net/ipv4/ip_tunnel_core.c

   1 /*
   2  * Copyright (c) 2013 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/types.h>
  22 #include <linux/kernel.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/in.h>
  26 #include <linux/if_arp.h>
  27 #include <linux/init.h>
  28 #include <linux/in6.h>
  29 #include <linux/inetdevice.h>
  30 #include <linux/netfilter_ipv4.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/if_ether.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/static_key.h>
  35
  36 #include <net/ip.h>
  37 #include <net/icmp.h>
  38 #include <net/protocol.h>
  39 #include <net/ip_tunnels.h>
  40 #include <net/arp.h>
  41 #include <net/checksum.h>
  42 #include <net/dsfield.h>
  43 #include <net/inet_ecn.h>
  44 #include <net/xfrm.h>
  45 #include <net/net_namespace.h>
  46 #include <net/netns/generic.h>
  47 #include <net/rtnetlink.h>
  48 #include <net/dst_metadata.h>
  49
  50 void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
  51                    __be32 src, __be32 dst, __u8 proto,
  52                    __u8 tos, __u8 ttl, __be16 df, bool xnet)
  53 {
  54         int pkt_len = skb->len - skb_inner_network_offset(skb);
  55         struct net *net = dev_net(rt->dst.dev);
  56         struct net_device *dev = skb->dev;
  57         struct iphdr *iph;
  58         int err;
  59
  60         skb_scrub_packet(skb, xnet);
  61
  62         skb_clear_hash(skb);
  63         skb_dst_set(skb, &rt->dst);
  64         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
  65
  66         /* Push down and install the IP header. */
  67         skb_push(skb, sizeof(struct iphdr));
  68         skb_reset_network_header(skb);
  69
  70         iph = ip_hdr(skb);
  71
  72         iph->version    =       4;
  73         iph->ihl        =       sizeof(struct iphdr) >> 2;
  74         iph->frag_off   =       df;
  75         iph->protocol   =       proto;
  76         iph->tos        =       tos;
  77         iph->daddr      =       dst;
  78         iph->saddr      =       src;
  79         iph->ttl        =       ttl;
  80         __ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);
  81
  82         err = ip_local_out(net, sk, skb);
  83         if (unlikely(net_xmit_eval(err)))
  84                 pkt_len = 0;
  85         iptunnel_xmit_stats(dev, pkt_len);
  86 }
  87 EXPORT_SYMBOL_GPL(iptunnel_xmit);
  88
  89 int iptunnel_pull_header(struct sk_buff *skb, int hdr_len, __be16 inner_proto,
  90                          bool xnet)
  91 {
  92         if (unlikely(!pskb_may_pull(skb, hdr_len)))
  93                 return -ENOMEM;
  94
  95         skb_pull_rcsum(skb, hdr_len);
  96
  97         if (inner_proto == htons(ETH_P_TEB)) {
  98                 struct ethhdr *eh;
  99
 100                 if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
 101                         return -ENOMEM;
 102
 103                 eh = (struct ethhdr *)skb->data;
 104                 if (likely(eth_proto_is_802_3(eh->h_proto)))
 105                         skb->protocol = eh->h_proto;
 106                 else
 107                         skb->protocol = htons(ETH_P_802_2);
 108
 109         } else {
 110                 skb->protocol = inner_proto;
 111         }
 112
 113         skb_clear_hash_if_not_l4(skb);
 114         skb->vlan_tci = 0;
 115         skb_set_queue_mapping(skb, 0);
 116         skb_scrub_packet(skb, xnet);
 117         return 0;
 118 }
 119 EXPORT_SYMBOL_GPL(iptunnel_pull_header);
 120
 121 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
 122                                              gfp_t flags)
 123 {
 124         struct metadata_dst *res;
 125         struct ip_tunnel_info *dst, *src;
 126
 127         if (!md || md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
 128                 return NULL;
 129
 130         res = metadata_dst_alloc(0, flags);
 131         if (!res)
 132                 return NULL;
 133
 134         dst = &res->u.tun_info;
 135         src = &md->u.tun_info;
 136         dst->key.tun_id = src->key.tun_id;
 137         if (src->mode & IP_TUNNEL_INFO_IPV6)
 138                 memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
 139                        sizeof(struct in6_addr));
 140         else
 141                 dst->key.u.ipv4.dst = src->key.u.ipv4.src;
 142         dst->mode = src->mode | IP_TUNNEL_INFO_TX;
 143
 144         return res;
 145 }
 146 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
 147
 148 struct sk_buff *iptunnel_handle_offloads(struct sk_buff *skb,
 149                                          int gso_type_mask)
 150 {
 151         int err;
 152
 153         if (likely(!skb->encapsulation)) {
 154                 skb_reset_inner_headers(skb);
 155                 skb->encapsulation = 1;
 156         }
 157
 158         if (skb_is_gso(skb)) {
 159                 err = skb_unclone(skb, GFP_ATOMIC);
 160                 if (unlikely(err))
 161                         goto error;
 162                 skb_shinfo(skb)->gso_type |= gso_type_mask;
 163                 return skb;
 164         }
 165
 166         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 167                 skb->ip_summed = CHECKSUM_NONE;
 168                 /* We clear encapsulation here to prevent badly-written
 169                  * drivers potentially deciding to offload an inner checksum
 170                  * if we set CHECKSUM_PARTIAL on the outer header.
