SNAP ||
 479             llc->ssap != LLC_SAP_SNAP ||
 480             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 481                 return htons(ETH_P_802_2);
 482
 483         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 484
 485         if (ntohs(llc->ethertype) >= ETH_P_802_3_MIN)
 486                 return llc->ethertype;
 487
 488         return htons(ETH_P_802_2);
 489 }
 490
 491 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 492                         int *key_lenp, int nh_len)
 493 {
 494         struct icmp6hdr *icmp = icmp6_hdr(skb);
 495         int error = 0;
 496         int key_len;
 497
 498         /* The ICMPv6 type and code fields use the 16-bit transport port
 499          * fields, so we need to store them in 16-bit network byte order.
 500          */
 501         key->ipv6.tp.src = htons(icmp->icmp6_type);
 502         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 503         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 504
 505         if (icmp->icmp6_code == 0 &&
 506             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 507              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 508                 int icmp_len = skb->len - skb_transport_offset(skb);
 509                 struct nd_msg *nd;
 510                 int offset;
 511
 512                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 513
 514                 /* In order to process neighbor discovery options, we need the
 515                  * entire packet.
 516                  */
 517                 if (unlikely(icmp_len < sizeof(*nd)))
 518                         goto out;
 519                 if (unlikely(skb_linearize(skb))) {
 520                         error = -ENOMEM;
 521                         goto out;
 522                 }
 523
 524                 nd = (struct nd_msg *)skb_transport_header(skb);
 525                 key->ipv6.nd.target = nd->target;
 526                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 527
 528                 icmp_len -= sizeof(*nd);
 529                 offset = 0;
 530                 while (icmp_len >= 8) {
 531                         struct nd_opt_hdr *nd_opt =
 532                                  (struct nd_opt_hdr *)(nd->opt + offset);
 533                         int opt_len = nd_opt->nd_opt_len * 8;
 534
 535                         if (unlikely(!opt_len || opt_len > icmp_len))
 536                                 goto invalid;
 537
 538                         /* Store the link layer address if the appropriate
 539                          * option is provided.  It is considered an error if
 540                          * the same link layer option is specified twice.
 541                          */
 542                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 543                             && opt_len == 8) {
 544                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 545                                         goto invalid;
 546                                 memcpy(key->ipv6.nd.sll,
 547                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 548                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 549                                    && opt_len == 8) {
 550                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 551                                         goto invalid;
 552                                 memcpy(key->ipv6.nd.tll,
 553                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 554                         }
 555
 556                         icmp_len -= opt_len;
 557                         offset += opt_len;
 558                 }
 559         }
 560
 561         goto out;
 562
 563 invalid:
 564         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 565         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 566         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 567
 568 out:
 569         *key_lenp = key_len;
 570         return error;
 571 }
 572
 573 /**
 574  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 575  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 576  * Ethernet header
 577  * @in_port: port number on which @skb was received.
 578  * @key: output flow key
 579  * @key_lenp: length of output flow key
 580  *
 581  * The caller must ensure that skb->len >= ETH_HLEN.
 582  *
 583  * Returns 0 if successful, otherwise a negative errno value.
 584  *
 585  * Initializes @skb header pointers as follows:
 586  *
 587  *    - skb->mac_header: the Ethernet header.
 588  *
 589  *    - skb->network_header: just past the Ethernet header, or just past the
 590  *      VLAN header, to the first byte of the Ethernet payload.
 591  *
 592  *    - skb->transport_header: If key->eth.type is ETH_P_IP or ETH_P_IPV6
 593  *      on output, then just past the IP header, if one is present and
 594  *      of a correct length, otherwise the same as skb->network_header.
 595  *      For other key->eth.type values it is left untouched.
 596  */
 597 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 598                  int *key_lenp)
 599 {
 600         int error = 0;
 601         int key_len = SW_FLOW_KEY_OFFSET(eth);
 602         struct ethhdr *eth;
 603
 604         memset(key, 0, sizeof(*key));
 605
 606         key->phy.priority = skb->priority;
 607         if (OVS_CB(skb)->tun_key)
 608                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 609         key->phy.in_port = in_port;
 610         key->phy.skb_mark = skb->mark;
 611
 612         skb_reset_mac_header(skb);
 613
 614         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 615          * header in the linear data area.
