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 "compat.h"
  49 #include "datapath.h"
  50 #include "tunnel.h"
  51 #include "vlan.h"
  52 #include "vport.h"
  53 #include "vport-internal_dev.h"
  54
  55 #define PORT_TABLE_SIZE  1024
  56
  57 static struct hlist_head *port_table __read_mostly;
  58
  59 /*
  60  * These are just used as an optimization: they don't require any kind of
  61  * synchronization because we could have just as easily read the value before
  62  * the port change happened.
  63  */
  64 static unsigned int key_local_remote_ports __read_mostly;
  65 static unsigned int key_remote_ports __read_mostly;
  66 static unsigned int key_multicast_ports __read_mostly;
  67 static unsigned int local_remote_ports __read_mostly;
  68 static unsigned int remote_ports __read_mostly;
  69 static unsigned int null_ports __read_mostly;
  70 static unsigned int multicast_ports __read_mostly;
  71
  72 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  73 #define rt_dst(rt) (rt->dst)
  74 #else
  75 #define rt_dst(rt) (rt->u.dst)
  76 #endif
  77
  78 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
  79 {
  80         return vport_from_priv(tnl_vport);
  81 }
  82
  83 static void free_config_rcu(struct rcu_head *rcu)
  84 {
  85         struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
  86         kfree(c);
  87 }
  88
  89 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
  90  * within an RCU callback.  Fortunately this part doesn't require waiting for
  91  * an RCU grace period.
  92  */
  93 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
  94 {
  95         ASSERT_RTNL();
  96         if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
  97                 struct in_device *in_dev;
  98                 in_dev = inetdev_by_index(port_key_get_net(&mutable->key), mutable->mlink);
  99                 if (in_dev)
 100                         ip_mc_dec_group(in_dev, mutable->key.daddr);
 101         }
 102 }
 103
 104 static void assign_config_rcu(struct vport *vport,
 105                               struct tnl_mutable_config *new_config)
 106 {
 107         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 108         struct tnl_mutable_config *old_config;
 109
 110         old_config = rtnl_dereference(tnl_vport->mutable);
 111         rcu_assign_pointer(tnl_vport->mutable, new_config);
 112
 113         free_mutable_rtnl(old_config);
 114         call_rcu(&old_config->rcu, free_config_rcu);
 115 }
 116
 117 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
 118 {
 119         bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
 120
 121         if (mutable->flags & TNL_F_IN_KEY_MATCH) {
 122                 if (mutable->key.saddr)
 123                         return &local_remote_ports;
 124                 else if (is_multicast)
 125                         return &multicast_ports;
 126                 else
 127                         return &remote_ports;
 128         } else {
 129                 if (mutable->key.saddr)
 130                         return &key_local_remote_ports;
 131                 else if (is_multicast)
 132                         return &key_multicast_ports;
 133                 else if (mutable->key.daddr)
 134                         return &key_remote_ports;
 135                 else
 136                         return &null_ports;
 137         }
 138 }
 139
 140 static u32 port_hash(const struct port_lookup_key *key)
 141 {
 142         return jhash2((u32 *)key, (PORT_KEY_LEN / sizeof(u32)), 0);
 143 }
 144
 145 static struct hlist_head *find_bucket(u32 hash)
 146 {
 147         return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
 148 }
 149
 150 static void port_table_add_port(struct vport *vport)
 151 {
 152         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 153         const struct tnl_mutable_config *mutable;
 154         u32 hash;
 155
 156         mutable = rtnl_dereference(tnl_vport->mutable);
 157         hash = port_hash(&mutable->key);
 158         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 159
 160         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 161 }
 162
 163 static void port_table_move_port(struct vport *vport,
 164                       struct tnl_mutable_config *new_mutable)
 165 {
 166         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 167         u32 hash;
 168
 169         hash = port_hash(&new_mutable->key);
 170         hlist_del_init_rcu(&tnl_vport->hash_node);
 171         hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
 172
 173         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 174         assign_config_rcu(vport, new_mutable);
 175         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
 176 }
 177
 178 static void port_table_remove_port(struct vport *vport)
 179 {
 180         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 181
 182         hlist_del_init_rcu(&tnl_vport->hash_node);
 183
 184         (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
 185 }
 186
 187 static struct vport *port_table_lookup(struct port_lookup_key *key,
 188                                        const struct tnl_mutable_config **pmutable)
 189 {
 190         struct hlist_node *n;
 191         struct hlist_head *bucket;
