netfilter: nf_conntrack: calculate per-protocol nlattr size
[linux-2.6.git] / net / ipv4 / netfilter / nf_conntrack_l3proto_ipv4.c
1
2 /* (C) 1999-2001 Paul `Rusty' Russell
3  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License version 2 as
7  * published by the Free Software Foundation.
8  */
9
10 #include <linux/types.h>
11 #include <linux/ip.h>
12 #include <linux/netfilter.h>
13 #include <linux/module.h>
14 #include <linux/skbuff.h>
15 #include <linux/icmp.h>
16 #include <linux/sysctl.h>
17 #include <net/route.h>
18 #include <net/ip.h>
19
20 #include <linux/netfilter_ipv4.h>
21 #include <net/netfilter/nf_conntrack.h>
22 #include <net/netfilter/nf_conntrack_helper.h>
23 #include <net/netfilter/nf_conntrack_l4proto.h>
24 #include <net/netfilter/nf_conntrack_l3proto.h>
25 #include <net/netfilter/nf_conntrack_core.h>
26 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
27 #include <net/netfilter/nf_nat_helper.h>
28 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
29
30 int (*nf_nat_seq_adjust_hook)(struct sk_buff *skb,
31                               struct nf_conn *ct,
32                               enum ip_conntrack_info ctinfo);
33 EXPORT_SYMBOL_GPL(nf_nat_seq_adjust_hook);
34
35 static bool ipv4_pkt_to_tuple(const struct sk_buff *skb, unsigned int nhoff,
36                               struct nf_conntrack_tuple *tuple)
37 {
38         const __be32 *ap;
39         __be32 _addrs[2];
40         ap = skb_header_pointer(skb, nhoff + offsetof(struct iphdr, saddr),
41                                 sizeof(u_int32_t) * 2, _addrs);
42         if (ap == NULL)
43                 return false;
44
45         tuple->src.u3.ip = ap[0];
46         tuple->dst.u3.ip = ap[1];
47
48         return true;
49 }
50
51 static bool ipv4_invert_tuple(struct nf_conntrack_tuple *tuple,
52                               const struct nf_conntrack_tuple *orig)
53 {
54         tuple->src.u3.ip = orig->dst.u3.ip;
55         tuple->dst.u3.ip = orig->src.u3.ip;
56
57         return true;
58 }
59
60 static int ipv4_print_tuple(struct seq_file *s,
61                             const struct nf_conntrack_tuple *tuple)
62 {
63         return seq_printf(s, "src=%pI4 dst=%pI4 ",
64                           &tuple->src.u3.ip, &tuple->dst.u3.ip);
65 }
66
67 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
68                             unsigned int *dataoff, u_int8_t *protonum)
69 {
70         const struct iphdr *iph;
71         struct iphdr _iph;
72
73         iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
74         if (iph == NULL)
75                 return -NF_DROP;
76
77         /* Conntrack defragments packets, we might still see fragments
78          * inside ICMP packets though. */
79         if (iph->frag_off & htons(IP_OFFSET))
80                 return -NF_DROP;
81
82         *dataoff = nhoff + (iph->ihl << 2);
83         *protonum = iph->protocol;
84
85         return NF_ACCEPT;
86 }
87
88 static unsigned int ipv4_confirm(unsigned int hooknum,
89                                  struct sk_buff *skb,
90                                  const struct net_device *in,
91                                  const struct net_device *out,
92                                  int (*okfn)(struct sk_buff *))
93 {
94         struct nf_conn *ct;
95         enum ip_conntrack_info ctinfo;
96         const struct nf_conn_help *help;
97         const struct nf_conntrack_helper *helper;
98         unsigned int ret;
99
100         /* This is where we call the helper: as the packet goes out. */
101         ct = nf_ct_get(skb, &ctinfo);
102         if (!ct || ctinfo == IP_CT_RELATED + IP_CT_IS_REPLY)
103                 goto out;
104
105         help = nfct_help(ct);
106         if (!help)
107                 goto out;
108
109         /* rcu_read_lock()ed by nf_hook_slow */
110         helper = rcu_dereference(help->helper);
111         if (!helper)
112                 goto out;
113
114         ret = helper->help(skb, skb_network_offset(skb) + ip_hdrlen(skb),
115                            ct, ctinfo);
116         if (ret != NF_ACCEPT)
117                 return ret;
118
119         if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status)) {
120                 typeof(nf_nat_seq_adjust_hook) seq_adjust;
121
122                 seq_adjust = rcu_dereference(nf_nat_seq_adjust_hook);
123                 if (!seq_adjust || !seq_adjust(skb, ct, ctinfo)) {
124                         NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
