net/ipv6/netfilter/nf_nat_l3proto_ipv6.c

   1 /*
   2  * Copyright (c) 2011 Patrick McHardy <kaber@trash.net>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * Development of IPv6 NAT funded by Astaro.
   9  */
  10 #include <linux/types.h>
  11 #include <linux/module.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/ipv6.h>
  14 #include <linux/netfilter.h>
  15 #include <linux/netfilter_ipv6.h>
  16 #include <net/secure_seq.h>
  17 #include <net/checksum.h>
  18 #include <net/ip6_checksum.h>
  19 #include <net/ip6_route.h>
  20 #include <net/ipv6.h>
  21
  22 #include <net/netfilter/nf_conntrack_core.h>
  23 #include <net/netfilter/nf_conntrack.h>
  24 #include <net/netfilter/nf_nat_core.h>
  25 #include <net/netfilter/nf_nat_l3proto.h>
  26 #include <net/netfilter/nf_nat_l4proto.h>
  27
  28 static const struct nf_nat_l3proto nf_nat_l3proto_ipv6;
  29
  30 #ifdef CONFIG_XFRM
  31 static void nf_nat_ipv6_decode_session(struct sk_buff *skb,
  32                                        const struct nf_conn *ct,
  33                                        enum ip_conntrack_dir dir,
  34                                        unsigned long statusbit,
  35                                        struct flowi *fl)
  36 {
  37         const struct nf_conntrack_tuple *t = &ct->tuplehash[dir].tuple;
  38         struct flowi6 *fl6 = &fl->u.ip6;
  39
  40         if (ct->status & statusbit) {
  41                 fl6->daddr = t->dst.u3.in6;
  42                 if (t->dst.protonum == IPPROTO_TCP ||
  43                     t->dst.protonum == IPPROTO_UDP ||
  44                     t->dst.protonum == IPPROTO_UDPLITE ||
  45                     t->dst.protonum == IPPROTO_DCCP ||
  46                     t->dst.protonum == IPPROTO_SCTP)
  47                         fl6->fl6_dport = t->dst.u.all;
  48         }
  49
  50         statusbit ^= IPS_NAT_MASK;
  51
  52         if (ct->status & statusbit) {
  53                 fl6->saddr = t->src.u3.in6;
  54                 if (t->dst.protonum == IPPROTO_TCP ||
  55                     t->dst.protonum == IPPROTO_UDP ||
  56                     t->dst.protonum == IPPROTO_UDPLITE ||
  57                     t->dst.protonum == IPPROTO_DCCP ||
  58                     t->dst.protonum == IPPROTO_SCTP)
  59                         fl6->fl6_sport = t->src.u.all;
  60         }
  61 }
  62 #endif
  63
  64 static bool nf_nat_ipv6_in_range(const struct nf_conntrack_tuple *t,
  65                                  const struct nf_nat_range *range)
  66 {
  67         return ipv6_addr_cmp(&t->src.u3.in6, &range->min_addr.in6) >= 0 &&
  68                ipv6_addr_cmp(&t->src.u3.in6, &range->max_addr.in6) <= 0;
  69 }
  70
  71 static u32 nf_nat_ipv6_secure_port(const struct nf_conntrack_tuple *t,
  72                                    __be16 dport)
  73 {
  74         return secure_ipv6_port_ephemeral(t->src.u3.ip6, t->dst.u3.ip6, dport);
  75 }
  76
  77 static bool nf_nat_ipv6_manip_pkt(struct sk_buff *skb,
  78                                   unsigned int iphdroff,
  79                                   const struct nf_nat_l4proto *l4proto,
  80                                   const struct nf_conntrack_tuple *target,
  81                                   enum nf_nat_manip_type maniptype)
  82 {
  83         struct ipv6hdr *ipv6h;
  84         __be16 frag_off;
  85         int hdroff;
  86         u8 nexthdr;
  87
  88         if (!skb_make_writable(skb, iphdroff + sizeof(*ipv6h)))
  89                 return false;
  90
  91         ipv6h = (void *)skb->data + iphdroff;
  92         nexthdr = ipv6h->nexthdr;
  93         hdroff = ipv6_skip_exthdr(skb, iphdroff + sizeof(*ipv6h),
  94                                   &nexthdr, &frag_off);
  95         if (hdroff < 0)
  96                 goto manip_addr;
  97
  98         if ((frag_off & htons(~0x7)) == 0 &&
  99             !l4proto->manip_pkt(skb, &nf_nat_l3proto_ipv6, iphdroff, hdroff,
 100                                 target, maniptype))
 101                 return false;
 102
 103         /* must reload, offset might have changed */
 104         ipv6h = (void *)skb->data + iphdroff;
 105
 106 manip_addr:
 107         if (maniptype == NF_NAT_MANIP_SRC)
 108                 ipv6h->saddr = target->src.u3.in6;