 192         u32 hash = port_hash(key);
 193         struct tnl_vport *tnl_vport;
 194
 195         bucket = find_bucket(hash);
 196
 197         hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
 198                 struct tnl_mutable_config *mutable;
 199
 200                 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 201                 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
 202                         *pmutable = mutable;
 203                         return tnl_vport_to_vport(tnl_vport);
 204                 }
 205         }
 206
 207         return NULL;
 208 }
 209
 210 struct vport *ovs_tnl_find_port(struct net *net, __be32 saddr, __be32 daddr,
 211                                 __be64 key, int tunnel_type,
 212                                 const struct tnl_mutable_config **mutable)
 213 {
 214         struct port_lookup_key lookup;
 215         struct vport *vport;
 216         bool is_multicast = ipv4_is_multicast(saddr);
 217
 218         port_key_set_net(&lookup, net);
 219         lookup.saddr = saddr;
 220         lookup.daddr = daddr;
 221
 222         /* First try for exact match on in_key. */
 223         lookup.in_key = key;
 224         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 225         if (!is_multicast && key_local_remote_ports) {
 226                 vport = port_table_lookup(&lookup, mutable);
 227                 if (vport)
 228                         return vport;
 229         }
 230         if (key_remote_ports) {
 231                 lookup.saddr = 0;
 232                 vport = port_table_lookup(&lookup, mutable);
 233                 if (vport)
 234                         return vport;
 235
 236                 lookup.saddr = saddr;
 237         }
 238
 239         /* Then try matches that wildcard in_key. */
 240         lookup.in_key = 0;
 241         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 242         if (!is_multicast && local_remote_ports) {
 243                 vport = port_table_lookup(&lookup, mutable);
 244                 if (vport)
 245                         return vport;
 246         }
 247         if (remote_ports) {
 248                 lookup.saddr = 0;
 249                 vport = port_table_lookup(&lookup, mutable);
 250                 if (vport)
 251                         return vport;
 252         }
 253
 254         if (is_multicast) {
 255                 lookup.saddr = 0;
 256                 lookup.daddr = saddr;
 257                 if (key_multicast_ports) {
 258                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
 259                         lookup.in_key = key;
 260                         vport = port_table_lookup(&lookup, mutable);
 261                         if (vport)
 262                                 return vport;
 263                 }
 264                 if (multicast_ports) {
 265                         lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
 266                         lookup.in_key = 0;
 267                         vport = port_table_lookup(&lookup, mutable);
 268                         if (vport)
 269                                 return vport;
 270                 }
 271         }
 272
 273         if (null_ports) {
 274                 lookup.daddr = 0;
 275                 lookup.saddr = 0;
 276                 lookup.in_key = 0;
 277                 lookup.tunnel_type = tunnel_type;
 278                 vport = port_table_lookup(&lookup, mutable);
 279                 if (vport)
 280                         return vport;
 281         }
 282         return NULL;
 283 }
 284
 285 static void ecn_decapsulate(struct sk_buff *skb)
 286 {
 287         if (unlikely(INET_ECN_is_ce(OVS_CB(skb)->tun_key->ipv4_tos))) {
 288                 __be16 protocol = skb->protocol;
 289
 290                 skb_set_network_header(skb, ETH_HLEN);
 291
 292                 if (protocol == htons(ETH_P_8021Q)) {
 293                         if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 294                                 return;
 295
 296                         protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 297                         skb_set_network_header(skb, VLAN_ETH_HLEN);
 298                 }
 299
 300                 if (protocol == htons(ETH_P_IP)) {
 301                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 302                             + sizeof(struct iphdr))))
 303                                 return;
 304
 305                         IP_ECN_set_ce(ip_hdr(skb));
 306                 }
 307 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 308                 else if (protocol == htons(ETH_P_IPV6)) {
 309                         if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 310                             + sizeof(struct ipv6hdr))))
 311                                 return;
 312
 313                         IP6_ECN_set_ce(ipv6_hdr(skb));
 314                 }
 315 #endif
 316         }
 317 }
 318
 319 /**
 320  *      ovs_tnl_rcv - ingress point for generic tunnel code
 321  *
 322  * @vport: port this packet was received on
 323  * @skb: received packet
 324  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
 325  *
 326  * Must be called with rcu_read_lock.
