datapath/tunnel.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/if_arp.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/ip.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/igmp.h>
  26 #include <linux/in.h>
  27 #include <linux/in_route.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/jhash.h>
  30 #include <linux/list.h>
  31 #include <linux/kernel.h>
  32 #include <linux/version.h>
  33 #include <linux/workqueue.h>
  34 #include <linux/rculist.h>
  35
  36 #include <net/dsfield.h>
  37 #include <net/dst.h>
  38 #include <net/icmp.h>
  39 #include <net/inet_ecn.h>
  40 #include <net/ip.h>
  41 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
  42 #include <net/ipv6.h>
  43 #endif
  44 #include <net/route.h>
  45 #include <net/xfrm.h>
  46
  47 #include "checksum.h"
  48 #include "datapath.h"
  49 #include "tunnel.h"
  50 #include "vlan.h"
  51 #include "vport.h"
  52 #include "vport-generic.h"
  53 #include "vport-internal_dev.h"
  54
  55 #ifdef NEED_CACHE_TIMEOUT
  56 /*
  57  * On kernels where we can't quickly detect changes in the rest of the system
  58  * we use an expiration time to invalidate the cache.  A shorter expiration
  59  * reduces the length of time that we may potentially blackhole packets while
  60  * a longer time increases performance by reducing the frequency that the
  61  * cache needs to be rebuilt.  A variety of factors may cause the cache to be
  62  * invalidated before the expiration time but this is the maximum.  The time
  63  * is expressed in jiffies.
  64  */
  65 #define MAX_CACHE_EXP HZ
  66 #endif
  67
  68 /*
  69  * Interval to check for and remove caches that are no longer valid.  Caches
  70  * are checked for validity before they are used for packet encapsulation and
  71  * old caches are removed at that time.  However, if no packets are sent through
  72  * the tunnel then the cache will never be destroyed.  Since it holds
  73  * references to a number of system objects, the cache will continue to use
  74  * system resources by not allowing those objects to be destroyed.  The cache
  75  * cleaner is periodically run to free invalid caches.  It does not
  76  * significantly affect system performance.  A lower interval will release
  77  * resources faster but will itself consume resources by requiring more frequent
  78  * checks.  A longer interval may result in messages being printed to the kernel
  79  * message buffer about unreleased resources.  The interval is expressed in
  80  * jiffies.
  81  */
  82 #define CACHE_CLEANER_INTERVAL (5 * HZ)
  83
  84 #define CACHE_DATA_ALIGN 16
  85 #define PORT_TABLE_SIZE  1024
  86
  87 static struct hlist_head *port_table __read_mostly;
  88 static int port_table_count;
  89
  90 static void cache_cleaner(struct work_struct *work);
  91 static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
  92
  93 /*
  94  * These are just used as an optimization: they don't require any kind of
  95  * synchronization because we could have just as easily read the value before
  96  * the port change happened.
  97  */
  98 static unsigned int key_local_remote_ports __read_mostly;
  99 static unsigned int key_remote_ports __read_mostly;
 100 static unsigned int key_multicast_ports __read_mostly;
 101 static unsigned int local_remote_ports __read_mostly;
 102 static unsigned int remote_ports __read_mostly;
 103 static unsigned int multicast_ports __read_mostly;
 104
 105 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
 106 #define rt_dst(rt) (rt->dst)
 107 #else
 108 #define rt_dst(rt) (rt->u.dst)
 109 #endif
 110
 111 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
 112 static struct hh_cache *rt_hh(struct rtable *rt)
 113 {
 114         struct neighbour *neigh = dst_get_neighbour_noref(&rt->dst);
 115         if (!neigh || !(neigh->nud_state & NUD_CONNECTED) ||
 116                         !neigh->hh.hh_len)
 117                 return NULL;
 118         return &neigh->hh;
 119 }
 120 #else
 121 #define rt_hh(rt) (rt_dst(rt).hh)
 122 #endif
 123
 124 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
 125 {
 126         return vport_from_priv(tnl_vport);
 127 }
 128
 129 /* This is analogous to rtnl_dereference for the tunnel cache.  It checks that
 130  * cache_lock is held, so it is only for update side code.
 131  */
 132 static struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
 133 {
 134         return rcu_dereference_protected(tnl_vport->cache,
 135                                  lockdep_is_held(&tnl_vport->cache_lock));
 136 }
 137
 138 static void schedule_cache_cleaner(void)
 139 {
 140         schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
 141 }
 142
 143 static void free_cache(struct tnl_cache *cache)
 144 {
 145         if (!cache)
 146                 return;
 147
 148         ovs_flow_put(cache->flow);
 149         ip_rt_put(cache->rt);
 150         kfree(cache);
 151 }
 152
 153 static void free_config_rcu(struct rcu_head *rcu)
 154 {
 155         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
 156         kfree(c);
 157 }
 158
 159 static void free_cache_rcu(struct rcu_head *rcu)
 160 {
 161         struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
 162         free_cache(c);
 163 }
 164
 165 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
 166  * within an RCU callback.  Fortunately this part doesn't require waiting for
 167  * an RCU grace period.