 616          */
 617         eth = eth_hdr(skb);
 618         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 619         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 620
 621         __skb_pull(skb, 2 * ETH_ALEN);
 622         /* We are going to push all headers that we pull, so no need to
 623          * update skb->csum here.
 624          */
 625
 626         if (vlan_tx_tag_present(skb))
 627                 key->eth.tci = htons(skb->vlan_tci);
 628         else if (eth->h_proto == htons(ETH_P_8021Q))
 629                 if (unlikely(parse_vlan(skb, key)))
 630                         return -ENOMEM;
 631
 632         key->eth.type = parse_ethertype(skb);
 633         if (unlikely(key->eth.type == htons(0)))
 634                 return -ENOMEM;
 635
 636         skb_reset_network_header(skb);
 637         __skb_push(skb, skb->data - skb_mac_header(skb));
 638
 639         /* Network layer. */
 640         if (key->eth.type == htons(ETH_P_IP)) {
 641                 struct iphdr *nh;
 642                 __be16 offset;
 643
 644                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 645
 646                 error = check_iphdr(skb);
 647                 if (unlikely(error)) {
 648                         if (error == -EINVAL) {
 649                                 skb->transport_header = skb->network_header;
 650                                 error = 0;
 651                         }
 652                         goto out;
 653                 }
 654
 655                 nh = ip_hdr(skb);
 656                 key->ipv4.addr.src = nh->saddr;
 657                 key->ipv4.addr.dst = nh->daddr;
 658
 659                 key->ip.proto = nh->protocol;
 660                 key->ip.tos = nh->tos;
 661                 key->ip.ttl = nh->ttl;
 662
 663                 offset = nh->frag_off & htons(IP_OFFSET);
 664                 if (offset) {
 665                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 666                         goto out;
 667                 }
 668                 if (nh->frag_off & htons(IP_MF) ||
 669                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 670                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 671
 672                 /* Transport layer. */
 673                 if (key->ip.proto == IPPROTO_TCP) {
 674                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 675                         if (tcphdr_ok(skb)) {
 676                                 struct tcphdr *tcp = tcp_hdr(skb);
 677                                 key->ipv4.tp.src = tcp->source;
 678                                 key->ipv4.tp.dst = tcp->dest;
 679                         }
 680                 } else if (key->ip.proto == IPPROTO_UDP) {
 681                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 682                         if (udphdr_ok(skb)) {
 683                                 struct udphdr *udp = udp_hdr(skb);
 684                                 key->ipv4.tp.src = udp->source;
 685                                 key->ipv4.tp.dst = udp->dest;
 686                         }
 687                 } else if (key->ip.proto == IPPROTO_ICMP) {
 688                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 689                         if (icmphdr_ok(skb)) {
 690                                 struct icmphdr *icmp = icmp_hdr(skb);
 691                                 /* The ICMP type and code fields use the 16-bit
 692                                  * transport port fields, so we need to store
 693                                  * them in 16-bit network byte order. */
 694                                 key->ipv4.tp.src = htons(icmp->type);
 695                                 key->ipv4.tp.dst = htons(icmp->code);
 696                         }
 697                 }
 698
 699         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 700                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 701                 struct arp_eth_header *arp;
 702
 703                 arp = (struct arp_eth_header *)skb_network_header(skb);
 704
 705                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 706                                 && arp->ar_pro == htons(ETH_P_IP)
 707                                 && arp->ar_hln == ETH_ALEN
 708                                 && arp->ar_pln == 4) {
 709
 710                         /* We only match on the lower 8 bits of the opcode. */
 711                         if (ntohs(arp->ar_op) <= 0xff)
 712                                 key->ip.proto = ntohs(arp->ar_op);
 713                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 714                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 715                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 716                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 717                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 718                 }
 719         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 720                 int nh_len;             /* IPv6 Header + Extensions */
 721
 722                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 723                 if (unlikely(nh_len < 0)) {
 724                         if (nh_len == -EINVAL)
 725                                 skb->transport_header = skb->network_header;
 726                         else
 727                                 error = nh_len;
 728                         goto out;
 729                 }
 730
 731                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 732                         goto out;
 733                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 734                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 735
 736                 /* Transport layer. */
 737                 if (key->ip.proto == NEXTHDR_TCP) {
 738                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 739                         if (tcphdr_ok(skb)) {
 740                                 struct tcphdr *tcp = tcp_hdr(skb);
 741                                 key->ipv6.tp.src = tcp->source;
 742                                 key->ipv6.tp.dst = tcp->dest;
 743                         }
 744                 } else if (key->ip.proto == NEXTHDR_UDP) {
 745                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 746                         if (udphdr_ok(skb)) {
 747                                 struct udphdr *udp = udp_hdr(skb);
 748                                 key->ipv6.tp.src = udp->source;
