[NETFILTER]: nf_conntrack: properly use RCU API for nf_ct_protos/nf_ct_l3protos arrays
[linux-2.6.git] / net / ipv6 / netfilter / nf_conntrack_proto_icmpv6.c
1 /*
2  * Copyright (C)2003,2004 USAGI/WIDE Project
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  * Author:
9  *      Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
10  *
11  * 16 Dec 2003: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
12  *      - ICMPv6 tracking support. Derived from the original ip_conntrack code
13  *        net/ipv4/netfilter/ip_conntrack_proto_icmp.c which had the following
14  *        copyright information:
15  *              (C) 1999-2001 Paul `Rusty' Russell
16  *              (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
17  */
18
19 #include <linux/types.h>
20 #include <linux/sched.h>
21 #include <linux/timer.h>
22 #include <linux/module.h>
23 #include <linux/netfilter.h>
24 #include <linux/in6.h>
25 #include <linux/icmpv6.h>
26 #include <linux/ipv6.h>
27 #include <net/ipv6.h>
28 #include <net/ip6_checksum.h>
29 #include <linux/seq_file.h>
30 #include <linux/netfilter_ipv6.h>
31 #include <net/netfilter/nf_conntrack_tuple.h>
32 #include <net/netfilter/nf_conntrack_l4proto.h>
33 #include <net/netfilter/nf_conntrack_core.h>
34 #include <net/netfilter/ipv6/nf_conntrack_icmpv6.h>
35
36 static unsigned long nf_ct_icmpv6_timeout __read_mostly = 30*HZ;
37
38 #if 0
39 #define DEBUGP printk
40 #else
41 #define DEBUGP(format, args...)
42 #endif
43
44 static int icmpv6_pkt_to_tuple(const struct sk_buff *skb,
45                                unsigned int dataoff,
46                                struct nf_conntrack_tuple *tuple)
47 {
48         struct icmp6hdr _hdr, *hp;
49
50         hp = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
51         if (hp == NULL)
52                 return 0;
53         tuple->dst.u.icmp.type = hp->icmp6_type;
54         tuple->src.u.icmp.id = hp->icmp6_identifier;
55         tuple->dst.u.icmp.code = hp->icmp6_code;
56
57         return 1;
58 }
59
60 /* Add 1; spaces filled with 0. */
61 static u_int8_t invmap[] = {
62         [ICMPV6_ECHO_REQUEST - 128]     = ICMPV6_ECHO_REPLY + 1,
63         [ICMPV6_ECHO_REPLY - 128]       = ICMPV6_ECHO_REQUEST + 1,
64         [ICMPV6_NI_QUERY - 128]         = ICMPV6_NI_QUERY + 1,
65         [ICMPV6_NI_REPLY - 128]         = ICMPV6_NI_REPLY +1
66 };
67
68 static int icmpv6_invert_tuple(struct nf_conntrack_tuple *tuple,
69                                const struct nf_conntrack_tuple *orig)
70 {
71         int type = orig->dst.u.icmp.type - 128;
72         if (type < 0 || type >= sizeof(invmap) || !invmap[type])
73                 return 0;
74
75         tuple->src.u.icmp.id   = orig->src.u.icmp.id;
76         tuple->dst.u.icmp.type = invmap[type] - 1;
77         tuple->dst.u.icmp.code = orig->dst.u.icmp.code;
78         return 1;
79 }
80
81 /* Print out the per-protocol part of the tuple. */
82 static int icmpv6_print_tuple(struct seq_file *s,
83                               const struct nf_conntrack_tuple *tuple)
84 {
85         return seq_printf(s, "type=%u code=%u id=%u ",
86                           tuple->dst.u.icmp.type,
87                           tuple->dst.u.icmp.code,
88                           ntohs(tuple->src.u.icmp.id));
89 }
90
91 /* Print out the private part of the conntrack. */
92 static int icmpv6_print_conntrack(struct seq_file *s,
93                                   const struct nf_conn *conntrack)
94 {
95         return 0;
96 }
97
98 /* Returns verdict for packet, or -1 for invalid. */
99 static int icmpv6_packet(struct nf_conn *ct,
100                        const struct sk_buff *skb,
101                        unsigned int dataoff,
102                        enum ip_conntrack_info ctinfo,
103                        int pf,
104                        unsigned int hooknum)
105 {
106         /* Try to delete connection immediately after all replies:
107            won't actually vanish as we still have skb, and del_timer
108            means this will only run once even if count hits zero twice
109            (theoretically possible with SMP) */
110         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) {
111                 if (atomic_dec_and_test(&ct->proto.icmp.count)
112                     && del_timer(&ct->timeout))
113                         ct->timeout.function((unsigned long)ct);
114         } else {
115                 atomic_inc(&ct->proto.icmp.count);
116                 nf_conntrack_event_cache(IPCT_PROTOINFO_VOLATILE, skb);