 168  */
 169 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
 170 {
 171         ASSERT_RTNL();
 172         if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
 173                 struct in_device *in_dev;
 174                 in_dev = inetdev_by_index(port_key_get_net(&mutable->key), mutable->mlink);
 175                 if (in_dev)
 176                         ip_mc_dec_group(in_dev, mutable->key.daddr);
 177         }
 178 }
 179
 180 static void assign_config_rcu(struct vport *vport,
 181                               struct tnl_mutable_config *new_config)
 182 {
 183         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 184         struct tnl_mutable_config *old_config;
 185
 186         old_config = rtnl_dereference(tnl_vport->mutable);
 187         rcu_assign_pointer(tnl_vport->mutable, new_config);
 188
 189         free_mutable_rtnl(old_config);
 190         call_rcu(&old_config->rcu, free_config_rcu);
 191 }
 192
 193 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
 194 {
 195         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 196         struct tnl_cache *old_cache;
 197
 198         old_cache = cache_dereference(tnl_vport);
 199         rcu_assign_pointer(tnl_vport->cache, new_cache);
 200
 201         if (old_cache)
 202                 call_rcu(&old_cache->rcu, free_cache_rcu);
 203 }
 204
 205 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 206 {
 207         bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
 208
 209         if (mutable->flags & TNL_F_IN_KEY_MATCH) {
 210                 if (mutable->key.saddr)
 211                         return &local_remote_ports;
 212                 else if (is_multicast)
 213                         return &multicast_ports;
 214                 else
 215                         return &remote_ports;
 216         } else {
 217                 if (mutable->key.saddr)
 218                         return &key_local_remote_ports;
 219                 else if (is_multicast)
 220                         return &key_multicast_ports;
 221                 else
 222                         return &key_remote_ports;
 223         }
 224 }
 225
 226 static u32 port_hash(const struct port_lookup_key *key)
 227 {
 228         return jhash2((u32 *)key, (PORT_KEY_LEN / sizeof(u32)), 0);
 229 }
 230
 231 static struct hlist_head *find_bucket(u32 hash)
 232 {
 233         return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
 234 }
 235
 236 static void port_table_add_port(struct vport *vport)
 237 {
 238         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 239         const struct tnl_mutable_config *mutable;
 240         u32 hash;
 241
 242         if (port_table_count == 0)
 243                 schedule_cache_cleaner();
 244
 245         mutable = rtnl_dereference(tnl_vport->mutable);
 246         hash = port_hash(&mutable->key);
 247         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 248         port_table_count++;
 249
 250         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 251 }
 252
 253 static void port_table_move_port(struct vport *vport,
 254                       struct tnl_mutable_config *new_mutable)
 255 {
 256         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 257         u32 hash;
 258
 259         hash = port_hash(&new_mutable->key);
 260         hlist_del_init_rcu(&tnl_vport->hash_node);
 261         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 262
 263         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 264         assign_config_rcu(vport, new_mutable);
 265         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 266 }
 267
 268 static void port_table_remove_port(struct vport *vport)
 269 {
 270         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 271
 272         hlist_del_init_rcu(&tnl_vport->hash_node);
 273
 274         port_table_count--;
 275         if (port_table_count == 0)
 276                 cancel_delayed_work_sync(&cache_cleaner_wq);
 277
 278         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 279 }
 280
 281 static struct vport *port_table_lookup(struct port_lookup_key *key,
 282                                        const struct tnl_mutable_config **pmutable)
 283 {
 284         struct hlist_node *n;
 285         struct hlist_head *bucket;
 286         u32 hash = port_hash(key);
 287         struct tnl_vport *tnl_vport;
 288
 289         bucket = find_bucket(hash);
 290
 291         hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
 292                 struct tnl_mutable_config *mutable;
 293
 294                 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 295                 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
 296                         *pmutable = mutable;
 297                         return tnl_vport_to_vport(tnl_vport);
 298                 }
 299         }
 300
 301         return NULL;
 302 }
 303
 304 struct vport *ovs_tnl_find_port(struct net *net, __be32 saddr, __be32 daddr,
 305                                 __be64 key, int tunnel_type,
 306                                 const struct tnl_mutable_config **mutable)
 307 {
 308         struct port_lookup_key lookup;
 309         struct vport *vport;
 310         bool is_multicast = ipv4_is_multicast(saddr);
 311
 312         port_key_set_net(&lookup, net);
 313         lookup.saddr = saddr;
 314         lookup.daddr = daddr;
 315
 316         /* First try for exact match on in_key. */
 317         lookup.in_key = key;
 318         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 319         if (!is_multicast && key_local_remote_ports) {
 320                 vport = port_table_lookup(&lookup, mutable);
 321                 if (vport)
 322                         return vport;
 323         }
 324         if (key_remote_ports) {
 325                 lookup.saddr = 0;
 326                 vport = port_table_lookup(&lookup, mutable);
 327                 if (vport)
 328                         return vport;
 329
 330                 lookup.saddr = saddr;
 331         }
 332
 333         /* Then try matches that wildcard in_key. */
 334         lookup.in_key = 0;
 335         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 336         if (!is_multicast && local_remote_ports) {
 337                 vport = port_table_lookup(&lookup, mutable);
 338                 if (vport)
 339                         return vport;
 340         }
 341         if (remote_ports) {
 342                 lookup.saddr = 0;
 343                 vport = port_table_lookup(&lookup, mutable);
 344                 if (vport)
 345                         return vport;
 346         }
 347
 348         if (is_multicast) {
 349                 lookup.saddr = 0;
 350                 lookup.daddr = saddr;
 351                 if (key_multicast_ports) {
 352                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 353                         lookup.in_key = key;
 354                         vport = port_table_lookup(&lookup, mutable);
 355                         if (vport)
 356                                 return vport;
 357                 }
 358                 if (multicast_ports) {
 359                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 360                         lookup.in_key = 0;
 361                         vport = port_table_lookup(&lookup, mutable);
 362                         if (vport)
 363                                 return vport;
 364                 }
 365         }
 366
 367         return NULL;
 368 }
 369
 370 static void ecn_decapsulate(struct sk_buff *skb, u8 tos)
 371 {
 372         if (unlikely(INET_ECN_is_ce(tos))) {
 373                 __be16 protocol = skb->protocol;
 374
 375                 skb_set_network_header(skb, ETH_HLEN);
 376
 377                 if (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                         skb_set_network_header(skb, VLAN_ETH_HLEN);
 383                 }
 384
 385                 if (protocol == htons(ETH_P_IP)) {
 386                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 387                             + sizeof(struct iphdr))))
 388                                 return;
 389
 390                         IP_ECN_set_ce(ip_hdr(skb));
 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, skb_network_offset(skb)
 395                             + sizeof(struct ipv6hdr))))
 396                                 return;
 397
 398                         IP6_ECN_set_ce(ipv6_hdr(skb));
 399                 }
 400 #endif
 401         }
 402 }
 403
 404 /**
 405  *      ovs_tnl_rcv - ingress point for generic tunnel code
 406  *
 407  * @vport: port this packet was received on
 408  * @skb: received packet
 409  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
 410  *
 411  * Must be called with rcu_read_lock.
 412  *
 413  * Packets received by this function are in the following state:
 414  * - skb->data points to the inner Ethernet header.
 415  * - The inner Ethernet header is in the linear data area.
 416  * - skb->csum does not include the inner Ethernet header.
 417  * - The layer pointers are undefined.
