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>
6 * (c) 2005-2012 Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
13 #include <linux/types.h>
14 #include <linux/netfilter.h>
15 #include <linux/skbuff.h>
16 #include <linux/proc_fs.h>
17 #include <linux/seq_file.h>
18 #include <linux/stddef.h>
19 #include <linux/slab.h>
20 #include <linux/err.h>
21 #include <linux/percpu.h>
22 #include <linux/kernel.h>
23 #include <linux/jhash.h>
24 #include <linux/moduleparam.h>
25 #include <linux/export.h>
26 #include <net/net_namespace.h>
28 #include <net/netfilter/nf_conntrack.h>
29 #include <net/netfilter/nf_conntrack_core.h>
30 #include <net/netfilter/nf_conntrack_expect.h>
31 #include <net/netfilter/nf_conntrack_helper.h>
32 #include <net/netfilter/nf_conntrack_tuple.h>
33 #include <net/netfilter/nf_conntrack_zones.h>
35 unsigned int nf_ct_expect_hsize __read_mostly;
36 EXPORT_SYMBOL_GPL(nf_ct_expect_hsize);
38 unsigned int nf_ct_expect_max __read_mostly;
40 static struct kmem_cache *nf_ct_expect_cachep __read_mostly;
42 /* nf_conntrack_expect helper functions */
43 void nf_ct_unlink_expect_report(struct nf_conntrack_expect *exp,
44 u32 portid, int report)
46 struct nf_conn_help *master_help = nfct_help(exp->master);
47 struct net *net = nf_ct_exp_net(exp);
49 NF_CT_ASSERT(master_help);
50 NF_CT_ASSERT(!timer_pending(&exp->timeout));
52 hlist_del_rcu(&exp->hnode);
53 net->ct.expect_count--;
55 hlist_del(&exp->lnode);
56 master_help->expecting[exp->class]--;
58 nf_ct_expect_event_report(IPEXP_DESTROY, exp, portid, report);
59 nf_ct_expect_put(exp);
61 NF_CT_STAT_INC(net, expect_delete);
63 EXPORT_SYMBOL_GPL(nf_ct_unlink_expect_report);
65 static void nf_ct_expectation_timed_out(unsigned long ul_expect)
67 struct nf_conntrack_expect *exp = (void *)ul_expect;
69 spin_lock_bh(&nf_conntrack_expect_lock);
70 nf_ct_unlink_expect(exp);
71 spin_unlock_bh(&nf_conntrack_expect_lock);
72 nf_ct_expect_put(exp);
75 static unsigned int nf_ct_expect_dst_hash(const struct nf_conntrack_tuple *tuple)
79 if (unlikely(!nf_conntrack_hash_rnd)) {
80 init_nf_conntrack_hash_rnd();
83 hash = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
84 (((tuple->dst.protonum ^ tuple->src.l3num) << 16) |
85 (__force __u16)tuple->dst.u.all) ^ nf_conntrack_hash_rnd);
87 return reciprocal_scale(hash, nf_ct_expect_hsize);
90 struct nf_conntrack_expect *
91 __nf_ct_expect_find(struct net *net, u16 zone,
92 const struct nf_conntrack_tuple *tuple)
94 struct nf_conntrack_expect *i;
97 if (!net->ct.expect_count)
100 h = nf_ct_expect_dst_hash(tuple);
101 hlist_for_each_entry_rcu(i, &net->ct.expect_hash[h], hnode) {
102 if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
103 nf_ct_zone(i->master) == zone)
108 EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
110 /* Just find a expectation corresponding to a tuple. */
111 struct nf_conntrack_expect *
112 nf_ct_expect_find_get(struct net *net, u16 zone,
113 const struct nf_conntrack_tuple *tuple)
115 struct nf_conntrack_expect *i;
118 i = __nf_ct_expect_find(net, zone, tuple);
119 if (i && !atomic_inc_not_zero(&i->use))
125 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
127 /* If an expectation for this connection is found, it gets delete from
128 * global list then returned. */
129 struct nf_conntrack_expect *
130 nf_ct_find_expectation(struct net *net, u16 zone,
131 const struct nf_conntrack_tuple *tuple)
133 struct nf_conntrack_expect *i, *exp = NULL;
136 if (!net->ct.expect_count)
139 h = nf_ct_expect_dst_hash(tuple);
140 hlist_for_each_entry(i, &net->ct.expect_hash[h], hnode) {
141 if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
142 nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
143 nf_ct_zone(i->master) == zone) {
151 /* If master is not in hash table yet (ie. packet hasn't left
152 this machine yet), how can other end know about expected?
