datapath/tunnel.c

   1 /*
   2  * Copyright (c) 2010 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 "datapath.h"
  33 #include "table.h"
  34 #include "tunnel.h"
  35 #include "vport.h"
  36 #include "vport-generic.h"
  37 #include "vport-internal_dev.h"
  38
  39 #ifdef NEED_CACHE_TIMEOUT
  40 /*
  41  * On kernels where we can't quickly detect changes in the rest of the system
  42  * we use an expiration time to invalidate the cache.  A shorter expiration
  43  * reduces the length of time that we may potentially blackhole packets while
  44  * a longer time increases performance by reducing the frequency that the
  45  * cache needs to be rebuilt.  A variety of factors may cause the cache to be
  46  * invalidated before the expiration time but this is the maximum.  The time
  47  * is expressed in jiffies.
  48  */
  49 #define MAX_CACHE_EXP HZ
  50 #endif
  51
  52 /*
  53  * Interval to check for and remove caches that are no longer valid.  Caches
  54  * are checked for validity before they are used for packet encapsulation and
  55  * old caches are removed at that time.  However, if no packets are sent through
  56  * the tunnel then the cache will never be destroyed.  Since it holds
  57  * references to a number of system objects, the cache will continue to use
  58  * system resources by not allowing those objects to be destroyed.  The cache
  59  * cleaner is periodically run to free invalid caches.  It does not
  60  * significantly affect system performance.  A lower interval will release
  61  * resources faster but will itself consume resources by requiring more frequent
  62  * checks.  A longer interval may result in messages being printed to the kernel
  63  * message buffer about unreleased resources.  The interval is expressed in
  64  * jiffies.
  65  */
  66 #define CACHE_CLEANER_INTERVAL (5 * HZ)
  67
  68 #define CACHE_DATA_ALIGN 16
  69
  70 /* Protected by RCU. */
  71 static struct tbl *port_table;
  72
  73 static void cache_cleaner(struct work_struct *work);
  74 DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
  75
  76 /*
  77  * These are just used as an optimization: they don't require any kind of
  78  * synchronization because we could have just as easily read the value before
  79  * the port change happened.
  80  */
  81 static unsigned int key_local_remote_ports;
  82 static unsigned int key_remote_ports;
  83 static unsigned int local_remote_ports;
  84 static unsigned int remote_ports;
  85
  86 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  87 #define rt_dst(rt) (rt->dst)
  88 #else
  89 #define rt_dst(rt) (rt->u.dst)
  90 #endif
  91
  92 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
  93 {
  94         return vport_from_priv(tnl_vport);
  95 }
  96
  97 static inline struct tnl_vport *tnl_vport_table_cast(const struct tbl_node *node)
  98 {
  99         return container_of(node, struct tnl_vport, tbl_node);
 100 }
 101
 102 static inline void schedule_cache_cleaner(void)
 103 {
 104         schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
 105 }
 106
 107 static void free_cache(struct tnl_cache *cache)
 108 {
 109         if (!cache)
 110                 return;
 111
 112         flow_put(cache->flow);
 113         ip_rt_put(cache->rt);
 114         kfree(cache);
 115 }
 116
 117 static void free_config_rcu(struct rcu_head *rcu)
 118 {
 119         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
 120         kfree(c);
 121 }
 122
 123 static void free_cache_rcu(struct rcu_head *rcu)
 124 {
 125         struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
 126         free_cache(c);
 127 }
 128
 129 static void assign_config_rcu(struct vport *vport,
 130                               struct tnl_mutable_config *new_config)
 131 {
 132         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 133         struct tnl_mutable_config *old_config;
 134
 135         old_config = tnl_vport->mutable;
 136         rcu_assign_pointer(tnl_vport->mutable, new_config);
 137         call_rcu(&old_config->rcu, free_config_rcu);
 138 }
 139
 140 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
 141 {
 142         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 143         struct tnl_cache *old_cache;
 144
 145         old_cache = tnl_vport->cache;
 146         rcu_assign_pointer(tnl_vport->cache, new_cache);
 147
 148         if (old_cache)
 149                 call_rcu(&old_cache->rcu, free_cache_rcu);
 150 }
 151
 152 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 153 {
 154         if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) {
 155                 if (mutable->port_config.saddr)
 156                         return &local_remote_ports;
 157                 else
 158                         return &remote_ports;
 159         } else {
 160                 if (mutable->port_config.saddr)
 161                         return &key_local_remote_ports;
 162                 else
 163                         return &key_remote_ports;
 164         }
 165 }
 166
 167 enum lookup_key {
 168         LOOKUP_TUNNEL_TYPE      = 0,
 169         LOOKUP_SADDR            = 1,
 170         LOOKUP_DADDR            = 2,
 171         LOOKUP_KEY              = 3,
 172 };
 173
 174 struct port_lookup_key {
 175         u32 vals[4];                    /* Contains enum lookup_key keys. */
 176         const struct tnl_mutable_config *mutable;
 177 };
 178
 179 /*
 180  * Modifies 'target' to store the rcu_dereferenced pointer that was used to do
 181  * the comparision.
