xref: /linux-6.15/net/netfilter/xt_socket.c (revision f5e4e7fd)
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/nf_tproxy_core.h>
23 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
24 
25 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
26 #define XT_SOCKET_HAVE_IPV6 1
27 #include <linux/netfilter_ipv6/ip6_tables.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 static bool
105 socket_match(const struct sk_buff *skb, struct xt_action_param *par,
106 	     const struct xt_socket_mtinfo1 *info)
107 {
108 	const struct iphdr *iph = ip_hdr(skb);
109 	struct udphdr _hdr, *hp = NULL;
110 	struct sock *sk = skb->sk;
111 	__be32 uninitialized_var(daddr), uninitialized_var(saddr);
112 	__be16 uninitialized_var(dport), uninitialized_var(sport);
113 	u8 uninitialized_var(protocol);
114 #ifdef XT_SOCKET_HAVE_CONNTRACK
115 	struct nf_conn const *ct;
116 	enum ip_conntrack_info ctinfo;
117 #endif
118 
119 	if (iph->protocol == IPPROTO_UDP || iph->protocol == IPPROTO_TCP) {
120 		hp = skb_header_pointer(skb, ip_hdrlen(skb),
121 					sizeof(_hdr), &_hdr);
122 		if (hp == NULL)
123 			return false;
124 
125 		protocol = iph->protocol;
126 		saddr = iph->saddr;
127 		sport = hp->source;
128 		daddr = iph->daddr;
129 		dport = hp->dest;
130 
131 	} else if (iph->protocol == IPPROTO_ICMP) {
132 		if (extract_icmp4_fields(skb, &protocol, &saddr, &daddr,
133 					&sport, &dport))
134 			return false;
135 	} else {
136 		return false;
137 	}
138 
139 #ifdef XT_SOCKET_HAVE_CONNTRACK
140 	/* Do the lookup with the original socket address in case this is a
141 	 * reply packet of an established SNAT-ted connection. */
142 
143 	ct = nf_ct_get(skb, &ctinfo);
144 	if (ct && !nf_ct_is_untracked(ct) &&
145 	    ((iph->protocol != IPPROTO_ICMP &&
146 	      ctinfo == IP_CT_ESTABLISHED_REPLY) ||
147 	     (iph->protocol == IPPROTO_ICMP &&
148 	      ctinfo == IP_CT_RELATED_REPLY)) &&
149 	    (ct->status & IPS_SRC_NAT_DONE)) {
150 
151 		daddr = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
152 		dport = (iph->protocol == IPPROTO_TCP) ?
153 			ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.tcp.port :
154 			ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.udp.port;
155 	}
156 #endif
157 
158 	if (!sk)
159 		sk = nf_tproxy_get_sock_v4(dev_net(skb->dev), protocol,
160 					   saddr, daddr, sport, dport,
161 					   par->in, NFT_LOOKUP_ANY);
162 	if (sk) {
163 		bool wildcard;
164 		bool transparent = true;
165 
166 		/* Ignore sockets listening on INADDR_ANY,
167 		 * unless XT_SOCKET_NOWILDCARD is set
168 		 */
169 		wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
170 			    sk->sk_state != TCP_TIME_WAIT &&
171 			    inet_sk(sk)->inet_rcv_saddr == 0);
172 
173 		/* Ignore non-transparent sockets,
174 		   if XT_SOCKET_TRANSPARENT is used */
175 		if (info->flags & XT_SOCKET_TRANSPARENT)
176 			transparent = ((sk->sk_state != TCP_TIME_WAIT &&
177 					inet_sk(sk)->transparent) ||
178 				       (sk->sk_state == TCP_TIME_WAIT &&
179 					inet_twsk(sk)->tw_transparent));
180 
181 		if (sk != skb->sk)
182 			xt_socket_put_sk(sk);
183 
184 		if (wildcard || !transparent)
185 			sk = NULL;
186 	}
187 
188 	pr_debug("proto %hhu %pI4:%hu -> %pI4:%hu (orig %pI4:%hu) sock %p\n",
189 		 protocol, &saddr, ntohs(sport),
190 		 &daddr, ntohs(dport),
191 		 &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
192 
193 	return (sk != NULL);
194 }
195 
196 static bool
197 socket_mt4_v0(const struct sk_buff *skb, struct xt_action_param *par)
198 {
199 	static struct xt_socket_mtinfo1 xt_info_v0 = {
200 		.flags = 0,
201 	};
202 
203 	return socket_match(skb, par, &xt_info_v0);
