netfilter: xt_connlimit: connlimit-above early loop termination
[linux-2.6.git] / net / netfilter / xt_connlimit.c
1 /*
2  * netfilter module to limit the number of parallel tcp
3  * connections per IP address.
4  *   (c) 2000 Gerd Knorr <kraxel@bytesex.org>
5  *   Nov 2002: Martin Bene <martin.bene@icomedias.com>:
6  *              only ignore TIME_WAIT or gone connections
7  *   (C) CC Computer Consultants GmbH, 2007
8  *
9  * based on ...
10  *
11  * Kernel module to match connection tracking information.
12  * GPL (C) 1999  Rusty Russell (rusty@rustcorp.com.au).
13  */
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/in.h>
16 #include <linux/in6.h>
17 #include <linux/ip.h>
18 #include <linux/ipv6.h>
19 #include <linux/jhash.h>
20 #include <linux/slab.h>
21 #include <linux/list.h>
22 #include <linux/module.h>
23 #include <linux/random.h>
24 #include <linux/skbuff.h>
25 #include <linux/spinlock.h>
26 #include <linux/netfilter/nf_conntrack_tcp.h>
27 #include <linux/netfilter/x_tables.h>
28 #include <linux/netfilter/xt_connlimit.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/netfilter/nf_conntrack_core.h>
31 #include <net/netfilter/nf_conntrack_tuple.h>
32 #include <net/netfilter/nf_conntrack_zones.h>
33
34 /* we will save the tuples of all connections we care about */
35 struct xt_connlimit_conn {
36         struct list_head list;
37         struct nf_conntrack_tuple tuple;
38 };
39
40 struct xt_connlimit_data {
41         struct list_head iphash[256];
42         spinlock_t lock;
43 };
44
45 static u_int32_t connlimit_rnd __read_mostly;
46 static bool connlimit_rnd_inited __read_mostly;
47
48 static inline unsigned int connlimit_iphash(__be32 addr)
49 {
50         return jhash_1word((__force __u32)addr, connlimit_rnd) & 0xFF;
51 }
52
53 static inline unsigned int
54 connlimit_iphash6(const union nf_inet_addr *addr,
55                   const union nf_inet_addr *mask)
56 {
57         union nf_inet_addr res;
58         unsigned int i;
59
60         for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i)
61                 res.ip6[i] = addr->ip6[i] & mask->ip6[i];
62
63         return jhash2((u32 *)res.ip6, ARRAY_SIZE(res.ip6), connlimit_rnd) & 0xFF;
64 }
65
66 static inline bool already_closed(const struct nf_conn *conn)
67 {
68         if (nf_ct_protonum(conn) == IPPROTO_TCP)
69                 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
70                        conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
71         else
72                 return 0;
73 }
74
75 static inline unsigned int
76 same_source_net(const union nf_inet_addr *addr,
77                 const union nf_inet_addr *mask,
78                 const union nf_inet_addr *u3, u_int8_t family)
79 {
80         if (family == NFPROTO_IPV4) {
81                 return (addr->ip & mask->ip) == (u3->ip & mask->ip);
82         } else {
83                 union nf_inet_addr lh, rh;
84                 unsigned int i;
85
86                 for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i) {
87                         lh.ip6[i] = addr->ip6[i] & mask->ip6[i];
88                         rh.ip6[i] = u3->ip6[i] & mask->ip6[i];
89                 }
90
91                 return memcmp(&lh.ip6, &rh.ip6, sizeof(lh.ip6)) == 0;
92         }
93 }
94
95 static int count_them(struct net *net,
96                       struct xt_connlimit_data *data,
97                       const struct nf_conntrack_tuple *tuple,
98                       const union nf_inet_addr *addr,
99                       const union nf_inet_addr *mask,
100                       u_int8_t family,
101                       unsigned int threshold)
102 {
103         const struct nf_conntrack_tuple_hash *found;
104         struct xt_connlimit_conn *conn;
105         struct xt_connlimit_conn *tmp;
106         struct nf_conn *found_ct;
107         struct list_head *hash;
108         bool addit = true;
109         int matches = 0;
110
111         if (family == NFPROTO_IPV6)
112                 hash = &data->iphash[connlimit_iphash6(addr, mask)];
113         else
114                 hash = &data->iphash[connlimit_iphash(addr->ip & mask->ip)];
115
116         rcu_read_lock();
117
118         /* check the saved connections */
119         list_for_each_entry_safe(conn, tmp, hash, list) {
120                 found    = nf_conntrack_find_get(net, NF_CT_DEFAULT_ZONE,
121                                                  &conn->tuple);
122                 found_ct = NULL;
123
124                 if (found != NULL)
125                         found_ct = nf_ct_tuplehash_to_ctrack(found);
126
127                 if (found_ct != NULL &&
128                     nf_ct_tuple_equal(&conn->tuple, tuple) &&
129                     !already_closed(found_ct))
130                         /*
131                          * Just to be sure we have it only once in the list.
132                          * We should not see tuples twice unless someone hooks
133                          * this into a table without "-p tcp --syn".
