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);
225 static inline int expect_matches(const struct nf_conntrack_expect *a,
226 const struct nf_conntrack_expect *b)
228 return a->master == b->master && a->class == b->class &&
229 nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
230 nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
231 nf_ct_zone(a->master) == nf_ct_zone(b->master);
234 /* Generally a bad idea to call this: could have matched already. */
235 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
237 spin_lock_bh(&nf_conntrack_expect_lock);
238 if (del_timer(&exp->timeout)) {
239 nf_ct_unlink_expect(exp);
240 nf_ct_expect_put(exp);
242 spin_unlock_bh(&nf_conntrack_expect_lock);
244 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
246 /* We don't increase the master conntrack refcount for non-fulfilled
247 * conntracks. During the conntrack destruction, the expectations are
248 * always killed before the conntrack itself */
249 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
251 struct nf_conntrack_expect *new;
253 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
258 atomic_set(&new->use, 1);
261 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
263 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
265 const union nf_inet_addr *saddr,
266 const union nf_inet_addr *daddr,
267 u_int8_t proto, const __be16 *src, const __be16 *dst)
271 if (family == AF_INET)
278 exp->expectfn = NULL;
280 exp->tuple.src.l3num = family;
281 exp->tuple.dst.protonum = proto;
284 memcpy(&exp->tuple.src.u3, saddr, len);
285 if (sizeof(exp->tuple.src.u3) > len)
286 /* address needs to be cleared for nf_ct_tuple_equal */
287 memset((void *)&exp->tuple.src.u3 + len, 0x00,
288 sizeof(exp->tuple.src.u3) - len);
289 memset(&exp->mask.src.u3, 0xFF, len);
290 if (sizeof(exp->mask.src.u3) > len)
291 memset((void *)&exp->mask.src.u3 + len, 0x00,
292 sizeof(exp->mask.src.u3) - len);
294 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
295 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
299 exp->tuple.src.u.all = *src;
300 exp->mask.src.u.all = htons(0xFFFF);
302 exp->tuple.src.u.all = 0;
303 exp->mask.src.u.all = 0;
306 memcpy(&exp->tuple.dst.u3, daddr, len);
307 if (sizeof(exp->tuple.dst.u3) > len)
308 /* address needs to be cleared for nf_ct_tuple_equal */
309 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
310 sizeof(exp->tuple.dst.u3) - len);
312 exp->tuple.dst.u.all = *dst;
314 #ifdef CONFIG_NF_NAT_NEEDED
315 memset(&exp->saved_addr, 0, sizeof(exp->saved_addr));
316 memset(&exp->saved_proto, 0, sizeof(exp->saved_proto));
319 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
321 static void nf_ct_expect_free_rcu(struct rcu_head *head)
323 struct nf_conntrack_expect *exp;
325 exp = container_of(head, struct nf_conntrack_expect, rcu);
326 kmem_cache_free(nf_ct_expect_cachep, exp);
329 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
331 if (atomic_dec_and_test(&exp->use))
332 call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
334 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
336 static int nf_ct_expect_insert(struct nf_conntrack_expect *exp)
338 struct nf_conn_help *master_help = nfct_help(exp->master);
339 struct nf_conntrack_helper *helper;
340 struct net *net = nf_ct_exp_net(exp);
341 unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
343 /* two references : one for hash insert, one for the timer */
344 atomic_add(2, &exp->use);
346 hlist_add_head(&exp->lnode, &master_help->expectations);
347 master_help->expecting[exp->class]++;
349 hlist_add_head_rcu(&exp->hnode, &net->ct.expect_hash[h]);
350 net->ct.expect_count++;
352 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
354 helper = rcu_dereference_protected(master_help->helper,
355 lockdep_is_held(&nf_conntrack_expect_lock));
357 exp->timeout.expires = jiffies +
358 helper->expect_policy[exp->class].timeout * HZ;
360 add_timer(&exp->timeout);
362 NF_CT_STAT_INC(net, expect_create);
366 /* Race with expectations being used means we could have none to find; OK. */
367 static void evict_oldest_expect(struct nf_conn *master,
368 struct nf_conntrack_expect *new)
370 struct nf_conn_help *master_help = nfct_help(master);
371 struct nf_conntrack_expect *exp, *last = NULL;
373 hlist_for_each_entry(exp, &master_help->expectations, lnode) {
374 if (exp->class == new->class)
378 if (last && del_timer(&last->timeout)) {
379 nf_ct_unlink_expect(last);
380 nf_ct_expect_put(last);
384 static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
386 const struct nf_conntrack_expect_policy *p;
387 struct nf_conntrack_expect *i;
388 struct nf_conn *master = expect->master;
389 struct nf_conn_help *master_help = nfct_help(master);
390 struct nf_conntrack_helper *helper;
391 struct net *net = nf_ct_exp_net(expect);
392 struct hlist_node *next;
400 h = nf_ct_expect_dst_hash(&expect->tuple);
401 hlist_for_each_entry_safe(i, next, &net->ct.expect_hash[h], hnode) {
402 if (expect_matches(i, expect)) {
403 if (del_timer(&i->timeout)) {
404 nf_ct_unlink_expect(i);
408 } else if (expect_clash(i, expect)) {
413 /* Will be over limit? */
414 helper = rcu_dereference_protected(master_help->helper,