153 Hence these are not the droids you are looking for (if
154 master ct never got confirmed, we'd hold a reference to it
155 and weird things would happen to future packets). */
156 if (!nf_ct_is_confirmed(exp->master))
159 /* Avoid race with other CPUs, that for exp->master ct, is
160 * about to invoke ->destroy(), or nf_ct_delete() via timeout
163 * The atomic_inc_not_zero() check tells: If that fails, we
164 * know that the ct is being destroyed. If it succeeds, we
165 * can be sure the ct cannot disappear underneath.
167 if (unlikely(nf_ct_is_dying(exp->master) ||
168 !atomic_inc_not_zero(&exp->master->ct_general.use)))
171 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
172 atomic_inc(&exp->use);
174 } else if (del_timer(&exp->timeout)) {
175 nf_ct_unlink_expect(exp);
178 /* Undo exp->master refcnt increase, if del_timer() failed */
179 nf_ct_put(exp->master);
184 /* delete all expectations for this conntrack */
185 void nf_ct_remove_expectations(struct nf_conn *ct)
187 struct nf_conn_help *help = nfct_help(ct);
188 struct nf_conntrack_expect *exp;
189 struct hlist_node *next;
191 /* Optimization: most connection never expect any others. */
195 spin_lock_bh(&nf_conntrack_expect_lock);
196 hlist_for_each_entry_safe(exp, next, &help->expectations, lnode) {
197 if (del_timer(&exp->timeout)) {
198 nf_ct_unlink_expect(exp);
199 nf_ct_expect_put(exp);
202 spin_unlock_bh(&nf_conntrack_expect_lock);
204 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
206 /* Would two expected things clash? */
207 static inline int expect_clash(const struct nf_conntrack_expect *a,
208 const struct nf_conntrack_expect *b)
210 /* Part covered by intersection of masks must be unequal,
211 otherwise they clash */
212 struct nf_conntrack_tuple_mask intersect_mask;
215 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
217 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
218 intersect_mask.src.u3.all[count] =
219 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
222 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask) &&
223 nf_ct_zone(a->master) == nf_ct_zone(b->master);
226 static inline int expect_matches(const struct nf_conntrack_expect *a,
227 const struct nf_conntrack_expect *b)
229 return a->master == b->master && a->class == b->class &&
230 nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
231 nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
232 nf_ct_zone(a->master) == nf_ct_zone(b->master);
235 /* Generally a bad idea to call this: could have matched already. */
236 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
238 spin_lock_bh(&nf_conntrack_expect_lock);
239 if (del_timer(&exp->timeout)) {
240 nf_ct_unlink_expect(exp);
241 nf_ct_expect_put(exp);
243 spin_unlock_bh(&nf_conntrack_expect_lock);
245 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
247 /* We don't increase the master conntrack refcount for non-fulfilled
248 * conntracks. During the conntrack destruction, the expectations are
249 * always killed before the conntrack itself */
250 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
252 struct nf_conntrack_expect *new;
254 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
259 atomic_set(&new->use, 1);
262 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
264 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
266 const union nf_inet_addr *saddr,
267 const union nf_inet_addr *daddr,
268 u_int8_t proto, const __be16 *src, const __be16 *dst)
272 if (family == AF_INET)
279 exp->expectfn = NULL;
281 exp->tuple.src.l3num = family;
282 exp->tuple.dst.protonum = proto;
285 memcpy(&exp->tuple.src.u3, saddr, len);
286 if (sizeof(exp->tuple.src.u3) > len)
287 /* address needs to be cleared for nf_ct_tuple_equal */
288 memset((void *)&exp->tuple.src.u3 + len, 0x00,
289 sizeof(exp->tuple.src.u3) - len);
290 memset(&exp->mask.src.u3, 0xFF, len);
291 if (sizeof(exp->mask.src.u3) > len)
292 memset((void *)&exp->mask.src.u3 + len, 0x00,
293 sizeof(exp->mask.src.u3) - len);
295 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
296 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
