1 /* 2 * Transparent proxy support for Linux/iptables 3 * 4 * Copyright (c) 2006-2010 BalaBit IT Ltd. 5 * Author: Balazs Scheidler, 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/ip.h> 16 #include <net/checksum.h> 17 #include <net/udp.h> 18 #include <net/tcp.h> 19 #include <net/inet_sock.h> 20 #include <net/inet_hashtables.h> 21 #include <linux/inetdevice.h> 22 #include <linux/netfilter/x_tables.h> 23 #include <linux/netfilter_ipv4/ip_tables.h> 24 25 #include <net/netfilter/ipv4/nf_defrag_ipv4.h> 26 27 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES) 28 #define XT_TPROXY_HAVE_IPV6 1 29 #include <net/if_inet6.h> 30 #include <net/addrconf.h> 31 #include <net/inet6_hashtables.h> 32 #include <linux/netfilter_ipv6/ip6_tables.h> 33 #include <net/netfilter/ipv6/nf_defrag_ipv6.h> 34 #endif 35 36 #include <linux/netfilter/xt_TPROXY.h> 37 38 enum nf_tproxy_lookup_t { 39 NFT_LOOKUP_LISTENER, 40 NFT_LOOKUP_ESTABLISHED, 41 }; 42 43 static bool tproxy_sk_is_transparent(struct sock *sk) 44 { 45 switch (sk->sk_state) { 46 case TCP_TIME_WAIT: 47 if (inet_twsk(sk)->tw_transparent) 48 return true; 49 break; 50 case TCP_NEW_SYN_RECV: 51 if (inet_rsk(inet_reqsk(sk))->no_srccheck) 52 return true; 53 break; 54 default: 55 if (inet_sk(sk)->transparent) 56 return true; 57 } 58 59 sock_gen_put(sk); 60 return false; 61 } 62 63 static inline __be32 64 tproxy_laddr4(struct sk_buff *skb, __be32 user_laddr, __be32 daddr) 65 { 66 struct in_device *indev; 67 __be32 laddr; 68 69 if (user_laddr) 70 return user_laddr; 71 72 laddr = 0; 73 rcu_read_lock(); 74 indev = __in_dev_get_rcu(skb->dev); 75 for_primary_ifa(indev) { 76 laddr = ifa->ifa_local; 77 break; 78 } endfor_ifa(indev); 79 rcu_read_unlock(); 80 81 return laddr ? laddr : daddr; 82 } 83 84 /* 85 * This is used when the user wants to intercept a connection matching 86 * an explicit iptables rule. In this case the sockets are assumed 87 * matching in preference order: 88 * 89 * - match: if there's a fully established connection matching the 90 * _packet_ tuple, it is returned, assuming the redirection 91 * already took place and we process a packet belonging to an 92 * established connection 93 * 94 * - match: if there's a listening socket matching the redirection 95 * (e.g. on-port & on-ip of the connection), it is returned, 96 * regardless if it was bound to 0.0.0.0 or an explicit 97 * address. The reasoning is that if there's an explicit rule, it 98 * does not really matter if the listener is bound to an interface 99 * or to 0. The user already stated that he wants redirection 100 * (since he added the rule). 101 * 102 * Please note that there's an overlap between what a TPROXY target 103 * and a socket match will match. Normally if you have both rules the 104 * "socket" match will be the first one, effectively all packets 105 * belonging to established connections going through that one. 106 */ 107 static inline struct sock * 108 nf_tproxy_get_sock_v4(struct net *net, struct sk_buff *skb, void *hp, 109 const u8 protocol, 110 const __be32 saddr, const __be32 daddr, 111 const __be16 sport, const __be16 dport, 112 const struct net_device *in, 113 const enum nf_tproxy_lookup_t lookup_type) 114 { 115 struct sock *sk; 116 struct tcphdr *tcph; 117 118 switch (protocol) { 119 case IPPROTO_TCP: 120 switch (lookup_type) { 121 case NFT_LOOKUP_LISTENER: 122 tcph = hp; 123 sk = inet_lookup_listener(net, &tcp_hashinfo, skb, 124 ip_hdrlen(skb) + 125 __tcp_hdrlen(tcph), 126 saddr, sport, 127 daddr, dport, 128 in->ifindex); 129 130 /* NOTE: we return listeners even if bound to 131 * 0.0.0.0, those are filtered out in 132 * xt_socket, since xt_TPROXY needs 0 bound 133 * listeners too 134 */ 