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,
  90                            __be16 inner_proto, bool raw_proto, 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 (!raw_proto && 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
 118         return iptunnel_pull_offloads(skb);
 119 }
 120 EXPORT_SYMBOL_GPL(__iptunnel_pull_header);
 121
 122 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
 123                                              gfp_t flags)
 124 {
 125         struct metadata_dst *res;
 126         struct ip_tunnel_info *dst, *src;
 127
 128         if (!md || md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
 129                 return NULL;
 130
 131         res = metadata_dst_alloc(0, flags);
 132         if (!res)
 133                 return NULL;
 134
 135         dst = &res->u.tun_info;
 136         src = &md->u.tun_info;
 137         dst->key.tun_id = src->key.tun_id;
 138         if (src->mode & IP_TUNNEL_INFO_IPV6)
 139                 memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
 140                        sizeof(struct in6_addr));
 141         else
 142                 dst->key.u.ipv4.dst = src->key.u.ipv4.src;
 143         dst->mode = src->mode | IP_TUNNEL_INFO_TX;
 144
 145         return res;
 146 }
 147 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
 148
 149 int iptunnel_handle_offloads(struct sk_buff *skb,
 150                              int gso_type_mask)
 151 {
 152         int err;
 153
 154         if (likely(!skb->encapsulation)) {
 155                 skb_reset_inner_headers(skb);
 156                 skb->encapsulation = 1;
 157         }
 158
 159         if (skb_is_gso(skb)) {
 160                 err = skb_unclone(skb, GFP_ATOMIC);
 161                 if (unlikely(err))
 162                         return err;
 163                 skb_shinfo(skb)->gso_type |= gso_type_mask;
 164                 return 0;
 165         }
 166
 167         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 168                 skb->ip_summed = CHECKSUM_NONE;
 169                 /* We clear encapsulation here to prevent badly-written
 170                  * drivers potentially deciding to offload an inner checksum
 171                  * if we set CHECKSUM_PARTIAL on the outer header.
 172                  * This should go away when the drivers are all fixed.
 173                  */
 174                 skb->encapsulation = 0;
 175         }
 176
 177         return 0;
 178 }
 179 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
 180
 181 /* Often modified stats are per cpu, other are shared (netdev->stats) */
 182 struct rtnl_link_stats64 *ip_tunnel_get_stats64(struct net_device *dev,
 183                                                 struct rtnl_link_stats64 *tot)
 184 {
 185         int i;
 186
 187         netdev_stats_to_stats64(tot, &dev->stats);
 188
 189         for_each_possible_cpu(i) {
 190                 const struct pcpu_sw_netstats *tstats =
 191                                                    per_cpu_ptr(dev->tstats, i);
 192                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
 193                 unsigned int start;
 194
 195                 do {
 196                         start = u64_stats_fetch_begin_irq(&tstats->syncp);
 197                         rx_packets = tstats->rx_packets;
 198                         tx_packets = tstats->tx_packets;
 199                         rx_bytes = tstats->rx_bytes;
 200                         tx_bytes = tstats->tx_bytes;
 201                 } while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
 202
 203                 tot->rx_packets += rx_packets;
 204                 tot->tx_packets += tx_packets;
 205                 tot->rx_bytes   += rx_bytes;
 206                 tot->tx_bytes   += tx_bytes;
 207         }
 208
 209         return tot;
 210 }
 211 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
 212
 213 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
 214         [LWTUNNEL_IP_ID]        = { .type = NLA_U64 },
 215         [LWTUNNEL_IP_DST]       = { .type = NLA_U32 },
 216         [LWTUNNEL_IP_SRC]       = { .type = NLA_U32 },
 217         [LWTUNNEL_IP_TTL]       = { .type = NLA_U8 },
 218         [LWTUNNEL_IP_TOS]       = { .type = NLA_U8 },
 219         [LWTUNNEL_IP_FLAGS]     = { .type = NLA_U16 },
 220 };
 221
 222 static int ip_tun_build_state(struct net_device *dev, struct nlattr *attr,
 223                               unsigned int family, const void *cfg,
 224                               struct lwtunnel_state **ts)
 225 {
 226         struct ip_tunnel_info *tun_info;
 227         struct lwtunnel_state *new_state;
 228         struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
 229         int err;
 230
 231         err = nla_parse_nested(tb, LWTUNNEL_IP_MAX, attr, ip_tun_policy);
 232         if (err < 0)
 233                 return err;
 234
 235         new_state = lwtunnel_state_alloc(sizeof(*tun_info));
 236         if (!new_state)
 237                 return -ENOMEM;
 238
 239         new_state->type = LWTUNNEL_ENCAP_IP;
 240
 241         tun_info = lwt_tun_info(new_state);
 242
 243         if (tb[LWTUNNEL_IP_ID])
 244                 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
 245
 246         if (tb[LWTUNNEL_IP_DST])
 247                 tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
 248
 249         if (tb[LWTUNNEL_IP_SRC])
 250                 tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
 251
 252         if (tb[LWTUNNEL_IP_TTL])
 253                 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
 254
 255         if (tb[LWTUNNEL_IP_TOS])
 256                 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
