xref: /linux-6.15/net/netfilter/xt_socket.c (revision bbeddf52)
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 && 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 	return socket_match(skb, par, NULL);
200 }
201 
202 static bool
203 socket_mt4_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
204 {
205 	return socket_match(skb, par, par->matchinfo);
206 }
207 
208 #ifdef XT_SOCKET_HAVE_IPV6
209 
210 static int
211 extract_icmp6_fields(const struct sk_buff *skb,
212 		     unsigned int outside_hdrlen,
213 		     int *protocol,
214 		     struct in6_addr **raddr,
215 		     struct in6_addr **laddr,
216 		     __be16 *rport,
217 		     __be16 *lport)
218 {
219 	struct ipv6hdr *inside_iph, _inside_iph;
220 	struct icmp6hdr *icmph, _icmph;
221 	__be16 *ports, _ports[2];
222 	u8 inside_nexthdr;
223 	__be16 inside_fragoff;
224 	int inside_hdrlen;
225 
226 	icmph = skb_header_pointer(skb, outside_hdrlen,
227 				   sizeof(_icmph), &_icmph);
228 	if (icmph == NULL)
229 		return 1;
230 
231 	if (icmph->icmp6_type & ICMPV6_INFOMSG_MASK)
232 		return 1;
233 
234 	inside_iph = skb_header_pointer(skb, outside_hdrlen + sizeof(_icmph), sizeof(_inside_iph), &_inside_iph);
235 	if (inside_iph == NULL)
236 		return 1;
237 	inside_nexthdr = inside_iph->nexthdr;
238 
239 	inside_hdrlen = ipv6_skip_exthdr(skb, outside_hdrlen + sizeof(_icmph) + sizeof(_inside_iph),
240 					 &inside_nexthdr, &inside_fragoff);
241 	if (inside_hdrlen < 0)
242 		return 1; /* hjm: Packet has no/incomplete transport layer headers. */
243 
244 	if (inside_nexthdr != IPPROTO_TCP &&
245 	    inside_nexthdr != IPPROTO_UDP)
246 		return 1;
247 
248 	ports = skb_header_pointer(skb, inside_hdrlen,
249 				   sizeof(_ports), &_ports);
250 	if (ports == NULL)
251 		return 1;
252 
253 	/* the inside IP packet is the one quoted from our side, thus
254 	 * its saddr is the local address */
255 	*protocol = inside_nexthdr;
256 	*laddr = &inside_iph->saddr;
257 	*lport = ports[0];
258 	*raddr = &inside_iph->daddr;
259 	*rport = ports[1];
260 
261 	return 0;
262 }
263 
264 static bool
265 socket_mt6_v1_v2(const struct sk_buff *skb, struct xt_action_param *par)
266 {
267 	struct ipv6hdr *iph = ipv6_hdr(skb);
268 	struct udphdr _hdr, *hp = NULL;
269 	struct sock *sk = skb->sk;
270 	struct in6_addr *daddr = NULL, *saddr = NULL;
271 	__be16 uninitialized_var(dport), uninitialized_var(sport);
272 	int thoff = 0, uninitialized_var(tproto);
273 	const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
274 
275 	tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
276 	if (tproto < 0) {
277 		pr_debug("unable to find transport header in IPv6 packet, dropping\n");
278 		return NF_DROP;
279 	}
280 
281 	if (tproto == IPPROTO_UDP || tproto == IPPROTO_TCP) {
282 		hp = skb_header_pointer(skb, thoff,
283 					sizeof(_hdr), &_hdr);
284 		if (hp == NULL)
285 			return false;
286 
287 		saddr = &iph->saddr;
288 		sport = hp->source;
289 		daddr = &iph->daddr;
290 		dport = hp->dest;
291 
292 	} else if (tproto == IPPROTO_ICMPV6) {
293 		if (extract_icmp6_fields(skb, thoff, &tproto, &saddr, &daddr,
294 					 &sport, &dport))
295 			return false;
296 	} else {
297 		return false;
298 	}
299 
300 	if (!sk)
301 		sk = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
302 					   saddr, daddr, sport, dport,
303 					   par->in, NFT_LOOKUP_ANY);
304 	if (sk) {
305 		bool wildcard;
306 		bool transparent = true;
307 
308 		/* Ignore sockets listening on INADDR_ANY
309 		 * unless XT_SOCKET_NOWILDCARD is set
310 		 */
311 		wildcard = (!(info->flags & XT_SOCKET_NOWILDCARD) &&
312 			    sk->sk_state != TCP_TIME_WAIT &&
313 			    ipv6_addr_any(&inet6_sk(sk)->rcv_saddr));
314 
315 		/* Ignore non-transparent sockets,
