datapath/tunnel.c

   1 /*
   2  * Copyright (c) 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include <linux/if_arp.h>
  10 #include <linux/if_ether.h>
  11 #include <linux/ip.h>
  12 #include <linux/if_vlan.h>
  13 #include <linux/in.h>
  14 #include <linux/in_route.h>
  15 #include <linux/jhash.h>
  16 #include <linux/kernel.h>
  17 #include <linux/version.h>
  18 #include <linux/workqueue.h>
  19
  20 #include <net/dsfield.h>
  21 #include <net/dst.h>
  22 #include <net/icmp.h>
  23 #include <net/inet_ecn.h>
  24 #include <net/ip.h>
  25 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
  26 #include <net/ipv6.h>
  27 #endif
  28 #include <net/route.h>
  29 #include <net/xfrm.h>
  30
  31 #include "actions.h"
  32 #include "checksum.h"
  33 #include "datapath.h"
  34 #include "table.h"
  35 #include "tunnel.h"
  36 #include "vlan.h"
  37 #include "vport.h"
  38 #include "vport-generic.h"
  39 #include "vport-internal_dev.h"
  40
  41 #ifdef NEED_CACHE_TIMEOUT
  42 /*
  43  * On kernels where we can't quickly detect changes in the rest of the system
  44  * we use an expiration time to invalidate the cache.  A shorter expiration
  45  * reduces the length of time that we may potentially blackhole packets while
  46  * a longer time increases performance by reducing the frequency that the
  47  * cache needs to be rebuilt.  A variety of factors may cause the cache to be
  48  * invalidated before the expiration time but this is the maximum.  The time
  49  * is expressed in jiffies.
  50  */
  51 #define MAX_CACHE_EXP HZ
  52 #endif
  53
  54 /*
  55  * Interval to check for and remove caches that are no longer valid.  Caches
  56  * are checked for validity before they are used for packet encapsulation and
  57  * old caches are removed at that time.  However, if no packets are sent through
  58  * the tunnel then the cache will never be destroyed.  Since it holds
  59  * references to a number of system objects, the cache will continue to use
  60  * system resources by not allowing those objects to be destroyed.  The cache
  61  * cleaner is periodically run to free invalid caches.  It does not
  62  * significantly affect system performance.  A lower interval will release
  63  * resources faster but will itself consume resources by requiring more frequent
  64  * checks.  A longer interval may result in messages being printed to the kernel
  65  * message buffer about unreleased resources.  The interval is expressed in
  66  * jiffies.
  67  */
  68 #define CACHE_CLEANER_INTERVAL (5 * HZ)
  69
  70 #define CACHE_DATA_ALIGN 16
  71
  72 static struct tbl __rcu *port_table __read_mostly;
  73
  74 static void cache_cleaner(struct work_struct *work);
  75 static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
  76
  77 /*
  78  * These are just used as an optimization: they don't require any kind of
  79  * synchronization because we could have just as easily read the value before
  80  * the port change happened.
  81  */
  82 static unsigned int key_local_remote_ports __read_mostly;
  83 static unsigned int key_remote_ports __read_mostly;
  84 static unsigned int local_remote_ports __read_mostly;
  85 static unsigned int remote_ports __read_mostly;
  86
  87 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  88 #define rt_dst(rt) (rt->dst)
  89 #else
  90 #define rt_dst(rt) (rt->u.dst)
  91 #endif
  92
  93 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
  94 {
  95         return vport_from_priv(tnl_vport);
  96 }
  97
  98 static inline struct tnl_vport *tnl_vport_table_cast(const struct tbl_node *node)
  99 {
 100         return container_of(node, struct tnl_vport, tbl_node);
 101 }
 102
 103 /* This is analogous to rtnl_dereference for the tunnel cache.  It checks that
 104  * cache_lock is held, so it is only for update side code.
 105  */
 106 static inline struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
 107 {
 108         return rcu_dereference_protected(tnl_vport->cache,
 109                                          lockdep_is_held(&tnl_vport->cache_lock));
 110 }
 111
 112 static inline void schedule_cache_cleaner(void)
 113 {
 114         schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
 115 }
 116
 117 static void free_cache(struct tnl_cache *cache)
 118 {
 119         if (!cache)
 120                 return;
 121
 122         flow_put(cache->flow);
 123         ip_rt_put(cache->rt);
 124         kfree(cache);
 125 }
 126
 127 static void free_config_rcu(struct rcu_head *rcu)
 128 {
 129         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
 130         kfree(c);
 131 }
 132
 133 static void free_cache_rcu(struct rcu_head *rcu)
 134 {
 135         struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
 136         free_cache(c);
 137 }
 138
 139 static void assign_config_rcu(struct vport *vport,
 140                               struct tnl_mutable_config *new_config)
 141 {
 142         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 143         struct tnl_mutable_config *old_config;
 144
 145         old_config = rtnl_dereference(tnl_vport->mutable);
 146         rcu_assign_pointer(tnl_vport->mutable, new_config);
 147         call_rcu(&old_config->rcu, free_config_rcu);
 148 }
 149
 150 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
 151 {
 152         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 153         struct tnl_cache *old_cache;
 154
 155         old_cache = cache_dereference(tnl_vport);
 156         rcu_assign_pointer(tnl_vport->cache, new_cache);
 157
 158         if (old_cache)
 159                 call_rcu(&old_cache->rcu, free_cache_rcu);
 160 }
 161
 162 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 163 {
 164         if (mutable->flags & TNL_F_IN_KEY_MATCH) {
 165                 if (mutable->saddr)
 166                         return &local_remote_ports;
 167                 else
 168                         return &remote_ports;
 169         } else {
 170                 if (mutable->saddr)
 171                         return &key_local_remote_ports;
 172                 else
 173                         return &key_remote_ports;
 174         }
 175 }
 176
 177 struct port_lookup_key {
 178         const struct tnl_mutable_config *mutable;
 179         __be64 key;
 180         u32 tunnel_type;
 181         __be32 saddr;
 182         __be32 daddr;
 183 };
 184
 185 /*
 186  * Modifies 'target' to store the rcu_dereferenced pointer that was used to do
 187  * the comparision.
