net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c

   1
   2 /* (C) 1999-2001 Paul `Rusty' Russell
   3  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   4  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/types.h>
  12 #include <linux/ip.h>
  13 #include <linux/netfilter.h>
  14 #include <linux/module.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/icmp.h>
  17 #include <linux/sysctl.h>
  18 #include <net/route.h>
  19 #include <net/ip.h>
  20
  21 #include <linux/netfilter_ipv4.h>
  22 #include <net/netfilter/nf_conntrack.h>
  23 #include <net/netfilter/nf_conntrack_helper.h>
  24 #include <net/netfilter/nf_conntrack_l4proto.h>
  25 #include <net/netfilter/nf_conntrack_l3proto.h>
  26 #include <net/netfilter/nf_conntrack_zones.h>
  27 #include <net/netfilter/nf_conntrack_core.h>
  28 #include <net/netfilter/nf_conntrack_seqadj.h>
  29 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  30 #include <net/netfilter/nf_nat_helper.h>
  31 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
  32 #include <net/netfilter/nf_log.h>
  33
  34 static int conntrack4_net_id __read_mostly;
  35 static DEFINE_MUTEX(register_ipv4_hooks);
  36
  37 struct conntrack4_net {
  38         unsigned int users;
  39 };
  40
  41 static bool ipv4_pkt_to_tuple(const struct sk_buff *skb, unsigned int nhoff,
  42                               struct nf_conntrack_tuple *tuple)
  43 {
  44         const __be32 *ap;
  45         __be32 _addrs[2];
  46         ap = skb_header_pointer(skb, nhoff + offsetof(struct iphdr, saddr),
  47                                 sizeof(u_int32_t) * 2, _addrs);
  48         if (ap == NULL)
  49                 return false;
  50
  51         tuple->src.u3.ip = ap[0];
  52         tuple->dst.u3.ip = ap[1];
  53
  54         return true;
  55 }
  56
  57 static bool ipv4_invert_tuple(struct nf_conntrack_tuple *tuple,
  58                               const struct nf_conntrack_tuple *orig)
  59 {
  60         tuple->src.u3.ip = orig->dst.u3.ip;
  61         tuple->dst.u3.ip = orig->src.u3.ip;
  62
  63         return true;
  64 }
  65
  66 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
  67                             unsigned int *dataoff, u_int8_t *protonum)
  68 {
  69         const struct iphdr *iph;
  70         struct iphdr _iph;
  71
  72         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  73         if (iph == NULL)
  74                 return -NF_ACCEPT;
  75
  76         /* Conntrack defragments packets, we might still see fragments
  77          * inside ICMP packets though. */
  78         if (iph->frag_off & htons(IP_OFFSET))
  79                 return -NF_ACCEPT;
  80
  81         *dataoff = nhoff + (iph->ihl << 2);
  82         *protonum = iph->protocol;
  83
  84         /* Check bogus IP headers */
  85         if (*dataoff > skb->len) {
  86                 pr_debug("nf_conntrack_ipv4: bogus IPv4 packet: "
  87                          "nhoff %u, ihl %u, skblen %u\n",
  88                          nhoff, iph->ihl << 2, skb->len);
  89                 return -NF_ACCEPT;
  90         }
  91
  92         return NF_ACCEPT;
  93 }
  94
  95 static unsigned int ipv4_helper(void *priv,
  96                                 struct sk_buff *skb,
  97                                 const struct nf_hook_state *state)
  98 {
  99         struct nf_conn *ct;
 100         enum ip_conntrack_info ctinfo;
 101         const struct nf_conn_help *help;
 102         const struct nf_conntrack_helper *helper;
 103
 104         /* This is where we call the helper: as the packet goes out. */
 105         ct = nf_ct_get(skb, &ctinfo);
 106         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 107                 return NF_ACCEPT;
 108
 109         help = nfct_help(ct);
 110         if (!help)
 111                 return NF_ACCEPT;
 112
 113         /* rcu_read_lock()ed by nf_hook_thresh */
 114         helper = rcu_dereference(help->helper);
 115         if (!helper)
 116                 return NF_ACCEPT;
 117
 118         return helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
 119                             ct, ctinfo);
 120 }
 121
 122 static unsigned int ipv4_confirm(void *priv,
 123                                  struct sk_buff *skb,
 124                                  const struct nf_hook_state *state)
 125 {
 126         struct nf_conn *ct;