 418  */
 419 void ovs_tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos)
 420 {
 421         struct ethhdr *eh;
 422
 423         skb_reset_mac_header(skb);
 424         eh = eth_hdr(skb);
 425
 426         if (likely(ntohs(eh->h_proto) >= 1536))
 427                 skb->protocol = eh->h_proto;
 428         else
 429                 skb->protocol = htons(ETH_P_802_2);
 430
 431         skb_dst_drop(skb);
 432         nf_reset(skb);
 433         skb_clear_rxhash(skb);
 434         secpath_reset(skb);
 435
 436         ecn_decapsulate(skb, tos);
 437         vlan_set_tci(skb, 0);
 438
 439         if (unlikely(compute_ip_summed(skb, false))) {
 440                 kfree_skb(skb);
 441                 return;
 442         }
 443
 444         ovs_vport_receive(vport, skb);
 445 }
 446
 447 static bool check_ipv4_address(__be32 addr)
 448 {
 449         if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
 450             || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
 451                 return false;
 452
 453         return true;
 454 }
 455
 456 static bool ipv4_should_icmp(struct sk_buff *skb)
 457 {
 458         struct iphdr *old_iph = ip_hdr(skb);
 459
 460         /* Don't respond to L2 broadcast. */
 461         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
 462                 return false;
 463
 464         /* Don't respond to L3 broadcast or invalid addresses. */
 465         if (!check_ipv4_address(old_iph->daddr) ||
 466             !check_ipv4_address(old_iph->saddr))
 467                 return false;
 468
 469         /* Only respond to the first fragment. */
 470         if (old_iph->frag_off & htons(IP_OFFSET))
 471                 return false;
 472
 473         /* Don't respond to ICMP error messages. */
 474         if (old_iph->protocol == IPPROTO_ICMP) {
 475                 u8 icmp_type, *icmp_typep;
 476
 477                 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
 478                                                 (old_iph->ihl << 2) +
 479                                                 offsetof(struct icmphdr, type) -
 480                                                 skb->data, sizeof(icmp_type),
 481                                                 &icmp_type);
 482
 483                 if (!icmp_typep)
 484                         return false;
 485
 486                 if (*icmp_typep > NR_ICMP_TYPES
 487                         || (*icmp_typep <= ICMP_PARAMETERPROB
 488                                 && *icmp_typep != ICMP_ECHOREPLY
 489                                 && *icmp_typep != ICMP_ECHO))
 490                         return false;
 491         }
 492
 493         return true;
 494 }
 495
 496 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 497                             unsigned int mtu, unsigned int payload_length)
 498 {
 499         struct iphdr *iph, *old_iph = ip_hdr(skb);
 500         struct icmphdr *icmph;
 501         u8 *payload;
 502
 503         iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
 504         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 505         payload = skb_put(nskb, payload_length);
 506
 507         /* IP */
 508         iph->version            =       4;
 509         iph->ihl                =       sizeof(struct iphdr) >> 2;
 510         iph->tos                =       (old_iph->tos & IPTOS_TOS_MASK) |
 511                                         IPTOS_PREC_INTERNETCONTROL;
 512         iph->tot_len            =       htons(sizeof(struct iphdr)
 513                                               + sizeof(struct icmphdr)
 514                                               + payload_length);
 515         get_random_bytes(&iph->id, sizeof(iph->id));
 516         iph->frag_off           =       0;
 517         iph->ttl                =       IPDEFTTL;
 518         iph->protocol           =       IPPROTO_ICMP;
 519         iph->daddr              =       old_iph->saddr;
 520         iph->saddr              =       old_iph->daddr;
 521
 522         ip_send_check(iph);
 523
 524         /* ICMP */
 525         icmph->type             =       ICMP_DEST_UNREACH;
 526         icmph->code             =       ICMP_FRAG_NEEDED;
 527         icmph->un.gateway       =       htonl(mtu);
 528         icmph->checksum         =       0;
 529
 530         nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
 531         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
 532                                             payload, payload_length,
 533                                             nskb->csum);
 534         icmph->checksum = csum_fold(nskb->csum);
 535 }
 536
 537 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 538 static bool ipv6_should_icmp(struct sk_buff *skb)
 539 {
 540         struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
 541         int addr_type;
 542         int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
 543         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
 544         __be16 frag_off;
 545
 546         /* Check source address is valid. */
 547         addr_type = ipv6_addr_type(&old_ipv6h->saddr);
 548         if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
 549                 return false;
 550
 551         /* Don't reply to unspecified addresses. */
 552         if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
 553                 return false;
 554
 555         /* Don't respond to ICMP error messages. */
 556         payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr, &frag_off);
 557         if (payload_off < 0)
 558                 return false;
 559
 560         if (nexthdr == NEXTHDR_ICMP) {
 561                 u8 icmp_type, *icmp_typep;
 562
 563                 icmp_typep = skb_header_pointer(skb, payload_off +
 564                                                 offsetof(struct icmp6hdr,
 565                                                         icmp6_type),
 566                                                 sizeof(icmp_type), &icmp_type);
 567
 568                 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
 569                         return false;
 570         }
 571
 572         return true;
 573 }
 574
 575 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
 576                             unsigned int mtu, unsigned int payload_length)
 577 {
 578         struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
 579         struct icmp6hdr *icmp6h;
 580         u8 *payload;
 581
 582         ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
 583         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
 584         payload = skb_put(nskb, payload_length);
 585
 586         /* IPv6 */
 587         ipv6h->version          =       6;
 588         ipv6h->priority         =       0;
 589         memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
 590         ipv6h->payload_len      =       htons(sizeof(struct icmp6hdr)
 591                                               + payload_length);
 592         ipv6h->nexthdr          =       NEXTHDR_ICMP;
 593         ipv6h->hop_limit        =       IPV6_DEFAULT_HOPLIMIT;
 594         ipv6h->daddr            =       old_ipv6h->saddr;
 595         ipv6h->saddr            =       old_ipv6h->daddr;
 596
 597         /* ICMPv6 */
 598         icmp6h->icmp6_type      =       ICMPV6_PKT_TOOBIG;
 599         icmp6h->icmp6_code      =       0;
 600         icmp6h->icmp6_cksum     =       0;
 601         icmp6h->icmp6_mtu       =       htonl(mtu);
 602
 603         nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
 604         nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
 605                                             payload, payload_length,
 606                                             nskb->csum);
 607         icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 608                                                 sizeof(struct icmp6hdr)
 609                                                 + payload_length,
 610                                                 ipv6h->nexthdr, nskb->csum);
 611 }
 612 #endif /* IPv6 */
 613
 614 bool ovs_tnl_frag_needed(struct vport *vport,
 615                          const struct tnl_mutable_config *mutable,
 616                          struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
 617 {
 618         unsigned int eth_hdr_len = ETH_HLEN;
 619         unsigned int total_length = 0, header_length = 0, payload_length;
 620         struct ethhdr *eh, *old_eh = eth_hdr(skb);
 621         struct sk_buff *nskb;
 622
 623         /* Sanity check */
 624         if (skb->protocol == htons(ETH_P_IP)) {
 625                 if (mtu < IP_MIN_MTU)
 626                         return false;
 627
 628                 if (!ipv4_should_icmp(skb))
 629                         return true;
 630         }
 631 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 632         else if (skb->protocol == htons(ETH_P_IPV6)) {
 633                 if (mtu < IPV6_MIN_MTU)
 634                         return false;
 635
 636                 /*
 637                  * In theory we should do PMTUD on IPv6 multicast messages but
 638                  * we don't have an address to send from so just fragment.
