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  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License version 2 as
   7  * published by the Free Software Foundation.
   8  */
   9
  10 #include <linux/types.h>
  11 #include <linux/ip.h>
  12 #include <linux/netfilter.h>
  13 #include <linux/module.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/icmp.h>
  16 #include <linux/sysctl.h>
  17 #include <net/route.h>
  18 #include <net/ip.h>
  19
  20 #include <linux/netfilter_ipv4.h>
  21 #include <net/netfilter/nf_conntrack.h>
  22 #include <net/netfilter/nf_conntrack_helper.h>
  23 #include <net/netfilter/nf_conntrack_l4proto.h>
  24 #include <net/netfilter/nf_conntrack_l3proto.h>
  25 #include <net/netfilter/nf_conntrack_zones.h>
  26 #include <net/netfilter/nf_conntrack_core.h>
  27 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  28 #include <net/netfilter/nf_nat_helper.h>
  29 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
  30 #include <net/netfilter/nf_log.h>
  31
  32 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  33                               struct nf_conn *ct,
  34                               enum ip_conntrack_info ctinfo);
  35 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  36
  37 static bool ipv4_pkt_to_tuple(const struct sk_buff *skb, unsigned int nhoff,
  38                               struct nf_conntrack_tuple *tuple)
  39 {
  40         const __be32 *ap;
  41         __be32 _addrs[2];
  42         ap = skb_header_pointer(skb, nhoff + offsetof(struct iphdr, saddr),
  43                                 sizeof(u_int32_t) * 2, _addrs);
  44         if (ap == NULL)
  45                 return false;
  46
  47         tuple->src.u3.ip = ap[0];
  48         tuple->dst.u3.ip = ap[1];
  49
  50         return true;
  51 }
  52
  53 static bool ipv4_invert_tuple(struct nf_conntrack_tuple *tuple,
  54                               const struct nf_conntrack_tuple *orig)
  55 {
  56         tuple->src.u3.ip = orig->dst.u3.ip;
  57         tuple->dst.u3.ip = orig->src.u3.ip;
  58
  59         return true;
  60 }
  61
  62 static int ipv4_print_tuple(struct seq_file *s,
  63                             const struct nf_conntrack_tuple *tuple)
  64 {
  65         return seq_printf(s, "src=%pI4 dst=%pI4 ",
  66                           &tuple->src.u3.ip, &tuple->dst.u3.ip);
  67 }
  68
  69 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
  70                             unsigned int *dataoff, u_int8_t *protonum)
  71 {
  72         const struct iphdr *iph;
  73         struct iphdr _iph;
  74
  75         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  76         if (iph == NULL)
  77                 return -NF_ACCEPT;
  78
  79         /* Conntrack defragments packets, we might still see fragments
  80          * inside ICMP packets though. */
  81         if (iph->frag_off & htons(IP_OFFSET))
  82                 return -NF_ACCEPT;
  83
  84         *dataoff = nhoff + (iph->ihl << 2);
  85         *protonum = iph->protocol;
  86
  87         /* Check bogus IP headers */
  88         if (*dataoff > skb->len) {
  89                 pr_debug("nf_conntrack_ipv4: bogus IPv4 packet: "
  90                          "nhoff %u, ihl %u, skblen %u\n",
  91                          nhoff, iph->ihl << 2, skb->len);
  92                 return -NF_ACCEPT;
  93         }
  94
  95         return NF_ACCEPT;
  96 }
  97
  98 static unsigned int ipv4_helper(unsigned int hooknum,
  99                                 struct sk_buff *skb,
 100                                 const struct net_device *in,
 101                                 const struct net_device *out,
 102                                 int (*okfn)(struct sk_buff *))
 103 {
 104         struct nf_conn *ct;
 105         enum ip_conntrack_info ctinfo;
 106         const struct nf_conn_help *help;
 107         const struct nf_conntrack_helper *helper;
 108         unsigned int ret;
 109
 110         /* This is where we call the helper: as the packet goes out. */
 111         ct = nf_ct_get(skb, &ctinfo);
 112         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 113                 return NF_ACCEPT;
 114
 115         help = nfct_help(ct);
 116         if (!help)
 117                 return NF_ACCEPT;
 118
 119         /* rcu_read_lock()ed by nf_hook_slow */