 257
 258         if (tb[LWTUNNEL_IP_FLAGS])
 259                 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
 260
 261         tun_info->mode = IP_TUNNEL_INFO_TX;
 262         tun_info->options_len = 0;
 263
 264         *ts = new_state;
 265
 266         return 0;
 267 }
 268
 269 static int ip_tun_fill_encap_info(struct sk_buff *skb,
 270                                   struct lwtunnel_state *lwtstate)
 271 {
 272         struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
 273
 274         if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
 275                          LWTUNNEL_IP_PAD) ||
 276             nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
 277             nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
 278             nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
 279             nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
 280             nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
 281                 return -ENOMEM;
 282
 283         return 0;
 284 }
 285
 286 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
 287 {
 288         return nla_total_size_64bit(8)  /* LWTUNNEL_IP_ID */
 289                 + nla_total_size(4)     /* LWTUNNEL_IP_DST */
 290                 + nla_total_size(4)     /* LWTUNNEL_IP_SRC */
 291                 + nla_total_size(1)     /* LWTUNNEL_IP_TOS */
 292                 + nla_total_size(1)     /* LWTUNNEL_IP_TTL */
 293                 + nla_total_size(2);    /* LWTUNNEL_IP_FLAGS */
 294 }
 295
 296 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
 297 {
 298         return memcmp(lwt_tun_info(a), lwt_tun_info(b),
 299                       sizeof(struct ip_tunnel_info));
 300 }
 301
 302 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
 303         .build_state = ip_tun_build_state,
 304         .fill_encap = ip_tun_fill_encap_info,
 305         .get_encap_size = ip_tun_encap_nlsize,
 306         .cmp_encap = ip_tun_cmp_encap,
 307 };
 308
 309 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
 310         [LWTUNNEL_IP6_ID]               = { .type = NLA_U64 },
 311         [LWTUNNEL_IP6_DST]              = { .len = sizeof(struct in6_addr) },
 312         [LWTUNNEL_IP6_SRC]              = { .len = sizeof(struct in6_addr) },
 313         [LWTUNNEL_IP6_HOPLIMIT]         = { .type = NLA_U8 },
 314         [LWTUNNEL_IP6_TC]               = { .type = NLA_U8 },
 315         [LWTUNNEL_IP6_FLAGS]            = { .type = NLA_U16 },
 316 };
 317
 318 static int ip6_tun_build_state(struct net_device *dev, struct nlattr *attr,
 319                                unsigned int family, const void *cfg,
 320                                struct lwtunnel_state **ts)
 321 {
 322         struct ip_tunnel_info *tun_info;
 323         struct lwtunnel_state *new_state;
 324         struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
 325         int err;
 326
 327         err = nla_parse_nested(tb, LWTUNNEL_IP6_MAX, attr, ip6_tun_policy);
 328         if (err < 0)
 329                 return err;
 330
 331         new_state = lwtunnel_state_alloc(sizeof(*tun_info));
 332         if (!new_state)
 333                 return -ENOMEM;
 334
 335         new_state->type = LWTUNNEL_ENCAP_IP6;
 336
 337         tun_info = lwt_tun_info(new_state);
 338
 339         if (tb[LWTUNNEL_IP6_ID])
 340                 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
 341
 342         if (tb[LWTUNNEL_IP6_DST])
 343                 tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
 344
 345         if (tb[LWTUNNEL_IP6_SRC])
 346                 tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
 347
 348         if (tb[LWTUNNEL_IP6_HOPLIMIT])
 349                 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
 350
 351         if (tb[LWTUNNEL_IP6_TC])
 352                 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
 353
 354         if (tb[LWTUNNEL_IP6_FLAGS])
 355                 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
 356
 357         tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
 358         tun_info->options_len = 0;
 359
 360         *ts = new_state;
 361
 362         return 0;
 363 }
 364
 365 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
 366                                    struct lwtunnel_state *lwtstate)
 367 {
 368         struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
 369
 370         if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
 371                          LWTUNNEL_IP6_PAD) ||
 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_TC, tun_info->key.tos) ||
 375             nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, 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_64bit(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         /* If you land here, make sure whether increasing ip_tunnel_info's
 402          * options_len is a reasonable choice with its usage in front ends
 403          * (f.e., it's part of flow keys, etc).
 404          */
 405         BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
 406
 407         lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
 408         lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
 409 }
 410
 411 struct static_key ip_tunnel_metadata_cnt = STATIC_KEY_INIT_FALSE;
 412 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
 413
 414 void ip_tunnel_need_metadata(void)
 415 {
 416         static_key_slow_inc(&ip_tunnel_metadata_cnt);
 417 }
 418 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
 419
 420 void ip_tunnel_unneed_metadata(void)
 421 {
 422         static_key_slow_dec(&ip_tunnel_metadata_cnt);
 423 }
 424 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);