 639                  */
 640                 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
 641                         return false;
 642
 643                 if (!ipv6_should_icmp(skb))
 644                         return true;
 645         }
 646 #endif
 647         else
 648                 return false;
 649
 650         /* Allocate */
 651         if (old_eh->h_proto == htons(ETH_P_8021Q))
 652                 eth_hdr_len = VLAN_ETH_HLEN;
 653
 654         payload_length = skb->len - eth_hdr_len;
 655         if (skb->protocol == htons(ETH_P_IP)) {
 656                 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
 657                 total_length = min_t(unsigned int, header_length +
 658                                                    payload_length, 576);
 659         }
 660 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 661         else {
 662                 header_length = sizeof(struct ipv6hdr) +
 663                                 sizeof(struct icmp6hdr);
 664                 total_length = min_t(unsigned int, header_length +
 665                                                   payload_length, IPV6_MIN_MTU);
 666         }
 667 #endif
 668
 669         payload_length = total_length - header_length;
 670
 671         nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
 672                              payload_length);
 673         if (!nskb)
 674                 return false;
 675
 676         skb_reserve(nskb, NET_IP_ALIGN);
 677
 678         /* Ethernet / VLAN */
 679         eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
 680         memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
 681         memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
 682         nskb->protocol = eh->h_proto = old_eh->h_proto;
 683         if (old_eh->h_proto == htons(ETH_P_8021Q)) {
 684                 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
 685
 686                 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
 687                 vh->h_vlan_encapsulated_proto = skb->protocol;
 688         } else
 689                 vlan_set_tci(nskb, vlan_get_tci(skb));
 690         skb_reset_mac_header(nskb);
 691
 692         /* Protocol */
 693         if (skb->protocol == htons(ETH_P_IP))
 694                 ipv4_build_icmp(skb, nskb, mtu, payload_length);
 695 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 696         else
 697                 ipv6_build_icmp(skb, nskb, mtu, payload_length);
 698 #endif
 699
 700         /*
 701          * Assume that flow based keys are symmetric with respect to input
 702          * and output and use the key that we were going to put on the
 703          * outgoing packet for the fake received packet.  If the keys are
 704          * not symmetric then PMTUD needs to be disabled since we won't have
 705          * any way of synthesizing packets.
 706          */
 707         if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
 708             (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
 709                 OVS_CB(nskb)->tun_id = flow_key;
 710
 711         if (unlikely(compute_ip_summed(nskb, false))) {
 712                 kfree_skb(nskb);
 713                 return false;
 714         }
 715
 716         ovs_vport_receive(vport, nskb);
 717
 718         return true;
 719 }
 720
 721 static bool check_mtu(struct sk_buff *skb,
 722                       struct vport *vport,
 723                       const struct tnl_mutable_config *mutable,
 724                       const struct rtable *rt, __be16 *frag_offp)
 725 {
 726         bool df_inherit = mutable->flags & TNL_F_DF_INHERIT;
 727         bool pmtud = mutable->flags & TNL_F_PMTUD;
 728         __be16 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
 729         int mtu = 0;
 730         unsigned int packet_length = skb->len - ETH_HLEN;
 731
 732         /* Allow for one level of tagging in the packet length. */
 733         if (!vlan_tx_tag_present(skb) &&
 734             eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 735                 packet_length -= VLAN_HLEN;
 736
 737         if (pmtud) {
 738                 int vlan_header = 0;
 739
 740                 /* The tag needs to go in packet regardless of where it
 741                  * currently is, so subtract it from the MTU.
 742                  */
 743                 if (vlan_tx_tag_present(skb) ||
 744                     eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
 745                         vlan_header = VLAN_HLEN;
 746
 747                 mtu = dst_mtu(&rt_dst(rt))
 748                         - ETH_HLEN
 749                         - mutable->tunnel_hlen
 750                         - vlan_header;
 751         }
 752
 753         if (skb->protocol == htons(ETH_P_IP)) {
 754                 struct iphdr *iph = ip_hdr(skb);
 755
 756                 if (df_inherit)
 757                         frag_off = iph->frag_off & htons(IP_DF);
 758
 759                 if (pmtud && iph->frag_off & htons(IP_DF)) {
 760                         mtu = max(mtu, IP_MIN_MTU);
 761
 762                         if (packet_length > mtu &&
 763                             ovs_tnl_frag_needed(vport, mutable, skb, mtu,
 764                                                 OVS_CB(skb)->tun_id))
 765                                 return false;
 766                 }
 767         }
 768 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 769         else if (skb->protocol == htons(ETH_P_IPV6)) {
 770                 /* IPv6 requires end hosts to do fragmentation
 771                  * if the packet is above the minimum MTU.
 772                  */
 773                 if (df_inherit && packet_length > IPV6_MIN_MTU)
 774                         frag_off = htons(IP_DF);
 775
 776                 if (pmtud) {
 777                         mtu = max(mtu, IPV6_MIN_MTU);
 778
 779                         if (packet_length > mtu &&
 780                             ovs_tnl_frag_needed(vport, mutable, skb, mtu,
 781                                                 OVS_CB(skb)->tun_id))
 782                                 return false;
 783                 }
 784         }
 785 #endif
 786
 787         *frag_offp = frag_off;
 788         return true;
 789 }
 790
 791 static void create_tunnel_header(const struct vport *vport,
 792                                  const struct tnl_mutable_config *mutable,
 793                                  const struct rtable *rt, void *header)
 794 {
 795         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 796         struct iphdr *iph = header;
 797
 798         iph->version    = 4;
 799         iph->ihl        = sizeof(struct iphdr) >> 2;
 800         iph->frag_off   = htons(IP_DF);
 801         iph->protocol   = tnl_vport->tnl_ops->ipproto;
 802         iph->tos        = mutable->tos;
 803         iph->daddr      = rt->rt_dst;
 804         iph->saddr      = rt->rt_src;
 805         iph->ttl        = mutable->ttl;
 806         if (!iph->ttl)
 807                 iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
 808
 809         tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
 810 }
 811
 812 static void *get_cached_header(const struct tnl_cache *cache)
 813 {
 814         return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
 815 }
 816
 817 #ifdef HAVE_RT_GENID
 818 static inline int rt_genid(struct net *net)
 819 {
 820         return atomic_read(&net->ipv4.rt_genid);
 821 }
 822 #endif
 823
 824 static bool check_cache_valid(const struct tnl_cache *cache,
 825                               const struct tnl_mutable_config *mutable)
 826 {
 827         struct hh_cache *hh;
 828
 829         if (!cache)
 830                 return false;
 831
 832         hh = rt_hh(cache->rt);
 833         return hh &&
 834 #ifdef NEED_CACHE_TIMEOUT
 835                 time_before(jiffies, cache->expiration) &&
 836 #endif
 837 #ifdef HAVE_RT_GENID
 838                 rt_genid(dev_net(rt_dst(cache->rt).dev)) == cache->rt->rt_genid &&
 839 #endif
 840 #ifdef HAVE_HH_SEQ
 841                 hh->hh_lock.sequence == cache->hh_seq &&
 842 #endif
 843                 mutable->seq == cache->mutable_seq &&
