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 void
  39 xt_socket_put_sk(struct sock *sk)
  40 {
  41         if (sk->sk_state == TCP_TIME_WAIT)
  42                 inet_twsk_put(inet_twsk(sk));
  43         else
  44                 sock_put(sk);
  45 }
  46
  47 static int
  48 extract_icmp4_fields(const struct sk_buff *skb,
  49                     u8 *protocol,
  50                     __be32 *raddr,
  51                     __be32 *laddr,
  52                     __be16 *rport,
  53                     __be16 *lport)
  54 {
  55         unsigned int outside_hdrlen = ip_hdrlen(skb);
  56         struct iphdr *inside_iph, _inside_iph;
  57         struct icmphdr *icmph, _icmph;
  58         __be16 *ports, _ports[2];
  59
  60         icmph = skb_header_pointer(skb, outside_hdrlen,
  61                                    sizeof(_icmph), &_icmph);
  62         if (icmph == NULL)
  63                 return 1;
  64
  65         switch (icmph->type) {
  66         case ICMP_DEST_UNREACH:
  67         case ICMP_SOURCE_QUENCH:
  68         case ICMP_REDIRECT:
  69         case ICMP_TIME_EXCEEDED:
  70         case ICMP_PARAMETERPROB:
  71                 break;
  72         default:
  73                 return 1;
  74         }
  75
  76         inside_iph = skb_header_pointer(skb, outside_hdrlen +
  77                                         sizeof(struct icmphdr),
  78                                         sizeof(_inside_iph), &_inside_iph);
  79         if (inside_iph == NULL)
  80                 return 1;
  81
  82         if (inside_iph->protocol != IPPROTO_TCP &&
  83             inside_iph->protocol != IPPROTO_UDP)
  84                 return 1;
  85
  86         ports = skb_header_pointer(skb, outside_hdrlen +
  87                                    sizeof(struct icmphdr) +
  88                                    (inside_iph->ihl << 2),
  89                                    sizeof(_ports), &_ports);
  90         if (ports == NULL)
  91                 return 1;
  92
  93         /* the inside IP packet is the one quoted from our side, thus
  94          * its saddr is the local address */
  95         *protocol = inside_iph->protocol;
  96         *laddr = inside_iph->saddr;
  97         *lport = ports[0];
  98         *raddr = inside_iph->daddr;
  99         *rport = ports[1];
 100
 101         return 0;
 102 }
 103
 104 /* "socket" match based redirection (no specific rule)
 105  * ===================================================
 106  *
 107  * There are connections with dynamic endpoints (e.g. FTP data
 108  * connection) that the user is unable to add explicit rules
 109  * for. These are taken care of by a generic "socket" rule. It is
 110  * assumed that the proxy application is trusted to open such
 111  * connections without explicit iptables rule (except of course the
 112  * generic 'socket' rule). In this case the following sockets are
 113  * matched in preference order:
 114  *
 115  *   - match: if there's a fully established connection matching the
 116  *     _packet_ tuple
 117  *
 118  *   - match: if there's a non-zero bound listener (possibly with a
 119  *     non-local address) We don't accept zero-bound listeners, since
 120  *     then local services could intercept traffic going through the
 121  *     box.
