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/inet_sock.h>
19#include <linux/inetdevice.h>
20#include <linux/netfilter/x_tables.h>
21#include <linux/netfilter_ipv4/ip_tables.h>
22
23#include <net/netfilter/ipv4/nf_defrag_ipv4.h>
24
25#if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
26#define XT_TPROXY_HAVE_IPV6 1
27#include <net/if_inet6.h>
28#include <net/addrconf.h>
29#include <linux/netfilter_ipv6/ip6_tables.h>
30#include <net/netfilter/ipv6/nf_defrag_ipv6.h>
31#endif
32
33#include <net/netfilter/nf_tproxy_core.h>
34#include <linux/netfilter/xt_TPROXY.h>
35
36static bool tproxy_sk_is_transparent(struct sock *sk)
37{
38	if (sk->sk_state != TCP_TIME_WAIT) {
39		if (inet_sk(sk)->transparent)
40			return true;
41		sock_put(sk);
42	} else {
43		if (inet_twsk(sk)->tw_transparent)
44			return true;
45		inet_twsk_put(inet_twsk(sk));
46	}
47	return false;
48}
49
50static inline __be32
51tproxy_laddr4(struct sk_buff *skb, __be32 user_laddr, __be32 daddr)
52{
53	struct in_device *indev;
54	__be32 laddr;
55
56	if (user_laddr)
57		return user_laddr;
58
59	laddr = 0;
60	rcu_read_lock();
61	indev = __in_dev_get_rcu(skb->dev);
62	for_primary_ifa(indev) {
63		laddr = ifa->ifa_local;
64		break;
65	} endfor_ifa(indev);
66	rcu_read_unlock();
67
68	return laddr ? laddr : daddr;
69}
70
71/**
72 * tproxy_handle_time_wait4 - handle IPv4 TCP TIME_WAIT reopen redirections
73 * @skb:	The skb being processed.
74 * @laddr:	IPv4 address to redirect to or zero.
75 * @lport:	TCP port to redirect to or zero.
76 * @sk:		The TIME_WAIT TCP socket found by the lookup.
77 *
78 * We have to handle SYN packets arriving to TIME_WAIT sockets
79 * differently: instead of reopening the connection we should rather
80 * redirect the new connection to the proxy if there's a listener
81 * socket present.
82 *
83 * tproxy_handle_time_wait4() consumes the socket reference passed in.
84 *
85 * Returns the listener socket if there's one, the TIME_WAIT socket if
86 * no such listener is found, or NULL if the TCP header is incomplete.
87 */
88static struct sock *
89tproxy_handle_time_wait4(struct sk_buff *skb, __be32 laddr, __be16 lport,
90			struct sock *sk)
91{
92	const struct iphdr *iph = ip_hdr(skb);
93	struct tcphdr _hdr, *hp;
94
95	hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
96	if (hp == NULL) {
97		inet_twsk_put(inet_twsk(sk));
98		return NULL;
99	}
100
101	if (hp->syn && !hp->rst && !hp->ack && !hp->fin) {
102		/* SYN to a TIME_WAIT socket, we'd rather redirect it
103		 * to a listener socket if there's one */
104		struct sock *sk2;
105
106		sk2 = nf_tproxy_get_sock_v4(dev_net(skb->dev), iph->protocol,
107					    iph->saddr, laddr ? laddr : iph->daddr,
108					    hp->source, lport ? lport : hp->dest,
109					    skb->dev, NFT_LOOKUP_LISTENER);
110		if (sk2) {
111			inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
112			inet_twsk_put(inet_twsk(sk));
113			sk = sk2;
114		}
115	}
116
117	return sk;
118}
119
120static unsigned int
121tproxy_tg4(struct sk_buff *skb, __be32 laddr, __be16 lport,
122	   u_int32_t mark_mask, u_int32_t mark_value)
123{
124	const struct iphdr *iph = ip_hdr(skb);
125	struct udphdr _hdr, *hp;
126	struct sock *sk;
127
128	hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
129	if (hp == NULL)
130		return NF_DROP;
131
132	/* check if there's an ongoing connection on the packet
133	 * addresses, this happens if the redirect already happened
134	 * and the current packet belongs to an already established
135	 * connection */
