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 void ipv4_print_tuple(struct seq_file *s,
  67                             const struct nf_conntrack_tuple *tuple)
  68 {
  69         seq_printf(s, "src=%pI4 dst=%pI4 ",
  70                    &tuple->src.u3.ip, &tuple->dst.u3.ip);
  71 }
  72
  73 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
  74                             unsigned int *dataoff, u_int8_t *protonum)
  75 {
  76         const struct iphdr *iph;
  77         struct iphdr _iph;
  78
  79         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  80         if (iph == NULL)
  81                 return -NF_ACCEPT;
  82
  83         /* Conntrack defragments packets, we might still see fragments
  84          * inside ICMP packets though. */
  85         if (iph->frag_off & htons(IP_OFFSET))
  86                 return -NF_ACCEPT;
  87
  88         *dataoff = nhoff + (iph->ihl << 2);
  89         *protonum = iph->protocol;
  90
  91         /* Check bogus IP headers */
  92         if (*dataoff > skb->len) {
  93                 pr_debug("nf_conntrack_ipv4: bogus IPv4 packet: "
  94                          "nhoff %u, ihl %u, skblen %u\n",
  95                          nhoff, iph->ihl << 2, skb->len);
  96                 return -NF_ACCEPT;
  97         }
  98
  99         return NF_ACCEPT;
 100 }
 101
 102 static unsigned int ipv4_helper(void *priv,
 103                                 struct sk_buff *skb,
 104                                 const struct nf_hook_state *state)
 105 {
 106         struct nf_conn *ct;
 107         enum ip_conntrack_info ctinfo;
 108         const struct nf_conn_help *help;
 109         const struct nf_conntrack_helper *helper;
 110
 111         /* This is where we call the helper: as the packet goes out. */
 112         ct = nf_ct_get(skb, &ctinfo);
 113         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 114                 return NF_ACCEPT;
 115
 116         help = nfct_help(ct);
 117         if (!help)
 118                 return NF_ACCEPT;
 119
 120         /* rcu_read_lock()ed by nf_hook_thresh */
 121         helper = rcu_dereference(help->helper);
 122         if (!helper)
 123                 return NF_ACCEPT;
 124
 125         return helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
 126                             ct, ctinfo);
 127 }
 128
 129 static unsigned int ipv4_confirm(void *priv,
 130                                  struct sk_buff *skb,
 131                                  const struct nf_hook_state *state)
 132 {
 133         struct nf_conn *ct;
 134         enum ip_conntrack_info ctinfo;
 135
 136         ct = nf_ct_get(skb, &ctinfo);
 137         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 138                 goto out;
 139
 140         /* adjust seqs for loopback traffic only in outgoing direction */
 141         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 142             !nf_is_loopback_packet(skb)) {
 143                 if (!nf_ct_seq_adjust(skb, ct, ctinfo, ip_hdrlen(skb))) {
 144                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
 145                         return NF_DROP;
 146                 }
 147         }
 148 out:
 149         /* We've seen it coming out the other side: confirm it */
 150         return nf_conntrack_confirm(skb);
 151 }
 152
 153 static unsigned int ipv4_conntrack_in(void *priv,
 154                                       struct sk_buff *skb,
 155                                       const struct nf_hook_state *state)
 156 {
 157         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 158 }
 159
 160 static unsigned int ipv4_conntrack_local(void *priv,
 161                                          struct sk_buff *skb,
 162                                          const struct nf_hook_state *state)
 163 {
 164         /* root is playing with raw sockets. */
 165         if (skb->len < sizeof(struct iphdr) ||
 166             ip_hdrlen(skb) < sizeof(struct iphdr))
 167                 return NF_ACCEPT;
 168         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 169 }
 170
 171 /* Connection tracking may drop packets, but never alters them, so
 172    make it the first hook. */
 173 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 174         {
 175                 .hook           = ipv4_conntrack_in,
 176                 .pf             = NFPROTO_IPV4,
 177                 .hooknum        = NF_INET_PRE_ROUTING,
 178                 .priority       = NF_IP_PRI_CONNTRACK,
 179         },
 180         {
 181                 .hook           = ipv4_conntrack_local,
 182                 .pf             = NFPROTO_IPV4,
 183                 .hooknum        = NF_INET_LOCAL_OUT,
 184                 .priority       = NF_IP_PRI_CONNTRACK,
 185         },
 186         {
 187                 .hook           = ipv4_helper,
 188                 .pf             = NFPROTO_IPV4,
 189                 .hooknum        = NF_INET_POST_ROUTING,
 190                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 191         },
 192         {
 193                 .hook           = ipv4_confirm,
 194                 .pf             = NFPROTO_IPV4,
 195                 .hooknum        = NF_INET_POST_ROUTING,
 196                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 197         },
 198         {
