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/igmp.h>
  14 #include <linux/in.h>
  15 #include <linux/in_route.h>
  16 #include <linux/inetdevice.h>
  17 #include <linux/jhash.h>
  18 #include <linux/list.h>
  19 #include <linux/kernel.h>
  20 #include <linux/version.h>
  21 #include <linux/workqueue.h>
  22 #include <linux/rculist.h>
  23
  24 #include <net/dsfield.h>
  25 #include <net/dst.h>
  26 #include <net/icmp.h>
  27 #include <net/inet_ecn.h>
  28 #include <net/ip.h>
  29 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
  30 #include <net/ipv6.h>
  31 #endif
  32 #include <net/route.h>
  33 #include <net/xfrm.h>
  34
  35 #include "actions.h"
  36 #include "checksum.h"
  37 #include "datapath.h"
  38 #include "tunnel.h"
  39 #include "vlan.h"
  40 #include "vport.h"
  41 #include "vport-generic.h"
  42 #include "vport-internal_dev.h"
  43
  44 #ifdef NEED_CACHE_TIMEOUT
  45 /*
  46  * On kernels where we can't quickly detect changes in the rest of the system
  47  * we use an expiration time to invalidate the cache.  A shorter expiration
  48  * reduces the length of time that we may potentially blackhole packets while
  49  * a longer time increases performance by reducing the frequency that the
  50  * cache needs to be rebuilt.  A variety of factors may cause the cache to be
  51  * invalidated before the expiration time but this is the maximum.  The time
  52  * is expressed in jiffies.
  53  */
  54 #define MAX_CACHE_EXP HZ
  55 #endif
  56
  57 /*
  58  * Interval to check for and remove caches that are no longer valid.  Caches
  59  * are checked for validity before they are used for packet encapsulation and
  60  * old caches are removed at that time.  However, if no packets are sent through
  61  * the tunnel then the cache will never be destroyed.  Since it holds
  62  * references to a number of system objects, the cache will continue to use
  63  * system resources by not allowing those objects to be destroyed.  The cache
  64  * cleaner is periodically run to free invalid caches.  It does not
  65  * significantly affect system performance.  A lower interval will release
  66  * resources faster but will itself consume resources by requiring more frequent
  67  * checks.  A longer interval may result in messages being printed to the kernel
  68  * message buffer about unreleased resources.  The interval is expressed in
  69  * jiffies.
  70  */
  71 #define CACHE_CLEANER_INTERVAL (5 * HZ)
  72
  73 #define CACHE_DATA_ALIGN 16
  74 #define PORT_TABLE_SIZE  1024
  75
  76 static struct hlist_head *port_table __read_mostly;
  77 static int port_table_count;
  78
  79 static void cache_cleaner(struct work_struct *work);
  80 static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
  81
  82 /*
  83  * These are just used as an optimization: they don't require any kind of
  84  * synchronization because we could have just as easily read the value before
  85  * the port change happened.
  86  */
  87 static unsigned int key_local_remote_ports __read_mostly;
  88 static unsigned int key_remote_ports __read_mostly;
  89 static unsigned int key_multicast_ports __read_mostly;
  90 static unsigned int local_remote_ports __read_mostly;
  91 static unsigned int remote_ports __read_mostly;
  92 static unsigned int multicast_ports __read_mostly;
  93
  94 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  95 #define rt_dst(rt) (rt->dst)
  96 #else
  97 #define rt_dst(rt) (rt->u.dst)
  98 #endif
  99
 100 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
 101 static struct hh_cache *rt_hh(struct rtable *rt)
 102 {
 103         struct neighbour *neigh = dst_get_neighbour(&rt->dst);
 104         if (!neigh || !(neigh->nud_state & NUD_CONNECTED) ||
 105                         !neigh->hh.hh_len)
 106                 return NULL;
 107         return &neigh->hh;
 108 }
 109 #else
 110 #define rt_hh(rt) (rt_dst(rt).hh)
 111 #endif
 112
 113 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
 114 {
 115         return vport_from_priv(tnl_vport);
 116 }
 117
 118 /* This is analogous to rtnl_dereference for the tunnel cache.  It checks that
 119  * cache_lock is held, so it is only for update side code.
 120  */
 121 static struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
 122 {
 123         return rcu_dereference_protected(tnl_vport->cache,
 124                                  lockdep_is_held(&tnl_vport->cache_lock));
 125 }
 126
 127 static void schedule_cache_cleaner(void)
 128 {
 129         schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
 130 }
 131
 132 static void free_cache(struct tnl_cache *cache)
 133 {
 134         if (!cache)
 135                 return;
 136
 137         flow_put(cache->flow);
 138         ip_rt_put(cache->rt);
 139         kfree(cache);
 140 }
 141
 142 static void free_config_rcu(struct rcu_head *rcu)
 143 {
 144         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
 145         kfree(c);
 146 }
 147
 148 static void free_cache_rcu(struct rcu_head *rcu)
 149 {
 150         struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
 151         free_cache(c);
 152 }
 153
 154 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
 155  * within an RCU callback.  Fortunately this part doesn't require waiting for
 156  * an RCU grace period.