 188  */
 189 static int port_cmp(const struct tbl_node *node, void *target, int unused)
 190 {
 191         const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node);
 192         struct port_lookup_key *lookup = target;
 193
 194         lookup->mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 195
 196         return (lookup->mutable->tunnel_type == lookup->tunnel_type &&
 197                 lookup->mutable->daddr == lookup->daddr &&
 198                 lookup->mutable->in_key == lookup->key &&
 199                 lookup->mutable->saddr == lookup->saddr);
 200 }
 201
 202 static u32 port_hash(struct port_lookup_key *k)
 203 {
 204         u32 x = jhash_3words((__force u32)k->saddr, (__force u32)k->daddr,
 205                              k->tunnel_type, 0);
 206         return jhash_2words((__force u64)k->key >> 32, (__force u32)k->key, x);
 207 }
 208
 209 static u32 mutable_hash(const struct tnl_mutable_config *mutable)
 210 {
 211         struct port_lookup_key lookup;
 212
 213         lookup.saddr = mutable->saddr;
 214         lookup.daddr = mutable->daddr;
 215         lookup.key = mutable->in_key;
 216         lookup.tunnel_type = mutable->tunnel_type;
 217
 218         return port_hash(&lookup);
 219 }
 220
 221 static void check_table_empty(void)
 222 {
 223         struct tbl *old_table = rtnl_dereference(port_table);
 224
 225         if (tbl_count(old_table) == 0) {
 226                 cancel_delayed_work_sync(&cache_cleaner_wq);
 227                 rcu_assign_pointer(port_table, NULL);
 228                 tbl_deferred_destroy(old_table, NULL);
 229         }
 230 }
 231
 232 static int add_port(struct vport *vport)
 233 {
 234         struct tbl *cur_table = rtnl_dereference(port_table);
 235         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 236         int err;
 237
 238         if (!port_table) {
 239                 struct tbl *new_table;
 240
 241                 new_table = tbl_create(TBL_MIN_BUCKETS);
 242                 if (!new_table)
 243                         return -ENOMEM;
 244
 245                 rcu_assign_pointer(port_table, new_table);
 246                 schedule_cache_cleaner();
 247
 248         } else if (tbl_count(cur_table) > tbl_n_buckets(cur_table)) {
 249                 struct tbl *new_table;
 250
 251                 new_table = tbl_expand(cur_table);
 252                 if (IS_ERR(new_table)) {
 253                         if (PTR_ERR(new_table) != -ENOSPC)
 254                                 return PTR_ERR(new_table);
 255                 } else {
 256                         rcu_assign_pointer(port_table, new_table);
 257                         tbl_deferred_destroy(cur_table, NULL);
 258                 }
 259         }
 260
 261         err = tbl_insert(rtnl_dereference(port_table), &tnl_vport->tbl_node,
 262                          mutable_hash(rtnl_dereference(tnl_vport->mutable)));
 263         if (err) {
 264                 check_table_empty();
 265                 return err;
 266         }
 267
 268         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 269
 270         return 0;
 271 }
 272
 273 static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable)
 274 {
 275         int err;
 276         struct tbl *cur_table = rtnl_dereference(port_table);
 277         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 278         u32 hash;
 279
 280         hash = mutable_hash(new_mutable);
 281         if (hash == tnl_vport->tbl_node.hash)
 282                 goto table_updated;
 283
 284         /*
 285          * Ideally we should make this move atomic to avoid having gaps in
 286          * finding tunnels or the possibility of failure.  However, if we do
 287          * find a tunnel it will always be consistent.
 288          */
 289         err = tbl_remove(cur_table, &tnl_vport->tbl_node);
 290         if (err)
 291                 return err;
 292
 293         err = tbl_insert(cur_table, &tnl_vport->tbl_node, hash);
 294         if (err) {
 295                 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 296                 check_table_empty();
 297                 return err;
 298         }
 299
 300 table_updated:
 301         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 302         assign_config_rcu(vport, new_mutable);
 303         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 304
 305         return 0;
 306 }
 307
 308 static int del_port(struct vport *vport)
 309 {
 310         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 311         int err;
 312
 313         err = tbl_remove(rtnl_dereference(port_table), &tnl_vport->tbl_node);
 314         if (err)
 315                 return err;
 316
 317         check_table_empty();
 318         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 319
 320         return 0;
 321 }
 322
 323 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key,
 324                             int tunnel_type,
 325                             const struct tnl_mutable_config **mutable)
 326 {
 327         struct port_lookup_key lookup;
 328         struct tbl *table = rcu_dereference_rtnl(port_table);
 329         struct tbl_node *tbl_node;
 330
 331         if (unlikely(!table))
 332                 return NULL;
 333
 334         lookup.saddr = saddr;
 335         lookup.daddr = daddr;
 336
 337         if (tunnel_type & TNL_T_KEY_EXACT) {
 338                 lookup.key = key;
 339                 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_MATCH;
 340
 341                 if (key_local_remote_ports) {
 342                         tbl_node = tbl_lookup(table, &lookup, sizeof(lookup),
 343                                               port_hash(&lookup), port_cmp);
 344                         if (tbl_node)
 345                                 goto found;
 346                 }
 347
 348                 if (key_remote_ports) {
 349                         lookup.saddr = 0;
 350
 351                         tbl_node = tbl_lookup(table, &lookup, sizeof(lookup),
 352                                               port_hash(&lookup), port_cmp);
 353                         if (tbl_node)
 354                                 goto found;
 355
 356                         lookup.saddr = saddr;
 357                 }
 358         }
 359
 360         if (tunnel_type & TNL_T_KEY_MATCH) {
 361                 lookup.key = 0;
 362                 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_EXACT;
 363
 364                 if (local_remote_ports) {
 365                         tbl_node = tbl_lookup(table, &lookup, sizeof(lookup),
 366                                               port_hash(&lookup), port_cmp);
 367                         if (tbl_node)
 368                                 goto found;
 369                 }
 370
 371                 if (remote_ports) {
 372                         lookup.saddr = 0;
 373
 374                         tbl_node = tbl_lookup(table, &lookup, sizeof(lookup),
 375                                               port_hash(&lookup), port_cmp);
 376                         if (tbl_node)
 377                                 goto found;
 378                 }
 379         }
 380
 381         return NULL;
 382
 383 found:
 384         *mutable = lookup.mutable;
 385         return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node));
 386 }
 387
 388 static void ecn_decapsulate(struct sk_buff *skb, u8 tos)
 389 {
 390         if (unlikely(INET_ECN_is_ce(tos))) {
 391                 __be16 protocol = skb->protocol;
 392
 393                 skb_set_network_header(skb, ETH_HLEN);
 394
 395                 if (protocol == htons(ETH_P_8021Q)) {
 396                         if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 397                                 return;
 398
 399                         protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 400                         skb_set_network_header(skb, VLAN_ETH_HLEN);
 401                 }
 402
 403                 if (protocol == htons(ETH_P_IP)) {
 404                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 405                             + sizeof(struct iphdr))))
 406                                 return;
 407
 408                         IP_ECN_set_ce(ip_hdr(skb));
 409                 }
 410 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 411                 else if (protocol == htons(ETH_P_IPV6)) {
 412                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 413                             + sizeof(struct ipv6hdr))))
 414                                 return;
 415
 416                         IP6_ECN_set_ce(ipv6_hdr(skb));
 417                 }
 418 #endif
 419         }
 420 }
 421
 422 /**
 423  *      tnl_rcv - ingress point for generic tunnel code
 424  *
 425  * @vport: port this packet was received on
 426  * @skb: received packet
 427  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
 428  *
 429  * Must be called with rcu_read_lock.
