SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   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
   9 #include <linux/types.h>
  10 #include <linux/ip.h>
  11 #include <linux/netfilter.h>
  12 #include <linux/module.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/icmp.h>
  15 #include <linux/sysctl.h>
  16 #include <net/route.h>
  17 #include <net/ip.h>
  18
  19 #include <linux/netfilter_ipv4.h>
  20 #include <net/netfilter/nf_conntrack.h>
  21 #include <net/netfilter/nf_conntrack_helper.h>
  22 #include <net/netfilter/nf_conntrack_l4proto.h>
  23 #include <net/netfilter/nf_conntrack_l3proto.h>
  24 #include <net/netfilter/nf_conntrack_core.h>
  25 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  26 #include <net/netfilter/nf_nat_helper.h>
  27
  28 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  29                               struct nf_conn *ct,
  30                               enum ip_conntrack_info ctinfo);
  31 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  32
  33 static bool ipv4_pkt_to_tuple(const struct sk_buff *skb, unsigned int nhoff,
  34                               struct nf_conntrack_tuple *tuple)
  35 {
  36         const __be32 *ap;
  37         __be32 _addrs[2];
  38         ap = skb_header_pointer(skb, nhoff + offsetof(struct iphdr, saddr),
  39                                 sizeof(u_int32_t) * 2, _addrs);
  40         if (ap == NULL)
  41                 return false;
  42
  43         tuple->src.u3.ip = ap[0];
  44         tuple->dst.u3.ip = ap[1];
  45
  46         return true;
  47 }
  48
  49 static bool ipv4_invert_tuple(struct nf_conntrack_tuple *tuple,
  50                               const struct nf_conntrack_tuple *orig)
  51 {
  52         tuple->src.u3.ip = orig->dst.u3.ip;
  53         tuple->dst.u3.ip = orig->src.u3.ip;
  54
  55         return true;
  56 }
  57
  58 static int ipv4_print_tuple(struct seq_file *s,
  59                             const struct nf_conntrack_tuple *tuple)
  60 {
  61         return seq_printf(s, "src=%u.%u.%u.%u dst=%u.%u.%u.%u ",
  62                           NIPQUAD(tuple->src.u3.ip),
  63                           NIPQUAD(tuple->dst.u3.ip));
  64 }
  65
  66 /* Returns new sk_buff, or NULL */
  67 static int nf_ct_ipv4_gather_frags(struct sk_buff *skb, u_int32_t user)
  68 {
  69         int err;
  70
  71         skb_orphan(skb);
  72
  73         local_bh_disable();
  74         err = ip_defrag(skb, user);
  75         local_bh_enable();
  76
  77         if (!err)
  78                 ip_send_check(ip_hdr(skb));
  79
  80         return err;
  81 }
  82
  83 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
  84                             unsigned int *dataoff, u_int8_t *protonum)
  85 {
  86         const struct iphdr *iph;
  87         struct iphdr _iph;
  88
  89         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  90         if (iph == NULL)
  91                 return -NF_DROP;
  92
  93         /* Conntrack defragments packets, we might still see fragments
  94          * inside ICMP packets though. */
  95         if (iph->frag_off & htons(IP_OFFSET))
  96                 return -NF_DROP;
  97
  98         *dataoff = nhoff + (iph->ihl << 2);
  99         *protonum = iph->protocol;
 100
 101         return NF_ACCEPT;
 102 }
 103
 104 static unsigned int ipv4_confirm(unsigned int hooknum,
 105                                  struct sk_buff *skb,
 106                                  const struct net_device *in,
 107                                  const struct net_device *out,
 108                                  int (*okfn)(struct sk_buff *))
 109 {
 110         struct nf_conn *ct;
 111         enum ip_conntrack_info ctinfo;
 112         const struct nf_conn_help *help;
 113         const struct nf_conntrack_helper *helper;
 114         unsigned int ret;
 115
 116         /* This is where we call the helper: as the packet goes out. */
 117         ct = nf_ct_get(skb, &ctinfo);
 118         if (!ct || ctinfo == IP_CT_RELATED + IP_CT_IS_REPLY)
 119                 goto out;
 120
 121         help = nfct_help(ct);
 122         if (!help)
 123                 goto out;
 124
 125         /* rcu_read_lock()ed by nf_hook_slow */
 126         helper = rcu_dereference(help->helper);
 127         if (!helper)
 128                 goto out;
 129
 130         ret = helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
 131                            ct, ctinfo);
 132         if (ret != NF_ACCEPT)
 133                 return ret;
 134
 135         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status)) {
 136                 typeof(nf_nat_seq_adjust_hook) seq_adjust;
 137
 138                 seq_adjust = rcu_dereference(nf_nat_seq_adjust_hook);
 139                 if (!seq_adjust || !seq_adjust(skb, ct, ctinfo))
 140                         return NF_DROP;
 141         }
 142 out:
 143         /* We've seen it coming out the other side: confirm it */
 144         return nf_conntrack_confirm(skb);
 145 }
 146
 147 static unsigned int ipv4_conntrack_defrag(unsigned int hooknum,
 148                                           struct sk_buff *skb,
 149                                           const struct net_device *in,
 150                                           const struct net_device *out,
 151                                           int (*okfn)(struct sk_buff *))
 152 {
 153         /* Previously seen (loopback)?  Ignore.  Do this before
 154            fragment check. */
 155         if (skb->nfct)
 156                 return NF_ACCEPT;
 157
 158         /* Gather fragments. */
 159         if (ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) {
 160                 if (nf_ct_ipv4_gather_frags(skb,
 161                                             hooknum == NF_INET_PRE_ROUTING ?