 844                 (!ovs_is_internal_dev(rt_dst(cache->rt).dev) ||
 845                 (cache->flow && !cache->flow->dead));
 846 }
 847
 848 static void __cache_cleaner(struct tnl_vport *tnl_vport)
 849 {
 850         const struct tnl_mutable_config *mutable =
 851                         rcu_dereference(tnl_vport->mutable);
 852         const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
 853
 854         if (cache && !check_cache_valid(cache, mutable) &&
 855             spin_trylock_bh(&tnl_vport->cache_lock)) {
 856                 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
 857                 spin_unlock_bh(&tnl_vport->cache_lock);
 858         }
 859 }
 860
 861 static void cache_cleaner(struct work_struct *work)
 862 {
 863         int i;
 864
 865         schedule_cache_cleaner();
 866
 867         rcu_read_lock();
 868         for (i = 0; i < PORT_TABLE_SIZE; i++) {
 869                 struct hlist_node *n;
 870                 struct hlist_head *bucket;
 871                 struct tnl_vport *tnl_vport;
 872
 873                 bucket = &port_table[i];
 874                 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node)
 875                         __cache_cleaner(tnl_vport);
 876         }
 877         rcu_read_unlock();
 878 }
 879
 880 static void create_eth_hdr(struct tnl_cache *cache, struct hh_cache *hh)
 881 {
 882         void *cache_data = get_cached_header(cache);
 883         int hh_off;
 884
 885 #ifdef HAVE_HH_SEQ
 886         unsigned hh_seq;
 887
 888         do {
 889                 hh_seq = read_seqbegin(&hh->hh_lock);
 890                 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
 891                 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
 892                 cache->hh_len = hh->hh_len;
 893         } while (read_seqretry(&hh->hh_lock, hh_seq));
 894
 895         cache->hh_seq = hh_seq;
 896 #else
 897         read_lock(&hh->hh_lock);
 898         hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
 899         memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
 900         cache->hh_len = hh->hh_len;
 901         read_unlock(&hh->hh_lock);
 902 #endif
 903 }
 904
 905 static struct tnl_cache *build_cache(struct vport *vport,
 906                                      const struct tnl_mutable_config *mutable,
 907                                      struct rtable *rt)
 908 {
 909         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 910         struct tnl_cache *cache;
 911         void *cache_data;
 912         int cache_len;
 913         struct hh_cache *hh;
 914
 915         if (!(mutable->flags & TNL_F_HDR_CACHE))
 916                 return NULL;
 917
 918         /*
 919          * If there is no entry in the ARP cache or if this device does not
 920          * support hard header caching just fall back to the IP stack.
 921          */
 922
 923         hh = rt_hh(rt);
 924         if (!hh)
 925                 return NULL;
 926
 927         /*
 928          * If lock is contended fall back to directly building the header.
 929          * We're not going to help performance by sitting here spinning.
 930          */
 931         if (!spin_trylock(&tnl_vport->cache_lock))
 932                 return NULL;
 933
 934         cache = cache_dereference(tnl_vport);
 935         if (check_cache_valid(cache, mutable))
 936                 goto unlock;
 937         else
 938                 cache = NULL;
 939
 940         cache_len = LL_RESERVED_SPACE(rt_dst(rt).dev) + mutable->tunnel_hlen;
 941
 942         cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
 943                         cache_len, GFP_ATOMIC);
 944         if (!cache)
 945                 goto unlock;
 946
 947         create_eth_hdr(cache, hh);
 948         cache_data = get_cached_header(cache) + cache->hh_len;
 949         cache->len = cache->hh_len + mutable->tunnel_hlen;
 950
 951         create_tunnel_header(vport, mutable, rt, cache_data);
 952
 953         cache->mutable_seq = mutable->seq;
 954         cache->rt = rt;
 955 #ifdef NEED_CACHE_TIMEOUT
 956         cache->expiration = jiffies + tnl_vport->cache_exp_interval;
 957 #endif
 958
 959         if (ovs_is_internal_dev(rt_dst(rt).dev)) {
 960                 struct sw_flow_key flow_key;
 961                 struct vport *dst_vport;
 962                 struct sk_buff *skb;
 963                 int err;
 964                 int flow_key_len;
 965                 struct sw_flow *flow;
 966
 967                 dst_vport = ovs_internal_dev_get_vport(rt_dst(rt).dev);
 968                 if (!dst_vport)
 969                         goto done;
 970
 971                 skb = alloc_skb(cache->len, GFP_ATOMIC);
 972                 if (!skb)
 973                         goto done;
 974
 975                 __skb_put(skb, cache->len);
 976                 memcpy(skb->data, get_cached_header(cache), cache->len);
 977
 978                 err = ovs_flow_extract(skb, dst_vport->port_no, &flow_key,
 979                                        &flow_key_len);
 980
 981                 consume_skb(skb);
 982                 if (err)
 983                         goto done;
 984
 985                 flow = ovs_flow_tbl_lookup(rcu_dereference(dst_vport->dp->table),
 986                                            &flow_key, flow_key_len);
 987                 if (flow) {
 988                         cache->flow = flow;
 989                         ovs_flow_hold(flow);
 990                 }
 991         }
 992
 993 done:
 994         assign_cache_rcu(vport, cache);
 995
 996 unlock:
 997         spin_unlock(&tnl_vport->cache_lock);
 998
 999         return cache;
1000 }
1001
1002 static struct rtable *__find_route(const struct tnl_mutable_config *mutable,
1003                                    u8 ipproto, u8 tos)
1004 {
1005         /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
1006          * router expect RT_TOS bits only. */
1007
1008 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
1009         struct flowi fl = { .nl_u = { .ip4_u = {
1010                                         .daddr = mutable->key.daddr,
1011                                         .saddr = mutable->key.saddr,
1012                                         .tos   = RT_TOS(tos) } },
1013                                         .proto = ipproto };
1014         struct rtable *rt;
1015
1016         if (unlikely(ip_route_output_key(port_key_get_net(&mutable->key), &rt, &fl)))
1017                 return ERR_PTR(-EADDRNOTAVAIL);
1018
1019         return rt;
1020 #else
1021         struct flowi4 fl = { .daddr = mutable->key.daddr,
1022                              .saddr = mutable->key.saddr,
1023                              .flowi4_tos = RT_TOS(tos),
1024                              .flowi4_proto = ipproto };
1025
1026         return ip_route_output_key(port_key_get_net(&mutable->key), &fl);
1027 #endif
1028 }
1029
1030 static struct rtable *find_route(struct vport *vport,
1031                                  const struct tnl_mutable_config *mutable,
1032                                  u8 tos, struct tnl_cache **cache)
1033 {
1034         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1035         struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
1036
1037         *cache = NULL;
1038         tos = RT_TOS(tos);
1039
1040         if (likely(tos == RT_TOS(mutable->tos) &&
1041             check_cache_valid(cur_cache, mutable))) {
1042                 *cache = cur_cache;
1043                 return cur_cache->rt;
1044         } else {
1045                 struct rtable *rt;
1046
1047                 rt = __find_route(mutable, tnl_vport->tnl_ops->ipproto, tos);
1048                 if (IS_ERR(rt))
1049                         return NULL;
1050
1051                 if (likely(tos == RT_TOS(mutable->tos)))
1052                         *cache = build_cache(vport, mutable, rt);
1053
1054                 return rt;
1055         }
1056 }
1057
1058 static bool need_linearize(const struct sk_buff *skb)
1059 {
1060         int i;
1061
1062         if (unlikely(skb_shinfo(skb)->frag_list))
1063                 return true;
1064
1065         /*
1066          * Generally speaking we should linearize if there are paged frags.