 430  *
 431  * Packets received by this function are in the following state:
 432  * - skb->data points to the inner Ethernet header.
 433  * - The inner Ethernet header is in the linear data area.
 434  * - skb->csum does not include the inner Ethernet header.
 435  * - The layer pointers are undefined.
 436  */
 437 void tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos)
 438 {
 439         struct ethhdr *eh;
 440
 441         skb_reset_mac_header(skb);
 442         eh = eth_hdr(skb);
 443
 444         if (likely(ntohs(eh->h_proto) >= 1536))
 445                 skb->protocol = eh->h_proto;
 446         else
 447                 skb->protocol = htons(ETH_P_802_2);
 448
 449         skb_dst_drop(skb);
 450         nf_reset(skb);
 451         skb_clear_rxhash(skb);
 452         secpath_reset(skb);
 453
 454         ecn_decapsulate(skb, tos);
 455         vlan_set_tci(skb, 0);
 456
 457         if (unlikely(compute_ip_summed(skb, false))) {
 458                 kfree_skb(skb);
 459                 return;
 460         }
 461
 462         vport_receive(vport, skb);
 463 }
 464
 465 static bool check_ipv4_address(__be32 addr)
 466 {
 467         if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
 468             || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
 469                 return false;
 470
 471         return true;
 472 }
 473
 474 static bool ipv4_should_icmp(struct sk_buff *skb)
 475 {
 476         struct iphdr *old_iph = ip_hdr(skb);
 477
 478         /* Don't respond to L2 broadcast. */
 479         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
 480                 return false;
 481
 482         /* Don't respond to L3 broadcast or invalid addresses. */
 483         if (!check_ipv4_address(old_iph->daddr) ||
 484             !check_ipv4_address(old_iph->saddr))
 485                 return false;
 486
 487         /* Only respond to the first fragment. */
 488         if (old_iph->frag_off & htons(IP_OFFSET))
 489                 return false;
 490
 491         /* Don't respond to ICMP error messages. */
 492         if (old_iph->protocol == IPPROTO_ICMP) {
 493                 u8 icmp_type, *icmp_typep;
 494
 495                 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
 496                                                 (old_iph->ihl << 2) +
 497                                                 offsetof(struct icmphdr, type) -
 498                                                 skb->data, sizeof(icmp_type),
 499                                                 &icmp_type);
 500
 501                 if (!icmp_typep)
 502                         return false;
 503
 504                 if (*icmp_typep > NR_ICMP_TYPES
 505                         || (*icmp_typep <= ICMP_PARAMETERPROB
 506                                 && *icmp_typep != ICMP_ECHOREPLY
 507                                 && *icmp_typep != ICMP_ECHO))
 508                         return false;
 509         }
 510
 511         return true;
 512 }
 513
 514 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 515                             unsigned int mtu, unsigned int payload_length)
 516 {
 517         struct iphdr *iph, *old_iph = ip_hdr(skb);
 518         struct icmphdr *icmph;
 519         u8 *payload;
 520
 521         iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
 522         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 523         payload = skb_put(nskb, payload_length);
 524
 525         /* IP */
 526         iph->version            =       4;
 527         iph->ihl                =       sizeof(struct iphdr) >> 2;
 528         iph->tos                =       (old_iph->tos & IPTOS_TOS_MASK) |
 529                                         IPTOS_PREC_INTERNETCONTROL;
 530         iph->tot_len            =       htons(sizeof(struct iphdr)
 531                                               + sizeof(struct icmphdr)
 532                                               + payload_length);
 533         get_random_bytes(&iph->id, sizeof(iph->id));
 534         iph->frag_off           =       0;
 535         iph->ttl                =       IPDEFTTL;
 536         iph->protocol           =       IPPROTO_ICMP;
 537         iph->daddr              =       old_iph->saddr;
 538         iph->saddr              =       old_iph->daddr;
 539
 540         ip_send_check(iph);
 541
 542         /* ICMP */
 543         icmph->type             =       ICMP_DEST_UNREACH;
 544         icmph->code             =       ICMP_FRAG_NEEDED;
 545         icmph->un.gateway       =       htonl(mtu);
 546         icmph->checksum         =       0;
 547
 548         nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
 549         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
 550                                             payload, payload_length,
 551                                             nskb->csum);
 552         icmph->checksum = csum_fold(nskb->csum);
 553 }
 554
 555 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 556 static bool ipv6_should_icmp(struct sk_buff *skb)
 557 {
 558         struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
 559         int addr_type;
 560         int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
 561         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 562
 563         /* Check source address is valid. */
 564         addr_type = ipv6_addr_type(&old_ipv6h->saddr);
 565         if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
 566                 return false;
 567
 568         /* Don't reply to unspecified addresses. */
 569         if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
 570                 return false;
 571
 572         /* Don't respond to ICMP error messages. */
 573         payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
 574         if (payload_off < 0)
 575                 return false;
 576
 577         if (nexthdr == NEXTHDR_ICMP) {
 578                 u8 icmp_type, *icmp_typep;
 579
 580                 icmp_typep = skb_header_pointer(skb, payload_off +
 581                                                 offsetof(struct icmp6hdr,
 582                                                         icmp6_type),
 583                                                 sizeof(icmp_type), &icmp_type);
 584
 585                 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
 586                         return false;
 587         }
 588
 589         return true;
 590 }
 591
 592 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 593                             unsigned int mtu, unsigned int payload_length)
 594 {
 595         struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
 596         struct icmp6hdr *icmp6h;
 597         u8 *payload;
 598
 599         ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
 600         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
 601         payload = skb_put(nskb, payload_length);
 602
 603         /* IPv6 */
 604         ipv6h->version          =       6;
 605         ipv6h->priority         =       0;
 606         memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
 607         ipv6h->payload_len      =       htons(sizeof(struct icmp6hdr)
 608                                               + payload_length);
 609         ipv6h->nexthdr          =       NEXTHDR_ICMP;
 610         ipv6h->hop_limit        =       IPV6_DEFAULT_HOPLIMIT;
 611         ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
 612         ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
 613
 614         /* ICMPv6 */
 615         icmp6h->icmp6_type      =       ICMPV6_PKT_TOOBIG;
 616         icmp6h->icmp6_code      =       0;
 617         icmp6h->icmp6_cksum     =       0;
 618         icmp6h->icmp6_mtu       =       htonl(mtu);
 619
 620         nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
 621         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
 622                                             payload, payload_length,
 623                                             nskb->csum);
 624         icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 625                                                 sizeof(struct icmp6hdr)
 626                                                 + payload_length,
 627                                                 ipv6h->nexthdr, nskb->csum);
 628 }
 629 #endif /* IPv6 */
 630
 631 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
 632                      struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
 633 {
 634         unsigned int eth_hdr_len = ETH_HLEN;
 635         unsigned int total_length = 0, header_length = 0, payload_length;
 636         struct ethhdr *eh, *old_eh = eth_hdr(skb);
 637         struct sk_buff *nskb;
 638
 639         /* Sanity check */
 640         if (skb->protocol == htons(ETH_P_IP)) {
 641                 if (mtu < IP_MIN_MTU)
 642                         return false;
 643
 644                 if (!ipv4_should_icmp(skb))
 645                         return true;
 646         }
 647 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 648         else if (skb->protocol == htons(ETH_P_IPV6)) {
 649                 if (mtu < IPV6_MIN_MTU)
 650                         return false;
 651
 652                 /*
 653                  * In theory we should do PMTUD on IPv6 multicast messages but
 654                  * we don't have an address to send from so just fragment.