117                 nf_ct_refresh_acct(ct, ctinfo, skb, nf_ct_icmpv6_timeout);
118         }
119
120         return NF_ACCEPT;
121 }
122
123 /* Called when a new connection for this protocol found. */
124 static int icmpv6_new(struct nf_conn *conntrack,
125                       const struct sk_buff *skb,
126                       unsigned int dataoff)
127 {
128         static u_int8_t valid_new[] = {
129                 [ICMPV6_ECHO_REQUEST - 128] = 1,
130                 [ICMPV6_NI_QUERY - 128] = 1
131         };
132         int type = conntrack->tuplehash[0].tuple.dst.u.icmp.type - 128;
133
134         if (type < 0 || type >= sizeof(valid_new) || !valid_new[type]) {
135                 /* Can't create a new ICMPv6 `conn' with this. */
136                 DEBUGP("icmpv6: can't create new conn with type %u\n",
137                        type + 128);
138                 NF_CT_DUMP_TUPLE(&conntrack->tuplehash[0].tuple);
139                 return 0;
140         }
141         atomic_set(&conntrack->proto.icmp.count, 0);
142         return 1;
143 }
144
145 static int
146 icmpv6_error_message(struct sk_buff *skb,
147                      unsigned int icmp6off,
148                      enum ip_conntrack_info *ctinfo,
149                      unsigned int hooknum)
150 {
151         struct nf_conntrack_tuple intuple, origtuple;
152         struct nf_conntrack_tuple_hash *h;
153         struct icmp6hdr _hdr, *hp;
154         unsigned int inip6off;
155         struct nf_conntrack_l4proto *inproto;
156         u_int8_t inprotonum;
157         unsigned int inprotoff;
158
159         NF_CT_ASSERT(skb->nfct == NULL);
160
161         hp = skb_header_pointer(skb, icmp6off, sizeof(_hdr), &_hdr);
162         if (hp == NULL) {
163                 DEBUGP("icmpv6_error: Can't get ICMPv6 hdr.\n");
164                 return -NF_ACCEPT;
165         }
166
167         inip6off = icmp6off + sizeof(_hdr);
168         if (skb_copy_bits(skb, inip6off+offsetof(struct ipv6hdr, nexthdr),
169                           &inprotonum, sizeof(inprotonum)) != 0) {
170                 DEBUGP("icmpv6_error: Can't get nexthdr in inner IPv6 header.\n");
171                 return -NF_ACCEPT;
172         }
173         inprotoff = nf_ct_ipv6_skip_exthdr(skb,
174                                            inip6off + sizeof(struct ipv6hdr),
175                                            &inprotonum,
176                                            skb->len - inip6off
177                                                     - sizeof(struct ipv6hdr));
178
179         if ((inprotoff < 0) || (inprotoff > skb->len) ||
180             (inprotonum == NEXTHDR_FRAGMENT)) {
181                 DEBUGP("icmpv6_error: Can't get protocol header in ICMPv6 payload.\n");
182                 return -NF_ACCEPT;
183         }
184
185         /* rcu_read_lock()ed by nf_hook_slow */
186         inproto = __nf_ct_l4proto_find(PF_INET6, inprotonum);
187
188         /* Are they talking about one of our connections? */
189         if (!nf_ct_get_tuple(skb, inip6off, inprotoff, PF_INET6, inprotonum,
190                              &origtuple, &nf_conntrack_l3proto_ipv6, inproto)) {
191                 DEBUGP("icmpv6_error: Can't get tuple\n");
192                 return -NF_ACCEPT;
193         }
194
195         /* Ordinarily, we'd expect the inverted tupleproto, but it's
196            been preserved inside the ICMP. */
197         if (!nf_ct_invert_tuple(&intuple, &origtuple,
198                                 &nf_conntrack_l3proto_ipv6, inproto)) {
199                 DEBUGP("icmpv6_error: Can't invert tuple\n");
200                 return -NF_ACCEPT;
201         }
202
203         *ctinfo = IP_CT_RELATED;
204
205         h = nf_conntrack_find_get(&intuple, NULL);
206         if (!h) {
207                 DEBUGP("icmpv6_error: no match\n");
208                 return -NF_ACCEPT;
209         } else {
210                 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
211                         *ctinfo += IP_CT_IS_REPLY;
212         }
213
214         /* Update skb to refer to this connection */
215         skb->nfct = &nf_ct_tuplehash_to_ctrack(h)->ct_general;
216         skb->nfctinfo = *ctinfo;
217         return -NF_ACCEPT;
218 }
219
220 static int
221 icmpv6_error(struct sk_buff *skb, unsigned int dataoff,
222              enum ip_conntrack_info *ctinfo, int pf, unsigned int hooknum)
223 {
224         struct icmp6hdr _ih, *icmp6h;
225
226         icmp6h = skb_header_pointer(skb, dataoff, sizeof(_ih), &_ih);