135 break; 136 case NFT_LOOKUP_ESTABLISHED: 137 sk = inet_lookup_established(net, &tcp_hashinfo, 138 saddr, sport, daddr, dport, 139 in->ifindex); 140 break; 141 default: 142 BUG(); 143 } 144 break; 145 case IPPROTO_UDP: 146 sk = udp4_lib_lookup(net, saddr, sport, daddr, dport, 147 in->ifindex); 148 if (sk) { 149 int connected = (sk->sk_state == TCP_ESTABLISHED); 150 int wildcard = (inet_sk(sk)->inet_rcv_saddr == 0); 151 152 /* NOTE: we return listeners even if bound to 153 * 0.0.0.0, those are filtered out in 154 * xt_socket, since xt_TPROXY needs 0 bound 155 * listeners too 156 */ 157 if ((lookup_type == NFT_LOOKUP_ESTABLISHED && (!connected || wildcard)) || 158 (lookup_type == NFT_LOOKUP_LISTENER && connected)) { 159 sock_put(sk); 160 sk = NULL; 161 } 162 } 163 break; 164 default: 165 WARN_ON(1); 166 sk = NULL; 167 } 168 169 pr_debug("tproxy socket lookup: proto %u %08x:%u -> %08x:%u, lookup type: %d, sock %p\n", 170 protocol, ntohl(saddr), ntohs(sport), ntohl(daddr), ntohs(dport), lookup_type, sk); 171 172 return sk; 173 } 174 175 #ifdef XT_TPROXY_HAVE_IPV6 176 static inline struct sock * 177 nf_tproxy_get_sock_v6(struct net *net, struct sk_buff *skb, int thoff, void *hp, 178 const u8 protocol, 179 const struct in6_addr *saddr, const struct in6_addr *daddr, 180 const __be16 sport, const __be16 dport, 181 const struct net_device *in, 182 const enum nf_tproxy_lookup_t lookup_type) 183 { 184 struct sock *sk; 185 struct tcphdr *tcph; 186 187 switch (protocol) { 188 case IPPROTO_TCP: 189 switch (lookup_type) { 190 case NFT_LOOKUP_LISTENER: 191 tcph = hp; 192 sk = inet6_lookup_listener(net, &tcp_hashinfo, skb, 193 thoff + __tcp_hdrlen(tcph), 194 saddr, sport, 195 daddr, ntohs(dport), 196 in->ifindex); 197 198 /* NOTE: we return listeners even if bound to 199 * 0.0.0.0, those are filtered out in 200 * xt_socket, since xt_TPROXY needs 0 bound 201 * listeners too 202 */ 203 break; 204 case NFT_LOOKUP_ESTABLISHED: 205 sk = __inet6_lookup_established(net, &tcp_hashinfo, 206 saddr, sport, daddr, ntohs(dport), 207 in->ifindex); 208 break; 209 default: 210 BUG(); 211 } 212 break; 213 case IPPROTO_UDP: 214 sk = udp6_lib_lookup(net, saddr, sport, daddr, dport, 215 in->ifindex); 216 if (sk) { 217 int connected = (sk->sk_state == TCP_ESTABLISHED); 218 int wildcard = ipv6_addr_any(&sk->sk_v6_rcv_saddr); 219 220 /* NOTE: we return listeners even if bound to 221 * 0.0.0.0, those are filtered out in 222 * xt_socket, since xt_TPROXY needs 0 bound 223 * listeners too 224 */ 225 if ((lookup_type == NFT_LOOKUP_ESTABLISHED && (!connected || wildcard)) || 226 (lookup_type == NFT_LOOKUP_LISTENER && connected)) { 227 sock_put(sk); 228 sk = NULL; 229 } 230 } 231 break; 232 default: 233 WARN_ON(1); 234 sk = NULL; 235 } 236 237 pr_debug("tproxy socket lookup: proto %u %pI6:%u -> %pI6:%u, lookup type: %d, sock %p\n", 238 protocol, saddr, ntohs(sport), daddr, ntohs(dport), lookup_type, sk); 239 240 return sk; 241 } 242 #endif 243 244 /** 245 * tproxy_handle_time_wait4 - handle IPv4 TCP TIME_WAIT reopen redirections 246 * @skb: The skb being processed. 247 * @laddr: IPv4 address to redirect to or zero. 248 * @lport: TCP port to redirect to or zero. 249 * @sk: The TIME_WAIT TCP socket found by the lookup. 250 * 251 * We have to handle SYN packets arriving to TIME_WAIT sockets 252 * differently: instead of reopening the connection we should rather 253 * redirect the new connection to the proxy if there's a listener 254 * socket present. 255 * 256 * tproxy_handle_time_wait4() consumes the socket reference passed in. 257 * 258 * Returns the listener socket if there's one, the TIME_WAIT socket if 259 * no such listener is found, or NULL if the TCP header is incomplete. 