net/ipv4/ip_tunnel_core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2013 Nicira, Inc.
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #include <linux/types.h>
   9 #include <linux/kernel.h>
  10 #include <linux/skbuff.h>
  11 #include <linux/netdevice.h>
  12 #include <linux/in.h>
  13 #include <linux/if_arp.h>
  14 #include <linux/init.h>
  15 #include <linux/in6.h>
  16 #include <linux/inetdevice.h>
  17 #include <linux/netfilter_ipv4.h>
  18 #include <linux/etherdevice.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/if_vlan.h>
  21 #include <linux/static_key.h>
  22
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/protocol.h>
  26 #include <net/ip_tunnels.h>
  27 #include <net/ip6_tunnel.h>
  28 #include <net/arp.h>
  29 #include <net/checksum.h>
  30 #include <net/dsfield.h>
  31 #include <net/inet_ecn.h>
  32 #include <net/xfrm.h>
  33 #include <net/net_namespace.h>
  34 #include <net/netns/generic.h>
  35 #include <net/rtnetlink.h>
  36 #include <net/dst_metadata.h>
  37
  38 const struct ip_tunnel_encap_ops __rcu *
  39                 iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
  40 EXPORT_SYMBOL(iptun_encaps);
  41
  42 const struct ip6_tnl_encap_ops __rcu *
  43                 ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
  44 EXPORT_SYMBOL(ip6tun_encaps);
  45
  46 void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
  47                    __be32 src, __be32 dst, __u8 proto,
  48                    __u8 tos, __u8 ttl, __be16 df, bool xnet)
  49 {
  50         int pkt_len = skb->len - skb_inner_network_offset(skb);
  51         struct net *net = dev_net(rt->dst.dev);
  52         struct net_device *dev = skb->dev;
  53         struct iphdr *iph;
  54         int err;
  55
  56         skb_scrub_packet(skb, xnet);
  57
  58         skb_clear_hash_if_not_l4(skb);
  59         skb_dst_set(skb, &rt->dst);
  60         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
  61
  62         /* Push down and install the IP header. */
  63         skb_push(skb, sizeof(struct iphdr));
  64         skb_reset_network_header(skb);
  65
  66         iph = ip_hdr(skb);
  67
  68         iph->version    =       4;
  69         iph->ihl        =       sizeof(struct iphdr) >> 2;
  70         iph->frag_off   =       ip_mtu_locked(&rt->dst) ? 0 : df;
  71         iph->protocol   =       proto;
  72         iph->tos        =       tos;
  73         iph->daddr      =       dst;
  74         iph->saddr      =       src;
  75         iph->ttl        =       ttl;
  76         __ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);
  77
  78         err = ip_local_out(net, sk, skb);
  79
  80         if (dev) {
  81                 if (unlikely(net_xmit_eval(err)))
  82                         pkt_len = 0;
  83                 iptunnel_xmit_stats(dev, pkt_len);
  84         }
  85 }
  86 EXPORT_SYMBOL_GPL(iptunnel_xmit);
  87
  88 int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len,
  89                            __be16 inner_proto, bool raw_proto, bool xnet)
  90 {
  91         if (unlikely(!pskb_may_pull(skb, hdr_len)))
  92                 return -ENOMEM;
  93
  94         skb_pull_rcsum(skb, hdr_len);
  95
  96         if (!raw_proto && inner_proto == htons(ETH_P_TEB)) {
  97                 struct ethhdr *eh;
  98
  99                 if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
 100                         return -ENOMEM;
 101
 102                 eh = (struct ethhdr *)skb->data;
 103                 if (likely(eth_proto_is_802_3(eh->h_proto)))
 104                         skb->protocol = eh->h_proto;
 105                 else
 106                         skb->protocol = htons(ETH_P_802_2);
 107
 108         } else {
 109                 skb->protocol = inner_proto;
 110         }
 111
 112         skb_clear_hash_if_not_l4(skb);
 113         __vlan_hwaccel_clear_tag(skb);
 114         skb_set_queue_mapping(skb, 0);
 115         skb_scrub_packet(skb, xnet);
 116
 117         return iptunnel_pull_offloads(skb);
 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->type != METADATA_IP_TUNNEL ||
 128             md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
 129                 return NULL;
 130
 131         src = &md->u.tun_info;
 132         res = metadata_dst_alloc(src->options_len, METADATA_IP_TUNNEL, flags);
 133         if (!res)
 134                 return NULL;
 135
 136         dst = &res->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->key.tun_flags = src->key.tun_flags;
 144         dst->mode = src->mode | IP_TUNNEL_INFO_TX;
 145         ip_tunnel_info_opts_set(dst, ip_tunnel_info_opts(src),
 146                                 src->options_len, 0);
 147
 148         return res;
 149 }
 150 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
 151
 152 int iptunnel_handle_offloads(struct sk_buff *skb,
 153                              int gso_type_mask)
 154 {
 155         int err;
 156
 157         if (likely(!skb->encapsulation)) {
 158                 skb_reset_inner_headers(skb);
 159                 skb->encapsulation = 1;
 160         }
 161
 162         if (skb_is_gso(skb)) {
 163                 err = skb_header_unclone(skb, GFP_ATOMIC);
 164                 if (unlikely(err))
 165                         return err;
 166                 skb_shinfo(skb)->gso_type |= gso_type_mask;
 167                 return 0;
 168         }
 169
 170         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 171                 skb->ip_summed = CHECKSUM_NONE;
 172                 /* We clear encapsulation here to prevent badly-written
 173                  * drivers potentially deciding to offload an inner checksum
 174                  * if we set CHECKSUM_PARTIAL on the outer header.
