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         size += sizeof(*sfa);
 211         if (size <= MAX_ACTIONS_BUFSIZE_KMALLOC)
 212                 sfa = kmalloc(size, GFP_KERNEL);
 213         else
 214                 sfa = vmalloc(size);
 215
 216         if (!sfa)
 217                 return ERR_PTR(-ENOMEM);
 218
 219         sfa->actions_len = 0;
 220         sfa->buf_size = size;
 221
 222         return sfa;
 223 }
 224
 225 void ovs_flow_actions_free(struct sw_flow_actions *sfa)
 226 {
 227         if (sfa->buf_size <= MAX_ACTIONS_BUFSIZE_KMALLOC)
 228                 kfree(sfa);
 229         else
 230                 vfree(sfa);
 231 }
 232
 233 struct sw_flow *ovs_flow_alloc(void)
 234 {
 235         struct sw_flow *flow;
 236
 237         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
 238         if (!flow)
 239                 return ERR_PTR(-ENOMEM);
 240
 241         spin_lock_init(&flow->lock);
 242         flow->sf_acts = NULL;
 243
 244         return flow;
 245 }
 246
 247 static struct hlist_head *find_bucket(struct flow_table *table, u32 hash)
 248 {
 249         hash = jhash_1word(hash, table->hash_seed);
 250         return flex_array_get(table->buckets,
 251                                 (hash & (table->n_buckets - 1)));
 252 }
 253
 254 static struct flex_array *alloc_buckets(unsigned int n_buckets)
 255 {
 256         struct flex_array *buckets;
 257         int i, err;
 258
 259         buckets = flex_array_alloc(sizeof(struct hlist_head *),
 260                                    n_buckets, GFP_KERNEL);
 261         if (!buckets)
 262                 return NULL;
 263
 264         err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
 265         if (err) {
 266                 flex_array_free(buckets);
 267                 return NULL;
 268         }
 269
 270         for (i = 0; i < n_buckets; i++)
 271                 INIT_HLIST_HEAD((struct hlist_head *)
 272                                         flex_array_get(buckets, i));
 273
 274         return buckets;
 275 }
 276
 277 static void free_buckets(struct flex_array *buckets)
 278 {
 279         flex_array_free(buckets);
 280 }
 281
 282 struct flow_table *ovs_flow_tbl_alloc(int new_size)
 283 {
 284         struct flow_table *table = kmalloc(sizeof(*table), GFP_KERNEL);
 285
 286         if (!table)
 287                 return NULL;
 288
 289         table->buckets = alloc_buckets(new_size);
 290
 291         if (!table->buckets) {
 292                 kfree(table);
 293                 return NULL;
 294         }
 295         table->n_buckets = new_size;
 296         table->count = 0;
 297         table->node_ver = 0;
 298         table->keep_flows = false;
 299         get_random_bytes(&table->hash_seed, sizeof(u32));
 300
 301         return table;
 302 }
 303
 304 void ovs_flow_tbl_destroy(struct flow_table *table)
 305 {
 306         int i;
 307
 308         if (!table)
 309                 return;
 310
 311         if (table->keep_flows)
 312                 goto skip_flows;
 313
 314         for (i = 0; i < table->n_buckets; i++) {
 315                 struct sw_flow *flow;
 316                 struct hlist_head *head = flex_array_get(table->buckets, i);
 317                 struct hlist_node *node, *n;
 318                 int ver = table->node_ver;
 319
 320                 hlist_for_each_entry_safe(flow, node, n, head, hash_node[ver]) {
 321                         hlist_del_rcu(&flow->hash_node[ver]);
 322                         ovs_flow_free(flow);
 323                 }
 324         }
 325
 326 skip_flows:
 327         free_buckets(table->buckets);
 328         kfree(table);
 329 }
 330
 331 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
 332 {
 333         struct flow_table *table = container_of(rcu, struct flow_table, rcu);
 334
 335         ovs_flow_tbl_destroy(table);
 336 }
 337
 338 void ovs_flow_tbl_deferred_destroy(struct flow_table *table)
 339 {
 340         if (!table)
 341                 return;
 342
 343         call_rcu(&table->rcu, flow_tbl_destroy_rcu_cb);
 344 }
 345
 346 struct sw_flow *ovs_flow_tbl_next(struct flow_table *table, u32 *bucket, u32 *last)
 347 {
 348         struct sw_flow *flow;
 349         struct hlist_head *head;
 350         struct hlist_node *n;
 351         int ver;
 352         int i;
 353
 354         ver = table->node_ver;
 355         while (*bucket < table->n_buckets) {
 356                 i = 0;
 357                 head = flex_array_get(table->buckets, *bucket);
 358                 hlist_for_each_entry_rcu(flow, n, head, hash_node[ver]) {
 359                         if (i < *last) {
 360                                 i++;
 361                                 continue;
 362                         }
 363                         *last = i + 1;
 364                         return flow;
 365                 }
 366                 (*bucket)++;
 367                 *last = 0;
 368         }
 369
 370         return NULL;
 371 }
 372
 373 static void __flow_tbl_insert(struct flow_table *table, struct sw_flow *flow)
 374 {
 375         struct hlist_head *head;
 376         head = find_bucket(table, flow->hash);
 377         hlist_add_head_rcu(&flow->hash_node[table->node_ver], head);
 378         table->count++;
 379 }
 380
 381 static void flow_table_copy_flows(struct flow_table *old, struct flow_table *new)
 382 {
 383         int old_ver;
 384         int i;
 385
 386         old_ver = old->node_ver;
 387         new->node_ver = !old_ver;
 388
 389         /* Insert in new table. */
 390         for (i = 0; i < old->n_buckets; i++) {
 391                 struct sw_flow *flow;
 392                 struct hlist_head *head;
 393                 struct hlist_node *n;
 394
 395                 head = flex_array_get(old->buckets, i);
 396
 397                 hlist_for_each_entry(flow, n, head, hash_node[old_ver])
 398                         __flow_tbl_insert(new, flow);
 399         }
 400         old->keep_flows = true;
 401 }
 402
 403 static struct flow_table *__flow_tbl_rehash(struct flow_table *table, int n_buckets)
 404 {
 405         struct flow_table *new_table;
 406
 407         new_table = ovs_flow_tbl_alloc(n_buckets);
 408         if (!new_table)
 409                 return ERR_PTR(-ENOMEM);
 410
 411         flow_table_copy_flows(table, new_table);
 412
 413         return new_table;
 414 }
 415
 416 struct flow_table *ovs_flow_tbl_rehash(struct flow_table *table)
 417 {
 418         return __flow_tbl_rehash(table, table->n_buckets);
 419 }
 420
 421 struct flow_table *ovs_flow_tbl_expand(struct flow_table *table)
 422 {
 423         return __flow_tbl_rehash(table, table->n_buckets * 2);
 424 }
 425
 426 void ovs_flow_free(struct sw_flow *flow)
 427 {
 428         if (unlikely(!flow))
 429                 return;
 430
 431         kfree((struct sf_flow_acts __force *)flow->sf_acts);
 432         kmem_cache_free(flow_cache, flow);
 433 }
 434
 435 /* RCU callback used by ovs_flow_deferred_free. */
 436 static void rcu_free_flow_callback(struct rcu_head *rcu)
 437 {
 438         struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
 439
 440         ovs_flow_free(flow);
 441 }
 442
 443 /* Schedules 'flow' to be freed after the next RCU grace period.
