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_DROP;
  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_DROP;
  83
  84         *dataoff = nhoff + (iph->ihl << 2);
  85         *protonum = iph->protocol;
  86
  87         return NF_ACCEPT;
  88 }
  89
  90 static unsigned int ipv4_confirm(unsigned int hooknum,
  91                                  struct sk_buff *skb,
  92                                  const struct net_device *in,
  93                                  const struct net_device *out,
  94                                  int (*okfn)(struct sk_buff *))
  95 {
  96         struct nf_conn *ct;
  97         enum ip_conntrack_info ctinfo;
  98         const struct nf_conn_help *help;
  99         const struct nf_conntrack_helper *helper;
 100         unsigned int ret;
 101
 102         /* This is where we call the helper: as the packet goes out. */
 103         ct = nf_ct_get(skb, &ctinfo);
 104         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 105                 goto out;
 106
 107         help = nfct_help(ct);
 108         if (!help)
 109                 goto out;
 110
 111         /* rcu_read_lock()ed by nf_hook_slow */
 112         helper = rcu_dereference(help->helper);
 113         if (!helper)
 114                 goto out;
 115
 116         ret = helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
 117                            ct, ctinfo);
 118         if (ret != NF_ACCEPT) {
 119                 nf_log_packet(NFPROTO_IPV4, hooknum, skb, in, out, NULL,
 120                               "nf_ct_%s: dropping packet", helper->name);
 121                 return ret;
 122         }
 123
 124         /* adjust seqs for loopback traffic only in outgoing direction */
 125         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 126             !nf_is_loopback_packet(skb)) {
 127                 typeof(nf_nat_seq_adjust_hook) seq_adjust;
 128
 129                 seq_adjust = rcu_dereference(nf_nat_seq_adjust_hook);
 130                 if (!seq_adjust || !seq_adjust(skb, ct, ctinfo)) {
 131                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
 132                         return NF_DROP;
 133                 }
 134         }
 135 out:
 136         /* We've seen it coming out the other side: confirm it */
 137         return nf_conntrack_confirm(skb);
 138 }
 139
 140 static unsigned int ipv4_conntrack_in(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         return nf_conntrack_in(dev_net(in), PF_INET, hooknum, skb);
 147 }
 148
 149 static unsigned int ipv4_conntrack_local(unsigned int hooknum,
 150                                          struct sk_buff *skb,
 151                                          const struct net_device *in,
 152                                          const struct net_device *out,
 153                                          int (*okfn)(struct sk_buff *))
 154 {
 155         /* root is playing with raw sockets. */
 156         if (skb->len < sizeof(struct iphdr) ||
 157             ip_hdrlen(skb) < sizeof(struct iphdr))
 158                 return NF_ACCEPT;
 159         return nf_conntrack_in(dev_net(out), PF_INET, hooknum, skb);
 160 }
 161
 162 /* Connection tracking may drop packets, but never alters them, so
 163    make it the first hook. */
 164 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 165         {
 166                 .hook           = ipv4_conntrack_in,
 167                 .owner          = THIS_MODULE,
 168                 .pf             = NFPROTO_IPV4,
 169                 .hooknum        = NF_INET_PRE_ROUTING,
 170                 .priority       = NF_IP_PRI_CONNTRACK,
 171         },
 172         {
 173                 .hook           = ipv4_conntrack_local,
 174                 .owner          = THIS_MODULE,
 175                 .pf             = NFPROTO_IPV4,
 176                 .hooknum        = NF_INET_LOCAL_OUT,
 177                 .priority       = NF_IP_PRI_CONNTRACK,
 178         },
 179         {
 180                 .hook           = ipv4_confirm,
 181                 .owner          = THIS_MODULE,
 182                 .pf             = NFPROTO_IPV4,
 183                 .hooknum        = NF_INET_POST_ROUTING,
 184                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 185         },
 186         {
 187                 .hook           = ipv4_confirm,
 188                 .owner          = THIS_MODULE,
 189                 .pf             = NFPROTO_IPV4,
 190                 .hooknum        = NF_INET_LOCAL_IN,
 191                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 192         },
 193 };
 194
 195 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 196 static int log_invalid_proto_min = 0;
 197 static int log_invalid_proto_max = 255;
 198
 199 static ctl_table ip_ct_sysctl_table[] = {
 200         {
 201                 .procname       = "ip_conntrack_max",
 202                 .data           = &nf_conntrack_max,
 203                 .maxlen         = sizeof(int),
 204                 .mode           = 0644,
