Loading...
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
16#include <linux/ip.h>
17#include <linux/ipv6.h>
18#include <linux/module.h>
19#include <linux/skbuff.h>
20#include <linux/netfilter/x_tables.h>
21#include <linux/netfilter/xt_connlimit.h>
22
23#include <net/netfilter/nf_conntrack.h>
24#include <net/netfilter/nf_conntrack_core.h>
25#include <net/netfilter/nf_conntrack_tuple.h>
26#include <net/netfilter/nf_conntrack_zones.h>
27#include <net/netfilter/nf_conntrack_count.h>
28
29static bool
30connlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
31{
32 struct net *net = xt_net(par);
33 const struct xt_connlimit_info *info = par->matchinfo;
34 struct nf_conntrack_tuple tuple;
35 const struct nf_conntrack_tuple *tuple_ptr = &tuple;
36 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
37 enum ip_conntrack_info ctinfo;
38 const struct nf_conn *ct;
39 unsigned int connections;
40 u32 key[5];
41
42 ct = nf_ct_get(skb, &ctinfo);
43 if (ct != NULL) {
44 tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
45 zone = nf_ct_zone(ct);
46 } else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
47 xt_family(par), net, &tuple)) {
48 goto hotdrop;
49 }
50
51 if (xt_family(par) == NFPROTO_IPV6) {
52 const struct ipv6hdr *iph = ipv6_hdr(skb);
53 union nf_inet_addr addr;
54 unsigned int i;
55
56 memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
57 &iph->daddr : &iph->saddr, sizeof(addr.ip6));
58
59 for (i = 0; i < ARRAY_SIZE(addr.ip6); ++i)
60 addr.ip6[i] &= info->mask.ip6[i];
61 memcpy(key, &addr, sizeof(addr.ip6));
62 key[4] = zone->id;
63 } else {
64 const struct iphdr *iph = ip_hdr(skb);
65 key[0] = (info->flags & XT_CONNLIMIT_DADDR) ?
66 iph->daddr : iph->saddr;
67
68 key[0] &= info->mask.ip;
69 key[1] = zone->id;
70 }
71
72 connections = nf_conncount_count(net, info->data, key, tuple_ptr,
73 zone);
74 if (connections == 0)
75 /* kmalloc failed, drop it entirely */
76 goto hotdrop;
77
78 return (connections > info->limit) ^ !!(info->flags & XT_CONNLIMIT_INVERT);
79
80 hotdrop:
81 par->hotdrop = true;
82 return false;
83}
84
85static int connlimit_mt_check(const struct xt_mtchk_param *par)
86{
87 struct xt_connlimit_info *info = par->matchinfo;
88 unsigned int keylen;
89
90 keylen = sizeof(u32);
91 if (par->family == NFPROTO_IPV6)
92 keylen += sizeof(struct in6_addr);
93 else
94 keylen += sizeof(struct in_addr);
95
96 /* init private data */
97 info->data = nf_conncount_init(par->net, par->family, keylen);
98
99 return PTR_ERR_OR_ZERO(info->data);
100}
101
102static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
103{
104 const struct xt_connlimit_info *info = par->matchinfo;
105
106 nf_conncount_destroy(par->net, par->family, info->data);
107}
108
109static struct xt_match connlimit_mt_reg __read_mostly = {
110 .name = "connlimit",
111 .revision = 1,
112 .family = NFPROTO_UNSPEC,
113 .checkentry = connlimit_mt_check,
114 .match = connlimit_mt,
115 .matchsize = sizeof(struct xt_connlimit_info),
116 .usersize = offsetof(struct xt_connlimit_info, data),
117 .destroy = connlimit_mt_destroy,
118 .me = THIS_MODULE,
119};
120
121static int __init connlimit_mt_init(void)
122{
123 return xt_register_match(&connlimit_mt_reg);
124}
125
126static void __exit connlimit_mt_exit(void)
127{
128 xt_unregister_match(&connlimit_mt_reg);
129}
130
131module_init(connlimit_mt_init);
132module_exit(connlimit_mt_exit);
133MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
134MODULE_DESCRIPTION("Xtables: Number of connections matching");
135MODULE_LICENSE("GPL");
136MODULE_ALIAS("ipt_connlimit");
137MODULE_ALIAS("ip6t_connlimit");
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/rbtree.h>
23#include <linux/module.h>
24#include <linux/random.h>
25#include <linux/skbuff.h>
26#include <linux/spinlock.h>
27#include <linux/netfilter/nf_conntrack_tcp.h>
