datapath/tunnel.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/if_arp.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/ip.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/igmp.h>
  26 #include <linux/in.h>
  27 #include <linux/in_route.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/jhash.h>
  30 #include <linux/list.h>
  31 #include <linux/kernel.h>
  32 #include <linux/version.h>
  33 #include <linux/workqueue.h>
  34 #include <linux/rculist.h>
  35
  36 #include <net/dsfield.h>
  37 #include <net/dst.h>
  38 #include <net/icmp.h>
  39 #include <net/inet_ecn.h>
  40 #include <net/ip.h>
  41 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
  42 #include <net/ipv6.h>
  43 #endif
  44 #include <net/route.h>
  45 #include <net/xfrm.h>
  46
  47 #include "checksum.h"
  48 #include "datapath.h"
  49 #include "tunnel.h"
  50 #include "vlan.h"
  51 #include "vport.h"
  52 #include "vport-internal_dev.h"
  53
  54 #define PORT_TABLE_SIZE  1024
  55
  56 static struct hlist_head *port_table __read_mostly;
  57
  58 /*
  59  * These are just used as an optimization: they don't require any kind of
  60  * synchronization because we could have just as easily read the value before
  61  * the port change happened.
  62  */
  63 static unsigned int key_local_remote_ports __read_mostly;
  64 static unsigned int key_remote_ports __read_mostly;
  65 static unsigned int key_multicast_ports __read_mostly;
  66 static unsigned int local_remote_ports __read_mostly;
  67 static unsigned int remote_ports __read_mostly;
  68 static unsigned int null_ports __read_mostly;
  69 static unsigned int multicast_ports __read_mostly;
  70
  71 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  72 #define rt_dst(rt) (rt->dst)
  73 #else
  74 #define rt_dst(rt) (rt->u.dst)
  75 #endif
  76
  77 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
  78 {
  79         return vport_from_priv(tnl_vport);
  80 }
  81
  82 static void free_config_rcu(struct rcu_head *rcu)
  83 {
  84         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
  85         kfree(c);
  86 }
  87
  88 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
  89  * within an RCU callback.  Fortunately this part doesn't require waiting for
  90  * an RCU grace period.
  91  */
  92 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
  93 {
  94         ASSERT_RTNL();
  95         if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
  96                 struct in_device *in_dev;
  97                 in_dev = inetdev_by_index(port_key_get_net(&mutable->key), mutable->mlink);
  98                 if (in_dev)
  99                         ip_mc_dec_group(in_dev, mutable->key.daddr);
 100         }
 101 }
 102
 103 static void assign_config_rcu(struct vport *vport,
 104                               struct tnl_mutable_config *new_config)
 105 {
 106         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 107         struct tnl_mutable_config *old_config;
 108
 109         old_config = rtnl_dereference(tnl_vport->mutable);
 110         rcu_assign_pointer(tnl_vport->mutable, new_config);
 111
 112         free_mutable_rtnl(old_config);
 113         call_rcu(&old_config->rcu, free_config_rcu);
 114 }
 115
 116 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 117 {
 118         bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
 119
 120         if (mutable->flags & TNL_F_IN_KEY_MATCH) {
 121                 if (mutable->key.saddr)
 122                         return &local_remote_ports;
 123                 else if (is_multicast)
 124                         return &multicast_ports;
 125                 else
 126                         return &remote_ports;
 127         } else {
 128                 if (mutable->key.saddr)
 129                         return &key_local_remote_ports;
 130                 else if (is_multicast)
 131                         return &key_multicast_ports;
 132                 else if (mutable->key.daddr)
 133                         return &key_remote_ports;
 134                 else
 135                         return &null_ports;
 136         }
 137 }
 138
 139 static u32 port_hash(const struct port_lookup_key *key)
 140 {
 141         return jhash2((u32 *)key, (PORT_KEY_LEN / sizeof(u32)), 0);
 142 }
 143
 144 static struct hlist_head *find_bucket(u32 hash)
 145 {
 146         return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
 147 }
 148
 149 static void port_table_add_port(struct vport *vport)
 150 {
 151         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 152         const struct tnl_mutable_config *mutable;
