datapath/tunnel.c

   1 /*
   2  * Copyright (c) 2007-2012 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/if_arp.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/ip.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/igmp.h>
  26 #include <linux/in.h>
  27 #include <linux/in_route.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/jhash.h>
  30 #include <linux/list.h>
  31 #include <linux/kernel.h>
  32 #include <linux/version.h>
  33 #include <linux/workqueue.h>
  34 #include <linux/rculist.h>
  35
  36 #include <net/dsfield.h>
  37 #include <net/dst.h>
  38 #include <net/icmp.h>
  39 #include <net/inet_ecn.h>
  40 #include <net/ip.h>
  41 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
  42 #include <net/ipv6.h>
  43 #endif
  44 #include <net/route.h>
  45 #include <net/xfrm.h>
  46
  47 #include "checksum.h"
  48 #include "datapath.h"
  49 #include "tunnel.h"
  50 #include "vlan.h"
  51 #include "vport.h"
  52 #include "vport-internal_dev.h"
  53
  54 #define PORT_TABLE_SIZE  1024
  55
  56 static struct hlist_head *port_table __read_mostly;
  57
  58 /*
  59  * These are just used as an optimization: they don't require any kind of
  60  * synchronization because we could have just as easily read the value before
  61  * the port change happened.
  62  */
  63 static unsigned int key_local_remote_ports __read_mostly;
  64 static unsigned int key_remote_ports __read_mostly;
  65 static unsigned int key_multicast_ports __read_mostly;
  66 static unsigned int local_remote_ports __read_mostly;
  67 static unsigned int remote_ports __read_mostly;
  68 static unsigned int null_ports __read_mostly;
  69 static unsigned int multicast_ports __read_mostly;
  70
  71 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  72 #define rt_dst(rt) (rt->dst)
  73 #else
  74 #define rt_dst(rt) (rt->u.dst)
  75 #endif
  76
  77 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
  78 {
  79         return vport_from_priv(tnl_vport);
  80 }
  81
  82 static void free_config_rcu(struct rcu_head *rcu)
  83 {
  84         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
  85         kfree(c);
  86 }
  87
  88 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
  89  * within an RCU callback.  Fortunately this part doesn't require waiting for
  90  * an RCU grace period.
  91  */
  92 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
  93 {
  94         ASSERT_RTNL();
  95         if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
  96                 struct in_device *in_dev;
  97                 in_dev = inetdev_by_index(port_key_get_net(&mutable->key), mutable->mlink);
  98                 if (in_dev)
  99                         ip_mc_dec_group(in_dev, mutable->key.daddr);
 100         }
 101 }
 102
 103 static void assign_config_rcu(struct vport *vport,
 104                               struct tnl_mutable_config *new_config)
 105 {
 106         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 107         struct tnl_mutable_config *old_config;
 108
 109         old_config = rtnl_dereference(tnl_vport->mutable);
 110         rcu_assign_pointer(tnl_vport->mutable, new_config);
 111
 112         free_mutable_rtnl(old_config);
 113         call_rcu(&old_config->rcu, free_config_rcu);
 114 }
 115
 116 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 117 {
 118         bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
 119
 120         if (mutable->flags & TNL_F_IN_KEY_MATCH) {
 121                 if (mutable->key.saddr)
 122                         return &local_remote_ports;
 123                 else if (is_multicast)
 124                         return &multicast_ports;
 125                 else
 126                         return &remote_ports;
 127         } else {
 128                 if (mutable->key.saddr)
 129                         return &key_local_remote_ports;
 130                 else if (is_multicast)
 131                         return &key_multicast_ports;
 132                 else if (mutable->key.daddr)
 133                         return &key_remote_ports;
 134                 else
 135                         return &null_ports;
 136         }
 137 }
 138
 139 static u32 port_hash(const struct port_lookup_key *key)
 140 {
 141         return jhash2((u32 *)key, (PORT_KEY_LEN / sizeof(u32)), 0);
 142 }
 143
 144 static struct hlist_head *find_bucket(u32 hash)
 145 {
 146         return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
 147 }
 148
 149 static void port_table_add_port(struct vport *vport)
 150 {
 151         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 152         const struct tnl_mutable_config *mutable;
 153         u32 hash;
 154
 155         mutable = rtnl_dereference(tnl_vport->mutable);
 156         hash = port_hash(&mutable->key);
 157         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 158
 159         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 160 }
 161
 162 static void port_table_move_port(struct vport *vport,
 163                       struct tnl_mutable_config *new_mutable)
 164 {
 165         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 166         u32 hash;
 167
 168         hash = port_hash(&new_mutable->key);
 169         hlist_del_init_rcu(&tnl_vport->hash_node);
 170         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 171
 172         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 173         assign_config_rcu(vport, new_mutable);
 174         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 175 }
 176
 177 static void port_table_remove_port(struct vport *vport)
 178 {
 179         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 180
 181         hlist_del_init_rcu(&tnl_vport->hash_node);
 182
 183         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 184 }