 122  */
 123 static struct sock *
 124 xt_socket_get_sock_v4(struct net *net, const u8 protocol,
 125                       const __be32 saddr, const __be32 daddr,
 126                       const __be16 sport, const __be16 dport,
 127                       const struct net_device *in)
 128 {
 129         switch (protocol) {
 130         case IPPROTO_TCP:
 131                 return __inet_lookup(net, &tcp_hashinfo,
 132                                      saddr, sport, daddr, dport,
 133                                      in->ifindex);
 134         case IPPROTO_UDP:
 135                 return udp4_lib_lookup(net, saddr, sport, daddr, dport,
 136                                        in->ifindex);
 137         }
 138         return NULL;
 139 }
 140
 141 static bool
 142 socket_match(const struct sk_buff *skb, struct xt_action_param *par,
 143              const struct xt_socket_mtinfo1 *info)
 144 {
 145         const struct iphdr *iph = ip_hdr(skb);
 146         struct udphdr _hdr, *hp = NULL;
 147         struct sock *sk = skb->sk;
 148         __be32 uninitialized_var(daddr), uninitialized_var(saddr);
 149         __be16 uninitialized_var(dport), uninitialized_var(sport);
 150         u8 uninitialized_var(protocol);
 151 #ifdef XT_SOCKET_HAVE_CONNTRACK
 152         struct nf_conn const *ct;
 153         enum ip_conntrack_info ctinfo;
 154 #endif
 155
 156         if (iph->protocol == IPPROTO_UDP || iph->protocol == IPPROTO_TCP) {
 157                 hp = skb_header_pointer(skb, ip_hdrlen(skb),
 158                                         sizeof(_hdr), &_hdr);
 159                 if (hp == NULL)
 160                         return false;
 161
 162                 protocol = iph->protocol;
 163                 saddr = iph->saddr;
 164                 sport = hp->source;
 165                 daddr = iph->daddr;
 166                 dport = hp->dest;
 167
 168         } else if (iph->protocol == IPPROTO_ICMP) {
 169                 if (extract_icmp4_fields(skb, &protocol, &saddr, &daddr,
 170                                         &sport, &dport))
 171                         return false;
 172         } else {
 173                 return false;
 174         }
 175
 176 #ifdef XT_SOCKET_HAVE_CONNTRACK
 177         /* Do the lookup with the original socket address in case this is a
 178          * reply packet of an established SNAT-ted connection. */
 179
 180         ct = nf_ct_get(skb, &ctinfo);
 181         if (ct && !nf_ct_is_untracked(ct) &&
 182             ((iph->protocol != IPPROTO_ICMP &&
 183               ctinfo == IP_CT_ESTABLISHED_REPLY) ||
 184              (iph->protocol == IPPROTO_ICMP &&
 185               ctinfo == IP_CT_RELATED_REPLY)) &&
 186             (ct->status & IPS_SRC_NAT_DONE)) {
 187
 188                 daddr = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 189                 dport = (iph->protocol == IPPROTO_TCP) ?
 190                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.tcp.port :
 191                         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.udp.port;
 192         }
 193 #endif
 194
 195         if (!sk)
 196                 sk = xt_socket_get_sock_v4(dev_net(skb->dev), protocol,
 197                                            saddr, daddr, sport, dport,
 198                                            par->in);
 199         if (sk) {
 200                 bool wildcard;
 201                 bool transparent = true;
 202
 203                 /* Ignore sockets listening on INADDR_ANY,
 204                  * unless XT_SOCKET_NOWILDCARD is set
 205                  */
 206                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 207                             sk->sk_state != TCP_TIME_WAIT &&
 208                             inet_sk(sk)->inet_rcv_saddr == 0);
 209
 210                 /* Ignore non-transparent sockets,
 211                    if XT_SOCKET_TRANSPARENT is used */
 212                 if (info && info->flags & XT_SOCKET_TRANSPARENT)
 213                         transparent = ((sk->sk_state != TCP_TIME_WAIT &&
 214                                         inet_sk(sk)->transparent) ||
 215                                        (sk->sk_state == TCP_TIME_WAIT &&
 216                                         inet_twsk(sk)->tw_transparent));
 217
 218                 if (sk != skb->sk)
 219                         xt_socket_put_sk(sk);
 220
 221                 if (wildcard || !transparent)
 222                         sk = NULL;
 223         }
 224
 225         pr_debug("proto %hhu %pI4:%hu -> %pI4:%hu (orig %pI4:%hu) sock %p\n",
 226                  protocol, &saddr, ntohs(sport),
 227                  &daddr, ntohs(dport),
 228                  &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
 229
 230         return (sk != NULL);
 231 }
 232
 233 static bool
 234 socket_mt4_v0(const struct sk_buff *skb, struct xt_action_param *par)
 235 {
 236         return socket_match(skb, par, NULL);
 237 }
 238
 239 static bool
 240 socket_mt4_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
 241 {
 242         return socket_match(skb, par, par->matchinfo);
