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_core.h>
  26 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  27 #include <net/netfilter/nf_nat_helper.h>
  28 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
  29
  30 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
  31                               struct nf_conn *ct,
  32                               enum ip_conntrack_info ctinfo);
  33 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
  34
  35 static bool ipv4_pkt_to_tuple(const struct sk_buff *skb, unsigned int nhoff,
  36                               struct nf_conntrack_tuple *tuple)
  37 {
  38         const __be32 *ap;
  39         __be32 _addrs[2];
  40         ap = skb_header_pointer(skb, nhoff + offsetof(struct iphdr, saddr),
  41                                 sizeof(u_int32_t) * 2, _addrs);
  42         if (ap == NULL)
  43                 return false;
  44
  45         tuple->src.u3.ip = ap[0];
  46         tuple->dst.u3.ip = ap[1];
  47
  48         return true;
  49 }
  50
  51 static bool ipv4_invert_tuple(struct nf_conntrack_tuple *tuple,
  52                               const struct nf_conntrack_tuple *orig)
  53 {
  54         tuple->src.u3.ip = orig->dst.u3.ip;
  55         tuple->dst.u3.ip = orig->src.u3.ip;
  56
  57         return true;
  58 }
  59
  60 static int ipv4_print_tuple(struct seq_file *s,
  61                             const struct nf_conntrack_tuple *tuple)
  62 {
  63         return seq_printf(s, "src=%pI4 dst=%pI4 ",
  64                           &tuple->src.u3.ip, &tuple->dst.u3.ip);
  65 }
  66
  67 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
  68                             unsigned int *dataoff, u_int8_t *protonum)
  69 {
  70         const struct iphdr *iph;
  71         struct iphdr _iph;
  72
  73         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  74         if (iph == NULL)
  75                 return -NF_DROP;
  76
  77         /* Conntrack defragments packets, we might still see fragments
  78          * inside ICMP packets though. */
  79         if (iph->frag_off & htons(IP_OFFSET))
  80                 return -NF_DROP;
  81
  82         *dataoff = nhoff + (iph->ihl << 2);
  83         *protonum = iph->protocol;
  84
  85         return NF_ACCEPT;
  86 }
  87
  88 static unsigned int ipv4_confirm(unsigned int hooknum,
  89                                  struct sk_buff *skb,
  90                                  const struct net_device *in,
  91                                  const struct net_device *out,
  92                                  int (*okfn)(struct sk_buff *))
  93 {
  94         struct nf_conn *ct;
  95         enum ip_conntrack_info ctinfo;
  96         const struct nf_conn_help *help;
  97         const struct nf_conntrack_helper *helper;
  98         unsigned int ret;
  99
 100         /* This is where we call the helper: as the packet goes out. */
 101         ct = nf_ct_get(skb, &ctinfo);
 102         if (!ct || ctinfo == IP_CT_RELATED + IP_CT_IS_REPLY)
 103                 goto out;
 104
 105         help = nfct_help(ct);
 106         if (!help)
 107                 goto out;
 108
 109         /* rcu_read_lock()ed by nf_hook_slow */
 110         helper = rcu_dereference(help->helper);
 111         if (!helper)
 112                 goto out;
 113
 114         ret = helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
 115                            ct, ctinfo);
 116         if (ret != NF_ACCEPT)
 117                 return ret;
 118
 119         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status)) {
 120                 typeof(nf_nat_seq_adjust_hook) seq_adjust;
 121
 122                 seq_adjust = rcu_dereference(nf_nat_seq_adjust_hook);
 123                 if (!seq_adjust || !seq_adjust(skb, ct, ctinfo))
 124                         return NF_DROP;
 125         }
 126 out:
 127         /* We've seen it coming out the other side: confirm it */
 128         return nf_conntrack_confirm(skb);
 129 }
 130
 131 static unsigned int ipv4_conntrack_in(unsigned int hooknum,
 132                                       struct sk_buff *skb,
 133                                       const struct net_device *in,
 134                                       const struct net_device *out,
 135                                       int (*okfn)(struct sk_buff *))
 136 {
 137         return nf_conntrack_in(dev_net(in), PF_INET, hooknum, skb);
 138 }
 139
 140 static unsigned int ipv4_conntrack_local(unsigned int hooknum,
 141                                          struct sk_buff *skb,
 142                                          const struct net_device *in,
 143                                          const struct net_device *out,
 144                                          int (*okfn)(struct sk_buff *))
 145 {
 146         /* root is playing with raw sockets. */
 147         if (skb->len < sizeof(struct iphdr) ||
 148             ip_hdrlen(skb) < sizeof(struct iphdr)) {
 149                 if (net_ratelimit())
 150                         printk("ipt_hook: happy cracking.\n");
 151                 return NF_ACCEPT;
 152         }
 153         return nf_conntrack_in(dev_net(out), PF_INET, hooknum, skb);
 154 }
 155
 156 /* Connection tracking may drop packets, but never alters them, so
 157    make it the first hook. */
 158 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 159         {
 160                 .hook           = ipv4_conntrack_in,
 161                 .owner          = THIS_MODULE,
 162                 .pf             = PF_INET,
 163                 .hooknum        = NF_INET_PRE_ROUTING,
 164                 .priority       = NF_IP_PRI_CONNTRACK,
 165         },
 166         {
 167                 .hook           = ipv4_conntrack_local,