204 }
205 
206 static bool
207 socket_mt4_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
208 {
209 	return socket_match(skb, par, par->matchinfo);
210 }
211 
212 #ifdef XT_SOCKET_HAVE_IPV6
213 
214 static int
215 extract_icmp6_fields(const struct sk_buff *skb,
216 		     unsigned int outside_hdrlen,
217 		     int *protocol,
218 		     struct in6_addr **raddr,
219 		     struct in6_addr **laddr,
220 		     __be16 *rport,
221 		     __be16 *lport)
222 {
223 	struct ipv6hdr *inside_iph, _inside_iph;
224 	struct icmp6hdr *icmph, _icmph;
225 	__be16 *ports, _ports[2];
226 	u8 inside_nexthdr;
227 	__be16 inside_fragoff;
228 	int inside_hdrlen;
229 
230 	icmph = skb_header_pointer(skb, outside_hdrlen,
231 				   sizeof(_icmph), &_icmph);
232 	if (icmph == NULL)
233 		return 1;
234 
235 	if (icmph->icmp6_type & ICMPV6_INFOMSG_MASK)
236 		return 1;
237 
238 	inside_iph = skb_header_pointer(skb, outside_hdrlen + sizeof(_icmph), sizeof(_inside_iph), &_inside_iph);
239 	if (inside_iph == NULL)
240 		return 1;
241 	inside_nexthdr = inside_iph->nexthdr;
242 
243 	inside_hdrlen = ipv6_skip_exthdr(skb, outside_hdrlen + sizeof(_icmph) + sizeof(_inside_iph),
244 					 &inside_nexthdr, &inside_fragoff);
245 	if (inside_hdrlen < 0)
246 		return 1; /* hjm: Packet has no/incomplete transport layer headers. */
247 
248 	if (inside_nexthdr != IPPROTO_TCP &&
249 	    inside_nexthdr != IPPROTO_UDP)
250 		return 1;
251 
252 	ports = skb_header_pointer(skb, inside_hdrlen,
253 				   sizeof(_ports), &_ports);
254 	if (ports == NULL)
255 		return 1;
256 
257 	/* the inside IP packet is the one quoted from our side, thus
258 	 * its saddr is the local address */
259 	*protocol = inside_nexthdr;
260 	*laddr = &inside_iph->saddr;
261 	*lport = ports[0];
262 	*raddr = &inside_iph->daddr;
263 	*rport = ports[1];
264 
265 	return 0;
266 }
267 
268 static bool
269 socket_mt6_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
270 {
271 	struct ipv6hdr *iph = ipv6_hdr(skb);
272 	struct udphdr _hdr, *hp = NULL;
273 	struct sock *sk = skb->sk;
274 	struct in6_addr *daddr = NULL, *saddr = NULL;
275 	__be16 uninitialized_var(dport), uninitialized_var(sport);
276 	int thoff = 0, uninitialized_var(tproto);
277 	const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
278 
279 	tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
280 	if (tproto < 0) {
281 		pr_debug("unable to find transport header in IPv6 packet, dropping\n");
282 		return NF_DROP;
283 	}
284 
285 	if (tproto == IPPROTO_UDP || tproto == IPPROTO_TCP) {
286 		hp = skb_header_pointer(skb, thoff,
287 					sizeof(_hdr), &_hdr);
288 		if (hp == NULL)
289 			return false;
290 
291 		saddr = &iph->saddr;
292 		sport = hp->source;
293 		daddr = &iph->daddr;
294 		dport = hp->dest;
295 
296 	} else if (tproto == IPPROTO_ICMPV6) {
297 		if (extract_icmp6_fields(skb, thoff, &tproto, &saddr, &daddr,
298 					 &sport, &dport))
299 			return false;
300 	} else {
301 		return false;
302 	}
303 
304 	if (!sk)
305 		sk = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
306 					   saddr, daddr, sport, dport,
307 					   par->in, NFT_LOOKUP_ANY);
308 	if (sk) {
309 		bool wildcard;
310 		bool transparent = true;
311 
312 		/* Ignore sockets listening on INADDR_ANY
313 		 * unless XT_SOCKET_NOWILDCARD is set
314 		 */
315 		wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
316 			    sk->sk_state != TCP_TIME_WAIT &&
317 			    ipv6_addr_any(&inet6_sk(sk)->rcv_saddr));
318 
319 		/* Ignore non-transparent sockets,
320 		   if XT_SOCKET_TRANSPARENT is used */
321 		if (info->flags & XT_SOCKET_TRANSPARENT)