 444  * The caller must hold rcu_read_lock for this to be sensible. */
 445 void ovs_flow_deferred_free(struct sw_flow *flow)
 446 {
 447         call_rcu(&flow->rcu, rcu_free_flow_callback);
 448 }
 449
 450 /* RCU callback used by ovs_flow_deferred_free_acts. */
 451 static void rcu_free_acts_callback(struct rcu_head *rcu)
 452 {
 453         struct sw_flow_actions *sf_acts = container_of(rcu,
 454                         struct sw_flow_actions, rcu);
 455         ovs_flow_actions_free(sf_acts);
 456 }
 457
 458 /* Schedules 'sf_acts' to be freed after the next RCU grace period.
 459  * The caller must hold rcu_read_lock for this to be sensible. */
 460 void ovs_flow_deferred_free_acts(struct sw_flow_actions *sf_acts)
 461 {
 462         call_rcu(&sf_acts->rcu, rcu_free_acts_callback);
 463 }
 464
 465 static int parse_vlan(struct sk_buff *skb, struct sw_flow_key *key)
 466 {
 467         struct qtag_prefix {
 468                 __be16 eth_type; /* ETH_P_8021Q */
 469                 __be16 tci;
 470         };
 471         struct qtag_prefix *qp;
 472
 473         if (unlikely(skb->len < sizeof(struct qtag_prefix) + sizeof(__be16)))
 474                 return 0;
 475
 476         if (unlikely(!pskb_may_pull(skb, sizeof(struct qtag_prefix) +
 477                                          sizeof(__be16))))
 478                 return -ENOMEM;
 479
 480         qp = (struct qtag_prefix *) skb->data;
 481         key->eth.tci = qp->tci | htons(VLAN_TAG_PRESENT);
 482         __skb_pull(skb, sizeof(struct qtag_prefix));
 483
 484         return 0;
 485 }
 486
 487 static __be16 parse_ethertype(struct sk_buff *skb)
 488 {
 489         struct llc_snap_hdr {
 490                 u8  dsap;  /* Always 0xAA */
 491                 u8  ssap;  /* Always 0xAA */
 492                 u8  ctrl;
 493                 u8  oui[3];
 494                 __be16 ethertype;
 495         };
 496         struct llc_snap_hdr *llc;
 497         __be16 proto;
 498
 499         proto = *(__be16 *) skb->data;
 500         __skb_pull(skb, sizeof(__be16));
 501
 502         if (ntohs(proto) >= 1536)
 503                 return proto;
 504
 505         if (skb->len < sizeof(struct llc_snap_hdr))
 506                 return htons(ETH_P_802_2);
 507
 508         if (unlikely(!pskb_may_pull(skb, sizeof(struct llc_snap_hdr))))
 509                 return htons(0);
 510
 511         llc = (struct llc_snap_hdr *) skb->data;
 512         if (llc->dsap != LLC_SAP_SNAP ||
 513             llc->ssap != LLC_SAP_SNAP ||
 514             (llc->oui[0] | llc->oui[1] | llc->oui[2]) != 0)
 515                 return htons(ETH_P_802_2);
 516
 517         __skb_pull(skb, sizeof(struct llc_snap_hdr));
 518
 519         if (ntohs(llc->ethertype) >= 1536)
 520                 return llc->ethertype;
 521
 522         return htons(ETH_P_802_2);
 523 }
 524
 525 static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key,
 526                         int *key_lenp, int nh_len)
 527 {
 528         struct icmp6hdr *icmp = icmp6_hdr(skb);
 529         int error = 0;
 530         int key_len;
 531
 532         /* The ICMPv6 type and code fields use the 16-bit transport port
 533          * fields, so we need to store them in 16-bit network byte order.
 534          */
 535         key->ipv6.tp.src = htons(icmp->icmp6_type);
 536         key->ipv6.tp.dst = htons(icmp->icmp6_code);
 537         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 538
 539         if (icmp->icmp6_code == 0 &&
 540             (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
 541              icmp->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 542                 int icmp_len = skb->len - skb_transport_offset(skb);
 543                 struct nd_msg *nd;
 544                 int offset;
 545
 546                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 547
 548                 /* In order to process neighbor discovery options, we need the
 549                  * entire packet.
 550                  */
 551                 if (unlikely(icmp_len < sizeof(*nd)))
 552                         goto out;
 553                 if (unlikely(skb_linearize(skb))) {
 554                         error = -ENOMEM;
 555                         goto out;
 556                 }
 557
 558                 nd = (struct nd_msg *)skb_transport_header(skb);
 559                 key->ipv6.nd.target = nd->target;
 560                 key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 561
 562                 icmp_len -= sizeof(*nd);
 563                 offset = 0;
 564                 while (icmp_len >= 8) {
 565                         struct nd_opt_hdr *nd_opt =
 566                                  (struct nd_opt_hdr *)(nd->opt + offset);
 567                         int opt_len = nd_opt->nd_opt_len * 8;
 568
 569                         if (unlikely(!opt_len || opt_len > icmp_len))
 570                                 goto invalid;
 571
 572                         /* Store the link layer address if the appropriate
 573                          * option is provided.  It is considered an error if
 574                          * the same link layer option is specified twice.
 575                          */
 576                         if (nd_opt->nd_opt_type == ND_OPT_SOURCE_LL_ADDR
 577                             && opt_len == 8) {
 578                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll)))
 579                                         goto invalid;
 580                                 memcpy(key->ipv6.nd.sll,
 581                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 582                         } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR
 583                                    && opt_len == 8) {
 584                                 if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll)))
 585                                         goto invalid;
 586                                 memcpy(key->ipv6.nd.tll,
 587                                     &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN);
 588                         }
 589
 590                         icmp_len -= opt_len;
 591                         offset += opt_len;
 592                 }
 593         }
 594
 595         goto out;
 596
 597 invalid:
 598         memset(&key->ipv6.nd.target, 0, sizeof(key->ipv6.nd.target));
 599         memset(key->ipv6.nd.sll, 0, sizeof(key->ipv6.nd.sll));
 600         memset(key->ipv6.nd.tll, 0, sizeof(key->ipv6.nd.tll));
 601
 602 out:
 603         *key_lenp = key_len;
 604         return error;
 605 }
 606
 607 /**
 608  * ovs_flow_extract - extracts a flow key from an Ethernet frame.
 609  * @skb: sk_buff that contains the frame, with skb->data pointing to the
 610  * Ethernet header
 611  * @in_port: port number on which @skb was received.
 612  * @key: output flow key
 613  * @key_lenp: length of output flow key
 614  *
 615  * The caller must ensure that skb->len >= ETH_HLEN.
 616  *
 617  * Returns 0 if successful, otherwise a negative errno value.
 618  *
 619  * Initializes @skb header pointers as follows:
 620  *
 621  *    - skb->mac_header: the Ethernet header.
 622  *
 623  *    - skb->network_header: just past the Ethernet header, or just past the
 624  *      VLAN header, to the first byte of the Ethernet payload.
