Merge tag 'perf-tools-fixes-for-v6.4-1-2023-05-20' of git://git.kernel.org/pub/scm...
[linux-block.git] / net / netfilter / nft_tproxy.c
CommitLineData
4ed8eb65
ME
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
15struct nft_tproxy {
4f16d25c
PNA
16 u8 sreg_addr;
17 u8 sreg_port;
18 u8 family;
4ed8eb65
ME
19};
20
21static 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
52f0f4e1
PNA
33 if (pkt->tprot != IPPROTO_TCP &&
34 pkt->tprot != IPPROTO_UDP) {
35 regs->verdict.code = NFT_BREAK;
36 return;
37 }
38
4ed8eb65
ME
39 hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
40 if (!hp) {
41 regs->verdict.code = NFT_BREAK;
42 return;
43 }
44
45 /* check if there's an ongoing connection on the packet addresses, this
46 * happens if the redirect already happened and the current packet
47 * belongs to an already established connection
48 */
49 sk = nf_tproxy_get_sock_v4(nft_net(pkt), skb, iph->protocol,
50 iph->saddr, iph->daddr,
51 hp->source, hp->dest,
52 skb->dev, NF_TPROXY_LOOKUP_ESTABLISHED);
53
54 if (priv->sreg_addr)
7278b3c1 55 taddr = nft_reg_load_be32(&regs->data[priv->sreg_addr]);
4ed8eb65
ME
56 taddr = nf_tproxy_laddr4(skb, taddr, iph->daddr);
57
58 if (priv->sreg_port)
7278b3c1 59 tport = nft_reg_load_be16(&regs->data[priv->sreg_port]);
4ed8eb65
ME
60 if (!tport)
61 tport = hp->dest;
62
63 /* UDP has no TCP_TIME_WAIT state, so we never enter here */
64 if (sk && sk->sk_state == TCP_TIME_WAIT) {
65 /* reopening a TIME_WAIT connection needs special handling */
66 sk = nf_tproxy_handle_time_wait4(nft_net(pkt), skb, taddr, tport, sk);
67 } else if (!sk) {
68 /* no, there's no established connection, check if
69 * there's a listener on the redirected addr/port
70 */
71 sk = nf_tproxy_get_sock_v4(nft_net(pkt), skb, iph->protocol,
72 iph->saddr, taddr,
73 hp->source, tport,
74 skb->dev, NF_TPROXY_LOOKUP_LISTENER);
75 }
76
77 if (sk && nf_tproxy_sk_is_transparent(sk))
78 nf_tproxy_assign_sock(skb, sk);
79 else
80 regs->verdict.code = NFT_BREAK;
81}
82
83#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
84static void nft_tproxy_eval_v6(const struct nft_expr *expr,
85 struct nft_regs *regs,
86 const struct nft_pktinfo *pkt)
87{
88 const struct nft_tproxy *priv = nft_expr_priv(expr);
89 struct sk_buff *skb = pkt->skb;
90 const struct ipv6hdr *iph = ipv6_hdr(skb);
2d7b4ace 91 int thoff = nft_thoff(pkt);
4ed8eb65 92 struct udphdr _hdr, *hp;
2d7b4ace 93 struct in6_addr taddr;
4ed8eb65
ME
94 __be16 tport = 0;
95 struct sock *sk;
96 int l4proto;
97
90d827f0
ME
98 memset(&taddr, 0, sizeof(taddr));
99
52f0f4e1
PNA
100 if (pkt->tprot != IPPROTO_TCP &&
101 pkt->tprot != IPPROTO_UDP) {
4ed8eb65
ME
102 regs->verdict.code = NFT_BREAK;
103 return;
104 }
105 l4proto = pkt->tprot;
106
107 hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
108 if (hp == NULL) {
109 regs->verdict.code = NFT_BREAK;
110 return;
111 }
112
113 /* check if there's an ongoing connection on the packet addresses, this
114 * happens if the redirect already happened and the current packet
115 * belongs to an already established connection
116 */
117 sk = nf_tproxy_get_sock_v6(nft_net(pkt), skb, thoff, l4proto,
118 &iph->saddr, &iph->daddr,
119 hp->source, hp->dest,
120 nft_in(pkt), NF_TPROXY_LOOKUP_ESTABLISHED);
121
122 if (priv->sreg_addr)
