0fe8fb0466ef2767741232f78305c69f26bc0f4a
[linux-2.6.git] / net / ipv4 / netfilter / nf_conntrack_proto_icmp.c
1 /* (C) 1999-2001 Paul `Rusty' Russell
2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8
9 #include <linux/types.h>
10 #include <linux/timer.h>
11 #include <linux/netfilter.h>
12 #include <linux/in.h>
13 #include <linux/icmp.h>
14 #include <linux/seq_file.h>
15 #include <net/ip.h>
16 #include <net/checksum.h>
17 #include <linux/netfilter_ipv4.h>
18 #include <net/netfilter/nf_conntrack_tuple.h>
19 #include <net/netfilter/nf_conntrack_l4proto.h>
20 #include <net/netfilter/nf_conntrack_core.h>
21
22 static unsigned long nf_ct_icmp_timeout __read_mostly = 30*HZ;
23
24 static int icmp_pkt_to_tuple(const struct sk_buff *skb,
25                              unsigned int dataoff,
26                              struct nf_conntrack_tuple *tuple)
27 {
28         struct icmphdr _hdr, *hp;
29
30         hp = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
31         if (hp == NULL)
32                 return 0;
33
34         tuple->dst.u.icmp.type = hp->type;
35         tuple->src.u.icmp.id = hp->un.echo.id;
36         tuple->dst.u.icmp.code = hp->code;
37
38         return 1;
39 }
40
41 /* Add 1; spaces filled with 0. */
42 static const u_int8_t invmap[] = {
43         [ICMP_ECHO] = ICMP_ECHOREPLY + 1,
44         [ICMP_ECHOREPLY] = ICMP_ECHO + 1,
45         [ICMP_TIMESTAMP] = ICMP_TIMESTAMPREPLY + 1,
46         [ICMP_TIMESTAMPREPLY] = ICMP_TIMESTAMP + 1,
47         [ICMP_INFO_REQUEST] = ICMP_INFO_REPLY + 1,
48         [ICMP_INFO_REPLY] = ICMP_INFO_REQUEST + 1,
49         [ICMP_ADDRESS] = ICMP_ADDRESSREPLY + 1,
50         [ICMP_ADDRESSREPLY] = ICMP_ADDRESS + 1
51 };
52
53 static int icmp_invert_tuple(struct nf_conntrack_tuple *tuple,
54                              const struct nf_conntrack_tuple *orig)
55 {
56         if (orig->dst.u.icmp.type >= sizeof(invmap)
57             || !invmap[orig->dst.u.icmp.type])
58                 return 0;
59
60         tuple->src.u.icmp.id = orig->src.u.icmp.id;
61         tuple->dst.u.icmp.type = invmap[orig->dst.u.icmp.type] - 1;
62         tuple->dst.u.icmp.code = orig->dst.u.icmp.code;
63         return 1;
64 }
65
66 /* Print out the per-protocol part of the tuple. */
67 static int icmp_print_tuple(struct seq_file *s,
68                             const struct nf_conntrack_tuple *tuple)
69 {
70         return seq_printf(s, "type=%u code=%u id=%u ",
71                           tuple->dst.u.icmp.type,
72                           tuple->dst.u.icmp.code,
73                           ntohs(tuple->src.u.icmp.id));
74 }
75
76 /* Print out the private part of the conntrack. */
77 static int icmp_print_conntrack(struct seq_file *s,
78                                 const struct nf_conn *conntrack)
79 {
80         return 0;
81 }
82
83 /* Returns verdict for packet, or -1 for invalid. */
84 static int icmp_packet(struct nf_conn *ct,
85                        const struct sk_buff *skb,
86                        unsigned int dataoff,
87                        enum ip_conntrack_info ctinfo,
88                        int pf,
89                        unsigned int hooknum)
90 {
91         /* Try to delete connection immediately after all replies:
92            won't actually vanish as we still have skb, and del_timer
93            means this will only run once even if count hits zero twice
94            (theoretically possible with SMP) */
95         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) {
96                 if (atomic_dec_and_test(&ct->proto.icmp.count)
97                     && del_timer(&ct->timeout))
98                         ct->timeout.function((unsigned long)ct);
99         } else {
100                 atomic_inc(&ct->proto.icmp.count);
101                 nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
102                 nf_ct_refresh_acct(ct, ctinfo, skb, nf_ct_icmp_timeout);
103         }
104
105         return NF_ACCEPT;
106 }
107
108 /* Called when a new connection for this protocol found. */
109 static int icmp_new(struct nf_conn *conntrack,
110                     const struct sk_buff *skb, unsigned int dataoff)
111 {
112         static const u_int8_t valid_new[] = {
113                 [ICMP_ECHO] = 1,
114                 [ICMP_TIMESTAMP] = 1,
115                 [ICMP_INFO_REQUEST] = 1,
116                 [ICMP_ADDRESS] = 1
117         };
118
119         if (conntrack->tuplehash[0].tuple.dst.u.icmp.type >= sizeof(valid_new)
120             || !valid_new[conntrack->tuplehash[0].tuple.dst.u.icmp.type]) {
121                 /* Can't create a new ICMP `conn' with this. */
122                 pr_debug("icmp: can't create new conn with type %u\n",
123                          conntrack->tuplehash[0].tuple.dst.u.icmp.type);
124                 NF_CT_DUMP_TUPLE(&conntrack->tuplehash[0].tuple);
