net/netfilter/xt_socket.c

   1 /*
   2  * Transparent proxy support for Linux/iptables
   3  *
   4  * Copyright (C) 2007-2008 BalaBit IT Ltd.
   5  * Author: Krisztian Kovacs
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  *
  11  */
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13 #include <linux/module.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/netfilter/x_tables.h>
  16 #include <linux/netfilter_ipv4/ip_tables.h>
  17 #include <net/tcp.h>
  18 #include <net/udp.h>
  19 #include <net/icmp.h>
  20 #include <net/sock.h>
  21 #include <net/inet_sock.h>
  22 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
  23
  24 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
  25 #define XT_SOCKET_HAVE_IPV6 1
  26 #include <linux/netfilter_ipv6/ip6_tables.h>
  27 #include <net/inet6_hashtables.h>
  28 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
  29 #endif
  30
  31 #include <linux/netfilter/xt_socket.h>
  32
  33 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  34 #define XT_SOCKET_HAVE_CONNTRACK 1
  35 #include <net/netfilter/nf_conntrack.h>
  36 #endif
  37
  38 static int
  39 extract_icmp4_fields(const struct sk_buff *skb,
  40                     u8 *protocol,
  41                     __be32 *raddr,
  42                     __be32 *laddr,
  43                     __be16 *rport,
  44                     __be16 *lport)
  45 {
  46         unsigned int outside_hdrlen = ip_hdrlen(skb);
  47         struct iphdr *inside_iph, _inside_iph;
  48         struct icmphdr *icmph, _icmph;
  49         __be16 *ports, _ports[2];
  50
  51         icmph = skb_header_pointer(skb, outside_hdrlen,
  52                                    sizeof(_icmph), &_icmph);
  53         if (icmph == NULL)
  54                 return 1;
  55
  56         switch (icmph->type) {
  57         case ICMP_DEST_UNREACH:
  58         case ICMP_SOURCE_QUENCH:
  59         case ICMP_REDIRECT:
  60         case ICMP_TIME_EXCEEDED:
  61         case ICMP_PARAMETERPROB:
  62                 break;
  63         default:
  64                 return 1;
  65         }
  66
  67         inside_iph = skb_header_pointer(skb, outside_hdrlen +
  68                                         sizeof(struct icmphdr),
  69                                         sizeof(_inside_iph), &_inside_iph);
  70         if (inside_iph == NULL)
  71                 return 1;
  72
  73         if (inside_iph->protocol != IPPROTO_TCP &&
  74             inside_iph->protocol != IPPROTO_UDP)
  75                 return 1;
  76
  77         ports = skb_header_pointer(skb, outside_hdrlen +
  78                                    sizeof(struct icmphdr) +
  79                                    (inside_iph->ihl << 2),
  80                                    sizeof(_ports), &_ports);
  81         if (ports == NULL)
  82                 return 1;
  83
  84         /* the inside IP packet is the one quoted from our side, thus
  85          * its saddr is the local address */
  86         *protocol = inside_iph->protocol;
  87         *laddr = inside_iph->saddr;
  88         *lport = ports[0];
  89         *raddr = inside_iph->daddr;
  90         *rport = ports[1];
  91
  92         return 0;
  93 }
  94
  95 /* "socket" match based redirection (no specific rule)
  96  * ===================================================
  97  *
  98  * There are connections with dynamic endpoints (e.g. FTP data
  99  * connection) that the user is unable to add explicit rules
 100  * for. These are taken care of by a generic "socket" rule. It is
 101  * assumed that the proxy application is trusted to open such
 102  * connections without explicit iptables rule (except of course the
 103  * generic 'socket' rule). In this case the following sockets are
 104  * matched in preference order:
 105  *
 106  *   - match: if there's a fully established connection matching the
 107  *     _packet_ tuple
 108  *
 109  *   - match: if there's a non-zero bound listener (possibly with a
 110  *     non-local address) We don't accept zero-bound listeners, since
 111  *     then local services could intercept traffic going through the
 112  *     box.