 749                                 key->ipv6.tp.dst = udp->dest;
 750                         }
 751                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 752                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 753                         if (icmp6hdr_ok(skb)) {
 754                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 755                                 if (error < 0)
 756                                         goto out;
 757                         }
 758                 }
 759         }
 760
 761 out:
 762         *key_lenp = key_len;
 763         return error;
 764 }
 765
 766 u32 ovs_flow_hash(const struct sw_flow_key *key, int key_len)
 767 {
 768         return jhash2((u32 *)key, DIV_ROUND_UP(key_len, sizeof(u32)), 0);
 769 }
 770
 771 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 772                                 struct sw_flow_key *key, int key_len)
 773 {
 774         struct sw_flow *flow;
 775         struct hlist_head *head;
 776         u32 hash;
 777
 778         hash = ovs_flow_hash(key, key_len);
 779
 780         head = find_bucket(table, hash);
 781         hlist_for_each_entry_rcu(flow, head, hash_node[table->node_ver]) {
 782
 783                 if (flow->hash == hash &&
 784                     !memcmp(&flow->key, key, key_len)) {
 785                         return flow;
 786                 }
 787         }
 788         return NULL;
 789 }
 790
 791 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 792 {
 793         struct hlist_head *head;
 794
 795         head = find_bucket(table, flow->hash);
 796         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 797         table->count++;
 798 }
 799
 800 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 801 {
 802         BUG_ON(table->count == 0);
 803         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 804         table->count--;
 805 }
 806
 807 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 808 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 809         [OVS_KEY_ATTR_ENCAP] = -1,
 810         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 811         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 812         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 813         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 814         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 815         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 816         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 817         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 818         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 819         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 820         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 821         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 822         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 823         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 824         [OVS_KEY_ATTR_TUNNEL] = -1,
 825 };
 826
 827 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 828                                   const struct nlattr *a[], u32 *attrs)
 829 {
 830         const struct ovs_key_icmp *icmp_key;
 831         const struct ovs_key_tcp *tcp_key;
 832         const struct ovs_key_udp *udp_key;
 833
 834         switch (swkey->ip.proto) {
 835         case IPPROTO_TCP:
 836                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 837                         return -EINVAL;
 838                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 839
 840                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 841                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 842                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 843                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 844                 break;
 845
 846         case IPPROTO_UDP:
 847                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 848                         return -EINVAL;
 849                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 850
 851                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 852                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 853                 swkey->ipv4.tp.src = udp_key->udp_src;
 854                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 855                 break;
 856
 857         case IPPROTO_ICMP:
 858                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 859                         return -EINVAL;
 860                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 861
 862                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 863                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 864                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 865                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 866                 break;
 867         }
 868
 869         return 0;
 870 }
 871
 872 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 873                                   const struct nlattr *a[], u32 *attrs)
 874 {
 875         const struct ovs_key_icmpv6 *icmpv6_key;
 876         const struct ovs_key_tcp *tcp_key;
 877         const struct ovs_key_udp *udp_key;
 878
 879         switch (swkey->ip.proto) {
 880         case IPPROTO_TCP:
 881                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 882                         return -EINVAL;
 883                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 884
 885                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 886                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 887                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 888                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 889                 break;
 890
 891         case IPPROTO_UDP:
 892                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 893                         return -EINVAL;
 894                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 895
 896                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 897                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 898                 swkey->ipv6.tp.src = udp_key->udp_src;
 899                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 900                 break;
 901
 902         case IPPROTO_ICMPV6:
 903                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 904                         return -EINVAL;