316 		   if XT_SOCKET_TRANSPARENT is used */
317 		if (info && info->flags & XT_SOCKET_TRANSPARENT)
318 			transparent = ((sk->sk_state != TCP_TIME_WAIT &&
319 					inet_sk(sk)->transparent) ||
320 				       (sk->sk_state == TCP_TIME_WAIT &&
321 					inet_twsk(sk)->tw_transparent));
322 
323 		if (sk != skb->sk)
324 			xt_socket_put_sk(sk);
325 
326 		if (wildcard || !transparent)
327 			sk = NULL;
328 	}
329 
330 	pr_debug("proto %hhd %pI6:%hu -> %pI6:%hu "
331 		 "(orig %pI6:%hu) sock %p\n",
332 		 tproto, saddr, ntohs(sport),
333 		 daddr, ntohs(dport),
334 		 &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
335 
336 	return (sk != NULL);
337 }
338 #endif
339 
340 static int socket_mt_v1_check(const struct xt_mtchk_param *par)
341 {
342 	const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
343 
344 	if (info->flags & ~XT_SOCKET_FLAGS_V1) {
345 		pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V1);
346 		return -EINVAL;
347 	}
348 	return 0;
349 }
350 
351 static int socket_mt_v2_check(const struct xt_mtchk_param *par)
352 {
353 	const struct xt_socket_mtinfo2 *info = (struct xt_socket_mtinfo2 *) par->matchinfo;
354 
355 	if (info->flags & ~XT_SOCKET_FLAGS_V2) {
356 		pr_info("unknown flags 0x%x\n", info->flags & ~XT_SOCKET_FLAGS_V2);
357 		return -EINVAL;
358 	}
359 	return 0;
360 }
361 
362 static struct xt_match socket_mt_reg[] __read_mostly = {
363 	{
364 		.name		= "socket",
365 		.revision	= 0,
366 		.family		= NFPROTO_IPV4,
367 		.match		= socket_mt4_v0,
368 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
369 				  (1 << NF_INET_LOCAL_IN),
370 		.me		= THIS_MODULE,
371 	},
372 	{
373 		.name		= "socket",
374 		.revision	= 1,
375 		.family		= NFPROTO_IPV4,
376 		.match		= socket_mt4_v1_v2,
377 		.checkentry	= socket_mt_v1_check,
378 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
379 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
380 				  (1 << NF_INET_LOCAL_IN),
381 		.me		= THIS_MODULE,
382 	},
383 #ifdef XT_SOCKET_HAVE_IPV6
384 	{
385 		.name		= "socket",
386 		.revision	= 1,
387 		.family		= NFPROTO_IPV6,
388 		.match		= socket_mt6_v1_v2,
389 		.checkentry	= socket_mt_v1_check,
390 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
391 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
392 				  (1 << NF_INET_LOCAL_IN),
393 		.me		= THIS_MODULE,
394 	},
395 #endif
396 	{
397 		.name		= "socket",
398 		.revision	= 2,
399 		.family		= NFPROTO_IPV4,
400 		.match		= socket_mt4_v1_v2,
401 		.checkentry	= socket_mt_v2_check,
402 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
403 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
404 				  (1 << NF_INET_LOCAL_IN),
405 		.me		= THIS_MODULE,
406 	},
407 #ifdef XT_SOCKET_HAVE_IPV6
408 	{
409 		.name		= "socket",
410 		.revision	= 2,
411 		.family		= NFPROTO_IPV6,
412 		.match		= socket_mt6_v1_v2,
413 		.checkentry	= socket_mt_v2_check,
414 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
415 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
416 				  (1 << NF_INET_LOCAL_IN),
417 		.me		= THIS_MODULE,
418 	},
419 #endif
420 };
421 
422 static int __init socket_mt_init(void)
423 {
424 	nf_defrag_ipv4_enable();
425 #ifdef XT_SOCKET_HAVE_IPV6
426 	nf_defrag_ipv6_enable();
427 #endif
428 
429 	return xt_register_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
430 }
431 
432 static void __exit socket_mt_exit(void)
433 {
434 	xt_unregister_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
435 }
436 
437 module_init(socket_mt_init);
438 module_exit(socket_mt_exit);
439 
440 MODULE_LICENSE("GPL");
441 MODULE_AUTHOR("Krisztian Kovacs, Balazs Scheidler");
442 MODULE_DESCRIPTION("x_tables socket match module");
443 MODULE_ALIAS("ipt_socket");
444 MODULE_ALIAS("ip6t_socket");
445