125                         return NF_DROP;
126                 }
127         }
128 out:
129         /* We've seen it coming out the other side: confirm it */
130         return nf_conntrack_confirm(skb);
131 }
132
133 static unsigned int ipv4_conntrack_in(unsigned int hooknum,
134                                       struct sk_buff *skb,
135                                       const struct net_device *in,
136                                       const struct net_device *out,
137                                       int (*okfn)(struct sk_buff *))
138 {
139         return nf_conntrack_in(dev_net(in), PF_INET, hooknum, skb);
140 }
141
142 static unsigned int ipv4_conntrack_local(unsigned int hooknum,
143                                          struct sk_buff *skb,
144                                          const struct net_device *in,
145                                          const struct net_device *out,
146                                          int (*okfn)(struct sk_buff *))
147 {
148         /* root is playing with raw sockets. */
149         if (skb->len < sizeof(struct iphdr) ||
150             ip_hdrlen(skb) < sizeof(struct iphdr))
151                 return NF_ACCEPT;
152         return nf_conntrack_in(dev_net(out), PF_INET, hooknum, skb);
153 }
154
155 /* Connection tracking may drop packets, but never alters them, so
156    make it the first hook. */
157 static struct nf_hook_ops ipv4_conntrack_ops[] __read_mostly = {
158         {
159                 .hook           = ipv4_conntrack_in,
160                 .owner          = THIS_MODULE,
161                 .pf             = PF_INET,
162                 .hooknum        = NF_INET_PRE_ROUTING,
163                 .priority       = NF_IP_PRI_CONNTRACK,
164         },
165         {
166                 .hook           = ipv4_conntrack_local,
167                 .owner          = THIS_MODULE,
168                 .pf             = PF_INET,
169                 .hooknum        = NF_INET_LOCAL_OUT,
170                 .priority       = NF_IP_PRI_CONNTRACK,
171         },
172         {
173                 .hook           = ipv4_confirm,
174                 .owner          = THIS_MODULE,
175                 .pf             = PF_INET,
176                 .hooknum        = NF_INET_POST_ROUTING,
177                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
178         },
179         {
180                 .hook           = ipv4_confirm,
181                 .owner          = THIS_MODULE,
182                 .pf             = PF_INET,
183                 .hooknum        = NF_INET_LOCAL_IN,
184                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM,
185         },
186 };
187
188 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
189 static int log_invalid_proto_min = 0;
190 static int log_invalid_proto_max = 255;
191
192 static ctl_table ip_ct_sysctl_table[] = {
193         {
194                 .ctl_name       = NET_IPV4_NF_CONNTRACK_MAX,
195                 .procname       = "ip_conntrack_max",
196                 .data           = &nf_conntrack_max,
197                 .maxlen         = sizeof(int),
198                 .mode           = 0644,
199                 .proc_handler   = proc_dointvec,
200         },
201         {
202                 .ctl_name       = NET_IPV4_NF_CONNTRACK_COUNT,
203                 .procname       = "ip_conntrack_count",
204                 .data           = &init_net.ct.count,
205                 .maxlen         = sizeof(int),
206                 .mode           = 0444,
207                 .proc_handler   = proc_dointvec,
208         },
209         {
210                 .ctl_name       = NET_IPV4_NF_CONNTRACK_BUCKETS,
211                 .procname       = "ip_conntrack_buckets",
212                 .data           = &nf_conntrack_htable_size,
213                 .maxlen         = sizeof(unsigned int),
214                 .mode           = 0444,
215                 .proc_handler   = proc_dointvec,
216         },
217         {
218                 .ctl_name       = NET_IPV4_NF_CONNTRACK_CHECKSUM,
219                 .procname       = "ip_conntrack_checksum",
220                 .data           = &init_net.ct.sysctl_checksum,
221                 .maxlen         = sizeof(int),
222                 .mode           = 0644,
223                 .proc_handler   = proc_dointvec,
224         },
225         {
226                 .ctl_name       = NET_IPV4_NF_CONNTRACK_LOG_INVALID,