 162                                             IP_DEFRAG_CONNTRACK_IN :
 163                                             IP_DEFRAG_CONNTRACK_OUT))
 164                         return NF_STOLEN;
 165         }
 166         return NF_ACCEPT;
 167 }
 168
 169 static unsigned int ipv4_conntrack_in(unsigned int hooknum,
 170                                       struct sk_buff *skb,
 171                                       const struct net_device *in,
 172                                       const struct net_device *out,
 173                                       int (*okfn)(struct sk_buff *))
 174 {
 175         return nf_conntrack_in(PF_INET, hooknum, skb);
 176 }
 177
 178 static unsigned int ipv4_conntrack_local(unsigned int hooknum,
 179                                          struct sk_buff *skb,
 180                                          const struct net_device *in,
 181                                          const struct net_device *out,
 182                                          int (*okfn)(struct sk_buff *))
 183 {
 184         /* root is playing with raw sockets. */
 185         if (skb->len < sizeof(struct iphdr) ||
 186             ip_hdrlen(skb) < sizeof(struct iphdr)) {
 187                 if (net_ratelimit())
 188                         printk("ipt_hook: happy cracking.\n");
 189                 return NF_ACCEPT;
 190         }
 191         return nf_conntrack_in(PF_INET, hooknum, skb);
 192 }
 193
 194 /* Connection tracking may drop packets, but never alters them, so
 195    make it the first hook. */
 196 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 197         {
 198                 .hook           = ipv4_conntrack_defrag,
 199                 .owner          = THIS_MODULE,
 200                 .pf             = PF_INET,
 201                 .hooknum        = NF_INET_PRE_ROUTING,
 202                 .priority       = NF_IP_PRI_CONNTRACK_DEFRAG,
 203         },
 204         {
 205                 .hook           = ipv4_conntrack_in,
 206                 .owner          = THIS_MODULE,
 207                 .pf             = PF_INET,
 208                 .hooknum        = NF_INET_PRE_ROUTING,
 209                 .priority       = NF_IP_PRI_CONNTRACK,
 210         },
 211         {
 212                 .hook           = ipv4_conntrack_defrag,
 213                 .owner          = THIS_MODULE,
 214                 .pf             = PF_INET,
 215                 .hooknum        = NF_INET_LOCAL_OUT,
 216                 .priority       = NF_IP_PRI_CONNTRACK_DEFRAG,
 217         },
 218         {
 219                 .hook           = ipv4_conntrack_local,
 220                 .owner          = THIS_MODULE,
 221                 .pf             = PF_INET,
 222                 .hooknum        = NF_INET_LOCAL_OUT,
 223                 .priority       = NF_IP_PRI_CONNTRACK,
 224         },
 225         {
 226                 .hook           = ipv4_confirm,
 227                 .owner          = THIS_MODULE,
 228                 .pf             = PF_INET,
 229                 .hooknum        = NF_INET_POST_ROUTING,
 230                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 231         },
 232         {
 233                 .hook           = ipv4_confirm,
 234                 .owner          = THIS_MODULE,
 235                 .pf             = PF_INET,
 236                 .hooknum        = NF_INET_LOCAL_IN,
 237                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 238         },
 239 };
 240
 241 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 242 static int log_invalid_proto_min = 0;
 243 static int log_invalid_proto_max = 255;
 244
 245 static ctl_table ip_ct_sysctl_table[] = {
 246         {
 247                 .ctl_name       = NET_IPV4_NF_CONNTRACK_MAX,
 248                 .procname       = "ip_conntrack_max",
 249                 .data           = &nf_conntrack_max,
 250                 .maxlen         = sizeof(int),
 251                 .mode           = 0644,
 252                 .proc_handler   = &proc_dointvec,
 253         },
 254         {
 255                 .ctl_name       = NET_IPV4_NF_CONNTRACK_COUNT,
 256                 .procname       = "ip_conntrack_count",
 257                 .data           = &init_net.ct.count,
 258                 .maxlen         = sizeof(int),
 259                 .mode           = 0444,
 260                 .proc_handler   = &proc_dointvec,
 261         },
 262         {
 263                 .ctl_name       = NET_IPV4_NF_CONNTRACK_BUCKETS,
 264                 .procname       = "ip_conntrack_buckets",
 265                 .data           = &nf_conntrack_htable_size,
 266                 .maxlen         = sizeof(unsigned int),