 625  *
 626  *    - skb->transport_header: If key->dl_type is ETH_P_IP or ETH_P_IPV6
 627  *      on output, then just past the IP header, if one is present and
 628  *      of a correct length, otherwise the same as skb->network_header.
 629  *      For other key->dl_type values it is left untouched.
 630  */
 631 int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key,
 632                  int *key_lenp)
 633 {
 634         int error = 0;
 635         int key_len = SW_FLOW_KEY_OFFSET(eth);
 636         struct ethhdr *eth;
 637
 638         memset(key, 0, sizeof(*key));
 639
 640         key->phy.priority = skb->priority;
 641         if (OVS_CB(skb)->tun_key)
 642                 memcpy(&key->tun_key, OVS_CB(skb)->tun_key, sizeof(key->tun_key));
 643         key->phy.in_port = in_port;
 644         key->phy.skb_mark = skb_get_mark(skb);
 645
 646         skb_reset_mac_header(skb);
 647
 648         /* Link layer.  We are guaranteed to have at least the 14 byte Ethernet
 649          * header in the linear data area.
 650          */
 651         eth = eth_hdr(skb);
 652         memcpy(key->eth.src, eth->h_source, ETH_ALEN);
 653         memcpy(key->eth.dst, eth->h_dest, ETH_ALEN);
 654
 655         __skb_pull(skb, 2 * ETH_ALEN);
 656
 657         if (vlan_tx_tag_present(skb))
 658                 key->eth.tci = htons(vlan_get_tci(skb));
 659         else if (eth->h_proto == htons(ETH_P_8021Q))
 660                 if (unlikely(parse_vlan(skb, key)))
 661                         return -ENOMEM;
 662
 663         key->eth.type = parse_ethertype(skb);
 664         if (unlikely(key->eth.type == htons(0)))
 665                 return -ENOMEM;
 666
 667         skb_reset_network_header(skb);
 668         __skb_push(skb, skb->data - skb_mac_header(skb));
 669
 670         /* Network layer. */
 671         if (key->eth.type == htons(ETH_P_IP)) {
 672                 struct iphdr *nh;
 673                 __be16 offset;
 674
 675                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
 676
 677                 error = check_iphdr(skb);
 678                 if (unlikely(error)) {
 679                         if (error == -EINVAL) {
 680                                 skb->transport_header = skb->network_header;
 681                                 error = 0;
 682                         }
 683                         goto out;
 684                 }
 685
 686                 nh = ip_hdr(skb);
 687                 key->ipv4.addr.src = nh->saddr;
 688                 key->ipv4.addr.dst = nh->daddr;
 689
 690                 key->ip.proto = nh->protocol;
 691                 key->ip.tos = nh->tos;
 692                 key->ip.ttl = nh->ttl;
 693
 694                 offset = nh->frag_off & htons(IP_OFFSET);
 695                 if (offset) {
 696                         key->ip.frag = OVS_FRAG_TYPE_LATER;
 697                         goto out;
 698                 }
 699                 if (nh->frag_off & htons(IP_MF) ||
 700                          skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 701                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 702
 703                 /* Transport layer. */
 704                 if (key->ip.proto == IPPROTO_TCP) {
 705                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 706                         if (tcphdr_ok(skb)) {
 707                                 struct tcphdr *tcp = tcp_hdr(skb);
 708                                 key->ipv4.tp.src = tcp->source;
 709                                 key->ipv4.tp.dst = tcp->dest;
 710                         }
 711                 } else if (key->ip.proto == IPPROTO_UDP) {
 712                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 713                         if (udphdr_ok(skb)) {
 714                                 struct udphdr *udp = udp_hdr(skb);
 715                                 key->ipv4.tp.src = udp->source;
 716                                 key->ipv4.tp.dst = udp->dest;
 717                         }
 718                 } else if (key->ip.proto == IPPROTO_ICMP) {
 719                         key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 720                         if (icmphdr_ok(skb)) {
 721                                 struct icmphdr *icmp = icmp_hdr(skb);
 722                                 /* The ICMP type and code fields use the 16-bit
 723                                  * transport port fields, so we need to store
 724                                  * them in 16-bit network byte order. */
 725                                 key->ipv4.tp.src = htons(icmp->type);
 726                                 key->ipv4.tp.dst = htons(icmp->code);
 727                         }
 728                 }
 729
 730         } else if ((key->eth.type == htons(ETH_P_ARP) ||
 731                    key->eth.type == htons(ETH_P_RARP)) && arphdr_ok(skb)) {
 732                 struct arp_eth_header *arp;
 733
 734                 arp = (struct arp_eth_header *)skb_network_header(skb);
 735
 736                 if (arp->ar_hrd == htons(ARPHRD_ETHER)
 737                                 && arp->ar_pro == htons(ETH_P_IP)
 738                                 && arp->ar_hln == ETH_ALEN
 739                                 && arp->ar_pln == 4) {
 740
 741                         /* We only match on the lower 8 bits of the opcode. */
 742                         if (ntohs(arp->ar_op) <= 0xff)
 743                                 key->ip.proto = ntohs(arp->ar_op);
 744                         memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src));
 745                         memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst));
 746                         memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN);
 747                         memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN);
 748                         key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
 749                 }
 750         } else if (key->eth.type == htons(ETH_P_IPV6)) {
 751                 int nh_len;             /* IPv6 Header + Extensions */
 752
 753                 nh_len = parse_ipv6hdr(skb, key, &key_len);
 754                 if (unlikely(nh_len < 0)) {
 755                         if (nh_len == -EINVAL)
 756                                 skb->transport_header = skb->network_header;
 757                         else
 758                                 error = nh_len;
 759                         goto out;
 760                 }
 761
 762                 if (key->ip.frag == OVS_FRAG_TYPE_LATER)
 763                         goto out;
 764                 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 765                         key->ip.frag = OVS_FRAG_TYPE_FIRST;
 766
 767                 /* Transport layer. */
 768                 if (key->ip.proto == NEXTHDR_TCP) {
 769                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 770                         if (tcphdr_ok(skb)) {
 771                                 struct tcphdr *tcp = tcp_hdr(skb);
 772                                 key->ipv6.tp.src = tcp->source;
 773                                 key->ipv6.tp.dst = tcp->dest;
