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