 655                  */
 656                 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
 657                         return false;
 658
 659                 if (!ipv6_should_icmp(skb))
 660                         return true;
 661         }
 662 #endif
 663         else
 664                 return false;
 665
 666         /* Allocate */
 667         if (old_eh->h_proto == htons(ETH_P_8021Q))
 668                 eth_hdr_len = VLAN_ETH_HLEN;
 669
 670         payload_length = skb->len - eth_hdr_len;
 671         if (skb->protocol == htons(ETH_P_IP)) {
 672                 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
 673                 total_length = min_t(unsigned int, header_length +
 674                                                    payload_length, 576);
 675         }
 676 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 677         else {
 678                 header_length = sizeof(struct ipv6hdr) +
 679                                 sizeof(struct icmp6hdr);
 680                 total_length = min_t(unsigned int, header_length +
 681                                                   payload_length, IPV6_MIN_MTU);
 682         }
 683 #endif
 684
 685         payload_length = total_length - header_length;
 686
 687         nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
 688                              payload_length);
 689         if (!nskb)
 690                 return false;
 691
 692         skb_reserve(nskb, NET_IP_ALIGN);
 693
 694         /* Ethernet / VLAN */
 695         eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
 696         memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
 697         memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
 698         nskb->protocol = eh->h_proto = old_eh->h_proto;
 699         if (old_eh->h_proto == htons(ETH_P_8021Q)) {
 700                 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
 701
 702                 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
 703                 vh->h_vlan_encapsulated_proto = skb->protocol;
 704         } else
 705                 vlan_set_tci(nskb, vlan_get_tci(skb));
 706         skb_reset_mac_header(nskb);
 707
 708         /* Protocol */
 709         if (skb->protocol == htons(ETH_P_IP))
 710                 ipv4_build_icmp(skb, nskb, mtu, payload_length);
 711 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 712         else
 713                 ipv6_build_icmp(skb, nskb, mtu, payload_length);
 714 #endif
 715
 716         /*
 717          * Assume that flow based keys are symmetric with respect to input
 718          * and output and use the key that we were going to put on the
 719          * outgoing packet for the fake received packet.  If the keys are
 720          * not symmetric then PMTUD needs to be disabled since we won't have
 721          * any way of synthesizing packets.
 722          */
 723         if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
 724             (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
 725                 OVS_CB(nskb)->tun_id = flow_key;
 726
 727         if (unlikely(compute_ip_summed(nskb, false))) {
 728                 kfree_skb(nskb);
 729                 return false;
 730         }
 731
 732         vport_receive(vport, nskb);
 733
 734         return true;
 735 }
 736
 737 static bool check_mtu(struct sk_buff *skb,
 738                       struct vport *vport,
 739                       const struct tnl_mutable_config *mutable,
 740                       const struct rtable *rt, __be16 *frag_offp)
 741 {
 742         bool df_inherit = mutable->flags & TNL_F_DF_INHERIT;
 743         bool pmtud = mutable->flags & TNL_F_PMTUD;
 744         __be16 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
 745         int mtu = 0;
 746         unsigned int packet_length = skb->len - ETH_HLEN;
 747
 748         /* Allow for one level of tagging in the packet length. */
 749         if (!vlan_tx_tag_present(skb) &&
 750             eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 751                 packet_length -= VLAN_HLEN;
 752
 753         if (pmtud) {
 754                 int vlan_header = 0;
 755
 756                 /* The tag needs to go in packet regardless of where it
 757                  * currently is, so subtract it from the MTU.
 758                  */
 759                 if (vlan_tx_tag_present(skb) ||
 760                     eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 761                         vlan_header = VLAN_HLEN;
 762
 763                 mtu = dst_mtu(&rt_dst(rt))
 764                         - ETH_HLEN
 765                         - mutable->tunnel_hlen
 766                         - vlan_header;
 767         }
 768
 769         if (skb->protocol == htons(ETH_P_IP)) {
 770                 struct iphdr *iph = ip_hdr(skb);
 771
 772                 if (df_inherit)
 773                         frag_off = iph->frag_off & htons(IP_DF);
 774
 775                 if (pmtud && iph->frag_off & htons(IP_DF)) {
 776                         mtu = max(mtu, IP_MIN_MTU);
 777
 778                         if (packet_length > mtu &&
 779                             tnl_frag_needed(vport, mutable, skb, mtu,
 780                                             OVS_CB(skb)->tun_id))
 781                                 return false;
 782                 }
 783         }
 784 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 785         else if (skb->protocol == htons(ETH_P_IPV6)) {
 786                 /* IPv6 requires end hosts to do fragmentation
 787                  * if the packet is above the minimum MTU.
