xref: /openbmc/linux/net/netfilter/xt_socket.c (revision 25985edc)
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 defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
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 defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
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;
111 	__be32 daddr, saddr;
112 	__be16 dport, sport;
113 	u8 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_IS_REPLY + IP_CT_ESTABLISHED) ||
147 	     (iph->protocol == IPPROTO_ICMP &&
148 	      ctinfo == IP_CT_IS_REPLY + IP_CT_RELATED)) &&
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 	sk = nf_tproxy_get_sock_v4(dev_net(skb->dev), protocol,
159 				   saddr, daddr, sport, dport, par->in, NFT_LOOKUP_ANY);
160 	if (sk != NULL) {
161 		bool wildcard;
162 		bool transparent = true;
163 
164 		/* Ignore sockets listening on INADDR_ANY */
165 		wildcard = (sk->sk_state != TCP_TIME_WAIT &&
166 			    inet_sk(sk)->inet_rcv_saddr == 0);
167 
168 		/* Ignore non-transparent sockets,
169 		   if XT_SOCKET_TRANSPARENT is used */
170 		if (info && info->flags & XT_SOCKET_TRANSPARENT)
171 			transparent = ((sk->sk_state != TCP_TIME_WAIT &&
172 					inet_sk(sk)->transparent) ||
173 				       (sk->sk_state == TCP_TIME_WAIT &&
174 					inet_twsk(sk)->tw_transparent));
175 
176 		xt_socket_put_sk(sk);
177 
178 		if (wildcard || !transparent)
179 			sk = NULL;
180 	}
181 
182 	pr_debug("proto %hhu %pI4:%hu -> %pI4:%hu (orig %pI4:%hu) sock %p\n",
183 		 protocol, &saddr, ntohs(sport),
184 		 &daddr, ntohs(dport),
185 		 &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
186 
187 	return (sk != NULL);
188 }
189 
190 static bool
191 socket_mt4_v0(const struct sk_buff *skb, struct xt_action_param *par)
192 {
193 	return socket_match(skb, par, NULL);
194 }
195 
196 static bool
197 socket_mt4_v1(const struct sk_buff *skb, struct xt_action_param *par)
198 {
199 	return socket_match(skb, par, par->matchinfo);
200 }
201 
202 #ifdef XT_SOCKET_HAVE_IPV6
203 
204 static int
205 extract_icmp6_fields(const struct sk_buff *skb,
206 		     unsigned int outside_hdrlen,
207 		     int *protocol,
208 		     struct in6_addr **raddr,
209 		     struct in6_addr **laddr,
210 		     __be16 *rport,
211 		     __be16 *lport)
212 {
213 	struct ipv6hdr *inside_iph, _inside_iph;
214 	struct icmp6hdr *icmph, _icmph;
215 	__be16 *ports, _ports[2];
216 	u8 inside_nexthdr;
217 	int inside_hdrlen;
218 
219 	icmph = skb_header_pointer(skb, outside_hdrlen,
220 				   sizeof(_icmph), &_icmph);
221 	if (icmph == NULL)
222 		return 1;
223 
224 	if (icmph->icmp6_type & ICMPV6_INFOMSG_MASK)
225 		return 1;
226 
227 	inside_iph = skb_header_pointer(skb, outside_hdrlen + sizeof(_icmph), sizeof(_inside_iph), &_inside_iph);
228 	if (inside_iph == NULL)
229 		return 1;
230 	inside_nexthdr = inside_iph->nexthdr;
231 
232 	inside_hdrlen = ipv6_skip_exthdr(skb, outside_hdrlen + sizeof(_icmph) + sizeof(_inside_iph), &inside_nexthdr);
233 	if (inside_hdrlen < 0)
234 		return 1; /* hjm: Packet has no/incomplete transport layer headers. */
235 
236 	if (inside_nexthdr != IPPROTO_TCP &&
237 	    inside_nexthdr != IPPROTO_UDP)
238 		return 1;
239 
240 	ports = skb_header_pointer(skb, inside_hdrlen,
241 				   sizeof(_ports), &_ports);
242 	if (ports == NULL)
243 		return 1;
244 
245 	/* the inside IP packet is the one quoted from our side, thus
246 	 * its saddr is the local address */
247 	*protocol = inside_nexthdr;
248 	*laddr = &inside_iph->saddr;
249 	*lport = ports[0];
250 	*raddr = &inside_iph->daddr;
251 	*rport = ports[1];
252 
253 	return 0;
254 }
255 
256 static bool
