sched: simplify __check_preempt_curr_fair()
[linux-2.6] / 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  *   Copyright © Jan Engelhardt <jengelh@gmx.de>, 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 #include <linux/in.h>
15 #include <linux/in6.h>
16 #include <linux/ip.h>
17 #include <linux/ipv6.h>
18 #include <linux/jhash.h>
19 #include <linux/list.h>
20 #include <linux/module.h>
21 #include <linux/random.h>
22 #include <linux/skbuff.h>
23 #include <linux/spinlock.h>
24 #include <linux/netfilter/nf_conntrack_tcp.h>
25 #include <linux/netfilter/x_tables.h>
26 #include <linux/netfilter/xt_connlimit.h>
27 #include <net/netfilter/nf_conntrack.h>
28 #include <net/netfilter/nf_conntrack_core.h>
29 #include <net/netfilter/nf_conntrack_tuple.h>
30
31 /* we will save the tuples of all connections we care about */
32 struct xt_connlimit_conn {
33         struct list_head list;
34         struct nf_conntrack_tuple tuple;
35 };
36
37 struct xt_connlimit_data {
38         struct list_head iphash[256];
39         spinlock_t lock;
40 };
41
42 static u_int32_t connlimit_rnd;
43 static bool connlimit_rnd_inited;
44
45 static inline unsigned int connlimit_iphash(__be32 addr)
46 {
47         if (unlikely(!connlimit_rnd_inited)) {
48                 get_random_bytes(&connlimit_rnd, sizeof(connlimit_rnd));
49                 connlimit_rnd_inited = true;
50         }
51         return jhash_1word((__force __u32)addr, connlimit_rnd) & 0xFF;
52 }
53
54 static inline unsigned int
55 connlimit_iphash6(const union nf_conntrack_address *addr,
56                   const union nf_conntrack_address *mask)
57 {
58         union nf_conntrack_address res;
59         unsigned int i;
60
61         if (unlikely(!connlimit_rnd_inited)) {
62                 get_random_bytes(&connlimit_rnd, sizeof(connlimit_rnd));
63                 connlimit_rnd_inited = true;
64         }
65
66         for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i)
67                 res.ip6[i] = addr->ip6[i] & mask->ip6[i];
68
69         return jhash2((u32 *)res.ip6, ARRAY_SIZE(res.ip6), connlimit_rnd) & 0xFF;
70 }
71
72 static inline bool already_closed(const struct nf_conn *conn)
73 {
74         u_int16_t proto = conn->tuplehash[0].tuple.dst.protonum;
75
76         if (proto == IPPROTO_TCP)
77                 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT;
78         else
79                 return 0;
80 }
81
82 static inline unsigned int
83 same_source_net(const union nf_conntrack_address *addr,
84                 const union nf_conntrack_address *mask,
85                 const union nf_conntrack_address *u3, unsigned int family)
86 {
87         if (family == AF_INET) {
88                 return (addr->ip & mask->ip) == (u3->ip & mask->ip);
89         } else {
90                 union nf_conntrack_address lh, rh;
91                 unsigned int i;
92
93                 for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i) {
94                         lh.ip6[i] = addr->ip6[i] & mask->ip6[i];
95                         rh.ip6[i] = u3->ip6[i] & mask->ip6[i];
96                 }
97
98                 return memcmp(&lh.ip6, &rh.ip6, sizeof(lh.ip6)) == 0;
99         }
100 }
101
102 static int count_them(struct xt_connlimit_data *data,
103                       const struct nf_conntrack_tuple *tuple,
104                       const union nf_conntrack_address *addr,
105                       const union nf_conntrack_address *mask,
106                       const struct xt_match *match)
107 {
108         struct nf_conntrack_tuple_hash *found;
109         struct xt_connlimit_conn *conn;
110         struct xt_connlimit_conn *tmp;
111         struct nf_conn *found_ct;
112         struct list_head *hash;
113         bool addit = true;
114         int matches = 0;
115
116
117         if (match->family == AF_INET6)
118                 hash = &data->iphash[connlimit_iphash6(addr, mask)];
119         else
120                 hash = &data->iphash[connlimit_iphash(addr->ip & mask->ip)];
121
122         read_lock_bh(&nf_conntrack_lock);
123
124         /* check the saved connections */
125         list_for_each_entry_safe(conn, tmp, hash, list) {
126                 found    = __nf_conntrack_find(&conn->tuple, NULL);
127                 found_ct = NULL;
128
129                 if (found != NULL)
130                         found_ct = nf_ct_tuplehash_to_ctrack(found);
131
132                 if (found_ct != NULL &&
133                     nf_ct_tuple_equal(&conn->tuple, tuple) &&
134                     !already_closed(found_ct))
135                         /*
136                          * Just to be sure we have it only once in the list.
137                          * We should not see tuples twice unless someone hooks
138                          * this into a table without "-p tcp --syn".