 788                  */
 789                 if (df_inherit && packet_length > IPV6_MIN_MTU)
 790                         frag_off = htons(IP_DF);
 791
 792                 if (pmtud) {
 793                         mtu = max(mtu, IPV6_MIN_MTU);
 794
 795                         if (packet_length > mtu &&
 796                             tnl_frag_needed(vport, mutable, skb, mtu,
 797                                             OVS_CB(skb)->tun_id))
 798                                 return false;
 799                 }
 800         }
 801 #endif
 802
 803         *frag_offp = frag_off;
 804         return true;
 805 }
 806
 807 static void create_tunnel_header(const struct vport *vport,
 808                                  const struct tnl_mutable_config *mutable,
 809                                  const struct rtable *rt, void *header)
 810 {
 811         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 812         struct iphdr *iph = header;
 813
 814         iph->version    = 4;
 815         iph->ihl        = sizeof(struct iphdr) >> 2;
 816         iph->frag_off   = htons(IP_DF);
 817         iph->protocol   = tnl_vport->tnl_ops->ipproto;
 818         iph->tos        = mutable->tos;
 819         iph->daddr      = rt->rt_dst;
 820         iph->saddr      = rt->rt_src;
 821         iph->ttl        = mutable->ttl;
 822         if (!iph->ttl)
 823                 iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
 824
 825         tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
 826 }
 827
 828 static inline void *get_cached_header(const struct tnl_cache *cache)
 829 {
 830         return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
 831 }
 832
 833 static inline bool check_cache_valid(const struct tnl_cache *cache,
 834                                      const struct tnl_mutable_config *mutable)
 835 {
 836         return cache &&
 837 #ifdef NEED_CACHE_TIMEOUT
 838                 time_before(jiffies, cache->expiration) &&
 839 #endif
 840 #ifdef HAVE_RT_GENID
 841                 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
 842 #endif
 843 #ifdef HAVE_HH_SEQ
 844                 rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq &&
 845 #endif
 846                 mutable->seq == cache->mutable_seq &&
 847                 (!is_internal_dev(rt_dst(cache->rt).dev) ||
 848                 (cache->flow && !cache->flow->dead));
 849 }
 850
 851 static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux)
 852 {
 853         struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node);
 854         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
 855         const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
 856
 857         if (cache && !check_cache_valid(cache, mutable) &&
 858             spin_trylock_bh(&tnl_vport->cache_lock)) {
 859                 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
 860                 spin_unlock_bh(&tnl_vport->cache_lock);
 861         }
 862
 863         return 0;
 864 }
 865
 866 static void cache_cleaner(struct work_struct *work)
 867 {
 868         schedule_cache_cleaner();
 869
 870         rcu_read_lock();
 871         tbl_foreach(rcu_dereference(port_table), cache_cleaner_cb, NULL);
 872         rcu_read_unlock();
 873 }
 874
 875 static inline void create_eth_hdr(struct tnl_cache *cache,
 876                                   const struct rtable *rt)
 877 {
 878         void *cache_data = get_cached_header(cache);
 879         int hh_len = rt_dst(rt).hh->hh_len;
 880         int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len;
 881
 882 #ifdef HAVE_HH_SEQ
 883         unsigned hh_seq;
 884
 885         do {
 886                 hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock);
 887                 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
 888         } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq));
 889
 890         cache->hh_seq = hh_seq;
 891 #else
 892         read_lock_bh(&rt_dst(rt).hh->hh_lock);
 893         memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
 894         read_unlock_bh(&rt_dst(rt).hh->hh_lock);
 895 #endif
 896 }
 897
 898 static struct tnl_cache *build_cache(struct vport *vport,
 899                                      const struct tnl_mutable_config *mutable,
 900                                      struct rtable *rt)
 901 {
 902         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 903         struct tnl_cache *cache;
 904         void *cache_data;
 905         int cache_len;
 906
 907         if (!(mutable->flags & TNL_F_HDR_CACHE))
 908                 return NULL;
 909
 910         /*
 911          * If there is no entry in the ARP cache or if this device does not
 912          * support hard header caching just fall back to the IP stack.
 913          */
 914         if (!rt_dst(rt).hh)
 915                 return NULL;
 916
 917         /*
 918          * If lock is contended fall back to directly building the header.
 919          * We're not going to help performance by sitting here spinning.
 920          */
 921         if (!spin_trylock_bh(&tnl_vport->cache_lock))
 922                 return NULL;
 923
 924         cache = cache_dereference(tnl_vport);
 925         if (check_cache_valid(cache, mutable))
 926                 goto unlock;
 927         else
 928                 cache = NULL;
 929
 930         cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen;
 931
 932         cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
 933                         cache_len, GFP_ATOMIC);
 934         if (!cache)
 935                 goto unlock;
 936
 937         cache->len = cache_len;
 938
 939         create_eth_hdr(cache, rt);
 940         cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len;
 941
 942         create_tunnel_header(vport, mutable, rt, cache_data);
 943
 944         cache->mutable_seq = mutable->seq;
 945         cache->rt = rt;
 946 #ifdef NEED_CACHE_TIMEOUT
 947         cache->expiration = jiffies + tnl_vport->cache_exp_interval;
 948 #endif
 949
 950         if (is_internal_dev(rt_dst(rt).dev)) {
 951                 struct sw_flow_key flow_key;
 952                 struct tbl_node *flow_node;
 953                 struct vport *dst_vport;
 954                 struct sk_buff *skb;
 955                 bool is_frag;
 956                 int err;
 957                 int flow_key_len;
 958
 959                 dst_vport = internal_dev_get_vport(rt_dst(rt).dev);
 960                 if (!dst_vport)
 961                         goto done;
 962
 963                 skb = alloc_skb(cache->len, GFP_ATOMIC);
 964                 if (!skb)
 965                         goto done;
 966
 967                 __skb_put(skb, cache->len);
 968                 memcpy(skb->data, get_cached_header(cache), cache->len);
 969
 970                 err = flow_extract(skb, dst_vport->port_no, &flow_key,
 971                                    &flow_key_len, &is_frag);
 972
 973                 consume_skb(skb);
 974                 if (err || is_frag)
 975                         goto done;
 976
 977                 flow_node = tbl_lookup(rcu_dereference(dst_vport->dp->table),
 978                                        &flow_key, flow_key_len,
 979                                        flow_hash(&flow_key, flow_key_len),
 980                                        flow_cmp);
 981                 if (flow_node) {
 982                         struct sw_flow *flow = flow_cast(flow_node);
 983
 984                         cache->flow = flow;
 985                         flow_hold(flow);
 986                 }
 987         }
 988
 989 done:
 990         assign_cache_rcu(vport, cache);
 991
 992 unlock:
 993         spin_unlock_bh(&tnl_vport->cache_lock);
 994
 995         return cache;
 996 }
 997
 998 static struct rtable *find_route(struct vport *vport,
 999                                  const struct tnl_mutable_config *mutable,
1000                                  u8 tos, struct tnl_cache **cache)
1001 {
1002         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1003         struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
1004
1005         *cache = NULL;
1006         tos = RT_TOS(tos);
1007
1008         if (likely(tos == mutable->tos && check_cache_valid(cur_cache, mutable))) {
1009                 *cache = cur_cache;
1010                 return cur_cache->rt;
1011         } else {
1012                 struct rtable *rt;
1013 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
1014                 struct flowi fl = { .nl_u = { .ip4_u =
1015                                               { .daddr = mutable->daddr,
1016                                                 .saddr = mutable->saddr,
1017                                                 .tos = tos } },
1018                                     .proto = tnl_vport->tnl_ops->ipproto };
1019
1020                 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
1021                         return NULL;
1022 #else
1023                 struct flowi4 fl = { .daddr = mutable->daddr,
1024                                      .saddr = mutable->saddr,
1025                                      .flowi4_tos = tos,
1026                                      .flowi4_proto = tnl_vport->tnl_ops->ipproto };
1027
1028                 rt = ip_route_output_key(&init_net, &fl);
1029                 if (IS_ERR(rt))
1030                         return NULL;
1031 #endif
1032
1033                 if (likely(tos == mutable->tos))
1034                         *cache = build_cache(vport, mutable, rt);
1035
1036                 return rt;
1037         }
1038 }
1039
1040 static inline bool need_linearize(const struct sk_buff *skb)
1041 {
1042         int i;
1043
1044         if (unlikely(skb_shinfo(skb)->frag_list))
1045                 return true;
1046
1047         /*
1048          * Generally speaking we should linearize if there are paged frags.