 153         u32 hash;
 154
 155         mutable = rtnl_dereference(tnl_vport->mutable);
 156         hash = port_hash(&mutable->key);
 157         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 158
 159         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 160 }
 161
 162 static void port_table_move_port(struct vport *vport,
 163                       struct tnl_mutable_config *new_mutable)
 164 {
 165         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 166         u32 hash;
 167
 168         hash = port_hash(&new_mutable->key);
 169         hlist_del_init_rcu(&tnl_vport->hash_node);
 170         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 171
 172         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 173         assign_config_rcu(vport, new_mutable);
 174         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 175 }
 176
 177 static void port_table_remove_port(struct vport *vport)
 178 {
 179         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 180
 181         hlist_del_init_rcu(&tnl_vport->hash_node);
 182
 183         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 184 }
 185
 186 static struct vport *port_table_lookup(struct port_lookup_key *key,
 187                                        const struct tnl_mutable_config **pmutable)
 188 {
 189         struct hlist_node *n;
 190         struct hlist_head *bucket;
 191         u32 hash = port_hash(key);
 192         struct tnl_vport *tnl_vport;
 193
 194         bucket = find_bucket(hash);
 195
 196         hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
 197                 struct tnl_mutable_config *mutable;
 198
 199                 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 200                 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
 201                         *pmutable = mutable;
 202                         return tnl_vport_to_vport(tnl_vport);
 203                 }
 204         }
 205
 206         return NULL;
 207 }
 208
 209 struct vport *ovs_tnl_find_port(struct net *net, __be32 saddr, __be32 daddr,
 210                                 __be64 key, int tunnel_type,
 211                                 const struct tnl_mutable_config **mutable)
 212 {
 213         struct port_lookup_key lookup;
 214         struct vport *vport;
 215         bool is_multicast = ipv4_is_multicast(saddr);
 216
 217         port_key_set_net(&lookup, net);
 218         lookup.saddr = saddr;
 219         lookup.daddr = daddr;
 220
 221         /* First try for exact match on in_key. */
 222         lookup.in_key = key;
 223         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 224         if (!is_multicast && key_local_remote_ports) {
 225                 vport = port_table_lookup(&lookup, mutable);
 226                 if (vport)
 227                         return vport;
 228         }
 229         if (key_remote_ports) {
 230                 lookup.saddr = 0;
 231                 vport = port_table_lookup(&lookup, mutable);
 232                 if (vport)
 233                         return vport;
 234
 235                 lookup.saddr = saddr;
 236         }
 237
 238         /* Then try matches that wildcard in_key. */
 239         lookup.in_key = 0;
 240         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 241         if (!is_multicast && local_remote_ports) {
 242                 vport = port_table_lookup(&lookup, mutable);
 243                 if (vport)
 244                         return vport;
 245         }
 246         if (remote_ports) {
 247                 lookup.saddr = 0;
 248                 vport = port_table_lookup(&lookup, mutable);
 249                 if (vport)
 250                         return vport;
 251         }
 252
 253         if (is_multicast) {
 254                 lookup.saddr = 0;
 255                 lookup.daddr = saddr;
 256                 if (key_multicast_ports) {
 257                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 258                         lookup.in_key = key;
 259                         vport = port_table_lookup(&lookup, mutable);
 260                         if (vport)
 261                                 return vport;
 262                 }
 263                 if (multicast_ports) {
 264                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 265                         lookup.in_key = 0;
 266                         vport = port_table_lookup(&lookup, mutable);
 267                         if (vport)
 268                                 return vport;
 269                 }
 270         }
 271
 272         if (null_ports) {