 113  */
 114 static struct sock *
 115 xt_socket_get_sock_v4(struct net *net, struct sk_buff *skb, const int doff,
 116                       const u8 protocol,
 117                       const __be32 saddr, const __be32 daddr,
 118                       const __be16 sport, const __be16 dport,
 119                       const struct net_device *in)
 120 {
 121         switch (protocol) {
 122         case IPPROTO_TCP:
 123                 return __inet_lookup(net, &tcp_hashinfo, skb, doff,
 124                                      saddr, sport, daddr, dport,
 125                                      in->ifindex);
 126         case IPPROTO_UDP:
 127                 return udp4_lib_lookup(net, saddr, sport, daddr, dport,
 128                                        in->ifindex);
 129         }
 130         return NULL;
 131 }
 132
 133 static bool xt_socket_sk_is_transparent(struct sock *sk)
 134 {
 135         switch (sk->sk_state) {
 136         case TCP_TIME_WAIT:
 137                 return inet_twsk(sk)->tw_transparent;
 138
 139         case TCP_NEW_SYN_RECV:
 140                 return inet_rsk(inet_reqsk(sk))->no_srccheck;
 141
 142         default:
 143                 return inet_sk(sk)->transparent;
 144         }
 145 }
 146
 147 static struct sock *xt_socket_lookup_slow_v4(struct net *net,
 148                                              const struct sk_buff *skb,
 149                                              const struct net_device *indev)
 150 {
 151         const struct iphdr *iph = ip_hdr(skb);
 152         struct sk_buff *data_skb = NULL;
 153         int doff = 0;
 154         __be32 uninitialized_var(daddr), uninitialized_var(saddr);
 155         __be16 uninitialized_var(dport), uninitialized_var(sport);
 156         u8 uninitialized_var(protocol);
 157 #ifdef XT_SOCKET_HAVE_CONNTRACK
 158         struct nf_conn const *ct;
 159         enum ip_conntrack_info ctinfo;
 160 #endif
 161
 162         if (iph->protocol == IPPROTO_UDP || iph->protocol == IPPROTO_TCP) {
 163                 struct udphdr _hdr, *hp;
 164
 165                 hp = skb_header_pointer(skb, ip_hdrlen(skb),
 166                                         sizeof(_hdr), &_hdr);
 167                 if (hp == NULL)
 168                         return NULL;
 169
 170                 protocol = iph->protocol;
 171                 saddr = iph->saddr;
 172                 sport = hp->source;
 173                 daddr = iph->daddr;
 174                 dport = hp->dest;
 175                 data_skb = (struct sk_buff *)skb;
 176                 doff = iph->protocol == IPPROTO_TCP ?
 177                         ip_hdrlen(skb) + __tcp_hdrlen((struct tcphdr *)hp) :
 178                         ip_hdrlen(skb) + sizeof(*hp);
 179
 180         } else if (iph->protocol == IPPROTO_ICMP) {
 181                 if (extract_icmp4_fields(skb, &protocol, &saddr, &daddr,
 182                                          &sport, &dport))
 183                         return NULL;
 184         } else {
 185                 return NULL;
 186         }
 187
 188 #ifdef XT_SOCKET_HAVE_CONNTRACK
 189         /* Do the lookup with the original socket address in
 190          * case this is a reply packet of an established
 191          * SNAT-ted connection.
 192          */
 193         ct = nf_ct_get(skb, &ctinfo);
 194         if (ct && !nf_ct_is_untracked(ct) &&
 195             ((iph->protocol != IPPROTO_ICMP &&
 196               ctinfo == IP_CT_ESTABLISHED_REPLY) ||
 197              (iph->protocol == IPPROTO_ICMP &&
 198               ctinfo == IP_CT_RELATED_REPLY)) &&
 199             (ct->status & IPS_SRC_NAT_DONE)) {
 200
 201                 daddr = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 202                 dport = (iph->protocol == IPPROTO_TCP) ?