 905                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 906
 907                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 908                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 909                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 910                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 911
 912                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 913                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 914                         const struct ovs_key_nd *nd_key;
 915
 916                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 917                                 return -EINVAL;
 918                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 919
 920                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 921                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 922                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 923                                sizeof(swkey->ipv6.nd.target));
 924                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 925                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 926                 }
 927                 break;
 928         }
 929
 930         return 0;
 931 }
 932
 933 static int parse_flow_nlattrs(const struct nlattr *attr,
 934                               const struct nlattr *a[], u32 *attrsp)
 935 {
 936         const struct nlattr *nla;
 937         u32 attrs;
 938         int rem;
 939
 940         attrs = 0;
 941         nla_for_each_nested(nla, attr, rem) {
 942                 u16 type = nla_type(nla);
 943                 int expected_len;
 944
 945                 if (type > OVS_KEY_ATTR_MAX || attrs & (1 << type))
 946                         return -EINVAL;
 947
 948                 expected_len = ovs_key_lens[type];
 949                 if (nla_len(nla) != expected_len && expected_len != -1)
 950                         return -EINVAL;
 951
 952                 attrs |= 1 << type;
 953                 a[type] = nla;
 954         }
 955         if (rem)
 956                 return -EINVAL;
 957
 958         *attrsp = attrs;
 959         return 0;
 960 }
 961
 962 int ovs_ipv4_tun_from_nlattr(const struct nlattr *attr,
 963                              struct ovs_key_ipv4_tunnel *tun_key)
 964 {
 965         struct nlattr *a;
 966         int rem;
 967         bool ttl = false;
 968
 969         memset(tun_key, 0, sizeof(*tun_key));
 970
 971         nla_for_each_nested(a, attr, rem) {
 972                 int type = nla_type(a);
 973                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
 974                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
 975                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
 976                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
 977                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
 978                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
 979                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
 980                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
 981                 };
 982
 983                 if (type > OVS_TUNNEL_KEY_ATTR_MAX ||
 984                         ovs_tunnel_key_lens[type] != nla_len(a))
 985                         return -EINVAL;
 986
 987                 switch (type) {
 988                 case OVS_TUNNEL_KEY_ATTR_ID:
 989                         tun_key->tun_id = nla_get_be64(a);
 990                         tun_key->tun_flags |= TUNNEL_KEY;
 991                         break;
 992                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
 993                         tun_key->ipv4_src = nla_get_be32(a);
 994                         break;
 995                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
 996                         tun_key->ipv4_dst = nla_get_be32(a);
 997                         break;
 998                 case OVS_TUNNEL_KEY_ATTR_TOS:
 999                         tun_key->ipv4_tos = nla_get_u8(a);
1000                         break;
1001                 case OVS_TUNNEL_KEY_ATTR_TTL:
1002                         tun_key->ipv4_ttl = nla_get_u8(a);
1003                         ttl = true;
1004                         break;
1005                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1006                         tun_key->tun_flags |= TUNNEL_DONT_FRAGMENT;
1007                         break;
1008                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1009                         tun_key->tun_flags |= TUNNEL_CSUM;
1010                         break;
1011                 default:
1012                         return -EINVAL;
1013
1014                 }
1015         }
1016         if (rem > 0)
1017                 return -EINVAL;
1018
1019         if (!tun_key->ipv4_dst)
1020                 return -EINVAL;
1021
1022         if (!ttl)
1023                 return -EINVAL;
1024
1025         return 0;
1026 }
1027
1028 int ovs_ipv4_tun_to_nlattr(struct sk_buff *skb,
1029                            const struct ovs_key_ipv4_tunnel *tun_key)
1030 {
1031         struct nlattr *nla;
1032
1033         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1034         if (!nla)
1035                 return -EMSGSIZE;
1036
1037         if (tun_key->tun_flags & TUNNEL_KEY &&
1038             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id))
1039                 return -EMSGSIZE;
1040         if (tun_key->ipv4_src &&
1041             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ipv4_src))
1042                 return -EMSGSIZE;
1043         if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ipv4_dst))
1044                 return -EMSGSIZE;
1045         if (tun_key->ipv4_tos &&
1046             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ipv4_tos))
1047                 return -EMSGSIZE;
1048         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ipv4_ttl))
1049                 return -EMSGSIZE;
1050         if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) &&
1051                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1052                 return -EMSGSIZE;
1053         if ((tun_key->tun_flags & TUNNEL_CSUM) &&
1054                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1055                 return -EMSGSIZE;
1056
1057         nla_nest_end(skb, nla);
1058         return 0;
1059 }
1060
1061 /**
1062  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
1063  * @swkey: receives the extracted flow key.