 273                 lookup.daddr = 0;
 274                 lookup.saddr = 0;
 275                 lookup.in_key = 0;
 276                 lookup.tunnel_type = tunnel_type;
 277                 vport = port_table_lookup(&lookup, mutable);
 278                 if (vport)
 279                         return vport;
 280         }
 281         return NULL;
 282 }
 283
 284 static void ecn_decapsulate(struct sk_buff *skb)
 285 {
 286         if (unlikely(INET_ECN_is_ce(OVS_CB(skb)->tun_key->ipv4_tos))) {
 287                 __be16 protocol = skb->protocol;
 288
 289                 skb_set_network_header(skb, ETH_HLEN);
 290
 291                 if (protocol == htons(ETH_P_8021Q)) {
 292                         if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 293                                 return;
 294
 295                         protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 296                         skb_set_network_header(skb, VLAN_ETH_HLEN);
 297                 }
 298
 299                 if (protocol == htons(ETH_P_IP)) {
 300                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 301                             + sizeof(struct iphdr))))
 302                                 return;
 303
 304                         IP_ECN_set_ce(ip_hdr(skb));
 305                 }
 306 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 307                 else if (protocol == htons(ETH_P_IPV6)) {
 308                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 309                             + sizeof(struct ipv6hdr))))
 310                                 return;
 311
 312                         IP6_ECN_set_ce(ipv6_hdr(skb));
 313                 }
 314 #endif
 315         }
 316 }
 317
 318 /**
 319  *      ovs_tnl_rcv - ingress point for generic tunnel code
 320  *
 321  * @vport: port this packet was received on
 322  * @skb: received packet
 323  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
 324  *
 325  * Must be called with rcu_read_lock.
 326  *
 327  * Packets received by this function are in the following state:
 328  * - skb->data points to the inner Ethernet header.
 329  * - The inner Ethernet header is in the linear data area.
 330  * - skb->csum does not include the inner Ethernet header.
 331  * - The layer pointers are undefined.
 332  */
 333 void ovs_tnl_rcv(struct vport *vport, struct sk_buff *skb)
 334 {
 335         struct ethhdr *eh;
 336
 337         skb_reset_mac_header(skb);
 338         eh = eth_hdr(skb);
 339
 340         if (likely(ntohs(eh->h_proto) >= 1536))
 341                 skb->protocol = eh->h_proto;
 342         else
 343                 skb->protocol = htons(ETH_P_802_2);
 344
 345         skb_dst_drop(skb);
 346         nf_reset(skb);
 347         skb_clear_rxhash(skb);
 348         secpath_reset(skb);
 349
 350         ecn_decapsulate(skb);
 351         vlan_set_tci(skb, 0);
 352
 353         if (unlikely(compute_ip_summed(skb, false))) {
 354                 kfree_skb(skb);
 355                 return;
 356         }
 357
 358         ovs_vport_receive(vport, skb);
 359 }
 360
 361 static struct rtable *find_route(struct net *net,
 362                 __be32 *saddr, __be32 daddr, u8 ipproto,
 363                 u8 tos)
 364 {
 365         struct rtable *rt;
 366         /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
 367          * router expect RT_TOS bits only. */
 368
 369 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 370         struct flowi fl = { .nl_u = { .ip4_u = {
 371                                         .daddr = daddr,
 372                                         .saddr = *saddr,
 373                                         .tos   = RT_TOS(tos) } },
 374                                         .proto = ipproto };
 375
 376         if (unlikely(ip_route_output_key(net, &rt, &fl)))
 377                 return ERR_PTR(-EADDRNOTAVAIL);
 378         *saddr = fl.nl_u.ip4_u.saddr;
 379         return rt;
 380 #else
 381         struct flowi4 fl = { .daddr = daddr,
 382                              .saddr = *saddr,
 383                              .flowi4_tos = RT_TOS(tos),
 384                              .flowi4_proto = ipproto };
 385
 386         rt = ip_route_output_key(net, &fl);
 387         *saddr = fl.saddr;
 388         return rt;
 389 #endif
 390 }
 391
 392 static bool need_linearize(const struct sk_buff *skb)
 393 {
 394         int i;
 395
 396         if (unlikely(skb_shinfo(skb)->frag_list))
 397                 return true;
 398
 399         /*
 400          * Generally speaking we should linearize if there are paged frags.
