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 bool ipv4_pkt_to_tuple(const struct sk_buff *skb, unsigned int nhoff,
  35                               struct nf_conntrack_tuple *tuple)
  36 {
  37         const __be32 *ap;
  38         __be32 _addrs[2];
  39         ap = skb_header_pointer(skb, nhoff + offsetof(struct iphdr, saddr),
  40                                 sizeof(u_int32_t) * 2, _addrs);
  41         if (ap == NULL)
  42                 return false;
  43
  44         tuple->src.u3.ip = ap[0];
  45         tuple->dst.u3.ip = ap[1];
  46
  47         return true;
  48 }
  49
  50 static bool ipv4_invert_tuple(struct nf_conntrack_tuple *tuple,
  51                               const struct nf_conntrack_tuple *orig)
  52 {
  53         tuple->src.u3.ip = orig->dst.u3.ip;
  54         tuple->dst.u3.ip = orig->src.u3.ip;
  55
  56         return true;
  57 }
  58
  59 static void ipv4_print_tuple(struct seq_file *s,
  60                             const struct nf_conntrack_tuple *tuple)
  61 {
  62         seq_printf(s, "src=%pI4 dst=%pI4 ",
  63                    &tuple->src.u3.ip, &tuple->dst.u3.ip);
  64 }
  65
  66 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
  67                             unsigned int *dataoff, u_int8_t *protonum)
  68 {
  69         const struct iphdr *iph;
  70         struct iphdr _iph;
  71
  72         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
  73         if (iph == NULL)
  74                 return -NF_ACCEPT;
  75
  76         /* Conntrack defragments packets, we might still see fragments
  77          * inside ICMP packets though. */
  78         if (iph->frag_off & htons(IP_OFFSET))
  79                 return -NF_ACCEPT;
  80
  81         *dataoff = nhoff + (iph->ihl << 2);
  82         *protonum = iph->protocol;
  83
  84         /* Check bogus IP headers */
  85         if (*dataoff > skb->len) {
  86                 pr_debug("nf_conntrack_ipv4: bogus IPv4 packet: "
  87                          "nhoff %u, ihl %u, skblen %u\n",
  88                          nhoff, iph->ihl << 2, skb->len);
  89                 return -NF_ACCEPT;
  90         }
  91
  92         return NF_ACCEPT;
  93 }
  94
  95 static unsigned int ipv4_helper(void *priv,
  96                                 struct sk_buff *skb,
  97                                 const struct nf_hook_state *state)
  98 {
  99         struct nf_conn *ct;
 100         enum ip_conntrack_info ctinfo;
 101         const struct nf_conn_help *help;
 102         const struct nf_conntrack_helper *helper;
 103
 104         /* This is where we call the helper: as the packet goes out. */
 105         ct = nf_ct_get(skb, &ctinfo);
 106         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 107                 return NF_ACCEPT;
 108
 109         help = nfct_help(ct);
 110         if (!help)
 111                 return NF_ACCEPT;
 112
 113         /* rcu_read_lock()ed by nf_hook_slow */
 114         helper = rcu_dereference(help->helper);
 115         if (!helper)
 116                 return NF_ACCEPT;
 117
 118         return helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
 119                             ct, ctinfo);
 120 }
 121
 122 static unsigned int ipv4_confirm(void *priv,
 123                                  struct sk_buff *skb,
 124                                  const struct nf_hook_state *state)
 125 {
 126         struct nf_conn *ct;
 127         enum ip_conntrack_info ctinfo;
 128
 129         ct = nf_ct_get(skb, &ctinfo);
 130         if (!ct || ctinfo == IP_CT_RELATED_REPLY)
 131                 goto out;
 132
 133         /* adjust seqs for loopback traffic only in outgoing direction */
 134         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
 135             !nf_is_loopback_packet(skb)) {
 136                 if (!nf_ct_seq_adjust(skb, ct, ctinfo, ip_hdrlen(skb))) {
 137                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
 138                         return NF_DROP;
 139                 }
 140         }
 141 out:
 142         /* We've seen it coming out the other side: confirm it */
 143         return nf_conntrack_confirm(skb);
 144 }
 145
 146 static unsigned int ipv4_conntrack_in(void *priv,
 147                                       struct sk_buff *skb,
 148                                       const struct nf_hook_state *state)
 149 {
 150         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 151 }
 152
 153 static unsigned int ipv4_conntrack_local(void *priv,
 154                                          struct sk_buff *skb,
 155                                          const struct nf_hook_state *state)
 156 {
 157         /* root is playing with raw sockets. */
 158         if (skb->len < sizeof(struct iphdr) ||
 159             ip_hdrlen(skb) < sizeof(struct iphdr))
 160                 return NF_ACCEPT;
 161
 162         if (ip_is_fragment(ip_hdr(skb))) /* IP_NODEFRAG setsockopt set */
 163                 return NF_ACCEPT;
 164
 165         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 166 }