 774                         }
 775                 } else if (key->ip.proto == NEXTHDR_UDP) {
 776                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 777                         if (udphdr_ok(skb)) {
 778                                 struct udphdr *udp = udp_hdr(skb);
 779                                 key->ipv6.tp.src = udp->source;
 780                                 key->ipv6.tp.dst = udp->dest;
 781                         }
 782                 } else if (key->ip.proto == NEXTHDR_ICMP) {
 783                         key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 784                         if (icmp6hdr_ok(skb)) {
 785                                 error = parse_icmpv6(skb, key, &key_len, nh_len);
 786                                 if (error < 0)
 787                                         goto out;
 788                         }
 789                 }
 790         }
 791
 792 out:
 793         *key_lenp = key_len;
 794         return error;
 795 }
 796
 797 static u32 ovs_flow_hash(const struct sw_flow_key *key, int key_start, int key_len)
 798 {
 799         return jhash2((u32 *)((u8 *)key + key_start),
 800                       DIV_ROUND_UP(key_len - key_start, sizeof(u32)), 0);
 801 }
 802
 803 static int flow_key_start(struct sw_flow_key *key)
 804 {
 805         if (key->tun_key.ipv4_dst)
 806                 return 0;
 807         else
 808                 return offsetof(struct sw_flow_key, phy);
 809 }
 810
 811 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *table,
 812                                 struct sw_flow_key *key, int key_len)
 813 {
 814         struct sw_flow *flow;
 815         struct hlist_node *n;
 816         struct hlist_head *head;
 817         u8 *_key;
 818         int key_start;
 819         u32 hash;
 820
 821         key_start = flow_key_start(key);
 822         hash = ovs_flow_hash(key, key_start, key_len);
 823
 824         _key = (u8 *) key + key_start;
 825         head = find_bucket(table, hash);
 826         hlist_for_each_entry_rcu(flow, n, head, hash_node[table->node_ver]) {
 827
 828                 if (flow->hash == hash &&
 829                     !memcmp((u8 *)&flow->key + key_start, _key, key_len - key_start)) {
 830                         return flow;
 831                 }
 832         }
 833         return NULL;
 834 }
 835
 836 void ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
 837                          struct sw_flow_key *key, int key_len)
 838 {
 839         flow->hash = ovs_flow_hash(key, flow_key_start(key), key_len);
 840         memcpy(&flow->key, key, sizeof(flow->key));
 841         __flow_tbl_insert(table, flow);
 842 }
 843
 844 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
 845 {
 846         hlist_del_rcu(&flow->hash_node[table->node_ver]);
 847         table->count--;
 848         BUG_ON(table->count < 0);
 849 }
 850
 851 /* The size of the argument for each %OVS_KEY_ATTR_* Netlink attribute.  */
 852 const int ovs_key_lens[OVS_KEY_ATTR_MAX + 1] = {
 853         [OVS_KEY_ATTR_ENCAP] = -1,
 854         [OVS_KEY_ATTR_PRIORITY] = sizeof(u32),
 855         [OVS_KEY_ATTR_IN_PORT] = sizeof(u32),
 856         [OVS_KEY_ATTR_SKB_MARK] = sizeof(u32),
 857         [OVS_KEY_ATTR_ETHERNET] = sizeof(struct ovs_key_ethernet),
 858         [OVS_KEY_ATTR_VLAN] = sizeof(__be16),
 859         [OVS_KEY_ATTR_ETHERTYPE] = sizeof(__be16),
 860         [OVS_KEY_ATTR_IPV4] = sizeof(struct ovs_key_ipv4),
 861         [OVS_KEY_ATTR_IPV6] = sizeof(struct ovs_key_ipv6),
 862         [OVS_KEY_ATTR_TCP] = sizeof(struct ovs_key_tcp),
 863         [OVS_KEY_ATTR_UDP] = sizeof(struct ovs_key_udp),
 864         [OVS_KEY_ATTR_ICMP] = sizeof(struct ovs_key_icmp),
 865         [OVS_KEY_ATTR_ICMPV6] = sizeof(struct ovs_key_icmpv6),
 866         [OVS_KEY_ATTR_ARP] = sizeof(struct ovs_key_arp),
 867         [OVS_KEY_ATTR_ND] = sizeof(struct ovs_key_nd),
 868         [OVS_KEY_ATTR_TUNNEL] = -1,
 869
 870         /* Not upstream. */
 871         [OVS_KEY_ATTR_TUN_ID] = sizeof(__be64),
 872 };
 873
 874 static int ipv4_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 875                                   const struct nlattr *a[], u64 *attrs)
 876 {
 877         const struct ovs_key_icmp *icmp_key;
 878         const struct ovs_key_tcp *tcp_key;
 879         const struct ovs_key_udp *udp_key;
 880
 881         switch (swkey->ip.proto) {
 882         case IPPROTO_TCP:
 883                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 884                         return -EINVAL;
 885                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 886
 887                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 888                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 889                 swkey->ipv4.tp.src = tcp_key->tcp_src;
 890                 swkey->ipv4.tp.dst = tcp_key->tcp_dst;
 891                 break;
 892
 893         case IPPROTO_UDP:
 894                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 895                         return -EINVAL;
 896                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 897
 898                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 899                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 900                 swkey->ipv4.tp.src = udp_key->udp_src;
 901                 swkey->ipv4.tp.dst = udp_key->udp_dst;
 902                 break;
 903
 904         case IPPROTO_ICMP:
 905                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMP)))
 906                         return -EINVAL;
 907                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMP);
 908
 909                 *key_len = SW_FLOW_KEY_OFFSET(ipv4.tp);
 910                 icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]);
 911                 swkey->ipv4.tp.src = htons(icmp_key->icmp_type);
 912                 swkey->ipv4.tp.dst = htons(icmp_key->icmp_code);
 913                 break;
 914         }
 915
 916         return 0;
 917 }
 918
 919 static int ipv6_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_len,
 920                                   const struct nlattr *a[], u64 *attrs)
 921 {
 922         const struct ovs_key_icmpv6 *icmpv6_key;
 923         const struct ovs_key_tcp *tcp_key;
 924         const struct ovs_key_udp *udp_key;
 925
 926         switch (swkey->ip.proto) {
 927         case IPPROTO_TCP:
 928                 if (!(*attrs & (1 << OVS_KEY_ATTR_TCP)))
 929                         return -EINVAL;
 930                 *attrs &= ~(1 << OVS_KEY_ATTR_TCP);
 931
 932                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 933                 tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]);
 934                 swkey->ipv6.tp.src = tcp_key->tcp_src;
 935                 swkey->ipv6.tp.dst = tcp_key->tcp_dst;
 936                 break;
 937
 938         case IPPROTO_UDP:
 939                 if (!(*attrs & (1 << OVS_KEY_ATTR_UDP)))
 940                         return -EINVAL;
 941                 *attrs &= ~(1 << OVS_KEY_ATTR_UDP);
 942
