[NETFILTER]: conntrack: add sysctl to disable checksumming
[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  * 16 Dec 2003: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
9  *      - enable working with Layer 3 protocol independent connection tracking.
10  *
11  * Derived from net/ipv4/netfilter/ip_conntrack_proto_icmp.c
12  */
13
14 #include <linux/types.h>
15 #include <linux/sched.h>
16 #include <linux/timer.h>
17 #include <linux/netfilter.h>
18 #include <linux/in.h>
19 #include <linux/icmp.h>
20 #include <linux/seq_file.h>
21 #include <net/ip.h>
22 #include <net/checksum.h>
23 #include <linux/netfilter_ipv4.h>
24 #include <net/netfilter/nf_conntrack_tuple.h>
25 #include <net/netfilter/nf_conntrack_protocol.h>
26 #include <net/netfilter/nf_conntrack_core.h>
27
28 unsigned long nf_ct_icmp_timeout = 30*HZ;
29
30 #if 0
31 #define DEBUGP printk
32 #else
33 #define DEBUGP(format, args...)
34 #endif
35
36 static int icmp_pkt_to_tuple(const struct sk_buff *skb,
37                              unsigned int dataoff,
38                              struct nf_conntrack_tuple *tuple)
39 {
40         struct icmphdr _hdr, *hp;
41
42         hp = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
43         if (hp == NULL)
44                 return 0;
45
46         tuple->dst.u.icmp.type = hp->type;
47         tuple->src.u.icmp.id = hp->un.echo.id;
48         tuple->dst.u.icmp.code = hp->code;
49
50         return 1;
51 }
52
53 /* Add 1; spaces filled with 0. */
54 static const u_int8_t invmap[] = {
55         [ICMP_ECHO] = ICMP_ECHOREPLY + 1,
56         [ICMP_ECHOREPLY] = ICMP_ECHO + 1,
57         [ICMP_TIMESTAMP] = ICMP_TIMESTAMPREPLY + 1,
58         [ICMP_TIMESTAMPREPLY] = ICMP_TIMESTAMP + 1,
59         [ICMP_INFO_REQUEST] = ICMP_INFO_REPLY + 1,
60         [ICMP_INFO_REPLY] = ICMP_INFO_REQUEST + 1,
61         [ICMP_ADDRESS] = ICMP_ADDRESSREPLY + 1,
62         [ICMP_ADDRESSREPLY] = ICMP_ADDRESS + 1
63 };
64
65 static int icmp_invert_tuple(struct nf_conntrack_tuple *tuple,
66                              const struct nf_conntrack_tuple *orig)
67 {
68         if (orig->dst.u.icmp.type >= sizeof(invmap)
69             || !invmap[orig->dst.u.icmp.type])
70                 return 0;
71
72         tuple->src.u.icmp.id = orig->src.u.icmp.id;
73         tuple->dst.u.icmp.type = invmap[orig->dst.u.icmp.type] - 1;
74         tuple->dst.u.icmp.code = orig->dst.u.icmp.code;
75         return 1;
76 }
77
78 /* Print out the per-protocol part of the tuple. */
79 static int icmp_print_tuple(struct seq_file *s,
80                             const struct nf_conntrack_tuple *tuple)
81 {
82         return seq_printf(s, "type=%u code=%u id=%u ",
83                           tuple->dst.u.icmp.type,
84                           tuple->dst.u.icmp.code,
85                           ntohs(tuple->src.u.icmp.id));
86 }
87
88 /* Print out the private part of the conntrack. */
89 static int icmp_print_conntrack(struct seq_file *s,
90                                 const struct nf_conn *conntrack)
91 {
92         return 0;
93 }
94
95 /* Returns verdict for packet, or -1 for invalid. */
96 static int icmp_packet(struct nf_conn *ct,
97                        const struct sk_buff *skb,
98                        unsigned int dataoff,
99                        enum ip_conntrack_info ctinfo,
100                        int pf,
101                        unsigned int hooknum)
102 {
103         /* Try to delete connection immediately after all replies:
104            won't actually vanish as we still have skb, and del_timer
105            means this will only run once even if count hits zero twice
106            (theoretically possible with SMP) */
107         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) {
108                 if (atomic_dec_and_test(&ct->proto.icmp.count)
109                     && del_timer(&ct->timeout))
110                         ct->timeout.function((unsigned long)ct);
111         } else {
112                 atomic_inc(&ct->proto.icmp.count);
113                 nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
114                 nf_ct_refresh_acct(ct, ctinfo, skb, nf_ct_icmp_timeout);
115         }
116
117         return NF_ACCEPT;
118 }
119
120 /* Called when a new connection for this protocol found. */
121 static int icmp_new(struct nf_conn *conntrack,
122                     const struct sk_buff *skb, unsigned int dataoff)
123 {
124         static const u_int8_t valid_new[] = {
125                 [ICMP_ECHO] = 1,
126                 [ICMP_TIMESTAMP] = 1,
127                 [ICMP_INFO_REQUEST] = 1,
128                 [ICMP_ADDRESS] = 1
129         };
130
131         if (conntrack->tuplehash[0].tuple.dst.u.icmp.type >= sizeof(valid_new)
132             || !valid_new[conntrack->tuplehash[0].tuple.dst.u.icmp.type]) {
133                 /* Can't create a new ICMP `conn' with this. */
134                 DEBUGP("icmp: can't create new conn with type %u\n",
