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
 133 socket_match(const struct sk_buff *skb, struct xt_action_param *par,
 134              const struct xt_socket_mtinfo1 *info)
 135 {
 136         const struct iphdr *iph = ip_hdr(skb);
 137         struct udphdr _hdr, *hp = NULL;
 138         struct sock *sk = skb->sk;
 139         __be32 uninitialized_var(daddr), uninitialized_var(saddr);
 140         __be16 uninitialized_var(dport), uninitialized_var(sport);
 141         u8 uninitialized_var(protocol);
 142 #ifdef XT_SOCKET_HAVE_CONNTRACK
 143         struct nf_conn const *ct;
 144         enum ip_conntrack_info ctinfo;
 145 #endif
 146
 147         if (iph->protocol == IPPROTO_UDP || iph->protocol == IPPROTO_TCP) {
 148                 hp = skb_header_pointer(skb, ip_hdrlen(skb),
 149                                         sizeof(_hdr), &_hdr);
 150                 if (hp == NULL)
 151                         return false;
 152
 153                 protocol = iph->protocol;
 154                 saddr = iph->saddr;
 155                 sport = hp->source;
 156                 daddr = iph->daddr;
 157                 dport = hp->dest;
 158
 159         } else if (iph->protocol == IPPROTO_ICMP) {
 160                 if (extract_icmp4_fields(skb, &protocol, &saddr, &daddr,
 161                                         &sport, &dport))
 162                         return false;
 163         } else {
 164                 return false;
 165         }
 166
 167 #ifdef XT_SOCKET_HAVE_CONNTRACK
 168         /* Do the lookup with the original socket address in case this is a
 169          * reply packet of an established SNAT-ted connection. */
 170
 171         ct = nf_ct_get(skb, &ctinfo);
 172         if (ct && !nf_ct_is_untracked(ct) &&
 173             ((iph->protocol != IPPROTO_ICMP &&
 174               ctinfo == IP_CT_ESTABLISHED_REPLY) ||
 175              (iph->protocol == IPPROTO_ICMP &&
 176               ctinfo == IP_CT_RELATED_REPLY)) &&
 177             (ct->status & IPS_SRC_NAT_DONE)) {
 178
 179                 daddr = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 180                 dport = (iph->protocol == IPPROTO_TCP) ?
 181                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.tcp.port :
 182                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.udp.port;
 183         }
 184 #endif
 185
 186         if (!sk)
 187                 sk = xt_socket_get_sock_v4(dev_net(skb->dev), protocol,
 188                                            saddr, daddr, sport, dport,
 189                                            par->in);
 190         if (sk) {
 191                 bool wildcard;
 192                 bool transparent = true;
 193
 194                 /* Ignore sockets listening on INADDR_ANY,
 195                  * unless XT_SOCKET_NOWILDCARD is set
 196                  */
 197                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 198                             sk->sk_state != TCP_TIME_WAIT &&
 199                             inet_sk(sk)->inet_rcv_saddr == 0);
 200
 201                 /* Ignore non-transparent sockets,
 202                    if XT_SOCKET_TRANSPARENT is used */
 203                 if (info->flags & XT_SOCKET_TRANSPARENT)
 204                         transparent = ((sk->sk_state != TCP_TIME_WAIT &&
 205                                         inet_sk(sk)->transparent) ||
 206                                        (sk->sk_state == TCP_TIME_WAIT &&
 207                                         inet_twsk(sk)->tw_transparent));
 208
 209                 if (sk != skb->sk)
 210                         sock_gen_put(sk);
 211
 212                 if (wildcard || !transparent)
 213                         sk = NULL;
 214         }
 215
 216         pr_debug("proto %hhu %pI4:%hu -> %pI4:%hu (orig %pI4:%hu) sock %p\n",
 217                  protocol, &saddr, ntohs(sport),
 218                  &daddr, ntohs(dport),
 219                  &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
 220
 221         return (sk != NULL);
 222 }
 223
 224 static bool
 225 socket_mt4_v0(const struct sk_buff *skb, struct xt_action_param *par)
 226 {
 227         static struct xt_socket_mtinfo1 xt_info_v0 = {
 228                 .flags = 0,
 229         };
 230
 231         return socket_match(skb, par, &xt_info_v0);
 232 }
 233
 234 static bool
 235 socket_mt4_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
 236 {
 237         return socket_match(skb, par, par->matchinfo);
 238 }
 239
 240 #ifdef XT_SOCKET_HAVE_IPV6
 241
 242 static int
 243 extract_icmp6_fields(const struct sk_buff *skb,
 244                      unsigned int outside_hdrlen,
 245                      int *protocol,
 246                      struct in6_addr **raddr,
 247                      struct in6_addr **laddr,
 248                      __be16 *rport,
 249                      __be16 *lport)
 250 {
 251         struct ipv6hdr *inside_iph, _inside_iph;
