1 /* Expectation handling for nf_conntrack. */
3 /* (C) 1999-2001 Paul `Rusty' Russell
4 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
5 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/types.h>
13 #include <linux/netfilter.h>
14 #include <linux/skbuff.h>
15 #include <linux/proc_fs.h>
16 #include <linux/seq_file.h>
17 #include <linux/stddef.h>
18 #include <linux/slab.h>
19 #include <linux/err.h>
20 #include <linux/percpu.h>
21 #include <linux/kernel.h>
22 #include <linux/jhash.h>
23 #include <net/net_namespace.h>
25 #include <net/netfilter/nf_conntrack.h>
26 #include <net/netfilter/nf_conntrack_core.h>
27 #include <net/netfilter/nf_conntrack_expect.h>
28 #include <net/netfilter/nf_conntrack_helper.h>
29 #include <net/netfilter/nf_conntrack_tuple.h>
31 unsigned int nf_ct_expect_hsize __read_mostly;
32 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
34 static unsigned int nf_ct_expect_hash_rnd __read_mostly;
35 unsigned int nf_ct_expect_max __read_mostly;
36 static int nf_ct_expect_hash_rnd_initted __read_mostly;
38 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
40 /* nf_conntrack_expect helper functions */
41 void nf_ct_unlink_expect(struct nf_conntrack_expect *exp)
43 struct nf_conn_help *master_help = nfct_help(exp->master);
44 struct net *net = nf_ct_exp_net(exp);
46 NF_CT_ASSERT(master_help);
47 NF_CT_ASSERT(!timer_pending(&exp->timeout));
49 hlist_del_rcu(&exp->hnode);
50 net->ct.expect_count--;
52 hlist_del(&exp->lnode);
53 master_help->expecting[exp->class]--;
54 nf_ct_expect_put(exp);
56 NF_CT_STAT_INC(expect_delete);
58 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect);
60 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
62 struct nf_conntrack_expect *exp = (void *)ul_expect;
64 spin_lock_bh(&nf_conntrack_lock);
65 nf_ct_unlink_expect(exp);
66 spin_unlock_bh(&nf_conntrack_lock);
67 nf_ct_expect_put(exp);
70 static unsigned int nf_ct_expect_dst_hash(const struct nf_conntrack_tuple *tuple)
74 if (unlikely(!nf_ct_expect_hash_rnd_initted)) {
75 get_random_bytes(&nf_ct_expect_hash_rnd, 4);
76 nf_ct_expect_hash_rnd_initted = 1;
79 hash = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
80 (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
81 (__force __u16)tuple->dst.u.all) ^ nf_ct_expect_hash_rnd);
82 return ((u64)hash * nf_ct_expect_hsize) >> 32;
85 struct nf_conntrack_expect *
86 __nf_ct_expect_find(struct net *net, const struct nf_conntrack_tuple *tuple)
88 struct nf_conntrack_expect *i;
92 if (!net->ct.expect_count)
95 h = nf_ct_expect_dst_hash(tuple);
96 hlist_for_each_entry_rcu(i, n, &net->ct.expect_hash[h], hnode) {
97 if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask))
102 EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
104 /* Just find a expectation corresponding to a tuple. */
105 struct nf_conntrack_expect *
106 nf_ct_expect_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
108 struct nf_conntrack_expect *i;
111 i = __nf_ct_expect_find(net, tuple);
112 if (i && !atomic_inc_not_zero(&i->use))
118 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
120 /* If an expectation for this connection is found, it gets delete from
121 * global list then returned. */
122 struct nf_conntrack_expect *
123 nf_ct_find_expectation(struct net *net, const struct nf_conntrack_tuple *tuple)
125 struct nf_conntrack_expect *i, *exp = NULL;
126 struct hlist_node *n;
129 if (!net->ct.expect_count)
132 h = nf_ct_expect_dst_hash(tuple);
133 hlist_for_each_entry(i, n, &net->ct.expect_hash[h], hnode) {
134 if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
135 nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)) {
143 /* If master is not in hash table yet (ie. packet hasn't left
144 this machine yet), how can other end know about expected?