136	sk = nf_tproxy_get_sock_v4(dev_net(skb->dev), iph->protocol,
137				   iph->saddr, iph->daddr,
138				   hp->source, hp->dest,
139				   skb->dev, NFT_LOOKUP_ESTABLISHED);
140
141	laddr = tproxy_laddr4(skb, laddr, iph->daddr);
142	if (!lport)
143		lport = hp->dest;
144
145	/* UDP has no TCP_TIME_WAIT state, so we never enter here */
146	if (sk && sk->sk_state == TCP_TIME_WAIT)
147		/* reopening a TIME_WAIT connection needs special handling */
148		sk = tproxy_handle_time_wait4(skb, laddr, lport, sk);
149	else if (!sk)
150		/* no, there's no established connection, check if
151		 * there's a listener on the redirected addr/port */
152		sk = nf_tproxy_get_sock_v4(dev_net(skb->dev), iph->protocol,
153					   iph->saddr, laddr,
154					   hp->source, lport,
155					   skb->dev, NFT_LOOKUP_LISTENER);
156
157	/* NOTE: assign_sock consumes our sk reference */
158	if (sk && tproxy_sk_is_transparent(sk)) {
159		/* This should be in a separate target, but we don't do multiple
160		   targets on the same rule yet */
161		skb->mark = (skb->mark & ~mark_mask) ^ mark_value;
162
163		pr_debug("redirecting: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n",
164			 iph->protocol, &iph->daddr, ntohs(hp->dest),
165			 &laddr, ntohs(lport), skb->mark);
166
167		nf_tproxy_assign_sock(skb, sk);
168		return NF_ACCEPT;
169	}
170
171	pr_debug("no socket, dropping: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n",
172		 iph->protocol, &iph->saddr, ntohs(hp->source),
173		 &iph->daddr, ntohs(hp->dest), skb->mark);
174	return NF_DROP;
175}
176
177static unsigned int
178tproxy_tg4_v0(struct sk_buff *skb, const struct xt_action_param *par)
179{
180	const struct xt_tproxy_target_info *tgi = par->targinfo;
181
182	return tproxy_tg4(skb, tgi->laddr, tgi->lport, tgi->mark_mask, tgi->mark_value);
183}
184
185static unsigned int
186tproxy_tg4_v1(struct sk_buff *skb, const struct xt_action_param *par)
187{
188	const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
189
190	return tproxy_tg4(skb, tgi->laddr.ip, tgi->lport, tgi->mark_mask, tgi->mark_value);
191}
192
193#ifdef XT_TPROXY_HAVE_IPV6
194
195static inline const struct in6_addr *
196tproxy_laddr6(struct sk_buff *skb, const struct in6_addr *user_laddr,
197	      const struct in6_addr *daddr)
198{
199	struct inet6_dev *indev;
200	struct inet6_ifaddr *ifa;
201	struct in6_addr *laddr;
202
203	if (!ipv6_addr_any(user_laddr))
204		return user_laddr;
205	laddr = NULL;
206
207	rcu_read_lock();
208	indev = __in6_dev_get(skb->dev);
209	if (indev)
210		list_for_each_entry(ifa, &indev->addr_list, if_list) {
211			if (ifa->flags & (IFA_F_TENTATIVE | IFA_F_DEPRECATED))
212				continue;
213
214			laddr = &ifa->addr;
215			break;
216		}
217	rcu_read_unlock();
218
219	return laddr ? laddr : daddr;
220}
221
222/**
223 * tproxy_handle_time_wait6 - handle IPv6 TCP TIME_WAIT reopen redirections
224 * @skb:	The skb being processed.
225 * @tproto:	Transport protocol.
226 * @thoff:	Transport protocol header offset.
227 * @par:	Iptables target parameters.
228 * @sk:		The TIME_WAIT TCP socket found by the lookup.
229 *
230 * We have to handle SYN packets arriving to TIME_WAIT sockets
231 * differently: instead of reopening the connection we should rather
232 * redirect the new connection to the proxy if there's a listener
233 * socket present.
234 *
235 * tproxy_handle_time_wait6() consumes the socket reference passed in.
236 *
237 * Returns the listener socket if there's one, the TIME_WAIT socket if
238 * no such listener is found, or NULL if the TCP header is incomplete.
