1 /* Expectation handling for nf_conntrack. */
3 /* (C) 1999-2001 Paul `Rusty' Russell
5 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
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,
92 const struct nf_conntrack_zone *zone,
93 const struct nf_conntrack_tuple *tuple)
95 struct nf_conntrack_expect *i;
98 if (!net->ct.expect_count)
101 h = nf_ct_expect_dst_hash(tuple);
102 hlist_for_each_entry_rcu(i, &net->ct.expect_hash[h], hnode) {
103 if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
104 nf_ct_zone_equal_any(i->master, zone))
109 EXPORT_SYMBOL_GPL(__nf_ct_expect_find);
111 /* Just find a expectation corresponding to a tuple. */
112 struct nf_conntrack_expect *
113 nf_ct_expect_find_get(struct net *net,
114 const struct nf_conntrack_zone *zone,
115 const struct nf_conntrack_tuple *tuple)
117 struct nf_conntrack_expect *i;
120 i = __nf_ct_expect_find(net, zone, tuple);
121 if (i && !atomic_inc_not_zero(&i->use))
127 EXPORT_SYMBOL_GPL(nf_ct_expect_find_get);
129 /* If an expectation for this connection is found, it gets delete from
130 * global list then returned. */
131 struct nf_conntrack_expect *
132 nf_ct_find_expectation(struct net *net,
133 const struct nf_conntrack_zone *zone,
134 const struct nf_conntrack_tuple *tuple)
136 struct nf_conntrack_expect *i, *exp = NULL;
139 if (!net->ct.expect_count)
142 h = nf_ct_expect_dst_hash(tuple);
143 hlist_for_each_entry(i, &net->ct.expect_hash[h], hnode) {
144 if (!(i->flags & NF_CT_EXPECT_INACTIVE) &&
145 nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask) &&
146 nf_ct_zone_equal_any(i->master, zone)) {
154 /* If master is not in hash table yet (ie. packet hasn't left
155 this machine yet), how can other end know about expected?
156 Hence these are not the droids you are looking for (if
157 master ct never got confirmed, we'd hold a reference to it
158 and weird things would happen to future packets). */
159 if (!nf_ct_is_confirmed(exp->master))
162 /* Avoid race with other CPUs, that for exp->master ct, is
163 * about to invoke ->destroy(), or nf_ct_delete() via timeout
166 * The atomic_inc_not_zero() check tells: If that fails, we
167 * know that the ct is being destroyed. If it succeeds, we
168 * can be sure the ct cannot disappear underneath.
170 if (unlikely(nf_ct_is_dying(exp->master) ||
171 !atomic_inc_not_zero(&exp->master->ct_general.use)))
174 if (exp->flags & NF_CT_EXPECT_PERMANENT) {
175 atomic_inc(&exp->use);
177 } else if (del_timer(&exp->timeout)) {
178 nf_ct_unlink_expect(exp);
181 /* Undo exp->master refcnt increase, if del_timer() failed */
182 nf_ct_put(exp->master);
187 /* delete all expectations for this conntrack */
188 void nf_ct_remove_expectations(struct nf_conn *ct)
190 struct nf_conn_help *help = nfct_help(ct);
191 struct nf_conntrack_expect *exp;
192 struct hlist_node *next;
194 /* Optimization: most connection never expect any others. */
198 spin_lock_bh(&nf_conntrack_expect_lock);
199 hlist_for_each_entry_safe(exp, next, &help->expectations, lnode) {
200 if (del_timer(&exp->timeout)) {
201 nf_ct_unlink_expect(exp);
202 nf_ct_expect_put(exp);
205 spin_unlock_bh(&nf_conntrack_expect_lock);
207 EXPORT_SYMBOL_GPL(nf_ct_remove_expectations);
209 /* Would two expected things clash? */
210 static inline int expect_clash(const struct nf_conntrack_expect *a,
211 const struct nf_conntrack_expect *b)
213 /* Part covered by intersection of masks must be unequal,
214 otherwise they clash */
215 struct nf_conntrack_tuple_mask intersect_mask;
218 intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
220 for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