 943                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 944                 udp_key = nla_data(a[OVS_KEY_ATTR_UDP]);
 945                 swkey->ipv6.tp.src = udp_key->udp_src;
 946                 swkey->ipv6.tp.dst = udp_key->udp_dst;
 947                 break;
 948
 949         case IPPROTO_ICMPV6:
 950                 if (!(*attrs & (1 << OVS_KEY_ATTR_ICMPV6)))
 951                         return -EINVAL;
 952                 *attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6);
 953
 954                 *key_len = SW_FLOW_KEY_OFFSET(ipv6.tp);
 955                 icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]);
 956                 swkey->ipv6.tp.src = htons(icmpv6_key->icmpv6_type);
 957                 swkey->ipv6.tp.dst = htons(icmpv6_key->icmpv6_code);
 958
 959                 if (swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) ||
 960                     swkey->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) {
 961                         const struct ovs_key_nd *nd_key;
 962
 963                         if (!(*attrs & (1 << OVS_KEY_ATTR_ND)))
 964                                 return -EINVAL;
 965                         *attrs &= ~(1 << OVS_KEY_ATTR_ND);
 966
 967                         *key_len = SW_FLOW_KEY_OFFSET(ipv6.nd);
 968                         nd_key = nla_data(a[OVS_KEY_ATTR_ND]);
 969                         memcpy(&swkey->ipv6.nd.target, nd_key->nd_target,
 970                                sizeof(swkey->ipv6.nd.target));
 971                         memcpy(swkey->ipv6.nd.sll, nd_key->nd_sll, ETH_ALEN);
 972                         memcpy(swkey->ipv6.nd.tll, nd_key->nd_tll, ETH_ALEN);
 973                 }
 974                 break;
 975         }
 976
 977         return 0;
 978 }
 979
 980 static int parse_flow_nlattrs(const struct nlattr *attr,
 981                               const struct nlattr *a[], u64 *attrsp)
 982 {
 983         const struct nlattr *nla;
 984         u64 attrs;
 985         int rem;
 986
 987         attrs = 0;
 988         nla_for_each_nested(nla, attr, rem) {
 989                 u16 type = nla_type(nla);
 990                 int expected_len;
 991
 992                 if (type > OVS_KEY_ATTR_MAX || attrs & (1ULL << type))
 993                         return -EINVAL;
 994
 995                 expected_len = ovs_key_lens[type];
 996                 if (nla_len(nla) != expected_len && expected_len != -1)
 997                         return -EINVAL;
 998
 999                 attrs |= 1ULL << type;
1000                 a[type] = nla;
1001         }
1002         if (rem)
1003                 return -EINVAL;
1004
1005         *attrsp = attrs;
1006         return 0;
1007 }
1008
1009 int ipv4_tun_from_nlattr(const struct nlattr *attr,
1010                          struct ovs_key_ipv4_tunnel *tun_key)
1011 {
1012         struct nlattr *a;
1013         int rem;
1014         bool ttl = false;
1015
1016         memset(tun_key, 0, sizeof(*tun_key));
1017
1018         nla_for_each_nested(a, attr, rem) {
1019                 int type = nla_type(a);
1020                 static const u32 ovs_tunnel_key_lens[OVS_TUNNEL_KEY_ATTR_MAX + 1] = {
1021                         [OVS_TUNNEL_KEY_ATTR_ID] = sizeof(u64),
1022                         [OVS_TUNNEL_KEY_ATTR_IPV4_SRC] = sizeof(u32),
1023                         [OVS_TUNNEL_KEY_ATTR_IPV4_DST] = sizeof(u32),
1024                         [OVS_TUNNEL_KEY_ATTR_TOS] = 1,
1025                         [OVS_TUNNEL_KEY_ATTR_TTL] = 1,
1026                         [OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT] = 0,
1027                         [OVS_TUNNEL_KEY_ATTR_CSUM] = 0,
1028                 };
1029
1030                 if (type > OVS_TUNNEL_KEY_ATTR_MAX ||
1031                         ovs_tunnel_key_lens[type] != nla_len(a))
1032                         return -EINVAL;
1033
1034                 switch (type) {
1035                 case OVS_TUNNEL_KEY_ATTR_ID:
1036                         tun_key->tun_id = nla_get_be64(a);
1037                         tun_key->tun_flags |= OVS_TNL_F_KEY;
1038                         break;
1039                 case OVS_TUNNEL_KEY_ATTR_IPV4_SRC:
1040                         tun_key->ipv4_src = nla_get_be32(a);
1041                         break;
1042                 case OVS_TUNNEL_KEY_ATTR_IPV4_DST:
1043                         tun_key->ipv4_dst = nla_get_be32(a);
1044                         break;
1045                 case OVS_TUNNEL_KEY_ATTR_TOS:
1046                         tun_key->ipv4_tos = nla_get_u8(a);
1047                         break;
1048                 case OVS_TUNNEL_KEY_ATTR_TTL:
1049                         tun_key->ipv4_ttl = nla_get_u8(a);
1050                         ttl = true;
1051                         break;
1052                 case OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT:
1053                         tun_key->tun_flags |= OVS_TNL_F_DONT_FRAGMENT;
1054                         break;
1055                 case OVS_TUNNEL_KEY_ATTR_CSUM:
1056                         tun_key->tun_flags |= OVS_TNL_F_CSUM;
1057                         break;
1058                 default:
1059                         return -EINVAL;
1060
1061                 }
1062         }
1063         if (rem > 0)
1064                 return -EINVAL;
1065
1066         if (!tun_key->ipv4_dst)
1067                 return -EINVAL;
1068
1069         if (!ttl)
1070                 return -EINVAL;
1071
1072         return 0;
1073 }
1074
1075 int ipv4_tun_to_nlattr(struct sk_buff *skb,
1076                         const struct ovs_key_ipv4_tunnel *tun_key)
1077 {
1078         struct nlattr *nla;
1079
1080         nla = nla_nest_start(skb, OVS_KEY_ATTR_TUNNEL);
1081         if (!nla)
1082                 return -EMSGSIZE;
1083
1084         if (tun_key->tun_flags & OVS_TNL_F_KEY &&
1085             nla_put_be64(skb, OVS_TUNNEL_KEY_ATTR_ID, tun_key->tun_id))
1086                 return -EMSGSIZE;
1087         if (tun_key->ipv4_src &&
1088             nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_SRC, tun_key->ipv4_src))
1089                 return -EMSGSIZE;
1090         if (nla_put_be32(skb, OVS_TUNNEL_KEY_ATTR_IPV4_DST, tun_key->ipv4_dst))
1091                 return -EMSGSIZE;
1092         if (tun_key->ipv4_tos &&
1093             nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TOS, tun_key->ipv4_tos))
1094                 return -EMSGSIZE;
1095         if (nla_put_u8(skb, OVS_TUNNEL_KEY_ATTR_TTL, tun_key->ipv4_ttl))
1096                 return -EMSGSIZE;
1097         if ((tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT) &&
1098                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
1099                 return -EMSGSIZE;
1100         if ((tun_key->tun_flags & OVS_TNL_F_CSUM) &&
1101                 nla_put_flag(skb, OVS_TUNNEL_KEY_ATTR_CSUM))
1102                 return -EMSGSIZE;
1103
1104         nla_nest_end(skb, nla);
1105         return 0;
1106 }
1107
1108 /**
1109  * ovs_flow_from_nlattrs - parses Netlink attributes into a flow key.
1110  * @swkey: receives the extracted flow key.
1111  * @key_lenp: number of bytes used in @swkey.
1112  * @attr: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1113  * sequence.