 167
 168 /* Connection tracking may drop packets, but never alters them, so
 169    make it the first hook. */
 170 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 171         {
 172                 .hook           = ipv4_conntrack_in,
 173                 .pf             = NFPROTO_IPV4,
 174                 .hooknum        = NF_INET_PRE_ROUTING,
 175                 .priority       = NF_IP_PRI_CONNTRACK,
 176         },
 177         {
 178                 .hook           = ipv4_conntrack_local,
 179                 .pf             = NFPROTO_IPV4,
 180                 .hooknum        = NF_INET_LOCAL_OUT,
 181                 .priority       = NF_IP_PRI_CONNTRACK,
 182         },
 183         {
 184                 .hook           = ipv4_helper,
 185                 .pf             = NFPROTO_IPV4,
 186                 .hooknum        = NF_INET_POST_ROUTING,
 187                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 188         },
 189         {
 190                 .hook           = ipv4_confirm,
 191                 .pf             = NFPROTO_IPV4,
 192                 .hooknum        = NF_INET_POST_ROUTING,
 193                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 194         },
 195         {
 196                 .hook           = ipv4_helper,
 197                 .pf             = NFPROTO_IPV4,
 198                 .hooknum        = NF_INET_LOCAL_IN,
 199                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 200         },
 201         {
 202                 .hook           = ipv4_confirm,
 203                 .pf             = NFPROTO_IPV4,
 204                 .hooknum        = NF_INET_LOCAL_IN,
 205                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 206         },
 207 };
 208
 209 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 210 static int log_invalid_proto_min = 0;
 211 static int log_invalid_proto_max = 255;
 212
 213 static struct ctl_table ip_ct_sysctl_table[] = {
 214         {
 215                 .procname       = "ip_conntrack_max",
 216                 .maxlen         = sizeof(int),
 217                 .mode           = 0644,
 218                 .proc_handler   = proc_dointvec,
 219         },
 220         {
 221                 .procname       = "ip_conntrack_count",
 222                 .maxlen         = sizeof(int),
 223                 .mode           = 0444,
 224                 .proc_handler   = proc_dointvec,
 225         },
 226         {
 227                 .procname       = "ip_conntrack_buckets",
 228                 .maxlen         = sizeof(unsigned int),
 229                 .mode           = 0444,
 230                 .proc_handler   = proc_dointvec,
 231         },
 232         {
 233                 .procname       = "ip_conntrack_checksum",
 234                 .maxlen         = sizeof(int),
 235                 .mode           = 0644,
 236                 .proc_handler   = proc_dointvec,
 237         },
 238         {
 239                 .procname       = "ip_conntrack_log_invalid",
 240                 .maxlen         = sizeof(unsigned int),
 241                 .mode           = 0644,
 242                 .proc_handler   = proc_dointvec_minmax,
 243                 .extra1         = &log_invalid_proto_min,
 244                 .extra2         = &log_invalid_proto_max,
 245         },
 246         { }
 247 };
 248 #endif /* CONFIG_SYSCTL && CONFIG_NF_CONNTRACK_PROC_COMPAT */
 249
 250 /* Fast function for those who don't want to parse /proc (and I don't
 251    blame them). */
 252 /* Reversing the socket's dst/src point of view gives us the reply
 253    mapping. */
 254 static int
 255 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 256 {
 257         const struct inet_sock *inet = inet_sk(sk);
 258         const struct nf_conntrack_tuple_hash *h;
 259         struct nf_conntrack_tuple tuple;
 260
 261         memset(&tuple, 0, sizeof(tuple));
 262
 263         lock_sock(sk);
 264         tuple.src.u3.ip = inet->inet_rcv_saddr;
 265         tuple.src.u.tcp.port = inet->inet_sport;
 266         tuple.dst.u3.ip = inet->inet_daddr;
 267         tuple.dst.u.tcp.port = inet->inet_dport;
 268         tuple.src.l3num = PF_INET;
 269         tuple.dst.protonum = sk->sk_protocol;
 270         release_sock(sk);
 271
 272         /* We only do TCP and SCTP at the moment: is there a better way? */
 273         if (tuple.dst.protonum != IPPROTO_TCP &&
 274             tuple.dst.protonum != IPPROTO_SCTP) {
 275                 pr_debug("SO_ORIGINAL_DST: Not a TCP/SCTP socket\n");
 276                 return -ENOPROTOOPT;
 277         }
 278
 279         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 280                 pr_debug("SO_ORIGINAL_DST: len %d not %Zu\n",
 281                          *len, sizeof(struct sockaddr_in));
 282                 return -EINVAL;
 283         }
 284
 285         h = nf_conntrack_find_get(sock_net(sk), &nf_ct_zone_dflt, &tuple);
 286         if (h) {
 287                 struct sockaddr_in sin;