 157  */
 158 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
 159 {
 160         ASSERT_RTNL();
 161         if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
 162                 struct in_device *in_dev;
 163                 in_dev = inetdev_by_index(&init_net, mutable->mlink);
 164                 if (in_dev)
 165                         ip_mc_dec_group(in_dev, mutable->key.daddr);
 166         }
 167 }
 168
 169 static void assign_config_rcu(struct vport *vport,
 170                               struct tnl_mutable_config *new_config)
 171 {
 172         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 173         struct tnl_mutable_config *old_config;
 174
 175         old_config = rtnl_dereference(tnl_vport->mutable);
 176         rcu_assign_pointer(tnl_vport->mutable, new_config);
 177
 178         free_mutable_rtnl(old_config);
 179         call_rcu(&old_config->rcu, free_config_rcu);
 180 }
 181
 182 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
 183 {
 184         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 185         struct tnl_cache *old_cache;
 186
 187         old_cache = cache_dereference(tnl_vport);
 188         rcu_assign_pointer(tnl_vport->cache, new_cache);
 189
 190         if (old_cache)
 191                 call_rcu(&old_cache->rcu, free_cache_rcu);
 192 }
 193
 194 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 195 {
 196         bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
 197
 198         if (mutable->flags & TNL_F_IN_KEY_MATCH) {
 199                 if (mutable->key.saddr)
 200                         return &local_remote_ports;
 201                 else if (is_multicast)
 202                         return &multicast_ports;
 203                 else
 204                         return &remote_ports;
 205         } else {
 206                 if (mutable->key.saddr)
 207                         return &key_local_remote_ports;
 208                 else if (is_multicast)
 209                         return &key_multicast_ports;
 210                 else
 211                         return &key_remote_ports;
 212         }
 213 }
 214
 215 static u32 port_hash(const struct port_lookup_key *key)
 216 {
 217         return jhash2((u32 *)key, (PORT_KEY_LEN / sizeof(u32)), 0);
 218 }
 219
 220 static struct hlist_head *find_bucket(u32 hash)
 221 {
 222         return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
 223 }
 224
 225 static void port_table_add_port(struct vport *vport)
 226 {
 227         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 228         const struct tnl_mutable_config *mutable;
 229         u32 hash;
 230
 231         if (port_table_count == 0)
 232                 schedule_cache_cleaner();
 233
 234         mutable = rtnl_dereference(tnl_vport->mutable);
 235         hash = port_hash(&mutable->key);
 236         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 237         port_table_count++;
 238
 239         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 240 }
 241
 242 static void port_table_move_port(struct vport *vport,
 243                       struct tnl_mutable_config *new_mutable)
 244 {
 245         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 246         u32 hash;
 247
 248         hash = port_hash(&new_mutable->key);
 249         hlist_del_init_rcu(&tnl_vport->hash_node);
 250         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 251
 252         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 253         assign_config_rcu(vport, new_mutable);
 254         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 255 }
 256
 257 static void port_table_remove_port(struct vport *vport)
 258 {
 259         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 260
 261         hlist_del_init_rcu(&tnl_vport->hash_node);
 262
 263         port_table_count--;
 264         if (port_table_count == 0)
 265                 cancel_delayed_work_sync(&cache_cleaner_wq);
 266
 267         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 268 }
 269
 270 static struct vport *port_table_lookup(struct port_lookup_key *key,
 271                                        const struct tnl_mutable_config **pmutable)
 272 {
 273         struct hlist_node *n;
 274         struct hlist_head *bucket;
 275         u32 hash = port_hash(key);
 276         struct tnl_vport *tnl_vport;
 277
 278         bucket = find_bucket(hash);
 279
 280         hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
 281                 struct tnl_mutable_config *mutable;
 282
 283                 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 284                 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
 285                         *pmutable = mutable;
 286                         return tnl_vport_to_vport(tnl_vport);
 287                 }
 288         }
 289
 290         return NULL;
 291 }
 292
 293 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key,
 294                             int tunnel_type,
 295                             const struct tnl_mutable_config **mutable)
 296 {
 297         struct port_lookup_key lookup;
 298         struct vport *vport;
 299         bool is_multicast = ipv4_is_multicast(saddr);
 300
 301         lookup.saddr = saddr;
 302         lookup.daddr = daddr;
 303
 304         /* First try for exact match on in_key. */
 305         lookup.in_key = key;
 306         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 307         if (!is_multicast && key_local_remote_ports) {
 308                 vport = port_table_lookup(&lookup, mutable);
 309                 if (vport)
 310                         return vport;
 311         }
 312         if (key_remote_ports) {
 313                 lookup.saddr = 0;
 314                 vport = port_table_lookup(&lookup, mutable);
 315                 if (vport)
 316                         return vport;
 317
 318                 lookup.saddr = saddr;
 319         }
 320
 321         /* Then try matches that wildcard in_key. */
 322         lookup.in_key = 0;
 323         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 324         if (!is_multicast && local_remote_ports) {
 325                 vport = port_table_lookup(&lookup, mutable);
 326                 if (vport)
 327                         return vport;
 328         }
 329         if (remote_ports) {
 330                 lookup.saddr = 0;
 331                 vport = port_table_lookup(&lookup, mutable);
 332                 if (vport)
 333                         return vport;
 334         }
 335
 336         if (is_multicast) {
 337                 lookup.saddr = 0;
 338                 lookup.daddr = saddr;
 339                 if (key_multicast_ports) {
 340                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 341                         lookup.in_key = key;
 342                         vport = port_table_lookup(&lookup, mutable);
 343                         if (vport)
 344                                 return vport;
 345                 }
 346                 if (multicast_ports) {
 347                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 348                         lookup.in_key = 0;
 349                         vport = port_table_lookup(&lookup, mutable);
 350                         if (vport)
 351                                 return vport;
 352                 }
 353         }
 354
 355         return NULL;
 356 }
 357
 358 static void ecn_decapsulate(struct sk_buff *skb, u8 tos)
 359 {
 360         if (unlikely(INET_ECN_is_ce(tos))) {
 361                 __be16 protocol = skb->protocol;
 362
 363                 skb_set_network_header(skb, ETH_HLEN);
 364
 365                 if (protocol == htons(ETH_P_8021Q)) {
 366                         if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 367                                 return;
 368
 369                         protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 370                         skb_set_network_header(skb, VLAN_ETH_HLEN);
 371                 }
 372
 373                 if (protocol == htons(ETH_P_IP)) {
 374                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 375                             + sizeof(struct iphdr))))
 376                                 return;
 377
 378                         IP_ECN_set_ce(ip_hdr(skb));
 379                 }
 380 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 381                 else if (protocol == htons(ETH_P_IPV6)) {
 382                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 383                             + sizeof(struct ipv6hdr))))
 384                                 return;
 385
 386                         IP6_ECN_set_ce(ipv6_hdr(skb));
 387                 }
 388 #endif
 389         }
 390 }
 391
 392 /**
 393  *      tnl_rcv - ingress point for generic tunnel code
 394  *
 395  * @vport: port this packet was received on
 396  * @skb: received packet
 397  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
 398  *
 399  * Must be called with rcu_read_lock.
 400  *
 401  * Packets received by this function are in the following state:
 402  * - skb->data points to the inner Ethernet header.
 403  * - The inner Ethernet header is in the linear data area.
 404  * - skb->csum does not include the inner Ethernet header.
 405  * - The layer pointers are undefined.