1114  */
1115 int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int *key_lenp,
1116                       const struct nlattr *attr)
1117 {
1118         const struct nlattr *a[OVS_KEY_ATTR_MAX + 1];
1119         const struct ovs_key_ethernet *eth_key;
1120         int key_len;
1121         u64 attrs;
1122         int err;
1123
1124         memset(swkey, 0, sizeof(struct sw_flow_key));
1125         key_len = SW_FLOW_KEY_OFFSET(eth);
1126
1127         err = parse_flow_nlattrs(attr, a, &attrs);
1128         if (err)
1129                 return err;
1130
1131         /* Metadata attributes. */
1132         if (attrs & (1 << OVS_KEY_ATTR_PRIORITY)) {
1133                 swkey->phy.priority = nla_get_u32(a[OVS_KEY_ATTR_PRIORITY]);
1134                 attrs &= ~(1 << OVS_KEY_ATTR_PRIORITY);
1135         }
1136         if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
1137                 u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
1138                 if (in_port >= DP_MAX_PORTS)
1139                         return -EINVAL;
1140                 swkey->phy.in_port = in_port;
1141                 attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
1142         } else {
1143                 swkey->phy.in_port = DP_MAX_PORTS;
1144         }
1145         if (attrs & (1 << OVS_KEY_ATTR_SKB_MARK)) {
1146                 uint32_t mark = nla_get_u32(a[OVS_KEY_ATTR_SKB_MARK]);
1147 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1148                 if (mark != 0)
1149                         return -EINVAL;
1150 #endif
1151                 swkey->phy.skb_mark = mark;
1152                 attrs &= ~(1 << OVS_KEY_ATTR_SKB_MARK);
1153         }
1154
1155         if (attrs & (1ULL << OVS_KEY_ATTR_TUN_ID) &&
1156             attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1157                 __be64 tun_id;
1158
1159                 err = ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1160                 if (err)
1161                         return err;
1162
1163                 if (!(swkey->tun_key.tun_flags & OVS_TNL_F_KEY))
1164                         return -EINVAL;
1165
1166                 tun_id = nla_get_be64(a[OVS_KEY_ATTR_TUN_ID]);
1167                 if (tun_id != swkey->tun_key.tun_id)
1168                         return -EINVAL;
1169
1170                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUN_ID);
1171                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1172         } else if (attrs & (1ULL << OVS_KEY_ATTR_TUNNEL)) {
1173
1174                 err = ipv4_tun_from_nlattr(a[OVS_KEY_ATTR_TUNNEL], &swkey->tun_key);
1175                 if (err)
1176                         return err;
1177
1178                 attrs &= ~(1ULL << OVS_KEY_ATTR_TUNNEL);
1179         }
1180
1181         /* Data attributes. */
1182         if (!(attrs & (1 << OVS_KEY_ATTR_ETHERNET)))
1183                 return -EINVAL;
1184         attrs &= ~(1 << OVS_KEY_ATTR_ETHERNET);
1185
1186         eth_key = nla_data(a[OVS_KEY_ATTR_ETHERNET]);
1187         memcpy(swkey->eth.src, eth_key->eth_src, ETH_ALEN);
1188         memcpy(swkey->eth.dst, eth_key->eth_dst, ETH_ALEN);
1189
1190         if (attrs & (1u << OVS_KEY_ATTR_ETHERTYPE) &&
1191             nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]) == htons(ETH_P_8021Q)) {
1192                 const struct nlattr *encap;
1193                 __be16 tci;
1194
1195                 if (attrs != ((1 << OVS_KEY_ATTR_VLAN) |
1196                               (1 << OVS_KEY_ATTR_ETHERTYPE) |
1197                               (1 << OVS_KEY_ATTR_ENCAP)))
1198                         return -EINVAL;
1199
1200                 encap = a[OVS_KEY_ATTR_ENCAP];
1201                 tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
1202                 if (tci & htons(VLAN_TAG_PRESENT)) {
1203                         swkey->eth.tci = tci;
1204
1205                         err = parse_flow_nlattrs(encap, a, &attrs);
1206                         if (err)
1207                                 return err;
1208                 } else if (!tci) {
1209                         /* Corner case for truncated 802.1Q header. */
1210                         if (nla_len(encap))
1211                                 return -EINVAL;
1212
1213                         swkey->eth.type = htons(ETH_P_8021Q);
1214                         *key_lenp = key_len;
1215                         return 0;
1216                 } else {
1217                         return -EINVAL;
1218                 }
1219         }
1220
1221         if (attrs & (1 << OVS_KEY_ATTR_ETHERTYPE)) {
1222                 swkey->eth.type = nla_get_be16(a[OVS_KEY_ATTR_ETHERTYPE]);
1223                 if (ntohs(swkey->eth.type) < 1536)
1224                         return -EINVAL;
1225                 attrs &= ~(1 << OVS_KEY_ATTR_ETHERTYPE);
1226         } else {
1227                 swkey->eth.type = htons(ETH_P_802_2);
1228         }
1229
1230         if (swkey->eth.type == htons(ETH_P_IP)) {
1231                 const struct ovs_key_ipv4 *ipv4_key;
1232
1233                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV4)))
1234                         return -EINVAL;
1235                 attrs &= ~(1 << OVS_KEY_ATTR_IPV4);
1236
1237                 key_len = SW_FLOW_KEY_OFFSET(ipv4.addr);
1238                 ipv4_key = nla_data(a[OVS_KEY_ATTR_IPV4]);
1239                 if (ipv4_key->ipv4_frag > OVS_FRAG_TYPE_MAX)
1240                         return -EINVAL;
1241                 swkey->ip.proto = ipv4_key->ipv4_proto;
1242                 swkey->ip.tos = ipv4_key->ipv4_tos;
1243                 swkey->ip.ttl = ipv4_key->ipv4_ttl;
1244                 swkey->ip.frag = ipv4_key->ipv4_frag;
1245                 swkey->ipv4.addr.src = ipv4_key->ipv4_src;
1246                 swkey->ipv4.addr.dst = ipv4_key->ipv4_dst;
1247
1248                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1249                         err = ipv4_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1250                         if (err)
1251                                 return err;
1252                 }
1253         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1254                 const struct ovs_key_ipv6 *ipv6_key;
1255
1256                 if (!(attrs & (1 << OVS_KEY_ATTR_IPV6)))
1257                         return -EINVAL;
1258                 attrs &= ~(1 << OVS_KEY_ATTR_IPV6);
1259
1260                 key_len = SW_FLOW_KEY_OFFSET(ipv6.label);
1261                 ipv6_key = nla_data(a[OVS_KEY_ATTR_IPV6]);
1262                 if (ipv6_key->ipv6_frag > OVS_FRAG_TYPE_MAX)
1263                         return -EINVAL;
1264                 swkey->ipv6.label = ipv6_key->ipv6_label;
1265                 swkey->ip.proto = ipv6_key->ipv6_proto;
1266                 swkey->ip.tos = ipv6_key->ipv6_tclass;
1267                 swkey->ip.ttl = ipv6_key->ipv6_hlimit;
1268                 swkey->ip.frag = ipv6_key->ipv6_frag;
1269                 memcpy(&swkey->ipv6.addr.src, ipv6_key->ipv6_src,
1270                        sizeof(swkey->ipv6.addr.src));
1271                 memcpy(&swkey->ipv6.addr.dst, ipv6_key->ipv6_dst,
1272                        sizeof(swkey->ipv6.addr.dst));
1273
1274                 if (swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1275                         err = ipv6_flow_from_nlattrs(swkey, &key_len, a, &attrs);
1276                         if (err)
1277                                 return err;
1278                 }
1279         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1280                    swkey->eth.type == htons(ETH_P_RARP)) {
1281                 const struct ovs_key_arp *arp_key;
1282
1283                 if (!(attrs & (1 << OVS_KEY_ATTR_ARP)))
1284                         return -EINVAL;
1285                 attrs &= ~(1 << OVS_KEY_ATTR_ARP);
1286
1287                 key_len = SW_FLOW_KEY_OFFSET(ipv4.arp);
1288                 arp_key = nla_data(a[OVS_KEY_ATTR_ARP]);
1289                 swkey->ipv4.addr.src = arp_key->arp_sip;
1290                 swkey->ipv4.addr.dst = arp_key->arp_tip;
1291                 if (arp_key->arp_op & htons(0xff00))
1292                         return -EINVAL;
1293                 swkey->ip.proto = ntohs(arp_key->arp_op);
1294                 memcpy(swkey->ipv4.arp.sha, arp_key->arp_sha, ETH_ALEN);
1295                 memcpy(swkey->ipv4.arp.tha, arp_key->arp_tha, ETH_ALEN);
1296         }
1297
1298         if (attrs)
1299                 return -EINVAL;
1300         *key_lenp = key_len;
1301
1302         return 0;
1303 }
1304
1305 /**
1306  * ovs_flow_metadata_from_nlattrs - parses Netlink attributes into a flow key.