139                          */
140                         addit = false;
141
142                 if (found == NULL) {
143                         /* this one is gone */
144                         list_del(&conn->list);
145                         kfree(conn);
146                         continue;
147                 }
148
149                 if (already_closed(found_ct)) {
150                         /*
151                          * we do not care about connections which are
152                          * closed already -> ditch it
153                          */
154                         list_del(&conn->list);
155                         kfree(conn);
156                         continue;
157                 }
158
159                 if (same_source_net(addr, mask, &conn->tuple.src.u3,
160                     match->family))
161                         /* same source network -> be counted! */
162                         ++matches;
163         }
164
165         read_unlock_bh(&nf_conntrack_lock);
166
167         if (addit) {
168                 /* save the new connection in our list */
169                 conn = kzalloc(sizeof(*conn), GFP_ATOMIC);
170                 if (conn == NULL)
171                         return -ENOMEM;
172                 conn->tuple = *tuple;
173                 list_add(&conn->list, hash);
174                 ++matches;
175         }
176
177         return matches;
178 }
179
180 static bool connlimit_match(const struct sk_buff *skb,
181                             const struct net_device *in,
182                             const struct net_device *out,
183                             const struct xt_match *match,
184                             const void *matchinfo, int offset,
185                             unsigned int protoff, bool *hotdrop)
186 {
187         const struct xt_connlimit_info *info = matchinfo;
188         union nf_conntrack_address addr, mask;
189         struct nf_conntrack_tuple tuple;
190         const struct nf_conntrack_tuple *tuple_ptr = &tuple;
191         enum ip_conntrack_info ctinfo;
192         const struct nf_conn *ct;
193         int connections;
194
195         ct = nf_ct_get(skb, &ctinfo);
196         if (ct != NULL)
197                 tuple_ptr = &ct->tuplehash[0].tuple;
198         else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
199                                     match->family, &tuple))
200                 goto hotdrop;
201
202         if (match->family == AF_INET6) {
203                 const struct ipv6hdr *iph = ipv6_hdr(skb);
204                 memcpy(&addr.ip6, &iph->saddr, sizeof(iph->saddr));
205                 memcpy(&mask.ip6, info->v6_mask, sizeof(info->v6_mask));
206         } else {
207                 const struct iphdr *iph = ip_hdr(skb);
208                 addr.ip = iph->saddr;
209                 mask.ip = info->v4_mask;
210         }
211
212         spin_lock_bh(&info->data->lock);
213         connections = count_them(info->data, tuple_ptr, &addr, &mask, match);
214         spin_unlock_bh(&info->data->lock);
215
216         if (connections < 0) {
217                 /* kmalloc failed, drop it entirely */
218                 *hotdrop = true;
219                 return false;
220         }
221
222         return (connections > info->limit) ^ info->inverse;
223
224  hotdrop:
225         *hotdrop = true;
226         return false;
227 }
228
229 static bool connlimit_check(const char *tablename, const void *ip,
230                             const struct xt_match *match, void *matchinfo,
231                             unsigned int hook_mask)
232 {
233         struct xt_connlimit_info *info = matchinfo;
234         unsigned int i;
235
236         if (nf_ct_l3proto_try_module_get(match->family) < 0) {
237                 printk(KERN_WARNING "cannot load conntrack support for "
238                        "address family %u\n", match->family);
239                 return false;
240         }
241
242         /* init private data */
243         info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
244         if (info->data == NULL) {
245                 nf_ct_l3proto_module_put(match->family);
246                 return false;
247         }
248
249         spin_lock_init(&info->data->lock);
250         for (i = 0; i < ARRAY_SIZE(info->data->iphash); ++i)
251                 INIT_LIST_HEAD(&info->data->iphash[i]);
252
253         return true;
254 }
255
256 static void connlimit_destroy(const struct xt_match *match, void *matchinfo)
257 {
258         struct xt_connlimit_info *info = matchinfo;
259         struct xt_connlimit_conn *conn;
260         struct xt_connlimit_conn *tmp;
261         struct list_head *hash = info->data->iphash;
262         unsigned int i;
263
264         nf_ct_l3proto_module_put(match->family);
265
266         for (i = 0; i < ARRAY_SIZE(info->data->iphash); ++i) {
267                 list_for_each_entry_safe(conn, tmp, &hash[i], list) {
268                         list_del(&conn->list);
269                         kfree(conn);
270                 }
271         }
272
273         kfree(info->data);
274 }
275
276 static struct xt_match connlimit_reg[] __read_mostly = {
277         {
278                 .name       = "connlimit",
279                 .family     = AF_INET,
280                 .checkentry = connlimit_check,
281                 .match      = connlimit_match,
282                 .matchsize  = sizeof(struct xt_connlimit_info),
283                 .destroy    = connlimit_destroy,
284                 .me         = THIS_MODULE,
285         },
286         {
287                 .name       = "connlimit",
288                 .family     = AF_INET6,
289                 .checkentry = connlimit_check,
290                 .match      = connlimit_match,
291                 .matchsize  = sizeof(struct xt_connlimit_info),
292                 .destroy    = connlimit_destroy,
293                 .me         = THIS_MODULE,
294         },
295 };
296
297 static int __init xt_connlimit_init(void)
298 {
299         return xt_register_matches(connlimit_reg, ARRAY_SIZE(connlimit_reg));
300 }
301
302 static void __exit xt_connlimit_exit(void)
303 {
304         xt_unregister_matches(connlimit_reg, ARRAY_SIZE(connlimit_reg));
305 }
306
307 module_init(xt_connlimit_init);
308 module_exit(xt_connlimit_exit);
309 MODULE_AUTHOR("Jan Engelhardt <jengelh@gmx.de>");
310 MODULE_DESCRIPTION("netfilter xt_connlimit match module");
311 MODULE_LICENSE("GPL");
312 MODULE_ALIAS("ipt_connlimit");
313 MODULE_ALIAS("ip6t_connlimit");