 406  */
 407 void tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos)
 408 {
 409         struct ethhdr *eh;
 410
 411         skb_reset_mac_header(skb);
 412         eh = eth_hdr(skb);
 413
 414         if (likely(ntohs(eh->h_proto) >= 1536))
 415                 skb->protocol = eh->h_proto;
 416         else
 417                 skb->protocol = htons(ETH_P_802_2);
 418
 419         skb_dst_drop(skb);
 420         nf_reset(skb);
 421         skb_clear_rxhash(skb);
 422         secpath_reset(skb);
 423
 424         ecn_decapsulate(skb, tos);
 425         vlan_set_tci(skb, 0);
 426
 427         if (unlikely(compute_ip_summed(skb, false))) {
 428                 kfree_skb(skb);
 429                 return;
 430         }
 431
 432         vport_receive(vport, skb);
 433 }
 434
 435 static bool check_ipv4_address(__be32 addr)
 436 {
 437         if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
 438             || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
 439                 return false;
 440
 441         return true;
 442 }
 443
 444 static bool ipv4_should_icmp(struct sk_buff *skb)
 445 {
 446         struct iphdr *old_iph = ip_hdr(skb);
 447
 448         /* Don't respond to L2 broadcast. */
 449         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
 450                 return false;
 451
 452         /* Don't respond to L3 broadcast or invalid addresses. */
 453         if (!check_ipv4_address(old_iph->daddr) ||
 454             !check_ipv4_address(old_iph->saddr))
 455                 return false;
 456
 457         /* Only respond to the first fragment. */
 458         if (old_iph->frag_off & htons(IP_OFFSET))
 459                 return false;
 460
 461         /* Don't respond to ICMP error messages. */
 462         if (old_iph->protocol == IPPROTO_ICMP) {
 463                 u8 icmp_type, *icmp_typep;
 464
 465                 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
 466                                                 (old_iph->ihl << 2) +
 467                                                 offsetof(struct icmphdr, type) -
 468                                                 skb->data, sizeof(icmp_type),
 469                                                 &icmp_type);
 470
 471                 if (!icmp_typep)
 472                         return false;
 473
 474                 if (*icmp_typep > NR_ICMP_TYPES
 475                         || (*icmp_typep <= ICMP_PARAMETERPROB
 476                                 && *icmp_typep != ICMP_ECHOREPLY
 477                                 && *icmp_typep != ICMP_ECHO))
 478                         return false;
 479         }
 480
 481         return true;
 482 }
 483
 484 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 485                             unsigned int mtu, unsigned int payload_length)
 486 {
 487         struct iphdr *iph, *old_iph = ip_hdr(skb);
 488         struct icmphdr *icmph;
 489         u8 *payload;
 490
 491         iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
 492         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 493         payload = skb_put(nskb, payload_length);
 494
 495         /* IP */
 496         iph->version            =       4;
 497         iph->ihl                =       sizeof(struct iphdr) >> 2;
 498         iph->tos                =       (old_iph->tos & IPTOS_TOS_MASK) |
 499                                         IPTOS_PREC_INTERNETCONTROL;
 500         iph->tot_len            =       htons(sizeof(struct iphdr)
 501                                               + sizeof(struct icmphdr)
 502                                               + payload_length);
 503         get_random_bytes(&iph->id, sizeof(iph->id));
 504         iph->frag_off           =       0;
 505         iph->ttl                =       IPDEFTTL;
 506         iph->protocol           =       IPPROTO_ICMP;
 507         iph->daddr              =       old_iph->saddr;
 508         iph->saddr              =       old_iph->daddr;
 509
 510         ip_send_check(iph);
 511
 512         /* ICMP */
 513         icmph->type             =       ICMP_DEST_UNREACH;
 514         icmph->code             =       ICMP_FRAG_NEEDED;
 515         icmph->un.gateway       =       htonl(mtu);
 516         icmph->checksum         =       0;
 517
 518         nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
 519         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
 520                                             payload, payload_length,
 521                                             nskb->csum);
 522         icmph->checksum = csum_fold(nskb->csum);
 523 }
 524
 525 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 526 static bool ipv6_should_icmp(struct sk_buff *skb)
 527 {
 528         struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
 529         int addr_type;
 530         int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
 531         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 532
 533         /* Check source address is valid. */
 534         addr_type = ipv6_addr_type(&old_ipv6h->saddr);
 535         if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
 536                 return false;
 537
 538         /* Don't reply to unspecified addresses. */
 539         if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
 540                 return false;
 541
 542         /* Don't respond to ICMP error messages. */
 543         payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
 544         if (payload_off < 0)
 545                 return false;
 546
 547         if (nexthdr == NEXTHDR_ICMP) {
 548                 u8 icmp_type, *icmp_typep;
 549
 550                 icmp_typep = skb_header_pointer(skb, payload_off +
 551                                                 offsetof(struct icmp6hdr,
 552                                                         icmp6_type),
 553                                                 sizeof(icmp_type), &icmp_type);
 554
 555                 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
 556                         return false;
 557         }
 558
 559         return true;
 560 }
 561
 562 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 563                             unsigned int mtu, unsigned int payload_length)
 564 {
 565         struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
 566         struct icmp6hdr *icmp6h;
 567         u8 *payload;
 568
 569         ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
 570         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
 571         payload = skb_put(nskb, payload_length);
 572
 573         /* IPv6 */
 574         ipv6h->version          =       6;
 575         ipv6h->priority         =       0;
 576         memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
 577         ipv6h->payload_len      =       htons(sizeof(struct icmp6hdr)
 578                                               + payload_length);
 579         ipv6h->nexthdr          =       NEXTHDR_ICMP;
 580         ipv6h->hop_limit        =       IPV6_DEFAULT_HOPLIMIT;
 581         ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
 582         ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
 583
 584         /* ICMPv6 */
 585         icmp6h->icmp6_type      =       ICMPV6_PKT_TOOBIG;
 586         icmp6h->icmp6_code      =       0;
 587         icmp6h->icmp6_cksum     =       0;
 588         icmp6h->icmp6_mtu       =       htonl(mtu);
 589
 590         nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
 591         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
 592                                             payload, payload_length,
 593                                             nskb->csum);
 594         icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 595                                                 sizeof(struct icmp6hdr)
 596                                                 + payload_length,
 597                                                 ipv6h->nexthdr, nskb->csum);
 598 }
 599 #endif /* IPv6 */
 600
 601 bool tnl_frag_needed(struct vport *vport,
 602                      const struct tnl_mutable_config *mutable,
 603                      struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
 604 {
 605         unsigned int eth_hdr_len = ETH_HLEN;
 606         unsigned int total_length = 0, header_length = 0, payload_length;
 607         struct ethhdr *eh, *old_eh = eth_hdr(skb);
 608         struct sk_buff *nskb;
 609
 610         /* Sanity check */
 611         if (skb->protocol == htons(ETH_P_IP)) {
 612                 if (mtu < IP_MIN_MTU)
 613                         return false;
 614
 615                 if (!ipv4_should_icmp(skb))
 616                         return true;
 617         }
 618 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 619         else if (skb->protocol == htons(ETH_P_IPV6)) {
 620                 if (mtu < IPV6_MIN_MTU)
 621                         return false;
 622
 623                 /*
 624                  * In theory we should do PMTUD on IPv6 multicast messages but
 625                  * we don't have an address to send from so just fragment.