 185
 186 static struct vport *port_table_lookup(struct port_lookup_key *key,
 187                                        const struct tnl_mutable_config **pmutable)
 188 {
 189         struct hlist_node *n;
 190         struct hlist_head *bucket;
 191         u32 hash = port_hash(key);
 192         struct tnl_vport *tnl_vport;
 193
 194         bucket = find_bucket(hash);
 195
 196         hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
 197                 struct tnl_mutable_config *mutable;
 198
 199                 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 200                 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
 201                         *pmutable = mutable;
 202                         return tnl_vport_to_vport(tnl_vport);
 203                 }
 204         }
 205
 206         return NULL;
 207 }
 208
 209 struct vport *ovs_tnl_find_port(struct net *net, __be32 saddr, __be32 daddr,
 210                                 __be64 key, int tunnel_type,
 211                                 const struct tnl_mutable_config **mutable)
 212 {
 213         struct port_lookup_key lookup;
 214         struct vport *vport;
 215         bool is_multicast = ipv4_is_multicast(saddr);
 216
 217         port_key_set_net(&lookup, net);
 218         lookup.saddr = saddr;
 219         lookup.daddr = daddr;
 220
 221         /* First try for exact match on in_key. */
 222         lookup.in_key = key;
 223         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 224         if (!is_multicast && key_local_remote_ports) {
 225                 vport = port_table_lookup(&lookup, mutable);
 226                 if (vport)
 227                         return vport;
 228         }
 229         if (key_remote_ports) {
 230                 lookup.saddr = 0;
 231                 vport = port_table_lookup(&lookup, mutable);
 232                 if (vport)
 233                         return vport;
 234
 235                 lookup.saddr = saddr;
 236         }
 237
 238         /* Then try matches that wildcard in_key. */
 239         lookup.in_key = 0;
 240         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 241         if (!is_multicast && local_remote_ports) {
 242                 vport = port_table_lookup(&lookup, mutable);
 243                 if (vport)
 244                         return vport;
 245         }
 246         if (remote_ports) {
 247                 lookup.saddr = 0;
 248                 vport = port_table_lookup(&lookup, mutable);
 249                 if (vport)
 250                         return vport;
 251         }
 252
 253         if (is_multicast) {
 254                 lookup.saddr = 0;
 255                 lookup.daddr = saddr;
 256                 if (key_multicast_ports) {
 257                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 258                         lookup.in_key = key;
 259                         vport = port_table_lookup(&lookup, mutable);
 260                         if (vport)
 261                                 return vport;
 262                 }
 263                 if (multicast_ports) {
 264                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 265                         lookup.in_key = 0;
 266                         vport = port_table_lookup(&lookup, mutable);
 267                         if (vport)
 268                                 return vport;
 269                 }
 270         }
 271
 272         if (null_ports) {
 273                 lookup.daddr = 0;
 274                 lookup.saddr = 0;
 275                 lookup.in_key = 0;
 276                 lookup.tunnel_type = tunnel_type;
 277                 vport = port_table_lookup(&lookup, mutable);
 278                 if (vport)
 279                         return vport;
 280         }
 281         return NULL;
 282 }
 283
 284 static void ecn_decapsulate(struct sk_buff *skb)
 285 {
 286         if (unlikely(INET_ECN_is_ce(OVS_CB(skb)->tun_key->ipv4_tos))) {
 287                 __be16 protocol = skb->protocol;
 288
 289                 skb_set_network_header(skb, ETH_HLEN);
 290
 291                 if (protocol == htons(ETH_P_8021Q)) {
 292                         if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 293                                 return;
 294
 295                         protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 296                         skb_set_network_header(skb, VLAN_ETH_HLEN);
 297                 }
 298
 299                 if (protocol == htons(ETH_P_IP)) {
 300                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 301                             + sizeof(struct iphdr))))
 302                                 return;
 303
 304                         IP_ECN_set_ce(ip_hdr(skb));
 305                 }
 306 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 307                 else if (protocol == htons(ETH_P_IPV6)) {
 308                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 309                             + sizeof(struct ipv6hdr))))
 310                                 return;
 311
 312                         IP6_ECN_set_ce(ipv6_hdr(skb));
 313                 }
 314 #endif
 315         }
 316 }
 317
 318 /**
 319  *      ovs_tnl_rcv - ingress point for generic tunnel code
 320  *
 321  * @vport: port this packet was received on
 322  * @skb: received packet
 323  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
 324  *
 325  * Must be called with rcu_read_lock.
 326  *
 327  * Packets received by this function are in the following state:
 328  * - skb->data points to the inner Ethernet header.
 329  * - The inner Ethernet header is in the linear data area.
 330  * - skb->csum does not include the inner Ethernet header.
 331  * - The layer pointers are undefined.