1067          * However, if all of the refcounts are 1 we know nobody else can
1068          * change them from underneath us and we can skip the linearization.
1069          */
1070         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1071                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
1072                         return true;
1073
1074         return false;
1075 }
1076
1077 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1078                                        const struct tnl_mutable_config *mutable,
1079                                        const struct rtable *rt)
1080 {
1081         int min_headroom;
1082         int err;
1083
1084         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1085                         + mutable->tunnel_hlen
1086                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1087
1088         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
1089                 int head_delta = SKB_DATA_ALIGN(min_headroom -
1090                                                 skb_headroom(skb) +
1091                                                 16);
1092                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
1093                                         0, GFP_ATOMIC);
1094                 if (unlikely(err))
1095                         goto error_free;
1096         }
1097
1098         forward_ip_summed(skb, true);
1099
1100         if (skb_is_gso(skb)) {
1101                 struct sk_buff *nskb;
1102
1103                 nskb = skb_gso_segment(skb, 0);
1104                 if (IS_ERR(nskb)) {
1105                         kfree_skb(skb);
1106                         err = PTR_ERR(nskb);
1107                         goto error;
1108                 }
1109
1110                 consume_skb(skb);
1111                 skb = nskb;
1112         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
1113                 /* Pages aren't locked and could change at any time.
1114                  * If this happens after we compute the checksum, the
1115                  * checksum will be wrong.  We linearize now to avoid
1116                  * this problem.
1117                  */
1118                 if (unlikely(need_linearize(skb))) {
1119                         err = __skb_linearize(skb);
1120                         if (unlikely(err))
1121                                 goto error_free;
1122                 }
1123
1124                 err = skb_checksum_help(skb);
1125                 if (unlikely(err))
1126                         goto error_free;
1127         }
1128
1129         set_ip_summed(skb, OVS_CSUM_NONE);
1130
1131         return skb;
1132
1133 error_free:
1134         kfree_skb(skb);
1135 error:
1136         return ERR_PTR(err);
1137 }
1138
1139 static int send_frags(struct sk_buff *skb,
1140                       const struct tnl_mutable_config *mutable)
1141 {
1142         int sent_len;
1143
1144         sent_len = 0;
1145         while (skb) {
1146                 struct sk_buff *next = skb->next;
1147                 int frag_len = skb->len - mutable->tunnel_hlen;
1148                 int err;
1149
1150                 skb->next = NULL;
1151                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1152
1153                 err = ip_local_out(skb);
1154                 skb = next;
1155                 if (unlikely(net_xmit_eval(err)))
1156                         goto free_frags;
1157                 sent_len += frag_len;
1158         }
1159
1160         return sent_len;
1161
1162 free_frags:
1163         /*
1164          * There's no point in continuing to send fragments once one has been
1165          * dropped so just free the rest.  This may help improve the congestion
1166          * that caused the first packet to be dropped.
1167          */
1168         ovs_tnl_free_linked_skbs(skb);
1169         return sent_len;
1170 }
1171
1172 int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
1173 {
1174         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1175         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1176
1177         enum vport_err_type err = VPORT_E_TX_ERROR;
1178         struct rtable *rt;
1179         struct dst_entry *unattached_dst = NULL;
1180         struct tnl_cache *cache;
1181         int sent_len = 0;
1182         __be16 frag_off = 0;
1183         u8 ttl;
1184         u8 inner_tos;
1185         u8 tos;
1186
1187         /* Validate the protocol headers before we try to use them. */
1188         if (skb->protocol == htons(ETH_P_8021Q) &&
1189             !vlan_tx_tag_present(skb)) {
1190                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1191                         goto error_free;
1192
1193                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1194                 skb_set_network_header(skb, VLAN_ETH_HLEN);
1195         }
1196
1197         if (skb->protocol == htons(ETH_P_IP)) {
1198                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1199                     + sizeof(struct iphdr))))
1200                         skb->protocol = 0;
1201         }
1202 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1203         else if (skb->protocol == htons(ETH_P_IPV6)) {
1204                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1205                     + sizeof(struct ipv6hdr))))
1206                         skb->protocol = 0;
1207         }
1208 #endif
1209
1210         /* ToS */
1211         if (skb->protocol == htons(ETH_P_IP))
1212                 inner_tos = ip_hdr(skb)->tos;
1213 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1214         else if (skb->protocol == htons(ETH_P_IPV6))
1215                 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1216 #endif
1217         else
1218                 inner_tos = 0;
1219
1220         if (mutable->flags & TNL_F_TOS_INHERIT)
1221                 tos = inner_tos;
1222         else
1223                 tos = mutable->tos;
1224
1225         /* Route lookup */
1226         rt = find_route(vport, mutable, tos, &cache);
1227         if (unlikely(!rt))
1228                 goto error_free;
1229         if (unlikely(!cache))
1230                 unattached_dst = &rt_dst(rt);
1231
1232         tos = INET_ECN_encapsulate(tos, inner_tos);
1233
1234         /* Reset SKB */
1235         nf_reset(skb);
1236         secpath_reset(skb);
1237         skb_dst_drop(skb);
1238         skb_clear_rxhash(skb);
1239
1240         /* Offloading */
1241         skb = handle_offloads(skb, mutable, rt);
1242         if (IS_ERR(skb))
1243                 goto error;
1244
1245         /* MTU */
1246         if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1247                 err = VPORT_E_TX_DROPPED;
1248                 goto error_free;
1249         }
1250
1251         /*
1252          * If we are over the MTU, allow the IP stack to handle fragmentation.