 626                  */
 627                 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
 628                         return false;
 629
 630                 if (!ipv6_should_icmp(skb))
 631                         return true;
 632         }
 633 #endif
 634         else
 635                 return false;
 636
 637         /* Allocate */
 638         if (old_eh->h_proto == htons(ETH_P_8021Q))
 639                 eth_hdr_len = VLAN_ETH_HLEN;
 640
 641         payload_length = skb->len - eth_hdr_len;
 642         if (skb->protocol == htons(ETH_P_IP)) {
 643                 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
 644                 total_length = min_t(unsigned int, header_length +
 645                                                    payload_length, 576);
 646         }
 647 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 648         else {
 649                 header_length = sizeof(struct ipv6hdr) +
 650                                 sizeof(struct icmp6hdr);
 651                 total_length = min_t(unsigned int, header_length +
 652                                                   payload_length, IPV6_MIN_MTU);
 653         }
 654 #endif
 655
 656         payload_length = total_length - header_length;
 657
 658         nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
 659                              payload_length);
 660         if (!nskb)
 661                 return false;
 662
 663         skb_reserve(nskb, NET_IP_ALIGN);
 664
 665         /* Ethernet / VLAN */
 666         eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
 667         memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
 668         memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
 669         nskb->protocol = eh->h_proto = old_eh->h_proto;
 670         if (old_eh->h_proto == htons(ETH_P_8021Q)) {
 671                 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
 672
 673                 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
 674                 vh->h_vlan_encapsulated_proto = skb->protocol;
 675         } else
 676                 vlan_set_tci(nskb, vlan_get_tci(skb));
 677         skb_reset_mac_header(nskb);
 678
 679         /* Protocol */
 680         if (skb->protocol == htons(ETH_P_IP))
 681                 ipv4_build_icmp(skb, nskb, mtu, payload_length);
 682 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 683         else
 684                 ipv6_build_icmp(skb, nskb, mtu, payload_length);
 685 #endif
 686
 687         /*
 688          * Assume that flow based keys are symmetric with respect to input
 689          * and output and use the key that we were going to put on the
 690          * outgoing packet for the fake received packet.  If the keys are
 691          * not symmetric then PMTUD needs to be disabled since we won't have
 692          * any way of synthesizing packets.
 693          */
 694         if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
 695             (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
 696                 OVS_CB(nskb)->tun_id = flow_key;
 697
 698         if (unlikely(compute_ip_summed(nskb, false))) {
 699                 kfree_skb(nskb);
 700                 return false;
 701         }
 702
 703         vport_receive(vport, nskb);
 704
 705         return true;
 706 }
 707
 708 static bool check_mtu(struct sk_buff *skb,
 709                       struct vport *vport,
 710                       const struct tnl_mutable_config *mutable,
 711                       const struct rtable *rt, __be16 *frag_offp)
 712 {
 713         bool df_inherit = mutable->flags & TNL_F_DF_INHERIT;
 714         bool pmtud = mutable->flags & TNL_F_PMTUD;
 715         __be16 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
 716         int mtu = 0;
 717         unsigned int packet_length = skb->len - ETH_HLEN;
 718
 719         /* Allow for one level of tagging in the packet length. */
 720         if (!vlan_tx_tag_present(skb) &&
 721             eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 722                 packet_length -= VLAN_HLEN;
 723
 724         if (pmtud) {
 725                 int vlan_header = 0;
 726
 727                 /* The tag needs to go in packet regardless of where it
 728                  * currently is, so subtract it from the MTU.
 729                  */
 730                 if (vlan_tx_tag_present(skb) ||
 731                     eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 732                         vlan_header = VLAN_HLEN;
 733
 734                 mtu = dst_mtu(&rt_dst(rt))
 735                         - ETH_HLEN
 736                         - mutable->tunnel_hlen
 737                         - vlan_header;
 738         }
 739
 740         if (skb->protocol == htons(ETH_P_IP)) {
 741                 struct iphdr *iph = ip_hdr(skb);
 742
 743                 if (df_inherit)
 744                         frag_off = iph->frag_off & htons(IP_DF);
 745
 746                 if (pmtud && iph->frag_off & htons(IP_DF)) {
 747                         mtu = max(mtu, IP_MIN_MTU);
 748
 749                         if (packet_length > mtu &&
 750                             tnl_frag_needed(vport, mutable, skb, mtu,
 751                                             OVS_CB(skb)->tun_id))
 752                                 return false;
 753                 }
 754         }
 755 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 756         else if (skb->protocol == htons(ETH_P_IPV6)) {
 757                 /* IPv6 requires end hosts to do fragmentation
 758                  * if the packet is above the minimum MTU.