 120         helper = rcu_dereference(help->helper);
 121         if (!helper)
 122                 return NF_ACCEPT;
 123
 124         ret = helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
 125                            ct, ctinfo);
 126         if (ret != NF_ACCEPT && (ret & NF_VERDICT_MASK) != NF_QUEUE) {
 127                 nf_log_packet(NFPROTO_IPV4, hooknum, skb, in, out, NULL,
 128                               "nf_ct_%s: dropping packet", helper->name);
 129         }
 130         return ret;
 131 }
 132
 133 static unsigned int ipv4_confirm(unsigned int hooknum,
 134                                  struct sk_buff *skb,
 135                                  const struct net_device *in,
 136                                  const struct net_device *out,
 137                                  int (*okfn)(struct sk_buff *))
 138 {
 139         struct nf_conn *ct;
 140         enum ip_conntrack_info ctinfo;
 141
 142         ct = nf_ct_get(skb, &ctinfo);
 143         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 144                 goto out;
 145
 146         /* adjust seqs for loopback traffic only in outgoing direction */
 147         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 148             !nf_is_loopback_packet(skb)) {
 149                 typeof(nf_nat_seq_adjust_hook) seq_adjust;
 150
 151                 seq_adjust = rcu_dereference(nf_nat_seq_adjust_hook);
 152                 if (!seq_adjust || !seq_adjust(skb, ct, ctinfo)) {
 153                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
 154                         return NF_DROP;
 155                 }
 156         }
 157 out:
 158         /* We've seen it coming out the other side: confirm it */
 159         return nf_conntrack_confirm(skb);
 160 }
 161
 162 static unsigned int ipv4_conntrack_in(unsigned int hooknum,
 163                                       struct sk_buff *skb,
 164                                       const struct net_device *in,
 165                                       const struct net_device *out,
 166                                       int (*okfn)(struct sk_buff *))
 167 {
 168         return nf_conntrack_in(dev_net(in), PF_INET, hooknum, skb);
 169 }
 170
 171 static unsigned int ipv4_conntrack_local(unsigned int hooknum,
 172                                          struct sk_buff *skb,
 173                                          const struct net_device *in,
 174                                          const struct net_device *out,
 175                                          int (*okfn)(struct sk_buff *))
 176 {
 177         /* root is playing with raw sockets. */
 178         if (skb->len < sizeof(struct iphdr) ||
 179             ip_hdrlen(skb) < sizeof(struct iphdr))
 180                 return NF_ACCEPT;
 181         return nf_conntrack_in(dev_net(out), PF_INET, hooknum, skb);
 182 }
 183
 184 /* Connection tracking may drop packets, but never alters them, so
 185    make it the first hook. */
 186 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 187         {
 188                 .hook           = ipv4_conntrack_in,
 189                 .owner          = THIS_MODULE,
 190                 .pf             = NFPROTO_IPV4,
 191                 .hooknum        = NF_INET_PRE_ROUTING,
 192                 .priority       = NF_IP_PRI_CONNTRACK,
 193         },
 194         {
 195                 .hook           = ipv4_conntrack_local,
 196                 .owner          = THIS_MODULE,
 197                 .pf             = NFPROTO_IPV4,
 198                 .hooknum        = NF_INET_LOCAL_OUT,
 199                 .priority       = NF_IP_PRI_CONNTRACK,
 200         },
 201         {
 202                 .hook           = ipv4_helper,
 203                 .owner          = THIS_MODULE,
 204                 .pf             = NFPROTO_IPV4,
 205                 .hooknum        = NF_INET_POST_ROUTING,
 206                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 207         },
 208         {
 209                 .hook           = ipv4_confirm,
 210                 .owner          = THIS_MODULE,
 211                 .pf             = NFPROTO_IPV4,
 212                 .hooknum        = NF_INET_POST_ROUTING,
 213                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 214         },