 182  */
 183 static int port_cmp(const struct tbl_node *node, void *target)
 184 {
 185         const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node);
 186         struct port_lookup_key *lookup = target;
 187
 188         lookup->mutable = rcu_dereference(tnl_vport->mutable);
 189
 190         return (lookup->mutable->tunnel_type == lookup->vals[LOOKUP_TUNNEL_TYPE]) &&
 191                lookup->mutable->port_config.daddr == lookup->vals[LOOKUP_DADDR] &&
 192                lookup->mutable->port_config.in_key == lookup->vals[LOOKUP_KEY] &&
 193                lookup->mutable->port_config.saddr == lookup->vals[LOOKUP_SADDR];
 194 }
 195
 196 static u32 port_hash(struct port_lookup_key *lookup)
 197 {
 198         return jhash2(lookup->vals, ARRAY_SIZE(lookup->vals), 0);
 199 }
 200
 201 static u32 mutable_hash(const struct tnl_mutable_config *mutable)
 202 {
 203         struct port_lookup_key lookup;
 204
 205         lookup.vals[LOOKUP_SADDR] = mutable->port_config.saddr;
 206         lookup.vals[LOOKUP_DADDR] = mutable->port_config.daddr;
 207         lookup.vals[LOOKUP_KEY] = mutable->port_config.in_key;
 208         lookup.vals[LOOKUP_TUNNEL_TYPE] = mutable->tunnel_type;
 209
 210         return port_hash(&lookup);
 211 }
 212
 213 static void check_table_empty(void)
 214 {
 215         if (tbl_count(port_table) == 0) {
 216                 struct tbl *old_table = port_table;
 217
 218                 cancel_delayed_work_sync(&cache_cleaner_wq);
 219                 rcu_assign_pointer(port_table, NULL);
 220                 tbl_deferred_destroy(old_table, NULL);
 221         }
 222 }
 223
 224 static int add_port(struct vport *vport)
 225 {
 226         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 227         int err;
 228
 229         if (!port_table) {
 230                 struct tbl *new_table;
 231
 232                 new_table = tbl_create(0);
 233                 if (!new_table)
 234                         return -ENOMEM;
 235
 236                 rcu_assign_pointer(port_table, new_table);
 237                 schedule_cache_cleaner();
 238
 239         } else if (tbl_count(port_table) > tbl_n_buckets(port_table)) {
 240                 struct tbl *old_table = port_table;
 241                 struct tbl *new_table;
 242
 243                 new_table = tbl_expand(old_table);
 244                 if (IS_ERR(new_table))
 245                         return PTR_ERR(new_table);
 246
 247                 rcu_assign_pointer(port_table, new_table);
 248                 tbl_deferred_destroy(old_table, NULL);
 249         }
 250
 251         err = tbl_insert(port_table, &tnl_vport->tbl_node, mutable_hash(tnl_vport->mutable));
 252         if (err) {
 253                 check_table_empty();
 254                 return err;
 255         }
 256
 257         (*find_port_pool(tnl_vport->mutable))++;
 258
 259         return 0;
 260 }
 261
 262 static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable)
 263 {
 264         int err;
 265         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 266         u32 hash;
 267
 268         hash = mutable_hash(new_mutable);
 269         if (hash == tnl_vport->tbl_node.hash)
 270                 goto table_updated;
 271
 272         /*
 273          * Ideally we should make this move atomic to avoid having gaps in
 274          * finding tunnels or the possibility of failure.  However, if we do
 275          * find a tunnel it will always be consistent.
 276          */
 277         err = tbl_remove(port_table, &tnl_vport->tbl_node);
 278         if (err)
 279                 return err;
 280
 281         err = tbl_insert(port_table, &tnl_vport->tbl_node, hash);
 282         if (err) {
 283                 check_table_empty();
 284                 return err;
 285         }
 286
 287 table_updated:
 288         assign_config_rcu(vport, new_mutable);
 289
 290         return 0;
 291 }
 292
 293 static int del_port(struct vport *vport)
 294 {
 295         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 296         int err;
 297
 298         err = tbl_remove(port_table, &tnl_vport->tbl_node);
 299         if (err)
 300                 return err;
 301
 302         check_table_empty();
 303         (*find_port_pool(tnl_vport->mutable))--;
 304
 305         return 0;
 306 }
 307
 308 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be32 key,
 309                             int tunnel_type,
 310                             const struct tnl_mutable_config **mutable)
 311 {
 312         struct port_lookup_key lookup;
 313         struct tbl *table = rcu_dereference(port_table);
 314         struct tbl_node *tbl_node;
 315
 316         if (unlikely(!table))
 317                 return NULL;
 318
 319         lookup.vals[LOOKUP_SADDR] = saddr;
 320         lookup.vals[LOOKUP_DADDR] = daddr;
 321
 322         if (tunnel_type & TNL_T_KEY_EXACT) {
 323                 lookup.vals[LOOKUP_KEY] = key;
 324                 lookup.vals[LOOKUP_TUNNEL_TYPE] = tunnel_type & ~TNL_T_KEY_MATCH;
 325
 326                 if (key_local_remote_ports) {
 327                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
 328                         if (tbl_node)
 329                                 goto found;
 330                 }
 331
 332                 if (key_remote_ports) {
 333                         lookup.vals[LOOKUP_SADDR] = 0;
 334
 335                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
 336                         if (tbl_node)
 337                                 goto found;
 338
 339                         lookup.vals[LOOKUP_SADDR] = saddr;
 340                 }
 341         }
 342
 343         if (tunnel_type & TNL_T_KEY_MATCH) {
 344                 lookup.vals[LOOKUP_KEY] = 0;
 345                 lookup.vals[LOOKUP_TUNNEL_TYPE] = tunnel_type & ~TNL_T_KEY_EXACT;
 346
 347                 if (local_remote_ports) {
 348                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
 349                         if (tbl_node)
 350                                 goto found;
 351                 }
 352
 353                 if (remote_ports) {
 354                         lookup.vals[LOOKUP_SADDR] = 0;
 355
 356                         tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
 357                         if (tbl_node)
 358                                 goto found;
 359                 }
 360         }
 361
 362         return NULL;
 363
 364 found:
 365         *mutable = lookup.mutable;
 366         return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node));
 367 }
 368
 369 static inline void ecn_decapsulate(struct sk_buff *skb)
 370 {
 371         u8 tos = ip_hdr(skb)->tos;
 372
 373         if (INET_ECN_is_ce(tos)) {
 374                 __be16 protocol = skb->protocol;
 375                 unsigned int nw_header = skb_network_offset(skb);
 376
 377                 if (skb->protocol == htons(ETH_P_8021Q)) {
 378                         if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 379                                 return;
 380
 381                         protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 382                         nw_header += VLAN_HLEN;
 383                 }
 384
 385                 if (protocol == htons(ETH_P_IP)) {
 386                         if (unlikely(!pskb_may_pull(skb, nw_header
 387                             + sizeof(struct iphdr))))
 388                                 return;
 389
 390                         IP_ECN_set_ce((struct iphdr *)(skb->data + nw_header));
 391                 }
 392 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 393                 else if (protocol == htons(ETH_P_IPV6)) {
 394                         if (unlikely(!pskb_may_pull(skb, nw_header
 395                             + sizeof(struct ipv6hdr))))
 396                                 return;
 397
 398                         IP6_ECN_set_ce((struct ipv6hdr *)(skb->data + nw_header));
 399                 }
 400 #endif
 401         }
 402 }
 403
 404 /* Called with rcu_read_lock. */
 405 void tnl_rcv(struct vport *vport, struct sk_buff *skb)
 406 {
 407         skb->pkt_type = PACKET_HOST;
 408         skb->protocol = eth_type_trans(skb, skb->dev);
 409
 410         skb_dst_drop(skb);
 411         nf_reset(skb);
 412         secpath_reset(skb);
 413         skb_reset_network_header(skb);
 414
 415         ecn_decapsulate(skb);
 416
 417         skb_push(skb, ETH_HLEN);