 759                  */
 760                 if (df_inherit && packet_length > IPV6_MIN_MTU)
 761                         frag_off = htons(IP_DF);
 762
 763                 if (pmtud) {
 764                         mtu = max(mtu, IPV6_MIN_MTU);
 765
 766                         if (packet_length > mtu &&
 767                             tnl_frag_needed(vport, mutable, skb, mtu,
 768                                             OVS_CB(skb)->tun_id))
 769                                 return false;
 770                 }
 771         }
 772 #endif
 773
 774         *frag_offp = frag_off;
 775         return true;
 776 }
 777
 778 static void create_tunnel_header(const struct vport *vport,
 779                                  const struct tnl_mutable_config *mutable,
 780                                  const struct rtable *rt, void *header)
 781 {
 782         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 783         struct iphdr *iph = header;
 784
 785         iph->version    = 4;
 786         iph->ihl        = sizeof(struct iphdr) >> 2;
 787         iph->frag_off   = htons(IP_DF);
 788         iph->protocol   = tnl_vport->tnl_ops->ipproto;
 789         iph->tos        = mutable->tos;
 790         iph->daddr      = rt->rt_dst;
 791         iph->saddr      = rt->rt_src;
 792         iph->ttl        = mutable->ttl;
 793         if (!iph->ttl)
 794                 iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
 795
 796         tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
 797 }
 798
 799 static void *get_cached_header(const struct tnl_cache *cache)
 800 {
 801         return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
 802 }
 803
 804 static bool check_cache_valid(const struct tnl_cache *cache,
 805                               const struct tnl_mutable_config *mutable)
 806 {
 807         struct hh_cache *hh;
 808
 809         if (!cache)
 810                 return false;
 811
 812         hh = rt_hh(cache->rt);
 813         return hh &&
 814 #ifdef NEED_CACHE_TIMEOUT
 815                 time_before(jiffies, cache->expiration) &&
 816 #endif
 817 #ifdef HAVE_RT_GENID
 818                 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
 819 #endif
 820 #ifdef HAVE_HH_SEQ
 821                 hh->hh_lock.sequence == cache->hh_seq &&
 822 #endif
 823                 mutable->seq == cache->mutable_seq &&
 824                 (!is_internal_dev(rt_dst(cache->rt).dev) ||
 825                 (cache->flow && !cache->flow->dead));
 826 }
 827
 828 static void __cache_cleaner(struct tnl_vport *tnl_vport)
 829 {
 830         const struct tnl_mutable_config *mutable =
 831                         rcu_dereference(tnl_vport->mutable);
 832         const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
 833
 834         if (cache && !check_cache_valid(cache, mutable) &&
 835             spin_trylock_bh(&tnl_vport->cache_lock)) {
 836                 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
 837                 spin_unlock_bh(&tnl_vport->cache_lock);
 838         }
 839 }
 840
 841 static void cache_cleaner(struct work_struct *work)
 842 {
 843         int i;
 844
 845         schedule_cache_cleaner();
 846
 847         rcu_read_lock();
 848         for (i = 0; i < PORT_TABLE_SIZE; i++) {
 849                 struct hlist_node *n;
 850                 struct hlist_head *bucket;
 851                 struct tnl_vport  *tnl_vport;
 852
 853                 bucket = &port_table[i];
 854                 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node)
 855                         __cache_cleaner(tnl_vport);
 856         }
 857         rcu_read_unlock();
 858 }
 859
 860 static void create_eth_hdr(struct tnl_cache *cache, struct hh_cache *hh)
 861 {
 862         void *cache_data = get_cached_header(cache);
 863         int hh_off;
 864
 865 #ifdef HAVE_HH_SEQ
 866         unsigned hh_seq;
 867
 868         do {
 869                 hh_seq = read_seqbegin(&hh->hh_lock);
 870                 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
 871                 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
 872                 cache->hh_len = hh->hh_len;
 873         } while (read_seqretry(&hh->hh_lock, hh_seq));
 874
 875         cache->hh_seq = hh_seq;
 876 #else
 877         read_lock(&hh->hh_lock);
 878         hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
 879         memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
 880         cache->hh_len = hh->hh_len;
 881         read_unlock(&hh->hh_lock);
 882 #endif
 883 }
 884
 885 static struct tnl_cache *build_cache(struct vport *vport,
 886                                      const struct tnl_mutable_config *mutable,
 887                                      struct rtable *rt)
 888 {
 889         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 890         struct tnl_cache *cache;
 891         void *cache_data;
 892         int cache_len;
 893         struct hh_cache *hh;
 894
 895         if (!(mutable->flags & TNL_F_HDR_CACHE))
 896                 return NULL;
 897
 898         /*
 899          * If there is no entry in the ARP cache or if this device does not
 900          * support hard header caching just fall back to the IP stack.
 901          */
 902
 903         hh = rt_hh(rt);
 904         if (!hh)
 905                 return NULL;
 906
 907         /*
 908          * If lock is contended fall back to directly building the header.
 909          * We're not going to help performance by sitting here spinning.
 910          */
 911         if (!spin_trylock(&tnl_vport->cache_lock))
 912                 return NULL;
 913
 914         cache = cache_dereference(tnl_vport);
 915         if (check_cache_valid(cache, mutable))
 916                 goto unlock;
 917         else
 918                 cache = NULL;
 919
 920         cache_len = LL_RESERVED_SPACE(rt_dst(rt).dev) + mutable->tunnel_hlen;
 921
 922         cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
 923                         cache_len, GFP_ATOMIC);
 924         if (!cache)
 925                 goto unlock;
 926
 927         create_eth_hdr(cache, hh);
 928         cache_data = get_cached_header(cache) + cache->hh_len;
 929         cache->len = cache->hh_len + mutable->tunnel_hlen;
 930
 931         create_tunnel_header(vport, mutable, rt, cache_data);
 932
 933         cache->mutable_seq = mutable->seq;
 934         cache->rt = rt;
 935 #ifdef NEED_CACHE_TIMEOUT
 936         cache->expiration = jiffies + tnl_vport->cache_exp_interval;
 937 #endif
 938
 939         if (is_internal_dev(rt_dst(rt).dev)) {
 940                 struct sw_flow_key flow_key;
 941                 struct vport *dst_vport;
 942                 struct sk_buff *skb;
 943                 int err;
 944                 int flow_key_len;
 945                 struct sw_flow *flow;
 946
 947                 dst_vport = internal_dev_get_vport(rt_dst(rt).dev);
 948                 if (!dst_vport)
 949                         goto done;
 950
 951                 skb = alloc_skb(cache->len, GFP_ATOMIC);
 952                 if (!skb)
 953                         goto done;
 954
 955                 __skb_put(skb, cache->len);
 956                 memcpy(skb->data, get_cached_header(cache), cache->len);
 957
 958                 err = flow_extract(skb, dst_vport->port_no, &flow_key,
 959                                    &flow_key_len);
 960
 961                 consume_skb(skb);
 962                 if (err)
 963                         goto done;
 964
 965                 flow = flow_tbl_lookup(rcu_dereference(dst_vport->dp->table),
 966                                          &flow_key, flow_key_len);
 967                 if (flow) {
 968                         cache->flow = flow;
 969                         flow_hold(flow);
 970                 }
 971         }
 972
 973 done:
 974         assign_cache_rcu(vport, cache);
 975
 976 unlock:
 977         spin_unlock(&tnl_vport->cache_lock);
 978
 979         return cache;
 980 }
 981
 982 static struct rtable *__find_route(const struct tnl_mutable_config *mutable,
 983                                    u8 ipproto, u8 tos)
 984 {
 985 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 986         struct flowi fl = { .nl_u = { .ip4_u = {
 987                                         .daddr = mutable->key.daddr,
 988                                         .saddr = mutable->key.saddr,
 989                                         .tos = tos } },
 990                             .proto = ipproto };
 991         struct rtable *rt;
 992
 993         if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
 994                 return ERR_PTR(-EADDRNOTAVAIL);
 995
 996         return rt;
 997 #else
 998         struct flowi4 fl = { .daddr = mutable->key.daddr,
 999                              .saddr = mutable->key.saddr,
1000                              .flowi4_tos = tos,
1001                              .flowi4_proto = ipproto };
1002
1003         return ip_route_output_key(&init_net, &fl);
1004 #endif
1005 }
1006
1007 static struct rtable *find_route(struct vport *vport,
1008                                  const struct tnl_mutable_config *mutable,
1009                                  u8 tos, struct tnl_cache **cache)
1010 {
1011         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1012         struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
1013
1014         *cache = NULL;
1015         tos = RT_TOS(tos);
1016
1017         if (likely(tos == mutable->tos &&
1018             check_cache_valid(cur_cache, mutable))) {
1019                 *cache = cur_cache;
1020                 return cur_cache->rt;
1021         } else {
1022                 struct rtable *rt;
1023
1024                 rt = __find_route(mutable, tnl_vport->tnl_ops->ipproto, tos);
1025                 if (IS_ERR(rt))
1026                         return NULL;
1027
1028                 if (likely(tos == mutable->tos))
1029                         *cache = build_cache(vport, mutable, rt);
1030
1031                 return rt;
1032         }
1033 }
1034
1035 static bool need_linearize(const struct sk_buff *skb)
1036 {
1037         int i;
1038
1039         if (unlikely(skb_shinfo(skb)->frag_list))
1040                 return true;
1041
1042         /*
1043          * Generally speaking we should linearize if there are paged frags.