1049          * However, if all of the refcounts are 1 we know nobody else can
1050          * change them from underneath us and we can skip the linearization.
1051          */
1052         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1053                 if (unlikely(page_count(skb_shinfo(skb)->frags[i].page) > 1))
1054                         return true;
1055
1056         return false;
1057 }
1058
1059 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1060                                        const struct tnl_mutable_config *mutable,
1061                                        const struct rtable *rt)
1062 {
1063         int min_headroom;
1064         int err;
1065
1066         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1067                         + mutable->tunnel_hlen
1068                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1069
1070         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
1071                 int head_delta = SKB_DATA_ALIGN(min_headroom -
1072                                                 skb_headroom(skb) +
1073                                                 16);
1074                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
1075                                         0, GFP_ATOMIC);
1076                 if (unlikely(err))
1077                         goto error_free;
1078         }
1079
1080         forward_ip_summed(skb, true);
1081
1082         if (skb_is_gso(skb)) {
1083                 struct sk_buff *nskb;
1084
1085                 nskb = skb_gso_segment(skb, 0);
1086                 if (IS_ERR(nskb)) {
1087                         kfree_skb(skb);
1088                         err = PTR_ERR(nskb);
1089                         goto error;
1090                 }
1091
1092                 consume_skb(skb);
1093                 skb = nskb;
1094         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
1095                 /* Pages aren't locked and could change at any time.
1096                  * If this happens after we compute the checksum, the
1097                  * checksum will be wrong.  We linearize now to avoid
1098                  * this problem.
1099                  */
1100                 if (unlikely(need_linearize(skb))) {
1101                         err = __skb_linearize(skb);
1102                         if (unlikely(err))
1103                                 goto error_free;
1104                 }
1105
1106                 err = skb_checksum_help(skb);
1107                 if (unlikely(err))
1108                         goto error_free;
1109         }
1110
1111         set_ip_summed(skb, OVS_CSUM_NONE);
1112
1113         return skb;
1114
1115 error_free:
1116         kfree_skb(skb);
1117 error:
1118         return ERR_PTR(err);
1119 }
1120
1121 static int send_frags(struct sk_buff *skb,
1122                       const struct tnl_mutable_config *mutable)
1123 {
1124         int sent_len;
1125
1126         sent_len = 0;
1127         while (skb) {
1128                 struct sk_buff *next = skb->next;
1129                 int frag_len = skb->len - mutable->tunnel_hlen;
1130                 int err;
1131
1132                 skb->next = NULL;
1133                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1134
1135                 err = ip_local_out(skb);
1136                 skb = next;
1137                 if (unlikely(net_xmit_eval(err)))
1138                         goto free_frags;
1139                 sent_len += frag_len;
1140         }
1141
1142         return sent_len;
1143
1144 free_frags:
1145         /*
1146          * There's no point in continuing to send fragments once one has been
1147          * dropped so just free the rest.  This may help improve the congestion
1148          * that caused the first packet to be dropped.
1149          */
1150         tnl_free_linked_skbs(skb);
1151         return sent_len;
1152 }
1153
1154 int tnl_send(struct vport *vport, struct sk_buff *skb)
1155 {
1156         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1157         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1158
1159         enum vport_err_type err = VPORT_E_TX_ERROR;
1160         struct rtable *rt;
1161         struct dst_entry *unattached_dst = NULL;
1162         struct tnl_cache *cache;
1163         int sent_len = 0;
1164         __be16 frag_off = 0;
1165         u8 ttl;
1166         u8 inner_tos;
1167         u8 tos;
1168
1169         /* Validate the protocol headers before we try to use them. */
1170         if (skb->protocol == htons(ETH_P_8021Q) &&
1171             !vlan_tx_tag_present(skb)) {
1172                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1173                         goto error_free;
1174
1175                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1176                 skb_set_network_header(skb, VLAN_ETH_HLEN);
1177         }
1178
1179         if (skb->protocol == htons(ETH_P_IP)) {
1180                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1181                     + sizeof(struct iphdr))))
1182                         skb->protocol = 0;
1183         }
1184 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1185         else if (skb->protocol == htons(ETH_P_IPV6)) {
1186                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1187                     + sizeof(struct ipv6hdr))))
1188                         skb->protocol = 0;
1189         }
1190 #endif
1191
1192         /* ToS */
1193         if (skb->protocol == htons(ETH_P_IP))
1194                 inner_tos = ip_hdr(skb)->tos;
1195 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1196         else if (skb->protocol == htons(ETH_P_IPV6))
1197                 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1198 #endif
1199         else
1200                 inner_tos = 0;
1201
1202         if (mutable->flags & TNL_F_TOS_INHERIT)
1203                 tos = inner_tos;
1204         else
1205                 tos = mutable->tos;
1206
1207         tos = INET_ECN_encapsulate(tos, inner_tos);
1208
1209         /* Route lookup */
1210         rt = find_route(vport, mutable, tos, &cache);
1211         if (unlikely(!rt))
1212                 goto error_free;
1213         if (unlikely(!cache))
1214                 unattached_dst = &rt_dst(rt);
1215
1216         /* Reset SKB */
1217         nf_reset(skb);
1218         secpath_reset(skb);
1219         skb_dst_drop(skb);
1220         skb_clear_rxhash(skb);
1221
1222         /* Offloading */
1223         skb = handle_offloads(skb, mutable, rt);
1224         if (IS_ERR(skb))
1225                 goto error;
1226
1227         /* MTU */
1228         if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1229                 err = VPORT_E_TX_DROPPED;
1230                 goto error_free;
1231         }
1232
1233         /*
1234          * If we are over the MTU, allow the IP stack to handle fragmentation.