 332  */
 333 void ovs_tnl_rcv(struct vport *vport, struct sk_buff *skb)
 334 {
 335         struct ethhdr *eh;
 336
 337         skb_reset_mac_header(skb);
 338         eh = eth_hdr(skb);
 339
 340         if (likely(ntohs(eh->h_proto) >= 1536))
 341                 skb->protocol = eh->h_proto;
 342         else
 343                 skb->protocol = htons(ETH_P_802_2);
 344
 345         skb_dst_drop(skb);
 346         nf_reset(skb);
 347         skb_clear_rxhash(skb);
 348         secpath_reset(skb);
 349
 350         ecn_decapsulate(skb);
 351         vlan_set_tci(skb, 0);
 352
 353         if (unlikely(compute_ip_summed(skb, false))) {
 354                 kfree_skb(skb);
 355                 return;
 356         }
 357
 358         ovs_vport_receive(vport, skb);
 359 }
 360
 361 static bool check_ipv4_address(__be32 addr)
 362 {
 363         if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
 364             || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
 365                 return false;
 366
 367         return true;
 368 }
 369
 370 static bool ipv4_should_icmp(struct sk_buff *skb)
 371 {
 372         struct iphdr *old_iph = ip_hdr(skb);
 373
 374         /* Don't respond to L2 broadcast. */
 375         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
 376                 return false;
 377
 378         /* Don't respond to L3 broadcast or invalid addresses. */
 379         if (!check_ipv4_address(old_iph->daddr) ||
 380             !check_ipv4_address(old_iph->saddr))
 381                 return false;
 382
 383         /* Only respond to the first fragment. */
 384         if (old_iph->frag_off & htons(IP_OFFSET))
 385                 return false;
 386
 387         /* Don't respond to ICMP error messages. */
 388         if (old_iph->protocol == IPPROTO_ICMP) {
 389                 u8 icmp_type, *icmp_typep;
 390
 391                 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
 392                                                 (old_iph->ihl << 2) +
 393                                                 offsetof(struct icmphdr, type) -
 394                                                 skb->data, sizeof(icmp_type),
 395                                                 &icmp_type);
 396
 397                 if (!icmp_typep)
 398                         return false;
 399
 400                 if (*icmp_typep > NR_ICMP_TYPES
 401                         || (*icmp_typep <= ICMP_PARAMETERPROB
 402                                 && *icmp_typep != ICMP_ECHOREPLY
 403                                 && *icmp_typep != ICMP_ECHO))
 404                         return false;
 405         }
 406
 407         return true;
 408 }
 409
 410 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 411                             unsigned int mtu, unsigned int payload_length)
 412 {
 413         struct iphdr *iph, *old_iph = ip_hdr(skb);
 414         struct icmphdr *icmph;
 415         u8 *payload;
 416
 417         iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
 418         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 419         payload = skb_put(nskb, payload_length);
 420
 421         /* IP */
 422         iph->version            =       4;
 423         iph->ihl                =       sizeof(struct iphdr) >> 2;
 424         iph->tos                =       (old_iph->tos & IPTOS_TOS_MASK) |
 425                                         IPTOS_PREC_INTERNETCONTROL;
 426         iph->tot_len            =       htons(sizeof(struct iphdr)
 427                                               + sizeof(struct icmphdr)
 428                                               + payload_length);
 429         get_random_bytes(&iph->id, sizeof(iph->id));
 430         iph->frag_off           =       0;
 431         iph->ttl                =       IPDEFTTL;
 432         iph->protocol           =       IPPROTO_ICMP;
 433         iph->daddr              =       old_iph->saddr;
 434         iph->saddr              =       old_iph->daddr;
 435
 436         ip_send_check(iph);
 437
 438         /* ICMP */
 439         icmph->type             =       ICMP_DEST_UNREACH;
 440         icmph->code             =       ICMP_FRAG_NEEDED;
 441         icmph->un.gateway       =       htonl(mtu);
 442         icmph->checksum         =       0;
 443
 444         nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
 445         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
 446                                             payload, payload_length,
 447                                             nskb->csum);
 448         icmph->checksum = csum_fold(nskb->csum);
 449 }
 450
 451 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 452 static bool ipv6_should_icmp(struct sk_buff *skb)
 453 {
 454         struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
 455         int addr_type;
 456         int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
 457         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 458         __be16 frag_off;
 459
 460         /* Check source address is valid. */
 461         addr_type = ipv6_addr_type(&old_ipv6h->saddr);
 462         if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
 463                 return false;
 464
 465         /* Don't reply to unspecified addresses. */
 466         if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
 467                 return false;
 468
 469         /* Don't respond to ICMP error messages. */
 470         payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr, &frag_off);
 471         if (payload_off < 0)
 472                 return false;
 473
 474         if (nexthdr == NEXTHDR_ICMP) {
 475                 u8 icmp_type, *icmp_typep;
 476
 477                 icmp_typep = skb_header_pointer(skb, payload_off +
 478                                                 offsetof(struct icmp6hdr,
 479                                                         icmp6_type),
 480                                                 sizeof(icmp_type), &icmp_type);
 481
 482                 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
 483                         return false;
 484         }
 485
 486         return true;
 487 }
 488
 489 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 490                             unsigned int mtu, unsigned int payload_length)
 491 {
 492         struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
 493         struct icmp6hdr *icmp6h;
 494         u8 *payload;
 495
 496         ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
 497         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
 498         payload = skb_put(nskb, payload_length);
 499
 500         /* IPv6 */
 501         ipv6h->version          =       6;
 502         ipv6h->priority         =       0;
 503         memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
 504         ipv6h->payload_len      =       htons(sizeof(struct icmp6hdr)
 505                                               + payload_length);
 506         ipv6h->nexthdr          =       NEXTHDR_ICMP;
 507         ipv6h->hop_limit        =       IPV6_DEFAULT_HOPLIMIT;
 508         ipv6h->daddr            =       old_ipv6h->saddr;
 509         ipv6h->saddr            =       old_ipv6h->daddr;
 510
 511         /* ICMPv6 */
 512         icmp6h->icmp6_type      =       ICMPV6_PKT_TOOBIG;
 513         icmp6h->icmp6_code      =       0;
 514         icmp6h->icmp6_cksum     =       0;
 515         icmp6h->icmp6_mtu       =       htonl(mtu);
 516
 517         nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
 518         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
 519                                             payload, payload_length,
 520                                             nskb->csum);
 521         icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 522                                                 sizeof(struct icmp6hdr)
 523                                                 + payload_length,
 524                                                 ipv6h->nexthdr, nskb->csum);
 525 }
 526 #endif /* IPv6 */
 527
 528 bool ovs_tnl_frag_needed(struct vport *vport,
 529                          const struct tnl_mutable_config *mutable,
 530                          struct sk_buff *skb, unsigned int mtu)
 531 {
 532         unsigned int eth_hdr_len = ETH_HLEN;
 533         unsigned int total_length = 0, header_length = 0, payload_length;
 534         struct ethhdr *eh, *old_eh = eth_hdr(skb);
 535         struct sk_buff *nskb;
 536
 537         /* Sanity check */
 538         if (skb->protocol == htons(ETH_P_IP)) {
 539                 if (mtu < IP_MIN_MTU)
 540                         return false;
 541
 542                 if (!ipv4_should_icmp(skb))
 543                         return true;
 544         }
 545 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 546         else if (skb->protocol == htons(ETH_P_IPV6)) {
 547                 if (mtu < IPV6_MIN_MTU)
 548                         return false;
 549
 550                 /*
 551                  * In theory we should do PMTUD on IPv6 multicast messages but
 552                  * we don't have an address to send from so just fragment.
