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, const u8 protocol,
 116                       const __be32 saddr, const __be32 daddr,
 117                       const __be16 sport, const __be16 dport,
 118                       const struct net_device *in)
 119 {
 120         switch (protocol) {
 121         case IPPROTO_TCP:
 122                 return __inet_lookup(net, &tcp_hashinfo,
 123                                      saddr, sport, daddr, dport,
 124                                      in->ifindex);
 125         case IPPROTO_UDP:
 126                 return udp4_lib_lookup(net, saddr, sport, daddr, dport,
 127                                        in->ifindex);
 128         }
 129         return NULL;
 130 }
 131
 132 static bool xt_socket_sk_is_transparent(struct sock *sk)
 133 {
 134         switch (sk->sk_state) {
 135         case TCP_TIME_WAIT:
 136                 return inet_twsk(sk)->tw_transparent;
 137
 138         case TCP_NEW_SYN_RECV:
 139                 return inet_rsk(inet_reqsk(sk))->no_srccheck;
 140
 141         default:
 142                 return inet_sk(sk)->transparent;
 143         }
 144 }
 145
 146 static struct sock *xt_socket_lookup_slow_v4(const struct sk_buff *skb,
 147                                              const struct net_device *indev)
 148 {
 149         const struct iphdr *iph = ip_hdr(skb);
 150         __be32 uninitialized_var(daddr), uninitialized_var(saddr);
 151         __be16 uninitialized_var(dport), uninitialized_var(sport);
 152         u8 uninitialized_var(protocol);
 153 #ifdef XT_SOCKET_HAVE_CONNTRACK
 154         struct nf_conn const *ct;
 155         enum ip_conntrack_info ctinfo;
 156 #endif
 157
 158         if (iph->protocol == IPPROTO_UDP || iph->protocol == IPPROTO_TCP) {
 159                 struct udphdr _hdr, *hp;
 160
 161                 hp = skb_header_pointer(skb, ip_hdrlen(skb),
 162                                         sizeof(_hdr), &_hdr);
 163                 if (hp == NULL)
 164                         return NULL;
 165
 166                 protocol = iph->protocol;
 167                 saddr = iph->saddr;
 168                 sport = hp->source;
 169                 daddr = iph->daddr;
 170                 dport = hp->dest;
 171
 172         } else if (iph->protocol == IPPROTO_ICMP) {
 173                 if (extract_icmp4_fields(skb, &protocol, &saddr, &daddr,
 174                                          &sport, &dport))
 175                         return NULL;
 176         } else {
 177                 return NULL;
 178         }
 179
 180 #ifdef XT_SOCKET_HAVE_CONNTRACK
 181         /* Do the lookup with the original socket address in
 182          * case this is a reply packet of an established
 183          * SNAT-ted connection.
 184          */
 185         ct = nf_ct_get(skb, &ctinfo);
 186         if (ct && !nf_ct_is_untracked(ct) &&
 187             ((iph->protocol != IPPROTO_ICMP &&
 188               ctinfo == IP_CT_ESTABLISHED_REPLY) ||
 189              (iph->protocol == IPPROTO_ICMP &&
 190               ctinfo == IP_CT_RELATED_REPLY)) &&
 191             (ct->status & IPS_SRC_NAT_DONE)) {
 192
 193                 daddr = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 194                 dport = (iph->protocol == IPPROTO_TCP) ?