28#include <linux/netfilter/x_tables.h>
29#include <linux/netfilter/xt_connlimit.h>
30#include <net/netfilter/nf_conntrack.h>
31#include <net/netfilter/nf_conntrack_core.h>
32#include <net/netfilter/nf_conntrack_tuple.h>
33#include <net/netfilter/nf_conntrack_zones.h>
34
35#define CONNLIMIT_SLOTS 256U
36
37#ifdef CONFIG_LOCKDEP
38#define CONNLIMIT_LOCK_SLOTS 8U
39#else
40#define CONNLIMIT_LOCK_SLOTS 256U
41#endif
42
43#define CONNLIMIT_GC_MAX_NODES 8
44
45/* we will save the tuples of all connections we care about */
46struct xt_connlimit_conn {
47 struct hlist_node node;
48 struct nf_conntrack_tuple tuple;
49 union nf_inet_addr addr;
50};
51
52struct xt_connlimit_rb {
53 struct rb_node node;
54 struct hlist_head hhead; /* connections/hosts in same subnet */
55 union nf_inet_addr addr; /* search key */
56};
57
58static spinlock_t xt_connlimit_locks[CONNLIMIT_LOCK_SLOTS] __cacheline_aligned_in_smp;
59
60struct xt_connlimit_data {
61 struct rb_root climit_root4[CONNLIMIT_SLOTS];
62 struct rb_root climit_root6[CONNLIMIT_SLOTS];
63};
64
65static u_int32_t connlimit_rnd __read_mostly;
66static struct kmem_cache *connlimit_rb_cachep __read_mostly;
67static struct kmem_cache *connlimit_conn_cachep __read_mostly;
68
69static inline unsigned int connlimit_iphash(__be32 addr)
70{
71 return jhash_1word((__force __u32)addr,
72 connlimit_rnd) % CONNLIMIT_SLOTS;
73}
74
75static inline unsigned int
76connlimit_iphash6(const union nf_inet_addr *addr,
77 const union nf_inet_addr *mask)
78{
79 union nf_inet_addr res;
80 unsigned int i;
81
82 for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i)
83 res.ip6[i] = addr->ip6[i] & mask->ip6[i];
84
85 return jhash2((u32 *)res.ip6, ARRAY_SIZE(res.ip6),
86 connlimit_rnd) % CONNLIMIT_SLOTS;
87}
88
89static inline bool already_closed(const struct nf_conn *conn)
90{
91 if (nf_ct_protonum(conn) == IPPROTO_TCP)
92 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
93 conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
94 else
95 return 0;
96}
97
98static int
99same_source_net(const union nf_inet_addr *addr,
100 const union nf_inet_addr *mask,
101 const union nf_inet_addr *u3, u_int8_t family)
102{
103 if (family == NFPROTO_IPV4) {
104 return ntohl(addr->ip & mask->ip) -
105 ntohl(u3->ip & mask->ip);
106 } else {
107 union nf_inet_addr lh, rh;
108 unsigned int i;
109
110 for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i) {
111 lh.ip6[i] = addr->ip6[i] & mask->ip6[i];
112 rh.ip6[i] = u3->ip6[i] & mask->ip6[i];
113 }
114
115 return memcmp(&lh.ip6, &rh.ip6, sizeof(lh.ip6));
116 }
117}
118
119static bool add_hlist(struct hlist_head *head,
120 const struct nf_conntrack_tuple *tuple,
121 const union nf_inet_addr *addr)
122{
123 struct xt_connlimit_conn *conn;
124
125 conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
126 if (conn == NULL)
127 return false;
128 conn->tuple = *tuple;
129 conn->addr = *addr;
130 hlist_add_head(&conn->node, head);
131 return true;
132}
133
134static unsigned int check_hlist(struct net *net,
135 struct hlist_head *head,
136 const struct nf_conntrack_tuple *tuple,
137 bool *addit)
138{
139 const struct nf_conntrack_tuple_hash *found;
140 struct xt_connlimit_conn *conn;
141 struct hlist_node *n;
142 struct nf_conn *found_ct;
143 unsigned int length = 0;
144
145 *addit = true;
146 rcu_read_lock();
147
148 /* check the saved connections */
149 hlist_for_each_entry_safe(conn, n, head, node) {
150 found = nf_conntrack_find_get(net, NF_CT_DEFAULT_ZONE,
151 &conn->tuple);
152 if (found == NULL) {
153 hlist_del(&conn->node);
154 kmem_cache_free(connlimit_conn_cachep, conn);
155 continue;
156 }
157
158 found_ct = nf_ct_tuplehash_to_ctrack(found);
159
160 if (nf_ct_tuple_equal(&conn->tuple, tuple)) {
161 /*
162 * Just to be sure we have it only once in the list.
