xref: /openbmc/linux/net/sched/act_nat.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
1 /*
2  * Stateless NAT actions
3  *
4  * Copyright (c) 2007 Herbert Xu <herbert@gondor.apana.org.au>
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License as published by the Free
8  * Software Foundation; either version 2 of the License, or (at your option)
9  * any later version.
10  */
11 
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/netfilter.h>
17 #include <linux/rtnetlink.h>
18 #include <linux/skbuff.h>
19 #include <linux/slab.h>
20 #include <linux/spinlock.h>
21 #include <linux/string.h>
22 #include <linux/tc_act/tc_nat.h>
23 #include <net/act_api.h>
24 #include <net/icmp.h>
25 #include <net/ip.h>
26 #include <net/netlink.h>
27 #include <net/tc_act/tc_nat.h>
28 #include <net/tcp.h>
29 #include <net/udp.h>
30 
31 
32 #define NAT_TAB_MASK	15
33 static struct tcf_common *tcf_nat_ht[NAT_TAB_MASK + 1];
34 static u32 nat_idx_gen;
35 static DEFINE_RWLOCK(nat_lock);
36 
37 static struct tcf_hashinfo nat_hash_info = {
38 	.htab	=	tcf_nat_ht,
39 	.hmask	=	NAT_TAB_MASK,
40 	.lock	=	&nat_lock,
41 };
42 
43 static int tcf_nat_init(struct rtattr *rta, struct rtattr *est,
44 			struct tc_action *a, int ovr, int bind)
45 {
46 	struct rtattr *tb[TCA_NAT_MAX];
47 	struct tc_nat *parm;
48 	int ret = 0;
49 	struct tcf_nat *p;
50 	struct tcf_common *pc;
51 
52 	if (rta == NULL || rtattr_parse_nested(tb, TCA_NAT_MAX, rta) < 0)
53 		return -EINVAL;
54 
55 	if (tb[TCA_NAT_PARMS - 1] == NULL ||
56 	    RTA_PAYLOAD(tb[TCA_NAT_PARMS - 1]) < sizeof(*parm))
57 		return -EINVAL;
58 	parm = RTA_DATA(tb[TCA_NAT_PARMS - 1]);
59 
60 	pc = tcf_hash_check(parm->index, a, bind, &nat_hash_info);
61 	if (!pc) {
62 		pc = tcf_hash_create(parm->index, est, a, sizeof(*p), bind,
63 				     &nat_idx_gen, &nat_hash_info);
64 		if (unlikely(!pc))
65 			return -ENOMEM;
66 		p = to_tcf_nat(pc);
67 		ret = ACT_P_CREATED;
68 	} else {
69 		p = to_tcf_nat(pc);
70 		if (!ovr) {
71 			tcf_hash_release(pc, bind, &nat_hash_info);
72 			return -EEXIST;
73 		}
74 	}
75 
76 	spin_lock_bh(&p->tcf_lock);
77 	p->old_addr = parm->old_addr;
78 	p->new_addr = parm->new_addr;
79 	p->mask = parm->mask;
80 	p->flags = parm->flags;
81 
82 	p->tcf_action = parm->action;
83 	spin_unlock_bh(&p->tcf_lock);
84 
85 	if (ret == ACT_P_CREATED)
86 		tcf_hash_insert(pc, &nat_hash_info);
87 
88 	return ret;
89 }
90 
91 static int tcf_nat_cleanup(struct tc_action *a, int bind)
92 {
93 	struct tcf_nat *p = a->priv;
94 
95 	return tcf_hash_release(&p->common, bind, &nat_hash_info);
96 }
97 
98 static int tcf_nat(struct sk_buff *skb, struct tc_action *a,
99 		   struct tcf_result *res)
100 {
101 	struct tcf_nat *p = a->priv;
102 	struct iphdr *iph;
103 	__be32 old_addr;
104 	__be32 new_addr;
105 	__be32 mask;
106 	__be32 addr;
107 	int egress;
108 	int action;
109 	int ihl;
110 
111 	spin_lock(&p->tcf_lock);
112 
113 	p->tcf_tm.lastuse = jiffies;
114 	old_addr = p->old_addr;
115 	new_addr = p->new_addr;
116 	mask = p->mask;
117 	egress = p->flags & TCA_NAT_FLAG_EGRESS;
118 	action = p->tcf_action;
119 
120 	p->tcf_bstats.bytes += skb->len;
121 	p->tcf_bstats.packets++;
122 
123 	spin_unlock(&p->tcf_lock);
124 
125 	if (unlikely(action == TC_ACT_SHOT))
126 		goto drop;
127 
128 	if (!pskb_may_pull(skb, sizeof(*iph)))
129 		goto drop;
130 
131 	iph = ip_hdr(skb);
132 
133 	if (egress)
134 		addr = iph->saddr;
135 	else
136 		addr = iph->daddr;
137 
138 	if (!((old_addr ^ addr) & mask)) {
139 		if (skb_cloned(skb) &&
140 		    !skb_clone_writable(skb, sizeof(*iph)) &&
141 		    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
142 			goto drop;
143 
144 		new_addr &= mask;
145 		new_addr |= addr & ~mask;
146 
147 		/* Rewrite IP header */
148 		iph = ip_hdr(skb);
149 		if (egress)
150 			iph->saddr = new_addr;
151 		else
152 			iph->daddr = new_addr;
153 
154 		nf_csum_replace4(&iph->check, addr, new_addr);
155 	}
156 
157 	ihl = iph->ihl * 4;
158 
159 	/* It would be nice to share code with stateful NAT. */
160 	switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
161 	case IPPROTO_TCP:
162 	{
163 		struct tcphdr *tcph;
164 
