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
 169         if (ip_is_fragment(ip_hdr(skb))) /* IP_NODEFRAG setsockopt set */
 170                 return NF_ACCEPT;
 171
 172         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 173 }
 174
 175 /* Connection tracking may drop packets, but never alters them, so
 176    make it the first hook. */
 177 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 178         {
 179                 .hook           = ipv4_conntrack_in,
 180                 .pf             = NFPROTO_IPV4,
 181                 .hooknum        = NF_INET_PRE_ROUTING,
 182                 .priority       = NF_IP_PRI_CONNTRACK,
 183         },
 184         {
 185                 .hook           = ipv4_conntrack_local,
 186                 .pf             = NFPROTO_IPV4,
 187                 .hooknum        = NF_INET_LOCAL_OUT,
 188                 .priority       = NF_IP_PRI_CONNTRACK,
 189         },
 190         {
 191                 .hook           = ipv4_helper,
 192                 .pf             = NFPROTO_IPV4,
 193                 .hooknum        = NF_INET_POST_ROUTING,
 194                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 195         },
 196         {
 197                 .hook           = ipv4_confirm,
 198                 .pf             = NFPROTO_IPV4,
 199                 .hooknum        = NF_INET_POST_ROUTING,
 200                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 201         },
 202         {
 203                 .hook           = ipv4_helper,
 204                 .pf             = NFPROTO_IPV4,
 205                 .hooknum        = NF_INET_LOCAL_IN,
 206                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 207         },
 208         {
 209                 .hook           = ipv4_confirm,
 210                 .pf             = NFPROTO_IPV4,
 211                 .hooknum        = NF_INET_LOCAL_IN,
 212                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 213         },
 214 };
 215
 216 /* Fast function for those who don't want to parse /proc (and I don't
 217    blame them). */
 218 /* Reversing the socket's dst/src point of view gives us the reply
 219    mapping. */
 220 static int
 221 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 222 {
 223         const struct inet_sock *inet = inet_sk(sk);
 224         const struct nf_conntrack_tuple_hash *h;
 225         struct nf_conntrack_tuple tuple;
 226
 227         memset(&tuple, 0, sizeof(tuple));
 228         tuple.src.u3.ip = inet->inet_rcv_saddr;
 229         tuple.src.u.tcp.port = inet->inet_sport;
 230         tuple.dst.u3.ip = inet->inet_daddr;
 231         tuple.dst.u.tcp.port = inet->inet_dport;
 232         tuple.src.l3num = PF_INET;
 233         tuple.dst.protonum = sk->sk_protocol;
 234
 235         /* We only do TCP and SCTP at the moment: is there a better way? */
 236         if (sk->sk_protocol != IPPROTO_TCP && sk->sk_protocol != IPPROTO_SCTP) {
 237                 pr_debug("SO_ORIGINAL_DST: Not a TCP/SCTP socket\n");
 238                 return -ENOPROTOOPT;
 239         }
 240
 241         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 242                 pr_debug("SO_ORIGINAL_DST: len %d not %zu\n",
 243                          *len, sizeof(struct sockaddr_in));
 244                 return -EINVAL;
 245         }
 246
 247         h = nf_conntrack_find_get(sock_net(sk), &nf_ct_zone_dflt, &tuple);
 248         if (h) {
 249                 struct sockaddr_in sin;
 250                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 251
 252                 sin.sin_family = AF_INET;
 253                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 254                         .tuple.dst.u.tcp.port;
 255                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 256                         .tuple.dst.u3.ip;
 257                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 258
 259                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 260                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 261                 nf_ct_put(ct);
 262                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 263                         return -EFAULT;
 264                 else
 265                         return 0;
 266         }
 267         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 268                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 269                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 270         return -ENOENT;
 271 }
 272
 273 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 274
 275 #include <linux/netfilter/nfnetlink.h>
 276 #include <linux/netfilter/nfnetlink_conntrack.h>
 277
 278 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 279                                 const struct nf_conntrack_tuple *tuple)
 280 {
 281         if (nla_put_in_addr(skb, CTA_IP_V4_SRC, tuple->src.u3.ip) ||
 282             nla_put_in_addr(skb, CTA_IP_V4_DST, tuple->dst.u3.ip))
 283                 goto nla_put_failure;
 284         return 0;
 285
 286 nla_put_failure:
 287         return -1;
 288 }
 289
 290 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 291         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 292         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 293 };
 294
 295 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 296                                 struct nf_conntrack_tuple *t)
 297 {
 298         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 299                 return -EINVAL;
 300
 301         t->src.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_SRC]);
 302         t->dst.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_DST]);
 303
 304         return 0;
 305 }
 306
 307 static int ipv4_nlattr_tuple_size(void)
 308 {
 309         return nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1);
 310 }
 311 #endif
 312
 313 static struct nf_sockopt_ops so_getorigdst = {
 314         .pf             = PF_INET,
 315         .get_optmin     = SO_ORIGINAL_DST,
