Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[sfrench/cifs-2.6.git] / net / sched / act_nat.c
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 const struct nla_policy nat_policy[TCA_NAT_MAX + 1] = {
44         [TCA_NAT_PARMS] = { .len = sizeof(struct tc_nat) },
45 };
46
47 static int tcf_nat_init(struct nlattr *nla, struct nlattr *est,
48                         struct tc_action *a, int ovr, int bind)
49 {
50         struct nlattr *tb[TCA_NAT_MAX + 1];
51         struct tc_nat *parm;
52         int ret = 0, err;
53         struct tcf_nat *p;
54         struct tcf_common *pc;
55
56         if (nla == NULL)
57                 return -EINVAL;
58
59         err = nla_parse_nested(tb, TCA_NAT_MAX, nla, nat_policy);
60         if (err < 0)
61                 return err;
62
63         if (tb[TCA_NAT_PARMS] == NULL)
64                 return -EINVAL;
65         parm = nla_data(tb[TCA_NAT_PARMS]);
66
67         pc = tcf_hash_check(parm->index, a, bind, &nat_hash_info);
68         if (!pc) {
69                 pc = tcf_hash_create(parm->index, est, a, sizeof(*p), bind,
70                                      &nat_idx_gen, &nat_hash_info);
71                 if (IS_ERR(pc))
72                     return PTR_ERR(pc);
73                 p = to_tcf_nat(pc);
74                 ret = ACT_P_CREATED;
75         } else {
76                 p = to_tcf_nat(pc);
77                 if (!ovr) {
78                         tcf_hash_release(pc, bind, &nat_hash_info);
79                         return -EEXIST;
80                 }
81         }
82
83         spin_lock_bh(&p->tcf_lock);
84         p->old_addr = parm->old_addr;
85         p->new_addr = parm->new_addr;
86         p->mask = parm->mask;
87         p->flags = parm->flags;
88
89         p->tcf_action = parm->action;
90         spin_unlock_bh(&p->tcf_lock);
91
92         if (ret == ACT_P_CREATED)
93                 tcf_hash_insert(pc, &nat_hash_info);
94
95         return ret;
96 }
97
98 static int tcf_nat_cleanup(struct tc_action *a, int bind)
99 {
100         struct tcf_nat *p = a->priv;
101
102         return tcf_hash_release(&p->common, bind, &nat_hash_info);
103 }
104
105 static int tcf_nat(struct sk_buff *skb, struct tc_action *a,
106                    struct tcf_result *res)
107 {
108         struct tcf_nat *p = a->priv;
109         struct iphdr *iph;
110         __be32 old_addr;
111         __be32 new_addr;
112         __be32 mask;
113         __be32 addr;
114         int egress;
115         int action;
116         int ihl;
117         int noff;
118
119         spin_lock(&p->tcf_lock);
120
121         p->tcf_tm.lastuse = jiffies;
122         old_addr = p->old_addr;
123         new_addr = p->new_addr;
124         mask = p->mask;
125         egress = p->flags & TCA_NAT_FLAG_EGRESS;
126         action = p->tcf_action;
127
128         p->tcf_bstats.bytes += qdisc_pkt_len(skb);
129         p->tcf_bstats.packets++;
130
131         spin_unlock(&p->tcf_lock);
132
133         if (unlikely(action == TC_ACT_SHOT))
134                 goto drop;
135
136         noff = skb_network_offset(skb);
137         if (!pskb_may_pull(skb, sizeof(*iph) + noff))
138                 goto drop;
139
140         iph = ip_hdr(skb);
141
142         if (egress)
143                 addr = iph->saddr;
144         else
145                 addr = iph->daddr;
146
147         if (!((old_addr ^ addr) & mask)) {
148                 if (skb_cloned(skb) &&
149                     !skb_clone_writable(skb, sizeof(*iph) + noff) &&
150                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
151                         goto drop;
152
153                 new_addr &= mask;
154                 new_addr |= addr & ~mask;
155
156                 /* Rewrite IP header */
157                 iph = ip_hdr(skb);
158                 if (egress)
159                         iph->saddr = new_addr;
160                 else
161                         iph->daddr = new_addr;
162
163                 csum_replace4(&iph->check, addr, new_addr);
164         } else if ((iph->frag_off & htons(IP_OFFSET)) ||
165                    iph->protocol != IPPROTO_ICMP) {
166                 goto out;
167         }
168
169         ihl = iph->ihl * 4;
170
171         /* It would be nice to share code with stateful NAT. */
172         switch (iph->frag_off & htons(IP_OFFSET) ? 0 : iph->protocol) {
173         case IPPROTO_TCP:
174         {
175                 struct tcphdr *tcph;
176
177                 if (!pskb_may_pull(skb, ihl + sizeof(*tcph) + noff) ||
178                     (skb_cloned(skb) &&
179                      !skb_clone_writable(skb, ihl + sizeof(*tcph) + noff) &&
180                      pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
181                         goto drop;
182
183                 tcph = (void *)(skb_network_header(skb) + ihl);
184                 inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr, 1);
185                 break;
186         }
187         case IPPROTO_UDP:
188         {
189                 struct udphdr *udph;