 327  *
 328  * Packets received by this function are in the following state:
 329  * - skb->data points to the inner Ethernet header.
 330  * - The inner Ethernet header is in the linear data area.
 331  * - skb->csum does not include the inner Ethernet header.
 332  * - The layer pointers are undefined.
 333  */
 334 void ovs_tnl_rcv(struct vport *vport, struct sk_buff *skb)
 335 {
 336         struct ethhdr *eh;
 337
 338         skb_reset_mac_header(skb);
 339         eh = eth_hdr(skb);
 340
 341         if (likely(ntohs(eh->h_proto) >= 1536))
 342                 skb->protocol = eh->h_proto;
 343         else
 344                 skb->protocol = htons(ETH_P_802_2);
 345
 346         skb_dst_drop(skb);
 347         nf_reset(skb);
 348         skb_clear_rxhash(skb);
 349         secpath_reset(skb);
 350
 351         ecn_decapsulate(skb);
 352         vlan_set_tci(skb, 0);
 353
 354         if (unlikely(compute_ip_summed(skb, false))) {
 355                 kfree_skb(skb);
 356                 return;
 357         }
 358
 359         ovs_vport_receive(vport, skb);
 360 }
 361
 362 static struct rtable *find_route(struct net *net,
 363                 __be32 *saddr, __be32 daddr, u8 ipproto,
 364                 u8 tos, u32 skb_mark)
 365 {
 366         struct rtable *rt;
 367         /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
 368          * router expect RT_TOS bits only. */
 369
 370 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
 371         struct flowi fl = { .nl_u = { .ip4_u = {
 372                                         .daddr = daddr,
 373                                         .saddr = *saddr,
 374 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20)
 375                                         .fwmark = skb_mark,
 376 #endif
 377                                         .tos   = RT_TOS(tos) } },
 378 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
 379                                         .mark = skb_mark,
 380 #endif
 381                                         .proto = ipproto };
 382
 383         if (unlikely(ip_route_output_key(net, &rt, &fl)))
 384                 return ERR_PTR(-EADDRNOTAVAIL);
 385         *saddr = fl.nl_u.ip4_u.saddr;
 386         return rt;
 387 #else
 388         struct flowi4 fl = { .daddr = daddr,
 389                              .saddr = *saddr,
 390                              .flowi4_tos = RT_TOS(tos),
 391                              .flowi4_mark = skb_mark,
 392                              .flowi4_proto = ipproto };
 393
 394         rt = ip_route_output_key(net, &fl);
 395         *saddr = fl.saddr;
 396         return rt;
 397 #endif
 398 }
 399
 400 static bool need_linearize(const struct sk_buff *skb)
 401 {
 402         int i;
 403
 404         if (unlikely(skb_shinfo(skb)->frag_list))
 405                 return true;
 406
 407         /*
 408          * Generally speaking we should linearize if there are paged frags.
 409          * However, if all of the refcounts are 1 we know nobody else can
 410          * change them from underneath us and we can skip the linearization.