 553                  */
 554                 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
 555                         return false;
 556
 557                 if (!ipv6_should_icmp(skb))
 558                         return true;
 559         }
 560 #endif
 561         else
 562                 return false;
 563
 564         /* Allocate */
 565         if (old_eh->h_proto == htons(ETH_P_8021Q))
 566                 eth_hdr_len = VLAN_ETH_HLEN;
 567
 568         payload_length = skb->len - eth_hdr_len;
 569         if (skb->protocol == htons(ETH_P_IP)) {
 570                 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
 571                 total_length = min_t(unsigned int, header_length +
 572                                                    payload_length, 576);
 573         }
 574 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 575         else {
 576                 header_length = sizeof(struct ipv6hdr) +
 577                                 sizeof(struct icmp6hdr);
 578                 total_length = min_t(unsigned int, header_length +
 579                                                   payload_length, IPV6_MIN_MTU);
 580         }
 581 #endif
 582
 583         payload_length = total_length - header_length;
 584
 585         nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
 586                              payload_length);
 587         if (!nskb)
 588                 return false;
 589
 590         skb_reserve(nskb, NET_IP_ALIGN);
 591
 592         /* Ethernet / VLAN */
 593         eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
 594         memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
 595         memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
 596         nskb->protocol = eh->h_proto = old_eh->h_proto;
 597         if (old_eh->h_proto == htons(ETH_P_8021Q)) {
 598                 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
 599
 600                 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
 601                 vh->h_vlan_encapsulated_proto = skb->protocol;
 602         } else
 603                 vlan_set_tci(nskb, vlan_get_tci(skb));
 604         skb_reset_mac_header(nskb);
 605
 606         /* Protocol */
 607         if (skb->protocol == htons(ETH_P_IP))
 608                 ipv4_build_icmp(skb, nskb, mtu, payload_length);
 609 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 610         else
 611                 ipv6_build_icmp(skb, nskb, mtu, payload_length);
 612 #endif
 613
 614         if (unlikely(compute_ip_summed(nskb, false))) {
 615                 kfree_skb(nskb);
 616                 return false;
 617         }
 618
 619         ovs_vport_receive(vport, nskb);
 620
 621         return true;
 622 }
 623
 624 static bool check_mtu(struct sk_buff *skb,
 625                       struct vport *vport,
 626                       const struct tnl_mutable_config *mutable,
 627                       const struct rtable *rt, __be16 *frag_offp,
 628                       int tunnel_hlen)
 629 {
 630         bool df_inherit;
 631         bool pmtud;
 632         __be16 frag_off;
 633         int mtu = 0;
 634         unsigned int packet_length = skb->len - ETH_HLEN;
 635
 636         if (OVS_CB(skb)->tun_key->ipv4_dst) {
 637                 df_inherit = false;
 638                 pmtud = false;
 639                 frag_off = OVS_CB(skb)->tun_key->tun_flags & OVS_TNL_F_DONT_FRAGMENT ?
 640                                   htons(IP_DF) : 0;
 641         } else {
 642                 df_inherit = mutable->flags & TNL_F_DF_INHERIT;
 643                 pmtud = mutable->flags & TNL_F_PMTUD;
 644                 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
 645         }
 646
 647         /* Allow for one level of tagging in the packet length. */
 648         if (!vlan_tx_tag_present(skb) &&
 649             eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 650                 packet_length -= VLAN_HLEN;
 651
 652         if (pmtud) {
 653                 int vlan_header = 0;
 654
 655                 /* The tag needs to go in packet regardless of where it
 656                  * currently is, so subtract it from the MTU.
 657                  */
 658                 if (vlan_tx_tag_present(skb) ||
 659                     eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 660                         vlan_header = VLAN_HLEN;
 661
 662                 mtu = dst_mtu(&rt_dst(rt))
 663                         - ETH_HLEN
 664                         - tunnel_hlen
 665                         - vlan_header;
 666         }
 667
 668         if (skb->protocol == htons(ETH_P_IP)) {
 669                 struct iphdr *iph = ip_hdr(skb);
 670
 671                 if (df_inherit)
 672                         frag_off = iph->frag_off & htons(IP_DF);
 673
 674                 if (pmtud && iph->frag_off & htons(IP_DF)) {
 675                         mtu = max(mtu, IP_MIN_MTU);
 676
 677                         if (packet_length > mtu &&
 678                             ovs_tnl_frag_needed(vport, mutable, skb, mtu))
 679                                 return false;
 680                 }
 681         }
 682 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 683         else if (skb->protocol == htons(ETH_P_IPV6)) {
 684                 /* IPv6 requires end hosts to do fragmentation
 685                  * if the packet is above the minimum MTU.