 171                  * This should go away when the drivers are all fixed.
 172                  */
 173                 skb->encapsulation = 0;
 174         }
 175
 176         return skb;
 177 error:
 178         kfree_skb(skb);
 179         return ERR_PTR(err);
 180 }
 181 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
 182
 183 /* Often modified stats are per cpu, other are shared (netdev->stats) */
 184 struct rtnl_link_stats64 *ip_tunnel_get_stats64(struct net_device *dev,
 185                                                 struct rtnl_link_stats64 *tot)
 186 {
 187         int i;
 188
 189         netdev_stats_to_stats64(tot, &dev->stats);
 190
 191         for_each_possible_cpu(i) {
 192                 const struct pcpu_sw_netstats *tstats =
 193                                                    per_cpu_ptr(dev->tstats, i);
 194                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
 195                 unsigned int start;
 196
 197                 do {
 198                         start = u64_stats_fetch_begin_irq(&tstats->syncp);
 199                         rx_packets = tstats->rx_packets;
 200                         tx_packets = tstats->tx_packets;
 201                         rx_bytes = tstats->rx_bytes;
 202                         tx_bytes = tstats->tx_bytes;
 203                 } while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
 204
 205                 tot->rx_packets += rx_packets;
 206                 tot->tx_packets += tx_packets;
 207                 tot->rx_bytes   += rx_bytes;
 208                 tot->tx_bytes   += tx_bytes;
 209         }
 210
 211         return tot;
 212 }
 213 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
 214
 215 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
 216         [LWTUNNEL_IP_ID]        = { .type = NLA_U64 },
 217         [LWTUNNEL_IP_DST]       = { .type = NLA_U32 },
 218         [LWTUNNEL_IP_SRC]       = { .type = NLA_U32 },
 219         [LWTUNNEL_IP_TTL]       = { .type = NLA_U8 },
 220         [LWTUNNEL_IP_TOS]       = { .type = NLA_U8 },
 221         [LWTUNNEL_IP_FLAGS]     = { .type = NLA_U16 },
 222 };
 223
 224 static int ip_tun_build_state(struct net_device *dev, struct nlattr *attr,
 225                               unsigned int family, const void *cfg,
 226                               struct lwtunnel_state **ts)
 227 {
 228         struct ip_tunnel_info *tun_info;
 229         struct lwtunnel_state *new_state;
 230         struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
 231         int err;
 232
 233         err = nla_parse_nested(tb, LWTUNNEL_IP_MAX, attr, ip_tun_policy);
 234         if (err < 0)
 235                 return err;
 236
 237         new_state = lwtunnel_state_alloc(sizeof(*tun_info));
 238         if (!new_state)
 239                 return -ENOMEM;
 240
 241         new_state->type = LWTUNNEL_ENCAP_IP;
 242
 243         tun_info = lwt_tun_info(new_state);
 244
 245         if (tb[LWTUNNEL_IP_ID])
 246                 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
 247
 248         if (tb[LWTUNNEL_IP_DST])
 249                 tun_info->key.u.ipv4.dst = nla_get_be32(tb[LWTUNNEL_IP_DST]);
 250
 251         if (tb[LWTUNNEL_IP_SRC])
 252                 tun_info->key.u.ipv4.src = nla_get_be32(tb[LWTUNNEL_IP_SRC]);
 253
 254         if (tb[LWTUNNEL_IP_TTL])
 255                 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
 256
 257         if (tb[LWTUNNEL_IP_TOS])
 258                 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
 259
 260         if (tb[LWTUNNEL_IP_FLAGS])
 261                 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
 262
 263         tun_info->mode = IP_TUNNEL_INFO_TX;
 264         tun_info->options_len = 0;
 265
 266         *ts = new_state;
 267
 268         return 0;
 269 }
 270
 271 static int ip_tun_fill_encap_info(struct sk_buff *skb,
 272                                   struct lwtunnel_state *lwtstate)
 273 {
 274         struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
 275
 276         if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id) ||
 277             nla_put_be32(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
 278             nla_put_be32(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
 279             nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
 280             nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
 281             nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
 282                 return -ENOMEM;
 283
 284         return 0;
 285 }
 286
 287 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
 288 {
 289         return nla_total_size(8)        /* LWTUNNEL_IP_ID */
 290                 + nla_total_size(4)     /* LWTUNNEL_IP_DST */
 291                 + nla_total_size(4)     /* LWTUNNEL_IP_SRC */
 292                 + nla_total_size(1)     /* LWTUNNEL_IP_TOS */
 293                 + nla_total_size(1)     /* LWTUNNEL_IP_TTL */