 205                 .proc_handler   = proc_dointvec,
 206         },
 207         {
 208                 .procname       = "ip_conntrack_count",
 209                 .data           = &init_net.ct.count,
 210                 .maxlen         = sizeof(int),
 211                 .mode           = 0444,
 212                 .proc_handler   = proc_dointvec,
 213         },
 214         {
 215                 .procname       = "ip_conntrack_buckets",
 216                 .data           = &init_net.ct.htable_size,
 217                 .maxlen         = sizeof(unsigned int),
 218                 .mode           = 0444,
 219                 .proc_handler   = proc_dointvec,
 220         },
 221         {
 222                 .procname       = "ip_conntrack_checksum",
 223                 .data           = &init_net.ct.sysctl_checksum,
 224                 .maxlen         = sizeof(int),
 225                 .mode           = 0644,
 226                 .proc_handler   = proc_dointvec,
 227         },
 228         {
 229                 .procname       = "ip_conntrack_log_invalid",
 230                 .data           = &init_net.ct.sysctl_log_invalid,
 231                 .maxlen         = sizeof(unsigned int),
 232                 .mode           = 0644,
 233                 .proc_handler   = proc_dointvec_minmax,
 234                 .extra1         = &log_invalid_proto_min,
 235                 .extra2         = &log_invalid_proto_max,
 236         },
 237         { }
 238 };
 239 #endif /* CONFIG_SYSCTL && CONFIG_NF_CONNTRACK_PROC_COMPAT */
 240
 241 /* Fast function for those who don't want to parse /proc (and I don't
 242    blame them). */
 243 /* Reversing the socket's dst/src point of view gives us the reply
 244    mapping. */
 245 static int
 246 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 247 {
 248         const struct inet_sock *inet = inet_sk(sk);
 249         const struct nf_conntrack_tuple_hash *h;
 250         struct nf_conntrack_tuple tuple;
 251
 252         memset(&tuple, 0, sizeof(tuple));
 253         tuple.src.u3.ip = inet->inet_rcv_saddr;
 254         tuple.src.u.tcp.port = inet->inet_sport;
 255         tuple.dst.u3.ip = inet->inet_daddr;
 256         tuple.dst.u.tcp.port = inet->inet_dport;
 257         tuple.src.l3num = PF_INET;
 258         tuple.dst.protonum = sk->sk_protocol;
 259
 260         /* We only do TCP and SCTP at the moment: is there a better way? */
 261         if (sk->sk_protocol != IPPROTO_TCP && sk->sk_protocol != IPPROTO_SCTP) {
 262                 pr_debug("SO_ORIGINAL_DST: Not a TCP/SCTP socket\n");
 263                 return -ENOPROTOOPT;
 264         }
 265
 266         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 267                 pr_debug("SO_ORIGINAL_DST: len %d not %Zu\n",
 268                          *len, sizeof(struct sockaddr_in));
 269                 return -EINVAL;
 270         }
 271
 272         h = nf_conntrack_find_get(sock_net(sk), NF_CT_DEFAULT_ZONE, &tuple);
 273         if (h) {
 274                 struct sockaddr_in sin;
 275                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 276
 277                 sin.sin_family = AF_INET;
 278                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 279                         .tuple.dst.u.tcp.port;
 280                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 281                         .tuple.dst.u3.ip;
 282                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 283
 284                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 285                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 286                 nf_ct_put(ct);
 287                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 288                         return -EFAULT;
 289                 else
 290                         return 0;
 291         }
 292         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 293                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 294                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 295         return -ENOENT;
 296 }
 297
 298 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 299
 300 #include <linux/netfilter/nfnetlink.h>
 301 #include <linux/netfilter/nfnetlink_conntrack.h>
 302
 303 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 304                                 const struct nf_conntrack_tuple *tuple)
 305 {
 306         NLA_PUT_BE32(skb, CTA_IP_V4_SRC, tuple->src.u3.ip);
 307         NLA_PUT_BE32(skb, CTA_IP_V4_DST, tuple->dst.u3.ip);
 308         return 0;
 309
 310 nla_put_failure:
 311         return -1;
 312 }
 313
 314 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 315         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 316         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 317 };
 318
 319 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 320                                 struct nf_conntrack_tuple *t)
 321 {
 322         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 323                 return -EINVAL;