 215         {
 216                 .hook           = ipv4_helper,
 217                 .owner          = THIS_MODULE,
 218                 .pf             = NFPROTO_IPV4,
 219                 .hooknum        = NF_INET_LOCAL_IN,
 220                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 221         },
 222         {
 223                 .hook           = ipv4_confirm,
 224                 .owner          = THIS_MODULE,
 225                 .pf             = NFPROTO_IPV4,
 226                 .hooknum        = NF_INET_LOCAL_IN,
 227                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 228         },
 229 };
 230
 231 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 232 static int log_invalid_proto_min = 0;
 233 static int log_invalid_proto_max = 255;
 234
 235 static ctl_table ip_ct_sysctl_table[] = {
 236         {
 237                 .procname       = "ip_conntrack_max",
 238                 .maxlen         = sizeof(int),
 239                 .mode           = 0644,
 240                 .proc_handler   = proc_dointvec,
 241         },
 242         {
 243                 .procname       = "ip_conntrack_count",
 244                 .maxlen         = sizeof(int),
 245                 .mode           = 0444,
 246                 .proc_handler   = proc_dointvec,
 247         },
 248         {
 249                 .procname       = "ip_conntrack_buckets",
 250                 .maxlen         = sizeof(unsigned int),
 251                 .mode           = 0444,
 252                 .proc_handler   = proc_dointvec,
 253         },
 254         {
 255                 .procname       = "ip_conntrack_checksum",
 256                 .maxlen         = sizeof(int),
 257                 .mode           = 0644,
 258                 .proc_handler   = proc_dointvec,
 259         },
 260         {
 261                 .procname       = "ip_conntrack_log_invalid",
 262                 .maxlen         = sizeof(unsigned int),
 263                 .mode           = 0644,
 264                 .proc_handler   = proc_dointvec_minmax,
 265                 .extra1         = &log_invalid_proto_min,
 266                 .extra2         = &log_invalid_proto_max,
 267         },
 268         { }
 269 };
 270 #endif /* CONFIG_SYSCTL && CONFIG_NF_CONNTRACK_PROC_COMPAT */
 271
 272 /* Fast function for those who don't want to parse /proc (and I don't
 273    blame them). */
 274 /* Reversing the socket's dst/src point of view gives us the reply
 275    mapping. */
 276 static int
 277 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 278 {
 279         const struct inet_sock *inet = inet_sk(sk);
 280         const struct nf_conntrack_tuple_hash *h;
 281         struct nf_conntrack_tuple tuple;
 282
 283         memset(&tuple, 0, sizeof(tuple));
 284         tuple.src.u3.ip = inet->inet_rcv_saddr;
 285         tuple.src.u.tcp.port = inet->inet_sport;
 286         tuple.dst.u3.ip = inet->inet_daddr;
 287         tuple.dst.u.tcp.port = inet->inet_dport;
 288         tuple.src.l3num = PF_INET;
 289         tuple.dst.protonum = sk->sk_protocol;
 290
 291         /* We only do TCP and SCTP at the moment: is there a better way? */
 292         if (sk->sk_protocol != IPPROTO_TCP && sk->sk_protocol != IPPROTO_SCTP) {
 293                 pr_debug("SO_ORIGINAL_DST: Not a TCP/SCTP socket\n");
 294                 return -ENOPROTOOPT;
 295         }
 296
 297         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 298                 pr_debug("SO_ORIGINAL_DST: len %d not %Zu\n",
 299                          *len, sizeof(struct sockaddr_in));
 300                 return -EINVAL;
 301         }
 302
 303         h = nf_conntrack_find_get(sock_net(sk), NF_CT_DEFAULT_ZONE, &tuple);
 304         if (h) {
 305                 struct sockaddr_in sin;
 306                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 307
 308                 sin.sin_family = AF_INET;
 309                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 310                         .tuple.dst.u.tcp.port;
 311                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 312                         .tuple.dst.u3.ip;
 313                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 314