300 exp->tuple.src.u.all = *src;
301 exp->mask.src.u.all = htons(0xFFFF);
303 exp->tuple.src.u.all = 0;
304 exp->mask.src.u.all = 0;
307 memcpy(&exp->tuple.dst.u3, daddr, len);
308 if (sizeof(exp->tuple.dst.u3) > len)
309 /* address needs to be cleared for nf_ct_tuple_equal */
310 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
311 sizeof(exp->tuple.dst.u3) - len);
313 exp->tuple.dst.u.all = *dst;
315 #ifdef CONFIG_NF_NAT_NEEDED
316 memset(&exp->saved_addr, 0, sizeof(exp->saved_addr));
317 memset(&exp->saved_proto, 0, sizeof(exp->saved_proto));
320 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
322 static void nf_ct_expect_free_rcu(struct rcu_head *head)
324 struct nf_conntrack_expect *exp;
326 exp = container_of(head, struct nf_conntrack_expect, rcu);
327 kmem_cache_free(nf_ct_expect_cachep, exp);
330 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
332 if (atomic_dec_and_test(&exp->use))
333 call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
335 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
337 static int nf_ct_expect_insert(struct nf_conntrack_expect *exp)
339 struct nf_conn_help *master_help = nfct_help(exp->master);
340 struct nf_conntrack_helper *helper;
341 struct net *net = nf_ct_exp_net(exp);
342 unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
344 /* two references : one for hash insert, one for the timer */
345 atomic_add(2, &exp->use);
347 hlist_add_head(&exp->lnode, &master_help->expectations);
348 master_help->expecting[exp->class]++;
350 hlist_add_head_rcu(&exp->hnode, &net->ct.expect_hash[h]);
351 net->ct.expect_count++;
353 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
355 helper = rcu_dereference_protected(master_help->helper,
356 lockdep_is_held(&nf_conntrack_expect_lock));
358 exp->timeout.expires = jiffies +
359 helper->expect_policy[exp->class].timeout * HZ;
361 add_timer(&exp->timeout);
363 NF_CT_STAT_INC(net, expect_create);
367 /* Race with expectations being used means we could have none to find; OK. */
368 static void evict_oldest_expect(struct nf_conn *master,
369 struct nf_conntrack_expect *new)
371 struct nf_conn_help *master_help = nfct_help(master);
372 struct nf_conntrack_expect *exp, *last = NULL;
374 hlist_for_each_entry(exp, &master_help->expectations, lnode) {
375 if (exp->class == new->class)
379 if (last && del_timer(&last->timeout)) {
380 nf_ct_unlink_expect(last);
381 nf_ct_expect_put(last);
385 static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
387 const struct nf_conntrack_expect_policy *p;
388 struct nf_conntrack_expect *i;
389 struct nf_conn *master = expect->master;
390 struct nf_conn_help *master_help = nfct_help(master);
391 struct nf_conntrack_helper *helper;
392 struct net *net = nf_ct_exp_net(expect);
393 struct hlist_node *next;
401 h = nf_ct_expect_dst_hash(&expect->tuple);
402 hlist_for_each_entry_safe(i, next, &net->ct.expect_hash[h], hnode) {
403 if (expect_matches(i, expect)) {
404 if (del_timer(&i->timeout)) {
405 nf_ct_unlink_expect(i);
409 } else if (expect_clash(i, expect)) {
414 /* Will be over limit? */
415 helper = rcu_dereference_protected(master_help->helper,
416 lockdep_is_held(&nf_conntrack_expect_lock));
418 p = &helper->expect_policy[expect->class];
419 if (p->max_expected &&
420 master_help->expecting[expect->class] >= p->max_expected) {
421 evict_oldest_expect(master, expect);
422 if (master_help->expecting[expect->class]
423 >= p->max_expected) {
430 if (net->ct.expect_count >= nf_ct_expect_max) {
431 net_warn_ratelimited("nf_conntrack: expectation table full\n");
438 int nf_ct_expect_related_report(struct nf_conntrack_expect *expect,
439 u32 portid, int report)
443 spin_lock_bh(&nf_conntrack_expect_lock);
444 ret = __nf_ct_expect_check(expect);
448 ret = nf_ct_expect_insert(expect);
451 spin_unlock_bh(&nf_conntrack_expect_lock);
452 nf_ct_expect_event_report(IPEXP_NEW, expect, portid, report);
455 spin_unlock_bh(&nf_conntrack_expect_lock);
458 EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
460 #ifdef CONFIG_NF_CONNTRACK_PROCFS
461 struct ct_expect_iter_state {
462 struct seq_net_private p;