1307  * @in_port: receives the extracted input port.
1308  * @tun_id: receives the extracted tunnel ID.
1309  * @key: Netlink attribute holding nested %OVS_KEY_ATTR_* Netlink attribute
1310  * sequence.
1311  *
1312  * This parses a series of Netlink attributes that form a flow key, which must
1313  * take the same form accepted by flow_from_nlattrs(), but only enough of it to
1314  * get the metadata, that is, the parts of the flow key that cannot be
1315  * extracted from the packet itself.
1316  */
1317
1318 int ovs_flow_metadata_from_nlattrs(struct sw_flow *flow, int key_len, const struct nlattr *attr)
1319 {
1320         struct ovs_key_ipv4_tunnel *tun_key = &flow->key.tun_key;
1321         const struct nlattr *nla;
1322         int rem;
1323         __be64 tun_id = 0;
1324
1325         flow->key.phy.in_port = DP_MAX_PORTS;
1326         flow->key.phy.priority = 0;
1327         flow->key.phy.skb_mark = 0;
1328         memset(tun_key, 0, sizeof(flow->key.tun_key));
1329
1330         nla_for_each_nested(nla, attr, rem) {
1331                 int type = nla_type(nla);
1332
1333                 if (type <= OVS_KEY_ATTR_MAX && ovs_key_lens[type] > 0) {
1334                         int err;
1335
1336                         if (nla_len(nla) != ovs_key_lens[type])
1337                                 return -EINVAL;
1338
1339                         switch (type) {
1340                         case OVS_KEY_ATTR_PRIORITY:
1341                                 flow->key.phy.priority = nla_get_u32(nla);
1342                                 break;
1343
1344                         case OVS_KEY_ATTR_TUN_ID:
1345                                 tun_id = nla_get_be64(nla);
1346
1347                                 if (tun_key->ipv4_dst) {
1348                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1349                                                 return -EINVAL;
1350                                         if (tun_key->tun_id != tun_id)
1351                                                 return -EINVAL;
1352                                         break;
1353                                 }
1354                                 tun_key->tun_id = tun_id;
1355                                 tun_key->tun_flags |= OVS_TNL_F_KEY;
1356
1357                                 break;
1358
1359                         case OVS_KEY_ATTR_TUNNEL:
1360                                 if (tun_key->tun_flags & OVS_TNL_F_KEY) {
1361                                         tun_id = tun_key->tun_id;
1362                                         err = ipv4_tun_from_nlattr(nla, tun_key);
1363                                         if (err)
1364                                                 return err;
1365
1366                                         if (!(tun_key->tun_flags & OVS_TNL_F_KEY))
1367                                                 return -EINVAL;
1368
1369                                         if (tun_key->tun_id != tun_id)
1370                                                 return -EINVAL;
1371                                 } else {
1372                                         err = ipv4_tun_from_nlattr(nla, tun_key);
1373                                         if (err)
1374                                                 return err;
1375                                 }
1376                                 break;
1377
1378                         case OVS_KEY_ATTR_IN_PORT:
1379                                 if (nla_get_u32(nla) >= DP_MAX_PORTS)
1380                                         return -EINVAL;
1381                                 flow->key.phy.in_port = nla_get_u32(nla);
1382                                 break;
1383
1384                         case OVS_KEY_ATTR_SKB_MARK:
1385 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
1386                                 if (nla_get_u32(nla) != 0)
1387                                         return -EINVAL;
1388 #endif
1389                                 flow->key.phy.skb_mark = nla_get_u32(nla);
1390                                 break;
1391                         }
1392                 }
1393         }
1394         if (rem)
1395                 return -EINVAL;
1396
1397         flow->hash = ovs_flow_hash(&flow->key,
1398                                    flow_key_start(&flow->key), key_len);
1399
1400         return 0;
1401 }
1402
1403 int ovs_flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb)
1404 {
1405         struct ovs_key_ethernet *eth_key;
1406         struct nlattr *nla, *encap;
1407
1408         if (swkey->phy.priority &&
1409             nla_put_u32(skb, OVS_KEY_ATTR_PRIORITY, swkey->phy.priority))
1410                 goto nla_put_failure;
1411
1412         if (swkey->tun_key.ipv4_dst &&
1413             ipv4_tun_to_nlattr(skb, &swkey->tun_key))
1414                 goto nla_put_failure;
1415
1416         if ((swkey->tun_key.tun_flags & OVS_TNL_F_KEY) &&
1417             nla_put_be64(skb, OVS_KEY_ATTR_TUN_ID, swkey->tun_key.tun_id))
1418                 goto nla_put_failure;
1419
1420         if (swkey->phy.in_port != DP_MAX_PORTS &&
1421             nla_put_u32(skb, OVS_KEY_ATTR_IN_PORT, swkey->phy.in_port))
1422                 goto nla_put_failure;
1423
1424         if (swkey->phy.skb_mark &&
1425             nla_put_u32(skb, OVS_KEY_ATTR_SKB_MARK, swkey->phy.skb_mark))
1426                 goto nla_put_failure;
1427
1428         nla = nla_reserve(skb, OVS_KEY_ATTR_ETHERNET, sizeof(*eth_key));
1429         if (!nla)
1430                 goto nla_put_failure;
1431         eth_key = nla_data(nla);
1432         memcpy(eth_key->eth_src, swkey->eth.src, ETH_ALEN);
1433         memcpy(eth_key->eth_dst, swkey->eth.dst, ETH_ALEN);
1434
1435         if (swkey->eth.tci || swkey->eth.type == htons(ETH_P_8021Q)) {
1436                 if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, htons(ETH_P_8021Q)) ||
1437                     nla_put_be16(skb, OVS_KEY_ATTR_VLAN, swkey->eth.tci))
1438                         goto nla_put_failure;
1439                 encap = nla_nest_start(skb, OVS_KEY_ATTR_ENCAP);
1440                 if (!swkey->eth.tci)
1441                         goto unencap;
1442         } else {
1443                 encap = NULL;
1444         }
1445
1446         if (swkey->eth.type == htons(ETH_P_802_2))
1447                 goto unencap;
1448
1449         if (nla_put_be16(skb, OVS_KEY_ATTR_ETHERTYPE, swkey->eth.type))
1450                 goto nla_put_failure;
1451
1452         if (swkey->eth.type == htons(ETH_P_IP)) {
1453                 struct ovs_key_ipv4 *ipv4_key;
1454
1455                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV4, sizeof(*ipv4_key));
1456                 if (!nla)
1457                         goto nla_put_failure;
1458                 ipv4_key = nla_data(nla);
1459                 ipv4_key->ipv4_src = swkey->ipv4.addr.src;