 195                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.tcp.port :
 196                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.udp.port;
 197         }
 198 #endif
 199
 200         return xt_socket_get_sock_v4(dev_net(skb->dev), protocol, saddr, daddr,
 201                                      sport, dport, indev);
 202 }
 203
 204 static bool
 205 socket_match(const struct sk_buff *skb, struct xt_action_param *par,
 206              const struct xt_socket_mtinfo1 *info)
 207 {
 208         struct sk_buff *pskb = (struct sk_buff *)skb;
 209         struct sock *sk = skb->sk;
 210
 211         if (!sk)
 212                 sk = xt_socket_lookup_slow_v4(skb, par->in);
 213         if (sk) {
 214                 bool wildcard;
 215                 bool transparent = true;
 216
 217                 /* Ignore sockets listening on INADDR_ANY,
 218                  * unless XT_SOCKET_NOWILDCARD is set
 219                  */
 220                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 221                             sk_fullsock(sk) &&
 222                             inet_sk(sk)->inet_rcv_saddr == 0);
 223
 224                 /* Ignore non-transparent sockets,
 225                  * if XT_SOCKET_TRANSPARENT is used
 226                  */
 227                 if (info->flags & XT_SOCKET_TRANSPARENT)
 228                         transparent = xt_socket_sk_is_transparent(sk);
 229
 230                 if (info->flags & XT_SOCKET_RESTORESKMARK && !wildcard &&
 231                     transparent)
 232                         pskb->mark = sk->sk_mark;
 233
 234                 if (sk != skb->sk)
 235                         sock_gen_put(sk);
 236
 237                 if (wildcard || !transparent)
 238                         sk = NULL;
 239         }
 240
 241         return sk != NULL;
 242 }
 243
 244 static bool
 245 socket_mt4_v0(const struct sk_buff *skb, struct xt_action_param *par)
 246 {
 247         static struct xt_socket_mtinfo1 xt_info_v0 = {
 248                 .flags = 0,
 249         };
 250
 251         return socket_match(skb, par, &xt_info_v0);
 252 }
 253
 254 static bool
 255 socket_mt4_v1_v2_v3(const struct sk_buff *skb, struct xt_action_param *par)
 256 {
 257         return socket_match(skb, par, par->matchinfo);
 258 }
 259
 260 #ifdef XT_SOCKET_HAVE_IPV6
 261
 262 static int
 263 extract_icmp6_fields(const struct sk_buff *skb,
 264                      unsigned int outside_hdrlen,
 265                      int *protocol,
 266                      const struct in6_addr **raddr,
 267                      const struct in6_addr **laddr,
 268                      __be16 *rport,
 269                      __be16 *lport,
 270                      struct ipv6hdr *ipv6_var)
 271 {
 272         const struct ipv6hdr *inside_iph;
 273         struct icmp6hdr *icmph, _icmph;
 274         __be16 *ports, _ports[2];
 275         u8 inside_nexthdr;
 276         __be16 inside_fragoff;
 277         int inside_hdrlen;
 278
 279         icmph = skb_header_pointer(skb, outside_hdrlen,
 280                                    sizeof(_icmph), &_icmph);
 281         if (icmph == NULL)
 282                 return 1;
 283
 284         if (icmph->icmp6_type & ICMPV6_INFOMSG_MASK)
 285                 return 1;
 286
 287         inside_iph = skb_header_pointer(skb, outside_hdrlen + sizeof(_icmph),
 288                                         sizeof(*ipv6_var), ipv6_var);
 289         if (inside_iph == NULL)
 290                 return 1;
 291         inside_nexthdr = inside_iph->nexthdr;
 292
 293         inside_hdrlen = ipv6_skip_exthdr(skb, outside_hdrlen + sizeof(_icmph) +
 294                                               sizeof(*ipv6_var),
 295                                          &inside_nexthdr, &inside_fragoff);
 296         if (inside_hdrlen < 0)
 297                 return 1; /* hjm: Packet has no/incomplete transport layer headers. */
 298
 299         if (inside_nexthdr != IPPROTO_TCP &&
 300             inside_nexthdr != IPPROTO_UDP)
 301                 return 1;
 302
 303         ports = skb_header_pointer(skb, inside_hdrlen,
 304                                    sizeof(_ports), &_ports);
 305         if (ports == NULL)
 306                 return 1;
 307
 308         /* the inside IP packet is the one quoted from our side, thus
 309          * its saddr is the local address */
 310         *protocol = inside_nexthdr;
 311         *laddr = &inside_iph->saddr;
 312         *lport = ports[0];
 313         *raddr = &inside_iph->daddr;
 314         *rport = ports[1];
 315
 316         return 0;
 317 }
 318
 319 static struct sock *
 320 xt_socket_get_sock_v6(struct net *net, const u8 protocol,
 321                       const struct in6_addr *saddr, const struct in6_addr *daddr,