260 */ 261 static struct sock * 262 tproxy_handle_time_wait4(struct net *net, struct sk_buff *skb, 263 __be32 laddr, __be16 lport, struct sock *sk) 264 { 265 const struct iphdr *iph = ip_hdr(skb); 266 struct tcphdr _hdr, *hp; 267 268 hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr); 269 if (hp == NULL) { 270 inet_twsk_put(inet_twsk(sk)); 271 return NULL; 272 } 273 274 if (hp->syn && !hp->rst && !hp->ack && !hp->fin) { 275 /* SYN to a TIME_WAIT socket, we'd rather redirect it 276 * to a listener socket if there's one */ 277 struct sock *sk2; 278 279 sk2 = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol, 280 iph->saddr, laddr ? laddr : iph->daddr, 281 hp->source, lport ? lport : hp->dest, 282 skb->dev, NFT_LOOKUP_LISTENER); 283 if (sk2) { 284 inet_twsk_deschedule_put(inet_twsk(sk)); 285 sk = sk2; 286 } 287 } 288 289 return sk; 290 } 291 292 /* assign a socket to the skb -- consumes sk */ 293 static void 294 nf_tproxy_assign_sock(struct sk_buff *skb, struct sock *sk) 295 { 296 skb_orphan(skb); 297 skb->sk = sk; 298 skb->destructor = sock_edemux; 299 } 300 301 static unsigned int 302 tproxy_tg4(struct net *net, struct sk_buff *skb, __be32 laddr, __be16 lport, 303 u_int32_t mark_mask, u_int32_t mark_value) 304 { 305 const struct iphdr *iph = ip_hdr(skb); 306 struct udphdr _hdr, *hp; 307 struct sock *sk; 308 309 hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr); 310 if (hp == NULL) 311 return NF_DROP; 312 313 /* check if there's an ongoing connection on the packet 314 * addresses, this happens if the redirect already happened 315 * and the current packet belongs to an already established 316 * connection */ 317 sk = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol, 318 iph->saddr, iph->daddr, 319 hp->source, hp->dest, 320 skb->dev, NFT_LOOKUP_ESTABLISHED); 321 322 laddr = tproxy_laddr4(skb, laddr, iph->daddr); 323 if (!lport) 324 lport = hp->dest; 325 326 /* UDP has no TCP_TIME_WAIT state, so we never enter here */ 327 if (sk && sk->sk_state == TCP_TIME_WAIT) 328 /* reopening a TIME_WAIT connection needs special handling */ 329 sk = tproxy_handle_time_wait4(net, skb, laddr, lport, sk); 330 else if (!sk) 331 /* no, there's no established connection, check if 332 * there's a listener on the redirected addr/port */ 333 sk = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol, 334 iph->saddr, laddr, 335 hp->source, lport, 336 skb->dev, NFT_LOOKUP_LISTENER); 337 338 /* NOTE: assign_sock consumes our sk reference */ 339 if (sk && tproxy_sk_is_transparent(sk)) { 340 /* This should be in a separate target, but we don't do multiple 341 targets on the same rule yet */ 342 skb->mark = (skb->mark & ~mark_mask) ^ mark_value; 343 344 pr_debug("redirecting: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n", 345 iph->protocol, &iph->daddr, ntohs(hp->dest), 346 &laddr, ntohs(lport), skb->mark); 347 348 nf_tproxy_assign_sock(skb, sk); 349 return NF_ACCEPT; 350 } 351 352 pr_debug("no socket, dropping: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n", 353 iph->protocol, &iph->saddr, ntohs(hp->source), 354 &iph->daddr, ntohs(hp->dest), skb->mark); 355 return NF_DROP; 356 } 357 358 static unsigned int 359 tproxy_tg4_v0(struct sk_buff *skb, const struct xt_action_param *par) 360 { 361 const struct xt_tproxy_target_info *tgi = par->targinfo; 362 363 return tproxy_tg4(par->net, skb, tgi->laddr, tgi->lport, tgi->mark_mask, tgi->mark_value); 364 } 365 366 static unsigned int 367 tproxy_tg4_v1(struct sk_buff *skb, const struct xt_action_param *par) 368 { 369 const struct xt_tproxy_target_info_v1 *tgi = par->targinfo; 370 371 return tproxy_tg4(par->net, skb, tgi->laddr.ip, tgi->lport, tgi->mark_mask, tgi->mark_value); 