 411          */
 412         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 413                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
 414                         return true;
 415
 416         return false;
 417 }
 418
 419 static struct sk_buff *handle_offloads(struct sk_buff *skb,
 420                                        const struct tnl_mutable_config *mutable,
 421                                        const struct rtable *rt,
 422                                        int tunnel_hlen)
 423 {
 424         int min_headroom;
 425         int err;
 426
 427         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 428                         + tunnel_hlen
 429                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 430
 431         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 432                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 433                                                 skb_headroom(skb) +
 434                                                 16);
 435                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 436                                         0, GFP_ATOMIC);
 437                 if (unlikely(err))
 438                         goto error_free;
 439         }
 440
 441         forward_ip_summed(skb, true);
 442
 443         if (skb_is_gso(skb)) {
 444                 struct sk_buff *nskb;
 445
 446                 nskb = __skb_gso_segment(skb, 0, false);
 447                 if (IS_ERR(nskb)) {
 448                         kfree_skb(skb);
 449                         err = PTR_ERR(nskb);
 450                         goto error;
 451                 }
 452
 453                 consume_skb(skb);
 454                 skb = nskb;
 455         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 456                 /* Pages aren't locked and could change at any time.
 457                  * If this happens after we compute the checksum, the
 458                  * checksum will be wrong.  We linearize now to avoid
 459                  * this problem.
 460                  */
 461                 if (unlikely(need_linearize(skb))) {
 462                         err = __skb_linearize(skb);
 463                         if (unlikely(err))
 464                                 goto error_free;
 465                 }
 466
 467                 err = skb_checksum_help(skb);
 468                 if (unlikely(err))
 469                         goto error_free;
 470         }
 471
 472         set_ip_summed(skb, OVS_CSUM_NONE);
 473
 474         return skb;
 475
 476 error_free:
 477         kfree_skb(skb);
 478 error:
 479         return ERR_PTR(err);
 480 }
 481
 482 static int send_frags(struct sk_buff *skb,
 483                       int tunnel_hlen)
 484 {
 485         int sent_len;
 486
 487         sent_len = 0;
 488         while (skb) {
 489                 struct sk_buff *next = skb->next;
 490                 int frag_len = skb->len - tunnel_hlen;
 491                 int err;
 492
 493                 skb->next = NULL;
 494                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 495
 496                 err = ip_local_out(skb);
 497                 skb = next;
 498                 if (unlikely(net_xmit_eval(err)))
 499                         goto free_frags;
 500                 sent_len += frag_len;
 501         }
 502
 503         return sent_len;
 504
 505 free_frags:
 506         /*
 507          * There's no point in continuing to send fragments once one has been
 508          * dropped so just free the rest.  This may help improve the congestion
 509          * that caused the first packet to be dropped.
 510          */
 511         ovs_tnl_free_linked_skbs(skb);
 512         return sent_len;
 513 }
 514
 515 /* Compute source UDP port for outgoing packet.
 516  * Currently we use the flow hash.
 517  */
 518 u16 ovs_tnl_get_src_port(struct sk_buff *skb)
 519 {
 520         int low;
 521         int high;
 522         unsigned int range;
 523         u32 hash = OVS_CB(skb)->flow->hash;
 524
 525         inet_get_local_port_range(&low, &high);
 526         range = (high - low) + 1;
 527         return (((u64) hash * range) >> 32) + low;
 528 }
 529
 530 int ovs_tnl_send(struct vport *vport, struct sk_buff *skb)
 531 {
 532         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 533         const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
 534         enum vport_err_type err = VPORT_E_TX_ERROR;
 535         struct rtable *rt;
 536         struct ovs_key_ipv4_tunnel tun_key;
 537         int sent_len = 0;
 538         int tunnel_hlen;
 539         __be16 frag_off;
 540         __be32 daddr;
 541         __be32 saddr;
 542         u32 skb_mark;
 543         u8 ttl;
 544         u8 tos;
 545
 546         /* Validate the protocol headers before we try to use them. */
 547         if (skb->protocol == htons(ETH_P_8021Q) &&
 548             !vlan_tx_tag_present(skb)) {
 549                 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
 550                         goto error_free;
 551
 552                 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 553                 skb_set_network_header(skb, VLAN_ETH_HLEN);
 554         }
 555
 556         if (skb->protocol == htons(ETH_P_IP)) {
 557                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 558                     + sizeof(struct iphdr))))
 559                         skb->protocol = 0;
 560         }
 561 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 562         else if (skb->protocol == htons(ETH_P_IPV6)) {
 563                 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
 564                     + sizeof(struct ipv6hdr))))
 565                         skb->protocol = 0;
 566         }
 567 #endif
 568
 569         /* If OVS_CB(skb)->tun_key is NULL, point it at the local tun_key here,
 570          * and zero it out.