 418         compute_ip_summed(skb, false);
 419
 420         vport_receive(vport, skb);
 421 }
 422
 423 static bool check_ipv4_address(__be32 addr)
 424 {
 425         if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
 426             || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
 427                 return false;
 428
 429         return true;
 430 }
 431
 432 static bool ipv4_should_icmp(struct sk_buff *skb)
 433 {
 434         struct iphdr *old_iph = ip_hdr(skb);
 435
 436         /* Don't respond to L2 broadcast. */
 437         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
 438                 return false;
 439
 440         /* Don't respond to L3 broadcast or invalid addresses. */
 441         if (!check_ipv4_address(old_iph->daddr) ||
 442             !check_ipv4_address(old_iph->saddr))
 443                 return false;
 444
 445         /* Only respond to the first fragment. */
 446         if (old_iph->frag_off & htons(IP_OFFSET))
 447                 return false;
 448
 449         /* Don't respond to ICMP error messages. */
 450         if (old_iph->protocol == IPPROTO_ICMP) {
 451                 u8 icmp_type, *icmp_typep;
 452
 453                 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
 454                                                 (old_iph->ihl << 2) +
 455                                                 offsetof(struct icmphdr, type) -
 456                                                 skb->data, sizeof(icmp_type),
 457                                                 &icmp_type);
 458
 459                 if (!icmp_typep)
 460                         return false;
 461
 462                 if (*icmp_typep > NR_ICMP_TYPES
 463                         || (*icmp_typep <= ICMP_PARAMETERPROB
 464                                 && *icmp_typep != ICMP_ECHOREPLY
 465                                 && *icmp_typep != ICMP_ECHO))
 466                         return false;
 467         }
 468
 469         return true;
 470 }
 471
 472 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 473                             unsigned int mtu, unsigned int payload_length)
 474 {
 475         struct iphdr *iph, *old_iph = ip_hdr(skb);
 476         struct icmphdr *icmph;
 477         u8 *payload;
 478
 479         iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
 480         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 481         payload = skb_put(nskb, payload_length);
 482
 483         /* IP */
 484         iph->version            =       4;
 485         iph->ihl                =       sizeof(struct iphdr) >> 2;
 486         iph->tos                =       (old_iph->tos & IPTOS_TOS_MASK) |
 487                                         IPTOS_PREC_INTERNETCONTROL;
 488         iph->tot_len            =       htons(sizeof(struct iphdr)
 489                                               + sizeof(struct icmphdr)
 490                                               + payload_length);
 491         get_random_bytes(&iph->id, sizeof(iph->id));
 492         iph->frag_off           =       0;
 493         iph->ttl                =       IPDEFTTL;
 494         iph->protocol           =       IPPROTO_ICMP;
 495         iph->daddr              =       old_iph->saddr;
 496         iph->saddr              =       old_iph->daddr;
 497
 498         ip_send_check(iph);
 499
 500         /* ICMP */
 501         icmph->type             =       ICMP_DEST_UNREACH;
 502         icmph->code             =       ICMP_FRAG_NEEDED;
 503         icmph->un.gateway       =       htonl(mtu);
 504         icmph->checksum         =       0;
 505
 506         nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
 507         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
 508                                             payload, payload_length,
 509                                             nskb->csum);
 510         icmph->checksum = csum_fold(nskb->csum);
 511 }
 512
 513 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 514 static bool ipv6_should_icmp(struct sk_buff *skb)
 515 {
 516         struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
 517         int addr_type;
 518         int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
 519         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 520
 521         /* Check source address is valid. */
 522         addr_type = ipv6_addr_type(&old_ipv6h->saddr);
 523         if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
 524                 return false;
 525
 526         /* Don't reply to unspecified addresses. */
 527         if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
 528                 return false;
 529
 530         /* Don't respond to ICMP error messages. */
 531         payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
 532         if (payload_off < 0)
 533                 return false;
 534
 535         if (nexthdr == NEXTHDR_ICMP) {
 536                 u8 icmp_type, *icmp_typep;
 537
 538                 icmp_typep = skb_header_pointer(skb, payload_off +
 539                                                 offsetof(struct icmp6hdr,
 540                                                         icmp6_type),
 541                                                 sizeof(icmp_type), &icmp_type);
 542
 543                 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
 544                         return false;
 545         }
 546
 547         return true;
 548 }
 549
 550 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 551                             unsigned int mtu, unsigned int payload_length)
 552 {
 553         struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
 554         struct icmp6hdr *icmp6h;
 555         u8 *payload;
 556
 557         ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
 558         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
 559         payload = skb_put(nskb, payload_length);
 560
 561         /* IPv6 */
 562         ipv6h->version          =       6;
 563         ipv6h->priority         =       0;
 564         memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
 565         ipv6h->payload_len      =       htons(sizeof(struct icmp6hdr)
 566                                               + payload_length);
 567         ipv6h->nexthdr          =       NEXTHDR_ICMP;
 568         ipv6h->hop_limit        =       IPV6_DEFAULT_HOPLIMIT;
 569         ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
 570         ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
 571
 572         /* ICMPv6 */
 573         icmp6h->icmp6_type      =       ICMPV6_PKT_TOOBIG;
 574         icmp6h->icmp6_code      =       0;
 575         icmp6h->icmp6_cksum     =       0;
 576         icmp6h->icmp6_mtu       =       htonl(mtu);
 577
 578         nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
 579         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
 580                                             payload, payload_length,
 581                                             nskb->csum);
 582         icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 583                                                 sizeof(struct icmp6hdr)
 584                                                 + payload_length,
 585                                                 ipv6h->nexthdr, nskb->csum);
 586 }
 587 #endif /* IPv6 */
 588
 589 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
 590                      struct sk_buff *skb, unsigned int mtu, __be32 flow_key)
 591 {
 592         unsigned int eth_hdr_len = ETH_HLEN;
 593         unsigned int total_length = 0, header_length = 0, payload_length;
 594         struct ethhdr *eh, *old_eh = eth_hdr(skb);
 595         struct sk_buff *nskb;
 596
 597         /* Sanity check */
 598         if (skb->protocol == htons(ETH_P_IP)) {
 599                 if (mtu < IP_MIN_MTU)
 600                         return false;
 601
 602                 if (!ipv4_should_icmp(skb))
 603                         return true;
 604         }
 605 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 606         else if (skb->protocol == htons(ETH_P_IPV6)) {
 607                 if (mtu < IPV6_MIN_MTU)
 608                         return false;
 609
 610                 /*
 611                  * In theory we should do PMTUD on IPv6 multicast messages but
 612                  * we don't have an address to send from so just fragment.