 315                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 316                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 317                 nf_ct_put(ct);
 318                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 319                         return -EFAULT;
 320                 else
 321                         return 0;
 322         }
 323         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 324                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 325                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 326         return -ENOENT;
 327 }
 328
 329 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 330
 331 #include <linux/netfilter/nfnetlink.h>
 332 #include <linux/netfilter/nfnetlink_conntrack.h>
 333
 334 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 335                                 const struct nf_conntrack_tuple *tuple)
 336 {
 337         if (nla_put_be32(skb, CTA_IP_V4_SRC, tuple->src.u3.ip) ||
 338             nla_put_be32(skb, CTA_IP_V4_DST, tuple->dst.u3.ip))
 339                 goto nla_put_failure;
 340         return 0;
 341
 342 nla_put_failure:
 343         return -1;
 344 }
 345
 346 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 347         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 348         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 349 };
 350
 351 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 352                                 struct nf_conntrack_tuple *t)
 353 {
 354         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 355                 return -EINVAL;
 356
 357         t->src.u3.ip = nla_get_be32(tb[CTA_IP_V4_SRC]);
 358         t->dst.u3.ip = nla_get_be32(tb[CTA_IP_V4_DST]);
 359
 360         return 0;
 361 }
 362
 363 static int ipv4_nlattr_tuple_size(void)
 364 {
 365         return nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1);
 366 }
 367 #endif
 368
 369 static struct nf_sockopt_ops so_getorigdst = {
 370         .pf             = PF_INET,
 371         .get_optmin     = SO_ORIGINAL_DST,
 372         .get_optmax     = SO_ORIGINAL_DST+1,
 373         .get            = &getorigdst,
 374         .owner          = THIS_MODULE,
 375 };
 376
 377 static int ipv4_init_net(struct net *net)
 378 {
 379 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 380         struct nf_ip_net *in = &net->ct.nf_ct_proto;
 381         in->ctl_table = kmemdup(ip_ct_sysctl_table,
 382                                 sizeof(ip_ct_sysctl_table),
 383                                 GFP_KERNEL);
 384         if (!in->ctl_table)
 385                 return -ENOMEM;
 386
 387         in->ctl_table[0].data = &nf_conntrack_max;
 388         in->ctl_table[1].data = &net->ct.count;
 389         in->ctl_table[2].data = &net->ct.htable_size;
 390         in->ctl_table[3].data = &net->ct.sysctl_checksum;
 391         in->ctl_table[4].data = &net->ct.sysctl_log_invalid;
 392 #endif
 393         return 0;
 394 }
 395
 396 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 397         .l3proto         = PF_INET,
 398         .name            = "ipv4",
 399         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 400         .invert_tuple    = ipv4_invert_tuple,
 401         .print_tuple     = ipv4_print_tuple,
 402         .get_l4proto     = ipv4_get_l4proto,
 403 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 404         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 405         .nlattr_tuple_size = ipv4_nlattr_tuple_size,
 406         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 407         .nla_policy      = ipv4_nla_policy,
 408 #endif
 409 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 410         .ctl_table_path  = "net/ipv4/netfilter",
 411 #endif
 412         .init_net        = ipv4_init_net,
 413         .me              = THIS_MODULE,
 414 };
 415
 416 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 417                   &nf_conntrack_htable_size, 0600);
 418
 419 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 420 MODULE_ALIAS("ip_conntrack");
 421 MODULE_LICENSE("GPL");
 422
 423 static int ipv4_net_init(struct net *net)
 424 {
 425         int ret = 0;
 426
 427         ret = nf_conntrack_l4proto_register(net,
 428                                             &nf_conntrack_l4proto_tcp4);