123 memcpy(&taddr, &regs->data[priv->sreg_addr], sizeof(taddr));
124 taddr = *nf_tproxy_laddr6(skb, &taddr, &iph->daddr);
125
126 if (priv->sreg_port)
7278b3c1 127 tport = nft_reg_load_be16(&regs->data[priv->sreg_port]);
4ed8eb65
ME
128 if (!tport)
129 tport = hp->dest;
130
131 /* UDP has no TCP_TIME_WAIT state, so we never enter here */
132 if (sk && sk->sk_state == TCP_TIME_WAIT) {
133 /* reopening a TIME_WAIT connection needs special handling */
134 sk = nf_tproxy_handle_time_wait6(skb, l4proto, thoff,
135 nft_net(pkt),
136 &taddr,
137 tport,
138 sk);
139 } else if (!sk) {
140 /* no there's no established connection, check if
141 * there's a listener on the redirected addr/port
142 */
143 sk = nf_tproxy_get_sock_v6(nft_net(pkt), skb, thoff,
144 l4proto, &iph->saddr, &taddr,
145 hp->source, tport,
146 nft_in(pkt), NF_TPROXY_LOOKUP_LISTENER);
147 }
148
149 /* NOTE: assign_sock consumes our sk reference */
150 if (sk && nf_tproxy_sk_is_transparent(sk))
151 nf_tproxy_assign_sock(skb, sk);
152 else
153 regs->verdict.code = NFT_BREAK;
154}
155#endif
156
157static void nft_tproxy_eval(const struct nft_expr *expr,
158 struct nft_regs *regs,
159 const struct nft_pktinfo *pkt)
160{
161 const struct nft_tproxy *priv = nft_expr_priv(expr);
162
163 switch (nft_pf(pkt)) {
164 case NFPROTO_IPV4:
165 switch (priv->family) {
166 case NFPROTO_IPV4:
167 case NFPROTO_UNSPEC:
168 nft_tproxy_eval_v4(expr, regs, pkt);
169 return;
170 }
171 break;
172#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
173 case NFPROTO_IPV6:
174 switch (priv->family) {
175 case NFPROTO_IPV6:
176 case NFPROTO_UNSPEC:
177 nft_tproxy_eval_v6(expr, regs, pkt);
178 return;
179 }
180#endif
181 }
182 regs->verdict.code = NFT_BREAK;
183}
184
185static const struct nla_policy nft_tproxy_policy[NFTA_TPROXY_MAX + 1] = {
186 [NFTA_TPROXY_FAMILY] = { .type = NLA_U32 },
187 [NFTA_TPROXY_REG_ADDR] = { .type = NLA_U32 },
188 [NFTA_TPROXY_REG_PORT] = { .type = NLA_U32 },
189};
190
191static int nft_tproxy_init(const struct nft_ctx *ctx,
192 const struct nft_expr *expr,
193 const struct nlattr * const tb[])
194{
195 struct nft_tproxy *priv = nft_expr_priv(expr);
196 unsigned int alen = 0;
197 int err;
198
199 if (!tb[NFTA_TPROXY_FAMILY] ||
200 (!tb[NFTA_TPROXY_REG_ADDR] && !tb[NFTA_TPROXY_REG_PORT]))
201 return -EINVAL;
202
203 priv->family = ntohl(nla_get_be32(tb[NFTA_TPROXY_FAMILY]));
204
205 switch (ctx->family) {
206 case NFPROTO_IPV4:
207 if (priv->family != NFPROTO_IPV4)
208 return -EINVAL;
209 break;
210#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
211 case NFPROTO_IPV6:
212 if (priv->family != NFPROTO_IPV6)
213 return -EINVAL;
214 break;
215#endif
216 case NFPROTO_INET:
217 break;
218 default:
219 return -EOPNOTSUPP;
220 }
221
222 /* Address is specified but the rule family is not set accordingly */
223 if (priv->family == NFPROTO_UNSPEC && tb[NFTA_TPROXY_REG_ADDR])
224 return -EINVAL;
225
226 switch (priv->family) {
227 case NFPROTO_IPV4:
c593642c 228 alen = sizeof_field(union nf_inet_addr, in);
4ed8eb65
ME
229 err = nf_defrag_ipv4_enable(ctx->net);
230 if (err)
231 return err;
232 break;
233#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
234 case NFPROTO_IPV6:
c593642c 235 alen = sizeof_field(union nf_inet_addr, in6);
4ed8eb65
ME
236 err = nf_defrag_ipv6_enable(ctx->net);
237 if (err)