 294                 + nla_total_size(2);    /* LWTUNNEL_IP_FLAGS */
 295 }
 296
 297 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
 298 {
 299         return memcmp(lwt_tun_info(a), lwt_tun_info(b),
 300                       sizeof(struct ip_tunnel_info));
 301 }
 302
 303 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
 304         .build_state = ip_tun_build_state,
 305         .fill_encap = ip_tun_fill_encap_info,
 306         .get_encap_size = ip_tun_encap_nlsize,
 307         .cmp_encap = ip_tun_cmp_encap,
 308 };
 309
 310 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
 311         [LWTUNNEL_IP6_ID]               = { .type = NLA_U64 },
 312         [LWTUNNEL_IP6_DST]              = { .len = sizeof(struct in6_addr) },
 313         [LWTUNNEL_IP6_SRC]              = { .len = sizeof(struct in6_addr) },
 314         [LWTUNNEL_IP6_HOPLIMIT]         = { .type = NLA_U8 },
 315         [LWTUNNEL_IP6_TC]               = { .type = NLA_U8 },
 316         [LWTUNNEL_IP6_FLAGS]            = { .type = NLA_U16 },
 317 };
 318
 319 static int ip6_tun_build_state(struct net_device *dev, struct nlattr *attr,
 320                                unsigned int family, const void *cfg,
 321                                struct lwtunnel_state **ts)
 322 {
 323         struct ip_tunnel_info *tun_info;
 324         struct lwtunnel_state *new_state;
 325         struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
 326         int err;
 327
 328         err = nla_parse_nested(tb, LWTUNNEL_IP6_MAX, attr, ip6_tun_policy);
 329         if (err < 0)
 330                 return err;
 331
 332         new_state = lwtunnel_state_alloc(sizeof(*tun_info));
 333         if (!new_state)
 334                 return -ENOMEM;
 335
 336         new_state->type = LWTUNNEL_ENCAP_IP6;
 337
 338         tun_info = lwt_tun_info(new_state);
 339
 340         if (tb[LWTUNNEL_IP6_ID])
 341                 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
 342
 343         if (tb[LWTUNNEL_IP6_DST])
 344                 tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
 345
 346         if (tb[LWTUNNEL_IP6_SRC])
 347                 tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
 348
 349         if (tb[LWTUNNEL_IP6_HOPLIMIT])
 350                 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
 351
 352         if (tb[LWTUNNEL_IP6_TC])
 353                 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
 354
 355         if (tb[LWTUNNEL_IP6_FLAGS])
 356                 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
 357
 358         tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
 359         tun_info->options_len = 0;
 360
 361         *ts = new_state;
 362
 363         return 0;
 364 }
 365
 366 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
 367                                    struct lwtunnel_state *lwtstate)
 368 {
 369         struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
 370
 371         if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id) ||
 372             nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
 373             nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
 374             nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.tos) ||
 375             nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.ttl) ||
 376             nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
 377                 return -ENOMEM;
 378
 379         return 0;
 380 }
 381
 382 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
 383 {
 384         return nla_total_size(8)        /* LWTUNNEL_IP6_ID */
 385                 + nla_total_size(16)    /* LWTUNNEL_IP6_DST */
 386                 + nla_total_size(16)    /* LWTUNNEL_IP6_SRC */
 387                 + nla_total_size(1)     /* LWTUNNEL_IP6_HOPLIMIT */
 388                 + nla_total_size(1)     /* LWTUNNEL_IP6_TC */
 389                 + nla_total_size(2);    /* LWTUNNEL_IP6_FLAGS */
 390 }
 391
 392 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
 393         .build_state = ip6_tun_build_state,
 394         .fill_encap = ip6_tun_fill_encap_info,
 395         .get_encap_size = ip6_tun_encap_nlsize,
 396         .cmp_encap = ip_tun_cmp_encap,
 397 };
 398
 399 void __init ip_tunnel_core_init(void)
 400 {
 401         lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
 402         lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
 403 }
 404
 405 struct static_key ip_tunnel_metadata_cnt = STATIC_KEY_INIT_FALSE;
 406 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
 407
 408 void ip_tunnel_need_metadata(void)
 409 {
 410         static_key_slow_inc(&ip_tunnel_metadata_cnt);
 411 }
 412 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
 413
 414 void ip_tunnel_unneed_metadata(void)
 415 {
 416         static_key_slow_dec(&ip_tunnel_metadata_cnt);
 417 }
 418 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);