 429         if (ret < 0) {
 430                 pr_err("nf_conntrack_l4proto_tcp4 :protocol register failed\n");
 431                 goto out_tcp;
 432         }
 433         ret = nf_conntrack_l4proto_register(net,
 434                                             &nf_conntrack_l4proto_udp4);
 435         if (ret < 0) {
 436                 pr_err("nf_conntrack_l4proto_udp4 :protocol register failed\n");
 437                 goto out_udp;
 438         }
 439         ret = nf_conntrack_l4proto_register(net,
 440                                             &nf_conntrack_l4proto_icmp);
 441         if (ret < 0) {
 442                 pr_err("nf_conntrack_l4proto_icmp4 :protocol register failed\n");
 443                 goto out_icmp;
 444         }
 445         ret = nf_conntrack_l3proto_register(net,
 446                                             &nf_conntrack_l3proto_ipv4);
 447         if (ret < 0) {
 448                 pr_err("nf_conntrack_l3proto_ipv4 :protocol register failed\n");
 449                 goto out_ipv4;
 450         }
 451         return 0;
 452 out_ipv4:
 453         nf_conntrack_l4proto_unregister(net,
 454                                         &nf_conntrack_l4proto_icmp);
 455 out_icmp:
 456         nf_conntrack_l4proto_unregister(net,
 457                                         &nf_conntrack_l4proto_udp4);
 458 out_udp:
 459         nf_conntrack_l4proto_unregister(net,
 460                                         &nf_conntrack_l4proto_tcp4);
 461 out_tcp:
 462         return ret;
 463 }
 464
 465 static void ipv4_net_exit(struct net *net)
 466 {
 467         nf_conntrack_l3proto_unregister(net,
 468                                         &nf_conntrack_l3proto_ipv4);
 469         nf_conntrack_l4proto_unregister(net,
 470                                         &nf_conntrack_l4proto_icmp);
 471         nf_conntrack_l4proto_unregister(net,
 472                                         &nf_conntrack_l4proto_udp4);
 473         nf_conntrack_l4proto_unregister(net,
 474                                         &nf_conntrack_l4proto_tcp4);
 475 }
 476
 477 static struct pernet_operations ipv4_net_ops = {
 478         .init = ipv4_net_init,
 479         .exit = ipv4_net_exit,
 480 };
 481
 482 static int __init nf_conntrack_l3proto_ipv4_init(void)
 483 {
 484         int ret = 0;
 485
 486         need_conntrack();
 487         nf_defrag_ipv4_enable();
 488
 489         ret = nf_register_sockopt(&so_getorigdst);
 490         if (ret < 0) {
 491                 printk(KERN_ERR "Unable to register netfilter socket option\n");
 492                 return ret;
 493         }
 494
 495         ret = register_pernet_subsys(&ipv4_net_ops);
 496         if (ret < 0) {
 497                 pr_err("nf_conntrack_ipv4: can't register pernet ops\n");
 498                 goto cleanup_sockopt;
 499         }
 500
 501         ret = nf_register_hooks(ipv4_conntrack_ops,
 502                                 ARRAY_SIZE(ipv4_conntrack_ops));
 503         if (ret < 0) {
 504                 pr_err("nf_conntrack_ipv4: can't register hooks.\n");
 505                 goto cleanup_pernet;
 506         }
 507 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 508         ret = nf_conntrack_ipv4_compat_init();
 509         if (ret < 0)
 510                 goto cleanup_hooks;
 511 #endif
 512         return ret;
 513 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 514  cleanup_hooks:
 515         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 516 #endif
 517  cleanup_pernet:
 518         unregister_pernet_subsys(&ipv4_net_ops);
 519  cleanup_sockopt:
 520         nf_unregister_sockopt(&so_getorigdst);
 521         return ret;
 522 }
 523
 524 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 525 {
 526         synchronize_net();
 527 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 528         nf_conntrack_ipv4_compat_fini();
 529 #endif
 530         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 531         unregister_pernet_subsys(&ipv4_net_ops);
 532         nf_unregister_sockopt(&so_getorigdst);
 533 }
 534
 535 module_init(nf_conntrack_l3proto_ipv4_init);
 536 module_exit(nf_conntrack_l3proto_ipv4_fini);
 537
 538 void need_ipv4_conntrack(void)
 539 {
 540         return;
 541 }
 542 EXPORT_SYMBOL_GPL(need_ipv4_conntrack);