163 * We should not see tuples twice unless someone hooks
164 * this into a table without "-p tcp --syn".
165 */
166 *addit = false;
167 } else if (already_closed(found_ct)) {
168 /*
169 * we do not care about connections which are
170 * closed already -> ditch it
171 */
172 nf_ct_put(found_ct);
173 hlist_del(&conn->node);
174 kmem_cache_free(connlimit_conn_cachep, conn);
175 continue;
176 }
177
178 nf_ct_put(found_ct);
179 length++;
180 }
181
182 rcu_read_unlock();
183
184 return length;
185}
186
187static void tree_nodes_free(struct rb_root *root,
188 struct xt_connlimit_rb *gc_nodes[],
189 unsigned int gc_count)
190{
191 struct xt_connlimit_rb *rbconn;
192
193 while (gc_count) {
194 rbconn = gc_nodes[--gc_count];
195 rb_erase(&rbconn->node, root);
196 kmem_cache_free(connlimit_rb_cachep, rbconn);
197 }
198}
199
200static unsigned int
201count_tree(struct net *net, struct rb_root *root,
202 const struct nf_conntrack_tuple *tuple,
203 const union nf_inet_addr *addr, const union nf_inet_addr *mask,
204 u8 family)
205{
206 struct xt_connlimit_rb *gc_nodes[CONNLIMIT_GC_MAX_NODES];
207 struct rb_node **rbnode, *parent;
208 struct xt_connlimit_rb *rbconn;
209 struct xt_connlimit_conn *conn;
210 unsigned int gc_count;
211 bool no_gc = false;
212
213 restart:
214 gc_count = 0;
215 parent = NULL;
216 rbnode = &(root->rb_node);
217 while (*rbnode) {
218 int diff;
219 bool addit;
220
221 rbconn = container_of(*rbnode, struct xt_connlimit_rb, node);
222
223 parent = *rbnode;
224 diff = same_source_net(addr, mask, &rbconn->addr, family);
225 if (diff < 0) {
226 rbnode = &((*rbnode)->rb_left);
227 } else if (diff > 0) {
228 rbnode = &((*rbnode)->rb_right);
229 } else {
230 /* same source network -> be counted! */
231 unsigned int count;
232 count = check_hlist(net, &rbconn->hhead, tuple, &addit);
233
234 tree_nodes_free(root, gc_nodes, gc_count);
235 if (!addit)
236 return count;
237
238 if (!add_hlist(&rbconn->hhead, tuple, addr))
239 return 0; /* hotdrop */
240
241 return count + 1;
242 }
243
244 if (no_gc || gc_count >= ARRAY_SIZE(gc_nodes))
245 continue;
246
247 /* only used for GC on hhead, retval and 'addit' ignored */
248 check_hlist(net, &rbconn->hhead, tuple, &addit);
249 if (hlist_empty(&rbconn->hhead))
250 gc_nodes[gc_count++] = rbconn;
251 }
252
253 if (gc_count) {
254 no_gc = true;
255 tree_nodes_free(root, gc_nodes, gc_count);
256 /* tree_node_free before new allocation permits
257 * allocator to re-use newly free'd object.
