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         return nf_conntrack_in(state->net, PF_INET, state->hook, skb);
 162 }
 163
 164 /* Connection tracking may drop packets, but never alters them, so
 165    make it the first hook. */
 166 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
 167         {
 168                 .hook           = ipv4_conntrack_in,
 169                 .pf             = NFPROTO_IPV4,
 170                 .hooknum        = NF_INET_PRE_ROUTING,
 171                 .priority       = NF_IP_PRI_CONNTRACK,
 172         },
 173         {
 174                 .hook           = ipv4_conntrack_local,
 175                 .pf             = NFPROTO_IPV4,
 176                 .hooknum        = NF_INET_LOCAL_OUT,
 177                 .priority       = NF_IP_PRI_CONNTRACK,
 178         },
 179         {
 180                 .hook           = ipv4_helper,
 181                 .pf             = NFPROTO_IPV4,
 182                 .hooknum        = NF_INET_POST_ROUTING,
 183                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 184         },
 185         {
 186                 .hook           = ipv4_confirm,
 187                 .pf             = NFPROTO_IPV4,
 188                 .hooknum        = NF_INET_POST_ROUTING,
 189                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 190         },
 191         {
 192                 .hook           = ipv4_helper,
 193                 .pf             = NFPROTO_IPV4,
 194                 .hooknum        = NF_INET_LOCAL_IN,
 195                 .priority       = NF_IP_PRI_CONNTRACK_HELPER,
 196         },
 197         {
 198                 .hook           = ipv4_confirm,
 199                 .pf             = NFPROTO_IPV4,
 200                 .hooknum        = NF_INET_LOCAL_IN,
 201                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
 202         },
 203 };
 204
 205 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 206 static int log_invalid_proto_min = 0;
 207 static int log_invalid_proto_max = 255;
 208
 209 static struct ctl_table ip_ct_sysctl_table[] = {
 210         {
 211                 .procname       = "ip_conntrack_max",
 212                 .maxlen         = sizeof(int),
 213                 .mode           = 0644,
 214                 .proc_handler   = proc_dointvec,
 215         },
 216         {
 217                 .procname       = "ip_conntrack_count",
 218                 .maxlen         = sizeof(int),
 219                 .mode           = 0444,
 220                 .proc_handler   = proc_dointvec,
 221         },
 222         {
 223                 .procname       = "ip_conntrack_buckets",
 224                 .maxlen         = sizeof(unsigned int),
 225                 .mode           = 0444,
 226                 .proc_handler   = proc_dointvec,
 227         },
 228         {
 229                 .procname       = "ip_conntrack_checksum",
 230                 .maxlen         = sizeof(int),
 231                 .mode           = 0644,
 232                 .proc_handler   = proc_dointvec,
 233         },
 234         {
 235                 .procname       = "ip_conntrack_log_invalid",
 236                 .maxlen         = sizeof(unsigned int),
 237                 .mode           = 0644,
 238                 .proc_handler   = proc_dointvec_minmax,
 239                 .extra1         = &log_invalid_proto_min,
 240                 .extra2         = &log_invalid_proto_max,
 241         },
 242         { }
 243 };
 244 #endif /* CONFIG_SYSCTL && CONFIG_NF_CONNTRACK_PROC_COMPAT */
 245
 246 /* Fast function for those who don't want to parse /proc (and I don't
 247    blame them). */
 248 /* Reversing the socket's dst/src point of view gives us the reply
 249    mapping. */
 250 static int
 251 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
 252 {
 253         const struct inet_sock *inet = inet_sk(sk);
 254         const struct nf_conntrack_tuple_hash *h;
 255         struct nf_conntrack_tuple tuple;
 256
 257         memset(&tuple, 0, sizeof(tuple));
 258         tuple.src.u3.ip = inet->inet_rcv_saddr;
 259         tuple.src.u.tcp.port = inet->inet_sport;
 260         tuple.dst.u3.ip = inet->inet_daddr;
 261         tuple.dst.u.tcp.port = inet->inet_dport;
 262         tuple.src.l3num = PF_INET;
 263         tuple.dst.protonum = sk->sk_protocol;
 264
 265         /* We only do TCP and SCTP at the moment: is there a better way? */
 266         if (sk->sk_protocol != IPPROTO_TCP && sk->sk_protocol != IPPROTO_SCTP) {
 267                 pr_debug("SO_ORIGINAL_DST: Not a TCP/SCTP socket\n");
 268                 return -ENOPROTOOPT;
 269         }
 270
 271         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
 272                 pr_debug("SO_ORIGINAL_DST: len %d not %Zu\n",
 273                          *len, sizeof(struct sockaddr_in));
 274                 return -EINVAL;
 275         }
 276
 277         h = nf_conntrack_find_get(sock_net(sk), &nf_ct_zone_dflt, &tuple);
 278         if (h) {
 279                 struct sockaddr_in sin;