 401          * However, if all of the refcounts are 1 we know nobody else can
 402          * change them from underneath us and we can skip the linearization.
 403          */
 404         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 405                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
 406                         return true;
 407
 408         return false;
 409 }
 410
 411 static struct sk_buff *handle_offloads(struct sk_buff *skb,
 412                                        const struct tnl_mutable_config *mutable,
 413                                        const struct rtable *rt,
 414                                        int tunnel_hlen)
 415 {
 416         int min_headroom;
 417         int err;
 418
 419         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 420                         + tunnel_hlen
 421                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 422
 423         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 424                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 425                                                 skb_headroom(skb) +
 426                                                 16);
 427                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 428                                         0, GFP_ATOMIC);
 429                 if (unlikely(err))
 430                         goto error_free;
 431         }
 432
 433         forward_ip_summed(skb, true);
 434
 435         if (skb_is_gso(skb)) {
 436                 struct sk_buff *nskb;
 437
 438                 nskb = skb_gso_segment(skb, 0);
 439                 if (IS_ERR(nskb)) {
 440                         kfree_skb(skb);
 441                         err = PTR_ERR(nskb);
 442                         goto error;
 443                 }
 444
 445                 consume_skb(skb);
 446                 skb = nskb;
 447         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 448                 /* Pages aren't locked and could change at any time.
 449                  * If this happens after we compute the checksum, the
 450                  * checksum will be wrong.  We linearize now to avoid
 451                  * this problem.
 452                  */
 453                 if (unlikely(need_linearize(skb))) {
 454                         err = __skb_linearize(skb);
 455                         if (unlikely(err))
 456                                 goto error_free;
 457                 }
 458
 459                 err = skb_checksum_help(skb);
 460                 if (unlikely(err))
 461                         goto error_free;
 462         }
 463
 464         set_ip_summed(skb, OVS_CSUM_NONE);
 465
 466         return skb;
 467
 468 error_free:
 469         kfree_skb(skb);
 470 error:
 471         return ERR_PTR(err);
 472 }
 473
 474 static int send_frags(struct sk_buff *skb,
 475                       int tunnel_hlen)
 476 {
 477         int sent_len;
 478
 479         sent_len = 0;
 480         while (skb) {
 481                 struct sk_buff *next = skb->next;
 482                 int frag_len = skb->len - tunnel_hlen;
 483                 int err;
 484
 485                 skb->next = NULL;
 486                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 487
 488                 err = ip_local_out(skb);
 489                 skb = next;
 490                 if (unlikely(net_xmit_eval(err)))
 491                         goto free_frags;
 492                 sent_len += frag_len;
 493         }
 494
 495         return sent_len;
 496
 497 free_frags:
 498         /*
 499          * There's no point in continuing to send fragments once one has been
 500          * dropped so just free the rest.  This may help improve the congestion
 501          * that caused the first packet to be dropped.
 502          */
 503         ovs_tnl_free_linked_skbs(skb);
 504         return sent_len;
 505 }
 506
 507 int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
 508 {
 509         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 510         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
 511         enum vport_err_type err = VPORT_E_TX_ERROR;
 512         struct rtable *rt;
 513         struct ovs_key_ipv4_tunnel tun_key;
 514         int sent_len = 0;
 515         int tunnel_hlen;
 516         __be16 frag_off;
 517         __be32 daddr;
 518         __be32 saddr;
 519         u8 ttl;
 520         u8 tos;
 521
 522         /* Validate the protocol headers before we try to use them. */
 523         if (skb->protocol == htons(ETH_P_8021Q) &&
 524             !vlan_tx_tag_present(skb)) {
 525                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 526                         goto error_free;
 527
 528                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 529                 skb_set_network_header(skb, VLAN_ETH_HLEN);
 530         }
 531
 532         if (skb->protocol == htons(ETH_P_IP)) {
 533                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 534                     + sizeof(struct iphdr))))
 535                         skb->protocol = 0;
 536         }
 537 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 538         else if (skb->protocol == htons(ETH_P_IPV6)) {
 539                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 540                     + sizeof(struct ipv6hdr))))
 541                         skb->protocol = 0;
 542         }
 543 #endif
 544
 545         /* If OVS_CB(skb)->tun_key is NULL, point it at the local tun_key here,
 546          * and zero it out.