 613                  */
 614                 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
 615                         return false;
 616
 617                 if (!ipv6_should_icmp(skb))
 618                         return true;
 619         }
 620 #endif
 621         else
 622                 return false;
 623
 624         /* Allocate */
 625         if (old_eh->h_proto == htons(ETH_P_8021Q))
 626                 eth_hdr_len = VLAN_ETH_HLEN;
 627
 628         payload_length = skb->len - eth_hdr_len;
 629         if (skb->protocol == htons(ETH_P_IP)) {
 630                 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
 631                 total_length = min_t(unsigned int, header_length +
 632                                                    payload_length, 576);
 633         }
 634 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 635         else {
 636                 header_length = sizeof(struct ipv6hdr) +
 637                                 sizeof(struct icmp6hdr);
 638                 total_length = min_t(unsigned int, header_length +
 639                                                   payload_length, IPV6_MIN_MTU);
 640         }
 641 #endif
 642
 643         total_length = min(total_length, mutable->mtu);
 644         payload_length = total_length - header_length;
 645
 646         nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
 647                              payload_length);
 648         if (!nskb)
 649                 return false;
 650
 651         skb_reserve(nskb, NET_IP_ALIGN);
 652
 653         /* Ethernet / VLAN */
 654         eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
 655         memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
 656         memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
 657         nskb->protocol = eh->h_proto = old_eh->h_proto;
 658         if (old_eh->h_proto == htons(ETH_P_8021Q)) {
 659                 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
 660
 661                 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
 662                 vh->h_vlan_encapsulated_proto = skb->protocol;
 663         }
 664         skb_reset_mac_header(nskb);
 665
 666         /* Protocol */
 667         if (skb->protocol == htons(ETH_P_IP))
 668                 ipv4_build_icmp(skb, nskb, mtu, payload_length);
 669 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 670         else
 671                 ipv6_build_icmp(skb, nskb, mtu, payload_length);
 672 #endif
 673
 674         /*
 675          * Assume that flow based keys are symmetric with respect to input
 676          * and output and use the key that we were going to put on the
 677          * outgoing packet for the fake received packet.  If the keys are
 678          * not symmetric then PMTUD needs to be disabled since we won't have
 679          * any way of synthesizing packets.
 680          */
 681         if ((mutable->port_config.flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
 682             (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
 683                 OVS_CB(nskb)->tun_id = flow_key;
 684
 685         compute_ip_summed(nskb, false);
 686         vport_receive(vport, nskb);
 687
 688         return true;
 689 }
 690
 691 static bool check_mtu(struct sk_buff *skb,
 692                       struct vport *vport,
 693                       const struct tnl_mutable_config *mutable,
 694                       const struct rtable *rt, __be16 *frag_offp)
 695 {
 696         int mtu;
 697         __be16 frag_off;
 698
 699         frag_off = (mutable->port_config.flags & TNL_F_PMTUD) ? htons(IP_DF) : 0;
 700         if (frag_off)
 701                 mtu = dst_mtu(&rt_dst(rt))
 702                         - ETH_HLEN
 703                         - mutable->tunnel_hlen
 704                         - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
 705         else
 706                 mtu = mutable->mtu;
 707
 708         if (skb->protocol == htons(ETH_P_IP)) {
 709                 struct iphdr *old_iph = ip_hdr(skb);
 710
 711                 frag_off |= old_iph->frag_off & htons(IP_DF);
 712                 mtu = max(mtu, IP_MIN_MTU);
 713
 714                 if ((old_iph->frag_off & htons(IP_DF)) &&
 715                     mtu < ntohs(old_iph->tot_len)) {
 716                         if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
 717                                 goto drop;
 718                 }
 719         }
 720 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 721         else if (skb->protocol == htons(ETH_P_IPV6)) {
 722                 unsigned int packet_length = skb->len - ETH_HLEN
 723                         - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
 724
 725                 mtu = max(mtu, IPV6_MIN_MTU);
 726
 727                 /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
 728                 if (packet_length > IPV6_MIN_MTU)
 729                         frag_off = htons(IP_DF);
 730
 731                 if (mtu < packet_length) {
 732                         if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
 733                                 goto drop;
 734                 }
 735         }
 736 #endif
 737
 738         *frag_offp = frag_off;
 739         return true;
 740
 741 drop:
 742         *frag_offp = 0;
 743         return false;
 744 }
 745
 746 static void create_tunnel_header(const struct vport *vport,
 747                                  const struct tnl_mutable_config *mutable,
 748                                  const struct rtable *rt, void *header)
 749 {
 750         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 751         struct iphdr *iph = header;
 752
 753         iph->version    = 4;
 754         iph->ihl        = sizeof(struct iphdr) >> 2;
 755         iph->frag_off   = htons(IP_DF);
 756         iph->protocol   = tnl_vport->tnl_ops->ipproto;
 757         iph->tos        = mutable->port_config.tos;
 758         iph->daddr      = rt->rt_dst;
 759         iph->saddr      = rt->rt_src;
 760         iph->ttl        = mutable->port_config.ttl;
 761         if (!iph->ttl)
 762                 iph->ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
 763
 764         tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
 765 }
 766
 767 static inline void *get_cached_header(const struct tnl_cache *cache)
 768 {
 769         return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
 770 }
 771
 772 static inline bool check_cache_valid(const struct tnl_cache *cache,
 773                                      const struct tnl_mutable_config *mutable)
 774 {
 775         return cache &&
 776 #ifdef NEED_CACHE_TIMEOUT
 777                 time_before(jiffies, cache->expiration) &&
 778 #endif
 779 #ifdef HAVE_RT_GENID
 780                 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
 781 #endif
 782 #ifdef HAVE_HH_SEQ
 783                 rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq &&
 784 #endif
 785                 mutable->seq == cache->mutable_seq &&
 786                 (!is_internal_dev(rt_dst(cache->rt).dev) ||
 787                 (cache->flow && !cache->flow->dead));
 788 }
 789
 790 static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux)
 791 {
 792         struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node);
 793         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
 794         const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