258 *
259 * This is a rare event; in most cases we will find
260 * existing node to re-use. (or gc_count is 0).
261 */
262 goto restart;
263 }
264
265 /* no match, need to insert new node */
266 rbconn = kmem_cache_alloc(connlimit_rb_cachep, GFP_ATOMIC);
267 if (rbconn == NULL)
268 return 0;
269
270 conn = kmem_cache_alloc(connlimit_conn_cachep, GFP_ATOMIC);
271 if (conn == NULL) {
272 kmem_cache_free(connlimit_rb_cachep, rbconn);
273 return 0;
274 }
275
276 conn->tuple = *tuple;
277 conn->addr = *addr;
278 rbconn->addr = *addr;
279
280 INIT_HLIST_HEAD(&rbconn->hhead);
281 hlist_add_head(&conn->node, &rbconn->hhead);
282
283 rb_link_node(&rbconn->node, parent, rbnode);
284 rb_insert_color(&rbconn->node, root);
285 return 1;
286}
287
288static int count_them(struct net *net,
289 struct xt_connlimit_data *data,
290 const struct nf_conntrack_tuple *tuple,
291 const union nf_inet_addr *addr,
292 const union nf_inet_addr *mask,
293 u_int8_t family)
294{
295 struct rb_root *root;
296 int count;
297 u32 hash;
298
299 if (family == NFPROTO_IPV6) {
300 hash = connlimit_iphash6(addr, mask);
301 root = &data->climit_root6[hash];
302 } else {
303 hash = connlimit_iphash(addr->ip & mask->ip);
304 root = &data->climit_root4[hash];
305 }
306
307 spin_lock_bh(&xt_connlimit_locks[hash % CONNLIMIT_LOCK_SLOTS]);
308
309 count = count_tree(net, root, tuple, addr, mask, family);
310
311 spin_unlock_bh(&xt_connlimit_locks[hash % CONNLIMIT_LOCK_SLOTS]);
312
313 return count;
314}
315
316static bool
317connlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
318{
319 struct net *net = dev_net(par->in ? par->in : par->out);
320 const struct xt_connlimit_info *info = par->matchinfo;
321 union nf_inet_addr addr;
322 struct nf_conntrack_tuple tuple;
323 const struct nf_conntrack_tuple *tuple_ptr = &tuple;
324 enum ip_conntrack_info ctinfo;
325 const struct nf_conn *ct;
326 unsigned int connections;
327
328 ct = nf_ct_get(skb, &ctinfo);
329 if (ct != NULL)
330 tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
331 else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
332 par->family, &tuple))
333 goto hotdrop;
334
335 if (par->family == NFPROTO_IPV6) {
336 const struct ipv6hdr *iph = ipv6_hdr(skb);
337 memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
338 &iph->daddr : &iph->saddr, sizeof(addr.ip6));
339 } else {
340 const struct iphdr *iph = ip_hdr(skb);
341 addr.ip = (info->flags & XT_CONNLIMIT_DADDR) ?
