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 indev = __in_dev_get_rcu(skb->dev); 74 for_primary_ifa(indev) { 75 laddr = ifa->ifa_local; 76 break; 77 } endfor_ifa(indev); 78 79 return laddr ? laddr : daddr; 80 } 81 82 /* 83 * This is used when the user wants to intercept a connection matching 84 * an explicit iptables rule. In this case the sockets are assumed 85 * matching in preference order: 86 * 87 * - match: if there's a fully established connection matching the 88 * _packet_ tuple, it is returned, assuming the redirection 89 * already took place and we process a packet belonging to an 90 * established connection 91 * 92 * - match: if there's a listening socket matching the redirection 93 * (e.g. on-port & on-ip of the connection), it is returned, 94 * regardless if it was bound to 0.0.0.0 or an explicit 95 * address. The reasoning is that if there's an explicit rule, it 96 * does not really matter if the listener is bound to an interface 97 * or to 0. The user already stated that he wants redirection 98 * (since he added the rule). 99 * 100 * Please note that there's an overlap between what a TPROXY target 101 * and a socket match will match. Normally if you have both rules the 102 * "socket" match will be the first one, effectively all packets 103 * belonging to established connections going through that one. 104 */ 105 static inline struct sock * 106 nf_tproxy_get_sock_v4(struct net *net, struct sk_buff *skb, void *hp, 107 const u8 protocol, 108 const __be32 saddr, const __be32 daddr, 109 const __be16 sport, const __be16 dport, 110 const struct net_device *in, 111 const enum nf_tproxy_lookup_t lookup_type) 112 { 113 struct sock *sk; 114 struct tcphdr *tcph; 115 116 switch (protocol) { 117 case IPPROTO_TCP: 118 switch (lookup_type) { 119 case NFT_LOOKUP_LISTENER: 120 tcph = hp; 121 sk = inet_lookup_listener(net, &tcp_hashinfo, skb, 122 ip_hdrlen(skb) + 123 __tcp_hdrlen(tcph), 124 saddr, sport, 125 daddr, dport, 126 in->ifindex, 0); 127 128 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 129 sk = NULL; 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, 0); 197 198 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 199 sk = NULL; 200 /* NOTE: we return listeners even if bound to 201 * 0.0.0.0, those are filtered out in 202 * xt_socket, since xt_TPROXY needs 0 bound 203 * listeners too 204 */ 205 break; 206 case NFT_LOOKUP_ESTABLISHED: 207 sk = __inet6_lookup_established(net, &tcp_hashinfo, 208 saddr, sport, daddr, ntohs(dport), 209 in->ifindex, 0); 210 break; 211 default: 212 BUG(); 213 } 214 break; 215 case IPPROTO_UDP: 216 sk = udp6_lib_lookup(net, saddr, sport, daddr, dport, 217 in->ifindex); 218 if (sk) { 219 int connected = (sk->sk_state == TCP_ESTABLISHED); 220 int wildcard = ipv6_addr_any(&sk->sk_v6_rcv_saddr); 221 222 /* NOTE: we return listeners even if bound to 223 * 0.0.0.0, those are filtered out in 224 * xt_socket, since xt_TPROXY needs 0 bound 225 * listeners too 226 */ 227 if ((lookup_type == NFT_LOOKUP_ESTABLISHED && (!connected || wildcard)) || 228 (lookup_type == NFT_LOOKUP_LISTENER && connected)) { 229 sock_put(sk); 230 sk = NULL; 231 } 232 } 233 break; 234 default: 235 WARN_ON(1); 236 sk = NULL; 237 } 238 239 pr_debug("tproxy socket lookup: proto %u %pI6:%u -> %pI6:%u, lookup type: %d, sock %p\n", 240 protocol, saddr, ntohs(sport), daddr, ntohs(dport), lookup_type, sk); 241 242 return sk; 243 } 244 #endif 245 246 /** 247 * tproxy_handle_time_wait4 - handle IPv4 TCP TIME_WAIT reopen redirections 248 * @skb: The skb being processed. 249 * @laddr: IPv4 address to redirect to or zero. 250 * @lport: TCP port to redirect to or zero. 251 * @sk: The TIME_WAIT TCP socket found by the lookup. 252 * 253 * We have to handle SYN packets arriving to TIME_WAIT sockets 254 * differently: instead of reopening the connection we should rather 255 * redirect the new connection to the proxy if there's a listener 256 * socket present. 