 127         enum ip_conntrack_info ctinfo;
 128
 129         ct = nf_ct_get(skb, &ctinfo);
 130         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 131                 goto out;
 132
 133         /* adjust seqs for loopback traffic only in outgoing direction */
 134         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 135             !nf_is_loopback_packet(skb)) {
 136                 if (!nf_ct_seq_adjust(skb, ct, ctinfo, ip_hdrlen(skb))) {
 137                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
 138                         return NF_DROP;
 139                 }
 140         }
 141 out:
 142         /* We've seen it coming out the other side: confirm it */
 143         return nf_conntrack_confirm(skb);
 144 }
 145
 146 static unsigned int ipv4_conntrack_in(void *priv,
 147                                       struct sk_buff *skb,
 148                                       const struct nf_hook_state *state)
 149 {
 150         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 151 }
 152
 153 static unsigned int ipv4_conntrack_local(void *priv,
 154                                          struct sk_buff *skb,
 155                                          const struct nf_hook_state *state)
 156 {
 157         if (ip_is_fragment(ip_hdr(skb))) /* IP_NODEFRAG setsockopt set */
 158                 return NF_ACCEPT;
 159
 160         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 161 }
 162
 163 /* Connection tracking may drop packets, but never alters them, so
 164    make it the first hook. */
 165 static const struct nf_hook_ops ipv4_conntrack_ops[] = {
 166         {
 167                 .hook           = ipv4_conntrack_in,
 168                 .pf             = NFPROTO_IPV4,
 169                 .hooknum        = NF_INET_PRE_ROUTING,
 170                 .priority       = NF_IP_PRI_CONNTRACK,
 171         },
 172         {
 173                 .hook           = ipv4_conntrack_local,
 174                 .pf             = NFPROTO_IPV4,
 175                 .hooknum        = NF_INET_LOCAL_OUT,
 176                 .priority       = NF_IP_PRI_CONNTRACK,
 177         },
 178         {
 179                 .hook           = ipv4_helper,
 180                 .pf             = NFPROTO_IPV4,
 181                 .hooknum        = NF_INET_POST_ROUTING,
 182                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 183         },
 184         {
 185                 .hook           = ipv4_confirm,
 186                 .pf             = NFPROTO_IPV4,
 187                 .hooknum        = NF_INET_POST_ROUTING,
 188                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 189         },
 190         {
 191                 .hook           = ipv4_helper,
 192                 .pf             = NFPROTO_IPV4,
 193                 .hooknum        = NF_INET_LOCAL_IN,
 194                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 195         },
 196         {
 197                 .hook           = ipv4_confirm,
 198                 .pf             = NFPROTO_IPV4,
 199                 .hooknum        = NF_INET_LOCAL_IN,
 200                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 201         },
 202 };
 203
 204 /* Fast function for those who don't want to parse /proc (and I don't
 205    blame them). */
 206 /* Reversing the socket's dst/src point of view gives us the reply
 207    mapping. */
 208 static int
 209 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 210 {
 211         const struct inet_sock *inet = inet_sk(sk);
 212         const struct nf_conntrack_tuple_hash *h;
 213         struct nf_conntrack_tuple tuple;
 214
 215         memset(&tuple, 0, sizeof(tuple));
 216
 217         lock_sock(sk);
 218         tuple.src.u3.ip = inet->inet_rcv_saddr;
 219         tuple.src.u.tcp.port = inet->inet_sport;
 220         tuple.dst.u3.ip = inet->inet_daddr;
 221         tuple.dst.u.tcp.port = inet->inet_dport;
 222         tuple.src.l3num = PF_INET;
 223         tuple.dst.protonum = sk->sk_protocol;
 224         release_sock(sk);
 225
 226         /* We only do TCP and SCTP at the moment: is there a better way? */
 227         if (tuple.dst.protonum != IPPROTO_TCP &&
 228             tuple.dst.protonum != IPPROTO_SCTP) {
 229                 pr_debug("SO_ORIGINAL_DST: Not a TCP/SCTP socket\n");