 288                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 289
 290                 sin.sin_family = AF_INET;
 291                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 292                         .tuple.dst.u.tcp.port;
 293                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 294                         .tuple.dst.u3.ip;
 295                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 296
 297                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 298                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 299                 nf_ct_put(ct);
 300                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 301                         return -EFAULT;
 302                 else
 303                         return 0;
 304         }
 305         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 306                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 307                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 308         return -ENOENT;
 309 }
 310
 311 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 312
 313 #include <linux/netfilter/nfnetlink.h>
 314 #include <linux/netfilter/nfnetlink_conntrack.h>
 315
 316 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 317                                 const struct nf_conntrack_tuple *tuple)
 318 {
 319         if (nla_put_in_addr(skb, CTA_IP_V4_SRC, tuple->src.u3.ip) ||
 320             nla_put_in_addr(skb, CTA_IP_V4_DST, tuple->dst.u3.ip))
 321                 goto nla_put_failure;
 322         return 0;
 323
 324 nla_put_failure:
 325         return -1;
 326 }
 327
 328 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 329         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 330         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 331 };
 332
 333 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 334                                 struct nf_conntrack_tuple *t)
 335 {
 336         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 337                 return -EINVAL;
 338
 339         t->src.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_SRC]);
 340         t->dst.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_DST]);
 341
 342         return 0;
 343 }
 344
 345 static int ipv4_nlattr_tuple_size(void)
 346 {
 347         return nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1);
 348 }
 349 #endif
 350
 351 static struct nf_sockopt_ops so_getorigdst = {
 352         .pf             = PF_INET,
 353         .get_optmin     = SO_ORIGINAL_DST,
 354         .get_optmax     = SO_ORIGINAL_DST+1,
 355         .get            = getorigdst,
 356         .owner          = THIS_MODULE,
 357 };
 358
 359 static int ipv4_init_net(struct net *net)
 360 {
 361 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 362         struct nf_ip_net *in = &net->ct.nf_ct_proto;
 363         in->ctl_table = kmemdup(ip_ct_sysctl_table,
 364                                 sizeof(ip_ct_sysctl_table),
 365                                 GFP_KERNEL);
 366         if (!in->ctl_table)
 367                 return -ENOMEM;
 368
 369         in->ctl_table[0].data = &nf_conntrack_max;
 370         in->ctl_table[1].data = &net->ct.count;
 371         in->ctl_table[2].data = &net->ct.htable_size;
 372         in->ctl_table[3].data = &net->ct.sysctl_checksum;
 373         in->ctl_table[4].data = &net->ct.sysctl_log_invalid;
 374 #endif
 375         return 0;
 376 }
 377
 378 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 379         .l3proto         = PF_INET,
 380         .name            = "ipv4",
 381         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 382         .invert_tuple    = ipv4_invert_tuple,
 383         .print_tuple     = ipv4_print_tuple,
 384         .get_l4proto     = ipv4_get_l4proto,
 385 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 386         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 387         .nlattr_tuple_size = ipv4_nlattr_tuple_size,
 388         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 389         .nla_policy      = ipv4_nla_policy,
 390 #endif
 391 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 392         .ctl_table_path  = "net/ipv4/netfilter",
 393 #endif
 394         .init_net        = ipv4_init_net,
 395         .me              = THIS_MODULE,
 396 };
 397
 398 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 399                   &nf_conntrack_htable_size, 0600);
 400
 401 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 402 MODULE_ALIAS("ip_conntrack");
 403 MODULE_LICENSE("GPL");
 404
 405 static int ipv4_net_init(struct net *net)
 406 {
 407         int ret = 0;
 408
 409         ret = nf_ct_l4proto_pernet_register(net, &nf_conntrack_l4proto_tcp4);
 410         if (ret < 0) {
 411                 pr_err("nf_conntrack_tcp4: pernet registration failed\n");
 412                 goto out_tcp;
 413         }
 414         ret = nf_ct_l4proto_pernet_register(net, &nf_conntrack_l4proto_udp4);