 252         struct icmp6hdr *icmph, _icmph;
 253         __be16 *ports, _ports[2];
 254         u8 inside_nexthdr;
 255         __be16 inside_fragoff;
 256         int inside_hdrlen;
 257
 258         icmph = skb_header_pointer(skb, outside_hdrlen,
 259                                    sizeof(_icmph), &_icmph);
 260         if (icmph == NULL)
 261                 return 1;
 262
 263         if (icmph->icmp6_type & ICMPV6_INFOMSG_MASK)
 264                 return 1;
 265
 266         inside_iph = skb_header_pointer(skb, outside_hdrlen + sizeof(_icmph), sizeof(_inside_iph), &_inside_iph);
 267         if (inside_iph == NULL)
 268                 return 1;
 269         inside_nexthdr = inside_iph->nexthdr;
 270
 271         inside_hdrlen = ipv6_skip_exthdr(skb, outside_hdrlen + sizeof(_icmph) + sizeof(_inside_iph),
 272                                          &inside_nexthdr, &inside_fragoff);
 273         if (inside_hdrlen < 0)
 274                 return 1; /* hjm: Packet has no/incomplete transport layer headers. */
 275
 276         if (inside_nexthdr != IPPROTO_TCP &&
 277             inside_nexthdr != IPPROTO_UDP)
 278                 return 1;
 279
 280         ports = skb_header_pointer(skb, inside_hdrlen,
 281                                    sizeof(_ports), &_ports);
 282         if (ports == NULL)
 283                 return 1;
 284
 285         /* the inside IP packet is the one quoted from our side, thus
 286          * its saddr is the local address */
 287         *protocol = inside_nexthdr;
 288         *laddr = &inside_iph->saddr;
 289         *lport = ports[0];
 290         *raddr = &inside_iph->daddr;
 291         *rport = ports[1];
 292
 293         return 0;
 294 }
 295
 296 static struct sock *
 297 xt_socket_get_sock_v6(struct net *net, const u8 protocol,
 298                       const struct in6_addr *saddr, const struct in6_addr *daddr,
 299                       const __be16 sport, const __be16 dport,
 300                       const struct net_device *in)
 301 {
 302         switch (protocol) {
 303         case IPPROTO_TCP:
 304                 return inet6_lookup(net, &tcp_hashinfo,
 305                                     saddr, sport, daddr, dport,
 306                                     in->ifindex);
 307         case IPPROTO_UDP:
 308                 return udp6_lib_lookup(net, saddr, sport, daddr, dport,
 309                                        in->ifindex);
 310         }
 311
 312         return NULL;
 313 }
 314
 315 static bool
 316 socket_mt6_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
 317 {
 318         struct ipv6hdr *iph = ipv6_hdr(skb);
 319         struct udphdr _hdr, *hp = NULL;
 320         struct sock *sk = skb->sk;
 321         struct in6_addr *daddr = NULL, *saddr = NULL;
 322         __be16 uninitialized_var(dport), uninitialized_var(sport);
 323         int thoff = 0, uninitialized_var(tproto);
 324         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 325
 326         tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
 327         if (tproto < 0) {
 328                 pr_debug("unable to find transport header in IPv6 packet, dropping\n");
 329                 return NF_DROP;
 330         }
 331
 332         if (tproto == IPPROTO_UDP || tproto == IPPROTO_TCP) {
 333                 hp = skb_header_pointer(skb, thoff,
 334                                         sizeof(_hdr), &_hdr);
 335                 if (hp == NULL)
 336                         return false;
 337
 338                 saddr = &iph->saddr;
 339                 sport = hp->source;
 340                 daddr = &iph->daddr;
 341                 dport = hp->dest;
 342
 343         } else if (tproto == IPPROTO_ICMPV6) {
 344                 if (extract_icmp6_fields(skb, thoff, &tproto, &saddr, &daddr,
 345                                          &sport, &dport))
 346                         return false;
 347         } else {
 348                 return false;
 349         }
 350
 351         if (!sk)
 352                 sk = xt_socket_get_sock_v6(dev_net(skb->dev), tproto,
 353                                            saddr, daddr, sport, dport,
 354                                            par->in);
 355         if (sk) {
 356                 bool wildcard;
 357                 bool transparent = true;
 358
 359                 /* Ignore sockets listening on INADDR_ANY
 360                  * unless XT_SOCKET_NOWILDCARD is set
 361                  */
 362                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 363                             sk->sk_state != TCP_TIME_WAIT &&
 364                             ipv6_addr_any(&sk->sk_v6_rcv_saddr));
 365
 366                 /* Ignore non-transparent sockets,
 367                    if XT_SOCKET_TRANSPARENT is used */
 368                 if (info->flags & XT_SOCKET_TRANSPARENT)
 369                         transparent = ((sk->sk_state != TCP_TIME_WAIT &&
 370                                         inet_sk(sk)->transparent) ||
