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