 415         if (ret < 0) {
 416                 pr_err("nf_conntrack_udp4: pernet registration failed\n");
 417                 goto out_udp;
 418         }
 419         ret = nf_ct_l4proto_pernet_register(net, &nf_conntrack_l4proto_icmp);
 420         if (ret < 0) {
 421                 pr_err("nf_conntrack_icmp4: pernet registration failed\n");
 422                 goto out_icmp;
 423         }
 424         ret = nf_ct_l3proto_pernet_register(net, &nf_conntrack_l3proto_ipv4);
 425         if (ret < 0) {
 426                 pr_err("nf_conntrack_ipv4: pernet registration failed\n");
 427                 goto out_ipv4;
 428         }
 429         return 0;
 430 out_ipv4:
 431         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_icmp);
 432 out_icmp:
 433         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_udp4);
 434 out_udp:
 435         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_tcp4);
 436 out_tcp:
 437         return ret;
 438 }
 439
 440 static void ipv4_net_exit(struct net *net)
 441 {
 442         nf_ct_l3proto_pernet_unregister(net, &nf_conntrack_l3proto_ipv4);
 443         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_icmp);
 444         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_udp4);
 445         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_tcp4);
 446 }
 447
 448 static struct pernet_operations ipv4_net_ops = {
 449         .init = ipv4_net_init,
 450         .exit = ipv4_net_exit,
 451 };
 452
 453 static int __init nf_conntrack_l3proto_ipv4_init(void)
 454 {
 455         int ret = 0;
 456
 457         need_conntrack();
 458         nf_defrag_ipv4_enable();
 459
 460         ret = nf_register_sockopt(&so_getorigdst);
 461         if (ret < 0) {
 462                 printk(KERN_ERR "Unable to register netfilter socket option\n");
 463                 return ret;
 464         }
 465
 466         ret = register_pernet_subsys(&ipv4_net_ops);
 467         if (ret < 0) {
 468                 pr_err("nf_conntrack_ipv4: can't register pernet ops\n");
 469                 goto cleanup_sockopt;
 470         }
 471
 472         ret = nf_register_hooks(ipv4_conntrack_ops,
 473                                 ARRAY_SIZE(ipv4_conntrack_ops));
 474         if (ret < 0) {
 475                 pr_err("nf_conntrack_ipv4: can't register hooks.\n");
 476                 goto cleanup_pernet;
 477         }
 478
 479         ret = nf_ct_l4proto_register(&nf_conntrack_l4proto_tcp4);
 480         if (ret < 0) {
 481                 pr_err("nf_conntrack_ipv4: can't register tcp4 proto.\n");
 482                 goto cleanup_hooks;
 483         }
 484
 485         ret = nf_ct_l4proto_register(&nf_conntrack_l4proto_udp4);
 486         if (ret < 0) {
 487                 pr_err("nf_conntrack_ipv4: can't register udp4 proto.\n");
 488                 goto cleanup_tcp4;
 489         }
 490
 491         ret = nf_ct_l4proto_register(&nf_conntrack_l4proto_icmp);
 492         if (ret < 0) {
 493                 pr_err("nf_conntrack_ipv4: can't register icmpv4 proto.\n");
 494                 goto cleanup_udp4;
 495         }
 496
 497         ret = nf_ct_l3proto_register(&nf_conntrack_l3proto_ipv4);
 498         if (ret < 0) {
 499                 pr_err("nf_conntrack_ipv4: can't register ipv4 proto.\n");
 500                 goto cleanup_icmpv4;
 501         }
 502
 503 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 504         ret = nf_conntrack_ipv4_compat_init();
 505         if (ret < 0)
 506                 goto cleanup_proto;
 507 #endif
 508         return ret;
 509 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 510  cleanup_proto:
 511         nf_ct_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 512 #endif
 513  cleanup_icmpv4:
 514         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 515  cleanup_udp4:
 516         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 517  cleanup_tcp4:
 518         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 519  cleanup_hooks:
 520         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 521  cleanup_pernet:
 522         unregister_pernet_subsys(&ipv4_net_ops);
 523  cleanup_sockopt:
 524         nf_unregister_sockopt(&so_getorigdst);
 525         return ret;
 526 }
 527
 528 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 529 {
 530         synchronize_net();
 531 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 532         nf_conntrack_ipv4_compat_fini();
 533 #endif
 534         nf_ct_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 535         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 536         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 537         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 538         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 539         unregister_pernet_subsys(&ipv4_net_ops);
 540         nf_unregister_sockopt(&so_getorigdst);
 541 }
 542
 543 module_init(nf_conntrack_l3proto_ipv4_init);
 544 module_exit(nf_conntrack_l3proto_ipv4_fini);