xref: /openbmc/linux/net/netfilter/nft_tproxy.c (revision 14474950)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #include <linux/module.h>
3 #include <linux/netfilter/nf_tables.h>
4 #include <net/netfilter/nf_tables.h>
5 #include <net/netfilter/nf_tables_core.h>
6 #include <net/netfilter/nf_tproxy.h>
7 #include <net/inet_sock.h>
8 #include <net/tcp.h>
9 #include <linux/if_ether.h>
10 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
11 #if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
12 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
13 #endif
14 
15 struct nft_tproxy {
16 	enum nft_registers      sreg_addr:8;
17 	enum nft_registers      sreg_port:8;
18 	u8			family;
19 };
20 
21 static void nft_tproxy_eval_v4(const struct nft_expr *expr,
22 			       struct nft_regs *regs,
23 			       const struct nft_pktinfo *pkt)
24 {
25 	const struct nft_tproxy *priv = nft_expr_priv(expr);
26 	struct sk_buff *skb = pkt->skb;
27 	const struct iphdr *iph = ip_hdr(skb);
28 	struct udphdr _hdr, *hp;
29 	__be32 taddr = 0;
30 	__be16 tport = 0;
31 	struct sock *sk;
32 
33 	hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
34 	if (!hp) {
35 		regs->verdict.code = NFT_BREAK;
36 		return;
37 	}
38 
39 	/* check if there's an ongoing connection on the packet addresses, this
40 	 * happens if the redirect already happened and the current packet
41 	 * belongs to an already established connection
42 	 */
43 	sk = nf_tproxy_get_sock_v4(nft_net(pkt), skb, iph->protocol,
44 				   iph->saddr, iph->daddr,
45 				   hp->source, hp->dest,
46 				   skb->dev, NF_TPROXY_LOOKUP_ESTABLISHED);
47 
48 	if (priv->sreg_addr)
49 		taddr = regs->data[priv->sreg_addr];
50 	taddr = nf_tproxy_laddr4(skb, taddr, iph->daddr);
51 
52 	if (priv->sreg_port)
53 		tport = nft_reg_load16(&regs->data[priv->sreg_port]);
54 	if (!tport)
55 		tport = hp->dest;
56 
57 	/* UDP has no TCP_TIME_WAIT state, so we never enter here */
58 	if (sk && sk->sk_state == TCP_TIME_WAIT) {
59 		/* reopening a TIME_WAIT connection needs special handling */
60 		sk = nf_tproxy_handle_time_wait4(nft_net(pkt), skb, taddr, tport, sk);
61 	} else if (!sk) {
62 		/* no, there's no established connection, check if
63 		 * there's a listener on the redirected addr/port
64 		 */
65 		sk = nf_tproxy_get_sock_v4(nft_net(pkt), skb, iph->protocol,
66 					   iph->saddr, taddr,
67 					   hp->source, tport,
68 					   skb->dev, NF_TPROXY_LOOKUP_LISTENER);
69 	}
70 
71 	if (sk && nf_tproxy_sk_is_transparent(sk))
72 		nf_tproxy_assign_sock(skb, sk);
73 	else
74 		regs->verdict.code = NFT_BREAK;
75 }
76 
77 #if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
78 static void nft_tproxy_eval_v6(const struct nft_expr *expr,
79 			       struct nft_regs *regs,
80 			       const struct nft_pktinfo *pkt)
81 {
82 	const struct nft_tproxy *priv = nft_expr_priv(expr);
83 	struct sk_buff *skb = pkt->skb;
84 	const struct ipv6hdr *iph = ipv6_hdr(skb);
85 	struct in6_addr taddr;
86 	int thoff = pkt->xt.thoff;
87 	struct udphdr _hdr, *hp;
88 	__be16 tport = 0;
89 	struct sock *sk;
90 	int l4proto;
91 
92 	memset(&taddr, 0, sizeof(taddr));
93 
94 	if (!pkt->tprot_set) {
95 		regs->verdict.code = NFT_BREAK;
96 		return;
97 	}
98 	l4proto = pkt->tprot;