 571          */
 572         if (!OVS_CB(skb)->tun_key) {
 573                 memset(&tun_key, 0, sizeof(tun_key));
 574                 OVS_CB(skb)->tun_key = &tun_key;
 575         }
 576
 577         tunnel_hlen = tnl_vport->tnl_ops->hdr_len(mutable, OVS_CB(skb)->tun_key);
 578         if (unlikely(tunnel_hlen < 0)) {
 579                 err = VPORT_E_TX_DROPPED;
 580                 goto error_free;
 581         }
 582         tunnel_hlen += sizeof(struct iphdr);
 583
 584         if (OVS_CB(skb)->tun_key->ipv4_dst) {
 585                 daddr = OVS_CB(skb)->tun_key->ipv4_dst;
 586                 saddr = OVS_CB(skb)->tun_key->ipv4_src;
 587                 tos = OVS_CB(skb)->tun_key->ipv4_tos;
 588                 ttl = OVS_CB(skb)->tun_key->ipv4_ttl;
 589                 frag_off = OVS_CB(skb)->tun_key->tun_flags &
 590                                 OVS_TNL_F_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 591         } else {
 592                 u8 inner_tos;
 593                 daddr = mutable->key.daddr;
 594                 saddr = mutable->key.saddr;
 595
 596                 if (unlikely(!daddr)) {
 597                         /* Trying to sent packet from Null-port without
 598                          * tunnel info? Drop this packet. */
 599                         err = VPORT_E_TX_DROPPED;
 600                         goto error_free;
 601                 }
 602
 603                 /* ToS */
 604                 if (skb->protocol == htons(ETH_P_IP))
 605                         inner_tos = ip_hdr(skb)->tos;
 606 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 607                 else if (skb->protocol == htons(ETH_P_IPV6))
 608                         inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
 609 #endif
 610                 else
 611                         inner_tos = 0;
 612
 613                 if (mutable->flags & TNL_F_TOS_INHERIT)
 614                         tos = inner_tos;
 615                 else
 616                         tos = mutable->tos;
 617
 618                 tos = INET_ECN_encapsulate(tos, inner_tos);
 619
 620                 /* TTL */
 621                 ttl = mutable->ttl;
 622                 if (mutable->flags & TNL_F_TTL_INHERIT) {
 623                         if (skb->protocol == htons(ETH_P_IP))
 624                                 ttl = ip_hdr(skb)->ttl;
 625 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 626                         else if (skb->protocol == htons(ETH_P_IPV6))
 627                                 ttl = ipv6_hdr(skb)->hop_limit;
 628 #endif
 629                 }
 630
 631                 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
 632         }
 633
 634         /* Route lookup */
 635         skb_mark = skb_get_mark(skb);
 636         rt = find_route(port_key_get_net(&mutable->key), &saddr, daddr,
 637                           tnl_vport->tnl_ops->ipproto, tos, skb_mark);
 638         if (IS_ERR(rt))
 639                 goto error_free;
 640
 641         /* Reset SKB */
 642         nf_reset(skb);
 643         secpath_reset(skb);
 644         skb_dst_drop(skb);
 645         skb_clear_rxhash(skb);
 646
 647         /* Offloading */
 648         skb = handle_offloads(skb, mutable, rt, tunnel_hlen);
 649         if (IS_ERR(skb)) {
 650                 skb = NULL;
 651                 goto err_free_rt;
 652         }
 653
 654         /* TTL Fixup. */
 655         if (!OVS_CB(skb)->tun_key->ipv4_dst) {
 656                 if (!(mutable->flags & TNL_F_TTL_INHERIT)) {
 657                         if (!ttl)
 658                                 ttl = ip4_dst_hoplimit(&rt_dst(rt));
 659                 }
 660         }
 661
 662         while (skb) {
 663                 struct iphdr *iph;
 664                 struct sk_buff *next_skb = skb->next;
 665                 skb->next = NULL;
 666
 667                 if (unlikely(vlan_deaccel_tag(skb)))
 668                         goto next;
 669
 670                 skb_push(skb, tunnel_hlen);
 671                 skb_reset_network_header(skb);