 175                  * This should go away when the drivers are all fixed.
 176                  */
 177                 skb->encapsulation = 0;
 178         }
 179
 180         return 0;
 181 }
 182 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
 183
 184 /* Often modified stats are per cpu, other are shared (netdev->stats) */
 185 void ip_tunnel_get_stats64(struct net_device *dev,
 186                            struct rtnl_link_stats64 *tot)
 187 {
 188         int i;
 189
 190         netdev_stats_to_stats64(tot, &dev->stats);
 191
 192         for_each_possible_cpu(i) {
 193                 const struct pcpu_sw_netstats *tstats =
 194                                                    per_cpu_ptr(dev->tstats, i);
 195                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
 196                 unsigned int start;
 197
 198                 do {
 199                         start = u64_stats_fetch_begin_irq(&tstats->syncp);
 200                         rx_packets = tstats->rx_packets;
 201                         tx_packets = tstats->tx_packets;
 202                         rx_bytes = tstats->rx_bytes;
 203                         tx_bytes = tstats->tx_bytes;
 204                 } while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
 205
 206                 tot->rx_packets += rx_packets;
 207                 tot->tx_packets += tx_packets;
 208                 tot->rx_bytes   += rx_bytes;
 209                 tot->tx_bytes   += tx_bytes;
 210         }
 211 }
 212 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
 213
 214 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
 215         [LWTUNNEL_IP_ID]        = { .type = NLA_U64 },
 216         [LWTUNNEL_IP_DST]       = { .type = NLA_U32 },
 217         [LWTUNNEL_IP_SRC]       = { .type = NLA_U32 },
 218         [LWTUNNEL_IP_TTL]       = { .type = NLA_U8 },
 219         [LWTUNNEL_IP_TOS]       = { .type = NLA_U8 },
 220         [LWTUNNEL_IP_FLAGS]     = { .type = NLA_U16 },
 221 };
 222
 223 static int ip_tun_build_state(struct nlattr *attr,
 224                               unsigned int family, const void *cfg,
 225                               struct lwtunnel_state **ts,
 226                               struct netlink_ext_ack *extack)
 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_deprecated(tb, LWTUNNEL_IP_MAX, attr,
 234                                           ip_tun_policy, extack);
 235         if (err < 0)
 236                 return err;
 237
 238         new_state = lwtunnel_state_alloc(sizeof(*tun_info));
 239         if (!new_state)
 240                 return -ENOMEM;
 241
 242         new_state->type = LWTUNNEL_ENCAP_IP;
 243
 244         tun_info = lwt_tun_info(new_state);
 245
 246 #ifdef CONFIG_DST_CACHE
 247         err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL);
 248         if (err) {
 249                 lwtstate_free(new_state);
 250                 return err;
 251         }
 252 #endif
 253
 254         if (tb[LWTUNNEL_IP_ID])
 255                 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
 256
 257         if (tb[LWTUNNEL_IP_DST])
 258                 tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
 259
 260         if (tb[LWTUNNEL_IP_SRC])
 261                 tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
 262
 263         if (tb[LWTUNNEL_IP_TTL])
 264                 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
 265
 266         if (tb[LWTUNNEL_IP_TOS])
 267                 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
 268
 269         if (tb[LWTUNNEL_IP_FLAGS])
 270                 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
 271
 272         tun_info->mode = IP_TUNNEL_INFO_TX;
 273         tun_info->options_len = 0;
 274
 275         *ts = new_state;
 276
 277         return 0;
 278 }
 279
 280 static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate)
 281 {
 282 #ifdef CONFIG_DST_CACHE
 283         struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
 284
 285         dst_cache_destroy(&tun_info->dst_cache);
 286 #endif
 287 }
 288
 289 static int ip_tun_fill_encap_info(struct sk_buff *skb,
 290                                   struct lwtunnel_state *lwtstate)
 291 {
 292         struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
 293
 294         if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
 295                          LWTUNNEL_IP_PAD) ||
 296             nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
 297             nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
 298             nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
 299             nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
 300             nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
 301                 return -ENOMEM;
 302
 303         return 0;
 304 }
 305
 306 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
 307 {
 308         return nla_total_size_64bit(8)  /* LWTUNNEL_IP_ID */
 309                 + nla_total_size(4)     /* LWTUNNEL_IP_DST */
 310                 + nla_total_size(4)     /* LWTUNNEL_IP_SRC */
 311                 + nla_total_size(1)     /* LWTUNNEL_IP_TOS */
 312                 + nla_total_size(1)     /* LWTUNNEL_IP_TTL */
 313                 + nla_total_size(2);    /* LWTUNNEL_IP_FLAGS */
 314 }
 315
 316 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