 322                       const __be16 sport, const __be16 dport,
 323                       const struct net_device *in)
 324 {
 325         switch (protocol) {
 326         case IPPROTO_TCP:
 327                 return inet6_lookup(net, &tcp_hashinfo,
 328                                     saddr, sport, daddr, dport,
 329                                     in->ifindex);
 330         case IPPROTO_UDP:
 331                 return udp6_lib_lookup(net, saddr, sport, daddr, dport,
 332                                        in->ifindex);
 333         }
 334
 335         return NULL;
 336 }
 337
 338 static struct sock *xt_socket_lookup_slow_v6(const struct sk_buff *skb,
 339                                              const struct net_device *indev)
 340 {
 341         __be16 uninitialized_var(dport), uninitialized_var(sport);
 342         const struct in6_addr *daddr = NULL, *saddr = NULL;
 343         struct ipv6hdr *iph = ipv6_hdr(skb);
 344         int thoff = 0, tproto;
 345
 346         tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
 347         if (tproto < 0) {
 348                 pr_debug("unable to find transport header in IPv6 packet, dropping\n");
 349                 return NULL;
 350         }
 351
 352         if (tproto == IPPROTO_UDP || tproto == IPPROTO_TCP) {
 353                 struct udphdr _hdr, *hp;
 354
 355                 hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
 356                 if (hp == NULL)
 357                         return NULL;
 358
 359                 saddr = &iph->saddr;
 360                 sport = hp->source;
 361                 daddr = &iph->daddr;
 362                 dport = hp->dest;
 363
 364         } else if (tproto == IPPROTO_ICMPV6) {
 365                 struct ipv6hdr ipv6_var;
 366
 367                 if (extract_icmp6_fields(skb, thoff, &tproto, &saddr, &daddr,
 368                                          &sport, &dport, &ipv6_var))
 369                         return NULL;
 370         } else {
 371                 return NULL;
 372         }
 373
 374         return xt_socket_get_sock_v6(dev_net(skb->dev), tproto, saddr, daddr,
 375                                      sport, dport, indev);
 376 }
 377
 378 static bool
 379 socket_mt6_v1_v2_v3(const struct sk_buff *skb, struct xt_action_param *par)
 380 {
 381         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 382         struct sk_buff *pskb = (struct sk_buff *)skb;
 383         struct sock *sk = skb->sk;
 384
 385         if (!sk)
 386                 sk = xt_socket_lookup_slow_v6(skb, par->in);
 387         if (sk) {
 388                 bool wildcard;
 389                 bool transparent = true;
 390
 391                 /* Ignore sockets listening on INADDR_ANY
 392                  * unless XT_SOCKET_NOWILDCARD is set
 393                  */
 394                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 395                             sk_fullsock(sk) &&
 396                             ipv6_addr_any(&sk->sk_v6_rcv_saddr));
 397
 398                 /* Ignore non-transparent sockets,
 399                  * if XT_SOCKET_TRANSPARENT is used
 400                  */
 401                 if (info->flags & XT_SOCKET_TRANSPARENT)
 402                         transparent = xt_socket_sk_is_transparent(sk);
 403
 404                 if (info->flags & XT_SOCKET_RESTORESKMARK && !wildcard &&
 405                     transparent)
 406                         pskb->mark = sk->sk_mark;
 407
 408                 if (sk != skb->sk)
 409                         sock_gen_put(sk);
 410
 411                 if (wildcard || !transparent)
 412                         sk = NULL;
 413         }
 414
 415         return sk != NULL;
 416 }
 417 #endif
 418
 419 static int socket_mt_v1_check(const struct xt_mtchk_param *par)
 420 {
 421         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 422
 423         if (info->flags & ~XT_SOCKET_FLAGS_V1) {
 424                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V1);
 425                 return -EINVAL;
 426         }
 427         return 0;
 428 }
 429
 430 static int socket_mt_v2_check(const struct xt_mtchk_param *par)
 431 {
 432         const struct xt_socket_mtinfo2 *info = (struct xt_socket_mtinfo2 *) par->matchinfo;
 433
 434         if (info->flags & ~XT_SOCKET_FLAGS_V2) {
 435                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V2);
 436                 return -EINVAL;
 437         }
 438         return 0;
 439 }
 440
 441 static int socket_mt_v3_check(const struct xt_mtchk_param *par)
 442 {
 443         const struct xt_socket_mtinfo3 *info =
 444                                     (struct xt_socket_mtinfo3 *)par->matchinfo;
 445
 446         if (info->flags & ~XT_SOCKET_FLAGS_V3) {
 447                 pr_info("unknown flags 0x%x\n",