1064  * @key_lenp: number of bytes used in @swkey.
1065  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1066  * sequence.
1067  */
1068 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1069                       const struct nlattr *attr)
1070 {
1071         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1072         const struct ovs_key_ethernet *eth_key;
1073         int key_len;
1074         u32 attrs;
1075         int err;
1076
1077         memset(swkey, 0, sizeof(struct sw_flow_key));
1078         key_len = SW_FLOW_KEY_OFFSET(eth);
1079
1080         err = parse_flow_nlattrs(attr, a, &attrs);
1081         if (err)
1082                 return err;
1083
1084         /* Metadata attributes. */
1085         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1086                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1087                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1088         }
1089         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1090                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1091                 if (in_port >= DP_MAX_PORTS)
1092                         return -EINVAL;
1093                 swkey->phy.in_port = in_port;
1094                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1095         } else {
1096                 swkey->phy.in_port = DP_MAX_PORTS;
1097         }
1098         if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1099                 swkey->phy.skb_mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1100                 attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1101         }
1102
1103         if (attrs & (1 << OVS_KEY_ATTR_TUNNEL)) {
1104                 err = ovs_ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1105                 if (err)
1106                         return err;
1107
1108                 attrs &= ~(1 << OVS_KEY_ATTR_TUNNEL);
1109         }
1110
1111         /* Data attributes. */
1112         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1113                 return -EINVAL;
1114         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1115
1116         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1117         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1118         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1119
1120         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1121             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1122                 const struct nlattr *encap;
1123                 __be16 tci;
1124
1125                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1126                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1127                               (1 << OVS_KEY_ATTR_ENCAP)))
1128                         return -EINVAL;
1129
1130                 encap = a[OVS_KEY_ATTR_ENCAP];
1131                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1132                 if (tci & htons(VLAN_TAG_PRESENT)) {
1133                         swkey->eth.tci = tci;
1134
1135                         err = parse_flow_nlattrs(encap, a, &attrs);
1136                         if (err)
1137                                 return err;
1138                 } else if (!tci) {
1139                         /* Corner case for truncated 802.1Q header. */
1140                         if (nla_len(encap))
1141                                 return -EINVAL;
1142
1143                         swkey->eth.type = htons(ETH_P_8021Q);
1144                         *key_lenp = key_len;
1145                         return 0;
1146                 } else {
1147                         return -EINVAL;
1148                 }
1149         }
1150
1151         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1152                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1153                 if (ntohs(swkey->eth.type) < ETH_P_802_3_MIN)
1154                         return -EINVAL;
1155                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1156         } else {
1157                 swkey->eth.type = htons(ETH_P_802_2);
1158         }
1159
1160         if (swkey->eth.type == htons(ETH_P_IP)) {
1161                 const struct ovs_key_ipv4 *ipv4_key;
1162
1163                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1164                         return -EINVAL;
1165                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1166
1167                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1168                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1169                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1170                         return -EINVAL;
1171                 swkey->ip.proto = ipv4_key->ipv4_proto;
1172                 swkey->ip.tos = ipv4_key->ipv4_tos;
1173                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1174                 swkey->ip.frag = ipv4_key->ipv4_frag;
1175                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1176                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1177
1178                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1179                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1180                         if (err)
1181                                 return err;
1182                 }
1183         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1184                 const struct ovs_key_ipv6 *ipv6_key;
1185
1186                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1187                         return -EINVAL;
1188                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1189
1190                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1191                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1192                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1193                         return -EINVAL;
1194                 swkey->ipv6.label = ipv6_key->ipv6_label;
1195                 swkey->ip.proto = ipv6_key->ipv6_proto;
1196                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1197                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1198                 swkey->ip.frag = ipv6_key->ipv6_frag;
1199                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1200                        sizeof(swkey->ipv6.addr.src));
1201                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1202                        sizeof(swkey->ipv6.addr.dst));
1203
1204                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1205                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1206                         if (err)
1207                                 return err;
1208                 }
1209         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1210                    swkey->eth.type == htons(ETH_P_RARP)) {
1211                 const struct ovs_key_arp *arp_key;
1212
1213                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1214                         return -EINVAL;
1215                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1216
1217                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1218                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1219                 swkey->ipv4.addr.src = arp_key->arp_sip;
1220                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1221                 if (arp_key->arp_op & htons(0xff00))
1222                         return -EINVAL;
1223                 swkey->ip.proto = ntohs(arp_key->arp_op);
1224                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1225                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1226         }
1227
1228         if (attrs)
1229                 return -EINVAL;
1230         *key_lenp = key_len;
1231
1232         return 0;
1233 }
1234
1235 /**
1236  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1237  * @flow: Receives extracted in_port, priority, tun_key and skb_mark.