221 intersect_mask.src.u3.all[count] =
222 a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
225 return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask) &&
226 nf_ct_zone_equal_any(a->master, nf_ct_zone(b->master));
229 static inline int expect_matches(const struct nf_conntrack_expect *a,
230 const struct nf_conntrack_expect *b)
232 return a->master == b->master && a->class == b->class &&
233 nf_ct_tuple_equal(&a->tuple, &b->tuple) &&
234 nf_ct_tuple_mask_equal(&a->mask, &b->mask) &&
235 nf_ct_zone_equal_any(a->master, nf_ct_zone(b->master));
238 /* Generally a bad idea to call this: could have matched already. */
239 void nf_ct_unexpect_related(struct nf_conntrack_expect *exp)
241 spin_lock_bh(&nf_conntrack_expect_lock);
242 if (del_timer(&exp->timeout)) {
243 nf_ct_unlink_expect(exp);
244 nf_ct_expect_put(exp);
246 spin_unlock_bh(&nf_conntrack_expect_lock);
248 EXPORT_SYMBOL_GPL(nf_ct_unexpect_related);
250 /* We don't increase the master conntrack refcount for non-fulfilled
251 * conntracks. During the conntrack destruction, the expectations are
252 * always killed before the conntrack itself */
253 struct nf_conntrack_expect *nf_ct_expect_alloc(struct nf_conn *me)
255 struct nf_conntrack_expect *new;
257 new = kmem_cache_alloc(nf_ct_expect_cachep, GFP_ATOMIC);
262 atomic_set(&new->use, 1);
265 EXPORT_SYMBOL_GPL(nf_ct_expect_alloc);
267 void nf_ct_expect_init(struct nf_conntrack_expect *exp, unsigned int class,
269 const union nf_inet_addr *saddr,
270 const union nf_inet_addr *daddr,
271 u_int8_t proto, const __be16 *src, const __be16 *dst)
275 if (family == AF_INET)
282 exp->expectfn = NULL;
284 exp->tuple.src.l3num = family;
285 exp->tuple.dst.protonum = proto;
288 memcpy(&exp->tuple.src.u3, saddr, len);
289 if (sizeof(exp->tuple.src.u3) > len)
290 /* address needs to be cleared for nf_ct_tuple_equal */
291 memset((void *)&exp->tuple.src.u3 + len, 0x00,
292 sizeof(exp->tuple.src.u3) - len);
293 memset(&exp->mask.src.u3, 0xFF, len);
294 if (sizeof(exp->mask.src.u3) > len)
295 memset((void *)&exp->mask.src.u3 + len, 0x00,
296 sizeof(exp->mask.src.u3) - len);
298 memset(&exp->tuple.src.u3, 0x00, sizeof(exp->tuple.src.u3));
299 memset(&exp->mask.src.u3, 0x00, sizeof(exp->mask.src.u3));
303 exp->tuple.src.u.all = *src;
304 exp->mask.src.u.all = htons(0xFFFF);
306 exp->tuple.src.u.all = 0;
307 exp->mask.src.u.all = 0;
310 memcpy(&exp->tuple.dst.u3, daddr, len);
311 if (sizeof(exp->tuple.dst.u3) > len)
312 /* address needs to be cleared for nf_ct_tuple_equal */
313 memset((void *)&exp->tuple.dst.u3 + len, 0x00,
314 sizeof(exp->tuple.dst.u3) - len);
316 exp->tuple.dst.u.all = *dst;
318 #ifdef CONFIG_NF_NAT_NEEDED
319 memset(&exp->saved_addr, 0, sizeof(exp->saved_addr));
320 memset(&exp->saved_proto, 0, sizeof(exp->saved_proto));
323 EXPORT_SYMBOL_GPL(nf_ct_expect_init);
325 static void nf_ct_expect_free_rcu(struct rcu_head *head)
327 struct nf_conntrack_expect *exp;
329 exp = container_of(head, struct nf_conntrack_expect, rcu);
330 kmem_cache_free(nf_ct_expect_cachep, exp);
333 void nf_ct_expect_put(struct nf_conntrack_expect *exp)
335 if (atomic_dec_and_test(&exp->use))
336 call_rcu(&exp->rcu, nf_ct_expect_free_rcu);
338 EXPORT_SYMBOL_GPL(nf_ct_expect_put);
340 static int nf_ct_expect_insert(struct nf_conntrack_expect *exp)
342 struct nf_conn_help *master_help = nfct_help(exp->master);
343 struct nf_conntrack_helper *helper;
344 struct net *net = nf_ct_exp_net(exp);
345 unsigned int h = nf_ct_expect_dst_hash(&exp->tuple);
347 /* two references : one for hash insert, one for the timer */