 109         else
 110                 ipv6h->daddr = target->dst.u3.in6;
 111
 112         return true;
 113 }
 114
 115 static void nf_nat_ipv6_csum_update(struct sk_buff *skb,
 116                                     unsigned int iphdroff, __sum16 *check,
 117                                     const struct nf_conntrack_tuple *t,
 118                                     enum nf_nat_manip_type maniptype)
 119 {
 120         const struct ipv6hdr *ipv6h = (struct ipv6hdr *)(skb->data + iphdroff);
 121         const struct in6_addr *oldip, *newip;
 122
 123         if (maniptype == NF_NAT_MANIP_SRC) {
 124                 oldip = &ipv6h->saddr;
 125                 newip = &t->src.u3.in6;
 126         } else {
 127                 oldip = &ipv6h->daddr;
 128                 newip = &t->dst.u3.in6;
 129         }
 130         inet_proto_csum_replace16(check, skb, oldip->s6_addr32,
 131                                   newip->s6_addr32, true);
 132 }
 133
 134 static void nf_nat_ipv6_csum_recalc(struct sk_buff *skb,
 135                                     u8 proto, void *data, __sum16 *check,
 136                                     int datalen, int oldlen)
 137 {
 138         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 139                 const struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 140
 141                 skb->ip_summed = CHECKSUM_PARTIAL;
 142                 skb->csum_start = skb_headroom(skb) + skb_network_offset(skb) +
 143                         (data - (void *)skb->data);
 144                 skb->csum_offset = (void *)check - data;
 145                 *check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 146                                           datalen, proto, 0);
 147         } else
 148                 inet_proto_csum_replace2(check, skb,
 149                                          htons(oldlen), htons(datalen), true);
 150 }
 151
 152 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 153 static int nf_nat_ipv6_nlattr_to_range(struct nlattr *tb[],
 154                                        struct nf_nat_range *range)
 155 {
 156         if (tb[CTA_NAT_V6_MINIP]) {
 157                 nla_memcpy(&range->min_addr.ip6, tb[CTA_NAT_V6_MINIP],
 158                            sizeof(struct in6_addr));
 159                 range->flags |= NF_NAT_RANGE_MAP_IPS;
 160         }
 161
 162         if (tb[CTA_NAT_V6_MAXIP])
 163                 nla_memcpy(&range->max_addr.ip6, tb[CTA_NAT_V6_MAXIP],
 164                            sizeof(struct in6_addr));
 165         else
 166                 range->max_addr = range->min_addr;
 167
 168         return 0;
 169 }
 170 #endif
 171
 172 static const struct nf_nat_l3proto nf_nat_l3proto_ipv6 = {
 173         .l3proto                = NFPROTO_IPV6,
 174         .secure_port            = nf_nat_ipv6_secure_port,
 175         .in_range               = nf_nat_ipv6_in_range,
 176         .manip_pkt              = nf_nat_ipv6_manip_pkt,
 177         .csum_update            = nf_nat_ipv6_csum_update,
 178         .csum_recalc            = nf_nat_ipv6_csum_recalc,
 179 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 180         .nlattr_to_range        = nf_nat_ipv6_nlattr_to_range,
 181 #endif
 182 #ifdef CONFIG_XFRM
 183         .decode_session = nf_nat_ipv6_decode_session,
 184 #endif
 185 };
 186
 187 int nf_nat_icmpv6_reply_translation(struct sk_buff *skb,
 188                                     struct nf_conn *ct,
 189                                     enum ip_conntrack_info ctinfo,
 190                                     unsigned int hooknum,
 191                                     unsigned int hdrlen)
 192 {
 193         struct {
 194                 struct icmp6hdr icmp6;
 195                 struct ipv6hdr  ip6;
 196         } *inside;
 197         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 198         enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
 199         const struct nf_nat_l4proto *l4proto;
 200         struct nf_conntrack_tuple target;
 201         unsigned long statusbit;
 202
 203         WARN_ON(ctinfo != IP_CT_RELATED && ctinfo != IP_CT_RELATED_REPLY);
 204
 205         if (!skb_make_writable(skb, hdrlen + sizeof(*inside)))