 672                 skb_set_transport_header(skb, sizeof(struct iphdr));
 673
 674                 if (next_skb)
 675                         skb_dst_set(skb, dst_clone(&rt_dst(rt)));
 676                 else
 677                         skb_dst_set(skb, &rt_dst(rt));
 678
 679                 /* Push IP header. */
 680                 iph = ip_hdr(skb);
 681                 iph->version    = 4;
 682                 iph->ihl        = sizeof(struct iphdr) >> 2;
 683                 iph->protocol   = tnl_vport->tnl_ops->ipproto;
 684                 iph->daddr      = daddr;
 685                 iph->saddr      = saddr;
 686                 iph->tos        = tos;
 687                 iph->ttl        = ttl;
 688                 iph->frag_off   = frag_off;
 689                 ip_select_ident(iph, &rt_dst(rt), NULL);
 690
 691                 /* Push Tunnel header. */
 692                 skb = tnl_vport->tnl_ops->build_header(vport, mutable,
 693                                                         &rt_dst(rt), skb, tunnel_hlen);
 694                 if (unlikely(!skb))
 695                         goto next;
 696
 697                 sent_len += send_frags(skb, tunnel_hlen);
 698
 699 next:
 700                 skb = next_skb;
 701         }
 702
 703         if (unlikely(sent_len == 0))
 704                 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
 705
 706         return sent_len;
 707
 708 err_free_rt:
 709         ip_rt_put(rt);
 710 error_free:
 711         ovs_tnl_free_linked_skbs(skb);
 712         ovs_vport_record_error(vport, err);
 713         return sent_len;
 714 }
 715
 716 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
 717         [OVS_TUNNEL_ATTR_FLAGS]    = { .type = NLA_U32 },
 718         [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
 719         [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
 720         [OVS_TUNNEL_ATTR_OUT_KEY]  = { .type = NLA_U64 },
 721         [OVS_TUNNEL_ATTR_IN_KEY]   = { .type = NLA_U64 },
 722         [OVS_TUNNEL_ATTR_TOS]      = { .type = NLA_U8 },
 723         [OVS_TUNNEL_ATTR_TTL]      = { .type = NLA_U8 },
 724         [OVS_TUNNEL_ATTR_DST_PORT] = { .type = NLA_U16 },
 725 };
 726
 727 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
 728  * zeroed. */
 729 static int tnl_set_config(struct net *net, struct nlattr *options,
 730                           const struct tnl_ops *tnl_ops,
 731                           const struct vport *cur_vport,
 732                           struct tnl_mutable_config *mutable)
 733 {
 734         const struct vport *old_vport;
 735         const struct tnl_mutable_config *old_mutable;
 736         struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
 737         int err;
 738
 739         port_key_set_net(&mutable->key, net);
 740         mutable->key.tunnel_type = tnl_ops->tunnel_type;
 741         if (!options)
 742                 goto out;
 743
 744         err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
 745         if (err)
 746                 return err;
 747
 748         /* Process attributes possibly useful for null_ports first */
 749         if (a[OVS_TUNNEL_ATTR_DST_PORT])
 750                 mutable->dst_port =
 751                         htons(nla_get_u16(a[OVS_TUNNEL_ATTR_DST_PORT]));
 752
 753         if (a[OVS_TUNNEL_ATTR_DST_IPV4])
 754                 mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
 755
 756         /* Skip the rest if configuring a null_port */
 757         if (!mutable->key.daddr)
 758                 goto out;
 759
 760         if (a[OVS_TUNNEL_ATTR_FLAGS])
 761                 mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS])
 762                         & TNL_F_PUBLIC;
 763
 764         if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
 765                 if (ipv4_is_multicast(mutable->key.daddr))
 766                         return -EINVAL;
 767                 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
 768         }
 769
 770         if (a[OVS_TUNNEL_ATTR_TOS]) {
 771                 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
