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				u16 zone,
138				bool *addit)
139{
140	const struct nf_conntrack_tuple_hash *found;
141	struct xt_connlimit_conn *conn;
142	struct hlist_node *n;
143	struct nf_conn *found_ct;
144	unsigned int length = 0;
145
146	*addit = true;
147	rcu_read_lock();
148
149	/* check the saved connections */
150	hlist_for_each_entry_safe(conn, n, head, node) {
151		found = nf_conntrack_find_get(net, zone, &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, u16 zone)
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, zone, &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, zone, &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, u16 zone)
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, zone);
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	u16 zone = NF_CT_DEFAULT_ZONE;
328
329	ct = nf_ct_get(skb, &ctinfo);
330	if (ct != NULL) {
331		tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
332		zone = nf_ct_zone(ct);
333	} else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
334				    par->family, &tuple)) {
335		goto hotdrop;
336	}
337
338	if (par->family == NFPROTO_IPV6) {
339		const struct ipv6hdr *iph = ipv6_hdr(skb);
340		memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
341		       &iph->daddr : &iph->saddr, sizeof(addr.ip6));
342	} else {
343		const struct iphdr *iph = ip_hdr(skb);
344		addr.ip = (info->flags & XT_CONNLIMIT_DADDR) ?
345			  iph->daddr : iph->saddr;
346	}
347
348	connections = count_them(net, info->data, tuple_ptr, &addr,
349	                         &info->mask, par->family, zone);
350	if (connections == 0)
351		/* kmalloc failed, drop it entirely */
352		goto hotdrop;
353
354	return (connections > info->limit) ^
355	       !!(info->flags & XT_CONNLIMIT_INVERT);
356
357 hotdrop:
358	par->hotdrop = true;
359	return false;
360}
361
362static int connlimit_mt_check(const struct xt_mtchk_param *par)
363{
364	struct xt_connlimit_info *info = par->matchinfo;
365	unsigned int i;
366	int ret;
367
368	if (unlikely(!connlimit_rnd)) {
369		u_int32_t rand;
370
371		do {
372			get_random_bytes(&rand, sizeof(rand));
373		} while (!rand);
374		cmpxchg(&connlimit_rnd, 0, rand);
375	}
376	ret = nf_ct_l3proto_try_module_get(par->family);
377	if (ret < 0) {
378		pr_info("cannot load conntrack support for "
379			"address family %u\n", par->family);
380		return ret;
381	}
382
383	/* init private data */
384	info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
385	if (info->data == NULL) {
386		nf_ct_l3proto_module_put(par->family);
387		return -ENOMEM;
388	}
389
390	for (i = 0; i < ARRAY_SIZE(info->data->climit_root4); ++i)
391		info->data->climit_root4[i] = RB_ROOT;
392	for (i = 0; i < ARRAY_SIZE(info->data->climit_root6); ++i)
393		info->data->climit_root6[i] = RB_ROOT;
394
395	return 0;
396}
397
398static void destroy_tree(struct rb_root *r)
399{
400	struct xt_connlimit_conn *conn;
401	struct xt_connlimit_rb *rbconn;
402	struct hlist_node *n;
403	struct rb_node *node;
404
405	while ((node = rb_first(r)) != NULL) {
406		rbconn = container_of(node, struct xt_connlimit_rb, node);
407
408		rb_erase(node, r);
409
410		hlist_for_each_entry_safe(conn, n, &rbconn->hhead, node)
411			kmem_cache_free(connlimit_conn_cachep, conn);
412
413		kmem_cache_free(connlimit_rb_cachep, rbconn);
414	}
415}
416
417static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
418{
419	const struct xt_connlimit_info *info = par->matchinfo;
420	unsigned int i;
421
422	nf_ct_l3proto_module_put(par->family);
423
424	for (i = 0; i < ARRAY_SIZE(info->data->climit_root4); ++i)
425		destroy_tree(&info->data->climit_root4[i]);
426	for (i = 0; i < ARRAY_SIZE(info->data->climit_root6); ++i)
427		destroy_tree(&info->data->climit_root6[i]);
428
429	kfree(info->data);
430}
431
432static struct xt_match connlimit_mt_reg __read_mostly = {
433	.name       = "connlimit",
434	.revision   = 1,
435	.family     = NFPROTO_UNSPEC,
436	.checkentry = connlimit_mt_check,
437	.match      = connlimit_mt,
438	.matchsize  = sizeof(struct xt_connlimit_info),
439	.destroy    = connlimit_mt_destroy,
440	.me         = THIS_MODULE,
441};
442
443static int __init connlimit_mt_init(void)
444{
445	int ret, i;
446
447	BUILD_BUG_ON(CONNLIMIT_LOCK_SLOTS > CONNLIMIT_SLOTS);
448	BUILD_BUG_ON((CONNLIMIT_SLOTS % CONNLIMIT_LOCK_SLOTS) != 0);
449
450	for (i = 0; i < CONNLIMIT_LOCK_SLOTS; ++i)
451		spin_lock_init(&xt_connlimit_locks[i]);
452
453	connlimit_conn_cachep = kmem_cache_create("xt_connlimit_conn",
454					   sizeof(struct xt_connlimit_conn),
455					   0, 0, NULL);
456	if (!connlimit_conn_cachep)
457		return -ENOMEM;
458
459	connlimit_rb_cachep = kmem_cache_create("xt_connlimit_rb",
460					   sizeof(struct xt_connlimit_rb),
461					   0, 0, NULL);
462	if (!connlimit_rb_cachep) {
463		kmem_cache_destroy(connlimit_conn_cachep);
464		return -ENOMEM;
465	}
466	ret = xt_register_match(&connlimit_mt_reg);
467	if (ret != 0) {
468		kmem_cache_destroy(connlimit_conn_cachep);
469		kmem_cache_destroy(connlimit_rb_cachep);
470	}
471	return ret;
472}
473
474static void __exit connlimit_mt_exit(void)
475{
476	xt_unregister_match(&connlimit_mt_reg);
477	kmem_cache_destroy(connlimit_conn_cachep);
478	kmem_cache_destroy(connlimit_rb_cachep);
479}
480
481module_init(connlimit_mt_init);
482module_exit(connlimit_mt_exit);
483MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
484MODULE_DESCRIPTION("Xtables: Number of connections matching");
485MODULE_LICENSE("GPL");
486MODULE_ALIAS("ipt_connlimit");
487MODULE_ALIAS("ip6t_connlimit");
488