 547          */
 548         if (!OVS_CB(skb)->tun_key) {
 549                 memset(&tun_key, 0, sizeof(tun_key));
 550                 OVS_CB(skb)->tun_key = &tun_key;
 551         }
 552
 553         tunnel_hlen = tnl_vport->tnl_ops->hdr_len(mutable, OVS_CB(skb)->tun_key);
 554         if (unlikely(tunnel_hlen < 0)) {
 555                 err = VPORT_E_TX_DROPPED;
 556                 goto error_free;
 557         }
 558         tunnel_hlen += sizeof(struct iphdr);
 559
 560         if (OVS_CB(skb)->tun_key->ipv4_dst) {
 561                 daddr = OVS_CB(skb)->tun_key->ipv4_dst;
 562                 saddr = OVS_CB(skb)->tun_key->ipv4_src;
 563                 tos = OVS_CB(skb)->tun_key->ipv4_tos;
 564                 ttl = OVS_CB(skb)->tun_key->ipv4_ttl;
 565                 frag_off = OVS_CB(skb)->tun_key->tun_flags &
 566                                 OVS_TNL_F_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 567         } else {
 568                 u8 inner_tos;
 569                 daddr = mutable->key.daddr;
 570                 saddr = mutable->key.saddr;
 571
 572                 if (unlikely(!daddr)) {
 573                         /* Trying to sent packet from Null-port without
 574                          * tunnel info? Drop this packet. */
 575                         err = VPORT_E_TX_DROPPED;
 576                         goto error_free;
 577                 }
 578
 579                 /* ToS */
 580                 if (skb->protocol == htons(ETH_P_IP))
 581                         inner_tos = ip_hdr(skb)->tos;
 582 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 583                 else if (skb->protocol == htons(ETH_P_IPV6))
 584                         inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
 585 #endif
 586                 else
 587                         inner_tos = 0;
 588
 589                 if (mutable->flags & TNL_F_TOS_INHERIT)
 590                         tos = inner_tos;
 591                 else
 592                         tos = mutable->tos;
 593
 594                 tos = INET_ECN_encapsulate(tos, inner_tos);
 595
 596                 /* TTL */
 597                 ttl = mutable->ttl;
 598                 if (mutable->flags & TNL_F_TTL_INHERIT) {
 599                         if (skb->protocol == htons(ETH_P_IP))
 600                                 ttl = ip_hdr(skb)->ttl;
 601 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 602                         else if (skb->protocol == htons(ETH_P_IPV6))
 603                                 ttl = ipv6_hdr(skb)->hop_limit;
 604 #endif
 605                 }
 606
 607                 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
 608         }
 609
 610         /* Route lookup */
 611         rt = find_route(port_key_get_net(&mutable->key), &saddr, daddr,
 612                           tnl_vport->tnl_ops->ipproto, tos);
 613         if (IS_ERR(rt))
 614                 goto error_free;
 615
 616         /* Reset SKB */
 617         nf_reset(skb);
 618         secpath_reset(skb);
 619         skb_dst_drop(skb);
 620         skb_clear_rxhash(skb);
 621
 622         /* Offloading */
 623         skb = handle_offloads(skb, mutable, rt, tunnel_hlen);
 624         if (IS_ERR(skb)) {
 625                 skb = NULL;
 626                 goto err_free_rt;
 627         }
 628
 629         /* TTL Fixup. */
 630         if (!OVS_CB(skb)->tun_key->ipv4_dst) {
 631                 if (!(mutable->flags & TNL_F_TTL_INHERIT)) {
 632                         if (!ttl)
 633                                 ttl = ip4_dst_hoplimit(&rt_dst(rt));
 634                 }
 635         }
 636
 637         while (skb) {
 638                 struct iphdr *iph;
 639                 struct sk_buff *next_skb = skb->next;
 640                 skb->next = NULL;
 641
 642                 if (unlikely(vlan_deaccel_tag(skb)))
 643                         goto next;
 644
 645                 skb_push(skb, tunnel_hlen);
 646                 skb_reset_network_header(skb);
 647                 skb_set_transport_header(skb, sizeof(struct iphdr));
 648
 649                 if (next_skb)