 686                  */
 687                 if (df_inherit && packet_length > IPV6_MIN_MTU)
 688                         frag_off = htons(IP_DF);
 689
 690                 if (pmtud) {
 691                         mtu = max(mtu, IPV6_MIN_MTU);
 692
 693                         if (packet_length > mtu &&
 694                             ovs_tnl_frag_needed(vport, mutable, skb, mtu))
 695                                 return false;
 696                 }
 697         }
 698 #endif
 699
 700         *frag_offp = frag_off;
 701         return true;
 702 }
 703
 704 static struct rtable *find_route(struct net *net,
 705                 __be32 *saddr, __be32 daddr, u8 ipproto,
 706                 u8 tos)
 707 {
 708         struct rtable *rt;
 709         /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
 710          * router expect RT_TOS bits only. */
 711
 712 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 713         struct flowi fl = { .nl_u = { .ip4_u = {
 714                                         .daddr = daddr,
 715                                         .saddr = *saddr,
 716                                         .tos   = RT_TOS(tos) } },
 717                                         .proto = ipproto };
 718
 719         if (unlikely(ip_route_output_key(net, &rt, &fl)))
 720                 return ERR_PTR(-EADDRNOTAVAIL);
 721         *saddr = fl.nl_u.ip4_u.saddr;
 722         return rt;
 723 #else
 724         struct flowi4 fl = { .daddr = daddr,
 725                              .saddr = *saddr,
 726                              .flowi4_tos = RT_TOS(tos),
 727                              .flowi4_proto = ipproto };
 728
 729         rt = ip_route_output_key(net, &fl);
 730         *saddr = fl.saddr;
 731         return rt;
 732 #endif
 733 }
 734
 735 static bool need_linearize(const struct sk_buff *skb)
 736 {
 737         int i;
 738
 739         if (unlikely(skb_shinfo(skb)->frag_list))
 740                 return true;
 741
 742         /*
 743          * Generally speaking we should linearize if there are paged frags.
 744          * However, if all of the refcounts are 1 we know nobody else can
 745          * change them from underneath us and we can skip the linearization.
 746          */
 747         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 748                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
 749                         return true;
 750
 751         return false;
 752 }
 753
 754 static struct sk_buff *handle_offloads(struct sk_buff *skb,
 755                                        const struct tnl_mutable_config *mutable,
 756                                        const struct rtable *rt,
 757                                        int tunnel_hlen)
 758 {
 759         int min_headroom;
 760         int err;
 761
 762         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 763                         + tunnel_hlen
 764                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 765
 766         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 767                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 768                                                 skb_headroom(skb) +
 769                                                 16);
 770                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 771                                         0, GFP_ATOMIC);
 772                 if (unlikely(err))
 773                         goto error_free;
 774         }
 775
 776         forward_ip_summed(skb, true);
 777
 778         if (skb_is_gso(skb)) {
 779                 struct sk_buff *nskb;
 780
 781                 nskb = skb_gso_segment(skb, 0);
 782                 if (IS_ERR(nskb)) {
 783                         kfree_skb(skb);
 784                         err = PTR_ERR(nskb);
 785                         goto error;
 786                 }
 787
 788                 consume_skb(skb);
 789                 skb = nskb;
 790         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 791                 /* Pages aren't locked and could change at any time.
 792                  * If this happens after we compute the checksum, the
 793                  * checksum will be wrong.  We linearize now to avoid
 794                  * this problem.
 795                  */
 796                 if (unlikely(need_linearize(skb))) {
 797                         err = __skb_linearize(skb);
 798                         if (unlikely(err))
 799                                 goto error_free;
 800                 }
 801
 802                 err = skb_checksum_help(skb);
 803                 if (unlikely(err))
 804                         goto error_free;
 805         }
 806
 807         set_ip_summed(skb, OVS_CSUM_NONE);
 808
 809         return skb;
 810
 811 error_free:
 812         kfree_skb(skb);
 813 error:
 814         return ERR_PTR(err);
 815 }
 816
 817 static int send_frags(struct sk_buff *skb,
 818                       int tunnel_hlen)
 819 {
 820         int sent_len;
 821
 822         sent_len = 0;
 823         while (skb) {
 824                 struct sk_buff *next = skb->next;
 825                 int frag_len = skb->len - tunnel_hlen;
 826                 int err;
 827
 828                 skb->next = NULL;
 829                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 830
 831                 err = ip_local_out(skb);
 832                 skb = next;
 833                 if (unlikely(net_xmit_eval(err)))
 834                         goto free_frags;
 835                 sent_len += frag_len;
 836         }
 837
 838         return sent_len;
 839
 840 free_frags:
 841         /*
 842          * There's no point in continuing to send fragments once one has been
 843          * dropped so just free the rest.  This may help improve the congestion
 844          * that caused the first packet to be dropped.