 795
 796         if (cache && !check_cache_valid(cache, mutable) &&
 797             spin_trylock_bh(&tnl_vport->cache_lock)) {
 798                 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
 799                 spin_unlock_bh(&tnl_vport->cache_lock);
 800         }
 801
 802         return 0;
 803 }
 804
 805 static void cache_cleaner(struct work_struct *work)
 806 {
 807         schedule_cache_cleaner();
 808
 809         rcu_read_lock();
 810         tbl_foreach(port_table, cache_cleaner_cb, NULL);
 811         rcu_read_unlock();
 812 }
 813
 814 static inline void create_eth_hdr(struct tnl_cache *cache,
 815                                   const struct rtable *rt)
 816 {
 817         void *cache_data = get_cached_header(cache);
 818         int hh_len = rt_dst(rt).hh->hh_len;
 819         int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len;
 820
 821 #ifdef HAVE_HH_SEQ
 822         unsigned hh_seq;
 823
 824         do {
 825                 hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock);
 826                 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
 827         } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq));
 828
 829         cache->hh_seq = hh_seq;
 830 #else
 831         read_lock_bh(&rt_dst(rt).hh->hh_lock);
 832         memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
 833         read_unlock_bh(&rt_dst(rt).hh->hh_lock);
 834 #endif
 835 }
 836
 837 static struct tnl_cache *build_cache(struct vport *vport,
 838                                      const struct tnl_mutable_config *mutable,
 839                                      struct rtable *rt)
 840 {
 841         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 842         struct tnl_cache *cache;
 843         void *cache_data;
 844         int cache_len;
 845
 846         if (!(mutable->port_config.flags & TNL_F_HDR_CACHE))
 847                 return NULL;
 848
 849         /*
 850          * If there is no entry in the ARP cache or if this device does not
 851          * support hard header caching just fall back to the IP stack.
 852          */
 853         if (!rt_dst(rt).hh)
 854                 return NULL;
 855
 856         /*
 857          * If lock is contended fall back to directly building the header.
 858          * We're not going to help performance by sitting here spinning.
 859          */
 860         if (!spin_trylock_bh(&tnl_vport->cache_lock))
 861                 return NULL;
 862
 863         cache = tnl_vport->cache;
 864         if (check_cache_valid(cache, mutable))
 865                 goto unlock;
 866         else
 867                 cache = NULL;
 868
 869         cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen;
 870
 871         cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
 872                         cache_len, GFP_ATOMIC);
 873         if (!cache)
 874                 goto unlock;
 875
 876         cache->len = cache_len;
 877
 878         create_eth_hdr(cache, rt);
 879         cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len;
 880
 881         create_tunnel_header(vport, mutable, rt, cache_data);
 882
 883         cache->mutable_seq = mutable->seq;
 884         cache->rt = rt;
 885 #ifdef NEED_CACHE_TIMEOUT
 886         cache->expiration = jiffies + tnl_vport->cache_exp_interval;
 887 #endif
 888
 889         if (is_internal_dev(rt_dst(rt).dev)) {
 890                 int err;
 891                 struct vport *vport;
 892                 struct dp_port *dp_port;
 893                 struct sk_buff *skb;
 894                 bool is_frag;
 895                 struct odp_flow_key flow_key;
 896                 struct tbl_node *flow_node;
 897
 898                 vport = internal_dev_get_vport(rt_dst(rt).dev);
 899                 if (!vport)
 900                         goto done;
 901
 902                 dp_port = vport_get_dp_port(vport);
 903                 if (!dp_port)
 904                         goto done;
 905
 906                 skb = alloc_skb(cache->len, GFP_ATOMIC);
 907                 if (!skb)
 908                         goto done;
 909
 910                 __skb_put(skb, cache->len);
 911                 memcpy(skb->data, get_cached_header(cache), cache->len);
 912
 913                 err = flow_extract(skb, dp_port->port_no, &flow_key, &is_frag);
 914
 915                 kfree_skb(skb);
 916                 if (err || is_frag)
 917                         goto done;
 918
 919                 flow_node = tbl_lookup(rcu_dereference(dp_port->dp->table),
 920                                        &flow_key, flow_hash(&flow_key),
 921                                        flow_cmp);
 922                 if (flow_node) {
 923                         struct sw_flow *flow = flow_cast(flow_node);
 924
 925                         cache->flow = flow;
 926                         flow_hold(flow);
 927                 }
 928         }
 929
 930 done:
 931         assign_cache_rcu(vport, cache);
 932
 933 unlock:
 934         spin_unlock_bh(&tnl_vport->cache_lock);
 935
 936         return cache;
 937 }
 938
 939 static struct rtable *find_route(struct vport *vport,
 940                                  const struct tnl_mutable_config *mutable,
 941                                  u8 tos, struct tnl_cache **cache)
 942 {
 943         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 944         struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
 945
 946         *cache = NULL;
 947         tos = RT_TOS(tos);
 948
 949         if (likely(tos == mutable->port_config.tos &&
 950                    check_cache_valid(cur_cache, mutable))) {
 951                 *cache = cur_cache;
 952                 return cur_cache->rt;
 953         } else {
 954                 struct rtable *rt;
 955                 struct flowi fl = { .nl_u = { .ip4_u =
 956                                               { .daddr = mutable->port_config.daddr,
 957                                                 .saddr = mutable->port_config.saddr,
 958                                                 .tos = tos } },
 959                                     .proto = tnl_vport->tnl_ops->ipproto };
 960
 961                 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
 962                         return NULL;
 963
 964                 if (likely(tos == mutable->port_config.tos))
 965                         *cache = build_cache(vport, mutable, rt);
 966
 967                 return rt;
 968         }
 969 }
 970
 971 static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom)
 972 {
 973         if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
 974                 struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16);
 975                 if (unlikely(!nskb)) {
 976                         kfree_skb(skb);
 977                         return ERR_PTR(-ENOMEM);
 978                 }
 979
 980                 set_skb_csum_bits(skb, nskb);
 981
 982                 if (skb->sk)
 983                         skb_set_owner_w(nskb, skb->sk);
 984
 985                 kfree_skb(skb);
 986                 return nskb;
 987         }
 988
 989         return skb;
 990 }
 991
 992 static inline bool need_linearize(const struct sk_buff *skb)
 993 {
 994         int i;
 995
 996         if (unlikely(skb_shinfo(skb)->frag_list))
 997                 return true;
 998
 999         /*
1000          * Generally speaking we should linearize if there are paged frags.