 650                         skb_dst_set(skb, dst_clone(&rt_dst(rt)));
 651                 else
 652                         skb_dst_set(skb, &rt_dst(rt));
 653
 654                 /* Push IP header. */
 655                 iph = ip_hdr(skb);
 656                 iph->version    = 4;
 657                 iph->ihl        = sizeof(struct iphdr) >> 2;
 658                 iph->protocol   = tnl_vport->tnl_ops->ipproto;
 659                 iph->daddr      = daddr;
 660                 iph->saddr      = saddr;
 661                 iph->tos        = tos;
 662                 iph->ttl        = ttl;
 663                 iph->frag_off   = frag_off;
 664                 ip_select_ident(iph, &rt_dst(rt), NULL);
 665
 666                 /* Push Tunnel header. */
 667                 skb = tnl_vport->tnl_ops->build_header(vport, mutable,
 668                                                         &rt_dst(rt), skb, tunnel_hlen);
 669                 if (unlikely(!skb))
 670                         goto next;
 671
 672                 sent_len += send_frags(skb, tunnel_hlen);
 673
 674 next:
 675                 skb = next_skb;
 676         }
 677
 678         if (unlikely(sent_len == 0))
 679                 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
 680
 681         return sent_len;
 682
 683 err_free_rt:
 684         ip_rt_put(rt);
 685 error_free:
 686         ovs_tnl_free_linked_skbs(skb);
 687         ovs_vport_record_error(vport, err);
 688         return sent_len;
 689 }
 690
 691 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
 692         [OVS_TUNNEL_ATTR_FLAGS]    = { .type = NLA_U32 },
 693         [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
 694         [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
 695         [OVS_TUNNEL_ATTR_OUT_KEY]  = { .type = NLA_U64 },
 696         [OVS_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
 697         [OVS_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
 698         [OVS_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
 699         [OVS_TUNNEL_ATTR_DST_PORT] = { .type = NLA_U16 },
 700 };
 701
 702 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
 703  * zeroed. */
 704 static int tnl_set_config(struct net *net, struct nlattr *options,
 705                           const struct tnl_ops *tnl_ops,
 706                           const struct vport *cur_vport,
 707                           struct tnl_mutable_config *mutable)
 708 {
 709         const struct vport *old_vport;
 710         const struct tnl_mutable_config *old_mutable;
 711         struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
 712         int err;
 713
 714         port_key_set_net(&mutable->key, net);
 715         mutable->key.tunnel_type = tnl_ops->tunnel_type;
 716         if (!options)
 717                 goto out;
 718
 719         err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
 720         if (err)
 721                 return err;
 722
 723         /* Process attributes possibly useful for null_ports first */
 724         if (a[OVS_TUNNEL_ATTR_DST_PORT])
 725                 mutable->dst_port =
 726                         htons(nla_get_u16(a[OVS_TUNNEL_ATTR_DST_PORT]));
 727
 728         if (a[OVS_TUNNEL_ATTR_DST_IPV4])
 729                 mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
 730
 731         /* Skip the rest if configuring a null_port */
 732         if (!mutable->key.daddr)
 733                 goto out;
 734
 735         if (a[OVS_TUNNEL_ATTR_FLAGS])
 736                 mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS])
 737                         & TNL_F_PUBLIC;
 738
 739         if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
 740                 if (ipv4_is_multicast(mutable->key.daddr))
 741                         return -EINVAL;
 742                 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
 743         }
 744
 745         if (a[OVS_TUNNEL_ATTR_TOS]) {
 746                 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
 747                 /* Reject ToS config with ECN bits set. */
 748                 if (mutable->tos & INET_ECN_MASK)