 845          */
 846         ovs_tnl_free_linked_skbs(skb);
 847         return sent_len;
 848 }
 849
 850 int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
 851 {
 852         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 853         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
 854         enum vport_err_type err = VPORT_E_TX_ERROR;
 855         struct rtable *rt;
 856         struct ovs_key_ipv4_tunnel tun_key;
 857         int sent_len = 0;
 858         int tunnel_hlen;
 859         __be16 frag_off = 0;
 860         __be32 daddr;
 861         __be32 saddr;
 862         u8 ttl;
 863         u8 tos;
 864
 865         /* Validate the protocol headers before we try to use them. */
 866         if (skb->protocol == htons(ETH_P_8021Q) &&
 867             !vlan_tx_tag_present(skb)) {
 868                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 869                         goto error_free;
 870
 871                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 872                 skb_set_network_header(skb, VLAN_ETH_HLEN);
 873         }
 874
 875         if (skb->protocol == htons(ETH_P_IP)) {
 876                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 877                     + sizeof(struct iphdr))))
 878                         skb->protocol = 0;
 879         }
 880 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 881         else if (skb->protocol == htons(ETH_P_IPV6)) {
 882                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 883                     + sizeof(struct ipv6hdr))))
 884                         skb->protocol = 0;
 885         }
 886 #endif
 887
 888         /* If OVS_CB(skb)->tun_key is NULL, point it at the local tun_key here,
 889          * and zero it out.
 890          */
 891         if (!OVS_CB(skb)->tun_key) {
 892                 memset(&tun_key, 0, sizeof(tun_key));
 893                 OVS_CB(skb)->tun_key = &tun_key;
 894         }
 895
 896         tunnel_hlen = tnl_vport->tnl_ops->hdr_len(mutable, OVS_CB(skb)->tun_key);
 897         if (unlikely(tunnel_hlen < 0)) {
 898                 err = VPORT_E_TX_DROPPED;
 899                 goto error_free;
 900         }
 901         tunnel_hlen += sizeof(struct iphdr);
 902
 903         if (OVS_CB(skb)->tun_key->ipv4_dst) {
 904                 daddr = OVS_CB(skb)->tun_key->ipv4_dst;
 905                 saddr = OVS_CB(skb)->tun_key->ipv4_src;
 906                 tos = OVS_CB(skb)->tun_key->ipv4_tos;
 907                 ttl = OVS_CB(skb)->tun_key->ipv4_ttl;
 908         } else {
 909                 u8 inner_tos;
 910                 daddr = mutable->key.daddr;
 911                 saddr = mutable->key.saddr;
 912
 913                 if (unlikely(!daddr)) {
 914                         /* Trying to sent packet from Null-port without
 915                          * tunnel info? Drop this packet. */
 916                         err = VPORT_E_TX_DROPPED;
 917                         goto error_free;
 918                 }
 919
 920                 /* ToS */
 921                 if (skb->protocol == htons(ETH_P_IP))
 922                         inner_tos = ip_hdr(skb)->tos;
 923 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 924                 else if (skb->protocol == htons(ETH_P_IPV6))
 925                         inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
 926 #endif
 927                 else
 928                         inner_tos = 0;
 929
 930                 if (mutable->flags & TNL_F_TOS_INHERIT)
 931                         tos = inner_tos;
 932                 else
 933                         tos = mutable->tos;
 934
 935                 tos = INET_ECN_encapsulate(tos, inner_tos);
 936
 937                 /* TTL */
 938                 ttl = mutable->ttl;
 939                 if (mutable->flags & TNL_F_TTL_INHERIT) {
 940                         if (skb->protocol == htons(ETH_P_IP))
 941                                 ttl = ip_hdr(skb)->ttl;
 942 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 943                         else if (skb->protocol == htons(ETH_P_IPV6))
 944                                 ttl = ipv6_hdr(skb)->hop_limit;
 945 #endif
 946                 }
 947
 948         }
 949
 950         /* Route lookup */
 951         rt = find_route(port_key_get_net(&mutable->key), &saddr, daddr,
 952                           tnl_vport->tnl_ops->ipproto, tos);
 953         if (IS_ERR(rt))
 954                 goto error_free;
 955
 956         /* Reset SKB */
 957         nf_reset(skb);
 958         secpath_reset(skb);
 959         skb_dst_drop(skb);
 960         skb_clear_rxhash(skb);
 961
 962         /* Offloading */
 963         skb = handle_offloads(skb, mutable, rt, tunnel_hlen);
 964         if (IS_ERR(skb)) {
 965                 skb = NULL;
 966                 goto err_free_rt;
 967         }
 968
 969         /* MTU */
 970         if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off, tunnel_hlen))) {
 971                 err = VPORT_E_TX_DROPPED;
 972                 goto err_free_rt;
 973         }
 974
 975         /* TTL Fixup. */
 976         if (!OVS_CB(skb)->tun_key->ipv4_dst) {
 977                 if (!(mutable->flags & TNL_F_TTL_INHERIT)) {
 978                         if (!ttl)
 979                                 ttl = ip4_dst_hoplimit(&rt_dst(rt));
 980                 }
 981         }
 982
 983         while (skb) {
 984                 struct iphdr *iph;
 985                 struct sk_buff *next_skb = skb->next;
 986                 skb->next = NULL;
 987
 988                 if (unlikely(vlan_deaccel_tag(skb)))
 989                         goto next;
 990
 991                 skb_push(skb, tunnel_hlen);
 992                 skb_reset_network_header(skb);
 993                 skb_set_transport_header(skb, sizeof(struct iphdr));
 994
 995                 if (next_skb)
 996                         skb_dst_set(skb, dst_clone(&rt_dst(rt)));
 997                 else
 998                         skb_dst_set(skb, &rt_dst(rt));
 999
1000                 /* Push IP header. */
1001                 iph = ip_hdr(skb);
1002                 iph->version    = 4;
1003                 iph->ihl        = sizeof(struct iphdr) >> 2;
1004                 iph->protocol   = tnl_vport->tnl_ops->ipproto;