134                          */
135                         addit = false;
136
137                 if (found == NULL) {
138                         /* this one is gone */
139                         list_del(&conn->list);
140                         kfree(conn);
141                         continue;
142                 }
143
144                 if (already_closed(found_ct)) {
145                         /*
146                          * we do not care about connections which are
147                          * closed already -> ditch it
148                          */
149                         nf_ct_put(found_ct);
150                         list_del(&conn->list);
151                         kfree(conn);
152                         continue;
153                 }
154
155                 if (same_source_net(addr, mask, &conn->tuple.src.u3, family)) {
156                         /* same source network -> be counted! */
157                         ++matches;
158                         if (matches > threshold) {
159                                 nf_ct_put(found_ct);
160                                 break;
161                         }
162                 }
163                 nf_ct_put(found_ct);
164         }
165
166         rcu_read_unlock();
167
168         if (addit) {
169                 /* save the new connection in our list */
170                 conn = kzalloc(sizeof(*conn), GFP_ATOMIC);
171                 if (conn == NULL)
172                         return -ENOMEM;
173                 conn->tuple = *tuple;
174                 list_add(&conn->list, hash);
175                 ++matches;
176         }
177
178         return matches;
179 }
180
181 static bool
182 connlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
183 {
184         struct net *net = dev_net(par->in ? par->in : par->out);
185         const struct xt_connlimit_info *info = par->matchinfo;
186         union nf_inet_addr addr;
187         struct nf_conntrack_tuple tuple;
188         const struct nf_conntrack_tuple *tuple_ptr = &tuple;
189         enum ip_conntrack_info ctinfo;
190         const struct nf_conn *ct;
191         int connections;
192
193         ct = nf_ct_get(skb, &ctinfo);
194         if (ct != NULL) {
195                 if (info->flags & XT_CONNLIMIT_DADDR)
196                         tuple_ptr = &ct->tuplehash[IP_CT_DIR_REPLY].tuple;
197                 else
198                         tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
199         } else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
200                                     par->family, &tuple)) {
201                 goto hotdrop;
202         }
203
204         if (par->family == NFPROTO_IPV6) {
205                 const struct ipv6hdr *iph = ipv6_hdr(skb);
206                 memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
207                        &iph->daddr : &iph->saddr, sizeof(addr.ip6));
208         } else {
209                 const struct iphdr *iph = ip_hdr(skb);
210                 addr.ip = (info->flags & XT_CONNLIMIT_DADDR) ?
211                           iph->daddr : iph->saddr;
212         }
213
214         spin_lock_bh(&info->data->lock);
215         connections = count_them(net, info->data, tuple_ptr, &addr,
216                                  &info->mask, par->family,
217                                  info->limit);
218         spin_unlock_bh(&info->data->lock);
219
220         if (connections < 0)
221                 /* kmalloc failed, drop it entirely */
222                 goto hotdrop;
223
224         return (connections > info->limit) ^
225                !!(info->flags & XT_CONNLIMIT_INVERT);
226
227  hotdrop:
228         par->hotdrop = true;
229         return false;
230 }
231
232 static int connlimit_mt_check(const struct xt_mtchk_param *par)
233 {
234         struct xt_connlimit_info *info = par->matchinfo;
235         unsigned int i;
236         int ret;
237
238         if (unlikely(!connlimit_rnd_inited)) {
239                 get_random_bytes(&connlimit_rnd, sizeof(connlimit_rnd));
240                 connlimit_rnd_inited = true;
241         }
242         ret = nf_ct_l3proto_try_module_get(par->family);
243         if (ret < 0) {
244                 pr_info("cannot load conntrack support for "
245                         "address family %u\n", par->family);
246                 return ret;
247         }
248
249         /* init private data */
250         info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
251         if (info->data == NULL) {
252                 nf_ct_l3proto_module_put(par->family);
253                 return -ENOMEM;
254         }
255
256         spin_lock_init(&info->data->lock);
257         for (i = 0; i < ARRAY_SIZE(info->data->iphash); ++i)
258                 INIT_LIST_HEAD(&info->data->iphash[i]);
259
260         return 0;
261 }
262
263 static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
264 {
265         const struct xt_connlimit_info *info = par->matchinfo;
266         struct xt_connlimit_conn *conn;
267         struct xt_connlimit_conn *tmp;
268         struct list_head *hash = info->data->iphash;
269         unsigned int i;
270
271         nf_ct_l3proto_module_put(par->family);
272
273         for (i = 0; i < ARRAY_SIZE(info->data->iphash); ++i) {
274                 list_for_each_entry_safe(conn, tmp, &hash[i], list) {
275                         list_del(&conn->list);
276                         kfree(conn);
277                 }
278         }
279
280         kfree(info->data);
281 }
282
283 static struct xt_match connlimit_mt_reg[] __read_mostly = {
284         {
285                 .name       = "connlimit",
286                 .revision   = 0,
287                 .family     = NFPROTO_UNSPEC,
288                 .checkentry = connlimit_mt_check,
289                 .match      = connlimit_mt,
290                 .matchsize  = sizeof(struct xt_connlimit_info),
291                 .destroy    = connlimit_mt_destroy,
292                 .me         = THIS_MODULE,
293         },
294         {
295                 .name       = "connlimit",
296                 .revision   = 1,
297                 .family     = NFPROTO_UNSPEC,
298                 .checkentry = connlimit_mt_check,
299                 .match      = connlimit_mt,
300                 .matchsize  = sizeof(struct xt_connlimit_info),
301                 .destroy    = connlimit_mt_destroy,
302                 .me         = THIS_MODULE,
303         },
304 };
305
306 static int __init connlimit_mt_init(void)
307 {
308         return xt_register_matches(connlimit_mt_reg,
309                ARRAY_SIZE(connlimit_mt_reg));
310 }
311
312 static void __exit connlimit_mt_exit(void)
313 {
314         xt_unregister_matches(connlimit_mt_reg, ARRAY_SIZE(connlimit_mt_reg));
315 }
316
317 module_init(connlimit_mt_init);
318 module_exit(connlimit_mt_exit);
319 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
320 MODULE_DESCRIPTION("Xtables: Number of connections matching");
321 MODULE_LICENSE("GPL");
322 MODULE_ALIAS("ipt_connlimit");
323 MODULE_ALIAS("ip6t_connlimit");