 749                         return -EINVAL;
 750         }
 751
 752         if (a[OVS_TUNNEL_ATTR_TTL])
 753                 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
 754
 755         if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
 756                 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
 757                 mutable->flags |= TNL_F_IN_KEY_MATCH;
 758         } else {
 759                 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
 760                 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
 761         }
 762
 763         if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
 764                 mutable->flags |= TNL_F_OUT_KEY_ACTION;
 765         else
 766                 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
 767
 768         mutable->mlink = 0;
 769         if (ipv4_is_multicast(mutable->key.daddr)) {
 770                 struct net_device *dev;
 771                 struct rtable *rt;
 772                 __be32 saddr = mutable->key.saddr;
 773
 774                 rt = find_route(port_key_get_net(&mutable->key),
 775                              &saddr, mutable->key.daddr,
 776                              tnl_ops->ipproto, mutable->tos);
 777                 if (IS_ERR(rt))
 778                         return -EADDRNOTAVAIL;
 779                 dev = rt_dst(rt).dev;
 780                 ip_rt_put(rt);
 781                 if (__in_dev_get_rtnl(dev) == NULL)
 782                         return -EADDRNOTAVAIL;
 783                 mutable->mlink = dev->ifindex;
 784                 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
 785         }
 786
 787 out:
 788         old_vport = port_table_lookup(&mutable->key, &old_mutable);
 789         if (old_vport && old_vport != cur_vport)
 790                 return -EEXIST;
 791
 792         return 0;
 793 }
 794
 795 struct vport *ovs_tnl_create(const struct vport_parms *parms,
 796                              const struct vport_ops *vport_ops,
 797                              const struct tnl_ops *tnl_ops)
 798 {
 799         struct vport *vport;
 800         struct tnl_vport *tnl_vport;
 801         struct tnl_mutable_config *mutable;
 802         int initial_frag_id;
 803         int err;
 804
 805         vport = ovs_vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
 806         if (IS_ERR(vport)) {
 807                 err = PTR_ERR(vport);
 808                 goto error;
 809         }
 810
 811         tnl_vport = tnl_vport_priv(vport);
 812
 813         strcpy(tnl_vport->name, parms->name);
 814         tnl_vport->tnl_ops = tnl_ops;
 815
 816         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
 817         if (!mutable) {
 818                 err = -ENOMEM;
 819                 goto error_free_vport;
 820         }
 821
 822         get_random_bytes(&initial_frag_id, sizeof(int));
 823         atomic_set(&tnl_vport->frag_id, initial_frag_id);
 824
 825         err = tnl_set_config(ovs_dp_get_net(parms->dp), parms->options, tnl_ops,
 826                              NULL, mutable);
 827         if (err)
 828                 goto error_free_mutable;
 829
 830         rcu_assign_pointer(tnl_vport->mutable, mutable);
 831
 832         port_table_add_port(vport);
 833         return vport;
 834
 835 error_free_mutable:
 836         free_mutable_rtnl(mutable);
 837         kfree(mutable);
 838 error_free_vport:
 839         ovs_vport_free(vport);
 840 error:
 841         return ERR_PTR(err);
 842 }
 843
 844 int ovs_tnl_set_options(struct vport *vport, struct nlattr *options)
 845 {
 846         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 847         const struct tnl_mutable_config *old_mutable;
 848         struct tnl_mutable_config *mutable;
 849         int err;
 850
 851         old_mutable = rtnl_dereference(tnl_vport->mutable);
 852         if (!old_mutable->key.daddr)
 853                 return -EINVAL;
 854
 855         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
 856         if (!mutable) {
 857                 err = -ENOMEM;
 858                 goto error;