 243 }
 244
 245 #ifdef XT_SOCKET_HAVE_IPV6
 246
 247 static int
 248 extract_icmp6_fields(const struct sk_buff *skb,
 249                      unsigned int outside_hdrlen,
 250                      int *protocol,
 251                      struct in6_addr **raddr,
 252                      struct in6_addr **laddr,
 253                      __be16 *rport,
 254                      __be16 *lport)
 255 {
 256         struct ipv6hdr *inside_iph, _inside_iph;
 257         struct icmp6hdr *icmph, _icmph;
 258         __be16 *ports, _ports[2];
 259         u8 inside_nexthdr;
 260         __be16 inside_fragoff;
 261         int inside_hdrlen;
 262
 263         icmph = skb_header_pointer(skb, outside_hdrlen,
 264                                    sizeof(_icmph), &_icmph);
 265         if (icmph == NULL)
 266                 return 1;
 267
 268         if (icmph->icmp6_type & ICMPV6_INFOMSG_MASK)
 269                 return 1;
 270
 271         inside_iph = skb_header_pointer(skb, outside_hdrlen + sizeof(_icmph), sizeof(_inside_iph), &_inside_iph);
 272         if (inside_iph == NULL)
 273                 return 1;
 274         inside_nexthdr = inside_iph->nexthdr;
 275
 276         inside_hdrlen = ipv6_skip_exthdr(skb, outside_hdrlen + sizeof(_icmph) + sizeof(_inside_iph),
 277                                          &inside_nexthdr, &inside_fragoff);
 278         if (inside_hdrlen < 0)
 279                 return 1; /* hjm: Packet has no/incomplete transport layer headers. */
 280
 281         if (inside_nexthdr != IPPROTO_TCP &&
 282             inside_nexthdr != IPPROTO_UDP)
 283                 return 1;
 284
 285         ports = skb_header_pointer(skb, inside_hdrlen,
 286                                    sizeof(_ports), &_ports);
 287         if (ports == NULL)
 288                 return 1;
 289
 290         /* the inside IP packet is the one quoted from our side, thus
 291          * its saddr is the local address */
 292         *protocol = inside_nexthdr;
 293         *laddr = &inside_iph->saddr;
 294         *lport = ports[0];
 295         *raddr = &inside_iph->daddr;
 296         *rport = ports[1];
 297
 298         return 0;
 299 }
 300
 301 static struct sock *
 302 xt_socket_get_sock_v6(struct net *net, const u8 protocol,
 303                       const struct in6_addr *saddr, const struct in6_addr *daddr,
 304                       const __be16 sport, const __be16 dport,
 305                       const struct net_device *in)
 306 {
 307         switch (protocol) {
 308         case IPPROTO_TCP:
 309                 return inet6_lookup(net, &tcp_hashinfo,
 310                                     saddr, sport, daddr, dport,
 311                                     in->ifindex);
 312         case IPPROTO_UDP:
 313                 return udp6_lib_lookup(net, saddr, sport, daddr, dport,
 314                                        in->ifindex);
 315         }
 316
 317         return NULL;
 318 }
 319
 320 static bool
 321 socket_mt6_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
 322 {
 323         struct ipv6hdr *iph = ipv6_hdr(skb);
 324         struct udphdr _hdr, *hp = NULL;
 325         struct sock *sk = skb->sk;
 326         struct in6_addr *daddr = NULL, *saddr = NULL;
 327         __be16 uninitialized_var(dport), uninitialized_var(sport);
 328         int thoff = 0, uninitialized_var(tproto);
 329         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 330
 331         tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
 332         if (tproto < 0) {
 333                 pr_debug("unable to find transport header in IPv6 packet, dropping\n");
 334                 return NF_DROP;
 335         }
 336
 337         if (tproto == IPPROTO_UDP || tproto == IPPROTO_TCP) {
 338                 hp = skb_header_pointer(skb, thoff,
 339                                         sizeof(_hdr), &_hdr);
 340                 if (hp == NULL)
 341                         return false;
 342
 343                 saddr = &iph->saddr;
 344                 sport = hp->source;
 345                 daddr = &iph->daddr;
 346                 dport = hp->dest;
 347
 348         } else if (tproto == IPPROTO_ICMPV6) {
 349                 if (extract_icmp6_fields(skb, thoff, &tproto, &saddr, &daddr,
 350                                          &sport, &dport))
 351                         return false;
 352         } else {
 353                 return false;
 354         }
 355
 356         if (!sk)
 357                 sk = xt_socket_get_sock_v6(dev_net(skb->dev), tproto,
 358                                            saddr, daddr, sport, dport,
 359                                            par->in);
 360         if (sk) {
 361                 bool wildcard;
 362                 bool transparent = true;
 363
 364                 /* Ignore sockets listening on INADDR_ANY
 365                  * unless XT_SOCKET_NOWILDCARD is set
 366                  */
 367                 wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
 368                             sk->sk_state != TCP_TIME_WAIT &&