 772                 /* Reject ToS config with ECN bits set. */
 773                 if (mutable->tos & INET_ECN_MASK)
 774                         return -EINVAL;
 775         }
 776
 777         if (a[OVS_TUNNEL_ATTR_TTL])
 778                 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
 779
 780         if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
 781                 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
 782                 mutable->flags |= TNL_F_IN_KEY_MATCH;
 783         } else {
 784                 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
 785                 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
 786         }
 787
 788         if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
 789                 mutable->flags |= TNL_F_OUT_KEY_ACTION;
 790         else
 791                 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
 792
 793         mutable->mlink = 0;
 794         if (ipv4_is_multicast(mutable->key.daddr)) {
 795                 struct net_device *dev;
 796                 struct rtable *rt;
 797                 __be32 saddr = mutable->key.saddr;
 798
 799                 rt = find_route(port_key_get_net(&mutable->key),
 800                              &saddr, mutable->key.daddr,
 801                              tnl_ops->ipproto, mutable->tos, 0);
 802                 if (IS_ERR(rt))
 803                         return -EADDRNOTAVAIL;
 804                 dev = rt_dst(rt).dev;
 805                 ip_rt_put(rt);
 806                 if (__in_dev_get_rtnl(dev) == NULL)
 807                         return -EADDRNOTAVAIL;
 808                 mutable->mlink = dev->ifindex;
 809                 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
 810         }
 811
 812 out:
 813         old_vport = port_table_lookup(&mutable->key, &old_mutable);
 814         if (old_vport && old_vport != cur_vport)
 815                 return -EEXIST;
 816
 817         return 0;
 818 }
 819
 820 struct vport *ovs_tnl_create(const struct vport_parms *parms,
 821                              const struct vport_ops *vport_ops,
 822                              const struct tnl_ops *tnl_ops)
 823 {
 824         struct vport *vport;
 825         struct tnl_vport *tnl_vport;
 826         struct tnl_mutable_config *mutable;
 827         int err;
 828
 829         vport = ovs_vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
 830         if (IS_ERR(vport)) {
 831                 err = PTR_ERR(vport);
 832                 goto error;
 833         }
 834
 835         tnl_vport = tnl_vport_priv(vport);
 836
 837         strcpy(tnl_vport->name, parms->name);
 838         tnl_vport->tnl_ops = tnl_ops;
 839
 840         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
 841         if (!mutable) {
 842                 err = -ENOMEM;
 843                 goto error_free_vport;
 844         }
 845
 846         err = tnl_set_config(ovs_dp_get_net(parms->dp), parms->options, tnl_ops,
 847                              NULL, mutable);
 848         if (err)
 849                 goto error_free_mutable;
 850
 851         rcu_assign_pointer(tnl_vport->mutable, mutable);
 852
 853         port_table_add_port(vport);
 854         return vport;
 855
 856 error_free_mutable:
 857         free_mutable_rtnl(mutable);
 858         kfree(mutable);
 859 error_free_vport:
 860         ovs_vport_free(vport);
 861 error:
 862         return ERR_PTR(err);
 863 }
 864
 865 int ovs_tnl_set_options(struct vport *vport, struct nlattr *options)
 866 {
 867         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 868         const struct tnl_mutable_config *old_mutable;
 869         struct tnl_mutable_config *mutable;
 870         int err;
 871
 872         old_mutable = rtnl_dereference(tnl_vport->mutable);
 873         if (!old_mutable->key.daddr)
 874                 return -EINVAL;
 875
 876         mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
 877         if (!mutable) {
 878                 err = -ENOMEM;
 879                 goto error;