165 		if (!pskb_may_pull(skb, ihl + sizeof(*tcph)) ||
166 		    (skb_cloned(skb) &&
167 		     !skb_clone_writable(skb, ihl + sizeof(*tcph)) &&
168 		     pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
169 			goto drop;
170 
171 		tcph = (void *)(skb_network_header(skb) + ihl);
172 		nf_proto_csum_replace4(&tcph->check, skb, addr, new_addr, 1);
173 		break;
174 	}
175 	case IPPROTO_UDP:
176 	{
177 		struct udphdr *udph;
178 
179 		if (!pskb_may_pull(skb, ihl + sizeof(*udph)) ||
180 		    (skb_cloned(skb) &&
181 		     !skb_clone_writable(skb, ihl + sizeof(*udph)) &&
182 		     pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
183 			goto drop;
184 
185 		udph = (void *)(skb_network_header(skb) + ihl);
186 		if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
187 			nf_proto_csum_replace4(&udph->check, skb, addr,
188 					       new_addr, 1);
189 			if (!udph->check)
190 				udph->check = CSUM_MANGLED_0;
191 		}
192 		break;
193 	}
194 	case IPPROTO_ICMP:
195 	{
196 		struct icmphdr *icmph;
197 
198 		if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + sizeof(*iph)))
199 			goto drop;
200 
201 		icmph = (void *)(skb_network_header(skb) + ihl);
202 
203 		if ((icmph->type != ICMP_DEST_UNREACH) &&
204 		    (icmph->type != ICMP_TIME_EXCEEDED) &&
205 		    (icmph->type != ICMP_PARAMETERPROB))
206 			break;
207 
208 		iph = (void *)(icmph + 1);
209 		if (egress)
210 			addr = iph->daddr;
211 		else
212 			addr = iph->saddr;
213 
214 		if ((old_addr ^ addr) & mask)
215 			break;
216 
217 		if (skb_cloned(skb) &&
218 		    !skb_clone_writable(skb,
219 					ihl + sizeof(*icmph) + sizeof(*iph)) &&
220 		    pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
221 			goto drop;
222 
223 		icmph = (void *)(skb_network_header(skb) + ihl);
224 		iph = (void *)(icmph + 1);
225 
226 		new_addr &= mask;
227 		new_addr |= addr & ~mask;
228 
229 		/* XXX Fix up the inner checksums. */
230 		if (egress)
231 			iph->daddr = new_addr;
232 		else
233 			iph->saddr = new_addr;
234 
235 		nf_proto_csum_replace4(&icmph->checksum, skb, addr, new_addr,
236 				       1);
237 		break;
238 	}
239 	default:
240 		break;
241 	}
242 
243 	return action;
244 
245 drop:
246 	spin_lock(&p->tcf_lock);
247 	p->tcf_qstats.drops++;
248 	spin_unlock(&p->tcf_lock);
249 	return TC_ACT_SHOT;
250 }
251 
252 static int tcf_nat_dump(struct sk_buff *skb, struct tc_action *a,
253 			int bind, int ref)
254 {
255 	unsigned char *b = skb_tail_pointer(skb);
256 	struct tcf_nat *p = a->priv;
257 	struct tc_nat *opt;
258 	struct tcf_t t;
259 	int s;
260 
261 	s = sizeof(*opt);
262 
263 	/* netlink spinlocks held above us - must use ATOMIC */
264 	opt = kzalloc(s, GFP_ATOMIC);
265 	if (unlikely(!opt))
266 		return -ENOBUFS;
267 
268 	opt->old_addr = p->old_addr;
269 	opt->new_addr = p->new_addr;
270 	opt->mask = p->mask;
271 	opt->flags = p->flags;
272 
273 	opt->index = p->tcf_index;
274 	opt->action = p->tcf_action;
275 	opt->refcnt = p->tcf_refcnt - ref;
276 	opt->bindcnt = p->tcf_bindcnt - bind;
277 
278 	RTA_PUT(skb, TCA_NAT_PARMS, s, opt);
279 	t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
280 	t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
281 	t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
282 	RTA_PUT(skb, TCA_NAT_TM, sizeof(t), &t);
283 
284 	kfree(opt);
285 
286 	return skb->len;
287 
288 rtattr_failure:
289 	nlmsg_trim(skb, b);
290 	kfree(opt);
291 	return -1;
292 }
293 
294 static struct tc_action_ops act_nat_ops = {
295 	.kind		=	"nat",
296 	.hinfo		=	&nat_hash_info,
297 	.type		=	TCA_ACT_NAT,
298 	.capab		=	TCA_CAP_NONE,
299 	.owner		=	THIS_MODULE,
300 	.act		=	tcf_nat,
301 	.dump		=	tcf_nat_dump,
302 	.cleanup	=	tcf_nat_cleanup,
303 	.lookup		=	tcf_hash_search,
304 	.init		=	tcf_nat_init,
305 	.walk		=	tcf_generic_walker
306 };
307 
308 MODULE_DESCRIPTION("Stateless NAT actions");
309 MODULE_LICENSE("GPL");
310 
311 static int __init nat_init_module(void)
312 {
313 	return tcf_register_action(&act_nat_ops);
314 }
315 
316 static void __exit nat_cleanup_module(void)
317 {
318 	tcf_unregister_action(&act_nat_ops);
319 }
320 
321 module_init(nat_init_module);
322 module_exit(nat_cleanup_module);
323