 168                 .owner          = THIS_MODULE,
 169                 .pf             = PF_INET,
 170                 .hooknum        = NF_INET_LOCAL_OUT,
 171                 .priority       = NF_IP_PRI_CONNTRACK,
 172         },
 173         {
 174                 .hook           = ipv4_confirm,
 175                 .owner          = THIS_MODULE,
 176                 .pf             = PF_INET,
 177                 .hooknum        = NF_INET_POST_ROUTING,
 178                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 179         },
 180         {
 181                 .hook           = ipv4_confirm,
 182                 .owner          = THIS_MODULE,
 183                 .pf             = PF_INET,
 184                 .hooknum        = NF_INET_LOCAL_IN,
 185                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 186         },
 187 };
 188
 189 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 190 static int log_invalid_proto_min = 0;
 191 static int log_invalid_proto_max = 255;
 192
 193 static ctl_table ip_ct_sysctl_table[] = {
 194         {
 195                 .ctl_name       = NET_IPV4_NF_CONNTRACK_MAX,
 196                 .procname       = "ip_conntrack_max",
 197                 .data           = &nf_conntrack_max,
 198                 .maxlen         = sizeof(int),
 199                 .mode           = 0644,
 200                 .proc_handler   = proc_dointvec,
 201         },
 202         {
 203                 .ctl_name       = NET_IPV4_NF_CONNTRACK_COUNT,
 204                 .procname       = "ip_conntrack_count",
 205                 .data           = &init_net.ct.count,
 206                 .maxlen         = sizeof(int),
 207                 .mode           = 0444,
 208                 .proc_handler   = proc_dointvec,
 209         },
 210         {
 211                 .ctl_name       = NET_IPV4_NF_CONNTRACK_BUCKETS,
 212                 .procname       = "ip_conntrack_buckets",
 213                 .data           = &nf_conntrack_htable_size,
 214                 .maxlen         = sizeof(unsigned int),
 215                 .mode           = 0444,
 216                 .proc_handler   = proc_dointvec,
 217         },
 218         {
 219                 .ctl_name       = NET_IPV4_NF_CONNTRACK_CHECKSUM,
 220                 .procname       = "ip_conntrack_checksum",
 221                 .data           = &init_net.ct.sysctl_checksum,
 222                 .maxlen         = sizeof(int),
 223                 .mode           = 0644,
 224                 .proc_handler   = proc_dointvec,
 225         },
 226         {
 227                 .ctl_name       = NET_IPV4_NF_CONNTRACK_LOG_INVALID,
 228                 .procname       = "ip_conntrack_log_invalid",
 229                 .data           = &init_net.ct.sysctl_log_invalid,
 230                 .maxlen         = sizeof(unsigned int),
 231                 .mode           = 0644,
 232                 .proc_handler   = proc_dointvec_minmax,
 233                 .strategy       = sysctl_intvec,
 234                 .extra1         = &log_invalid_proto_min,
 235                 .extra2         = &log_invalid_proto_max,
 236         },
 237         {
 238                 .ctl_name       = 0
 239         }
 240 };
 241 #endif /* CONFIG_SYSCTL && CONFIG_NF_CONNTRACK_PROC_COMPAT */
 242
 243 /* Fast function for those who don't want to parse /proc (and I don't
 244    blame them). */
 245 /* Reversing the socket's dst/src point of view gives us the reply
 246    mapping. */
 247 static int
 248 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 249 {
 250         const struct inet_sock *inet = inet_sk(sk);
 251         const struct nf_conntrack_tuple_hash *h;
 252         struct nf_conntrack_tuple tuple;
 253
 254         memset(&tuple, 0, sizeof(tuple));
 255         tuple.src.u3.ip = inet->rcv_saddr;
 256         tuple.src.u.tcp.port = inet->sport;
 257         tuple.dst.u3.ip = inet->daddr;
 258         tuple.dst.u.tcp.port = inet->dport;
 259         tuple.src.l3num = PF_INET;
 260         tuple.dst.protonum = IPPROTO_TCP;
 261
 262         /* We only do TCP at the moment: is there a better way? */
 263         if (strcmp(sk->sk_prot->name, "TCP")) {
 264                 pr_debug("SO_ORIGINAL_DST: Not a TCP socket\n");
 265                 return -ENOPROTOOPT;
 266         }
 267
 268         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 269                 pr_debug("SO_ORIGINAL_DST: len %d not %Zu\n",
 270                          *len, sizeof(struct sockaddr_in));
 271                 return -EINVAL;
 272         }
 273
 274         h = nf_conntrack_find_get(sock_net(sk), &tuple);
 275         if (h) {
 276                 struct sockaddr_in sin;
 277                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 278
 279                 sin.sin_family = AF_INET;
 280                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 281                         .tuple.dst.u.tcp.port;
 282                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 283                         .tuple.dst.u3.ip;
 284                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 285
 286                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 287                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 288                 nf_ct_put(ct);
 289                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 290                         return -EFAULT;
 291                 else
 292                         return 0;
 293         }
 294         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 295                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 296                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 297         return -ENOENT;
 298 }
 299