1460                 ipv4_key->ipv4_dst = swkey->ipv4.addr.dst;
1461                 ipv4_key->ipv4_proto = swkey->ip.proto;
1462                 ipv4_key->ipv4_tos = swkey->ip.tos;
1463                 ipv4_key->ipv4_ttl = swkey->ip.ttl;
1464                 ipv4_key->ipv4_frag = swkey->ip.frag;
1465         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1466                 struct ovs_key_ipv6 *ipv6_key;
1467
1468                 nla = nla_reserve(skb, OVS_KEY_ATTR_IPV6, sizeof(*ipv6_key));
1469                 if (!nla)
1470                         goto nla_put_failure;
1471                 ipv6_key = nla_data(nla);
1472                 memcpy(ipv6_key->ipv6_src, &swkey->ipv6.addr.src,
1473                                 sizeof(ipv6_key->ipv6_src));
1474                 memcpy(ipv6_key->ipv6_dst, &swkey->ipv6.addr.dst,
1475                                 sizeof(ipv6_key->ipv6_dst));
1476                 ipv6_key->ipv6_label = swkey->ipv6.label;
1477                 ipv6_key->ipv6_proto = swkey->ip.proto;
1478                 ipv6_key->ipv6_tclass = swkey->ip.tos;
1479                 ipv6_key->ipv6_hlimit = swkey->ip.ttl;
1480                 ipv6_key->ipv6_frag = swkey->ip.frag;
1481         } else if (swkey->eth.type == htons(ETH_P_ARP) ||
1482                    swkey->eth.type == htons(ETH_P_RARP)) {
1483                 struct ovs_key_arp *arp_key;
1484
1485                 nla = nla_reserve(skb, OVS_KEY_ATTR_ARP, sizeof(*arp_key));
1486                 if (!nla)
1487                         goto nla_put_failure;
1488                 arp_key = nla_data(nla);
1489                 memset(arp_key, 0, sizeof(struct ovs_key_arp));
1490                 arp_key->arp_sip = swkey->ipv4.addr.src;
1491                 arp_key->arp_tip = swkey->ipv4.addr.dst;
1492                 arp_key->arp_op = htons(swkey->ip.proto);
1493                 memcpy(arp_key->arp_sha, swkey->ipv4.arp.sha, ETH_ALEN);
1494                 memcpy(arp_key->arp_tha, swkey->ipv4.arp.tha, ETH_ALEN);
1495         }
1496
1497         if ((swkey->eth.type == htons(ETH_P_IP) ||
1498              swkey->eth.type == htons(ETH_P_IPV6)) &&
1499              swkey->ip.frag != OVS_FRAG_TYPE_LATER) {
1500
1501                 if (swkey->ip.proto == IPPROTO_TCP) {
1502                         struct ovs_key_tcp *tcp_key;
1503
1504                         nla = nla_reserve(skb, OVS_KEY_ATTR_TCP, sizeof(*tcp_key));
1505                         if (!nla)
1506                                 goto nla_put_failure;
1507                         tcp_key = nla_data(nla);
1508                         if (swkey->eth.type == htons(ETH_P_IP)) {
1509                                 tcp_key->tcp_src = swkey->ipv4.tp.src;
1510                                 tcp_key->tcp_dst = swkey->ipv4.tp.dst;
1511                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1512                                 tcp_key->tcp_src = swkey->ipv6.tp.src;
1513                                 tcp_key->tcp_dst = swkey->ipv6.tp.dst;
1514                         }
1515                 } else if (swkey->ip.proto == IPPROTO_UDP) {
1516                         struct ovs_key_udp *udp_key;
1517
1518                         nla = nla_reserve(skb, OVS_KEY_ATTR_UDP, sizeof(*udp_key));
1519                         if (!nla)
1520                                 goto nla_put_failure;
1521                         udp_key = nla_data(nla);
1522                         if (swkey->eth.type == htons(ETH_P_IP)) {
1523                                 udp_key->udp_src = swkey->ipv4.tp.src;
1524                                 udp_key->udp_dst = swkey->ipv4.tp.dst;
1525                         } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
1526                                 udp_key->udp_src = swkey->ipv6.tp.src;
1527                                 udp_key->udp_dst = swkey->ipv6.tp.dst;
1528                         }
1529                 } else if (swkey->eth.type == htons(ETH_P_IP) &&
1530                            swkey->ip.proto == IPPROTO_ICMP) {
1531                         struct ovs_key_icmp *icmp_key;
1532
1533                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMP, sizeof(*icmp_key));
1534                         if (!nla)
1535                                 goto nla_put_failure;
1536                         icmp_key = nla_data(nla);
1537                         icmp_key->icmp_type = ntohs(swkey->ipv4.tp.src);
1538                         icmp_key->icmp_code = ntohs(swkey->ipv4.tp.dst);
1539                 } else if (swkey->eth.type == htons(ETH_P_IPV6) &&
1540                            swkey->ip.proto == IPPROTO_ICMPV6) {
1541                         struct ovs_key_icmpv6 *icmpv6_key;
1542
1543                         nla = nla_reserve(skb, OVS_KEY_ATTR_ICMPV6,
1544                                                 sizeof(*icmpv6_key));
1545                         if (!nla)
1546                                 goto nla_put_failure;
1547                         icmpv6_key = nla_data(nla);
1548                         icmpv6_key->icmpv6_type = ntohs(swkey->ipv6.tp.src);
1549                         icmpv6_key->icmpv6_code = ntohs(swkey->ipv6.tp.dst);
1550
1551                         if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION ||
1552                             icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) {
1553                                 struct ovs_key_nd *nd_key;
1554
1555                                 nla = nla_reserve(skb, OVS_KEY_ATTR_ND, sizeof(*nd_key));
1556                                 if (!nla)
1557                                         goto nla_put_failure;
1558                                 nd_key = nla_data(nla);
1559                                 memcpy(nd_key->nd_target, &swkey->ipv6.nd.target,
1560                                                         sizeof(nd_key->nd_target));
1561                                 memcpy(nd_key->nd_sll, swkey->ipv6.nd.sll, ETH_ALEN);
1562                                 memcpy(nd_key->nd_tll, swkey->ipv6.nd.tll, ETH_ALEN);
1563                         }
1564                 }
1565         }
1566
1567 unencap:
1568         if (encap)
1569                 nla_nest_end(skb, encap);
1570
1571         return 0;
1572
1573 nla_put_failure:
1574         return -EMSGSIZE;
1575 }
1576
1577 /* Initializes the flow module.
1578  * Returns zero if successful or a negative error code. */
1579 int ovs_flow_init(void)
1580 {
1581         flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
1582                                         0, NULL);
1583         if (flow_cache == NULL)
1584                 return -ENOMEM;
1585
1586         return 0;
1587 }
1588
1589 /* Uninitializes the flow module. */
1590 void ovs_flow_exit(void)
1591 {
1592         kmem_cache_destroy(flow_cache);
1593 }