1005                 iph->daddr      = daddr;
1006                 iph->saddr      = saddr;
1007                 iph->tos        = tos;
1008                 iph->ttl        = ttl;
1009                 iph->frag_off   = frag_off;
1010                 ip_select_ident(iph, &rt_dst(rt), NULL);
1011
1012                 /* Push Tunnel header. */
1013                 skb = tnl_vport->tnl_ops->build_header(vport, mutable,
1014                                                         &rt_dst(rt), skb, tunnel_hlen);
1015                 if (unlikely(!skb))
1016                         goto next;
1017
1018                 sent_len += send_frags(skb, tunnel_hlen);
1019
1020 next:
1021                 skb = next_skb;
1022         }
1023
1024         if (unlikely(sent_len == 0))
1025                 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
1026
1027         return sent_len;
1028
1029 err_free_rt:
1030         ip_rt_put(rt);
1031 error_free:
1032         ovs_tnl_free_linked_skbs(skb);
1033         ovs_vport_record_error(vport, err);
1034         return sent_len;
1035 }
1036
1037 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1038         [OVS_TUNNEL_ATTR_FLAGS]    = { .type = NLA_U32 },
1039         [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1040         [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1041         [OVS_TUNNEL_ATTR_OUT_KEY]  = { .type = NLA_U64 },
1042         [OVS_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
1043         [OVS_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
1044         [OVS_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
1045         [OVS_TUNNEL_ATTR_DST_PORT] = { .type = NLA_U16 },
1046 };
1047
1048 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
1049  * zeroed. */
1050 static int tnl_set_config(struct net *net, struct nlattr *options,
1051                           const struct tnl_ops *tnl_ops,
1052                           const struct vport *cur_vport,
1053                           struct tnl_mutable_config *mutable)
1054 {
1055         const struct vport *old_vport;
1056         const struct tnl_mutable_config *old_mutable;
1057         struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1058         int err;
1059
1060         port_key_set_net(&mutable->key, net);
1061         mutable->key.tunnel_type = tnl_ops->tunnel_type;
1062         if (!options)
1063                 goto out;
1064
1065         err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1066         if (err)
1067                 return err;
1068
1069         if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1070                 return -EINVAL;
1071
1072         mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1073         mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1074
1075         if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
1076                 if (ipv4_is_multicast(mutable->key.daddr))
1077                         return -EINVAL;
1078                 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1079         }
1080
1081         if (a[OVS_TUNNEL_ATTR_TOS]) {
1082                 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1083                 /* Reject ToS config with ECN bits set. */
1084                 if (mutable->tos & INET_ECN_MASK)
1085                         return -EINVAL;
1086         }
1087
1088         if (a[OVS_TUNNEL_ATTR_TTL])
1089                 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1090
1091         if (a[OVS_TUNNEL_ATTR_DST_PORT])
1092                 mutable->dst_port =
1093                         htons(nla_get_u16(a[OVS_TUNNEL_ATTR_DST_PORT]));
1094
1095         if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1096                 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1097                 mutable->flags |= TNL_F_IN_KEY_MATCH;
1098         } else {
1099                 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1100                 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1101         }
1102
1103         if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1104                 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1105         else
1106                 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1107
1108         mutable->mlink = 0;
1109         if (ipv4_is_multicast(mutable->key.daddr)) {
1110                 struct net_device *dev;
1111                 struct rtable *rt;
1112                 __be32 saddr = mutable->key.saddr;
1113
1114                 rt = find_route(port_key_get_net(&mutable->key),
1115                              &saddr, mutable->key.daddr,
1116                              tnl_ops->ipproto, mutable->tos);
1117                 if (IS_ERR(rt))
1118                         return -EADDRNOTAVAIL;
1119                 dev = rt_dst(rt).dev;
1120                 ip_rt_put(rt);
1121                 if (__in_dev_get_rtnl(dev) == NULL)
1122                         return -EADDRNOTAVAIL;
1123                 mutable->mlink = dev->ifindex;
1124                 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
1125         }
1126
1127 out:
1128         old_vport = port_table_lookup(&mutable->key, &old_mutable);
1129         if (old_vport && old_vport != cur_vport)
1130                 return -EEXIST;
1131
1132         return 0;
1133 }
1134
1135 struct vport *ovs_tnl_create(const struct vport_parms *parms,
1136                              const struct vport_ops *vport_ops,
1137                              const struct tnl_ops *tnl_ops)
1138 {
1139         struct vport *vport;
1140         struct tnl_vport *tnl_vport;
1141         struct tnl_mutable_config *mutable;
1142         int initial_frag_id;
1143         int err;
1144
1145         vport = ovs_vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1146         if (IS_ERR(vport)) {
1147                 err = PTR_ERR(vport);
1148                 goto error;
1149         }
1150
1151         tnl_vport = tnl_vport_priv(vport);
1152
1153         strcpy(tnl_vport->name, parms->name);
1154         tnl_vport->tnl_ops = tnl_ops;
1155
1156         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1157         if (!mutable) {
1158                 err = -ENOMEM;
1159                 goto error_free_vport;