227         if (icmp6h == NULL) {
228                 if (LOG_INVALID(IPPROTO_ICMPV6))
229                 nf_log_packet(PF_INET6, 0, skb, NULL, NULL, NULL,
230                               "nf_ct_icmpv6: short packet ");
231                 return -NF_ACCEPT;
232         }
233
234         if (nf_conntrack_checksum && hooknum == NF_IP6_PRE_ROUTING &&
235             nf_ip6_checksum(skb, hooknum, dataoff, IPPROTO_ICMPV6)) {
236                 nf_log_packet(PF_INET6, 0, skb, NULL, NULL, NULL,
237                               "nf_ct_icmpv6: ICMPv6 checksum failed\n");
238                 return -NF_ACCEPT;
239         }
240
241         /* is not error message ? */
242         if (icmp6h->icmp6_type >= 128)
243                 return NF_ACCEPT;
244
245         return icmpv6_error_message(skb, dataoff, ctinfo, hooknum);
246 }
247
248 #if defined(CONFIG_NF_CT_NETLINK) || \
249     defined(CONFIG_NF_CT_NETLINK_MODULE)
250
251 #include <linux/netfilter/nfnetlink.h>
252 #include <linux/netfilter/nfnetlink_conntrack.h>
253 static int icmpv6_tuple_to_nfattr(struct sk_buff *skb,
254                                   const struct nf_conntrack_tuple *t)
255 {
256         NFA_PUT(skb, CTA_PROTO_ICMPV6_ID, sizeof(u_int16_t),
257                 &t->src.u.icmp.id);
258         NFA_PUT(skb, CTA_PROTO_ICMPV6_TYPE, sizeof(u_int8_t),
259                 &t->dst.u.icmp.type);
260         NFA_PUT(skb, CTA_PROTO_ICMPV6_CODE, sizeof(u_int8_t),
261                 &t->dst.u.icmp.code);
262
263         return 0;
264
265 nfattr_failure:
266         return -1;
267 }
268
269 static const size_t cta_min_proto[CTA_PROTO_MAX] = {
270         [CTA_PROTO_ICMPV6_TYPE-1] = sizeof(u_int8_t),
271         [CTA_PROTO_ICMPV6_CODE-1] = sizeof(u_int8_t),
272         [CTA_PROTO_ICMPV6_ID-1]   = sizeof(u_int16_t)
273 };
274
275 static int icmpv6_nfattr_to_tuple(struct nfattr *tb[],
276                                 struct nf_conntrack_tuple *tuple)
277 {
278         if (!tb[CTA_PROTO_ICMPV6_TYPE-1]
279             || !tb[CTA_PROTO_ICMPV6_CODE-1]
280             || !tb[CTA_PROTO_ICMPV6_ID-1])
281                 return -EINVAL;
282
283         if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
284                 return -EINVAL;
285
286         tuple->dst.u.icmp.type =
287                         *(u_int8_t *)NFA_DATA(tb[CTA_PROTO_ICMPV6_TYPE-1]);
288         tuple->dst.u.icmp.code =
289                         *(u_int8_t *)NFA_DATA(tb[CTA_PROTO_ICMPV6_CODE-1]);
290         tuple->src.u.icmp.id =
291                         *(__be16 *)NFA_DATA(tb[CTA_PROTO_ICMPV6_ID-1]);
292
293         if (tuple->dst.u.icmp.type < 128
294             || tuple->dst.u.icmp.type - 128 >= sizeof(invmap)
295             || !invmap[tuple->dst.u.icmp.type - 128])
296                 return -EINVAL;
297
298         return 0;
299 }
300 #endif
301
302 #ifdef CONFIG_SYSCTL
303 static struct ctl_table_header *icmpv6_sysctl_header;
304 static struct ctl_table icmpv6_sysctl_table[] = {
305         {
306                 .ctl_name       = NET_NF_CONNTRACK_ICMPV6_TIMEOUT,
307                 .procname       = "nf_conntrack_icmpv6_timeout",
308                 .data           = &nf_ct_icmpv6_timeout,
309                 .maxlen         = sizeof(unsigned int),
310                 .mode           = 0644,
311                 .proc_handler   = &proc_dointvec_jiffies,
312         },
313         {
314                 .ctl_name       = 0
315         }
316 };
317 #endif /* CONFIG_SYSCTL */
318
319 struct nf_conntrack_l4proto nf_conntrack_l4proto_icmpv6 =
320 {
321         .l3proto                = PF_INET6,
322         .l4proto                = IPPROTO_ICMPV6,
323         .name                   = "icmpv6",
324         .pkt_to_tuple           = icmpv6_pkt_to_tuple,
325         .invert_tuple           = icmpv6_invert_tuple,
326         .print_tuple            = icmpv6_print_tuple,
327         .print_conntrack        = icmpv6_print_conntrack,
328         .packet                 = icmpv6_packet,
329         .new                    = icmpv6_new,
330         .error                  = icmpv6_error,
331 #if defined(CONFIG_NF_CT_NETLINK) || \
332     defined(CONFIG_NF_CT_NETLINK_MODULE)
333         .tuple_to_nfattr        = icmpv6_tuple_to_nfattr,
334         .nfattr_to_tuple        = icmpv6_nfattr_to_tuple,
335 #endif
336 #ifdef CONFIG_SYSCTL
337         .ctl_table_header       = &icmpv6_sysctl_header,
338         .ctl_table              = icmpv6_sysctl_table,
339 #endif
340 };
341
342 EXPORT_SYMBOL(nf_conntrack_l4proto_icmpv6);