1253          * Fragmentation is a slow path anyways.
1254          */
1255         if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1256                      cache)) {
1257                 unattached_dst = &rt_dst(rt);
1258                 dst_hold(unattached_dst);
1259                 cache = NULL;
1260         }
1261
1262         /* TTL */
1263         ttl = mutable->ttl;
1264         if (!ttl)
1265                 ttl = ip4_dst_hoplimit(&rt_dst(rt));
1266
1267         if (mutable->flags & TNL_F_TTL_INHERIT) {
1268                 if (skb->protocol == htons(ETH_P_IP))
1269                         ttl = ip_hdr(skb)->ttl;
1270 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1271                 else if (skb->protocol == htons(ETH_P_IPV6))
1272                         ttl = ipv6_hdr(skb)->hop_limit;
1273 #endif
1274         }
1275
1276         while (skb) {
1277                 struct iphdr *iph;
1278                 struct sk_buff *next_skb = skb->next;
1279                 skb->next = NULL;
1280
1281                 if (unlikely(vlan_deaccel_tag(skb)))
1282                         goto next;
1283
1284                 if (likely(cache)) {
1285                         skb_push(skb, cache->len);
1286                         memcpy(skb->data, get_cached_header(cache), cache->len);
1287                         skb_reset_mac_header(skb);
1288                         skb_set_network_header(skb, cache->hh_len);
1289
1290                 } else {
1291                         skb_push(skb, mutable->tunnel_hlen);
1292                         create_tunnel_header(vport, mutable, rt, skb->data);
1293                         skb_reset_network_header(skb);
1294
1295                         if (next_skb)
1296                                 skb_dst_set(skb, dst_clone(unattached_dst));
1297                         else {
1298                                 skb_dst_set(skb, unattached_dst);
1299                                 unattached_dst = NULL;
1300                         }
1301                 }
1302                 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1303
1304                 iph = ip_hdr(skb);
1305                 iph->tos = tos;
1306                 iph->ttl = ttl;
1307                 iph->frag_off = frag_off;
1308                 ip_select_ident(iph, &rt_dst(rt), NULL);
1309
1310                 skb = tnl_vport->tnl_ops->update_header(vport, mutable,
1311                                                         &rt_dst(rt), skb);
1312                 if (unlikely(!skb))
1313                         goto next;
1314
1315                 if (likely(cache)) {
1316                         int orig_len = skb->len - cache->len;
1317                         struct vport *cache_vport;
1318
1319                         cache_vport = ovs_internal_dev_get_vport(rt_dst(rt).dev);
1320                         skb->protocol = htons(ETH_P_IP);
1321                         iph = ip_hdr(skb);
1322                         iph->tot_len = htons(skb->len - skb_network_offset(skb));
1323                         ip_send_check(iph);
1324
1325                         if (cache_vport) {
1326                                 if (unlikely(compute_ip_summed(skb, true))) {
1327                                         kfree_skb(skb);
1328                                         goto next;
1329                                 }
1330
1331                                 OVS_CB(skb)->flow = cache->flow;
1332                                 ovs_vport_receive(cache_vport, skb);
1333                                 sent_len += orig_len;
1334                         } else {
1335                                 int xmit_err;
1336
1337                                 skb->dev = rt_dst(rt).dev;
1338                                 xmit_err = dev_queue_xmit(skb);
1339
1340                                 if (likely(net_xmit_eval(xmit_err) == 0))
1341                                         sent_len += orig_len;
1342                         }
1343                 } else
1344                         sent_len += send_frags(skb, mutable);
1345
1346 next:
1347                 skb = next_skb;
1348         }
1349
1350         if (unlikely(sent_len == 0))
1351                 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
1352
1353         goto out;
1354
1355 error_free:
1356         ovs_tnl_free_linked_skbs(skb);
1357 error:
1358         ovs_vport_record_error(vport, err);
1359 out:
1360         dst_release(unattached_dst);
1361         return sent_len;
1362 }
1363
1364 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1365         [OVS_TUNNEL_ATTR_FLAGS]    = { .type = NLA_U32 },
1366         [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1367         [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1368         [OVS_TUNNEL_ATTR_OUT_KEY]  = { .type = NLA_U64 },
1369         [OVS_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
1370         [OVS_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
1371         [OVS_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
1372 };
1373
1374 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
1375  * zeroed. */
1376 static int tnl_set_config(struct net *net, struct nlattr *options,
1377                           const struct tnl_ops *tnl_ops,
1378                           const struct vport *cur_vport,
1379                           struct tnl_mutable_config *mutable)
1380 {
1381         const struct vport *old_vport;
1382         const struct tnl_mutable_config *old_mutable;
1383         struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1384         int err;
1385
1386         if (!options)
1387                 return -EINVAL;
1388
1389         err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1390         if (err)
1391                 return err;
1392
1393         if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1394                 return -EINVAL;
1395
1396         mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1397
1398         port_key_set_net(&mutable->key, net);
1399         mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1400         if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
1401                 if (ipv4_is_multicast(mutable->key.daddr))
1402                         return -EINVAL;
1403                 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1404         }
1405
1406         if (a[OVS_TUNNEL_ATTR_TOS]) {
1407                 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1408                 /* Reject ToS config with ECN bits set. */
1409                 if (mutable->tos & INET_ECN_MASK)
1410                         return -EINVAL;
1411         }
1412
1413         if (a[OVS_TUNNEL_ATTR_TTL])
1414                 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1415
1416         mutable->key.tunnel_type = tnl_ops->tunnel_type;
1417         if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1418                 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1419                 mutable->flags |= TNL_F_IN_KEY_MATCH;
1420         } else {
1421                 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1422                 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1423         }
1424
1425         if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1426                 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1427         else
1428                 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1429
1430         mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1431         if (mutable->tunnel_hlen < 0)
1432                 return mutable->tunnel_hlen;
1433
1434         mutable->tunnel_hlen += sizeof(struct iphdr);
1435
1436         old_vport = port_table_lookup(&mutable->key, &old_mutable);
1437         if (old_vport && old_vport != cur_vport)
1438                 return -EEXIST;
1439
1440         mutable->mlink = 0;