466 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
468 struct net *net = seq_file_net(seq);
469 struct ct_expect_iter_state *st = seq->private;
470 struct hlist_node *n;
472 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
473 n = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
480 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
481 struct hlist_node *head)
483 struct net *net = seq_file_net(seq);
484 struct ct_expect_iter_state *st = seq->private;
486 head = rcu_dereference(hlist_next_rcu(head));
487 while (head == NULL) {
488 if (++st->bucket >= nf_ct_expect_hsize)
490 head = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
495 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
497 struct hlist_node *head = ct_expect_get_first(seq);
500 while (pos && (head = ct_expect_get_next(seq, head)))
502 return pos ? NULL : head;
505 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
509 return ct_expect_get_idx(seq, *pos);
512 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
515 return ct_expect_get_next(seq, v);
518 static void exp_seq_stop(struct seq_file *seq, void *v)
524 static int exp_seq_show(struct seq_file *s, void *v)
526 struct nf_conntrack_expect *expect;
527 struct nf_conntrack_helper *helper;
528 struct hlist_node *n = v;
531 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
533 if (expect->timeout.function)
534 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
535 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
538 seq_printf(s, "l3proto = %u proto=%u ",
539 expect->tuple.src.l3num,
540 expect->tuple.dst.protonum);
541 print_tuple(s, &expect->tuple,
542 __nf_ct_l3proto_find(expect->tuple.src.l3num),
543 __nf_ct_l4proto_find(expect->tuple.src.l3num,
544 expect->tuple.dst.protonum));
546 if (expect->flags & NF_CT_EXPECT_PERMANENT) {
547 seq_printf(s, "PERMANENT");
550 if (expect->flags & NF_CT_EXPECT_INACTIVE) {
551 seq_printf(s, "%sINACTIVE", delim);
554 if (expect->flags & NF_CT_EXPECT_USERSPACE)
555 seq_printf(s, "%sUSERSPACE", delim);
557 helper = rcu_dereference(nfct_help(expect->master)->helper);
559 seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
560 if (helper->expect_policy[expect->class].name)
562 helper->expect_policy[expect->class].name);
570 static const struct seq_operations exp_seq_ops = {
571 .start = exp_seq_start,
572 .next = exp_seq_next,
573 .stop = exp_seq_stop,
577 static int exp_open(struct inode *inode, struct file *file)
579 return seq_open_net(inode, file, &exp_seq_ops,
580 sizeof(struct ct_expect_iter_state));
583 static const struct file_operations exp_file_ops = {
584 .owner = THIS_MODULE,
588 .release = seq_release_net,
590 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
592 static int exp_proc_init(struct net *net)
594 #ifdef CONFIG_NF_CONNTRACK_PROCFS
595 struct proc_dir_entry *proc;
597 proc = proc_create("nf_conntrack_expect", 0440, net->proc_net,
601 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
605 static void exp_proc_remove(struct net *net)
607 #ifdef CONFIG_NF_CONNTRACK_PROCFS
608 remove_proc_entry("nf_conntrack_expect", net->proc_net);
609 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
612 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
614 int nf_conntrack_expect_pernet_init(struct net *net)
618 net->ct.expect_count = 0;
619 net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
620 if (net->ct.expect_hash == NULL)
623 err = exp_proc_init(net);
629 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
634 void nf_conntrack_expect_pernet_fini(struct net *net)
636 exp_proc_remove(net);
637 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
640 int nf_conntrack_expect_init(void)
642 if (!nf_ct_expect_hsize) {
643 nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
644 if (!nf_ct_expect_hsize)
645 nf_ct_expect_hsize = 1;
647 nf_ct_expect_max = nf_ct_expect_hsize * 4;
648 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
649 sizeof(struct nf_conntrack_expect),
651 if (!nf_ct_expect_cachep)
656 void nf_conntrack_expect_fini(void)
658 rcu_barrier(); /* Wait for call_rcu() before destroy */
659 kmem_cache_destroy(nf_ct_expect_cachep);