348 atomic_add(2, &exp->use);
350 hlist_add_head(&exp->lnode, &master_help->expectations);
351 master_help->expecting[exp->class]++;
353 hlist_add_head_rcu(&exp->hnode, &net->ct.expect_hash[h]);
354 net->ct.expect_count++;
356 setup_timer(&exp->timeout, nf_ct_expectation_timed_out,
358 helper = rcu_dereference_protected(master_help->helper,
359 lockdep_is_held(&nf_conntrack_expect_lock));
361 exp->timeout.expires = jiffies +
362 helper->expect_policy[exp->class].timeout * HZ;
364 add_timer(&exp->timeout);
366 NF_CT_STAT_INC(net, expect_create);
370 /* Race with expectations being used means we could have none to find; OK. */
371 static void evict_oldest_expect(struct nf_conn *master,
372 struct nf_conntrack_expect *new)
374 struct nf_conn_help *master_help = nfct_help(master);
375 struct nf_conntrack_expect *exp, *last = NULL;
377 hlist_for_each_entry(exp, &master_help->expectations, lnode) {
378 if (exp->class == new->class)
382 if (last && del_timer(&last->timeout)) {
383 nf_ct_unlink_expect(last);
384 nf_ct_expect_put(last);
388 static inline int __nf_ct_expect_check(struct nf_conntrack_expect *expect)
390 const struct nf_conntrack_expect_policy *p;
391 struct nf_conntrack_expect *i;
392 struct nf_conn *master = expect->master;
393 struct nf_conn_help *master_help = nfct_help(master);
394 struct nf_conntrack_helper *helper;
395 struct net *net = nf_ct_exp_net(expect);
396 struct hlist_node *next;
404 h = nf_ct_expect_dst_hash(&expect->tuple);
405 hlist_for_each_entry_safe(i, next, &net->ct.expect_hash[h], hnode) {
406 if (expect_matches(i, expect)) {
407 if (del_timer(&i->timeout)) {
408 nf_ct_unlink_expect(i);
412 } else if (expect_clash(i, expect)) {
417 /* Will be over limit? */
418 helper = rcu_dereference_protected(master_help->helper,
419 lockdep_is_held(&nf_conntrack_expect_lock));
421 p = &helper->expect_policy[expect->class];
422 if (p->max_expected &&
423 master_help->expecting[expect->class] >= p->max_expected) {
424 evict_oldest_expect(master, expect);
425 if (master_help->expecting[expect->class]
426 >= p->max_expected) {
433 if (net->ct.expect_count >= nf_ct_expect_max) {
434 net_warn_ratelimited("nf_conntrack: expectation table full\n");
441 int nf_ct_expect_related_report(struct nf_conntrack_expect *expect,
442 u32 portid, int report)
446 spin_lock_bh(&nf_conntrack_expect_lock);
447 ret = __nf_ct_expect_check(expect);
451 ret = nf_ct_expect_insert(expect);
454 spin_unlock_bh(&nf_conntrack_expect_lock);
455 nf_ct_expect_event_report(IPEXP_NEW, expect, portid, report);
458 spin_unlock_bh(&nf_conntrack_expect_lock);
461 EXPORT_SYMBOL_GPL(nf_ct_expect_related_report);
463 #ifdef CONFIG_NF_CONNTRACK_PROCFS
464 struct ct_expect_iter_state {
465 struct seq_net_private p;
469 static struct hlist_node *ct_expect_get_first(struct seq_file *seq)
471 struct net *net = seq_file_net(seq);
472 struct ct_expect_iter_state *st = seq->private;
473 struct hlist_node *n;
475 for (st->bucket = 0; st->bucket < nf_ct_expect_hsize; st->bucket++) {
476 n = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
483 static struct hlist_node *ct_expect_get_next(struct seq_file *seq,
484 struct hlist_node *head)
486 struct net *net = seq_file_net(seq);
487 struct ct_expect_iter_state *st = seq->private;
489 head = rcu_dereference(hlist_next_rcu(head));
490 while (head == NULL) {
491 if (++st->bucket >= nf_ct_expect_hsize)
493 head = rcu_dereference(hlist_first_rcu(&net->ct.expect_hash[st->bucket]));
498 static struct hlist_node *ct_expect_get_idx(struct seq_file *seq, loff_t pos)
500 struct hlist_node *head = ct_expect_get_first(seq);