125                 return 0;
126         }
127         atomic_set(&conntrack->proto.icmp.count, 0);
128         return 1;
129 }
130
131 extern struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4;
132 /* Returns conntrack if it dealt with ICMP, and filled in skb fields */
133 static int
134 icmp_error_message(struct sk_buff *skb,
135                  enum ip_conntrack_info *ctinfo,
136                  unsigned int hooknum)
137 {
138         struct nf_conntrack_tuple innertuple, origtuple;
139         struct {
140                 struct icmphdr icmp;
141                 struct iphdr ip;
142         } _in, *inside;
143         struct nf_conntrack_l4proto *innerproto;
144         struct nf_conntrack_tuple_hash *h;
145         int dataoff;
146
147         NF_CT_ASSERT(skb->nfct == NULL);
148
149         /* Not enough header? */
150         inside = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_in), &_in);
151         if (inside == NULL)
152                 return -NF_ACCEPT;
153
154         /* Ignore ICMP's containing fragments (shouldn't happen) */
155         if (inside->ip.frag_off & htons(IP_OFFSET)) {
156                 pr_debug("icmp_error_message: fragment of proto %u\n",
157                          inside->ip.protocol);
158                 return -NF_ACCEPT;
159         }
160
161         /* rcu_read_lock()ed by nf_hook_slow */
162         innerproto = __nf_ct_l4proto_find(PF_INET, inside->ip.protocol);
163
164         dataoff = ip_hdrlen(skb) + sizeof(inside->icmp);
165         /* Are they talking about one of our connections? */
166         if (!nf_ct_get_tuple(skb, dataoff, dataoff + inside->ip.ihl*4, PF_INET,
167                              inside->ip.protocol, &origtuple,
168                              &nf_conntrack_l3proto_ipv4, innerproto)) {
169                 pr_debug("icmp_error_message: ! get_tuple p=%u",
170                          inside->ip.protocol);
171                 return -NF_ACCEPT;
172         }
173
174         /* Ordinarily, we'd expect the inverted tupleproto, but it's
175            been preserved inside the ICMP. */
176         if (!nf_ct_invert_tuple(&innertuple, &origtuple,
177                                 &nf_conntrack_l3proto_ipv4, innerproto)) {
178                 pr_debug("icmp_error_message: no match\n");
179                 return -NF_ACCEPT;
180         }
181
182         *ctinfo = IP_CT_RELATED;
183
184         h = nf_conntrack_find_get(&innertuple);
185         if (!h) {
186                 /* Locally generated ICMPs will match inverted if they
187                    haven't been SNAT'ed yet */
188                 /* FIXME: NAT code has to handle half-done double NAT --RR */
189                 if (hooknum == NF_IP_LOCAL_OUT)
190                         h = nf_conntrack_find_get(&origtuple);
191
192                 if (!h) {
193                         pr_debug("icmp_error_message: no match\n");
194                         return -NF_ACCEPT;
195                 }
196
197                 /* Reverse direction from that found */
198                 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
199                         *ctinfo += IP_CT_IS_REPLY;
200         } else {
201                 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
202                         *ctinfo += IP_CT_IS_REPLY;
203         }
204
205         /* Update skb to refer to this connection */
206         skb->nfct = &nf_ct_tuplehash_to_ctrack(h)->ct_general;
207         skb->nfctinfo = *ctinfo;
208         return -NF_ACCEPT;
209 }
210
211 /* Small and modified version of icmp_rcv */
212 static int
213 icmp_error(struct sk_buff *skb, unsigned int dataoff,
214            enum ip_conntrack_info *ctinfo, int pf, unsigned int hooknum)
215 {
216         struct icmphdr _ih, *icmph;
217
218         /* Not enough header? */
219         icmph = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_ih), &_ih);
220         if (icmph == NULL) {
221                 if (LOG_INVALID(IPPROTO_ICMP))
222                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
223                                       "nf_ct_icmp: short packet ");
224                 return -NF_ACCEPT;
225         }
226
227         /* See ip_conntrack_proto_tcp.c */
228         if (nf_conntrack_checksum && hooknum == NF_IP_PRE_ROUTING &&
229             nf_ip_checksum(skb, hooknum, dataoff, 0)) {
230                 if (LOG_INVALID(IPPROTO_ICMP))
231                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
232                                       "nf_ct_icmp: bad HW ICMP checksum ");
233                 return -NF_ACCEPT;
234         }
235
236         /*
237          *      18 is the highest 'known' ICMP type. Anything else is a mystery
238          *
239          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
240          *                discarded.