1235          * Fragmentation is a slow path anyways.
1236          */
1237         if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1238                      cache)) {
1239                 unattached_dst = &rt_dst(rt);
1240                 dst_hold(unattached_dst);
1241                 cache = NULL;
1242         }
1243
1244         /* TTL */
1245         ttl = mutable->ttl;
1246         if (!ttl)
1247                 ttl = ip4_dst_hoplimit(&rt_dst(rt));
1248
1249         if (mutable->flags & TNL_F_TTL_INHERIT) {
1250                 if (skb->protocol == htons(ETH_P_IP))
1251                         ttl = ip_hdr(skb)->ttl;
1252 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1253                 else if (skb->protocol == htons(ETH_P_IPV6))
1254                         ttl = ipv6_hdr(skb)->hop_limit;
1255 #endif
1256         }
1257
1258         while (skb) {
1259                 struct iphdr *iph;
1260                 struct sk_buff *next_skb = skb->next;
1261                 skb->next = NULL;
1262
1263                 if (unlikely(vlan_deaccel_tag(skb)))
1264                         goto next;
1265
1266                 if (likely(cache)) {
1267                         skb_push(skb, cache->len);
1268                         memcpy(skb->data, get_cached_header(cache), cache->len);
1269                         skb_reset_mac_header(skb);
1270                         skb_set_network_header(skb, rt_dst(rt).hh->hh_len);
1271
1272                 } else {
1273                         skb_push(skb, mutable->tunnel_hlen);
1274                         create_tunnel_header(vport, mutable, rt, skb->data);
1275                         skb_reset_network_header(skb);
1276
1277                         if (next_skb)
1278                                 skb_dst_set(skb, dst_clone(unattached_dst));
1279                         else {
1280                                 skb_dst_set(skb, unattached_dst);
1281                                 unattached_dst = NULL;
1282                         }
1283                 }
1284                 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1285
1286                 iph = ip_hdr(skb);
1287                 iph->tos = tos;
1288                 iph->ttl = ttl;
1289                 iph->frag_off = frag_off;
1290                 ip_select_ident(iph, &rt_dst(rt), NULL);
1291
1292                 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1293                 if (unlikely(!skb))
1294                         goto next;
1295
1296                 if (likely(cache)) {
1297                         int orig_len = skb->len - cache->len;
1298                         struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1299
1300                         skb->protocol = htons(ETH_P_IP);
1301                         iph = ip_hdr(skb);
1302                         iph->tot_len = htons(skb->len - skb_network_offset(skb));
1303                         ip_send_check(iph);
1304
1305                         if (cache_vport) {
1306                                 if (unlikely(compute_ip_summed(skb, true))) {
1307                                         kfree_skb(skb);
1308                                         goto next;
1309                                 }
1310
1311                                 OVS_CB(skb)->flow = cache->flow;
1312                                 vport_receive(cache_vport, skb);
1313                                 sent_len += orig_len;
1314                         } else {
1315                                 int xmit_err;
1316
1317                                 skb->dev = rt_dst(rt).dev;
1318                                 xmit_err = dev_queue_xmit(skb);
1319
1320                                 if (likely(net_xmit_eval(xmit_err) == 0))
1321                                         sent_len += orig_len;
1322                         }
1323                 } else
1324                         sent_len += send_frags(skb, mutable);
1325
1326 next:
1327                 skb = next_skb;
1328         }
1329
1330         if (unlikely(sent_len == 0))
1331                 vport_record_error(vport, VPORT_E_TX_DROPPED);
1332
1333         goto out;
1334
1335 error_free:
1336         tnl_free_linked_skbs(skb);
1337 error:
1338         vport_record_error(vport, err);
1339 out:
1340         dst_release(unattached_dst);
1341         return sent_len;
1342 }
1343
1344 static const struct nla_policy tnl_policy[ODP_TUNNEL_ATTR_MAX + 1] = {
1345         [ODP_TUNNEL_ATTR_FLAGS]    = { .type = NLA_U32 },
1346         [ODP_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1347         [ODP_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1348         [ODP_TUNNEL_ATTR_OUT_KEY]  = { .type = NLA_U64 },
1349         [ODP_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
1350         [ODP_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
1351         [ODP_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
1352 };
1353
1354 /* Sets ODP_TUNNEL_ATTR_* fields in 'mutable', which must initially be zeroed. */
1355 static int tnl_set_config(struct nlattr *options, const struct tnl_ops *tnl_ops,
1356                           const struct vport *cur_vport,
1357                           struct tnl_mutable_config *mutable)
1358 {
1359         const struct vport *old_vport;
1360         const struct tnl_mutable_config *old_mutable;
1361         struct nlattr *a[ODP_TUNNEL_ATTR_MAX + 1];
1362         int err;
1363
1364         if (!options)
1365                 return -EINVAL;
1366
1367         err = nla_parse_nested(a, ODP_TUNNEL_ATTR_MAX, options, tnl_policy);
1368         if (err)
1369                 return err;
1370
1371         if (!a[ODP_TUNNEL_ATTR_FLAGS] || !a[ODP_TUNNEL_ATTR_DST_IPV4])
1372                 return -EINVAL;
1373
1374         mutable->flags = nla_get_u32(a[ODP_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1375
1376         if (a[ODP_TUNNEL_ATTR_SRC_IPV4])
1377                 mutable->saddr = nla_get_be32(a[ODP_TUNNEL_ATTR_SRC_IPV4]);
1378         mutable->daddr = nla_get_be32(a[ODP_TUNNEL_ATTR_DST_IPV4]);
1379
1380         if (a[ODP_TUNNEL_ATTR_TOS]) {
1381                 mutable->tos = nla_get_u8(a[ODP_TUNNEL_ATTR_TOS]);
1382                 if (mutable->tos != RT_TOS(mutable->tos))
1383                         return -EINVAL;
1384         }
1385
1386         if (a[ODP_TUNNEL_ATTR_TTL])
1387                 mutable->ttl = nla_get_u8(a[ODP_TUNNEL_ATTR_TTL]);
1388
1389         mutable->tunnel_type = tnl_ops->tunnel_type;
1390         if (!a[ODP_TUNNEL_ATTR_IN_KEY]) {
1391                 mutable->tunnel_type |= TNL_T_KEY_MATCH;
1392                 mutable->flags |= TNL_F_IN_KEY_MATCH;
1393         } else {
1394                 mutable->tunnel_type |= TNL_T_KEY_EXACT;
1395                 mutable->in_key = nla_get_be64(a[ODP_TUNNEL_ATTR_IN_KEY]);
1396         }
1397
1398         if (!a[ODP_TUNNEL_ATTR_OUT_KEY])
1399                 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1400         else