 206                 return 0;
 207         if (nf_ip6_checksum(skb, hooknum, hdrlen, IPPROTO_ICMPV6))
 208                 return 0;
 209
 210         inside = (void *)skb->data + hdrlen;
 211         if (inside->icmp6.icmp6_type == NDISC_REDIRECT) {
 212                 if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
 213                         return 0;
 214                 if (ct->status & IPS_NAT_MASK)
 215                         return 0;
 216         }
 217
 218         if (manip == NF_NAT_MANIP_SRC)
 219                 statusbit = IPS_SRC_NAT;
 220         else
 221                 statusbit = IPS_DST_NAT;
 222
 223         /* Invert if this is reply direction */
 224         if (dir == IP_CT_DIR_REPLY)
 225                 statusbit ^= IPS_NAT_MASK;
 226
 227         if (!(ct->status & statusbit))
 228                 return 1;
 229
 230         l4proto = __nf_nat_l4proto_find(NFPROTO_IPV6, inside->ip6.nexthdr);
 231         if (!nf_nat_ipv6_manip_pkt(skb, hdrlen + sizeof(inside->icmp6),
 232                                    l4proto, &ct->tuplehash[!dir].tuple, !manip))
 233                 return 0;
 234
 235         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 236                 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 237                 inside = (void *)skb->data + hdrlen;
 238                 inside->icmp6.icmp6_cksum = 0;
 239                 inside->icmp6.icmp6_cksum =
 240                         csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
 241                                         skb->len - hdrlen, IPPROTO_ICMPV6,
 242                                         skb_checksum(skb, hdrlen,
 243                                                      skb->len - hdrlen, 0));
 244         }
 245
 246         nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
 247         l4proto = __nf_nat_l4proto_find(NFPROTO_IPV6, IPPROTO_ICMPV6);
 248         if (!nf_nat_ipv6_manip_pkt(skb, 0, l4proto, &target, manip))
 249                 return 0;
 250
 251         return 1;
 252 }
 253 EXPORT_SYMBOL_GPL(nf_nat_icmpv6_reply_translation);
 254
 255 unsigned int
 256 nf_nat_ipv6_fn(void *priv, struct sk_buff *skb,
 257                const struct nf_hook_state *state,
 258                unsigned int (*do_chain)(void *priv,
 259                                         struct sk_buff *skb,
 260                                         const struct nf_hook_state *state,
 261                                         struct nf_conn *ct))
 262 {
 263         struct nf_conn *ct;
 264         enum ip_conntrack_info ctinfo;
 265         struct nf_conn_nat *nat;
 266         enum nf_nat_manip_type maniptype = HOOK2MANIP(state->hook);
 267         __be16 frag_off;
 268         int hdrlen;
 269         u8 nexthdr;
 270
 271         ct = nf_ct_get(skb, &ctinfo);
 272         /* Can't track?  It's not due to stress, or conntrack would
 273          * have dropped it.  Hence it's the user's responsibilty to
 274          * packet filter it out, or implement conntrack/NAT for that
 275          * protocol. 8) --RR
 276          */
 277         if (!ct)
 278                 return NF_ACCEPT;
 279
 280         nat = nfct_nat(ct);
 281
 282         switch (ctinfo) {
 283         case IP_CT_RELATED:
 284         case IP_CT_RELATED_REPLY:
 285                 nexthdr = ipv6_hdr(skb)->nexthdr;
 286                 hdrlen = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr),
 287                                           &nexthdr, &frag_off);
 288
 289                 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
 290                         if (!nf_nat_icmpv6_reply_translation(skb, ct, ctinfo,
 291                                                              state->hook,
 292                                                              hdrlen))
 293                                 return NF_DROP;
 294                         else
 295                                 return NF_ACCEPT;
 296                 }
 297                 /* Fall thru... (Only ICMPs can be IP_CT_IS_REPLY) */
 298         case IP_CT_NEW:
 299                 /* Seen it before?  This can happen for loopback, retrans,
 300                  * or local packets.