 316         .get_optmax     = SO_ORIGINAL_DST+1,
 317         .get            = getorigdst,
 318         .owner          = THIS_MODULE,
 319 };
 320
 321 static int ipv4_hooks_register(struct net *net)
 322 {
 323         struct conntrack4_net *cnet = net_generic(net, conntrack4_net_id);
 324         int err = 0;
 325
 326         mutex_lock(&register_ipv4_hooks);
 327
 328         cnet->users++;
 329         if (cnet->users > 1)
 330                 goto out_unlock;
 331
 332         err = nf_defrag_ipv4_enable(net);
 333         if (err) {
 334                 cnet->users = 0;
 335                 goto out_unlock;
 336         }
 337
 338         err = nf_register_net_hooks(net, ipv4_conntrack_ops,
 339                                     ARRAY_SIZE(ipv4_conntrack_ops));
 340
 341         if (err)
 342                 cnet->users = 0;
 343  out_unlock:
 344         mutex_unlock(&register_ipv4_hooks);
 345         return err;
 346 }
 347
 348 static void ipv4_hooks_unregister(struct net *net)
 349 {
 350         struct conntrack4_net *cnet = net_generic(net, conntrack4_net_id);
 351
 352         mutex_lock(&register_ipv4_hooks);
 353         if (cnet->users && (--cnet->users == 0))
 354                 nf_unregister_net_hooks(net, ipv4_conntrack_ops,
 355                                         ARRAY_SIZE(ipv4_conntrack_ops));
 356         mutex_unlock(&register_ipv4_hooks);
 357 }
 358
 359 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 360         .l3proto         = PF_INET,
 361         .name            = "ipv4",
 362         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 363         .invert_tuple    = ipv4_invert_tuple,
 364         .print_tuple     = ipv4_print_tuple,
 365         .get_l4proto     = ipv4_get_l4proto,
 366 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 367         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 368         .nlattr_tuple_size = ipv4_nlattr_tuple_size,
 369         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 370         .nla_policy      = ipv4_nla_policy,
 371 #endif
 372         .net_ns_get      = ipv4_hooks_register,
 373         .net_ns_put      = ipv4_hooks_unregister,
 374         .me              = THIS_MODULE,
 375 };
 376
 377 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 378                   &nf_conntrack_htable_size, 0600);
 379
 380 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 381 MODULE_ALIAS("ip_conntrack");
 382 MODULE_LICENSE("GPL");
 383
 384 static struct nf_conntrack_l4proto *builtin_l4proto4[] = {
 385         &nf_conntrack_l4proto_tcp4,
 386         &nf_conntrack_l4proto_udp4,
 387         &nf_conntrack_l4proto_icmp,
 388 #ifdef CONFIG_NF_CT_PROTO_DCCP
 389         &nf_conntrack_l4proto_dccp4,
 390 #endif
 391 #ifdef CONFIG_NF_CT_PROTO_SCTP
 392         &nf_conntrack_l4proto_sctp4,
 393 #endif
 394 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
 395         &nf_conntrack_l4proto_udplite4,
 396 #endif
 397 };
 398
 399 static int ipv4_net_init(struct net *net)
 400 {
 401         int ret = 0;
 402
 403         ret = nf_ct_l4proto_pernet_register(net, builtin_l4proto4,
 404                                             ARRAY_SIZE(builtin_l4proto4));
 405         if (ret < 0)
 406                 return ret;
 407         ret = nf_ct_l3proto_pernet_register(net, &nf_conntrack_l3proto_ipv4);
 408         if (ret < 0) {
 409                 pr_err("nf_conntrack_ipv4: pernet registration failed\n");
 410                 nf_ct_l4proto_pernet_unregister(net, builtin_l4proto4,
 411                                                 ARRAY_SIZE(builtin_l4proto4));
 412         }
 413         return ret;
 414 }
 415
 416 static void ipv4_net_exit(struct net *net)
 417 {
 418         nf_ct_l3proto_pernet_unregister(net, &nf_conntrack_l3proto_ipv4);
 419         nf_ct_l4proto_pernet_unregister(net, builtin_l4proto4,
 420                                         ARRAY_SIZE(builtin_l4proto4));
 421 }
 422
 423 static struct pernet_operations ipv4_net_ops = {
 424         .init = ipv4_net_init,
 425         .exit = ipv4_net_exit,
 426         .id = &conntrack4_net_id,
 427         .size = sizeof(struct conntrack4_net),
 428 };
 429
 430 static int __init nf_conntrack_l3proto_ipv4_init(void)
 431 {
 432         int ret = 0;
 433
 434         need_conntrack();
 435
 436         ret = nf_register_sockopt(&so_getorigdst);
 437         if (ret < 0) {
 438                 pr_err("Unable to register netfilter socket option\n");
 439                 return ret;
 440         }
 441
 442         ret = register_pernet_subsys(&ipv4_net_ops);
 443         if (ret < 0) {
 444                 pr_err("nf_conntrack_ipv4: can't register pernet ops\n");
 445                 goto cleanup_sockopt;
 446         }
 447
 448         ret = nf_ct_l4proto_register(builtin_l4proto4,
 449                                      ARRAY_SIZE(builtin_l4proto4));
 450         if (ret < 0)
 451                 goto cleanup_pernet;
 452
 453         ret = nf_ct_l3proto_register(&nf_conntrack_l3proto_ipv4);
 454         if (ret < 0) {
 455                 pr_err("nf_conntrack_ipv4: can't register ipv4 proto.\n");
 456                 goto cleanup_l4proto;
 457         }
 458
 459         return ret;
 460 cleanup_l4proto:
 461         nf_ct_l4proto_unregister(builtin_l4proto4,
 462                                  ARRAY_SIZE(builtin_l4proto4));
 463  cleanup_pernet:
 464         unregister_pernet_subsys(&ipv4_net_ops);
 465  cleanup_sockopt:
 466         nf_unregister_sockopt(&so_getorigdst);
 467         return ret;
 468 }
 469
 470 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 471 {
 472         synchronize_net();
 473         nf_ct_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 474         nf_ct_l4proto_unregister(builtin_l4proto4,
 475                                  ARRAY_SIZE(builtin_l4proto4));
 476         unregister_pernet_subsys(&ipv4_net_ops);
 477         nf_unregister_sockopt(&so_getorigdst);
 478 }
 479
 480 module_init(nf_conntrack_l3proto_ipv4_init);
 481 module_exit(nf_conntrack_l3proto_ipv4_fini);