190
191                 if (!pskb_may_pull(skb, ihl + sizeof(*udph) + noff) ||
192                     (skb_cloned(skb) &&
193                      !skb_clone_writable(skb, ihl + sizeof(*udph) + noff) &&
194                      pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
195                         goto drop;
196
197                 udph = (void *)(skb_network_header(skb) + ihl);
198                 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
199                         inet_proto_csum_replace4(&udph->check, skb, addr,
200                                                  new_addr, 1);
201                         if (!udph->check)
202                                 udph->check = CSUM_MANGLED_0;
203                 }
204                 break;
205         }
206         case IPPROTO_ICMP:
207         {
208                 struct icmphdr *icmph;
209
210                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + noff))
211                         goto drop;
212
213                 icmph = (void *)(skb_network_header(skb) + ihl);
214
215                 if ((icmph->type != ICMP_DEST_UNREACH) &&
216                     (icmph->type != ICMP_TIME_EXCEEDED) &&
217                     (icmph->type != ICMP_PARAMETERPROB))
218                         break;
219
220                 if (!pskb_may_pull(skb, ihl + sizeof(*icmph) + sizeof(*iph) +
221                                         noff))
222                         goto drop;
223
224                 icmph = (void *)(skb_network_header(skb) + ihl);
225                 iph = (void *)(icmph + 1);
226                 if (egress)
227                         addr = iph->daddr;
228                 else
229                         addr = iph->saddr;
230
231                 if ((old_addr ^ addr) & mask)
232                         break;
233
234                 if (skb_cloned(skb) &&
235                     !skb_clone_writable(skb, ihl + sizeof(*icmph) +
236                                              sizeof(*iph) + noff) &&
237                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
238                         goto drop;
239
240                 icmph = (void *)(skb_network_header(skb) + ihl);
241                 iph = (void *)(icmph + 1);
242
243                 new_addr &= mask;
244                 new_addr |= addr & ~mask;
245
246                 /* XXX Fix up the inner checksums. */
247                 if (egress)
248                         iph->daddr = new_addr;
249                 else
250                         iph->saddr = new_addr;
251
252                 inet_proto_csum_replace4(&icmph->checksum, skb, addr, new_addr,
253                                          0);
254                 break;
255         }
256         default:
257                 break;
258         }
259
260 out:
261         return action;
262
263 drop:
264         spin_lock(&p->tcf_lock);
265         p->tcf_qstats.drops++;
266         spin_unlock(&p->tcf_lock);
267         return TC_ACT_SHOT;
268 }
269
270 static int tcf_nat_dump(struct sk_buff *skb, struct tc_action *a,
271                         int bind, int ref)
272 {
273         unsigned char *b = skb_tail_pointer(skb);
274         struct tcf_nat *p = a->priv;
275         struct tc_nat opt;
276         struct tcf_t t;
277
278         opt.old_addr = p->old_addr;
279         opt.new_addr = p->new_addr;
280         opt.mask = p->mask;
281         opt.flags = p->flags;
282
283         opt.index = p->tcf_index;
284         opt.action = p->tcf_action;
285         opt.refcnt = p->tcf_refcnt - ref;
286         opt.bindcnt = p->tcf_bindcnt - bind;
287
288         NLA_PUT(skb, TCA_NAT_PARMS, sizeof(opt), &opt);
289         t.install = jiffies_to_clock_t(jiffies - p->tcf_tm.install);
290         t.lastuse = jiffies_to_clock_t(jiffies - p->tcf_tm.lastuse);
291         t.expires = jiffies_to_clock_t(p->tcf_tm.expires);
292         NLA_PUT(skb, TCA_NAT_TM, sizeof(t), &t);
293
294         return skb->len;
295
296 nla_put_failure:
297         nlmsg_trim(skb, b);
298         return -1;
299 }
300
301 static struct tc_action_ops act_nat_ops = {
302         .kind           =       "nat",
303         .hinfo          =       &nat_hash_info,
304         .type           =       TCA_ACT_NAT,
305         .capab          =       TCA_CAP_NONE,
306         .owner          =       THIS_MODULE,
307         .act            =       tcf_nat,
308         .dump           =       tcf_nat_dump,
309         .cleanup        =       tcf_nat_cleanup,
310         .lookup         =       tcf_hash_search,
311         .init           =       tcf_nat_init,
312         .walk           =       tcf_generic_walker
313 };
314
315 MODULE_DESCRIPTION("Stateless NAT actions");
316 MODULE_LICENSE("GPL");
317
318 static int __init nat_init_module(void)
319 {
320         return tcf_register_action(&act_nat_ops);
321 }
322
323 static void __exit nat_cleanup_module(void)
324 {
325         tcf_unregister_action(&act_nat_ops);
326 }
327
328 module_init(nat_init_module);
329 module_exit(nat_cleanup_module);