 300 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 301
 302 #include <linux/netfilter/nfnetlink.h>
 303 #include <linux/netfilter/nfnetlink_conntrack.h>
 304
 305 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 306                                 const struct nf_conntrack_tuple *tuple)
 307 {
 308         NLA_PUT_BE32(skb, CTA_IP_V4_SRC, tuple->src.u3.ip);
 309         NLA_PUT_BE32(skb, CTA_IP_V4_DST, tuple->dst.u3.ip);
 310         return 0;
 311
 312 nla_put_failure:
 313         return -1;
 314 }
 315
 316 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 317         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 318         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 319 };
 320
 321 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 322                                 struct nf_conntrack_tuple *t)
 323 {
 324         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 325                 return -EINVAL;
 326
 327         t->src.u3.ip = nla_get_be32(tb[CTA_IP_V4_SRC]);
 328         t->dst.u3.ip = nla_get_be32(tb[CTA_IP_V4_DST]);
 329
 330         return 0;
 331 }
 332 #endif
 333
 334 static struct nf_sockopt_ops so_getorigdst = {
 335         .pf             = PF_INET,
 336         .get_optmin     = SO_ORIGINAL_DST,
 337         .get_optmax     = SO_ORIGINAL_DST+1,
 338         .get            = &getorigdst,
 339         .owner          = THIS_MODULE,
 340 };
 341
 342 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 343         .l3proto         = PF_INET,
 344         .name            = "ipv4",
 345         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 346         .invert_tuple    = ipv4_invert_tuple,
 347         .print_tuple     = ipv4_print_tuple,
 348         .get_l4proto     = ipv4_get_l4proto,
 349 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 350         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 351         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 352         .nla_policy      = ipv4_nla_policy,
 353 #endif
 354 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 355         .ctl_table_path  = nf_net_ipv4_netfilter_sysctl_path,
 356         .ctl_table       = ip_ct_sysctl_table,
 357 #endif
 358         .me              = THIS_MODULE,
 359 };
 360
 361 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 362                   &nf_conntrack_htable_size, 0600);
 363
 364 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 365 MODULE_ALIAS("ip_conntrack");
 366 MODULE_LICENSE("GPL");
 367
 368 static int __init nf_conntrack_l3proto_ipv4_init(void)
 369 {
 370         int ret = 0;
 371
 372         need_conntrack();
 373         nf_defrag_ipv4_enable();
 374
 375         ret = nf_register_sockopt(&so_getorigdst);
 376         if (ret < 0) {
 377                 printk(KERN_ERR "Unable to register netfilter socket option\n");
 378                 return ret;
 379         }
 380
 381         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_tcp4);
 382         if (ret < 0) {
 383                 printk("nf_conntrack_ipv4: can't register tcp.\n");
 384                 goto cleanup_sockopt;
 385         }
 386
 387         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_udp4);
 388         if (ret < 0) {
 389                 printk("nf_conntrack_ipv4: can't register udp.\n");
 390                 goto cleanup_tcp;
 391         }
 392
 393         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_icmp);
 394         if (ret < 0) {
 395                 printk("nf_conntrack_ipv4: can't register icmp.\n");
 396                 goto cleanup_udp;
 397         }
 398
 399         ret = nf_conntrack_l3proto_register(&nf_conntrack_l3proto_ipv4);
 400         if (ret < 0) {
 401                 printk("nf_conntrack_ipv4: can't register ipv4\n");
 402                 goto cleanup_icmp;
 403         }
 404
 405         ret = nf_register_hooks(ipv4_conntrack_ops,
 406                                 ARRAY_SIZE(ipv4_conntrack_ops));
 407         if (ret < 0) {
 408                 printk("nf_conntrack_ipv4: can't register hooks.\n");
 409                 goto cleanup_ipv4;
 410         }
 411 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 412         ret = nf_conntrack_ipv4_compat_init();
 413         if (ret < 0)
 414                 goto cleanup_hooks;
 415 #endif
 416         return ret;
 417 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 418  cleanup_hooks:
 419         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 420 #endif
 421  cleanup_ipv4:
 422         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 423  cleanup_icmp:
 424         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 425  cleanup_udp:
 426         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 427  cleanup_tcp:
 428         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 429  cleanup_sockopt:
 430         nf_unregister_sockopt(&so_getorigdst);
 431         return ret;
 432 }
 433
 434 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 435 {
 436         synchronize_net();
 437 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 438         nf_conntrack_ipv4_compat_fini();
 439 #endif
 440         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 441         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 442         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 443         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 444         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 445         nf_unregister_sockopt(&so_getorigdst);
 446 }
 447
 448 module_init(nf_conntrack_l3proto_ipv4_init);
 449 module_exit(nf_conntrack_l3proto_ipv4_fini);
 450
 451 void need_ipv4_conntrack(void)
 452 {
 453         return;
 454 }
 455 EXPORT_SYMBOL_GPL(need_ipv4_conntrack);