415 lockdep_is_held(&nf_conntrack_expect_lock));
417 p = &helper->expect_policy[expect->class];
418 if (p->max_expected &&
419 master_help->expecting[expect->class] >= p->max_expected) {
420 evict_oldest_expect(master, expect);
421 if (master_help->expecting[expect->class]
422 >= p->max_expected) {
429 if (net->ct.expect_count >= nf_ct_expect_max) {
430 net_warn_ratelimited("nf_conntrack: expectation table full\n");
437 int nf_ct_expect_related_report(struct nf_conntrack_expect *expect,
438 u32 portid, int report)
442 spin_lock_bh(&nf_conntrack_expect_lock);
443 ret = __nf_ct_expect_check(expect);
447 ret = nf_ct_expect_insert(expect);
450 spin_unlock_bh(&nf_conntrack_expect_lock);
451 nf_ct_expect_event_report(IPEXP_NEW, expect, portid, report);
454 spin_unlock_bh(&nf_conntrack_expect_lock);
457 EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
459 #ifdef CONFIG_NF_CONNTRACK_PROCFS
460 struct ct_expect_iter_state {
461 struct seq_net_private p;
465 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
467 struct net *net = seq_file_net(seq);
468 struct ct_expect_iter_state *st = seq->private;
469 struct hlist_node *n;
471 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
472 n = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
479 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
480 struct hlist_node *head)
482 struct net *net = seq_file_net(seq);
483 struct ct_expect_iter_state *st = seq->private;
485 head = rcu_dereference(hlist_next_rcu(head));
486 while (head == NULL) {
487 if (++st->bucket >= nf_ct_expect_hsize)
489 head = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
494 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
496 struct hlist_node *head = ct_expect_get_first(seq);
499 while (pos && (head = ct_expect_get_next(seq, head)))
501 return pos ? NULL : head;
504 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
508 return ct_expect_get_idx(seq, *pos);
511 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
514 return ct_expect_get_next(seq, v);
517 static void exp_seq_stop(struct seq_file *seq, void *v)
523 static int exp_seq_show(struct seq_file *s, void *v)
525 struct nf_conntrack_expect *expect;
526 struct nf_conntrack_helper *helper;
527 struct hlist_node *n = v;
530 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
532 if (expect->timeout.function)
533 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
534 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
537 seq_printf(s, "l3proto = %u proto=%u ",
538 expect->tuple.src.l3num,
539 expect->tuple.dst.protonum);
540 print_tuple(s, &expect->tuple,
541 __nf_ct_l3proto_find(expect->tuple.src.l3num),
542 __nf_ct_l4proto_find(expect->tuple.src.l3num,
543 expect->tuple.dst.protonum));
545 if (expect->flags & NF_CT_EXPECT_PERMANENT) {
546 seq_printf(s, "PERMANENT");
549 if (expect->flags & NF_CT_EXPECT_INACTIVE) {
550 seq_printf(s, "%sINACTIVE", delim);
553 if (expect->flags & NF_CT_EXPECT_USERSPACE)
554 seq_printf(s, "%sUSERSPACE", delim);
556 helper = rcu_dereference(nfct_help(expect->master)->helper);
558 seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
559 if (helper->expect_policy[expect->class].name)
561 helper->expect_policy[expect->class].name);
564 return seq_putc(s, '\n');
567 static const struct seq_operations exp_seq_ops = {
568 .start = exp_seq_start,
569 .next = exp_seq_next,
570 .stop = exp_seq_stop,
574 static int exp_open(struct inode *inode, struct file *file)
576 return seq_open_net(inode, file, &exp_seq_ops,
577 sizeof(struct ct_expect_iter_state));
580 static const struct file_operations exp_file_ops = {
581 .owner = THIS_MODULE,
585 .release = seq_release_net,
587 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
589 static int exp_proc_init(struct net *net)
591 #ifdef CONFIG_NF_CONNTRACK_PROCFS
592 struct proc_dir_entry *proc;
594 proc = proc_create("nf_conntrack_expect", 0440, net->proc_net,
598 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
602 static void exp_proc_remove(struct net *net)
604 #ifdef CONFIG_NF_CONNTRACK_PROCFS
605 remove_proc_entry("nf_conntrack_expect", net->proc_net);
606 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
609 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
611 int nf_conntrack_expect_pernet_init(struct net *net)
615 net->ct.expect_count = 0;
616 net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
617 if (net->ct.expect_hash == NULL)
620 err = exp_proc_init(net);
626 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
631 void nf_conntrack_expect_pernet_fini(struct net *net)
633 exp_proc_remove(net);
634 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
637 int nf_conntrack_expect_init(void)
639 if (!nf_ct_expect_hsize) {
640 nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
641 if (!nf_ct_expect_hsize)
642 nf_ct_expect_hsize = 1;
644 nf_ct_expect_max = nf_ct_expect_hsize * 4;
645 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
646 sizeof(struct nf_conntrack_expect),
648 if (!nf_ct_expect_cachep)
653 void nf_conntrack_expect_fini(void)
655 rcu_barrier(); /* Wait for call_rcu() before destroy */
656 kmem_cache_destroy(nf_ct_expect_cachep);