 267                 .mode           = 0444,
 268                 .proc_handler   = &proc_dointvec,
 269         },
 270         {
 271                 .ctl_name       = NET_IPV4_NF_CONNTRACK_CHECKSUM,
 272                 .procname       = "ip_conntrack_checksum",
 273                 .data           = &nf_conntrack_checksum,
 274                 .maxlen         = sizeof(int),
 275                 .mode           = 0644,
 276                 .proc_handler   = &proc_dointvec,
 277         },
 278         {
 279                 .ctl_name       = NET_IPV4_NF_CONNTRACK_LOG_INVALID,
 280                 .procname       = "ip_conntrack_log_invalid",
 281                 .data           = &nf_ct_log_invalid,
 282                 .maxlen         = sizeof(unsigned int),
 283                 .mode           = 0644,
 284                 .proc_handler   = &proc_dointvec_minmax,
 285                 .strategy       = &sysctl_intvec,
 286                 .extra1         = &log_invalid_proto_min,
 287                 .extra2         = &log_invalid_proto_max,
 288         },
 289         {
 290                 .ctl_name       = 0
 291         }
 292 };
 293 #endif /* CONFIG_SYSCTL && CONFIG_NF_CONNTRACK_PROC_COMPAT */
 294
 295 /* Fast function for those who don't want to parse /proc (and I don't
 296    blame them). */
 297 /* Reversing the socket's dst/src point of view gives us the reply
 298    mapping. */
 299 static int
 300 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 301 {
 302         const struct inet_sock *inet = inet_sk(sk);
 303         const struct nf_conntrack_tuple_hash *h;
 304         struct nf_conntrack_tuple tuple;
 305
 306         memset(&tuple, 0, sizeof(tuple));
 307         tuple.src.u3.ip = inet->rcv_saddr;
 308         tuple.src.u.tcp.port = inet->sport;
 309         tuple.dst.u3.ip = inet->daddr;
 310         tuple.dst.u.tcp.port = inet->dport;
 311         tuple.src.l3num = PF_INET;
 312         tuple.dst.protonum = IPPROTO_TCP;
 313
 314         /* We only do TCP at the moment: is there a better way? */
 315         if (strcmp(sk->sk_prot->name, "TCP")) {
 316                 pr_debug("SO_ORIGINAL_DST: Not a TCP socket\n");
 317                 return -ENOPROTOOPT;
 318         }
 319
 320         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 321                 pr_debug("SO_ORIGINAL_DST: len %d not %Zu\n",
 322                          *len, sizeof(struct sockaddr_in));
 323                 return -EINVAL;
 324         }
 325
 326         h = nf_conntrack_find_get(&tuple);
 327         if (h) {
 328                 struct sockaddr_in sin;
 329                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 330
 331                 sin.sin_family = AF_INET;
 332                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 333                         .tuple.dst.u.tcp.port;
 334                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 335                         .tuple.dst.u3.ip;
 336                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 337
 338                 pr_debug("SO_ORIGINAL_DST: %u.%u.%u.%u %u\n",
 339                          NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
 340                 nf_ct_put(ct);
 341                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 342                         return -EFAULT;
 343                 else
 344                         return 0;
 345         }
 346         pr_debug("SO_ORIGINAL_DST: Can't find %u.%u.%u.%u/%u-%u.%u.%u.%u/%u.\n",
 347                  NIPQUAD(tuple.src.u3.ip), ntohs(tuple.src.u.tcp.port),
 348                  NIPQUAD(tuple.dst.u3.ip), ntohs(tuple.dst.u.tcp.port));
 349         return -ENOENT;
 350 }
 351
 352 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 353
 354 #include <linux/netfilter/nfnetlink.h>
 355 #include <linux/netfilter/nfnetlink_conntrack.h>
 356
 357 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 358                                 const struct nf_conntrack_tuple *tuple)
 359 {
 360         NLA_PUT_BE32(skb, CTA_IP_V4_SRC, tuple->src.u3.ip);
 361         NLA_PUT_BE32(skb, CTA_IP_V4_DST, tuple->dst.u3.ip);
 362         return 0;
 363
 364 nla_put_failure:
 365         return -1;
 366 }
 367
 368 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 369         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 370         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 371 };
 372
 373 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 374                                 struct nf_conntrack_tuple *t)
 375 {
 376         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 377                 return -EINVAL;
 378
 379         t->src.u3.ip = nla_get_be32(tb[CTA_IP_V4_SRC]);