322 			transparent = ((sk->sk_state != TCP_TIME_WAIT &&
323 					inet_sk(sk)->transparent) ||
324 				       (sk->sk_state == TCP_TIME_WAIT &&
325 					inet_twsk(sk)->tw_transparent));
326 
327 		if (sk != skb->sk)
328 			xt_socket_put_sk(sk);
329 
330 		if (wildcard || !transparent)
331 			sk = NULL;
332 	}
333 
334 	pr_debug("proto %hhd %pI6:%hu -> %pI6:%hu "
335 		 "(orig %pI6:%hu) sock %p\n",
336 		 tproto, saddr, ntohs(sport),
337 		 daddr, ntohs(dport),
338 		 &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
339 
340 	return (sk != NULL);
341 }
342 #endif
343 
344 static int socket_mt_v1_check(const struct xt_mtchk_param *par)
345 {
346 	const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
347 
348 	if (info->flags & ~XT_SOCKET_FLAGS_V1) {
349 		pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V1);
350 		return -EINVAL;
351 	}
352 	return 0;
353 }
354 
355 static int socket_mt_v2_check(const struct xt_mtchk_param *par)
356 {
357 	const struct xt_socket_mtinfo2 *info = (struct xt_socket_mtinfo2 *) par->matchinfo;
358 
359 	if (info->flags & ~XT_SOCKET_FLAGS_V2) {
360 		pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V2);
361 		return -EINVAL;
362 	}
363 	return 0;
364 }
365 
366 static struct xt_match socket_mt_reg[] __read_mostly = {
367 	{
368 		.name		= "socket",
369 		.revision	= 0,
370 		.family		= NFPROTO_IPV4,
371 		.match		= socket_mt4_v0,
372 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
373 				  (1 << NF_INET_LOCAL_IN),
374 		.me		= THIS_MODULE,
375 	},
376 	{
377 		.name		= "socket",
378 		.revision	= 1,
379 		.family		= NFPROTO_IPV4,
380 		.match		= socket_mt4_v1_v2,
381 		.checkentry	= socket_mt_v1_check,
382 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
383 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
384 				  (1 << NF_INET_LOCAL_IN),
385 		.me		= THIS_MODULE,
386 	},
387 #ifdef XT_SOCKET_HAVE_IPV6
388 	{
389 		.name		= "socket",
390 		.revision	= 1,
391 		.family		= NFPROTO_IPV6,
392 		.match		= socket_mt6_v1_v2,
393 		.checkentry	= socket_mt_v1_check,
394 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
395 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
396 				  (1 << NF_INET_LOCAL_IN),
397 		.me		= THIS_MODULE,
398 	},
399 #endif
400 	{
401 		.name		= "socket",
402 		.revision	= 2,
403 		.family		= NFPROTO_IPV4,
404 		.match		= socket_mt4_v1_v2,
405 		.checkentry	= socket_mt_v2_check,
406 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
407 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
408 				  (1 << NF_INET_LOCAL_IN),
409 		.me		= THIS_MODULE,
410 	},
411 #ifdef XT_SOCKET_HAVE_IPV6
412 	{
413 		.name		= "socket",
414 		.revision	= 2,
415 		.family		= NFPROTO_IPV6,
416 		.match		= socket_mt6_v1_v2,
417 		.checkentry	= socket_mt_v2_check,
418 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
419 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
420 				  (1 << NF_INET_LOCAL_IN),
421 		.me		= THIS_MODULE,
422 	},
423 #endif
424 };
425 
426 static int __init socket_mt_init(void)
427 {
428 	nf_defrag_ipv4_enable();
429 #ifdef XT_SOCKET_HAVE_IPV6
430 	nf_defrag_ipv6_enable();
431 #endif
432 
433 	return xt_register_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
434 }
435 
436 static void __exit socket_mt_exit(void)
437 {
438 	xt_unregister_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
439 }
440 
441 module_init(socket_mt_init);
442 module_exit(socket_mt_exit);
443 
444 MODULE_LICENSE("GPL");
445 MODULE_AUTHOR("Krisztian Kovacs, Balazs Scheidler");
446 MODULE_DESCRIPTION("x_tables socket match module");
447 MODULE_ALIAS("ipt_socket");
448 MODULE_ALIAS("ip6t_socket");
449