145 Hence these are not the droids you are looking for (if
146 master ct never got confirmed, we'd hold a reference to it
147 and weird things would happen to future packets). */
148 if (!nf_ct_is_confirmed(exp->master))
151 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
152 atomic_inc(&exp->use);
154 } else if (del_timer(&exp->timeout)) {
155 nf_ct_unlink_expect(exp);
162 /* delete all expectations for this conntrack */
163 void nf_ct_remove_expectations(struct nf_conn *ct)
165 struct nf_conn_help *help = nfct_help(ct);
166 struct nf_conntrack_expect *exp;
167 struct hlist_node *n, *next;
169 /* Optimization: most connection never expect any others. */
173 hlist_for_each_entry_safe(exp, n, next, &help->expectations, lnode) {
174 if (del_timer(&exp->timeout)) {
175 nf_ct_unlink_expect(exp);
176 nf_ct_expect_put(exp);
180 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
182 /* Would two expected things clash? */
183 static inline int expect_clash(const struct nf_conntrack_expect *a,
184 const struct nf_conntrack_expect *b)
186 /* Part covered by intersection of masks must be unequal,
187 otherwise they clash */
188 struct nf_conntrack_tuple_mask intersect_mask;
191 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
193 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
194 intersect_mask.src.u3.all[count] =
195 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
198 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask);
201 static inline int expect_matches(const struct nf_conntrack_expect *a,
202 const struct nf_conntrack_expect *b)
204 return a->master == b->master && a->class == b->class
205 && nf_ct_tuple_equal(&a->tuple, &b->tuple)
206 && nf_ct_tuple_mask_equal(&a->mask, &b->mask);
209 /* Generally a bad idea to call this: could have matched already. */
210 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
212 spin_lock_bh(&nf_conntrack_lock);
213 if (del_timer(&exp->timeout)) {
214 nf_ct_unlink_expect(exp);
215 nf_ct_expect_put(exp);
217 spin_unlock_bh(&nf_conntrack_lock);
219 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
221 /* We don't increase the master conntrack refcount for non-fulfilled
222 * conntracks. During the conntrack destruction, the expectations are
223 * always killed before the conntrack itself */
224 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
226 struct nf_conntrack_expect *new;
228 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
233 atomic_set(&new->use, 1);
234 INIT_RCU_HEAD(&new->rcu);
237 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
239 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
241 const union nf_inet_addr *saddr,
242 const union nf_inet_addr *daddr,
243 u_int8_t proto, const __be16 *src, const __be16 *dst)
247 if (family == AF_INET)
254 exp->expectfn = NULL;
256 exp->tuple.src.l3num = family;
257 exp->tuple.dst.protonum = proto;
260 memcpy(&exp->tuple.src.u3, saddr, len);
261 if (sizeof(exp->tuple.src.u3) > len)
262 /* address needs to be cleared for nf_ct_tuple_equal */
263 memset((void *)&exp->tuple.src.u3 + len, 0x00,
264 sizeof(exp->tuple.src.u3) - len);
265 memset(&exp->mask.src.u3, 0xFF, len);
266 if (sizeof(exp->mask.src.u3) > len)
267 memset((void *)&exp->mask.src.u3 + len, 0x00,
268 sizeof(exp->mask.src.u3) - len);
270 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
271 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
275 exp->tuple.src.u.all = *src;
276 exp->mask.src.u.all = htons(0xFFFF);
278 exp->tuple.src.u.all = 0;
279 exp->mask.src.u.all = 0;
282 memcpy(&exp->tuple.dst.u3, daddr, len);
283 if (sizeof(exp->tuple.dst.u3) > len)