135                        conntrack->tuplehash[0].tuple.dst.u.icmp.type);
136                 NF_CT_DUMP_TUPLE(&conntrack->tuplehash[0].tuple);
137                 return 0;
138         }
139         atomic_set(&conntrack->proto.icmp.count, 0);
140         return 1;
141 }
142
143 extern struct nf_conntrack_l3proto nf_conntrack_l3proto_ipv4;
144 /* Returns conntrack if it dealt with ICMP, and filled in skb fields */
145 static int
146 icmp_error_message(struct sk_buff *skb,
147                  enum ip_conntrack_info *ctinfo,
148                  unsigned int hooknum)
149 {
150         struct nf_conntrack_tuple innertuple, origtuple;
151         struct {
152                 struct icmphdr icmp;
153                 struct iphdr ip;
154         } _in, *inside;
155         struct nf_conntrack_protocol *innerproto;
156         struct nf_conntrack_tuple_hash *h;
157         int dataoff;
158
159         NF_CT_ASSERT(skb->nfct == NULL);
160
161         /* Not enough header? */
162         inside = skb_header_pointer(skb, skb->nh.iph->ihl*4, sizeof(_in), &_in);
163         if (inside == NULL)
164                 return -NF_ACCEPT;
165
166         /* Ignore ICMP's containing fragments (shouldn't happen) */
167         if (inside->ip.frag_off & htons(IP_OFFSET)) {
168                 DEBUGP("icmp_error_message: fragment of proto %u\n",
169                        inside->ip.protocol);
170                 return -NF_ACCEPT;
171         }
172
173         innerproto = __nf_ct_proto_find(PF_INET, inside->ip.protocol);
174         dataoff = skb->nh.iph->ihl*4 + sizeof(inside->icmp);
175         /* Are they talking about one of our connections? */
176         if (!nf_ct_get_tuple(skb, dataoff, dataoff + inside->ip.ihl*4, PF_INET,
177                              inside->ip.protocol, &origtuple,
178                              &nf_conntrack_l3proto_ipv4, innerproto)) {
179                 DEBUGP("icmp_error_message: ! get_tuple p=%u",
180                        inside->ip.protocol);
181                 return -NF_ACCEPT;
182         }
183
184         /* Ordinarily, we'd expect the inverted tupleproto, but it's
185            been preserved inside the ICMP. */
186         if (!nf_ct_invert_tuple(&innertuple, &origtuple,
187                                 &nf_conntrack_l3proto_ipv4, innerproto)) {
188                 DEBUGP("icmp_error_message: no match\n");
189                 return -NF_ACCEPT;
190         }
191
192         *ctinfo = IP_CT_RELATED;
193
194         h = nf_conntrack_find_get(&innertuple, NULL);
195         if (!h) {
196                 /* Locally generated ICMPs will match inverted if they
197                    haven't been SNAT'ed yet */
198                 /* FIXME: NAT code has to handle half-done double NAT --RR */
199                 if (hooknum == NF_IP_LOCAL_OUT)
200                         h = nf_conntrack_find_get(&origtuple, NULL);
201
202                 if (!h) {
203                         DEBUGP("icmp_error_message: no match\n");
204                         return -NF_ACCEPT;
205                 }
206
207                 /* Reverse direction from that found */
208                 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
209                         *ctinfo += IP_CT_IS_REPLY;
210         } else {
211                 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
212                         *ctinfo += IP_CT_IS_REPLY;
213         }
214
215         /* Update skb to refer to this connection */
216         skb->nfct = &nf_ct_tuplehash_to_ctrack(h)->ct_general;
217         skb->nfctinfo = *ctinfo;
218         return -NF_ACCEPT;
219 }
220
221 /* Small and modified version of icmp_rcv */
222 static int
223 icmp_error(struct sk_buff *skb, unsigned int dataoff,
224            enum ip_conntrack_info *ctinfo, int pf, unsigned int hooknum)
225 {
226         struct icmphdr _ih, *icmph;
227
228         /* Not enough header? */
229         icmph = skb_header_pointer(skb, skb->nh.iph->ihl*4, sizeof(_ih), &_ih);
230         if (icmph == NULL) {
231                 if (LOG_INVALID(IPPROTO_ICMP))
232                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
233                                       "nf_ct_icmp: short packet ");
234                 return -NF_ACCEPT;
235         }
236
237         /* See ip_conntrack_proto_tcp.c */
238         if (nf_conntrack_checksum && hooknum == NF_IP_PRE_ROUTING &&
239             nf_ip_checksum(skb, hooknum, dataoff, 0)) {
240                 if (LOG_INVALID(IPPROTO_ICMP))
241                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
242                                       "nf_ct_icmp: bad HW ICMP checksum ");
243                 return -NF_ACCEPT;
244         }
245
246         /*
247          *      18 is the highest 'known' ICMP type. Anything else is a mystery
248          *
249          *      RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
250          *                discarded.