 230                 return -ENOPROTOOPT;
 231         }
 232
 233         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 234                 pr_debug("SO_ORIGINAL_DST: len %d not %zu\n",
 235                          *len, sizeof(struct sockaddr_in));
 236                 return -EINVAL;
 237         }
 238
 239         h = nf_conntrack_find_get(sock_net(sk), &nf_ct_zone_dflt, &tuple);
 240         if (h) {
 241                 struct sockaddr_in sin;
 242                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 243
 244                 sin.sin_family = AF_INET;
 245                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 246                         .tuple.dst.u.tcp.port;
 247                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 248                         .tuple.dst.u3.ip;
 249                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 250
 251                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 252                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 253                 nf_ct_put(ct);
 254                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 255                         return -EFAULT;
 256                 else
 257                         return 0;
 258         }
 259         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 260                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 261                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 262         return -ENOENT;
 263 }
 264
 265 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 266
 267 #include <linux/netfilter/nfnetlink.h>
 268 #include <linux/netfilter/nfnetlink_conntrack.h>
 269
 270 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 271                                 const struct nf_conntrack_tuple *tuple)
 272 {
 273         if (nla_put_in_addr(skb, CTA_IP_V4_SRC, tuple->src.u3.ip) ||
 274             nla_put_in_addr(skb, CTA_IP_V4_DST, tuple->dst.u3.ip))
 275                 goto nla_put_failure;
 276         return 0;
 277
 278 nla_put_failure:
 279         return -1;
 280 }
 281
 282 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 283         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 284         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 285 };
 286
 287 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 288                                 struct nf_conntrack_tuple *t)
 289 {
 290         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 291                 return -EINVAL;
 292
 293         t->src.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_SRC]);
 294         t->dst.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_DST]);
 295
 296         return 0;
 297 }
 298 #endif
 299
 300 static struct nf_sockopt_ops so_getorigdst = {
 301         .pf             = PF_INET,
 302         .get_optmin     = SO_ORIGINAL_DST,
 303         .get_optmax     = SO_ORIGINAL_DST+1,
 304         .get            = getorigdst,
 305         .owner          = THIS_MODULE,
 306 };
 307
 308 static int ipv4_hooks_register(struct net *net)
 309 {
 310         struct conntrack4_net *cnet = net_generic(net, conntrack4_net_id);
 311         int err = 0;
 312
 313         mutex_lock(&register_ipv4_hooks);
 314
 315         cnet->users++;
 316         if (cnet->users > 1)
 317                 goto out_unlock;
 318
 319         err = nf_defrag_ipv4_enable(net);
 320         if (err) {
 321                 cnet->users = 0;
 322                 goto out_unlock;
 323         }
 324
 325         err = nf_register_net_hooks(net, ipv4_conntrack_ops,
 326                                     ARRAY_SIZE(ipv4_conntrack_ops));
 327
 328         if (err)
 329                 cnet->users = 0;
 330  out_unlock:
 331         mutex_unlock(&register_ipv4_hooks);
 332         return err;
 333 }
 334
 335 static void ipv4_hooks_unregister(struct net *net)
 336 {
 337         struct conntrack4_net *cnet = net_generic(net, conntrack4_net_id);
 338
 339         mutex_lock(&register_ipv4_hooks);
 340         if (cnet->users && (--cnet->users == 0))
 341                 nf_unregister_net_hooks(net, ipv4_conntrack_ops,
 342                                         ARRAY_SIZE(ipv4_conntrack_ops));
 343         mutex_unlock(&register_ipv4_hooks);
 344 }
 345
 346 const struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 = {