 371                                        (sk->sk_state == TCP_TIME_WAIT &&
 372                                         inet_twsk(sk)->tw_transparent));
 373
 374                 if (sk != skb->sk)
 375                         sock_gen_put(sk);
 376
 377                 if (wildcard || !transparent)
 378                         sk = NULL;
 379         }
 380
 381         pr_debug("proto %hhd %pI6:%hu -> %pI6:%hu "
 382                  "(orig %pI6:%hu) sock %p\n",
 383                  tproto, saddr, ntohs(sport),
 384                  daddr, ntohs(dport),
 385                  &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
 386
 387         return (sk != NULL);
 388 }
 389 #endif
 390
 391 static int socket_mt_v1_check(const struct xt_mtchk_param *par)
 392 {
 393         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 394
 395         if (info->flags & ~XT_SOCKET_FLAGS_V1) {
 396                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V1);
 397                 return -EINVAL;
 398         }
 399         return 0;
 400 }
 401
 402 static int socket_mt_v2_check(const struct xt_mtchk_param *par)
 403 {
 404         const struct xt_socket_mtinfo2 *info = (struct xt_socket_mtinfo2 *) par->matchinfo;
 405
 406         if (info->flags & ~XT_SOCKET_FLAGS_V2) {
 407                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V2);
 408                 return -EINVAL;
 409         }
 410         return 0;
 411 }
 412
 413 static struct xt_match socket_mt_reg[] __read_mostly = {
 414         {
 415                 .name           = "socket",
 416                 .revision       = 0,
 417                 .family         = NFPROTO_IPV4,
 418                 .match          = socket_mt4_v0,
 419                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 420                                   (1 << NF_INET_LOCAL_IN),
 421                 .me             = THIS_MODULE,
 422         },
 423         {
 424                 .name           = "socket",
 425                 .revision       = 1,
 426                 .family         = NFPROTO_IPV4,
 427                 .match          = socket_mt4_v1_v2,
 428                 .checkentry     = socket_mt_v1_check,
 429                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 430                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 431                                   (1 << NF_INET_LOCAL_IN),
 432                 .me             = THIS_MODULE,
 433         },
 434 #ifdef XT_SOCKET_HAVE_IPV6
 435         {
 436                 .name           = "socket",
 437                 .revision       = 1,
 438                 .family         = NFPROTO_IPV6,
 439                 .match          = socket_mt6_v1_v2,
 440                 .checkentry     = socket_mt_v1_check,
 441                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 442                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 443                                   (1 << NF_INET_LOCAL_IN),
 444                 .me             = THIS_MODULE,
 445         },
 446 #endif
 447         {
 448                 .name           = "socket",
 449                 .revision       = 2,
 450                 .family         = NFPROTO_IPV4,
 451                 .match          = socket_mt4_v1_v2,
 452                 .checkentry     = socket_mt_v2_check,
 453                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 454                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 455                                   (1 << NF_INET_LOCAL_IN),
 456                 .me             = THIS_MODULE,
 457         },
 458 #ifdef XT_SOCKET_HAVE_IPV6
 459         {
 460                 .name           = "socket",
 461                 .revision       = 2,
 462                 .family         = NFPROTO_IPV6,
 463                 .match          = socket_mt6_v1_v2,
 464                 .checkentry     = socket_mt_v2_check,
 465                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 466                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 467                                   (1 << NF_INET_LOCAL_IN),
 468                 .me             = THIS_MODULE,
 469         },
 470 #endif
 471 };
 472
 473 static int __init socket_mt_init(void)
 474 {
 475         nf_defrag_ipv4_enable();
 476 #ifdef XT_SOCKET_HAVE_IPV6
 477         nf_defrag_ipv6_enable();
 478 #endif
 479
 480         return xt_register_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 481 }
 482
 483 static void __exit socket_mt_exit(void)
 484 {
 485         xt_unregister_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 486 }
 487
 488 module_init(socket_mt_init);
 489 module_exit(socket_mt_exit);
 490
 491 MODULE_LICENSE("GPL");
 492 MODULE_AUTHOR("Krisztian Kovacs, Balazs Scheidler");
 493 MODULE_DESCRIPTION("x_tables socket match module");
 494 MODULE_ALIAS("ipt_socket");
 495 MODULE_ALIAS("ip6t_socket");