 317 {
 318         return memcmp(lwt_tun_info(a), lwt_tun_info(b),
 319                       sizeof(struct ip_tunnel_info));
 320 }
 321
 322 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
 323         .build_state = ip_tun_build_state,
 324         .destroy_state = ip_tun_destroy_state,
 325         .fill_encap = ip_tun_fill_encap_info,
 326         .get_encap_size = ip_tun_encap_nlsize,
 327         .cmp_encap = ip_tun_cmp_encap,
 328         .owner = THIS_MODULE,
 329 };
 330
 331 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
 332         [LWTUNNEL_IP6_ID]               = { .type = NLA_U64 },
 333         [LWTUNNEL_IP6_DST]              = { .len = sizeof(struct in6_addr) },
 334         [LWTUNNEL_IP6_SRC]              = { .len = sizeof(struct in6_addr) },
 335         [LWTUNNEL_IP6_HOPLIMIT]         = { .type = NLA_U8 },
 336         [LWTUNNEL_IP6_TC]               = { .type = NLA_U8 },
 337         [LWTUNNEL_IP6_FLAGS]            = { .type = NLA_U16 },
 338 };
 339
 340 static int ip6_tun_build_state(struct nlattr *attr,
 341                                unsigned int family, const void *cfg,
 342                                struct lwtunnel_state **ts,
 343                                struct netlink_ext_ack *extack)
 344 {
 345         struct ip_tunnel_info *tun_info;
 346         struct lwtunnel_state *new_state;
 347         struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
 348         int err;
 349
 350         err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr,
 351                                           ip6_tun_policy, extack);
 352         if (err < 0)
 353                 return err;
 354
 355         new_state = lwtunnel_state_alloc(sizeof(*tun_info));
 356         if (!new_state)
 357                 return -ENOMEM;
 358
 359         new_state->type = LWTUNNEL_ENCAP_IP6;
 360
 361         tun_info = lwt_tun_info(new_state);
 362
 363         if (tb[LWTUNNEL_IP6_ID])
 364                 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
 365
 366         if (tb[LWTUNNEL_IP6_DST])
 367                 tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
 368
 369         if (tb[LWTUNNEL_IP6_SRC])
 370                 tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
 371
 372         if (tb[LWTUNNEL_IP6_HOPLIMIT])
 373                 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
 374
 375         if (tb[LWTUNNEL_IP6_TC])
 376                 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
 377
 378         if (tb[LWTUNNEL_IP6_FLAGS])
 379                 tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
 380
 381         tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
 382         tun_info->options_len = 0;
 383
 384         *ts = new_state;
 385
 386         return 0;
 387 }
 388
 389 static int ip6_tun_fill_encap_info(struct sk_buff *skb,
 390                                    struct lwtunnel_state *lwtstate)
 391 {
 392         struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
 393
 394         if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
 395                          LWTUNNEL_IP6_PAD) ||
 396             nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
 397             nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
 398             nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) ||
 399             nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) ||
 400             nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
 401                 return -ENOMEM;
 402
 403         return 0;
 404 }
 405
 406 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
 407 {
 408         return nla_total_size_64bit(8)  /* LWTUNNEL_IP6_ID */
 409                 + nla_total_size(16)    /* LWTUNNEL_IP6_DST */
 410                 + nla_total_size(16)    /* LWTUNNEL_IP6_SRC */
 411                 + nla_total_size(1)     /* LWTUNNEL_IP6_HOPLIMIT */
 412                 + nla_total_size(1)     /* LWTUNNEL_IP6_TC */
 413                 + nla_total_size(2);    /* LWTUNNEL_IP6_FLAGS */
 414 }
 415
 416 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
 417         .build_state = ip6_tun_build_state,
 418         .fill_encap = ip6_tun_fill_encap_info,
 419         .get_encap_size = ip6_tun_encap_nlsize,
 420         .cmp_encap = ip_tun_cmp_encap,
 421         .owner = THIS_MODULE,
 422 };
 423
 424 void __init ip_tunnel_core_init(void)
 425 {
 426         /* If you land here, make sure whether increasing ip_tunnel_info's
 427          * options_len is a reasonable choice with its usage in front ends
 428          * (f.e., it's part of flow keys, etc).
 429          */
 430         BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
 431
 432         lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
 433         lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
 434 }
 435
 436 DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt);
 437 EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
 438
 439 void ip_tunnel_need_metadata(void)
 440 {
 441         static_branch_inc(&ip_tunnel_metadata_cnt);
 442 }
 443 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
 444
 445 void ip_tunnel_unneed_metadata(void)
 446 {
 447         static_branch_dec(&ip_tunnel_metadata_cnt);
 448 }
 449 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);