284 /* address needs to be cleared for nf_ct_tuple_equal */
285 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
286 sizeof(exp->tuple.dst.u3) - len);
288 exp->tuple.dst.u.all = *dst;
290 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
292 static void nf_ct_expect_free_rcu(struct rcu_head *head)
294 struct nf_conntrack_expect *exp;
296 exp = container_of(head, struct nf_conntrack_expect, rcu);
297 kmem_cache_free(nf_ct_expect_cachep, exp);
300 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
302 if (atomic_dec_and_test(&exp->use))
303 call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
305 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
307 static void nf_ct_expect_insert(struct nf_conntrack_expect *exp)
309 struct nf_conn_help *master_help = nfct_help(exp->master);
310 struct net *net = nf_ct_exp_net(exp);
311 const struct nf_conntrack_expect_policy *p;
312 unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
314 atomic_inc(&exp->use);
316 hlist_add_head(&exp->lnode, &master_help->expectations);
317 master_help->expecting[exp->class]++;
319 hlist_add_head_rcu(&exp->hnode, &net->ct.expect_hash[h]);
320 net->ct.expect_count++;
322 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
324 p = &master_help->helper->expect_policy[exp->class];
325 exp->timeout.expires = jiffies + p->timeout * HZ;
326 add_timer(&exp->timeout);
328 atomic_inc(&exp->use);
329 NF_CT_STAT_INC(expect_create);
332 /* Race with expectations being used means we could have none to find; OK. */
333 static void evict_oldest_expect(struct nf_conn *master,
334 struct nf_conntrack_expect *new)
336 struct nf_conn_help *master_help = nfct_help(master);
337 struct nf_conntrack_expect *exp, *last = NULL;
338 struct hlist_node *n;
340 hlist_for_each_entry(exp, n, &master_help->expectations, lnode) {
341 if (exp->class == new->class)
345 if (last && del_timer(&last->timeout)) {
346 nf_ct_unlink_expect(last);
347 nf_ct_expect_put(last);
351 static inline int refresh_timer(struct nf_conntrack_expect *i)
353 struct nf_conn_help *master_help = nfct_help(i->master);
354 const struct nf_conntrack_expect_policy *p;
356 if (!del_timer(&i->timeout))
359 p = &master_help->helper->expect_policy[i->class];
360 i->timeout.expires = jiffies + p->timeout * HZ;
361 add_timer(&i->timeout);
365 int nf_ct_expect_related(struct nf_conntrack_expect *expect)
367 const struct nf_conntrack_expect_policy *p;
368 struct nf_conntrack_expect *i;
369 struct nf_conn *master = expect->master;
370 struct nf_conn_help *master_help = nfct_help(master);
371 struct net *net = nf_ct_exp_net(expect);
372 struct hlist_node *n;
376 NF_CT_ASSERT(master_help);
378 spin_lock_bh(&nf_conntrack_lock);
379 if (!master_help->helper) {
383 h = nf_ct_expect_dst_hash(&expect->tuple);
384 hlist_for_each_entry(i, n, &net->ct.expect_hash[h], hnode) {
385 if (expect_matches(i, expect)) {
386 /* Refresh timer: if it's dying, ignore.. */
387 if (refresh_timer(i)) {
391 } else if (expect_clash(i, expect)) {
396 /* Will be over limit? */
397 p = &master_help->helper->expect_policy[expect->class];
398 if (p->max_expected &&
399 master_help->expecting[expect->class] >= p->max_expected) {
400 evict_oldest_expect(master, expect);
401 if (master_help->expecting[expect->class] >= p->max_expected) {
407 if (net->ct.expect_count >= nf_ct_expect_max) {
410 "nf_conntrack: expectation table full\n");
415 nf_ct_expect_insert(expect);
416 nf_ct_expect_event(IPEXP_NEW, expect);
419 spin_unlock_bh(&nf_conntrack_lock);
422 EXPORT_SYMBOL_GPL(nf_ct_expect_related);
424 #ifdef CONFIG_PROC_FS
425 struct ct_expect_iter_state {