257 socket_mt6_v1(const struct sk_buff *skb, struct xt_action_param *par)
258 {
259 	struct ipv6hdr *iph = ipv6_hdr(skb);
260 	struct udphdr _hdr, *hp = NULL;
261 	struct sock *sk;
262 	struct in6_addr *daddr, *saddr;
263 	__be16 dport, sport;
264 	int thoff, tproto;
265 	const struct xt_socket_mtinfo1 *info = (struct xt_socket_mtinfo1 *) par->matchinfo;
266 
267 	tproto = ipv6_find_hdr(skb, &thoff, -1, NULL);
268 	if (tproto < 0) {
269 		pr_debug("unable to find transport header in IPv6 packet, dropping\n");
270 		return NF_DROP;
271 	}
272 
273 	if (tproto == IPPROTO_UDP || tproto == IPPROTO_TCP) {
274 		hp = skb_header_pointer(skb, thoff,
275 					sizeof(_hdr), &_hdr);
276 		if (hp == NULL)
277 			return false;
278 
279 		saddr = &iph->saddr;
280 		sport = hp->source;
281 		daddr = &iph->daddr;
282 		dport = hp->dest;
283 
284 	} else if (tproto == IPPROTO_ICMPV6) {
285 		if (extract_icmp6_fields(skb, thoff, &tproto, &saddr, &daddr,
286 					 &sport, &dport))
287 			return false;
288 	} else {
289 		return false;
290 	}
291 
292 	sk = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
293 				   saddr, daddr, sport, dport, par->in, NFT_LOOKUP_ANY);
294 	if (sk != NULL) {
295 		bool wildcard;
296 		bool transparent = true;
297 
298 		/* Ignore sockets listening on INADDR_ANY */
299 		wildcard = (sk->sk_state != TCP_TIME_WAIT &&
300 			    ipv6_addr_any(&inet6_sk(sk)->rcv_saddr));
301 
302 		/* Ignore non-transparent sockets,
303 		   if XT_SOCKET_TRANSPARENT is used */
304 		if (info && info->flags & XT_SOCKET_TRANSPARENT)
305 			transparent = ((sk->sk_state != TCP_TIME_WAIT &&
306 					inet_sk(sk)->transparent) ||
307 				       (sk->sk_state == TCP_TIME_WAIT &&
308 					inet_twsk(sk)->tw_transparent));
309 
310 		xt_socket_put_sk(sk);
311 
312 		if (wildcard || !transparent)
313 			sk = NULL;
314 	}
315 
316 	pr_debug("proto %hhd %pI6:%hu -> %pI6:%hu "
317 		 "(orig %pI6:%hu) sock %p\n",
318 		 tproto, saddr, ntohs(sport),
319 		 daddr, ntohs(dport),
320 		 &iph->daddr, hp ? ntohs(hp->dest) : 0, sk);
321 
322 	return (sk != NULL);
323 }
324 #endif
325 
326 static struct xt_match socket_mt_reg[] __read_mostly = {
327 	{
328 		.name		= "socket",
329 		.revision	= 0,
330 		.family		= NFPROTO_IPV4,
331 		.match		= socket_mt4_v0,
332 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
333 				  (1 << NF_INET_LOCAL_IN),
334 		.me		= THIS_MODULE,
335 	},
336 	{
337 		.name		= "socket",
338 		.revision	= 1,
339 		.family		= NFPROTO_IPV4,
340 		.match		= socket_mt4_v1,
341 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
342 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
343 				  (1 << NF_INET_LOCAL_IN),
344 		.me		= THIS_MODULE,
345 	},
346 #ifdef XT_SOCKET_HAVE_IPV6
347 	{
348 		.name		= "socket",
349 		.revision	= 1,
350 		.family		= NFPROTO_IPV6,
351 		.match		= socket_mt6_v1,
352 		.matchsize	= sizeof(struct xt_socket_mtinfo1),
353 		.hooks		= (1 << NF_INET_PRE_ROUTING) |
354 				  (1 << NF_INET_LOCAL_IN),
355 		.me		= THIS_MODULE,
356 	},
357 #endif
358 };
359 
360 static int __init socket_mt_init(void)
361 {
362 	nf_defrag_ipv4_enable();
363 #ifdef XT_SOCKET_HAVE_IPV6
364 	nf_defrag_ipv6_enable();
365 #endif
366 
367 	return xt_register_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
368 }
369 
370 static void __exit socket_mt_exit(void)
371 {
372 	xt_unregister_matches(socket_mt_reg, ARRAY_SIZE(socket_mt_reg));
373 }
374 
375 module_init(socket_mt_init);
376 module_exit(socket_mt_exit);
377 
378 MODULE_LICENSE("GPL");
379 MODULE_AUTHOR("Krisztian Kovacs, Balazs Scheidler");
380 MODULE_DESCRIPTION("x_tables socket match module");
381 MODULE_ALIAS("ipt_socket");
382 MODULE_ALIAS("ip6t_socket");
383