1001          * However, if all of the refcounts are 1 we know nobody else can
1002          * change them from underneath us and we can skip the linearization.
1003          */
1004         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1005                 if (unlikely(page_count(skb_shinfo(skb)->frags[0].page) > 1))
1006                         return true;
1007
1008         return false;
1009 }
1010
1011 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1012                                        const struct tnl_mutable_config *mutable,
1013                                        const struct rtable *rt)
1014 {
1015         int min_headroom;
1016         int err;
1017
1018         forward_ip_summed(skb);
1019
1020         err = vswitch_skb_checksum_setup(skb);
1021         if (unlikely(err))
1022                 goto error_free;
1023
1024         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1025                         + mutable->tunnel_hlen;
1026
1027         if (skb_is_gso(skb)) {
1028                 struct sk_buff *nskb;
1029
1030                 /*
1031                  * If we are doing GSO on a pskb it is better to make sure that
1032                  * the headroom is correct now.  We will only have to copy the
1033                  * portion in the linear data area and GSO will preserve
1034                  * headroom when it creates the segments.  This is particularly
1035                  * beneficial on Xen where we get a lot of GSO pskbs.
1036                  * Conversely, we avoid copying if it is just to get our own
1037                  * writable clone because GSO will do the copy for us.
1038                  */
1039                 if (skb_headroom(skb) < min_headroom) {
1040                         skb = check_headroom(skb, min_headroom);
1041                         if (unlikely(IS_ERR(skb))) {
1042                                 err = PTR_ERR(skb);
1043                                 goto error;
1044                         }
1045                 }
1046
1047                 nskb = skb_gso_segment(skb, 0);
1048                 kfree_skb(skb);
1049                 if (unlikely(IS_ERR(nskb))) {
1050                         err = PTR_ERR(nskb);
1051                         goto error;
1052                 }
1053
1054                 skb = nskb;
1055         } else {
1056                 skb = check_headroom(skb, min_headroom);
1057                 if (unlikely(IS_ERR(skb))) {
1058                         err = PTR_ERR(skb);
1059                         goto error;
1060                 }
1061
1062                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1063                         /*
1064                          * Pages aren't locked and could change at any time.
1065                          * If this happens after we compute the checksum, the
1066                          * checksum will be wrong.  We linearize now to avoid
1067                          * this problem.
1068                          */
1069                         if (unlikely(need_linearize(skb))) {
1070                                 err = __skb_linearize(skb);
1071                                 if (unlikely(err))
1072                                         goto error_free;
1073                         }
1074
1075                         err = skb_checksum_help(skb);
1076                         if (unlikely(err))
1077                                 goto error_free;
1078                 } else if (skb->ip_summed == CHECKSUM_COMPLETE)
1079                         skb->ip_summed = CHECKSUM_NONE;
1080         }
1081
1082         return skb;
1083
1084 error_free:
1085         kfree_skb(skb);
1086 error:
1087         return ERR_PTR(err);
1088 }
1089
1090 static int send_frags(struct sk_buff *skb,
1091                       const struct tnl_mutable_config *mutable)
1092 {
1093         int sent_len;
1094         int err;
1095
1096         sent_len = 0;
1097         while (skb) {
1098                 struct sk_buff *next = skb->next;
1099                 int frag_len = skb->len - mutable->tunnel_hlen;
1100
1101                 skb->next = NULL;
1102
1103                 memset(&IPCB(skb)->opt, 0, sizeof(IPCB(skb)->opt));
1104                 IPCB(skb)->flags = 0;
1105
1106                 err = ip_local_out(skb);
1107                 if (likely(net_xmit_eval(err) == 0))
1108                         sent_len += frag_len;
1109                 else {
1110                         skb = next;
1111                         goto free_frags;
1112                 }
1113
1114                 skb = next;
1115         }
1116
1117         return sent_len;
1118
1119 free_frags:
1120         /*
1121          * There's no point in continuing to send fragments once one has been
1122          * dropped so just free the rest.  This may help improve the congestion
1123          * that caused the first packet to be dropped.