 324
 325         t->src.u3.ip = nla_get_be32(tb[CTA_IP_V4_SRC]);
 326         t->dst.u3.ip = nla_get_be32(tb[CTA_IP_V4_DST]);
 327
 328         return 0;
 329 }
 330
 331 static int ipv4_nlattr_tuple_size(void)
 332 {
 333         return nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1);
 334 }
 335 #endif
 336
 337 static struct nf_sockopt_ops so_getorigdst = {
 338         .pf             = PF_INET,
 339         .get_optmin     = SO_ORIGINAL_DST,
 340         .get_optmax     = SO_ORIGINAL_DST+1,
 341         .get            = &getorigdst,
 342         .owner          = THIS_MODULE,
 343 };
 344
 345 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 346         .l3proto         = PF_INET,
 347         .name            = "ipv4",
 348         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 349         .invert_tuple    = ipv4_invert_tuple,
 350         .print_tuple     = ipv4_print_tuple,
 351         .get_l4proto     = ipv4_get_l4proto,
 352 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 353         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 354         .nlattr_tuple_size = ipv4_nlattr_tuple_size,
 355         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 356         .nla_policy      = ipv4_nla_policy,
 357 #endif
 358 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 359         .ctl_table_path  = nf_net_ipv4_netfilter_sysctl_path,
 360         .ctl_table       = ip_ct_sysctl_table,
 361 #endif
 362         .me              = THIS_MODULE,
 363 };
 364
 365 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 366                   &nf_conntrack_htable_size, 0600);
 367
 368 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 369 MODULE_ALIAS("ip_conntrack");
 370 MODULE_LICENSE("GPL");
 371
 372 static int __init nf_conntrack_l3proto_ipv4_init(void)
 373 {
 374         int ret = 0;
 375
 376         need_conntrack();
 377         nf_defrag_ipv4_enable();
 378
 379         ret = nf_register_sockopt(&so_getorigdst);
 380         if (ret < 0) {
 381                 printk(KERN_ERR "Unable to register netfilter socket option\n");
 382                 return ret;
 383         }
 384
 385         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_tcp4);
 386         if (ret < 0) {
 387                 pr_err("nf_conntrack_ipv4: can't register tcp.\n");
 388                 goto cleanup_sockopt;
 389         }
 390
 391         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_udp4);
 392         if (ret < 0) {
 393                 pr_err("nf_conntrack_ipv4: can't register udp.\n");
 394                 goto cleanup_tcp;
 395         }
 396
 397         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_icmp);
 398         if (ret < 0) {
 399                 pr_err("nf_conntrack_ipv4: can't register icmp.\n");
 400                 goto cleanup_udp;
 401         }
 402
 403         ret = nf_conntrack_l3proto_register(&nf_conntrack_l3proto_ipv4);
 404         if (ret < 0) {
 405                 pr_err("nf_conntrack_ipv4: can't register ipv4\n");
 406                 goto cleanup_icmp;
 407         }
 408
 409         ret = nf_register_hooks(ipv4_conntrack_ops,
 410                                 ARRAY_SIZE(ipv4_conntrack_ops));
 411         if (ret < 0) {
 412                 pr_err("nf_conntrack_ipv4: can't register hooks.\n");
 413                 goto cleanup_ipv4;
 414         }
 415 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 416         ret = nf_conntrack_ipv4_compat_init();
 417         if (ret < 0)
 418                 goto cleanup_hooks;
 419 #endif
 420         return ret;
 421 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 422  cleanup_hooks:
 423         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 424 #endif
 425  cleanup_ipv4:
 426         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 427  cleanup_icmp:
 428         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 429  cleanup_udp:
 430         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 431  cleanup_tcp:
 432         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 433  cleanup_sockopt:
 434         nf_unregister_sockopt(&so_getorigdst);
 435         return ret;
 436 }
 437
 438 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 439 {
 440         synchronize_net();
 441 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 442         nf_conntrack_ipv4_compat_fini();
 443 #endif
 444         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 445         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 446         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 447         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 448         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 449         nf_unregister_sockopt(&so_getorigdst);
 450 }
 451
 452 module_init(nf_conntrack_l3proto_ipv4_init);
 453 module_exit(nf_conntrack_l3proto_ipv4_fini);
 454
 455 void need_ipv4_conntrack(void)
 456 {
 457         return;
 458 }
 459 EXPORT_SYMBOL_GPL(need_ipv4_conntrack);