 203                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.tcp.port :
 204                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.udp.port;
 205         }
 206 #endif
 207
 208         return xt_socket_get_sock_v4(net, data_skb, doff, protocol, saddr,
 209                                      daddr, sport, dport, indev);
 210 }
 211
 212 static bool
 213 socket_match(const struct sk_buff *skb, struct xt_action_param *par,
 214              const struct xt_socket_mtinfo1 *info)
 215 {
 216         struct sk_buff *pskb = (struct sk_buff *)skb;
 217         struct sock *sk = skb->sk;
 218
 219         if (!sk)
 220                 sk = xt_socket_lookup_slow_v4(par->net, skb, par->in);
 221         if (sk) {
 222                 bool wildcard;
 223                 bool transparent = true;
 224
 225                 /* Ignore sockets listening on INADDR_ANY,
 226                  * unless XT_SOCKET_NOWILDCARD is set
 227                  */
 228                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 229                             sk_fullsock(sk) &&
 230                             inet_sk(sk)->inet_rcv_saddr == 0);
 231
 232                 /* Ignore non-transparent sockets,
 233                  * if XT_SOCKET_TRANSPARENT is used
 234                  */
 235                 if (info->flags & XT_SOCKET_TRANSPARENT)
 236                         transparent = xt_socket_sk_is_transparent(sk);
 237
 238                 if (info->flags & XT_SOCKET_RESTORESKMARK && !wildcard &&
 239                     transparent)
 240                         pskb->mark = sk->sk_mark;
 241
 242                 if (sk != skb->sk)
 243                         sock_gen_put(sk);
 244
 245                 if (wildcard || !transparent)
 246                         sk = NULL;
 247         }
 248
 249         return sk != NULL;
 250 }
 251
 252 static bool
 253 socket_mt4_v0(const struct sk_buff *skb, struct xt_action_param *par)
 254 {
 255         static struct xt_socket_mtinfo1 xt_info_v0 = {
 256                 .flags = 0,
 257         };
 258
 259         return socket_match(skb, par, &xt_info_v0);
 260 }
 261
 262 static bool
 263 socket_mt4_v1_v2_v3(const struct sk_buff *skb, struct xt_action_param *par)
 264 {
 265         return socket_match(skb, par, par->matchinfo);
 266 }
 267
 268 #ifdef XT_SOCKET_HAVE_IPV6
 269
 270 static int
 271 extract_icmp6_fields(const struct sk_buff *skb,
 272                      unsigned int outside_hdrlen,
 273                      int *protocol,
 274                      const struct in6_addr **raddr,
 275                      const struct in6_addr **laddr,
 276                      __be16 *rport,
 277                      __be16 *lport,
 278                      struct ipv6hdr *ipv6_var)
 279 {
 280         const struct ipv6hdr *inside_iph;
 281         struct icmp6hdr *icmph, _icmph;
 282         __be16 *ports, _ports[2];
 283         u8 inside_nexthdr;
 284         __be16 inside_fragoff;
 285         int inside_hdrlen;
 286
 287         icmph = skb_header_pointer(skb, outside_hdrlen,
 288                                    sizeof(_icmph), &_icmph);
 289         if (icmph == NULL)
 290                 return 1;
 291
 292         if (icmph->icmp6_type & ICMPV6_INFOMSG_MASK)
 293                 return 1;
 294
 295         inside_iph = skb_header_pointer(skb, outside_hdrlen + sizeof(_icmph),
 296                                         sizeof(*ipv6_var), ipv6_var);
 297         if (inside_iph == NULL)
 298                 return 1;
 299         inside_nexthdr = inside_iph->nexthdr;
 300
 301         inside_hdrlen = ipv6_skip_exthdr(skb, outside_hdrlen + sizeof(_icmph) +
 302                                               sizeof(*ipv6_var),
 303                                          &inside_nexthdr, &inside_fragoff);
 304         if (inside_hdrlen < 0)
 305                 return 1; /* hjm: Packet has no/incomplete transport layer headers. */
 306
 307         if (inside_nexthdr != IPPROTO_TCP &&
 308             inside_nexthdr != IPPROTO_UDP)
 309                 return 1;
 310
 311         ports = skb_header_pointer(skb, inside_hdrlen,
 312                                    sizeof(_ports), &_ports);
 313         if (ports == NULL)
 314                 return 1;
 315
 316         /* the inside IP packet is the one quoted from our side, thus
 317          * its saddr is the local address */
 318         *protocol = inside_nexthdr;
 319         *laddr = &inside_iph->saddr;
 320         *lport = ports[0];
 321         *raddr = &inside_iph->daddr;
 322         *rport = ports[1];
 323
 324         return 0;
 325 }
 326
 327 static struct sock *
 328 xt_socket_get_sock_v6(struct net *net, struct sk_buff *skb, int doff,
 329                       const u8 protocol,
 330                       const struct in6_addr *saddr, const struct in6_addr *daddr,
 331                       const __be16 sport, const __be16 dport,