1124          */
1125         tnl_free_linked_skbs(skb);
1126         return sent_len;
1127 }
1128
1129 int tnl_send(struct vport *vport, struct sk_buff *skb)
1130 {
1131         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1132         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1133
1134         enum vport_err_type err = VPORT_E_TX_ERROR;
1135         struct rtable *rt;
1136         struct dst_entry *unattached_dst = NULL;
1137         struct tnl_cache *cache;
1138         int sent_len = 0;
1139         __be16 frag_off;
1140         u8 ttl;
1141         u8 inner_tos;
1142         u8 tos;
1143
1144         /* Validate the protocol headers before we try to use them. */
1145         if (skb->protocol == htons(ETH_P_8021Q)) {
1146                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1147                         goto error_free;
1148
1149                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1150                 skb_set_network_header(skb, VLAN_ETH_HLEN);
1151         }
1152
1153         if (skb->protocol == htons(ETH_P_IP)) {
1154                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1155                     + sizeof(struct iphdr))))
1156                         skb->protocol = 0;
1157         }
1158 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1159         else if (skb->protocol == htons(ETH_P_IPV6)) {
1160                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1161                     + sizeof(struct ipv6hdr))))
1162                         skb->protocol = 0;
1163         }
1164 #endif
1165
1166         /* ToS */
1167         if (skb->protocol == htons(ETH_P_IP))
1168                 inner_tos = ip_hdr(skb)->tos;
1169 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1170         else if (skb->protocol == htons(ETH_P_IPV6))
1171                 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1172 #endif
1173         else
1174                 inner_tos = 0;
1175
1176         if (mutable->port_config.flags & TNL_F_TOS_INHERIT)
1177                 tos = inner_tos;
1178         else
1179                 tos = mutable->port_config.tos;
1180
1181         tos = INET_ECN_encapsulate(tos, inner_tos);
1182
1183         /* Route lookup */
1184         rt = find_route(vport, mutable, tos, &cache);
1185         if (unlikely(!rt))
1186                 goto error_free;
1187         if (unlikely(!cache))
1188                 unattached_dst = &rt_dst(rt);
1189
1190         /* Reset SKB */
1191         nf_reset(skb);
1192         secpath_reset(skb);
1193         skb_dst_drop(skb);
1194
1195         /* Offloading */
1196         skb = handle_offloads(skb, mutable, rt);
1197         if (unlikely(IS_ERR(skb)))
1198                 goto error;
1199
1200         /* MTU */
1201         if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1202                 err = VPORT_E_TX_DROPPED;
1203                 goto error_free;
1204         }
1205
1206         /*
1207          * If we are over the MTU, allow the IP stack to handle fragmentation.
1208          * Fragmentation is a slow path anyways.
1209          */
1210         if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1211                      cache)) {
1212                 unattached_dst = &rt_dst(rt);
1213                 dst_hold(unattached_dst);
1214                 cache = NULL;
1215         }
1216
1217         /* TTL */
1218         ttl = mutable->port_config.ttl;
1219         if (!ttl)
1220                 ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
1221
1222         if (mutable->port_config.flags & TNL_F_TTL_INHERIT) {
1223                 if (skb->protocol == htons(ETH_P_IP))
1224                         ttl = ip_hdr(skb)->ttl;
1225 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1226                 else if (skb->protocol == htons(ETH_P_IPV6))
1227                         ttl = ipv6_hdr(skb)->hop_limit;
1228 #endif
1229         }
1230
1231         while (skb) {
1232                 struct iphdr *iph;
1233                 struct sk_buff *next_skb = skb->next;
1234                 skb->next = NULL;
1235
1236                 if (likely(cache)) {
1237                         skb_push(skb, cache->len);
1238                         memcpy(skb->data, get_cached_header(cache), cache->len);
1239                         skb_reset_mac_header(skb);
1240                         skb_set_network_header(skb, rt_dst(rt).hh->hh_len);
1241
1242                 } else {
1243                         skb_push(skb, mutable->tunnel_hlen);
1244                         create_tunnel_header(vport, mutable, rt, skb->data);
1245                         skb_reset_network_header(skb);
1246
1247                         if (next_skb)
1248                                 skb_dst_set(skb, dst_clone(unattached_dst));
1249                         else {
1250                                 skb_dst_set(skb, unattached_dst);
1251                                 unattached_dst = NULL;
1252                         }
1253                 }
1254                 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1255
1256                 iph = ip_hdr(skb);
1257                 iph->tos = tos;
1258                 iph->ttl = ttl;
1259                 iph->frag_off = frag_off;
1260                 ip_select_ident(iph, &rt_dst(rt), NULL);
1261
1262                 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1263                 if (unlikely(!skb))
1264                         goto next;
1265
1266                 if (likely(cache)) {
1267                         int orig_len = skb->len - cache->len;
1268                         struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1269
1270                         skb->protocol = htons(ETH_P_IP);
1271                         iph->tot_len = htons(skb->len - skb_network_offset(skb));
1272                         ip_send_check(iph);
1273
1274                         if (cache_vport) {
1275                                 OVS_CB(skb)->flow = cache->flow;
1276                                 compute_ip_summed(skb, true);
1277                                 vport_receive(cache_vport, skb);
1278                                 sent_len += orig_len;
1279                         } else {
1280                                 int err;
1281
1282                                 skb->dev = rt_dst(rt).dev;
1283                                 err = dev_queue_xmit(skb);
1284
1285                                 if (likely(net_xmit_eval(err) == 0))
1286                                         sent_len += orig_len;
1287                         }
1288                 } else
1289                         sent_len += send_frags(skb, mutable);
1290
1291 next:
1292                 skb = next_skb;
1293         }
1294
1295         if (unlikely(sent_len == 0))
1296                 vport_record_error(vport, VPORT_E_TX_DROPPED);
1297
1298         goto out;
1299
1300 error_free:
1301         tnl_free_linked_skbs(skb);
1302 error:
1303         dst_release(unattached_dst);
1304         vport_record_error(vport, err);
1305 out:
1306         return sent_len;
1307 }
1308
1309 static int set_config(const void __user *uconfig, const struct tnl_ops *tnl_ops,