429 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
431 struct net *net = &init_net;
432 struct ct_expect_iter_state *st = seq->private;
433 struct hlist_node *n;
435 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
436 n = rcu_dereference(net->ct.expect_hash[st->bucket].first);
443 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
444 struct hlist_node *head)
446 struct net *net = &init_net;
447 struct ct_expect_iter_state *st = seq->private;
449 head = rcu_dereference(head->next);
450 while (head == NULL) {
451 if (++st->bucket >= nf_ct_expect_hsize)
453 head = rcu_dereference(net->ct.expect_hash[st->bucket].first);
458 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
460 struct hlist_node *head = ct_expect_get_first(seq);
463 while (pos && (head = ct_expect_get_next(seq, head)))
465 return pos ? NULL : head;
468 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
472 return ct_expect_get_idx(seq, *pos);
475 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
478 return ct_expect_get_next(seq, v);
481 static void exp_seq_stop(struct seq_file *seq, void *v)
487 static int exp_seq_show(struct seq_file *s, void *v)
489 struct nf_conntrack_expect *expect;
490 struct hlist_node *n = v;
493 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
495 if (expect->timeout.function)
496 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
497 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
500 seq_printf(s, "l3proto = %u proto=%u ",
501 expect->tuple.src.l3num,
502 expect->tuple.dst.protonum);
503 print_tuple(s, &expect->tuple,
504 __nf_ct_l3proto_find(expect->tuple.src.l3num),
505 __nf_ct_l4proto_find(expect->tuple.src.l3num,
506 expect->tuple.dst.protonum));
508 if (expect->flags & NF_CT_EXPECT_PERMANENT) {
509 seq_printf(s, "PERMANENT");
512 if (expect->flags & NF_CT_EXPECT_INACTIVE)
513 seq_printf(s, "%sINACTIVE", delim);
515 return seq_putc(s, '\n');
518 static const struct seq_operations exp_seq_ops = {
519 .start = exp_seq_start,
520 .next = exp_seq_next,
521 .stop = exp_seq_stop,
525 static int exp_open(struct inode *inode, struct file *file)
527 return seq_open_private(file, &exp_seq_ops,
528 sizeof(struct ct_expect_iter_state));
531 static const struct file_operations exp_file_ops = {
532 .owner = THIS_MODULE,
536 .release = seq_release_private,
538 #endif /* CONFIG_PROC_FS */
540 static int exp_proc_init(void)
542 #ifdef CONFIG_PROC_FS
543 struct proc_dir_entry *proc;
545 proc = proc_net_fops_create(&init_net, "nf_conntrack_expect", 0440, &exp_file_ops);
548 #endif /* CONFIG_PROC_FS */
552 static void exp_proc_remove(void)
554 #ifdef CONFIG_PROC_FS
555 proc_net_remove(&init_net, "nf_conntrack_expect");
556 #endif /* CONFIG_PROC_FS */
559 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0600);
561 int nf_conntrack_expect_init(struct net *net)
565 if (!nf_ct_expect_hsize) {
566 nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
567 if (!nf_ct_expect_hsize)
568 nf_ct_expect_hsize = 1;
570 nf_ct_expect_max = nf_ct_expect_hsize * 4;
572 net->ct.expect_count = 0;
573 net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize,
574 &net->ct.expect_vmalloc);
575 if (net->ct.expect_hash == NULL)
578 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
579 sizeof(struct nf_conntrack_expect),
581 if (!nf_ct_expect_cachep)
584 err = exp_proc_init();
591 kmem_cache_destroy(nf_ct_expect_cachep);
593 nf_ct_free_hashtable(net->ct.expect_hash, net->ct.expect_vmalloc,
599 void nf_conntrack_expect_fini(struct net *net)
602 kmem_cache_destroy(nf_ct_expect_cachep);
603 nf_ct_free_hashtable(net->ct.expect_hash, net->ct.expect_vmalloc,