503 while (pos && (head = ct_expect_get_next(seq, head)))
505 return pos ? NULL : head;
508 static void *exp_seq_start(struct seq_file *seq, loff_t *pos)
512 return ct_expect_get_idx(seq, *pos);
515 static void *exp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
518 return ct_expect_get_next(seq, v);
521 static void exp_seq_stop(struct seq_file *seq, void *v)
527 static int exp_seq_show(struct seq_file *s, void *v)
529 struct nf_conntrack_expect *expect;
530 struct nf_conntrack_helper *helper;
531 struct hlist_node *n = v;
534 expect = hlist_entry(n, struct nf_conntrack_expect, hnode);
536 if (expect->timeout.function)
537 seq_printf(s, "%ld ", timer_pending(&expect->timeout)
538 ? (long)(expect->timeout.expires - jiffies)/HZ : 0);
541 seq_printf(s, "l3proto = %u proto=%u ",
542 expect->tuple.src.l3num,
543 expect->tuple.dst.protonum);
544 print_tuple(s, &expect->tuple,
545 __nf_ct_l3proto_find(expect->tuple.src.l3num),
546 __nf_ct_l4proto_find(expect->tuple.src.l3num,
547 expect->tuple.dst.protonum));
549 if (expect->flags & NF_CT_EXPECT_PERMANENT) {
550 seq_printf(s, "PERMANENT");
553 if (expect->flags & NF_CT_EXPECT_INACTIVE) {
554 seq_printf(s, "%sINACTIVE", delim);
557 if (expect->flags & NF_CT_EXPECT_USERSPACE)
558 seq_printf(s, "%sUSERSPACE", delim);
560 helper = rcu_dereference(nfct_help(expect->master)->helper);
562 seq_printf(s, "%s%s", expect->flags ? " " : "", helper->name);
563 if (helper->expect_policy[expect->class].name)
565 helper->expect_policy[expect->class].name);
573 static const struct seq_operations exp_seq_ops = {
574 .start = exp_seq_start,
575 .next = exp_seq_next,
576 .stop = exp_seq_stop,
580 static int exp_open(struct inode *inode, struct file *file)
582 return seq_open_net(inode, file, &exp_seq_ops,
583 sizeof(struct ct_expect_iter_state));
586 static const struct file_operations exp_file_ops = {
587 .owner = THIS_MODULE,
591 .release = seq_release_net,
593 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
595 static int exp_proc_init(struct net *net)
597 #ifdef CONFIG_NF_CONNTRACK_PROCFS
598 struct proc_dir_entry *proc;
602 proc = proc_create("nf_conntrack_expect", 0440, net->proc_net,
607 root_uid = make_kuid(net->user_ns, 0);
608 root_gid = make_kgid(net->user_ns, 0);
609 if (uid_valid(root_uid) && gid_valid(root_gid))
610 proc_set_user(proc, root_uid, root_gid);
611 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
615 static void exp_proc_remove(struct net *net)
617 #ifdef CONFIG_NF_CONNTRACK_PROCFS
618 remove_proc_entry("nf_conntrack_expect", net->proc_net);
619 #endif /* CONFIG_NF_CONNTRACK_PROCFS */
622 module_param_named(expect_hashsize, nf_ct_expect_hsize, uint, 0400);
624 int nf_conntrack_expect_pernet_init(struct net *net)
628 net->ct.expect_count = 0;
629 net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize, 0);
630 if (net->ct.expect_hash == NULL)
633 err = exp_proc_init(net);
639 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
644 void nf_conntrack_expect_pernet_fini(struct net *net)
646 exp_proc_remove(net);
647 nf_ct_free_hashtable(net->ct.expect_hash, nf_ct_expect_hsize);
650 int nf_conntrack_expect_init(void)
652 if (!nf_ct_expect_hsize) {
653 nf_ct_expect_hsize = nf_conntrack_htable_size / 256;
654 if (!nf_ct_expect_hsize)
655 nf_ct_expect_hsize = 1;
657 nf_ct_expect_max = nf_ct_expect_hsize * 4;
658 nf_ct_expect_cachep = kmem_cache_create("nf_conntrack_expect",
659 sizeof(struct nf_conntrack_expect),
661 if (!nf_ct_expect_cachep)
666 void nf_conntrack_expect_fini(void)
668 rcu_barrier(); /* Wait for call_rcu() before destroy */
669 kmem_cache_destroy(nf_ct_expect_cachep);