1044          * However, if all of the refcounts are 1 we know nobody else can
1045          * change them from underneath us and we can skip the linearization.
1046          */
1047         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1048                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
1049                         return true;
1050
1051         return false;
1052 }
1053
1054 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1055                                        const struct tnl_mutable_config *mutable,
1056                                        const struct rtable *rt)
1057 {
1058         int min_headroom;
1059         int err;
1060
1061         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1062                         + mutable->tunnel_hlen
1063                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1064
1065         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
1066                 int head_delta = SKB_DATA_ALIGN(min_headroom -
1067                                                 skb_headroom(skb) +
1068                                                 16);
1069                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
1070                                         0, GFP_ATOMIC);
1071                 if (unlikely(err))
1072                         goto error_free;
1073         }
1074
1075         forward_ip_summed(skb, true);
1076
1077         if (skb_is_gso(skb)) {
1078                 struct sk_buff *nskb;
1079
1080                 nskb = skb_gso_segment(skb, 0);
1081                 if (IS_ERR(nskb)) {
1082                         kfree_skb(skb);
1083                         err = PTR_ERR(nskb);
1084                         goto error;
1085                 }
1086
1087                 consume_skb(skb);
1088                 skb = nskb;
1089         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
1090                 /* Pages aren't locked and could change at any time.
1091                  * If this happens after we compute the checksum, the
1092                  * checksum will be wrong.  We linearize now to avoid
1093                  * this problem.
1094                  */
1095                 if (unlikely(need_linearize(skb))) {
1096                         err = __skb_linearize(skb);
1097                         if (unlikely(err))
1098                                 goto error_free;
1099                 }
1100
1101                 err = skb_checksum_help(skb);
1102                 if (unlikely(err))
1103                         goto error_free;
1104         }
1105
1106         set_ip_summed(skb, OVS_CSUM_NONE);
1107
1108         return skb;
1109
1110 error_free:
1111         kfree_skb(skb);
1112 error:
1113         return ERR_PTR(err);
1114 }
1115
1116 static int send_frags(struct sk_buff *skb,
1117                       const struct tnl_mutable_config *mutable)
1118 {
1119         int sent_len;
1120
1121         sent_len = 0;
1122         while (skb) {
1123                 struct sk_buff *next = skb->next;
1124                 int frag_len = skb->len - mutable->tunnel_hlen;
1125                 int err;
1126
1127                 skb->next = NULL;
1128                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1129
1130                 err = ip_local_out(skb);
1131                 skb = next;
1132                 if (unlikely(net_xmit_eval(err)))
1133                         goto free_frags;
1134                 sent_len += frag_len;
1135         }
1136
1137         return sent_len;
1138
1139 free_frags:
1140         /*
1141          * There's no point in continuing to send fragments once one has been
1142          * dropped so just free the rest.  This may help improve the congestion
1143          * that caused the first packet to be dropped.
1144          */
1145         tnl_free_linked_skbs(skb);
1146         return sent_len;
1147 }
1148
1149 int tnl_send(struct vport *vport, struct sk_buff *skb)
1150 {
1151         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1152         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1153
1154         enum vport_err_type err = VPORT_E_TX_ERROR;
1155         struct rtable *rt;
1156         struct dst_entry *unattached_dst = NULL;
1157         struct tnl_cache *cache;
1158         int sent_len = 0;
1159         __be16 frag_off = 0;
1160         u8 ttl;
1161         u8 inner_tos;
1162         u8 tos;
1163
1164         /* Validate the protocol headers before we try to use them. */
1165         if (skb->protocol == htons(ETH_P_8021Q) &&
1166             !vlan_tx_tag_present(skb)) {
1167                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1168                         goto error_free;
1169
1170                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1171                 skb_set_network_header(skb, VLAN_ETH_HLEN);
1172         }
1173
1174         if (skb->protocol == htons(ETH_P_IP)) {
1175                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1176                     + sizeof(struct iphdr))))
1177                         skb->protocol = 0;
1178         }
1179 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1180         else if (skb->protocol == htons(ETH_P_IPV6)) {
1181                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1182                     + sizeof(struct ipv6hdr))))
1183                         skb->protocol = 0;
1184         }
1185 #endif
1186
1187         /* ToS */
1188         if (skb->protocol == htons(ETH_P_IP))
1189                 inner_tos = ip_hdr(skb)->tos;
1190 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1191         else if (skb->protocol == htons(ETH_P_IPV6))
1192                 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1193 #endif
1194         else
1195                 inner_tos = 0;
1196
1197         if (mutable->flags & TNL_F_TOS_INHERIT)
1198                 tos = inner_tos;
1199         else
1200                 tos = mutable->tos;
1201
1202         tos = INET_ECN_encapsulate(tos, inner_tos);
1203
1204         /* Route lookup */
1205         rt = find_route(vport, mutable, tos, &cache);
1206         if (unlikely(!rt))
1207                 goto error_free;
1208         if (unlikely(!cache))
1209                 unattached_dst = &rt_dst(rt);
1210
1211         /* Reset SKB */
1212         nf_reset(skb);
1213         secpath_reset(skb);
1214         skb_dst_drop(skb);
1215         skb_clear_rxhash(skb);
1216
1217         /* Offloading */
1218         skb = handle_offloads(skb, mutable, rt);
1219         if (IS_ERR(skb))
1220                 goto error;
1221
1222         /* MTU */
1223         if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1224                 err = VPORT_E_TX_DROPPED;
1225                 goto error_free;
1226         }
1227
1228         /*
1229          * If we are over the MTU, allow the IP stack to handle fragmentation.