 199                 .hook           = ipv4_helper,
 200                 .pf             = NFPROTO_IPV4,
 201                 .hooknum        = NF_INET_LOCAL_IN,
 202                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 203         },
 204         {
 205                 .hook           = ipv4_confirm,
 206                 .pf             = NFPROTO_IPV4,
 207                 .hooknum        = NF_INET_LOCAL_IN,
 208                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 209         },
 210 };
 211
 212 /* Fast function for those who don't want to parse /proc (and I don't
 213    blame them). */
 214 /* Reversing the socket's dst/src point of view gives us the reply
 215    mapping. */
 216 static int
 217 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 218 {
 219         const struct inet_sock *inet = inet_sk(sk);
 220         const struct nf_conntrack_tuple_hash *h;
 221         struct nf_conntrack_tuple tuple;
 222
 223         memset(&tuple, 0, sizeof(tuple));
 224         tuple.src.u3.ip = inet->inet_rcv_saddr;
 225         tuple.src.u.tcp.port = inet->inet_sport;
 226         tuple.dst.u3.ip = inet->inet_daddr;
 227         tuple.dst.u.tcp.port = inet->inet_dport;
 228         tuple.src.l3num = PF_INET;
 229         tuple.dst.protonum = sk->sk_protocol;
 230
 231         /* We only do TCP and SCTP at the moment: is there a better way? */
 232         if (sk->sk_protocol != IPPROTO_TCP && sk->sk_protocol != IPPROTO_SCTP) {
 233                 pr_debug("SO_ORIGINAL_DST: Not a TCP/SCTP socket\n");
 234                 return -ENOPROTOOPT;
 235         }
 236
 237         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 238                 pr_debug("SO_ORIGINAL_DST: len %d not %zu\n",
 239                          *len, sizeof(struct sockaddr_in));
 240                 return -EINVAL;
 241         }
 242
 243         h = nf_conntrack_find_get(sock_net(sk), &nf_ct_zone_dflt, &tuple);
 244         if (h) {
 245                 struct sockaddr_in sin;
 246                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 247
 248                 sin.sin_family = AF_INET;
 249                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 250                         .tuple.dst.u.tcp.port;
 251                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 252                         .tuple.dst.u3.ip;
 253                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 254
 255                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 256                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 257                 nf_ct_put(ct);
 258                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 259                         return -EFAULT;
 260                 else
 261                         return 0;
 262         }
 263         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 264                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 265                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 266         return -ENOENT;
 267 }
 268
 269 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 270
 271 #include <linux/netfilter/nfnetlink.h>
 272 #include <linux/netfilter/nfnetlink_conntrack.h>
 273
 274 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 275                                 const struct nf_conntrack_tuple *tuple)
 276 {
 277         if (nla_put_in_addr(skb, CTA_IP_V4_SRC, tuple->src.u3.ip) ||
 278             nla_put_in_addr(skb, CTA_IP_V4_DST, tuple->dst.u3.ip))
 279                 goto nla_put_failure;
 280         return 0;
 281
 282 nla_put_failure:
 283         return -1;
 284 }
 285
 286 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 287         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 288         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 289 };
 290
 291 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 292                                 struct nf_conntrack_tuple *t)
 293 {
 294         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 295                 return -EINVAL;
 296
 297         t->src.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_SRC]);
 298         t->dst.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_DST]);
 299
 300         return 0;
 301 }
 302
 303 static int ipv4_nlattr_tuple_size(void)
 304 {
 305         return nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1);
 306 }
 307 #endif
 308
 309 static struct nf_sockopt_ops so_getorigdst = {
 310         .pf             = PF_INET,
 311         .get_optmin     = SO_ORIGINAL_DST,
 312         .get_optmax     = SO_ORIGINAL_DST+1,
 313         .get            = getorigdst,
 314         .owner          = THIS_MODULE,
 315 };
 316
 317 static int ipv4_hooks_register(struct net *net)
 318 {
 319         struct conntrack4_net *cnet = net_generic(net, conntrack4_net_id);
 320         int err = 0;
 321
 322         mutex_lock(&register_ipv4_hooks);
 323
 324         cnet->users++;
 325         if (cnet->users > 1)
 326                 goto out_unlock;
 327
 328         err = nf_defrag_ipv4_enable(net);
 329         if (err) {
 330                 cnet->users = 0;
 331                 goto out_unlock;
 332         }
 333
 334         err = nf_register_net_hooks(net, ipv4_conntrack_ops,