 859         }
 860
 861         mutable->seq = old_mutable->seq + 1;
 862
 863         /* Parse the others configured by userspace. */
 864         err = tnl_set_config(ovs_dp_get_net(vport->dp), options, tnl_vport->tnl_ops,
 865                              vport, mutable);
 866         if (err)
 867                 goto error_free;
 868
 869         if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
 870                 port_table_move_port(vport, mutable);
 871         else
 872                 assign_config_rcu(vport, mutable);
 873
 874         return 0;
 875
 876 error_free:
 877         free_mutable_rtnl(mutable);
 878         kfree(mutable);
 879 error:
 880         return err;
 881 }
 882
 883 int ovs_tnl_get_options(const struct vport *vport, struct sk_buff *skb)
 884 {
 885         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 886         const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 887
 888         if (mutable->dst_port && nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT,
 889                                              ntohs(mutable->dst_port)))
 890                 goto nla_put_failure;
 891
 892         /* Skip the rest for null_ports */
 893         if (!mutable->key.daddr)
 894                 return 0;
 895
 896         if (nla_put_be32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr))
 897                 goto nla_put_failure;
 898         if (nla_put_u32(skb, OVS_TUNNEL_ATTR_FLAGS,
 899                         mutable->flags & TNL_F_PUBLIC))
 900                 goto nla_put_failure;
 901         if (!(mutable->flags & TNL_F_IN_KEY_MATCH) &&
 902             nla_put_be64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key))
 903                 goto nla_put_failure;
 904         if (!(mutable->flags & TNL_F_OUT_KEY_ACTION) &&
 905             nla_put_be64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key))
 906                 goto nla_put_failure;
 907         if (mutable->key.saddr &&
 908             nla_put_be32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr))
 909                 goto nla_put_failure;
 910         if (mutable->tos && nla_put_u8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos))
 911                 goto nla_put_failure;
 912         if (mutable->ttl && nla_put_u8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl))
 913                 goto nla_put_failure;
 914
 915         return 0;
 916
 917 nla_put_failure:
 918         return -EMSGSIZE;
 919 }
 920
 921 static void free_port_rcu(struct rcu_head *rcu)
 922 {
 923         struct tnl_vport *tnl_vport = container_of(rcu,
 924                                                    struct tnl_vport, rcu);
 925
 926         kfree((struct tnl_mutable __force *)tnl_vport->mutable);
 927         ovs_vport_free(tnl_vport_to_vport(tnl_vport));
 928 }
 929
 930 void ovs_tnl_destroy(struct vport *vport)
 931 {
 932         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 933         struct tnl_mutable_config *mutable;
 934
 935         mutable = rtnl_dereference(tnl_vport->mutable);
 936         port_table_remove_port(vport);
 937         free_mutable_rtnl(mutable);
 938         call_rcu(&tnl_vport->rcu, free_port_rcu);
 939 }
 940
 941 const char *ovs_tnl_get_name(const struct vport *vport)
 942 {
 943         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 944         return tnl_vport->name;
 945 }
 946
 947 void ovs_tnl_free_linked_skbs(struct sk_buff *skb)
 948 {
 949         while (skb) {
 950                 struct sk_buff *next = skb->next;
 951                 kfree_skb(skb);
 952                 skb = next;
 953         }
 954 }
 955
 956 int ovs_tnl_init(void)
 957 {
 958         int i;
 959
 960         port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
 961                              GFP_KERNEL);
 962         if (!port_table)
 963                 return -ENOMEM;
 964
 965         for (i = 0; i < PORT_TABLE_SIZE; i++)
 966                 INIT_HLIST_HEAD(&port_table[i]);
 967
 968         return 0;
 969 }
 970
 971 void ovs_tnl_exit(void)
 972 {
 973         kfree(port_table);
 974 }