1310                       const struct vport *cur_vport,
1311                       struct tnl_mutable_config *mutable)
1312 {
1313         const struct vport *old_vport;
1314         const struct tnl_mutable_config *old_mutable;
1315
1316         if (copy_from_user(&mutable->port_config, uconfig, sizeof(struct tnl_port_config)))
1317                 return -EFAULT;
1318
1319         if (mutable->port_config.daddr == 0)
1320                 return -EINVAL;
1321
1322         if (mutable->port_config.tos != RT_TOS(mutable->port_config.tos))
1323                 return -EINVAL;
1324
1325         mutable->tunnel_hlen = tnl_ops->hdr_len(&mutable->port_config);
1326         if (mutable->tunnel_hlen < 0)
1327                 return mutable->tunnel_hlen;
1328
1329         mutable->tunnel_hlen += sizeof(struct iphdr);
1330
1331         mutable->tunnel_type = tnl_ops->tunnel_type;
1332         if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) {
1333                 mutable->tunnel_type |= TNL_T_KEY_MATCH;
1334                 mutable->port_config.in_key = 0;
1335         } else
1336                 mutable->tunnel_type |= TNL_T_KEY_EXACT;
1337
1338         old_vport = tnl_find_port(mutable->port_config.saddr,
1339                                   mutable->port_config.daddr,
1340                                   mutable->port_config.in_key,
1341                                   mutable->tunnel_type,
1342                                   &old_mutable);
1343
1344         if (old_vport && old_vport != cur_vport)
1345                 return -EEXIST;
1346
1347         if (mutable->port_config.flags & TNL_F_OUT_KEY_ACTION)
1348                 mutable->port_config.out_key = 0;
1349
1350         return 0;
1351 }
1352
1353 struct vport *tnl_create(const char *name, const void __user *config,
1354                          const struct vport_ops *vport_ops,
1355                          const struct tnl_ops *tnl_ops)
1356 {
1357         struct vport *vport;
1358         struct tnl_vport *tnl_vport;
1359         int initial_frag_id;
1360         int err;
1361
1362         vport = vport_alloc(sizeof(struct tnl_vport), vport_ops);
1363         if (IS_ERR(vport)) {
1364                 err = PTR_ERR(vport);
1365                 goto error;
1366         }
1367
1368         tnl_vport = tnl_vport_priv(vport);
1369
1370         strcpy(tnl_vport->name, name);
1371         tnl_vport->tnl_ops = tnl_ops;
1372
1373         tnl_vport->mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1374         if (!tnl_vport->mutable) {
1375                 err = -ENOMEM;
1376                 goto error_free_vport;
1377         }
1378
1379         vport_gen_rand_ether_addr(tnl_vport->mutable->eth_addr);
1380         tnl_vport->mutable->mtu = ETH_DATA_LEN;
1381
1382         get_random_bytes(&initial_frag_id, sizeof(int));
1383         atomic_set(&tnl_vport->frag_id, initial_frag_id);
1384
1385         err = set_config(config, tnl_ops, NULL, tnl_vport->mutable);
1386         if (err)
1387                 goto error_free_mutable;
1388
1389         spin_lock_init(&tnl_vport->cache_lock);
1390
1391 #ifdef NEED_CACHE_TIMEOUT
1392         tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1393                                         (net_random() % (MAX_CACHE_EXP / 2));
1394 #endif
1395
1396         err = add_port(vport);
1397         if (err)
1398                 goto error_free_mutable;
1399
1400         return vport;
1401
1402 error_free_mutable:
1403         kfree(tnl_vport->mutable);
1404 error_free_vport:
1405         vport_free(vport);
1406 error:
1407         return ERR_PTR(err);
1408 }
1409
1410 int tnl_modify(struct vport *vport, const void __user *config)
1411 {
1412         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1413         struct tnl_mutable_config *mutable;
1414         int err;
1415
1416         mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1417         if (!mutable) {
1418                 err = -ENOMEM;
1419                 goto error;
1420         }
1421
1422         err = set_config(config, tnl_vport->tnl_ops, vport, mutable);
1423         if (err)
1424                 goto error_free;
1425
1426         mutable->seq++;
1427
1428         err = move_port(vport, mutable);
1429         if (err)
1430                 goto error_free;
1431
1432         return 0;
1433
1434 error_free:
1435         kfree(mutable);
1436 error:
1437         return err;
1438 }
1439
1440 static void free_port_rcu(struct rcu_head *rcu)
1441 {
1442         struct tnl_vport *tnl_vport = container_of(rcu, struct tnl_vport, rcu);
1443
1444         spin_lock_bh(&tnl_vport->cache_lock);
1445         free_cache(tnl_vport->cache);
1446         spin_unlock_bh(&tnl_vport->cache_lock);
1447
1448         kfree(tnl_vport->mutable);
1449         vport_free(tnl_vport_to_vport(tnl_vport));
1450 }
1451
1452 int tnl_destroy(struct vport *vport)
1453 {
1454         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1455         const struct tnl_mutable_config *old_mutable;
1456
1457         if (vport == tnl_find_port(tnl_vport->mutable->port_config.saddr,
1458             tnl_vport->mutable->port_config.daddr,
1459             tnl_vport->mutable->port_config.in_key,
1460             tnl_vport->mutable->tunnel_type,
1461             &old_mutable))
1462                 del_port(vport);
1463
1464         call_rcu(&tnl_vport->rcu, free_port_rcu);
1465
1466         return 0;
1467 }
1468
1469 int tnl_set_mtu(struct vport *vport, int mtu)
1470 {
1471         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1472         struct tnl_mutable_config *mutable;
1473
1474         mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1475         if (!mutable)
1476                 return -ENOMEM;
1477
1478         mutable->mtu = mtu;
1479         assign_config_rcu(vport, mutable);
1480
1481         return 0;
1482 }
1483
1484 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1485 {
1486         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1487         struct tnl_mutable_config *mutable;
1488
1489         mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1490         if (!mutable)
1491                 return -ENOMEM;
1492
1493         memcpy(mutable->eth_addr, addr, ETH_ALEN);
1494         assign_config_rcu(vport, mutable);
1495
1496         return 0;
1497 }
1498
1499 const char *tnl_get_name(const struct vport *vport)
1500 {
1501         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1502         return tnl_vport->name;
1503 }
1504
1505 const unsigned char *tnl_get_addr(const struct vport *vport)
1506 {
1507         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1508         return rcu_dereference(tnl_vport->mutable)->eth_addr;
1509 }
1510
1511 int tnl_get_mtu(const struct vport *vport)
1512 {
1513         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1514         return rcu_dereference(tnl_vport->mutable)->mtu;
1515 }
1516
1517 void tnl_free_linked_skbs(struct sk_buff *skb)
1518 {
1519         if (unlikely(!skb))
1520                 return;
1521
1522         while (skb) {
1523                 struct sk_buff *next = skb->next;
1524                 kfree_skb(skb);
1525                 skb = next;
1526         }
1527 }