342 iph->daddr : iph->saddr;
343 }
344
345 connections = count_them(net, info->data, tuple_ptr, &addr,
346 &info->mask, par->family);
347 if (connections == 0)
348 /* kmalloc failed, drop it entirely */
349 goto hotdrop;
350
351 return (connections > info->limit) ^
352 !!(info->flags & XT_CONNLIMIT_INVERT);
353
354 hotdrop:
355 par->hotdrop = true;
356 return false;
357}
358
359static int connlimit_mt_check(const struct xt_mtchk_param *par)
360{
361 struct xt_connlimit_info *info = par->matchinfo;
362 unsigned int i;
363 int ret;
364
365 if (unlikely(!connlimit_rnd)) {
366 u_int32_t rand;
367
368 do {
369 get_random_bytes(&rand, sizeof(rand));
370 } while (!rand);
371 cmpxchg(&connlimit_rnd, 0, rand);
372 }
373 ret = nf_ct_l3proto_try_module_get(par->family);
374 if (ret < 0) {
375 pr_info("cannot load conntrack support for "
376 "address family %u\n", par->family);
377 return ret;
378 }
379
380 /* init private data */
381 info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
382 if (info->data == NULL) {
383 nf_ct_l3proto_module_put(par->family);
384 return -ENOMEM;
385 }
386
387 for (i = 0; i < ARRAY_SIZE(info->data->climit_root4); ++i)
388 info->data->climit_root4[i] = RB_ROOT;
389 for (i = 0; i < ARRAY_SIZE(info->data->climit_root6); ++i)
390 info->data->climit_root6[i] = RB_ROOT;
391
392 return 0;
393}
394
395static void destroy_tree(struct rb_root *r)
396{
397 struct xt_connlimit_conn *conn;
398 struct xt_connlimit_rb *rbconn;
399 struct hlist_node *n;
400 struct rb_node *node;
401
402 while ((node = rb_first(r)) != NULL) {
403 rbconn = container_of(node, struct xt_connlimit_rb, node);
404
405 rb_erase(node, r);
406
407 hlist_for_each_entry_safe(conn, n, &rbconn->hhead, node)
408 kmem_cache_free(connlimit_conn_cachep, conn);
409
410 kmem_cache_free(connlimit_rb_cachep, rbconn);
411 }
412}
413
414static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
415{
416 const struct xt_connlimit_info *info = par->matchinfo;
417 unsigned int i;
418
419 nf_ct_l3proto_module_put(par->family);
420
421 for (i = 0; i < ARRAY_SIZE(info->data->climit_root4); ++i)
422 destroy_tree(&info->data->climit_root4[i]);
423 for (i = 0; i < ARRAY_SIZE(info->data->climit_root6); ++i)
424 destroy_tree(&info->data->climit_root6[i]);
425
426 kfree(info->data);
427}
428
429static struct xt_match connlimit_mt_reg __read_mostly = {
430 .name = "connlimit",
431 .revision = 1,
432 .family = NFPROTO_UNSPEC,
433 .checkentry = connlimit_mt_check,
434 .match = connlimit_mt,
435 .matchsize = sizeof(struct xt_connlimit_info),
436 .destroy = connlimit_mt_destroy,
437 .me = THIS_MODULE,
438};
439
440static int __init connlimit_mt_init(void)
441{
442 int ret, i;
443
444 BUILD_BUG_ON(CONNLIMIT_LOCK_SLOTS > CONNLIMIT_SLOTS);
445 BUILD_BUG_ON((CONNLIMIT_SLOTS % CONNLIMIT_LOCK_SLOTS) != 0);
446
447 for (i = 0; i < CONNLIMIT_LOCK_SLOTS; ++i)
448 spin_lock_init(&xt_connlimit_locks[i]);
449
450 connlimit_conn_cachep = kmem_cache_create("xt_connlimit_conn",
451 sizeof(struct xt_connlimit_conn),
452 0, 0, NULL);
453 if (!connlimit_conn_cachep)
454 return -ENOMEM;
455
456 connlimit_rb_cachep = kmem_cache_create("xt_connlimit_rb",
457 sizeof(struct xt_connlimit_rb),
458 0, 0, NULL);
459 if (!connlimit_rb_cachep) {
460 kmem_cache_destroy(connlimit_conn_cachep);
461 return -ENOMEM;
462 }
463 ret = xt_register_match(&connlimit_mt_reg);
464 if (ret != 0) {
465 kmem_cache_destroy(connlimit_conn_cachep);
466 kmem_cache_destroy(connlimit_rb_cachep);
467 }
468 return ret;
469}
470
471static void __exit connlimit_mt_exit(void)
472{
473 xt_unregister_match(&connlimit_mt_reg);
474 kmem_cache_destroy(connlimit_conn_cachep);
475 kmem_cache_destroy(connlimit_rb_cachep);
476}
477
478module_init(connlimit_mt_init);
479module_exit(connlimit_mt_exit);
480MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
481MODULE_DESCRIPTION("Xtables: Number of connections matching");
482MODULE_LICENSE("GPL");
483MODULE_ALIAS("ipt_connlimit");
484MODULE_ALIAS("ip6t_connlimit");