227                 .procname       = "ip_conntrack_log_invalid",
228                 .data           = &init_net.ct.sysctl_log_invalid,
229                 .maxlen         = sizeof(unsigned int),
230                 .mode           = 0644,
231                 .proc_handler   = proc_dointvec_minmax,
232                 .strategy       = sysctl_intvec,
233                 .extra1         = &log_invalid_proto_min,
234                 .extra2         = &log_invalid_proto_max,
235         },
236         {
237                 .ctl_name       = 0
238         }
239 };
240 #endif /* CONFIG_SYSCTL && CONFIG_NF_CONNTRACK_PROC_COMPAT */
241
242 /* Fast function for those who don't want to parse /proc (and I don't
243    blame them). */
244 /* Reversing the socket's dst/src point of view gives us the reply
245    mapping. */
246 static int
247 getorigdst(struct sock *sk, int optval, void __user *user, int *len)
248 {
249         const struct inet_sock *inet = inet_sk(sk);
250         const struct nf_conntrack_tuple_hash *h;
251         struct nf_conntrack_tuple tuple;
252
253         memset(&tuple, 0, sizeof(tuple));
254         tuple.src.u3.ip = inet->rcv_saddr;
255         tuple.src.u.tcp.port = inet->sport;
256         tuple.dst.u3.ip = inet->daddr;
257         tuple.dst.u.tcp.port = inet->dport;
258         tuple.src.l3num = PF_INET;
259         tuple.dst.protonum = IPPROTO_TCP;
260
261         /* We only do TCP at the moment: is there a better way? */
262         if (strcmp(sk->sk_prot->name, "TCP")) {
263                 pr_debug("SO_ORIGINAL_DST: Not a TCP socket\n");
264                 return -ENOPROTOOPT;
265         }
266
267         if ((unsigned int) *len < sizeof(struct sockaddr_in)) {
268                 pr_debug("SO_ORIGINAL_DST: len %d not %Zu\n",
269                          *len, sizeof(struct sockaddr_in));
270                 return -EINVAL;
271         }
272
273         h = nf_conntrack_find_get(sock_net(sk), &tuple);
274         if (h) {
275                 struct sockaddr_in sin;
276                 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
277
278                 sin.sin_family = AF_INET;
279                 sin.sin_port = ct->tuplehash[IP_CT_DIR_ORIGINAL]
280                         .tuple.dst.u.tcp.port;
281                 sin.sin_addr.s_addr = ct->tuplehash[IP_CT_DIR_ORIGINAL]
282                         .tuple.dst.u3.ip;
283                 memset(sin.sin_zero, 0, sizeof(sin.sin_zero));
284
285                 pr_debug("SO_ORIGINAL_DST: %pI4 %u\n",
286                          &sin.sin_addr.s_addr, ntohs(sin.sin_port));
287                 nf_ct_put(ct);
288                 if (copy_to_user(user, &sin, sizeof(sin)) != 0)
289                         return -EFAULT;
290                 else
291                         return 0;
292         }
293         pr_debug("SO_ORIGINAL_DST: Can't find %pI4/%u-%pI4/%u.\n",
294                  &tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
295                  &tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
296         return -ENOENT;
297 }
298
299 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
300
301 #include <linux/netfilter/nfnetlink.h>
302 #include <linux/netfilter/nfnetlink_conntrack.h>
303
304 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
305                                 const struct nf_conntrack_tuple *tuple)
306 {
307         NLA_PUT_BE32(skb, CTA_IP_V4_SRC, tuple->src.u3.ip);
308         NLA_PUT_BE32(skb, CTA_IP_V4_DST, tuple->dst.u3.ip);
309         return 0;
310
311 nla_put_failure:
312         return -1;
313 }
314
315 static const struct nla_policy ipv4_nla_policy[CTA_IP_MAX+1] = {
316         [CTA_IP_V4_SRC] = { .type = NLA_U32 },
317         [CTA_IP_V4_DST] = { .type = NLA_U32 },
318 };
319
320 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
321                                 struct nf_conntrack_tuple *t)
322 {
323         if (!tb[CTA_IP_V4_SRC] || !tb[CTA_IP_V4_DST])
324                 return -EINVAL;
325
326         t->src.u3.ip = nla_get_be32(tb[CTA_IP_V4_SRC]);
327         t->dst.u3.ip = nla_get_be32(tb[CTA_IP_V4_DST]);
328
329         return 0;
330 }
331
332 static int ipv4_nlattr_tuple_size(void)
333 {
334         return nla_policy_len(ipv4_nla_policy, CTA_IP_MAX + 1);
335 }
336 #endif
337
338 static struct nf_sockopt_ops so_getorigdst = {
339         .pf             = PF_INET,
340         .get_optmin     = SO_ORIGINAL_DST,
341         .get_optmax     = SO_ORIGINAL_DST+1,
342         .get            = &getorigdst,