372 } 373 374 #ifdef XT_TPROXY_HAVE_IPV6 375 376 static inline const struct in6_addr * 377 tproxy_laddr6(struct sk_buff *skb, const struct in6_addr *user_laddr, 378 const struct in6_addr *daddr) 379 { 380 struct inet6_dev *indev; 381 struct inet6_ifaddr *ifa; 382 struct in6_addr *laddr; 383 384 if (!ipv6_addr_any(user_laddr)) 385 return user_laddr; 386 laddr = NULL; 387 388 rcu_read_lock(); 389 indev = __in6_dev_get(skb->dev); 390 if (indev) 391 list_for_each_entry(ifa, &indev->addr_list, if_list) { 392 if (ifa->flags & (IFA_F_TENTATIVE | IFA_F_DEPRECATED)) 393 continue; 394 395 laddr = &ifa->addr; 396 break; 397 } 398 rcu_read_unlock(); 399 400 return laddr ? laddr : daddr; 401 } 402 403 /** 404 * tproxy_handle_time_wait6 - handle IPv6 TCP TIME_WAIT reopen redirections 405 * @skb: The skb being processed. 406 * @tproto: Transport protocol. 407 * @thoff: Transport protocol header offset. 408 * @par: Iptables target parameters. 409 * @sk: The TIME_WAIT TCP socket found by the lookup. 410 * 411 * We have to handle SYN packets arriving to TIME_WAIT sockets 412 * differently: instead of reopening the connection we should rather 413 * redirect the new connection to the proxy if there's a listener 414 * socket present. 415 * 416 * tproxy_handle_time_wait6() consumes the socket reference passed in. 417 * 418 * Returns the listener socket if there's one, the TIME_WAIT socket if 419 * no such listener is found, or NULL if the TCP header is incomplete. 420 */ 421 static struct sock * 422 tproxy_handle_time_wait6(struct sk_buff *skb, int tproto, int thoff, 423 const struct xt_action_param *par, 424 struct sock *sk) 425 { 426 const struct ipv6hdr *iph = ipv6_hdr(skb); 427 struct tcphdr _hdr, *hp; 428 const struct xt_tproxy_target_info_v1 *tgi = par->targinfo; 429 430 hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr); 431 if (hp == NULL) { 432 inet_twsk_put(inet_twsk(sk)); 433 return NULL; 434 } 435 436 if (hp->syn && !hp->rst && !hp->ack && !hp->fin) { 437 /* SYN to a TIME_WAIT socket, we'd rather redirect it 438 * to a listener socket if there's one */ 439 struct sock *sk2; 440 441 sk2 = nf_tproxy_get_sock_v6(par->net, skb, thoff, hp, tproto, 442 &iph->saddr, 443 tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr), 444 hp->source, 445 tgi->lport ? tgi->lport : hp->dest, 446 skb->dev, NFT_LOOKUP_LISTENER); 447 if (sk2) { 448 inet_twsk_deschedule_put(inet_twsk(sk)); 449 sk = sk2; 450 } 451 } 452 453 return sk; 454 } 455 456 static unsigned int 457 tproxy_tg6_v1(struct sk_buff *skb, const struct xt_action_param *par) 458 { 459 const struct ipv6hdr *iph = ipv6_hdr(skb); 460 const struct xt_tproxy_target_info_v1 *tgi = par->targinfo; 461 struct udphdr _hdr, *hp; 462 struct sock *sk; 463 const struct in6_addr *laddr; 464 __be16 lport; 465 int thoff = 0; 466 int tproto; 467 468 tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL); 469 if (tproto < 0) { 470 pr_debug("unable to find transport header in IPv6 packet, dropping\n"); 471 return NF_DROP; 472 } 473 474 hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr); 475 if (hp == NULL) { 476 pr_debug("unable to grab transport header contents in IPv6 packet, dropping\n"); 477 return NF_DROP; 478 } 479 480 /* check if there's an ongoing connection on the packet 481 * addresses, this happens if the redirect already happened 482 * and the current packet belongs to an already established 483 * connection */ 484 sk = nf_tproxy_get_sock_v6(par->net, skb, thoff, hp, tproto, 485 &iph->saddr, &iph->daddr, 486 hp->source, hp->dest, 487 par->in, NFT_LOOKUP_ESTABLISHED); 488 489 laddr = tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr); 490 