1160         }
1161
1162         random_ether_addr(mutable->eth_addr);
1163
1164         get_random_bytes(&initial_frag_id, sizeof(int));
1165         atomic_set(&tnl_vport->frag_id, initial_frag_id);
1166
1167         err = tnl_set_config(ovs_dp_get_net(parms->dp), parms->options, tnl_ops,
1168                              NULL, mutable);
1169         if (err)
1170                 goto error_free_mutable;
1171
1172         rcu_assign_pointer(tnl_vport->mutable, mutable);
1173
1174         port_table_add_port(vport);
1175         return vport;
1176
1177 error_free_mutable:
1178         free_mutable_rtnl(mutable);
1179         kfree(mutable);
1180 error_free_vport:
1181         ovs_vport_free(vport);
1182 error:
1183         return ERR_PTR(err);
1184 }
1185
1186 int ovs_tnl_set_options(struct vport *vport, struct nlattr *options)
1187 {
1188         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1189         const struct tnl_mutable_config *old_mutable;
1190         struct tnl_mutable_config *mutable;
1191         int err;
1192
1193         old_mutable = rtnl_dereference(tnl_vport->mutable);
1194         if (!old_mutable->key.daddr)
1195                 return -EINVAL;
1196
1197         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1198         if (!mutable) {
1199                 err = -ENOMEM;
1200                 goto error;
1201         }
1202
1203         /* Copy fields whose values should be retained. */
1204         mutable->seq = old_mutable->seq + 1;
1205         memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1206
1207         /* Parse the others configured by userspace. */
1208         err = tnl_set_config(ovs_dp_get_net(vport->dp), options, tnl_vport->tnl_ops,
1209                              vport, mutable);
1210         if (err)
1211                 goto error_free;
1212
1213         if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1214                 port_table_move_port(vport, mutable);
1215         else
1216                 assign_config_rcu(vport, mutable);
1217
1218         return 0;
1219
1220 error_free:
1221         free_mutable_rtnl(mutable);
1222         kfree(mutable);
1223 error:
1224         return err;
1225 }
1226
1227 int ovs_tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1228 {
1229         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1230         const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1231
1232         if (nla_put_u32(skb, OVS_TUNNEL_ATTR_FLAGS,
1233                       mutable->flags & TNL_F_PUBLIC) ||
1234             nla_put_be32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr))
1235                 goto nla_put_failure;
1236
1237         if (!(mutable->flags & TNL_F_IN_KEY_MATCH) &&
1238             nla_put_be64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key))
1239                 goto nla_put_failure;
1240         if (!(mutable->flags & TNL_F_OUT_KEY_ACTION) &&
1241             nla_put_be64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key))
1242                 goto nla_put_failure;
1243         if (mutable->key.saddr &&
1244             nla_put_be32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr))
1245                 goto nla_put_failure;
1246         if (mutable->tos && nla_put_u8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos))
1247                 goto nla_put_failure;
1248         if (mutable->ttl && nla_put_u8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl))
1249                 goto nla_put_failure;
1250         if (mutable->dst_port && nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT,
1251                                              ntohs(mutable->dst_port)))
1252                 goto nla_put_failure;
1253
1254         return 0;
1255
1256 nla_put_failure:
1257         return -EMSGSIZE;
1258 }
1259
1260 static void free_port_rcu(struct rcu_head *rcu)
1261 {
1262         struct tnl_vport *tnl_vport = container_of(rcu,
1263                                                    struct tnl_vport, rcu);
1264
1265         kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1266         ovs_vport_free(tnl_vport_to_vport(tnl_vport));
1267 }
1268
1269 void ovs_tnl_destroy(struct vport *vport)
1270 {
1271         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1272         struct tnl_mutable_config *mutable;
1273
1274         mutable = rtnl_dereference(tnl_vport->mutable);
1275         port_table_remove_port(vport);
1276         free_mutable_rtnl(mutable);
1277         call_rcu(&tnl_vport->rcu, free_port_rcu);
1278 }
1279
1280 int ovs_tnl_set_addr(struct vport *vport, const unsigned char *addr)
1281 {
1282         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1283         struct tnl_mutable_config *old_mutable, *mutable;
1284
1285         old_mutable = rtnl_dereference(tnl_vport->mutable);
1286         mutable = kmemdup(old_mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1287         if (!mutable)
1288                 return -ENOMEM;
1289
1290         old_mutable->mlink = 0;
1291
1292         memcpy(mutable->eth_addr, addr, ETH_ALEN);
1293         assign_config_rcu(vport, mutable);
1294
1295         return 0;
1296 }
1297
1298 const char *ovs_tnl_get_name(const struct vport *vport)
1299 {
1300         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1301         return tnl_vport->name;
1302 }
1303
1304 const unsigned char *ovs_tnl_get_addr(const struct vport *vport)
1305 {
1306         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1307         return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1308 }
1309
1310 void ovs_tnl_free_linked_skbs(struct sk_buff *skb)
1311 {
1312         while (skb) {
1313                 struct sk_buff *next = skb->next;
1314                 kfree_skb(skb);
1315                 skb = next;
1316         }
1317 }
1318
1319 int ovs_tnl_init(void)
1320 {
1321         int i;
1322
1323         port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1324                              GFP_KERNEL);
1325         if (!port_table)
1326                 return -ENOMEM;
1327
1328         for (i = 0; i < PORT_TABLE_SIZE; i++)
1329                 INIT_HLIST_HEAD(&port_table[i]);
1330
1331         return 0;
1332 }
1333
1334 void ovs_tnl_exit(void)
1335 {
1336         kfree(port_table);
1337 }