1238  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1239  * sequence.
1240  *
1241  * This parses a series of Netlink attributes that form a flow key, which must
1242  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1243  * get the metadata, that is, the parts of the flow key that cannot be
1244  * extracted from the packet itself.
1245  */
1246 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow,
1247                                    const struct nlattr *attr)
1248 {
1249         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1250         const struct nlattr *nla;
1251         int rem;
1252
1253         flow->key.phy.in_port = DP_MAX_PORTS;
1254         flow->key.phy.priority = 0;
1255         flow->key.phy.skb_mark = 0;
1256         memset(tun_key, 0, sizeof(flow->key.tun_key));
1257
1258         nla_for_each_nested(nla, attr, rem) {
1259                 int type = nla_type(nla);
1260
1261                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1262                         int err;
1263
1264                         if (nla_len(nla) != ovs_key_lens[type])
1265                                 return -EINVAL;
1266
1267                         switch (type) {
1268                         case OVS_KEY_ATTR_PRIORITY:
1269                                 flow->key.phy.priority = nla_get_u32(nla);
1270                                 break;
1271
1272                         case OVS_KEY_ATTR_TUNNEL:
1273                                 err = ovs_ipv4_tun_from_nlattr(nla, tun_key);
1274                                 if (err)
1275                                         return err;
1276                                 break;
1277
1278                         case OVS_KEY_ATTR_IN_PORT:
1279                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1280                                         return -EINVAL;
1281                                 flow->key.phy.in_port = nla_get_u32(nla);
1282                                 break;
1283
1284                         case OVS_KEY_ATTR_SKB_MARK:
1285                                 flow->key.phy.skb_mark = nla_get_u32(nla);
1286                                 break;
1287                         }
1288                 }
1289         }
1290         if (rem)
1291                 return -EINVAL;
1292         return 0;
1293 }
1294
1295 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1296 {
1297         struct ovs_key_ethernet *eth_key;
1298         struct nlattr *nla, *encap;
1299
1300         if (swkey->phy.priority &&
1301             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1302                 goto nla_put_failure;
1303
1304         if (swkey->tun_key.ipv4_dst &&
1305             ovs_ipv4_tun_to_nlattr(skb, &swkey->tun_key))
1306                 goto nla_put_failure;
1307
1308         if (swkey->phy.in_port != DP_MAX_PORTS &&
1309             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1310                 goto nla_put_failure;
1311
1312         if (swkey->phy.skb_mark &&
1313             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1314                 goto nla_put_failure;
1315
1316         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1317         if (!nla)
1318                 goto nla_put_failure;
1319         eth_key = nla_data(nla);
1320         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1321         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1322
1323         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1324                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1325                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1326                         goto nla_put_failure;
1327                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1328                 if (!swkey->eth.tci)
1329                         goto unencap;
1330         } else {
1331                 encap = NULL;
1332         }
1333
1334         if (swkey->eth.type == htons(ETH_P_802_2))
1335                 goto unencap;
1336
1337         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1338                 goto nla_put_failure;
1339
1340         if (swkey->eth.type == htons(ETH_P_IP)) {
1341                 struct ovs_key_ipv4 *ipv4_key;
1342
1343                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1344                 if (!nla)
1345                         goto nla_put_failure;
1346                 ipv4_key = nla_data(nla);
1347                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1348                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1349                 ipv4_key->ipv4_proto = swkey->ip.proto;
1350                 ipv4_key->ipv4_tos = swkey->ip.tos;
1351                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1352                 ipv4_key->ipv4_frag = swkey->ip.frag;
1353         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1354                 struct ovs_key_ipv6 *ipv6_key;
1355
1356                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1357                 if (!nla)
1358                         goto nla_put_failure;
1359                 ipv6_key = nla_data(nla);
1360                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1361                                 sizeof(ipv6_key->ipv6_src));
1362                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1363                                 sizeof(ipv6_key->ipv6_dst));
1364                 ipv6_key->ipv6_label = swkey->ipv6.label;
1365                 ipv6_key->ipv6_proto = swkey->ip.proto;