 380         t->dst.u3.ip = nla_get_be32(tb[CTA_IP_V4_DST]);
 381
 382         return 0;
 383 }
 384 #endif
 385
 386 static struct nf_sockopt_ops so_getorigdst = {
 387         .pf             = PF_INET,
 388         .get_optmin     = SO_ORIGINAL_DST,
 389         .get_optmax     = SO_ORIGINAL_DST+1,
 390         .get            = &getorigdst,
 391         .owner          = THIS_MODULE,
 392 };
 393
 394 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 395         .l3proto         = PF_INET,
 396         .name            = "ipv4",
 397         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 398         .invert_tuple    = ipv4_invert_tuple,
 399         .print_tuple     = ipv4_print_tuple,
 400         .get_l4proto     = ipv4_get_l4proto,
 401 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 402         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 403         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 404         .nla_policy      = ipv4_nla_policy,
 405 #endif
 406 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 407         .ctl_table_path  = nf_net_ipv4_netfilter_sysctl_path,
 408         .ctl_table       = ip_ct_sysctl_table,
 409 #endif
 410         .me              = THIS_MODULE,
 411 };
 412
 413 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 414                   &nf_conntrack_htable_size, 0600);
 415
 416 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 417 MODULE_ALIAS("ip_conntrack");
 418 MODULE_LICENSE("GPL");
 419
 420 static int __init nf_conntrack_l3proto_ipv4_init(void)
 421 {
 422         int ret = 0;
 423
 424         need_conntrack();
 425
 426         ret = nf_register_sockopt(&so_getorigdst);
 427         if (ret < 0) {
 428                 printk(KERN_ERR "Unable to register netfilter socket option\n");
 429                 return ret;
 430         }
 431
 432         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_tcp4);
 433         if (ret < 0) {
 434                 printk("nf_conntrack_ipv4: can't register tcp.\n");
 435                 goto cleanup_sockopt;
 436         }
 437
 438         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_udp4);
 439         if (ret < 0) {
 440                 printk("nf_conntrack_ipv4: can't register udp.\n");
 441                 goto cleanup_tcp;
 442         }
 443
 444         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_icmp);
 445         if (ret < 0) {
 446                 printk("nf_conntrack_ipv4: can't register icmp.\n");
 447                 goto cleanup_udp;
 448         }
 449
 450         ret = nf_conntrack_l3proto_register(&nf_conntrack_l3proto_ipv4);
 451         if (ret < 0) {
 452                 printk("nf_conntrack_ipv4: can't register ipv4\n");
 453                 goto cleanup_icmp;
 454         }
 455
 456         ret = nf_register_hooks(ipv4_conntrack_ops,
 457                                 ARRAY_SIZE(ipv4_conntrack_ops));
 458         if (ret < 0) {
 459                 printk("nf_conntrack_ipv4: can't register hooks.\n");
 460                 goto cleanup_ipv4;
 461         }
 462 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 463         ret = nf_conntrack_ipv4_compat_init();
 464         if (ret < 0)
 465                 goto cleanup_hooks;
 466 #endif
 467         return ret;
 468 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 469  cleanup_hooks:
 470         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 471 #endif
 472  cleanup_ipv4:
 473         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 474  cleanup_icmp:
 475         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 476  cleanup_udp:
 477         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 478  cleanup_tcp:
 479         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 480  cleanup_sockopt:
 481         nf_unregister_sockopt(&so_getorigdst);
 482         return ret;
 483 }
 484
 485 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 486 {
 487         synchronize_net();
 488 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 489         nf_conntrack_ipv4_compat_fini();
 490 #endif
 491         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 492         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 493         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 494         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 495         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 496         nf_unregister_sockopt(&so_getorigdst);
 497 }
 498
 499 module_init(nf_conntrack_l3proto_ipv4_init);
 500 module_exit(nf_conntrack_l3proto_ipv4_fini);
 501
 502 void need_ipv4_conntrack(void)
 503 {
 504         return;
 505 }
 506 EXPORT_SYMBOL_GPL(need_ipv4_conntrack);