99 
100 	hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
101 	if (hp == NULL) {
102 		regs->verdict.code = NFT_BREAK;
103 		return;
104 	}
105 
106 	/* check if there's an ongoing connection on the packet addresses, this
107 	 * happens if the redirect already happened and the current packet
108 	 * belongs to an already established connection
109 	 */
110 	sk = nf_tproxy_get_sock_v6(nft_net(pkt), skb, thoff, l4proto,
111 				   &iph->saddr, &iph->daddr,
112 				   hp->source, hp->dest,
113 				   nft_in(pkt), NF_TPROXY_LOOKUP_ESTABLISHED);
114 
115 	if (priv->sreg_addr)
116 		memcpy(&taddr, &regs->data[priv->sreg_addr], sizeof(taddr));
117 	taddr = *nf_tproxy_laddr6(skb, &taddr, &iph->daddr);
118 
119 	if (priv->sreg_port)
120 		tport = nft_reg_load16(&regs->data[priv->sreg_port]);
121 	if (!tport)
122 		tport = hp->dest;
123 
124 	/* UDP has no TCP_TIME_WAIT state, so we never enter here */
125 	if (sk && sk->sk_state == TCP_TIME_WAIT) {
126 		/* reopening a TIME_WAIT connection needs special handling */
127 		sk = nf_tproxy_handle_time_wait6(skb, l4proto, thoff,
128 						 nft_net(pkt),
129 						 &taddr,
130 						 tport,
131 						 sk);
132 	} else if (!sk) {
133 		/* no there's no established connection, check if
134 		 * there's a listener on the redirected addr/port
135 		 */
136 		sk = nf_tproxy_get_sock_v6(nft_net(pkt), skb, thoff,
137 					   l4proto, &iph->saddr, &taddr,
138 					   hp->source, tport,
139 					   nft_in(pkt), NF_TPROXY_LOOKUP_LISTENER);
140 	}
141 
142 	/* NOTE: assign_sock consumes our sk reference */
143 	if (sk && nf_tproxy_sk_is_transparent(sk))
144 		nf_tproxy_assign_sock(skb, sk);
145 	else
146 		regs->verdict.code = NFT_BREAK;
147 }
148 #endif
149 
150 static void nft_tproxy_eval(const struct nft_expr *expr,
151 			    struct nft_regs *regs,
152 			    const struct nft_pktinfo *pkt)
153 {
154 	const struct nft_tproxy *priv = nft_expr_priv(expr);
155 
156 	switch (nft_pf(pkt)) {
157 	case NFPROTO_IPV4:
158 		switch (priv->family) {
159 		case NFPROTO_IPV4:
160 		case NFPROTO_UNSPEC:
161 			nft_tproxy_eval_v4(expr, regs, pkt);
162 			return;
163 		}
164 		break;
165 #if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
166 	case NFPROTO_IPV6:
167 		switch (priv->family) {
168 		case NFPROTO_IPV6:
169 		case NFPROTO_UNSPEC:
170 			nft_tproxy_eval_v6(expr, regs, pkt);
171 			return;
172 		}
173 #endif
174 	}
175 	regs->verdict.code = NFT_BREAK;
176 }
177 
178 static const struct nla_policy nft_tproxy_policy[NFTA_TPROXY_MAX + 1] = {
179 	[NFTA_TPROXY_FAMILY]   = { .type = NLA_U32 },
180 	[NFTA_TPROXY_REG_ADDR] = { .type = NLA_U32 },
181 	[NFTA_TPROXY_REG_PORT] = { .type = NLA_U32 },
182 };
183 
184 static int nft_tproxy_init(const struct nft_ctx *ctx,
185 			   const struct nft_expr *expr,
186 			   const struct nlattr * const tb[])
187 {
188 	struct nft_tproxy *priv = nft_expr_priv(expr);
189 	unsigned int alen = 0;
190 	int err;
191 
192 	if (!tb[NFTA_TPROXY_FAMILY] ||
193 	    (!tb[NFTA_TPROXY_REG_ADDR] && !tb[NFTA_TPROXY_REG_PORT]))
194 		return -EINVAL;
195 
196 	priv->family = ntohl(nla_get_be32(tb[NFTA_TPROXY_FAMILY]));
197 
198 	switch (ctx->family) {
199 	case NFPROTO_IPV4:
200 		if (priv->family != NFPROTO_IPV4)
201 			return -EINVAL;