 280                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
 281
 282                 sin.sin_family = AF_INET;
 283                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 284                         .tuple.dst.u.tcp.port;
 285                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
 286                         .tuple.dst.u3.ip;
 287                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
 288
 289                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
 290                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
 291                 nf_ct_put(ct);
 292                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
 293                         return -EFAULT;
 294                 else
 295                         return 0;
 296         }
 297         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
 298                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
 299                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
 300         return -ENOENT;
 301 }
 302
 303 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 304
 305 #include <linux/netfilter/nfnetlink.h>
 306 #include <linux/netfilter/nfnetlink_conntrack.h>
 307
 308 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
 309                                 const struct nf_conntrack_tuple *tuple)
 310 {
 311         if (nla_put_in_addr(skb, CTA_IP_V4_SRC, tuple->src.u3.ip) ||
 312             nla_put_in_addr(skb, CTA_IP_V4_DST, tuple->dst.u3.ip))
 313                 goto nla_put_failure;
 314         return 0;
 315
 316 nla_put_failure:
 317         return -1;
 318 }
 319
 320 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
 321         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
 322         [CTA_IP_V4_DST] = { .type = NLA_U32 },
 323 };
 324
 325 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
 326                                 struct nf_conntrack_tuple *t)
 327 {
 328         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
 329                 return -EINVAL;
 330
 331         t->src.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_SRC]);
 332         t->dst.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_DST]);
 333
 334         return 0;
 335 }
 336
 337 static int ipv4_nlattr_tuple_size(void)
 338 {
 339         return nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1);
 340 }
 341 #endif
 342
 343 static struct nf_sockopt_ops so_getorigdst = {
 344         .pf             = PF_INET,
 345         .get_optmin     = SO_ORIGINAL_DST,
 346         .get_optmax     = SO_ORIGINAL_DST+1,
 347         .get            = getorigdst,
 348         .owner          = THIS_MODULE,
 349 };
 350
 351 static int ipv4_init_net(struct net *net)
 352 {
 353 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 354         struct nf_ip_net *in = &net->ct.nf_ct_proto;
 355         in->ctl_table = kmemdup(ip_ct_sysctl_table,
 356                                 sizeof(ip_ct_sysctl_table),
 357                                 GFP_KERNEL);
 358         if (!in->ctl_table)
 359                 return -ENOMEM;
 360
 361         in->ctl_table[0].data = &nf_conntrack_max;
 362         in->ctl_table[1].data = &net->ct.count;
 363         in->ctl_table[2].data = &net->ct.htable_size;
 364         in->ctl_table[3].data = &net->ct.sysctl_checksum;
 365         in->ctl_table[4].data = &net->ct.sysctl_log_invalid;
 366 #endif
 367         return 0;
 368 }
 369
 370 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
 371         .l3proto         = PF_INET,
 372         .name            = "ipv4",
 373         .pkt_to_tuple    = ipv4_pkt_to_tuple,
 374         .invert_tuple    = ipv4_invert_tuple,
 375         .print_tuple     = ipv4_print_tuple,
 376         .get_l4proto     = ipv4_get_l4proto,
 377 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
 378         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
 379         .nlattr_tuple_size = ipv4_nlattr_tuple_size,
 380         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
 381         .nla_policy      = ipv4_nla_policy,
 382 #endif
 383 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 384         .ctl_table_path  = "net/ipv4/netfilter",
 385 #endif
 386         .init_net        = ipv4_init_net,
 387         .me              = THIS_MODULE,
 388 };
 389
 390 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
 391                   &nf_conntrack_htable_size, 0600);
 392
 393 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
 394 MODULE_ALIAS("ip_conntrack");
 395 MODULE_LICENSE("GPL");
 396
 397 static int ipv4_net_init(struct net *net)
 398 {
 399         int ret = 0;
 400
 401         ret = nf_ct_l4proto_pernet_register(net, &nf_conntrack_l4proto_tcp4);
 402         if (ret < 0) {
 403                 pr_err("nf_conntrack_tcp4: pernet registration failed\n");
 404                 goto out_tcp;
 405         }
 406         ret = nf_ct_l4proto_pernet_register(net, &nf_conntrack_l4proto_udp4);