238 return err;
239 break;
240#endif
241 case NFPROTO_UNSPEC:
242 /* No address is specified here */
243 err = nf_defrag_ipv4_enable(ctx->net);
244 if (err)
245 return err;
033eab53 246#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
4ed8eb65
ME
247 err = nf_defrag_ipv6_enable(ctx->net);
248 if (err)
249 return err;
033eab53 250#endif
4ed8eb65
ME
251 break;
252 default:
253 return -EOPNOTSUPP;
254 }
255
256 if (tb[NFTA_TPROXY_REG_ADDR]) {
4f16d25c
PNA
257 err = nft_parse_register_load(tb[NFTA_TPROXY_REG_ADDR],
258 &priv->sreg_addr, alen);
4ed8eb65
ME
259 if (err < 0)
260 return err;
261 }
262
263 if (tb[NFTA_TPROXY_REG_PORT]) {
4f16d25c
PNA
264 err = nft_parse_register_load(tb[NFTA_TPROXY_REG_PORT],
265 &priv->sreg_port, sizeof(u16));
4ed8eb65
ME
266 if (err < 0)
267 return err;
268 }
269
270 return 0;
271}
272
de8c1211
FW
273static void nft_tproxy_destroy(const struct nft_ctx *ctx,
274 const struct nft_expr *expr)
275{
276 const struct nft_tproxy *priv = nft_expr_priv(expr);
277
278 switch (priv->family) {
279 case NFPROTO_IPV4:
280 nf_defrag_ipv4_disable(ctx->net);
281 break;
282#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
283 case NFPROTO_IPV6:
284 nf_defrag_ipv6_disable(ctx->net);
285 break;
286#endif
287 case NFPROTO_UNSPEC:
288 nf_defrag_ipv4_disable(ctx->net);
289#if IS_ENABLED(CONFIG_NF_TABLES_IPV6)
290 nf_defrag_ipv6_disable(ctx->net);
291#endif
292 break;
293 }
294}
295
4ed8eb65 296static int nft_tproxy_dump(struct sk_buff *skb,
7d34aa3e 297 const struct nft_expr *expr, bool reset)
4ed8eb65
ME
298{
299 const struct nft_tproxy *priv = nft_expr_priv(expr);
300
301 if (nla_put_be32(skb, NFTA_TPROXY_FAMILY, htonl(priv->family)))
302 return -1;
303
304 if (priv->sreg_addr &&
305 nft_dump_register(skb, NFTA_TPROXY_REG_ADDR, priv->sreg_addr))
306 return -1;
307
308 if (priv->sreg_port &&
309 nft_dump_register(skb, NFTA_TPROXY_REG_PORT, priv->sreg_port))
310 return -1;
311
312 return 0;
313}
314
18bbc321
FW
315static int nft_tproxy_validate(const struct nft_ctx *ctx,
316 const struct nft_expr *expr,
317 const struct nft_data **data)
318{
319 return nft_chain_validate_hooks(ctx->chain, 1 << NF_INET_PRE_ROUTING);
320}
321
4ed8eb65
ME
322static struct nft_expr_type nft_tproxy_type;
323static const struct nft_expr_ops nft_tproxy_ops = {
324 .type = &nft_tproxy_type,
325 .size = NFT_EXPR_SIZE(sizeof(struct nft_tproxy)),
326 .eval = nft_tproxy_eval,
327 .init = nft_tproxy_init,
de8c1211 328 .destroy = nft_tproxy_destroy,
4ed8eb65 329 .dump = nft_tproxy_dump,
b2d30654 330 .reduce = NFT_REDUCE_READONLY,
18bbc321 331 .validate = nft_tproxy_validate,
4ed8eb65
ME
332};
333
334static struct nft_expr_type nft_tproxy_type __read_mostly = {
335 .name = "tproxy",
336 .ops = &nft_tproxy_ops,
337 .policy = nft_tproxy_policy,
338 .maxattr = NFTA_TPROXY_MAX,
339 .owner = THIS_MODULE,
340};
341
342static int __init nft_tproxy_module_init(void)
343{
344 return nft_register_expr(&nft_tproxy_type);
345}
346
347static void __exit nft_tproxy_module_exit(void)
348{
349 nft_unregister_expr(&nft_tproxy_type);
350}
351
352module_init(nft_tproxy_module_init);
353module_exit(nft_tproxy_module_exit);
354
355MODULE_LICENSE("GPL");
356MODULE_AUTHOR("Máté Eckl");
357MODULE_DESCRIPTION("nf_tables tproxy support module");
358MODULE_ALIAS_NFT_EXPR("tproxy");