 335                                     ARRAY_SIZE(ipv4_conntrack_ops));
 336
 337         if (err)
 338                 cnet->users = 0;
 339  out_unlock:
 340         mutex_unlock(&register_ipv4_hooks);
 341         return err;
 342 }
 343
 344 static void ipv4_hooks_unregister(struct net *net)
 345 {
 346         struct conntrack4_net *cnet = net_generic(net, conntrack4_net_id);
 347
 348         mutex_lock(&register_ipv4_hooks);
 349         if (cnet->users && (--cnet->users == 0))
 350                 nf_unregister_net_hooks(net, ipv4_conntrack_ops,
 351                                         ARRAY_SIZE(ipv4_conntrack_ops));
 352         mutex_unlock(&register_ipv4_hooks);
 353 }
 354
 355 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 356         .l3proto         = PF_INET,
 357         .name            = "ipv4",
 358         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 359         .invert_tuple    = ipv4_invert_tuple,
 360         .print_tuple     = ipv4_print_tuple,
 361         .get_l4proto     = ipv4_get_l4proto,
 362 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 363         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 364         .nlattr_tuple_size = ipv4_nlattr_tuple_size,
 365         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 366         .nla_policy      = ipv4_nla_policy,
 367 #endif
 368         .net_ns_get      = ipv4_hooks_register,
 369         .net_ns_put      = ipv4_hooks_unregister,
 370         .me              = THIS_MODULE,
 371 };
 372
 373 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 374                   &nf_conntrack_htable_size, 0600);
 375
 376 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 377 MODULE_ALIAS("ip_conntrack");
 378 MODULE_LICENSE("GPL");
 379
 380 static struct nf_conntrack_l4proto *builtin_l4proto4[] = {
 381         &nf_conntrack_l4proto_tcp4,
 382         &nf_conntrack_l4proto_udp4,
 383         &nf_conntrack_l4proto_icmp,
 384 #ifdef CONFIG_NF_CT_PROTO_DCCP
 385         &nf_conntrack_l4proto_dccp4,
 386 #endif
 387 #ifdef CONFIG_NF_CT_PROTO_SCTP
 388         &nf_conntrack_l4proto_sctp4,
 389 #endif
 390 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
 391         &nf_conntrack_l4proto_udplite4,
 392 #endif
 393 };
 394
 395 static int ipv4_net_init(struct net *net)
 396 {
 397         int ret = 0;
 398
 399         ret = nf_ct_l4proto_pernet_register(net, builtin_l4proto4,
 400                                             ARRAY_SIZE(builtin_l4proto4));
 401         if (ret < 0)
 402                 return ret;
 403         ret = nf_ct_l3proto_pernet_register(net, &nf_conntrack_l3proto_ipv4);
 404         if (ret < 0) {
 405                 pr_err("nf_conntrack_ipv4: pernet registration failed\n");
 406                 nf_ct_l4proto_pernet_unregister(net, builtin_l4proto4,
 407                                                 ARRAY_SIZE(builtin_l4proto4));
 408         }
 409         return ret;
 410 }
 411
 412 static void ipv4_net_exit(struct net *net)
 413 {
 414         nf_ct_l3proto_pernet_unregister(net, &nf_conntrack_l3proto_ipv4);
 415         nf_ct_l4proto_pernet_unregister(net, builtin_l4proto4,
 416                                         ARRAY_SIZE(builtin_l4proto4));
 417 }
 418
 419 static struct pernet_operations ipv4_net_ops = {
 420         .init = ipv4_net_init,
 421         .exit = ipv4_net_exit,
 422         .id = &conntrack4_net_id,
 423         .size = sizeof(struct conntrack4_net),
 424 };
 425
 426 static int __init nf_conntrack_l3proto_ipv4_init(void)
 427 {
 428         int ret = 0;
 429
 430         need_conntrack();
 431
 432         ret = nf_register_sockopt(&so_getorigdst);
 433         if (ret < 0) {
 434                 pr_err("Unable to register netfilter socket option\n");
 435                 return ret;
 436         }
 437
 438         ret = register_pernet_subsys(&ipv4_net_ops);
 439         if (ret < 0) {
 440                 pr_err("nf_conntrack_ipv4: can't register pernet ops\n");
 441                 goto cleanup_sockopt;
 442         }
 443
 444         ret = nf_ct_l4proto_register(builtin_l4proto4,
 445                                      ARRAY_SIZE(builtin_l4proto4));
 446         if (ret < 0)
 447                 goto cleanup_pernet;
 448
 449         ret = nf_ct_l3proto_register(&nf_conntrack_l3proto_ipv4);
 450         if (ret < 0) {
 451                 pr_err("nf_conntrack_ipv4: can't register ipv4 proto.\n");
 452                 goto cleanup_l4proto;
 453         }
 454
 455         return ret;
 456 cleanup_l4proto:
 457         nf_ct_l4proto_unregister(builtin_l4proto4,
 458                                  ARRAY_SIZE(builtin_l4proto4));
 459  cleanup_pernet:
 460         unregister_pernet_subsys(&ipv4_net_ops);
 461  cleanup_sockopt:
 462         nf_unregister_sockopt(&so_getorigdst);
 463         return ret;
 464 }
 465
 466 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 467 {
 468         synchronize_net();
 469         nf_ct_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 470         nf_ct_l4proto_unregister(builtin_l4proto4,
 471                                  ARRAY_SIZE(builtin_l4proto4));
 472         unregister_pernet_subsys(&ipv4_net_ops);
 473         nf_unregister_sockopt(&so_getorigdst);
 474 }
 475
 476 module_init(nf_conntrack_l3proto_ipv4_init);
 477 module_exit(nf_conntrack_l3proto_ipv4_fini);