 407         if (ret < 0) {
 408                 pr_err("nf_conntrack_udp4: pernet registration failed\n");
 409                 goto out_udp;
 410         }
 411         ret = nf_ct_l4proto_pernet_register(net, &nf_conntrack_l4proto_icmp);
 412         if (ret < 0) {
 413                 pr_err("nf_conntrack_icmp4: pernet registration failed\n");
 414                 goto out_icmp;
 415         }
 416         ret = nf_ct_l3proto_pernet_register(net, &nf_conntrack_l3proto_ipv4);
 417         if (ret < 0) {
 418                 pr_err("nf_conntrack_ipv4: pernet registration failed\n");
 419                 goto out_ipv4;
 420         }
 421         return 0;
 422 out_ipv4:
 423         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_icmp);
 424 out_icmp:
 425         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_udp4);
 426 out_udp:
 427         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_tcp4);
 428 out_tcp:
 429         return ret;
 430 }
 431
 432 static void ipv4_net_exit(struct net *net)
 433 {
 434         nf_ct_l3proto_pernet_unregister(net, &nf_conntrack_l3proto_ipv4);
 435         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_icmp);
 436         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_udp4);
 437         nf_ct_l4proto_pernet_unregister(net, &nf_conntrack_l4proto_tcp4);
 438 }
 439
 440 static struct pernet_operations ipv4_net_ops = {
 441         .init = ipv4_net_init,
 442         .exit = ipv4_net_exit,
 443 };
 444
 445 static int __init nf_conntrack_l3proto_ipv4_init(void)
 446 {
 447         int ret = 0;
 448
 449         need_conntrack();
 450         nf_defrag_ipv4_enable();
 451
 452         ret = nf_register_sockopt(&so_getorigdst);
 453         if (ret < 0) {
 454                 printk(KERN_ERR "Unable to register netfilter socket option\n");
 455                 return ret;
 456         }
 457
 458         ret = register_pernet_subsys(&ipv4_net_ops);
 459         if (ret < 0) {
 460                 pr_err("nf_conntrack_ipv4: can't register pernet ops\n");
 461                 goto cleanup_sockopt;
 462         }
 463
 464         ret = nf_register_hooks(ipv4_conntrack_ops,
 465                                 ARRAY_SIZE(ipv4_conntrack_ops));
 466         if (ret < 0) {
 467                 pr_err("nf_conntrack_ipv4: can't register hooks.\n");
 468                 goto cleanup_pernet;
 469         }
 470
 471         ret = nf_ct_l4proto_register(&nf_conntrack_l4proto_tcp4);
 472         if (ret < 0) {
 473                 pr_err("nf_conntrack_ipv4: can't register tcp4 proto.\n");
 474                 goto cleanup_hooks;
 475         }
 476
 477         ret = nf_ct_l4proto_register(&nf_conntrack_l4proto_udp4);
 478         if (ret < 0) {
 479                 pr_err("nf_conntrack_ipv4: can't register udp4 proto.\n");
 480                 goto cleanup_tcp4;
 481         }
 482
 483         ret = nf_ct_l4proto_register(&nf_conntrack_l4proto_icmp);
 484         if (ret < 0) {
 485                 pr_err("nf_conntrack_ipv4: can't register icmpv4 proto.\n");
 486                 goto cleanup_udp4;
 487         }
 488
 489         ret = nf_ct_l3proto_register(&nf_conntrack_l3proto_ipv4);
 490         if (ret < 0) {
 491                 pr_err("nf_conntrack_ipv4: can't register ipv4 proto.\n");
 492                 goto cleanup_icmpv4;
 493         }
 494
 495 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 496         ret = nf_conntrack_ipv4_compat_init();
 497         if (ret < 0)
 498                 goto cleanup_proto;
 499 #endif
 500         return ret;
 501 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 502  cleanup_proto:
 503         nf_ct_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 504 #endif
 505  cleanup_icmpv4:
 506         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 507  cleanup_udp4:
 508         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 509  cleanup_tcp4:
 510         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 511  cleanup_hooks:
 512         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 513  cleanup_pernet:
 514         unregister_pernet_subsys(&ipv4_net_ops);
 515  cleanup_sockopt:
 516         nf_unregister_sockopt(&so_getorigdst);
 517         return ret;
 518 }
 519
 520 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
 521 {
 522         synchronize_net();
 523 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
 524         nf_conntrack_ipv4_compat_fini();
 525 #endif
 526         nf_ct_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
 527         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_icmp);
 528         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_udp4);
 529         nf_ct_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
 530         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
 531         unregister_pernet_subsys(&ipv4_net_ops);
 532         nf_unregister_sockopt(&so_getorigdst);
 533 }
 534
 535 module_init(nf_conntrack_l3proto_ipv4_init);
 536 module_exit(nf_conntrack_l3proto_ipv4_fini);