257 * 258 * tproxy_handle_time_wait4() consumes the socket reference passed in. 259 * 260 * Returns the listener socket if there's one, the TIME_WAIT socket if 261 * no such listener is found, or NULL if the TCP header is incomplete. 262 */ 263 static struct sock * 264 tproxy_handle_time_wait4(struct net *net, struct sk_buff *skb, 265 __be32 laddr, __be16 lport, struct sock *sk) 266 { 267 const struct iphdr *iph = ip_hdr(skb); 268 struct tcphdr _hdr, *hp; 269 270 hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr); 271 if (hp == NULL) { 272 inet_twsk_put(inet_twsk(sk)); 273 return NULL; 274 } 275 276 if (hp->syn && !hp->rst && !hp->ack && !hp->fin) { 277 /* SYN to a TIME_WAIT socket, we'd rather redirect it 278 * to a listener socket if there's one */ 279 struct sock *sk2; 280 281 sk2 = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol, 282 iph->saddr, laddr ? laddr : iph->daddr, 283 hp->source, lport ? lport : hp->dest, 284 skb->dev, NFT_LOOKUP_LISTENER); 285 if (sk2) { 286 inet_twsk_deschedule_put(inet_twsk(sk)); 287 sk = sk2; 288 } 289 } 290 291 return sk; 292 } 293 294 /* assign a socket to the skb -- consumes sk */ 295 static void 296 nf_tproxy_assign_sock(struct sk_buff *skb, struct sock *sk) 297 { 298 skb_orphan(skb); 299 skb->sk = sk; 300 skb->destructor = sock_edemux; 301 } 302 303 static unsigned int 304 tproxy_tg4(struct net *net, struct sk_buff *skb, __be32 laddr, __be16 lport, 305 u_int32_t mark_mask, u_int32_t mark_value) 306 { 307 const struct iphdr *iph = ip_hdr(skb); 308 struct udphdr _hdr, *hp; 309 struct sock *sk; 310 311 hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr); 312 if (hp == NULL) 313 return NF_DROP; 314 315 /* check if there's an ongoing connection on the packet 316 * addresses, this happens if the redirect already happened 317 * and the current packet belongs to an already established 318 * connection */ 319 sk = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol, 320 iph->saddr, iph->daddr, 321 hp->source, hp->dest, 322 skb->dev, NFT_LOOKUP_ESTABLISHED); 323 324 laddr = tproxy_laddr4(skb, laddr, iph->daddr); 325 if (!lport) 326 lport = hp->dest; 327 328 /* UDP has no TCP_TIME_WAIT state, so we never enter here */ 329 if (sk && sk->sk_state == TCP_TIME_WAIT) 330 /* reopening a TIME_WAIT connection needs special handling */ 331 sk = tproxy_handle_time_wait4(net, skb, laddr, lport, sk); 332 else if (!sk) 333 /* no, there's no established connection, check if 334 * there's a listener on the redirected addr/port */ 335 sk = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol, 336 iph->saddr, laddr, 337 hp->source, lport, 338 skb->dev, NFT_LOOKUP_LISTENER); 339 340 /* NOTE: assign_sock consumes our sk reference */ 341 if (sk && tproxy_sk_is_transparent(sk)) { 342 /* This should be in a separate target, but we don't do multiple 343 targets on the same rule yet */ 344 skb->mark = (skb->mark & ~mark_mask) ^ mark_value; 345 346 pr_debug("redirecting: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n", 347 iph->protocol, &iph->daddr, ntohs(hp->dest), 348 &laddr, ntohs(lport), skb->mark); 349 350 nf_tproxy_assign_sock(skb, sk); 351 return NF_ACCEPT; 352 } 353 354 pr_debug("no socket, dropping: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n", 355 iph->protocol, &iph->saddr, ntohs(hp->source), 356 &iph->daddr, ntohs(hp->dest), skb->mark); 357 return NF_DROP; 358 } 359 360 static unsigned int 361 tproxy_tg4_v0(struct sk_buff *skb, const struct xt_action_param *par) 362 { 363 const struct xt_tproxy_target_info *tgi = par->targinfo; 364 365 return tproxy_tg4(xt_net(par), skb, tgi->laddr, tgi->lport, 366 tgi->mark_mask, tgi->mark_value); 367 } 368 369 static unsigned int 370 tproxy_tg4_v1(struct