241          */
242         if (icmph->type > NR_ICMP_TYPES) {
243                 if (LOG_INVALID(IPPROTO_ICMP))
244                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
245                                       "nf_ct_icmp: invalid ICMP type ");
246                 return -NF_ACCEPT;
247         }
248
249         /* Need to track icmp error message? */
250         if (icmph->type != ICMP_DEST_UNREACH
251             && icmph->type != ICMP_SOURCE_QUENCH
252             && icmph->type != ICMP_TIME_EXCEEDED
253             && icmph->type != ICMP_PARAMETERPROB
254             && icmph->type != ICMP_REDIRECT)
255                 return NF_ACCEPT;
256
257         return icmp_error_message(skb, ctinfo, hooknum);
258 }
259
260 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
261
262 #include <linux/netfilter/nfnetlink.h>
263 #include <linux/netfilter/nfnetlink_conntrack.h>
264
265 static int icmp_tuple_to_nfattr(struct sk_buff *skb,
266                                 const struct nf_conntrack_tuple *t)
267 {
268         NFA_PUT(skb, CTA_PROTO_ICMP_ID, sizeof(u_int16_t),
269                 &t->src.u.icmp.id);
270         NFA_PUT(skb, CTA_PROTO_ICMP_TYPE, sizeof(u_int8_t),
271                 &t->dst.u.icmp.type);
272         NFA_PUT(skb, CTA_PROTO_ICMP_CODE, sizeof(u_int8_t),
273                 &t->dst.u.icmp.code);
274
275         return 0;
276
277 nfattr_failure:
278         return -1;
279 }
280
281 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
282         [CTA_PROTO_ICMP_TYPE-1] = sizeof(u_int8_t),
283         [CTA_PROTO_ICMP_CODE-1] = sizeof(u_int8_t),
284         [CTA_PROTO_ICMP_ID-1]   = sizeof(u_int16_t)
285 };
286
287 static int icmp_nfattr_to_tuple(struct nfattr *tb[],
288                                 struct nf_conntrack_tuple *tuple)
289 {
290         if (!tb[CTA_PROTO_ICMP_TYPE-1]
291             || !tb[CTA_PROTO_ICMP_CODE-1]
292             || !tb[CTA_PROTO_ICMP_ID-1])
293                 return -EINVAL;
294
295         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
296                 return -EINVAL;
297
298         tuple->dst.u.icmp.type =
299                         *(u_int8_t *)NFA_DATA(tb[CTA_PROTO_ICMP_TYPE-1]);
300         tuple->dst.u.icmp.code =
301                         *(u_int8_t *)NFA_DATA(tb[CTA_PROTO_ICMP_CODE-1]);
302         tuple->src.u.icmp.id =
303                         *(__be16 *)NFA_DATA(tb[CTA_PROTO_ICMP_ID-1]);
304
305         if (tuple->dst.u.icmp.type >= sizeof(invmap)
306             || !invmap[tuple->dst.u.icmp.type])
307                 return -EINVAL;
308
309         return 0;
310 }
311 #endif
312
313 #ifdef CONFIG_SYSCTL
314 static struct ctl_table_header *icmp_sysctl_header;
315 static struct ctl_table icmp_sysctl_table[] = {
316         {
317                 .ctl_name       = NET_NF_CONNTRACK_ICMP_TIMEOUT,
318                 .procname       = "nf_conntrack_icmp_timeout",
319                 .data           = &nf_ct_icmp_timeout,
320                 .maxlen         = sizeof(unsigned int),
321                 .mode           = 0644,
322                 .proc_handler   = &proc_dointvec_jiffies,
323         },
324         {
325                 .ctl_name = 0
326         }
327 };
328 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
329 static struct ctl_table icmp_compat_sysctl_table[] = {
330         {
331                 .ctl_name       = NET_IPV4_NF_CONNTRACK_ICMP_TIMEOUT,
332                 .procname       = "ip_conntrack_icmp_timeout",
333                 .data           = &nf_ct_icmp_timeout,
334                 .maxlen         = sizeof(unsigned int),
335                 .mode           = 0644,
336                 .proc_handler   = &proc_dointvec_jiffies,
337         },
338         {
339                 .ctl_name = 0
340         }
341 };
342 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
343 #endif /* CONFIG_SYSCTL */
344
345 struct nf_conntrack_l4proto nf_conntrack_l4proto_icmp =
346 {
347         .l3proto                = PF_INET,
348         .l4proto                = IPPROTO_ICMP,
349         .name                   = "icmp",
350         .pkt_to_tuple           = icmp_pkt_to_tuple,
351         .invert_tuple           = icmp_invert_tuple,
352         .print_tuple            = icmp_print_tuple,
353         .print_conntrack        = icmp_print_conntrack,
354         .packet                 = icmp_packet,
355         .new                    = icmp_new,
356         .error                  = icmp_error,
357         .destroy                = NULL,
358         .me                     = NULL,
359 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
360         .tuple_to_nfattr        = icmp_tuple_to_nfattr,
361         .nfattr_to_tuple        = icmp_nfattr_to_tuple,
362 #endif
363 #ifdef CONFIG_SYSCTL
364         .ctl_table_header       = &icmp_sysctl_header,
365         .ctl_table              = icmp_sysctl_table,
366 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
367         .ctl_compat_table       = icmp_compat_sysctl_table,
368 #endif
369 #endif
370 };
371 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_icmp);