 301                  */
 302                 if (!nf_nat_initialized(ct, maniptype)) {
 303                         unsigned int ret;
 304
 305                         ret = do_chain(priv, skb, state, ct);
 306                         if (ret != NF_ACCEPT)
 307                                 return ret;
 308
 309                         if (nf_nat_initialized(ct, HOOK2MANIP(state->hook)))
 310                                 break;
 311
 312                         ret = nf_nat_alloc_null_binding(ct, state->hook);
 313                         if (ret != NF_ACCEPT)
 314                                 return ret;
 315                 } else {
 316                         pr_debug("Already setup manip %s for ct %p\n",
 317                                  maniptype == NF_NAT_MANIP_SRC ? "SRC" : "DST",
 318                                  ct);
 319                         if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 320                                 goto oif_changed;
 321                 }
 322                 break;
 323
 324         default:
 325                 /* ESTABLISHED */
 326                 WARN_ON(ctinfo != IP_CT_ESTABLISHED &&
 327                         ctinfo != IP_CT_ESTABLISHED_REPLY);
 328                 if (nf_nat_oif_changed(state->hook, ctinfo, nat, state->out))
 329                         goto oif_changed;
 330         }
 331
 332         return nf_nat_packet(ct, ctinfo, state->hook, skb);
 333
 334 oif_changed:
 335         nf_ct_kill_acct(ct, ctinfo, skb);
 336         return NF_DROP;
 337 }
 338 EXPORT_SYMBOL_GPL(nf_nat_ipv6_fn);
 339
 340 unsigned int
 341 nf_nat_ipv6_in(void *priv, struct sk_buff *skb,
 342                const struct nf_hook_state *state,
 343                unsigned int (*do_chain)(void *priv,
 344                                         struct sk_buff *skb,
 345                                         const struct nf_hook_state *state,
 346                                         struct nf_conn *ct))
 347 {
 348         unsigned int ret;
 349         struct in6_addr daddr = ipv6_hdr(skb)->daddr;
 350
 351         ret = nf_nat_ipv6_fn(priv, skb, state, do_chain);
 352         if (ret != NF_DROP && ret != NF_STOLEN &&
 353             ipv6_addr_cmp(&daddr, &ipv6_hdr(skb)->daddr))
 354                 skb_dst_drop(skb);
 355
 356         return ret;
 357 }
 358 EXPORT_SYMBOL_GPL(nf_nat_ipv6_in);
 359
 360 unsigned int
 361 nf_nat_ipv6_out(void *priv, struct sk_buff *skb,
 362                 const struct nf_hook_state *state,
 363                 unsigned int (*do_chain)(void *priv,
 364                                          struct sk_buff *skb,
 365                                          const struct nf_hook_state *state,
 366                                          struct nf_conn *ct))
 367 {
 368 #ifdef CONFIG_XFRM
 369         const struct nf_conn *ct;
 370         enum ip_conntrack_info ctinfo;
 371         int err;
 372 #endif
 373         unsigned int ret;
 374
 375         /* root is playing with raw sockets. */
 376         if (skb->len < sizeof(struct ipv6hdr))
 377                 return NF_ACCEPT;
 378
 379         ret = nf_nat_ipv6_fn(priv, skb, state, do_chain);
 380 #ifdef CONFIG_XFRM
 381         if (ret != NF_DROP && ret != NF_STOLEN &&
 382             !(IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) &&
 383             (ct = nf_ct_get(skb, &ctinfo)) != NULL) {
 384                 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 385
 386                 if (!nf_inet_addr_cmp(&ct->tuplehash[dir].tuple.src.u3,