1230          * Fragmentation is a slow path anyways.
1231          */
1232         if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1233                      cache)) {
1234                 unattached_dst = &rt_dst(rt);
1235                 dst_hold(unattached_dst);
1236                 cache = NULL;
1237         }
1238
1239         /* TTL */
1240         ttl = mutable->ttl;
1241         if (!ttl)
1242                 ttl = ip4_dst_hoplimit(&rt_dst(rt));
1243
1244         if (mutable->flags & TNL_F_TTL_INHERIT) {
1245                 if (skb->protocol == htons(ETH_P_IP))
1246                         ttl = ip_hdr(skb)->ttl;
1247 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1248                 else if (skb->protocol == htons(ETH_P_IPV6))
1249                         ttl = ipv6_hdr(skb)->hop_limit;
1250 #endif
1251         }
1252
1253         while (skb) {
1254                 struct iphdr *iph;
1255                 struct sk_buff *next_skb = skb->next;
1256                 skb->next = NULL;
1257
1258                 if (unlikely(vlan_deaccel_tag(skb)))
1259                         goto next;
1260
1261                 if (likely(cache)) {
1262                         skb_push(skb, cache->len);
1263                         memcpy(skb->data, get_cached_header(cache), cache->len);
1264                         skb_reset_mac_header(skb);
1265                         skb_set_network_header(skb, cache->hh_len);
1266
1267                 } else {
1268                         skb_push(skb, mutable->tunnel_hlen);
1269                         create_tunnel_header(vport, mutable, rt, skb->data);
1270                         skb_reset_network_header(skb);
1271
1272                         if (next_skb)
1273                                 skb_dst_set(skb, dst_clone(unattached_dst));
1274                         else {
1275                                 skb_dst_set(skb, unattached_dst);
1276                                 unattached_dst = NULL;
1277                         }
1278                 }
1279                 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1280
1281                 iph = ip_hdr(skb);
1282                 iph->tos = tos;
1283                 iph->ttl = ttl;
1284                 iph->frag_off = frag_off;
1285                 ip_select_ident(iph, &rt_dst(rt), NULL);
1286
1287                 skb = tnl_vport->tnl_ops->update_header(vport, mutable,
1288                                                         &rt_dst(rt), skb);
1289                 if (unlikely(!skb))
1290                         goto next;
1291
1292                 if (likely(cache)) {
1293                         int orig_len = skb->len - cache->len;
1294                         struct vport *cache_vport;
1295
1296                         cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1297                         skb->protocol = htons(ETH_P_IP);
1298                         iph = ip_hdr(skb);
1299                         iph->tot_len = htons(skb->len - skb_network_offset(skb));
1300                         ip_send_check(iph);
1301
1302                         if (cache_vport) {
1303                                 if (unlikely(compute_ip_summed(skb, true))) {
1304                                         kfree_skb(skb);
1305                                         goto next;
1306                                 }
1307
1308                                 OVS_CB(skb)->flow = cache->flow;
1309                                 vport_receive(cache_vport, skb);
1310                                 sent_len += orig_len;
1311                         } else {
1312                                 int xmit_err;
1313
1314                                 skb->dev = rt_dst(rt).dev;
1315                                 xmit_err = dev_queue_xmit(skb);
1316
1317                                 if (likely(net_xmit_eval(xmit_err) == 0))
1318                                         sent_len += orig_len;
1319                         }
1320                 } else
1321                         sent_len += send_frags(skb, mutable);
1322
1323 next:
1324                 skb = next_skb;
1325         }
1326
1327         if (unlikely(sent_len == 0))
1328                 vport_record_error(vport, VPORT_E_TX_DROPPED);
1329
1330         goto out;
1331
1332 error_free:
1333         tnl_free_linked_skbs(skb);
1334 error:
1335         vport_record_error(vport, err);
1336 out:
1337         dst_release(unattached_dst);
1338         return sent_len;
1339 }
1340
1341 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1342         [OVS_TUNNEL_ATTR_FLAGS]    = { .type = NLA_U32 },
1343         [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1344         [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1345         [OVS_TUNNEL_ATTR_OUT_KEY]  = { .type = NLA_U64 },
1346         [OVS_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
1347         [OVS_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
1348         [OVS_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
1349 };
1350
1351 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
1352  * zeroed. */
1353 static int tnl_set_config(struct nlattr *options, const struct tnl_ops *tnl_ops,
1354                           const struct vport *cur_vport,
1355                           struct tnl_mutable_config *mutable)
1356 {
1357         const struct vport *old_vport;
1358         const struct tnl_mutable_config *old_mutable;
1359         struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1360         int err;
1361
1362         if (!options)
1363                 return -EINVAL;
1364
1365         err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1366         if (err)
1367                 return err;
1368
1369         if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1370                 return -EINVAL;
1371
1372         mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1373
1374         mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1375         if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
1376                 if (ipv4_is_multicast(mutable->key.daddr))
1377                         return -EINVAL;
1378                 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1379         }
1380
1381         if (a[OVS_TUNNEL_ATTR_TOS]) {
1382                 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1383                 if (mutable->tos != RT_TOS(mutable->tos))
1384                         return -EINVAL;
1385         }
1386
1387         if (a[OVS_TUNNEL_ATTR_TTL])
1388                 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1389
1390         mutable->key.tunnel_type = tnl_ops->tunnel_type;
1391         if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1392                 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1393                 mutable->flags |= TNL_F_IN_KEY_MATCH;
1394         } else {
1395                 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1396                 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1397         }
1398
1399         if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1400                 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1401         else
1402                 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1403
1404         mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1405         if (mutable->tunnel_hlen < 0)
1406                 return mutable->tunnel_hlen;
1407
1408         mutable->tunnel_hlen += sizeof(struct iphdr);
1409
1410         old_vport = port_table_lookup(&mutable->key, &old_mutable);
1411         if (old_vport && old_vport != cur_vport)
1412                 return -EEXIST;
1413
1414         mutable->mlink = 0;
1415         if (ipv4_is_multicast(mutable->key.daddr)) {