1401                 mutable->out_key = nla_get_be64(a[ODP_TUNNEL_ATTR_OUT_KEY]);
1402
1403         mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1404         if (mutable->tunnel_hlen < 0)
1405                 return mutable->tunnel_hlen;
1406
1407         mutable->tunnel_hlen += sizeof(struct iphdr);
1408
1409         old_vport = tnl_find_port(mutable->saddr, mutable->daddr,
1410                                   mutable->in_key, mutable->tunnel_type,
1411                                   &old_mutable);
1412
1413         if (old_vport && old_vport != cur_vport)
1414                 return -EEXIST;
1415
1416         return 0;
1417 }
1418
1419 struct vport *tnl_create(const struct vport_parms *parms,
1420                          const struct vport_ops *vport_ops,
1421                          const struct tnl_ops *tnl_ops)
1422 {
1423         struct vport *vport;
1424         struct tnl_vport *tnl_vport;
1425         struct tnl_mutable_config *mutable;
1426         int initial_frag_id;
1427         int err;
1428
1429         vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1430         if (IS_ERR(vport)) {
1431                 err = PTR_ERR(vport);
1432                 goto error;
1433         }
1434
1435         tnl_vport = tnl_vport_priv(vport);
1436
1437         strcpy(tnl_vport->name, parms->name);
1438         tnl_vport->tnl_ops = tnl_ops;
1439
1440         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1441         if (!mutable) {
1442                 err = -ENOMEM;
1443                 goto error_free_vport;
1444         }
1445
1446         vport_gen_rand_ether_addr(mutable->eth_addr);
1447
1448         get_random_bytes(&initial_frag_id, sizeof(int));
1449         atomic_set(&tnl_vport->frag_id, initial_frag_id);
1450
1451         err = tnl_set_config(parms->options, tnl_ops, NULL, mutable);
1452         if (err)
1453                 goto error_free_mutable;
1454
1455         spin_lock_init(&tnl_vport->cache_lock);
1456
1457 #ifdef NEED_CACHE_TIMEOUT
1458         tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1459                                        (net_random() % (MAX_CACHE_EXP / 2));
1460 #endif
1461
1462         rcu_assign_pointer(tnl_vport->mutable, mutable);
1463
1464         err = add_port(vport);
1465         if (err)
1466                 goto error_free_mutable;
1467
1468         return vport;
1469
1470 error_free_mutable:
1471         kfree(mutable);
1472 error_free_vport:
1473         vport_free(vport);
1474 error:
1475         return ERR_PTR(err);
1476 }
1477
1478 int tnl_set_options(struct vport *vport, struct nlattr *options)
1479 {
1480         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1481         const struct tnl_mutable_config *old_mutable;
1482         struct tnl_mutable_config *mutable;
1483         int err;
1484
1485         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1486         if (!mutable) {
1487                 err = -ENOMEM;
1488                 goto error;
1489         }
1490
1491         /* Copy fields whose values should be retained. */
1492         old_mutable = rtnl_dereference(tnl_vport->mutable);
1493         mutable->seq = old_mutable->seq + 1;
1494         memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1495
1496         /* Parse the others configured by userspace. */
1497         err = tnl_set_config(options, tnl_vport->tnl_ops, vport, mutable);
1498         if (err)
1499                 goto error_free;
1500
1501         err = move_port(vport, mutable);
1502         if (err)
1503                 goto error_free;
1504
1505         return 0;
1506
1507 error_free:
1508         kfree(mutable);
1509 error:
1510         return err;
1511 }
1512
1513 int tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1514 {
1515         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1516         const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1517
1518         NLA_PUT_U32(skb, ODP_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC);
1519         NLA_PUT_BE32(skb, ODP_TUNNEL_ATTR_DST_IPV4, mutable->daddr);
1520
1521         if (!(mutable->flags & TNL_F_IN_KEY_MATCH))
1522                 NLA_PUT_BE64(skb, ODP_TUNNEL_ATTR_IN_KEY, mutable->in_key);
1523         if (!(mutable->flags & TNL_F_OUT_KEY_ACTION))
1524                 NLA_PUT_BE64(skb, ODP_TUNNEL_ATTR_OUT_KEY, mutable->out_key);
1525         if (mutable->saddr)
1526                 NLA_PUT_BE32(skb, ODP_TUNNEL_ATTR_SRC_IPV4, mutable->saddr);
1527         if (mutable->tos)
1528                 NLA_PUT_U8(skb, ODP_TUNNEL_ATTR_TOS, mutable->tos);
1529         if (mutable->ttl)
1530                 NLA_PUT_U8(skb, ODP_TUNNEL_ATTR_TTL, mutable->ttl);
1531
1532         return 0;
1533
1534 nla_put_failure:
1535         return -EMSGSIZE;
1536 }
1537
1538 static void free_port_rcu(struct rcu_head *rcu)
1539 {
1540         struct tnl_vport *tnl_vport = container_of(rcu,
1541                                                    struct tnl_vport, rcu);
1542
1543         free_cache((struct tnl_cache __force *)tnl_vport->cache);
1544         kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1545         vport_free(tnl_vport_to_vport(tnl_vport));
1546 }
1547
1548 int tnl_destroy(struct vport *vport)
1549 {
1550         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1551         const struct tnl_mutable_config *mutable, *old_mutable;
1552
1553         mutable = rtnl_dereference(tnl_vport->mutable);
1554
1555         if (vport == tnl_find_port(mutable->saddr, mutable->daddr,
1556                                    mutable->in_key, mutable->tunnel_type,
1557                                    &old_mutable))
1558                 del_port(vport);
1559
1560         call_rcu(&tnl_vport->rcu, free_port_rcu);
1561
1562         return 0;
1563 }
1564
1565 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1566 {
1567         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1568         struct tnl_mutable_config *mutable;
1569
1570         mutable = kmemdup(rtnl_dereference(tnl_vport->mutable),
1571                           sizeof(struct tnl_mutable_config), GFP_KERNEL);
1572         if (!mutable)
1573                 return -ENOMEM;
1574
1575         memcpy(mutable->eth_addr, addr, ETH_ALEN);
1576         assign_config_rcu(vport, mutable);
1577
1578         return 0;
1579 }
1580
1581 const char *tnl_get_name(const struct vport *vport)
1582 {
1583         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1584         return tnl_vport->name;
1585 }
1586
1587 const unsigned char *tnl_get_addr(const struct vport *vport)
1588 {
1589         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1590         return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1591 }
1592
1593 void tnl_free_linked_skbs(struct sk_buff *skb)
1594 {
1595         while (skb) {
1596                 struct sk_buff *next = skb->next;
1597                 kfree_skb(skb);
1598                 skb = next;
1599         }
1600 }