sk_buff *skb, const struct xt_action_param *par) 371 { 372 const struct xt_tproxy_target_info_v1 *tgi = par->targinfo; 373 374 return tproxy_tg4(xt_net(par), skb, tgi->laddr.ip, tgi->lport, 375 tgi->mark_mask, tgi->mark_value); 376 } 377 378 #ifdef XT_TPROXY_HAVE_IPV6 379 380 static inline const struct in6_addr * 381 tproxy_laddr6(struct sk_buff *skb, const struct in6_addr *user_laddr, 382 const struct in6_addr *daddr) 383 { 384 struct inet6_dev *indev; 385 struct inet6_ifaddr *ifa; 386 struct in6_addr *laddr; 387 388 if (!ipv6_addr_any(user_laddr)) 389 return user_laddr; 390 laddr = NULL; 391 392 indev = __in6_dev_get(skb->dev); 393 if (indev) { 394 read_lock_bh(&indev->lock); 395 list_for_each_entry(ifa, &indev->addr_list, if_list) { 396 if (ifa->flags & (IFA_F_TENTATIVE | IFA_F_DEPRECATED)) 397 continue; 398 399 laddr = &ifa->addr; 400 break; 401 } 402 read_unlock_bh(&indev->lock); 403 } 404 405 return laddr ? laddr : daddr; 406 } 407 408 /** 409 * tproxy_handle_time_wait6 - handle IPv6 TCP TIME_WAIT reopen redirections 410 * @skb: The skb being processed. 411 * @tproto: Transport protocol. 412 * @thoff: Transport protocol header offset. 413 * @par: Iptables target parameters. 414 * @sk: The TIME_WAIT TCP socket found by the lookup. 415 * 416 * We have to handle SYN packets arriving to TIME_WAIT sockets 417 * differently: instead of reopening the connection we should rather 418 * redirect the new connection to the proxy if there's a listener 419 * socket present. 420 * 421 * tproxy_handle_time_wait6() consumes the socket reference passed in. 422 * 423 * Returns the listener socket if there's one, the TIME_WAIT socket if 424 * no such listener is found, or NULL if the TCP header is incomplete. 425 */ 426 static struct sock * 427 tproxy_handle_time_wait6(struct sk_buff *skb, int tproto, int thoff, 428 const struct xt_action_param *par, 429 struct sock *sk) 430 { 431 const struct ipv6hdr *iph = ipv6_hdr(skb); 432 struct tcphdr _hdr, *hp; 433 const struct xt_tproxy_target_info_v1 *tgi = par->targinfo; 434 435 hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr); 436 if (hp == NULL) { 437 inet_twsk_put(inet_twsk(sk)); 438 return NULL; 439 } 440 441 if (hp->syn && !hp->rst && !hp->ack && !hp->fin) { 442 /* SYN to a TIME_WAIT socket, we'd rather redirect it 443 * to a listener socket if there's one */ 444 struct sock *sk2; 445 446 sk2 = nf_tproxy_get_sock_v6(xt_net(par), skb, thoff, hp, tproto, 447 &iph->saddr, 448 tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr), 449 hp->source, 450 tgi->lport ? tgi->lport : hp->dest, 451 skb->dev, NFT_LOOKUP_LISTENER); 452 if (sk2) { 453 inet_twsk_deschedule_put(inet_twsk(sk)); 454 sk = sk2; 455 } 456 } 457 458 return sk; 459 } 460 461 static unsigned int 462 tproxy_tg6_v1(struct sk_buff *skb, const struct xt_action_param *par) 463 { 464 const struct ipv6hdr *iph = ipv6_hdr(skb); 465 const struct xt_tproxy_target_info_v1 *tgi = par->targinfo; 466 struct udphdr _hdr, *hp; 467 struct sock *sk; 468 const struct in6_addr *laddr; 469 __be16 lport; 470 int thoff = 0; 471 int tproto; 472 473 tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL); 474 if (tproto < 0) { 475 pr_debug("unable to find transport header in IPv6 packet, dropping\n"); 476 return NF_DROP; 477 } 478 479 hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr); 480 if (hp == NULL) { 481 pr_debug("unable to grab transport header contents in IPv6 packet, dropping\n"); 482 return NF_DROP; 483 } 484 485 /* check if there's an ongoing connection on the packet 486 * addresses, this happens if the redirect already happened 487 * and the current packet belongs to an already established 488 * connection */ 489 sk = nf_tproxy_get_sock_v6(xt_net(par), skb, thoff, hp, tproto, 490 &iph->saddr, &iph->daddr, 491 hp->source, hp->dest, 492 xt_in(par), NFT_LOOKUP_ESTABLISHED); 