 369                             ipv6_addr_any(&inet6_sk(sk)->rcv_saddr));
 370
 371                 /* Ignore non-transparent sockets,
 372                    if XT_SOCKET_TRANSPARENT is used */
 373                 if (info && info->flags & XT_SOCKET_TRANSPARENT)
 374                         transparent = ((sk->sk_state != TCP_TIME_WAIT &&
 375                                         inet_sk(sk)->transparent) ||
 376                                        (sk->sk_state == TCP_TIME_WAIT &&
 377                                         inet_twsk(sk)->tw_transparent));
 378
 379                 if (sk != skb->sk)
 380                         xt_socket_put_sk(sk);
 381
 382                 if (wildcard || !transparent)
 383                         sk = NULL;
 384         }
 385
 386         pr_debug("proto %hhd %pI6:%hu -> %pI6:%hu "
 387                  "(orig %pI6:%hu) sock %p\n",
 388                  tproto, saddr, ntohs(sport),
 389                  daddr, ntohs(dport),
 390                  &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
 391
 392         return (sk != NULL);
 393 }
 394 #endif
 395
 396 static int socket_mt_v1_check(const struct xt_mtchk_param *par)
 397 {
 398         const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
 399
 400         if (info->flags & ~XT_SOCKET_FLAGS_V1) {
 401                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V1);
 402                 return -EINVAL;
 403         }
 404         return 0;
 405 }
 406
 407 static int socket_mt_v2_check(const struct xt_mtchk_param *par)
 408 {
 409         const struct xt_socket_mtinfo2 *info = (struct xt_socket_mtinfo2 *) par->matchinfo;
 410
 411         if (info->flags & ~XT_SOCKET_FLAGS_V2) {
 412                 pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V2);
 413                 return -EINVAL;
 414         }
 415         return 0;
 416 }
 417
 418 static struct xt_match socket_mt_reg[] __read_mostly = {
 419         {
 420                 .name           = "socket",
 421                 .revision       = 0,
 422                 .family         = NFPROTO_IPV4,
 423                 .match          = socket_mt4_v0,
 424                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 425                                   (1 << NF_INET_LOCAL_IN),
 426                 .me             = THIS_MODULE,
 427         },
 428         {
 429                 .name           = "socket",
 430                 .revision       = 1,
 431                 .family         = NFPROTO_IPV4,
 432                 .match          = socket_mt4_v1_v2,
 433                 .checkentry     = socket_mt_v1_check,
 434                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 435                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 436                                   (1 << NF_INET_LOCAL_IN),
 437                 .me             = THIS_MODULE,
 438         },
 439 #ifdef XT_SOCKET_HAVE_IPV6
 440         {
 441                 .name           = "socket",
 442                 .revision       = 1,
 443                 .family         = NFPROTO_IPV6,
 444                 .match          = socket_mt6_v1_v2,
 445                 .checkentry     = socket_mt_v1_check,
 446                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 447                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 448                                   (1 << NF_INET_LOCAL_IN),
 449                 .me             = THIS_MODULE,
 450         },
 451 #endif
 452         {
 453                 .name           = "socket",
 454                 .revision       = 2,
 455                 .family         = NFPROTO_IPV4,
 456                 .match          = socket_mt4_v1_v2,
 457                 .checkentry     = socket_mt_v2_check,
 458                 .matchsize      = sizeof(struct xt_socket_mtinfo1),
 459                 .hooks          = (1 << NF_INET_PRE_ROUTING) |
 460                                   (1 << NF_INET_LOCAL_IN),
 461                 .me             = THIS_MODULE,
 462         },
 463 #ifdef XT_SOCKET_HAVE_IPV6
 464         {
 465                 .name           = "socket",
 466                 .revision       = 2,
 467                 .family         = NFPROTO_IPV6,
 468                 .match          = socket_mt6_v1_v2,
 469                 .checkentry     = socket_mt_v2_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 #endif
 476 };
 477
 478 static int __init socket_mt_init(void)
 479 {
 480         nf_defrag_ipv4_enable();
 481 #ifdef XT_SOCKET_HAVE_IPV6
 482         nf_defrag_ipv6_enable();
 483 #endif
 484
 485         return xt_register_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 486 }
 487
 488 static void __exit socket_mt_exit(void)
 489 {
 490         xt_unregister_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
 491 }
 492
 493 module_init(socket_mt_init);
 494 module_exit(socket_mt_exit);
 495
 496 MODULE_LICENSE("GPL");
 497 MODULE_AUTHOR("Krisztian Kovacs, Balazs Scheidler");
 498 MODULE_DESCRIPTION("x_tables socket match module");
 499 MODULE_ALIAS("ipt_socket");
 500 MODULE_ALIAS("ip6t_socket");