 347         .l3proto         = PF_INET,
 348         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 349         .invert_tuple    = ipv4_invert_tuple,
 350         .get_l4proto     = ipv4_get_l4proto,
 351 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 352         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 353         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 354         .nla_policy      = ipv4_nla_policy,
 355         .nla_size        = NLA_ALIGN(NLA_HDRLEN + sizeof(u32)) + /* CTA_IP_V4_SRC */
 356                            NLA_ALIGN(NLA_HDRLEN + sizeof(u32)),  /* CTA_IP_V4_DST */
 357 #endif
 358         .net_ns_get      = ipv4_hooks_register,
 359         .net_ns_put      = ipv4_hooks_unregister,
 360         .me              = THIS_MODULE,
 361 };
 362
 363 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 364                   &nf_conntrack_htable_size, 0600);
 365
 366 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 367 MODULE_ALIAS("ip_conntrack");
 368 MODULE_LICENSE("GPL");
 369
 370 static const struct nf_conntrack_l4proto * const builtin_l4proto4[] = {
 371         &nf_conntrack_l4proto_tcp4,
 372         &nf_conntrack_l4proto_udp4,
 373         &nf_conntrack_l4proto_icmp,
 374 #ifdef CONFIG_NF_CT_PROTO_DCCP
 375         &nf_conntrack_l4proto_dccp4,
 376 #endif
 377 #ifdef CONFIG_NF_CT_PROTO_SCTP
 378         &nf_conntrack_l4proto_sctp4,
 379 #endif
 380 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
 381         &nf_conntrack_l4proto_udplite4,
 382 #endif
 383 };
 384
 385 static int ipv4_net_init(struct net *net)
 386 {
 387         return nf_ct_l4proto_pernet_register(net, builtin_l4proto4,
 388                                              ARRAY_SIZE(builtin_l4proto4));
 389 }
 390
 391 static void ipv4_net_exit(struct net *net)
 392 {
 393         nf_ct_l4proto_pernet_unregister(net, builtin_l4proto4,
 394                                         ARRAY_SIZE(builtin_l4proto4));
 395 }
 396
 397 static struct pernet_operations ipv4_net_ops = {
 398         .init = ipv4_net_init,
 399         .exit = ipv4_net_exit,
 400         .id = &conntrack4_net_id,
 401         .size = sizeof(struct conntrack4_net),
 402 };
 403
 404 static int __init nf_conntrack_l3proto_ipv4_init(void)
 405 {
 406         int ret = 0;
 407
 408         need_conntrack();
 409
 410 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 411         if (WARN_ON(nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1) !=
 412             nf_conntrack_l3proto_ipv4.nla_size))
 413                 return -EINVAL;
 414 #endif
 415         ret = nf_register_sockopt(&so_getorigdst);
 416         if (ret < 0) {
 417                 pr_err("Unable to register netfilter socket option\n");
 418                 return ret;
 419         }
 420
 421         ret = register_pernet_subsys(&ipv4_net_ops);
 422         if (ret < 0) {
 423                 pr_err("nf_conntrack_ipv4: can't register pernet ops\n");
 424                 goto cleanup_sockopt;
 425         }
 426
 427         ret = nf_ct_l4proto_register(builtin_l4proto4,
 428                                      ARRAY_SIZE(builtin_l4proto4));
 429         if (ret < 0)
 430                 goto cleanup_pernet;
 431
 432         ret = nf_ct_l3proto_register(&nf_conntrack_l3proto_ipv4);
 433         if (ret < 0) {
 434                 pr_err("nf_conntrack_ipv4: can't register ipv4 proto.\n");
 435                 goto cleanup_l4proto;
 436         }
 437
 438         return ret;
 439 cleanup_l4proto:
 440         nf_ct_l4proto_unregister(builtin_l4proto4,
 441                                  ARRAY_SIZE(builtin_l4proto4));
 442  cleanup_pernet:
 443         unregister_pernet_subsys(&ipv4_net_ops);
 444  cleanup_sockopt:
 445         nf_unregister_sockopt(&so_getorigdst);
 446         return ret;
 447 }
 448
 449 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 450 {
 451         synchronize_net();
 452         nf_ct_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 453         nf_ct_l4proto_unregister(builtin_l4proto4,
 454                                  ARRAY_SIZE(builtin_l4proto4));
 455         unregister_pernet_subsys(&ipv4_net_ops);
 456         nf_unregister_sockopt(&so_getorigdst);
 457 }
 458
 459 module_init(nf_conntrack_l3proto_ipv4_init);
 460 module_exit(nf_conntrack_l3proto_ipv4_fini);