202 		break;
203 #if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
204 	case NFPROTO_IPV6:
205 		if (priv->family != NFPROTO_IPV6)
206 			return -EINVAL;
207 		break;
208 #endif
209 	case NFPROTO_INET:
210 		break;
211 	default:
212 		return -EOPNOTSUPP;
213 	}
214 
215 	/* Address is specified but the rule family is not set accordingly */
216 	if (priv->family == NFPROTO_UNSPEC && tb[NFTA_TPROXY_REG_ADDR])
217 		return -EINVAL;
218 
219 	switch (priv->family) {
220 	case NFPROTO_IPV4:
221 		alen = sizeof_field(union nf_inet_addr, in);
222 		err = nf_defrag_ipv4_enable(ctx->net);
223 		if (err)
224 			return err;
225 		break;
226 #if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
227 	case NFPROTO_IPV6:
228 		alen = sizeof_field(union nf_inet_addr, in6);
229 		err = nf_defrag_ipv6_enable(ctx->net);
230 		if (err)
231 			return err;
232 		break;
233 #endif
234 	case NFPROTO_UNSPEC:
235 		/* No address is specified here */
236 		err = nf_defrag_ipv4_enable(ctx->net);
237 		if (err)
238 			return err;
239 #if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
240 		err = nf_defrag_ipv6_enable(ctx->net);
241 		if (err)
242 			return err;
243 #endif
244 		break;
245 	default:
246 		return -EOPNOTSUPP;
247 	}
248 
249 	if (tb[NFTA_TPROXY_REG_ADDR]) {
250 		priv->sreg_addr = nft_parse_register(tb[NFTA_TPROXY_REG_ADDR]);
251 		err = nft_validate_register_load(priv->sreg_addr, alen);
252 		if (err < 0)
253 			return err;
254 	}
255 
256 	if (tb[NFTA_TPROXY_REG_PORT]) {
257 		priv->sreg_port = nft_parse_register(tb[NFTA_TPROXY_REG_PORT]);
258 		err = nft_validate_register_load(priv->sreg_port, sizeof(u16));
259 		if (err < 0)
260 			return err;
261 	}
262 
263 	return 0;
264 }
265 
266 static int nft_tproxy_dump(struct sk_buff *skb,
267 			   const struct nft_expr *expr)
268 {
269 	const struct nft_tproxy *priv = nft_expr_priv(expr);
270 
271 	if (nla_put_be32(skb, NFTA_TPROXY_FAMILY, htonl(priv->family)))
272 		return -1;
273 
274 	if (priv->sreg_addr &&
275 	    nft_dump_register(skb, NFTA_TPROXY_REG_ADDR, priv->sreg_addr))
276 		return -1;
277 
278 	if (priv->sreg_port &&
279 	    nft_dump_register(skb, NFTA_TPROXY_REG_PORT, priv->sreg_port))
280 			return -1;
281 
282 	return 0;
283 }
284 
285 static struct nft_expr_type nft_tproxy_type;
286 static const struct nft_expr_ops nft_tproxy_ops = {
287 	.type		= &nft_tproxy_type,
288 	.size		= NFT_EXPR_SIZE(sizeof(struct nft_tproxy)),
289 	.eval		= nft_tproxy_eval,
290 	.init		= nft_tproxy_init,
291 	.dump		= nft_tproxy_dump,
292 };
293 
294 static struct nft_expr_type nft_tproxy_type __read_mostly = {
295 	.name		= "tproxy",
296 	.ops		= &nft_tproxy_ops,
297 	.policy		= nft_tproxy_policy,
298 	.maxattr	= NFTA_TPROXY_MAX,
299 	.owner		= THIS_MODULE,
300 };
301 
302 static int __init nft_tproxy_module_init(void)
303 {
304 	return nft_register_expr(&nft_tproxy_type);
305 }
306 
307 static void __exit nft_tproxy_module_exit(void)
308 {
309 	nft_unregister_expr(&nft_tproxy_type);
310 }
311 
312 module_init(nft_tproxy_module_init);
313 module_exit(nft_tproxy_module_exit);
314 
315 MODULE_LICENSE("GPL");
316 MODULE_AUTHOR("Máté Eckl");
317 MODULE_DESCRIPTION("nf_tables tproxy support module");
318 MODULE_ALIAS_NFT_EXPR("tproxy");
319