239 */
240static struct sock *
241tproxy_handle_time_wait6(struct sk_buff *skb, int tproto, int thoff,
242			 const struct xt_action_param *par,
243			 struct sock *sk)
244{
245	const struct ipv6hdr *iph = ipv6_hdr(skb);
246	struct tcphdr _hdr, *hp;
247	const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
248
249	hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
250	if (hp == NULL) {
251		inet_twsk_put(inet_twsk(sk));
252		return NULL;
253	}
254
255	if (hp->syn && !hp->rst && !hp->ack && !hp->fin) {
256		/* SYN to a TIME_WAIT socket, we'd rather redirect it
257		 * to a listener socket if there's one */
258		struct sock *sk2;
259
260		sk2 = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
261					    &iph->saddr,
262					    tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr),
263					    hp->source,
264					    tgi->lport ? tgi->lport : hp->dest,
265					    skb->dev, NFT_LOOKUP_LISTENER);
266		if (sk2) {
267			inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
268			inet_twsk_put(inet_twsk(sk));
269			sk = sk2;
270		}
271	}
272
273	return sk;
274}
275
276static unsigned int
277tproxy_tg6_v1(struct sk_buff *skb, const struct xt_action_param *par)
278{
279	const struct ipv6hdr *iph = ipv6_hdr(skb);
280	const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
281	struct udphdr _hdr, *hp;
282	struct sock *sk;
283	const struct in6_addr *laddr;
284	__be16 lport;
285	int thoff = 0;
286	int tproto;
287
288	tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
289	if (tproto < 0) {
290		pr_debug("unable to find transport header in IPv6 packet, dropping\n");
291		return NF_DROP;
292	}
293
294	hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
295	if (hp == NULL) {
296		pr_debug("unable to grab transport header contents in IPv6 packet, dropping\n");
297		return NF_DROP;
298	}
299
300	/* check if there's an ongoing connection on the packet
301	 * addresses, this happens if the redirect already happened
302	 * and the current packet belongs to an already established
303	 * connection */
304	sk = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
305				   &iph->saddr, &iph->daddr,
306				   hp->source, hp->dest,
307				   par->in, NFT_LOOKUP_ESTABLISHED);
308
309	laddr = tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr);
310	lport = tgi->lport ? tgi->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_wait6(skb, tproto, thoff, par, 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_v6(dev_net(skb->dev), tproto,
320					   &iph->saddr, laddr,
321					   hp->source, lport,
322					   par->in, 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 & ~tgi->mark_mask) ^ tgi->mark_value;
329
330		pr_debug("redirecting: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n",
331			 tproto, &iph->saddr, ntohs(hp->source),
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 %pI6:%hu -> %pI6:%hu, mark: %x\n",
339		 tproto, &iph->saddr, ntohs(hp->source),
340		 &iph->daddr, ntohs(hp->dest), skb->mark);
341
342	return NF_DROP;
343}
344
345static int tproxy_tg6_check(const struct xt_tgchk_param *par)
346{
347	const struct ip6t_ip6 *i = par->entryinfo;
348
349	if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP)
350	    && !(i->flags & IP6T_INV_PROTO))
351		return 0;
352
353	pr_info("Can be used only in combination with "
354		"either -p tcp or -p udp\n");
355	return -EINVAL;
356}
357#endif
358
359static int tproxy_tg4_check(const struct xt_tgchk_param *par)
360{
361	const struct ipt_ip *i = par->entryinfo;
362
363	if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP)
364	    && !(i->invflags & IPT_INV_PROTO))
365		return 0;
366
367	pr_info("Can be used only in combination with "
368		"either -p tcp or -p udp\n");
369	return -EINVAL;
370}
371
372static struct xt_target tproxy_tg_reg[] __read_mostly = {
373	{
374		.name		= "TPROXY",
375		.family		= NFPROTO_IPV4,
376		.table		= "mangle",
377		.target		= tproxy_tg4_v0,
378		.revision	= 0,
379		.targetsize	= sizeof(struct xt_tproxy_target_info),
380		.checkentry	= tproxy_tg4_check,
381		.hooks		= 1 << NF_INET_PRE_ROUTING,
382		.me		= THIS_MODULE,
383	},
384	{
385		.name		= "TPROXY",
386		.family		= NFPROTO_IPV4,
387		.table		= "mangle",
388		.target		= tproxy_tg4_v1,
389		.revision	= 1,
390		.targetsize	= sizeof(struct xt_tproxy_target_info_v1),
391		.checkentry	= tproxy_tg4_check,
392		.hooks		= 1 << NF_INET_PRE_ROUTING,
393		.me		= THIS_MODULE,
394	},
395#ifdef XT_TPROXY_HAVE_IPV6
396	{
397		.name		= "TPROXY",
398		.family		= NFPROTO_IPV6,
399		.table		= "mangle",
400		.target		= tproxy_tg6_v1,
401		.revision	= 1,
402		.targetsize	= sizeof(struct xt_tproxy_target_info_v1),
403		.checkentry	= tproxy_tg6_check,
404		.hooks		= 1 << NF_INET_PRE_ROUTING,
405		.me		= THIS_MODULE,
406	},
407#endif
408
409};
410
411static int __init tproxy_tg_init(void)
412{
413	nf_defrag_ipv4_enable();
414#ifdef XT_TPROXY_HAVE_IPV6
415	nf_defrag_ipv6_enable();
416#endif
417
418	return xt_register_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg));
419}
420
421static void __exit tproxy_tg_exit(void)
422{
423	xt_unregister_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg));
424}
425
426module_init(tproxy_tg_init);
427module_exit(tproxy_tg_exit);
428MODULE_LICENSE("GPL");
429MODULE_AUTHOR("Balazs Scheidler, Krisztian Kovacs");
430MODULE_DESCRIPTION("Netfilter transparent proxy (TPROXY) target module.");
431MODULE_ALIAS("ipt_TPROXY");
432MODULE_ALIAS("ip6t_TPROXY");
433