1366                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1367                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1368                 ipv6_key->ipv6_frag = swkey->ip.frag;
1369         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1370                    swkey->eth.type == htons(ETH_P_RARP)) {
1371                 struct ovs_key_arp *arp_key;
1372
1373                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1374                 if (!nla)
1375                         goto nla_put_failure;
1376                 arp_key = nla_data(nla);
1377                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1378                 arp_key->arp_sip = swkey->ipv4.addr.src;
1379                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1380                 arp_key->arp_op = htons(swkey->ip.proto);
1381                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1382                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1383         }
1384
1385         if ((swkey->eth.type == htons(ETH_P_IP) ||
1386              swkey->eth.type == htons(ETH_P_IPV6)) &&
1387              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1388
1389                 if (swkey->ip.proto == IPPROTO_TCP) {
1390                         struct ovs_key_tcp *tcp_key;
1391
1392                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1393                         if (!nla)
1394                                 goto nla_put_failure;
1395                         tcp_key = nla_data(nla);
1396                         if (swkey->eth.type == htons(ETH_P_IP)) {
1397                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1398                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1399                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1400                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1401                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1402                         }
1403                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1404                         struct ovs_key_udp *udp_key;
1405
1406                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1407                         if (!nla)
1408                                 goto nla_put_failure;
1409                         udp_key = nla_data(nla);
1410                         if (swkey->eth.type == htons(ETH_P_IP)) {
1411                                 udp_key->udp_src = swkey->ipv4.tp.src;
1412                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1413                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1414                                 udp_key->udp_src = swkey->ipv6.tp.src;
1415                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1416                         }
1417                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1418                            swkey->ip.proto == IPPROTO_ICMP) {
1419                         struct ovs_key_icmp *icmp_key;
1420
1421                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1422                         if (!nla)
1423                                 goto nla_put_failure;
1424                         icmp_key = nla_data(nla);
1425                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1426                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1427                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1428                            swkey->ip.proto == IPPROTO_ICMPV6) {
1429                         struct ovs_key_icmpv6 *icmpv6_key;
1430
1431                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1432                                                 sizeof(*icmpv6_key));
1433                         if (!nla)
1434                                 goto nla_put_failure;
1435                         icmpv6_key = nla_data(nla);
1436                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1437                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1438
1439                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1440                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1441                                 struct ovs_key_nd *nd_key;
1442
1443                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1444                                 if (!nla)
1445                                         goto nla_put_failure;
1446                                 nd_key = nla_data(nla);
1447                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1448                                                         sizeof(nd_key->nd_target));
1449                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1450                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1451                         }
1452                 }
1453         }
1454
1455 unencap:
1456         if (encap)
1457                 nla_nest_end(skb, encap);
1458
1459         return 0;
1460
1461 nla_put_failure:
1462         return -EMSGSIZE;
1463 }
1464
1465 /* Initializes the flow module.
1466  * Returns zero if successful or a negative error code. */
1467 int ovs_flow_init(void)
1468 {
1469         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1470                                         0, NULL);
1471         if (flow_cache == NULL)
1472                 return -ENOMEM;
1473
1474         return 0;
1475 }
1476
1477 /* Uninitializes the flow module. */
1478 void ovs_flow_exit(void)
1479 {
1480         kmem_cache_destroy(flow_cache);
1481 }