1441         if (ipv4_is_multicast(mutable->key.daddr)) {
1442                 struct net_device *dev;
1443                 struct rtable *rt;
1444
1445                 rt = __find_route(mutable, tnl_ops->ipproto, mutable->tos);
1446                 if (IS_ERR(rt))
1447                         return -EADDRNOTAVAIL;
1448                 dev = rt_dst(rt).dev;
1449                 ip_rt_put(rt);
1450                 if (__in_dev_get_rtnl(dev) == NULL)
1451                         return -EADDRNOTAVAIL;
1452                 mutable->mlink = dev->ifindex;
1453                 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
1454         }
1455
1456         return 0;
1457 }
1458
1459 struct vport *ovs_tnl_create(const struct vport_parms *parms,
1460                              const struct vport_ops *vport_ops,
1461                              const struct tnl_ops *tnl_ops)
1462 {
1463         struct vport *vport;
1464         struct tnl_vport *tnl_vport;
1465         struct tnl_mutable_config *mutable;
1466         int initial_frag_id;
1467         int err;
1468
1469         vport = ovs_vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1470         if (IS_ERR(vport)) {
1471                 err = PTR_ERR(vport);
1472                 goto error;
1473         }
1474
1475         tnl_vport = tnl_vport_priv(vport);
1476
1477         strcpy(tnl_vport->name, parms->name);
1478         tnl_vport->tnl_ops = tnl_ops;
1479
1480         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1481         if (!mutable) {
1482                 err = -ENOMEM;
1483                 goto error_free_vport;
1484         }
1485
1486         random_ether_addr(mutable->eth_addr);
1487
1488         get_random_bytes(&initial_frag_id, sizeof(int));
1489         atomic_set(&tnl_vport->frag_id, initial_frag_id);
1490
1491         err = tnl_set_config(ovs_dp_get_net(parms->dp), parms->options, tnl_ops,
1492                              NULL, mutable);
1493         if (err)
1494                 goto error_free_mutable;
1495
1496         spin_lock_init(&tnl_vport->cache_lock);
1497
1498 #ifdef NEED_CACHE_TIMEOUT
1499         tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1500                                        (net_random() % (MAX_CACHE_EXP / 2));
1501 #endif
1502
1503         rcu_assign_pointer(tnl_vport->mutable, mutable);
1504
1505         port_table_add_port(vport);
1506         return vport;
1507
1508 error_free_mutable:
1509         free_mutable_rtnl(mutable);
1510         kfree(mutable);
1511 error_free_vport:
1512         ovs_vport_free(vport);
1513 error:
1514         return ERR_PTR(err);
1515 }
1516
1517 int ovs_tnl_set_options(struct vport *vport, struct nlattr *options)
1518 {
1519         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1520         const struct tnl_mutable_config *old_mutable;
1521         struct tnl_mutable_config *mutable;
1522         int err;
1523
1524         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1525         if (!mutable) {
1526                 err = -ENOMEM;
1527                 goto error;
1528         }
1529
1530         /* Copy fields whose values should be retained. */
1531         old_mutable = rtnl_dereference(tnl_vport->mutable);
1532         mutable->seq = old_mutable->seq + 1;
1533         memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1534
1535         /* Parse the others configured by userspace. */
1536         err = tnl_set_config(ovs_dp_get_net(vport->dp), options, tnl_vport->tnl_ops,
1537                              vport, mutable);
1538         if (err)
1539                 goto error_free;
1540
1541         if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1542                 port_table_move_port(vport, mutable);
1543         else
1544                 assign_config_rcu(vport, mutable);
1545
1546         return 0;
1547
1548 error_free:
1549         free_mutable_rtnl(mutable);
1550         kfree(mutable);
1551 error:
1552         return err;
1553 }
1554
1555 int ovs_tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1556 {
1557         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1558         const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1559
1560         if (nla_put_u32(skb, OVS_TUNNEL_ATTR_FLAGS,
1561                       mutable->flags & TNL_F_PUBLIC) ||
1562             nla_put_be32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr))
1563                 goto nla_put_failure;
1564
1565         if (!(mutable->flags & TNL_F_IN_KEY_MATCH) &&
1566             nla_put_be64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key))
1567                 goto nla_put_failure;
1568         if (!(mutable->flags & TNL_F_OUT_KEY_ACTION) &&
1569             nla_put_be64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key))
1570                 goto nla_put_failure;
1571         if (mutable->key.saddr &&
1572             nla_put_be32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr))
1573                 goto nla_put_failure;
1574         if (mutable->tos && nla_put_u8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos))
1575                 goto nla_put_failure;
1576         if (mutable->ttl && nla_put_u8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl))
1577                 goto nla_put_failure;
1578
1579         return 0;
1580
1581 nla_put_failure:
1582         return -EMSGSIZE;
1583 }
1584
1585 static void free_port_rcu(struct rcu_head *rcu)
1586 {
1587         struct tnl_vport *tnl_vport = container_of(rcu,
1588                                                    struct tnl_vport, rcu);
1589
1590         free_cache((struct tnl_cache __force *)tnl_vport->cache);
1591         kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1592         ovs_vport_free(tnl_vport_to_vport(tnl_vport));
1593 }
1594
1595 void ovs_tnl_destroy(struct vport *vport)
1596 {
1597         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1598         struct tnl_mutable_config *mutable;
1599
1600         mutable = rtnl_dereference(tnl_vport->mutable);
1601         port_table_remove_port(vport);
1602         free_mutable_rtnl(mutable);
1603         call_rcu(&tnl_vport->rcu, free_port_rcu);
1604 }
1605
1606 int ovs_tnl_set_addr(struct vport *vport, const unsigned char *addr)
1607 {
1608         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1609         struct tnl_mutable_config *old_mutable, *mutable;
1610
1611         old_mutable = rtnl_dereference(tnl_vport->mutable);
1612         mutable = kmemdup(old_mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1613         if (!mutable)
1614                 return -ENOMEM;
1615
1616         old_mutable->mlink = 0;
1617
1618         memcpy(mutable->eth_addr, addr, ETH_ALEN);
1619         assign_config_rcu(vport, mutable);
1620
1621         return 0;
1622 }
1623
1624 const char *ovs_tnl_get_name(const struct vport *vport)
1625 {
1626         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1627         return tnl_vport->name;
1628 }
1629
1630 const unsigned char *ovs_tnl_get_addr(const struct vport *vport)
1631 {
1632         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1633         return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1634 }
1635
1636 void ovs_tnl_free_linked_skbs(struct sk_buff *skb)
1637 {
1638         while (skb) {
1639                 struct sk_buff *next = skb->next;
1640                 kfree_skb(skb);
1641                 skb = next;
1642         }
1643 }
1644
1645 int ovs_tnl_init(void)
1646 {
1647         int i;
1648
1649         port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1650                              GFP_KERNEL);
1651         if (!port_table)
1652                 return -ENOMEM;
1653
1654         for (i = 0; i < PORT_TABLE_SIZE; i++)
1655                 INIT_HLIST_HEAD(&port_table[i]);
1656
1657         return 0;
1658 }
1659
1660 void ovs_tnl_exit(void)
1661 {
1662         kfree(port_table);
1663 }