 332                       const struct net_device *in)
 333 {
 334         switch (protocol) {
 335         case IPPROTO_TCP:
 336                 return inet6_lookup(net, &tcp_hashinfo, skb, doff,
 337                                     saddr, sport, daddr, dport,
 338                                     in->ifindex);
 339         case IPPROTO_UDP:
 340                 return udp6_lib_lookup(net, saddr, sport, daddr, dport,
 341                                        in->ifindex);
 342         }
 343
 344         return NULL;
 345 }
 346
 347 static struct sock *xt_socket_lookup_slow_v6(struct net *net,
 348                                              const struct sk_buff *skb,
 349                                              const struct net_device *indev)
 350 {
 351         __be16 uninitialized_var(dport), uninitialized_var(sport);
 352         const struct in6_addr *daddr = NULL, *saddr = NULL;
 353         struct ipv6hdr *iph = ipv6_hdr(skb);
 354         struct sk_buff *data_skb = NULL;
 355         int doff = 0;
 356         int thoff = 0, tproto;
 357
 358         tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
 359         if (tproto < 0) {
 360                 pr_debug("unable to find transport header in IPv6 packet, dropping\n");
 361                 return NULL;
 362         }
 363
 364         if (tproto == IPPROTO_UDP || tproto == IPPROTO_TCP) {
 365                 struct udphdr _hdr, *hp;
 366
 367                 hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
 368                 if (hp == NULL)
 369                         return NULL;
 370
 371                 saddr = &iph->saddr;
 372                 sport = hp->source;
 373                 daddr = &iph->daddr;
 374                 dport = hp->dest;
 375                 data_skb = (struct sk_buff *)skb;
 376                 doff = tproto == IPPROTO_TCP ?
 377                         thoff + __tcp_hdrlen((struct tcphdr *)hp) :
 378                         thoff + sizeof(*hp);
 379
 380         } else if (tproto == IPPROTO_ICMPV6) {
 381                 struct ipv6hdr ipv6_var;
 382
 383                 if (extract_icmp6_fields(skb, thoff, &tproto, &saddr, &daddr,
 384                                          &sport, &dport, &ipv6_var))
 385                         return NULL;
 386         } else {
 387                 return NULL;
 388         }
 389
 390         return xt_socket_get_sock_v6(net, data_skb, doff, tproto, saddr, daddr,
 391                                      sport, dport, indev);
 392 }
 393
 394 static bool
 395 socket_mt6_v1_v2_v3(const struct sk_buff *skb, struct xt_action_param *par)
 396 {
 397         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 398         struct sk_buff *pskb = (struct sk_buff *)skb;
 399         struct sock *sk = skb->sk;
 400
 401         if (!sk)
 402                 sk = xt_socket_lookup_slow_v6(par->net, skb, par->in);
 403         if (sk) {
 404                 bool wildcard;
 405                 bool transparent = true;
 406
 407                 /* Ignore sockets listening on INADDR_ANY
 408                  * unless XT_SOCKET_NOWILDCARD is set
 409                  */
 410                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 411                             sk_fullsock(sk) &&
 412                             ipv6_addr_any(&sk->sk_v6_rcv_saddr));
 413
 414                 /* Ignore non-transparent sockets,
 415                  * if XT_SOCKET_TRANSPARENT is used
 416                  */
 417                 if (info->flags & XT_SOCKET_TRANSPARENT)
 418                         transparent = xt_socket_sk_is_transparent(sk);
 419
 420                 if (info->flags & XT_SOCKET_RESTORESKMARK && !wildcard &&
 421                     transparent)
 422                         pskb->mark = sk->sk_mark;
 423
 424                 if (sk != skb->sk)
 425                         sock_gen_put(sk);
 426
 427                 if (wildcard || !transparent)
 428                         sk = NULL;
 429         }
 430
 431         return sk != NULL;
 432 }
 433 #endif
 434
 435 static int socket_mt_v1_check(const struct xt_mtchk_param *par)
 436 {
 437         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 438
 439         if (info->flags & ~XT_SOCKET_FLAGS_V1) {
 440                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V1);
 441                 return -EINVAL;
 442         }
 443         return 0;
 444 }
 445
 446 static int socket_mt_v2_check(const struct xt_mtchk_param *par)
 447 {
 448         const struct xt_socket_mtinfo2 *info = (struct xt_socket_mtinfo2 *) par->matchinfo;
 449
 450         if (info->flags & ~XT_SOCKET_FLAGS_V2) {
 451                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V2);
 452                 return -EINVAL;
 453         }
 454         return 0;
 455 }
 456
 457 static int socket_mt_v3_check(const struct xt_mtchk_param *par)
 458 {
 459         const struct xt_socket_mtinfo3 *info =