 448                         info->flags & ~XT_SOCKET_FLAGS_V3);
 449                 return -EINVAL;
 450         }
 451         return 0;
 452 }
 453
 454 static struct xt_match socket_mt_reg[] __read_mostly = {
 455         {
 456                 .name           = "socket",
 457                 .revision       = 0,
 458                 .family         = NFPROTO_IPV4,
 459                 .match          = socket_mt4_v0,
 460                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 461                                   (1 << NF_INET_LOCAL_IN),
 462                 .me             = THIS_MODULE,
 463         },
 464         {
 465                 .name           = "socket",
 466                 .revision       = 1,
 467                 .family         = NFPROTO_IPV4,
 468                 .match          = socket_mt4_v1_v2_v3,
 469                 .checkentry     = socket_mt_v1_check,
 470                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 471                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 472                                   (1 << NF_INET_LOCAL_IN),
 473                 .me             = THIS_MODULE,
 474         },
 475 #ifdef XT_SOCKET_HAVE_IPV6
 476         {
 477                 .name           = "socket",
 478                 .revision       = 1,
 479                 .family         = NFPROTO_IPV6,
 480                 .match          = socket_mt6_v1_v2_v3,
 481                 .checkentry     = socket_mt_v1_check,
 482                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 483                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 484                                   (1 << NF_INET_LOCAL_IN),
 485                 .me             = THIS_MODULE,
 486         },
 487 #endif
 488         {
 489                 .name           = "socket",
 490                 .revision       = 2,
 491                 .family         = NFPROTO_IPV4,
 492                 .match          = socket_mt4_v1_v2_v3,
 493                 .checkentry     = socket_mt_v2_check,
 494                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 495                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 496                                   (1 << NF_INET_LOCAL_IN),
 497                 .me             = THIS_MODULE,
 498         },
 499 #ifdef XT_SOCKET_HAVE_IPV6
 500         {
 501                 .name           = "socket",
 502                 .revision       = 2,
 503                 .family         = NFPROTO_IPV6,
 504                 .match          = socket_mt6_v1_v2_v3,
 505                 .checkentry     = socket_mt_v2_check,
 506                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 507                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 508                                   (1 << NF_INET_LOCAL_IN),
 509                 .me             = THIS_MODULE,
 510         },
 511 #endif
 512         {
 513                 .name           = "socket",
 514                 .revision       = 3,
 515                 .family         = NFPROTO_IPV4,
 516                 .match          = socket_mt4_v1_v2_v3,
 517                 .checkentry     = socket_mt_v3_check,
 518                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 519                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 520                                   (1 << NF_INET_LOCAL_IN),
 521                 .me             = THIS_MODULE,
 522         },
 523 #ifdef XT_SOCKET_HAVE_IPV6
 524         {
 525                 .name           = "socket",
 526                 .revision       = 3,
 527                 .family         = NFPROTO_IPV6,
 528                 .match          = socket_mt6_v1_v2_v3,
 529                 .checkentry     = socket_mt_v3_check,
 530                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 531                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 532                                   (1 << NF_INET_LOCAL_IN),
 533                 .me             = THIS_MODULE,
 534         },
 535 #endif
 536 };
 537
 538 static int __init socket_mt_init(void)
 539 {
 540         nf_defrag_ipv4_enable();
 541 #ifdef XT_SOCKET_HAVE_IPV6
 542         nf_defrag_ipv6_enable();
 543 #endif
 544
 545         return xt_register_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 546 }
 547
 548 static void __exit socket_mt_exit(void)
 549 {
 550         xt_unregister_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 551 }
 552
 553 module_init(socket_mt_init);
 554 module_exit(socket_mt_exit);
 555
 556 MODULE_LICENSE("GPL");
 557 MODULE_AUTHOR("Krisztian Kovacs, Balazs Scheidler");
 558 MODULE_DESCRIPTION("x_tables socket match module");
 559 MODULE_ALIAS("ipt_socket");
 560 MODULE_ALIAS("ip6t_socket");