343         .owner          = THIS_MODULE,
344 };
345
346 struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4 __read_mostly = {
347         .l3proto         = PF_INET,
348         .name            = "ipv4",
349         .pkt_to_tuple    = ipv4_pkt_to_tuple,
350         .invert_tuple    = ipv4_invert_tuple,
351         .print_tuple     = ipv4_print_tuple,
352         .get_l4proto     = ipv4_get_l4proto,
353 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
354         .tuple_to_nlattr = ipv4_tuple_to_nlattr,
355         .nlattr_tuple_size = ipv4_nlattr_tuple_size,
356         .nlattr_to_tuple = ipv4_nlattr_to_tuple,
357         .nla_policy      = ipv4_nla_policy,
358 #endif
359 #if defined(CONFIG_SYSCTL) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
360         .ctl_table_path  = nf_net_ipv4_netfilter_sysctl_path,
361         .ctl_table       = ip_ct_sysctl_table,
362 #endif
363         .me              = THIS_MODULE,
364 };
365
366 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
367                   &nf_conntrack_htable_size, 0600);
368
369 MODULE_ALIAS("nf_conntrack-" __stringify(AF_INET));
370 MODULE_ALIAS("ip_conntrack");
371 MODULE_LICENSE("GPL");
372
373 static int __init nf_conntrack_l3proto_ipv4_init(void)
374 {
375         int ret = 0;
376
377         need_conntrack();
378         nf_defrag_ipv4_enable();
379
380         ret = nf_register_sockopt(&so_getorigdst);
381         if (ret < 0) {
382                 printk(KERN_ERR "Unable to register netfilter socket option\n");
383                 return ret;
384         }
385
386         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_tcp4);
387         if (ret < 0) {
388                 printk("nf_conntrack_ipv4: can't register tcp.\n");
389                 goto cleanup_sockopt;
390         }
391
392         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_udp4);
393         if (ret < 0) {
394                 printk("nf_conntrack_ipv4: can't register udp.\n");
395                 goto cleanup_tcp;
396         }
397
398         ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_icmp);
399         if (ret < 0) {
400                 printk("nf_conntrack_ipv4: can't register icmp.\n");
401                 goto cleanup_udp;
402         }
403
404         ret = nf_conntrack_l3proto_register(&nf_conntrack_l3proto_ipv4);
405         if (ret < 0) {
406                 printk("nf_conntrack_ipv4: can't register ipv4\n");
407                 goto cleanup_icmp;
408         }
409
410         ret = nf_register_hooks(ipv4_conntrack_ops,
411                                 ARRAY_SIZE(ipv4_conntrack_ops));
412         if (ret < 0) {
413                 printk("nf_conntrack_ipv4: can't register hooks.\n");
414                 goto cleanup_ipv4;
415         }
416 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
417         ret = nf_conntrack_ipv4_compat_init();
418         if (ret < 0)
419                 goto cleanup_hooks;
420 #endif
421         return ret;
422 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
423  cleanup_hooks:
424         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
425 #endif
426  cleanup_ipv4:
427         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
428  cleanup_icmp:
429         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
430  cleanup_udp:
431         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
432  cleanup_tcp:
433         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
434  cleanup_sockopt:
435         nf_unregister_sockopt(&so_getorigdst);
436         return ret;
437 }
438
439 static void __exit nf_conntrack_l3proto_ipv4_fini(void)
440 {
441         synchronize_net();
442 #if defined(CONFIG_PROC_FS) && defined(CONFIG_NF_CONNTRACK_PROC_COMPAT)
443         nf_conntrack_ipv4_compat_fini();
444 #endif
445         nf_unregister_hooks(ipv4_conntrack_ops, ARRAY_SIZE(ipv4_conntrack_ops));
446         nf_conntrack_l3proto_unregister(&nf_conntrack_l3proto_ipv4);
447         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_icmp);
448         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_udp4);
449         nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_tcp4);
450         nf_unregister_sockopt(&so_getorigdst);
451 }
452
453 module_init(nf_conntrack_l3proto_ipv4_init);
454 module_exit(nf_conntrack_l3proto_ipv4_fini);
455
456 void need_ipv4_conntrack(void)
457 {
458         return;
459 }
460 EXPORT_SYMBOL_GPL(need_ipv4_conntrack);