 880         }
 881
 882         /* Parse the others configured by userspace. */
 883         err = tnl_set_config(ovs_dp_get_net(vport->dp), options, tnl_vport->tnl_ops,
 884                              vport, mutable);
 885         if (err)
 886                 goto error_free;
 887
 888         if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
 889                 port_table_move_port(vport, mutable);
 890         else
 891                 assign_config_rcu(vport, mutable);
 892
 893         return 0;
 894
 895 error_free:
 896         free_mutable_rtnl(mutable);
 897         kfree(mutable);
 898 error:
 899         return err;
 900 }
 901
 902 int ovs_tnl_get_options(const struct vport *vport, struct sk_buff *skb)
 903 {
 904         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 905         const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
 906
 907         if (mutable->dst_port && nla_put_u16(skb, OVS_TUNNEL_ATTR_DST_PORT,
 908                                              ntohs(mutable->dst_port)))
 909                 goto nla_put_failure;
 910
 911         /* Skip the rest for null_ports */
 912         if (!mutable->key.daddr)
 913                 return 0;
 914
 915         if (nla_put_be32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr))
 916                 goto nla_put_failure;
 917         if (nla_put_u32(skb, OVS_TUNNEL_ATTR_FLAGS,
 918                         mutable->flags & TNL_F_PUBLIC))
 919                 goto nla_put_failure;
 920         if (!(mutable->flags & TNL_F_IN_KEY_MATCH) &&
 921             nla_put_be64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key))
 922                 goto nla_put_failure;
 923         if (!(mutable->flags & TNL_F_OUT_KEY_ACTION) &&
 924             nla_put_be64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key))
 925                 goto nla_put_failure;
 926         if (mutable->key.saddr &&
 927             nla_put_be32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr))
 928                 goto nla_put_failure;
 929         if (mutable->tos && nla_put_u8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos))
 930                 goto nla_put_failure;
 931         if (mutable->ttl && nla_put_u8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl))
 932                 goto nla_put_failure;
 933
 934         return 0;
 935
 936 nla_put_failure:
 937         return -EMSGSIZE;
 938 }
 939
 940 static void free_port_rcu(struct rcu_head *rcu)
 941 {
 942         struct tnl_vport *tnl_vport = container_of(rcu,
 943                                                    struct tnl_vport, rcu);
 944
 945         kfree((struct tnl_mutable __force *)tnl_vport->mutable);
 946         ovs_vport_free(tnl_vport_to_vport(tnl_vport));
 947 }
 948
 949 void ovs_tnl_destroy(struct vport *vport)
 950 {
 951         struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 952         struct tnl_mutable_config *mutable;
 953
 954         mutable = rtnl_dereference(tnl_vport->mutable);
 955         port_table_remove_port(vport);
 956         free_mutable_rtnl(mutable);
 957         call_rcu(&tnl_vport->rcu, free_port_rcu);
 958 }
 959
 960 const char *ovs_tnl_get_name(const struct vport *vport)
 961 {
 962         const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
 963         return tnl_vport->name;
 964 }
 965
 966 void ovs_tnl_free_linked_skbs(struct sk_buff *skb)
 967 {
 968         while (skb) {
 969                 struct sk_buff *next = skb->next;
 970                 kfree_skb(skb);
 971                 skb = next;
 972         }
 973 }
 974
 975 int ovs_tnl_init(void)
 976 {
 977         int i;
 978
 979         port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
 980                              GFP_KERNEL);
 981         if (!port_table)
 982                 return -ENOMEM;
 983
 984         for (i = 0; i < PORT_TABLE_SIZE; i++)
 985                 INIT_HLIST_HEAD(&port_table[i]);
 986
 987         return 0;
 988 }
 989
 990 void ovs_tnl_exit(void)
 991 {
 992         kfree(port_table);
 993 }