251          */
252         if (icmph->type > NR_ICMP_TYPES) {
253                 if (LOG_INVALID(IPPROTO_ICMP))
254                         nf_log_packet(PF_INET, 0, skb, NULL, NULL, NULL,
255                                       "nf_ct_icmp: invalid ICMP type ");
256                 return -NF_ACCEPT;
257         }
258
259         /* Need to track icmp error message? */
260         if (icmph->type != ICMP_DEST_UNREACH
261             && icmph->type != ICMP_SOURCE_QUENCH
262             && icmph->type != ICMP_TIME_EXCEEDED
263             && icmph->type != ICMP_PARAMETERPROB
264             && icmph->type != ICMP_REDIRECT)
265                 return NF_ACCEPT;
266
267         return icmp_error_message(skb, ctinfo, hooknum);
268 }
269
270 #if defined(CONFIG_NF_CT_NETLINK) || \
271     defined(CONFIG_NF_CT_NETLINK_MODULE)
272
273 #include <linux/netfilter/nfnetlink.h>
274 #include <linux/netfilter/nfnetlink_conntrack.h>
275
276 static int icmp_tuple_to_nfattr(struct sk_buff *skb,
277                                 const struct nf_conntrack_tuple *t)
278 {
279         NFA_PUT(skb, CTA_PROTO_ICMP_ID, sizeof(u_int16_t),
280                 &t->src.u.icmp.id);
281         NFA_PUT(skb, CTA_PROTO_ICMP_TYPE, sizeof(u_int8_t),
282                 &t->dst.u.icmp.type);
283         NFA_PUT(skb, CTA_PROTO_ICMP_CODE, sizeof(u_int8_t),
284                 &t->dst.u.icmp.code);
285
286         return 0;
287
288 nfattr_failure:
289         return -1;
290 }
291
292 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
293         [CTA_PROTO_ICMP_TYPE-1] = sizeof(u_int8_t),
294         [CTA_PROTO_ICMP_CODE-1] = sizeof(u_int8_t),
295         [CTA_PROTO_ICMP_ID-1]   = sizeof(u_int16_t)
296 };
297
298 static int icmp_nfattr_to_tuple(struct nfattr *tb[],
299                                 struct nf_conntrack_tuple *tuple)
300 {
301         if (!tb[CTA_PROTO_ICMP_TYPE-1]
302             || !tb[CTA_PROTO_ICMP_CODE-1]
303             || !tb[CTA_PROTO_ICMP_ID-1])
304                 return -EINVAL;
305
306         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
307                 return -EINVAL;
308
309         tuple->dst.u.icmp.type = 
310                         *(u_int8_t *)NFA_DATA(tb[CTA_PROTO_ICMP_TYPE-1]);
311         tuple->dst.u.icmp.code =
312                         *(u_int8_t *)NFA_DATA(tb[CTA_PROTO_ICMP_CODE-1]);
313         tuple->src.u.icmp.id =
314                         *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_ICMP_ID-1]);
315
316         if (tuple->dst.u.icmp.type >= sizeof(invmap)
317             || !invmap[tuple->dst.u.icmp.type])
318                 return -EINVAL;
319
320         return 0;
321 }
322 #endif
323
324 struct nf_conntrack_protocol nf_conntrack_protocol_icmp =
325 {
326         .list                   = { NULL, NULL },
327         .l3proto                = PF_INET,
328         .proto                  = IPPROTO_ICMP,
329         .name                   = "icmp",
330         .pkt_to_tuple           = icmp_pkt_to_tuple,
331         .invert_tuple           = icmp_invert_tuple,
332         .print_tuple            = icmp_print_tuple,
333         .print_conntrack        = icmp_print_conntrack,
334         .packet                 = icmp_packet,
335         .new                    = icmp_new,
336         .error                  = icmp_error,
337         .destroy                = NULL,
338         .me                     = NULL,
339 #if defined(CONFIG_NF_CT_NETLINK) || \
340     defined(CONFIG_NF_CT_NETLINK_MODULE)
341         .tuple_to_nfattr        = icmp_tuple_to_nfattr,
342         .nfattr_to_tuple        = icmp_nfattr_to_tuple,
343 #endif
344 };
345
346 EXPORT_SYMBOL(nf_conntrack_protocol_icmp);