 460                                     (struct xt_socket_mtinfo3 *)par->matchinfo;
 461
 462         if (info->flags & ~XT_SOCKET_FLAGS_V3) {
 463                 pr_info("unknown flags 0x%x\n",
 464                         info->flags & ~XT_SOCKET_FLAGS_V3);
 465                 return -EINVAL;
 466         }
 467         return 0;
 468 }
 469
 470 static struct xt_match socket_mt_reg[] __read_mostly = {
 471         {
 472                 .name           = "socket",
 473                 .revision       = 0,
 474                 .family         = NFPROTO_IPV4,
 475                 .match          = socket_mt4_v0,
 476                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 477                                   (1 << NF_INET_LOCAL_IN),
 478                 .me             = THIS_MODULE,
 479         },
 480         {
 481                 .name           = "socket",
 482                 .revision       = 1,
 483                 .family         = NFPROTO_IPV4,
 484                 .match          = socket_mt4_v1_v2_v3,
 485                 .checkentry     = socket_mt_v1_check,
 486                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 487                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 488                                   (1 << NF_INET_LOCAL_IN),
 489                 .me             = THIS_MODULE,
 490         },
 491 #ifdef XT_SOCKET_HAVE_IPV6
 492         {
 493                 .name           = "socket",
 494                 .revision       = 1,
 495                 .family         = NFPROTO_IPV6,
 496                 .match          = socket_mt6_v1_v2_v3,
 497                 .checkentry     = socket_mt_v1_check,
 498                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 499                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 500                                   (1 << NF_INET_LOCAL_IN),
 501                 .me             = THIS_MODULE,
 502         },
 503 #endif
 504         {
 505                 .name           = "socket",
 506                 .revision       = 2,
 507                 .family         = NFPROTO_IPV4,
 508                 .match          = socket_mt4_v1_v2_v3,
 509                 .checkentry     = socket_mt_v2_check,
 510                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 511                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 512                                   (1 << NF_INET_LOCAL_IN),
 513                 .me             = THIS_MODULE,
 514         },
 515 #ifdef XT_SOCKET_HAVE_IPV6
 516         {
 517                 .name           = "socket",
 518                 .revision       = 2,
 519                 .family         = NFPROTO_IPV6,
 520                 .match          = socket_mt6_v1_v2_v3,
 521                 .checkentry     = socket_mt_v2_check,
 522                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 523                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 524                                   (1 << NF_INET_LOCAL_IN),
 525                 .me             = THIS_MODULE,
 526         },
 527 #endif
 528         {
 529                 .name           = "socket",
 530                 .revision       = 3,
 531                 .family         = NFPROTO_IPV4,
 532                 .match          = socket_mt4_v1_v2_v3,
 533                 .checkentry     = socket_mt_v3_check,
 534                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 535                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 536                                   (1 << NF_INET_LOCAL_IN),
 537                 .me             = THIS_MODULE,
 538         },
 539 #ifdef XT_SOCKET_HAVE_IPV6
 540         {
 541                 .name           = "socket",
 542                 .revision       = 3,
 543                 .family         = NFPROTO_IPV6,
 544                 .match          = socket_mt6_v1_v2_v3,
 545                 .checkentry     = socket_mt_v3_check,
 546                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 547                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 548                                   (1 << NF_INET_LOCAL_IN),
 549                 .me             = THIS_MODULE,
 550         },
 551 #endif
 552 };
 553
 554 static int __init socket_mt_init(void)
 555 {
 556         nf_defrag_ipv4_enable();
 557 #ifdef XT_SOCKET_HAVE_IPV6
 558         nf_defrag_ipv6_enable();
 559 #endif
 560
 561         return xt_register_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 562 }
 563
 564 static void __exit socket_mt_exit(void)
 565 {
 566         xt_unregister_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 567 }
 568
 569 module_init(socket_mt_init);
 570 module_exit(socket_mt_exit);
 571
 572 MODULE_LICENSE("GPL");
 573 MODULE_AUTHOR("Krisztian Kovacs, Balazs Scheidler");
 574 MODULE_DESCRIPTION("x_tables socket match module");
 575 MODULE_ALIAS("ipt_socket");
 576 MODULE_ALIAS("ip6t_socket");