lport = tgi->lport ? tgi->lport : hp->dest; 491 492 /* UDP has no TCP_TIME_WAIT state, so we never enter here */ 493 if (sk && sk->sk_state == TCP_TIME_WAIT) 494 /* reopening a TIME_WAIT connection needs special handling */ 495 sk = tproxy_handle_time_wait6(skb, tproto, thoff, par, sk); 496 else if (!sk) 497 /* no there's no established connection, check if 498 * there's a listener on the redirected addr/port */ 499 sk = nf_tproxy_get_sock_v6(par->net, skb, thoff, hp, 500 tproto, &iph->saddr, laddr, 501 hp->source, lport, 502 par->in, NFT_LOOKUP_LISTENER); 503 504 /* NOTE: assign_sock consumes our sk reference */ 505 if (sk && tproxy_sk_is_transparent(sk)) { 506 /* This should be in a separate target, but we don't do multiple 507 targets on the same rule yet */ 508 skb->mark = (skb->mark & ~tgi->mark_mask) ^ tgi->mark_value; 509 510 pr_debug("redirecting: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n", 511 tproto, &iph->saddr, ntohs(hp->source), 512 laddr, ntohs(lport), skb->mark); 513 514 nf_tproxy_assign_sock(skb, sk); 515 return NF_ACCEPT; 516 } 517 518 pr_debug("no socket, dropping: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n", 519 tproto, &iph->saddr, ntohs(hp->source), 520 &iph->daddr, ntohs(hp->dest), skb->mark); 521 522 return NF_DROP; 523 } 524 525 static int tproxy_tg6_check(const struct xt_tgchk_param *par) 526 { 527 const struct ip6t_ip6 *i = par->entryinfo; 528 529 if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP) && 530 !(i->invflags & IP6T_INV_PROTO)) 531 return 0; 532 533 pr_info("Can be used only in combination with " 534 "either -p tcp or -p udp\n"); 535 return -EINVAL; 536 } 537 #endif 538 539 static int tproxy_tg4_check(const struct xt_tgchk_param *par) 540 { 541 const struct ipt_ip *i = par->entryinfo; 542 543 if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP) 544 && !(i->invflags & IPT_INV_PROTO)) 545 return 0; 546 547 pr_info("Can be used only in combination with " 548 "either -p tcp or -p udp\n"); 549 return -EINVAL; 550 } 551 552 static struct xt_target tproxy_tg_reg[] __read_mostly = { 553 { 554 .name = "TPROXY", 555 .family = NFPROTO_IPV4, 556 .table = "mangle", 557 .target = tproxy_tg4_v0, 558 .revision = 0, 559 .targetsize = sizeof(struct xt_tproxy_target_info), 560 .checkentry = tproxy_tg4_check, 561 .hooks = 1 << NF_INET_PRE_ROUTING, 562 .me = THIS_MODULE, 563 }, 564 { 565 .name = "TPROXY", 566 .family = NFPROTO_IPV4, 567 .table = "mangle", 568 .target = tproxy_tg4_v1, 569 .revision = 1, 570 .targetsize = sizeof(struct xt_tproxy_target_info_v1), 571 .checkentry = tproxy_tg4_check, 572 .hooks = 1 << NF_INET_PRE_ROUTING, 573 .me = THIS_MODULE, 574 }, 575 #ifdef XT_TPROXY_HAVE_IPV6 576 { 577 .name = "TPROXY", 578 .family = NFPROTO_IPV6, 579 .table = "mangle", 580 .target = tproxy_tg6_v1, 581 .revision = 1, 582 .targetsize = sizeof(struct xt_tproxy_target_info_v1), 583 .checkentry = tproxy_tg6_check, 584 .hooks = 1 << NF_INET_PRE_ROUTING, 585 .me = THIS_MODULE, 586 }, 587 #endif 588 589 }; 590 591 static int __init tproxy_tg_init(void) 592 { 593 nf_defrag_ipv4_enable(); 594 #ifdef XT_TPROXY_HAVE_IPV6 595 nf_defrag_ipv6_enable(); 596 #endif 597 598 return xt_register_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg)); 599 } 600 601 static void __exit tproxy_tg_exit(void) 602 { 603 xt_unregister_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg)); 604 } 605 606 module_init(tproxy_tg_init); 607 module_exit(tproxy_tg_exit); 608 MODULE_LICENSE("GPL"); 609 MODULE_AUTHOR("Balazs Scheidler, Krisztian Kovacs"); 610 MODULE_DESCRIPTION("Netfilter transparent proxy (TPROXY) target module."); 611 MODULE_ALIAS("ipt_TPROXY"); 612 MODULE_ALIAS("ip6t_TPROXY"); 613