493 494 laddr = tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr); 495 lport = tgi->lport ? tgi->lport : hp->dest; 496 497 /* UDP has no TCP_TIME_WAIT state, so we never enter here */ 498 if (sk && sk->sk_state == TCP_TIME_WAIT) 499 /* reopening a TIME_WAIT connection needs special handling */ 500 sk = tproxy_handle_time_wait6(skb, tproto, thoff, par, sk); 501 else if (!sk) 502 /* no there's no established connection, check if 503 * there's a listener on the redirected addr/port */ 504 sk = nf_tproxy_get_sock_v6(xt_net(par), skb, thoff, hp, 505 tproto, &iph->saddr, laddr, 506 hp->source, lport, 507 xt_in(par), NFT_LOOKUP_LISTENER); 508 509 /* NOTE: assign_sock consumes our sk reference */ 510 if (sk && tproxy_sk_is_transparent(sk)) { 511 /* This should be in a separate target, but we don't do multiple 512 targets on the same rule yet */ 513 skb->mark = (skb->mark & ~tgi->mark_mask) ^ tgi->mark_value; 514 515 pr_debug("redirecting: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n", 516 tproto, &iph->saddr, ntohs(hp->source), 517 laddr, ntohs(lport), skb->mark); 518 519 nf_tproxy_assign_sock(skb, sk); 520 return NF_ACCEPT; 521 } 522 523 pr_debug("no socket, dropping: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n", 524 tproto, &iph->saddr, ntohs(hp->source), 525 &iph->daddr, ntohs(hp->dest), skb->mark); 526 527 return NF_DROP; 528 } 529 530 static int tproxy_tg6_check(const struct xt_tgchk_param *par) 531 { 532 const struct ip6t_ip6 *i = par->entryinfo; 533 int err; 534 535 err = nf_defrag_ipv6_enable(par->net); 536 if (err) 537 return err; 538 539 if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP) && 540 !(i->invflags & IP6T_INV_PROTO)) 541 return 0; 542 543 pr_info_ratelimited("Can be used only with -p tcp or -p udp\n"); 544 return -EINVAL; 545 } 546 #endif 547 548 static int tproxy_tg4_check(const struct xt_tgchk_param *par) 549 { 550 const struct ipt_ip *i = par->entryinfo; 551 int err; 552 553 err = nf_defrag_ipv4_enable(par->net); 554 if (err) 555 return err; 556 557 if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP) 558 && !(i->invflags & IPT_INV_PROTO)) 559 return 0; 560 561 pr_info_ratelimited("Can be used only with -p tcp or -p udp\n"); 562 return -EINVAL; 563 } 564 565 static struct xt_target tproxy_tg_reg[] __read_mostly = { 566 { 567 .name = "TPROXY", 568 .family = NFPROTO_IPV4, 569 .table = "mangle", 570 .target = tproxy_tg4_v0, 571 .revision = 0, 572 .targetsize = sizeof(struct xt_tproxy_target_info), 573 .checkentry = tproxy_tg4_check, 574 .hooks = 1 << NF_INET_PRE_ROUTING, 575 .me = THIS_MODULE, 576 }, 577 { 578 .name = "TPROXY", 579 .family = NFPROTO_IPV4, 580 .table = "mangle", 581 .target = tproxy_tg4_v1, 582 .revision = 1, 583 .targetsize = sizeof(struct xt_tproxy_target_info_v1), 584 .checkentry = tproxy_tg4_check, 585 .hooks = 1 << NF_INET_PRE_ROUTING, 586 .me = THIS_MODULE, 587 }, 588 #ifdef XT_TPROXY_HAVE_IPV6 589 { 590 .name = "TPROXY", 591 .family = NFPROTO_IPV6, 592 .table = "mangle", 593 .target = tproxy_tg6_v1, 594 .revision = 1, 595 .targetsize = sizeof(struct xt_tproxy_target_info_v1), 596 .checkentry = tproxy_tg6_check, 597 .hooks = 1 << NF_INET_PRE_ROUTING, 598 .me = THIS_MODULE, 599 }, 600 #endif 601 602 }; 603 604 static int __init tproxy_tg_init(void) 605 { 606 return xt_register_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg)); 607 } 608 609 static void __exit tproxy_tg_exit(void) 610 { 611 xt_unregister_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg)); 612 } 613 614 module_init(tproxy_tg_init); 615 module_exit(tproxy_tg_exit); 616 MODULE_LICENSE("GPL"); 617 MODULE_AUTHOR("Balazs Scheidler, Krisztian Kovacs"); 618 MODULE_DESCRIPTION("Netfilter transparent proxy (TPROXY) target module."); 619 MODULE_ALIAS("ipt_TPROXY"); 620 MODULE_ALIAS("ip6t_TPROXY"); 621