 387                                       &ct->tuplehash[!dir].tuple.dst.u3) ||
 388                     (ct->tuplehash[dir].tuple.dst.protonum != IPPROTO_ICMPV6 &&
 389                      ct->tuplehash[dir].tuple.src.u.all !=
 390                      ct->tuplehash[!dir].tuple.dst.u.all)) {
 391                         err = nf_xfrm_me_harder(state->net, skb, AF_INET6);
 392                         if (err < 0)
 393                                 ret = NF_DROP_ERR(err);
 394                 }
 395         }
 396 #endif
 397         return ret;
 398 }
 399 EXPORT_SYMBOL_GPL(nf_nat_ipv6_out);
 400
 401 unsigned int
 402 nf_nat_ipv6_local_fn(void *priv, struct sk_buff *skb,
 403                      const struct nf_hook_state *state,
 404                      unsigned int (*do_chain)(void *priv,
 405                                               struct sk_buff *skb,
 406                                               const struct nf_hook_state *state,
 407                                               struct nf_conn *ct))
 408 {
 409         const struct nf_conn *ct;
 410         enum ip_conntrack_info ctinfo;
 411         unsigned int ret;
 412         int err;
 413
 414         /* root is playing with raw sockets. */
 415         if (skb->len < sizeof(struct ipv6hdr))
 416                 return NF_ACCEPT;
 417
 418         ret = nf_nat_ipv6_fn(priv, skb, state, do_chain);
 419         if (ret != NF_DROP && ret != NF_STOLEN &&
 420             (ct = nf_ct_get(skb, &ctinfo)) != NULL) {
 421                 enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 422
 423                 if (!nf_inet_addr_cmp(&ct->tuplehash[dir].tuple.dst.u3,
 424                                       &ct->tuplehash[!dir].tuple.src.u3)) {
 425                         err = ip6_route_me_harder(state->net, skb);
 426                         if (err < 0)
 427                                 ret = NF_DROP_ERR(err);
 428                 }
 429 #ifdef CONFIG_XFRM
 430                 else if (!(IP6CB(skb)->flags & IP6SKB_XFRM_TRANSFORMED) &&
 431                          ct->tuplehash[dir].tuple.dst.protonum != IPPROTO_ICMPV6 &&
 432                          ct->tuplehash[dir].tuple.dst.u.all !=
 433                          ct->tuplehash[!dir].tuple.src.u.all) {
 434                         err = nf_xfrm_me_harder(state->net, skb, AF_INET6);
 435                         if (err < 0)
 436                                 ret = NF_DROP_ERR(err);
 437                 }
 438 #endif
 439         }
 440         return ret;
 441 }
 442 EXPORT_SYMBOL_GPL(nf_nat_ipv6_local_fn);
 443
 444 static int __init nf_nat_l3proto_ipv6_init(void)
 445 {
 446         int err;
 447
 448         err = nf_nat_l4proto_register(NFPROTO_IPV6, &nf_nat_l4proto_icmpv6);
 449         if (err < 0)
 450                 goto err1;
 451         err = nf_nat_l3proto_register(&nf_nat_l3proto_ipv6);
 452         if (err < 0)
 453                 goto err2;
 454         return err;
 455
 456 err2:
 457         nf_nat_l4proto_unregister(NFPROTO_IPV6, &nf_nat_l4proto_icmpv6);
 458 err1:
 459         return err;
 460 }
 461
 462 static void __exit nf_nat_l3proto_ipv6_exit(void)
 463 {
 464         nf_nat_l3proto_unregister(&nf_nat_l3proto_ipv6);
 465         nf_nat_l4proto_unregister(NFPROTO_IPV6, &nf_nat_l4proto_icmpv6);
 466 }
 467
 468 MODULE_LICENSE("GPL");
 469 MODULE_ALIAS("nf-nat-" __stringify(AF_INET6));
 470
 471 module_init(nf_nat_l3proto_ipv6_init);
 472 module_exit(nf_nat_l3proto_ipv6_exit);