1416                 struct net_device *dev;
1417                 struct rtable *rt;
1418
1419                 rt = __find_route(mutable, tnl_ops->ipproto, mutable->tos);
1420                 if (IS_ERR(rt))
1421                         return -EADDRNOTAVAIL;
1422                 dev = rt_dst(rt).dev;
1423                 ip_rt_put(rt);
1424                 if (__in_dev_get_rtnl(dev) == NULL)
1425                         return -EADDRNOTAVAIL;
1426                 mutable->mlink = dev->ifindex;
1427                 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
1428         }
1429
1430         return 0;
1431 }
1432
1433 struct vport *tnl_create(const struct vport_parms *parms,
1434                          const struct vport_ops *vport_ops,
1435                          const struct tnl_ops *tnl_ops)
1436 {
1437         struct vport *vport;
1438         struct tnl_vport *tnl_vport;
1439         struct tnl_mutable_config *mutable;
1440         int initial_frag_id;
1441         int err;
1442
1443         vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1444         if (IS_ERR(vport)) {
1445                 err = PTR_ERR(vport);
1446                 goto error;
1447         }
1448
1449         tnl_vport = tnl_vport_priv(vport);
1450
1451         strcpy(tnl_vport->name, parms->name);
1452         tnl_vport->tnl_ops = tnl_ops;
1453
1454         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1455         if (!mutable) {
1456                 err = -ENOMEM;
1457                 goto error_free_vport;
1458         }
1459
1460         vport_gen_rand_ether_addr(mutable->eth_addr);
1461
1462         get_random_bytes(&initial_frag_id, sizeof(int));
1463         atomic_set(&tnl_vport->frag_id, initial_frag_id);
1464
1465         err = tnl_set_config(parms->options, tnl_ops, NULL, mutable);
1466         if (err)
1467                 goto error_free_mutable;
1468
1469         spin_lock_init(&tnl_vport->cache_lock);
1470
1471 #ifdef NEED_CACHE_TIMEOUT
1472         tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1473                                        (net_random() % (MAX_CACHE_EXP / 2));
1474 #endif
1475
1476         rcu_assign_pointer(tnl_vport->mutable, mutable);
1477
1478         port_table_add_port(vport);
1479         return vport;
1480
1481 error_free_mutable:
1482         free_mutable_rtnl(mutable);
1483         kfree(mutable);
1484 error_free_vport:
1485         vport_free(vport);
1486 error:
1487         return ERR_PTR(err);
1488 }
1489
1490 int tnl_set_options(struct vport *vport, struct nlattr *options)
1491 {
1492         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1493         const struct tnl_mutable_config *old_mutable;
1494         struct tnl_mutable_config *mutable;
1495         int err;
1496
1497         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1498         if (!mutable) {
1499                 err = -ENOMEM;
1500                 goto error;
1501         }
1502
1503         /* Copy fields whose values should be retained. */
1504         old_mutable = rtnl_dereference(tnl_vport->mutable);
1505         mutable->seq = old_mutable->seq + 1;
1506         memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1507
1508         /* Parse the others configured by userspace. */
1509         err = tnl_set_config(options, tnl_vport->tnl_ops, vport, mutable);
1510         if (err)
1511                 goto error_free;
1512
1513         if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1514                 port_table_move_port(vport, mutable);
1515         else
1516                 assign_config_rcu(vport, mutable);
1517
1518         return 0;
1519
1520 error_free:
1521         free_mutable_rtnl(mutable);
1522         kfree(mutable);
1523 error:
1524         return err;
1525 }
1526
1527 int tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1528 {
1529         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1530         const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1531
1532         NLA_PUT_U32(skb, OVS_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC);
1533         NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr);
1534
1535         if (!(mutable->flags & TNL_F_IN_KEY_MATCH))
1536                 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key);
1537         if (!(mutable->flags & TNL_F_OUT_KEY_ACTION))
1538                 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key);
1539         if (mutable->key.saddr)
1540                 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr);
1541         if (mutable->tos)
1542                 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos);
1543         if (mutable->ttl)
1544                 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl);
1545
1546         return 0;
1547
1548 nla_put_failure:
1549         return -EMSGSIZE;
1550 }
1551
1552 static void free_port_rcu(struct rcu_head *rcu)
1553 {
1554         struct tnl_vport *tnl_vport = container_of(rcu,
1555                                                    struct tnl_vport, rcu);
1556
1557         free_cache((struct tnl_cache __force *)tnl_vport->cache);
1558         kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1559         vport_free(tnl_vport_to_vport(tnl_vport));
1560 }
1561
1562 void tnl_destroy(struct vport *vport)
1563 {
1564         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1565         struct tnl_mutable_config *mutable;
1566
1567         mutable = rtnl_dereference(tnl_vport->mutable);
1568         port_table_remove_port(vport);
1569         free_mutable_rtnl(mutable);
1570         call_rcu(&tnl_vport->rcu, free_port_rcu);
1571 }
1572
1573 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1574 {
1575         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1576         struct tnl_mutable_config *old_mutable, *mutable;
1577
1578         old_mutable = rtnl_dereference(tnl_vport->mutable);
1579         mutable = kmemdup(old_mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1580         if (!mutable)
1581                 return -ENOMEM;
1582
1583         old_mutable->mlink = 0;
1584
1585         memcpy(mutable->eth_addr, addr, ETH_ALEN);
1586         assign_config_rcu(vport, mutable);
1587
1588         return 0;
1589 }
1590
1591 const char *tnl_get_name(const struct vport *vport)
1592 {
1593         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1594         return tnl_vport->name;
1595 }
1596
1597 const unsigned char *tnl_get_addr(const struct vport *vport)
1598 {
1599         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1600         return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1601 }
1602
1603 void tnl_free_linked_skbs(struct sk_buff *skb)
1604 {
1605         while (skb) {
1606                 struct sk_buff *next = skb->next;
1607                 kfree_skb(skb);
1608                 skb = next;
1609         }
1610 }
1611
1612 int tnl_init(void)
1613 {
1614         int i;
1615
1616         port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1617                         GFP_KERNEL);
1618         if (!port_table)
1619                 return -ENOMEM;
1620
1621         for (i = 0; i < PORT_TABLE_SIZE; i++)
1622                 INIT_HLIST_HEAD(&port_table[i]);
1623
1624         return 0;
1625 }
1626
1627 void tnl_exit(void)
1628 {
1629         int i;
1630
1631         for (i = 0; i < PORT_TABLE_SIZE; i++) {
1632                 struct tnl_vport *tnl_vport;
1633                 struct hlist_head *hash_head;
1634                 struct hlist_node *n;
1635
1636                 hash_head = &port_table[i];
1637                 hlist_for_each_entry(tnl_vport, n, hash_head, hash_node) {
1638                         BUG();
1639                         goto out;
1640                 }
1641         }
1642 out:
1643         kfree(port_table);
1644 }